STIGMA: THE PSYCHOLOGY AND ECONOMICS OF SUPERFUND
Prepared by:
William Schulze, Project Director
Kent Messer, Katherine Hackett
Department of Applied Economics and Management
Cornell University
Ithaca, New York 14853
Trudy Cameron, Graham Crawford
University of Oregon
Eugene, Oregon 97403
Gary McClelland
University of Colorado
Boulder, Colorado 80309
Preparedfor:
U.S. ENVIRONMENTAL PROTECTION AGENCY
CR 824393-01-0
July 2004
Project Officer
Dr. Alan Carlin
National Center for Environmental Economics
Office of Policy, Economics and Innovation
U.S. Environmental Protection Agency
Washington, DC 20460
* This research was supported by the USEPA under cooperative agreement CR 824393-01-0. We do wish to thank
Alan Carlin for his patience and support and Kip Viscusi lor his thoughtful comments. We also would like to thank
Christian Coerds, Rachel Deming, Brian Hurd, Eleanor Smith, and Matt Todaro for their support on this project and
the participants of the Risk Perception, Valuation and Policy conference at the University of Central Florida and the
2004 AERE Workshop in Estes Park, Colorado for their helpful feedback.
-------
DISCLAIMER
Although prepared with partial EPA funding, this report has neither been reviewed nor
approved by the U.S. Environmental Protection Agency for publication as an EPA report. The
contents do not necessarily reflect the views or policies of the U.S. Environmental Protection
Agency, nor does mention of trade names or commercial products constitute endorsement or
recommendation for use.
-------
TABLE OF CONTENTS
ABSTRACT 7
CHAPTER 1 OVERVIEW AND EXECUTIVE SUMMARY 8
1.1 Introduction 8
1.2 Case Studies 9
1.3 Expert Error 18
1.4 Events, Perceptual Cues, Risk Perception, and Stigma 21
1.5 Stigma and Property Values 24
1.6 Policy Implications 40
CHAPTER 2 HISTORY OF CURRENT SUPERFUND LEGISLATION 44
2.1 Overview of Superfund Legislation 44
2.2 Legislative Background 45
2.3 Comprehensive Environmental Response, Compensation, and Liability Act 47
2.4 Implementation of Superfund: 1980-1985 48
2.5 1985: The Expiration of Superfund 50
2.6 Superfund Amendments and Reauthorizations 51
2.7 Superfund Reforms and Successes 53
2.8 Conclusion 55
CHAPTER 3 OPERATING INDUSTRIES, INC. LANDFILL 56
3.1 Overview 56
3.2 History of the Landfill 59
CHAPTER 4 WOBURN, MASSACHUSETTS 69
4.1 Overview 69
4.2 History of Wobum and its Superfund Sites 72
CHAPTER 5 MONTCLAIR, NEW JERSEY 91
5.1 Overview 91
5.2 Timeline and History 92
CHAPTER 6 EAGLE MINE, COLORADO 105
6.1 Overview 105
6.2 History and Timeline 107
CHAPTER 7 EXPERT ERROR AND THE PSYCHOLOGY OF RISK AND STIGMA... 120
7.1 Expert Error 120
7.1.1 Love Canal, Niagara, New York 121
7.1.2 Times Beach, Missouri 125
7.1.3 The Defective Dalkon Shield 128
7.1.4 The Discovery of Cold Fusion 131
7.1.5 The Failure of Biosphere 2 132
7.1.6 The Three Mile Island Accident 134
7.1.7 Union Carbide Accident in Bhopal, India 137
7.2 Contradictory Information in the News 140
7.3 Events, Perceptual Cues, Risk Perception, and Stigma 142
-------
CHAPTER 8 PROPERTY VALUE, APPROACH, AND DATA 145
8.1 Introduction |. 145
8.1.1 Objective versus Subjective Risk 147
8.1.2 Distance Effects over Time 147
8.1.3 Endogenous Socio-demographics 149
8.1.4 Endogenous Housing Stock Attributes 150
8.1.5 Environmental Justice/Equity 151
8.2 The Sample 152
8.2.1 Descriptive Statistics, Exclusions !, 154
8.2.2 Extent of the Market 155
8.3 Hedonic Property- Value Models 156
8.4 Control Variables ''. 158
8.4.1 Annual Dummy Variables 158
8.4.2 Distance to the Superfund Site 159
8.4.3 Housing Characteristics 160
8.4.4 Neighborhood Characteristics 163
8.4.5 Other Local Amenities and Disamenities 166
CHAPTER 9 PROPERTY VALUE RESULTS J 173
9.1 Classes of Hedonic Property Value Models 173
9.2 Auxiliary Models Time-Varying Demographic Patterns 174
9.2.1 Montclair ! 180
9.2.2 OH 182
9.2.3 Wobum 183
9.2.4 Eagle Mine 184
9.2.5 Synthesis 184
9.3 Auxiliary Models: Time-Varying Housing Attributes 185
9.3.1 Montclair ; 186
9.3.2 OH 187
9.3.3 Woburn 188
9.3.4 Eagle Mine 188
9.3.5 Synthesis 189
9.4 Hedonic Property Value Models with Time-Varying Proximity Effects 189
9.4.1 Montclair i 190
9.4.2 Oil 196
9.4.3 Wobum 201
9.4.4 Eagle Mine 208
9.5 Synthesis and Conclusions 212
CHAPTER 10 CONCLUSION: STIGMA AND PROPERTY VALUES 214
CHAPTER 11 REFERENCES '. 231
APPENDIX A-MONTCLAIR 242
APPENDIX B-OII LANDFILL 284
APPENDIX C-WOBURN 327
APPENDIX D - EAGLE MINE 365
-------
TABLES
Table 1.1 Key Dates and Statistics
Table 1,2 Coefficient Determinants
Table 1.3 Number and Description of Events
Table 1.4 Psychological Model, Dependent Variable '
Table 1.5 Cleanup Scenarios
Table 8.1 Montclair Housing Characteristics
Table 8.2 Oil Housing Characteristics
Table 8.3 Wobum Housing Characteristics
Table 8.4 Eagle Mine Housing Characteristics
Table 8.5 Neighborhood Characteristic Variables
Table 9.1 Montclair Census Tract Proportion Coefficient
Table 9.2 Oil Census Tract Proportion Coefficients
Table 9.3 Woburn Census Tract Proportion Coefficients
Table 9.4 Montclair Housing Attribute Coefficient
Table 9.5 Oil Housing Attribute Coefficient
Table 9.6 Woburn Housing Attribute Coefficients
Table 9.7 Montclair
Table 9.8 Montclair (with lot size interactions)
Table 9.9 Oil Landfill
Table 9.10 Oil Landfill (with lot size interactions)
Table 9.11 Wobum
Table 9.12 Eagle Mine
Table 10.1 Distance Coefficients
Table 10.2 Number and Description of Events
Table 10.3 Psychological Model, Dependent Variable R' ~
Table 10.4 Cleanup Scenarios
11
31
35
..37
..41
161
162
162
163
164
180
182
183
186
187
188
191
194
197
199
202
211
220
223
225
229
-------
FlGURpS
Figure 1.1 The Effect of Stigma on Equilibrium Housing Prices 26
Figure 1.2 Discriminative Auction Market 27
Figure 1.3 Relative Property Value over Time for Wobum, Massachusetts 34
Figure 1.4 Relative Property Value over Time for Oil Landfill, California 34
Figure 1.5 Relative Property Value over Time for Montclair, New Jersey (outside of area) 34
Figure 1.6 Relative Property Value over Time for Eagle Mine, Colorado 38
Figure 1.7 Relative Property Value over Time for Montclair, New Jersey (inside of area) 39
Figure 1.8 Relative Property Value over Time for Woburn, Massachusetts with and without
socio-demographic variables j 40
Figure 1.9 Policy Simulations Using the Oil Landfill History 42
Figure 2.1 Superfund Budget (1981-2004) 45
Figure 3.1 OH Landfill Vicinity 56
Figure 3.2 Oil Landfill : 57
Figure 4.1 Woburn Vicinity 69
Figure 4.2 Industri-Plex and Wells G&H Sites 70
Figure 5.1 West Orange, Montclair, Glen Ridge Sites 92
Figure 6.1 Eagle Mine Site 106
Figure 9.1 Changes in Socio-demographics near Superfund site over time 178
Figure 9.2 Woburn Model 4 206
Figure 10.1 The Effect of Stigma on Equilibrium Housing Prices 215
Figure 10.2 Discriminative Auction Market 217
Figure 10.3 Relative Property Value over Time for Oil Landfill, California 221
Figure 10.4 Relative Property Value over Time for Montclair, New Jersey (outside of area)... 221
Figure 10.5 Relative Property' Value over Time for Woburn, Massachusetts 221
Figure 10.6 Relative Property Value over Time for Montclair, New Jersey (inside of area) 227
Figure 10.7 Relative Property Value over Time for Eagle Mine, Colorado 227
Figure 10.8 Relative Property Value over Time for Woburn, Massachusetts with and without
socio-demographic variables 227
Figure 10.9 Policy Simulations using the Oil Landfill History 228
-------
Abstract
This study documents the long-term impacts of Superfund cleanup on property values in
communities neighboring prominent Superfund sites. To understand the impacts, one must
integrate the psychology of risk perceptions and stigma with the economics of property values
that capture those perceptions. The research specifically examines the sale prices of nearly
35,000 homes for up to a thirty-year period near six very large Superfund sites. To our
knowledge, no property value studies have examined sites in multiple areas with large property
value losses over the length of time used here. The results we obtain for these very large sites are
both surprising and inconsistent with most prior work. The principal result is it that, when
cleanup is delayed for ten, fifteen, and even up to twenty years, the discounted present value of
the cleanup is mostly lost, most likely because sites are stigmatized and the homes in the
surrounding communities are shunned. The psychological model developed suggests that, for
very large sites, expedited cleanup and simplifying the process to reduce the number of
stigmatizing events that attract attention to sites would reduce property losses.
-------
Chapter 1
i
Overview and Executive Summary
1.1 Introduction
This study attempts to evaluate the benefits (as captured in residential property values) of
hazardous waste cleanup conducted under the Comprehensive Environmental Response,
Compensation, and Liability Act (CERCLA), commonly known as Superfund. When this
legislation was passed in 1983, following Love Canal, the public imagined that the
Environmental Protection Agency (EPA) would begin immediate cleanup of sites deemed
hazardous to human and environmental health, using tax money collected from the petroleum
and chemical industries. However, CERCLA's provision of joint and several liability requires
that all previous and current owners could be responsible for cleanup cost, regardless of the
amount of hazardous waste deposited at the site. Thus the legal complexity of CERCLA in
establishing fair and just responsibility substantially delayed cleanup at many listed Superfund
sites (as described in detail in Chapter 2, which provides a brief history of Superfund).
This research documents the consequences of that delay on property values in
communities neighboring prominent Superfund sites. To understand those consequences, one
must integratp the psychology of risk perceptions and stigma with the economics of property
values that capture those perceptions. To explore the possibility that stigma can help explain
public reaction to potentially hazardous sites, six Superfund sites in four geographic areas are
examined: the Operating Industries, Inc. landfill site near the communities of Monterey Park and
Montebello, California; the radium pollution in Montclair, Glen Ridge, and East/West Orange
Townships in northern New Jersey; the Industri-Plex and water Wells G & H in Woburn,
Massachusetts, and the Eagle Mine outside Vail, Colorado. The research specifically examines
the sale prices of nearly 35,000 homes for up to a thjrty-year period, and describes the history of
each site. It should be noted that many Superfund sites have shown no or small property value
losses in surrounding communities. The sites selected for this study all have shown large losses
at some point in time. Furthermore, to our knowledge, no prior property value study has
examined sites in multiple areas with large property value losses over the length of time used
-------
9
here. The results we obtain are both surprising and inconsistent with most prior work that looks
at shorter time periods (e.g., McClelland, Schulze and Hurd 1990; Gayer, Hamilton, and Viscusi,
2000; Gayer and Viscusi, 2002). For our prominent sites, one can draw a variety of conclusions
depending on what part of the history of property values are examined. Our results are more
consistent with studies that look beyond the complete cleanup which suggest property values
may only recover after cleanup is complete (Kohlhase, 1991; Dale et al., 1999).
In summary, the principal result is it that over the long term, when cleanup is delayed for
ten, fifteen, and even up to twenty years, the discounted present value of benefits of cleanup are
mostly lost because sites are stigmatized and the homes in the surrounding communities are
shunned. Additionally, the research documents how trends in the socio-demographic
composition of the communities near the sites differed from the trends in communities farther
from the site.
This chapter summarizes the key findings of the study and is organized as follows. The
second section briefly describes the six Superfund sites in four geographic areas throughout the
U.S. The third section discusses why residents of communities neighboring Superfund sites may
not completely believe the opinion of scientific experts regarding the health risks associated with
the sites. The fourth section outlines what is known about the psychology of risk perceptions and
stigma. The fifth section integrates the psychology of stigma with economic hedonic property
value approach which, as noted by Adams and Cantor (2001), is anontrivial task. Finally, the
sixth section presents our conclusions.
1.2 Case Studies
Operating Industries, Inc. Landfill: The OH Landfill covers 190 acres and is located 10 miles (16
kilometers) east of Los Angeles between the communities of Monterey Park and Montebello,
California The Pomona Freeway (Route 60) divides the site into two parcels; one 45-acre area
lies north of the freeway and the other 145-acre parcel lies south of the freeway. The landfill is in
the city of Monterey and the city of Montebello borders the southern end and portions of the
northern section of the landfill. Throughout its operating life, from 1948 to 1984, the landfill
received 30 million cubic yards of residential and commercial refuse, industrial wastes, liquid
wastes, and a variety of hazardous wastes. The EPA determined that approximately 4,000
-------
10
different parties sent waste to the landfill at one point or another. In October 1984, the landfill
was closed and proposed for listing on the National Priority List (NPL) In June 1986, the landfill
was officially listed as a NPL Superfund site, and experts estimated that the cleanup could take
as long as 45 years, and more than $600 million to (tomplete. As of 2002, the EPA had reached
settlements with almost 4,000 parties to pay for the icleanup work, with the total settlements
reaching over $600 million (Table 1.1).
OH Landfill and Neighboring Community
In the early 1980's, residents near the landfill formed Homeowners to Eliminate Landfill
Problems (HELP) to address increasing odor and potential health problems at the site, as well as
specific issues such as leachate seepage, methane gas buildup, declining property values, and
land use after closure of the site. This organization, comprised of 460 dues-paying families, was
an essential force in the eventual closing of the landfill. Community council meetings became
volatile as residents protested the "assaulting stench" of the air. "We could never open the
[house] windows," said Montebello resident Phyllis Lee. As another resident stated, "Some
nights I wake up coughing at two, three, four o'clock in the morning. The methane gas is so
strong that I have a hard time breathing."
-------
11
Table 1.1 Key Dates and Statistics
NPL
Site Name Discovery Listing
Operating Industries, Inc. Landfill 1978 19X5
Los Angeles, California
Monlclair, West Orange, & Glen Ridge 1983 1985
New Jersey
Industriplex and Water Wells G & H 1979 1983
Wobum, Massachusetts
Eagle Mine 1984 1986
Colorado
Homes
Dates & Descriptions of in Clean-up
Major Clean-up Phases Sample Cost
1988
1997
1991
1993-1995
19%
1992-1993
1989-1991
1996
1997
Drilling of wells 9,200 $600m
and groundwater
treatment
Construction of
cap on landfill
Phase 1 12,444 S200m
Phase 2& 3
Phase4&5
Main cleanup on 1 1,940 SKOm
both sites
Problematic 1,087 $70m +
State-led cleanup $0.7m/yr
Removal of
contaminated soils
Tailing piles capped
Total
Property
Value Loss
39.5%
8.9%
14%
15.3%
According to Katharine Shrine, assistant regional counsel for the EPA Region 9, "This
site is basically a 300-foot-tall, 190-acre mountain of every kind of disposable item in the
world." Residents say the landfill is so large that it interferes with television reception.
Approximately 53,000 people live within three miles of the sites, 23,000 within one mile of the
site, and 2,150 within 1000 feet of the landfill. Three schools are located within 1 mile of the
landfill. The area consists of heavy residential development and mostly middle income and
multi-racial neighborhoods.
For the Operating Industries, Inc. (Oil) Landfill case study, we were able to obtain data
on selling prices, housing characteristics, and Census information for nearly three decades (1970
to 1999). The length of this sample enables an examination of how proximity to the landfill
affected housing prices well before the problems began to arise in the late 1970's. A relatively
large footprint was selected in this study. The broader neighborhood surrounding the Oil Landfill
site includes 9,279 dwellings between 60 meters and about 8.5 kilometers (5.3 miles) from the
boundary of the site. Chapter 3 presents a more detailed history of the OH Landfill site.
-------
12
Montclair, West Orange, and Glen Ridge, New Jersey: Montclair, Glen Ridge, and East and
West Orange Townships are located about eight miles from Newark Airport in northern New
Jersey. These towns are densely populated, and are focated in one of the most densely populated
regions of the United States. Approximately 50,000 people live within one mile of the Superfund
sites. The Montclair/West Orange Radium Superfund site consists of 366 residential properties
on 120 acres in Montclair and West Orange. The Glen Ridge Radium Superfund site is
comprised of 306 properties on 90 acres of residential land in Glen Ridge and East Orange. The
soil at both sites is contaminated with radium, a naturally occurring element that can result in
high levels of radon gas and gamma radiation in nearby homes. Several plants occupied the area,
the largest of which was the U.S. Radium Corporation (formerly the Radium Luminous
Materials Corporation) which operated between 191j5 and 1926. Because of its luminescent
properties, radium was added to the paint that was used for numbers on watch dials and
instruments, which became especially popular during World War I. The Center for Disease
Control and the New Jersey Department of Health declared these sites to be a public health
hazard due to concerns about lung cancer. Montclair/West Orange and Glen Ridge were listed on
the NPL for Superfund sites in 1985 because of their proximity to radium waste generated by
radium processing. These plants had operated in thei area after the turn of the 20th century and an
estimated 200,000 cubic yards of contaminated material were placed on private and public areas
in the communities.
A USEPA contractor takes gamma radiation measurements in Montclair.
New Jersey Department of Environmental Protection officials were planning to notify
local government officials and residents of their findings in early December 1983. However,
-------
13
despite a request by officials to hold the story until official notification had been made, a
November 30th television news report broke the story early. According to the New York Times
(October 16,1984) article published one year later, "[Many] residents of the three communities -
Montclair, West Orange and Glen Ridge - were not told about the problem until... technicians,
wearing protective gear began taking soil and air samples in and around their homes." A couple
of news reports, referred to the radium contamination in New Jersey as "another Love Canal,"
since both residential areas were built on contaminated soil.
Initial attempts to remove the contaminated soil were hampered by the lack of a suitable
waste depository, resulting in 4,902 drums and 33 containers of soil being stored for nearly two
years on the yards of partially excavated properties in Montclair. In 1999, nearly 20 years after
the initial identification of the problem and 12 years after being put on the NPL, cleanup
activities continued to occur as the streets were replaced and the EPA continued to investigate
the possibility of additional groundwater contamination. By 1998, a total of $175 million had
been spent to remediate over 300 houses and remove 80,000 cubic yards (or 5,000 large truck
loads) of contaminated soil. In 2004, estimates of total cleanup exceeded $200 million (Table
1.1).
For this case study of the radium contamination in the communities of Montclair, Glen
Ridge, and East and West Orange in northern New Jersey, we were able to obtain good data on
selling prices, housing characteristics, and Census information for one decade (1987 to 1997),
which started just two years after the sites were listed on the NPL. This data enabled us to
examine the change over time of housing prices during the lengthy multi-phase cleanup process.
The data for this case study showed two different patterns of affects on housing prices.
For homes that neighbored the affected communities, but did not experience the contamination
themselves, there was a general decrease in property values as described below. For the homes
that were within the affected communities, the swings in property value changes were greater,
and the initial remediation efforts appear to have caused a temporary recovery in property value,
however, this recovery does not appear permanent. One possible explanation for this recovery in
property values is that the process of remediation often involved some remodeling of the homes
directly, such as a new garage and/or landscape. Therefore, the cleanup not only removed
-------
14
potential hazards, but directly improved affected homes. Chapter 4 provides a more detailed
description of these sites.
Industri-Plex and Water Wells G & H, Woburn, Massachusetts: Woburn is a historic city
(founded in 1640) of about 35,000 people located 12 miles northwest of Boston. The community
is predominantly blue-collar because of its industrial heritage. In the mid-1800s, Woburn became
known for shoe manufacturing. Local manufacturing activity later shifted from shoes to leather
production, and Wobum became a leader in the U.S. tanning industry by 1865. By 1884, Wobum
was home to 26 large tanneries that employed approximately 1,500 employees and produced
$4.5 million worth of leather. At the peak of Woburh's tanning industry, from 1900 to 1934, an
estimated 2,000 to 4,000 tons of chromium was dumped directly into Woburn's water resources,
as well as 65 to 140 tons of copper, 85 to 175 tons of lead, and 40 to 75 tons of zinc.
Abandoned 55-gallon Drum with the Entire Side Corroded; Found Near Wells G & H.
Wobum is also the location of two large Superfund sites: Wells G & H and Industri-Plex.
Together the sites cover almost 600 acres of land in the 14 square mile community. Both sites are
located in the section of Woburn east of Main Street, a low, swampy area that includes many
streams and the Aberjona River. This section of Wobum, referred to as East Woburn, is a mix of
industrial and residential areas. Roughly 13,000 households are located within two miles of the
Industri-Plex site, and homes are located within 1,000 feet of the site. Approximately 34,000
people live within three miles of both sites. While the two sites are distinct from each other, the
pollution problems at both sites were discovered within a few months of each other. Both sites
were evaluated by the EPA and added to the NPL in the early 1980s (Table 1.1).
-------
15
Throughout Wobura's history, more than 100 companies used the Aberjona River, which
flows through the city, for industrial waste disposal. Companies dumped wastes on land, into
lagoons and ponds adjacent to the river, as well as directly into the river itself. From 1853 to
1931, compounds and chemicals such as acetic acid, sulfuric acid, lead, arsenic, chromium,
benzene and toluene were dumped behind buildings, used as fill for low spots, and included in
construction material for dikes and levees. Wobum has a long history of public health problems,
including elevated rates of kidney and liver cancer, colon-rectal cancer, child and adult leukemia,
male breast cancer, melanoma, multiple myeloma, and brain and lung cancer.
The 330-acre Wells G & H site is located near the Aberjona River, about one and a
quarter miles downstream (south) of the Industri-Plex site. It once ranked as the tenth worst site
on the EPA's NPL list. The site is the location of two drinking water wells for the city of
Wobum, which were built in 1964 (Well G) and 1967 (Well H). These wells were located near
an automobile graveyard, an industrial barrel cleaning and reclamation company, a waste oil
refinery, a tannery, a dry cleaner, and a machinery manufacturer. Despite public complaints
about the water from these wells, Woburn continued to use the wells, especially during the
summer. Both wells were finally closed in 1979 after testing showed that the water was
contaminated. Soil and groundwater at the site are contaminated with volatile organic
compounds (VOCs), such as trichlorethylene (TCE) and tetrachloroethylene (also called
perchlorethylne, PCE, or 'perc'). Land in this area is zoned for industrial and commercial use,
with some areas for residential and recreational use.
The Industri-Plex site, the location of Wobum's most intensive industrial activity since
the 1850s, consists of 245 acres in an industrial park and once ranked as the fifth worst site on
the NPL. This area is located one mile northwest of the intersection of Interstate 93 and Route
128 and is bordered by the communities of Wilmington and Reading. Two tributaries of the
Aberjona River flow through the Industri-Plex site. Of the 245 acres at the site, one-third was
contaminated and 60 acres were used for commercial purposes throughout the remediation of the
site. Contamination at the Industri-Plex site includes heavy metals and hydrocarbons. In the soil,
the contamination was primarily arsenic, lead, and chromium and in the water the contamination
was primarily benzene, toluene, arsenic, and chromium. Additionally, hydrogen sulfide gas
emanating from wastes and buried animal hides from the tanneries, once permeated the air.
-------
16
The discovery of two major hazardous waste problems in one town prompted strong
media interest as well as the active response and involvement of Wobum's residents. Area
newspapers and TV stations ran multi-part stories about Wobum, alluding to it as a "toxic
wasteland." Millions of dollars and several years were devoted to the Woburn court case which
i
commanded front-page national media attention. The book describing the lawsuit, A Civil Action,
I
was published in 1996 and became a bestseller. In 1999, the book was made into a movie
starring John Travolta.
For Wobum, Massachusetts, we were able to obtain data on selling prices, housing
characteristics, and Census information from 1978 to 1997 on 12,444 homes. Therefore, the
sample begins one year before the discovery of comWination at Industri-Plex and Wells G & H
and extends throughout the lengthy litigation and cleanup activities. The Wobum case most
clearly demonstrates the importance of accounting for socio-demographic change when
conducting economic studies on the value of neighboring homes. When these factors are
included, it becomes evident that part of the decline in relative values for homes near the two
sites is related to a general deterioration of the neighborhoods. If these factors are not controlled
for in the analysis, the property affects of proximity to the sites may be overstated. However, the
sites themselves are the likely cause of neighborhood deterioration. Chapter 5 provides a detailed
history of these sites.
i
Eagle Mine, Colorado: Eagle Mine is centrally located between Vail and Beaver Creek ski areas,
approximately 100 miles west of Denver, Colorado. Eagle Mine lies between the small towns of
Mintum and Red Cliff, just off U.S. Highway 24 and was once one of trie nation's top producers
of zinc. The property consists of approximately 6,000 acres, 340 of which are contaminated with
toxic waste. Most of the contamination originates from areas located along the Eagle River, and
includes: the abandoned mining town of Oilman located on a cliff just above the mine, the old
Eagle Mine processing plant in Belden, two ponds cbntaining wastes from the smelting of ore,
Maloit Park, Rex Flats, various waste rock and roaster piles, and an elevated pipeline. The Eagle
River (a major tributary of the Colorado River), Cros's Creek, and several other tributaries run
through the site.
-------
17
Warning sign at the entrance to Rex Flats & OTP.
The Eagle Mine site is contaminated with eight to ten million tons of hazardous
substances including arsenic, nickel, chromium, zinc, manganese, cadmium, copper, and lead.
The main cause of Eagle River contamination came from acid mine drainage, which occurs when
sulfide minerals, such as pyrite, are exposed to oxygen and water and then oxidize. This process
creates sulfuric acid, which contaminated soil, groundwater, and surface water surrounding Eagle
Mine, producing water with low pH levels. Acid drainage at Eagle Mine resulted from
precipitation flowing through the waste piles that accumulated from nearly 100 years of mining.
As Eagle Mine acid drainage seeped into ground and surface water, it killed aquatic life and
vegetation growing along the water's edge and contaminated the river with zinc, lead,
manganese, and cadmium. Not only did this contamination threaten brown trout, the most
populous fish in this segment of the river, but it also permanently stained the rocks in and along
the river bright orange, providing Mintum and Red Cliff residents with a constant reminder of
the contamination at Eagle River.
State studies conducted in 1984 revealed dangerously high levels of cadmium,, copper,
lead, and zinc in local water resources. Mintum, with a population of 1,500, is the closest town
and draws drinking water from Cross Creek and two wells located within 2,000 feet of the mine
tailings. While Eagle Mine had a history of environmental problems dating back to 1957, the
majority of the problems arose after the mine closed in 1984. In March 1985, Ray Merry, the
Eagle Mine Environmental Health Officer, ordered the 14 families remaining in Oilman to leave
the site because of potential human health hazards. By July, all families had left the area and
-------
18
Oilman became a ghost town. A gate prohibiting entrance to the town read "Town for Sale."
Eagle Mine was placed on the NPL in June 1986.
As the cleanup began, public concern about the possibility of adverse human health
effects intensified. Although the EPA chose not to endorse the State of Colorado's cleanup plan
because it was skeptical of the plan's long-term effejctiveness, the State forged ahead with the
cleanup of the Eagle River site fearing the worsening of public health and environmental
damages that might result from continued acid mine drainage. However, the State's decision to
pump tailings pond water back into the mine, using jthe mine as a holding tank, proved to be
disastrous and caused even more pollution to infiltrate the Eagle River. A dry winter caused mine
seepage to make up most of river water, and the river turned orange. As a result, fish populations
declined dramatically. Samples taken from the river;that fall revealed zinc levels were 255 times
higher than fish tolerance thresholds. No fish lived in the river, and contamination was turning
the Eagle River various colors.
For the Eagle Mine, near Vail, Colorado, we were able to obtain data from 1,087 owner
occupied properties downstream of the Eagle Mine over a 24-year period (1976 to 1999).
Unfortunately, the data available from the Eagle County Assessor's office does not span enough
distinct Census tracts for the differences in socio-deniographic characteristics across these tracts
to be useful in explaining the variation in housing prices. A challenge with this area is that,
unlike the other three cases, a high percent of the homes are recreational and not owner occupied.
There is substantial evidence that areas most effected by the pollution from Eagle Mine, such as
Mintum, did not experience rapid development growth that occurred in other areas of the Vail
area, even though they were in closer proximity to Vail resort. Due to the lack of socio-
demographic data and the fact that Eagle Mine affeqted a mountain community where the main
pollution was observed in a river, not just the original point source, the data from Eagle Mine
was not included in the psychological model and analysis described below. Chapter 6 describes
the Eagle mine site and history in more detail.
1.3 Expert Error
Gayer, Hamilton and Viscusi (2000) argue thait residents living near Superfund sites judge
risks to be of a magnitude consistent with EPA expert opinions and that these judgments are
-------
19
reflected in property values. The research presented here suggests quite the opposite. However,
the sites studied here are much larger and likely to attract more attention. This section documents
many cases of expert error to help explain why expert opinion plays a limited role in explaining
residents' risk beliefs. Thus, the judgments of experts are only one component of the mix of news
media stories and perceptual cues received by the typical citizen. Even if statements by scientific
experts were accepted as credible, they would compete with a mix of the other signals and
perceptual cues. As simply one component, such statements are unlikely to be the primary
determinant of individual risk beliefs. Thus, risk beliefs determined largely by media stories and
other perceptual cues are unlikely to be easily changed by the pronouncements of a few scientists
(Fischhoff, 1989).
Furthermore, it is unlikely that statements by scientific experts will be accepted as
completely credible. Even when different experts are in essential agreement, the news media
often focuses on those aspects where experts disagree (Wilkins and Patterson, 1990), thus
lowering the perceived credibility of experts. In a study examining news coverage of Three Mile
Island and Chernobyl, Rubin (1987) found that news stories tended to dichotomize events rather
than blend a continuum of information to recipients. The result is that the public discredits
information it receives from experts because it appears that experts cannot agree among
themselves and, therefore, do not really know the risk that a site presents.
Despite the ideal that science discovers absolute truths, for every health or environment
related article there appears to be a corresponding article that rejects the tenets of the previously
publicized claim. Numerous famous examples exist where experts from academia, government,
and industry have made errors and misestimates:
• Soil contamination at Love Canal, Niagara, New York
• Dioxin contamination in Times Beach, Missouri
• The defective Dalkon Shield for birth control
• The false discovery of Cold Fusion
• The failures at Biosphere 2
• The near nuclear meltdown at Three Mile Island
• The Union Carbide Accident in Bhopal, India
-------
20
These examples, which are described in detail in Chapter 7, are not just relegated to the
past, as the costly search for weapons of mass destruction in Iraq, to date, has yet to support early
claims by intelligence experts. i
News about human and environmental health is omnipresent, yet much of this
information is contradictory. Nearly every day newspapers, magazines, and television shows
report new information that tends to further obscure issues rather than clarify them. A cursory
I
survey of two major national newspapers conducted between September 1,1999, and November
1,1999, yielded several articles mat contested previpusly reported claims or presented evidence
of scientific or expert misjudgment and error. These articles reported the following:
• "Studies Bolster Link between Diet Drugs, Heart-Valve Leaks." Contrary to the previous
claims of the manufacturer, the diet drugs Redux and fen-phen can cause permanent heart
damage (Wall Street Journal, September 10,1999).
• "Questions for Drug Maker on Honesty of Test Results: FBI Asks About Diet Product's
Approval." A drug manufacturer did not report to the Federal Drug Administration all
relevant test results prior to petitioning for approval of a drug (New York Times,
September 10,1999).
• "Tobacco Industry- Accused of Fraud" For more than forty years, the tobacco industry
suppressed evidence that tobacco use causes .cancer (New York Times, September 23,
1999).
• "Japanese Fuel Plant Spews Radiation after Accident." Trained operators of a nuclear
power plant in Japan poured more than six times the required amount of uranium into a
tank, resulting in a nuclear chain reaction (New York Times, October 1,1999).
• "Two Teams, Two Measures Equaled One Lost Spacecraft." The Mars Orbiter burned in
space because the spacecraft's creator used imperial measurements when the spacecraft's
i
navigational team used metric measurements (New York Times, October 1, 1999).
• "Drug May Be Cause of Veterans' Illness: Pentagon Survey Links Gulf War Syndrome
to Nerve-Gas Antidote." Persian Gulf War soldiers who were given a drug to protect
them from nerve gas attacks suffer from damage to areas of the brain that control
reflexes, movement, memory, and emotion (New York Times, October 19,1999).
-------
21
• "Testing in Nevada Desert is Tied to Cancers." Soldiers who participated in nuclear tests
for the military in the 1950s have higher than normal death rates and an increased
likelihood of developing leukemia and prostrate and nasal cancer (New York Times,
October 26,1999).
Due to this steady flow of events and news stories that present contradictory, inaccurate,
or incomplete expert evidence, the public is unlikely to accept expert evidence as absolutely
accurate all the time. The frequency of events as well as the ambiguity and uncertainty of
experts, government officials, and the media, as demonstrated by these examples, leads to doubt
and skepticism on behalf of the public. The implication is that residents living near Superfund
sites are forced to construct their own risk beliefs based on perceptual cues and media coverage.
McClelland et al. (1990) surveyed residents near Oil about their risk beliefs and found a bimodal
response with more than half believing that living near the site was as dangerous as smoking
more than one pack of cigarettes per day, with an incremental annual risk of death of
approximately 1/100. Most of the remaining residents viewed the risk as trivial. Assuming
typical values for statistical life and assuming three people per home, the discounted present
value of the risk for the residents that assessed the risk as similar to smoking exceeds the price
paid by these residents for their homes! Residents who responded this way did report that they
were desperate to sell and sought immediate cleanup.
1.4 Events, Perceptual Cues, Risk Perception, and Stigma
Given the doubts that people will inevitably have with respect to the credibility of expert
risk assessment, perceived risks will be based on personal and community judgments derived
from other sources of information. Events that are associated with a Superfund site will lead to
perceptual cues and media attention that will most likely elevate perceived risk and stigmatize
the site for reasons documented below. Some of the most important determinants of risk beliefs
are perceptual cues. Perceptual cues are physical aspects of a site that are perceived by local
residents, and are suggestive of risk. Examples of perceptual cues include odors emanating from
landfills, unusual odors or flavors in well water, unusual soil or water coloration at the site, and a
heavy volume of truck traffic going in and out of the site. Ironically, some actions taken by
authorities to minimize public health and safety risks tend to exacerbate risk beliefs by providing
-------
22
clear cues that some risk is present. Erecting chain link fences, posting 24-hour guards, placing
warning signs, conducting on-site tests (especially bjy workers wearing protective clothing) are
all cues to residents that risk levels may be higher than they thought. Such actions, which may be
necessary, almost never lower risk beliefs. Proximity to a site increases the frequency and
duration of contact with, or observation of, perceptual cues, which contributes directly to the
intensity of risk beliefs.
The effects of strong perceptual cues are well illustrated by the Oil Landfill. Initially,
concern about high volumes of truck traffic and odob (produced by decomposition in the
landfill) prompted local residents to organize and confront problems associated with the site.
McClelland at al. (1990) found a significant correlation between recognition of these perceptual
cues and the high risk beliefs of many residents living near the site. Several of the perceptual
cues were removed or reduced by (a) installing wells to extract the methane gas for commercial
use and (b) closing the site, which eliminated most of the truck traffic. Even though these actions
did not address risks that hazardous substances would migrate into local neighborhoods, the risk
estimates of many residents dropped dramatically after the principal perceptual cues were
removed. McClelland et al. also demonstrated that tikere were significant property value losses
associated with these risk beliefs.
Attention given to a site in the media, apart from the actual content of news stories, is
itself a perceptual cue that risks may be high. Many studies have shown that frequent exposure to
media reports about a site increases the likelihood thiat residents will believe the site is very
risky. The specific risk at a site and perhaps the site itself will usually be unfamiliar to residents.
That in itself increases risk beliefs (Wilkins and Patterson, 1987). But more importantly, it means
that residents are almost totally dependent on the news media for information about the risk.
Reflecting the concerns of their consumers, the news' media often focus on aspects that
accentuate dread, such as the uncontrollability of the risk and the frightful worst outcome (e.g.,
dying of cancer), rather than on information about trie low probabilities of the risk and how those
probabilities compare to other risks that residents accept.
The signals that the media sends to the public regarding risks from hazardous waste sites
are important, but the way in which the public interprets this information is equally important. A
key feature of how news coverage is interpreted by residents is whether there is an easily
-------
23
identifiable "villain" responsible for the hazardous waste problems at the site. For example, if the
responsible party is a corporation whose primary business activity is outside the community, then
it is more easily portrayed as a villain than a local business which has strong affiliations to the
community. Russell et al. (1991) found that the more important a site's potentially responsible
parties (PRPs) were to the local economy, the more skeptical residents living near the site were
that it needed to be cleaned up. Personal familiarity with a site also influences how news reports
are interpreted. The greater the prior familiarity, the less risk beliefs are likely to be elevated by
news stories.
The largest PRP for the OH Landfill was an outside corporation that had not provided
significant employment or other economic benefits for the residents who lived nearby. Most of
the waste, especially hazardous waste, was generated and brought to Oil from outside the
community. Oil was primarily a commercial landfill serving many interests outside of the
community. In short, conditions were ripe for news stories to elevate risk concerns significantly.
How a risk affects the community, society, and the economy will depend on individual
and group perceptions of the risk (Slovic et al., 1991; Kunreuther and Slovic, 2001). There can
be a compounding or "rippling" effect as more and more individuals respond to the risk
(Kasperson et al., 1988). Or, as Dr. Paul Slovic describes it, interactions among individuals can
produce a "social amplification of the original risk concern." The greater the population living
near a site, the greater the potential for compounding or social amplification.
When residents or potential buyers are extraordinarily fearful of a site, they may respond
by shunning the site. This behavioral response has been labeled stigmatization and has been
explored in a number of experiments that suggest that if risks are perceived as being excessive,
people replace calculations of risk versus benefit with a simple heuristic of shunning, the
avoidance of the stigmatized object.
Stigma has been shown to have a number of key properties. Laboratory experiments
testing these properties have involved dipping a medically sterilized cockroach into glasses of
juice and gauging subjects' willingness to drink the juice after the cockroach has been removed
(Rozin, 2001). First, stigma shares many of the psychological characteristics of contagion, where
contagion is associated with touch or physical contact. For example, while subjects refused to
drink the juice if the sterilized cockroach was dipped into the glass, they would drink the juice if
-------
24
the cockroach was just placed near it. Second, stigma appears to be permanent. Subjects refused
to drink the juice even if it had been in the freezer for one year. Third, stigma appeared to be
insensitive to dose. Reductions in the duration of contact between juice and cockroach had little
effect. Any contact was sufficient for subjects to shijin the juice. Fourth, the source of contagion
is usually unknown. Thus, while shunning may have evolved from an adaptive response to avoid
contaminated food, it can be triggered in inappropriate circumstances. For example, subjects who
saw sugar water placed in a clean empty jar and then saw a cyanide label placed on the jar still
tended to refuse to drink the sugar water. Finally, subjects tend to medicalize the risk, arguing
that the stigmatization was the result of a fear of health effects.
The possibility that Superfund sites might be stigmatized could have a major impact on
the prospects for successful cleanup of contaminated sites. If such sites are permanently shunned
because, like the "cockroached" juice, they are viewed as permanently stigmatized, property
values may not recover immediately once cleanup is in progress (since future improvements
should be capitalized into home values) or even when cleanup is completed.
1.5 Stigma and Property Values
The possibility that stigma may cause large losses in property values has been noted by
other researchers (e.g., Dale et al., 1999; Adams and Cantor, 2001) and the EPA (Harris, 2004).
In contrast to'the hedonic approach (Rosen, 1974; aijid for application to hazardous sites see
Bartik, 1998; Harris, 2004; Harrison and Stock, 1984; Ketkar, 1992; Kolhase, 1991;
Mendelsohn, et al., 1992; Michaels and Smith, 1990; etc.) where risk is treated as one of many
attributes that contribute to a determination of sale price, stigma is likely to effect property
values in a rather different and more direct manner. Upon learning of the contamination
potentially affecting their community, some current home owners may simply be unwilling to
continue to live in their home, and likewise, potential buyers will be unwilling to consider
buying a home in that community. If some owners and buyers have lexicographic preferences,
the standard hedonic model fails since it relies on a tradeoff between risk and home prices.
Rather, shunning by both current owners and potential home buyers will reduce the total demand
for housing for a neighborhood near a site as shown in Figure 1.1. Imagine that the total demand
-------
25
for homes in a particular fully built-out neighborhood with H existing homes is Q(P) where Q is
the number of desired homes, P is the sale price, and quantity demanded falls with price, Q'< 0.
If, for example, homes were sold in a competitive uniform price auction, the equilibrium price,
Pe, is obtained by solving H=Q(P), so Pe=Q"'(H). Now consider the case where a fraction f of
home buyers and owners shun a neighborhood because of a nearby Superfund site. The usual
hedonic model cannot handle this phenomenon because the hedonic price adjustment for these
individuals, either through very high subjective risk beliefs (assuming conventional values of
statistical life) or shunning would give homes a risk deficit greater than or equal to the value of
the home. In other words, in either case the perceived costs of staying in the home are greater
than the entire value of the home and the observed behavior would be identical. This implies that
fraction f of current owners will sell and that the number of potential buyers will be reduced by
fraction f as well. As shown in Figure 1.1, since we have defined total demand for the
neighborhood to include current owners, the equilibrium price will now be determined by the
solving H=(l-f)Q(P), so Pe*=Q'1(H/(l-f)) and Pe* < Pe for f > 0. If f falls with distance from the
site, as is likely since perceptual cues decline with distance, then property values will rise with
distance, ceteris paribus. Of course, relative demand for housing that is more distant from the
site will increase, but presumably this increase in demand will fall on a much larger group of
homes, resulting in a negligible increase in prices of homes farther from the site.
The next question is, since a hedonic analysis is used to incorporate normal attributes for
predicting property prices, how can downward sloping demand be incorporated into the analysis?
The answer proposed here is that hedonic models predict an average price based on home and
community attributes, but do not take into account individual buyer characteristics, including
bidding errors, which will affect the willingness to pay for homes in a particular area. So, for
example, relative to a predicted hedonic price, PH, one particular individual will be willing to pay
more because grandmother happens to live in the neighborhood and another particular individual
will be willing to pay less because of a random error in bidding strategy. Clearly no hedonic
market can exist for such attributes since they are buyer specific, and these sale price deviations
will appear as part of the error term in the estimated hedonic equation. Thus, for homes with a
particular set of hedonic attributes in a homogenous neighborhood with a mean sale price of PH,
there exists an array of values for homes among potential buyers, V, with a cumulative
-------
26
distribution function of Q(V). Presumably, the H buyers with the highest individual values will
own homes in the area.
Figure 1.1 The Effect of Stigma on Equilibrium Housing Prices
To further understand the property value market, we model the market itself as a
discriminative auction to account for the fact that identical homes in the same neighborhood can,
in fact, sell for different prices depending on unobserved individual buyer errors and other
attributes (see Cox et al, 1984, for a discussion of the relevant theory and an experimental test of
this auction). Approximating the property value market with an appropriate auction where
multiple buyers compete for available homes solves! the potential problem associated with
modeling real estate sales as bilateral negotiations where some sellers potentially have no value.
Rather, in a discriminative auction other potential biiyers provide competition that maintains the
price at a higher level than that which would be predicted by bilateral negotiation. The properties
of a discriminative auction are well understood, and this auction provides a reasonable
-------
27
approximation of the real estate market under the special circumstances where homes near a site
are stigmatized.
As previously discussed, sellers in our model have essentially no value for the homes
they are selling since they shun the site. Thus, any price they can get for the home* is acceptable.
This corresponds to an auction situation where buyers bid on H homes put up for sale, and the H
bidders with highest bids obtain the homes for the prices bid. Figure 1.2 shows this market in the
context of total demand where all homes in a neighborhood are potentially up for sale. Note that
the bids in a discriminative auction (shown as the lower step function) fall below the true values
(upper step function). Note also, that compared to the price that would be obtained in a uniform
price auction giving a price, Pe, in a discriminative auction there is a distribution of bids and sale
prices around the equilibrium price, since buyers pay accepted bid prices. In a discriminative
auction, it is well known that if'buyers are risk neutral, the average of the accepted bids will
equal the uniform price, so revenue neutrality exists in theory between uniform price and
discriminative auctions. Note also that risk aversion will increase bids in a discriminative auction
and bring them closer to true values because buyers trade off the gain in consumer surplus of a
lower accepted bid against the reduced probability of having their lower bid accepted. The lower
bid curve shown in Figure 1.2 assumes risk neutrality and plausibly provides a lower bound for
bids in a real estate market.
Figure 1.2 Discriminative Auction Market
-------
28
With these concepts in mind, we can then tup to the hedonic model used to estimate
property values at each of our study sites. The hedohic model estimated to explain property
values uses a logarithmic specification and takes the form:
0.1)
SPRlCEit =
Here, P, is an area-wide price index for owner-occupied housing in year t, DISTit is the distance
of each dwelling from the Superfund site in question. The coefficient associated with this
variable will be allowed to differ across years by interacting the constant distance measure with
yearly dummy variables. The vector An is property attributes and Sit is a vector of (interpolated)
time-varying characteristics of the Census tract in which the dwelling is located, and AT is a
vector of the logarithms of the distances from the dwelling to a potentially relevant set of other
spatially differentiated local amenities or disamenities, calculated at time T, the end of the
sample period, rather than contemporaneously.
Taking the logarithms of both sides of the equation yields a version of this model that is
appropriate for estimation:
(1.2)
LSPRICEit = \nPt + bltLDI$Tit
where LSPRICEit denotes the logarithm of the observed selling price, \r\P, will be captured as an
intercept for the first year in the sample and a set of intercept shifters activated by year dummy
variables. The variables of key interest are the LDISTa, which consist of a vector of logged
distances from the dwelling to the Superfund site interacted with yearly dummies in order to
permit year-varying elasticities of housing prices with respect to distance to the site. Geographic
Information Systems (GIS) techniques were used to measure distances from the homes to the
closest Superfund site in the specific year, t, that the sales price was observed and the distance to
other local amenities or disamenities as they existed jn year T.
An ideal sample of data would consist of transactions data and housing structural
characteristics, neighborhood characteristics, distanqes to all relevant amenities and disamenities,
all collected contemporaneously with the time of sale. This ideal data would also include
-------
29
analogous information (except for selling price) about houses that did not sell in these periods,
either because they were not for sale, or they did not find a buyer. This would allow the
researcher to control for non-random selection into the pool of dwellings actually observed to be
transacted.
When a researcher has data like these data over a number of years, it is possible to control
for many unobserved housing and neighborhood characteristics that do not vary across time by
using the so-called "repeat sales" method. When a house has sold more than once in the observed
time period, the difference in the selling price can be explained in terms of differences in any
explanatory variables that have also changed over time. This method for eliminating all the time-
invariant characteristics from the analysis was first proposed by Bailey, et al. (1963), and has
recently been used to analyze the influence of news stories about Superfund sites on housing
prices (Gayer and Viscusi, 2002). One disadvantage of this method is that the sample of repeat-
sales dwellings over-represents houses with greater turnover and excludes dwellings that have
been sold only once during the window of time for which data are available. There is also a
problem that any remodeling or updating of the property that is not captured by the quantity
variables typically recorded in multiple listing service data will go unacknowledged in the
process of dropping all structural characteristics by differencing over time.
In this study, we use a source of data that over-samples houses that have been sold only
once over the time period in question. Our data roughly reflect the current status of dwellings.
The data are provided, for the most part, by Experian, a company which provides information to
direct mail marketers and others. These data are updated at fairly regular intervals, although not
simultaneously. Anyone buying these records gets the most recent information available. For
each street address in the sample, most records include information on the date when the house
was purchased and the price that was paid at that time. For different localities, there are different
quantities of structural information in the data set. From the same data supplier, all fields will be
available for all localities, but for any given locality, blocks of fields will be blank. Blank fields
differ across localities, possibly reflecting different public recording requirements.
In some cases, notably the Eagle County files sought for the Eagle Mine site near Vail,
Colorado, the missing data problem was so severe that, despite the appearance of over 5,000
house transactions in the data, there were less than 50 with sufficient data for estimation of a
-------
30
basic hedonic property value model. Part of the problem is that a large share of dwellings is not
owner-occupied. In that case, we sought and received data from the Eagle County Assessors
office. There were roughly 1,400 observations for owner-occupied units, lying between 2.6 and
19.3 kilometers of the nearest part of the Eagle Mine site. About 57% were owner-occupied but
were not single-family dwellings. Other problems existed with this data. For example, the data
indicates that mere are no current owner-occupied uhits in the vicinity of the middle school
which is only 1,500 feet from a tailings pile. It would have been vastly preferable to have
acquired the same assessor's office information for each year during the time span of interest (in
this case, from 1976 to 1999). However, data that are "obsolete" from the point of view of the
assessor's office are apparently not retained merely for the convenience of researchers who wish
to understand time patterns in property values.
An obvious disadvantage of our sample is that in all of our data sets we only observe
selling prices for the most recent sale of a house. If a house is in an area where turnover is high,
there will be more recent sales and fewer earlier sales. For analytical purposes, it would be
preferable to have data on all sales in all years and selling price in those years, but such data do
not exist. Data could be purchased from Experian every year, if a future study could be
anticipated, but retrospectively, the data are not available. The data are collected primarily for
current marketing purposes and records are updated without saving their previous values.
Historical modeling is not a use anticipated by the providers of the data Consequently, there
may be some systematic sampling. We observe earlier transactions prices only for houses which
are still occupied by the owners who purchased them at that earlier date. We do not observe
many early transactions prices for houses in neighborhoods where there has been a lot of
turnover. It must be a maintained hypothesis that rates of turnover are uncorrelated with
identification and cleanup of Superfund sites. This may be a strenuous assumption, but there are
few alternatives. So it will be necessary to speculate upon the types of biases this non-random
selection is likely to produce in the effects of distance from a Superfund site on housing
transactions prices. Chapter 8 presents a description of the property value approach and data used
in the study, while Chapter 9 presents the property value results.
However, a distinct advantage exists of only having one observation for each home in the
sample. By only having one observation per house and controlling for area-wide price index with
-------
31
dummy variables, we ensure that each observation is independent. Therefore, the coefficient b»
(the effect of distance from the Superfund site on property values) can be observed over time by
looking at the hedonic estimates for each year over the 20-30 years of observations that have
been obtained for each of the sites. To dampen noise, we average bit the coefficients over three-
year intervals. To get time trends in property values as affected by the site, we normalize both by
the initial three-year period property value effect, t=0, and by distance. Thus, we ask the
question, at a minimum distance from the site, DISTmin, how do property values compare to price
at distance D/STmax (the boundary of the available data), which was chosen to be sufficiently far
away such that no effects of the site should be present, and to the magnitude of this effect in the
initial period. The relative property value effect, normalized by base period and by property
values at a large distance is defined as
(1.3)
Thus, the index for each site starts at 1.0 (or 100% in the figures below) and either decreases or
increases in successive three-year periods from this value. Table 1.2 presents the results for each
of the case studies.
Table 1.2 Coefficient Determinants
on
Montclair
(Outside)
Montclair
(Inside)
Time
Period
1970-1972
1973-1975
1976-1978
1979-1981
1982-1984
1985-1987
1988-1990
1991-1993
1994-1996
1997-1999
1987-1989
1990-1992
1993-1995
1996-1997
1987-1989
1990-1992
1993-1995
Average
Distance
Coefficient
-0.133
-0.136
-0.086
-0.099
-0.015
-0.039
0.013
0.015
0.015
0.027
-0.022
0.009
0.031
0.064
0.102
0.174
0.094
Normalized
Value
100.00%
100.79%
86.25%
89.65%
68.94%
74.28%
63.03%
62.65%
62.65%
60.46%
100.00%
90.65%
84.81%
76.51%
100.0%
92.9%
100.8%
-------
32
1996-1997
Woburn
1978-1979
0.191
-0.166
1980-1982 -0.115
1983-1985 |-0.154
1986-1988 -0.157
Eagle
1989-1991
1992-1994
1995-1997
1976-1982
1983-1988
1989-1994
1995-1999
-0.134
-0.111
[-0.106
-0.814
2.134
4.815
1.966
91.1%
100.00%
87.96%
97.01%
97.85%
92.35%
87.12%
86.04%
100.00%
83.88%
71.48%
84.72%
As can be seen in Figures 1.4, 1.5, and 1.6 presented below, relative property values of
the three metropolitan case studies (Oil in Los Angeles, Industri-Plex and Wells G&H in
Wobum, and Montclair, New Jersey) tend to follow an overall declining trend consistent with the
i
notion of progressive stigmatization of the site as suggested by arguments from psychology. This
result is in contrast to a number of earlier studies that examined property values over shorter time
periods (Carroll et al, 1996; Kiel, 1995; Kiel and Zajbel, 2001).
Our concluding chapter, Chapter 10, attempts to explain the long term downward trends
observed in relative property values shown in Figures 1.3-1.5 using a psychological model. If the
trend is driven by f, the fraction of home owners and potential buyers who shun homes near the
site, a model of the determination of f over time is needed. From the discussion of the
psychology of risk perception and stigma, the determination of the fraction of shunners will be
driven by media attention and perceptual cues resulting from activity at the site, which are in turn
driven by "events" such as EPA announcements, discovery, NPL-listing, and cleanup. Thus, it is
plausible that the percentage change between periods in the fraction of the population who shun
the site is a linear function of events of type j occurring during the prior interval, characterized
by the discrete dummy variable (or index summarizing a number of dummy variables), Ej,t-i,
thus
(1.4)
So, in a period with no events, Ej>t.i= 0 Vj, we hypothesize that a is negative and f will decline,
thereby raising home values, because some people who know about the site will leave the area
-------
33
(perhaps because of job opportunities elsewhere) and some new potential buyers will move into
the area who will have no awareness of the site. Other events, such as cleanup activities, might,
(a) raise awareness and thereby increase the fraction of the population who shun the site, or
alternatively, (b) reduce the fraction of shunners by convincing people who know about the site
that it is now safe. This latter possibility is unlikely in that the notion that, "once contaminated,
always contaminated" is part of the psychology of stigmatization. Note, also, that changes in
perceived risk for those who may not shun the site will likely follow a similar model.
There is no available data on f, so the model specified above cannot be estimated directly.
However, if one assumes a constant elasticity of demand, TI
-------
Figure 1.4 Relative Property Value over Time
for Oil Landfill, California
1970- 1073- 1970- 1979- 1982- IMS- 1988- 1991- 1864- 1997-
1972 197S 1678 1981 1884 1947 1990 1993 1«W 1999
34
Figure 1.5 Relative Property Value over Time for
Montclair, New Jersey (outside of area)
A
H 60%
20%
19S4 PAM« 1 :l*an-up
1985 phai»3 cE*an-up
1996 PhfifM
1993-1995
1996-1997
Figure 1.3 Relative Prdperty Value over Time
for Woburn, Massachusetts
2 tm.
S
f
1978-1979 199>1S$2 1861-1985 1960-1968 1989-1961 1982-1994 1995-191
-------
35
To employ this transformation we need to know the relevant elasticities of demand that depend
on the error distribution in bids. Since we do not have this information, we assume that the
elasticities are all -1.0, consistent with a linear approximation of the relationship between f and
the change in R over time.
Table 1.3 Number and Description of Events
Event Type
EPA Action
State Government Action
Local Government
Action
Public Action
Potentially Responsible
Party Action
Remediation Action
EPA Announcement
Site Incident
TOTAL
Number of Events
on
11
6
10
2
7
6
12
5
Montclair
3
1
1
1
0
4
3
2
Woburn
14
4
0
9
0
3
8
12
59 15 I 50
TOTAL
28
11
11
12
7
13
23
19
124
Table 1.4 presents a psychological model using the data shown in Figures 1.3,1.4, and
1.5 of relative property values over time for the three metropolitan sites. Note, as mentioned
earlier. Eagle Mine was excluded from this analysis because the socio-demographic information
for the homes were unavailable. Since all of the home sale observations were independent, a
simple linear regression could be used with 18 observations of changes in relative property value
(R'* - R~^ ) over the three-year periods for the three sites. For Discovery, NPL Listing, and the
Beginning of Major Phases of Cleanup, dummy variables were used. The variable "Events" was
-------
36
derived by summing the number of major announcements and actions described in EPA
published reports for the relevant three-year interval for each of the three case studies (Table
1.3). Events are defined as followed:
• EPA Action - Includes site investigations, orders, notifications/decisions,
remediation, legal actions, and regulations the EPA.
• State Government Action - Includes site investigations, orders, resolutions,
remediation, lawsuits, reports, and regulations by state agencies.
• Lpcal Government Action - Includes sjte investigations, orders, resolutions,
remediation, lawsuits, reports, and regulations by local cities, county, and school
districts.
• Public Action - Include the creation of public interest groups, major public meetings
and protests, lawsuits by the residents, and the hiring of technical advisors for the
community.
• Potentially Responsible Parties Action] - Include site operation and closure, and
committees formed. Lawsuits by PRPs.
• Remediation Action - Includes containment of contaminati ons, remediation efforts
and site improvements.
• EPA Announcement - Includes official Consent Decrees, Record of Decisions
(RODs), and announcements of settlements with PRPs.
• Site Incident - Includes general site facts, reports and studies regarding the
contaminants and occurrences at the site.
The analysis across the three sites shows that discovery, cleanup itself, and the number of
events all negatively affect property values by drawing attention to the site and possibly
increasing the number of owners and potential buyers who shun the site thereafter (Table 1.4).
Thus, the effect of any event, publicity or site infonriation, good or bad, appears to increase the
fraction of the current home owners and potential buyers that stigmatize and consequently shun
the communities neighboring the sites. In other words, at least within the observed period of the
-------
37
studies, all news is bad news and causes relatively permanent property value losses as an
increasing fraction of original owners leave and more potential buyers shun the site. The only
good news in the study is that property values did significantly recover for a short period after
sites were listed on the NPL. But, it is likely that as soon as it was realized that EPA could not
immediately clean up the sites, the process of stigmatization began with consequent reduction in
property values. All of these coefficients except the constant are significant at better than 1%
level.
Table 1.4 Psychological Model, Dependent Variable ' M
Model
(Constant)
Discovery
NPL - Listing
Clean-up Begins
Number of Events
B
0.078
-0.160
0.105
-0.096
-0.016
t
3.578
-4.493
4.097
-4.753
-6.156
P
0.003
0.001
0.001
0.001
0.000
N=18
R2 = 0.855
Rather than property losses reversing immediately once cleanup has begun, we see no
permanent recovery in property values within the time period of our data and speculate that
recovery will only occur as the local population gradually moves away, events cease, and
perceptual cues and media attention disappear, so more buyers are uninformed. Note that
McClelland et al. (1990) found that most buyers were uninformed in spite of reporting
requirements. The positive intercept in the psychological model (significant at better than the 5%
level) indicates that property values will increase at a linear rate of about 12% every three-years
if no actions are taken and no news is generated by the site. Thus, at Oil one could expect a
complete recovery in about a decade if no news is generated from the site and recovery might
occur in about half that time for the other sites.
-------
38
Figure 1.6 Relative Property Value over Time for Eagle Mini;, Colorado
1976-1092
The sites excluded from the model are also of some independent interest. First, although
the Eagle Mine (see Figure 1.6) has very different characteristics from the three sites discussed
i
above, it shows a similar pattern in that relative property values decline for most of the period
analyzed. Given the small amount of data available along the Eagle River, we are forced to use
six-year rather than three-year periods for the analysis but do confirm the general pattern shown
|
above. Second, the "inside" Montclair property value estimates do not use distance as an
explanatory variable since the homes themselves are within the Superfund site. Yearly dummy
variables averaged over the same three-year intervals used in the outside-Montclair analysis
show that, unsurprisingly, cleanup itself does have a positive impact on property values (Figure
1.7). Third, another interesting result in the property value studies is the effect of including
socio-demographic variables. As shown in Figure 8, these make a large difference in the
magnitude of property losses at the Woburn site. Negative socio-demographic trends, that may
-------
39
be the result of the progressive stigmatization of the site, also take a substantial toll on property
values (that are not included in the psychological model), but possibly should be included in any
damage assessment. These results suggest a different trend than observed by Kiel and Zabel
(2001) which did not account for these socio-demographic affects.
Figure 1.7 Relative Property Value over Time for Montclair, New Jersey (inside of area)
100%
80%-
•3 60%-
I
I
« 40% -
20%-
0%
1987-1989
1991 Phiss 1 eWa
1990-1992
1993-1995
1996-1997
-------
40
Figure 1.8 Relative Property Value over Time for Wo burn, Massachusetts
with and without socio-demographic variables.
1978-1979 1980-1992 1963-1919 1986-1988 1989-1991 1992-1994 1995-1997
|—»— Mod«M with Sociodemoflnphica -• •*•• -jMcdel 2 v»ltl\aH Soclod«mocraphte'r|
1.6 Policy Implications
Since economic benefits are based on discounted present value, the benefits of delayed
cleanup for homes surrounding sites are likely to be negligible where cleanup takes 20 years and
another 5-10 years may be needed after cleanup is complete for property values to recover. The
principal policy conclusion becomes evident from the results of the psychological model, which
suggest that the promise of a prompt cleanup raises property values, while an increase in the
number of events that are the root cause of perceptual cues and media attention decreases
-------
41
property values. Thus, an expedited cleanup should occur as quickly as possible after a site has
been determined to be hazardous and this cleanup should be conducted in a way that does not
arouse excessive attention. Otherwise the neighborhoods surrounding the site will-likely be
stigmatized resulting in quasi-permanent economic damages.
Table 1.5 Cleanup Scenarios
Scenario A
Scenario B
Scenario C
Scenario D
Time
Horizon
33 years
33 years
24 years
1 5 years
Events
All
25% Fewer
25% Fewer
25% Fewer
Discovery
1978
1978
1978
1978
NPL
Listing
1985
1985
1982
1979
Cleanup time
periods
1988-1990
& 1997- 1999
1988-1990
& 1997-1999
1985-1987
& 1988- 1990
1979-1981
Recovery
time periods
2002-2005
2002-2005
1990-1995
1982-1987
Final % of
Original Value
64.5%
85.2%
95.6%
100.0%
Using the history of the OH and the corresponding dates and events in a simulation, the
potential benefits of these policies becomes evident (Figure 1.9). As shown in Table 1.5, this
simulation considers four different scenarios and includes an extrapolation of a recovery in
property values after cleanup is complete where there are no further events. Given the legislative
history of Superfund, some of these scenarios are clearly fanciful, but the results are nevertheless
suggestive as to what potential benefits could be obtained by expediting the cleanup process and
reducing the number of events that drive perceptual cues, media attention, and social
amplification. These results support several of the suggestions made by Kunreuther and Slovic
(Chapter 21,2001) for reducing stigma. In particular, they suggest prevention of the occurrence
of stigmatizing events and the reduction of the number of stigmatizing messages and thus
reducing social amplification.
Note that these results contrast with those of Gayer, Hamilton and Viscusi (2000) and
Gayer and Viscusi (2002) who argue that media attention supports learning that leads to a
lowering of public risk perceptions more consistent with scientific evidence for smaller sites. No
credible evidence supports a significant long-term health risk to residents living near Oil
(McClelland et al. 1990). Yet the actual property value losses are enormous. One difference is
-------
42
that this study focuses on prominent sites while the jtwo studies cited above focused on less
prominent sites. Note, however, that most potential benefits from cleanup are likely to come
from prominent, sites. Also, both Wobum and Montclair are associated with demonstrable long-
term health risks, yet property losses are much smaller than at OIL Finely, property value losses
seem to be greatest when cleanup finishes, when risks should be at their lowest.
Figure 1.9 Policy Simulations Usitg the Oil Landfill History
120% -i
100%-
80%-
£
Q.
'jai
O
60% •
40% -
20%-
0%
IS
A
fe
B
.*!
at
fe
-•—A: Baseline - 33 years
•«s-B: 33 years; 25% fewer events
•-*••• C: 24 years; 25% fewer events
-*-D: 15 years; 25% fewer frvents
-------
43
It is interesting to note that Carol Browner did in fact institute reforms to USEPA policy
in 1995 to at least partly attempt to avoid the pattern shown in this study. EPA began to work
with PRPs in an attempt to negotiate sufficient cleanup at potential Superfund sites to avoid
having sites listed on the NPL. These reforms may, in fact, have represented an optimal response
given the difficulty stigma presents for neighborhoods surrounding Superfund sites. It should
also be noted that the enormously costly process of litigation and delayed cleanup that has
occurred under the Superfund program has provided strong incentives for industry to avoid
creating new hazardous waste sites. However, for residents living near very large Superfund
sites, as they have often stated, the program has failed in spite of EPA's best efforts. In this
regard, it should be noted that when CERCLA was passed, little or none of the work in
psychology necessary to understand the phenomena described here had been completed. In fact,
much of the relevant work was motivated by Superfund.
This study raises several questions for future research. First, are smaller sites truly
different as the work by Gayer, Hamilton and Viscusi suggests? Second, although the
psychological model developed here is statistically significant, it is based on data from just three
sites. Additional work to incorporate both larger sites, as well as smaller sites, and additional
explanatory variables would be worthwhile in our judgment. Finally, more research to
understand and prevent stigmatization is warranted.
-------
44
Chapter 2
History of Current Superfund Legislation
2.1 Overview of Superfund Legislation
Superfund is one of the most controversial pieces of legislation ever implemented by the
EPA, and it has been tainted for years by skepticism and uncertainty. Congress approved the
Superfund bill in 1980, despite uncertainty about the number of sites, costs of cleanup, and
availability of appropriate technology. The number bf sites in need of cleanup and the costs of
the program skyrocketed beyond the original expectations of Congress or the EPA, due to
underestimates of the extent of the hazardous waste iproblem. The bill was reauthorized twice,
but in December 1995 it was not renewed. The taxing authority of the fund expired in December
1995, leaving only fines, penalties, and interest as working income for EPA's Superfund
program.
Since 1995, the Superfund program has been primarily funded by the Trust Fund, and the
General Fund that supplements the Trust Fund. Since 2000, an average total of $1.3 billion is
appropriated to Superfund each year with more money being allocated from the General Fund
than the Trust Fund (Figure 2.1). The Superfund truit is used in cases where responsible
polluting parties cannot or will not pay for cleanup. In some cases where the EPA has tapped the
Superfund trust for cleanup, EPA cleans up the property itself and then sues the responsible
polluting party for triple the damages. The 2004 request is $ 1.1 billion from the General Fund
and only $290 Million from the Trust Fund. Changep in the way the Superfund program and how
the EPA cleans hazardous areas are underway. On April 12,2004, the National Advisory
Council for Environmental Policy and Technology released its Final Report on the how the
Superfund program could be improved.
-------
45
Figure 2.1 Superfund Budget (1981-2004)
$1,8001
$1,500
$1,200
Millions $
$900
$600
$300
SO
HGeneveral Revenues
Q Trust Fund
S
a
Year
2.2 Legislative Background
The driving force for Superfund legislation came from 1978 report by Michael Brown, a
local reporter for the Niagara Gazette newspaper, on Love Canal, the toxic waste site near
Niagara Falls. He described children coming home from the playground with hard pimples on
their bodies, women giving birth to deformed and mentally retarded children, and many other
horrible consequences of Hooker Chemical and Plastics Corporation's (now Occidental
Petroleum) disposal of over 20,000 tons of toxic wastes in an unlined canal (Bamett, 1994). The
thought of a country filled with sites similar to Love Canal angered the American public. Shortly
after the issuance of Brown's report, Congress began hearings on Love Canal and other waste
disposal sites throughout the country.
The Carter Administration favored a new piece of legislation addressing the toxic waste
issue because existing legislation, such as the Clean Air Act of 1970 and the Clean Water Act of
-------
46
1972, regulated the emission of chemical pollutants, but did not consider the impact of chemical
pollutants and toxic waste on human and ecological!health. A key question in the legislative
debate was whether the funds should come from general government revenues or from the
financial contributions of both the offending industry and the government In mid-1979,
President Carter spoke in favor of a bill with a $1.6 billion fund, financed 80% by industry and
20% by government imposing strict, joint and several liability on responsible polluting parties
(Bamett, 1994). Joint and several liability means that a company, individual, or some
combination thereof, can be held responsible for the entire cleanup of a hazardous waste site
regardless of the amount of pollution contributed to the site. That is. there is no proportionality
j
with respect to liability. This "deep pockets" principle was highly controversial.
In 1979, several key issues divided the proppsed Superfund legislation in the House and
Senate. These included the size of the fund and if the legislation should be limited to abandoned
hazardous waste sites or also include provisions for oil and hazardous substance spills. The
House preferred two separate bills totaling $1.9 billion, while the Senate preferred a $4.1 billion
bill that could be used for both emergency removals and more costly, long lasting projects. Both
the House and Senate, however, agreed the legislation should be financed primarily through
taxes imposed on chemical feed stocks. The chemical industry disputed this overall tax on the
industry and advocated financing the fund through the federal treasury, responsible parties, and
state contributions. The final Superfund bill did not include other House and Senate
recommendations regarding public participation in the siting of hazardous wastes or determining
satisfactory levels of cleanup.
The Comprehensive Environmental Response, Compensation, and Liability Act
(CERCLA) of 1980 (commonly referred to as Superfund) was signed into Public Law 96-510 by
President Carter on December 11,1980. It passed by wide margins in both the Senate (78 to 9)
and House (274 to 94). The final Superfund bill passed under several unusual circumstances. The
Senate Finance Committee sent the bill forward without a formal recommendation and President
Carter pushed for it to be signed quickly, before Reagan took office. The bill, because it was
rushed, was not based upon an accurate assessment of the problem, and suffered from a lack of
research as well as inaccurate estimates of the potential costs and the number of toxic waste sites.
As noted in a 1985 report funded by the Cato Institute, the law "was based upon
-------
47
misunderstandings and distortions of the situation, and zipped through a lame duck Congress in a
spirit of vengeance against the polluters" (Delong, 1995).
2.3 Comprehensive Environmental Response, Compensation, and
Liability Act
CERCLA created a five-year, $1.6 billion trust fund for the cleanup of active or abandoned
hazardous waste disposal sites that posed an immediate threat to public health and the
environment or when a responsible party would not take action. Superfund called for two
different types of cleanups: (1) Remedial Actions, which ere long-term cleanup for sites on the
National Priority List (NPL), and (2) Removal Actions, which are short-term cleanups of
immediate threats. Removal Actions did not require listing on the NPL.
The trust was funded by the chemical and petroleum industries, which were required to pay
into the fund. The chemical industry ended up paying 85% of the $1.38 billion paid by these two
industries. The States were required to contribute 10% of the total cleanup costs, provide long
term maintenance at sites, and provide disposal capacity for waste removal (Office of
Technological Assessment, 1984). The other component of the Superfund trust included a series
of taxes. Specifically, the bill created a $0.79 per barrel tax on US refineries, crude oil and
petroleum imports, and domestically produced crude oil. Hazardous chemicals and wastes were
taxed at rates varying from $0.22 to $4.87 per ton. After September 30,1983, there was an
additional excise tax of $2.13 per dry weight ton of hazardous waste received at qualified
hazardous waste disposal sites (Office of Technological Assessment, 1984). The taxing authority
of the CERCLA legislation expired September 30,1985, though it was reauthorized shortly
thereafter.
Under the CERCLA legislation the EPA was required to assess the nation's hazardous
waste sites according to the hazard ranking system, which is a numerically based screening
system that determines the threat of waste sites to human health. A minimum of 400 of the worst
hazardous waste sites were to be placed on the NPL, and each state was given the opportunity to
place a site on the list as well. The bill also gave the EPA and the Justice Department legal
authority to identify polluting parties, enforce liability, and enlist contributions for cleanup.
-------
48
Superfiund imposed strict, joint, and several liability, and the generators of hazardous waste
substances and the owners and operators of waste facilities, past and present, were all liable to
pay for the costs of cleanups. The process for cleanup entailed listing a site on the NPL. Then
scientists conducted a detailed examination of the sjte called a remedial investigation. This was
followed by a feasibility study, an ascription of the most appropriate cleanup remedies. These
two processes are often referred to as the "remedial investigation/feasibility study" or "RI/FS".
Once the appropriate remedies were selected, the Elj'A prepared, or appointed an agency to
prepare specifications and timelines for the cleanup of the site and remediation begins.
2.4 Implementation of Superfund: 1980-1985
The Reagan Administration took office only weeks after the Superfund bill was signed,
and it promoted an ideology that was incompatible with many of Superfund's legislative
provisions. The objective of the Reagan Program was to promote economic growth by reducing
the size and number of federal spending programs, cutting taxes, and reforming regulatory
agencies (Barnett, 1994). Reagan reduced government spending and emphasized voluntary
compliance with environmental regulations.
Funding Superfund presented an immediate problem. The actual costs of cleanups were
much higher than the original estimates of $7 to $10 million (Barnett, 1994). By 1992, 42 of the
50 states lacked adequate resources to fulfill their obligations to pay 10% of the costs of cleanup
(Office of Technology and Assessment, 1983). Duriing the first three years of the program,
appropriations for Superfund also fell far below those initially authorized by Congress. Of the
$960 million authorized in the first three years, $74.4 million was appropriated in 1981, $26.6
million in 1982, and $210 million in 1983. Most furiding for the program was devoted to legal
battles with polluters as well, and the federal government wrote off $270 million because
responsible parlies could not be found or were unable to pay (New York Times, 1993). As a
result of these funding setbacks, elements of the Superfund program were cut back in 1981 and
1983. The EPA's total outlays were cut by one third (Shabecoff, ] 983), the abatement and
control staff declined 21%, the enforcement staff debreased 33%, and the research and
development staff declinedl 6% (Crandall and Portney, 1984: 68).
-------
49
While the funding for Superfund was less than originally anticipated, the number of sites
identified for the National Priority List far exceeded expectations. The legislation called for an
identification of at least 400 sites to be added to the NPL. The EPA quickly identified 16,200
potentially hazardous sites and placed 546 sites on the proposed National Priority List. The first
official National Priority List, released in September of 1983, included 406 sites (Environmental
Protection Agency, 1997). The following year, the EPA added 132 more sites to the List bringing
the total up to 538 (Environmental Protection Agency, 1997). While the list continued to grow,
only 119 site removals and two full cleanups were completed by mid-1983 (Davis, 1983).
Various government and private agencies re-estimated the number of sites in need of cleanup.
The new figures were much larger than originally anticipated, and ranged from EPA's initial
estimate of 2,000 sites to the Office of Technological Assessment's estimate of over 10,000 sites
(Office of Technological Assessment, 1985). As of March 31, 2004,1,239 sites were listed on
the National Priorities List (NPL).
To further complicate the bill's implementation, a scandal concerning the manipulation of
hazardous waste programs by some EPA officials erupted in 1982. Top officials were charged
with using political criteria to determine Superfund spending and making "sweetheart" deals with
industry members for partial cleanup, perjury, and the manipulation and destruction of
government files. EPA Administrator Anne Burford and 13 other top EPA officials, including
Rita Lavelle (head of the Superfund program), were forced to resign in March of 1983. This
"Sewergate" scandal dominated the front pages of major newspapers for the first three months of
the year, causing the EPA to lose a considerable amount of legitimacy and some of its own
morale. This scandal, furthermore, undermined the momentum of the already struggling
program.
After the resignation of Administrator Anne Burford, William Ruckelshaus took office in
1983. Ruckelshaus adopted a more aggressive approach to remediation and often used lawsuits
to force industry members into compliance. The EPA financed 102 waste removals within the
first six months of Ruckleshaus' term and an additional 157 waste removals from 1984 to 1985.
The EPA's combined outlays in fiscal years 1984 and 1985 totaled over $600 million, a 261%
increase over the $233 million spent between 1981 and 1983 (Bamett, 1994).
-------
50
2.5 1985: The Expiration of Superfund
Superfund legislation expired on September 30,1985. Debates about reauthorization of the
bill continued for over a year because of uncertainty about how to improve the program's
performance. Reauthorization issues included concerns over health and environmental risks at
waste sites, questions about risk assessment, frustration with lengthy cleanups, and burgeoning
costs. Many cleanup cases were embroiled in legal battles which often expanded into complex
webs of multiple party lawsuits, further compounding Superfund's problems. The public's
expectations of the program were initially very high, but they found the rate and success of
cleanups disappointing (Office of Technological Assessment, 1984). Changes to the Superfund
legislation were imminent and necessary.
The House and Senate proposed different reauthorization bills in 1985. The House was
overwhelmingly (391 to 33) in favor of a bill creating a five-year, $10.1 billion dollar program.
Revenues would include $3.1 billion from a tax on petroleum companies, $2 billion from a tax
on chemical companies, and $2 billion from a tax on toxic wastes. The Senate proposed a five-
year, $7.5 billion program. General revenues would provide approximately $1 billion with the
remaining $6.5 billion to come from special taxes on manufacturers and processors of raw
materials with annual sales of $5 million or more (Shabecoff, 1985). Environmentalists argued
that the bill would not raise enough money and was described by the Washington representative
of the Sierra Club as "a missed opportunity to build an effective program" (Shabecoff, 1985).
The EPA anticipated that the debate between Congress and the House would continue past
the bill's expiration date and began to halt or slow cleanup at 57 sites in August of 1985 (New
York Times, 1985). As the authorization debate continued through April, the Agency reduced its
response rate to toxic waste emergencies by 80%. Although EPA Administrator Lee Thomas said
that $5 billion was the most EPA was capable of spending over the next five years, the House
and Senate remained firm in their higher requests of $10.1 billion and $7.5 billion respectively
(New York Times, 1985). As a result of the lengthyidebates in 1986, appropriations for the
program were reduced from the expected $900 million to $206 million (Shabecoff, 1986).
-------
51
2.6 Superfund Amendments and Reauthorizations
A conference committee finally approved a plan on July 31,1986 for a five-year, $9 billion
fund. The bill included $500 million for regulating leaky underground storage tanks; created a
new, broad based tax on corporations earning more than $2 million a year; and increased taxes
on oil products (Shabecoff, 1986). Other provisions imposed new regulations throughout all EPA
offices with respect to deadlines and cleanup standards, increased public involvement by
requiring industry to provide local residents with information about chemicals used, and required
the EPA to provide technical assistance grants to communities proximate to NPL sites.
President Reagan signed this bill known as the Superfund Amendments and
Reauthorization Act of 1986 (SARA) on October 17,1986. After the expiration of Superfund's
taxing authority, the CERCLA legislation was extended two more times: first for a four-year
period at which time $5.1 billion was authorized for the program and second through December
1995.
The 1986 reauthorization failed to solve many of the Superfund's problems. The EPA
still lacked adequate financial resources to cleanup many waste sites due to difficulties collecting
money from responsible parties. The liability scheme caused funds to be tied up in legal battles.
The EPA spends approximately 15-18% of the Superfund budget on the legal enforcement of its
cleanup mandates. There were often over 100 PRPs at large sites, creating complex webs of
lawsuits (Barnett, 1994). Responsible parties found it worthwhile to litigate in hopes of
spreading costs among many responsible parties (as opposed to settling) because the average cost
of cleanup at a site was approximately $30 million. A 1993 report funded by the Cato Institute
indicated that, in 1989, insurance companies spent an average of $470 million on costs related to
the Superfund program, $410 million of which went to defending their policy holders (New York
Times, 1992). According to a RAND Corporation survey in 1994, legal fighting over prior
liability was costing a total of $1 billion a year, while fewer than 200 of the most serious 1200
sites had been cleaned up (Quint, 1994).
To help remedy inefficient spending on legal fees, the Congressional Budget Office
(CBO) studied the costs of repealing prior liability, a provision in the Superfund legislation that
holds a company legally responsible for cleanups when the waste was dumped before Superfund
-------
52
was enacted. The CBO found that repealing prior liability would reduce transaction costs and
increase efficiency of the program. It might also, however, reduce the speed of cleanup or require
lowering cleanup standards. Additionally, federal government spending on Superfund would
need to increase by $1.6 billion per year and the government would incur an additional one-time
cost of almost $7.5 billion for PRPs' ongoing expenses plus $6 billion for past costs (Committee
for the National Institute for the Environment, 1997). Consequently, many government officials
remain opposed to the repeal of prior liability. They believe that the general public should not
have to pay for the costs of hazardous waste cleanup and that tax money should be reserved for
orphan sites and emergency actions.
Other challenges with the Superfund program remain as well. Cleanup times were
criticized as being too slow. According to the General Accounting Office (GAO), the average
time to cleanup a site increased from 3.9 years in 1986 to 10.6 years in 1996 to 11.5 years in
1999. The time it took from discovery of a problem site to its final listing increased from 5.8
years in the 1986-1990 time period to 9.4 years in 1,996 (GAO, 1998). The GAO concluded that
these increases are due to the legislation's ambiguous cleanup requirements (GAO, 1998). The
structure of the program was fundamentally flawed because it neither provided contractors with
incentives for cost effective remediation nor encouraged innovation. Responsible parties were
reluctant to try new technology because they feared inadequate results and the possibility of
having to conduct a second costly remediation. The! legislation also required copious amounts of
paperwork and authorizations. Critics also were concerned about the EPA's risk assessment
procedures, claiming that the true risks of most people living near sites are overstated, resulting
in costly remedies and little gain in risk reduction. Polluters and their insurance companies are
dissatisfied with the law's retroactive provisions, requirements to reach "gold plated" cleanups,
and disregard for cost when selecting cleanup remedies (Center for Hazardous Waste
Management, 1997).
In December 1995, Superfund lost the ability to tax, and this taxing authority has not
been renewed. Since taxes on the industry could not finance the Superfund Trust Fund, it began
to rely heavily on appropriations from general revenue. Other income also included fines,
penalties, and cost recoveries. Since 1995, there have been several bills debated in Congress, yet
none have been approved.
-------
53
Reauthorization was a top environmental priority for the 105th and 106th Congresses,
according to Congressional leaders, but action on Superfund legislation never materialized.
President Clinton toured a Superfund site in Wallington, NJ in March 1996 to show his concern
about Superfund's shortcomings. Reauthorization issues include the size of the fund, broad
liability scheme, high contractor costs, and slow pace of cleanups. Most Republicans, and some
Democrats, continue to support repealing prior liability (Government and Commerce, 1997).
There are also disagreements over who should be required to pay for cleanups, the stringency of
cleanup currently required by the law, whether cleanup results in too few benefits for the costs,
and whether or not to limit the National Priority List. Another issue that has gained more
attention recently is the damage to natural resources, such as rivers, caused by hazardous wastes
(Government and Commerce, 1997). The Senate proposed to allow the addition of only 90 more
sites (30 annually for three years) to the National Priority List. The House proposed to allow 125
sites over an eight-year period, declining from 30 sites in the first year to 10 sites in each of the
last three years.
2.7 Superfund Reforms and Successes
Under the leadership of Carol Browner, in 1995, the EPA implemented three series of
more than 45 administrative reforms designed to strengthen Superfund by targeting its problems
with cleanups and enforcement. The reforms sought the goal of being "faster, fairer, more
efficient" and were similar to provision of the 1994 reauthorization legislation that died in the
previous Congress. These reforms expanded beyond the scope of the reforms in the
reauthorization of 1986 and have been even more successful than originally anticipated
(Nakamura and Church, 2003).
While the federal government maintains control of determining appropriate remedies for
hazardous waste sites, states have recently taken on more responsibility for the cleanup of
Superfund sites. States have developed the capacity and technical expertise necessary for
successful remediation of Superfund sites. Currently, the federal government enters into
cooperative agreements with states, on a site-by-site basis, which authorizes the states to conduct
cleanup activities.
-------
54
In 1995, the EPA also implemented a new "l|>rownfields" economic redevelopment
program. Brownfields are sites contaminated with hazardous waste that are possible candidates
for the National Priority List. These sites often remain abandoned for long periods of time
because they are undesirable to lenders and developers who fear that they will assume liability or
that the land will remain undervalued. To date, the brownfields program has been highly
successful and helps states, communities, and other stakeholders work together to safely and
efficiently cleanup and reuse brownfields (Environmental Protection Agency, 1997).
The Superfund program has been much mork successful in recent years as a result of the
aforementioned reforms. As of 2004, more than 82% of the sites on the final Superfund NPL
were either undergoing or had completed cleanup (NACEPT, 2004). Furthermore, the number of
sites added to the National Priority List is currently declining, while the number of sites deleted
is increasing. Only 12 sites were added in 2003, as ppposed to 162 additions in 1986. The total
number of sites listed as possible National Priority tist sites also decreased over the past years.
In 1988, there were 378 sites on the proposed list. In 2003, 54 sites wers on the list. As of 2003,
the total number of sites on the National Priority list! is 1,572 while 274 sites have been deleted
from the NPL list. Deletion from the NPL list means that the remedial goals had been achieved
even if operation and maintenance of the site continues (NACEPT, 2004).
Responsible parties were also paying a higher percentage of the cleanup costs than in the
past. Polluters were now contributing more than 75% of long term cleanup costs compared to
37% in 1987 (EPA, 1997). This has saved taxpayers: a total of over $12 billion and resulted in the
EPA obtaining over $7.00 in private cleanup commitments for every $1.00 spent in Superfund
enforcement. The EPA also reached settlements with more than 14,000 smaller polluting parties
(such as small businesses and municipal governments) and gave over $457 million in
compensation to responsible parties willing to negotiate long-term cleanup settlements
(Superfund Administrative Reforms, 1996). Moreover, the Justice Department collected $790
million for cleanup activities from responsible parties in 1996. As a result, the EPA was able to
preserve the Superfund budget for sites where responsible parties could not be identified.
In July 2001, the Superfund Subcommittee of the National Advisory Council for
Environmental Policy and Technology was formed to evaluate the Superfund Program. In April
12, 2004, the Final Report was released which recommend improvements to three main areas:
-------
55
what types of sites should be listed on the NPL, how to measure performance, and what to do
about "mega sites" (cleanup costs greater than $50 million).
2.8 Conclusion
Superfund has been controversial since its inception. Many of the problems with the
program, such as high legal costs, inadequate funding, slow pace of cleanups, and high
contractor costs, plagued the program from the outset because it was developed in response to an
emergency situation of uncertain proportions. However, Superfund did have some success in the
1990s. As of 2004, over 82% of the sites on the final Superfund NPL were either undergoing
construction or have completed cleanup. Responsible parties are paying more of the costs of
cleanups, allowing the EPA to conserve its budget for orphaned hazardous waste sites. In the
next chapters, detailed site histories are provided for the six "mega-sites" examined in this study.
-------
56
Chapter 3
Operating Industries, Inc. Landfill
3.1
Overview
The Landfill covers 190 acres and is located 10 miles (16 kilometers) east of Los Angeles
between the communities of Monterey Park and Montebello, California (Figure 3.1). The
Pomona Freeway (Route 60) divides the site into two parcels, one 45-acre area lies north of the
freeway and the other 145-acre parcel lies south of the freeway (Figure 3.2). The landfill is in the
city of Monterey and the city of Montebello borders the southern end and portions of the
northern section of the landfill. Throughout its operating life, from 194U to 1984, the landfill
received 30 million cubic yards of residential and commercial refuse, industrial wastes, liquid
wastes, and a variety of hazardous wastes. The EPA determined that approximately 4,000
different parties sent waste to the landfill at one point or another. In October 1984, the landfill
was closed and proposed for listing on the National Priority List (NPL). In June 1986, the landfill
was officially listed as a NPL Superfund site, and experts estimated that the cleanup could take
as much as 45 years and more than $600 million to complete. As of 2002, the EPA had reached
settlement with more than 1,250 parties to pay for the cleanup work, with the total settlements
reaching more than $600 million.
Figure 3.1 Oil Landfill Vicinity
-------
57
Figure 3.2 Oil Landfill
P^J.«^^WI^.^I^I £^l,,,,l--^
|*?**«4*4™| | I £, i S i !M l^v/*^' | J
$ -v- • > * •'
J % ?^ i v =?
<}r - cfS^txi ;?
.gs ;:•>:•?* >>
^ \***ist'Jt »v < "• *>»' : <•*.
-------
58
The landfill remains a particularly prominent feature of the area, lowering more than 300 feet
above the surrounding community. '•
On the road to completion: Oil's final landfill cover along the Pomona Freeway.
According to Kathenne Shrine, assistant regional counsel for the EPA Region 9, "This
site is basically a 300-foot-tall, 190-acre mountain of every kind of disposable item in the
world." Residents say the landfill is so large that it interferes with television reception.
Approximately 53,000 people live within three miles (4.8 kilometers) of the sites, 23,000 within
one mile (1.6 kilometers) of the site, and 2,150 withjn 1000 feet (0.3 kilometers) of the landfill.
Three schools are located within 1 mile (1.6 kilometers) of the landfill. The area consists of
heavy residential development and mostly middle income and multi-racial neighborhoods.
Oil Landfill and Neighboring Community.
-------
59
3.2 History of the Landfill
Before 1970
In 1948, the Monterey Park Disposal Company began municipal dump operations in a
former stone and sand quarry. At this time there was very little development neighboring the site.
Operating Industries, Inc. (Oil) purchased the site four years later in January 1952. OH continued
municipal operations, but began accepting industrial wastes at the site as well.
1970-1972
In the early 1970's, the population of southern California blossomed and development
pressure in areas surrounding Los Angeles increased dramatically. To help alleviate this
pressure, the land surrounding the landfill was approved by Montebello for residential use.
Promised closure of the landfill by the city and proposed development of a golf course on the site
prompted rapid development of the area.
1973-1975
The Pomona Freeway was built in 1974 and intersected the 190-acre landfill. Soon the
middle income communities of Monterey Park and Montebello encircled the Oil Landfill.
Landfill activities were restricted to the larger southern area (South Parcel) of the landfill, closest
to Montebello. In compensation for the closure of the North Parcel, the city increased the height
limits on the South Parcel of the landfill.
1976-1978
The increase in the height limit on the South Parcel led to increased erosion, mudslides,
and ultimately exposed refuse. Starting in 1978, leachate seepages were observed periodically on
the slopes of the South Parcel of the Oil Landfill. The leachate contained both organic
constituents (such as volatile organic compounds, semi-volatile organic compounds, oil, and
grease) and inorganic constituents (such as metals, ammonia, chloride, and high levels of total
dissolved solids). Large amounts of landfill gas, generated by the natural decomposition of
organic and hazardous wastes, were also reported at the site. Tests of the landfill gas found its
primary components to be methane, nitrogen, carbon dioxide, and volatile organic compounds
-------
60
(approximately 0,05 percent). Threats of contamination at Oil stemmed from exposure to toxic
compounds (such as trichloroethane and toluene) and carcinogenic compounds (such as vinyl
chloride, benzene, trichloroethylene, and carbon tetrachloride) via air, groundwater, soil, and
leachate.
Before the EPA's involvement, numerous state, regional, and local agencies were
involved with the Oil site. The earliest government intervention began in March 1978 when the
South Coast Air Quality Management District (SCXQMD) issued an order requiring Oil to
follow proper maintenance and disposal proceduresn One year later, Oil hired Getty Synthetic
Fuels to collect methane gas generated by the landfill. Getty Synthetic Fuels then removed the
i
methane gas from the site and refined it for commercial purposes but these activities, including
drilling, exacerbated odor problems
1979-1981
In 1980, the Los Angeles County Department of Health Services (LACDOHS) realized
that the original methane gas collection system by itgelf was insufficient and directed OH to
institute a second gas control system.
Residents near the landfill formed Homeowners to Eliminate Landfill Problems (HELP)
to address increasing odor and potential health problems at the site, as well as specific issues
such as leachate seepage, methane gas buildup, declining property values, and land use after
closure of the site. This organization, comprised of 460 dues-paying families, was an essential
j
force in the eventual closing of the landfill. According to testimony from Montebello
Councilwoman Norma Lopez-Reid:
"... residents living near the Operating Industries Landfill came home each evening to an
area filled with migrating gases, that made them suffer from headaches, nauseating odors,
and grass-less yards due to the hazardous liquid waste, called leachate, that seeped out of
the ground. These difficult circumstances made the quality of life in this bedroom
community decrease considerably, we couldn't even open our windows on hot summer
nights. Little did our residents know the extent to which companies, large and small, had
been allowed to dump incredible amounts of hazardous waste, including carcinogens,
into the landfill that was only supposed to contain regular trash.'7
-------
61
In the early 1980's, community council meetings became volatile as residents protested
the "assaulting stench" of the air. "We could never open the [house] windows," said Montebello
resident Phyllis Lee. As another resident stated, "Some nights I wake up coughing at two, three,
four o'clock in the morning. The methane gas is so strong that I have a hard time breathing."
On March 5, 1981, the Montebello School District passed a resolution objecting to the
landfill odors and ordering an investigation of potential health risks. Later that year, county
health officials cited OH for not controlling the migration of potentially hazardous gasses. The
Oil Landfill was ordered to temporarily shut down.
1982-1984
Despite resident complaints, heavy criticism from state health and air quality officials,
possible $1,000 per day operating fines, and the previous temporary order to shutdown, OH
reopened and continued to operate. Finally, in January 1983, OH ceased accepting hazardous
liquid waste. In April, they stopped accepting any liquid waste. Also in April, offsite levels of
vinyl chloride gas (a known carcinogen) were measured at 19 parts per billion (the state
regulated level at that time for vinyl chloride was 10 parts per billion), however, in a random
sample of 12 homes elevated levels of vinyl chloride gas were not detected. In June, a buildup of
methane within the landfill caused several underground fires. Potentially explosive levels of
methane were also discovered underneath city streets adjacent to the landfill. Levels of air-borne
vinyl chloride in excess of EPA and state health standards were also detected around the site.
However, the LA County Department of Health released a 1983 study showing no pattern of
school absence and that there was not excess mortality around the Oil Landfill compared to other
areas of Los Angeles.
In January' 1984, the State of California placed the Oil Landfill on the California
Hazardous Waste Priority List.
In August 1984, the LACDOHS cited Oil for allowing landfill gas concentrations which
exceeded the lower explosive limit (5% methane in air) to migrate beyond landfill boundaries.
That same month, California Department of Health Services (CADOH) also issued its first
Remedial Action Order against Oil. It required Oil to completely phase out the on-site disposal
-------
62
of leachate and to provide plans for implementing leachate collection and treatment systems, site
characterization and groundwater monitoring programs, landfill gas collection and monitoring
programs, and slope stability corrective measures,
In October 1984. after four years of legal battles, public hearings, and tremendous
community public hearings, SC AQMD issued a secjond order of abatement requiring the landfill
to close permanently, thereby ending the disposal of all solid wastes. This order also instructed
OH to install a landfill gas emission control system,1 a permanent leachate control system, and
also to perform specified landfill maintenance. Soon after the closure of the landfill, the owner of
OH declared bankruptcy.
Also in October, the EPA proposed Oil for the federal Superfund NPL, making it eligible
for federal Superfund money. Likewise, the California Regional Water Quality Control Board
(RWQCB) issued its own abatement order requiring that the on-site disposal of leachate be
phased out.
In December 1984, the EPA dug six wells around the OH site to test for possible
groundwater contamination. The test results showed1 organics and trace mineral contamination in
three of the wells, but no pesticide contamination. Fprtunately, the drinking water used by the
neighboring residents of the landfill came from a number of municipal water companies, which
did not operate any wells located on or near the site' However, based on their initial tests, the
EPA decided to conduct further testing to determine the specific location and potential
movement of the groundwater contamination. The EPA installed an additional 24 wells around
the perimeter of the site, which tested positively forisoil and groundwater contamination.
1985-1987
In April 1985, while the Oil Landfill awaited its final federal NPL listing, the EPA began
its Remedial Investigation/Feasibility Study (Rl/FS) which assessed and prioritized remedial
actions for the site. This did not, however, assuage state concerns about the site. One month later,
the California Waste Management Board (CWMB) joined the CADOHS and filed a joint suit
against OH for not complying with the CADOHS's first Remedial Action Order issued in August
1984.
-------
63
In June 1986, the EPA placed the landfill on the NPL of Superfund sites and assumed
responsibility for all remedial activities at the site. Shortly after the NPL listing of the site, the
California Department of Health Services conducted an extensive epidemiological investigation
comparing health symptoms of residents near the site with those of control communities (Satin et
al., 1986). The results indicated no significant differences between the health of local residents
and that of control communities.
The EPA began its search for Oil's PRPs concurrent with the NPL listing. Because of the
nature of landfills, the EPA estimated that as many as 4,000 companies were potentially liable
for dumping hazardous waste at the Oil Landfill during its operable years. Although not all of
these PRPs contributed to the cleanup of the landfill, the first of several cleanup agreements was
signed in May 1989 between the EPA and over 110 polluting companies. Valued at
approximately $66 million, this First Partial Consent Decree required site control and monitoring
activities and construction of an interim leachate treatment facility. In return for their immediate
cooperation and financial contributions, the Consent Decree released from future liability several
large national corporations including Mobil, Exxon, and General Motors. A group of PRPs then
organized the Oil Steering Committee, of which Oil was not a part, to handle legal and
environmental issues at the site. This committee eventually formed a corporation called the
Coalition Undertaking Remedial Efforts, Inc. (CURE), which would remediate leachate at the
site according to the established leachate management plan.
The EPA signed its first Record of Decision (ROD) for the Oil site in July 1987,
authorizing short-term control and management activities to prevent further contamination and
exposure to potential health risks. One such action included fencing the site and posting a guard
at the entrance to ensure that no trespassers come into contact with the contaminants. Other
emergency measures included gas migration control measures, slope stability, leachate control
measures, erosion control, and runoff and drainage improvements.
Once these emergency measures were in place, the EPA signed the second ROD, in
November, for control and cleanup of leachate at the site. The EPA proposed several alternative
plans and submitted its draft plan for public review. As it often did for area residents, the EPA
extended the 30-day public comment period to allow ample time for all interested parties to
respond. Based on these public comments, the EPA decided to replace the system of off-site
-------
64
leachate treatment with an on-site treatment plant because it found this alternative to be more
acceptable, more cost effective, and more protective of public health and the environment. While
the surrounding communities were supportive of this decision, they were greatly concerned that
the plant might be used to process liquid wastes from other sites as well. The EPA assured local
citizens that only liquids generated from the Oil sit$ would be treated at the on-site plant, and
then preceded with its remedial plan. The plant woijld treat 43,200 gallons of OH leachate per
day, test it for compliance, and then discharge it into the Los Angeles County Sanitation District
sewer system. Natural attenuation and degradation of contaminants would also help reduce
groundwater contaminant levels beyond the site boundary, and the EPA agreed to routinely
monitor groundwater under and near the site.
1988-1990
Landfill gas migration was the second major problem the EPA addressed at the Oil
Landfill. The EPA's third ROD for Oil called for special gas migration and treatment studies so
that gas control systems could be improved prior to the final site cleanup. The results of these
studies called for the design and construction of a njew gas flare facility (thermal destruction
facility), new gas piping and extraction wells, use of existing extraction wells until they are no
longer functional, discontinuing use of the air dike system, construction of additional gas
monitoring probes, and the installation of gas extraction wells on the North Parcel. Fifty gas
wells were also installed in the South Parcel to help control gas and liquid migration. During this
time, the EPA notified residents that workers wearihg protective gear and loud drilling noises
would be present at the site for several months. The EPA estimated that with these improvements
70% more of the landfill gas would be collected, but that the landfill still would not reach the
EPA's goal for surface emissions until the final landfill cover was put into place.
The cost of the leachate treatment facility was estimated to be SI. 6 million, with annual
operation and maintenance cost of $700,000. Although the plant was originally scheduled for
construction in the summer of 1988, it was delayed almost a year because of the lack of
appropriate funding and the long processes of public comment and facility design and
finalization. Likewise, plant operation, scheduled to begin in early 1989, did not begin until
August 1994.
-------
65
In 1989, a Consent Decree was signed by the EPA and more than 100 companies that
disposed wastes during the operation of the landfill. These companies formed a committee to
examine the issues facing the cleanup effort. Also in 1989, over 100 businesses and public
entities filed a lawsuit against other PRP's to share in the cleanup costs.
As part of the gas control system, the EPA amended the ROD to add a landfill cover that
sought to better control gas emissions and improve surface water management. The existing
cover was highly variable in its thickness and ability to limit surface emissions and odor. A new
landfill cover would make gas control remedies more effective and efficient, reduce the amount
of gas escaping into the atmosphere, and facilitate the cost effectiveness of the final site remedy.
The new landfill cover sought to keep water and oxygen out of the landfill and prevent erosion
and run-off from the landfill's slopes. The cover also was designed to improve the appearance of
the site. Vegetation has been planted, whenever possible, over the cover. The EPA estimated that
the construction of the landfill cover would cost between $61 million and $116 million dollars.
1991-1993
In 1991, the EPA extended the settlement offer to another 154 PRPs, which resulted in a
Second Partial Consent Decree similar to the first. In August, 63 of the 154 companies signed the
Second Partial Consent Decree worth $8.5 million. Those that did not sign the agreement denied
the charges, asserted that they dumped only non-hazardous materials at the site, or maintained
that their refuse dumping records didn't exist or were unrecoverable. A third settlement between
the EPA and approximately 178 PRPs was reached in December 1991. It required the
defendants, later organized into a corporation called the New CURE, to implement major
portions of the gas control and landfill cover remedies, improvements worth $130 million.
In addition, the EPA sent letters to more than 50 additional PRPs informing them of their
liability at the site. During this time, several private companies brought suit against nearby
communities to make them liable for contaminants dumped at the landfill. To recoup legal costs
resulting from these accusations, one city raised trash collection rates and several other cities
sued trash haulers.
In November 1991, the EPA installed another 21 wells beyond the perimeter of the site,
which determined that groundwater contamination had not spread beyond the boundaries of the
-------
66
landfill. In total, the EPA constructed 75 wells and conducted six major hydrologic
investigations over the course of 28 years.
In April 1992, Judge Mariana R. Pfaelzer of jMontebello, California, enforced the
previous $130 million agreement. The EPA also entered into another Consent Decree with a
variety of contributing companies to begin the initial cleanup activities related to gas control and
landfill cover as designed by the EPA.
Implementation of gas control remedies beg^i in 1992. The construction, operation, and
maintenance of these remedies were estimated to cost $73 million over a 30-year period. From
1993 to March 1994, during the construction of the new gas flare facility, low levels of vinyl
chloride escaped into the atmosphere because the temperature of the old gas flare treatment
system was insufficient to incinerate the gas. Although they had no way of monitoring emissions,
the EPA maintained there was no problem with air quality. However, the EPA still tested 197
Montebello homes for vinyl chlorine. Four percent of the homes tested positively and the EPA
installed gas management systems to aerate the hcrries and prevent further the contamination
from entering the houses. The EPA is required to monitor these homes for ten years.
1994-1996
In 1994, the leachate treatment facility began operations. A settlement was reached in
1994 that resolved the 1989 lawsuit filed by 137 businesses and public entities that had already
contributed more than $200 million to the cleanup efforts.
In April 1995, a Fourth Partial Consent Decijee was reached between the United States,
the State of California, several private companies, 14 cities and municipalities that disposed of
municipal waste at the landfill, and those who transported municipal waste to the landfill. The
agreement provided $51 million for the Final Remedy and construction of a Thermal Destruction
Facility on the North Parcel.
In March 1996, thirty companies signed a final Fifth Partial Consent Decree to pay $18.7
million for interim site costs and the Final Remedy. In total, over $270 million had been
collected from almost 400 different entities for the cleanup of the OH Landfill. This estimate
represents only private money contributed to site cleanup. An additional estimated $165 million
of public dollars will be spent for site remediation.
-------
67
In September 1996, the fourth Oil ROD for Final Site Remedy was signed. The EPA
agreed to evaluate and monitor the site and the success of their cleanup efforts every five years
after the remedial action plan is implemented. The EPA also decided to replace the gas flare
disposal system with a new landfill gas thermal destruction facility, which would be used to treat
or destroy landfill gas produced at the site.
1997-1999
In mid-February, 50 new testing wells were installed on the southern and western
perimeter of the former OH Landfill near Montebello. These wells were designed to better
control gas and liquid migration from the landfill. The EPA advised residents to not interpret the
workers protective clothing as an indicator of a "hazard for the neighborhood". Additionally they
emphasized that during the drilling process "there will be no contamination hazard to nearby
residents during these activities" (emphasis in original).
Additionally, construction of the landfill cover began. The EPA mailed out a special
notice to residents informing them of the upcoming constructions and possible disturbances that
may result. Construction of the cover involved moving six million cubic yards of soil. The old
dirt on the landfill was removed and was replaced with a six-foot-thick "monocover"of clean
soil. A variety of native vegetation including grasses and shrubs was added to the slopes of the
landfill and the flat top of the landfill was covered by a multi-layer "geosynthetic clay liner" (a
system of woven matting and clay). The objective of both covers was to prevent rainwater from
entering the landfill and to stop gas from escaping.
In March 1997, the EPA issued a universal administrative order requiring seven more
companies, which collectively dumped six million gallons of hazardous substance into to the Oil
Landfill, to contribute to Oil's remediation. Specifically, these companies are responsible for
maintaining the on-site leachate treatment facility until December 1999, which will cost a total of
approximately three million dollars.
In 1999, the EPA reached a settlement with 327 parties which allegedly disposed small
amounts of waste at Oil. These parties contributed between 4,200 and 110,000 gallons of waste.
After 1999
-------
68
In 2000, the landfill cap and the new gas control systems (Thermal Destruction Facility)
were essentially completed, gas flares were replace<| and the treatment of groundwater
commenced. Development of a shopping center at the North Parcel of the landfill began. This
North Parcel was not significantly affected by the hazardous waste and is one of the largest
pieces of undeveloped property in the Los Angeles ^rea.
In December 2001, a $340 million settlement was signed with 161 PRP's. This was the
eighth settlement since 1986. Settlements up to this point had totaled over $600 million.
Also in 2001, the construction of the ground] water remedy was completed and the EPA
ended its in-home monitoring and random sampling programs after it found no evidence of a
problem in any of the houses for several years.
The eighth Consent Decree was approved imMay 2002, which outlined the final cleanup
remedies.
-------
69
Chapter 4
Woburn, Massachusetts
4.1
Overview
Wobum is a historic city (founded in 1640) of about 35,000 people located 12 miles (19.3
kilometers) northwest of Boston (Figure 4.1). The community is predominantly blue-collar
because of its industrial heritage. It is also the Socation of two large Superfund sites: Wells G &
H and Industri-Plex. Together the sites cover almost 600 acres of land in the 14 square mile (36.3
square kilometer) community. Both sites are located in the section of Woburn east of Main
Street, a low, swampy area that includes many streams and the Aberjona River (Figure 4.2). This
section of Wobum, referred to as East Woburn, is a mix of industrial and residential areas. For
the Industri-Plex site, homes are located within 1,000 feet and 13,000 households are within a
two mile (3.2 kilometer) radius. Approximately 34,000 people live within three miles (4.8
kilometer) of both sites. While the two sites are distinct from each other, the pollution problems
at both sites were discovered within a few months of each other. Both sites were evaluated by the
EPA and added to the NPL in the early 1980s.
Figure 4.1 Woburn Vicinity
V
-------
70
Figure 4.2 Industri-Plex and Wells G&H Sites
-------
71
Throughout Wobum's history, more than 100 companies used the Aberjona River, which
flows through the city, for industrial waste disposal. Companies dumped wastes on land, into
lagoons and ponds adjacent to the river, as well as directly into the river itself. From 1853 to
1931, compounds and chemicals such as acetic acid, sulfuric acid, lead, arsenic, chromium,
benzene and toluene were dumped behind buildings, used as fill for low spots, and included in
construction material for dikes and levees. Woburn has a long history of public health problems,
including elevated rates of kidney and liver cancer, colon-rectal cancer, child and adult leukemia,
male breast cancer, melanoma, multiple myeloma, and brain and lung cancer.
The 330-acre Wells G & H site is located near the Aberjona River, about one and a
quarter miles (2 kilometers) downstream (south) of the Industri-Plex site. It once ranked as the
tenth worst site on the EPA's NPL list. The site is the location of two drinking water wells for
the city of Woburn, which were built in 1964 (Well G) and 1967 (Well H). These wells were
located near an automobile graveyard, an industrial barrel cleaning and reclamation company, a
waste oil refinery, a tannery, a dry cleaner, and a machinery manufacturer. Despite public
complaints about the water from these wells, Wobum continued to use the wells, especially
during the summer. Both wells were finally closed in 1979 after testing showed that the water
was contaminated. Soil and groundwater at the site are contaminated with volatile organic
compounds (VOCs), such as trichlorethylene (TCE) and tetrachloroethylene (also called
perchlorethylne, PCE, or 'perc'). Land in this area is zoned for industrial and commercial use,
with some areas for residential and recreational use.
The Industri-Plex site, the location of Woburn's most intensive industrial activity since
the 1850s, consists of 245 acres in an industrial park and once ranked as the fifth worst site on
the EPA's NPL. This area is located one mile (1.6 kilometers) northwest of the intersection of
Interstate 93 and Route 128 and is bordered by the communities of Wilmington and Reading.
Two tributaries of the Aberjona River flow through the Industri-Plex site. Of the 245 acres at the
site, one-third was contaminated and 60 acres were used for commercial purposes throughout the
remediation of the site. Contamination at the Industri-Plex site includes heavy metals and
hydrocarbons. In the soil, the contamination was primarily arsenic, lead, and chromium and in
the water the contamination was primarily benzene, toluene, arsenic, and chromium.
-------
72
Additionally, hydrogen sulfide gas emanating from buried animal hides from the tanneries and
wastes once permeated the air.
Redevelopment plans are now underway for, the Industri-Plex site and consist of a
Regional Transportation Center (train station, park and ride), a recycling center, highway
interchange, an office park, and retail space. This development is expected to yield between
12,000 and 16,000 jobs by 2010, relieve traffic congestion in Woburn, and help the Boston area
come into compliance with EPA air emissions standards by enhancing public transportation in
the area. However, as of 1999, the City of Wobum estimated that the number of permanent jobs
at the redeveloped site was 4,315.
Groundwater remediation and monitoring now constitute the bulk of the remaining work
to be done. As of December 1998, three groundwater treatment plants operating at the site had
pumped and treated more than 150 million gallons of water. In addition, all the contaminated soil
has been removed (approximately 150 tons) and 1,360 pounds of VOCs have been destroyed.
Although major strides in the remediation of the property have taken place, according to EPA
lawyers, complete cleanup of the site will cost approximately $80 million and could take another
20-30 years because of the site's extensive groundwater contamination.
4.2 History of Woburn and its Superfund Sites
1600-1700's
Woburn was incorporated as a town in 1640 and shortly thereafter became a center of
manufacturing in New England, because its location) was ideal for industry because of its
accessibility to major transportation through ways (roadways and seaports) and proximity to the
consumer market of Boston. In 1648, the first tannery opened in Wobum.
1800's
In the mid-1800s, Wobum became known for shoe manufacturing. Local manufacturing
activity later shifted from shoes to leather production, and Woburn became a leader in the U.S.
tanning industry by 1865. By 1884, Wobum was home to 26 large tanneries that employed
approximately 1,500 employees and produced $4.5 trillion worth of leather. At the peak of
-------
73
Woburn's tanning industry, from 1900 to 1934, an estimated 2,000 to 4,000 tons of chromium
was dumped directly into Wobum's water resources, as well as 65 to 140 tons of copper, 85 to
175 tons of lead, and 40 to 75 tons of zinc.
Numerous chemical manufacturing firms occupied the Induslri-Plex land, which is
upstream from water Wells G & H, for almost 150 years. In the 1800s, several firms operated on
the site, producing chemicals for the local tanning, textile, and paper industries. In 1863,
theMerrimac Chemical Company (Merrimac) became the leading industry on the site.
1900's
From 1863 to 1931, Merrimac produced lead-arsenic insecticides, TNT and other
explosives, dyes, and organic chemicals such as phenol, benzene, and toluene. Between 1900 and
1914, Merrimac was one of the largest producers of arsenic insecticides in the country. In 1915,
as part of the war effort, industry in Wobum began to diversify to include munitions, chemicals,
and insecticides. (Until the war, Germany had been the major source of these chemicals.) By
1917, Merrimac had grown into the largest chemical manufacturer in New England.
1920's
The City of Wobum built a sewer due to concerns about pollution of the Aberjona River
and Upper Mystic Lake. Monsanto acquired Merrimac in 1929 and moved the chemical
operations off the Industri-Plex site in 1931.
1930's
In 1934, Monsanto sold the Industri-Plex land. Between 1934 and 1968, the companies
that occupied the Industri-Plex site manufactured glue and gelatin by extracting collagen from
animal hides and chrome-tanned leather. Wastes generated by these processes included animal
hide residues and metals such as arsenic, lead, and chromium. These wastes were generally
deposited on top of existing waste deposits. Waste piles, including the animal hides, covered tens
of acres and reached heights of 40 to 50 feet above the natural grade.
19SO's
-------
74
In the 1950s, East Wobum began discussing the need for additional water supplies for the
city's expanding population. In 1958, the city hired |an engineering consultant to examine the
possibility of utilizing the Aberjona River water for drinking. Although the consultant concluded
that the water was too heavily contaminated to be safe for drinking, the city began constructing
two new wells, Wells G & H. i
1960's
Water Wells G & H, located on the east side of town, began operating in 1964 and 1967,
respectively, to provide Woburn's growing population with drinking water. At that time, these
two wells supplied 30% of the city's drinking water. The 330-acre Wells G & H site contains
five contaminated properties bounded by Route 128|to the south, Interstate 93 to the east, the
Boston and Maine Railroad to the west, and Salem Street to the south. The Aberjona River also
flows through the site, through on-site wetlands found immediately adjacent to both sides of the
river, and into the Mystic Lakes. The Mystic Lakes fire a popular recreational destination
including swimming and fishing.
Shortly after the installation of the two wells, residents of East Wobum complained that
the water smelled and tasted funny, corroded their pipes, discolored their dishwashers, and
stained clothing and fixtures. Prompted by citizen complaints, city officials tested the water from
the two wells. Test results only revealed high levels of salts in the water and officials
downplayed Citizen's concerns about the water.
In 1967, the Massachusetts State Department of Health (MSDH) recommended that Well
G & H be closed because of concerns about bacteria and only recommended their use in
conjunction with continuous chlorination. Soon residents reported concerns about their water
tasting like "bleach". In the spring and summer of 1968, residents complained about their "red
water," but city officials claimed it was due to the city's unlined old cast iron pipe. MSDH gave
the city permission to add sodium hexametaphosphate ("Calgon") to Well G & H to remedy the
problem and to adjust the water's pH content. In an attempt to find a long term water supply
solution that did not involve the use of Wells G & H, the Woburn City Council authorized the
Mayor to negotiate with the Metropolitan District Commission (MDC) about joining its water
system with the neighboring town of Stoneham. Starting in February 1969, the City of Woburn
-------
75
increased the chlorine feed at Well G by 50%. Due to resident's complaints, Well G was closed
for the winter starting in October.
At Industri-Plex, Stauffer Chemical Company (Stauffer) bought the last of the glue
manufacturers in the 1960s and ceased operation in 1968, In 1968, Mark Philip Realty Trust (MP
Trust), a real estate developer, bought the Industri-Plex land from Stauffer to develop it as an
industrial park, MP Trust began preparing for construction of a shopping mall and an industrial
park. In 1969, the project started illegally without a permit, though a permit was obtained a year
later. The construction at Industri-Plex involved moving piles of waste accumulated over 130
years and filling in low-lying areas and wetlands.
Early 1970's
Throughout the early 1970s, despite the repeated reassurances from city and state
officials, the citizens of East Wobum continued to express concerns about the water from Wells
G & H. In response to public pressure, Well G was frequently turned off in the winter when
water was more abundant only to be called into service again during hot, dry summers.
Voluntary water restrictions were put into place in 1972 and 1973 to avoid use of the wells,
while the city continued to work with MDC to provide a long-term water solution. In the summer
of 1974, water shortages forced the city to consider activating both Wells G & H, which caused a
"storm of protest" from residents. In 1975, a fire destroying the MDC pumping state at Spot
Pond in Stoneham interrupted construction of the water main connecting Woburn and MDC.
In 1975, as part of the development of the south end of Industri-Plex, 20 acres of animal
hide piles and animal glue waste were disrupted releasing hydrogen sulfide fumes into the air.
This "rotten egg" smell, extremely potent at times, elicited numerous complaints from citizens of
Wobum and the neighboring Town of Reading, downwind of Industri-Plex. Citizens and the
media referred to the omni-present stench as the "Wobum odor." This odor at times prevented
children from playing outside during noon recess at school and residents of affected areas
claimed that they could not use their yards. When the odor was strong, citizens working outside
complained of nausea, burning eyes, and difficulty breathing. Residents even mentioned that the
airborne chemicals caused the exterior paint on their houses to peel. Large open-air pits of waste
at the site allowed humans and animals to come into direct contact with the contaminants.
-------
76
According to one Wobum mother, "It was only a fiVe-minute walk and you see the open pits
filled with chemicals. We used to go blueberry pickling where they found the arsenic."
The Massachusetts Department of Environmental Quality Engineering (MDEQE) [now
the Massachusetts Department of Environmental Protection] issued MP Trust numerous violation
notices, orders to halt construction, and requests to cleanup wastes at the Industri-Plex site. In
1977, MDEQE and the Town of Reading filed a lawsuit against MP Trust demanding that MP
Trust be prohibited from disturbing the two parcels Iwhere the glue waste was buried. However,
MP Trust continued its construction because it had the permission of Massachusetts Department
of Health (MDOH) (which was responsible for hazardous waste management at that time) to
excavate and dispose of hazardous wastes on the site. "It [remediation] would make the land too
expensive to develop. It [the waste] can stay there as far as I'm concerned," said William
D'Annolfo, owner of MP Trust.
1978-1979
The major pollution problems at both of Woburn's Superfund sites were discovered
within six months of each other.
Abandoned 55-gallon Drum with the Entire Side Corroded; Found Near Wells G & H.
Wells G&H. In May 1979, construction workers discovered that 184 55-gallon barrels of waste
had been dumped near the Wells G & H. Immediately after this discovery, the MDEQE tested
the wells for possible contamination. These tests repealed high concentrations of several VOCs,
known carcinogens in animals, but indicated that the barrels were not the source. Officials were
particularly concerned about TCE and perc in the well water because TCE levels tested as high
-------
77
as 267 parts per billion and VOC levels tested as high as 100 parts per billion. (The EPA
considered anything more than 27 parts per billion and 5 parts per billion, respectively, to be
hazardous.) Additionally, river sediments were found to be contaminated with polycyclic
aromatic hydrocarbons (PAHs) and heavy metals such as chromium, zinc, mercury, and arsenic.
Adjacent soils also contained PAHs, poly-chlorinated biphenyls (PCBs), VOCs, and pesticides.
Anyone coming into contact, swallowing, or ingesting this groundwater, soil, or river sediments
would be at risk.
Although the EPA discovered these problems with Wells G & H in July of 1979, Wobum
officials and residents were not notified of the problem by the EPA, COE, or MDEQE until an
enterprising reporter released the information in a local newspaper story. The problems described
in the news story and the lack of notification by the official organizations involved, generated
much outrage and distrust on the part of local citizens. Likewise, contamination at Industri-Plex
was documented in federal and state COE and EPA records in August 1979, but local officials
were notified of the contamination not through official channels but, again, by a local newspaper
article reporting the results of EPA investigations conducted earlier that summer.
Industri-Plex. At Industri-Plex, the Massachusetts Department of Environmental Protection
(MDEP) asked the Army Corps of Engineers (COE) to investigate alleged wetlands violations
and help control the activities of MP Trust. After conducting a preliminary survey of the site, the
COE solicited the help of EPA. In late 1979, based on their discovery of illegal filling of
wetlands, the EPA and the U.S. Attorney's office (on behalf of the Army Corps of Engineers)
obtained a court order against MP Trust to stop development at the Industri-Plex site.
Additionally, the EPA discovered pits of buried animal hair and barrels of slaughterhouse waste.
In December, regional EPA officials requested funds for the installation of a permanent air
monitoring station for North Woburn.
Groundwater was contaminated with arsenic and VOCs, including benzene and toluene,
and soil was extensively contaminated with arsenic, chromium, and lead. Benzene has been
proven to cause leukemia. In fact, EPA investigations revealed a football field-sized arsenic pit
that rose 40 feet into the air. Arsenic was used in the production of lead-arsenate, an insecticide
that was replaced by DDT in the 1940s. Arsenic is a known human carcinogen and can cause
-------
! 78
skin tumors when ingested, and lung tumors when inhaled. Arsenic is also linked to
chromosomal damage in humans and animals. Measured arsenic concentrations in this pit
reached as high as 1,100 parts per million (ppm) and debris from the pit was detected on the
slopes of Route 93, a half-mile away (0.8 kilometers). Although EPA officials were uncertain
about when the arsenic was dumped, they believed it dates between 18!)9 and 1934, meaning that
the arsenic has been in Wobum soil for 85 to 100 years. Other contaminants permeated the site
as well. Recorded levels of chromium reached 3,000 ppm in one place and 78,000 ppm in
another. The concentration of lead was as high as I,i200 ppm. At the time, the standards for both
of these contaminants were 0.05 ppm.
Community Reaction. The discovery of two major hazardous waste problems in one town
prompted strong media interest as well as the active response and involvement of Wobum's
residents. Area newspapers and TV stations ran multi-part stories about Wobum, alluding to it as
a "toxic wasteland." Local newspapers and magazines featured articles with headlines such as:
i
"Lagoon of Arsenic Discovered in North Woburn" (The Daily Times, September 10,1979),
"Chasing A Radioactive Ghost" (The Daily Times, October 16,1979), 'Deaths From Cancer
Increase in Wobum" (The Daily Times, December 12, 1979). In particular, The Daily Times
published two notorious articles about Wobum's hazardous waste contamination, which reported
higher rates of adult and childhood leukemia, bone and skin cancer, prostate cancer in men, and
breast cancer. However, the estimates quoted in the article were not confirmed by MDOH until
the folio wing, spring. Interestingly, in its final report^ the MDOH used the same statistics
reported in The Daily Times months earlier.
One east Woburn resident, Anne Anderson, began to suspect a link between the well
water and her son's leukemia. "From the time we moved here, the water was bad in the summer.
It had an unpleasant odor and a terrible taste," she later recalled. "My mother brought jars of
MDC water when she came to visit. The kids used to always ask for 'Nana water.' It was like a
mother's milk, for God's sake. She still brings it when she visits." Anderson began recognizing
some of her neighbors at the hospital, who were alsb there with children suffering from
leukemia. "I just don't see where all the leukemia cases in our area aren't correlated," she said.
"It seems they have to be. The thing that strikes me is there are two neighbors off of Pine Street
-------
79
who have children with leukemia. A year later people on the other side of the street had their
child diagnosed and two people we know personally were diagnosed." Her husband continued,
"Before, in all of my life, I knew of only one child with leukemia. But these are all in Woburn."
Anderson had requested that city officials test the water, but she was informed that it was not
standard procedure to perform such tests on the basis of one individual's request. Unsatisfied
with this response, Anderson, with the help of her minister Reverend Bruce Young, convened a
meeting in September 1979 for parents of children with leukemia. Attendees of that meeting
counted the number of local leukemia cases and mapped the homes of the sick children—eight
leukemia cases were clustered within one square mile (1.6 square kilometers) in East Wobum,
six in a six-block square served by Wells G & H.1 Sparked by these findings, the citizens of
Wobum formed the group For A Cleaner Environment (FACE) in October 1979. Two months
later (December), FACE and the doctor treating Wobum children with leukemia convinced the
Wobum City Council to contact the Centers for Disease Control (CDC). After examining the
situation, the CDC found that Wobum's childhood leukemia rate was two to three times that of
the national average and four times the average of other communities the size of Wobum. As
CDC described it, Wobum had the most persistent leukemia cluster it had ever seen.
Later that December, MDOH released a preliminary report on the second five-year study2
of the health effects of Woburn's drinking water, which contradicted the clustering of leukemia
in East Wobum and the pronouncements of CDC. It stated that Woburn had a higher than normal
incidence of many cancers but that there was no "association between environmental hazards and
the incidence of childhood leukemia." (As later determined, the state had used in its
calculations, a population estimate for Wobum, taken from the 1970 Census, which was much
greater than Wobum's population at the time the study was conducted.) When MDOH corrected
this inaccuracy, its calculations revealed several statistically significant rates of cancer and
leukemia.
1980-1982
1 Between 1964 and 1997,28 leukemia cases were diagnosed in Woburn. Of these 28 cases, 16 resulted in death.
(The last case of documented childhood leukemia in Woburn was reported in 1986.)
2 Dr. Robert Tuthill and Dr. Leslie Lipworth, of the University of Massachusetts, conducted the first five-year study,
which found a slightly elevated but statistically insignificant increase in Wobum's rates of cancer and leukemia.
-------
80
Wells G&H. Initial EPA investigations of potential contamination at Wells G & H began in
1981. Per these investigations, the EPA divided the site into three areas, or "operable units", and
identified five likely sources of pollution. The operable units included five properties inside the
site boundary, the area immediately surrounding the wells, and a segment of the Aberjona River
and adjacent wetlands. Three of the sources of pollution that EPA identified were W.R. Grace,
UniFirst, and the John J. Riley Tannery (Riley) which had been purchased by Beatrice and then
again purchased by Wildwood Conservation Corporation. Grace operated an equipment
manufacturing plant located about 2,500 feet northeast of the wells; the firm used solvents at
several points in the manufacturing process. The Riley Tannery, and an adjacent 15 acre
property, was bought by Beatrice in 1978 and sold back to Riley in 1983, but Beatrice retained
legal liability for environmental matters at the tanne(ry property. UniFirst, located about 2,000
feet north of the wells, used perc as part of its industrial dry-cleaning business. The other two
sources of contamination were New England Plastics and Olympia Nominee Trust. Final testing
conducted in September 1988 confirmed that groundwater contamination emanated from
pollution at these properties. On December 30, 1982, the EPA proposed adding Wells G & H to
theNPL.
Installation of a groundwater monitoring well near Wells G&H.
Industri-Plex. In 1980, the EPA allocated $150,000 for an investigation of the Industri-Plex site,
which revealed major pollution problems. In May of 1980, ajudge ordered MP Trust to halt
construction until it designed, with the help of MDEQE, an appropriate cleanup plan for the site.
-------
81
Also in 1980, MDEP placed a latex cover over the inorganic wastes at the site. At that time, the
site contained streams, ponds, remnant manufacturing, buildings, a warehouse, office buildings,
and waste piles.
The EPA began negotiating remediation with the primary polluting parties. On October
23,1981, the EPA proposed the Industri-Plex site for inclusion on the NPL. The EPA installed
chain link fence. This fence was subsequently damaged by ATVs and was not permanently fixed
until 1986. During this time period, illegal dumping occurred at the Industri-Plex site.
Unlike the lengthy lawsuit with MP Trust over the Industri-Plex site, negotiations with
Stauffer were expeditious. In May 1982, Stauffer signed a Consent Decree with the EPA and
MDEQE to undertake a remedial investigation and feasibility study (RI/FS) for the site.
Community Reaction. In May 1980, the CDC and the National Institute for Occupational Safety
and Health initiated a more detailed study of Wobum's rates of leukemia, which confirmed the
presence of elevated levels of kidney cancer and childhood leukemia. However, the final report
stated that the results of the study were inconclusive because of the lack of data prior to 1979. It
also failed to attribute elevated levels of leukemia to hazardous wastes, "The information
gathered thus far fails to provide evidence establishing an association between environmental
hazards and the incidence of childhood leukemia... in Woburn." The public was outraged and
felt betrayed by this persistent stonewalling from governmental agencies. According to local
residents, state and city health officials worked to preserve public health in theory only. The
media continued to document concerns about the sites including "Workers Near Waste Site
Complain of Headaches, Fatigue" (The Daily Times, July 2, 1980), and "Toxic Waste: One Year
Later, Still No Answers" (The Daily Times, August 1, 1980).
The reports fueled community activism and led FACE to question the validity of the
reports. Seven months after their initial meeting, FACE convened a group of state and federal
agency representatives to discuss the plight of Woburn residents. At that meeting, the EPA
agreed to investigate the wells. An EPA report released later that year confirmed what the public
already knew, that high levels of contamination were present in groundwater, particularly in the
areas of Wells G & H. This was the first of FACE's many victories, as the group was
instrumental in the remediation of both the Wells G & H and the Industri-Plex sites.
-------
, 82
In 1982, eight Woburn families filed a highlv publicized $400 million lawsuit against
several industries alleging that they had contaminated the aquifer for the two wells, and that this
contamination caused the high rate of childhood leujkemia and other health problems in Wobum.
While the court case proceeded, FACE continued its grassroots advocacy work and held
numerous public meetings to mobilize community leaders and local residents. Two professors
studying the clustering of disease heard of FACE'S struggles and invited activists Anne
Anderson and Reverend Bruce Young to present the Wobum case at the Harvard School of
Public Health (HSPH). As a result of this presentatipn, the HSPH and FACE collaborated on a
more detailed study of environmental contaminants at Wobum. The HSPH designed and
administered'a public health survey of the area withithe help of FACE, which coordinated 235
volunteers to implement the survey. Between April and September, 54% (3,257 households) of
all Woburn residents answered the survey. The results revealed a clear linkage between
leukemia, fetal and newborn deaths, birth defects, and childhood illnesses within the
I
neighborhoods that received most of their water from Wells G & H. The survey also found that
the well water caused ten times the expected rate of ptillborn births. As Reverend Young
described the situation, "For seven years we were told that the burden of proof was upon us as
independent citizens to gather the statistics.... All omr work was done independent of the
Commonwealth of Massachusetts. They offered no support, and were in fact one of our
adversaries in this battle to prove that we had a problem."
Millions of dollars and several years were devoted to the Wobum court case which
j
commanded front-page national media attention. The book describing the lawsuit, A Civil Action,
was published in 1996 and became a bestseller. In 1999, the book was made into a movie
starring John Travolta
1983-1985
Wells G&H. On September 8,1983, Wells G & H 'were officially listed on the NPL. In 1983,
the EPA issued its first order requiring Grace, Beatrice, and UniFirst to begin initial
investigations on the contamination at their properties affecting Wells G & H. In 1985, the
EPA's second order mandated that Wildwood Conservation Corporation fence its property and
hire a 24-hour guard to prevent any additional human contact with the contaminants present on
-------
83
that property. An EPA Technical Assistance Grant, awarded to FACE, allowed the community to
hire a consultant who could interpret technical information and reports about the site. FACE
heavily utilized the expertise of their consultant, providing community members the opportunity
to actively participate in the development of the Record of Decision (ROD) and final remediation
guidelines for the site.
Industri-Plex. On September 8,1983, the EPA placed the Industri-Plex site on the NPL. Stauffer
completed the RI/FS in April 1985 and found that arsenic contamination was even greater than
initially suspected. Stauffer's investigation also revealed that the northeast section of the
property would require only groundwater monitoring and might be appropriate for future
development. Finally, the report concluded that although Wells G & H and Industri-Plex were
hydraulically connected, contamination present at Wells G & H is the result of pollution dumped
not at the Industri-Plex site, but at a location south of Route 128. The EPA ordered a full-blown
investigation of the entire 330-acre Wells G & H site. Stauffer chemical signed a consent order
with EPA to pay for its apportioned share of the remediation efforts. In May 1985, the parties
approved decrees requiring MP Trust to investigate and cleanup the site, but MP Trust never
undertook these activities citing financial concerns.
1986-1988
Wells G & H. In 1986, after five years, the $400 million lawsuit filed by the eight Woburn
families went to trial. The initial trial lasted only 80 days and none of the surviving plaintiffs
ever took the witness stand to talk about their loss resulting from contamination at Wells G & H.
Wobum residents were embittered by the results of the verdict:
• UniFirst Corporation (UniFirst) settled for $1.05 million prior to the trial without
admitting any wrongdoing.
• Although ajury found Grace & Company (Grace) negligent, a district judge dismissed
the ruling because of inconsistencies in the evidence. Grace eventually settled for $8
million without admitting any wrongdoing. After lawyer's fees, each family received
approximately $300,000.
-------
84
• A jury dismissed the charges against Beatrice Foods (Beatrice), but the judge reopened
the case because of legal misconduct on the behalf of Beatrice's lawyers.
In 1986, an EPA administrative order required Olympia Nominee Trust to remove all
drums and debris from the western portion of its property. Additionally, in 1987, EPA issued an
administrative order requiring Uni First to install monitoring wells and remove contaminants near
Wells G & H. Also in that year, the U.S. Geological Survey reports that approximately 50% of
the water for Wells G & H originated, from the polluted Aberjona River. In 1988, the EPA
conducted a detailed investigation showing the groundwater contamination from the five
properties near the wells.
Industri-Plex.. Also in 1986, to restrict access to the Industri-Plex site, the EPA ordered the fence
to be fixed and a 10,000-foot extension. A year and a half after the RJ/FS was complete, in
September the ROD was finalized; the EPA published its ROD describing the remedies selected
for the Industri-Plex site. The remedy consisted of five elements:
• The "soil remedy" called for installation of a permeable cap over 105 acres to prevent
physical contact with soils and sediments contaminated with high concentrations of lead,
arsenic, or chromium.
• The "air remedy" called for placement of an impermeable cap over five acres of the site
to prevent water infiltration and gas release, and installation of a gas collection and
treatment system.
• Interim groundwater treatment of a benzene/toluene "hot spot" on the site to reduce
concentrations and limit migration of the chemicals.
• Investigation into and development of a plan1 for treatment of groundwater and surface
water.
• Implementation of institutional controls to limit the future use of the site because
available cleanup technology could not provide the safety necessary for unrestricted use.
-------
85
The total cost of site investigations and remedial actions was estimated to be $50 million.
Although cleanup remedies for the Industri-Plex site were identified in 1986, actual cleanup
activities did not begin in until 1993.
The fence around the Industri-Plex site was completed in 1988, but shortly thereafter dirt
bikes and ATV riders again destroyed a section of the fence, and several barrels of waste were
dumped illegally at the site. Three months after the repair, the fence and posted warning signs
were demolished by vandalism.
1989-1991
Wells G&H. In 1989, the EPA granted Woburn a Technical Assistance Grant enabling the
community to hire a technical advisor to help them better understand the technical aspects of the
contamination and remediation efforts and take an active part in decision making processes for
Wells G & H. On September 14, after incorporating issues mentioned in the public comment
phase, the EPA released its final ROD. The ROD addressed the properties contained within the
site, the accompanying groundwater contamination, and the subsequent investigations of the
other two operable units of the site.
In July 1991, after only four months of negotiation, the EPA finalized a "record-
breaking" settlement with four of the five PRPs for the Wells G&H site (Grace, UniFirst,
Beatrice, and North Eastern Plastics). At the time, it was the most expensive Superfund
settlement ever achieved in New England. Although an agreement with the fifth PRP, Olympia
Nominee Trust, was never reached, the comprehensive cleanup of the G&H Wells site began
immediately upon the closing of this multi-million dollar deal. The settlement stipulated that the
companies would:
* Clean up their own properties simultaneously, at a collective cost of approximately $68.4
million,
• Provide funding for EPA's oversight of cleanup activities, valued at $6.4 million,
• Conduct a risk assessment of the area immediately surrounding Wells G&H, and
• Reimburse the EPA for its investigation studies, which cost approximately $2.65 million.
t
-------
86
Due to the amount of contamination present at the site, agency officials knew long-term
remedial plans for the site would be critical and remediation of groundwater resources would be
extensive. These plans included excavating and incinerating 2,100 cubic yards of contaminated
soils, extracting toxins from soil vapors (this entailed, literally, suctioning the toxins out of soil),
and pumping groundwater from the underground aquifer and returning it after treatment.
Although the problems at both the Wells G & H and Industri-Plex sites were identified in
1979, local citizens grew increasingly frustrated by ihe 14 year delay in remediating Industri-
Plex and the 13 year delay in remediating Wells G & H.
Industri-Plex. Trespassing on the Industri-Plex site ceased after the Industri-Plex Site Remedial
Trust (ISRT) established its office on the site in 1989 and posted 24-hour security guards.
After five years of arduous negotiation, MP Trust signed a Consent Decree to investigate
contamination at the site and resolve wetland infilling violations. Unable to comply with terms of
this agreement, MP Trust filed for bankruptcy. A Consent Decree was signed for Industri-Plex
site in April 1989, Monsanto, Stauffer-ICI (ICI Americas, Inc, purchased Stauffer in 1987), and
twenty smaller other PRPs established the Industri-Plex Site Remedial Trust (ISRT) to
implement the agreement. In addition to forming a remedial trust, this Consent Decree allowed
Monsanto and Stauffer-ICI to create a Custodial Trust which would technically own title to
contaminated areas of the Industri-Plex property, protecting Monsanto said Stauffer-ICI from
liability relative to the site, attempt to avoid conflicts among PRPs, and set-up a mechanism to
sell the land after completion of the remediation of s,ite. A key feature of this agreement required
that the recently bankrupt MP Trust sell all of its Industri-Plex holdings to fund its share of
remediation of the toxic contamination and in return; to have no additional liability.
1992-1994
Wells G&H. One year after signing their multi-million dollar agreement with the EPA, in
September 1992, UniFirst and Grace began groundwater remediation, and Wildwood and New
England Plastics began soil excavation. This progress was viewed as a mixed blessing.
Government officials applauded the PRPs for ultimately accepting responsibility and then acting
expediently, but residents felt betrayed. According tp Gretchen Latowski, director of FACE,
-------
87
"The Wobum experience is the ultimate failure of Superfund. It took 12 years since the problems
at the wells were first identified before anything was done or any responsibility taken."
Excavation of contaminated soil near Wells G & H.
Industri-Plex. Although cleanup remedies for the Industri-Plex site were identified in 1986,
actual cleanup activities did not begin in until 1993 when construction began on the permeable
and impermeable cap for Industri-Plex site. By the summer of 1994, EPA had approved 100% of
the remedial design. Also in the early 1990's the EPA and the ISRT amended the groundwater
remedies listed in the Industri-Plex ROD (Element 3) because of unanticipated contamination,
advancing technology, and cost efficiency. As a result of this change, groundwater was treated
by a pilot oxygenation and bioremediation process as opposed to the original remedial
prescription to pump groundwater, strip it of contamination, and return it to the aquifer.
Industri-Plex during remediation: site with cap on contaminated soils.
t
1995-1997
Wells G&H. On April 27, 1995, Jeffery Purvis, Chief of the Community Assessment Unit of
the Bureau of Environmental Health Assessment, reported on the status and conditions of Wells
-------
88
G & H. Though not clearly identified, the well field was easily accessible and a No Trespassing
sign was posted. However, the area was not fenced |and rolls of fold fencing and piles of concrete
piping lay in the area. The site did not appear to have been accessed recently, though trash,
including clothing, furniture, tires, and other debris,j littered the site. In 1995, the UniFirst
Corporation and W.R. Grace and Company installed LJV-oxidation systems to treat groundwater
in area bedrock.
The bestselling book, A Civil Action, was published in 1996.
After extensive but fruitless efforts to bring Olympia Nominee Trust to the table, the EPA
in 1997 agreed to remediate that part of the site with money from the Superfund trust. Initial tests
of the Olympia Nominee Trust property began in September 1997.
Industri-Plex. By 1997 the soil and air "remedies" ordered in 1986 for Industri-Plex were in
place (Elements 1 and 2), the groundwater treatment to reduce the benzene/toluene "hot spot"
was only partially implemented (Element 3), and the instructional controls to determine future
use of the site were still not finalized (Element 5).
1998 and later
Wells G & H. In 2000, North Eastern Plastics completed the remediation of the soil and water
contaminants related to Wells G & H. In 2003, six years after reaching agreement with the other
four responsible parties, EPA reached an agreement with Olympia Nominee Trust, the fifth
source of pollution for Wells G & H. The EPA entered into administrative order by consent with
1 i
all parties this year to address PAH and PCP contamination.
Phase II investigation of the groundwater contamination, beyond the five sources
continues. As part of Phase III, which focuses on the Aberjona River, the EPA prepared a risk
assessment.
Industri-Plex. At the Industri-Plex site, the gas collection and treatment system were completed
in December 1998. Groundwater remediation and monitoring constitute the bulk of the
remaining work to be done. As of December 1998, the three groundwater treatment plants
t
-------
89
operating at the site had pumped and treated more than 150 million gallons of water. In addition,
all the contaminated soil was removed (approximately 150 tons) and 1,360 pounds of VOCs had
been destroyed. Although major strides in the remediation of the property have taken place,
according to EPA lawyers, complete cleanup of the site will cost approximately $80 million and
could take another 20-30 years because of the site's extensive groundwater contamination. An
investigation of groundwater contamination at the Industri-Plex site began in 1999. Initially it
was expected to be completed by early 2000, however, it was not finished until 2003.
State and local governments and EPA officials have been successfully working together
to promote commercial redevelopment of the Industri-Plex site. For example, with the support of
state and local officials, the EPA pursued prospective purchaser agreements which limit the
liability of property purchasers and, in some cases, offered previously contaminated properties at
reduced rates. Plans included having Home Depot and Target serve as anchor stores of the 110
acres of commercial development at the site, with 35 acres being devoted to a regional
transportation center, and 100 acres being preserved as wetlands and open space. As one
development engineer recently observed, foxes and snapping turtles now frequent Industri-Plex
wetlands.
Industri-Plex after remediation: regional transportation center.
Though soil data was collected in 1995 and 1997, additional sediment and surface soil
sampling was conducted in 2000 and 2001 to fill in data gaps of the Human Health and
Ecological Risk Assessments. In 2002, sediment and soil samples were collected along the
-------
90
Aberjona River to evaluate the potential impacts from the river, including on the cranberry bog,
on flood source soil conditions, and on areas the City of Woburn was interested in developing in
the future.
-------
91
Chapter 5
Montclair, New Jersey
5.1
Overview
Montclair, Glen Ridge, and East and West Orange Townships are located about eight
miles (12.9 kilometers) from Newark Airport in northern New Jersey. These towns are densely
populated, and are located in one of the most densely populated regions of the United States. The
Montclair/West Orange Radium Superfund site consists of 366 residential properties on 120
acres in Montclair and West Orange (Figure 5.1). The Glen Ridge Radium Superfund site is
comprised of 306 properties on 90 acres of residential land in Glen Ridge and East Orange. The
soil at both sites is contaminated with radium, a naturally occurring element which can result in
high levels of radon gas and gamma radiation in nearby homes. The Center for Disease Control
(CDC) and the New Jersey Department of Health (NJDOH) declared these sites to be a public
health hazard due to concerns about lung cancer. Montclair/West Orange and Glen Ridge were
listed on the NPL for Superfund sites in 1985 because of their proximity to radium waste
generated by radium processing. These plants had operated in the area after the turn of the 20
century, and an estimated 200,000 cubic yards of contaminated material were placed on private
and public areas in the communities.
Initial attempts to remove the contaminated soil were hampered by the lack of suitable
waste depository, resulting in 4,902 drums and 33 containers of soil being stored for nearly two
years on the yards of partially excavated properties in Montclair. In 1999, nearly 20 years after
the initial identification of the problem and 12 years after being put on the NPL, cleanup
activities continued to occur as the streets are replaced and the EPA continued to investigate the
possibility of additional groundwater contamination. By 1998, a total of $175 million had been
spent to remediate over 300 houses and remove 80,000 cubic yards (or 5,000 large truck loads)
of contaminated soil. In 2004, estimates of total cleanup exceeded $200 million.
-------
92
Figure 5.1 West Orange, Montclair, Glen Ridge Sites
•::*!. r*-s=."-*/w=.-. ;•- -'•
^m<'m-'^:'
5.2 Timeline and History
1915-1926
Northern New Jersey was a center for radium processing and the dial-painting industry in
the 1900's. Several plants occupied the area, the largest of which was the U.S. Radium
Corporation (formerly the Radium Luminous Materials Corporation) which operated between
1915 and 1926. Because of its luminescent properties, radium was added to the paint that was
used for numbers on watch dials and instruments, which became especially popular during
World War I. At various times, U.S. Radium Corporation employed between 100 to 300 young
women as watch dial painters. Corporate supervisors instructed the painters to lick the tips of
their brushes to create the fine point needed for the detailed work. As a result of this process, the
-------
93
dial painters ingested a significant amount of radium. Since clothes were often speckled with this
luminous paint, many of the dial painters glowed in the dark by the time they returned home
from work. Women sometimes painted their fingernails, buttons, or eyelids with the paint for its
special glow-in-the-dark effects. One worker reportedly painted her teeth with the luminous paint
in preparation for a date after work one evening. Because of their exposure to radium, some of
the workers developed bone diseases, such as necrosis of the jaw, anemia, and cancer. Many
factors, including health problems and the discovery of richer uranium ore in the Belgian Congo
led to the closing, and eventual destruction, of New Jersey's radium plants.
"POISONED! - as They Chatted Merrily at Their Work
Painting the Luminous Numbers on Watches, the Radium Accumulated in Their Bodies, and
Without Warning Began to Bombard and Destroy Teeth, Jaws and Finger Bones. Marking
Fifty Young Factory Girls for Painful, Lingering,
But Inevitable Death"'
f
Source: Hearst Sunday supplement American Weekly, February' 28, 1926.
-------
94
The U.S. Radium Corporation extracted and|purified radium from camotite ore found in
New Jersey spils. It processed approximately one and a half to two tons of ore each day, and
generated large amounts of radium contaminated waste. This waste was stored temporarily on-
site, and then transported and dumped on nearby rural areas.
1930's
During the 1930's, the contaminated soil was used as fill to prepare approximately 200
acres of low-lying areas in Essex County, New Jersey, for residential development.
i
Contaminated soil was also mixed with cement for foundations and sidewalks. An estimated
200,000 cubic yards of contaminated material is believed to have been disposed of on private and
public lands within the communities before the area was fully developed. This area,
contaminated by radium infill, eventually gave rise,to the townships of Montclair, Glen Ridge,
and East & West Orange.
The Montclair site was almost entirely residential with some small businesses and a park
nearby. Within a half-mile (0.8 kilometer) of Montblair were five schools, a hospital, one health
care facility, and a nursing home. West Orange was also predominantly residential with some
businesses located on the north/northeast side of the site. Located within one half-mile of the
West Orange site were two schools and one hospital. The Glen Ridge site was primarily
residential, and encompassed a park and had several small businesses nearby. No hospitals were
located near the Glen Ridge site, but three schools exist within a half-mile.
A USEPA contractor takes gamma radiation measurements in Montclair.
t
-------
95
Pre-1987
The New Jersey Department of Environmental Protection (NJDEP) discovered radium
contamination in these communities in 1979 during an investigation of the former radium
processing facilities. In 1981, the NJDEP requested EPA funding to perform an aerial gamma
radiation survey of 12 square miles (19.3 kilometers) of Essex County to identify possible
contamination from offsite disposal of radium processing waste. This survey identified three
distinct areas of elevated radiation in the townships of Montclair, Glen Ridge, and West Orange.
The NJDEP conducted additional screening surveys and ground readings in Glen Ridge
in July 1983, and in Montclair and West Orange in October 1983. As part of this screening, 17
homes in Montclair and 10 homes in Glen Ridge underwent additional testing for radiation, 19 of
which were found to exceed federal safety standards for radium (13 in Montclair, 6 in Glen
Ridge).
NJDEP officials were planning to notify local government officials and residents of their
findings in early December 1983. However, despite a request by NJDEP officials to hold the
story until official notification had been made, a November 30th television news report broke the
story early. According to the New York Times (October 16, 1984) article published one year later,
"[Many] residents of the three communities - Montclair, West Orange and Glen Ridge - were
not told about the problem until... technicians, wearing protective gear began taking soil and air
samples in and around their homes." Within a few days of the television news report, New
Jersey Governor Kean convened a news conference to make an official announcement about the
radium contamination. That month, local newspapers reported extensively on the areas of radium
contamination in their communities. Response to the announcement of radium contamination
was immediate, widespread, and occurred at many levels from local neighborhood residents to
federal agencies. With this heightened public concern, the EPA immediately installed temporary
radon ventilation systems in 38 homes and gamma radiation shielding in 12 homes.
A combined federal and state task force formed in December 1983 to devise a
comprehensive sampling plan that would better define the areas of contamination and provide a
benchmark for remedial action. This plan included an initial screening of "grab" samples which
showed above background radon levels. Long term units for continuous sampling and
-------
96
monitoring were later installed. Technicians in protective suits performed surface gamma
radiation surveys on the properties surrounding affected homes. Soil samples were also collected.
In response to local citizens' and officials' concerns, two schools in Montclair were tested but no
contamination was found.
At the local level, community members received different messages from different
agencies about the health risks involved. NJDEP Commissioner Hughey portrayed radon as
strictly an environmental problem. A New Jersey Department of Health (NJDOH) representative
said that the only known health risk was lung cancef, and a NJDOH representative was made
available to meet with affected families for advice. According to the EPA, risks associated with
radon were equivalent to the risks associated with cigarette smoking. A couple of news reports,
however, referred to the radium contamination in N|sw Jersey as "another Love Canal," since
both residential areas were built on contaminated soil. Even EPA officials expressed great
concern, among themselves, about this case, because it identified the first residences built
directly on contaminated ground. In response to thi^ mixed information, the Montclair Township
Council formed its own task force in December 1983, which held its first organizational meeting
that month. Montclair residents from the contaminated neighborhoods also formed a Radiation
Ad Hoc Committee in December 1983.
Within a month of the November 30, 1983, news report, three bills to aid the affected
communities were introduced into the New Jersey state legislature. One bill requested cleanup
funds, another victim compensation, and the third would require the NJDEP to investigate any
potentially affected homes at the owner's request and provide a certificate of clearance if the
property was not contaminated. The latter bill was passed in January 1984, and by July 20,1984,
some residents received certification that their homes had been tested and were clean.
The EPA began field investigations in Glen Ridge and Montclair in January 1984 to
determine the boundary of the contamination and to quantify gamma radiation and radon levels
in the affected areas. Residents were asked for perrnission to collect samples and were then
provided with the results for their homes. Field investigations continued throughout the fall of
1984. By the end of the year, the EPA had completed all radon source characterization surveys in
the three townships.
-------
97
The NJDOH conducted an epidemiological assessment of the three radium sites and
found a possible, but not statistically significant, increase in lung cancer among white males at
these sites. As a result of this study, the CDC and NJDOH declared these sites to be a public
health hazard. The Centers for Disease Control (CDC) released a Public Health Advisory for
Glen Ridge and Montclair which quantified health risks and recommended appropriate remedial
actions. The CDC divided homes into four categories based on their levels of contamination and
the actions necessary to reduce human exposure contamination:
• Level I homes required remedial action within two days and restricted smoking and time
spent in high radon level areas of the homes.
• Level II homes necessitated remedial action in 1 -3 months.
• Level III homes necessitate remedial action within 1-2 years.
• Level IV homes required no action.
Prescribed actions included the installation of remedial systems such as dilution air fans,
air filtration systems, and sealing foundations. The plan also suggested additional studies to
determine the boundaries of the contaminated areas, locate and characterize the source of the
contamination, and assess the potential for groundwater and vegetation contamination.
All this testing and EPA attention prompted local realtors to gather at a Task Force
meeting to discuss the potential impact of radon on the housing market. Reactions were mixed.
Some realtors reported that there was no decrease in the number of homes sold. However, other
realtors reported, anecdotally, a decrease in the selling prices of the homes i/the homes were
known to be in affected areas. Fearing lower property values, local residents felt strongly that
they should not have to pay full property taxes on their homes. In response to these concerns, the
Essex County Board of Taxation granted to petitioners in 1984 tax relief on 39 properties in
Montclair: 20% relief if there was soil contamination and 50% relief if radon levels required
installation of ventilation systems. The County also granted tax relief for 22 properties in Glen
Ridge, but the town appealed this in State Court. In West Orange, tax relief was granted for eight
properties - including some adjacent to contaminated properties.
In June, the media agitated already deepening concerns with a spate of negative publicity
which included local newspaper articles and several special news features on major television
-------
98
stations. Residents and officials of Montclair grew more concerned, and decided to hire a private
consultant, David Rosenbaum, a former EPA official. Many of Rosenbaum's views were
published in the local newspaper, including: 1) levels of radiation are the highest ever
determined in a dwelling in the U.S.; 2) residents in these homes were exposed to "some of the
highest concentrations of any carcinogen ever recorded"; and 3) "Affected residents are in
considerably more danger than the people who once, lived in the Love Canal region in NY."
Rosenbaum concluded that the contaminated soil should be dumped in the ocean and that the
i
EPA could approve this solution (Montclair Times, October 11.1984). Residents of Montclair,
Glen Ridge, and West Orange filed suit against the U.S. Radium Corporation. The 7-year legal
battle went to the State Supreme Court and ended im 1991 when 237 residents received a $4.2
million settlement (average $18,000/house) from remnants of the U.S. Radium Corporation.
The entire backyard of this property in Montclair was excavated during cleanup.
A joint EPA/NJDEP task force identified 12 homes in the three communities for a
proposed pilot study involving soil excavation and removal. The EPA decided to postpone the
pilot study until the completion of the requisite feasibility study, scheduled to begin later that
year. The NJDEP, however, planned to proceed with the pilot study on its own. The search for an
acceptable disposal site stalled progress of the pilpt program and ultimately generated a great
deal of public anger and distrust of the NJDEP, In August, the NJDEP proposed to temporarily
-------
99
store contaminated soil from the pilot study at the National Guard Armory in West Orange.
Residents of West Orange strongly opposed this plan and the proposal was withdrawn. The
Montclair Township Council was then asked to help locate a disposal site in Montclair for the
radioactive waste. In response, the Council passed a resolution stating that the Township would
under no circumstances comply with the request because: 1) complete cleanup was the only
acceptable alternative; 2) high population density and development pressures excluded the
possibility of local storage sites; and 3) securing a disposal site was a state and federal
responsibility.
As the search for a suitable storage site for the contaminated soil became a major
problem, the NJDEP considered several other options, but significant resistance from people
living in or near the potential sites caused each of these possibilities to be abandoned. For
example, protests and demonstrations were held to prevent storage of contaminated soil at a
dump site on the Montclair State College campus. During this time, questions were raised about
the desirability of a pilot project. Those in favor of the project said that it would shorten the time
it would take the EPA to start cleanup because the project would demonstrate the feasibility of
the cleanup approach.
In the fall of 1984, the NJDEP and federal officials considered simply buying and
fencing-off the contaminated properties. This option deeply concerned the townships, which
soon deemed it the least desirable alternative because it would transform whole neighborhoods
into "radium dumps." Finally, in September 1984, the Montclair Township Council and
community task forces announced their intention to undertake legal action against the EPA and
the NJDEP to facilitate timely removal of contaminated soil.
The Montclair/West Orange and Glen Ridge sites were added to the proposed NPL in
October 1984, and the EPA began its Remediation Investigation/Feasibility Study (RI/FS) the
following month. Three months later (January 1985), both sites were added to the final NPL. In
April, the EPA finalized its RI/FS and submitted it to the public for final approval.
Also in April, the New Jersey Governor signed a bill appropriating eight million dollars
for the pilot study (this was later increased to $15 million). With this money, the NJDEP was
able to locate a storage site in Beatty, Nevada, and the pilot study began with four homes in Glen
Ridge in June of 1985. In August, five Montclair families were temporarily moved from their
-------
100
I
homes for an expected two months and excavation pf the contaminated soil began. Prior to its
shipment to the permanent storage site, approximately two-thirds of the contaminated soil
excavated for the pilot project (9,500 drums and 51 (containers) was stored in Keamy, New
Jersey, and the remainder of the soil (4,902 drums and 33 containers) was stored in the yards of
the partially excavated properties in Montclair.
Immediately preceding tiie completion of the pilot project and soil shipment, the state of
Nevada revoked the NJDEP's disposal permit. In October 1985, the U.S. Supreme Court directed
the NJDEP to look for a disposal site within the static. Once again there was no place to store the
contaminated soil from the complete remediation of four homes in Glen Ridge and the partial
remediation of five homes in Montclair. Consequenjtry, almost 5,000 containers of contaminated
soil remained in the yards of the partially excavated Montclair homes.
In July 1986, the NJDEP made plans to ship, the excavated soil 10 an abandoned quarry in
Vemon, New Jersey, where it would be blended with clean dirt to bring radium levels down to
acceptable levels. Thousands of Vemon residents vigorously protested against this plan by
obtaining temporary restraining orders and demonstrating with chants of "Hell no, we won't
glow." Several state and federal lawsuits were also filed against the NJDEP, and the NJDEP
dropped the plan in November. As the search for a storage site dragged on, the plight of the
relocated Montclair families continued to be the subject of media attention. Three hundred
people from Montclair and surrounding communities rallied in support of the indefinitely
displaced families.
1987-1989
The NJDEP made several offers to buy these five properties from their owners at market
value - as if the homes were not contaminated -- and to pay relocation costs, but these offers
were refused. The NJDEP also offered to bury the contaminated soil more deeply in the yards
and install filtering systems. This offer was also refused. The Township of Montclair again filed
suit in State Supreme Court to force the NJDEP to remove the barrels. Judgment was passed in
March 1987 requiring the NJDEP to start removing the barrels by May 15, 1987.
Eventually, the EPA negotiated a disposal sife, and in December 1987 the NJDEP spent
almost four million dollars to ship the barrels of soil to Oak Ridge, TN where the soil was mixed
-------
101
with radioactive waste from power plants and shipped to a storage site in Washington State.
Likewise, the soil stored at Keamy, NJ was shipped out of state in the summer of 1988.
An excavator removes radium-tainted soil from Barrows Field in Glen Ridge.
In April 1989, the EPA released its draft remediation plan and held public meetings for
discussion and comment. The EPA's $53 million action plan called for a five-tiered approach to
remediation based on the level of contamination found in the homes. In June, the first Record of
Decision (ROD) was signed. It established five classifications related to the level of
contamination and subsequent required remediation:
Tier A (23 homes): Complete soil removal and replacement of the most contaminated
areas.
Tier B (75 homes): Covering of contaminated soil and installation of radon control
systems in homes with very high radiation levels.
Tier C (65 homes): Installation of anti-radiation devices in less severely contaminated
homes, which would also be subject to deed restrictions and other
controls.
Tier D (296 homes): Monitoring homes with low levels of radon.
Tier E: The homes had no evidence of radium contamination and would
receive no further action.
-------
102
Tlie public staunchly opposed this plan and all proposed remedial efforts short of complete
removal of contaminated soil. In response to public concerns, the EPA installed a fence around
two of the sites to prevent the public from coming irpo contact wi th hazardous materials. They
began removing soil at the most contaminated homes and extended the comment period on the
plan, while deferring decisions on the less contaminated homes.
In 1989, Dr. William Kinnard of the Real Estate Counseling Group of Connecticut, Inc.
was retained to conduct a market research study of dll single-family residential property sales
within the three radium Superfund site areas. This analysis identified, reported, and measured the
actual market sales behavior of homebuyers and sellers using a total of 1,423 housing sales in
three different locations from July 1,1980 to June 30,1989. In one location, Dr. Kinnard found a
statistically significant decrease in property values and volume of housing sales after the public
announcement of contamination discovered at these sites. In the other two locations, the rates of
property value appreciation and housing sales volume increased more slowly than in locations
without Superfund sites. Evidence from Kennard's study also suggests that the housing market
response to a known Superfund site is a direct function of the speed, proximity, and apparent
effectiveness of any remediation or cleanup efforts.
1990-1992
In January 1990, the twice-extended public comment period ended and by June, the
public agreed] to a revised $250 million plan and thejsecond ROD was signed. This plan included
removal of the first 15 feet of contaminated soil from approximately 400 homes in the three
towns. Cleanup efforts would be spread out over a maximum of 10 years, and radiation-
ventilating devices would be installed in homes in the interim. As part of this plan, the EPA
i
would also replace existing radon units with higher efficiency radon units.
In 1991, as discussed above, after seven years of litigation, 237 residents received a $4.2
million settlement (average $18,000/house) from the remnants of U.S. Radium Corporation.
While the pilot study involving a limited nurhber of affected homes started in 1984, the
major cleanup activities of other areas began in 1991. The cleanup was divided into seven phases
based on the severity of contamination, owner access agreements, and location. Radon mitigation
systems were,maintained in almost 40 homes throughout each phase of the cleanup and until
-------
103
remediation was complete. Each phase of the remedial plan required access to properties to
perform a design survey, which included gamma radiation surveys, installation of radon and
alpha detectors, and soil sampling and drilling. If the survey found no evidence of contamination,
the property owner was given the results of the test, and a follow-up radon test was conducted
one year later. If the follow-up test confirmed the lack of radium contamination, the property
owner was given a final report summarizing all results. If contamination was present in the initial
or any of the follow-up surveys, the EPA remediated the property. After remediation was
complete, monitoring and sampling were done for one year to evaluate the success of the
cleanup. Once the property was cleared of all contamination, a detailed summary package was
provided to each homeowner, which included details of excavation and results of the testing.
Over the course of cleanup, approximately 100 families were temporarily relocated.
The cleanup process was highly disruptive to the neighborhood as it involved, in some
cases, the installation of building supports and the use of large machinery to remove
contaminated soil. The Pilot Phase and Phase I entailed the cleanup of 56 properties, temporary
relocation of 22 families, and removal of over 15,000 cubic yards of contaminated soil. After the
cleanup was complete, houses, property, driveways, and sidewalks were restored to at least their
original condition and in many cases were improved by enhanced landscaping and sidewalk
and/or garage replacement. Additionally, the amount of radon remaining in the soil at these
locations after remediation was well below the natural level of radon contained in most New
Jersey soils
1993-1995
In 1993, Phase HA was underway, which included the cleanup of 26 additional
properties. In 1994, Phase IIB started which called for the remediation of 53 properties. During
this time period, EPA had still not made a final decision regarding remediating the three
communities' streets. Phase III was completed in 1995 and consisted of remediation of 54
homes.
1996-1997
-------
104
Phase IV and Phase V were completed in 1996, and included the partial demolition and
rehabilitation of 55 homes. Phase VII of the cleanup began in the summer of 1997, which
included continuing remediation of properties with post excavation radon levels above normal
and beginning remediation of six additional homes. 'Remediation at 441 properties complete at
Montclair/West Orange, but 20 additional properties were discovered to need remediation.
After 1997
I
By 1998, a total of $175 million had been spent to remediate 300 houses and remove
80,000 cubic yards (or 5,000 large truck loads) of contaminated soil. In the fall of 1998, a two-
year remediation plan for the streets finalized and began in 1999, as part of Phase VI. Also, an
additional 30 homes were rehabilitated and 35,000 cubic yards of soil were removed as part of
I
Phase VI, which was completed in 2001. In 1999, the EPA began testing groundwater for
possible contamination, and a January 2003 Remedial Investigation Report revealed that elevated
levels of radon were found in the groundwater. Phase VIII was completed by the end of 2003.
Three additional properties require remediation and are included in Phase IX, currently
scheduled for completion in January 2005.
-------
105
Chapter 6
Eagle Mine, Colorado
6.1 Overview
Eagle Mine is centrally located between Vail and Beaver Creek ski areas, approximately
100 miles (160 kilometers) west of Denver, Colorado and about 14 miles (22.5 kilometers)
southeast of Vail, Colorado (Figure 6.1). Once one of the nation's top producers of zinc, the
mine lies between the small towns of Minturn and Red Cliff, just off U.S. Highway 24. The
property' consists of approximately 6,000 acres, 340 of which are contaminated with toxic waste.
Most of the contamination originates from areas located along the Eagle River, and includes: the
abandoned mining town of Oilman located on a cliff just above the mine, the old Eagle Mine
processing plant in Belden, two ponds containing wastes from the smelting of ore, Maloit Park,
Rex Flats, various waste rock and roaster piles, and an elevated pipeline. The Eagle River (a
major tributary of the Colorado River), Cross Creek, and several other tributaries run through the
site.
The Eagle Mine site is contaminated with eight to ten million tons of hazardous
substances including arsenic, nickel, chromium, zinc, manganese, cadmium, copper, and lead.
The main cause of Eagle River contamination came from acid mine drainage, which occurs when
sulfide minerals, such as pyrite, are exposed to oxygen and water and then oxidize. This process
creates sulfuric acid, which contaminated soil, groundwater, and surface water surrounding Eagle
Mine, producing water with low pH levels. Acid drainage at Eagle Mine resulted from
precipitation flowing through the waste piles that accumulated from nearly 100 years of mining.
As Eagle Mine acid drainage seeped into ground and surface water, it killed aquatic life and
vegetation growing along the water's edge and contaminated the river with zinc, lead,
manganese, and cadmium. Not only did this contamination threaten brown trout, the most
populous fish in this segment of the river, but it also permanently stained the rocks in and along
the river bright orange, providing Mintum and Red Cliff residents with a constant reminder of
the contamination at Eagle River.
-------
106
Figure 6.1 Eagle Mine Site
ll
-------
107
Eagle River across from CTP, stained rock.
State studies conducted in 1984 revealed dangerously high levels of cadmium, copper,
lead, and zinc in local water resources. Minturn, with a population of 1500, is the closest town
and draws drinking water from Cross Creek and two wells located within 2000 feet of the mine
tailings4 While Eagle Mine had a history of environmental problems dating back to 1957, the
majority of the problems arose after the mine closed in 1984. Eagle Mine was placed on the
national Superfund priorities list (NPL) in June 1986.
6.2 History and Timeline
Mining in Colorado began with the discovery of silver-lead and gold-silver in 1879, and
played an important role in the economic development of many Colorado mountains. In the mid
1890s, anew ore was discovered in Battle Mountain, and the bulk of mine extractions shifted
from gold and silver to zinc. Zinc extracted by independent miners was shipped off site until
1905, when a zinc ore processing plant was built near Belden, approximately a half-mile (0.8
kilometers) southeast of Oilman. The plant heated the ore to extreme temperatures and then
extracted the zinc using magnets, a process called roasting. After the roasting was complete, a
Most mine tailings are disposed of in an on-site compound, and are typically comprised of 40-70% effluent liquids
used in the mining process and 30-60% solids.
-------
108
tramway transported roasting wastes across the Eagle River and dumped them into three piles on
the west side of the canyon in direct sight of Gilmaii. In addition to these piles, two more waste
piles were located on the east side of Eagle River.
North face of Gilman and waste rock pile.
Around 1912, the Empire Zinc Company began buying small independent mines. By
1915, the company had consolidated the small independent operations into one business which it
named the Eagle Mine. Gilman became a company town and, at its peak, was home to over 400
residents.
The New Jersey Zinc Company constructed an underground flotation mill to process the
zinc ore from Eagle Mine. The ore was ground into & powder and then mixed with water and
treated with chemicals to bring the zinc to the surface of the mixture. The waste from this
process consisted of slurry, which was transported north via an underground pipe and dumped in
the Old Tailing Pile (OTP) and Rex Flats areas just ivest of the Eagle River. In 1928, the zinc
-------
109
operation ceased due to falling prices resulting from the Great Depression. Mining of gold and
silver ore continued, but the ores were shipped off-site for processing.
Ownership of the Eagle Mine Superfund site is quite complex. In 1938, Empire Zinc
Company merged with the New Jersey Zinc Company, making it the new owner of Eagle Mine.
Later in 1966, New Jersey Zinc Company merged with Gulf & Western Industries, Inc., which
later changed its name to Gulf+Western, Incorporated.
Because it was used to harden steel, a valuable commodity during the war, zinc
production resumed in the early 1940s. Once again the tailings were deposited at the Old Tailing
Pile (OTP). By 1946, the pile had reached capacity and a New Tailings Pile (NTP) was
established one-half mile (0.8 kilometers) northeast of the OTP, near the confluence of Cross
Creek and the Eagle River, just south of Minturn Middle School and Maloit Park.
Approximately, 75,000 tons of waste, covering 15-20 acres, were dumped at this site.
Long before Eagle Mine was listed on the NPL of Superfund sites, state officials
expressed concern about the amount of hazardous pollution originating at the site. In fact, a
March 17, 1957 Denver Post article reported that a zinc mine in Oilman, CO was asked to pay
$15,000 for 75,000 trout that suffocated in the Eagle River due to an oil fuel spillage. Another
Denver Post article (October 20,1974) reported that the New Jersey Zinc Company paid "$3,308
for trout and other fish which died July 19 after 12,000 gallons of liquid wastes and 100,000
pounds of mill tailings polluted the Eagle River".
Peter Seibert and Bob Parker founded Vail ski resort in 1962 less than 14 miles (22.5
kilometers) from Gilman and the Eagle Mine. The resort opened for business on January 10,
1963 and 12 skiers bought lift tickets. Within two years, more than 14,000 skiers visited the
resort. Vail's reputation blossomed in the mid 1970s when Gerald Ford became president. Ford
had lived part-time in the Vail area since the 1960s, and during his tenure as president, the resort
became known as the "Western White House". Starting in the 1970s, Vail experienced explosive
growth and building construction. Golf courses, tennis courts, and other sports activities attracted
summer tourists almost as plentiful as Vail's winter tourists.
-------
110
1976-1982
Zinc production continued until 1977 when, due to falling zinc prices, the operation was
shut down again and more than 150 mineworkers were laid off. Most feimilies moved out of
Oilman at this time.
By 1981, one million tourists visited Vail report annually.
1983-1988
On September 1,1983, Glenn Miller, a Colorado businessman, bought the Eagle Mine
property for $17.5 million with the intention of developing it into a ski resort. Unable to finance
this venture, Miller sold to Battle Mountain Corporation (BMC) 1,400 acres of the property
including Rex Flats, the tailings ponds, 70 homes, a|bowling alley, and several business offices.
In December 1983, the State of Colorado filed a lawsuit against Gulf+Western and New
Jersey Zinc for contaminating the Eagle River. At this time, the State also initiated a preliminary
risk assessment of the site. As the State and PRPs w|ere unable to come to an out-of-court
agreement, the complaint evolved into a court-ordered negotiation regarding who was
responsible for the $80 million cleanup of the 7.5 million tons of tailings at the site.
In 1984, while the State of Colorado's lawsuit awaited settlement, the new owner of the
site, Glenn Miller, lost his financial backing and lost the balance of the property due to
nonpayment of taxes. These lands were sold at Eagle County tax sales.
Copper-silver ore mining continued sporadically until 1984 when all mining operations at
the Eagle Mine halted. At that time the mine began to fill with water because of the many
fissures and cracks inside the mine. The Colorado Department of Public Health and Environment
(CDPHE) estimated that approximately 250 gallons of water per minute entered the mine. If the
mine flooded, the water would come in contact with transformers in the mine which contained an
estimated 3,000 pounds of the known-carcinogen poly chlorinated biphenyls (PCBs), which could
then potentially seep into local water supplies.
In June, the Public Service Company of Colorado informed the EPA that it was going to
turn off the electric power supplied to the Eagle Mine because Mr. Miller was unable to pay the
mounting electric bill, which was in excess of $90,0|00. This posed a problem because the mine
was being pumped to prevent it from flooding and coming into contact with PCBs. The EPA
-------
Ill
intervened and agreed to pay approximately $1,000 a day to keep the power on, until they could
perform an emergency removal of the PCB-laden transformers. Before the end of the month, the
EPA drained PCBs from all three of the transformers. The transformers remained in the mine,
however, to help prevent the mine from collapsing. The EPA also built dikes inside the mine to
divert water from entering the Eagle River.
In October 1984, the EPA added Eagle Mine to the list of proposed NPL sites. However,
action was delayed because Congress was unable to garner support for additional Superfund
funding. The State proceeded by developing a cleanup proposal for the site.
In March 1985, Ray Merry, the Eagle Mine Environmental Health Officer, ordered the 14
families remaining in Oilman to leave the site because of potential human health hazards. By
July, all families had left the area and Oilman became a ghost town. A gate prohibiting entrance
to the town read "Town for Sale." In December, the Colorado Department of Health conducted a
Remedial Investigation/Feasibility Study (RI/FS).
Warning sign at the entrance to Rex Flats & OTP.
The EPA placed Eagle Mine on the National Priority List on June 10, 1986. The EPA
then formally designated the State of Colorado to act as lead agency for the cleanup of the site,
but both agencies retained the right to take independent actions.
-------
112
In 1986, the State of Colorado filed $50 million lawsuit against Paramount (owner of
Gulf+ Western and now Viacom). The lawsuit was resolved in 1988 and the parties entered into a
Consent Decree and Remedial Action Plan (RAP). It was estimated that the cleanup would cost
$150 million; and take ten years to complete. The agreement drafted by Colorado State health
officials set acceptable zinc standards and pH levels for the river and relevant soils and required:
1. Plugging the mine portals to stop the production of acid mine drainage;
2. Removal of the roaster piles and reprocejssing of the tailings;
3. Collection and treatment of mine water and groundwater;
4. Revegetating the waste removal areas and the Consolidate Tailings Pile (CTP); and
5. Long-term monitoring of the site.
1989 -1994
As the cleanup began, public concern about the possibility of adverse human health
effects intensified. In March 1989, fearing that no assessment had been conducted, federal EPA
officials conducted their own investigation of potenjtial contamination at Mintum Middle School,
located 400 yards from the mine. A Vail Daily article (July 18,1989) reported on this
investigation and indicated that "Richard 'Dick' Parachini [CDPHE Project Manager] told the
governor that dust levels [from the cleanup] are 'right on the break point' of what is generally
considered environmentally safe. Breathing the heafy metals present at the Mintum site, in high
enough levels and over a period of years, is expected to greatly increase cancer risks." This
news generated alarm among local residents because the school was located less than one mile
(1.6 kilometers) from the 70-acre CTP, which was used for waste dumping until 1977. The EPA
convened a well-attended public meeting regarding [the issue. Approximately 1,000 irate citizens
attended this meeting, which EPA officials characterized as a "lynch mob". At this meeting, the
state agreed to amend its RAP to include the construction of a permanent waste water treatment
plant that would reduce the level of zinc in the river and raise the pH level of the river. This plant
would be operable by July 1,1990. Additional plugs were installed in the mine to stop water
from pouring directly out of the mine and into the river. Unfortunately, the volume of waste
-------
113
water needing treatment exceeded the capacity of the new plant, allowing contaminated water to
seep into the river. Viacom was forced to begin construction on a second water treatment plant.
In the interim, the local school board hired a private environmental consultant, Leonard
Slosky, to evaluate air quality and dust emanating from the roasting piles. Slosky installed 15 air
monitoring stations in and around the school and enlisted the help of several school teachers as
well. Every day for six weeks, select teachers wore small compact devices that resembled
portable headphone stereos. At the end of each day, a field technician analyzed the filters for
toxic residue. Slosky confirmed the presence of heavy metals in and around the school, but
concluded that the amount of metals present fell far below hazardous levels. For example,
arsenic found at the site presented the greatest potential health risk, but Slosky likened the
children's chance of developing arsenic-related cancer to that of smoking eight cigarettes in a
lifetime.
Although the EPA chose not to endorse the state's RAP because it was skeptical of the
plan's long-term effectiveness, the State forged ahead with the cleanup of the Eagle River site
fearing the worsening of public health and environmental damages that might result from
continued acid mine drainage. However, the State's decision to pump tailings pond water back
into the mine, using the mine as a holding tank, proved to be disastrous and caused even more
pollution to infiltrate the Eagle River. A dry winter caused mine seepage to make up most of
river water, and the river turned orange. As a result, fish populations declined dramatically.
Samples taken from the river that fall revealed zinc levels were 255 times higher than fish
tolerance thresholds. No fish lived in the river, and contamination was turning the Eagle River
various colors.
Media headlines reinforced the environmental and health related fears of local residents:
"Eagle 'cleanup' casts doubt on state" Denver Post April 20, 1990; "Eagle River fish population
smaller than expected" Denver Post April 24,1990; "Fish fading in river polluted by mine"
Rocky Mountain News April 27, 1990; "Mine cleanup again called inadequate" Denver Post July
13, 1990; "Polluted water flows to Minturn wetlands" Rocky Mountain News July 13, 1990; and
"More problems foul cleanup of zinc mine -- mine's toxic metals pollute Eagle River" Rocky
Mountain News August 28,1990. The cleanup was obviously failing, and local citizens felt state
and federal agencies were shirking their responsibilities. In an article published in the Rocky
-------
114
Mountain News on April 18,1990, a letter from the Minturn town council to Governor Roy
Romer stated, "Our river continues to run from murky green to sickly red... Above the mine the
river is crystal clear. The town of Minturn believes that we are being held hostage to a
bureaucratic nightmare. Minturn is a small town with very limited resources. We find ourselves
unable to get any action from our state officials who are supposed to be acting on our behalf."
In April, Paramount's new facility began to treat polluted water State and PRPs amended
RAP to add a^ chemical water treatment plant and install additional mine plugs. Colorado
Department of Wildlife began monitoring the fish and macroinvertebrates populations in the
Eagle River on an annual basis. At the start of this process, virtually no fish were found in the
Eagle River.
In September, a newly formed Eagle County Oversight Committee citizens group, joined
by Trout Unlimited, Eagle River White Water, Inc., and the Gore Creek Flyfisher, filed a $300
million class action suit against Paramount Communications, Inc. for water, air, and soil
contamination originating from the Eagle Mine property. The organizations claimed that Eagle
Mine seepage was contaminating the Eagle River as1 well as the Mintum Middle School. They
added that faulty water treatment plans were causing as much as 40 gallons of contaminated
water per minute to be dumped into Maloit Park weilands. From Paramount, the plaintiffs sought
damages for potential health risks, compensation of economic harm due to lower property
values, funds for removal of the tailings, and "exemplary damages for wanton and reckless
disregard" of residents' rights. Cindy Cacioppo, a member of the Committee, said "We decided a
lawsuit was needed for people to recover damages because there's nothing in Superfund that
allows citizens to recoup."
i
In 1991, the EPA became increasingly concerned about the site and notified the State of
Colorado that Paramount was in violation of six different aspects of the Clean Water Act because
of the mine seepage and discharge from the waste piles that had contaminated the river the
previous year. The water treatment plant was replaced and the EPA conducted a risk assessment
for PCBs.
Remediation efforts removed five piles of waste materials from the ore roasting plant
near Belden. This waste was relocated to the CTP, and former waste piles were revegetated.
-------
115
During this time period, Vail Associates started silently acquiring options to buy portions
of Eagle Mine property.
In June 1992, the EPA decided to conduct a Feasibility Study Addendum, which found
that the Eagle River ecosystem continued to suffer severely from heavy metal contamination.
Part of the study included additional risk assessments that resulted in a more comprehensive
investigation of the health offish and other natural resources. Additional soil studies and risk
assessments of ground water quality were conducted in Maloit Park, Minturn Middle School, and
the town of Oilman. The Feasibility Study Addendum identified the following activities to
supplement the State's RAP: collect additional seepage from the Rock Creek drainage; monitor
former roaster pile areas and waste rock piles; continue operation of the Water Treatment Plant;
collect additional groundwater from CTP and treat it at the Water Treatment Plant; and cleanup
Maloit Park wetlands area Risks to human health were also reviewed, but no appreciable threat
to drinking water was found, and heavy metals in soil and dust were found to be well within
acceptable standards.
Regardless of these findings, Viacom later moved Minturn's drinking wells upstream of
their old location. Finally, the EPA concluded that the potential risk of PCB ingestion from the
15 pounds of PCBs remaining in the transformers was low. The EPA agreed to continue to
monitor for potential PCB contamination, which has never been found. In June of 1992, the EPA
proposed a second and preferred remedial plan for the site, which included an alternative cleanup
for each of the individual areas contributing metals to the Eagle River. Cleanup activities focused
on removing the 150,000 tons of tailings deposited at Rex Flats and the one million tons of
tailings deposited at the Old Tailings Pile (OTP). All of these tailings were relocated to the
consolidated tailing pile (CTP). After the removal of the tailings, revegetation efforts were
undertaken. Restrictions were put into place to restrict the use of groundwater below the OLP.
Efforts were made to control seepage, surface water drainage, and groundwater flow in Rock
Creek Canyon. The seepage and drainage were collected for treatment.
In 1992, recognizing an opportunity for even more expansion, Vail Associates covertly
funded Turkey Creek, LLC's payment of three years of back taxes on over half of the Eagle
Mine site. Payment of these back taxes guaranteed Turkey Creek LLC and Vail Associates a first
bid option on the Eagle Mine property.
-------
116
In March, 1993, the EPA issued a Record off Decision (ROD) that required additional site
investigations and remedial actions to be implemented. The ROD included modifications of the
established remediation standard, proposed monitoring of additional metals, collection of
additional groundwater seepage, monitoring runoff, accelerating the capping of the CTP,
removal of the contaminated material from the Maloit Park wetlands, development of a
monitoring plan, and implementation of an inspection and maintenance plan.
The 1993 risk assessment determined that soils in Maloit Park Wetlands contained
elevated levels of arsenic, cadmium and lead. In June, the State amended the remedial plan for
the third time. Viacom was to permanently remove pond water from the top of the CTP,
implement a sludge dewatering system, and construct a sludge disposal cell.
In the summer of 1994, the EPA issued a unilateral administrative order that consisted of
additional monitoring and testing of the site, which was amended again in 1995 to add a work
plan for Maloit Park waste removal and restoration.
1995-1999
By 1995, more than 2.1 million tourists visit|the Vail Valley annually.
In August, 1995, the State of Colorado, US EPA, and Viacom agree to a Three-Party
Consent Decree and Statement of Work to implement the 1993 Record of Decision. The
agreement caUed for sampling of water quality, along with assessments of the aquatic insect and
fish populations in the Eagle River to determine the effectiveness of the cleanup actions.
Additionally,'the three parties agreed to investigate the adaptation of biological-based cleanup
standard for the site.
In addition to the actions taken in 1992, more efforts were made to control seepage,
surface water drainage and groundwater flow in Rodk Creek Canyon.
At Rex Flats and OLP, more than 800 cubic yards of zinc concentrates were removed and
moved to the consolidate tailing pile. To prevent grqund water contamination, remediation
efforts involve intercepting and diverting 100-200 gallons of clean water per minute from the
Eagle Mine.
In 1996, the contaminated soils in Maloit Park Wetlands were removed. Clean soil was
used to cover the previously contaminated area and revegetation efforts were undertaken.
-------
117
On February 21,1986, 32 cars of an 82-car train derailed on the Tennessee Pass heading
towards Mintum. Two crew members on the train were killed and another was injured.
According to the EPA, four tank cars ruptured, spilling approximately 54,000 gallons of sulfuric
acid. Five to six acres of nearby trees and vegetation were blackened by the sulfuric acid that
went over an embankment and across the two lane highway. Other contaminates spilled
including Triethylene glycol (antifreeze) and small amounts of diesel fuel. However, the overall
environmental damage appeared to be less than originally feared. Tests of the Eagle River
revealed no significant levels pollution in the river or other water sources.
By 1997, the main tailing pile was capped. Residents reported increasingly better water
quality in the Eagle River. In August, the EPA awarded "Environmental Achievement Awards"
to both the Eagle River Environmental and Business Alliance and Viacom. The Eagle River
Environmental Business Alliance was acknowledge for their efforts in the successful cleanup of
the site by "keeping area residents informed, providing technical input, and discussing with
people their concerns about a hazardous waste cleanup in their neighborhood." Viacom was
lauded for their cleanup efforts that had "gone beyond legal requirements, furnishing the town of
Mintum with a safe water supply, voluntarily cleaning up large amounts of hazardous materials,
planning to intercept clean water flowing onto its site and keeping a skeptical public informed
about the cleanup."
In 1998, the EPA issued a final ROD ensuring it would provide ongoing monitoring of
the site. In 1999, state and federal authorities formally sought to change the cleanup agreement to
include pumping of groundwater to keep it from filling the mine and complicating treatment of
contaminated water from the mine.
Although they denied their intentions for years, Vail Associates revealed, under oath in
March, its intention to buy and develop the Oilman property. By April, Vail was already 93%
built out, so it turned its eyes on neighboring communities. While the remediation and
bankruptcy proceedings for the Oilman property progressed, Vail filed suit against Mintum for
its under-utilized water rights, which it would use for its controversial back bowl ski trail
expansion. Much to the dismay of Mintum residents, Vail won the rights to 4,76 cubic feet per
second of running water during the driest months from October to April. The value of these
water rights is estimated to be $14 to $16 million. The enormity of the Vail resort, as well as the
-------
118
i
wielding of its political power, caused a tremendous amount of animosity between Vail
Associates and the small nearby communities. In June, six environmental groups sued the US
Forest Service over the expansion of Vail resort and its threat to wildlife.
With its option to purchase portions of the Eagle Mine property, as well as an easement
from the Forest Service and entitlement to some of Minium's water supply, Vail Associates
began construction of its controversial 885-acre back bowl expansion in July 1999. The
expansion opened in the 2000-2001 ski season and brought skiers within a mile (1.6 kilometers)
of the Oilman property. Local residents feared that surroundings communities were destined for
yet another Vail-controlled real estate expansion similar to that of Bachelor Gulch where ski-in,
ski-out homes were sold for $750,000 each. According to local residents, such a surge in housing
and rental rates would threaten the stability of the blue-collar families and communities already
struggling to afford Vail Valley's ever-increasing cost of living.
However, in November, Vail Associates announced its plan to protect some of Eagle
County's last remaining open space, the Eagle Mine property. Some in the Colorado
environmental community viewed the announcement as public relations effort. According to Ted
Zukoski of the Land and Water Fund, "We're going to be watching this very closely to make
sure this isn't just a green-washing effort to make people feel warn and fuzzy about a big
development. It's potentially a step in the right direction, but we're going to have to wait and see
how far it goes."
2000 and Beyond
According to the first 5-Year Review completed in October, 2002, the cleanup of the
Eagle Mine sfte, both the federal and the state portiqns, were essentially complete. The review
concluded that public health risks had been removed and restoration of the Eagle River had
progressed significantly. Eight million tons of wastej rock, tailings, and roaster debris were
moved to the Consolidated Tailings Pile, and the CTP was capped and revegetated. The tailings
from Rex Flats and the Old Tailings Pile adjacent tolRex Flats had been removed, and the area
was revegetated. The roaster piles that were directly across the canyon from Oilman had been
moved to the CTP. Maloit Park wetland had been cleaned and a barbwire fence was constructed
around it.
-------
119
Rock Creek no longer flowed directly into the Eagle River; its water continues to be
treated at the water treatment plant before being released. Also, approximately 250 gallons per
minute (gpm) of water from the mine is being treated before being released into the river. In
total, the water treatment plant treats about 360,000 gpm every day. In October, 2001, The
Denver Post reported that as a result of the 14-year effort and a cost of $70 million that the Eagle
River once again ran clean enough for a healthy fish population. Groundwater remediation
efforts would continue in order to cleanup an estimated 700 million gallons of contaminated
groundwater in the 70 miles (113 kilometers) of tunnels within the mine. The annual cost will be
approximately $750,000.
The only indication of past contamination is permanent oxidized manganese and iron
stains, which give the rocks along the river's edge a rusty brown "bathtub ring". The results of
the annual biological assessment of the Eagle Mine site show dramatic improvements in the
Eagle River aquatic community such as higher numbers of fish and macroinvertebrates.
-------
120
Chapter 7
Expert Error and the Psychology of Risk and Stigma
7.1 Expert Error
Gayer, Hamilton and Viscusi (2000) argue that residents living near Superfund sites
judge risks to be of a magnitude consistent with EPA expert opinions and that these judgments
are reflected in property values. The research presented here suggests quite the opposite.
However, the sites studied here are much larger and, likely to attract more attention. This section
documents many cases of expert error to help explain why expert opinion plays a limited role in
explaining residents' risk beliefs. Thus, the judgments of experts are only one component of the
mix of news media stories and perceptual cues received by the typical citizen. Even if statements
by scientific experts were accepted as credible, they would compete with a mix of the other
signals and perceptual cues. As simply one component, such statements are unlikely to be the
primary determinant of individual risk beliefs. Thus, risk beliefs determined largely by media
stories and other perceptual cues are unlikely to be easily changed by the pronouncements of a
few scientists (Fischhoff, 1989).
Furthermore, it is unlikely that statements by scientific experts will be accepted as
completely credible. Even when different experts are in essential agreement, the news media
often focuses on those aspects where experts disagree (Wilkins and Patterson, 1990), thus
lowering the perceived credibility of experts. In a stpdy examining news coverage of Three Mile
Island and Chernobyl, Rubin (1987) found that news stories tended to dichotomize events rather
than blend a continuum of information to recipients.! The result is that the public discredits
information it receives from experts because it appears that experts cannot agree among
themselves and, therefore, do not really know the ri$k that a site presents.
Despite the ideal that science discovers absolute truths, for even.' health or environment
related article there appears to be a corresponding article that rejects the tenets of the previously
publicized claim. Numerous famous examples exist, which are described in detail below, where
experts from academia, government, and industry have made errors and misestimates:
• Soil contamination at Love Canal, Niagara, New York
-------
121
Dioxin contamination in Times Beach, Missouri
The defective Dalkon Shield for birth control
The false discovery of Cold Fusion
The failures at Biosphere 2
The near nuclear meltdown at Three Mile Island
The Union Carbide Accident in Bhopal, India
These examples are not just relegated to the past, as the costly search for weapons of
mass destruction in Iraq, to date, has yet to support early claims by intelligence experts. Each of
the short descriptions below serves to illustrate the characteristics and media attention that such
failures attract.
7.1.1 Love Canal, Niagara, New York
Love Canal is permanently etched in the collective conscious of America, and these
words were synonymous with hazardous waste contamination, cancer, and distrust of authorities.
Love Canal brought about anew understanding of the potential health effects of hazardous waste
as well as Superfund legislation designed to deal with chemical disposal sites.
Located in Niagara Falls, New York, Love Canal is named for William Love who began
digging a canal in 1896 for a proposed hydroelectric power plant. Love abandoned the project
when he declared bankruptcy, and in 1920 the city of Niagara Falls purchased the site for use as
a landfill. In 1942 Hooker Chemicals and Plastics Corporation (now Occidental Chemical
Corporation) purchased the landfill for their own disposal purposes. From 1942 to 1953, Hooker
dumped into Love Canal 21,800 tons of toxic waste including more than 400 different chemicals,
11 known carcinogens, PCBs, dioxins, pesticides such as DDT and lindane (both of which have
been banned in the United States), heavy metals, and multiple solvents. Three years after
Hooker's dumping began, an internal memo from an engineer foreshadowed the disaster to
come, "[Love Canal is a] quagmire which will be a potential source of lawsuits."
Once the site reached capacity, Hooker covered the 16-acre toxic waste site with a 40-
acre clay seal to prevent chemical seepage. (A 1981 EPA report confirmed that Hooker's waste
disposal techniques required only minor adjustments to come into compliance with the hazardous
-------
122
waste disposal standards in place at the time.) Undter threats of acquisition by eminent domain
and extreme pressure from the city school board, Hooker reluctantly sold the site to the New
York State Department of Education for $1.00, on the condition that Hooker would be
indemnified from any future liability concerning th^ site. Because of the potential dangers
associated with the site, Hooker insisted that deed restrictions accompany the property transfer
and repeatedly warned the school board of potential health hazards at the site. Hooker also
stressed that under no circumstances should the land be excavated or trie clay cap be jeopardized.
Despite these warnings, the city constructed a school on site and sold the remaining parcels of
land to real estate developers. The community of Lpve Canal was bom
Love Canal residents first began complaining of chemical odors and residues in the
1960s. By 1976 chemical seepage had infiltrated neighborhood creeks, sewer lines, sump pumps,
and soil - even the air inside several Love Canal homes. That year, the New York Department of
Environmental Conservation initiated the first environmental testing of Love Canal which found
contaminated groundwater, soil, and air. Once the results of that research were released, local
and national Jnedia responded quickly: "Vapors fro^n Love Canal Pose Serious Threats" (Courier
Express Niagara, May 15,1978), "Toxic Exposure at Love Canal Called Chronic" (Courier
Express Niagara, May. 25,1978), "Wider Range of Illnesses Expected'' (Courier Express
Niagara, August 4, 1978), "Upstate Waste Site May Endanger Lives" (New York Times, August
2,1978), and "The Devil's Brew in Love Canal" (Fortune, November 19, 1979). Heightened
alarm among community members and media attention prompted the New York State
Department of Health to test Love Canal homes close to the disposal site for environmental
contamination. Two years later, the State Department of Health declared a state of emergency,
ordered the school to be closed, and recommended ^n evacuation of the 239 homes that tested
positive for environmental contamination. This news spread rampantly throughout the
community causing a widespread panic and loss of property values of homes adjacent to and
outside the immediate canal area. Fearing for their health, the lives of their children, and their
futures, the remaining 660 families pressured both New York State Governor Hugh Carey and
President Jimmy Carter to expand the evacuation area.
One year later, in February 1979, Dr. Beverly Paigen, a biologist with Roswell Park
Memorial Institute in Buffalo, conducted a study which revealed that between 1974 and 1978:
-------
123
56% of the children born at Love Canal had birth defects; miscarriages had increased 300%;
urinary tract disorders had increased 300%; and the frequency of asthma, epilepsy, suicide, and
hyperactivty had increased. Dr. Paigen also claimed to have evidence that these conditions
subsided once residents moved away from Love Canal. These findings fueled the Love Canal
panic, even though Dr. Paigen's research was not a scientific controlled study but, instead, based
on anecdotal evidence from personal interviews with Love Canal residents. Dr. Paigen's research
was thoroughly discredited at that time by the NY Department of Health. A governor's panel
charged with reviewing her work found that, "[Dr. Paigen's research] falls short of the mark as
an exercise of epidemiology. She [Dr. Paigen] believes fervently that her observations prove the
existence of multiple disease states directly attributable to chemical pollution, but her data cannot
be taken as scientific evidence for her conclusions. The study is based largely on anecdotal
information provided by questionnaires submitted to a narrowly selected group of residents.
There are no adequate control groups, the illnesses cited as caused by chemical pollution were
not medically validated.... This panel finds the Paigen report literally impossible to interpret. It
cannot be taken seriously as a piece of sound epidemiological research...."
However, two studies conducted in 1980 by the EPA initially seemed to confirm portions
of Dr. Paigen's research and found chromosomal irregularities and nerve damage among Love
Canal residents. Upon release of these findings, chaos broke loose at Love Canal and two EPA
officials were involuntarily detained. That evening, Lois Gibbs, a member of the Love Canal
Homeowners Association, phoned the White House to inform them of their hostages. Pressured
by the unfavorable findings of the research and extreme political pressure from local residents,
the President issued orders on May 20,1980, to permanently relocate all families that wished to
leave. In total, approximately 950 families (2,500 residents) evacuated the area, leaving the
government with $3-5 million in relocation costs. These relocations eventually became
permanent costing the government over $30 million.
The integrity of the two 1980 EPA reports as well as the validity of their findings have
since been questioned by the Center for Disease Control (Morbidity and Mortality Weekly, May
1983), American Medical Association (March 1984), National Research Council, and New York
State Department of Health on the basis of the lack of control group (adjusted for by comparing
Love Canal results with a control group from a previous unpublished experiment), incorrect
-------
124
statistical analysis, small sample sizes, inadequate experimental methodology, report release
prior to peer review, and drawing conclusions that in some cases were not supported by the
evidence. For example, the chromosome study actually found a lower rate of chromosomal
damage among Love Canal residents than the control group. According to a 1981 New York
Times article, "it may well turn out that the public hjas suffered less from the chemicals [at Love
Canal] than from the hysteria generated by flimsy research irresponsibly handled."
Research conducted in 1982 by EPA found no unusually high levels of contamination
outside the area immediately surrounding the canal, confirming the results of several previously
conducted reports. This EPA report was considered to be highly controversial and was eventually
dismissed. (However, another EPA study conducted in 1987 confirmed the results of the 1982
study.)
Despite the lack of evidence to support such an action, Love Canal was declared a
national Superfund priority on September 1,1983. federal and state remediation activities were
expensive and highly intrusive. Fences were erected around the site and bulldozers demolished
the abandoned homes and school within. Leachate treatment plants, high temperature
incineration, excavation and off-site disposal of contaminants and hydraulic cleaning of sewers
and culverts removed wastes and hazardous toxins from the site. Although twenty thousand tons
of waste currently remains at the site, the area was declared "habitable" in September 1988.
Initial redevelopment of the site was difficult because local banks were hesitant to grant home
mortgages for fear of being held liable for environmental contamination. Although the value of
the homes was approximately 20% lower than comparable markets. 239 of the 240 homes in the
Love Canal neighborhood, now called Black Creek [Village, have been successfully rehabilitated
and sold. Approximately 30% of the purchasers are original Love Canal residents.
In 1991, the Committee on Environmental Epidemiology of the National Research
Council thoroughly reviewed all Love Canal research and reports and concluded that there was
no definitive link between the health conditions of Love Canal residents and the chemical
seepage from the canal, with the possible exception of decreased birth weights and heights. Legal
settlements are starting to be resolved as well. In 1998,2,300 Love Canal families received
between payments ranging from $83 to $400,000 frm Occidental Chemical Corporation.
-------
125
Two quotes will serve to summarize public reaction to the site that eventually led to the
Superfund program:
"Love Canal doesn 't end with this generation's cancer or even with the next
generation's birth defects. For many residents, the damage is permanent in their
genes and their children's. The mutated genes will affect all of their descendents,
one generation after another, "
Lois Gibbs, (executive director of the Center for Health, Environment and Justice
and former Love Canal resident) Who's Poisoning America, p. 270.
"It is not enough for industry and government to act in good faith - their mistakes
are counted in human lives."
Glamour, November 1980
7.1.2 Times Beach, Missouri
All that remains of Times Beach, Missouri, a small community once located 17 miles
west of St. Louis, is a legacy of an environmental disaster. In 1972 and 1973, the city of Times
Beach hired Russell Bliss to manage air-borne dust from its unpaved roads. During that time,
Northeastern Pharmaceutical and Chemical Corporation also hired Bliss to dispose of their
wastes, including dioxin yielded from the production of a then popular skin cleanser called
hexachlorophene. In an attempt to complete both tasks efficiently, Bliss mixed Northeastern
Pharmaceutical and Chemical Corporation's wastes with oil and sprayed the mixture on Times
Beach roads. Days later animals started dying, and months later children got sick. After Bliss
sprayed Shenandoah Stables' roads, several horses died and the proprietor's daughter became
very ill. In November 1982, the EPA found Bliss' oil mixture to be contaminated with dioxin, a
known human carcinogen. Dioxin is an unintentional hazardous byproduct of many common
industrial processes such as the bleaching of paper and wood pulp; production of herbicides and
wood preservatives; and incomplete combustion of wood and industrial and municipal wastes.
One month later, the nearby Meramee River flooded. As the water receded, experts predicted
-------
126
that it would redistribute dioxin throughout the city. Consequently, not only would Times Beach
roads be contaminated, but the entire Times Beach community might aiso be laden with dioxin
contamination.
During this time, several studies emerged supporting the highly carcinogenic nature and
potential dangers associated with dioxin. Based on this research, in 1982, the Centers for Disease
Control and other experts recommended completely evacuating Times Beach. On March 4, 1983,
Times Beach was proposed for Superfund's NPL. The town was officially closed in April 1985
and six months later, on September 8, 1983, Times peach was placed on the final NPL. By the
end of 1986, the federal government had spent $33 million to permanently relocate all 2,240
Times Beach residents. The title to the town was conveyed to the State of Missouri, and any
remaining parcels of land were purchased by the Federal Emergency Management Agency
(FEMA). As part of the remediation of the site, almost all buildings in the city were demolished
and the entire area was enclosed by a chain-link fence.
Doubtful of the severity of the adverse health impacts associated with dioxin (such as
cancer and infertility), as well as the proposed pathway's of human exposure, many scientists and
experts characterized the Times Beach relocation asl an over reaction. An article in the Wall
Street Journal written the week of the evacuation supported this sentiment, "There are two
dangers with toxic wastes. One is the very real threat to health posed by the chemicals
themselves. The second is that a hysterical exaggeration of that threat will needlessly frighten
i
people and drive them from their homes." Considering the best available research at that time,
the Times Beach evacuation was indeed an over-reaction because it was based primarily on the
analysis of soil samples, rather than the potential human health risks and exposure pathways of
dioxin.
In 1991, several scientific experts, including Dr. Vemon Houk of the Centers for Disease
Control, reversed their initial conclusions about the toxicity of dioxin and their recommendations
to evacuate Times Beach. This reversal was based on new research, which wholly contradicted
previous conclusions about dioxin. The new research found dioxin to be less harmful to human
health than originally suspected, making the Times (Jeach evacuation seem that overly drastic
and unwarranted. As Houk stated, "Times Beach was an over-reaction. It was based on the best
available scientific information we had at the time, ft turns out that we were in error.... The only
-------
127
thing I would have done differently, I would have said we may be wrong. If we're going to be
wrong, we'll be in the wrong side of protecting human health. I don't think we ever said we may
be wrong."
Upon learning of the new research, industry representatives complained vociferously
about the over-regulation of dioxin and the exorbitant costs associated with its stringent
regulation. Prompted by these complaints and under the direction of William Reilly, the EPA
Administrator under President Bush, Sr., the EPA undertook an extensive series of highly
technical experiments on the toxicity of dioxin. Three years later, to the surprise and dismay of
industry representatives, these experiments reaffirmed the link between dioxin and cancer even
at very low levels of exposure. These experiments also revealed that dioxin bioaccumulates in
living tissue and can cause stunted fetal development, suppression of the immune system,
interference with regulatory hormones, and increased likelihood of developing endometriosis and
diabetes. However, like the second round of dioxin research, this research also revealed that the
major pathway of dioxin exposure is not environmental but through the ingestion of dairy foods
which contain small amounts of the compound.
Armed with this new knowledge, the EPA devised a plan for the remediation of Times
Beach contamination that included the construction of an on-site thermal destruction plant.
Incineration of the dioxin-contaminated soil began in March 1996. Community action groups,
such as the Times Beach Action Group and Dioxin Incinerator Response Group, strongly
opposed this incinerator fearing that burning the dioxin might spread contamination rather than
reduce it. Research studies conducted on a Jacksonville, Arkansas incinerator — similar to the
one constructed at Times Beach — confirmed these fears: blood levels of dioxin among residents
living near the Arkansas incinerator were 22 times higher than before incineration began. The
Arkansas studies further concluded that these elevated levels of dioxin caused increased
incidences of diabetes among residents living near the incineration plant. In 1993, Missouri state
officials confirmed that the Times Beach incinerator was also producing more dioxin than it was
destroying, and the EPA disbanded the plant in 1997. Remediation of the site cost $200 million
and was completed in 1997 with the closing of the incineration plant. The property, now a 40-
acre state park was named Route 66.
-------
128
Although more than two decades have passed since the Times Beach evacuation, many
former residents still speculate about the true effects of dioxin on their families and friends. One
Times Beach resident recalls several now-dead community members that suffered from cancer,
immune deficiencies, miscarriages, and suicides. Sre laments, "I'm so tired of death. There's not
a day [that] goes by that I don't wonder if all this is coincidence - or dioxin."
7.1.3 The Defective Dalkon Shield
In the 1960s, many women became concerned about the possible adverse health effects,
such as cancer and strokes, associated with oral contraceptives and began seeking alternatives. In
response, several pharmaceutical companies invested heavily in the development of intrauterine
devices (lUDs) as a potentially substitute for birth control pills. A.H. Robins (Robins) decided to
enter the IUD market in 1970 by acquiring Dalkon Corporation, a manufacturer of the Dalkon
Shield. Robins had no experience in the development of contraceptive devices and was best
known for its non-prescriptive remedies such as flea and tick collars and cough medicine.
Because Robins had neither obstetricians nor gynecologists on its staff (nor an appropriate
department at that time), it assigned the production £nd assembly of its IUD, the Dalkon Shield
(Shield), to its Chap Stick division.
Other than one research study conducted by Dr. Hugh Davis, a co-inventor of the Shield,
Robins had conducted no testing of the Shield in women or animals when it entered the market
in 1971. Yet A.H. Robins positioned the Dalkon Shjeld as the "Cadillac of contraceptives" and
"the truly superior modern contraceptive". Dr. Davis' research boasted, among other things, that
the Shield was five times safer than other lUDs. After its release in January 1971, the popularity
of the Shield blossomed as a total of 4.5 million Dalkon Shields were sold to women worldwide
by 1975. However, as early as the summer of 1972, JRobins began receiving reports of the
Shield's ineffectiveness in preventing pregnancy as well an increased incidence of pelvic
infection among its users. In 1974, Robins was ordered to stop producing the Dalkon Shield and
required to recall the product. By the time most Dalkon Shields had been removed, reports
documented a total of 15 premature Shield-related djeaths and an estimated 90,000 Shield-related
injuries including sterility, pelvic inflammatory disease (PID), septic abortions, hemorrhaging,
perforated uteri, and birth defects in children.
-------
129
Even though the shortcomings of the Dalkon Shield were well known to A.H. Robins, the
company continued to produce and distribute its product. Evidence contrary to the purported
comfort, effectiveness, and safety of the Shield was suppressed. For example, research conducted
by Dr. Davis claimed the Shield's failure rate in preventing pregnancy was 1.1%. Davis did not
reveal, however, that participants in his study used a backup method of birth control for three
months after the Shield was inserted. An independent researcher later found the Shield's failure
rate to be 3-5%, making it inferior to other lUDs and the pill in preventing pregnancy. Robins
was also forewarned in several memos and conversations of the Shield's tendency to cause
infection. R.W. Nickless, management coordinator for pharmaceutical products, wrote a memo
to 39 officials at Robins on June 29, 1970, detailing the Shield's propensity for wicking and
infection. A July 28,1971 memo written by Wayne Crowder, a quality control supervisor in
Robins' Chap Stick Department, reinforced these findings as well as the need to address the issue
immediately. Crowder's memo was ignored and his position later eliminated. Evidence presented
by Crowder was confirmed several months later by Irwin Lerner, the inventor of the Dalkon
Shield, in an October 11,1971, conversation with Kenneth Moore, Shield project coordinator. In
1972, Dr. Thad Earl, an investor in Dalkon Corporation prior to its acquisition by Robins, also
sent Robins a memo in 1972 warning that women who became pregnant while using the Shield
needed to have it removed immediately to prevent infection. However, Robins warned neither
physicians nor Shield users of the dangers associated with the Shield for another three years,
despite being alerted to the potentially dire consequences of its use.
The Centers for Disease Control (CDC) substantiated the results and warnings contained
in previous memos, correspondence, and conversations. Its study conducted from 1976 to 1978
found that, depending on the length of use, the risk of developing PID among Shield users was
five to ten times higher than for non-Shield IUD users. Two studies conducted in 1985 in Boston
and Seattle yielded similar results.
Researchers found the defective component of the Shield to be the unique design of the
string, or "tail", which facilitates the removal of lUDs. Previously, a single piece of nylon
formed the tails of lUDs. But the tail of the Shield was comprised of many strands of nylon
encased in a nylon sheath to prevent the spread of bacteria But the nylon sheath was left open at
both ends of the Shield, allowing bacteria to spread up, or "wick" into, the tail into the uterus.
-------
130
Although Robins had been warned of this wi eking tendency numerous times, it was reluctant to
withdraw the Shield from the market because it generated profit margins of 40% in the United
States and 70% internationally. j
In 1974, Dr. Howard Tatum, an independenj researcher, testified before Congress about
the relationship between Dalkon Shields and PID. Biased on this testimony, FDA officials
pressured Robins into halting production of the product. Robins stopped distributing the Shield
in the United States in June 1974, but continued to market the Shield abroad for another 10
months. Finally, in April 1975, Robins stopped international distribution of the Shield. By the
time distribution of the Shield ceased, it had been used by 2.8 million American women and
another 1.7 million women worldwide. Fearing legal repercussions, however, Robins refused to
recall Shield? already in use for several more years. In 1983, when FDA officials suspected that
most Shields had already been removed, Robins publicly recalled the Shield and offered to pay
for the removal of any still in use. Over 4,000 women accepted this offer within the first two
months of the announcement.
By 1984, more than 10,000 claims for Shield-related injuries had been filed against
Robins. According to U.S. District Court Judge Miles Lord, who presided over 21 Shield court
cases, "The only conceivable reasons you [Robins] have not recalled this product are that it
would hurt your balance sheet and alert women wh have already been harmed that you may be
liable for their injuries. You [E. Claibome Robins Jr. (President and CEO), William Forest
(General Counsel), and Dr. Carl Lunsford (Director: of Research)] have taken the bottom line as
your guiding beacon and the low road as your route." Robins' former general attorney, Roger
Tuttle, echoed this sentiment, "Robins entered a therapeutic area with no prior experience, no
trained personnel, and reliance on statistics from an admittedly biased source. Although the
device was based on sound scientific principles, Robins over-promoted it without sufficient
clinical testing in an effort to ride the crest of a marketing wave for financial gain."
Largely as a result of litigation over the Dalkon Shield, A.H. Robins filed for Chapter 11
bankruptcy in 1985 and established a multi-million dollar trust for unresolved complaints three
years later. Prior to the court trial regarding Robins' negligence, 12 boxes of correspondence
attributing PID in Dalkon Shield users to the wicking effects of the Shield's tail disappeared, and
Aetna Life Insurance canceled its contract with Robins.
-------
131
In 1989, American Home Products purchased A.H. Robins. As of July 1992,115,000 of
the 137,000 claims that have been finalized have received a settlement of $1,000 or less from the
trust. Since the problems with Dalkon Shield, product labeling and package inserts for all lUDs
include extensive and comprehensive information regarding user profiles and potential side-
effects.
7.1.4 The Discovery of Cold Fusion
Imagine a world of equitable nations where resource-poor countries no longer struggle to
survive on dwindling natural resources; where industrialized countries are not tethered to the vast
oil riches of the Middle East; where energy consumption does not necessarily mean
environmental degradation. This is the world promised by cold fusion, the remarkable discovery
of two highly respected University of Utah chemists.
Prior to this discovery, scientists, physicists, and chemists around the world had deemed
cold fusion impossible because the fusion of two atoms required extreme heat temperatures and
expensive heavy metals such as uranium. In March 1989, Stanley Pons and Martin Fleischmann
held a press conference claiming to have had detected bursts of excess heat and the appearance
of neutrons that exceeded background levels in their cold fusion experiments. This
unprecedented discovery astounded the world and promised the world great amounts of energy
generated simply and inexpensively. Cold fusion meant an abundance of energy and the end of
all potential future energy crises.
The day of their announcement, Pons and Fleischmann intimated that their results could
be easily replicated and scaled up for a nuclear reactor, despite the fact that their experiments
substantiated neither claim. A worldwide flurry of media articles, accusations, press conferences,
confirmations, and objections ensued as governmental agencies, scientists, and industrial
representatives raced to replicate the results of Pons's and Fleischmann' experiments.
Many experts remained skeptical of cold fusion, because of the inability to replicate their
results. The few experts who claimed to successfully repeat Pons's and Fleischmann's
experiments later retracted their results. As was later determined, neuron detectors, used by Pons
and Fleischmann, are very inaccurate and often detect "neutrons" that are actually small
variances in temperature, humidity, or electric power surges. In the end, Pons's and
-------
132
Fleischmann's results were rejected because they ntyl only lied about the amount of energy
produced in their experiments, but they also failed to conduct control experiments to establish
accurate baseline data.
Despite the credibility of the scientists who announced cold fusion, the integrity of the
institutions that supported them, and the scientific fervor that surrounded the incidence, cold
fusion's skeptics prevailed, and Pons's and Fleischmann's experiments are now largely
recognized as a scientific hoax. According to one CalTech researcher, "[Cold fusion] has been
cast out by the scientific establishment. Between cold fusion and respectable science there is
[now] virtually no communication at all."
7.1.5 The Failure of Biosphere 2
On September 26,1991, eight "biospherians" entered the world's first self-contained,
human-constructed, completely independent ecosystem. A massive media blitz highlighted the
lofty goals of this experiment, to test the ability of numans to construct and survive
independently in a self-contained environment that would provide everything necessary to
sustain life. Hie construction of Biosphere 2 took six years and cost $200 million. It contained
five distinct ecosystems (rain forest, ocean and coral reef, fog desert, marsh, and grasslands) and
3,800 different species of animals. Scientists and engineers designed the structure to replicate as
closely as possible the metabolic and biologic functions of the first biosphere, earth. Once they
entered, the biospherians would remain inside the three-acre Biosphere 2 for two years, growing
and harvesting their own food, managing their own wastewater systems, monitoring ecological
systems, etc. According to the project goals, any contact whatsoever with the outside world, be
it importing food or allowing air to escape into the Earth's atmosphere, would completely
destroy the experiment. Only in the event of a medical emergency would the biospherians be
allowed to leave Biosphere 2.
Discover heralded Biosphere 2 as "the most exciting scientific project to be undertaken in
the United States since President Kennedy launched us toward the moon." The New York Times
and the Boston Globe followed Discover'* enthusiastic lead. As they reported, hummingbirds
pollinated the flowers within; a colony of termites aided the decomposition of vegetation; and
bugs such as ladybugs, lacewings, and spiders minimized insect damage to Biosphere 2 crops.
-------
133
No outside energy entered the structure. Energy was, instead, provided by an internal solar
power plant. Sensors placed throughout the structure provided 24-hour monitoring of the balance
within Biosphere 2, including the mixture of gasses in the air, emotional and mental stress of the
biospherians, and aerobic rate of microbes in the soil. Lastly, Biosphere 2 was completely sealed
off to prevent any atmospheric exchange between the earth and Biosphere 2.
The purposes of the experiment, as portrayed by the media, varied from developing the
new science of "biospherics" to "developing] the technology necessary to colonize other planets
with biosphere structures" (New Republic). The public later learned that indeed Space Biosphere
Ventures (SBV), the owner of Biosphere 2, intended to develop and sell this type of technology
to NASA and the European Space Agency. SBV also had other intentions, to develop an
extensive 2,500-acre theme park adjacent to Biosphere 2. As the knowledge of these plans
became more common, it strained the credibility of the program among the public as well as
many scientists nationwide.
Within weeks of the biospherians' entrance, problems arose inside Biosphere 2. One
biospherian, Jane Poynter, cut off the tip of her finger while using the thresher. She left
Biosphere 2 to seek medical attention and returned two days later with a duffel bag purportedly
containing fresh food and new sealant to patch Biosphere 2's air leaks. SBV denied these rumors
for three months until they admitted that Poynter had indeed returned with a duffel bag
containing items such as plastic bags, film, and computer parts. Two months later, Marc Cooper
of the Village Voice confirmed and reported that SBV had installed a carbon dioxide scrubber in
Biosphere 2 just before its closure. Later that month, SBV secretly injected Biopshere 2 with
600,000 cubic feet of outside air to relieve its falling atmospheric pressure. Upon learning of this
injection, Biospherians Linda Leigh and Roy Walford threatened to leave Biopshere 2 unless a
public announcement of the injection was made.
These incidents led SBV to hire a panel of scientists to review how Biosphere 2 science
was being conducted. During this time, the amount of oxygen inside the structure was dropping
from a normal 21% to 14%, approximately the amount of oxygen found at an altitude of 17,500
feet. SBV had no choice but to breach the structure's seal once again and inject oxygen-enriched
air into the Biosphere 2. Despite the recommendations of the science review panel, business
remained as usual at Biosphere 2 and by the end of April 1993, all of the science panel review
-------
134
members had resigned. As the chair of the science review panel, Thomas Lovejoy of the
Smithsonian Institution indicated, "The Biospherians will soldier on, but their two-year
experiment in self-sufficiency is starting to look less like science and more like a $150 million
stunt." Two years after their entrance, the biospherians emerged from Biosphere 2 as planned,
but were greeted with much less public and media enthusiasm. Fifteen to thirty percent of
Biosphere 2 species had gone "locally extinct" whil|e other populations exploded. Fruit trees had
produced little to no fruit, and all seven species of frogs disappeared. Despite these problems and
repeatedly breaking Biosphere 2's atmospheric seal, SBV proclaimed Biosphere 2 a success.
Biosphere 2's reputation and credibility nevpr recovered from the deliberate public
deceptions of its management and the media ridicule that followed. After remaining dormant for
two years, Columbia University's Lamont-Doherty Earth Observatory took over Biosphere 2 and
is currently using it as a hands-on research and educational center.
7.1.6 The Three Mile Island Accident
Although it led to no immediate deaths to plant workers or citizsns of nearby
communities, the accident at Three Mile Island on March 28, 1979, was the most serious nuclear
power plant accident in the history of the United States. Unbeknownst to the 140,000 residents of
Harrisburg, Pennsylvania ten miles away, a combination of human and system errors caused the
nuclear core bf the Three Mile Island nuclear poweif plant to dangerously overheat and melt.
At about 4:00 a.m., primary water pumps at the plant shut down allowing the cooling
waters circulating through the core of the plant to escape. The emergency backup water coolers
failed because their flow valves had not been reopened after routine testing two days prior.
Pressure immediately began to build in the main nuclear portion of the plant. A safety pressure
release valve opened to relieve the pressure, but later failed to close causing the pressure of the
system to fal^ below normal. This combination of decreasing pressure and delayed water cooling
produced errbneous readings in the control room, iri response to these faulty readings, plant
operators shut off the cooling waters, and the temperature within the reactor core climbed above
4,300 degrees Fahrenheit, 900 degrees below the complete meltdown threshold. This extreme
heat caused fuel to melt through the concrete containment floor of the raactor, and as a result,
radiation leaked into other areas inside the plant as well as into the outside environment. There
-------
135
was great uncertainty about how to properly contain and minimize the exposure to radiation.
However, predicting the next reaction of the core under such stress became a much greater
concern. One possibility was that pressure within the core would continue to increase, causing an
explosion of radioactive gas and debris. According to Walter Cronkite on CBS Evening News,
"The world has never known a day quite like today. It faced the considerable uncertainties and
dangers of the worst nuclear power plant accident of the atomic age. And the horror tonight is
that it could get much worse."
With these impacts and uncertainties in mind, all non-essential staff were evacuated by
11:00 a,m. that day. Over the next two days, efforts to halt toxic releases to the environment
failed. Two days later on March 30, Pennsylvania Governor Thomburgh called for the
evacuation of all preschool children and pregnant women within a five-mile radius of the plant
and ordered everyone within a ten-mile radius of the plant to remain indoors with their windows
closed. Unlike the April 1986 accident at Chernobyl, the nuclear power plant at Three Mile
Island was fortunately encased in a protective dome that prevented the leakage of large amounts
of radiation. Estimates of immediate exposure to radiation ranged from one millirem of radiation
to 100 millirems of radiation per person. (A standard x-ray exposes an individual to
approximately six millirems.) Phone calls to the Governor's Three Mile Island hot-line reported
dramatic effects related to radiation exposure, including stillborn and deformed pets and
livestock, unexplainable livestock deaths, radiation poisoning, and mutated vegetation. Two
dentists eight miles northwest of Three Mile Island also found that all dental x-rays of their
patients' teeth taken within two days of the Three Mile Island accident were fogged or banded.
After the Three Mile Island accident, residents also found dandelion and maple tree mutations
comparable to what was later found in Germany after the Chernobyl accident.
The health affects of the Three Mile Island accident were greatly disputed. After the
accident, several hundred people in the area reported hair loss, eye irritations, skin rashes,
headaches, menstrual irregularities, blistered noses and lips, nausea, vomiting, and a number of
livestock and pet deaths. However, initial studies of the area conducted by the U.S. Nuclear
Regulatory Commission (NRC); U.S. EPA; U.S. Department of Health, Education and Welfare
(now Health and Human Services); National Cancer Institute; Pennsylvania Department of
Energy; and State of Pennsylvania revealed the presence of "very little off-site releases of
-------
136
radioactivity." These reports further stated that "comprehensive investigations and assessments
by several well-respected organizations [conclude] that in spite of serious damage to the reactor,
most of the radiation was contained and that the actual release had negligible effects on the
physical health of individuals and the environment". In fact, according to one report, "residents
within 20 miles downwind of the plant had fewer cancer deaths than expected during the five-
year period." The report of the President's Commission on the accident at Three Mile Island
found the only adverse health effect of the Three Mile Island accident to be psychological. A
1990 study conducted by the Division of Epidemiology at Columbia University confirmed that
mere were no significant adverse health impacts resulting from the Three Mile Island accident.
Finally, Dr. Steve Wing, associate professoi
of epidemiology at. the University of North
Carolina, School of Public Health, led a ten-year study (1975-1985) of residents within ten miles
of Three Milje Island. His study found two to ten tir ics more lung cancer and leukemia among
residents downwind of the plant than among those Upwind. Dr. Wing also reanalyzed data from
the 1990 Columbia University study that concluded no significant increase in cancer due to the
Three Mile Island accident. According to Dr. Wing's analysis of the data and adjustment for pre-
accident cases of cancer, the 1990 Columbia University study does reveal "a striking increase in
cancers downwind from Three Mile Island." Dr. Wing further stated, "The cancer findings,
along with studies of animal, plant, and chromosomal damage in the Three Mile Island area
residents, all point to much higher radiation levels than were previously reported. If you say that
there was not high radiation, then you are left with higher cancer rates downwind of the plume
than are otherwise unexplainable."
The cleanup of the nuclear power plant took nearly twelve years and cost approximately
$973 million. As a result of the incident at Three Mile Island, the U.S. Nuclear Regulatory
Commission significantly tightened its regulatory standards and oversight of nuclear power
plants. Yet, ljhe public still perceives that nuclear pdwer poses a significant threat to public
safety. This distrust of nuclear power and scientific estimates is not unwarranted. According to
the Reactor Safety Study conducted in 1975 by Professor Norman Rasmussen of the
Massachusetts Institute of Technology with the fm4ncial support of the NRC, the probability of
-------
137
any nuclear power plant accident happening was ,000001 accidents per 1,000 reactor years.5
However, five years after the Three Mile Island accident, another NRC-sponsored study
increased this likelihood of a nuclear power plant accident to 1.7 to 4.5 accidents per 1,000
reactor years.
No new nuclear reactors have been built since the accident at Three Mile Island
Recently, however, as of 2004, two groups of companies have formed with the intentions of
applying for licenses to build new nuclear power plants. Though neither group actually has plans
yet to build the power plant, they intend to work with the U.S. Department of Energy to obtain a
license for an advanced nuclear power reactor.
7.1.7 Union Carbide Accident in Bhopal, India
At approximately 1:00 a.m. on December 3,1984, hundreds of residents of Bhopal, India,
sought medical attention for persistent coughing, extreme difficulty breathing, fever, eye tearing,
vomiting, difficulty keeping their eyes open, and brief spells of blindness. Because of the number
of complaints and the symptoms described, doctors at Hamidia Hospital immediately suspected
the release of toxic chemicals from the nearby pesticide-producing Union Carbide plant. When
questioned, the Chief Medical Officer for Union Carbide, Dr. L.D. Loya, admitted to an
accidental gas emission of methyl isocyanate (MIC) the night before. Dr. Loya, however, denied
that MIC was toxic and poisonous even though Union Carbide's plant manual clearly stated that
"MIC is a poison to human beings by inhalation, swallowing, and skin contact." Dr. Loya
explained that any temporary side effects experienced, such as agitated eyes and strained
breathing, would soon subside. Over the course of the next weeks and months, Union Carbide
maintained this position and claimed that MIC was "nothing more than a potent tear gas." They
refused to divulge any information regarding the composition or toxicity of the gas. Likewise,
Medical research conducted by the Indian Government regarding the Union Carbide-MIC
accident was also immediately classified as confidential by the government under the Official
Secrets Act. The effects of this "non-poisonous" gas, however, included approximately 3,800
Reactor years are the cumulative number of years in which ALL nuclear power plants have been operational. For
example, two nuclear power plants operating a total of two years each yields Four reactor years.
-------
138
j
immediate deaths and the hospitalization of 200,000 more people. One thousand animals were
also killed by the accident and another 6,000 harmed.
Studies conducted by the Indian Council of Medical Research (3CMR) revealed that
40,000 new cases of asthma were reported three mojnths after the accident. Five years after the
accident victims continued to display chronic deterioration of the lungs, gastrointestinal
disorders, partial and complete blindness, impaired jmmune systems, neurological disorders,
menstrual irregularities, reproductive disorders (including stillbirths and deformities), and post
traumatic stress symptoms. Additional ICMR studie|S also found: three times more people
suffered from MIC-related respiratory disorders in 1991 than in 1987; a three-fold increase in
pulmonary tuberculosis and cataracts among those exposed to MIC (as compared to an
unexposed control group of Bhopal residents); a 300% increase over the national average for
spontaneous abortions; delayed motor and language skills among children conceived or bom
after the accident; and the likelihood of permanent damage given the 10-year persistence of
symptoms. The International Medical Commission
-------
139
tank causing a highly exothermic reaction and the formation of a high-pressure MIC gas bubble.
This increase in pressure went undetected by plant operators. Apparently, the MIC pressure
value gauge was giving abnormally low-pressure readings (i.e. 2 psi instead of 20 psi) prior to
the accident. Two hours before the accident, the gauge indicated an increase in pressure from 3
psi to 10 psi, but plant operators decided that, like many others in the plant, the gauge was faulty.
Thirty minutes after the MIC reaction began, operators suspected a gaseous leak not
because of the plant's monitoring systems, but because their eyes began to tear. Unable to see or
breath, operators immediately abandoned their control panels. The refrigeration capacity of the
MIC storage tank had been turned off to conserve electricity, and all of the plant's backup safety
systems failed or were delayed allowing a massive amount of the gas to escape the plant.
Although a safety alarm sounded an hour after the spill, most of the damage had already been
done. Before residents could escape, MIC gas contaminated a 20 square kilometer area. People
within two and a half kilometers of the plant sustained lethal injuries, and people within a four
kilometer radius sustained severe injuries.
Union Carbide officials never notified local authorities of the toxicity of the chemicals
they produced. Consequently, no evacuation or emergency medical procedures were in place,
and no public knowledge of how to deal with the toxic gas cloud existed. Health officials
estimate that hundreds of lives could have been saved had the public been instructed to breathe
through a damp cloth. The catastrophe might also have been avoided if Union Carbide had: 1)
heeded the safety warnings issued after the death of one plant worker in December 1981, the
injury of 28 workers in January 1982, or the October 1982 spill of MIC, hydrochloric acid, and
chloroform and/or 2) repaired the "61 hazards, 30 of them major and 11 in the dangerous
phosgene/MIC units" as reported in a May 1982 safety audit. This outright negligence on the
behalf of Union Carbide affected almost a third of all people living in Bhopal at the time. To
date, approximately 20,000 people have died from MIC exposure.
Victims of the Union Carbide accident filed a $3 billion lawsuit against the corporation,
but Union Carbide adamantly denied liability blaming a fictitious disgruntled worker for the
accident. The Central Bureau of Investigation (CBI), with the help of plant workers, developed a
strong case supporting a connection between the MIC incident and negligent management. On
December 7, 1984, Union Carbide's Bhopal CEO, Warren Anderson, and 11 other corporate
-------
i 140
officials were arrested for culpable homicide, grievous hurt, and death and poisoning of animals.
Warren Anderson bailed himself out of jail for $2,000 (US) and has never returned to India.
In private, out-of-court deliberations and fearing the loss of other transnational
corporations, the Indian government settled on a payment of $470 million for the survivors of the
accident (an average payment of $940) as well as a full liability release for Union Carbide. In an
attempt to create goodwill, Union Carbide agreed to set up the Bhopal Hospital Trust, a hospital
to treat the victims of the 1984 accident. Today, Union Carbide's plant in Bhopal no longer
operates, but the site remains, never cleaned up and still potentially releases dangerous chemicals
into the environment. The Indian government has not been able to extradite Warren Anderson,
and in 2002 the Indian government was prevented by Indian courts from reducing the charges
against Union Carbide officials in an attempt to speed extradition from the United States.
7.2 Contradictory Information in the News
As the examples presented above show, news about human and environmental health is
omnipresent, yet much of this information is contradictory. Nearly everyday, newspapers,
magazines, and television shows report new information that tends to further obscure issues
rather than clarify them. A cursory survey of two major national newspapers conducted between
September 1,1999, and November 1,1999, yielded several articles that contested previously
reported claims or presented evidence of scientific or expert misjudgment and error. These
articles reported the following:
• "Studies Bolster Link between Diet Drugs, Heart-Valve Leaks." Contrary to the previous
claims of the manufacturer, the diet drugs Redux and fen-phen can cause permanent heart
damage (Wall Street Journal, September 10,j 1999).
• "Questions for Drug Maker on Honesty of Test Results: FBI Asks About Diet Product's
Approval." A drug manufacturer did not report to the Federal Drug Administration all
relevant test results prior to petitioning for approval of a drug (New York Times,
September 10, 1999).
• "Tobacco Industry Accused of Fraud Lawsuit by U.S." For more than forty years, the
tobacco industry suppressed evidence that tobacco use causes cancer (New York Times,
September 23, 1999).
-------
141
• "Japanese Fuel Plant Spews Radiation after Accident." Trained operators of a nuclear
power plant in Japan poured more than six times the required amount of uranium into a
tank, resulting in a nuclear chain reaction (New York Times, October 1, 1999).
• "Two Teams, Two Measures Equaled One Lost Spacecraft." The Mars Orbiter burned in
space because the spacecraft's creator used imperial measurements when the spacecraft's
navigational team used metric measurements (New York Times, October 1,1999).
• "Drug May Be Cause of Veterans' Illness: Pentagon Survey Links Gulf War Syndrome
to Nerve-Gas Antidote." Persian Gulf War soldiers who were given a drug to protect
them from nerve gas attacks suffer from damage to areas of the brain that control
reflexes, movement, memory, and emotion (New York Times, October 19, 1999).
• "Testing in Nevada Desert is Tied to Cancers." Soldiers who participated in nuclear tests
for the military in the 1950s have higher than normal death rates and an increased
likelihood of developing leukemia and prostrate and nasal cancer (New York Times,
October 26, 1999).
Due to this steady flow of events and news stories that present contradictory, inaccurate,
or incomplete expert evidence, the public is unlikely to accept expert evidence as absolutely
accurate all the time. The frequency of events, as well as the ambiguity and uncertainty of
experts, government officials, and the media, as demonstrated by these short case studies, leads
to doubt and skepticism on behalf of the public. The implication is that residents living near
Superfund sites are forced to construct their own risk beliefs based on perceptual cues and media
coverage. McClelland et al. (1990) surveyed residents near Oil about their risk beliefs and found
a bimodal response with more than half believing that living near the site was as dangerous as
smoking more than one pack of cigarettes per day, with an incremental annual risk of death of
approximately 1/100. Most of the remaining residents viewed the risk as trivial. Assuming
typical values for statistical life and assuming three people per home, the discounted present
value of the risk for the residents that assessed the risk as similar to smoking exceeds the price
paid by these residents for their homes! Residents who responded this way did report that they
were desperate to sell and sought immediate cleanup.
-------
142
7.3 Events, Perceptual Cues, Risk Perception, and Stigma
Given the doubts that people will inevitably have with respect to the credibility of expert
risk assessment, perceived risks will be based on personal and community judgments derived
from other sources of information. Events that are associated with a Superfund site will lead to
perceptual cues and media attention, that will most likely elevate perceived risk and stigmatize
the site for reasons documented below. Some of the,most important determinants of risk beliefs
are perceptual cues. Perceptual cues are physical aspects of a site that are perceived by local
residents, and are suggestive of risk. Examples of perceptual cues include odors emanating from
landfills, unusual odors or flavors in well water, unusual soil or water coloration at the site, and a
heavy volume of truck traffic going in and out of thf site. Ironically, the actions taken by
authorities to minimize public health and safety risks tend to exacerbate risk beliefs by providing
clear cues that some risk is present. Erecting chain link fences, posting 24-hour guards, placing
warning signs, conducting on-site tests (especially by workers wearing protective clothing) are
all cues to residents that risk levels may be higher than they thought. Such actions, which may be
necessary, almost never lower risk beliefs. Proximity to a site increases the frequency and
duration of contact with, or observation of, perceptual cues, which contributes directly to the
intensity of risk beliefs.
The effects of strong perceptual cues are well illustrated by the Oil Landfill. Initially,
concern about high volumes of truck traffic and odo|rs (produced by decomposition in the
landfill) prompted local residents to organize and confront problems associated with the site.
McClelland et al. (1990) found a significant correlation between recognition of these perceptual
cues and the high risk beliefs of many residents living near the site. Several of the perceptual
cues were removed or reduced by (a) installing wells to extract the methane gas for commercial
use and (b) closing the site, which eliminated most of the truck traffic. Even though these actions
did not address risks that hazardous substances would migrate into local neighborhoods, the risk
estimates of many residents dropped dramatically aijter the principal perceptual cues were
removed. McClelland et al. also demonstrated that there were significant property value losses
associated with these risk beliefs.
-------
143
Attention given to a site in the media, apart from the actual content of news stories, is
itself a perceptual cue that risks may be high. Many studies have shown that frequent exposure to
media reports about a site increases the likelihood that residents will believe the site is very
risky. The specific risk at a site and perhaps the site itself will usually be unfamiliar to residents.
That in itself increases risk beliefs (Wilkins and Patterson, 1987). But more importantly, it means
that residents are almost totally dependent on the news media for information about the risk.
Reflecting the concerns of their consumers, the news media often focus on aspects that
accentuate dread, such as the uncontrollability of the risk and the frightful worst outcome (e.g.,
dying of cancer), rather than on information about the low probabilities of the risk and how those
probabilities compare to other risks that residents accept.
The signals that the media sends to the public regarding risks from hazardous waste sites
are important, but the way in which the public interprets this information is equally important. A
key feature of how news coverage is interpreted by residents is whether there is an easily
identifiable "villain" responsible for the hazardous waste problems at the site. For example, if the
responsible party is a corporation whose primary business activity is outside the community, then
it is more easily portrayed as a villain than a local business which has strong affiliations to the
community. Russell et al. (1991) found that the more important a site's PRPs were to the local
economy, the more skeptical residents living near the site were that it needed to be cleaned up.
Personal familiarity with a site also influences how news reports are interpreted. The greater the
prior familiarity, the less risk beliefs are likely to be elevated by news stories.
The largest PRP for the Oil Landfill was an outside corporation that had not provided
significant employment or other economic benefits for the residents who lived nearby. Most of
the waste, especially hazardous waste, was generated and brought to OH from outside the
community. OH was primarily a commercial landfill serving many interests outside of the
community. In short, conditions were ripe for news stories to elevate risk concerns significantly.
How a risk affects the community, society, and the economy will depend on individual
and group perceptions of the risk (Slovic et al.. 1991; Kunreuther and Slovic, 2001). There can
be a compounding or "rippling" effect as more and more individuals respond to the risk
(Kasperson et al., 1988). Or, as Dr. Paul Slovic describes it, interactions among individuals can
-------
144
produce a "social amplification of the original risk concern." The greater the population living
near a site, the greater the potential for compounding or social amplification.
When residents or potential buyers are extraordinarily fearful of a site, they respond by
shunning the site. This behavioral response has been labeled stigmatizaiion and has been
explored in a number of experiments that suggest that if risks are perceived as being excessive,
people replace calculations of risk versus benefit with a simple heuristic of shunning, the
avoidance of the stigmatized object.
Stigma has been shown to have a number of key properties. Laboratory experiments
testing these properties have involved dipping a medically sterilized cockroach into glasses of
juice and gauging subjects' willingness to drink the Ijuice after the cockroach has been removed
(Rozin, 2001). First, stigma shares many of the psychological characteristics of contagion, where
contagion is associated with touch or physical conta:t. For example, while subjects refused to
drink the juice if the sterilized cockroach was dippetf into the glass, they would drink the juice if
the cockroach was just placed near it. Second, stigma appears to be permanent. Subjects refused
to drink the juice even if it had been in the freezer for one year. Third, stigma appeared to be
insensitive to dose. Reductions in the duration of contact between juice and cockroach had little
effect. Any contact was sufficient for subjects to shun the juice. Fourth, the source of contagion
is usually unknown. Thus, while shunning may have evolved from an adaptive response to avoid
contaminated food, it can be triggered in inappropriate circumstances. For example, subjects who
saw sugar water placed in a clean empty jar and then saw a cyanide label placed on the jar still
tended to refuse to drink the sugar water. Finally, subjects tend to medicalize the risk, arguing
that the stigmatization was the result of a fear of health effects.
The possibility that Superfund sites might be stigmatized could have a major impact on
the prospects for successful cleanup of contaminated sites. If such sites are permanently shunned
because, like,the "cockroached" juice, they are viewed as permanently stigmatized, property
values may riot recover immediately once cleanup i$ in progress (since future improvements
should be capitalized into home values) or even when cleanup is completed.
-------
145
Chapter 8
Property Value, Approach, and Data
8.1 Introduction
In this chapter, we undertake a comparison of hedonic property value models across four
different Superfund sites: The Montclair, NJ, area radium sites, the Oil landfill in the County of
Los Angeles, CA, the Wells G&H and Industri-Plex sites in Wobum, MA, and the Eagle Mine
site near Vail, CO. Our goal is to clarify whether the effects on property values of distance from
a Superfund site vary over time in a manner that is related to the progress of remediation at that
site. Distance is treated as a proxy for perceived risk. We control for area-wide variations in
housing prices and we also use GIS techniques to measure distances to other local amenities or
disamenities, which can confound the effects of proximity to the Superfund site. Some
provocative results stem from our extensive efforts to control for variations in the socio-
demographic characteristics of the surrounding neighborhood as measured by trends inferred
from tract-level data from multiple decennial Censuses. Neighborhood change occurs and may
be brought about in large part by the episode of "taint" precipitated by the identification and
remediation of the Superfund site. These shifting socio-demographics can easily confound
evidence about the rebound of property values that one would expect to observe following the
cleanup of a Superfund site.
There is considerable public concern about how hazardous waste sites impact property
values in their neighborhoods, both in the short run and in the long run. For a careful review of
empirical studies that assess the negative effects on property values of locally undesirable land
uses (LULUs) such as waste sites, hazardous manufacturing facilities, and electric utility plants,
see Farber (1998). The typical implicit or explicit goal in this literature is to establish a monetary
measure of the loss of welfare experienced by people who live near these sites. This monetary
loss can be interpreted as the amount of money that would be required to compensate these
individuals for this loss, or alternatively as a partial measure of the social benefits that would
ensue from affecting a cleanup.
-------
146
We examine distance and time patterns in property values in the vicinity of four different
sites on the NPL for the Comprehensive Environmental Response. Compensation, and Liability
Act (CERCLA, or "Superfund") sites in four different states. These include sites in
• Woburn, MA (the Wells G & H site that wa^ the subject of the book and motion picture,
A Civil Action, and the nearby Industri-Plex site);
• Montclair, NJ (three large areas with radium contamination of the soil);
• Monterey Park, CA (the Operating Industries, Inc (Oil) landfill site), also featuring some
of thq most dramatic demographic changes in Los Angeles County over the time period
in question; and
• Vail, CO (the Eagle Mine site, upstream on the Eagle River from a sizeable fraction of
housing along Interstate 70 west of Vail). Vail is notable as a winter resort area.
The key insights drawn from this work stem from our measurement of trends in both
socio-demographic characteristics and housing stock characteristics at the neighborhood level.
Taking advantage of the large data increment provided by the 2000 U.S. Census, we construct
approximate time-series for the set of conformable Census-tract-level variables in order to span a
31 year time period, based on the 1970, 1980,1990 and 2000 Censuses. We consider
neighborhood change, both in terms of socio-demographics (ethnicity, age, and household
composition) and the housing stock (owner- versus renter-occupancy and vacancy rates, as well
as shifts over time and with distance from the Superfund site in the characteristics of homes in
our large sample of transactions). ;
Our findings, borne out across several sites, suggest that endogenous neighborhood
change can be precipitated by the identification and remediation of a Superfund site. While the
objective or subjective risk associated with the site may initially account for decrements in
housing prices around the site, these lower prices aljso make the neighborhoods more accessible
to lower income groups, younger families, minorities, and non-traditional households. The
increased presence of these groups can replace Superfund site risks as an explanation for
systematically lower property values near the site over time, even after the site has been cleaned
up.
-------
147
A number of threads in the literature on hedonic property values must be acknowledged
explicitly, since each has some bearing on the approach we take here.
8.1.1 Objective versus Subjective Risk
One body of work in this literature emphasizes the effects of objectively measured risk on
property values. In a series of related papers concerning assorted hazardous waste sites and
housing prices in Greater Grand Rapids, Michigan, Gayer (2000), Gayer and Viscusi (2002), and
Gayer, et al. (2002) use detailed calculations of objective risk based partly on distance but also
on assumed transport of pollutants.
An alternative approach typically involves the use of distance from the site as a proxy for
either objective or subjective risk. This research falls into this second category. Even scientists
have difficulty assessing the true levels of risk from a hazardous waste site with any degree of
accuracy. Furthermore, home buyers tend to be less than perfectly well-informed about
environmental risks (see Hite, 1998), and it is the perceptions of home buyers and sellers, rather
than the facts, that will determine the prices observed for housing transactions. However,
McClelland, et al. (1990) explore the distinction between subjective and objective risks to
homeowners and their differential effects in hedonic property value models for the OH landfill.
They find that distance to the site and odor levels are not statistically significant when included
in a model in addition to a neighborhood average of homeowner risk levels.
8.1.2 Distance Effects over Time
In this research, we address some interesting empirical results concerning distance effects
over time in the hedonic property value literature concerning hazardous waste sites. Farber
(1998) reports that the literature suggests that the post-announcement impact of proximity to a
site is roughly $3500 per mile (in 1993 dollars). There is evidence in some cases that property
values that have been temporarily depressed by the announcement of Superfund status rebound
fully after cleanup (see Kohlhase, 1991).
A number of researchers have looked for this rebound effect. In particular, we will refer
to a style of hedonic inquiry represented in work by Kiel (1995) and Kiel and Zabel (2001), in
two papers by Kiel and McClain (1995), one paper by Dale, et al. (1999) and one by Carroll, et
-------
148
al. (1996). In these works, the authors estimate separate hedonic property value functions for
discrete intervals of time separated by mileposts in the process of identifying and remediating a
hazardous site.
Two of the Kiel papers (Kiel, 1995; and Kiel and Zabel, 2001) concentrate on a pair of
Superfund sites in Woburn, MA, Wells G &H (the subject of the book and motion picture
entitled A Civil Action) and the nearby Industri-Plex site. They identify six time periods: 1975-
76,1977-81, 1982-84, 1985-88,1989-91 and 1992. (These sites are also one of the cases
examined in our research, albeit for the time period 1978-97, and our data source is different.
Kiel provides a very thorough background for the Woburn sites.) The earlier paper controls for a
variety of structural characteristics of the house itself, and for the logarithm of minimum distance
from the Superfund sites, finding strong evidence of a positive price gradient away from these
sites in the second, fourth and fifth periods. The latqr paper also controls for two Census tract
variables: the proportion of unemployed workers and median household income in nominal
dollars. The paper does not appear to indicate whether more than one decennial Census was used
to create values for these variables. The variables are billed as being necessary to control for
omitted variables bias, and there is little further discussion of their contribution, since their
coefficients are not statistically significant.
Kiel and McClain (1995) consider the timeline for the siting of an incinerator in North
Andover, MA. These papers seek to discern different distance effects on housing prices in four
different time intervals, 1989-90, 1981-84, 1985-88 and 1989-92.
Dale and his coauthors focus on the RSR lead smelter site in Dallas, Texas, identifying
five time periods: 1979-80, 1981-84, 1985-86, 1987-90, and 1991-95. They use a sample of over
200,000 house sales distributed across these time periods, at an average distance of 11.8 miles
from the smelter. The geographic scope of the sample subsumes 14 school districts. These
authors interpolate between 1980 and 1990 tract-level Census data, and extrapolate based on the
growth rates in variables for 1979 and the years beyond 1990. They control only for the percent
of the Census tract below the poverty line, the percent Hispanic, and the percent black. These
Census variables are the only ones that vary over time, beyond the yearly dummy variables
included in the model. A simple distance variable is the key explanatory variable in these
-------
149
models, and the authors find that house prices increase with distance during the first three time
periods, but fall with distance in the last two.
8.1.3 Endogenous Socio-demographics
Each of these substantial studies that focuses on the time pattern of distance effects on
property values around a locally undesirable land use controls for two or three Census tract level
characteristics, if not in earlier papers from the project, then in later ones. But these earlier
studies were hampered by the absence of the 2000 Census data needed to construct plausible
trends over time in neighborhood characteristics beyond 1990. These researchers have been
limited to extrapolations based on the 1980 and 1990 Census data sets. The present work was
also delayed considerably in its completing while the authors awaited the release of the year
2000 Census results.
None of the earlier papers addressing the time pattern of distance effects reports any
exploration of whether population characteristics near the hazardous waste site also vary
systematically with distance as well as with time. Concerns about omitted variables bias are
acknowledged as justifications for including a few socio-demographic variables, but there are no
reports of scrutiny of the correlations of these Census variables with distances over time.
Why might we expect socio-demographics, potentially, to be correlated with distance in
ways that change over time? Housing prices are expected to be lower, the closer a property lies
to a newly identified Superfund site. These lower prices may result in dwellings being sold to
new owners who differ systematically from the existing population. If neighborhoods with
greater proportions of residents with the characteristics of these new arrivals are typically
associated with lower housing prices, this transition in the neighborhood may result in property
values in this area failing to completely recover their original trajectories over time.
The existing studies which control for time-varying neighborhood demographics when
measuring the effect on property prices of distance from a hazardous waste site may conclude,
from the estimated coefficient on distance that property values have rebounded from an episode
of Superfund designation and cleanup. However, demographics may be endogenously
determined by this process. The evolution of the neighborhood over this time period may leave it
with a different socio-demographic mix than prior to the episode. If these socio-demographic
-------
150
shifts have a negative effect on housing prices, these prices may not fully recover. Most models
attempt to make welfare inferences concerning the ijosses in capitalized housing values to pre-
existing owners based on the dynamics of the distance coefficient. What matters, however, is the
actual effect on housing prices.
8.1.4 Endogenous Housing Stock Attributes
In addition to localized demographic changes that differ from trends in the broader
community, the housing stock in the area near the Hazardous waste site may also be affected
differentially. There may be a shift in tenure from owner occupancy to more rental occupancy,
and there may be changes in vacancy rates. If homeowners are less inclined to remodel houses
nearest the site during the Superfund identification and remediation process, and developers are
less inclined to replace older houses or construct new dwellings for sale in this area, or if new
dwellings here are systematically different from new dwellings in area; beyond the influence of
the hazardous waste site, then these changes in the housing stock in neighborhoods nearer the
site can also contribute to sustained lower housing prices.
It is appropriate to control for demographic and housing stock changes, but all these
hedonic studies implicitly assume that these changes are exogenous, and therefore do not bother
to scrutinize them. We provide evidence that these variables are endogenously determined, and
changes in their levels are correlated with distance from the site and dynamically related to the
identification and remediation process. The full effpct on housing prices of "proximity to a
hazardous waste site" over time is captured not just by the simple distance coefficient, but also in
part by the full complement of socio-demographic and housing stock variables whose values are
also affected by the identification and remediation process.
McCluskey and Rausser (2001) utilize a dynamic, discrete-time model to analyze the
evolution of perceived risk around a hazardous wasjte site and its effect on property values.
Perceived risk enters the model as a state equation that involves a media coverage variable, and
is an unobserved variable whose values are imputed from the model. (Gayer and Viscusi, 2002;
also explore media coverage (newspaper stories) and their effect on property value changes in
the vicinity of Superfund sites.) McCluskey and Rausser's results suggest that media coverage
and high prior risk perception increase current perceived risk, which in turn lowers property
-------
151
values. However, the pattern of evolution of these imputed perceived risks over time is derived
from a specification that controls for distance from the site, but not for any changes in
demographics, which could also account for systematic shifts in housing prices. Perceived risk is
inferred to remain high if housing prices remain low. But if housing prices remain depressed near
the site because of changes in neighborhood socio-demographics precipitated by the Superfund
identification and remediation, such a model could falsely conclude that perceived risk remains
high.
8.1.5 Environmental Justice/Equity
This research also contributes to the environmental justice literature by explicitly
examining the effects on the socio-demographic mix of neighborhoods over space and time in
response to the identification and cleanup of three different Superfund sites. (The Vail, CO, case
has a settlement pattern that is too localized to allow for enough Census tracts for reliable
analysis.)
The neighborhoods containing hazardous waste sites tend to be lower-income and
possibly more non-white in their racial makeup. Bowen (2002) offers a critical review of the
existing environmental justice literature and concludes, on the basis of studies that he identifies
to be of relatively high quality, that
"... it appears to be that hazardous sites are located in white working-class
neighborhoods with residents heavily concentrated in industrial occupations,
living in somewhat less expensive than average homes."
He acknowledges the possible presence of other patterns at the subnational level, but that these
vary in their character from region to region.
When a spatial pattern of concentration of hazardous facilities in low-income or minority
communities can be established, the question arises of which came first. Do industries or
governments seeking to locate hazardous facilities disproportionately choose low income or
minority neighborhoods, or does the tendency of these facilities to reduce the prices of nearby
properties attract lower income home-buyers over time, and is ethnicity sufficiently correlated
-------
152
with income to produce this observed spatial patterii. Single cross-sections of data do not afford
an opportunity to discern which came first, the low-income or minority neighborhood, or the
hazardous waste site. It is necessary to determine how neighborhoods change over time, both
close to the hazardous facility and elsewhere. A disbussion of the issues is presented in Liu
(1997), and in Been (1994) and Been and Gupta (1997).
Graham, et al. (1999) explore the siting of coke plants and oil refineries. They identify
the year of the siting decision and retrieve historical Census data for the decennial Census
preceding that year (or the earliest adequate Census; data, if the siting decision preceded the
advent of sufficiently detailed Census information)] They conclude that market and non-market
mechanisms, such as redlining, block-busting and other legal and illegal activities, may dominate
the original coke plant and oil refinery siting decisions as explanations for the 1990 proportion of
non-white residents near these facilities.
Graham, et al. (1999) cite "market dynamic^ theory" as predicting, over time, that
hazardous or unattractive residential areas will lose high-income residents and attract low-
income residents (due to the relatively depressed property values in these areas.)
8.2 The Sample
An ideal sample of data would consist of transactions data and housing structural
characteristics, neighborhood characteristics, distances to all relevant amenities and disamenities,
all collected contemporaneously with the time of sale. This ideal data would also include
analogous information (except for selling price) abjaut houses that did not sell in these periods,
either because they were not for sale, or they did not find a buyer. This would allow the
researcher to control for non-random selection into the pool of dwellings actually observed to be
transacted.
When a researcher has data like these data <|>ver a number of years, it is possible to control
for many unobserved housing and neighborhood characteristics that do not vary across time by
using the so-called "repeat sales" method. When a house has sold more than once in the observed
time period, the difference in the selling price can be explained in terms of differences in any
explanatory variables that have also changed over time. This method for eliminating all the time-
invariant characteristics from the analysis was first'proposed by Bailey, et al. (1963), and has
-------
153
recently be used to good effect to analyze the influence of news stories about Superfund sites on
housing prices (Gayer and Viscusi, 2002). One disadvantage of this method is that the sample of
repeat-sales dwellings over-represents houses with greater turnover and excludes dwellings that
have been sold only once during the window of time for which data are available. There is also a
problem that any remodeling or updating of the property that is not captured by the quantity
variables typically recorded in multiple listing service data will go unacknowledged in the
process of dropping all structural characteristics by differencing over time.
We use a source of data that over-samples houses that have been sold only once over the
time period in question. Our data roughly reflect the current status of dwellings. The data are
provided, for the most part, by Experian, a company which provides information to direct mail
marketers and others. These data are updated at fairly regular intervals, although not
simultaneously. Anyone buying these records gets the most recent information available. For
each street address in the sample, most records include information on the date when the house
was purchased and the price that was paid at that time. For different localities, there are different
quantities of structural information in the data set. From the same data supplier, all fields will be
available for all localities, but for any given locality, blocks of fields will be blank. Blank fields
differ across localities, possibly reflecting different public recording requirements.
In some cases, notably the Eagle County files sought for the Eagle Mine site near Vail,
Colorado, the missing data problem was so severe that, despite the appearance of over 5,000
house transactions in the data, there were less than 50 with sufficient data for estimation of a
basic hedonic property value model. It seems that a large share of dwellings are not owner-
occupied. In that case, we sought and received data from the Eagle County Assessors office.
There were roughly 1400 observations for owner-occupied units, lying between 2.6 and 19.3
kilometers of the nearest part of the Eagle Mine site. About 57% were owner-occupied but were
not single-family dwellings. While the site description for the Eagle Mine site indicates that one
of the main deposits of tailings was within 1500 feet of a middle school, there are apparently no
current owner-occupied units in the vicinity. It would have been vastly preferable to have
acquired the same assessor's office information for each year during the time span of interest (in
this case, from 1976 to 1999). However, data that are "obsolete" from the point of view of the
-------
154
assessor's office are apparently not retained merely Ifor the convenience of researchers who wish
to understand time patterns in property values. ,
The feature that has the greatest potential to compromise these data, then, is the fact that,
in all of our data sets, we only observe selling prices for the most recent sale of a house. If a
house is in an area where turnover is high, there will be more recent sa'ies and fewer earlier sales.
For analytical purposes, it would be preferable to hkve data on all sales in all years and selling
price in those years, but such data do not exist. Data could be purchased from Experian every
year, if a future study could be anticipated, but retrospectively, the data are not available. The
data are collected primarily for current marketing purposes and records are updated without
saving their previous values. Historical modeling is not a use anticipated by the providers of the
data
Consequently, there may be some systematic sampling. We observe earlier transactions
prices only for houses which are still occupied by the owners who purchased them at that earlier
date. We do not observe many early transactions prices for houses in neighborhoods where there
has been a lot of turnover. It must be a maintained hypothesis that rates of turnover are
uncorrelated with identification and cleanup of Superfund sites. This may be a strenuous
assumption, but there are few alternatives. So it wiil be necessary to speculate upon the types of
biases this non-random selection is likely to produte in the effects of distance from a Superfund
site on housing transactions prices.
From these difficulties, we can glean an important item for the future research agenda.
i
8.2.1 Descriptive Statistics, Exclusions
For all of our sites, we exclude dwellings which are not owner-occupied, or if the selling
price or lot size or other key variables are missing, We also omit houses which could not be
successfully geolocated based on address information. For each site, we also limit the sample to
the range of years for which data are sufficiently plentiful. We exclude dwellings with very
unusual characteristics, such as very large values for number of floors, total numbers of rooms,
numbers of bedrooms, numbers of bathrooms, square footage, and lot size, or extraordinarily
high or low selling prices. We are attempting to model housing prices for a generous range of
"typical" dwellings in the geographic areas in question. Very unusual dwellings are therefore
-------
155
excluded. Each of these exclusions accounts for a very small relative number of dwellings, and
these unusual values may. in many places, be simply errors in data entry.
Selection on housing sale prices was conducted as judiciously as possible. Year by year
distributions of house prices, in levels and in log form, were scrutinized for obvious outliers.
Typically there were at most one or two deleted outliers in each year (aggregating across all
Census tracts). Prices are in nominal terms, so it is important not to exclude outliers based on
their identification from a marginal distribution.
Details are presented in the appendix for each site, along with complete descriptive
statistics for each of the classes of variables used in our models.
8.2.2 Extent of the Market
It is at the discretion of the researcher to decide upon the extent of the market that may be
influenced by proximity to a Supertund site. In some cases, if the disamenity is out of the line of
sight, it will have negligible effects. In other cases, where it may contribute to unhealthful air
quality or other obvious externalities, the geographic scope of its effects can be much more far-
reaching. Furthermore, perception of the disamenity can be directional. For example, it can be
influenced by prevailing winds (as in the case of the "Woburn odor"), or by the direction of
water flow in a river (some houses in the Eagle Mine sample are downstream of the mine, others
are not).
We selected a relatively large footprint for our initial models for each sample. For
Montclair, the 11,982 houses in the sample range from zero to about 6.7 kilometers away from
the site and a number of houses are actually located on top of areas of contaminated soil. The
broader neighborhood for the OH landfill site includes 9,279 dwellings between 60 meters and
about 8.5 kilometers from the boundary of the site. In Wobum, we characterize distances in
terms of the distance from the nearer of the two sites, since they are in such close proximity. The
sites are located in a non-residential area, so the range of distances represented among the 12,444
houses in the sample is from about 375 meters to about 8.4 kilometers.
The Eagle Mine site is anomalous. This is the mountainous territory around Vail, and
most settlement is clustered along either the Eagle River or Gore Creek, which runs west through
-------
156
the town of Vail and into the Eagle River at a point'downstream of the Eagle Mine site. Our
sample of assessor's data should be relatively complete, but there is a very high proportion of
non-owner-occupied properties in the region. There are only a handful of houses within six
kilometers of the Superfund site, and we delete these because of their potential to have an
inordinate effect on the price gradient in different years. The 1,087 owner occupied properties in
the Eagle Mine sample lie between 6.09 and about 13 kilometers from the downstream portion of
the site. No houses in the sample are upstream of the site. Furthermore, the numbers of
transactions in each of 1976 through 1999 are sufficiently small that we needed to constrain the
site distance^effects to be equal across three-year intervals in order to discern any reliable effects.
The size of the sampling area in a study such as this should be sufficiently large that the
distance premium in housing prices should arguably be zero near the boundary. We expect that
statistically discernible distance premia should emerge only for houses considerably closer to
each site.
8.3 Hedonic Property Value Models
Hedonic property values models have been used widely in literature, so we do not
undertake in this research to explain their theoretical justification or limitations for their use.
Many papers contain clear expositions of the underlying intuition. One recent example is Gayer
and Viscusi (2002).
Most housing attributes are dummy variables or small integer values, with the exception
of square footage and lot size. We retain square footage and lot size in linear form. However, all
key continuous variables, including selling price StRICE, the value of improvements, IMPVAL,
and all distances, are logged. This allows sufficient flexibility for us to see diminishing marginal
increases in housing prices with distance from a Superfund site, culminating in an essentially flat
distance profiles beyond the radius at which furthef increases in distance have a negligible effect
on property values.
Our basic model seeks to explain variations over space (i) and rime (t) in nominal selling
prices of dwellings, SPRICEit. For some variables, the only available data correspond to the
status of the structure in the last year of the sample|, which we will denote as t=T. Our generic
model takes the form:
-------
157
SPRICE = P • DIST
Here, Pt is an area-wide price index for owner-occupied housing in year t, DISTit is the distance
of each dwelling from the Superfund site in question, defined in a manner appropriate to the
case. The coefficient associated with this variable will be allowed to differ across years by
interacting the constant distance measure with yearly dummy variables. The variable v. signifies
lot size. This variable also appears as an element of the vector AiT of property attributes. Sit is a
vector of (interpolated) time- varying characteristics of the Census tract in which the dwelling is
located, and Z)T is a vector of the logarithms of the distances from the dwelling to a potentially
relevant set of other spatially differentiated local amenities or disamenities, calculated at time T,
the end of the sample period, rather than contemporaneously.
Taking logarithms of both sides of the equation yields a version of this model that is
appropriate for estimation:
LSPRICEit =
(A
o
where LSPRICE,, denotes the logarithm of the observed selling price, ln(.^) will be captured as
an intercept for the first year in the sample and a set of intercept shifters activated by year
dummy variables. The variables of key interest are the LDISTit, which consist of a vector of
logged distances from the dwelling to the Superfund site interacted with yearly dummies in order
to permit year-varying elasticities of housing prices with respect to distance to the site. The
control variables include the property attributes AiT, the Sa vector of proportions of the number
of persons, households, or dwellings in the Census tract falling into specific categories, and
theD,T set of logged distances to other local features, the identities of which vary according to
the case in question. The effects of these other logged distances are constrained to be constant
over the sample period.
-------
158
The version of the estimating specification ih Equation 8.2 highlights the special role of
lot size,v., in the modeling. Each locational amenifr
or disamenity should be permitted to affect
not just the price of the property overall, but its price per unit area. In our specifications, lot size
can affect property values outright, but it can also shift the marginal effects of amenities and
disamenities upon house prices. Our primary specifications restrict /?1U, /731, and /?,, to be zero,
but in the Appendices, we explore the consequences of allowing these parameters to be non-zero.
8.4
Control Variables
The main objective in this study is to determine whether there are statistically detectable
effects on the selling prices of houses due to proximity to a Superfund site, and whether these
effects, if any, vary over time as the site is identified and remediated. Before the incremental
effects of such proximity can be isolated, it is necessary to control for other factors which might
influence these prices. If any of these factors is correlated with distance from the site, or varies
overtime, or especially both, then the apparent timewise variations in prices due to proximity to
the site could be distorted by omitted variables bias.
In this study, we use four main classes of cpvariates to control for systematic variations in
housing prices due to heterogeneity other than the Affects of proximity to the Superfund site over
time.
8.4.1 Annual Dummy Variables
The average price of housing in a region will rise and fall over time in response to
regional business cycles that affect area-wide housing demand and other exogenous
macroeconomic factors, such as interest rates. To tike extent that housing prices vary over time
independently of the dwelling's distance from the local Superfund site, we control for all these
implicit time-varying factors with a set of yearly dummy variables in each of our models. Our
dependent variable in all cases is the log of the most recent selling price of the dwelling. The
annual dummy variables associated with all but the earliest period (the omitted category)
therefore bear coefficients mat can be interpreted a£ the area-wide percentage difference in
housing prices, relative to the first year of data, in each year of the sample.
-------
159
Appendices A through D contain descriptive statistics on the frequencies of observations
in each year of the period spanned by each sample. Given the systematic selection on house sale
data for houses still occupied by the same owner since the time of the sale, there are fewer
observations with earlier sale dates and proportionately more observations with recent sale dates.
8.4.2 Distance to the Superfund Site
Montclair: For Montclair, the radium sites spanned three large footprints, and housing had been
built on top of much of the affected areas. Distance is captured by a dummy variable for whether
the house in question was on top of one of these sites (INSIDE) interacted with yearly dummies.
It is also captured by distance from the boundary of the closest site, for houses which are not on
top of a site. Both of these variables—the dummy for being inside or outside a boundary, and
distance from the nearest boundary, if outside—are also interacted with lot size in a set of
auxiliary models. This allows for overall distance effects on selling price to be influenced by lot
size.
*
OH: In the OH case, there is one well-defined site with all housing external to that site.
Distance from the boundary of that site interacted with yearly dummies. It is also interacted with
lot size in auxiliary models.
Woburn: In the Wobum case, there are two fairly distinct sites, so the distances are calculated to
each of them separately. But there are relatively few housing transactions right near these sites,
since they are not located in residential areas. We went to considerable lengths to attempt to
capture the independent effects on housing prices of proximity to each of these sites: Wells G &
H, and Industri-Plex. However, the two sites are in too-close proximity relative to the
distribution of housing locations, so we were forced, in the end, to consider the distance from
each house to the nearest of the two sites.
The Industri-Plex site is known to have had a distinct olfactory externality downwind that
we have attempted to accommodate. We have included variables measuring the absolute latitude
and absolute longitude position of each dwelling. These variables allow for an underlying spatial
profile in the form of a plane that rises most steeply in any arbitrary direction, whatever the data
-------
160
j
dictate. On top of this spatial profile can be stacked an additional spatial profile of housing prices
that could rise with distance from the Superfund sites. If the distance premium is not symmetrical
around the site, the combination of these two patterns will allow the level curves of housing
prices relative to the site location to be elliptical and asymmetric around the site, as opposed to
circular and centered on the site.
Preview: If anything, the strategy of using latitude and longitude (separately for each
year of the sample) produced a suggestion that housing prices in many years seem definitely to
rise towards the south and rise towards the east (although possibly at a decreasing rate the further
north one goes). These absolute position effects are strengthened when distance from the nearest
Superfund site is also included in the model.
Eagle Mine: There are four major parts to the Eagle Mine site, each of which we mapped with
our GIS software and included as polygon features. 'In determining the distance from each house
in the sample to the Eagle Mine site, we used the distance to the boundary of the nearest feature.
In almost all cases, this is the feature that is furthest1 downstream towards the populated areas
along the Eagle River and Gore Creek, north of their point of confluence.
8.4.3 Housing Characteristics
The dependent variable in all of our models js the logarithm of the selling price of the
house.
All hedonic property value studies include ajt least some characteristics of the dwelling
itself and the property upon which it is situated. For each of our samples, the available structural
characteristics are as follows:
Montclair: In the Montclair case, the number of non-missing structural characteristics in the
data is unfortunately rather small. We therefore experiment with using the information in the
property tax assessment distinction between land vajlue and the value of improvements. In the
assessment data, the total (most recent) assessed value of most properties is divided into land
value and the value of improvements. In principle, the land value should reflect the value of the
location of the dwelling and the value of the improvements should reflect the quality of the
-------
161
structure itself. The impact of proximity to a Superfund site should show up the assessed value of
the land, and should vary over time as the perceived negative impact of proximity changes. But
the assessor's 1997 value of the improvements should reflect the size and quality of the structure
itself. This assessed value of the improvements will reflect any remodeling or additions that have
taken place since the last sale of the dwelling, so this is an imperfect control for the structural
quality of the dwelling when it was last sold. However, it is worth exploring the assessed value
of the structure as a rough proxy for missing structural information.
In some of the models we report, we have opted to employ the estimate of the value of the
improvements to the property as a regressor in the class of housing characteristics. We continue
to use all available and complete measures of the physical characteristics of the house. These are
typically quantity measures, rather than quality measures (e.g. number of bathrooms, rather than
how expensive the fixtures, flooring, and countertops might have been), so we use the
improvements value as a proxy for the "fit and finish" and the quality of the materials embodied
in the structure. Higher quality architectural features in a dwelling would be more likely to be
captured by the improvements value than by strict counts of bedrooms or floor space.
The quality of the most recent assessed value of improvements as a proxy for housing
quality at the time of the last sale will deteriorate with the elapsed time between that sale and the
most recent assessment.
Table 8.1 Montclair Housing Characteristics
t
Variable
impval
knovvflr
floors
ageknown
age20
age30
age40
ageSO
age60
age70
Definitional 1940)
assessed value of improvements, 1997
=1 if number of floors is known, =0 otherwise
= number of floors, if known
=1 if age of structure is known, =0 otherwise
=1 if age>=10 and age <20, =0 otherwise (omitted category = age<10)
=1 if age>=20 and age <30, =0 otherwise
=1 if age>=30 and age <40, =0 otherwise
=1 if age>=40 and age <50, =0 otherwise
=1 if age>=40 and age <60, =0 otherwise
=1 if age>=40 and age <70, =0 otherwise
-------
162
ageTOplus
lot size
=1 if age>=70, =0 otherwise
Size of lot, in ratio to sample average lot size
= 9232 sq. ft.
OH: The OH data are drawn from the same Experiin source as the Montclair data, but many
more data fields are non-empty in these data. Thus, a wider range of explanatory variables
capturing structural characteristics can be entertained.
Table 8.2 OH Housing Characteristics
Variable
notold
age
age2
sqft
sqft2
hedrms
bthrms
sqft bed
sqftbth
{place
knowflr
floors
lot size
Definition (n=9211)
=1 if structure was built after 1900, =0 otherwise
Age of structure if built after 1 900
Age of structure, squared
Square feet of floor space, in 'OOOs
Square feet of floor space, squared
Number of bedrooms !
Number of bathrooms
Interaction term: sqft * bedims
Interaction term: sqft * bthrms
=1 if at least one fireplace, =0 otherwise
=1 if number of floors is report, *=0 otherwise
Number of floors, if data not missing
Size of lot, in ratio to sample average lot size = 6199 sq. ft.
Woburn: The explanatory variables available for the Woburn model are the same as those
available for the OH site, although lot size will be normalized on the mean for this different
sample.
Table 8.3 Woburn Housing Characteristics
Variable
notold
age
age2
sqft
sqft2
Definition (n=l 2444)
=1 if structure was built after 1900, =0 otherwise
Age of structure if built after 1 900
Age of structure, squared '
Square feet of floor space, in 'OOOs
Square feet of floor space, squared
t
-------
163
bedrms
bthrms
sqftbed
sqftbth
fplace
knowflr
floors
lot size
Number of bedrooms
Number of bathrooms
Interaction term: sqft * bedrms
Interaction term: sqft * bthrms
=1 if at least one fireplace, =0 otherwise
=1 if number of floors is report, =0 otherwise
Number of floors, if data not missing
Size of lot, in ratio to sample average lot size = 15129 sq. ft
Eagle Mine: Given that the Eagle Mine data are drawn from the Eagle County assessor's office,
rather than from the Experian data, the available variables are somewhat different.
Table 8.4 Eagle Mine Housing Characteristics
t
Variable
sfd
age
age2
bedrms
bthrms
notwdframe
heatelec
constgood
constfair
downstream
lot size
Definition (n=1087)
=1 if single-family dwelling, =0 if condominium
Age of dwelling
Age of dwelling, squared
Number of bedrooms
Number of bathrooms
=1 if not wood-frame construction, =0 otherwise
=1 if electric heating, 0 otherwise
=1 if construction quality rated as "good" or better, =0 otherwise
=1 if construction quality rated as "fair" or worse, =0 otherwise
=1 if nearest to Eagle River, =0 if nearest to Gore Creek
Size of lot, in ratio to sample average lot size = 5154 sq. ft.
8.4.4 Neighborhood Characteristics
We define the neighborhood in which a house is located as synonymous with its Census
tract. Been and Gupta (1997) give a very thorough rationale for the desirability and tractability of
using Census tracts when there is a need to quantify socio-demographic and other very local
characteristics over time. Since neighborhoods change over time, and might be expected to
change as a result of the discovery and remediation of a Superfund site, it is important to
distinguish between changes over time in the effect on housing prices of mere proximity to the
-------
164
site, versus the influence of changing local demographics on property values. Hedonic property
value studies which ignore changing demographics are essentially estimating reduced form
model, which confounds the proximity effects with the changing demographics. It is possible
that the pure effects of proximity to a site are completely resolved with remediation, but the
effects of changing demographics as a result of the experience are more permanent.
We have been careful to collect Census data from all relevant decennial Censuses and
interpolated all of the conformable socio-demographic characteristics. For some sites, we needed
the 1970, 1980,1990 and 2000 Census counts in order to interpolate a series between, for
example, 1978 and 1997. Obviously, the decennial interval is problematic and these
characteristics will inevitably be measured with some error. Errors-in-variables attenuation may
lead to underestimates of the effect of neighborhood characteristics on housing prices.
The 2000 Census will offer greater resolutioji for a number of these variables, but the
complete 2000 Census data at the Census tract level are not yet available at the time of this
writing. We strived to achieve comparability across the different Censuses in these data, subject
to the constraints imposed by the available data in each year. Counts were collected for each
Census tract in each of four Census years, and categories were aggregated until they conformed
and the data could be pooled and used in an algorithjm to interpolate approximate values for each
variable in each year between Census years. This procedure resulted in a Census dataset that
could be merged with housing transactions by Census tract and year, so that the approximate
current neighborhood mix could be used to explain housing prices in each year.
The Census variables we constructed that conformed across all four Census years and
could be computed for each Census tract for each of our four sample areas are described in the
following table. A smoothing algorithm was used to "connect the dots" in each Census year and
to impute approximate values for each inter-Census year. These approximated time series will
not accurately represent inter-Census variations in e|ach variable, but this procedure seems to
dominate the'use of just one single year of Census year or the use of all three or four spanning
Census years for each of our samples.
Table 8.5 Neighborhood Characteristic Variables
Variable
tract
Description
1
Each tract number represented in th|e sample
t
-------
165
year
population
households
housing units
males
females
white
black
other
age_under5
age_5_29
age_30_64
age_65_up
marhh_chd
mhh_child
fhh_child
vacant
owner_occ
renter_occ
Actual data for each Census year; interpolated data for between-Census years
Total population of the tract in each year
Number of households in each tract in each year
Number of housing units in each tract in each year
Number of males
Number of females
Number identifying race as "white" (omitted category)
Number identifying race as "black"
Number identifying as "other race"
Number of persons aged under 5
Number of persons aged between 5 and 29
Number of persons aged between 30 and 64 (omitted category)
Number of persons aged 65 or older
Number of households consisting of married heads of household with children
Number of households consisting of male head of household with children
Number of households consisting of female head of household with children
Number of housing units vacant
Number of housing units owner-occupied (omitted category)
Number of housing units renter-occupied
These Census variables were converted to analogous percentage variables, prefixed with the
letter "p", based upon the appropriate denominator (either population, households, or housing
units). Where percentages sum approximately to 100%, the majority or (typically omitted)
category is dropped. For example, we will arbitrarily drop pmale, pwhite, page_30_64,
pmarhh_chd, and powner_occ. This leaves a typical vector of socio-demographic variables in the
same Census tract as each observation that includes:
[pfemale, pblack, pother, page_under5, page_5_29, page_65_up, pmhh_child, pfhh_child,
pvacant, and prenter_occ].
We anticipate that these Census tract characteristics may be very important to the
problem of sorting out the variations over time in the effect of proximity to a Superfund site on
housing values. Over the time horizons involved in our different cases (which range from 11
years to almost 30 years), there is a substantial scope for demographic shifts. Initial decreases in
housing prices due to the recognition of an environmental disamenity can make the
-------
166
neighborhood accessible to lower-income households, who may be prepared to accept the
disamenity in exchange for more housing at the same price, or cheaper housing, than they can
obtain elsewhere. Neighborhood characteristics are not independent of-the "taint" due to a
Superfund site. It is important to ascertain whether tie observation that housing price gradients
often tend to rise as one moves away from a Superfund site, even after remediation, may be due
to filtering-down of this housing stock that occurs during the period when taint is maximum.
It is entirely possible that, controlling for neighborhood changes that ensue from an
episode of major environmental taint, the eventual effect of proximity to the site is actually
positive (the price gradient moving away from the site is negative following remediation).
Homeowners! may value proximity to a cleaned up sjite more than they value proximity to other
less-certifiably safe features.
The fact that Census data are available only at ten-year intervals has been an impediment
to addressing this research issue. It was necessary td wait for the availability of the year 2000
Census data to be able to interpolate between the 1990 Census tract information and the 2000
Census information in order to construct usable data, for the last seven to nine years of housing
sales in our various data sets.
8.4.5 Other Local Amenities and Disamenities
Many earlier hedonic property value studies have included only one or two other
distances in their models (such as distance to the nearest shopping center, or the central business
district), or even no other distance variables at all. In the last few years, a number of
environmental aspects of spatial data have been examined by researchers who are interested in
determining their potential effects on property values. Acharya and Bennett (2001) use a sample
of about 4000 houses in New Haven county in Conrjecticut between 1995 and 1997 to explore
whether open space and land-use diversity affect housing prices. Diversity, richness, evenness
and dominance measures are used to quantify the patterns of land use within a fixed radius of
each dwelling. Among the spatial features they include as controls are distances to open space,
lakes, streams, the ocean, parks, and highways. They also use Census block group data for the
percentage of white households, the crime rate, average income, percentage college-bound
students, to quantify neighborhood characteristics. However, they do not attempt to isolate
f
-------
167
variations over time in the effects of perceived environmental quality on housing prices. The
marginal effects are assumed to be static.
In this research, based on our geo-location of each property using GIS software, we have
measured for each house the distance to a wide variety of other topographic features, land uses,
and institutions. Some of these are common to all four of our samples, such as distance to the
nearest park and distance to the nearest school or shopping center. Others are unique to each site.
Major freeways may be close enough to matter, or they may not, and airports or the flight paths
implied by their runway configurations may be close enough to affect housing prices, or they
may not. Based on a careful examination of the features in the region of each sample, we have
identified and measured distances for a wide variety of things that could plausibly affect housing
sale prices. The "other distance" variables relevant to each Superfund site are discussed in detail
in the appendices devoted to each site, but will be summarized briefly below.
Concerning precedents in the literature for controlling for certain classes of variables, we
can identify the following:
Summits: Benson, et al. (1998) assess the influence on housing prices of a variety of views,
differentiated by both type and quality. They find that depending upon the particular view,
willingness to pay for this type of amenity is quite high. While map proximity to a summit does
not translate into the presence of a view, there may be some correlation.
Schools: "Close to schools" is considered by many home-buyers to be a positive feature of
housing location, but Clauretie and Neill (2000) find that proximity to year-round schools, tends
to decrease housing values.
Roads: Spatiotemporal fluctuations in location premiums associated with a major urban
highway construction project in a study by Vadali and Sohn (2001). Boamet and Chalermpong
(2001) consider the effect of introducing new toll roads that increase accessibility.
-------
168
Airports: Bspey and Lopez (2000) find the airport] in their study to be a disamenity. This stands
in contrast to earlier work by Tomkins, et al. (1998), who found proximity to one particular
airport to be a net positive amenity. i
Railroads: Strand and Vagnes (2001) study the relationship between the price of residential
properties and proximity to railroads in Oslo. Bowes and Ihlanfeldt (2001) consider the effect of
commuter rail stations on the value of nearby properties.
Parks: Open spaces are not synonymous with parks, but work by Smith, et al. (2002) and by
Geoghegan (2002) considers the impact of open spaces on housing prices, suggesting that urban
parks may play this role to a certain extent in the more heavily settled examples in this research.
Water Bodies: Poor, et al. (2001), Spalatro and Ptovencher (2001), and Mahan, et al. (2000) all
consider hedonic property models with water features incorporated.
One shortcoming of these data concerning other distances is that they are "snapshot" data
based on the present geographic configuration of the local area. We have no way (at reasonable
cost) to reconstruct the appearance, during our sample periods, of new airports or new parks, for
example. We must rely upon the assumption that each of these features has remained fairly
constant over time (the 11 to 30 years of our samples) so that its effects are independent of time.
Of course, one could interact each of these distances with annual dummy variables to distinguish
year-specific, distance effects for each, but the number of regressors would rapidly become even
more unmanageable than it is at present.
It must be acknowledged, however, that the emergence (part-way though our sample
period) of one of the shopping malls present in our sample in 2000 could distort the apparent
effect of distance from a Superfund site in that year. Suppose a shopping mall appears inside the
sample area, close to the Superfund site, half-way through the sample period. Suppose further
that the presence of the shopping mall causes housing values to increase with proximity to the
mall. Prior to the introduction of this mall, the prevailing effect might be the negative effect of
proximity to the Superfund site. Proximity to the yet non-existent mall would have no effect.
-------
169
After the mall appears, however, proximity to the mall will increase prices, but proximity to the
Superfund site will decrease it, and the effects essentially wash out. The coefficient on the
distance from the mall location would be the average over the whole time period of the zero
effects without the mall and the positive effects with the mall, which would be overall positive
but smaller than their "true" effects. The coefficients on proximity to the Superfund site are
allowed to vary by year, however.
In the hedonic property value literature, there is some discussion about whether local
amenities and disamenities should make a fixed difference in the price of a property, regardless
of the size of that property, or whether these factors should in fact affect the per-unit-area price
of the property. We deem it unlikely that either one of these assumptions is entirely credible.
Hence, whenever a local amenity or disamenity enters a model, we explore expanded models
with both the amenity/disamenity distance and this distance interacted with the lot size of the
property in question. The effect of the amenity on house price (the derivative of expected
property price with respect to the amenity or disamenity) is therefore permitted to depend, in a
linear fashion, on the size of the property in question. If no such dependence is present in the
data, the coefficient on the property size interaction term will be indistinguishable from zero.
The "typical" magnitude of a non-constant marginal effect can sometimes be difficult to
appreciate when pondering regression results. Thus we undertake a convenient normalization.
Whatever our estimating sample, we first scale the raw data on lot size by the sample average of
the lot size variable, so that the sample mean of the scaled variable is one. Then, at the mean lot
size in the sample, the marginal effect of an amenity or disamenity that is interacted with lot size
will be given simply by the sum of the coefficient on the level of that amenity and the coefficient
on the interaction term. If these two coefficients are of opposite signs, the coefficient with the
larger absolute value will determine the sign of the effect at the means of the data. This
convention is observed for all of our local amenity and disamenity effects: neighborhood socio-
demographic and housing stock characteristics, distances from other potentially relevant
amenities and disamenities (parks, freeways, etc.), and the effects of distances from the
Superfund site in each year.
-------
170
Montclair We will mention the other distances
in the order in which they appear in the
estimating models. By using an additively separate specification in the logarithms of "distance to
the nearest X," we are of course imposing the strong assumption that the distance gradients
relative to all amenities or disamenities of the same type are identical, ;md that the effect on
house prices of proximity to one site is not affected
by proximity to another. These are strong
assumptions, but given the number of candidate variables, a considerably larger data set would
probably be required to discern the magnitudes of these interactions.
For Montclair, we include distances to the nearest summit of land, to the nearest school
and nearest retail center (shopping mall). We also use distance to the nearest hospital, church,
and cemetery. With respect to surface transportation we control for distance from a railroad, a
major road, Interstate 280 and the Garden State Parkway. We include distance from parks and
major bodies of water. Among institutions, we have distance to the nearest college or university
and distance to the nearest golf or country club. In addition to controlling for distance from the
nearest airport, we include a separate airport distance unique to Newark International Airport.
The details for each of these amenities or disameniljies are included in Appendix A.
Oil: For our housing price model for the Oil landfill area, we measure and control for
distances from the nearest school, retail center (shopping mall), hospital, church and cemetery.
We also control for distance from the nearest railroad and for distances from a number of
specific Southern California freeways: Interstate 5,| Interstate 605, Interstate 10 and State Route
60. The effects for these distances are in addition to those of the distance from the nearest major
road, which may sometimes be one of these freeways, but will often be a large "surface street."
We control for distance to the nearest "river" (often seasonal in Southern California) and the
nearest major body of water (typically anon-seasorjal river constrained within concrete for flood
control). The distance to the nearest park is measured, with additional controls for distance from
the largest regional park, the Whittier Narrows recreation area. There is one college campus that
may be relevant, the California State University' at Los Angeles, and we also control for
distances from the nearest country club or golf course.
Full details of the features captured by the "other distance" variables for the OH landfill
site are included in Appendix B.
-------
171
Woburn: For Wobum, we control for distance from the nearest summit of land, as well as
distances from the nearest school, retail center (shopping mall), hospital, church and cemetery.
Along with distance to the nearest railroad, we include a number of different measures of
proximity to roads: the distance to Interstates 93 and 95, the distance to the nearest principal
artery, the distance to the nearest other principal road, the distance to the nearest road including
smaller roads.
Air transportation corridors could also have an effect on housing prices in this area. We
control for distance from each of Boston's four airports: Logan airport, Beverly Municipal
airport, "Tew-Mac" airport and Hanscom Air Force Base. In addition to the simple distance
measures, we also plotted the potential flight paths over the sample area, based on the trajectories
of runways at each airport. We then computed the distance from each house to the nearest flight
path associated with each airport.
Finally, we include distances to the nearest park, major body of water, and country club.
Details concerning each of these types of "other distances" we have computed for the dwellings
in the Wobum area sample are contained in Appendix C.
t
Eagle Mine: The universe of available and potentially relevant "other distance" variables
considered in the Eagle mine analysis is displayed below. While we measured distances for
variables in the similar classes as those used for other Superfund sites, the distances to many
features, for the Eagle Mine sample, were great enough that one would not expect them to have
an influence on variations in housing prices within the sample.
Measured distances include distances to the nearest summit (and there are about 20
named summits in the sample area) and to the nearest river, either Eagle River or Gore Creek.
Elevation might plausibly explain housing values, but horizontal distances to the nearest summit
are likely to be a poor proxy for elevation in this terrain. We also collected information about
distances to the nearest school and retail area, hospital, church and cemetery, as usual. Only
schools, however, are contained in the sample area and some of the other distances are extremely
large (up to 50 miles). The effects of distances from the nearest railroad or road will be
confounded by proximity to the nearest river, since the highways and railroad follow either the
-------
172
Eagle River, or Gore Creek, or both, due to the topography of the area. There are no "major
bodies of water" in the area other than the rivers. "Recareas" include three golf clubs, and these
may have independent distance effects. Other points of local interest are too heterogeneous to be
combined (e.g. campgrounds, ranger stations). The one exception is distance to the Vail ski area.
We did measure distances from each site to the nearest town for seven distinct towns
(Avon, Eagle, Eagle-Vail, Leadfille, Mintum, Redcjiff, and Vail itself) However, the linear
arrangement of houses and towns along Interstate 70, leaves very little independence in these
distances. The main distance variable we retain in the model is the distance to the Vail ski area.
A full description of these "other distances" I for the Eagle Mine sample is included in
Appendix D.
t
-------
173
Chapter 9
Property Value Results
9.1 Classes of Hedonic Property Value Models
Our most basic specification, called Model 1, explores for time-varying proximity effects
in a generic specification of the following sort:
(9.1) LSPRICEa = lntf) + (fl0r)ID/S7; + &A,T +ett
This model is designed to mimic the most rudimentary type of specification that a researcher
might first explore when looking for time-varying distance effects. Distance is interpreted as a
proxy for perceived risk. When we estimate each of our models with time-varying distance
effects, we test the hypothesis that the profile of the distance premium (i.e. the perceive risk
premium) is the same in all years in the sample, since a simpler model yet would not even
distinguish distance effects that may vary over time.
The key argument of this research is that time-varying socio-demographics, both near and
further away from the Superfund site, can have a systematic effect on housing prices. To
determine the effects of demographics when other distances are not controlled, the generic
2 i
version of Mode! 2 is:
(9.2) LSPRICE,, =
However, the question of whether there are any general proximity effects, at all, cannot be
reliably ascertained without controlling for proximity to other amenities and disamenities. The
apparent magnitude of the proximity effects across years may be rendered generally higher or
lower by controlling for time-constant effects of proximity to other features. Model 3 will take
the general form:
t
(9.3) LSPRICE,, = ln(P,) + (P^LDIST, + p2AiT + (/?40)Ar +
-------
174
This model is similar to the type of model estimated by many previous researchers, although
most models of this form control for a rather limited selection of other distances.
Our most complete general model is Model 4, corresponding to Equation 8.2 above,
which employs all five classes of regressors: the set of year dummies that reveals the area-wide
price index, ln(Pt); the complete set of interactions between the logarithms of distance from the
Superfund site and annual dummy variables, giving (fr^LDISTj.; the structural characteristics,
Afj.; the other distances, DlT, and the time-varying demographics, Sit. In selected cases, we will
include other or different variables, such as potential lot size shifters on the key coefficients, or,
in the case of Woburn, latitude and longitude shifters for the entire price function, or
downstream/non-downstream variables, as in the case of Eagle Mine.
Before we delve into the details of the estimated hedonic properly value models,
however, it is important to look at some auxiliary models. These models specifically examine the
trends over time in the socio-demographic characteristics of the populations both near, and
further away from, each Superfund site.
9.2 Auxiliary Models Time-Varying Demographic Patterns
There has been considerable interest in the environmental equity/justice literature about
whether hazardous waste sites are selected because the population in the area is relatively lower
income, has a larger proportion of minorities, or unpikely to organize to resist a siting decision.
i I
The Superfund sites in our sample are not new siting decisions. These sites were in place before
much of the residential development around them began to occur. [See the approximate timelines
listed above.] Much of the population around the site has accumulated during or since the time
the site was in operation. The environmentally relevant "events" in our cases are not decisions to
site a new facility, but the news surrounding each sjte's designation as a Superfund site and the
i
resulting deliberation about cleanup strategies and Actual remediation. So in this context, it is not
so much the siting decision that matters, but the consequences over time of the publicity about
the site's Superfund status, and therefore public awareness of the site and the hazards that it may
represent. Perceived risk is assumed to vary over time, and should alsc vary over space in a way
that is approximately correlated with possible exposure.
-------
175
It is difficult to reconstruct reliable time-series of neighborhood socio-demographic
characteristics. As Gayer (2000) points out, Census block groups allow a more refined measure
of a dwelling's neighborhood than the Census tract. However, even Census tracts pose problems
for splicing together data across different Censuses. Tracts tend to be split as density increases,
and it is enough of a challenge to construct a time series at the tract level, since coding categories
change across time. We constructed a set of characteristics that aggregated socio-demographic
groups sufficiently to afford a match across decennial Censuses and aggregated Census tracts to
a common denominator across years. This leads to a loss of some resolution both in detail and
spatially, compared to what is available if one relies solely on the 1990 Census, for example,6
However, we deem it important for us to use a comparable set of variables across the three
samples where this strategy is feasible. We also need to approximate neighborhood
characteristics for years that span the 1970's, 1980's and 1990's, which requires conformation
across four different Censuses.
One of the most significant empirical enquiries into the effects of hazardous waste sites
on local demographics is described in Been and Gupta (1997). These researchers identify 608
commercial hazardous waste treatment, storage, and disposal facilities opened between 1970 and
1990. They collect Census data for the Census prior to the opening, and for 1990 for each of
these tracts. As controls, they draw a random 5% sample of all tracts in the U.S. They analyze
"before" demographics and "after" demographics, but of primary interest in the context of the
present work is their study of the difference-in-differences between the host tracts and all other
tracts, between the Census prior to the facility opening and 1990. As multivariate analyses, these
researchers pooled the host and non-host Census tracts and regressed the 1990 values for each
demographic characteristic (in percentage terms) against the value of this variable in the Census
prior to the siting, along with a dummy variable for whether the tract hosted a site.
Using these empirical techniques, Been and Gupta (1997) conclude that their study "does
not support the argument that market dynamics following a siting of a TSDF change the racial,
ethnic, or socioeconomic characteristics of host neighborhoods. The analysis suggests that the
areas surrounding TSDFs sited in the 1970s and 1980s are growth areas: in host areas, the
t
' Only "short-form" 2000 Census statistics are available at the Census tract level as of this writing.
-------
176
number of vacant housing units was lower than in sample areas, and the percentage of housing
built in the prior decade was higher. Such growth suggests that the market for land in the host
areas is active and should respond to any nuisance Created by the TSDFs. It also may suggest that
the burdens of the TSDF are being off-set by the benefits, such as increased employment
opportunities."
As Liu (1997) prescribes, any assessment of] market dynamics as a potential explanation
for neighborhood change around a locally undesirable land use requires controlling both for the
characteristics of the neighborhood before a siting decision and for changes in other
neighborhoods. The Been and Gupta (1997) approach uses randomly selected Census tracts
elsewhere in the country as a control for what is happening in "other neighborhoods." Here, we
control for patterns in other neighborhoods by enlisting the broader area around the host tract as
control tracts. Rather than looking for discrete differences in socio-demographic characteristics
between a host tract and tracts that are greatly displaced in terms of distance, we look for patterns
in demographics over time that differ continuously with distance from our Superfund sites. If the
socio-demographic patterns near the site are indistinguishable from those farther away the
distance gradient relative to the site will be flat. Depending upon conditions at the beginning of
our sample periods, there may be other reasons why we might observe a positive or negative
distance gradient in socio-demographic characteristics. What matters, however, is how this
gradient changes over time. If white residents tend t)o move out, and non-white residents to move
in, in the wake of publicity about Superfund designation and remediation, then the distance
gradient for whites should be observed to become relatively more positively sloped (or less
negatively sloped) with time. Likewise, the distance gradient for non-whites would be expected
to become less positively sloped (or more negatively sloped) over time.
In the auxiliary models described in this section, we explore the simple trends in
demographics near our four Superfund sites over time. Each observation is a house sale, and for
each observation we know the log of the distance from the site in question. As dependent
variables, we use the Census tract proportions of inhabitants in each class, pX, where X is a
Census category. These are the neighborhood characteristics associated with each house in our
main sample. In future work, we plan to aggregate the analysis to consider each Census tract as a
unit of observation, but for now we preserve houses as observations since this will most clearly
t
-------
177
highlight the multicollinearities present in our main hedonic property value models. For the
purposes of these main models, we interpolate the information for each Census tract across time
to provide approximate time series for each variable.
We assess the propensity for lower-income families, non-whites, and non-traditional
families to "come to the nuisance," possibly attracted by lower housing prices brought about by
taint from the Superfund site. The models we examine regress proportions for each socio-
demographic and housing tenure characteristic, across Census tracts and over time, against a
measure of distance, a time trend, and an interaction between distance and time. We use the log
of distance, since this transformation is important in our main hedonic property value models.
The log transformation allows any effect of proximity to the Superfund site to dissipate with
distance until, in the limit, further increases in distance have very little effect on socio-
demographic proportions. This is a reasonable maintained hypothesis, since the "reach" of
influence of any particular Superfund site must be finite.
Our specifications take the following simple form:
(9.4) pXit = a0 + a,LDISTit + a2t + a,t • LDISTU + v,,
Using the log of distance from the site permits the marginal effect of distance on proportions to
decline with distance, since as distance increases, the same percent change in distance
corresponds to a greater and greater increment of distance. This functional form assumes that the
effect of a percent change in distance is constant.
The effect of the log of distance from the Superfund site on the neighborhood proportion
of a given group is given by a, + aj. The effect of the passage of time on the neighborhood
proportion of a given group is given by a, +a3LDISTa. If the distance profile for the proportion
of residents of a particular type is merely shifting upwards or downward everywhere in the
region, we would expect to see a2 nonzero, but a3 = 0. In contrast, if the distance profile is
changing systematically over time, we will see a3 < 0 if this group is becoming relatively more
numerous near the Superfund site, and a3 > 0 if this group is becoming relatively less numerous
near the Superfund site (Figure 9.1).
-------
178
Figure 9.1 Changes in Socio-demographics near Superfund site over time
Census proportion
of group X
Area-wide shift plus localized change:
(change in intercept AND slope)
Area-wide shift:
Growth with no change in spatial gradient
(only intercept shifts)
Base year distance gradient
distance from s te
In the Montclair case, there is not only a distance variable measured from the perimeter
of the nearest site for houses outside the sites, there is also a dummy variable indicating whether
the dwelling in question is inside the site boundaries. In this case, Equation 9.4 will be adapted
to:
(9.5)
pXit = or0 + a, LDIST^ + a2t+a3t- LDISTfl
+ a, INSIDErt + aj • INSIDE,, + vu
In reporting these results, we concentrate on the estimated value of a3 (and a5 in the Montclair
case) for each type of Census group, X „ for each of the Superfund sites for which we are able to
employ Census tract characteristics (i.e. all sites except Eagle Mine). In the case of groups
becoming relatively more numerous throughout our sample areas, independent of distance from
the Superfund site, any effects of these changes on) housing prices will be picked up by the
-------
179
estimated area-wide price index, Pt, that is captured by the set of year dummies in the model for
LSPRICEit.
In perusing the estimated values of a} in the tables that follow, keep in mind that the
dependent variable is constant for all houses in a particular Census tract in a particular year. To
the extent that this variable does not perfectly reflect the "neighborhood" that is relevant to the
house sale in question, there will be a degree of error in this dependent variable. The explanatory
variables, distance and year, are observation-specific.7
It must also be acknowledged that these specifications constrain the time trends to be
monotonic. If they are non-monotonic, perhaps first increasing then decreasing, for example, or
vice versa, this would tend to obscure any trends.
(9.6) pXit = afl + a^LDlST^ + a2t + a3t • LDISTU + a/ + asr • LDISTit + vit
This specification yields a distance gradient for Census tract characteristic X in the form of
dpXit
(9.7)
8pXu _ g|
~
, .t, or
dLDIST 3
8DISTit DISTa
This implies that the effect will decrease with distance, and is quadratic in time, if the
as coefficient is significantly different from zero. The fitted turning point occurs at
t - -a3 /(2a5). If t* lies within the range of the data, the fitted time trend in the Superfund
There is an alternative strategy, which we are currently exploring. We find the effective center of gravity for the all
the houses in each Census tract across all years by computing the mean latitude and mean longitude of these houses.
We then locate these centers of gravity in our GIS software and use ArcMap to measure the distance from each tract
centroid to the boundary (or center) of the Superfund site. The number of observations can then be limited to just
actual Census years and individual or merged tracts that can be conformed across all Census years. In these more
limited data, we lose the resolution on distance for each house, but avoid the spurious estimation precision that
accompanies the use of the same Census tract information for all houses in a tract and smoothes the time profile by
interpolating. These results are not yet finalized.
-------
180
distance profile of Census characteristic X changes £ign. If t" lies beyond the range of the data,
the trend is approximately monotonic.
We have explored the data for evidence of significance in quadratic terms and found
some evidence. But one must bear in mind that the underlying real data consist of only four
decennial snapshots of neighborhood characteristics. The interpolated data are only approximate.
Given the quality of the data, we do not deem it judicious to push too hard on this generalization,
In this research, we report only the key coefficient estimates, and their standard errors,
for the a3 coefficient in Equation 9.4. The rest of the results for each model for each Census
tract attribute are presented in Appendices A through D. The appendices also illustrate the
progression of fitted distance profiles over time, with the heavier line in each graph depicting the
first-year in the sample and the progression being captured by representative intervening years
between then and the end of the sample period.
9.2.1 Montclair
There are significant numbers of observations from each of over 60 different Census
tracts represented in the Montclair sample, so there is considerable variation in tract
characteristics across the sample in any one year, and also in the interpolated time series within
each tract.
The Montclair model for each Census tract characteristic differentiates between distance
effects, implied by a3, and the effect of being on topj of one of the radium sites, as. Only these
i
two coefficients for each model are presented in the) following table. Trie rest of the results for
i
each model are presented in Appendix A.
Table 9.1 Montclair Census Traict Proportion Coefficient
Census tract
proportion
pfemale
pwhite
pblack
pother
«3 Coefficient
(robust t-test statistic)
.0000813 1.33
.0028637 3.00***
-.0033634 -3.39***
.0003952 1.85*
-------
181
page_under5
page_5_29
page_30_64
page_65_up
pmarhh_chd
pmhh_child
pfhh_child
powner_occ
pvacant
prenter_occ
-.000146 -3.98***
-.0005365 -3.00***
-.000451 -3.33***
.001029 6.10***
-.0005364 -1.79*
-2.45e-06 -0.06
-.0003317 -1.90*
-.0002585 -0.26
-.0000826 -1.11
.0003412 0.36
-.0000739 -0.38
.0012068 1.68*
-.0007805 -1.16
-.000403 -0.57
.0002539 0.18
.0002933 1.68*
.002167 2.54***
-.0023418 -0.45
-.0002405 -0.50
.0025831 0.53
Over the 1987-1997 time period for which we have data for the Montclair area, we will first
consider changes in the distance profile of relative concentration for different groups. The «3
coefficient if negative, conveys that the group in question has been becoming more concentrated
closer to the Superfund site. Groups that have become statistically significantly more prevalent
nearer the site include: blacks, children under 5, young people aged 5-29 and people aged 30-64,
as well as (possibly) married heads of household with children and female-headed households
with children. Positive «3 coefficients indicate that a group has become relatively less numerous
closer to the site. These groups include whites and seniors, and possibly other non-white ethnic
groups.
For houses inside the contaminated areas, a positive as coefficient implies an increase in
the proportion of that group over time. The relative share of female residents has become
statistically significantly higher overtime here, as has the proportion of blacks, female-headed
households with children, and possibly young persons between the ages of 5 and 29 and male
heads of household with children. In contrast, whites have tended to move away from the
contaminated areas.
One notable feature of these results is that there seem to have been no systematic effects
on housing tenure or vacancy rates, either on top of the Montclair sites, or close to them.
-------
182
9.2.2 Oil i
There are over 50 different Census tracts around the Oil landfill. This creates a diversity
of different values for neighborhood effects within any one year, and more variation across years
in the interpolated Census tract data.
In the 1960's and 1970's, Asian emigrants began to populate what had been a mostly
white bedroom community. By the 1980s, racial tension between longtime residents and new
immigrants became significant, leading to a controversial law requiring English on business
signs, Chinese books in the public library and attempts to make English the city's official
language. Racial problems have now mostly subsided and Monterey Park has become the only
city in the San Gabriel Valley with an Asian majority.
The pattern in this Southern California community is somewhat different from that in the
Montclair and Woburn cases. Our conformable Census measures do not include a distinct
i
i
category for Asian ethnic groups, so these populations are captured by the "pother" category.
Again, only the key «3 coefficient for each model ijs presented in the following table. The
remainder of the results for each model appear in Appendix B.
i
Table 9.2 Oil Census Tract Proportion Coefficients
Census tract
proportion
pfemale
pwhite
pblack
pother
page_under5
page_5_29
page_30_64
page_65_up
pmarhh_chd
pmhh_child
pfhh_child
powner occ
pvacant
prenter_occ
a3 Coefficient (with robust
t-test statistic)
-.0000198 -1.27
-.0018906 -10.81***
.000019 2.57***
,00i0202 11.42***
.0000226 0.92
.0002393 4.20***
.0003549 6.10***
-.0004772 -10.35***
.0009305 9.33***
-.0002995 -8.45***
.0000927 2.21***
.0009346 4.05***
.00p279 15.02***
-.0012124 -5.54***
-------
183
Over the 1970-1999 time period for which we have data for the vicinity of the Oil landfill
site, groups that have become relatively more numerous nearer the site include: whites, seniors,
and male heads of household with children (although this last group is very small everywhere).
The relative increase in whites near the site could be the flip side of a relative increase in Asian
groups at locations further way from the Superfund site. In an area experiencing a wave of
immigration, if immigrants avoid tainted neighborhoods and settle away from them, the older
racial mix of inhabitants will persist nearer the site. Renter-occupied housing has also become
more common near the site.
The groups that have become significantly relatively less numerous nearer the site
include blacks, and especially other non-whites (including Asians), as well young people aged 5-
29, middle aged persons aged 30-64, married heads of household with children and female heads
of household with children. Owner occupied housing has become less prevalent near the site, but
so have vacant properties. There has been little discernible change in the relative abundance near
the site of women or children under 5.
Graphical depictions of the implications of these models are presented along with
complete regression results in Appendix B.
9.2.3 Woburn
Our estimating sample contains significant amounts of data from 22 different Census
tracts in the Woburn area. Again, only the estimates for «3 for each model are presented in the
following table. The estimates of these key coefficients control for the baseline trend in
concentrations with distance, and for trends in the area-wide concentration of each group. See
Appendix C for more details on the remaining coefficients in each model.
Table 9.3 Woburn Census Tract Proportion Coefficients
Census tract
proportion
pfemale
pwhite
pblack
pother
«3 Coefficient (with robust
t-test statistic)
.0000174 0.42
.0019515 14.34***
-.0003281 -14.38***
-.0024137 -14.64***
-------
184
page_under5
page_5_29
page_30_64
page_65_up
pmarhh_chd
pmhh_child
pfhh_child
powner_occ
pvacant
prenter_occ
.0(»207;2 5.33***
-.00035-:
.0001 H
-.00006(
8 -1.93*
8 0.93
7 -0.41
.0016572 6.66***
-.000145 -11.20***
-.0006561 -10.69***
-2.66911 -1.54
-.0003236 -9.13***
-.00334$3 -7.69***
Over the 1978-1997 time period for which we have data for the Woburn area, groups that
have become relatively more numerous nearer the site include: blacks, other non-white groups,
the age 5-29 group, male heads of household with children, female headed households with
children, vacant properties, and rental properties. Groups that have become relatively less
numerous near the site include: whites, children under 5, and households with children headed by
married couples. Groups for which there has been little discernible change in the relative
abundance near the site are: females, persons between the ages of 30 and 64, seniors, owner
occupants.
Woburn was incorporated in 1652 and leather, tanning, and boot and shoe production was
the main industry from the mid-1800's to 1915. Suburban growth began in the mid-1900's and
has continued. The site has been described extensively by other authors, including Kiel (1995)
and Kiel and Zabel (2001).
9.2.4 Eagle Mine
There are insufficient numbers of Census tracts represented around the Eagle Mine site to
allow an analysis of trends in Census tract attributes associated with house sales over time.
9.2.5 Synthesis
It seems eminently clear that there is a strong tendency for fundamental neighborhood
change in the wake of a Superfund identification aid remediation process. The property prices
-------
185
we use from our analysis stem from house sales, and every time there is a house sale, the
occupants of that dwelling typically change. Who moves out and where they choose to go, and
who moves in, determines the change in the composition of the community in the vicinity of a
site. If the negative price shock accompanying a Superfund designation and the cleanup process
make housing in the vicinity of the site more affordable to lower income households,
unconventional households, ethnic minorities, or absentee landlords, a sufficient number of sales
can detectibly alter the makeup of the community.
There is a considerable literature in urban economics concerning the mechanisms of
neighborhood change (invasion-succession, tipping). The precipitating agent for the process, in
our cases, seems likely to have been the identification of the Superfund site.
Empirical models may fail to control for neighborhood change over time as a Superfund
identification and remediation process takes place. This can lead to omitted variables bias that
creates the impression that the Superfund process "taints" a neighborhood long after the site
itself has been cleaned up. In reality, what accounts for the persistent negative price differential
closer to the site could be the gradient in socio-demographic and income classes approaching the
site.
With site-induced neighborhood change, this "income-socio-demographic" gradient will
masquerade as a persistent "Superfund site proximity" gradient. When the site is clean, it may be
that nobody in the neighborhood or beyond is the least worried about any residual hazard. In fact,
having been certifiably cleaned, the site may even appear safer and more environmentally
attractive than competing uncertified areas elsewhere in the region. The true post-cleanup
"environmental gradient" might even display higher property prices near the cleaned site.
However, if one fails to control for the changed "income-socio-demographic gradient," it is
possible to misidentify the phenomenon as a persistent taint or perceived risk due to the site.
9.3 Auxiliary Models: Time-Varying Housing Attributes
The housing stock in a region can change more slowly than the characteristics of the
population. Many houses are remodeled and updated, sometimes to include additional bathrooms
or perhaps bedrooms, lots are subdivided in order to permit increases in density. In this section,
we examine the trends over time in the average characteristics of houses sold, as a function of
-------
186
their distance from each Superfund site. To the extent that the housing stock around a Superfund
site is not upgraded or renovated at the same rate as| dwellings in the more distant surrounding
area, this decline in the relative quality of the housing stock can account for persistent negative
price differentials in the region nearest the site. In the text of this research, we report only the
most significant effects. For complete models, see tjie Appendices A-D associate with each site.
The same basic estimating models used for Census tract proportions of different groups
in the neighborhood around the site are used in this section (either Equation 9.4, or 9.5 for
Montclair). Now, however, the dependent variables! are not all proportions of the population of
persons, households, or structures. Instead, the dependent variables are discrete or continuous
measures of structural attributes of each house itself.
93.1 Montclair
For the Montclair site, the data provide very few housing attributes to use in the property
value model. We exploit the age variable as completely as possible and substitute the most recent
tax assessor's "value of improvements" as a proxy for the current quality of the housing stock.
i
Interpretation of "impval" as a dependent variable is therefore somewhat problematic. We do not
observe its value contemporaneously with the sale of the house, but only currently.
Table 9.4 Montclair Housing Attribute Coefficient
Housing Attribute
knowflr
floors*
ageknown
agelO*
age20*
age30*
age40*
age50*
age60*
age70*
age70plus*
age*
«3 Coefficient
(robust t-test statistic)
-.0032718 -2.30***
.0029175 1.-H5
-.0045979 -2.71***
-.0002659 -O.kl
-.0009382 -1.80*
I
-.0005861 -O.'Sl
-.0111006 -5,79***
.0055656 2.93***
.0016836 0.194
.0053225 2.12**
.0003191 OJ10
.2577415 1.61
j
flfs Coefficient
(robust t-test statistic)
-.0065508 -0.73
.0123721 1.00
-.000424 -0.04
.0126552 1.52
-.002362:5 -0.76
-.0050581 -0.59
-.013702? -1.13
-.0093004 -0.87
-.002016 -0.33
.0208763 1.55
-.0010915 -0.06
.1463102 0.13
-------
187
lot size
.0002315 0.08
.015799
1.41
* in first column: if data observed;
The statistically insignificant as coefficients indicate that none of the information about age,
floors, or lot size for the Montclair properties displays any tendency to change systematically
over time for houses located on top of any of the radium contaminated sites.
The age dummy variables, capturing decade intervals of age for each house at the time it
was sold for the observed price, show a few notable patterns. The relative proportion of houses
that are less than 40 years old at the time of sale (age40=l) seems to have been increasing nearer
the site. The relative proportion of houses more than 40 years old at the time of sale has been
falling nearer the sites.
9.3.2 Oil
Table 9.5 Oil Housing Attribute Coefficient
Housing Attribute
age
sqft
bedrms
bthrms
fplace
knowflr
floors*
lot size
-------
188
zoning increases density and makes a neighborhood less attractive to many potential
homeowners. Lot sizes do not appear to be shrinking farther away from the site.
93.3 Woburn
Table 9.6 Woburn Housing Attribute Coefficients
Housing Attribute
notold
age*
sqft
bedrms
bthrms
t'place
knowflr
floors*
lot size
OT3 Coefficient (with robust
t-test statistic)
6.06e-06 0.01
-.3173891 -3.99***
.0026118 1.00
.0042233 1.33
.0109726 3.75***
-.0060986 -3.52***
-.0152776 -8.89***
-.00512
-.0026f
* calculated only for observations
49 -1.59
•04 -1.04
where data are observed
Over time,, houses which are sold nearer the isite are becoming differentially older than
those sold elsewhere in the sample area. Number of baths per house are increasing over time at
points more distant from the site, but not nearby the site. Fireplaces are becoming a more
commonplace feature in houses sold at points farther from the site. All this points towards a
conclusion that the housing stock nearest the site is not undergoing the amount of renewal and
there is not as much new construction near the site as there is elsewhere. If the housing stock
near the Superfund site is in decline compared to the stock elsewhere, then it is not surprising if
observed housing prices persist in being lower near the site than elsewhere in the sample.
93.4 Eagle Mine
The Eagle Mine data come from a different source than the data for the other sites, so the
available structural variables are somewhat different.
-------
189
9.3.5 Synthesis
In none of our examples does it appear that the housing stock nearest the Superfund site
is being renewed and upgraded at the same rate as housing at locations further removed from the
site. It may be the case that after a sufficient period of time has passed following a cleanup
project, and the site is designated as "safe" that homeowners in the area will again undertake to
accelerate maintenance of the housing stock to bring it back into line with the typical housing
stock in the surrounding area. However, if lower-income homeowners have moved into the area,
and if rental rates have increased, this may not set the stage for such accelerated renewal of the
stock. Despite cleanup, housing price may remain lower than the surrounding area due to
deferred maintenance and slower remodeling schedules or teardowns and replacements.
As in the case of the "income-socio-demographic" gradient created by earlier price
differentials due to the Superfund identification and cleanup process, we may see a "deferred
maintenance" gradient come into being relative to the location of the Superfund site. To the
extent that degradation of the housing stock accompanies a Superfund experience and the
attendant income and socio-demographic changes, and persists beyond the end of the cleanup
process, this factor may also masquerade as a persistent environmental taint.
9.4 Hedonic Property Value Models with Time-Varying Proximity Effects
Due to the very large number of control variables, and hence estimated coefficients, in
each of the models we estimate, we limit our discussion in the text of this research to the
properties of the key set of year-specific coefficients that describe the elasticity of selling prices
with respect to distance from the Superfund site, by year.
Our key results will be displayed in tables that merely note the presence or absence of the
other types of regressors in the specification. An extensive appendix for each site contains the
full specifications and other supporting statistical results. In the text of this research, we limit our
attention to specifications that constrain the lot size effects to be zero. In some cases, these lot
size effects are individually statistically significant, but including this generalization in the model
does not alter the general weight of the findings in any case, so we do not emphasize those
results.
-------
190
9.4.1 Montclair
The format of the estimating model for the Montclair sample is somewhat different from
the generic specification. Suppressing the $„, /?31, and /?„, parameters on the lot size interaction
terms, the Montclair estimating model is:
LSPRICEa =
(9.8)
Recall that since there are houses being sold that lie inside the boundaries of the radium sites in
the Montclair area, we include year-specific dummy variables, INSIDEit, for these houses. The
coefficients on these variables may also be permitte^ to vary systematically with lot size, but this
interaction is suppressed in this equation. For houses outside the radium sites, we compute the
distance from the boundary of the site to each house.
If the radium contamination negatively affects housing prices, we expect the
coefficients ft^t > m some or all years, will be negatij/e. If proximity to the radium contamination
depresses housing prices in some or all years, we expect the coefficients /?10,to be positive in
those years, reflecting the increase in average selling prices at locations further away from the
boundaries of these sites.
The change in neighborhood characteristics over time, in the vicinity of a Superfund site,
can contribute to the changes in observed property values that have nothing to do with levels of
perceived risk over time.
Table 9.7 presents the portion of the results f[or the Montclair sample concerning the price
i I
differentials for being on top of the site, as well as the distance coefficients intended to capture
perceived risks. Complete results are presented in Appendix A.
-------
Table 9.7 Montclair
191
insideS?
msideSS
inside89
insideQO
inside91
inside92
inside93
inside94
inside95
inside96
inside97
Idis87
Idisgg
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
idis97
structure
other
distances
neighborhood
characteristics
years
Model 1
Coefficient t-statistic
-.0657 (-.41)
-.1609 (-.92)
-.0995 (-2.23)**
-.02656 (-.56)
-.5115 (-2.28)**
-.08959 (-1.06)
-.2383 (-2)**
-.06177 (-1.58)
-.05908 (-1.33)
-.1324 (-1.94)*
-.1899 (-.98)
-.02189 (-.84)
.03968 (1.83)*
.008715 (.87)
.03492 (2.62)***
.04532 (3.8)***
.0368 (3.04)***
.03403 (3.74)***
.05929 (4.19)***
.0551 (5.34)***
,04956 (4.46)***
.09538 (5.62)***
yes
no
no
yes
Model 2
Coefficient t-stutistic
-.06151 (-.38)
-.1263 (-.7)
-.07226 (-1.36)
.04012 (.76)
-.4604 (-2.1)**
-.02405 (-.29)
-.1378 (-1.13)
-.0458 (-1.27)
-.0421 1 (-.95)
-.1386 (-2.09)**
-.205 (-1.1)
-.06202 (-2.35)***
.01417 (.7)
-.02332 (-2.34)***
-.0007486 (-.06)
.02125 (1.83)*
.008775 (.8)
.01032 (1,19)
.03949 (2.73)***
.03157 (3.16)***
.03363 (3.13)***
.08996 (5.26)***
yes
no
yes
yes
Model 3
Coefficient t-statistic
-.0907 (-.56)
-.2015 (-1.15)
-.1029 (-2.08)**
-.03043 (-.67)
-.5175 (-2.29)**
-.1016 (-1.11)
-.2408 (-2.02)**
-.03933 (-.87)
-.06601 (-1.35)
-.131 (-1.96)**
-.246 (-1.21)
-.03166 (-1.24)
.0374 (1.76)*
-.003609 (-.33)
.02711 (1.96)**
.03412 (2.78)***
.02854 (2.32)**
.02741 (2.88)***
.04881 (3.31)***
.04518 (4.1)***
.04136 (3.57)***
.08389 (4.96)***
yes
yes
no
yes
Model 4
Coefficient t-statistic
-.06221 (-.38)
-.1822 (-1.01)
-.06194 (-1.1)
.01035 (.19)
-.4786 (-2.13)**
-.05251 (-.58)
-.1809 (-1.48)
-.041 (-.97)
-.05923 (-1.14)
-.1389 (-2)**
-.243 (-1.23)
-.06222 (-2.39)***
.02204 (1.09)
-.02613 (-2.31)**
-.00106 (-.08)
.01903 (1.57)
.01049 (.89)
.01467 (1.59)
.04154 (2.74)***
.03649 (3.37)***
.03739 (3.27)***
.09065 (5.37)***
yes
yes
yes
yes
-------
192
9.4.1.1 Model 1: No Census Variables or Other Distances
At the Montclair site, the problem was ideraified prior to the beginning of our data. Hie
first Record of Decision was issued in 1989, whereupon remediation could begin. Remediation
was essentially completed by 1997, at the end of our sample period. Ws are looking for
statistically significant price differentials associated with a house being on top of one of the
radium contaminated areas, and/or positive coefficients on the log distance variables, suggesting
a premium for houses at locations which are exposed to lower risks. We find negative
coefficients on the INSIDE dummies in all years, and statistically significant coefficients in
1989, die year of the first ROD, in 1991, in 1993, and again in 1996. The distance premia in this
j
model are all positive after 1987 and statistically significantly different from zero in all years
after 1989. These results would suggest significant perceived risks closer to the site.
9.4.1.2 Model 2: Including Census Variables
When we control for variations over time in the socio-demographic makeup of the
population near die site, compared to changes elsewhere in the study area, we find that the point
estimates for all of the decrements in house value oij top of the site shrink in size up until 1996,
and only the decrements for 1991 and 1996 remain individually significant. The distance effects
actually change sign in the early years, becoming statistically significantly negative in 1987 and
1989, and they lose their statistical significance in 1990 through 1993. If distance captures
perceived risk, this perceived risk is only evident Chousing prices during 1994-1997, and its
magnitude is diminished from what was suggested in Model 1.
In the rest of the results for Model 2, reported in Appendix A, it is notable that the
proportion of blacks in the neighborhood makes a very strongly significant positive difference in
housing prices in this context. Appendix A also sho^vs changes in the degree of racial integration
across the different Census tracts over time. There are several predominantly black communities
and several predominantly white communities. Some of these remain strongly segregated
throughout the sample period, but a wide range of other communities in the area has become
considerably more integrated. Blacks, in particular, have moved nearer to the Superfund site in
considerable numbers. It also seems somewhat counter-intuitive that housing prices are highest
-------
193
where there is a higher proportion of vacant dwellings and a higher proportion of renter-occupied
units.
9.4.1.3 Model 3: Including Other Distances
When we include only other distances in Model 3, and leave out the Census variables, the
patterns in terms of the Superfund proximity variables exhibited in Model 1 are mostly restored,
although the point estimates of the distance (perceived risk) effects shrink very slightly. The
other distance effects, reported in Appendix A, seem to imply that high points of land (for New
Jersey), retail centers, hospitals, roads, Interstate 280, and especially the Garden State Parkway,
major water bodies (again, this is New Jersey), and Newark International Airport are considered
to be disamenities. Desirable features include: cemeteries, parks, colleges, and airports other
than Newark International. All of these seem to qualify as examples of open space, to some
extent.
9.4.1,4 Model 4: Both Census Data and Other Distances
This final model wherein the coefficients on neighborhood characteristics and other
distances do not depend upon lot sizes more or less preserves the same results obtained for the
Superfund site proximity variables attained separately in Models 2 and 3.
9.4.1.5 Models with Lot size Interaction Terms
y.4.1.3 Moaeis witn Lor size interaction i erms
In Table 9.8, we present just the coefficients on the Superfund proximity variables, but in
these four models, we now interact all of the proximity variables, the neighborhood
characteristics variables, and other distances variables with lot size to determine whether the
premium or discount associate with different attributes is independent of lot size or changes with
the size of the parcel in question. Recall that lot sizes are scaled to equal one at the means of the
data, to aid in the interpretation of the estimates. At the means of the data, the effective slope
coefficient will be the sum of the base coefficient and the coefficient on the lot size interaction
term. These models can reveal whether the losses in property values near a Superfund site are
borne disproportionately by homeowners selling smaller properties, or whether they are borne
disproportionately by homeowners selling larger properties.
-------
Table 9.8 Montclair (with lot size interactions)
194
inside87
insideSS
inside89
inside90
inside91
inside92
inside93
inside94
mside95
inside96
inside97
Idis87
Idis88
Idis89
Idis90
Idis9l
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
v inside 87
vinsideSS
vinside89
vinside90
vinside91
vinside92
vinside93
vinside94
vinside95
vinside96
vinside97
vldis87
vldisSS
Model 1
Coefficient t-statistic
1.204 (1.57)
-.2893 (-1.16)
-.156 (-2.17)**
.03244 (.51)
-.8026 (-2.52)***
.1341 (1.11)
-.4908 (-1.95)*
.08437 (1.05)
-.01186 (-.08)
.02536 (.28)
.1995 (.62)
-2.032 (-1.5)
.2354 (.71)
.08432 (1.63)
-.08558 (-1.2)
.4007 (2.08)**
-.2878 (-1.33)
.2283 (1-65)*
-.2021 (-2.08)**
-.07835 (-.31)
-.1843 (-1.85)*
-.5096 (-.86)
-.04527 (-1.34)
.02927 (1.04)
.04006 (2.54)***
.0554 (2.81)***
.06625 (3,6)***
.05323 (2.85)***
.07795 (4.94)***
.05488 (2.63)***
.0653 (3.54)***
,07897 (4.55)***
.06813 (2.3)**
.02719 (.94)
.01052 (.53)
Model 2
Coefficient t-statistic
1.328 (1.74)*
-.314 (-1.23)
i
-.09118 (-1.09)
.07578 (1.04)
-.7539 (-2.39)**'
.1988 (1.65)*
-.4436 (-1.83)* :
.1083 (1.37)
-.01322 (-.09)
.04732 (.54)
.1772 (.54)
-2.211 (-1.63)
.3521 (1.02)
.005509 (.08)
-.07026 (-.67)
.409 (2.1)**
-.3188 (-1.47)
.2368 (1.7)*
-.2102 (-2.07)**
-.0282 (-.11)
-.2274 (-2.16)**
-.491 (-.81)
-.06673 (-1.98)**
.001667 (.06)
.01334 (.8)
.02067 (1.15)
.05084 (2.85)***!
.02485 (1.44)
.05211 (3.38)***
.04984 (2.36)***
.04575 (2.19)** ;
.07134 (4.05)***
.08531 (2.76)***
.002705 (.09)
.0135 (.68) ,
Model 3
Coefficient t-statistk
1.312 (1.76)*
-.2724 (-1.09)
-.1442 (-1.79)*
.08889 (1.11)
-.8318 (-2.6)***
.1736 (1.39)
-.4005 (-1.64)
.1253 (1.38)
.03542 (.2)
.0682 (.75)
.235 (.72)
-2.242 (-1.7)"
.1583 (.48)
.06731 (.85)
-.1523 (-1.92)*
.4514 (2.3)**
-.3377 (-1.52)
.1408 (1.07)
-.2235 (-1.92)*
-.14 (-,43)
-.2392 (-2.16)**
-.598 (-1)
-.06854 (-2.06)**
.01633 (.59)
.01808 (1.13)
.01132 (.57)
.03852 (2.16)**
.02719 (1.47)
.04443 (2.76)***
.02964 (1.36)
.02722 (1.37)
,04789 (2.66)***
.05896 (2.05)**
.03459 (1.151
.02157 (1.04)
Model 4
Coefficient t-statistic
1.32 (1.72)*
-.2131 (-.81)
-.05622 (-.54)
.115 (1.41)
-.7885 (-2.45)***
.2425 (1.9)*
-.3638 (-1.5)
.1475 (1.56)
.001388 (.01)
.03598 (.36)
.1947 (.57)
-2.195 (-1.61)
.1111 (.31)
.002244 (.02)
-.1619 (-1.69)*
.4508 (2.24)**
-.3933 (-1.73)*
.1374 (1)
-.2371 (-1.84)*
-.05203 (-.19)
-.1943 (-1.52)
-.5287 (-.84)
-.08702 (-2.61)***
-.007 (-.26)
-.009747 (-.53)
-.02621 (-1.33)
.02363 (1.23)
.00332 (.18)
.03258 (1.97)**
.02317 (1.02)
.01563 (.74)
.04409 (2.26)**
.0669 (2.22)**
.02197 (.75)
.03402 (1.66)*
-------
195
vidis89
vldis90
v!dis91
vldis92
vldis93
vldis94
vldis95
vldis96
vldis97
structure
distance from
nearest site
interacted
with lot size
other
distances
other
distances
interacted
neighborhood
characteristics
neighborhood
characteristics
interacted
with lot size
years
-.0331 (-2.56)***
-.02587 (-1.45)
-.02399 (-1.46)
-.01942 (-1.14)
-.04774 (-3.15)***
.004041 (.26)
-.0126 (-.5)
-.03439 (-2.23)**
.02978 (1.1)
yes
yes
no
no
no
no
yes
-.03629 (-2.35)***
-.02958 (-1.75)*
-.03508 (-2.14)**
-.02103 (-1.25)
-.04496 (-2.89)***
-.01731 (-1.09)
-.0219 (-.71)
-.0479 (-2.81)***
.0002769 (.01)
yes
yes
no
no
yes
yes
yes
-.02188 (-1.66)*
.01631 (.9)
-.006006 (-.38)
-.0006561 (-.04)
-.01848 (-1,15)
.01786 (1.05)
.01796 (.65)
-.01098 (-.67)
.02711 (1.01)
yes
yes
yes
yes
no
no
yes
-.01317 (-.76)
.02515 (1.29)
-.004871 (-.25)
.00637 (.33)
-.01703 (-.99)
.01603 (.86)
.01531 (.51)
-.01636 (-.81)
.01808 (.63)
yes
yes
yes
yes
yes
yes
yes
In the version of Model 1 with lot size interaction terms, the coefficient on the interaction
terms for being on top of the site is statistically significant and positive in all time periods after
1988, suggesting that the impact of house prices of being on top of the site gets more positive
(less negative) as lot sizes get bigger, and these results control for lot size itself. Where
significant, the distance premium for a decrease in perceived risk, which can also be interpreted
as a proximity discount for increased perceived risk, gets smaller as lot sizes get larger. This
-------
196
means that houses on smaller lots at any given distance from the Superfund site experience a
larger relative decrease in value than do houses on large lots.
In the more complete specification in Model 4, however, with both neighborhood and
other distances in the model, lot size effects on the distance premium are much less apparent.
Appendix A also displays the lot size effects on all
)f the other variables. Over half of the
Census tract variables, and over half of the other distance variables retain statistically significant
lot size effects. In general, the lot size interaction terms have coefficients bearing the opposite
sign to the main effects, indicating that the housing price premiums or discounts associated with
location decrease in absolute magnitude as lot size grows.
9.4.2 Oil
For the Oil site, the generic estimating formula suffices. There are no special features to these
data The estimates for the distance effects are contained in Table 9.9.
9.4.2.1 Mocfel 1: No Census Variables or Other Distances
In this model, with no controls for socio-denjiographic change or distances to other
amenities or disamenities, the fitted model creates the impression that the locale of the landfill
was systematically more desirable than the surrounding area until about 1983. In 1984, the
landfill was closed and the site was proposed for the NPL. Following this, there is evidence of a
proximity premium in 1987,1992, and again in 1999, although a proximity discount is evident at
the 10% level in 1989,1991 and 1997. A Consent decree was signed in 1989, and construction
of some of the remediation measures began in 1991, 1996 was the year of the final Record of
Decision and landfill cover work began in 1997, so activity at the site would have been apparent
then.
-------
Table 9.9 Oil Landfill
197
Idis70
Idis71
Idis72
Idis73
Idis74
Idis75
Idis76
Idis77
Idis78
Idis79
IdisSO
IdisSl
Idis82
Idis83
Idis84
Id.s85
ldisS6
Idis87
IdisSS
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
structure
Model 1
Coefficient t-statistrc
-.09106 (-1.34)
-.1911 (-2.86)*"
-.1452 (-3.34)***
..1446 H.15)***
-.1544 (-3.18)***
-.1339 (-5.8)***
-.08965 (-5.91)***
-.09486 (-4.65)***
-.07947 H.49)***
-.02105 (-.53)
-.1574 (-3.79)***
-.1389 (-1.05)
.1391 (1.69)*
-.1294 (-3.28)***
-.02547 (-.53)
-.05666 (-1-26)
.001428 (.04)
-.05865 (-2.15)**
-.0136 (-.25)
.08966 (1.7)*
-.03567 (-.96)
,1127 (1.8)*
-.0556 (-2.59)***
-.02067 (-.94)
-.01194 (-.61)
.001991 (.05)
.04686 (1.25)
.0606 (1.75)*
-.004532 (-.34)
.02549 (2.15)**
yes
Model 2
Coefficient t-statistic
-.07008 (-.99)
-.153 (-2.28)**
-.129 (-2.96)***
-.1135 (-3.18)***
-.1353 (-2.78)***
-.1044 (-4.47)***
-.0682 (-4.31)***
-.07244 (-3.67)***
-.05892 (-3.35)***
.0009394 (.02)
-.147 (-3.56)***
-.1049 (-.8)
.1327 (1.56)
-.1166 (-2.98)***
-.009132 (-.19)
-.03818 (-.85)
.01342 (.43)
-.04202 (-1.56)
.0005704 (.01)
.1019 (1.98)**
-.0164 (-.45)
.1236 (1.95)*
-.03762 (-1.8)*
-.005524 (-.25)
.01012 (.54)
.01688 (.42)
.06034 (1.67)*
.08704 (2.56)***
.009177 (.72)
.02755 (2.22)**
yes
Model 3
Coefficient t-statistie
-.1274 (-1.71)*
-.2088 (-2.98)***
-.1767 (-3.81)***
-.1642 (-4.17)***
-.1842 (-3.5)***
-.1461 (-5.55)***
-.1182 (-6.14)***
-.1215 (-5.12)***
-.09687 (-4.36)***
-.04875 (-1.13)
-.1827 (-4.29)***
-.1629 (-1.22)
.104 (1.16)
-.1617 (-3.81)***
-.05933 (-1.17)
-.08788 (-1.85)*
-.03696 (-1.05)
-.08905 (-2.8)***
-.04601 (-.81)
.05101 (1.03)
-.07362 (-1.91)*
.07644 (1.1 1)
-.08487 (-3.25)***
-.05289 (-1.94)*
-.04331 (-1.74)*
-.0317 (-.72)
.01493 (.37)
.03561 (.96)
-.03436 (-1.76)*
-.01498 (-.75)
yes
Model 4
Coefficient t-statistic
-.08746 (-1.17)
-.1682 (-2.39)***
-.1442 (-3.03)***
-.1307 (-3.27)***
-.1533 (-2.88)***
-.1234 (-4.38)***
-.09191 H.42)***
-.09261 (-3.69)***
-.07437 (-3.09)***
-.01745 (-.4)
-.1584 (-3.65)***
-.12 (-.89)
.1175 (1.32)
-.1318 (-3.1)***
-.03113 (-.61)
-.05643 (-1.18)
-.005591 (-.16)
-.05456 (-1.67)*
-.01318 (-.23)
.08643 (1.74)*
-.03331 (-.86)
.111 (1.61)
-.04724 (-1 .74)*
-.01805 (-.64)
-.003816 (-.15)
.003596 (.08)
.0459 (1.13)
.0734 (1.98)**
-.004028 (-.2)
.01017 (.48)
yes
-------
198
other
distances
neighborhood
characteristics
years
no
no
yes
no
yes
yes
yes
no
yes
yes
yes
yes
9.4.2.2 Model 2: Including Census Variables
When we include Census variables, there ar^ no statistically significant proximity premia
after the site Closure in 1984, with the possible exception of 1992: where the distance effect is
negative and significant, but only at the 10% level. The apparent proximity premia suggested by
Model 1 for two other years after the landfill closure all disappear. Proximity discounts become
more strongly significant in 1989,1991,1997, 1999, and possibly in 1996, which was the year of
the final ROD.
The coefficients on the Census tract variables are presented with the full results in
Appendix B. Housing prices are enhanced in tracts ^vith a higher proportion of females, but are
lowered when there are higher proportions of children under 5, married heads of household with
children, and male heads of household with children. These Census data are very highly
correlated, so one cannot be certain that the independent effects of each variable are being
accurately captured. See the auxiliary R-squared values for each one of the Census variables,
presented in Appendix B.
9.4.2.3 Model 3: Including Other Distances
If we introduce into Model 1 only a set of other distance variables, not the set of Census
tract variables, the apparent proximity premia in the vicinity of the landfill site, evident in Model
1, reappear. These effects are strongly significant in many years prior to 1984, and considerably
less so afterwards. Again, there is little evidence in this model of any increase in perceived risk
nearer the site.
The distance variables suggest that in this area, churches, Interstate 10, and golf and
country clubs are amenities, while cemeteries, Interjtates 5 and 605, railroads, rivers (this is
Southern California, where many riverbanks are concrete to protect against flash floods), roads,
-------
199
and California State University at Los Angeles are all considered disamenities, as may be the
Whittier Narrows Recreation Area.
Table 9.10 Oil Landfill (with lot size interactions)
Idis70
Idis71
tdis72
Idis73
Idis74
Idis75
Idis76
Idis77
Idis78
Idis79
IdisSO
IdisSl
Idis82
Idis83
Idis84
Idis85
Idis86
Idis87
Idis88
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
vldis70
vldis71
vldis72
Model 1
Coefficient t-statistic
-.27 (-1,83)*
-.1485 (-1.2)
-.0259 (-.28)
-.07654 (-.91)
-.1172 (-1.31)
-.005313 (-.09)
-.1384 (-3.02)***
-.09239 (-2.24)**
.01811 (.41)
-.02743 (-.34)
-.1004 (-1.29)
.3079 (1.1)
.2359 (1.9)*
-.07051 (-.74)
.1435 (1.6)
.06277 (.89)
.1827 (1.79)*
-.09714 (-1.95)*
-.05173 (-.65)
.1526 (2.01)**
.107 (1.07)
.3504 (3.37)***
-.009454 (-.2)
.009659 (.24)
-.02342 (-.48)
.1405 (2.49)***
.1152 (1.63)
.09979 (1.97)**
.03436 (1.07)
.03965 (1.6)
.1847 (1.37)
-.03955 (-.47)
-.1149 (-1.42)
Model 2
Coefficient t-statistic
-.2827 (-1.8)*
-.175 (-1.27)
-.09312 (-.93)
-.1058 (-1.21)
-.1764 (-1.86)*
-.04556 (-.73)
-.1926 (-4.28)***
-.1263 (-2.97)***
-.06694 (-1.51)
-.0795 (-.95)
-.1876 (-2.28)**
.2662 (.95)
.1086 (.86)
-.0834 (-.92)
.08438 (.97)
.01631 (.23)
.154 (1.44)
-.131 (-2.58)***
-.05799 (-.73)
.1368 (1.9)*
.1163 (1.17)
.3204 (3.04)***
-.01275 (-.26)
.0063 (.16)
-.004869 (-.1)
.1549 (2.67)***
.1441 (2.05)**
.1283 (2.56)***
.05402 (1.59)
.05542 (1.92)*
.2119 (1.48)
.01896 (.2)
-.03491 (-.4)
Model 3
Coefficient t-statistic
-.2855 (-1.83)*
-.1823 (-1.31)
-.09609 (-.94)
-.08494 (-.99)
-.194 (-1.92)*
-.04565 (-.61)
-.1906 (-3.08)***
-.1348 (-2.2)**
-.03481 (-.56)
-.06993 (-.75)
-.1599 (-1.78)*
.2934 (1.01)
.1516 (1.07)
-.08315 (-.86)
.05283 (.53)
-.00965 (-.11)
.113 (1)
-.1757 (-2.47)***
-.1172 (-1.28)
.08203 (1.02)
,0691 (.67)
.2788 (2.33)***
-.08638 (-1.34)
-.07089 (-1.18)
-.08747 (-1.28)
.08279 (1.12)
.03524 (.42)
.04271 (.65)
-.03338 (-.62)
-.03476 (-.68)
.1554 (1.1)
-.02667 (-.26)
-.08224 (-.94)
Model 4
Coefficient t-statistic
-.2893 (-1 .76)*
-.1592 (-1.04)
-.09796 (-.85)
-.08642 (-.87)
-.2076 (-1.92)*
-.07106 (-.89)
-.2316 (-3.67)***
-.1678 (-2.52)***
-.1049 (-1.64)
-.1102 (-1.16)
-.2106 (-2.22)**
.2646 (.91)
.06886 (.48)
-.09621 (-.98)
-.005956 (-.06)
-.02657 (-.3)
.1158 (1.01)
-.1703 (-2.25)**
-.1056 (-1.09)
.09581 (1.15)
.09904 (.95)
.2747 (2.28)**
-.05117 (-.72)
-.03885 (-.58)
-.03493 (-.47)
,1349 (1.67)*
.1013 (1.12)
.111 (1.53)
.0321 (.51)
.03708 (.61)
.2052 (1.37)
-.007716 (-.07)
-.0461 1 (-.45)
-------
200
vldis?3
vldis74
vldis75
vldis76
vldis77
vldis78
vldis79
vldisSO
vldisSl
vidis82
vldis83
vldis84
vtdis85
vldis86
vldis87
vldisSS
vldis89
vldis90
vldis91
vldis92
vldis93
vldis94
vldis95
vldis96
vldis97
vldis98
vldis99
structure
distance from
nearest site
interacted
with lot size
other
distances
other
distances
interacted
-.P6112 (-.76)
-.0304 (-.44)
-.1062 (-2.25)**
.04763 (1.08)
-.004628 (-.15)
-.089 (-2.05)**
.005398 (.08)
-.05687 (-.81)
-.4912 (-1.56)
-.07137 (-1.17)
-.05983 (-.63)
-.1691 (-2.12)**
-.1112 (-2.57)***
-.1753 (-1.73)*
.03552 (.87)
.03658 (.58)
-.06469 (-1.12)
-.1528 (-1.37)
-.2477 (-2.63)***
-.04449 (-1.18)
-.03445 (-.97)
.01093 (.27)
-.,1418 (-2.64)***
-.071 (-1.6)
-.04198 (-1.26)
-.03999 (-1.38)
-.01624 (-.69)
yes
yes
no
no
-.008749 (-.11)
.03737 (.49)
-.04743 (-1)
.1328 (2.99)***
.05602 (1.72)*
.009527 (.21)
.07723 (1.13)
.04746 (.65)
-.4108 (-1.29)
.02564 (.39)
-.02607 (-.29)
-.09 (-1.22)
-.04642 (-1.1)
-.1323 (-1.25)
.088 (2.07)**
.05722 (.9) !
-.03303 (-.61)
-.1389 (-1.26)
-.2025 (-2.16)**
-.02181 (-.53)
-.01395 (-.43)
.01633 (.36)
-.1395 (-2.55)***i
-.08543 (-1.86)*
-.04459 (-1.28)
-.0434 (-1 .37)
-.02705 (-1 .04)
yes
yes
no
no
-.08803 (-1.14)
.01076 (.13)
-.09497 (-1.64)
.07517 (1.23)
.008887 (.17)
-.06312 (-1.08)
.01691 (.21)
-.03085 (-.39)
-.5125 (-1.57)
-.04998 (-.65)
-.08017 (-.87)
-.119 (-1.36)
-.07743 (-1.25)
-.1492 (-1.33)
.08476 (1.36)
.05822 (.77)
-.0361 (-.52)
-.1618 (-1.45)
-.2142 (-2.08)**
-.001723 (-.03)
.01704 (.32)
.03404 (.57)
-.1218 (-1.81)*
-.02421 (-.39)
-.01321 (-.25)
-.007049 (-.14)
.01472 (.32)
yes
yes
yes
yes
-.04798 (-.55)
.05655 (.64)
-.04278 (-.67)
.1502 (2.36)***
.07814 (1.29)
.03014 (.48)
.08485 (1.05)
.05572 (.65)
-.423 (-1.29)
.03959 (.47)
-.0243 (-.25)
-.02818 (-.31)
-.0189 (-.3)
-.1136 (-1.01)
.1174 (1.73)*
.08596 (1.06)
-.00754 (-.1)
-.1448 (-1.28)
-.1698 (-1.65)*
.003867 (.06)
.02118 (.34)
.02696 (.4)
-.1381 (-1.82)*
-.05795 (-.84)
-.04349 (-.7)
-.0393 (-.68)
-.03046 (-.54)
yes
yes
yes
yes
-------
201
neighborhood
characteristics
neighborhood
characteristics
interacted
with lot size
years
no
no
yes
yes
yes
yes
no
no
yes
yes
yes
9.4.2.4 Model 4: Both Census Data and Other Distances
When both Census variables and other distances are included in the model, there are no
strongly significant proximity premia after 1983. There is some evidence (at the 10% level) of
proximity premia in 1987 and 1992, and some evidence of a proximity discount in 1989 and
1997, the year a Consent Decree was signed and the year landfill cover work began respectively.
The only strongly significant demographic effects in this model remain the positive
effects of higher percentages of females and the negative effects of children under 5. There may
be a modest decrease in housing values accompanying greater renter-occupancy.
The only remaining apparent disamenities are the 1-605 freeway, rivers, roads, and the
Cal State campus. The only remaining significant amenity is the MO freeway.
9.4.2.5 Models with Lot Size Interaction Terms
Distance effects for models with lot size interaction terms are displayed in Table 9.10. As
was the case for the Montclair models, the inclusion of lot size interaction terms tends to lead to
lot size diminishing the absolute magnitude of the effect. The interaction terms typically have the
opposite sign from the baseline effect of any variable.
9.4.3 Woburn
Very conveniently, Kiel (1995) and Kiel and Zabel (2001) have labeled their six different
phases of the process at the Wobum sites: pre-discovery (1975-76), discovery (1977-81), EPA
announcement of Superfund listing (1982-84), cleanup discussion (1985-88), cleanup
announcement (1989-91), and cleanup (1992). Our results are detailed in Table 9.11.
-------
202
9.4.3.1 Model 1: No Census Variables or Other Distances
The most interesting feature of our Woburn results is that our simplest models tend to
confirm the findings of Kiel and Zabel (2001), who find that housing prices increase with
distance from the Woburn sites. In Model 1, we see
are insignificantly different from zero up through th
that the distance elasticities in housing prices
j end of the cleanup discussion. Beginning in
1990, however, there is evidence of a positive distance elasticity. The effect appears to dip in
1994, and again in 1997 (probably because of our smaller sample size covering only part of
1997). Otherwise, the distance elasticity ranges between 0.05 and 0.10.
Table 9.11 Woburn
Idisw78
Idisw79
IdiswSO
IdiswSl
Idisw82
Idisw83
Idisvv84
IdiswSS
Idisw86
Idisw87
IdiswSS
Idisw89
Idisw90
Idisw91
Idisw92
Idisw93
Idis\v94
Idisw95
Idisw96
Idisw97
structure
other
distances
Model 1
Coefficient t-statistic
-i02852 (-.74)
*.1323 (-1.78)*
-.0005082 (-.01)
-.05335 (-.63)
.007758 (.14)
-.09781 (-1.8)*
.04437 (1.17)
.1042 (2.37)***
-.008799 (-.2)
-.01143 (-.34)
.04308 (1.54)
103717 (1.36)
#5218 (1.93)*
,05028 (1.76)*
.07371 (2.63)***
.07933 (3.6)***
.04168 (1.67)*
.1095 (4.41)***
.1032 (5.65)***
.04631 (1.28)
yes
no
Model 2
Coefficient t-statistic
-.07515 (-2.01)**
-.1633 (-2.24)** '.
-.02039 (-.34)
-.1036 (-1.24)
-.04167 (-.77)
-.1412 (-2.74)***
-.1057 (-2.68)***
-.02891 (-.69)
-.101 (-2.39)***
-.1306 (-3.84)***
-.07725 (-2.64)***
-.08354 (-2.89)***
-.07328 (-2.62)***
-.06244 (-2.2)**
-.03901 (-1.32)
-.02932 (-1.26)
-.08882 (-3.48)***
-.01088 (-.43)
-.01269 (-.65)
-.0573 (-1.59)
yes
no
Mode)3
Coefficient t-statistic
-.0816 (-1.93)*
-.1858 (-2.47)***
-.05354 (-.84)
-.1366 (-1.58)
-.09406 (-1.66)*
-.1711 (-3.06)***
-.1 122 (-2.64)***
-.01804 (-.39)
-.1165 (-2.47)***
-.1179 (-3.11)"**
-.08086 (-2.39)"**
-.08138 (-2.46)"**
-.0873 (-2.66)"**
-.08535 (-2.52)-**
-.05477 (-1.54)
-.03415 (-1.19)
-.1062 (-3.54)"**
-.03558 (-1.19)
-.04663 (-1.83)"
-.1077 (-2.79)"**
yes
yes
Model 4
Coefficient t-statistic
-.1184 (-2.67)***
-.2127 (-2.75)***
-.07207 (-1.11)
-.1565 (-1.79)*
-.1157 (-1.97)**
-.1991 (-3.56)***
-.1772 (-3.91)***
-.0843 (-1.79)*
-.1607 (-3.35)***
-.1775 (-4.39)***
-.1327 (-3.64)***
-.1365 (-3.84)***
-.1376 (-3.79)***
-.1279 (-3.47)***
-.0982 (-2.55)***
-.08197 (-2.63)***
-.1525 (-4.79)***
-.08104 (-2.48)***
-.0921 (-3.22)***
-.1448 (-3.54)***
yes
yes
-------
203
neighborhood
characteristics
years
no
yes
yes
yes
no
yes
yes
yes
9.4.3.2 Model 2: Including Census Variables
Model 2, however shows what happens when we introduce our time-varying Census tract
information. What once were insignificant or positive distance elasticities now turn negative and
significant in many cases. The effect is dramatic. This model does not control for distances to
other amenities and disamenities, so we will not yet attempt to interpret individual Census tract
characteristics coefficients. However, only the proportion of black and the proportion of male
heads-of-household fail to make a statistically significant contribution. This is due to the tiny
absolute numbers of these groups in the tracts represented in our sample. In contrast, the only
neighborhood variables that Kiel and Zabel (2001) control for are the proportion of unemployed
in the Census tract and median household income in the Census tract. They find that the
unemployment rate influences housing prices only in the 1982-84 period, and median income
influences housing prices only in the 1989-91 period.
Why does the inclusion of time-varying Census tract information have such a profound
influence on the distance elasticity of housing prices? The answer seems to lie in the different
trends over time in the characteristics of Census tract nearest the site versus farther away.
Appendix C presents a full set of regression models and fitted time-and-distance profiles for the
neighborhood characteristics associated with each house in our sample. Since our main hedonic
price models employ the logarithms of distance, we use the log of distance in these specifications
as well. The most substantial socio-demographic effects we discover include:
• The proportion of whites near the site fell more than it did further away.
• The proportion of blacks, while remaining small, grew much more near the site than
elsewhere in the sample area.
• The proportion of other ethnic groups grew faster near the site than elsewhere.
* About a 30% growth elsewhere in the sample in the proportion of children under 5
whereas the population of young children nearest the site increases hardly at all.
• The proportion of 5-29 year olds shrank more slowly near the site than elsewhere.
-------
zw
• The population of prime-aged 30-64 year olds grew more slowly near the site than
elsewhere, as did the population of seniors.
• There was no discernible difference in the n te of decrease over time in the proportion of
married heads of household with children by distance from site.
• The proportion of male heads of household with children, while very small, grew more
quickly near the site than further away.
« Female headed households with children grew, close to the site, but declined as a
proportion of the population further away.
Owner-occupancy fell near the site, but remained more or less constant further away.
near the sjte, but remained relatively constant farther
!
• Renter-occupancy grew over time
away.
• There was no discernible difference in the gtowth in vacancy rates across the sample
area.
It must be noted that the suite of Census variables at our disposal are very highly correlated with
one another. The appendix reports the R-squared values for auxiliary regressions conducted
among the Census variables used in our Models, and these R-squared values are all over 60%
and range as high as almost 96%. As a consequence!, it will be difficult to attribute variations in
i [
housing prices to the independent effects of each of these variables. Collectively, however, they
make a considerable difference (for micro-data) to the R-squared value between Model 1 and
Model 2, boasting it from 0.49 to 0.53.
9.4.3.3 Model 3: Including Other Distances
Instead of controlling for different and shifting Census tract characteristics, we include
distances to other amenities in Model 3. These distances do not vary over time, but their presence
in the model also causes the previously significant site distance elasticities to change sign and
even become significantly negative. Among the set pf other distances, auxiliary R-squared values
reveal that the distances to the nearest d_retail, d_hospital, d_church, and distances to the four
airports (all of which lie outside the sample area) are highly correlated with the rest of the
variables in the model. The coefficients on these particular distance variables are prone to
multicollinearity problems.
-------
205
When we do not control for socio-demographic characteristics, it appears that the housing
prices increase with distance from: hospitals, churches, railroads, "other" primary roads, smaller
roads, Interstate 93, and all four airports. Housing prices appear to decrease with distance from:
schools, retail centers, cemeteries, principal arteries, the flight-path of Tew-Mac airport, parks,
major water bodies, and golf and country clubs. Each of these effects could be argued to be
plausible.
9,4.3.4 Model 4: Both Census Data and Other Distances
In Model 4, where we include both time-varying Census tract characteristics and other
distance variables, we find that housing prices seem to vary significantly with a number of
individual Census characteristics, although the resolution on these variables is doubtful because
of the high multicollinearity. Most plausible are the findings that housing prices tend to vary
negatively with the proportion of young people aged 5-29 in the population, and positively with
the proportion of married heads of household with children. Many of the "other distance"
coefficients lose their individual significance, leaving only the results that prices no longer
significantly increase with distance from: churches, and one of the four airports. Prices no longer
significantly decrease with distance from: cemeteries, the flight-path of Tew-Mac airport, or
golf and country clubs, but they now decrease with distances from summits. This seems
plausible, since proximity to a summit is likely to increase the chance of the dwelling having a
view.
The striking effect of including these Census tract attributes and other distances in Model
4 is that the site-distance elasticity of housing prices is now negative and significant at the 10%
level for all but one year in the 1978-1997 interval. It is significant at the 5% level in all but four
years. Next, we need to consider the time profile of these distance elasticities. How do they
change over time?
-------
206
.055484 -
-.364265 -
Figure 9.2 Woburn Model 4
- betal
-betal uc
- betaljc
-zeio
78 80 62 64 86 &8 90 92 94 96 98
yjear
Model 4 fitted site-distance elasticities: Wobum
9.4.3.5 Synthesis
In a naive specification, there do seem to be measurable impacts of proximity to the
nearest of two Woburn Superfund sites. However, when we control for other distances, it is
plausible that the apparent negative effect of proximity to the Superfund sites is just a
manifestation of greater distance from other desirable amenities or greater proximity to other
undesirable disamenities, including physical features as well as neighborhood socio-demographic
effects.
What then of the apparent variations in the effects of proximity to the nearest of the two
Wobum Superfund sites over time? There appears p be a substantial likelihood that the negative
i
effect of proximity to these sites towards the end of the sample period in a naive model like
Model 1 may be due to neighborhood transition. Lower housing prices in the vicinity of the
Superfund sites can be explained in part by demographic trends in that areas that differ from
those in the broader sample area.
There is evidence that Superfund site identification and remediation may at first lower
housing prices, but this impact in turn initiates a pattern of in-migration by socio-demographic
groups that previously would have been unable to afford housing in this area. Traditional higher-
income groups will be inclined to buy elsewhere and lower-income groups will have an
opportunity to move in. However, their growing presence may then become the dominant factor
-------
207
keeping downward pressure on housing prices, even though the Superfund remediation takes
place.
9.4.3.6 Models with Latitude and Longitude Variables
The Woburn models can differ from the generic model if we allow the absolute location
of each house to systematically affect the sales price. This permits a tilted planar spatial pattern
in housing prices to overlay the systematic effects due to distance from each house to the nearest
of the two Wobum sites; Wells G&H, or the Industri-Plex facility. The rationale for allowing this
generalization is that it may pick up the asymmetric spatial effects due to the "characteristic
Woburn odor" that was apparently carried generally eastward from the area of the sites by the
prevailing winds. The most general model for the Wobum site, short of including lot size effects,
is:
LSPRICE, =
(9 9)
Here, LLU denotes vectors of interaction terms between year dummies and the latitude and the
longitude of the house location (in decimal degrees, to six decimal places). We also consider
models wherein these latitude and longitude dummies in each year are interacted with the
distance variable. The models reported in the text of this research do not include this
generalization. These results are reported in Appendix C along with other more general models.
Appendix C details models that include latitude and longitude variables by year, and
latitude and longitude by year also interacted with log(distance from site), in addition to Census
variables and other distance variables. Housing prices appear to be significantly increasing to the
east and increasing to the south, overall (or, roughly increasing in the direction of Boston). This
is not surprising, and undoubtedly captures an accessibility effect as well as any directional
gradient in distance effect for the Wobum sites. Adding these variables to Model 4, discussed
above, does not alter its findings concerning the separate effects of the year-specific site-distance
variables.
However, there is a glimmer of something interesting when we also interact the site-
distance variables with the latitude and longitude variables. The coefficient on site distance then
-------
208
becomes a liaear function of the absolute location of the house. For these models, the years 1985
and 1988 develop positive and significant main distance effects, negative and marginally
significant latitude*distance effects and longitude*distance effects. The implied formulas for the
elasticity of housing price with respect to distance are as follows in these two years:
1985: 0.25 - 5.4 * latitude -3.2 * longitude
1988: 0.35 - 4.3 * latitude - 2.6 * longitude
Latitude increases to the north and longitude increases to the east. However, mean latitude in the
sample is 42.50474 degrees and mean longitude is -71.13818. Thus, at the means of the data, the
overall effect of distance on house prices is still negjative, rather than positive.
9.4.4 Eagle Mine
The Eagle Mine housing sample from the Eagle County Assessor's office does not span
enough distinct Census tracts for the differences in socio-demographic characteristics across
these tracts to be useful in explaining the variation jn housing prices. Only Census tracts 9534,
9535, 9536 and 9537 are represented in the estimating sample, and the total populations for each
of these tracts were only 6162, 2480,166 and 1134 persons at the time of the 1990 Census. If we
attenuate the footprint of the data to attempt to get a better picture of the more-nearby housing
price effects, we drop to just two Census tracts, which makes of the Census data perfectly
collinear. In -the spirit of the models used for the other sites, the richest Eagle mine estimating
model would thus be just:
(9.10)
LSPRICEU =
(/?40)Ar
However, there is more information that can be broiught into play in this case. Eagle River flows
NNW through the mine site. Several kilometers doWnstream from the mine site, it is joined by
Gore Creek, which flows in from the direction of Vail. The houses in our sample are split
between those lying on the Eagle River, downstream of the Eagle Mine site, and on Gore Creek,
which is not affected by the mine site. Thus, we differentiate between houses in these two
-------
209
groups, using a set of time-specific dummy variables for houses located downstream of the mine
site on Eagle River, rather than Gore Creek, DOWNST^. We will also interact these timewise
dummy variables with the distances from the Eagle Mine site, LDISTjt to yield a richest model
of the form:
This functional form allows the effect of proximity to the mine to depend on distance in the
following way:
dLSPRICEit = DOWNSTR
OLDIST, 1C* 2'
The elasticity of selling prices with respect to distance from the Eagle Mine site will be just $0/
in year t for houses on Gore Creek. For houses on the Eagle River, downstream of the mine site,
the elasticity of selling prices will be /710/ + /2,. If the estimated parameter ?2t in year t is
insignificantly different from zero, being downstream of the mine does not affect the elasticity of
selling price with respect to distance. If /?,<,, is zero and y2t is positive, then there is no premium
from being further from the mine site if the house is not downstream from the mine on Eagle
River, but there is a distance effect if the house is downstream.
In the housing data for this site, there are insufficient numbers of observations to permit
entirely separate distance coefficients to be estimated for each individual year. We retain the
yearly dummy variables to control for area-wide increases in housing prices in each year, but we
constrain the distance and downstream effects to be constant across roughly three-year intervals.
To conform with some of the main benchmark years in the site's history, and to create sufficient
observations while aggregating as little as possible, we hold the downstream and distance effects
constant across the following sets ofyears: 1976-1979, 1980-1982, 1983-1985, 1986-1988,
1989-1991,1992-1994, 1995-1997, and 1998-1999.
-------
210
The site was listed on the NPL in 1986 and 4 remediation plan was approved in 1988, so
there would have been much publicity in the area in
agreement between the agencies involved and the responsible party to evaluate and possiblely
construct a groundwater extraction system, although
the 1986-1988 period. 1996 saw an
the capping of the main tailings pile was
completed in 1997. This discussion might also have created awareness of the problem in the
1995-1997 period. In 1999, state and federal authorities formally sought to change the cleanup
agreement to include pumping of groundwater to keep it from filling the mine and complicating
treatment of contaminated water from the mine. The lead-up to this re-opening of the agreement
might be expected to influence housing prices in the, 1998-99 period.
Empirical results for the distance coefficient^ are presented in Table 9.12, with complete
results relegated to Appendix D.
9.4.4.1 Model 1: No Control for Other Distances
As usual, Model 1 considers distance effects over time, controlling for structural
attributes of the dwelling and general appreciation in housing prices, but not for any other
distances. This model suggests that distance from true mine site mattered little to housing prices
along Gore Creek, but did affect housing prices dovJnslream from the mine along the Eagle River
in some years. House prices increased with distance1 from the mine along the Eagle River in the
1976-1979 period, and in the 1986-1988 period. In 1992-1994, it seems that there were distance
effects along Gore Creek, but not along the Eagle River. This is difficult to explain. Thus we
consider a richer model, which also controls for other distances that may affect housing prices.
9.4.4.2 Model 2: Controlling for Other Distances
We include in Model 2 the logarithms of distance to the Vail ski: area, distance to the
nearest recreational area (golf course or country clup), distance to the nearest railroad, and
distance to the nearest river. The coefficients on the distances to each of these features bears the
expected sign and all are statistically significant. The Vail ski area is an amenity, with housing
prices decreasing as one moves away from it. Likewise, golf courses and country clubs are
amenities, as are rivers. In contrast, proximity to a railway is a disamenity. Housing prices rise as
one moves away from the railroad.
-------
211
Table 9.12 Eagle Mine
Idis79
Idis82
Idis85
IdisSS
Idis91
Idis94
Idis97
Idis99
downldist79
downldist82
downldistSS
downldistSS
downldist91
structure
other
distances
years
Model
Coefficient
-.181
-.3655
-.007467
-.1457
.1182
.4586
.05625
.1029
6.481
.9653
-1.42
1.716
-.5503
1
t- statistic
(-.57)
(-1.35)
(-.02)
(-7)
(.7)
(2.66)***
(.62)
(1.2)
(5.56)***
(1.22)
(-1.73)*
(2.11)**
(-.89)
yes
no
yes
Model 3
Coefficient
-.3849
-.4108
.1481
-.216
.002615
.3133
-.1811
-.0487
5.702
1.172
-1.084
2.599
.4107
t-statistic
(-.96)
(-1.47)
(.37)
(-.93)
(.01)
(1.5)
(-1.06)
(-.27)
(4.57)***
(1.47)
(-1.21)
(3.01)***
(.5)
yes
yes
yes
In this model, there are no individually statistically significant distance effects for houses
along Gore Creek, which are not likely to be exposed to any contamination from the Eagle River.
Downstream, however, there are significant positive distance effects in each of four different
time intervals. The largest effect appears to be in the 1976-1979 period, before the site was listed
on the NPL. If we knew more about accessibility of residential areas downstream area in this
period, it might be possible to say more about this observation. This area, further away from
Vail, may have been less accessible in that time period. Significant distance effects appear next
in the 1986-1988 interval, during which the site was proposed for the NPL and the remediation
plan was approved. This would have been a period of high publicity. The next discernible effect
came between 1995-1997. This is when the groundwater flooding problem became apparent and
the government agencies involved began to consider groundwater extraction in order to reduce
-------
212
the amount of contaminated water that had to be treated. A further effect is apparent in 1998-
1999 housing prices. In this period, the EPA formally proposed a change to the prior agreement
concerning remediation plans. It is not surprising that houses downstream of the site might
reflect this concern in their selling prices.
9.4.4.3 Models with Lot size Interaction Terms
Appendix D also details a set of models whejre the context of the dwelling, as opposed to
its attributes, affects not the unit price of the house, but the price per square foot of the land that
it occupies. We include both the usual lot size independent distance effects, and an interaction
term with lot size that allows us to see whether lot size affects the size of the premium for
distance from the site. Our models show that there are no strongly significant effects of lot size
on the "other' distances." If anything, the premium for being closer to the Vail ski area or the
local recreation areas (golf and country clubs) is enhanced when lot sizes are larger. (Lot sizes in
this model are normalized to one to permit evaluation of compound coefficients at mean lot size.)
However, the premium for being further downstream of the Eagle Mine site, now observed for
the 1976-1979, 1986-1988 periods and the 1989-1991 periods, is positive. For the first period it
increases with lot size, but for the two later periods the premium diminishes with lot size.
The positive effects of greater distance observed for the last two periods in Model 2
above disappear when we move to this more complex model. In this model, however, two
downstream simple dummy variables disappear due to multicollinearity and their effects are
absorbed by some of the interaction terms.
9.5 Synthesis and Conclusions
This research contributes four additional case studies to the literature on the time-varying
effects of localized environmental risks on housing prices. Over the relatively long time horizons
involved in Superfund identification and remediation processes, we find that the apparent time
patterns in proximity effects on housing prices seem to be confounded by systematic changes in
neighborhood composition in the vicinity of these sites. There is some evidence that housing
tenure patterns and the housing stock near the site are also altered by the process.
A "reduced form" type specification where individual house selling prices are modeled as
depending only upon structural variables, an area-wide price index for housing, and proximity to
-------
213
a Superfund site interacted with time dummies will indeed document observed patterns in
housing prices during a Superfund identification and remediation process. However, such a
reduced form cannot distinguish between the effect of perceived risk at each distance and the
effects of changing socio-demographic or housing stock variables at each distance. To isolate the
effects of perceived risk, one must control for these other effects, but at the same time, recognize
that these other changes are not exogenous.
The implicit experiment imbedded in an estimated distance effect is a change in the risk
associated with increased distance from the site. At a great enough distance, the risk is presumed
to go to zero, and so should the property value differential. When, over a long time horizon,
property value distance profiles do not return to a zero slope when the risk is reduced essentially
to zero, neighborhood change is a potential explanation for persistent price differentials with
distance. It is not at all possible to conclude that perceived risk does not respond to cleanup.
In some cases, especially the Wobum case, we find that controlling for timewise
variation in neighborhood characteristics such as gender, ethnicity, the age distribution, family
structures and housing tenure reveals very little in the way of a remaining distance profile, so
that any inferences about persistent risk perceptions are difficult to make.
-------
214
Chapter
it.
Conclusion: Stigma and Property Values
The possibility that stigma may cause large losses in property values has been noted by
other researchers (e.g., Dale et al., 1999; Adams and Cantor, 2001) and the EPA (Harris, 2004).
In contrast to the hedonic approach (Rosen, 1974; and for application to hazardous sites see
Bartik, 1998; Harris, 2004; Harrison and Stock, 1984; Ketkar, 1992; Kolhase, 1991;
Mendelsohn, :et al., 1992; Michaels and Smith, 199Q; etc.) where risk is treated as one of many
attributes thaf contribute to a determination of sale price, stigma is likely to effect property
values in a rather different and more direct manner. Upon learning of the contamination
potentially affecting their community, some current home owners may simply be unwilling to
continue to live in their home, and likewise, potential buyers will be unwilling to consider
buying a home in that community. If some owners ajid buyers have lexicographic preferences,
the standard hedonic model fails since it relies on a tradeoff between risk and home prices.
Rather, shunning by both current owners and potential home buyers will reduce the total demand
for housing for a neighborhood near a site as shown in Figure 10.1. Imagine that the total
demand for homes in a particular fully built-out neighborhood with H existing homes is Q(P)
where Q is the number of desired homes, P is the sate price, and quantity demanded falls with
price, Q'< 0. If, for example, homes were sold in a competitive uniform price auction, the
equilibrium price, Pc, is obtained by solving H=Q(P), so Pe=Q"'(H). Now consider the case where
a fraction f of home buyers and owners shun a neighborhood because of a nearby Superfund site.
The usual heionic model cannot handle this phenomenon because the hedonic price adjustment
for these individuals, either through very high subjective risk beliefs (assuming conventional
values of statistical life) or shunning would give homes a risk deficit greater than or equal to the
value of the home. In other words, in either case the perceived costs of staying in the home are
greater than the entire value of the home and the observed behavior would be identical. This
implies that fraction f of current owners will sell and that the number of potential buyers will be
reduced by fraction fas well. As shown in Figure 10.1, since we have defined total demand for
the neighborhood to include current owners, the equilibrium price will now be determined by the
solving H=(l-f)Q(P)> so PB*=Q'I(H/(l-f)) and Pe* < Pc for f > 0. If f falls with distance from the
-------
215
site, as is likely since perceptual cues decline with distance, then property values will rise with
distance, ceteris paribus. Of course, relative demand for housing that is more distant from the
site will increase, but presumably this increase in demand will fall on a much larger group of
homes, resulting in a negligible increase in prices of homes farther from the site.
Figure 10.1 The Effect of Stigma on Equilibrium Housing Prices
The next question is, since a hedonic analysis is used to incorporate normal attributes for
predicting property prices, how can downward sloping demand be incorporated into the analysis?
The answer proposed here is that hedonic models predict an average price based on home and
community attributes, but do not take into account individual buyer characteristics, including
bidding errors, which will affect the willingness to pay for homes in a particular area. So, for
example, relative to a predicted hedonic price, PH, one particular individual will be willing to pay
more because grandmother happens to live in the neighborhood and another particular individual
will be willing to pay less because of a random error in bidding strategy. Clearly no hedonic
market can exist for such attributes since they are buyer specific, and these sale price deviations
will appear as part of the error term in the estimated hedonic equation. Thus, for homes with a
particular set of hedonic attributes in a homogenous neighborhood with a mean sale price of PH,
-------
216
there exists an array of values for homes among potential buyers, V, with a cumulative
distribution function of Q(V). Presumably, the H buyers with the highest individual values will
own homes in the area. i
To further understand the property value market, we model the market itself as a
discriminative auction to account for the fact that identical homes in the same neighborhood can,
in fact, sell for different prices depending on unobserved individual buyer errors and other
attributes (see Cox et al, 1984, for a discussion of the relevant theory and an experimental test of
this auction). Approximating the property value market with an appropriate auction where
multiple buyers compete for available homes solves the potential problem associated with
modeling real estate sales as bilateral negotiations where some sellers potentially have no value.
Rather, in a discriminative auction other potential buyers provide competition that maintains the
price at a higher level than that which would be predicted by bilateral negotiation. The properties
of a discriminative auction are well understood, and this auction provides a reasonable
approximation of the real estate market under the special circumstances where homes near a site
are stigmatized.
As previously discussed, sellers in our model may have little or no value for the homes
they are selling since they shun the site. Thus, any price they can get for the home is acceptable.
This corresponds to an auction situation where buyers bid on H homes put up for sale, and the H
bidders with highest bids obtain the homes for the prices bid. Figure 10.2 shows this market in
the context of total demand where all homes in a neighborhood are potentially up for sale. Note
that the bids in a discriminative auction (shown as the lower step function) fall below the true
values (uppeif step function). Note also, that compared to the price that would be obtained in a
uniform price auction giving a price, Pe, in a discriminative auction there is a distribution of bids
and sale prices around the equilibrium price, since bpyers pay accepted bid prices. In a
discriminative auction, it is well known that if buyers are risk neutral, the average of the accepted
bids will equal the uniform price, so revenue neutrality exists in theory between uniform price
and discriminative auctions. Note also that risk aversion will increase bids in a discriminative
auction and bring them closer to true values because buyers trade off the gain in consumer
surplus of a lower accepted bid against the reduced probability of having their lower bid
-------
217
accepted. The lower bid curve shown in Figure 10.2 assumes risk neutrality and plausibly
provides a lower bound for bids in a real estate market.
Figure 10.2 Discriminative Auction Market
X ^
l_
•->••• Id |
With these concepts in mind, we can then turn to the hedonic model used to estimate
property values at each of our study sites described in Chapters 8 and 9. The hedonic model
estimated to explain property values uses a logarithmic specification and takes the form:
(10.1)
SPR1CE, = pniST*>'eb*A'Teb's"ebiD'7e£"
Here, Pt is an area-wide price index for owner-occupied housing in year t, DISTit is the distance
of each dwelling from the Superfund site in question. The coefficient associated with this
variable will be allowed to differ across years by interacting the constant distance measure with
yearly dummy variables. The vector A,r is property attributes and Slt is a vector of (interpolated)
time-varying characteristics of the Census tract in which the dwelling is located, and D,T is a
vector of the logarithms of the distances from the dwelling to a potentially relevant set of other
spatially differentiated local amenities or disamenities, calculated at time T, the end of the
sample period, rather than contemporaneously.
-------
218
Taking the logarithms of both sides of the equation yields a version of this model that is
appropriate for estimation:
(10.2)
LSPRICE,, =
b4tDiT
e
where LSPRICEit denotes the logarithm of the observed selling price, ln/>, will be captured as an
intercept for the first year in the sample and a set of intercept shifters activated by year dummy
variables. The variables of key interest are iheLDISTit, which consist of a vector of logged
distances from the dwelling to the Superfund site interacted with yearly dummies in order to
permit year-varying elasticities of housing prices wjth respect to distance to the site. Geographic
Information Systems techniques were used to measure distances from the homes to the closest
Superfund site in the specific year, t, that the sales p'rice was observed and the distance to other
local amenities or disamenities as they existed in year T.
As discussed previously, n obvious disadvantage of our sample described in previous
chapters is that in all of our data sets we only observe selling prices for the most recent sale of a
house. If a house is in an area where turnover is high, there will be more recent sales and fewer
earlier sales. For analytical purposes, it would be preferable to have data on all sales in all years
and selling price in those years, but such data do nojt exist. Data could be purchased from
Experian every year, if a future study could be antidipated, but retrospectively, the data are not
available. The data are collected primarily for current marketing purposes and records are
updated without saving their previous values. Historical modeling is not a use anticipated by the
providers of the data Consequently, there may be some systematic sampling. We observe earlier
transactions prices only for houses which are still occupied by the owners who purchased them at
that earlier date. We do not observe many early transactions prices for houses in neighborhoods
where there has been a lot of turnover. It must be a maintained hypothesis that rates of turnover
are uncorrelatted with identification and cleanup of Superfund sites. This may be a strenuous
assumption, but there are few alternatives. So it will be necessary to speculate upon the types of
biases this non-random selection is likely to produce in the effects of distance from a Superfund
site on housing transactions prices.
-------
219
However, a distinct advantage exists of only having one observation for each home in the
sample. By only having one observation per house and controlling for area-wide price index with
dummy variables, we ensure that each observation is independent. Therefore, the coefficient bu
(the effect of distance from the Superfund site on property values) can be observed over time by
looking at the hedonic estimates for each year over the 20-30 years of observations that have
been obtained for each of the sites. To dampen noise, we average bu the coefficients over three-
year intervals. To get time trends in property values as affected by the site, we normalize both by
the initial three-year period property value effect, t=0, and by distance. Thus, we ask the
question, at a minimum distance from the site, DISTmin, how do property values compare to price
at distance DISTm!ai (the boundary of the available data), which was chosen to be sufficiently far
away such that no effects of the site should be present, and to the magnitude of this effect in the
initial period. The relative property value effect, normalized by base period and by property
values at a large distance is defined as
(10.3)
Thus, the index for each site starts at 1.0 (or 100% in the figures below) and either
decreases or increases in successive three-year periods from this value. Table 10.1 presents the
results for each of the case studies.
As can be seen in the Figures 10,3, 10.4 and 10.5 presented below, relative property
values of the three metropolitan case studies (Oil in Los Angeles, Industri-Plex and Wells G&H
in Wobum, and Montclair, New Jersey) tend to follow an overall declining trend consistent with
the notion of progressive stigmatization of the site as suggested by arguments from psychology.
This result is in contrast to a number of earlier studies that examined property values over shorter
time periods (Carroll et al., 1996; Kiel, 1995; Kiel andZabel, 2001).
-------
220
Table 10.1 Distance Coefficients
Oil Landfill
Montclair
(Outside)
Montclair
(Inside)
Woburn
Eagle
Time period
1970-1972
1973-1975
1976-1978
1979-1981
1982-1984
1985-1987
1988-1990
1991-1993
1994-1996
1997-1999
1987-1989
1990-1992
1993-1995
1996-1997
1987-1989
1990-1992
1993-1995
1996-1997
1978-1979
1980-1982
1983-1985
1986-1988
1989-1991
1992-1994
1995-1997
1976-1982
1983-1988
1989-1994
1995-1999
Alvg. Distance
Coefficient
-0.133
-0.136
: -0.086
-0.099
-0.015
1 -0.039
0.013
0.015
0.015
0.027
-0.022
0.009
0.031
0.064
'• 0.102
0.174
0.094
• 0.191
-0.166
-0.115
-0.154
-0.157
-0.134
-0.111
-0.106
-0.814
2.134
4.815
1.966
Normalized
Value
100.00%
100.79%
$6.25%
£9.65%
68.94%
74.28%
63.03%
62.65%
62.65%
60.46%
100.00%
90.65%
84.81%
76.51%
100.0%
92.9%
100.8%
91.1%
130.00%
87.96%
97.01%
97.85%
92.35%
87.12%
86.04%
100.00%
83.88%
71.48%
84.72%
-------
221
Figure 10.3 Relative Property Value over
Time for OD Landfill, California
a,
I
1
Figure 10.4 Relative Property Value over Time
for Montctair, New Jersey (outside of area)
120% T N...A ,.
IKS Prttixti
1970- 1873. 1S79- 1979- 1982-
1972 1975 1978 1981 19M
1905-
1957
198S-
1990
1991-
1993
1997-
1999
Figure 10.5 Relative Property Value over Time for
Woburn, Massachusetts
C.
?
a.
5
I
f
1971-1(79 1900-1982 1983-1995 1989-1988
I992-1W4 1M6-1987
-------
222
What expjains the long term downward trends observed in relative property values shown
in Figures 10,3-10.5? If the trend is driven by f, theifraction of home owners and potential buyers
who shun homes near the site, a model of the determination of f over time is needed. From the
discussion of the psychology of risk perception and stigma, the determination of the fraction of
shunners will be driven by media attention and perceptual cues resulting from activity at the site,
which are in turn driven by "events" such as EPA ajinouncements, discovery, NPL-listing, and
cleanup. Thus, it is plausible that the percentage change between periods in the fraction of the
population who shun the site is a linear function of events of type j occurring during the prior
interval, characterized by the discrete dummy variable (or index summarizing a number of
dummy variables), Ej, t-i, thus
(10.4)
So, in a period with no events, Ejit.i= 0 Vj, we hypothesize that a is negative and f will decline,
thereby raising home values, because some people who know about the site will leave the area
(perhaps because of job opportunities elsewhere) anld some new potential buyers will move into
the area who will have no awareness of the site. Other events, such as cleanup activities, might,
(a) raise awareness and thereby increase the fraction of the population who shun the site, or
alternatively,, (b) reduce the fraction of shunners by convincing people who know about the site
that it is now safe. This latter possibility is unlikely in that the notion that, "once contaminated,
always contaminated" is part of the psychology of stigmatization. Note, also, that changes in
perceived risk for those who may not shun the site >yill likely follow a similar model.
There is no available data on f, so the model specified above cannot be estimated directly.
However, if one assumes a constant elasticity of demand, ti<0, and risk neutrality, a simple
transformation exists between ft and Rt as defined above: /, = 1 - R~*. Thus, the equation
describing movement in ft can be rewritten as:
(10.5)
-------
223
To employ this transformation we need to know the relevant elasticities of demand that
depend. Since we do not have this information, we assume that the elasticities are all -1.0,
consistent with a linear approximation of the relationship between f and the change in R. over
time.
Table 10.2 Number and Description of Events
Event Type
EPA Action
State Government Action
Local Government
Action
Public Action
Potentially Responsible
Party Action
Remediation Action
EPA Announcement
Site Incident
TOTAL
Number of Events
on
11
6
10
2
7
6
12
5
Montclair
3
1
1
1
0
4
3
2
Wobum
14
4
0
9
0
3
8
12
59 15 ! 50
TOTAL
28
11
11
12
7
13
23
19
124
Table 10.2 presents a psychological model using the data shown in Figures 10.3,10.4,
and 10.5 of relative property values over time for the three metropolitan sites. Note, as
mentioned earlier, Eagle Mine was excluded from this analysis because the socio-demographic
information for the homes were unavailable. Since all of the home sale observations were
independent, a simple linear regression could be used with 18 observations of changes in relative
property value (R^ -R^}) over the three-year periods for the three sites. For Discovery, NPL
Listing, and the Beginning of Major Phases of Cleanup, dummy variables were used. The
variable "Events" was derived by summing the number of major announcements and actions
described in EPA published reports for the relevant three-year interval for each of the three case
-------
224
studies (Table 10,4). Note that such events will be highly correlated with and drive important
perceptual cues defined in Chapter 7 such as noise, odor, truck traffic, visible on site activity, and
media coverage. Events are defined as follows:
• EPA Action - Includes site investigations, orders, notifications/decisions,
remediation, legal actions, and regulations by the EPA.
• State Government Action - Includejs
remediation, lawsuits, reports, and
site investigations, orders, resolutions,
ations by state agencies.
rejuli
• Local Government Action - Includes site investigations, orders, resolutions,
remediation, lawsuits, reports, and regulations by local cities, county, and school
districts.
• Public Action - Include the creation of public interest groups, major meetings
and protests, lawsuits by the residents, and the hiring of technical advisors for the
community.
• i Potentially Responsible Parties Action - Include site operation and closure,
committees formed, and lawsuits by PRPs.
• Remediation Action - Includes containment of contaminations, remediation
efforts and site improvements.
• EPA Announcement - Includes official Consent Decrees, Record of Decisions
(RODs), and announcements of settlements with PRPs.
Site Incident - Includes general site facts, reports and studies regarding the
contaminants and occurrences at the site.
-------
225
The analysis across the three sites shows that discovery, cleanup itself, and the number of
events all negatively affect property values by drawing attention to the site and possibly
increasing the number of owners and potential buyers who shun the site thereafter (Table 10.3).
Thus, the effect of any events, publicity or site information, good or bad, appears to increase the
fraction of the current home owners and potential buyers that stigmatize and consequently shun
the communities neighboring the sites. In other words, at least within the observed period of the
studies, all news is bad news and causes relatively permanent property value losses as an
increasing fraction of original owners leave and more potential buyers shun the site. The only
good news in the study is that property values did significantly recover for a short period after
sites were listed on the NPL. But, it is likely that as soon as it was realized that EPA could not
immediately clean up the sites, the process of stigmatization began with consequent reduction in
property values. Given the small sample size, it is remarkable that all of these coefficients are
significant at better than the 1% level.
n _ n
Table 10.3 Psychological Model, Dependent Variable ' '-'
Model
(Constant)
Discovery
NPL -Listing
Clean-up Begins
Number of Events
B
0.078
-0.160
0.105
-0.096
-0.016
t
3.578
-4.493
4.097
-4.753
-6.156
P
0.003
0.001
0.001
0.001
0.000
N=18
R2 = 0.855
Rather than property losses reversing immediately once cleanup begins, we see no
permanent recovery in property values within the time period of our data and speculate that
recovery will only occur as the local population gradually moves away, events cease, and
perceptual cues and media attention disappear, so more buyers are uninformed. Note that
McClelland et al. (1990) found that most buyers were uninformed in spite of reporting
requirements. The positive intercept in the psychological model (significant at the 5% level)
-------
226
above. Second, the "inside" Montclair property valu
explanatory variable since the homes themselves are
indicates that property values will increase at a linear rate of about 12% every three-years if no
actions are taken and no news is generated by the site. Thus, at Oil one could expect a complete
recovery in about a decade if no news is generated from the site and recovery might occur in
about half that time for the other sites.
The sites excluded from the model are also of some independent interest. First, although
Eagle Mine (see Figure 10.7) has very different characteristics from the three sites discussed
above, it shows a similar pattern in that relative property' values decline for most of the period
analyzed. Given the small amount of data available along the Eagle River, we are forced to use
six-year rather than three-year periods for the analysis but do confirm the general pattern shown
e estimates do not use distance as an
within the Superfund site. Yearly dummy
variables averaged over the same three-year intervals used in the outside-Montclair analysis
show that, unsurprisingly, cleanup itself does have a positive impact on property values (Figure
10.6). Third, another interesting result in the property value studies is the effect of including
socio-demographic variables. As shown in Figure 10.8, these make a large difference in the
magnitude of property losses at the Woburn site. Negative socio-demographic trends, that may
be the result of the progressive stigmatization of the site, also take a substantial toll on property
values (that are not included in the psychological model), but possibly should be included in any
damage assessment. These results suggest a different trend than observed by Kiel and Zabe!
(2001) which did not account for these socio-demographic affects.
Since economic benefits are based on discounted present value, the benefits of delayed
cleanup for homes surrounding sites are likely to be negligible where cleanup takes 20 years and
another 5-10 years may be needed after cleanup is complete for property values to recover. The
principal policy conclusion becomes evident from the results of the psychological model which
suggest that the promise of a prompt cleanup raises property values, while an increase in the
number of evients that are the root causes of perceptual cues and media attention decreases
property valUes. Thus, an expedited cleanup should ^ccur as quickly as possible after a site has
been determined to be hazardous and this cleanup should be conducted in a way that does not
arouse excessive attention. Otherwise the neighborhoods surrounding the site will likely be
stigmatized resulting in quasi-permanent economic damages.
-------
227
Figure 10.6 Relative Property Value over Time for Figure 10.7 Relative Property Value over Time for
Montclair, New Jersey (inside of area) Eagle Mine, Colorado
s eo%
C.
I
I
60U
1863-1986 1989-18W
Figure 10.8 Relative Property Value over Time for Wobum,
Massachusetts with and without socio-demographic variables
I 40%
1983 NR kstng, twin
j WHS C 4 H F|-npn5,,j *fl.
1978-1979 1980-1092 1SB3-1B85 1988-1968 1868-1W1 1902-19M 1095-199?
-------
228
Using the history of the Oil and the corresponding events and dates in a simulation, the
potential benefits of these policies becomes evident (Figure 10.9)
Figure 10.9 Policy Simulations usifig the Oil Landfill History
120%
100%
> 80%
f
a.
I
6
60%
40%
20%
0%
-*-A: Baseline - 33 years
••*•• B: 33 years; 25% fewer events
••••• C: 24 years; 25% fewer events
--*-- D: 15 years; 25% fewer events
As shown in Table 10.4, this simulation considers four different scenarios and includes an
extrapolation of a recovery in property values after cleanup is complete where there are no
further events. Given the legislative history of Superfund, some of these scenarios are clearly
fanciful, but the results are nevertheless suggestive as to what potential benefits could be
obtained by expediting the cleanup process and reducing the number of events that drive
perceptual cues, media attention, and social amplification. These results support several of the
suggestions made by Kunreuther and Slovic (Chapter 21,2001) for reducing stigma. In
-------
229
particular, they suggest prevention of the occurrence of stigmatizing events and the reduction of
the number of stigmatizing messages and thus reducing social amplification.
Table 10.4 Cleanup Scenarios
Scenario A
Scenario B
Scenario C
Scenario D
Time
Horizon
33 years
33 years
24 years
1 5 years
Events
All
25%
Fewer
25%
Fewer
25%
Fewer
Discovery
1978
1978
1978
1978
NPL
Listing
1985
1985
1982
1979
Cleanup
time periods
1988-1990
& 1997-1999
1988-1990
& 1997-1999
1985-1987
& 1988- 1990
1979-1981
Recovery
time periods
2002-2005
2002-2005
1990-1995
1982-1987
Final % of
Original Value
64.5%
85.2%
95.6%
100.0%
Note that these results directly contrast with those of Gayer, Hamilton and Viscusi (2000)
and Gayer and Viscusi (2002) who argue that media attention supports learning that leads to a
lowering of public risk perceptions more consistent with scientific evidence for smaller sites. No
credible evidence supports a significant long-term health risk to residents living near Oil
(McClelland et at. 1990). Yet the actual property value losses are enormous. One difference is
that this study focuses on prominent sites while the two studies cited above focused on less
prominent sites. Note that most potential benefits from cleanup are likely to come from
prominent sites.
It is interesting to note that Carol Browner did in fact institute reforms to USEPA policy
in 1995 to at least partly attempt to avoid the pattern shown in this study. EPA began to work
with PRPs in an attempt to negotiate sufficient cleanup at potential Superfund sites to avoid
having sites listed on the NPL. These reforms may, in fact, have represented an optimal response
given the difficulty stigma presents for neighborhoods surrounding Superfund sites. It should
also be noted that the enormously costly process of litigation and delayed cleanup that has
occurred under the Superfund program has provided strong incentives for industry to avoid
creating new hazardous waste sites. However, for residents living near very large Superfund
sites, as they have often stated, the program has failed in spite of EPA's best efforts. In this
regard, it should be noted that when CERCLA was passed, little or none of the work in
-------
230
psychology necessary to understand the phenomena described here had been completed. In fact,
much of the relevant work was motivated by Superfund sites and other hazardous facilities.
This study raises several questions for future research. First, are smaller sites truly
different as the work by Gayer, Hamilton and Visciisi suggests? Second, although the
psychological model developed here is statistically significant, it is based on data from just three
sites. Additional work to incorporate both larger sites, as well as smaller sites, and additional
explanatory variables would be worthwhile in our judgment. Finally, more research to
understand and prevent stigmatization is warranted.
-------
231
Chapter 11
References
1. Achaiya, G., and L. L. Bennett. "Valuing open space and land-use patterns in urban
watersheds." Journal of Real Estate Finance and Economics 22: 221-237 (2001).
2. Adams, G. and R. Cantor. "Risk, Stigma, and Property Value-What are People Afraid
of?" In Risk, Media and Stigma - Understanding Public Challenges to Modern Science
and Technology, by J. Flynn, P. Slovic and H. Kunreuther, London, UK & Sterling, VA:
Earthscan Publications Ltd., 2001, 175-185.
3. Allen, John. Biosphere 2: The Human Experiment. New York: Viking Penguin, 1991.
4. Bailey, M. J., R. F. Muth, and H. O. Hourse. "A Regression Method for Real Estate Price
Index Construction."yowr«a/ of the American Statistical Association 58: 933-942 (1963).
5. Barnett, Harold C. Toxic Debts and the Super/and Dilemma. North Carolina: University
of North Carolina Press, 1994.
6. Bartik, T.J. "Measuring the Benefits of Amenity Improvements in Hedonic Price
Models." Land Economics 64 (2): 172-183 (1988).
7. Been, V. "Locally Undesirable Land Uses in Minority Neighborhoods - Disproportionate
Siting or Market Dynamics." Yale Law Journal 103: 1383-1422 (1994).
8. Been, V., and F. Gupta. "Coming to the nuisance or going to the barrios? A longitudinal
analysis of environmental justice claims." Ecology Law Quarterly 24: 1 -56 (1997).
9. Benson, E. D., et al. "Pricing residential amenities: The value of a view." Journal of Real
Estate Finance and Economics 16: 55-73(1998).
10. Boamet, M. G., and S. Chalermpong. "New highways, house prices, and urban
development: A case study of toll roads in Orange County, CA." Housing Policy Debate
12:575-605(2001).
11. Bowen, W. "An analytical review of environmental justice research: What do we really
know?" Environmental Management 29: 3-15 (2002).
12. Bowes, D. R, and K. R. Ihlanfeldt. "Identifying the impacts of rail transit stations on
residential property values. "Journal of Urban Economics 50: 1-25 (2001).
-------
232
13. Browner, Carol, Letter to The Honorable Thomas Bliley Chairman Committee on
Commerce. United States House of Representatives.
14. Bureau of National Affairs. "Remediation Complete at Times Beach: Site Helped Insure
Passage of Statute." Environmental Reporte^, July 11, 1997.
15. Carroll, T. M., et al. "The economic impact of a transient hazard on property values: The
1988 PEPCON explosion in Henderson, Nevada." Journal of Real Estate Finance and
Economics 13: 143-167 (1996).
16. Center for Hazardous Waste Management Goals and Indicators of Progress in
Superjund.. Chicago, 111.: Illinois Institute of Technology Research Unit, 1989.
17. Chakravarty, Subrata, "Tunnel Vision: A.H.j Robins thought it c ould jump into the
pharmaceutical big leagues overnight with tne Dalkon Shield and now it is paying
dearly." Forbes May 21,1984: 214 (1984).
18. Chapman, Duane. Energy Resources and Energy Corporations. Ithaca: Cornell
University Press, 1983.
19. Clauretie, T. M., and H. R. Neill. "Year-round school schedules and residential property
values." Journal of Real Estate Finance and Economics 20: 311-322 (2000).
20. Committee for the National Institute for the Environment. SuperfundReauthorization
Issues in the 105th Congress. Congressional Research Service, April 7,1997.
21. Congressional Research Service. EPA: FY1998 Budget. Issue Brief, March 5, 1997.
22. Congressional Research Service. Super fund Fact Book. Issue Brief, May 26,1997.
23. Cox, J. C., V. L. Smith, and J. M. Walker. "The Theory and Behavior of Multiple Unit
Discriminative Auction." Journal of Finance 39 (9): 983-1010 (1984).
24. Cushman, John Jr. "Gingrich, Like the President Calls for A "New Environmentalism'."
New York Times, 25. April 1996.
25. Cushman, John Jr. "Program to Clean Toxic Waste Sites is Left in Turmoil." New York
Times, 15. January 1996.
26. Dale, L., J.C. Murdoch, M.A. Thayer, and P.A. Waddell. "Do property values rebound
from environmental stigmas? Evidence from Dallas." Land Economics 75: 311-326
(1999).
27. Davis, Charles E. Politics of Hazardous Wqste. New Jersey: Prentice Hall, 1983.
-------
233
28. Delong, James V, Policy Analysis; Privatizing Superfund How to Clean Up Hazardous
Waste, Cato Institute, 1995.
29. Dewdney, A.K. Yes, We Have No Neutrons: An Eye-Opening Tour Thought the Twists
and Turns of Bad Science. New York: John Wiley & Sons, Inc. 1997.
30. Downs, Peter. "The EPA Fights Environmentalists." The Progressive 62 (9): 14 1998.
31. Environmental Protection Agency .1997 Supplementary Materials: National Priorities
List, Proposed Rule and Final Rule. EPA/9320.7-061. U.S Environmental Protection
Agency, 1997.
32. Environmental Protection Agency. Superjund Facts: The Program At Work. May 7,1997.
33. Environmental Protection Agency. Superfund Administrative Reforms Annual Report
Fiscal Year 1996. December 20,1996.
34. Environmental Protection Agency. The Clinton Administration's Superfund Legislative
Reform Principles. May 7, 1997.
35. Environmental Protection Agency. SuperfundReauthorization. EPA Hearings by the
House Transportation and Infrastructure Committee, Subcommittee on Water Resources
and Environment. March 12, 1997.
36. Espey, M., and H. Lopez. "The impact of airport noise and proximity on residential
property values." Growth and Change 31: 408-419 (2000).
37. Farber, S. "Undesirable facilities and property values: A summary of empirical studies."
Ecological Economics 24: 1-14 (1998).
38. Fischhoff, B. "Risk: A Guide to Controversy." In Appendix to National Research
Council. Improving Risk Communications. Washington, D.C.: National Academy Press,
1989,211-319.
39. Foster, K., D. Bernstein, and P. Huber. Phantom Risk: Scientific Interference and the
Law. Cambridge, Mass.: MIT Press, 1993.
40. Franklin, Ben. A. "Plan for a 5-Year Toxic Waste Program is Praised." New York Times,
2. August 1986.
41. GAO. Superfimd Program: Current Status and Future Fiscal Challenges. GAO-03-850.
31, July 31 2003.
42. GAO. Superfund: Information on EPA 'a Administrative Reforms. Washington D.C., May
30, 1997.
-------
234
43. GAO. Super-fund: Time to Assess and Champ Hazardous Waste, Sites Exceed Program
Goals. Washington D.C., February 13, 1997.
44. GAO. Superfund Legal Expenses for Cleanup Related Activities of Major US
Corporations. Washington D.C., 1994.
45. Gayer, T. "Neighborhood Demographics ana the Distribution of Hazardous Waste Risks:
An Instrumental Variables Estimation." Journal of Regulatory Economics 17: 131-155
(2000).
46. Gayer, T., J.T. Hamilton, and W.K. Viscusi. "Private Values of Risk Tradeoffs at
Superfund Sites: Housing Market Evidence on Learning about Risk." The Review of
Economics and Statistics 82 (3): 439-451 (2000).
i !
47. Gayer, T., J. T. Hamilton, and W. K. Viscusi. "The Market Value of Reducing Cancer
Risk: Hedonic Housing Prices with Changing Information." Southern Economic Journal
69: 266-289 (2002).
48. Gayer, T., and W. K. Viscusi. "Housing Price Responses to Newspaper Publicity of
Hazardous Waste Sites." Resource and Energy Economics 24: 33-51 (2002).
i
49. Geoghegan, J. "The Value of Open Spaces in Residential Land Use." Land Use Policy 19:
91-98 (2002).
50. Gibbs, Lois M. Dying from Dioxin: A Citizen's Guide to Reclaiming Our Health and
Rebuilding Democracy. Boston: South End Press, 1995.
51. Gordon, Meryl. "A Cash Settlement, but No jApology." New York Times, 20. February
1999.'
52. Graham, J. D., et al. "Who lives near coke plants and oil refineries? An exploration of the
environmental inequity hypothesis." Risk Analysis 19: 171-186 (1999).
53. Hamilton, J.T. and W. Kip Viscusi. Calculating Risks: The Spatial and Political
Dimensions of Hazardous Waste Policy. Cambridge, Massachusetts: MIT Press, 1999.
54. Harris, J.D. "Property Values, Stigma, & Superfund." Working Paper for the US EPA.
http://ww\v.epa.gov/superfund/programs/recycle/property.htm. (2004).
55. Harrison, D. and J. Stock. Hedonic Housing Values, Local Public Goods, and the Benefits
of Hazardous Waste Cleanup. Discussion Paper Series. Energy and Environmental Policy
Center; Harvard University, 1984.
-------
235
56. Hazarika, Sanjoy. "Gas Leak in India Kills at Least 410 in City of Bhopal." New York
Times, 3. December 1984.
57. Hicks, Karen. Surviving the Dalkon Shield IUD, Women V. The Pharmaceutical Industry.
New York: Teacher's College Press, 1994.
58. Hite, D. "Information and bargaining in markets for environmental quality." Land
Economics 74: 303-316 (1998).
59. Hogan, Dan. "A new Life for Biosphere 2." Current Science 82 (12): 8-9 (1997).
60. Howard, Phillip K. The Death of Common Sense How Law is Suffocating America. New
York: Random House, 1994.
61. Hewlett, D. and R. Tyson, "Toxicity of Times Beach No Longer in Doubt." USA Today,
13. September 1994.
62. Kasperson, R., O. Renn, P. Siovic, H. Brown, J. Emel, R. Goble, J. Kasperson, and S.
Ratick. "The Social Amplification of Risk: A Conceptual Framework." Risk Analysis 8
(2): 177-187(1988).
63. Ketkar, K. "Hazardous waste sites and property values in the state of New Jersey."
Applied Economics 24: 647-659 (1992).
64. Kiel, K. A. "Measuring the Impact of the Discovery and Cleaning of Identified
Hazardous-Waste Sites On House Values." Land Economics 71: 428-435 (1995).
65. Kiel, K. A., and K. T. McClain. "House Prices During Siting Decision Stages - the Case
of an Incinerator From Rumor Through Operation." Journal of Environmental Economics
and Management 28: 241-255 (1995).
66. Kiel, K.A., and J. Zabel. "Estimating the economic benefits of cleaning up Superfund
sites: The case of Woburn, Massachusetts." Journal of Real Estate Finance and
Economics 22,163-184 (2001).
67. Kohlhase, J. "The Impact of Toxic Waste Sites on Housing Values." Journal of Urban
Economics 30: 1-26(1991).
68. Kolata, Gina. "Love Canal: False Alarm Cause By Botched Study." Science, Volume 208,
June 30,1980.
69. Kunreuther, H. and P. Siovic. "Coping with Stigma: Challenges and Opportunities." In
Risk, Media and Stigma - Understanding Public Challenges to Modern Science and
Technology, by J. Flynn, P. Siovic and H. Kunreuther, London, UK & Sterling, VA:
Earthscan Publications Ltd., 2001,331-352.
-------
236
70. Liebarman, Adam. Facts Versus Fears: A
Review of the Greatest Unfounded Health
Scares of Recent Times. New York: Americin Council on Science and Health, September
1997.
71. Liu, F. "Dynamics and causation of environmental equity, locally unwanted land uses,
and neighborhood changes." Environmental Management 21: 643-656 (1997).
72. Mahan, B. L., S. Polasky, and R. M. Adams. "Valuing urban wetlands: A property price
approach." Land Economics 76: 100-113 (2000).
73. Maztmanian Daniel and Morell, David. Beyond Superfund America's Toxics Policy for the
199Q*&. Colorado: Westview Press, 1992.
74. McClelland, G., W. Schulze, and B. Kurd. "iThe Effect of Risk Beliefs on Property
Values: A Case Study of a Hazardous Waste Site." Risk Analysis 10: 485-497 (1990).
75. McCluskey, J. J., and G. C. Rausser. "Estimation of perceived risk and its effect on
property values." Land Economics 77: 42-55 (2001).
76. Mendelsohn, R., D. Hellerstein, M. Huguenin, R. Unsworth, and R. Brazee. "Measuring
Hazardous Waste Damages with Panel Models." Journal of Environmental Economic and
Management 22: 259-271 (1992).
77. Michaels, G. and V. Kerry Smith. "Market Segmentation and Valuing Amenities with
Hedonic Models: The Case of Hazardous Waste Sites. "Journal of Urban Economics 28:
223-242(1990).
78. Milstein, Michael. "Breaking Up the Biosphere." Earth 1 (2): 31 (1998).
79. Mintz, Morton. At Any Cost: Corporate Greed, Women, and the Dalkon Shield. New
York: Pantheon Books, 1985.
80. Mitchell, Alison. "Clinton Asks Tax Breaks For Toxic Waste Cleanup." New York Times,
12. March 1996. !
81. Morris, Betsy. "Monsanto Unit Stops Marketing Its lUDs in U.S." The Wall Street
Journal, 3. February 1986.
82. Moss, T. H. and D. L. Sills. The Three Mile Island Nuclear Accident: Lessons and
Implications. New York: The New York Academy of Sciences, 1981.
83. Nakamura, R. T. and W. C. Church. Taming,Regulation: Superfund and the Challenge of
Regulatory Reform. Washington, D.C.: The ferookings Institution, 2003.
-------
237
84. Napoli, Maiyann. "Look Back in Anger: The DBS and Dalkon Shield Scandals."Ms.: 40,
May/June 1996.
85. National Advisory Council for Environmental Policy and Technology. Final Report of the
Super fund Subcommittee ofNACEPT. April 12, 2004.
86. New York Times. "US Writes Off Cleanup Costs of Toxic Sites." June 21, 1993.
87. New York Times. "Cleanup Gets Little of Superfund Settlements." April 26,1992.
88. New York Times. "Toxic Site Cleanup Reported Lagging." September 10,1989.
89. New York Times. "The Superfund: Locked in a Lovefest" July 19,1986.
90. New York Times. "Senate Refuses to Cut Toxic Waste Cleanup." Sept 21,1985.
91. New York Times. "EPA Halts Work On Toxic Cleanups." August 17, 1985.
92. Office of Technological Assessment. Superfund Strategy. Washington, D.C., 1985.
93. Office of Technological Assessment. Background and Issues Relating to the
Reauthorization of Superfund: Scheduled for Hearings before the Committee of Finance
on September 19 and 25, 1984. Staff of the Joint Committee on Taxation. Washington,
D.C., 1984.
94. Quint, Michael. "A Superfund Plan Divides the Insurance Industry." New York Times, 10.
January 1994.
95, Poor, P. J., et al. "Objective versus subjective measures of water clarity in hedonic
property value models." Land Economics 77: 482-493 (2001).
96, Preston, Jennifer. "Showcasing an Issue, Clinton Plans A Visit to A Bergen County
Dumpsite Today." New York Times, 11. March 1996.
97. Probst, K. N. and P. R. Portnoy. Assigning Liability for the Superfund Cleanup: An
Analysis of Policy Options. Washington, D.C.: Resources for the Future, 1992.
98. Ramanan, T. "The Bhopal Tragedy Revisited." Risk Management 39 (10): 62 (1992).
99. Ray, Dixie Lee. Environmental Overkill: Whatever Happened to Common Sense?
Washington, D.C.: Regnery Gateway, 1993.
100. Rosen. S. "Hedonic Prices and Implicit Markets: Product Differentiation in Pure
Competition." Journal of Political Economy 82: 34-55 (1974).
-------
238
101. Roziri, P. "Technological Stigma: Some Perspectives from the Study of Contagion." In
Risk, Media and Stigma - Understanding Public Challenges to Modern Science and
Technology, by J. Flynn, P. Slovic and H. Kunreuther, London, UK & Sterling, VA;
Earthscan Publications Ltd., 2001,31-40.
102. Rubiri, D. "How the Media Reported on Thrfce Mile Island and Chernobyl." Journal of
Communication 37 (3): 42-57 (1987). '
103. Russell, M., E. Colglazier, and M. English. Hazardous Waste Remediation: The Task
Ahead. Knoxville, Tennessee: Waste Management and Education Institute, University of
Tennessee, 1991.
104. Schnejider, Keith. "Times Beach Warning: Rjegrets a Decade Later." New York Times, 15.
August 1991.
105. Schneider, Keith. "US Backing Away from Saying Dioxin is a Deadly Peril." Wen- York
Times, 15. August 1991.
106. Shabeeoff, Philip. "Reagan Signs Renewal of Toxic Waste Project." New York Times, 16.
October 1986.
107. Shabecoff, Philip. "Lack of Money Might Halt Toxic Cleanup Soon, U.S. Officials Say."
New York Times, 26. January 1986.
108. Shabecoff, Philip. "House Passes $10 Billion Measure to Bolster Toxic Waste Program."
New York Times, 11. December 1985.
109. Shabecoff, Philip. "Senate Backs Bill For Cleaning Up Hazardous Dumps." New York
Times, 27. September 1985.
110. Shabecoff, Philip. "E.P.A. Says It Maps New Top Priority " New York Times, 21.
November 1983.
111. Silverstone, Sally. Eating In: From the Fields to the Kitchen in Biosphere 2. Oracle, AZ:
The Biosphere Press, 1993.
112. Slovic, P., M. Layman, N. Kraus, J. Flynn, J. Chalmers, and G. Gesell. "Perceived Risk,
Stigma, and Potential Economic Impacts of a High-Level Nuclear Waste Repository in
Nevada." Risk Analysis 11 (4): 683-696(1991).
1 i
113. Smith, V. K., C. Poulos, and H. Kim. "Treating open space as an urban amenity."
Resource and Energy Economics 24: 107-129 (2002).
114. Spalatro, F., and B. Provencher. "An analysis of minimum frontage zoning to preserve
lakefront amenities." Land Economics 77: 469-481 (2001).
-------
239
115. State Brief: "Love Canal Agency: Our Work Is Done." Ithaca Journal, December 9,
1999.
116. "Statement of Carol M. Browner Administrator of US Environmental Protection Agency"
before Committee on Environment and Public Works U.S. Senate. March 5, 1997.
117. Strand, J., and M. Vagnes. "The relationship between property values and railroad
proximity: a study based on hedonic prices and real estate brokers' appraisals."
Transportation 28: 137-156 (2001).
118. "Times Beach Toxic Site Cleaned." Chemical Market Reporter 252 (2): 44 (1997).
119. Tomkins, J., et al. "Noise versus access: The impact of an airport in an urban property
market." Urban Studies 35: 243-258 (1998).
120. Uzumeri, Mustafa and Charles Snyder. "Information Technology and Accelerated
Science: The Cases of the Pentium Flaw and the Dalkon Shield." California Management
Review 38 (2): 44 (1996).
121. Vadali, S. R., and C. Sohn. "Using a geographic information system to track changes in
spatially segregated location premiums - Alternative method for assessing residential land
use impact of transportation projects." In Transportation Research Record, by TRB,
National Research Council, Washington, D.C., 2001, 180-192.
122. Whelan, Elizabeth. Toxic Terror: The Truth Behind the Cancer Scares. Buffalo, New
York: Prometheus Books, 1992.
123. White, Jocelyn. "Superfund: Pouring Money Down A Hole." New York Times. 17. April
1992.
124. Wilkins, L. and P. Patterson. "Risk Analysis and the Construction of News.'Vowr«a/ of
Communication 37 (3): 80-92 (1987).
125. Wilkins, L. and P. Patterson. "Risky Business: Covering Slow-Onset Hazards as Rapidly
Developing News." Political Communication and Persuasion 1 (2): 11-23 (1990).
126. Zurilla, D. "Reflections of a Dalkon Shield Arbitrator." Dispute Resolution Journal,
February 1998.
-------
240
Website References: ,
127. http://caltech.edu/goosstein/fusion.html (September 9, 1999).
128. http://enviroweb.org/Three Mile Islanda/PressP2.html#cont (September 9,1999).
129. http://research.et.byu.edu/woburn/photos/plaices/ (March 17.2000).
130. http://usnews.com/usnews/issue/990329/29Tfhree Mile Island.htm (September 9,1999).
131. http://yosemite.epa.gov/r9/sfund/sphotos.nsf/57079n ed!9a0334882565cd006099c5/7f22
b96c0346d2eb882568fl00010bad?OpenDocument (October 5, 2000).
132. http://www.acsh.org/publications/reports/island_0399.html (September 20, 1999).
133. http://www-bcf.usc.edu/~meshkati/humanfattors.html (1991).
134. http://www.biospherics.org/.
135. http://www.cdc.gov/epo/mmwr/preview/mmwmtml/00000072.htm (May 06,1983).
136. http://www.crc.losrios.cc.ca.us/~hodappd/14/wastedoc.html.
137. http://www.crl.com.
138. http://www.consumerlaert.org/fumento/supdr.htm (September 14,1999).
139. http://www.coopamerica.org/isf/tobi/tobi-agenda/union-carbisehtni.
140. http://www.corpwatch. org/trac/bhopal/nightmare.htm.
141. http://www.corpwatch.org/trac/feature.india/profiles/bhopal.
i |
142. http://www.earthbase.org/home/timeline/1942/love/ (September 14,1999).
143. http://www.earthbase.org/home/timeline/1983/times_beach/.
144. http://www. earthbase. org/home/timeline/1984/bhopal/.
145. http://www.envirolink.org/enviroarts/arts_a(id_activism?loisGibbs.html (September, 14,
1999). '
146. http://www.enviroweb.org/issues/dioxin.
-------
241
147. http://www.enviroweb.org/Three Mile Islanda/Mmaple.html (September 9,1999).
148. http://www.enviroweb.org/Three Mile Islanda/wing4.htm (September 9,1999).
149. http://www.epa.gov/oerrpage/superfund/sites/npl/nar833.htm.
150. http://www.epa.gov/superfund/pics/.
151. http://www.epa.gov/region01/superfund/sites/wellsgh/photo.html.
152. http://www.epa.gov/region02/superrhd/site_sum/0201290c.htm (September 14,1999).
153. http://www.epa/gov/region07/programs/nplfacts/TOC.html#missouri.
154. http://www.epaosc.net/site_profile.asp?sitejd=0146.
155. http://www.home.stlnet.com/~cdstelzer/diox39.html.
156. http://www.msnbc.com/news/150998.asp#BODY (September 14, 1999).
157. http://www.nrc.gov/OPA/gmo/tip/tip9809.htm (September 9, 1999).
158. http://www.nj .gov/dep/srp/publications/site_status/l 998/html/98highli2.htm.
159. http://www.nytimes.com/reuters/news/news-utilities-nuclear.html (April 5, 2004).
160. http://www.onlineethics.0rg/text/cases/l.canal/history.html (September 14,1999).
161. http://www.opcw.nl/chemhaz/chemacci.htm.
162. http://ww\v,pathfinder.com/time/daily/0,2960,1098,00.html (September 14,1999).
163. http://www.pathfmder.com/time/daily/0,2960J1740,OO.html (September 14,1999).
164. http://www.prioritiesforhealth.com/1004/lovecanal.html (September 14,1999).
165. http://www.uwmc.uwc.edu/geography/350/cleanEPA.htm (June 18, 2004).
-------
-------
242
Appendix A - Montclair Radium Sites
Contents:
1 CRITERIA FOR EXCLUSION FROM RAW SAMPLE 244
2 ANNUAL COUNTS IN SAMPLE 244
3 DESCRIPTIVE STATISTICS 244
3.1 Housing prices and distances from site 244
3.2 Structural variables 246
3.3 Census tract attributes 247
3.4 Other distances 247
4 COLLINEARITIES 249
4.1 Time patterns in average site distances in sample 249
4.2 Time trend in average lot sizes 249
4.3 Distance to site vs. structural variables 249
4.4 Distance to site vs. Census tract attributes 250
4.5 Distance to site vs. other distances 250
5 TRENDS IN THE DISTANCE GRADIENT 251
5.1 Structural variables 251
5.1.1 Floors known? 251
5.1.2 Floors 252
5.1.3 Age known? 252
5.1.4 Age 252
5.1.5 Lotsize 253
5.2 Census tract attributes 253
5.2.1 Females 253
5.2.2 Whites 254
5.2.3 Blacks 255
5.2.4 Other ethnic groups 257
5.2.5 Children under 5 258
5.2.6 Persons between 5 and 29 259
5.2.7 Persons between 30 and 64 260
5.2.8 Persons 65 and older 261
5.2.9 Married heads of household 261
5.2.10 Male-headed of household with children 262
5.2.11 Female-headed households with children 263
5.2.12 Owner-occupancy 264
5.2.13 Renter-occupancy 265
5.2.14 Vacancy rates 266
6 COMPLETE REGRESSION RESULTS - No LOT SIZE INTERACTIONS 266
6.1 Just structural characteristics and year dummies 266
6.2 Including Census tract attributes 268
-------
243
6.3 Including other distances 269
6.4 Including both other distances and tract alkributes 271
7 COMPLETE REGRESSION RESULTS - Wi]ra LOT SIZE INTERACTIONS 273
7.1 Just structural characteristics and year dummies 273
7.2 Including Census tract attributes : 275
7.3 Including other distances , 277
7.4 Including both other distances and tract attributes 280
-------
244
Chapter 1 Criteria for exclusion from raw sample
For Montclair, we drop observations for which
• Sale price is greater than $2 million or missing
• House has more than four floors
• Land area is greater than 75,000 square feet or missing
• The street address given places them outside the census tracts in which they are supposed
to lie (assumed typographical errors in addresses)
• The address could not be geolocated using GIS software
• The sale year is prior to 1987 or after 1997
• An assessors estimate of the value of improvements is missing
Chapter 2 Annual counts in sample
year I Freq. Percent Cum.
— — +
87 I
83 I
89 I
90 1
91 1
92 1
93 I
94 I
95 1
96 1
97 I
490
793
814
997
1030
1152
1348
1505
1425
1665
826
4 .
6.
6.
7.
8.
9.
11.
12.
11.
13.
6.
10
68
82
43
63
65
29
60
93
94
92
4
10
17
25
33
43
54
67
79
S3
100
.10
.79
.60
.03
.66
.31
.60
.20
.14
.08
.00
Total I 11940 100.00
Chapter 3 Descriptive statistics
3.1 Housing prices and distances from site
Variable | Obs Mean Std. Dev. Min Max
dist I 11940 2.958726 1.653603 0 6.716527
sprice I 11940 188345.9 108008.5 2000 1250000
-------
245
1.36+06-
1.26+06-
1.16+06-
1.0e+06-
900000
800000
700000
600000
500000
400000
300000
200000
100000
. \
'-.>••: ••*'••
«^|tdk-.^
I :%42»«i
.....
\
i
3
dist
Distance from nearest Montclair site (km)
1.36+06-
1.2e+06-
1.16+06-
1.06+06-
900000 -
800000 -
70QQQQ-
600000 -
500000 -
400000 -
300000-
200000-
100000-
T
T
1 4
dist
Distance from nearest Montclair site (km)
T
-------
246
153063-
.9
1
u.
0-
2000 1.3e+06
SPRICE
Marginal distribution of house prices: Montclair
044066- n
o-
dist
671653
Marginal distribution of distances: Montclair
3.2 Structural variables
Variable
Obs
Mean
Std. Dev.
Min
Max
knowf Ir
f locrs
limpval
agekncwn
age20
age 30
age 4 0
age 50
age 60
age 70
sge70plus
lotsize
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
.9752931
1.448936
10.96357
.4040201
.0040201
.0117253
.0323293
.0330321
.0324958
.0932161
.1913735
1.001101
.33CKC01
.7510S19
.7406981
.4907219
.0632793
.1076512
.1768779
.1788584
.1773202
.2907472
.3933989
.907426
0
0
S. 006368
0
0
0
0
0
0
0
0
.0009748
1
4
13. 61376
1
1
1
1
1
1
1
1
8.068128
-------
3.3 Census tract attributes
Variable
Cbs
Mean
Std. Dev.
Min
3.4 Other distances
Max
pfemales
pblack
pother
page underS
page 5 29
page 65 up
pmarhh chd
pmhh chile)
pfhh child.
pvacantj
prenter oca
11940
11340
11940
11940
11940
11940
11940
11940
11940
11940
11940
.5312597
.1981348
.0856179
.0658442
.3131265
.1546853
.2536222
.0133316
.0617661
.0365
.3480293
.0137262
.2875085
.0566641
.0109725
.0482451
.0437867
.0729032
.0100036 :
.053228 ;
.0238177 :
.2366121
.47829S4
.0009133
.0120837
.0337553
.2308882
.0421896
.0747633
.0012361
.0167364
.0095438
.0275876
.600673
.9759917
.3221:607
.1124166
.4530005
.251818
.4308186
.05
-------
248
d_parks
d_mj water
d_colleges
d_cclubs
d_airports
d_newark_i
coefficient on this variable is a proximity effect in addition to
proximity from the nearest main roads, d njrds.
Distance from the nearest park. There are about 23 park areas that
could be the nearest park for houses in the sample.
Distance from the nearest body of water. There are no significant
bodies of water in the sample area. The Pompton River runs to the
north and to the east of the sample area, and the Cedar Grove and
Great Notch reservoirs lie to the north, in adjacent zip codes.
Distance from the nearest college or university. Upsala College and
Bloomfield College lie inside the sample area. Seton Hall lies in an
adjacent zip code to the south. NJ Institute of Technology and
Rutgers campuses lie to the southeast, Caldwell College to the west,
and Montclair State is adjacent to the northern perimeter of our
sample.
Distance to the nearest country club. There are ten country clubs
with significant amounts of land within the sample area, mostly in
the West Orange and Bloomfield zip codes.
Distance from the nearest airport. Essex County airport lies to the
northwest of the sample area It is a smaller regional airport with
two runways. Newark International airport about equidistant from
the center of our sample area, but to the southeast. None of the
main runways of either of these two airports would produce flight
paths that intersect the sample area.
Distance from Newark International Airport. This distance will be
correlated with the distance from other disamenities (or amenities)
associated with the location of the airport.
Variable
Obs
Mean
Std. Dev.
Min
Max
d summits
d school
d re t a i 1
d hospital
d church
d cemetery
d railroad
d njrds
d_i280
d gspkwy
d parks
d m j wa t e r
d colleges
d cclubs
d airports
d newark i
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
11940
3679.74
547.4059
8348.043
2269.5
1220 .129
1865.784
1122.386
158. 9385
3291 .807
2767.276
773.363
4557 .643
2197.54
1531.028
8096.859
11529.63
1617.395
391.6714
1929.518
1132.129
603.5879
1097.649
946.2423
140.2323
2224.808
2050.373
493.792
1739.214
1074.472
1095.087
1477.857
2697.796
281.3663
12.21375
3757.655
65.12099
10.8354
64.37774
. 3640983
.0163899
3.563274
6.2S9956
.0404199
394.7092
1
.0299353
3930.727
5343.567
7515.158
2300.027
11851.8
5554.716
3225.218
4720.175
4178.208
936.6699
8773.392
7662.895
2549.764
8069.31
5100.181
5288.419
10653.95
16899.11
-------
249
Chapter 4 Collinearities
i
4.1 Time patterns in average site distances in sample
Regression with
4.2
Idist
yearBS
year89
year90
year91
year92
year93
year94
year95
year96
year 97
cons
robust standard errors '
Coe f .
-. 207^507
-.2492569
-.1644464
-.1574261
-.1746706
-.2351696
-.2040609
-.1721154
-.1787973
-.1899618
1 .015294
Robust
Std. Err.
.0468216
.0474433
.0450486
.0448733
.043588
.0434252
.0420796
.0417705
.0405406
.0445582
.0339332
-4
-5
-3
-3
-4
-5
-4
-4
-4
-4
29
t .
.44
.25
.65
.51
.01
.42
.85
.12
.41
.26
.88
P>|t|
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
Numbe rot" obs
F( 10, 11929)
Prob > F
R-sqi^are i
Root MSE
[95% Conf.
-.2995286
-.3422.535
-.2527489
-. 245385
-.2601102
-.3202')01
-.2865-;33
-.2539925
-.2582(J36
-.2773(131
.9486813
11940
3.94
= 0.0000
= 0.0026
- .91947
Interval ]
-.1159729
-.1562604
-.0761439
-.0694672
-.0892309
-.1500491
-.1215779
-.0902334
-.0993311
-.1026205
1.081907
Time trend in average lot sizes
Regression with
4.3
lotsize
yearSS
year89
year 90
year91
year92
year93
year94
year95
year96
year97
cons
Distance
Regression with
robust standard errors
Coef .
.1030169
.1249482
.1082126
.1178118
.1766366
.1886342
.1402731
.123381
.1520162
.0667246
.8709363
to site vs.
Robust
Std. Err.
.0454197
.0478034
.0461802
.0424183
.0439886
.0419542
.0392404
.0397198
.0399157
.0440867
.0327428
structural
2
2
i
2
4
4
3
3
3
1
26
r
.27
.61
.34
.78 .
.02
.50 ,
.57
.11
-SI
.51
.60
p> 1 1 i
0
0
0
0
0
0
0
0
0
0
0
.023
.009
.019
.005
.000
.000
.000
.002
.000
.130
.000
Number of obs
F( 10, 11929)
Prob > F
R-squarec
Root MSE
;9b* Conf.
.013987
.0312457
.017691S
.034665
.0904117
.1063971
.0633555
.0455238
.0737749
-.01969S6
.8067:b5
11940
3.02
= 0.0008
= 0.0021
= .90684
Interval]
.1920469
.2186506
.1937333
.2009586
.2628615
.2708713
.2171908
.2012383
.2302574
.1531418
.9351176
variables
robust standard errors
Number of obs
F( 12, Il'i27)
Prob > F
R-squared
11940
73.27
= 0.0000
= 0.0614
-------
250
Root MSE
.89201
Idist
knowflr
floors
limpval
age known
age20
ageSO
age40
age50
age 60
ageTO
age70plus
lotsize
cons
-.
.
-.
.
-.
— .
4.4 Distance to
Coef.
5621402
1351645
1153885
0110498
1802057
0183933
3357685
4380861
5016029
1647473
0327322
0468758
8240943
site vs.
Robust
Std. Err.
.0368365
.0151117
.009619
.0803332
.124023
.1036304
.0903477
.0925754
.0866531
.0850763
.0824492
.0090222
.1079984
P> 1 1 I
15.
-8.
12.
-0 .
1.
-0.
3.
4 .
5.
1 .
-0.
5 .
-7.
26
34
30
14
45
18
72
73
79
94
40
20
63
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
a.
a.
000
000
000
891
146
859
000
000
000
053
691
000
000
[95% Conf.
.4899347
-.1647859
.0965336
-.1685159
-.0628995
-.2215257
.1586722
.2566231
.3317487
-.0020162
-.1943459
.0291908
-1.035789
Interval]
-.
.
.
6343457
1055431
1342433
1464163
4233109
.184739
5128647
.619549
.671457
3315108
1288816
0645608
6123997
Census tract attributes
Regression with robust standard errors
Idist
pfemales
pblack
pother
page_under5
page 5 29
page 65 up
pmarhh chd
prrihh child
pfhh child
pvacant
prenter occ
cons
4
-1
-3
1
"~ ^
-3
.
Coef.
.607242
.051875
.264204
.089476
.142977
.173556
7647791
-23.87896
10.31201
2
-.
4.5 Distance to
.370267
8083711
5929649
site vs.
Robust
Std. Err.
. 9148456
.0813007
.2138823
1.250295
.4941953
.4744836
.2641258
2.266044
.6031825
.3116062
.0882834
.4528341
5.
-12.
-15.
0.
-10.
-6.
2.
-10.
17.
7.
-9.
1.
t
04
94
26
87
41
69
90
54
10
61
16
29
p>
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
It 1
000
000
000
384
000
000
004
000
000
000
000
198
Number of obs
F< 11, 11928)
Prob > F
R-squared
Root MSE
[95% Conf.
2.813995
-1.211237
-3.683448
-1.361306
-6.11168
-4.103621
.2470496
-28.32078
9.129675
1.759468
-.9814209
-.3046637
=
=
=
=
=
11940
312.02
0.0000
0.1783
.8346
Interval ]
6
-
-2
3
-4
-2
1
.400488
.892512
.844959
.540259
.174273
.243491
.282509
-19.43715
11.49435
2
-.
1
.981065
6353214
.470594
other distances
Regression with robust standard errors
Idist
Coef.
Robust
Std. Err.
t P>l t 1
Number of obs
F( 16, 11923)
Prob > F
R-squared
Root MSE
!95* Conf.
=
=
=
=
11940
767.69
0.0000
0.5011
. 65044
Interval ]
-------
251
Id summits
Id school
Id retail
Id hospital
Id church
Id cemetery
ld_railroad
Id njrds
Id_i280
Id gspkwy
Id parks
Id mjwater
Id colleges
Id cclubs
Id airports
ld_newark i
cons
.5830631
.0040095
-.8080889
.1947804
-.1235253
.5175183
-.0574414
.000.3375
.1593406
.083741
.0631714
-.9691344
.2875994
-.2339437
-.2033038
-1.141372
17.13824
.0225069
.0093627
.0934561
.0111483
.0108221
.0170457
.007
.0046745
.OC833
.0099501
.0073117
.C27243
.0143331
.0124952
.0631761
.071517
1.199418
25
0
-8
17
-11
30
_Cj
0
19
c
£
-35
2C
-16
_3
-15
14
.91 1
.43
.65
.47 .
.41
.36
.63
.07 !
.03
.42
.64
.57 '
.07
.72
.22 •
.96
.29
0
0
0
0
0
0
0
0
0
c
0
0
0
0
0
0
0
.000
.668
.000
.000
.000
.000
.000
.942
.oco
.oco
.000
.000
.000
.000
.001
.000
.000
.5389459
-.014343
-.991278
.172328
-.1447384
.4841064
-.C811 525
-.0088.253
.1426124
.C642371
.0488392
-1.022535
.2595042
-.2534362
-.3271393
-1.231357
14.78'18
.6271802
.022362
-.6243958
.2166328
-.1023122
.5509313
-.05372C4
.0095002
.1752687
.1032449
.0775036
-.9157337
.3156945
-.2094511
-.0794682
-1.001187
19.48929
Chapter 5 Trends in the distance gradient
These models use individual houses as observations. We associate with each house the
proportion of each group in the Census tract that contains the house. The right-hand side
variables are the measured distance of the house itself from the Woburn site, a time trend,
starting at 1 in the first period of the data, and an interaction term between distance and time.
The simple trend picks up the trend over time in the concentration of the group in question
throughout the sample area. The "Idisw" variable, Distance to the nearer of the Wells G&H sites
or the Industri-Plex site, picks up any baseline distance gradient in the concentration of the group
in question as a function of distance from the nearest Superfund site. The key variable is the
interaction term, which tells how the distance gradient is shifting over time. If the distance
gradient is becoming either less positive or more negative, the concentration of the group in
question nearer the Superfund site is growing, relative to the concentration further away.
5.1 Structural variables
There are very few available structural variables for!each house. In lieu of a longer list of
structural variables, we employ the current assessed value of improvements as a proxy for
housing quality. It is not reasonable to assess how these values change with the time of sale of
the house. Given the paucity of data on housing attributes for the Montclair sample, we cannot
conclude much about the condition of the housing stock over time.
5.1.1 Floors known?
Regression with
1
knowflr I
inside I
robust standard errors
Robust
Coef. 3td. Err. t
-.0771242 .0547592 -1.41
Number oi obs
F( 5, 11934)
Prob > F
R-squared
Root: MSE
P>|t| [95% Conf.
0.159 -.1844592
1194C
41.1 =
= O.OOOC
= 0.022:
= .3267;
Interval
.030210E
-------
252
Idist I
trend I
insidey I
Idisty I
cons 1
5.1.2 Floors
Regression with
1
floors I
inside I
Idist I
trend I
insidey I
Idisty I
cons I
.0641301
-.001299
-.0065508
-.0032718
.8472986
.0090269 7.10
.0017916 -0.73
.0090075 -0.73
.0014203 -2.30
.0114769 73. S3
robust standard errors
Coef.
-.2215466
.0477309
.0020551
-.0016867
-.0028708
1.416048
Robust
Std. Err. t
.1054497 -2.10
.0190507 2.51
.0037851 0.54
.0172702 -0.10
.0029731 -0.97
.0243248 58.21
0.000 .0464359
0.468 -.0048108
0.467 -.0242069
0.021 -.0060559
0.000 .824802
Number of obs
F< 5, 11934)
Prob > F
R-squared
Root MSE
P>itl [95% Conf.
0.036 -.4282452
0.012 .0103885
0.587 -.0053644
0.922 -.0355391
0.334 -.0086987
0.000 1.368367
.0818243
.0022128
.0111053
-.0004877
.8697951
11940
7.90
= 0.0000
= 0.0040
= .74973
Interval )
-.0148481
.0850734
.0094746
.0321657
.002957
1.463729
5.1.3 Age known?
Regression with
age known I
inside I
Idist 1
trend I
insidey I
Idisty 1
cons I
5.1.4 Age
Regression with
age I
robust standard errors
Coef.
.0129883
.0898056
-.0010347
-.000424
-.0045979
.356786
Robust
Std. Err. t
.0659437 0.20
.01086 3.27
.0021129 -0.49
.010727 -0.04
.001698 -2.71
.013576 26.28
robust standard errors
Coef.
Robust
Std. Err. t
Number of obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
P>|t| [95% Conf.
0.844 -.1162722
0.000 .0685183
0.624 -.0051763
0.968 -.0214506
0.007 -.0079263
0.000 .3301749
Number of obs
F( 5, 4818)
Prob > F
R-squared
Root MSE
P>itl [95% Conf.
11940
40.13
= 0.0000
= 0.0155
= .48699
Interval ]
.1422488
.1110929
.0031069
.0206026
-.0012696
.3833972
4824
53.85
= 0.0000
= 0.0607
= 22.792
Interval)
-------
253
inside
insidey
Idist
Idisty
trend
cons
|
1
1
1
1
-13.64081
.1463102
-6.439359
.2577415
1.03418
67 .0986
6.437296
1 .092536
1 .063588
.1603724
.2088739
1 .389429
-2
0
-6
1
4
48
.12
.13
.05 i
.61
.95
.29
0
0
0
0
0
0
.034
.893
.000
.103
.000
.000
-26.26085
-1.99!J56
-3.524478
-.0566'517
. 624^92
54.37.163
-1.020772
2.23818
-4.35424
.5721446
1.443668
69.82251
5.1.5 Lotsfce
Regression with
lotsize
inside
Idist
trend
insidey
Idisty
cons
1
I
1
1
1
1
1
1
robust standard errors
Coef.
-.2707097
.0504139
.0039132
.015799
.0002315
.9397404
Robust
Std. Err.
.0593119
.017611
.0039927
.0112238
.0027266
.0261798
-4
2
0
1
0
35
t
.56
.86
.98 •
.41
.08
.90
P>lt|
C
C
0
C
C
0
.000
.004
.326
.159
.932
.000
Number o:.: obs
F( 5, 11934)
?rob > F
R-squared
Roo-: MSE
[95% Corif.
-.3869"06
.0158933
-.0038936
-.0062015
-.0051:32
.8384236
11940
24.43
=> 0.0000
= 0.0043
- .90567
Interval |
-.1544488
.0849344
.01172
.0377996
.0055762
.9910571
5.2 Census tract attributes
5.2.1 Females
Regression with
pfemales 1
Idist 1
trend 1
Idisty 1
inside. |
insidey I
cons 1
robust standard errors
Coef.
-.0004869
-.0001192
.3000813
-.0070261
.00078
.5320221
Robust
Std. Err.
.0004075
.0000789
.0000612
.0014803
.0002652
.00053
-1
-1
1
-4
2
1003
t.
.19
.51
.33
.75
.94
.79
P> 1 1 I
0.232
0.131
0.184
0.000
0.003
0.000
Number of obs
F( 5, 11934)
Prob > F
R-squarec.
Root MSE
, 95% Conf .
-.00126:57
-.0002739
-.0000387
-.0099278
.0002601
.5309E32
11940
6.34
= 0.0000
= 0.0016
= .01372
Interval 1
.0003119
.0000355
.0002012
-.0041244
.0012999
.533061
-------
254
- pf emales_87
- pf emales_94
- pfemales_91
- pf emales~97
.54-
.52-
dist
8.4
Fitted pfemales by distance from nearest Montclair site (km)
5.2.2 Whites
Regression with robust standard errors
pwhite I
Idist 1
Irond I
Idisty 1
inside I
insidey 1
cons I
Coef .
.0416303
-.0102532
.0028637
.1553385
-.0171576
.7254027
Robust
Std. Err.
.0061867
.0013299
.0009531
.0225094
.0044719
.0086937
6
-7
3
6
-3
83.
t
.73
.71
.00
.90
.34
.44
P>|t 1
0
0
0
0
0
0
.000
.000
.003
.000
.000
.000
Number of obs - 11940
F( 5, 11934) = 127.80
Prob > F - 0.0000
R-squared = 0.0462
Root MSE = .26929
[95% Conf.
.0295033
-.0128601
.0009954
.1112165
-.025923
.7083617
, Interval]
.0537572
-.0076463
.004732
.1994605
-.0083921
.7424438
-------
255
- p\Miite_87
~
-pvttiite 91
- pv*ite~97
.85-
I
.35-
T
dist
I8.4
Fitted pwhite by distance from nearest Montclair site (km)
5.2.3 Blacks
Regression with robust standard errors
Number of obs = 11940
F{ 5, Ij934) = 85.93
Prob > F = 0.0000
R-squarecl = 0.0246
Root MSE = .284
pblack 1
Idist I
trend I
Idisty 1
inside I
insidey I
cons' I
Coef .
-.0253767
.0066725
-.0033634
-.1937636
.0184994
.1989742
Robust
3td. Err.
.0064643
.0013978
.0009932
.0216261
.0044745
.0091774
_•}
4
-3
-8
4
21
-_
.93
.77
.39
.96
.13
.68
P>|t|
0.000
0.000
0.001
0.000
0.000
0.000
[95% Conf.
-.0380477
.0039226
-.0053102
-.2361543
.0097286
.180585
Interval ]
-.0127057
.0094124
-.0014167
-.1513729
.0272703
.2169634
-------
256
-pblack 87
-pblaclT94
-pblack 91
- pblaclf 97
.4-
g.
2
08-
T
dist
8.4
Fitted pblack by distance from nearest Montclair site (km)
We note that some census tracts in the Montclair area were predominantly white and others were
predominantly black in data interpolated for 1987. For 1997, it is clear that some of these
communities are becoming much more integrated, but others remain segretaged.
1 -
.9-
100
120140
160
TRACT
200
220
Percent black by census tract: Montclair 1987
-------
257
1 -
.9-
.8-
.7-
.6-
.5-
.4-
.3-
.2-
.1 -
. .V
c *
100
120
140
ib
TRACT
200 220
Percent black by census tract: Montclair 1997
5.2.4 Other ethnic groups
Regression with
pother |
Idist |
trend i
Idisty |
inside |
insidey |
cons' |
robust standard errors
Coef .
-.0150917
.0036705
.0003952
.0369728
-.0013925
.074682
Robust
Std. Err.
.001525
.0002875
.0002132
.0095714
.0013491
.0020093
-9
12
1
3
-1
37
t
.90
.77
.85
.96
.03
.17
P>|t|
• o.ooo
0.000
0.064
0.000
0.302
0.000
Number or obs
F{ 5, l:..934)
Prob > F
R-squared
Root; MSE
'95% Conf.
-,0180E:09
.0031C71
-.0000227
.0182112
-.00403-69
.0707435
11940
= 269.55
= 0.0000
- 0.0948
= .05392
Interval]
-.0121024
.004234
.0008132
.0557344
.0012519
.0786205
-------
258
- potfief_87
- pother_94
- pother_91
- pother_97
.14-
S
a
.04-
dist
Fitted pother by distance from nearest Montclair site (km)
5.2.5 Children under 5
Regression with
I
page underS I
Idist I
trend I
Jdisty I
inside I
insidey I
cons I
robust standard errors
Coef .
.0018929
.0008964
-.000146
-.0009087
-.0000739
.0593659
Robust
Std. Err.
.0002475
.0000495
.0000367
.001223
.0001942
.0003327
1
18
-3
-0
-0
179
t
.65
,12
.9B
.74
.38
.92
P> 1 1 I
0.000
0.000
0.000
0.458
0.703
0.000
Number of obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
[95% Conf.
.0014079
.0007994
-.0002179
-.003306
-.0004545
.0592136
11940
= 119.51
= 0.0000
= 0.0500
.0107
Interval |
.002378
.0009934
-.0000741
.0014887
.0003067
.060518
-------
259
- page_underS_67
- page_under5~94
- page_under5_91
- page_unde6_97
.07-
S
Q.
.05-
dist
8.4
itted page_under5 by distance from nearest Montclair site, (krr
5.2.6 Persons between 5 and 29
Regression with
I
page_5_29 I
Idist I
trend I
Idisty I
inside I
insidey |
cons I
robust standard errors
Coef .
-.0089331
-.0022751
-.0005365
-.002263
.0012068
.3359905
Robust
Std. Err.
.0011781
.0002302
.0001789
.0046821
.0007183
.0015184
-7
-9
-3
-0
1
221
i
.58
.88
.00
.48
.68
.27
P>|t|
, 0.000
0.000
0.003
0.629
0.093
0.000
Number ol obs
F( 5, 11934)
Prob > F
R-squarec
Root MSE
| 95% Conf .
-.0112425
-.0027262
-.0008E72
-.0114407
-.0002C13
.3330141
11940
= 187.85
= 0.0000
= 0.0795
.0463
Interval]
-.0066238
-.0018239
-.0001853
.0069148
.0026148
.3339669
-------
260
-page_5_29_B7
-page_5_29_94
-page_5 29 91
.36-
I
S
CL
.26-
dist
84
Fitted page_5_29 by distance from nearest Montclair site (km)
5.2.7 Persons between 30 and 64
Regression with
I
page 30 64 I
Idist I
t rend I
Idisty I
inside !
insidey I
cons I
robust standard errors
Coef .
.009835
.0031315
-.000451
-.0015717
-.0007805
.4425372
Robust
Std. Err.
.OOOS88
.00017
.0001355
.004051
.0006714
.0011031
11
18
_3
-0
-1
401
t
.08
.42
.33
.39
.16
.13
P> 1 1 I
0
0
0
0
0
0
.000
.000
.001
.698
.245
.000
Number of obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
[95% Conf.
.0080944
.0027983
-.0007166
-.0095122
-.0020965
.4403749
11940
= 242.79
= 0.0000
= 0.1034
= .03069
Interval]
.0115755
.0034647
-.0001854
.0063688
.0005354
.4446994
- pa9e_30_64_87
-paae_30_64_94
-page 30 64 91
.49-
T
dist
84
-"itted page_30_64 by distance from nearest Montclair site (km
-------
261
5.2.8 Persons 65 and older
Regression with robust standard errors Number o:.: obs = 11940
E'( 5, l:,934) = 35.27
?rob > F = 0.0000
R-squared - 0.0144
Roo-; MSE = .04348
1 Robust
page 65 up 1 Coet". Std. Err. t . P> 1 1 1 [95% Conf. Interval]
Idist 1 -.0016329 .0011066 -1.48 0.140 -.0038M9 .0005362
trend 1 -.001663 .0002146 -7.75 0.000 -.0020E36 -.0012424
Idisty 1 .001029 .0001688 6.10 0.000 . 0306982 .0013599
inside I .0032911 .0047883 0.69 0.492 -.0360948 .0126769
insidey I -.000403 .00071 -0.57 0.570 -.0317947 .0009887
cons 1 .1606655 .0014503 110.78 0.000 .1578126 .1635084
Proportion
'-» S
T
dist
8.4
ritted page_65_up by distance from nearest Montclairsite (km
5.2.9 Married heads of household
Regression with
I
pmarhh chd I
Idist I
trend I
Idisty I
inside I
insidey I
cons I
robust standard errors
Coef .
.0239196
.0015687
-.0005364
.0078763
.0002539
.2271155
Robust
Std. Err.
.0013363
.0003609
.0002992
.0075662
.0013898
.0021676
13
4
-1
1
0
104
t
.C3
.35
.79
.04
.18
.78
P>|t 1
0
0
0
3
0
3
.000
.000
.073
.298
.855
.000
Number of obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
[95% Cjnf.
.0203233
.0008613
--C01123
-.0069516
-.00247 33
.2223655
11940
= 148.78
- 0.0000
= 0.0703
= .07031
Interval ]
.027519
.0022761
.0000501
.0227072
.0029781
.2313644
-------
262
- pmarhh_chd_87
- pmarhh_chd_94
- pmarhh_chd_91
- pmarhh_chd_97
.15-
T
dist
8.4
:itted pmartih_chd by distance from nearest Montclair site (km
5.2.10 Male-headed of household with children
Regres
pmhh
i
sion with
1
child I
idist 1
t re nd I
Idisty 1
inside I
nsidey 1
cons 1
robust standard errors
Coef .
-.0028063
.0001108
-2.45e-06
-.0047882
.0002933
.0151105
Robust
Std. Err.
.0002924
.0000555
.000042
.0011136
.0001744
.0003973
-9,
2,
-0,
-4,
1 ,
39,
t
.60
.00
,06
.30
.63
.01
P> 1 1 1
0
0
0
0
0
0
.000
.046
.953
.000
.093
.000
Number of obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
[95% Conf.
-.0033795
2.07e-06
-.0000848
-.0069711
-.0000486
.0143512
11940
= 122.95
= 0.0000
= 0.0677
= .00966
Interval]
-.0022331
.0002196
.0000799
-.0026053
.0006352
.0158697
-------
263
- pmhh_child_87
- pmhh_child_94
- pmhh_ctli1d_91
- pmhh_chi!d_97
.03-
S
D_
0-
dist
84
ritted pmhh_child by distance from nearest Montclair site (km
5.2.11 Female-headed households with children
Regression with robust standard errors
Number of obs = 11940
F( 5, 11934) = 89.64
Prob > F = 0.0000
R-squareci = 0.0290
Root. MSE = .05246
1
pfhh child 1
Idist |
trend I
Idisty |
inside I
insidey |
consl |
Ccef .
-.0078861
-.0000687
-.0003317
-.0179512
.002167
.0704102
Robust
Std. Err.
.0011973
.0002414
.0001748
.005672
.0008539
.0016435
-6
-0
-1
-3
2
42
-
.59
.28
.90
.16
.54
.34
P>|t I
0.000
0.776
0.058
0.002
0.011
0.000
| SS^ Conf.
-.0102229
-.0005419
-.0006743
-.0290^93
.0004532
.0671687
Interval ]
-.0055392
.0004045
.0000109
-.0066331
.0038408
.0736317
-------
264
-pfhh_child_87
- pf hh_child_94
-pfhh child_91
-pfh(Tchild_97
.1 -
.04-.
dist
8.4
Fitted pfhh_child by distance from nearest Montclair site (km)
5.2.12 Owner-occupancy
Regression with robust standard errors
1
powner occ I
Idist 1
trend I
Idisty I
inside I
insidey I
cons I
Coef .
.0857379
.0050299
-.0002585
.0196921
-.0023418
.516553
Robust
Std. Err.
.0063471
.0012501
.0010134
.0333314
.0051767
.008328
12
4.
-0
0
-0
62
t
.52
.02
.26
.59
.45
.03
P>lt|
0.
0,
0.
0,
0,
0,
,000
.000
,799
.555
.651
,000
Numbe r o f obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
[95% Conf.
.0723165
.0025794
-.0022448
-.0457409
-.012489
.5002288
11940
= 207.33
= 0.0000
= 0.0975
= .23883
Interval]
.0991593
.0074803
.0017278
.0851252
.0078054
.5328771
-------
265
- powner_occ_87
- powner_occ_94
-powner occ_91
- powner~occ_97
.75-
disl
84
:itted powner_acc by distance from nearest Montclairsite (km
5.2.13 Renter-occupancy
Regression with
1
prenter occ 1
Idist 1
trend I
Idisty 1
inside I
insidey I
cons1 |
robust standard errors
Coef .
-.0327912
-.0042791
.0003412
-.0187592
.0025831
.4397649
Robust
Std. Err.
.0064599
.0011782
.0009554
.0315779
.OOJ9126
.0078506
-12
-3
0
-0
0
56
-_
.82
.63
.36
.59
.53
.02
P>|t 1
0.000
. 0.000
0.721
0.552
0.599
o.ooo
Number of obs
F( 5, 11934)
Prob > F
R-squared
Root MSE
195% tonf.
-.0954535
-.0065885
-.0015315
-.0806571
-.0070465
.4243764
11940
= 213.39
= 0.0000
= 0.1006
= .22444
Interval]
-.0701283
-.0019696
.002214
.0431387
.0122126
.4551535
-prenter occ_87
- prenteCocc_94
.7-
.2-
- prenter occ_91
- prenter~occ_97
T
dist
:itted prenter_occ by distance from nearest Montclairsite (km
-------
266
5.2.14 Vacancy rates
Regression with robust standard errors Number of obs = 11940
F( 5, 11934) - 84.03
Prob > F = 0.0000
R-squared = 0.0268
Root MSE = .0235
1 Robust
pvacant I Coef. Std. Err. t P> 1 t I 195% Conf. Interval]
Idist I -.0029473 .0005111 -5.77 C.OOO -.0039492 -.0019454
trend 1 -.0007481 .0000914 -8. IS 0.000 -.0009273 -.0005699
Idisty 1 -.0000826 .0000747 -1.11 0.269 -.0002291 .0000638
inside I -.0009375 .0034032 -0.28 0.783 -.0076083 .0057333
insidey I -.0002405 .0004838 -0.50 0.619 -.0011889 .0007079
cons I .0436624 .0006134 71.18 0.000 .04246 .0448648
.06-
c
<£
02-
— i — pvacant_87 pvacant_91
dist
84
Fitted pvacant by distance from nearest Montclairsite (km)
Chapter 6 Complete regression results - No lot size interactions
6.1 Just structural characteristics and year dummies
Regression with
1
Isprice 1
knowflr 1
robust standard
errors
Robust
Coef. Std. Err. t
-.2462875 .0210282 -11.71
Number of obs
F( 44, 11895)
Prob > F
R-squared
Root MSE
P>|t| [95* Conf.
0.000 -.2875061
11940
= 169.26
= 0.0000
= 0.4645
= .45092
Interval)
-.2050688
-------
267
!
floors .0681763 .0096046 7.10 '
lirapval 1 .5031077 .0086643 58.07
ageknown -.2537312 .0644329 -3.94
age20
age 30
age 40
age50
age60
age 70
age70plus
lotsize
inside87
insideSS
inside89
inside90
inside91
inside 92,
inside 93
inside94
inside95
inside 96
inside 97
Idis87
IdisSS
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
year88
year99
year90
year91
year92
year93
year 94
year95
year96
year97
.0065962 .0957836 0.37
.0495509 .0821105 0.60 1
.3073331 .0636384 4. 46
.2960261 .0666703 4.44
.3650611 .0664993 5.49
.2700766 .0657613 4.11 \
.2313334 .065123 3.55
.C738371 .0065051 12.13
-.C657024 .1603274 -0.41
-.1609055 .1739843 -0.92
-.0995028 .0447094 -2.23
-.0265615 .0470461 -0.56
-.5115417 .2246727 -2.28
-.0895896 .0845382 -1.06
-.238346 .1188784 -2.00
-.0617714 .0391401 -1.58
-.05903 .0443651 -1.33
-.1324379 .0683929 -1.94
-.1898961 .1928148 -0.98 '
-.0218912 .0260457 -0.34
.0396776 .0216239 1.83
.0087154 .009982 0.87
.034921 .0133111 2.62
.0453182 .0119244 3.90
.0368011 .0121178 3.04
.0340323 .0091016 3.74
.0592907 .0141601 4.19 '
.0550966 .0103257 5.34
.0495602 .0111017 4.46
.0953849 .0169856 5.62
.1179787 .0440608 2.68
.2189199 .0368393 5.94
.1273144 .0387591 3.28
.0349477 .0381631 0.92
.0380014 .0381022 1.00 •
.0406369 .036635 1.11
.0428305 .0331448 1.12
.0546726 .0366021 1.49
.0227944 .0375643 0.61
-.0727253 .0421038 -1.73
_cons 6.424755 .1016539 63.20
0.000
0.000
0.000
0.945
0.546
0.000
0.000
0.000
0.000
0.000
0 .000
0.682
0.355
0.026
0.572
0.023
0.289
0.045
0.115
0.183
0.053
0.325
0.401
0.067
0.383
0.009
0.000
0.002
0.000
0.000
0.000
0.000
0.000
0.007
o.ooc
0.001
0 .360
0.319
0.268
0.262
0.135
0.544
0.084
0.000
. 0493497
.4861:44
-.33001-03
-.1811553
-.111299
.1728456
.1653403
.2347116
.1411736
.1037217
.066136
-.3799702
-.5019432
-.1871405
-.1187795
-.95191-69
-.2552S-83
-.471302
-.13349-25
-.1460423
-.2664S91
-.5678446
-.0729451
-.0027C87
-.0108509
.0088292
.0219446
.0130482
.0161917
.0315345
.0348566
.027799
.0620503
.0316123
.1467088
.0513402
-.0398582
-.0366851
-.0312717
-.0319396
-.0170735
-.0508376
-.1552556
€.225497
.087003
.52C0911
-.1274321
.1943476
.2105008
.4419306
.4267113
.4954106
.3989795
.359035
.0916383
.2485654
.1801322
-.0118651
.0656565
-.0711465
.0761192
-.0053249
.0149497
.0278828
.0016233
.1880524
.0291627
.0820639
.0282817
.0610129
.06S6919
.060554
.0518729
.0870469
.0753367
.0713214
.1286795
.204345
.291131
.2032836
.1097536
.1126879
.1125454
.1176005
.1264187
.0964266
.009805
6.624013
Hypothesis
P-value Reject?
ofF-test
All structural attribute slopes simultaneously zero
All year-specific coefficient on INSIDE simultaneously zero
All year-specific coefficients on INSIDE the same
All year-specific slopes on LDIST simultaneously z$
All year-specific slope on LDIST the same
ro
0.0000
O.OCS4
0.5731
0.0000
0.0005
NO
-------
268
6.2 Including Census tract attributes
Regression with
1
Isprice |
kr.owflr 1
floors I
limpval 1
age known I
age20 I
age30 I
age 40 I
ageSO I
age 60 I
age 70 I
age70plu3 |
lotsize I
insid«j97 I
inside88 I
inside89 1
inside90 1
inside91 I
ins ids 92 I
inside93 !
inside94 !
inside95 I
insids96 I
inside97 t
Idis87 i
Idis88 I
Idis39 I
Idis90 1
Idis91 1
Idis92 I
Idis93 1
Idis94 I
Idis95 1
Idis96 I
Idis97 |
pfemales 1
pblack I
pother I
page under 5 1
page_5_29 1
page 65 up 1
pmarhh chd I
pmhh child 1
pfhh_child |
robust standard errors
Coef ,
-.1688066
.0784902
.4750606
-.1790821
-.036708
-.0301581
.1971785
.1909351
.2689345
.1945324
.1718968
.0650683
-.0615138
-.1262801
-.0722595
.0401207
-.4604077
-.0240486
-.1377516
-.0458004
-.0421126
-.1386125
-.2050098
-.0620205
.0141716
-.0233212
-.0007486
.0212482
.0087746
.010319
.0394947
.0315719
.0336331
.0899648
.2013268
.6770098
.5524539
2.670647
-1.220749
-.79264
1.341849
-.9407862
-3.579822
Robust
Std. Err.
.0213291
.0095354
.0124775
.0583229
.0862812
.0767926
.0624207
.0603937
.0599962
.0589539
.0581463
.0065243
.1626518
.18143
.0531107
.0527627
.219591
.0933429
.1221886
.0359594
.0442928
.0664374
.1963174
.0263429
.0201659
.0099723
.0123961
.0116264
.0109771
.00969
.0144622
.0100001
.010753
.0171062
.6286908
.0586664
.1250948
.3953561
.2574557
.3140823
.2016146
1.298238
.4373976
t
-7 .91
8.23
33.07
-3.07
-0.43
-0.39
3.16
3.16
4 .48
3.30
2.96
9.97
-0.38
-0.70
-1.36
0.76
-2.10
-0.29
-1.13
-1 .27
-0.95
-2.09
-1.10
-2.35
0.70
-2.34
-0.06
1.83
0.80
1.19
2.73
3.16
3.13
5.26
0.32
11.54
4 .42
2.98
-4.74
-2.52
6.66
-0.72
-8. 18
P>|t I
0.000
0.000
0.000
0.002
0.671
0.695
0.002
0.002
0.000
0.001
0 .003
0.000
0 .705
0.486
0.174
0.447
0.036
0.774
0.260
0.203
0.342
0.037
0.271
0.019
0.482
0.019
0.952
0.068
0.424
0.235
0.006
0.002
0.002
0.000
0.749
0.000
0.000
0.003
0.000
0.012
0.000
0.469
0.000
Number of obs = 11940
F( 55, 11884) = 161.68
Prob > F = 0.0000
R-squared = 0.4936
Root MSE = .4387
[95% Conf,
-.2106152
.0597992
.4506026
-.2934045
-.2058332
-.1806842
.0748236
.0725536
.1513322
.0789731
.0579205
.0522796
-.3803379
-.4819126
-.1763651
-.0633029
-.890842
-.1883944
-.3772613
-.1162847
-.1289338
-.2688407
-.5702223
-.1136569
-.0253568
-.0429685
-.0250469
-.0015415
-.0127423
-.0067147
.0111463
.0119701
.0125555
.0564338
-1.03101
.5620141
.3072476
.9156029
-1.725404
-1.408293
.9466511
-3.485546
-4.437193
. Interval]
-.126998
.0971813
.4995186
-.0647598
.1324172
.120368
.3195334
.3093165
.3865369
.3100917
.2858731
.0778571
.2573104
.2293524
.0318461
.1435443
-.0299734
.1402973
.1017581
.0246839
.0447086
-.0083843
.1602028
-.0103941
.0537001
-.0037739
.0235498
.0440379
.0302914
.027352S
.0678431
.0511737
.0547106
.1234958
1.433664
.7920055
.7976602
4.425692
-.7160935
-.1769874
1.737046
1.603973
-2.722451
-------
269
pvacant 1 .9293185 .26619'! 3.49 0.000
prenter occ 1 .1776243 .0565373 3.141 0.002
yeat-88 I .0904015 .0430697 2.10, 0.036
year89 I .1801503 .0364891 4.94 0.000
year90 I .0879317 .0382724 2.30 , 0.022
year91 1 -.0093839 .0380023 -0.25 0.305
year92 -.0222973 .0377707 -C.59 0.555
year93 -.0277533 .0370234 -C.75' 0.453
year94 -.0532313 .0388661 -1.37 0.171
year95 -.053087 .037926 -1.40' 0.162
year96 i -.1112112 .0393976 -2.83 C.005
year97 1 -.2126013 .043806 -4.85' 0.000
cons I 6.5S805 .3058625 21.57 0.000
Hypothesis
All structural attribute slopes simultaneously zero
All year-specific coefficient on INSIDE simultaneously zero
All year-specific coefficients on INSIDE the same
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All Census tract characteristic effects simultaneously zero
6.3 Including other distances
Regression with robust standard errors
Robust
Isprice Coef. Std. Err. t P> 1 t 1
knowflr -.2362626 .0227698 -10.38 0.000
floors .0433051 .0102289 4.23 0.000
limpval .4977374 .0104624 47.57 0.000
ageknown -.1995411 .0607206 -3.29 0.001
age20 -.0030888 .0903818 -0.03 . 0.973
age30 .0294952 .0763365 0.39 0.699
age40 .2355597 .0644895 3.65 0.000
ageSO .2290158 .0625756 3.66 0.000
age60 .3014699 .0623585 4.83 0.000
age70 .2375039 .0613852 3.87 0.000
age70plus .1937161 .0606292 3.20 0.001
lotsize .0688024 .0065148 10.56 0.000
.4075347 1.451102
.066704 .2885446
.C059778 .1748252
.1086257 .251675
.0129115 .1629519
-.0838746 .0651069
-.0963545 .051739
-.1003302 .0448137
-.1294.52 .0229527
-.1274381 .021254
-.1885.569 -.0340854
-.2984682 -.1267344
5.99:351 7.197591
P-value Reject?
ofF-test
c.oooo
0.1124 NO
0.4442 NO
0.0000
0.0000
0.0000
Number of obs = 11940
F( €0, 11879) = 143.60
Prot > F = 0.0000
R-squared = 0.4943
Root MSE = .43846
|95% Conif. Interval]
-.2808951 -.1916301
.023255 .0633553
.4772234 .5182453
-.3185635 -.0805188
-.1802519 .1740743
-.12013-39 .1791273
.1091437 .3619697
.1063574 .3516742
.1792359 .4237028
.1171739 .3578289
.0748'73 .3125593
.05603:!4 .0815723
-------
270
insideS? | -.090702
insideSS
inside89
inside 90
iriside91
inside92
inside93
inside94
inside95
inside96
inside97
Idis87
Idis88
Idis89
Idis90
Idis91
Idis92
idis93
Idis94
Idis95
Idis96
Idis97
Id summits
Id school
id_retail
Id hospital
Id church
Id cemetery
Id railroad
Id njrds
Id i280
Id gspkwy
Id parks
Id mjwa'er
Id colleges
Id cclubs
id airports
Id newark i
year88
year89
year90
year91
year92
year93
year94
year95
ye a r 9 6
year97
cons
-.2015083
-.1028902
-.0304311
-.5174714
-.101642
-.2407762
-.0393263
-.0660106
-.1309703
-.2459959
-.0316554
.0374025
-.0036091
.0271119
.0341177
.028539
.0274066
.0488147
.0451795
.0413617
.0838871
.0460893
.0119404
.6365502
.0692428
-.0020441
-.0242507
.0061729
.03C119
.012097
.1202201
-.0175926
.1307918
-.073431
.0075314
-.566893
.3909264
.1147391
,2172775
.108937
.0285168
.0230723
.0347034
.0390628
.0456108
.0165675
-.0642
-.1845058
.1621694
.1755523
.0494611
.0456801
.2262181
.0914085
.1189361
.045037
.0490507
.066828
.2036548
.0254472
.0212966
.010969
.0138659
.0122519
.0122753
.0095124
.0147332
.0110321
.0115783
.0168984
.0166768
.0068243
.080798
.0094311
.0086553
.0097184
.0067071
.0034949
.0073877
.0124874
.0046738
.0201436
.0117311
.005722
-0.56
-1.15
-2.08
-0.67
-2.29
-1.11
-2.02
-0 .87
-1.35
-1 .96
-1.21
-1.24
1.76
-0.33
1.96
2.78
2.32
2 .66
3.31
4 .10
3.57
4 .96
2.76
1.75
7.38
7.30
-0,24
-2.50
0.92
9.62
1.64
9.63
-3.76
6.49
-6.23
1.32
.0512933 -11.05
.0527823
.0433964
.0364684
.0385283
.0377344
.0374975
.0361729
.0376351
.0361571
.0371395
.0419882
.9263968
7.41
2.64
5.96
2.83
0.76
0.62
0.96
1.04
1.26
0.45
-1.53
-0.20
0.576
0.251
0.038
0.505
0.022
0.266
0.043
0.383
0.178
0.050
0.227
0.214
0.079
0.742
0.051
0.005
0.020
0.004
0.001
0.000
0.000
0.000
0.006
0.080
0.000
0.000
0.813
0.013
0.357
0.000
0.102
0.000
0.000
0.000
0.000
0.188
0.000
0.000
0.008
0.000
0.005
0.450
0.538
0.337
0.299
0.207
0.656
0.125
0.842
-.4085806
-.5456195
-.1998422
-.1199715
-.960896
-.2808177
-.4739103
-.1276061
-.1621579
-.2619646
-.6451926
-.081536
-.0043423
-.0251101
-.0000676
.010102
.0044774
.0087608
.0199352
.0235547
.0186664
.0507636
.0134
-.0014363
.4781729
.0506583
-.01901
-.0433004
-.0069741
.0232683
-.0023841
.0957426
-.026754
.0913071
-.0965238
-.0036846
-.6674362
.2874645
.029675
.1457935
.0334151
-.0454487
-.0504289
-.0362014
-.0347081
-.0252631
-.056232
-.1463077
-2.000395
.2271766
.1426028
-.0059383
.0591093
-.0740469
.0775336
-.007642
.0489536
.0301367
.000023
.1532009
.0182252
.0791473
.0178918
.0542913
.0581334
.0526006
.0460524
.0776943
.0668042
.064057
.1170107
.0787786
.0253171
.7949276
.0878273
.0149217
-.005201
.01932
.0369696
.0265781
.1446975
-.0084312
.1702765
-.0503382
.0187474
-.4663497
.4943884
.1998033
.2887614
.1844588
.1024824
.0965736
.1056083
.1128336
.1164846
.0893671
.0179077
1.631383
Hypothesis
P-value Reject?
ofF-test
All structural attribute slopes simultaneously
zero
0.0000
-------
271
All year-specific coefficient on INSIDE simultaneously zero
All year-specific coefficients on INSIDE the same
All year-specific slopes on LDIST simultaneously z^ro
All year-specific slope on LDIST the same
All other distance effects simultaneously zero ,
0.0209
0.4535 NO
0.0000
0.0003
0.0000
6.4 Including both other distances and tract attributes
Regression with robust standard errors Number of obs = 11940
F( 71, 11868) = 137.29
Prob > F = 0.0000
R-squared = 0.5124
Root MSE = .43073
Isprice
knowf Ir
floors
limpval
ageknowri
age 20
age30
age 40
age 50
age 60
age 70
age70plus
lotsiae
inside87
inside 8 8
inside89
inside90
inside 91
inside 92
inside93
inside94
inside95
inside96 •
inside97
Idis87
Idis88
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Coef .
-.1481448
.0408979
.4976241
-.1293435
-.0363002
-.0468389
.1572034
.1650789
.2469087
.1769108
.1420149
.0594497
-.0622082
- .1821917
-.0619429
.0103467
-.4785792
-.0525078
-.130911
-.040997
-.0592325
-.13893
-.2430417
-.0622194
.0220357
-.026132
-.OC106C4
.0190349
.0104912
.01 4 6"' 33
.04154
.C364873
.037389
.0906506
Robust
Std. Err.
.023507
.0102978
.013299
.0557219
.0836071
.0720979
.0592558
.0572808
.0568836
.0558059
.0548805
.0064778
.1630904
.1812667
.0564081
.0557159
.2246065
.0908439
.1219092
.0420785
.0521817
.0693742
.1980546
.0259862
.0201606
.0113241
.0134562
.0121289
.0117362
.0092452
.0151521
.0108252
.0114299
.0168807
-6
3
37
-2
-0
-0
2
2
4
3
2
9
-0
-1
-1
0
-2
-0
-1
-0
-1
-2
-1
-2
1
-2
-0
1
0
1
2
"5
3
5
t
.30
.97
.42
.32
.43
.65
.65
.38
.34
.17
.59
.19
.33
.01
.10
.19
.13
.58
.48
.97
.14
.00
.23
.39
.09
.31
,C8
.57
.89
.59
.74
.37
.27
.37
P>lt|
0.000
0.000
0.000
0.020
, 0.664
' 0.516
0.008
0.004
0.000
i 0.002
0.010
0.000
0.703
0.315
0.272
0.853
0.033
' 0.563
0.138
0.330
0.256
. 0.045
0.220
0.017
. 0.274
' 0.021
0.937
0.117
0.371
0.113
0.006
0.001
0.001
, 0.000
:95% C=nf
-.1942224
.0207124
.4715558
-.2385675
-.2001837
-.1881627
.0410524
.0527991
.1354075
.0675222
.0344432
.0467522
-.3818921
-.5375042
-.1725119
-.0983656
-.9188448
-.2305767
-.419873
-.12347?8
-.1615173
-.2749148
-.6312613
-.1131555
-.0174824
-.0483291
-.0274367
-.0047336
-.0125137
-.0034438
.0118333
.0152632
.0149814
.0575616
Interval]
-.1020672
.0610833
.5236924
-.0201194
.1275834
.0944849
.2733544
.2773587
.3584099
.2862995
.2495896
.0721472
.2574757
.1731207
.0486262
.1195589
-.0383136
.1255612
.0580511
.041483S
.0430522
-.0029451
.1451778
-.0112822
.0615538
-.003935
.0253159
.0428095
.0334962
.0327954
.0712407
.0577064
.0597935
.1237395
-------
272
ld_summits 1 .0419269 .0174385 2.40 0.016
Id school
Id retail
Id hospital
Id church
Id cemetery
Id railroad
Id njrds
ld_i2SO
Id qspkwy
Id parks
Id mjwater
Id colleges
Id cclubs
Id airports
Id newark i
pfsmales
pblack
pother
page underS
page 5 29
page 65 up
pmarhh chd
pmhh child
pfhh child
pvacant
prenter ccc
yearSS
yearS9
year90
year91
year 92
year93
year94
yearSS
year96
year97
cons
.0066978 .0069586 0.96 0.336
.2545124 .0881352 2.89 0.004
.0424212 .0098504 4.31 0.000
.0104202 .0089931 1.16 0.247
-.0192373 .0099644 -1.93 0.054
-.0016612 .0068 -0.24 0.807
.0270953 .0035152 7.71 0.000
-.005972 .0073289 -0.76 0.446
.0739805 .013462 5.50 0.000
-.0083094 .0046428 -1.79 0.074
.1070762 .0213203 5.02 0.000
-.0233721 .0118601 -1.97 0.049
.0099117 .0058941 1.69 0.093
-.4220757 .05471 -7.71 0.000
.1830013 .0576743 3.17 0.002
.454854 .6400132 0.71 0.477
.6873239 .0628156 10.94 0.000
.9665518 .1326575 7.29 0.000
1.77971 .9829002 1.81 3.070
-.8105554 .2867132 -2.83 0.005
-1.335181 .3309936 -4.03 0.000
1.075264 .243632 4.41 0.000
-1.687813 1.299352 -1.30 0.194
-3.47872 .4418469 -7.87 0.000
.4643934 .2714068 1.71 0.087
.1652722 .063607 2.60 0.009
.0862001 .042869 2.01 0.044
.1781398 .0365535 4.87 0.000
.0761685 .038544 1.98 0.048
-.0161497 .0378644 -0.43 0.670
-.0337583 .0377255 -0.89 0.371
-.0344591 .0370058 -0.93 0.352
-.061627 .0388062 -1.59 0.112
-.0677887 .0378429 -1.73 0.073
-.1212633 .0397723 -3.05 0.002
-.2137039 .0440602 -4.85 0.000
4.130283 1.093149 3.80 0.000
.0077445 .0761093
-.0069421 .0203377
.081753 .4272719
.0231128 .0617296
-.0072079 .0280482
-.038769 .0002945
-.0149903 .0116679
.020195 .0339756
-.021318 .009374
.0475928 .1003683
-.01741 .0007912
.065285 .1488674
-.0466198 -.0001244
-.0016418 .0214652
-.5293164 -.3148351
.0699507 .2960528
-.7996767 1.709385
.5641951 .3104527
.7065213 1.226582
-.1469349 3.706356
-1.37256 -.2485505
-1.983982 -.6863791
.597705 1.552822
-4.234756 .8531301
-4.344812 -2.612628
-.0676085 .9963952
.040592 .2899524
.0021699 .1702303
.106539 .2499406
.000616 .151721
-.0903701 .0580707
-.1077065 .0401899
-.1069965 .0380783
-.1376935 .0144395
-.141967 .0063897
-.1992235 -.0433032
-.3000692 -.1273387
1.997333 6.263234
Hypothesis
All structural attribute slopes simultaneously zero
All year-specific coefficient on INSIDE simultaneously zero
All year-specific coefficients on INSIDE the same
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All other distance effects simultaneously zero
All Census tract characteristic effects simultaneously zero
P-value Reject?
ofF-test
0.0000
0.1400 |s|Q
0.5290 NO
0.0000
0.0000
0.0000
0.0000
-------
273
Chapter 7 Complete regression results - With lot size interactions
7.1 Just structural characteristics and yt^ar dummies
Regression with robust standard errors
1
Isprice |
knowfir |
floors |
limpval 1
ageknown |
age20 |
age30 I
aqe40 |
age50 I
age 60 I
age70 :
ageTOplus ,
lotsize i
inside87 |
insideSS 1
insideS9 ;
inside 90. |
inside91 |
inside 92 I
inside93' |
inside94 1
inside95 |
inside 96 |
inside 97 |
vinsideB? |
vinsideSS |
vinside39 1
vinside90 I
vinside91 |
vinside92 I
vinside93 |
v inside 94 I
vinside95 I
vinside96 I
vinside97 |
Idis87 I
Idis88 I
Idis89 I
idis90 1
Idis91 |
Idis92 1
Cos f .
-.2461375
.0684279
.5027855
-.2661599
.0199907
.0603632
.3243322
.31029
.3760373
.23C8749
.2433283
.0925949
1.20434
-.2393118
-.1560025
.0324357
-.8025334
.1341343
-.4907818
.08437
-.0113645
.0253598
.1994603
-2.031502
.2353674
.0343223
-.0855838
.4007411
-.2877699
.2283016
-.2021044
-.0783487
-.1842632
-.5096444
-.0452699
.0292719
.0400567
.055396
.0662461
.0532283
Robust
Std. Err.
.0210599
.0095993
.0086359
.0645674
.0954586
.0815477
.0634936
.0669149
.0665557
.0658646
.0652726
.0036944
.7655295
.2491193
.071952
.0631234
.3186441
.120401
.2521055
.0802709
.1420805
.089303
.3223943
1.352263
.3331005
.0518359
.0713869
.192845
.2157495
.1382501
.097027
.2558219
.0995791
.5947695
.0339335
.0281842
.0157842
.0197471
.0183999
.0186971
t
-11.69
7.13
58.22
-4.12
0.21 ;
0.74
4 .74
4.64
5.65
4.26
3.73
10.65
1.57
-1.16
-2.17
0.51
-2.52
1.11
-1.95
1.05
-0.08
0.28
0.62
-1.50
0.71
1.53
-1.20
2.08
-1.33
1.65
-2.08
-0.31
-1.85
-0.86
-1.34
1 .04
2.54
2.81
3 .60
2 . 35
P>|t I
C.OOO
0 .000
C.OOO
0.000
0.834
0.459
0.000
0.000
0.000
0.000
0.000
0 .000
0.116
0.246
0.030
0.607
0.012
0.265
0.052
0.293
0.933
0.776
0.536
0.133
0.480
0.104
. 0.231
0.038
0.182
0.099
0.037
0.759
0.064
0.392
0.181
0.299
0.011
0.005
0.000
0.004
Number o:~. obs
F( 56, i:..373)
Prob > F
R-squared
Roo- MSE
[95% Conf.
-.2874: 84
.0496118
.48581.77
-.3927i.26
-.1671238
-.0994*36
.190U36
-1791258
.2456276
.1517(.96
„ 1153&32
.0755525
-.2962227
-.7776264
-.2970<01
-.0912S64
-1.427178
-.IOIS'13
-. 9849^99
-. 0729741
-.2903*55
-. 1496637
-.4324552
-4.682158
-.41756-41
-.0172E46
-.2255139
.0227333
-.7106742
-.0426913
-.3922932
-.5798C15
-.3794545
-1.67E49
-.1115891
-.0259737
.009117
.0166884
.0301793
.0165794
11940
= 119.00
= 0.0000
= 0.4674
.4501
Interval]
-.2043566
.0872441
.5197133
-.1395972
.2071053
.22021
.4586403
.4414542
.5065479
.4099803
.3712733
.1096374
2.704904
.1990029
-.0149648
.1561678
-.1779887
.3701399
.0033863
.2417142
.2666365
.2004084
.8314059
.6191541
.3882939
.1859291
.0543462
,778749
.1351345
.4992944
-.0119155
.4231041
.0109282
.6562013
.0210493
.0845175
.0709964
.0941036
.1023129
.0898782
-------
274
Idis93
Idis94
Idis95
Idis96
Idis97
vlciis37
vidis88
vldis89
vldis90
vldisSl
vldis92
vldis93
vldis94
vldis95
vldis96
vldis97
.0779527
.0548778
.0652971
.078968
.0681285
.0271892
.010517
-.0330968
-.0258729
-.0239932
-.0194233
-.047743
.0040406
-.0125951
-.034393
.029776
.0157865
.0208902
.0184302
.0173553
.0296495
.0290303
.0198821
.0129333
.0177925
.0164347
.0170738
.0151712
.0155776
.0252384
.0154318
.0269676
4 .94
2.63
3.54
4.55
2.30
94
53
56
45
46
14
15
0.26
-0.50
-2.23
1.10
0.000
0.009
0.000
0.000
0.022
0.349
0.597
0.011
146
144
0.255
0.002
0.795
0.618
0.026
0.270
.0470086
.0139297
.0291709
.0449488
.0100106
.0297149
.0284553
.0584482
.0607295
.0562079
.0528907
-.077481
.0264941
.0620665
.0646419
.0230849
.1088967
.095826
.1014232
.1129872
.1262464
.0840934
.0494892
.0077453
.0089838
.0082216
.0140441
-.018005
.0345752
.0368764
.0041441
.0826368
year88
year89
year90
year91
year92
year 93
year94
year95
year 96
year 97
cons
.1182232
.2210754
.1333214
.0392233
.0426791
.0466043
.0432732
.0580416
.0302123
-.0719657
6.413598
.0446438
.0374153
.0395382
.0388796
.0388566
. 037367
.0386575
.0377463
.0382153
.0426243
.1015964
2.65
5.91
3 .37
1.01
1.10
1 .25
1.12
1.54
0.79
-1 .69
63.13
0.008
0.000
0.001
0 .313
0.272
C .212
0.263
0.124
0.429
0.091
0.000
.0307141
.1477353
.0558201
-.036987
-.0334861
-.0266406
-.0325018
-.0159474
-.044696
-.1555163
6.214452
.2057322
.2944154
.2108227
.1154336
.1188444
.1198502
.1190432
.1320305
.1051206
.0115849
6.612743
Hypothesis
P-value
ofF-test
Reject?
All structural attribute slopes simultaneously zero
All lotsize-independent year-specific coefficient on INSIDE
simultaneously zero
All lotsize-independent year-specific coefficients on INSIDE the
same
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same
All lotsize-dependent year-specific coefficient on INSIDE
simultaneously zero
All lotsize-dependent year-specific coefficients on INSIDE the
same
All lotsize-dependent year-specific slopes on LDIST
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on
vX Idist variables)
0.0000
0.0326
0.0216
0.1086
0.0153
0.0099
0.0033
0.0703
-------
275
7.2 Including Census tract attributes
Regression with robust standard errors '
Isprice
knowf Ir
floors
limpval
age known
age 20
age 30
age 40
age 50
age 60
age 70
age70plus
lotsize
inside37
inside88
insideS9
inside90
inside91
inside 92
inside93
inside94
inside95
inside96
inside97
vinsideS?
vinside88
vinside89
vinside90
vinside91
vinside92
v inside 93
vinside94
vinside95
vinside96
vinside97
Idis87'
IdisSS
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Gcef .
-.1555917
.0799018
.4513043
-.1667545
-.0547812
-.0460286
.1732292
.1754747
.2577204
.184442
.1503967
1.898886
1.327893
-.3140372
-.0911767
.0757837
-.7538S22
.1987656
-.4435937
.1083205
-.0132165
.0473243
. 17T2172
-2.210784
.3521388
.0055391
-.0702629
.4089645
-.3187379
.2367958
-.2102388
-.0281995
-.2273566
-.4909729
-.0667347
.0016666
.0133427
.0206669
.0503369
.0248532
.0521144
.0498406
.045747
.0713424
Robust
Std. Err.
.0212656
.0093892
.0128258
.0604198
.0889524
.077327
-C637876
.0623812
.0619217
.06094
.0600749
.4284057
.7613485
.2556629
.0836671
.0729481
-31534B2
.1203457
.2426716
.0790341
.1424972
.0878266
.3288637
1.353887
.3457122
.0687858
.1051623
.194346
.2174231
.1396171
.1017983
.2625748
.1052115
.604846
.0336392
.0266233
.0166562
.0179299
.0178128
-C1726
.0154312
.0211036
.0209164
.0176271
t
-7.32
8.51
35.19
-2.76
-0.62
-0.60
2.72
2.81
4 .16
3.03
2.50
4.43
1.74
-1 .23
-1 .09
1 .04
-2.39
1.65
-1.83
1.37
-0.09
0 .54
0.54
-1.63
1.02
0.08
-0.67
2.10
-1.47
1.70
-2.07
-0.11
-2.16
-0.81
-1.98
0.06
0.80
1.15
2.65
1.44
3.38
2.36
2.19
4.C5
P>|t I
0.000
0.000
0.000
0.006
0.538
0.552
0.007
0.005
0.000
0.002
0.012
0.000
0.081
0.219
0.276
0.299
0.017
0.099
0.068
0.171
0.926
0.590
0 .590
0.103
0.308
0.936
0.504
i 0.035
0.143
0.090
0.039
' 0.914
0.031
0.417
0.047
0.950
0.423
0.249
0.004
0.150
0.001
0.018
0.029
3.000
Number of obs = 11940
F( 88, 11851) = 118.41
Prob > F = O.OOCO
R-squarec. = 0.5042
Root MSB = .43469
;95>% C'onf
-. 1972',5S
.0614975
.4261(136
-.2851E71
-.229K25
-.1976022
.048195
.0531972
.1363437
.0649696
.0326401
1.05914
-.1644749
-.8151784
-.255178
-.0672C65
-1.372016
-.0371317
-.9192699
-.0465992
-.2925343
-. 1248303
-.4674096
-4.864624
-.3255439
-.1293224
-.2763933
.0280145
-.7449723
-.0368765
-.4097833
-.5428891
-.4335834
-1.67657
-.1326729
-.0505136
-.0193052
-.0144736
.0159239
-.OC89732
.0218667
.0084711
.00474 74
.0367934
Interval ]
-.1139077
.0983062
.4764451
-.0483219
.1195801
.1055451
.2992633
.2977522
.3790972
.3038944
.2681533
2.738631
2.820261
.187104
.0728246
.2187739
-.1357479
.4346629
.0320825
.2632403
.2661013
.2194789
.3218439
.4430563
1.029761
.1403406
.1358724
.7899146
.107397
.5104682
-.0106974
.4864901
-.0211249
.6946245
-.0007964
.0538527
.0459916
.0558125
.0857528
.0586356
.0823621
.0912071
.0867466
.1053943
-------
276
Idis97
.0953133
.0309396
2.76 0.006
.0246666
.14596
vldis37
vldis88
vldis89
vldis90
vldis91
vldis92
vldis93
vldis94
vldis95
vldis96
vldiE97
pfemales
pblack
pother
page underS
page_5_29
page 65 up
pmarhh chd
pmhh child
pfhhTchild
pvacant
prenter occ
vpfemales
vpblack
vpother
vpage underS
vpage 5 29
vpage 65 up
v pmarhh chd
vpmhh child
vpfhh child
vpvacant
vprenter occ
yearSS
year89
year90
year91
year92
year93
year94
year95
year96
year97
cons
.0027051 .0286978 0.09
.0135 .0198033 0.68
-.036293 .0154413 -2.35
-.0295837 .0168912 -1.75
-.0350836 .0164293 -2.14
-.0210323 .0168633 -1.25
-.0449558 .0155513 -2.89
-.0173126 .0153303 -1.09
-.0218964 .0308617 -0.71
-.0478982 .0170374 -2.81
.0002769 .0285389 0.01
3.633995 .9776639 3.72
.5734917 .0784868 7.31
.2065206 .1902086 1.09
1.657506 1.442782 1.15
-1.163295 .4363086 -2.67
-1.359091 .4858882 -2.80
.65092 .3487365 1.37
1.162086 2.290889 0.51
-4.248994 .7425809 -5.72
.7488725 .5147223 1.45
.0053371 .0965389 0.06
-4.575072 .9002597 -5.08
.0309387 .0706607 0.44
.2929666 .147287 1.99
-.4123094 1.104966 -0.37
.1748894 .3880707 0.45
1.137465 .4297104 2.65
1.089086 .2997905 3.63
-3.252622 2.244686 -1.45
1.517444 .9471985 1.79
.0602551 .5079868 C.12
.2529799 .0917453 2.76
.0911239 .0432864 2.11
.1851009 .0368067 5.03
.1015647 .0385181 2.64
.0000576 .0383678 0.00
-.0109949 .0381852 -0.29
-.0168257 .0373153 -0.45
-.0349802 .038968 -0.90
-.0295593 .039832 -0.74
-.0798571 .039797 -2.01
-.1876873 .0440131 -4.26
5.40201 .4651399 11 .61
0,925
0.495
0.019
0.080
0.033
0.212
0,004
0,274
0.478
0,005
0.992
0.000
0.000
0.278
0.251
0.008
0.005
0.062
0.612
0.000
0.146
0,956
0,000
0.661
0.047
0.709
0.652
0.008
0 .000
0 .147
0.073
0.906
0.006
0.035
0.000
0.008
0.999
0.773
0.652
0.369
0.458
0.045
0.000
0.000
-.0535473 .0589575
-.0253177 .0523176
-.0665606 -.0060254
-.0626932 .0035258
-.0672877 -.0028795
-.0540969 .0120324
-.075439 -.0144726
-.0483426 .0137173
-.0823905 .0385977
-.0812943 -.014502
-.0556641 .0562178
1.722613 5.555377
.4196446 .7273388
-.1663195 .5793607
-1.170584 4.485597
-2.018531 -.3080584
-2.311512 -.4066707
-.0326603 1.334501
-3.328433 5.652604
-5.704574 -2.793413
-.2600688 1.757814
-.1838949 .1945691
-6.339729 -2.810415
-.1075178 .1694952
.0042599 .5816733
-2.578029 1.75341
-.5857929 .9355716
.2951622 1.979766
.5014475 1.676725
-7.652574 1.14733
-.1432044 3.178092
-.9354823 1.055993
.0731441 .4323157
.0062754 .1759724
.1129538 .257248
.0260629 .1770665
-.0751495 .0752647
-.0858442 .0638543
-.0899698 .0563184
-.1113638 .0414035
-.1076366 .048518
-.1578658 -.0018484
-.27397 -.1014045
4.490259 6.31376
Hypothesis
All structural attribute slopes simultaneously zero
All lotsize-independent year-specific coefficient on
simultaneously zero
INSIDE
All lotsize-independent year-specific coefficients on INSIDE t
P-value Reject?
ofF-test
0.0000
0.0331
le 0.0253
-------
277
same ,
AH lotsize-independent year-specific slopes on LDIfST
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same
All lotsize-independent Census tract characteristic effects
simultaneously zero
All lotsize-dependent year-specific coefficient on INSIDE
simultaneously zero
All lotsize-dependent year-specific coefficients on INSIDE the
same
All lotsize-dependent year-specific slopes on LDIST
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on
vX Idist variables)
All lotsize-dependent Census tract characteristic effects
simultaneously zero (on vX Census tract variables)
< .0000
0.0120
C.0000
0.0145
O.C143
0.0050
0.2525
0.0000
NO
7.3 Including other distances
Regression with robust standard errors
Isprice
knowf lr
floors
limpvai
age known
age20
age 30
age 40
age50
age 60
age70
age70plus
lotsize
inside87
inside98
inside89
inside90
inside91
Coe f .
-.2297369
.0386152
.4813311
-.2347059
.0510878
.0663445
.2449608
.246457
.3300137
.2690778
.2177839
5.198291
1.312467
-.2723894
-.144157
.0898942
-.8317921
Robust
Std. Err.
.0222194
.0103033
.0113756
.0631998
.0911383
.0767563
.0664987
.065636
.0648841
.0639952
.0633959
2.090635
.7447861
.2503217
.0806899
.0901202
.3199761
-10
3
42
-3
0
0
3
3
5
4
3
2
1
-1
-1
1
-2
t
.29
.75
.31
.71
.56
.36
.58
.75
.09
.20
.44
.49
.".'6
.09
.79
.11
.60
P>|t I
0.000
•o.ooo
0.000
0.000
0.575
0.387
0.000
'o.ooo
.0.000
0.000
0.001
0.013
0.078
'0.277
,0.074
0.267
0.009
Number of obs = 11940
r( 93, 11-341) = 102.65
Prob > F = 0.0000
R-sq-jared = 0.5067
Root MSE = .43377
[95* Conf
-.2722906
.01S4J 9
.4590:-3
-.35856:8
-.127555:2
-.0841105
.1146124
.1177996
.20283C1
.1436367
.C935175
1.100303
-.1474363
-.7630611
-.3023225
-.0681545
-1.458993
Interval ]
-.1851833
.0588114
.5036292
-.1108239
.2297339
.2167995
.3753093
.3751144
.4571972
.3945189
.3420503
9.296279
2.77237
.2182823
.0140084
.2459429
-.2045864
-------
278
inside92
inside 93
inside 94
inside 95
inside 96
inside97
vinsideS?
vinsideSS
vinside89
vinsideSO
vinside91
vinside92
vir.side93
vinside94
vinside95
vinside96
vinside97
Idis87
Idis88
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
vldis87
vldisSS
vldis89
vldis90
vldis91
vldis92
vldis93
vldis94
vldio95
vldis96
vldis97
Id summits
Id school
id retail
Id hospital
Id church
Id cemetery
Id railroad
Id rijrds
Id 1280
Id gspkwy
Id parks
Id mjwater
Id col leges
Id cclubs
Id airports
Id newark i
.1736454
-.4005431
.1252535
.0354209
.0681992
.2350139
-2.241971
.1582947
.0673103
-.152343
.4513601
-.3377249
.1407771
-.2234914
-.1399916
-.2391652
-.5979581
-.0685386
.0163296
.0180767
.0113164
.03852
.027194
.0444265
.0296375
.027221
.0478906
.0589567
.0345949
.0215694
-.0218843
.0163053
-.0060064
-.0006561
-.0184821
.0178603
.0179614
-.0109826
.027111
.0373709
.0365305
.8859575
.1218503
.0163629
-.0137802
-.0294455
.0063738
-.00512
.171849
.0139228
.1372817
-.0785524
.0009817
-.7112129
.5865961
.125308
.2437214
.091058
.1787S43
.0913403
.3245386
1.317233
.3332254
.0794468
.0794524
.196025
.2224905
.1316803
.1167025
.328934
-110798
.6008897
.0332642
.0276973
.0160616
.0197947
.0178379
.0185139
.0160717
.0217526
.0198239
.0179807
.0287878
.0300081
.0206786
.0131949
.0180541
.0159728
.0177764
.0160643
.0170651
.0277529
.0165121
.0267199
.029433
.0116968
.1729206
.0145481
.0143884
.016551
.0105898
.0056671
.013585
.0231S9
.0083375
.0370056
.0237257
.0082758
.1201315
.1197834
1.39
-1.64
1 .33
0.20
0.75
0.72
-1.70
0.48
0.85
-1.92
2.30
-1 .52
1 .07
-1.92
-0.43
-2.16
-1.00
-2.06
0.59
1.13
0.57
2.16
1.47
2.76
1,36
1.37
2.66
2,05
1,15
1.04
-1.66
0.90
-0.33
-0.04
-1.15
1.05
0.65
-0.67
1.01
1.27
3.12
5.12
8.38
1.14
-0.83
-2.78
1.12
-0.38
7.41
1.67
3.71
-3.31
0.12
-5.92
4.90
0.166
0.100
0.169
0.843
0.455
0.469
0.089
0.635
0.397
0.055
0.021
0.129
0.285
0.056
0.670
0.031
0.320
0.039
0.555
0.260
0.568
0.031
0.142
0.006
0.173
0.170
O.OOS
0.041
0.249
0.297
0.097
0.366
0.707
0.971
0.250
0.295
0.518
0.506
0.310
0.204
0.002
0.000
0.000
0.255
0 .405
0.005
0.261
0.706
0 .000
0.095
0.000
0.001
0.906
0.000
0.000
-.0719788
-.8782772
-.0532352
-.3150258
-.1108427
-.4011351
-4.823963
-.4948818
-.088418
-.3080826
.0671188
-.7738429
-.1173379
-.4522474
-.7847562
-.4563475
-1.775801
-.133742
-.0379617
-.0134066
-.0274845
.0035547
-.0090964
.0129233
-.013001
-.011637
.0126455
.0025279
-.0242259
-.0189641
-.0477484
-.0190831
-.0373157
-.0355008
-.0499707
-.0155901
-.0364388
-.043349
-.0252644
-.0203226
.0136029
.5470046
.0933336
-.0118407
-.0462229
-.0502033
-.0047347
-.0317489
.1263947
-.0024199
.0647447
-.1250586
-.0152402
-.9466904
.3518006
.4192696
.077191
.3037422
.3858675
.2472412
.8711629
.3400214
.8114712
.2230396
.0033967
.8356013
.0983931
.398B921
.0052646
.5047731
-.0219829
.5798846
-.0033352
.0706209
.0495601
.0501174
.0734852
.0634844
.0759296
.0722761
.0660791
.0831356
.1153854
.0934157
.0621029
.0039798
.0516948
.0253029
.0341887
.0130064
.0513103
.0723616
.0213839
.0794864
.0950644
.0594581
1.22491
.1503669
.0445664
.0186625
-.0086877
.0174824
.0215089
.2173033
.0302656
.2098187
-.0320462
.0172037
-.4757354
.8213913
.0160069
.0258883
C.62 0.536
-.0347383
.0667522
-------
279
vld_school I -.0186214
vld retail
vld hospital
vld church
vld cemetery
vld railroad
vld njrds
vld_i280
vld gspkwy
vld parks
vld mjwater
vld colleges
vld cclubs
vld airports
vld newark i
year88
yearS9
year90
year91
year92
year93
year94
year95
year 96
year 97
cons
-.3242198
-.067021
-.0139537
-.0117227
.0408327
.023313
.0165044
-.0748724
-.0272249
-.0417069
.0249413
-.0003378
.1606291
-.2736358
.1048026
.2069C7
.1027585
.0220043
.0142346
.027096
.0276894
.0386965
.0107121
-.0767808
-3.550678
.0095079
.169203
.0146772
.0114934
.0151627
.0088443
.0045355
.0117868
.0267435
.007566
.0349399
.0238649
.0040357
.1132629
.1249954
.0438016
.0368331
.0391785
.0383307
.038081
.0367563
.0379951
.0369751
.0376805
.042187
1.89115
-1.96
-1.92 i
-4.57
-1.21
-0.77
4.62
5.14
1.40
-2.80
-3.60
-1.19
1.05
-0.08
1.46
-2.19
2.39
5. 52
2.62
0.57
0.37
0.74
0.73
1.05
0.28
-1.32
-1.88
0
0
0
C
C
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.050
.055
.000
.225
.439
.000
.000
.161
.005
.000
.233
.296
.933
.145
.029
.017
.000
.009
.566
.709
.461
.466
.295
.776
.069
.060
-.03725.84
-.65S8fc:55
-.0957!>08
-.C364>'27
-.0414442
.0234964
.0144:27
-,OD65!:97
-.12725:39
-.0420J-55
-.1101S43
-.0218377
-.0082^84
-.0555C-42
-,5186S'73
.0189442
.1347C81
.0259622
-.0531302
-.0604304
-.0449^25
-.0467673
-.0337607
-.0631479
-.1594743
-7.257642
.0000155
.0074458
-.0382512
.0085753
.0179987
.0581691
.0322033
.0396086
-.0224508
-.0123942
.0267809
.0717204
.0075729
.3767624
-.0286743
.190661
.2791059
.1795549
.0971387
.0888796
.0991445
.102166
.1111737
.0845721
.0059127
.156287
Hypothesis
P-value
ofF-test
Reject?
All structural attribute slopes simultaneously zero
All lotsize-independent year-specific coefficient on JNSIDE
simultaneously zero
All lotsize-independent year-specific coefficients on INSIDE the
same !
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
All lotsize-independent year-specific slope on LDISt" the same
All lotsize-independent other distance effects simultaneously zero
All lotsize-dependent year-specific coefficient on INSIDE
simultaneously zero
All lotsize-dependent year-specific coefficients on INSIDE the
same
All lotsize-dependent year-specific slopes on LDIST
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on
vX Idist variables)
0.0000
0.0190
0.0130
0.0092
0.1169
0.0000
0.0385
0.0119
0..1252
0. 1312
NO
NO
NO
-------
280
All lotsize-dependent other distance effects simultaneously zero (on ° • 000°
vX "other distance" variables)
7.4 Including both other distances and tract attributes
Regression with robust standard errors Number of obs
F(120, 11S19)
Prob > F
R-squared
1
Isprice I
knowf 1 r 1
floors I
limpval I
age known 1
age 20 I
age 30 1
age4C I
age50 i
age 60 I
age?0 I
ageVOplus 1
lotsize I
inside87 |
insideSS I
inside89 ]
inside90 I
inside91 1
inside92 i
inside93 1
inside94 1
inside95 1
inside96 I
inside97 1
v inside 8 7 |
vinside88 1
v ins ide8 9 I
v inside 90 1
v inside 91 1
vinside92 1
vinside93 I
vinside94 1
vinside95 1
v inside 9 6 1
v inside 97 |
Idis87 I
Idis88 1
Idis89 1
Idis90 I
Coef .
-.1388045
.0468302
.4685191
-.1020435
-.0367788
-.056402
.1163051
.1307326
.2194256
.1489352
.1039544
11.05272
1.320297
-.2130517
-.0562239
.1149565
-.7884622
.2425379
-.3638381
.147494
.0013879
.0359847
.1947032
-2.194676
.1110547
.0022437
-.1618678
.450845
-.3932505
.137383
-.237142
-.0520266
-.1943162
-.5286984
-.0870175
-.0069996
-.0097467
-.0262071
Robust
Std. Err.
.0234666
.0101169
.0140783
.0599523
.0867948
.0740602
.0631492
.061948
.0613041
.0603619
.059647
2.193922
.7673162
.2620469
.10345
.081788
.3218548
.1279812
.2431987
.0946385
.1543916
.1000822
.3393725
1 . 362436
.3607654
.1253706
.0959758
.2013335
.2273961
.1375244
.1286066
.279238:
.1277085
.6271795
.033402
.0266664
.0182342
.0197768
-5
4
33
-1
-0
-0
1
2
3
2
1
C.
1
-0
-0
1
•"!
~ £.
1
-1
1
0
0
0
-1
0
0
-1
2
-I
1
-1
-0
-1
-0
-2
-0
-0
-1
t
.91
.63
.28
.70
.42
.76
.84
.11
.58
.47
.74
.06
.72
.81
.54
.41
.45
.90
.50
.56
.01
.36
.57
. 61
.31
.02
.69
.24
. 7 3
.00
.84
.19
.52
.84
.61
.26
.53
.33
P>l 1 1
0.000
0.000
0.000
0.089
0.672
3.446
0.066
0.035
0.000
0.014
0.081
0.000
0.085
0.416
0.587
0.160
0.014
0.058
0.135
0.119
0.993
0.719
0.566
0.107
0.758
0.986
0.092
0.025
0.084
0.318
0.065
0.852
0.128
0.399
0.009
0.793
0.593
0.135
Root MSE
[95% Conf.
-.1848029
.0269994
.4409233
-.2195598
-.206911
-.2015722
-.0074776
.0093044
.0992594
.030616
-.0129635
6.772071
-.1837694
-.7267068
-.2590029
-.0453615
-1.419351
-.0083264
-.8405476
-.0380132
-.3012451
-.1601929
-.4705227
-4 .865275
-.5961049
-.2435032
-.3499962
.0561982
-.8389844
-.1321875
-.4892321
-.5993793
-.444646
-1.758074
-.1524909
-.0592702
-.0454887
-.064973
11940
= 101.32
= 0.0000
= 0.5295
= .42402
Interval)
-.092806
.0666611
.496115
.0154728
.1333533
.0887681
.2400879
.2521609
.3395918
.2672545
.2208724
15.33337
2.824363
.3006035
.1465552
.2752745
-.1575737
.4934022
.1128714
.3330011
.3040208
.2321623
.8599291
.475923
.8182143
.2479907
.0262606
.8454917
.0524834
.4069536
.0149481
.495326
.0560135
.7006768
-.0215441
.0452709
.0259953
.0125587
-------
281
Idis91 1
Idis92 I
Idis93 1
Idis94 1
Idis95 1
Idis96 I
Idis97 I
vldisB? I
vldisSB 1
vldis89 i
vldis90 1
vldis91 I
vldis92 1
vldis93 1
vldis94 1
vldis95 1
vldis96 !
vldis97 1
Id summits 1
Id school I
Id retail 1
Id hospital [
Id church !
Id cemetery I
Id railroad 1
Id njrdg 1
Id_i280 I
Id gspkwy I
Id parks 1
Id mjwater 1
Id colleges I
ld_ccluba 1
Id airports I
Id newark i I
vld summits I
vld school 1
vld retaii I
vld hospital 1
vld church 1
vld cemetery 1
vld_railroad 1
vld njrds 1
vld_i280 I
vld gspkwy 1
vld parks 1
vld mjwater i
vld colleges I
vld cclubs ',
vld airports I
vld newark i 1
pfemales 1
pblack I
pother [
page under 5 1
page_5_29 1
page_65 up 1
ptnarhh chd 1
. 0236279
.0033196
.0325805
.0231713
.0156275
.0440692
.0669015
.0213704
.0340216
-.0131749
.0251521
-.0048707
.0063697
-.0170301
.0160286
.0153054
-.0163576
.0180802
.0658995
.038532
.3609476
.0708335
.0125312
-.0016792
-.0296702
.0048456
.0136775
.109992
.0157448
.1177259
-.0120485
.0004339
-.3845297
.4456733
-.0298753
-.0294116
-.2153227
-.0559326
.0036147
-.0241847
.036741
.0235222
-.0355665
-.0504349
-.0203045
-.1098461
.0071572
.0008703
-.0649777
-.552884
2.946455
.6455517
.5025301
1.104882
-.8263215
-2.008093
-.0910406
.0191811
.0186659
.0165168
.0227175
.0209928
.0195189
.0301052
.0292732
.0204872
.0173746
.0195516
.0192534
.0190284
.0172681
.0185496
.0301445
.020216
.0288953
.0301136
.0120029
.1826239
.0158687
.0150399
.017265
.0106432
.0056867
.0145797
.0240495
.0076826
.0383034
.0226507
.0087763
.1326254
.1218155
.0247342
.0095838
.1699369
.0145134
.0122057
.0163083
.0084999
.0046235
.0123809
.0268557
.0067555
.0374935
.0217797
.0049006
.1225234
.1222448
1.GC9645
.0912565
.2143407
1.544262
.4643796
.5136042
.4179533
:..23 '
0.18
1.97
1.02 !
0.74
2.26
2.22 ,
0.75
1 .66
-0.76 !
1 .29
-0.25
0.33 '
-C .99
C .86
C.51
-C .81
C.63
2.19
3.21
1.98
4.46 '
0.83
-0.10
-2.79
0.85
0.94
4.57 •
2.05
3.07
-0.53 .
0.05
-2.90
3.66
-1.21
-3.07
-1.27
-3.85
0.31
-1.48
4.32
5.09
-2.87
-1.88
-3.01
-2.93
0.33
0.18
-0.53
-4.52 ;
2.92
7.07 ,
2.34
0.72
-1 .78
-3.91
-0.22
0.218
0.859
0.049
0.308
0.457
0.024
0.026
0.453
0.097
0.448
0.198
0 . SCO
0.739
C.324
0.368
0.612
0.418
0.532
0.029
C.001
0.048
0.000
0.405
0.923
0.005
0.394
0.348
0.000
0.040
0.002
0.595
0.961
0.004
0.000
0.227
0.002
0.205
C.OOO
C.755
0.138
0.000
0.000
0.004
0.060
0.003
0.003
0.742
0 .859
0.596
0.000
0.004
0.000
0.019
0.474
0.075
0.000
0.828
-.0139703
-.0332686
.0002348
-.0213588
-.0255219
.0058289
.0078305
-.C 354)99
-.C061367
-.047232
-.C131723
-.0426106
-.030929
-.C508796
-.0203317
-.0437329
-.C559343
-.0385595
.0068718
.0150)43
.0029747
.0397282
-.0169-195
-.0355;!14
-.0505327
-.0063013
-.0149012
.062851
.0006057
.0426449
-.0564475
-.0167691
-.6444974
.2068943
-.0783!. 84
-.0481975
-.548il27
-.0843B13
-.0201: 04
-.0561516
.0200":99
. 01441-93
-.0598351
-.1030'.' 64
-.0335^63
-.1333:-! 96
-.0355I545
-.0037::56
-.3D5K38
-.792504
.9573E43
. 4666";38
.0824^69
-1.922126
-1.736E82
-3.014642
-.910298
.061226
.0399079
.0649562
.0677014
.0567763
.0823494
.1259126
.0793507
.0741799
.0208822
.0634765
.0328692
.0436684
.0168193
.0523888
.0743936
.023269
.0747198
.1249272
.0620598
.7189204
.1019388
.0420119
.032163
-.0088077
.0159924
.0422562
.1571329
.0308039
.1923069
.0323505
.0176368
-.124562
.6844517
.0186078
-.0106258
.1177816
-.0274839
.0277398
.0077822
.0534021
.0325851
-.011298
.0022067
-.0070627
-.0363527
.049849
.0104762
.1751883
-.3132641
4.925526
.8244295
.9227232
4.131839
.0839389
-1.001344
.7282163
-------
282
pmhh child
pfhh_child
pvacant
prenr.er occ
vpfemalss
vpblack
vpother
vpage underS
vpage 5 29
vpage 65 up
vpmarhh chd
vprahh child
vpfhh^child
vpvacant
vprenter occ
year88
year89
year90
year91
year92
year93
ye a r 9 4
year 95
year96
year 97
cons
.5381206
-4 .029349
.4903542
-.0574815
-3.239823
-.111944
.5074429
-1.070621
.06907
1.263968
1.677632
-2.802156
1.604859
-.526921
.3158721
.0756214
.1722892
.0830788
-.0165673
-.0324988
-.0345966
-.0563537
-.0525721
-.0994709
-.2015484
-1.241569
2.239511
.7243345
.5223224
.1097548
.9289654
.0852464
.184935
1.285217
.4157292
.4582809
.3684749
2.123441
.7723147
.5277025
.102198
.0430731
.0368321
.0338464
.0383278
.038089
.037293
.0390625
.0389484
.0403243
.0442536
2.044169
0
-5
0
-0
-3
-1
2
-0
0
2
4
-1
2
-1
3
1
4
2
— u
-0
-0
-1
-1
-2
-4
-0
.24
.56
.34
.52
.49
.31
.74
.83
.17
.76
.55
.32
.08
.00
.09
.76
.68
.14
.43
.85
.93
.44
.35
.47
.55
.61
o.aio
0.000
0.348
0.600
0.000
0.189
0.006
0.405
0.868
0.006
0.000
0.187
0.038
0.318
0 .002
0.079
0.000
0.032
0.666
0.394
0.354
0.149
0.177
0.014
0.000
0 .544
-3.85169
-5.449164
-.5334836
-.272619
-5.060748
-.279041
.1449398
-3.589858
-.7458276
.3656619
.9553604
-6.964451
.0909948
-1.561305
.1155471
-.0088088
.1000922
.0069334
-.0916962
-.1071596
-.1076971
-.1329226
-.1289175
-.1785132
-.2882928
-5.248477
4.927932
-2.609534
1.514192
.1576561
-1.418897
.0551529
.8699459
1.448616
.8839676
2.162274
2.399903
1.360139
3.11S723
.5074629
.516197
.1600517
.2444863
.1592243
.0585616
.0421619
.0385038
.0202151
.0237733
-.0204285
-.1148041
2.765339
-------
283
Hypothesis
P-value
ofF-test
Reject?
All structural attribute slopes simultaneously zero !
All lotsize-independent year-specific coefficient on,INSIDE
simultaneously zero
All lotsize-independent year-specific coefficients on INSIDE the
same
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same
All lotsize-independent other distance effects simultaneously zero
All lotsize-independent Census tract characteristic effects
simultaneously zero
All lotsize-dependent year-specific coefficient on INSIDE
simultaneously zero ,
All lotsize-dependent year-specific coefficients on INSIDE the
same
All lotsize-dependent year-specific slopes on LDIST
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST ,the same (on
vX Idist variables)
All lotsize-dependent other distance effects simultaneously zero (on
vX "other distance" variables)
All lotsize-dependent Census tract characteristic effects
simultaneously zero (on vX Census tract variables)
C:.0000
C .0376
C .0427
0.0042
0.0040
C.OCOO
C .0000
C.0355
C.0478
0.2052
0.1606
0.0000
o.cooo
NO
NO
-------
284
Appendix B - OH Landfill Site
Contents:
1 CRITERIA FOR EXCLUSION FROM RAW SAMPLES 286
2 ANNUAL COUNTS IN SAMPLE 286
3 DESCRIPTIVE STATISTICS 287
3.1 Housing prices and distances from the site 287
3.2 Structural variables 287
3.2.1 R2 for auxiliary regressions among variables 288
3.3 Census tract attributes 288
3.3.1 R2 for auxiliary regressions among variables 288
3.4 Other distances 288
3.4.1 R2 for auxiliary regressions among variables 290
4 COLLINEARITIES 290
4.1 Time patterns in average site distances in sample 290
4.2 Time trend in average lot sizes 291
4.3 Distance to site vs. structural variables 292
4.4 Distance to site vs. Census tract attributes 292
4.5 Distance to site vs. other distances 293
5 TRENDS IN THE DISTANCE GRADIENT 293
5.1 Structural variables 294
5.1.1 Built post-1900 294
5.1.2 Age if built post-1900 294
5.1.3 Square footage 294
5.1.4 Bedrooms 295
5.1.5 Bathrooms 295
5.1.6 Fireplace(s)? 295
5.1.7 Floors recorded? 296
5.1.8 Floors 296
5.1.9 Lotsize 296
5.2 Census tract attributes 297
5.2.1 Females 297
5.2.2 Whites 297
5.2.3 Blacks 298
5.2.4 Other ethnic groups 299
5.2.5 Children under 5 300
5.2.6 Persons between 5 and 29 300
5.2.7 Persons between 30 and 64 301
5.2.8 Persons 65 and older 302
5.2.9 Married heads of household 302
5.2.10 Male-headed of household with children 303
5.2.11 Female-headed households with children 304
-------
285
5.2.12 Owner-occupancy , 304
5.2.13 Renter-occupancy ! 305
5.2.14 Vacancy rates 306
6 COMPLETE REGRESSION RESULTS - No LOT SIZE INTERACTIONS 306
6.1 Jus} structural characteristics and year durjimies 306
6.2 Including Census tract attributes 308
6.3 Including other distances 310
6.4 Including both other distances and tract attributes 312
7 COMPLETE REGRESSION RESULTS - WI|H LOT SIZE INTERACTIONS 314
7.1 Just structural characteristics and year dummies 314
7.2 Including Census tract attributes 317
7.3 Including other distances 319
7.4 Including both other distances and tract attributes 322
-------
286
Chapter 1 Criteria for Exclusion from Raw Samples
For OH, we drop observations for all the same reasons as Montclair. For OH, however, there are
many more structural characteristics of the house available, and we prefer to use these to control
for structural differences. For Oil, we drop observations for which
• no data are available concerning the year the dwelling was built, which is used to
determine its age at the time of the last sale.
• Number of floors exceeds four
• Square footage of the dwelling is missing
• Number of bedrooms or bathrooms is missing, or number of full baths is greater than five
• Presence or absence of a fireplace is not recorded
• Square footage of dwelling exceeds 5000.
• Lotsize is greater than 25,000 square feet (e.g. 250' x 100')
Chapter 2 Annual counts in sample
year
70
71
72
73
74
75
76
77
78'
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
Freq.
99
138
189
203
201
209
242
259
225
228
130
82
98
166
200
229
303
382
415
398
331
343
312
364
432
396
467
484
745
941
Percent
1.07
1.50
2.05
2.20
2.18
2.27
2.63
2.81
2.44
2.48
1.41
0.89
1.06
1.80
2.17
2.49
3.29
4.15
4.51
4.32
3.59
3.72
3.39
3.95
4.69
4.30
5.07
5.25
8.09
10.22
Cum.
1.07
2.57
4.62
6.83
9.01
11.28
13.91
16.72
19.16
21.64
23.05
23.94
25.00
26.80
28. 98
31.46
34.75
38.90
43.40
47.73
51.32
55.04
58.43
62.38
67.07
71.37
76.44
81.70
89.78
100.00
-------
287
Total |
9211
100.00
Chapter 3 Descriptive statistics
3.1 Housing prices and distances from the site
Variable |
Obs
Mean Std. Dev.
Min
Max
dist | 9211 3.367053 1.853798 .0635377 8.467858
sprice I 9211 130252.1 75643.73 2000 593000
041088-
o-
.063538
dist
8.4
Marginal distribution of distances: OH
.086297-
0-
2000
sprice
593^)00
Marginal distribution of house prices: OH
3.2 Structural variables
Variable t Obs
Mean Std. Dev.
Min
Max
-------
288
notold
age
age2
sqft
sqft2
bedrms
bthrms
sqftbed
sqftbth
fplace
knowf Ir
floors
lotsize
9211
9211
9211
9211
9211
9211
9211
9211
9211
9211
9211
9211
9211
.9997829
32.53849
1405.443
1.373521
2.131615
2.903811
1.912116
4 .256291
2.884616
.396591
.8059928
.9280751
1
.0147346
18.62063
1293.041
.4950583
1.739636
.8580447
.8345881
2.66714
2.1323
.4892163
.3954559
.5609922
.4556344
0
0
0
.3
.09
1
1
.3
.3
0
0
0
.1294409
1
91
8281
4.8
23.04
7
5
27
19.2
1
1
3
4.045029
3.2.1 R for auxiliary regressions among variables
3.3 Census tract attributes
Variable
pfemales
pblack
pother
page underS
page_5_29
page 65 up
pmarhh chd
pmhh child
pfhh_child
pvacant
prenter occ
Obs
9211
9211
9211
9211
9211
9211
9211
9211
9211
9211
9211
Mean
.5118072
.0059348
.5082869
.073255
.4069791
.1073373
.3142498
.0258387
.0784943
.0296959
.3972171
Std. Dev.
.0118325
.0054257
.1999439
.0177446
.0487373
.0394719
.0640051
.0166391
.0324638
.0137419
.1705825
Min
.4678834
0
.003553
.0382657
.2603768
.0255308
.1093058
0
0
0
.090209
Max
.5899951
.0882786
.8814761
.136191
.5292445
.2438971
.5365998
.1202512
.1863905
.1009516
.7281437
3.3.1 R for auxiliary regressions among variables
3.4 Other distances
Distance variable
d_school
d_retail
djiospital
Description (in kilometers)
Distance to nearest school. There are dozens of schools in the
sample area.
Distance to nearest retail center. Montebello Mall lies about three-
quarters of a mile to the east of the eastern edge of the site.
Distance to nearest hospital. There are four hospitals that lie within
the boundaries of our sample area, and a further 12 lying just
outside the sample area that will sometimes be the nearest hospital
to some houses in our sample.
-------
289
d church
Distance to nearest church. O|nly Saint Alphonsus Catholic Church
lies in our sample area, in the western portion. There are no other
churches within the sample ar^, but there are dozens of religious
facilities clustered in each of three areas, one staning about a mile
to the southwest, one starting about two miles to the southeast of
the sample area, and one starting about a mile to the northeast.
ting
JR
d_cemetery
Distance to nearest cemetery.! Resurrection Cemetery lies just to
the north of the landfill site. Savannah Cemetery lies in the
Northeast Corner, and there are no less than six cemeteries at the
western extreme of our sample area. Ten other cemeteries, outside
the sample area, may serve as the closest cemetery to some houses
in the sample.
d i5
Distance to Interstate 5 freeway. The Golden State Freeway runs
just outside the southwestern nerimeter of our sample area, forming
this boundary.
d i605
Distance to Interstate 605 freeway. This freeway Jbrms the
southeastern boundary of our sample area.
d ilO
Distance to Interstate 10 freewjay. The San Bernardino Freeway
runs along the northern edge of our sample, except for a subset of
houses in one Census tract that spans the freeway.
d railroad
Distance to nearest railroad. A dense array of railroad tracks
occupies an area 7 miles east-west by 2.5 miles north-south
adjacent to the southwest portion of our sample area. Railroad
tracks border our sample area to the north and the southeast, and
three east-west lines cut through the sample area at different
latitudes, but none of these lines approaches more closely than
about 1.4 miles from the landfill site.
jfill;
3H101
d s60
Distance to state route 60 (Pomona Freeway). This freeway runs
east-west, spitting the landfill site into its northern and southern
portions and splitting our sample of houses roughly in half as well.
d rivers
Distance to nearest minor river or streambed. There are few natural
rivers in Southern California. The Alhambra, Rubio, and Eaton
Wash features cut through the northeastern comer of our sample
area, and the Whittier Narrows dam creates some water features.
See dm] water for substantial waterways.
d cards
Distance to nearest road (CA roads feature). In addition to the
freeways for which distances are also included individually, this
group of features includes about 7 major roads running east-west
and about 7 major roads running north-south through the sample
area
d whittiern
Distance to Whittier Narrows recreation area. This is a large
recreation area sitting to the east of our sample area, roughly 2.5
miles by 1 miles in size, but not completely contiguous. It lies
about another % of a mile further to the east of the landfill site than
the Montebello Mall.
-------
290
d_parks
d_mj water
d_csula
d_cclubs
Variable |
d school
d retail
d hospital
d church
d cemetery
d 15
d 1605
d_ilO
d railroad
d_s60
d rivers
d cards
d whittiern
d parks
d mjwater
d csula
d cclubs
Distance to nearest park. There are roughly two dozen small parks
within the boundaries of our sample area, and more just to the
outside of the area which may be the closest parks to some houses
in the sample.
Distance to nearest major body of water. The San Gabriel River
corresponds geographically to the 1-605 Freeway in this area, so it
will be extremely unlikely that we can distinguish between the
effects of this freeway and the effects of this River. However, this
category of features also includes the Rio Hondo River, which
parallels the San Gabriel River about 1.75 miles to the northwest.
The Los Angeles River lies at least a mile outside our sample area,
to the west and southwest
Distance to the campus of the California State University at LA.
This major public urban university lies just outside the northwest
corner of our sample area.
Distance to nearest country club/golf course. There are four golf or
country clubs within our sample area, and another three outside the
boundary that may serve as the closest facilities.
Obs Mean Std. Dev. Min Max
9211 .4695382 .2278552 .0210058 1.570117
9211 4.546856 1.95908 .3162922 9.420857
9211 1.849731 .7941503 .0317994 3.996492
9211 3.041624 1.283886 .0635077 6.233308
9211 2.535859 1.04748 .0507924 4.412509
9211 5.223122 2.794447 .1060339 12.28935
9211 5.256435 2.773785 .2184427 11.02538
9211 4.678754 3.495584 .0385008 13.30645
9211 1.333081 .9336122 .0020506 3.661991
9211 2.973134 2.092419 .0089229 9.187186
9211 2.762804 1.824602 .0013621 7.937554
9211 .2527654 .1971348 6.50e-07 1.167369
9211 4.498945 2.148109 .2511083 9.383685
9211 .6886061 .397285 3.90e-07 1.988177
9211 2.407134 1.820432 .0022402 7.016768
9211 7.118163 3.014083 .483409 13.94009
9211 2.250993 1.238161 .0000936 5.479274
3.4.1 R2 for auxiliary regressions among variables
Chapter 4 Co (linearities
4.1 Time patterns in average site distances in sample
Regression with robust standard errors
Number of obs
F( 29, 9131)
Prob > F
9211
1 .94
0.0019
-------
291
Idist
year71
year72
year73
year74
year75
year76
year77
year78
year79
yearSO
yearBl
yearS2
year93
yearS4
year85
year 36
year87
yearSa
year99
year90
year91
year92
year93
year94
year95
year96
year97
year98
year99
cons
Coe f .
.0317235
.0859397
.0032975
.0767495
-.Q753499
-.1300573
-.170214
-.1477047
-.0895337
-.1115134
.0920412
.0197791
-.0919169
-.0593357
.1188389
.0284096
.0622672
.0567704
-.0147613
-.030246
-.0163813
.0391751
.0579568
.0319384
-.0086662
.0268945
.0447263
.0191437
.0230723
1.0002C4
Robust
Std. Err.
.0883472
.0840053
.0832593
.0816442
.0397324
.0956679
.0909799
.0908943
.0866816
.108103
.0962669
.0995868
.094225
.0850137
.0812613
.0817358
.0768907
.076419
.0794248
.0805885
.0783395
.0799508
.0783126
.0759538
.0774514
.0771292
.0746531
.0737975
.0723098
.0686039
t
0.36
1 .02
C.04
C.94
-C .35
-1 .36
-1.87
-1.63
-1.03
-1.03
0.96
0.20
-0.99
-0.70
1.46
0.35
0.81
0.74
-0.19
-0.38
-0.21
0.49
0.74
0.42
-0.11
0.35
0.60
0.26
0.32
14.58
P>|t I
' 0.720
0.306
. 0.968
0.347
0.393
, 0.174
0.061
0 .104
i 0.302
0.302
0.339
: 0.843
0.329
0.485
0.144
0.728
0.418
0.458
0.853
0.707
0.834
0.624
. 0.459
0.675
0.911
0.727
0.549
0.795
0.750
0.000
R-squarei
Root. MSE
[95% :onf.
-.1414366
-.C737 504
-.1599392
-. 0832913
-.2517 154
-.3175376
-.3435.347
-.3258">76
-. 2594489
-.3234 ..94
-.0966o33
-.1754332
-.2766189
-.2259B14
-.0404!>13
-.1318L.03
-.0884!.»56
-.0930:.:7S
-.1704M6
-.1882! 74
-.1699lt 1
3.403
0.189
0.713
0.441
0.034
0.863
0.909
0.166
0.955
Number of obs
F( 29, £181)
Prob > F
R-squarec
Root MSE
[95% Cor.f.
-.0498378
-.0286421
-.0688469
-.0532611
.007331
-.0912786
-.0782955
-.02739
-.0941773
9211
5.56
= 0.0000
= 0.0154
= .45283
Interval ]
.1240091
.1449926
.1006451
.1222456
.1867444
.0765397
.0879401
.1593283
.0889044
-------
292
year 80
year 81
year82
year93
year 3 4
year35
year86
year87
year 98
year89
year90
year91
year92
year 93
year94
year95
year96
year 97
year98
year 9 9
ccns
-.0004551
-.0857179
-.0294701
-.0527158
-.0916612
.0041223
-.0403184
-.0925032
-.0485246
-.0928933
-.1342887
-.113602
-.1057711
-.0716032
-.0892971
-.0976527
-.1178293
-.0796451
-.0983243
-.0897773
1.059879
.0-624263
.0482712
.0702935
.0476387
.0454278
.0458008
.0432267
.0403962
. 041859
.0403567
.041311
.0424817
.0424013
.0421767
.0395896
.0407658
.0400843
.0416097
.0387418
.0378823
.0343544
-0
-1
-0
-1
_2
0
-0
-2
-1
_2
-3
-2
-2
-1
-2
-2
-2
-1
-2
-2
30
.01
.78
.42
.11
.32
.39
.93
.29
.16
.30
.25
.67
.49
.70
.26
.40
.94
.91
.54
.37
.85
0
0
0
0
0
0
0
0
0
0
0
0
0
D
0
0
0
0
0
0
0
. 994
.076
.675
.269
.044
.928
.351
.022
.246
.021
.001
.008
.013
.090
.024
.017
.003
.056
.011
.018
.000
-.1228245
-.1803402
-.167261
-.1460982
-.1807098
-.0856575
-.1250523
-.1716887
-.1305775
-.1720018
-.2152675
-.1968756
-. 188887
-.1542789
-.1669016
-.1775628
-.1964035
-.1612094
-.174267
-.1640351
.9925371
.1219143
.0089043
.1083208
.0406665
-.0026126
.093902
.0444154
-.0133177
.0335282
-.0137858
-.0533099
-.0303283
-.0226551
.0110726
-.0116926
-.0177426
-.0392551
.0019193
-.0223817
-.0155196
1.127222
4.3 Distance to site vs. structural variables
Reqression with robust standard errors
Idist 1 Coef.
notold -1.164896
age
age2
sqft
sqf t2
bedrtns
bthrrns
sqftbed
sqf toth
fplace
knowf Ir
floors
lotsise
cons
.0169089
-.0001336
-.2965837
.102883
.1511174
-.0651158
-.0620597
.0075439
-.1613489
.3965481
-.4768071
.0003271
2.108913
4.4 Distance to site vs.
Robust
Std. Err.
.1536876
.0014887
.000018
.0615899
.016596S
.0313248
.0364991
.019966
.0236997
.0164944
.0462004
.0336813
.0173777
.1713417
_-i
11.
-7.
-4.
6.
4.
-1.
-3 .
0.
-9.
8.
-14.
0.
12.
t
34
36
42
82
20
82
78
11
32
78
53
16
02
31
P> 1 1 1
0.
0.
0.
0.
0.
0.
0.
0.
0 .
0.
0.
0.
0.
0.
000
000
000
000
000
000
074
002
750
000
000
000
985
000
Number of. obs
F( 13, 9197)
Prob > F
R-squared
Root MSE
195% Conf.
-1.475959
.0139908
-.0001689
-.4173135
.0703691
.0897138
-.1366621
-.1011975
-.0389128
-.1936816
.305985
-.5428308
-.0347173
1.773045
9211
= 125.27
= 0.0000
= 0.2048
= .66652
Interval)
-.8538331
.019827
-.0000983
-.1758538
.1353969
.212521
.0064306
-.0229219
.0540005
-.1290163
.4871112
-.4107833
.0353714
2.444781
Census tract attributes
Regression with robust standard errors
Number of obs
F( 11, 9199)
Prob > F
R-squared
9211
= 227.77
= 0.0000
= 0.2171
-------
293
Root MSE
= .66128
1
Idist I
pfemales 1
pblack I
pother 1
page underS I
page_5_29 ;
page 65 up I
pmarhh chd I
pmhh child I
pfhh child I
pvacant 1
prenter occ 1
cons I
Coef .
-3.289739
6.708423
.6246435
22.60372
4.512753
10.09821
1 .996983
-11.40863
-2.727324
-7.573115
-.7672199
-1.948006
Robust
Std. Err.
.9466478
1.635602
.070074
1.054946
.5028299
.4471518
.198402
.5669918
.3696495
.6748127
.0758307
.5272493
-2
4
£
21
8
22
10
-20
-7
-11
-10
_ 3
1
t
.48 |
.10 !
.91
.43
.97
.58 '
.60
.12
.38
.22
.12
.69
P>lt 1
0
0
0
0
0
0
0
0
0
0
0
0
.001
.000
.000
.000
.000
.000
.003
.003
.000
.000
.000
.000
[55% Conf.
-5.145379
3. 50.228
.4372-528
20. 5. "53
3.527095
9.221597
1.627 ;79
-12.52006
-3.451-319
-9.8951398
-.9158'549
-2.981S32
Interval]
-1.434099
9.914566
.7620041
24.67165
5.498411
10.97473
2.366298
-10.2972
-2.002728
-6.250332
-.6185748
-.9144807
4.5 Distance to site vs. other distances
Regression with robust standard errors
Number of obs = 9211
F( 17, ;'193) = 1793.39
Prob > F = 0.0000
R-squareci = 0.8109
Root MSE - .32513
Idist
Id school
ld_retail
Id hospital
Id church
Id cemetery
Id 15
Id 1605
Id ilO
Id railroad
Id_s60
Id rivers,
Id cards
Id whittle rn
Id parks
Id rajwater
Id csula
Id cclubs
cons
Coef.
-.1322357
1.136024
-.1399455
-.3041922
-.0447594
-.135625
-.1456176
.0109707
-.0655516
.2688995
-.0012389
-.0202639
-.3163222
-.0039177
-.0689012
-.4005061
-.0363624
1.6459
Robust
Std. Err.
.0083541
.0340795
.0073205
.0135102
.0072388
.0069391
.0098454
.0067179
.0038653
.0127635
.0067469
.0016522
.028066
.0027232
.005308
.0137442
.0064279
.0636562
-21
34
-19
-22
-6
-19
-14
1
-16
21
-0
-12
-29
-1
-12
-29
-5
25
t
.31
.80
.12
.52
.18
.55
.79
.63
.36
.07
.18
.26
.09
.44
.98
.14
.73
. 86
P>|t I
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.102
o.ooc
0.000
0.854
0.000
0.000
0.150
0.000
1 o.ooo
0.000
0 .000
195% Conf.
-.1936117
] .119221
-.1542952
-.3306752
-.0539491
-.149227]
-.1649168
-.0021579
-.0731284
.2438801
-.0144642
-.0235027
-.8713379
-.0092557
-.079306
-.4274478
-.0494625
1 .521119
Interval]
-.1653593
1.252828
-.1255957
-.2777092
-.0305693
-.1220229
-.1263184
.0241394
-.0579748
.2939198
.0119865
-.0170252
-.7613066
.0014203
-.0584964
-.3735644
-.0242622
1.77068
Chapter 5 Trends in the distance gradient
These models use individual houses as observations. We associate with each house the
proportion of each group in the Census tract that contains the house. The right-hand side
variables are the measured distance of the house itself from the OH landfill site, a time trend, and
-------
294
an interaction term between distance and time. The simple trend picks up the trend over time in
the concentration of the group in question throughout the sample area. The "Idist" variable,
distance to the boundary of the landfill site, picks up any baseline distance gradient in the
concentration of the group in question as a function of distance from the site. The key variable is
the interaction term, which tells how the distance gradient is shifting over time. If the distance
gradient is becoming either less positive or more negative, the concentration of the group in
question nearer the Superfund site is growing, relative to the concentration further away.
5.1 Structural variables
5.1.1 Built post-1900
Regression with
i
notold I
Idist I
Idisty !
trend I
cons I
robust standard errors
Coef .
-.000214
-1.50e-06
6.72e-06
.9999013
Robust
Std. Err.
.0003268
.0000179
4.95e-06
.0000736
t
-0,65
-0.08
1 .36
P> 1 1 I
0.513
0.933
0.175
0.000
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
[95% Conf.
-.0008547
-.0000365
-2.99e-06
.999757
9211
0.67
= 0.5726
= 0.0002
= .01474
Interval)
.0004266
.0000335
.0000164
1.000046
5.1.2 Age if built post-1900
Regression with robust standard errors
Number of obs = 9211
F( 3, 9207) = 1546.57
Prob > F = 0.0000
R-squared = 0.2226
Root MSE = 16.421
1
age 1 Coe f .
Idist ! 7.436829
Idisty I -.0009755
trend [ .7720479
cons i 10.7639
5.1.3 Square footage
Robust
Std. Err. t
.30763B5 24.17
.016739 -0.06
.0229092 33.70
.407078 26.44
Regression with robust standard errors
P>|t! [95% Conf.
0.000 6.83379
0.954 -.0337876
0.000 .7271408
0.000 9.965937
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
Interval 1
S. 039869
.0318366
.816955
11.56186
9211
- 210.03
= 0.0000
= 0.0653
.4787
Robust
-------
295
sqft I
Idist I
Idisty I
trend 1
cons 1
Coef.
-.1779509
.0007751
-.0033883
1.609737
Std. Err.
.0146573
.0007512
.0009721
.019037
t I
-12.14 i
1 .03
-4 .00 |
84 .56
P>l 1 1
0.000
0.302
0.000
0.000
[95% Conf.
-.2066325
-.0006974
-.0057938
1.57242
, Interval]
-.1492192
.0022476
-.0019829
1.647053
5.1.4 Bedrooms
Regression with
1
bedrms I
Idist 1
Idisty i
trend 1
cons I
robust standard errors
Coef .
-.2387678
.0020337
-.007976
3.253231
Robust
Std. Err.
.0264259
.0013125
.0016546
.0331528
-9
1
-4
99
t
.04 .
.55
.32
.13
P>|t I
0.000
0.121
0.000
0.000
Number of obs
F( 3, )207)
Prob > F
R-squared
Roo- MSE
[95% Conf.
-.2905684
-.0005:91
-.01121 94
3.138244
9211
= 112.74
= 0.0000
= 0.0356
= .84276
Interval]
-.1869672
.0046064
-.0047326
3.318218
5.1.5 Bathrooms
Regression with
1
bthrms 1
Idist 1
Idisty I
trend |
cons 1
robust standard errors
Coe f .
-.3548969
.0006031
.0024499
2.211612
Robust
Std. Err.
.0224883
.0011436
.0014982
.0295537
-15
0
1
74
t
.78
.53
.64
.?3
P>|t I
0.000
0.598
. 0.102
0.000
Number of obs
F( 3, S207)
Prob > F
R-scuared
Root MSE
[95% Conf.
-.398979
-.0016337
-.000437
2.15358
9211
= 382.90
= 0.0000
= 0.0953
= .79396
Interval ]
-.3108149
.0028449
.0053868
2.269544
5.1.6 Fireplace(s)?
Regression with robust standard errors
Number of obs = 9211
F( 3, 9207) = 252.56
Prob > F = 0.0000
R-squared = 0.0626
Root MSE - .47374
Robust
1_
-------
296
fplace
Coef. Std. Err.
P>|t|
[95% Conf. Interval]
Idist 1
Idisty I
trend I
cons I
-.1819811 .0131149 -13.88
.0022162 .0006919 3.20
-.0086033 .0009755 -9.83
.6971983 .0165422 42.15
0.000 -.2076891
0.001 .00086
C.OOO -.0103195
0.000 .664772
-.156273
.0035725
-.0068871
.7296246
5.1.7 Floors recorded?
Regression with
1
knowfir I
Idist I
Idisty 1
trend 1
cons I
5.1.8 Floors
Regression with
floors 1
idist 1
Idisty 1
trand 1
cons I
5.1.9 Lotsize
Regression with
robust standard errors
Robust
Coef. Std. Err. t
-.0554189 .0092847 -5.97
.002548 .0005351 4.76
-.0089382 .0006572 -13.60
.9794968 .0107698 90.95
robust standard errors
Robust
Coef. Std. Err. t
-.2379908 .0169375 -14.05
.0038295 .0009294 4.12
-.0101069 .0012234 -8.26
1.282427 .0221347 57.94
robust standard errors
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
P>|t | [95% Conf.
0.000 -.0736189
0.000 .0014991
0.000 -.0102265
0.000 .9583857
Number of obs
Ft 3, 9207)
Prob > F
R-squared
Root MSE
P>|t| [95% Conf.
0.000 -.271192
0.000 .0020076
0.000 -.012505
0.000 1.239038
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
9211
97.87
= 0.0000
= 0.0217
.3912
Interval ]
-.0372188
.0035968
-.00765
1.000608
9211
= 191.70
= 0.0000
= 0.0640
= .54283
Interval 1
-.2047896
.0056515
-.0077088
1.325816
9211
41.13
= 0.0000
= 0.0126
= .45282
Robust
-------
lotsize
Coef.
Std. Err.
P> 111
[95% Conf. Interval]
Idist I
Idisty !
trend I
cons 1
-.0303502
.0026144
-.0080879
1.131305
.0126484
.0006575
.0008551
.0162807
-2.40 .
3 . 98 '
-9.46
69.49 |
0.016
O.OCO
0.000
0.000
-.0551439
.0013256
-.0097641
1.099391
-.0055565
.0039033
-.0064118
1.163219
5.2 Census tract attributes
5.2.1 Females
297
Regression with robust standard errors Nuitiber of obs = 9211
F( 3, 3207) = 20.26
Prob > F = 0.0000
R-squared = 0.0063
Root MSE - .0118
1 Robust
pfemaleg I Coef. Std. Err. t P>|tl [95* Conf. Interval)
Idist I -.0005075 .0003176 -1.60 0.110 -.00113 .0001149
trend 1 .0001003 .0000197 5.09 0.000 . 0000iil7 .0001389
Idisty I -.0000198 .0000155 -1.27 0.203 -. 00001.02 .0000107
cons 1 .5108233 .0003917 1303.98 0.000 .5100ii04 .5115962
.52-
cL
.5-
T
dist
Fitted pfemales by distance from OH site (km)
8.5
5.2.2 Whites
Regression with robust standard errors
Number of obs = 9211
F( 3, 9207) = 4210.76
Prob > F = 0.0000
R-squared = 0.5570
Root MSE = .13148
-------
1
pwhits I
Idist \
trend 1
Idisty I
cons I
Coef .
.0830403
-.0147613
-.0018906
.7094765
Robust
Std. Err.
.0033537
.0002263
.0001748
.0042602
24
-65
-10
166
t
.16
.24
.81
.54
e>it i
0.000
C.OOO
0.000
0.000
[95% Conf.
.0764663
-.0152048
-.0022333
.7011256
Interval ]
.0896142
-.0143177
-.0015479
.7178273
298
-pwtiite 70
- pv*ite_84
-pvrtiite 77
0-
T
8.5
dist
Fitted pwhite by distance from Oil site (km)
5.2.3 Blacks
Regression with
pblack 1
Idist I
trend |
Idisty 1
cons i
robust standard errors
Coef.
-.0007355
.0000988
.OOC019
.0047176
Robust
Std. Err.
.0001735
5.59e-06
7.39e-06
.OGC1203
-4
15
2
39
t
.53
.88
.57
.22
P>
0.
0.
0.
0.
1 1 1
000
000
010
000
Number of obs
F( 3, 9207}
Prob > F
R-squared
Root MSE
[95% Conf.
-.0011256
.0000779
4.49e-06
.0044818
9211
= 121.95
= 0.0000
= 0.0329
= .00534
Interval j
-.0004455
.0000998
.0000335
.0049533
-------
299
-pbtack 70
-pblack_84
-pblack 77
- pblackl91
.01 -
0-
0
dist
8.5
Fitted pblack by distance from OH site (km)
5.2.4 Other ethnic groups
Regression with robust standard errors
Number o:: obs = 9211
r( 3, '5237) = 4321.06
Prob > F = 0.0000
R-squared = 0.5660
Root MSE = .13174
pother
Idist
trend
Idisty
cons
1 i
1
1 Coef.
I -.0950076
1 .014992
; .0020202
1 .277383
nnthnr H4
Robust
Std. Err.
.0034053
.0002305
.0001769
.0043922
.< . oother 91
-24
65
11
63
t
.96
.03
.42
.15
P>|t 1
0
0
0
0
.000
.000
.000
.000
[95% Conf.
--0916&28
.0145'
.00167
.2687-,
01
34
34
, Interval]
-.0733324
.0154439
.002367
.2859927
1 -
0-
6
dist
Fitted pother by distance from OH site (km)
-------
5.2.5 Children under 5
300
Regression with robust standard errors
Number of obs = 9211
F{ 3, 9207) = 510.68
Prob > F = 0.0000
R-squared = 0.1224
Root MSE = .01663
page underS
Coef.
Robust
Std. Err.
P>lt I
[95% Conf. Interval]
Idist I .0063994 .0005023
trend I -.0004558 .0000322
Idisty 1 .0000226 .0000246
cons I .0798627 .0006704
12.74 0.000
-14.14 0.000
0.92 0.357
119.14 0.000
.0054147 .0073841
-.000519 -.0003927
-.0000256 .0000708
.07B5487 .0811768
- page_underS_70
- page_under5_84
- page_underS_77
- page_undei5_91
1 -
.s
I
a.
.04-
dist
85
Fitted page_under5 by distance from OH site (km)
5.2.6 Persons between 5 and 29
Regression with
page 5 29
Idist
trend
Idisty
cons
I
I
I
I
I
I
robust standard errors
Coef.
.009715
-.0025506
.0002393
.4400477
Robust
Std. Err.
.0010478
.0000747
.0000569
.0014033
t
9.27
-34.15
4.20
313.59
P>lt 1
0
0
0
0
.000
.000
.000
.000
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
195% Conf.
.0076612
-.002697
.0001277
.4372969
9211
= 894.76
= 0.0000
= 0.2081
= .04333
Interval)
.0117689
-.0024042
.0003508
.4427984
-------
301
2
s
-page 5_29_70
- pagej>_29_84
3-
T
- page 5 29_77
- page~5l29_91 |
dist
85
Fitted page_5_29 by distance from Oil site (km)
5.2.7 Persons between 30 and 64
Regression with
page_
I
30_64 I
Idist I
trend I
Idisty I
consj I
robust standard errors
Coef .
-.0240991
.0010147
.0003549
. 401974
Robust
Std. Err.
.001275
.0000792
.0000592
.0017344
-18
12
6
231
t
.90 .
.97
.10
.77
P> 1 1 1
0
0
0
0
.000
.000
.000
.000
Number of obs
F{ 3, 3207)
Prob > F
R-square'i
Root MSE
195*
-.0265
Con if.
:»84
.0008iil4
.0002
.3985
•;03
'•'42
9211
= 750.46
= 0.0000
- 0.1712
= .03882
Interval ]
-.0215998
.0011681
.000469
.4053737
• page_30_64 70
. page_30_6
-------
302
5.2.8 Persons 65 and older
Regression with
page_65 up !
Idist 1
t rend 1
Idisty I
cons I
robust standard errors
Coef .
.0053883
.0026715
-.0004772
.061345
Robust
Std. Err.
.00078
.0000623
.0000461
.0010659
6,
42.
-10
57
t
.91
,54
.35
.55
P> 1 1 1
0.000
0.000
0.000
0.000
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
|95% Conf.
.0038593
.0025484
-.0005676
.0592556
9211
= 981.78
= 0.0000
= 0.2373
= .03448
Interval]
.0069172
.0027946
-.0003869
.0634343
- page_65_up_70
- page_65_up_84
-page_65_up_77
-page~65~up_91
.16-
04-
dist
8.5
Fitted pageJ35_up by distance from Oil site (km)
5.2.9 Married heads of household
Regression with
pmarhh chd
Idist
trend
idisty
cons
I
I
I
E
I
robust standard errors
Coe f .
.0007957
-.0007432
.0009305
.3097138
Robust
Std. Err.
.002086
.0001309
.0000997
.0026691
t
0.38
-5. 68
9.33
116.04
P>
0.
0 .
0.
0.
It 1
703
000
000
000
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
[95% Conf.
-.0032933
-.0009998
.0007351
.3044818
9211
- 231.93
= 0.0000
= 0.0505
= .06238
Interval )
.0048848
-.0004866
.0011259
.3149458
-------
303
S
Q.
- pmarhti_chd_70
- pmarhh~chd_84
- pmarhh chd_77
- pmarhh~chd_91
.36-
dist
8.5
Fitted pmarhh_chd by distance from OH site (km)
5.2.10 Male-headed of household with children
Regression with robust standard errors
Number of obs = 9211
F( 3, 9207) = 1268.58
Prcb > F = 0.0000
R-squared = 0.2514
Root MSE = .0144
pmhh_child
:oef .
Robust
Std. Err.
F>|t I
[95% Conf. Interval]
Idist I .0040426 .0004067
trend I .0012375 .0000438
Idisty I -.0002995 .0000355
cons I .0044591 .0005502
9.94 0.000
25.38 0.000
-8.45 0.000
8.10 C.OOO
.0032.152 .0043399
.001K13 .0013331
-.000369 -.00023
.0033?i05 .0055377
- pmhh_child_70
• pmhh_child_B4
.06-
-.02-
- pmhh_child_77
• pmhh~child~91
T
dist
Fitted pmhh_child by distance from Oil site (km)
85
-------
304
5.2.11 Female-headed households with children
Regression with
pfhh_child I
Idist 1
trend |
Idisty 1
cons I
robust standard errors
Cosf .
.0048499
.0012721
.0000927
.0482733
Robust
Std. Err.
.0008963
.00005
.000042
.0010703
t P>
5.47 C.
25.42 0.
2.21 0.
45.10 0 .
1 t I
000
000
027
000
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
(95% Conf.
.0031125
.001174
.0000104
.0461753
9211
= 582.34
= 0.0000
= 0.1572
= .02981
Interval J
.0065873
.0013702
.000175
.0503714
-pfhh child 70
-pfhh_child>l
-pfhh child 77
-pfhh~chikf91
.1 -
.a
I
s
0-
tflSt
8.5
Fitted pfhh_child by distance from OH site (km)
5.2.12 Owner-occupancy
Regression with
powner occ
Idist
trend
Idist y
cons
1
1
1
1
1
1
robust standard errors
Coef .
.0033151
-.0044307
.0009346
.634293
Robust
Std. Err. t P>
Number of obs = 9211
F( 3, 9207) - 143.27
Prob > F = 0.0000
R-squared = 0.0361
Root MSE = .17435
t| [95* Conf. Interval!
.0043922 0.75 0.450 -.0052946 .0119243
.0002689 -16.48 0.000 -.0049578 -.0039035
.0002308 4.05 0.000 .0004821 .001387
.0051611 122.90 0.000 .6241761 .6444098
-------
305
-powner_occ 70
- powner_occji4
- powner_occ_77
- powner~occ_91 I
.4-
T
dist
! 8.5
Fitted powner_occ by distance from Oil site (km)
5.2.13 Renter-occupancy
Regression with
1
prenter occ 1
Idist 1
trend I
Idisty 1
cons 1
robus*. standard errors
Coe f .
.0040978
.004771
-.0012124
.3274087
. . prenter occ 70
Robust
Std. Err. t P> 1 t I
.0041508 G.99 C.324
.0002528 18.87 . 0.000
.0002187 -5.54 0.000
.0048078 68.10 0.000
prenter_occ_77
prenter occ 91
Number of obs = 9211
F( 3, 3207) = 162.12
Prob > F = 0.0000
R-squared = 0.0394
Root MSE = .16722
[95% Conf. Interval)
-.0040J86 .0122342
.0042754 .0052666
-. 001541 -.0007837
.3179343 .3368331
53-
.28-
dist
IBS
Fitted prenter_occ by distance from OH site (km)
-------
306
5.2.14 Vacancy rates
Regression with
1
pvacant I
Idist !
trend I
Idisty I
cons t
robust standard errors
Coef .
-.0074489
-.000332
.000279
.0380797
Robust
Std. Err.
.0003894
.0000258
.0000186
.0005776
-19
-12
15
65
t
.13
.89
.02
.92
P>|t I
0.000
0.000
0.000
0.000
Number of obs
F( 3, 9207)
Prob > F
R-squared
Root MSE
[95% Conf.
-.0082122
-.0003825
.0002426
.0369474
9211
= 132.89
= 0.0000
= 0.0373
= .01349
Interval ]
-.0066856
-.0002815
.0003154
.039212
-pvacant_70
-pvacant~84
-pvacant_77
-pvacant_91
06-
§
I
01-
dist
8.5
Fitted pvacant by distance from Oil site (km)
Chapter 6 Complete regression results - No lot size interactions
6.1 Just structural characteristics and year dummies
Regression with
1
Isprice I
notold 1
age I
age2 1
sqt't 1
sqff.2 1
robust standard errors
Coef.
. 9932554
-.0117411
.0001031
.5206245
-.0924193
Robust
Std. Err.
.1623343
.001059
.0000131
.0611521
.0242179
6.
-11.
7.
8.
-3,
t
.10
.09
.84
.51
.82
P>
0.
0.
0.
0.
0.
Itl
000
000
000
000
000
Number of obs
F( 72, 9138)
Prob > F
R-squared
Root MSE
(95* Conf.
.6740629
-.013817
.0000773
.4007527
-.1398917
9211
= 349.56
= 0.0000
= 0.6916
= .45753
Interval ]
1.312448
-.0096653
.0001288
.6404963
-.0449468
-------
307
bedrms
bthrms
sqftbed
sqftbth
fplace
knowf Ir
floors
lotsize
Idis70
Idis71
Idis72
Idis73
Idis74
ldis7B
Idis76
Idis77
ldis7B
Idis79
IdisBO
Idis81
Idis82
Idis83
Idis84
Idis85
Idis86
Idis87
Idis88
Idis89
Idis90
ldis9J
Idis92
Idis93
Idis94
Idis95
Idis96
Idis9l
Idis98
Idis99
year71
year72
year 73
year74
year75
year76
year77
year78
year79
yearSO
yearSl
yearB2
year83
year84
year 85
year86
year87
year88
year89
year90
year91
.07C2122
-.1172286
-.0386692
.0735727
.0709999
.1081705
-.0431627
.1345719
-.0910615
-.1911183
-.1452377
-.1446323
-.1543859
-.133872
-.0896486
-.0948632
-.0794666
-.0210521
-.1573536
-.1388767
.139C67
-.1294389
-.0254665
-.056664
.0014281
-.0586542
-.013604
.0896625
-.0356742
.1126622
-.0556043
-.0206699
-.0119439
.OC19911
.0468581
.0606048
-.004532
.0254911
.1859305
.2292775
.391793
.3391404
.562629
.6670246
.905679
1.182555
1.210936
1.296118
1.19299
1.064098
1 .551468
1.50C138
1.63059
1.703408
1.339991
1.918535
1.93175
2.179907
2.086958
.0283334
.0308256
.0185536
.0208992
.0121874
.0331057
.0208639
.0149515
.0679057
.0668653
.0434544
.0348781
.0485435
.0230817
.0151661
.0204214
.0176889
.040024
.0415527
.1325139
.0825229
.0394743
.0476241
.045126
.0319232
.0272223
.0538295
.0526624
.0370301
.0625144
.0214439
.0220965
.0194802
.0397644
.037562
.0346603
.0134719
.0118365
.1090268
.0952193
.0911333
.099699
.0910229
.0890339
.08994
.087627
.0984302
.0995991
.1745181
.1375375
.0950517
.1030402
.1062313
.0953771
.0919629
.10588
.1104169
.0997584
.1128417
2.48.
-3.801
-2.09
3.52,
5.82
3.27
-2 . 07 '
12.34
-1.34 '
-2.86
-3.34
-4 .15
-3.18
-5.80 :
-5.91
-4.65
-4.49 '
-0.53
-3.79
-1.05
1.69
-3.28 •
-0.53
-1 .26
0.04
-2.15
-0.25 ,
1 .70
-C .96
1 .80
-2 .59
-C.94
-C.61 '
C.05
1 .25
1.75
-0.34
2.15
1.71
2.41
4 .30
3.40
6.18
7.49
10.07
13.50
12.30
13.01 •
6.84
7 .74
16.32
14 .56
15.35
17.86
20.55
18.12 .
17.50
21.35
18.49 !
0.013
0 .000
0.037
0.000
0.000
0.001
0.039
0.000
0.180
0.004
0.001
0.000
O.OC1
0.000
o.oco
0.000
0.000
3.599
3.000
0.295
0.092
0.001
0.593
0.209
0.964
0.031
0.800
0.089
0.335
0.072
0.010
0.350
0.540
0.960
0.212
0.080
C.737
0.031
0.088
0.016
0.000
0.001
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
.0144.724
-.1776-536
-.075(385
.0326056
.0470099
.0432759
~. 084 C 60 6
.1552637
-.2241713
-.3221892
-.2304181
-.213C012
-.2495421
-.1791172
-.1193775
-.1348956
-.1141408
-.0995061
-.2388062
-.3986335
-.0226964
-.2068173
-.1188203
-.1451212
-.0611485
-.1120161
-.1191218
-.0135677
-.1082614
-.0098799
-.0976391
-.0639341
-.0501294
-.075356
-.0267718
-.0073371
-.03 :94
.002289
-.0277:s64
.0426265
.213i:>62
.1437081
.3842033
.4924S84
.7293765
1.010786
1.018041
1.100882
.850B95
.7944933
1.365: 45
1.298:57
1 .422:53
'-.516«48
1.709',23
:. .710987
:.7i5:;08
1.984: 59
1.865"; 63
.125752
-.0568035
-.0023
.1145398
.0947898
.1730652
-.0022649
.2138802
.0420488
-.0600474
-.0600574
-.0762634
-.0592297
-.0886267
-.0599197
-.0548327
-.0447924
.0574039
-.0759011
.1208801
.3008304
-.0520604
.0678874
.0317931
.0640047
-.0052923
.0919139
.1928927
.0369131
.2352044
-.0135694
.0226442
.0262415
.0799332
.1204879
.1295467
.0218759
.0486932
.3996474
.4159286
.5704397
.5345728
.7410541
.8415608
1.081982
1.354324
1.403931
1.491355
1.5350S4
1.333702
1.737791
1.70212
1.838827
1.890369
2.070259
2.126084
2.148192
2.375456
2.308153
-------
308
yearS2 I
year£3 1
year 94 I
year95 |
year96 1
year97 |
year98 I
year99 1
cons I
2
2
^
9
2
2
3
.349472
.251359
.233253
2.13355
.116251
.051297
2.1765
.227537
.267537
.0886803
.0904794
.0886592
.098177
.098453
.0981948
.0876417
.087748
. 1923249
26
24
25
21
21
20
24
25
42
.49
.69
.25
.73
.50
.89
.83
.39
.99
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.000
.000
.000
.000
.000
2
2
2
1
1
1
2
2
7
.175639
.074499
.064461
.941101
.923261
.358813
.004703
.055531
.890537
2.
2.
2.
2.
2.
2.
2.
2.
8,
523305
429219
412045
325999
309241
243781
348298
399542
644536
All
All
All
Hypothesis P-v
ofF
alue Reject @
-test 5% level?
structural attribute slopes simultaneously zero o . oooo
year-specific slopes on LDIST simultaneously zero o . oooo
year-specific slope on LDIST the same o . oooo
6.2 Including Census tract attributes
Regression with robust standard errors
Isprice
not eld
age
age2
soft
sqft2
bed cms
bthrras
sqftbed
sqftbth
fplace
knowflr
floors
lotsize
ldi37C
ldi;;71
Idis72
Idis73
Idis74
Idis75
Idis76
ldi.-77
Idis78
Idis79
IdisSO
Coef .
.9311328
-.0139166
.0001407
.4534617
-.0770347
.0683652
-.1222492
-.0352973
.06746
.0460746
.0915762
-.0295288
.1928214
-.0700751
-.1529778
-.1290165
-.1135123
-.1352636
-.1043923
-.068202
-.0724416
-.0589154
.0009394
-.147009
Robust
Std. Err.
.2032774
.0011205
.0000145
.0613922
.0239452
.0281657
.0309686
.0183522
.0208661
.012244
.0328087
.0208388
.0150136
.0707616
.0672047
.0436163
.0357361
.0486705
.0233475
.0153228
.0197573
.0176079
.0401474
.0413456
4
— 12
9
7
-3
2
-3
-1
3
3
1
-1
12
-0
_2
-2
-3
—2
-4
-4
-3
"•3
0
-3
t
.58
.42
.71
.39
.22
.43
.95
.92
.23
.76
.79
.42
.18
.99
.28
,96
.13
.78
.47
.31
.67
.35
.02
.56
P>|t I
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.001
.015
.000
.054
.001
.000
.005
.157
.000
.322
.023
.003
.001
.005
.000
.000
.000
.001
.981
.000
Number of obs = 9211
F{ 33, 9127) - 328.50
Prob > F = 0.0000
R-squared = 0.6997
Root MSE = .45174
[95% Conf
.5326636
-.016113
.0001123
.3331193
-.1239726
.013154
-.1829546
-.0712723
.0265578
.0220736
.0272639
-.0703775
.1533914
-.2087836
-.2847141
-.2145141
-.1835631
-.2306686
-.1501587
-.0992183
-.1111704
-.0934309
-.0777585
-.2280556
Interval]
1.329602
-.0117203
.0001692
.573804
-.0300968
.1235763
-.0615439
.0006766
.1033622
.0700757
.1558886
.01132
.2122515
.0686335
-.0212415
-.0435188
-.0434615
-.0398586
-.0586258
-.0371856
-.0337129
-.0243999
.0796374
-.0659624
-------
309
Idis81
Idis82
Idis83
Idis84
IdisSS
Idis86
Idis87
Idis99
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
pfemales
pblack
pother
page under5
page_5_29
page 65 up
pmarhh chd
pmhh child
pfhh_child
pvacant
prenter occ
year71
year72
year73
year74
year7S
year76
year77
year 7$
year79
year80
yearSl
year82
year83
year84
year85
year 86
year87
year88
year99
year90
year91
year92
year93
year94
year95
year96
year97
year98
year99
-.1C4919
.1326903
-.1165535
-.0051316
-.3381543
.3134244
-.0420247
,0005704
.1013377
-.0163981
.1236039
-.0376243
-.0055244
.0101151
.0168815
.0603449
.087042
.009177
.02755
2.747538
-1 .852562
-.0160533
-2.693753
-.2775C36
-.4607948
-.3810539
-.8717903
.3196293
.350096
.0058759
.1545344
.2277507
.3767912
.3498717
.5612083
.673521
.9169083
1.192805
1.233654
1.337426
1.185371
1 .C96352
] .586889
1.524242
1.653669
1.722433
1 .914389
1.950266
1.964867
2.206596
2.117776
2.367983
2.271292
2.237837
2.135686
2.123062
2.03405
2.169597
2.22936
.1317158
.0951089
.0390932
.0472946
.0447816
.0313907
.0268712
.0537216
. 0515"?77
.036353
.0632254
.0208473
.0218501
.0188412
.0397943
.0361389
.0339601
.0127379
.012406
.7050241
1.552152
.0617295
.8482998
.4197C39
.366569
.2313454
.3919165
.4317667
.5082449
.0578392
.1121572
.0984653
.0960932
.105628
.0989128
.0993022
.102101
.1023776
.1123752
.1123766
.1808823
.1523085
.1105634
.1178005
.1224609
.1123297
.1105949
.1233513
.1225789
.119912
.1311657
.1072534
.1095807
.1071719
.1174229
.1137846
.1151111
.1055907
.1068837
-3.80'
1.56
-2.98
-3.19i
-3.35
0.43
-1.56
0.01 '
1.98
-0.45;
1.95
-1.80
-0.25
0.54
0.42
1.67
2.56 :
0.72
2.22
3.90
-:. .19
-0.26
-3 . 17
-0.66
-1.26
-1.65 '
-2.22
0.74
0.69 '
0.10
1 .33 .
2.31
:- .92
3.31 .
E.67
6.73
6 .98
11.65 '
1C. 93
11 .90
€.55 •
7.20
14.35
12.94 ,
13.50
15.33
17.31
15.81
16.03
18.40
16.15
22.08
20 .73
20.88
18.19
18.66
17.67
20.55
20.86
0.426
0.119
0.003
0.847
0.394
0.669
0.118
0.992
0.043
0.652
0.051
0.071
0.800
0.591
0.671
0.095
0.010
0.471
0.026
0.000
0.233
0.795
0.002
0.509
0.209
0.100
0.026
0.459
0.491
0.919
0.168
0.021
0.000
0.001
0 .000
0.000
0 .000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0 .000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
-.3631116
-.0341417
-. 193185
-.101E396
-.125^663
-.0481084
-.0946983
-.1C47358
.0006338
-. 0876581
-.0003321
-.0784906
-.0483555
-.0263178
-.0611242
-.0104954
.0204726
-.0157922
.0032315
1.365532
-4.895127
-.137057
-4.353611
-1 .100217
-1.179352
-.8345427
-1.640335
-.5267301
-.6461778
-.1075019
-.0653187
.0347367
.183427
.1428172
.367317
.4788565
.7167675
.9921225
1.013374
1.117143
.8308314
.7977936
1.37316
1.293327
1.413318
1.502.242
1.697598
1.70347
1.724.J95
1.971542
1.860662
2.157''37
2.056.189
2.027756
1.90!i51
1.900019
1 -808.;.07
1.9621.15
2.019H44
.1532735
.2995233
-.039922
.0835764
.0495976
.0749572
.0106489
.1058767
.2030416
.0548619
.2475398
.003242
.0373066
.047043
.0948873
.1311851
.1536114
.0341462
.0518684
4.129543
1.190003
.1049504
-1,027896
.54521
.2577625
.0724349
-.1035467
1.165989
1.34637
.1192538
.3743876
.4207647
.5651555
.5569261
.7550997
.8681756
1.117049
1.393488
1.453935
1.55771
1.539941
1.394911
1.333618
1.755158
1.893719
1.942625
2.13118
2.192062
2.205149
2.441651
2.37489
2.578238
2.436094
2.447918
2.365861
2.346105
2.259694
2.376573
2.438876
-------
310
cons I 7.501291 .4593043
16.33 0.000
6.600951
8.401631
AH
All
All
All
Hypothesis P-v:
ofF
due Reject @
-test 5% level?
structural attribute slopes simultaneously zero o . oooo
year-specific slopes on LDIST simultaneously zero o . oooo
year-specific slope on LDIST the same o . oooo
Census tract characteristic effects simultaneously zero o . oooo
6.3 Including other distances
Regression with robust standard errors
Isprice
noto Id
age
age2
sqft
sqft2
bedrms
bthrms
sqf tbed
sqftbth
fplace
knowf Ir
floors
lotsize
Idis70
Idis71
Idis72
Idis73
Idis74
Idis75
Idis76
Idis77
ldis-73
Idis79
IdisBO
IdisSl
Idis82
Idis83
Idis84
Idis85
Cosf .
1.001211
-.0130169
.0001135
.4567512
-.077342
.0698487
-.1118675
-.0363684
.0640609
.060934
.0898725
-.02346B4
.1839993
-.1273846
-.2088247
-.1767336
-.1642089
-.1841929
-.1461116
-.1182289
-.1214819
-.0968739
-.0487484
-.1826636
-.1628725
.1040464
-.1617318
-.0593314
-.0878825
Robust
Std. Err.
.1977214
.0011016
.0000139
.0604111
.0233334
.0281143
.0306591
.0183891
.0206107
.0121563
.0330778
.020879
.0152313
.0742849
.0700265
.0463757
.0393582
.0525649
.0263498
.0192546
.023741
.0222307
.0431883
.042573S
.1332603
.089668
.042452
.0505849
.0475759
5
-11
9
7
— 2
2
-3
-1
3
5
2
-1
12
-1
— '.?
_ t
-4
_3
-5
-6
-5
-4
-1
-4
-1
1
_ j
-1
-1
r
.06
.82
.19
.56
. 31
.48
.65
.98
.11
.01
.72
.12
.08
.71
.98
.81
.17
.50
.55
.14
.12
.36
.13
.29
.22
.16
.81
.17
. 65
P>|t I
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.001
.013
.000
.048
.002
.000
.007
.261
.000
.086
.003
.000
.000
.000
.000
.000
.000
.000
.259
.000
.222
.246
.000
.241
.065
Number of obs = 9211
F( 89, 9121) = 295.22
Prob > F = 0.0000
R-squared = 0.6983
Root MSE = .45296
[95% Conf
.6136332
-.0151762
.0000863
.3383319
-.1230808
.0147384
-.1719662
-.0724151
.0236593
.0371049
.0250325
-.0643959
.1541426
-.2729995
-.3460924
-.2676403
-.2413597
-.2872319
-.1977632
-.1559726
-.1680195
-.140451
-.1334072
-.2661275
-.4240926
-.071723
-.2449473
-.1584891
-.1811419
Interval J
1.3BS79
-.0108575
.0001408
.5751705
-.0316032
.124959
-.0517687
-.0003217
.1044624
.0847631
.1547125
.0174591
.2138559
.0182304
-.071557
-.0858269
-.0870581
-.0811539
-.09446
-.0804853
-.0749412
-.0532968
.0359105
-.0991996
.0983476
.2798158
-.0785163
.0398263
.0053769
-------
311
Idis86
Idis87
Idis88
Idis89
Idis90
IdisSl
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
Id school
ld_retail
Id hospital
Id church
Id cemetery
Id 15
Id 1605
ld_110
Id railroad
Id_s60
Id rivers
Id cards
Id whittiern
Id parks
Id mjwater
Id csula
Id cclubs
year71
year72
year73
year74
year75
year76
year77
year78
year79
yearSO
yearS 1
year 82
year83
year 84
ye a r 8 5
year86
year 87
yearSfi
year89
year90
year 91
year92
year93
year94
year95
year 96
year97
year98
-.0369571
-.089046
-.0460065
.0510108
-.0736167
.076444
-.0848735
-.0528936
-.C433073
-.C316994
.014926
.0356117
-.0343556
-.0149839
.007193
.0476416
-.0142779
-.0327921
.0624055
.0569511
.0598053
-.0319407
.0261523
.0105443
.0263751
.0163784
.0725796
-.0062075
-.0137087
.1005911
-.0249674
. 1550973
.2168663
.3770868
.3377187
.5337655
.6657166
.8984543
1.159694
1.204526
1.295781
1.181369
1.05633
1.554778
1.498359
1.623616
1.706289
1.887947
1.917709
1 .941076
2.19021
2.Q8-M04
2.34227
2.246327
2.230809
2.135439
2.115704
2.045975
2.178776
.0350382
.031852
.0568567
.0495855
.038476
.0687475
.0261137
.0272822
.0249046
.0437923
.040119
.037133
.0195687
.0200047
.0091592
.0380076
.0110092
.0158547
.0130525
.0155329
.0154894
.0112692
.0074647
.0131657
.0125085
.002612
.0366962
.0048469
.0100736
.0283902
.0111858
.1147808
.100495
.0971175
.105214
.0971142
.0955676
.0957098
.0939853
.1041701
.1046955
.1760032
.1470438
.1010053
.1084669
.1107628
.1003312
.0973966
.1110131
.1152451
.1047923
.1178713
.0944174
.0963544
.0945247
.1038255
.1033699
. 103889
.0936741
-1.05
-2.801
-C.91
1.03,
-1.91
1.11,
-3.25
-1.94,
-1.74'
-C.72
0.37'
C.96
-1.76i
-0.75
0.79
1.25
-1.30'
-2.07
4.86'
3.67
3.86'
-2.83
3.50.
0.80
2.11.
5.27
1.98
-1.28'
-1.36
3.54
-2.23
1.35
2.16.
3.88
3.21
5.50
6.97
9.39 '
12.34
11.56
12.38
6.71 ,
7 .19
15.39 .
13.81
14 .66
17 .01 :
IS. 38
17.27
16.84
20.90
17.71
24.81
23.31
23.60
20.57
20.47
19.69 '
23.26
0.292
0.005
0.418
0.304
0.056
0.266
0.001
0.053
0.032
0.469
0.710
0.338
0.079
0.454
0.432
0.210
0.195
0.039
0.000
0.000
0.000
0.005
0.000
0.423
0.035
0.000
0.048
0.200
0.174
0.000
0.026
0.177
0.031
O.OCC
0 .001
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0 .000
0 .000
0 .000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
-.1056399
-.151483
-.1574583
-.0461878
-.1490382
-.0583166
-.1360622
-.1063729
-.0921258
-.1171)421
-.0637163
-.0371773
-.0727147
-.0541976
-.010'-6U
-.0268619
-.035(1584
-.0638708
.0378196
.0265031
.029^,425
-.OS3--309
.0115:198
-.0152634
.0018556
.0112582
.0006467
-.0157036
-.0334551
.04494
-.0466941
-.0695987
.0198736
.1867147
.1314758
.3433999
.4783826
.7108417
.9754614
1.000329
1.090555
.8363637
.7636409
1.356785
1.28574
1.406496
1.509618
1.697328
1. 700099
1.715169
1.984793
1.35535
2.157:91
2.057-151
2.04552
1.931 '31 8
1.913D76
1.842H29
1. 995154
.0317257
-.0266091
.0654453
.1482094
.0018048
.2112046
-.0336848
.0005856
.0055113
.0541433
.0935682
.1084007
.0040036
.0242293
.0251471
.1221451
.0073026
-.0017134
.0889913
.0873991
.0901681
-.0097504
.0407849
.036352
.0508946
.0214936
.1445124
.0032936
.0060378
.1562423
-.OC30407
.3800934
.4138589
.5674588
.5439616
.7241311
.3530505
1.086067
1.343926
1.408723
1 .501008
1.526375
1.345118
1.752771
1.710979
1.840736
1.902961
2.078866
2.13532
2.166982
2.395626
2.318458
2.527349
2.435204
2.416099
2.33896
2.318332
2.24962
2.362399
-------
312
year99 I 2.239947 .0935955 23.93 0.000 2.056479 2.423416
cons I 7.920597 .2422357 32.7D 0.000 7.445761 8.395433
All
All
All
All
Hypothesis P-v;
ofF
due Reject @
-test 5% level?
structural attribute slopes simultaneously zero o . oooo
year-specific slopes on LDIST simultaneously zero o . oooo
year-specific slope on LDIST the same o . oooo
other distance effects simultaneously zero o . o o o o
6.4 Including both other distances and tract attributes
Regression with robust standard errors
\
Isprice Coef .
notioid .950093
age
age2
sqft
sqft2
bedrms
bthrms
sqf tbed
sqftbth
f place
knowf Ir
floors
lotsize
Idis70
Idis71
Idis72
Idis73
Idis74
Idis75
Idis76
Idis77
Idis79
Idis79
IdisSO
idis 8 1
Idis82
ldi?83
-.0144509
.0001413
.4376641
-.0711489
.0666789
-.1108145
-.0348318
.0610939
.0460638
.0709631
-.0176153
.1772867
-.0874608
-.1681922
-.1442436
-.1307234
-.1532548
-.1233935
-.0919088
-.0926112
-.0743741
-.0174455
-.1583898
-.1199943
.1175408
-.1318143
Robust
Std. Err.
.2208615
.0011484
.0000149
.0607247
.0232073
.0278111
.0309102
.0181468
.0207373
.0123143
.0329042
.0207367
.0150853
.074724
.0703947
.0475576
.0400035
.053254
.0281761
.0207793
.0251135
.0240642
.0436411
.0433717
.1351424
.0892852
.0425434
4
-12
9
7
-3
2
— 3
-1
2
3
o
-0
11
-1
-2
-3
-3
-2
-4
-4
_3
-3
-0
_3
-0
1
-3
t
.30
.58
.52
.21
.:7
.40
.59
.92
.95
.74
.16
.85
7 5
.17
.39
.03
.27
.88
.33
.42
.69
.09
.40
.65
.89
.32
.10
P
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
>ltl
.000
.000
.000
.000
.002
.017
.000
.055
.003
.000
.031
.396
.000
.242
.017
.002
.001
.004
.000
.000
.000
.002
.689
.000
.375
.188
.002
Number of obs = 9211
F(100, 9110) = 282.52
Prob > F = 0.0000
R-squared = 0.7021
Root MSE = .45036
[95% Conf
.517145
-.0167019
.0001122
.31863
-.1166404
.0121628
-.1714053
-.0704036
.0204442
.021925
.0064635
-.0582638
.1477152
-.2339366
-.3061815
-.2374672
-.2091393
-.2576445
-.1786249
-.1326409
-.1418393
-.1215453
-.1029918
-.243408
-.3849037
-.0574781
-.2152186
Interval ]
1.383021
-.0121999
.0001704
.5566982
-.0256574
.1211949
-.0502236
.00074
.1017436
.0702026
.1354628
.0230333
.2068581
.059015
-.0302028
-.0510201
-.0523075
-.0488651
-.0691621
-.0511767
-.0433831
-.0272029
.0681008
-.0733716
.1449152
.2925598
-.0484099
-------
313
Idis84
Idis85
Idis86
Idis87
Idis88
Idis89
Idis30
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
pfemalas
pblack
pother
page underS
page_5_29
page 65 up
pmarhh chd
prahh child
pfhh~child
pvacant
prenter occ
Id school
ld~retail
Id hospital
Id church
Id cemetery
Id 15
Id 1605
ld_110
Id railroad
Id_s60
Id rivers
Id cards
Id whittiern
Id parks
Id mjwater
Id csula
ld_cclubs
year 71
year72
year73
year74
year75
year76
year77
year78
year? 9
year 80
year 81
year82
year83
year84
-.0311296
-.0564328
-.00155914
-.0545578
-.0131781
.0864298
-.033315
.1109849
-.0472435
-.0180475
-.0038159
.0035955
.0459027
.0734012
-.0040276
.0101689
1.64439
-.6662112
-.166078
-2.453882
.2766206
-.0038523
-.3355717
-.371012
-.1693681
.4702613
-.1179529
.0105924
-.0014932
.0026951
-.0171211
.0242556
-.0023033
.0702899
-.040326
.0097142
.0071373
.0427592
.0158733
.0150716
-.0046835
-.0172807
.0823504
-.0122514
.1588892
.2353756
.4002031
.3827154
.5991426
.7303127
.9775283
1.251705
1.293517
1.40171
1.256444
1.163924
1.659741
1.603451
.C510418
,0476447
.035916
.0325945
.0572352
.0497325
.0386042
.0687847
.0270797
.0280293
.0256759
.0440419
.0405971
.0371506
.0203746
.0211885
.7910702
1.855665
.0879576
.9112003
.4944557
.4456985
.2471527
.4212083
.4980972
.5623476
.0662052
.0094185
.0402856
.0113842
.0164738
.015568
.0176145
.0227337
.0129888
.0078631
.0145953
.0134791
.0026404
.0401881
.0049686
.0111405
.0356364
.0116803
.1141079
.0995915
.0976332
.1078895
.1029001
.1043307
.1078208
.1091715
.1193009
.1192476
.1872487
.1587979
.1185764
.125543
-C.61
-1.18
-o.i el
-1.67
-0.23
1.741
-0.86
1.61
-1.741
-0.64
-0.15
0.08:
1.13
1.98
-0.201
0.48
2.08
-0.36
-1.89
-2.691
0.56
-0.01
-1.36!
-0.38
-0.34
0.84'
-1.78
1.12'
-0.04
3.24
-1.04'
1.56
-0.16
3.09:
-3.10
1.24
0.49'
3.17
6.01
0.38 '
-0.94
-1.55
2.31 '
-1.05
1.39
2.36
4.10
3.55
5.82
"'.00
9.07
11.47
10.88
11.75.
6.71
7.33
14.00 •
12.77
0.542
0.236
0.876
0.094
0.818
0.082
0.388
0.107
0.031
0.520
0.882
0.935
0.258
0.048
0.843
0.631
0.038
0.72C
0.059
0.007
0.576
0.993
0.175
0.378
0.733
0.403
0.075
0.261
0.970
0.813
0.299
0.119
0.874
0.002
0.002
0.217
0.625
0.002
0.000
0.708
0.346
0.121
0.021
0.294
0.164
0.013
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
-.131183
-.1493272
-.0759948
-.1181504
-.1253719
-.011057
-.108 3879
-.0233486
-.1003259
-.072-3911
-.054:464
-.0827366
-.033-5766
.0005776
-.0439664
-.0313653
.0937146
-4 .30373
-.338<:948
-4 .240039
-. 692(i234
-.8775214
-.8200465
-1. 196675
-1.14625
-.6320662
-.2477299
-.01787
-.080-: 671
-.019€20<
-.049^134
-.0062612
-.037I-.316
.0251267
-.06E787
-.0056991
-.0214729
.016337
.01C69S
-.0637062
-.0144231
-.0391186
.012495
-.0351474
-.064788
.0401539
.2038251
.1712279
.3974354
.5258011
.7661754
1.0377C4
1.06466
1.167957
.8893941
.8526443
1.427305
1.357349
.0689237
.0369616
.0648119
.0093348
.0990158
.1839166
.0423579
.2458184
.0058388
.0368962
.0465146
.0899276
.1254821
.1462248
.0359113
.0517031
3.195065
2.971303
.0063387
-.6677252
1.245865
.8698167
.1489031
.4546508
.8065141
1.572589
.0118241
.0290549
.0774707
.0250106
.0151712
.0547724
.031725
.1148531
-.014865
.0251276
.0357474
.0691814
.0210495
.0938493
.005056
.0045571
.1522058
.0106445
.3825663
.4305974
.5915911
.594203
.8008498
.9348243
1 .188881
1 .465706
1.532373
1.635462
1.623493
1.475203
1.892178
1.849554
-------
314
year85
year86
yearS?
year88
year89
year90
year91
year92
year 93
year 9 4
year95
year96
year97
year 98
year £9
cons
1.
1.
9
2.
2.
2.
2.
2.
2
2.
2.
2.
2.
2.
2.
7.
73555S
812839
0053S9
044687
063719
309781
214955
462505
.36903
342507
246307
235956
150304
291612
356637
635599
,1302581
.1205921
.1199308
.1317092
.1330474
.1294271
.1391015
. 1178963
.120211
.1175678
.1272988
.1239523
.1258222
.1171559
.1181444
.487419
13
15
16
15
15
17
15
20
19
19
17
18
17
19
19
15
.32
.03
.72
.52
.51
.85
.92
.99
.71
.92
.65
.04
.09
.56
.96
.67
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
1.
1.
1.
1.
1.
o
1.
2.
2
2.
1.
1.
1.
2
2.
6.
480223
576452
770298
786508
802916
056075
942285
231402
.13339
112048
996773
992982
903664
.06196
127048
680148
1
2
O
2
2
2
2
2
2
2
2
2
2
2
8
.990893
.049227
2.24048
.302867
.324522
.563497
.487625
.693608
.604671
.572966
.495841
.478931
.396943
.521264
.590227
.591049
All
All
All
All
All
Hypothesis P-vs
ofF
ilue Reject @
-test 5% level?
structural attribute slopes simultaneously zero o . oooo
year-specific slopes on LDIST simultaneously zero o . oooo
year-specific slope on LDIST the same o . oooo
other distance effects simultaneously zero o . o o o o
Census tract characteristic effects simultaneously zero o . oooo
Chapter 7 Complete regression results - With lot size interactions
7.1 Just structural characteristics and year dummies
Regression with robust standard errors
Isprice I
no to Id 1
age |
aqe2 I
sqft I
sqft2 1
bedrms I
bthrtns 1
sqftbed I
sqftbth I
Coef .
.9988356
-.0116319
.0001005
.5209625
-.0943504
.0622435
-.1142465
-.0343525
.0699198
Robust
Std. Err.
.1701005
.0010565
.0000132
.0610285
.0239555
.0282524
.0311376
.0185268
.0211335
r-.
-11
7
9
-3
2
-3
-1
n
t
.37
.01
.61
.54
.94
.20
.67
.85
.31
F>|t|
u
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.000
.028
.000
.064
.001
Number of obs = 9211
F(102, 9108) = 260.26
Prob > F = 0.0000
R-squared = 0.6943
Root MSE = .45625
!95% Conf.
.6654003
-.0137029
.0000747
.401333
-.1413086
.0068625
-.1752831
-.0706693
.0284935
. Interval]
1 .332271
-.009561
.0001264
.6405921
-.0473923
.1176246
-.0532099
.0019643
.1113461
-------
315
fplace
knowf Ir
floors
lotsize
IdislO
Idisll
Idis72
Idis73
Idis74
Idis75
ldis?6
Idis77
Ic(is78
Idis79
IdisSO
IdisSl
Idis8,2
Idis93
Id is 8.4
IdisBS
Idis86
Idis87
Idis88
Idis89
Idis90
Idis91
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
vldis70
vldis7l
vldis7£
vldis73
vldis74
vldis75
vldis76
vldis77
vldis78
vldis79
vldisSO
vldisSl
vldis82
vldis83
vldis84
vldis8S
vldisS6
vldis87
vldis88
vldis89
vldis90
vldis91
vldis92
vldis93
vldis94
.0697511
.1009321
-.0382573
,2435939
-.2699964
-.1484632
- .0258969
-.076537
-.1172235
-.0053131
-.1384437
-.0923926
.0181055
-.0274318
-.1004064
.3079565
.2359055
-.0705092
.1435025
.0627717
.1827039
-.0971432
-.0517317
.1526253
.1069724
.3504297
-.0094538
.0096591
-.0234191
.1405376
.1152141
.099792
.0343551
.0396E28
.1845399
-.0395453
-.1149431
-.0611215
-.0303966
-.1061756
.0476323
-.0046285
-.088997
.OC53975
-.0568727
-.4911717
-.0713744
-.0598326
-.1691355
-.1112468
-.175329
.0355179
.0365849
-.0646925
-.152776
-.247712
-.0444869
-.0344438
.0109283
.0121702
.0332847
.0208782
.0258695
.1477605
.1242143
.0931448
.0841739
.0895339
.0619588
.0458815
.0413224
.0441523
.0817553
.0776648
.2908649
.1239274
.0952766
.0898464
.0706398
.1023375
.049881
.0791622
.0757524
.0996711
.1040185
.0462287
.0405834
.048441
.0563654
.0705214
.0505463
.0320405
.0248585
.134578
.084321
.0808118
.080204
.0685864
.0471282
.0439517
.0313049
.043431
.066298
.0700676
.3155184
.0611182
.0955887
.0797681
.0433113
.1011475
.0406835
.0634283
.057548
.111741
.094244
.0377843
.0355211
.040978
5.73
3.03i
-1.83
9.42.
-1.83
-1.20!
-0.23
-0.91|
-1.31
-0.09
-3.021
-2.24
0.41'
-0.34
-1.29!
1.10
1.90,
-3.74
1.60.
0.89
1.79
-1.95
-0.65
2.01
1.07
3.37
-0.20
0.24
-0.48
2.49
1.63 '
1.97
1.07
1.60
1.37
-0.47
-1.42
-0.76
-0.44
-2.25
1 .08 •
-C.15
-2.05 .
C.08
-C.81
-1.56
-1.17
-0.63
-2.12
-2.57
-1.73
0.87
0.58
-1.12
-1.37
-2.63
-1.18 '
-0.97
0.27
0.000
0.002
0.067
0.000
0.068
0.232
0.781
0.363
0.190
0.932
0.003
0 .025
0.682
0.737
0.196
0.273
0.057
0.459
0.110
0.374
0.074
0.052
0.513
0.044
0.283
0.001
0.838
0.812
0.629
0.013
0.102
3.048
0.284
0.111
0.170
0.639
0.155
0.446
0 . 658
C.024
0.279
0.882
0.040
0.935
0.417
0.120
C.243
0.531
0.034
0.010
0.083
0.383
0.564
0.261
0.172
0.009
0.239
0.332
0.790
.045!?947
.035>.i865
-.0791833
.192:i838
-.559I.S401
-.3919511
-.2C8.-=316
-.2415.368
-.292"298
-.126' 664
-.228:^318
--.173I-.938
-.068443
-.187f,906
--.2526469
-.2427017
-.00702
-.257^:727
-. 0326165
-.075698
-.017S-006
-.194S'212
-.2065074
.0041337
-.0884052
.1465301
-.100C724
-.0698935
-.1183742
.0300488
-.0230237
.00071
-.0284515
-.0090754
-. 0791142
-.2048335
-.2733323
-.2183393
-.1648413
-.1985574
-.0385229
-.0659931
-.1741315
-.1245615
-.1942208
-1.109559
-.1911799
-.2472079
-.3254988
-.1961.167
-.3736008
-.0442309
-.0877489
-.1774995
-.3718135
-.4324515
-.1185527
-.1040"82
-.0693978
.0936076
.1661777
.0026686
.294309
.0196474
.0950248
.1566877
.0884628
.0582827
.1161401
-.0485056
-.0113913
.1046539
.1328269
.0518341
.8584146
.4788309
.1162543
.3196216
.2012415
.3833085
.0006347
.1034441
.3011169
.3C23501
.5543294
.0811649
.0892117
.0715361
.2510263
.253452
.198874
.0971616
.088381
.4484918
.1257429
.0434662
.0960963
.104048
-.0137938
.1337874
.0567362
-.0038625
.1353566
.0804755
.1273153
.0434311
.1275427
-.0127721
-.026347
.0229428
.1152668
.1609187
.0481145
.0662616
-.0629726
.0295789
.0351805
.0912543
-------
316
vldis95
vldis96
vldis97
vidis93
vldis99
year71
year72
year? 3
year74
ye a r 7 5
ye a r 7 6
year77
year78
year? 9
yearSO
year81
year82
year 83
yearS4
year £5
year 66
year 67
year 88
year89
year90
year91
year 92
year93
year94
year95
ye a r 9 6
year97
year98
year 99
cons
-.1417644 .0536202
-.0709976 .0443198
-.0419938 .0332385
-.0399856 .028989
-.0162408 .0233971
.197388 .1101488
.2434049 .0970738
.399473 .0928372
.3445368 .1014587
.5585659 .0930854
.6772712 .0913233
.9209143 .0914585
1.192673 .0892969
1.224925 .1003313
1.313545 .1014507
1.233937 .1794308
1.053051 .1407317
1.571066 .0965234
1.512784 .1047431
1.641528 .1075886
1.722435 .0971564
1.912529 .0935988
1.936966 .1065382
1.951962 .11225
2.203294 .1008454
2.107972 .1128124
2.365667 .0902217
2.274759 .0921271
2.258237 .0902125
2.154008 .0994145
2.137511 .0994495
2.073228 .0993347
2.196374 .089126
2.248859 .0893183
8.200771 .1999337
-2.64 0.008
-1.60 0.109
-1.26 0.207
-1.33 0.168
-0.69 0.488
1.79 0.073
2.56 0.011
4. 3D 0.000
3.40 0.001
6.00 0.000
7.42 0.000
10.07 0.000
13.36 0.000
12.21 0.000
12.95 0.000
6.88 0.000
7.48 0.000
16.28 0.000
14.44 0.000
15.26 0.000
17.73 0.000
20.43 0.000
18.18 0.000
17.39 0.000
21.85 0.000
13.69 0.000
26.22 0.000
24.69 0.000
25.03 C.OOO
21.67 0.000
21.49 0.000
20.87 0.000
24.64 0.000
25.13 0.000
41.02 0.000
.2468721 -.0366568
.1578744 .0158792
.1071388 .0231711
.0968105 .0168393
.0621045 .0296228
.0185283 .4133044
.0581184 .4386914
.2173737 .5315723
.1456549 .5434186
.3760975 .7410342
.4982571 .8562854
.7416352 1.100193
1.017632 1.367715
1.028254 1.421597
1.114679 1.512411
.8822121 1.585662
.7771856 1.328917
1.381858 1.760273
1.307464 1.713104
1.43063 1.852426
1.531987 1.912884
1.729054 2.096003
1.728127 2.145804
1.731927 2.171997
2.005615 2.400974
1.886834 2.329109
2.188812 2.542521
2.09417 2.455349
2.08145 2.435124
1.959133 2.348883
1.942568 2.332455
1.87851 2.267947
2.021668 2.371081
2.073775 2.423943
7.808856 B. 592686
Hypothesis
All structural attribute slopes simultaneously
zero
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same
All lotsize-dependent year-specific slopes on
LDIST
P-value Reject @
ofF-test 5% level?
0.0000
0.0000
0.0000
0.0176
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on
vX Idist variables)
0.0342
-------
317
7.2 Including Census tract attributes
Regression with robust standard errors '
Isprice
notold
age
age2
sqft
sqf tt2
bedrms
bthrrns
sqftbad
sqftbth
fplace
knowflr
f loots
lotsize
ldis?0
Idis71
Idis72 1
Idis73 1
Idis74 ;
Idis75
Idis76
Idis77
ldis7S
Idis79
IdisSO
IdisSl
Idis8j2
Idis83
Idis8<4
Idis85
Idis86
Idis87 1
IdisBB
Idis89
ldis9D
IdisSl
Idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis98
Idis99
vldis70
vldis71 !
vldis72 !
vldis73
1
| Coe f .
I .9165522
| -.0131541
| .0001286
I .4608774
I -.0807706
1 .0612303
-.1176677
t -.0314833
! .0640482
! .0439777
! .0783876
I -.0235397
1 1.724679
I -.2826613
1 -.1750469
I -.0931239
I -.1058343
| -.1763794
I -.0455593
1 -.1925953
-.1263481
1 -.0669354
[ -.0794996
! -.1375689
I .2662066
1 .1086019
| -.0833953
1 .0843799
I .0163124
1 .1540321
1 -.1310279
I -.0579894
I .1368076
1 .1163003
1 .3203736
-.0127485
I .0063002
[ -.0048699
1 .1543803
1 .1441416
I .1233079
1 .0540219
I .0554174
1 .2118703
I .0389604
1 -.0349145
I -.0087492
Robust
Std. Err.
.2052025
.0011457
.000015
.0615074
.0235728
.027982
.0312911
.0182654
.02111
.0122687
.0333356
.0209715
.9375752
.156844
.1380508
. 100128
.0875336
.0950248
.0625504
.0450261
.0425396
.0443613
.0835965
.0822936
.279905
.1269789
.0907232
.0868951
.0699395
.1066145
.0507077
.0799171
.072131
.0989829
.1053835
.0486472
.0384857
.0511578
.0580075
.0703128
.0501519
.0338889
.0238101
.1432855
.0957227
.0881181
.0785328
i
t
4.47'
-11.48
8.58'
7.49
-3.43
2.19
-3.76,
-1.72
3.03
3.581
2.35
-1.12,
1.84
-1.801
-1.27
-0.93
-1.21
-1.86.
-0.73
-4.28
-2. 97'
-1.51
-D.95
-2.28
0.95.
0.86
-0.92
0.97 '
0.23
1.44 •
-2.58
-0.73 .
1.90
1.17
3 . 04 '
-0.26
0.16.
-0.10
2.67
2.05
2.56
1.59
1.92
1 .48
0.20
-0.40 '
-C .11
P> 1 1 I
0.000
0.000
0.000
0.000
0.001
0.029
0.000
0.035
0.002
0.000
0.019
0.262
0.066
0.072
0.205
0.352
0.227
0.063
0.466
0.000
0.003
0.131
0.342
0.023
0.342
0.392
0.358
0.332
0.816
0.149
0.010
0.468
0.058
0.240
0.002
0.793
0.870
0.924
0.008
0.040
0.011
0.111
0.054
0.139
0.843
0.692
0.911
Number of obs = 9211
F(124, 9036} = 236.96
Prob > F = 0.0000
R-squarsd = 0.7034
Root MSE = .45001
[95% Conf,
.5143091
-.0153999
.000:3992
.340309
-.1269786
.0063793
-. 1790053
-.0672876
.022 5678
.0199282
.0130423
-,064i5435
-.1131791
-.590:.109
-.4456576
-.2893974
-.2774139
-.3626495
-.163:721
-.280E566
-.2097354
-.153E-965
-.243::675
-.348E827
-. 2824702
-.140J052
-.2612332
-.085954
-.1207347
-.0545563
-.2304264
-.2146449
-.0045855
-.0777285
.1137982
-.1081078
-.0691404
-.1051498
.0411726
.0063128
.0299988
-.012408
-.0010569
-.0690311
-.1686777
-.2076459
-.1626312
, Interval)
1.318795
-.0109083
.000158
.5314458
-.0345627
.1160814
-.0563301
.0043209
.1054286
.0680271
.1437329
.0175691
3.562539
.0247883
.0955638
.1031495
.0657513
.0098907
.0770536
-.104334
-.0429608
.0200197
.0843683
-.026255
.8148835
.3575091
.0944425
.2547137
.1534096
.3630206
-.0316293
.098666
.2782007
.3103291
.5269491
.0826109
.0817409
.095412
.268588
.2819705
.226617
.1204518
.1118916
.4927427
.2065984
.1378169
.1451927
-------
318
vldis74
vldis75
vldis76
vldis77
vldis78
vldis79
vldisSO
vldisSl
vldis82
vldis83
vldis34
vldisSS
vldis86
vldis87
vldisSS
vldis89
vldis90
vldis91
vldis92
vldis93
vldis94
vldis95
vldis96
vldis97
vldis98
vldis99
pfemales
pblack
pother
page under5
page 5 29
page 65 up
pmarhh chd
pmhh child
pfhh_child
pvacant
prenter occ
vpfemales
vpblack
vpother
vpage underS
vpage 5 29
vpage 65 up
v pmarhh chd
vpmhh child
vpfhh child
vpvacant
vprenter occ
ye a r 7 1
year72
vear73
year74
year75
year76
year77
year78
year 7 9
year 80
.0373692
-.0474257
.1327652
.0560184
.0095267
.0772341
.0474642
-.410782
.0256398
-.0260686
-.0900032
-.0464233
-.1323016
.0979971
.057224
-.033028
-.1389267
-.2024679
-.0218133
-.0139467
.0163334
-.1395255
-.0854278
-.0445903
-.0433964
-.0270545
3.93342
-3.091772
-.0013977
-5.258563
2.75681
.4459353
-.5782048
-.8582164
.1185717
1.428592
-.1759728
-.9480066
1.245737
-.0138705
1.985041
-3.098925
-1.057589
.3038902
.0868051
.3086173
-1 .204047
.2321561
.1646225
.2426421
.3830431
.3589601
.5575864
.6507012
.9052132
1.186598
1.219498
1 .319258
.0757932
.0472024
.0443804
.0326274
.0446826
.0681129
.0724889
.3177079
.0664738
.0903692
.0740601
.0422408
.1055433
.0425103
.0637339
.0542153
.1102413
.0937774
.041006
.0325591
.0449405
.0547954
.0458419
.0347812
.0316081
.0259879
1.9916
2.240053
.1657788
1.906766
.9891335
.8712632
.4886966
1.040405
.8544424
1.004489
.1486005
1.835389
2.104473
.1498806
1.67961
.8805155
.7917475
.4465291
1.060696
.7840376
.9984553
.1488887
.1147839
.1008523
.0988288
.1084108
.1019989
.1026361
.1048552
.1052646
.114578
.1150346
0
-1
2
1
0
1
0
-1
0
-0
-1
-1
-1
2
0
-0
-1
-2
-0
-0
0
— 2
-1
-1
-1
-1
1
-1
-0
-2
2
0
— 1
-0
0
1
-1
-0
D
-0
1
-3
-1
0
0
0
-1
1
1
2
3
3
5
6
8
11
1C
11
.49
.00
.99
.72
.21
.13
.65
.29
.39
.29
.22
.ID
.25
.07
.90
.61
.26
.16
.53
.43
.36
.55
.86
.28
.37
.04
.98
.38
.Cl
.76
.79
.51
.18
.32
.14
.42
.18
.52
.59
.09
.18
.52
.34
.68
.08
.39
.21
.56
.43
.41
.88
.31
.47
.34
.63
.27
.64
.47
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
c
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
3
0
.622
.315
.003
.086
.831
.257
.513
.196
.700
.773
.224
.272
.210
.038
.369
.542
.208
.031
.595
.668
.716
.011
.062
.200
.170
.298
.048
.168
.993
.006
.005
.609
.237
.409
.890
.155
.236
.606
.554
.926
.237
.000
.182
.498
.935
.694
.228
.119
.152
.016
.000
.001
.000
.000
.000
.000
.000
.000
-.
-.
-.
-.
-.
-.
— 1
-.
-.
-.
-.
-.
,
-.
-.
-.
-.
-.
-.
1112026
1399531
0457696
0079387
0780612
0562826
0946304
.033561
1046638
2032125
2351777
1292249
3391902
0046674
0677088
1393022
3550245
3862927
1021942
-.07777
-.07176
2469367
1752883
1127698
1053553
0779967
0294355
-7.48278
-.
-8
-1
3263614
.996253
8178857
.261937
.
,
,
.
,
,
,
.
.
-.
.
,
-.
,
-
7
1
,
-1
4
2
-1.53616
-2
-1
-.
- .
-4
-2
-.
-1
-4
-2
-1
-1
-3
.
.
.
.
1
.397644
.556328
5404324
4672631
.545781
.879503
3076703
.307372
.824933
.609592
5753278
.992397
.228273
.161245
0596993
0603798
0449489
1893164
1464505
3576456
4494133
6996734
9802556
9948995
.093764
1
1
3
•
2
5
.
5
-1
.
1
2
1
.
.
.
.
1
1
1
.185941
0451016
2197608
1199755
0971147
2107507
1895589
2119969
1559434
1510753
0551712
0363782
.074587
1713268
1821567
0732463
0771711
0186431
0585676
0498766
1044269
0321143
0044327
0235882
0185626
0238878
.837405
.299235
3235661
.520873
.695734
.153807
3797505
.181212
.793471
.397617
1153175
.649768
.370977
2799292
.277455
.372916
4944139
.183108
.166007
.845507
7531509
5240115
3896249
4403353
5767697
5714697
7575272
8519891
.110753
1.39294
.444097
.544751
-------
319
yearSl
year82
year83
year84
ysar85
year86
year87
yearSS
year89
year90
year91
year92
year93
year94
year95
year96
ysar97
year98
year99
cons
1.18804
1.0S4203
1 .563661
1.439866
1.634104
1 -7C6003
1.901796
1.936037
1.948422
2.192643
2.104079
2.350348
2.261341
2.226663
2.124498
2.111936
2.028976
2.158611
2.219835
5.889568
.1853593
.1588885
.1131756
.1201688
.1246051
.1150238
.1133351
.1249827
.124541
.1221637
.1321843
.1099423
.1121414
.1099758
.1196479
.1167747
.1176057
.1083457
.1097631
1.063314
6.41
6.82!
13.82
12.48|
13.11'
14.83
16.781
15.49
15.64
17.95'
15.92
21.38
20.17'
20.25
17 .76,
18.09
17.25
19.92'
20.22
5.54
0.000
0.000
O.ODO
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0 .000
0.000
0.000
0.000
0.000
.824.5942
.77:>746
1.34 .312
1.264308
1.3 a 985
1.4:5053
1.67':>633
1.69:043
1.70.1294
1. 953174
1.84. ',963
2.13 F
R-squared =
Root MSE
9211
207.46
0.0000
0.7022
.45117
-------
320
Isprice
notold
age
age2
sqft
sqft2
bedrms
bthrms
sqf tbed
sqftbth
f place
knowf Ir
floors
lotsi ze
Idis70
Idis71
Idis72
Idis73
Idis74
Idis75
idis7S
Idis77
Idis73
idis79
Idis33
IdisSl
Idis92
Idis83
Idis84
Idis85
Idis86
Idis87
IdisSS
Idis89
Idis90
Idis91
idis92
Idis93
Idis94
Idis95
Idis96
Idis97
Idis93
Idis99
vldis70
vldis71
vldis72
vldis73
vldis74
vldis75
vldis76
v.ldis77
vldis78
vldis79
vldisSO
vldisSl
vldis82
Coef .
1.01078
-.0130358
.0001119
.448072
-.0731531
.0623636
-.0943
-.0324151
.053809
.0601806
.0754882
-.0168327
.3340394
-.2855172
-.1822907
-.0960865
-.0849355
-.1939962
-.045649
-.19058G4
-.1348433
-.0348141
-.0699274
-.1598652
.2933968
.1516452
-.0831549
.0528265
-.00965
.113026
-.1757187
-.1172299
.0820262
.0690992
.2787667
-.0863779
-.0708935
-.0874664
.0327853
.0352414
.0427095
-.0333916
-.0347635
.1553786
-.0266688
-.0822359
-.0880326
.0107562
-.0949748
.0751702
.0088869
-.0631169
.0169076
-.0308488
-.5124926
-.0499772
Robust
Std. Err.
.1963429
.0011239
.0000141
.060142
.0228597
.027589
.0316053
.0180257
.0212689
.0122113
.0333205
.0209767
.262537
.1559739
.1391968
.102256
.0859546
.1008932
.0742886
.0618659
.0611903
.0616751
.0937735
.0995759
.2913203
.1423549
.0962946
.1002795
.0858904
.1135165
.0711654
.0919078
.0807829
.102748
.1197444
.0642588
.060235
.068083
.0740736
.0341704
.0654582
.0538653
.0509786
.1409644
.1023727
.0878412
.0772817
.080406
.0530861
.0612246
.0536453
.0584521
.0794405
.0799577
.3266528
. 076536
5
-11
7
7
-3
•~>
-2
-1
2
4
2
-0
1
_ -I
-1
-0
-0
-1
-0
-3
-2
-0
-0
-1
1
1
-0
0
-0
1
-2
-1
1
0
2
-1
-1
-1
1
0
0
-0
-C
1
-0
-0
-1
0
-1
1
0
-1
0
-0
-1
-0
t
.15
.60
.93
.45
.20
.26
.98
.80
.53
.93
.27
.80
.27
.83
.31
.94
.99
.92
.61
.03
.20
.55
.75
.73
.01
.07
.85
.53
.11
.00
.47
.23
.02
.67
.33
.34
.13
.23
. 12
.42
.65
.62
. 63
. 10
.26
.94
.14
. 13
.64
.23
.17
.08
.21
.39
.57
.65
P> 1 t I
0.000
0.000
0.000
0.000
0.001
0.024
0.003
0.072
0.011
0.000
0.024
0.422
0.203
0.067
0.190
0.347
0.323
0.055
0.539
0 .002
0.029
0.572
0.456
0.074
0.314
0,287
0,388
0.598
0.911
0.319
0.014
0.202
0.310
0.501
0.020
0.179
0.239
0.199
0.264
0.675
0 .514
0 .535
0 .495
0 .270
0.794
0.349
0.255
0.894
0.102
0.220
0.868
0.280
0.831
0.700
0.117
0.514
[95% Conf.
.6259035
-.0152389
.0000842
.3301802
-.1179634
.0082829
-.1562535
-.0677495
.0121172
.0362437
.0101724
-.0579517
-.1805923
-.5912612
-.4551479
-.2965313
-.253426
-.3917697
-.1912713
-.3118515
-.2547901
-.1557112
-.2537446
-.3354542
-.2776567
-.1274024
-.2719141
-.1437439
-.1780145
-.109492
-.3152189
-.29739
-.0763266
-.1323099
.0440408
-.2123396
-.1889677
-.2209245
-.0624157
-.1297516
-.0556032
-.1389698
-.134693
-.1209433
-.2273423
-.2544245
-.2395222
-.1468577
-.2088366
-.0448439
-.0962701
-.1776962
-.1388138
-.187584
-1.152806
-.2000051
Interval 1
1.395656
-.0108326
.0001396
.5659638
-.0283429
.1164442
-.0323465
.0029194
.0955008
.0841175
.140804
.0242863
.8486711
.0202268
.0905665
.1043584
.0835549
.0037773
.0999734
-.0693094
-.0148965
.036083
.1138898
.0157238
.8644503
.4306928
.1056043
.2493968
.1587146
.335544
-.0362185
.0629301
.2403789
.2705084
.5134927
.0395837
.0471807
.0459917
.2279863
.2002343
.1710223
.0722066
.0651661
.4317006
.1740047
.0899528
.0634569
.1683702
.0188871
.1951843
.1140438
.0514624
.1726289
.1258863
.1278204
.1000506
-------
321
vldis83
vldis84
vldis85
vldis86
vldisS?
vldisBS
vldis89
vldis90
vldisSl
vldis92
vldis93
vldis94
vldis95
vldis96
vldis97
vldis98
vldis99
Id school
ld_retail
Id hospital
Id church
Id cemetery
Id 15
Id 1605
ld_ilO
Id railroad
Id_s60
Id rivers
Id cards
Id whittiern
Id parks
Id mjwater
Id csula
Id cclubs
vld school
vld retajLl
vld_hospital
vld church
vld cemetery
vld i5
vld 1605
vld_110
vld railroad
vld_s6C
vld rivers
vld cards
vld whitti-n
vld parks
vld mjwater
vld csula
vld cclubs
year 171
year72
year 7 3
year74
year75
year? 6
-.0801731
-.1139937
-.0774273
-.1491694
.0847634
.0532239
-.0361026
-.1617859
-.2141515
-.0017229
.0170388
.034036
-.1217679
-.0242093
-.0132117
-.007C4S6
.0147186
.05835
.2126186
.01DC744
.0692553
.1146273
- .0408457
.0033587
-.09C095
.0742574
-.0254397
-.0022064
.0239115
.0164331
.0035618
.0235234
.2097668
-.047513
-.0536143
-.1630339
-.0306094
-.1054151
-.0500317
.0955708
.0487181
.0535919
-.0491097
.0354948
.0326098
-.008712
.0473927
-.0138063
-.0349612
-.1055273
.025362
.1631364
.23412
.3935531
.3445819
.5432421
.6636308
.0925115
.0974959
.061781
.1117435
.0622988
.0755884
.0695848
.1117712
.1026371
.0550318
.0526479
.0592979
.0673704
.0616169
.0524804
.0488948
.0459458
.0229996
.0907346
.0252366
.0514981
.0348863
.046C772
.0406476
.0331572
.0225946
.0363207
.0348716
.0046567
.0923855
.0158883
.0221965
.0690151
.0154859
.0227146
.0806322
.0236917
.0574642
.0338297
.0464233
.0403704
.0320213
.0223083
.0370346
.0348624
.005189
.0874666
.0167664
.0186473
.0651511
.0174375
.1156831
.1019729
.0982892
.1061608
.0985958
.0974931
-0.87
-1.36
-1 .25
-1.33
1.36
0.77
-0.52
-1.45
-2.09
-0.03
0.32
0.57]
-1.81J
-0.33,
-0.23
-0.14
0.32
2.56
2.35
0.40
1.34
3.29
-0.89
0.22
-2.72
3.29
-0.70
-0.06
5.13
0.19
0.541
1 .0^
3.0^
-3.01
-2.36;
-2.02
-1.29
-1.83
-1.48
2.06
1.21
1.67
-2.20
0.96
0 .94
-: . 68
0.54
-0.82
-'..81
-1.62
1.45
"-•41
2.3t|
4 .00
3.24
5.51
6.81
0.386
0.174
0.210
0.182
0.174
0.441
0.604
0.148
0.037
0.975
0.747
0.566
0.071
0.694
0.801
0.885
0.749
0.011
0.019
0.690
0.179
0.001
0.375
0.826
0.007
0.001
0.484
0.950
0.000
0.859
0.590
0.289
0.002
0.002
0.018
0.043
0.196
0.067
0.139
0.040
0.223
0.094
0.028
0.333
0.350
0.093
0.588
0.410
0.061
0.105
0.146
0.159
0.022
0.000
0.001
0.000
0.000
-.2615165
-.2905104
-.1933319
-.3682119
-.0373563
-.0893463
-.1725044
-.3803826
-.4157354
-.1095975
-.085555
-.0822013
-.2533292
-.1449922
-.1160852
-.1023935
-.0753456
.0137655
.0349532
-.0393951
-.0316925
.0462428
-.1311674
-.0711118
-.1550906
.0299666
-.0966365
-.0705625
.0147833
-.1646634
-.0225829
-.0199867
.0744817
-.0778688
-.0981404
-.3210913
-.0770504
-.2180579
-.1163455
.0045697
-.030417
-.0091771
-.092S39
-.0371014
-.0357284
-.0188835
-.1240615
-.0466722
-.0715141
-.233238
-.0088194
-.0636266
.0342301
.2008842
.1364827
.3499722
.4725223
.1011702
.052513
.0436774
.0698731
.206E83
.2063941
.1002992
.0573107
-.0125675
.1061518
.1206326
.1502732
.0102934
.0965737
.0896617
,0887962
.1047828
.1039345
.3906789
.0595438
.1702031
.1830126
.0494761
.0890293
-.0250995
.1185479
.0457571
.0661498
.0330396
.1975296
.0397065
.0670335
.345052
-.0171572
-.0090891
-.0049766
.0158317
.0072276
.0162821
.1865719
.1273531
.1163608
-.0053804
.108091
.100948
.0014596
.2188469
.0190597
.0015918
.0221835
.0595433
.3899014
.4340099
.586222
.552681
.736512
.8547392
-------
322
year? 7
year78
year79
yearSO
yearSl
year82
year 83
year84
year85
ye a r 8 6
year87
yearSS
yearB 9
year90
year91
year 92
year93
year94
year95
yearS6
yearS7
ysar58
year99
cons
.
1
1
I
i
1
1
1
1
1
2
2
2
O
£.
2
2
<•}
o
2
1
9082953
1.17338
.216334
.313444
.223729
.075728
.567677
.513282
.635502
1 .72384
1.90264
.941813
.955631
.213139
.105524
2.36137
.259504
.254821
.156261
.132398
.064529
.200241
.260399
.777638
.096925
.0950978
.1052138
.106073
.1815608
.1550198
.1019548
.1098191
.1120046
.1017301
.0985698
.1102744
.1168134
.1058574
.1181024
.09562
.0977135
.0956137
.1048449
.1042553
.1045225
.0949103
.09494
.336108
9
12
11
12
6
6
15
13
11
16
19
17
16
20
17
24
23
23
20
20
19
23
23
23
.37
.34
.56
.38
.74
.94
.38
.78
.60
.95
.30
.61
.•74
.91
.83
.70
.12
.58
.57
.45
.75
.19
.81
.14
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.
1
1
.
.
1
1
1
1
1
1
1
2
1
2
2
2
1
1
1
2
2
7183004
.986967
.010091
.105517
8678288
7718546
.367823
.298012
.415948
.524426
.709421
.725651
.726651
.005634
.874016
.173934
.067963
.067396
.950742
.928034
.959641
.014195
.074295
7.11879
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
8
1.09829
.359793
.422577
.521371
.579629
.379602
.767532
.728552
.855056
.923254
.095859
.157976
.184612
.420643
.337031
.548807
.451044
.442245
.361781
.336762
.269417
.386287
.446502
.436485
Hypothesis
P-value
ofF-test
Reject @
5% level?
All structural attribute slopes simultaneously zero
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same
0.0000
0.0000
0.0001
All lotsize-independent other distance effects simultaneously zero ° • 000°
All lotsize-dependent year-specific slopes on LDIST
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on
vX Idist variables)
0.0182
0.0139
All lotsize-dependent other distance effects simultaneously zero (on ° • °006
vX "other distance" variables)
7.4 Including both other distances and tract attributes
Regression with robust standard errors
Number of obs
F(158, 9052)
Prob > F
R-squared
Root MSE
9211
193.83
0.0000
0.7072
.44791
-------
323
Isprice
notold
age
ags2
sqft
sqft2
bedrms
bthrms
sqftbad
sqftbth
fplace
knowf Ir
floors
lotsige
IdisTO
Idis71
Idis72
Idis73
Idis74
Idis75
Idis76
Idis77
Idis78
Idis79
Idis30
Idis31
Idis82
Idis83
Idis84
IdisSS
Idis86
Idis87
IdisSB
Idis89
Idis9|0
Idis91
Idis92
Idis93
Idis94
Idis95
Idis9l6
Idis97
Idis98
Idis99
vldis7Q
vldis71
vldis72
vldis73
vldis74
vldis75
vldis76
vldis77
vldis78
vldis79
vldisSO
Cosf .
.9396044
-.0139859
.000133
.430146
-.0643951
.0603665
-.0878012
-.032318
.0485405
. 04 4 20 4 S
.0454479
-.0057551
2.133716
-.2892864
-.1591525
-.0979534
-.0864205
-.207558
-.0710578
-.2315977
-.1678442
-.1049015
-.11C2457
-.21C5626
.264586
.0683614
-.0962C69
-.0059556
-.0265716
.1159009
-.1703324
-.1055557
.0953057
.09S0395
.2746674
-.0511714
-.03994S3
-.0349283
.1349149
.1012796
.1109927
.0320976
.0370607
.2051993
-.0077163
-.0461071
-.0479757
.0565544
-.0427771
.1501732
.0781389
.0301396
.0843537
.0557155
Robust
Std. Err.
.22367
.0012051
.0000155
.0606969
.023C478
.0274557
.0317455
.0179173
.0213351
.0123183
.0335581
.0210582
.9663641
.1646555
.1523035
.1151836
.0983882
.108093
.0793885
.0630613
.0666591
.0641114
.0948561
.0949694
.2906608
.1424117
.0983969
.1036861
.0876936
.1149114
.0757173
.0963463
.0329579
.1041006
.1204551
.0713493
.0675568
.0747392
.0906216
.0908195
.0724285
.0625171
.0609756
.1498543
.1104356
.1014559
.0874913
.0377758
.0638557
.0637333
.0605014
.0634212
.0809932
.0852435
r
t
4.11
-11.61,
8.561
7.09
-2.82
2.22
-2.77
-1.80!
2.28'
3.59
1.35
-0.27
2.21.
-1.76
-1.04
-0.85
-0.87
-1.92,
-0.89
-3.67
-2.52
-1.64
-1.16,
-2.22'
0.91
0.48
-0.98
-0.06
-0.30:
1.01
-2.25
-1.09
1.15
0.95
2.28
-0.72
-D.58
-3.47
1.67
1.12
1.53
0.51
3.61
1.37,
-3.071
-3.45
-0.55
3.64
-3.67
2.36i
1.29
3.48
1.05
3.65
P> 1 1 1
0.000
0.000
0.000
0.000
C.005
C.026
0.006
0.071
0.023
0.000
0.176
0.785
0.027
0.079
0.298
0.395
0.382
0.055
0.374
0.000
0.012
0.102
0.245
0.027
0.363
D.629
0.328
0.954
0.762
0.314
0.024
0.276
0.248
0.341
0.023
0.473
0.565
0.640
0.094
0.265
0.125
0.608
0.543
0.171
0.944
0.650
0.533
0.519
0.503
0.018
0.197
0.635
0.295
0.513
195* Conf.
.4911i596
~.016:'481
.000: 026
.311: 663
-.:i( 074
.007: 471
-.150(1295
-.OC744
.006"-189
.020d582
-.020:- 336
-.0470339
.2394244
-.612(i483
-.4586819
-.323-493
-.2802637
-.4194446
-.2276572
-.35! 212
-.2985 112
-.2301743
-.296:852
-.3967241
-.305:748
-.210;: 97 9
-.289(371
-.209:.037
-.1984708
-.1094514
-.3187555
-.295:363
-.066^-106
-.105C213
.0385482
-.191C302
-.1712763
-.183434
-.0231217
-.0767472
-.030^836
-.0
-------
324
vldisSl
vldis82
vldis83
vldis84
vldis85
vldis86
vldisSl
vldis83
vldis89
vldis90
vldis91
vldis92
vldis93
vldis94
vldis95
vldis9S
vldis97
vldis98
vldis99
Id school
ld_retail
Id hospital
Id church
Id cemetery
Id 15
Id i605
Id 110
Id railroad
Id_s60
Id rivers
Id cards
Id whittle rn
Id parkH
id mj water
Id csula
Id cclubs
vld school
vld retail
vld hospital
vld church
vld cemetery
vld 15
vld 1605
vld_110
vld railroad
vid s60
vld rivers
vld cards
vld whitti~n
vld parks
vld mjwater
vld csuls
vld cclubs
pferr.ales
pblack
pother
page ur.de rE
page 5 29
-.4229702
.0395891
-.0242993
-.0281755
-.0189046
-.1135712
.1174411
.085962
-.0075404
-.1447628
-.1698419
.0038675
.0211849
.0269625
-.1380712
-.0579474
-.0434914
-.0393046
-.0304552
.076556
.1410083
.0038849
.0567614
.0943059
-.0496932
.0515217
-.1494311
.062625
-.0233059
.044171
.0277613
-.010857
.00142
.0452407
.2458509
-.033089
-.067924
-.1492354
-.0070525
-.0767225
-.0649201
.0454299
.0169224
.1035502
-.0513127
.0354789
.0042554
-.0144015
.0146674
-.0053955
-.0571488
-.154431
.0228168
3.149603
-1.893343
-.281395
-4.450687
3.058623
.3271879
.0843534
.095707
.0894657
.0639991
.1123787
.0677692
,0808687
.0730036
.1130671
.1028258
.0629433
.0620146
.0673223
.0757118
.0691739
.0619631
.0581896
.0561175
.0238608
.0959807
.0270081
.0533221
.039173
.0504129
.0561445
.0381768
.0234534
.0411942
.037762
.0048259
.0959999
.0153337
.0247954
.0796422
.018603
.0237941
.0838446
.0252526
.0596684
.0365769
.0506654
.0562335
.0358445
.0233714
.0422519
.037204
.0054509
.0886792
.0160706
.0212625
.0743566
.0219965
2.153247
2.643731
.2120095
2.176262
1 .126971
-1
0
-0
-0
-0
-1
1
1
-0
-1
-1
0
0
0
-1
-0
-0
-0
-0
3
1
0
1
2
-0
0
-3
2
-0
1
5
-0
0
1
3
-1
-2
-1
-0
-1
-1
0
0
2
-2
0
0
-2
0
-0
_9
-2
1
1
-0
-1
-2
2
.29
.47
.25
.31
.33
.01
.73
.06
.10
.28
.65
.06
.34
.40
.82
.84
.70
.68
.54
.21
.47
.14
.06
.41
.99
.92
.91
.67
.57
.11
.75
• + J-
.09
. 32
.09
.78
.35
.78
.23
.25
.77
.90
.30
.99
.20
.84
.11
.64
.17
.34
.69
.OS
.04
.46
.72
.33
.05
.71
0.196
0.639
0.800
0.753
0.768
0.312
0.083
0.288
0.918
0.200
0.099
0.951
0.733
0.689
0.068
0.402
0.483
0.499
0.587
0.001
0.142
0 .886
0.287
0.016
0.324
0.359
0.000
O.OOS
0.572
0.242
0.000
0.910
0.926
0.063
0.002
0.075
0.004
0.075
0.780
0.199
0.076
0.370
0.763
0.004
0.028
0.401
0 . 909
0.008
O.S69
0.737
0.007
0.038
0.300
0.144
0.474
0.184
0.041
0 .007
-1.064332
-.1257625
-.2119066
-.2035484
-.1443574
-.3338589
-.0154018
-.0725589
-.1506439
-.3664
-.3714038
-.1195156
-.1003778
-.1050044
-.2864834
-.1935439
-.1649531
-.1533694
-.1404583
.0297834
-.0471355
-.0490572
-.047762
.0175179
-.148514
-.0585342
-.2242663
.0166511
-.1040559
-.0298511
.0183019
-.1990384
-.0286375
-.0033638
.0697342
-.0695551
-.1145658
-.3135897
-.0565533
-.1936861
-.1366192
-.0538967
-.093308
.0332868
-.0971259
-.0473444
-.0666729
-.0250864
-.1591639
-.0368975
-.0988281
-.3001868
-.0203014
-1.071249
-7.075653
-.6969816
-6.716692
.8495047
.2183921
.2049407
.163308
.1471975
.1065482
.1067164
.2502841
.244483
.1355631
.0768743
.0317199
.1272507
.1427476
.1589295
.0103409
.077649
.0779703
.0747602
.0795478
.1233285
.3291521
.0568269
.1612849
.1710939
.0491275
.1615777
-.074596
.1085989
.0574441
.118193
.0372217
.1773244
.0314774
.0938452
.4019675
.0033771
-.0212822
.0151189
.0424434
.0402411
.006779
.1447445
.1271527
.1738135
-.0054996
.1183021
.0771837
-.0037165
.1884937
.0261064
-.0154696
-.0086752
.065935
7.370454
3.288968
.1341917
-.1846827
5.267741
-------
325
page 65 up
pmarhh chd
pmhh child
pfhh_child
pvacant
prenter occ
vpfemales
vpblack
vpothfer
vpage underB
vpage 5 29
vpage 65 up
vpmarhh chd
vpmhh child
vpfhh child
vpvacant
vprenter occ
year71
year72
yearVS
year74
year75
year76
year77
year78
year? 9
yearSO
yearSl
year82
yearG3
year84
year85
year86
year87
yearSS
yean89
yearSO
yearj91
year92
yeac93
yeat94
year95
ye a r 9 6
year97
year98
.8124726 .9695418
-.6169049 .5570739
1.007679 1.121035
1.186137 1.082091
1.203104 1.174282
-.5459779 .1711734
-1.409647 1.93081
1.33834 2.72497
.1237247 .1929762
:. 594432 1.9326
-2.909213 1.026244
-.9587272 .8971892
.3797277 .5167106
-1.408249 1.137144
-1.135848 1.026224
-.9868546 1.150016
.4592367 .1703943
.1664823 .1162305
.2514694 .101474
.4092673 .0995113
.3848241 .1096513
.5964986 .1050497
.7026122 .1070529
.9613523 .1099833
1.247636 .1111936
1.289347 .1206733
1.382417 .1213021
1.254109 .1923185
1.159287 .16687
1.633089 .1204857
1.595514 .127342
1.706921 .1324946
1.791261 .1225671
1.982614 .1219792
2.C28457 .1328252
2.040198 .1340566
2.293965 .1311899
2.1.97231 .1396465
2.443586 .1199116
2.349368 .1219154
2.330382 .1195678
2.238097 .1288106
2.220733 .127
2.14241 .1274552
2.282842 .1191668
year99 2.3B0795 .1204582
cons 5.7Q4457 1.097236
0.841
-1.11
C.90
i.iol
1.02
-3.20
|
-0.71
0.49
0.67
0.83
-2.83
-LOT
0.73
-1.24
-l.H
-0.86
2.69
1.43
2.48
4-H
3.51
5.68
6.5$
8.74
11.22
10.68
11.40
£.52
£.95
13. 5i
12.45
12.88
14. el
16.25
15.27
15.22
17.49
15.73
20. 3B
19.29
19.49
17.36
17.49
16.81
19. 16
19.52
5.20
0.402
0.268
0.369
0.273
0.306
0.001
0.477
0.623
0.505
0.409
0.005
0.285
0.464
0.216
0.268
0.391
O.C07
0.152
0.013
0.000
0.000
0.000
0.000
0.000
o.ooc
o.ooc
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0 .000
0.000
0.000
0.000
-1.083048 2.712994
-1.70-3896 .475086
-1.189803 3.205161
-.9353068 3.307281
-1.093754 3.504962
-.8823165 -.2114394
-5.292484 2.473189
-4.003218 6.679898
-.2495522 .5070016
-2.193851 5.382815
-4.920888 -.8975467
-2.717421 .7999664
-.6341419 1.391597
-3.63731 .8208107
-3.147478 .8757819
-3.241145 1.267436
.1242452 .7942292
-.0613558 .3943204
.0525574 .4503313
.2142027 .6043318
.1698829 .5997654
.39C5774 .8024199
.4927643 .9124602
.7457601 1.176944
1. 029672 1.465601
1.0*2801 1.525894
1. 144637 1.620196
. 9711202 1.631996
.8321837 1.48639
1.396909 1.869268
1.3;-5395 1.83513?
1.4*7201 1. 96664
1. 551002 2.03152
1.7--.3507 2-22172:
1.7(18089 2.288324
1.7"7417 2.302979
2.036705 2.551025
1. 923492 2.4709^
2.208533 2.67864
2.1X0582 2.588153
2. 096003 2.564762
1.985599 2.490595
1.9''1784 2.469681
1.832559 2.392251
2.019248 2.516436
2.11466 2.58691
3.5J3626 7.855289
Hypothesis
All structural attribute slopes simultaneously
zero
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
j
All lotsize-independent year-specific slope on LDIST
the same
P-value Reject @
ofF-test 5% level?
0.0000
0.0000
0.0000
-------
326
All lotsize-independent other distance effects simultaneously zero
All lotsize-independent Census tract characteristic effects
simultaneously zero
All lotsize-dependent year-specific slopes on LDIST
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on
vX Idist variables)
All lotsize-dependent other distance effects simultaneously zero (on
vX "other distance" variables)
All lotsize-dependent Census tract characteristic effects
simultaneously zero (on vX Census tract variables)
0.0000
0.0000
0.0105
0.0077
0.0001
0.0000
-------
327
Appendix C - Woburn Sites (Wells G&H and Industri-Plex)
Contents
i
1 CRITERIA FOR EXCLUSION FROM RAW SAMPLE 329
2 ANNUAL COUNTS IN SAMPLE 329
3 DESCRIPTIVE STATISTICS 330
3.1 Housing prices and distances from nearest site 330
3.2 Structural variables 331
3.2.1 R2 for auxiliary regressions among these variables 332
3.3 Census tract attributes f 332
3.3.1 R2 for auxiliary regressions amonglthese variables 332
3.4 Other distances 332
3.4.1 R2 for auxiliary regressions among these variables 335
4 COLLINEARITIES t 335
4.1 Time patterns in average site distances in sample 335
4.2 Time trend in average lot sizes 336
4.3 Distance to site vs. structural variables 337
4.4 Distance to site vs. Census tract attributqs 337
4.5 Distance to site vs. other distances 338
5 TRENDS IN THE DISTANCE GRADIENT 338
5.1 Structural variables 339
5.1.1 Built post-1900 339
5.1.2 Age if built post-1900 339
5.1.3 Square footage 339
5.1.4 ! Bedrooms 340
5.1.5 Bathrooms 340
5.1.6 Fireplace(s)? 340
5.1.7 Floors recorded? 340
5.1.8 Floors 341
5.1.9 Lotsize 341
5.2 Census tract attributes 341
5.2.1 Females 341
5.2.2 Whites 342
5.2.3 Blacks 343
5.2.4 Other ethnic groups 344
5.2.5 Children under 5 344
5.2.6 Persons between 5 and 29 345
5.2.7 Persons between 30 and 64 346
5.2.8 Persons 65 and older 346
5.2.9 Married heads of household 347
5.2.10 Male-headed of household with children 348
5.2.11 Female-headed households with children 348
-------
328
5.2.12 Owner-occupancy 349
5.2.13 Renter-occupancy 350
5.2.14 Vacancy rates 350
6 COMPLETE REGRESSION RESULTS - No LOT SIZE INTERACTIONS 351
6.1 Just structural characteristics and year dummies 351
6.2 Including Census tract attributes 352
6.3 Including other distances 354
6.4 Including both other distances and tract attributes 356
7 COMPLETE REGRESSION RESULTS - MODELS EXPLORING ABSOLUTE
DIRECTIONAL EFFECTS 358
7.1 Including latitude and longitude linear shifters 358
7.2 Including latitude and longitude linearly and as site distance interactions 360
-------
329
Chapter 1 Criteria for exclusion from raw sample
Observations are excluded from the estimating sample for the Woburn site if:
the recorded selling price is zero or there is no record of the current assessed value of
improvements for the property
the total number of rooms exceeds 15
the number of bedrooms is zero or greater than 8
the house has more than four stories
the number of baths, including fractions, exceeds 5
the land area exceeds 75,000 square feet
the building area exceeds 5000 square feet
the year of the recorded sale is outside the 1978-1997 window
the most current assessed value of the dwelling is less than $8000 (if the log of this value
is less than 9; affects 10 observations)
the recorded sale price is less than about $1100 (if the log of this value is less than 7;
early house sales in this sample involve a lot of extremely low prices that are too
numerous to be either coding errors or non-arm's-length sales, this criterion affects 18
observations).
the recorded sale price is greater than $1 million (affects 5 observations)
the census tract is number 3585 (the data contain only what appears to be 31 replications
of the same transaction, at the same price, in the same year)
Chapter 2 Annual counts in sample
YEAR
78
79
90
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
Freq.
316
253
203
143
174
284
465
590
586
684
652
644
612
813
1077
1195
1255
1021
1146
301
Percent
2.54
2.07
1.67
1.15
1.40
2.28
3.90
4.74
4.71
5.50
5.24
5.18
4.92
6.53
9.65
9.60
10.09
8.20
9.21
2.42
Cum.
?.54
4.61
6.23
7.43
8.83
11.11
15.01
19.75
24.46
2P.96
35.20
40.37
45.29
5J.82
6D.48
70.08
80.17
8B.37
97.58
100.00
-------
330
Total i
12444
100.00
Chapter 3 Descriptive statistics
3.1 Housing prices and distances from nearest site
.036082 -
0-
.374486
8 40859
distwel
Marginal distribution of distances: Wobum
.142398-
o-
1500 935060
SPRICE
Marginal distribution of house prices: Woburn
-------
331
LOe+06-
900000 -
800000 -
1 700000-
c
I 600000-
u
g" 500000-
u 400000-
3
I 300000-
200000 -
100000-
Tl2 3 4 4 ! 7
Distance from nearest Wobum site (km)
1 Oe+06-
900000 -
„ 800000-
| 700000-
°. 600000-
S. 500000-
| 400000-
"v 300000-
"" 200000-
100000-
6 \
i r
dlst
J
Distance from nearest Wobum site (km)
3.2 Structural variables
Variable
Obs
Mean
Std. Dev.
Min
Max
notold
age
age2
sqft
sqft2
bedrms
bthrms
sqftbed
sqftbth
fplace
knowf Ir
floors
lotsize
12444
12444
12444
12444
12444
12444
12444
12444
12444
12444
12444
12444
12444
.8825137
29.22059
1431.947
1.764122
3.585322
3.206606
1.916747
5.994829
3.759232
.3726294
.5279653
.9040501
1
.322C119
24.04477
173C .919
.6879198
3.127169
.8423902
.7942649
3.798772
2.999084.
.482.5241;
.4992374!
.942C839
.6486305
0
0
0
.408
.166464
1
1
.408
.408
0
0
0
.0725759
1
95
9025
4. 981
24. 31036
7
5
34.72
24.8
1
1
3
4. 35228
-------
332
3.2.1 R2 for auxiliary regressions among these variables
Presented in order of variable list above.
.4683797170314082
.9220851573928605
.9118813156209173
.9581688612114861
.9805588416981539
.9080785263944619
.9357735922711881
.9782196655164073
.9794293001361518
.6306254215308172
.8866865498149391
.8684375445278907
.1760773201957944
3.3 Census tract attributes
Variable
Obs
Mean
Std. Dev.
Min
Max
1
p females
pblack
pother
page unde r5
page_5 29
page 65 up
pmarhh chd
prahh child
pfhh child
pvacant
prenter occ
12444
12444
12444
12444
12444
12444
12444
12444
12444
12444
12444
.5163597
.0073106
.0536386
.066371
.3474652
.1266392
.3083403
.0086841
.0464293
.0306772
.2269974
.0125727
.0067375
.0963819
.0140333
.050334
.0366465
.0750423
.0053736
.0314776
.0209923
.1426593
.4956687
.0007966
.003478
.0370092
.2429152
.0594679
.1766845
.0034562
.0180311
.0073603
.0308584
.5579294
.0391236
.6231612
.1198748
.4951487
.2443653
.5188977
.0361781
.2152134
.1417197
.656051
3.3.1 R2 for auxiliary regressions among these variables
Presented in order of variable list above.
.8333379383483428
.6423130939986187
.6499138244909342
.7940032426653801
.9540897077185154
.9312561902433929
.9694583283248606
.6241497143798369
.8195435645337803
.6963054354151017
.9587034947733537
3.4 Other distances
-------
333
Distance variable
d_summits
d_school
d_retail
d_hospital
d_church
d_cemetety
d_railroad
d_prinarte
d_othpriro
d_ma_rds
d_i95
Description ,
Distance from the nearest summit of land. There are about three
doxen minor summits in the sample area, and none of them are very
high. No house is at an altitude greater than about 3200 feet (verify
units, meters?)
Distance from the nearest school. There are about 76 different schools
in the sample area.
Distance to the nearest retail icenter. There are no major retail centers
within the sample area. (Houses in the sample are nearest to either
Fresh Pond Mall, Meadow Glenn Mall, or Northshore Mall, among
malls that are presently active. We do not have historical data
concerning their level of activity in the period 1987-97.) All three of
these centers are closer to Bcjston than the sample area, so the effect of
this variable will partially capture proximity to Boston's central
business district.
Distance to the nearest hospital. There are 3 hospitals inside the
sample area, one in each of the three southernmost zip codes, nearest
to Boston, of the seven zip codes in the sample area. Thus some of
the effects of proximity to Boston's central business district will
confound the effects of proxijnity to a hospital.
Distance from the nearest church. There are no churches at all inside
the sample area. All recorded churches are much closer to Boston's
city center. Thus, the effects of this variable will partially capture
proximity to Boston's central business district.
Distance to the nearest cemetery. There are thirteen cemeteries either
within the sample area, with at least one in each of the seven zip
codes.
Railroads cut through six of the seven zip codes m the sample area.
(Burlington Northern and Union Pacific are recorded as the owners of
these railroads.) There are two main routes, each running northwest
out of the Boston area
"Principal arteries are defined as significant roads that are not
designated as freeways. There are two basic north-south routes
cutting through the sample area, other than the two freeways, which
are not classed as principal arteries, [might want to drop this specially
constructed variable. .. different intuition than ESRI variables!
"Other primary roads" consists of a network of roads that criss-cross
the sample area, but do no include quiet residential streets.
Distance from the closest main Massachusetts roads. This includes
Interstates 93 and 95, if they happen to be the nearest main roads
(which they usually will not be).
Distance from Interstate 95, an east-west freeway that runs roughly
across the center of the sample area. The coefficient on this variable
is a proximity effect in addition to proximity from the nearest main
-------
334
d_i93
d_fp_tewma
d_fp_milit
d_fp_logan
d_fp_bevmu
d_parks
d_mj water
d_cclubs
d tewmac
d militarv
d logan
d bevmuni
road, d ma rds.
Distance from Interstate 93, a north-south freeway that runs roughly
up and down the center of the sample area. The coefficient on this
variable is a proximity effect in addition to proximity from the nearest
main road, d ma rds.
Distance from the closest of the two flight paths associated with Tew-
Mac airport, a small airport just outside the sample area, to the
northwest.
Distance from the closest of the two flight paths associated with
Hanscom Air Force Base, just outside the sample area to the
southwest.
Distance from the one flight path associated with Logan International
Airport (Boston) that cuts across the sample area. The center of the
sample area is about 1 5 kilometers northwest of Logan Airport.
Distance from the one flight path for Beverly Municipal Airport, a
small airport to the east of the sample area.
Distance from the nearest park. The most extensive park areas are on
the southern and southeastern boundaries of the sample area. There
are only eight very small parks scattered within the sample area, other
than these large parks areas in the south. Three external parks may be
the closest park for some houses near the boundaries of the sample
area
Distance from the nearest body of water. There are lakes associated
with the major park areas on the south and south-east boundaries of
the sample area, three bodies of water just north of the sample area,
and two or three lakes outside parks in the southern and eastern
portions of the sample area
Distance to the nearest country club. There are four country clubs
inside the sample area, in the three most southern zip codes. There are
two on the eastern boundary, or just outside it, and two near Hanscom
Airforce Base to the southwest, but outside the sample area.
Distance from Tew-Mac Airport.
Distance from Hanscom Air Force Base
Distance from Logan International Airport.
Distance from Beverly Municipal Airport.
Variable
Mean
Std. Cev.
Kin
Max
-------
337
4.3 Distance to site vs. structural variables
Regression with robust standard errors
Idisw |
notold |
age 1
age2 ;
sqft |
sqft2 1
bedrms |
bthrms |
sqftbed i
sqftbth 1
fplace |
knowflr 1
floors 1
lot size |
cons 1
Cos f .
.1323117
-.0027551
.OC0026
-.1913799
.015791
.039S46:1
-.1056481
-.0236376
.0512712
.2887278
-.4661392
.1193639
.0916SC1
1.406325
Robust
Std. Err.
.0175011
.0006524
5.299-06
.0286959
.0087448
.0165267
.0200206
.0071075
.0089206
.0163352
.0289458
.0137333
.0072345
.0416209
7
-5
3
-6
5
2
-5
-3
5
17
-16
Q
12
33
»-
.56
,76
.13
.67
.24
.39
.28
.33
.75
.68
.10
.62
.67
.79
P>
0.
C .
c .
0.
0.
0.
0.
0.
0.
0 .
0.
0.
0.
0.
1 t 1
000
000
002
000
000
017
000
001
000
000
000
000
(K)0
000
Number i
F( 13, :
Prob > i
R-squar<
Root MSI
195*
.098(
-.005C
;.73t
-.247*
.028*
.oca:
-- . 1 4 4 f.
-.037.'
.033:
.256',
>f obs
.2430)
id
Conf .
i069
|339
-06
.282
• 498
513
•916
694
S55
083
-.522F775
.051'
.077'
1.32^
445
693
742
12444
- 128.97
= 0.0000
= O.;052
.4722
Interval '
.1666165
-.0024762
.CC00422
-.1351315
.0629323
.0719415
-.0664046
-.0097058
.0687563
.3207473
-.409401
.1452834
.105331
1.487909
4.4 Distance to site vs. Census tract attributes
Regression with robust standard errors
Nur.ber c f obs = 12444
F{ 11, ;24?2) = 1039.33
Prob > F = 0.0000
R-squared = 0.4990
Root MSE = .3533
1
Idisw |
p females 1
pblack 1
pother I
page under5 1
page 5 29 1
page 65 up 1
pma rhh c hd |
pinhh child 1
pfhh child 1
pvacant 1
prenter occ |
cons I
25
10
I.
6.
1 .
-9.
-2
53
-29
21
-3.
-10
Coef .
.53699
.47T38
900744
731207
727-713
046912
.20178
.36124
.25011
.46273
117636
.32135
Robust
Std. Err.
.6144916
1.206333
.3224596
.6627516
.3253576
.2977327
.2470664
1.605203
.6321976
.6624228
.1324616
.3140708
41.
o .
5 .
10.
5.
-30.
-8.
33.
-46.
32.
-23.
-32.
t
64
63
89
23
31
39
91
24
27
40
54
66
P> 1 :. 1 [3
0
0
0
0
0
0
0
0
0
0
3
0
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
. 0 0 0
.000
24
3.
1.
5.
1.
— 9
-2.
50
-30
20
-3.
-10
5* Conf.
.3E251
111675
268674
481111
089S61
630515
6SC069
.25479
.45932
-If 428
377261
.93747
, Interval]
26
12
2.
3.
2.
-8
-1.
56
-28
22
-2.
-9.
.79147
.34209
532815
080303
365464
.46331
717492
.50769
.01091
.76118
=57991
706219
-------
338
4.5 Distance to site vs. other distances
Regression with robust standard errors
Number of obs = 12444
F( 23, 12420) = 5527.36
Prob > F - 0.0000
R-squared = 0.8940
Root MSE = .17012
Idisw
Id summits
Id school
Id retail
Id hospital
Id church
Id cemetery
Id railroad
Id prinarte
Id othpriro
Id rca roads
Id 195
Id_i93
Id fp tewma
id fp milit
Id fp log an
Id fp be virtu
Id parks
Id mjwater
Id cclubs
Id tewmac
Id military
Id logan
Id bevmuni
cons
Coef .
-.0370292
.0324862
1.320319
.1263315
-3.610246
.1680658
.0572693
-.0071236
-.0223408
.0062531
.2831022
.1343621
.004121
-.0357545
-.0042021
-.0410567
-.0147874
.0030723
-.016519
.3481346
-1.227163
.1453099
-3.979154
64.21698
Robust
Std. Err.
.0034191
.0026741
.0584987
.0077495
.0514098
.0046991
.0022626
.0024505
.0019421
.0012634
.0074579
.0035001
.0019537
.0023531
.0038821
.00303
.0026756
.0037919
.0024433
.0351869
.0504958
.0573186
.0894777
1.9738
-10
12
22
16
-70
35
25
-2
-11
4
37
38
2
-15
-1
-13
~ 5
0
-6
9
-24
2
-44
32
t
.33
.15
.57
.30
.22
.77
.31
.91
.50
.95
.96
.39
.11
.19
.09
.55
.53
.31
.76
.39
.30
.54
.47
.53
P> 1 1 I
G
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.000
.000
.000
.000
.000
.004
.000
.000
.000
.000
.035
.000
.279
.000
.000
.418
.000
.000
.000
.011
.000
.000
[95% Conf.
-.0437312
.0272447
1.205653
.1111412
-3.711017
.1588549
.0528342
-.011927
-.0261476
.0037766
.2684835
.1275013
.0002914
-.0403669
-.0118116
-.046996
-.0200319
-.0043604
-.0213084
.2791628
-1.326143
.0329566
-4.154544
60.34802
Interval ]
-.0303273
.0377278
1.434986
.1415217
-3.509475
.1772767
.0617044
-.0023203
-.0185341
.0087296
.2977209
.1412229
.0079506
-.031142
.0034073
-.0351173
-.0095429
.010505
-.0117297
.4171064
-1.128184
.2576633
-3.803764
68.08593
Chapter 5 Trends in the distance gradient
These models use individual houses as observations. We associate with each house the
proportion of each group in the Census tract that contains the house. The right-hand side
variables are the measured distance of the house itself from the Woburn site, a time trend,
starting at 1 in the first period of the data, and an interaction term between distance and time.
The simple trend picks up the trend over time in the concentration of the group in question
throughout the sample area. The "Idisw" variable, distance to the nearer of the Wells G&H sites
or the Industri-Plex site, picks up any baseline distance gradient in the concentration of the group
in question as a function of distance from the nearest Superfund site. The key variable is the
interaction term, which tells how the distance gradient is shifting over time. If the distance
gradient is becoming either less positive or more negative, the concentration of the group in
question nearer the Superfund site is growing, relative to the concentration further away.
-------
339
5.1 Structural variables
5.1.1 Built post-1900
Regression with
5.1.2
1
notold 1
Idisw I
trend I
Idiswy |
cons 1
robust standard errors
Coef .
.0511654
-.0025208
6.06e-06
.8455241
Robust
Std. Err. t
.0139318 3.67
.001649 -1.53
.0010607 0.01
.0216549 3!). 05
Nurr.be r of obs
F( 3, 12440)
Prcb > F
R-squared
Roct MSE
P>|t| [95% Conf.
0.000 .023857
0.126 -.0057532
0.995 -.0020732
0.000 .803077
12444
41.84
= 0.0000
- 0.0081
= .32075
Interval 1
.0784739
.0007116
.0020853
.8879711
Age if built post-1900
Regression with
5.1.3
I
age 1
Idisw !
trend 1
Idiswy I
cons 1
robust standard errors
Coef.
-.8602572
1.293745
-.3173891
23.7504
Robust
Std. Err. t
1.018054 -0,85
.1155265 11.24
.0795552 -3.99
1.495089 15.89
Nuir.be r of obs
F( 3, 10978)
Prcb > F
R-squared
Roct MSE
P>|t| [95% Ccnf.
0.398 -2.855327
0.000 1.072252
0.000 -.4733316
0.000 20.81976
10982
= 211.40
= 0.0000
= 0.0492
— 22 37^
Interval I
1.135313
1.525196
-.1614466
26.68104
Square footage
Regression with
sqft 1
Idisw [
trend 1
Idiswy I
cons i
robust standard errors
Coef.
.1432007
-.0096499
.0026113
1.651138
Robust
Std. Err. t
.0342994 4.18
.0032731 -2.95
.0025989 1 .00
.0435167 37.94
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
P>|t| [95% Conf.
0.000 .0759686
0.003 -.0160657
0.315 -.0024325
0.000 1.565339
12444
88.66
= 0.0000
= 0.0187
= .68154
Interval ]
.2104328
-.0032341
.0077061
1.736438
-------
340
5.1.4 Bedrooms
Regression with robust standard errors
bedrms I Coef.
Idisw I .0809946
trend I -.0136365
Idiswy | .0042233
_cons I 3.199047
5.1.5 Bathrooms
Robust
Std. Err. t
.0417319 1.94
.0043075 -3.17
.0031858 1.33
.0565213 56.58
Regression with robust standard errors
bthrms 1 Coef.
Idisw 1 .0593045
trend I -.0222378
Idiswy I .0109726
_cons I 1.934167
5.1.6 Fireplace(s)?
Robust
Std. Err. t
.0387366 1.53
.0038683 -5.75
.0029239 3.75
.0516204 37.47
Regression with robust standard errors
f place I Coef.
idisw | .1507521
trend I .0145869
Idiswy I -.0060986
cons I .0950634
5.1.7 Floors recorded?
Regression with robust st
Robust
Std. Err. t
.0225073 6.70
.0023447 6.22
.0017332 -3.52
.0304548 3.12
andard errors
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
P>|t| [95% Conf.
0.053 -.0009064
0.002 -.0220799
0.1S5 -.0020214
0.000 3.087257
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
P>|t I j95% Conf.
0.126 -.0167232
0.000 -.0298204
0.000 .0052412
0.000 1.832983
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
P>|t| [95% Conf.
0.000 .1066343
0.000 .0099909
0.000 -.0094958
0.002 .0353672
Number of obs
F( 3, 12440)
12444
37.65
= 0.0000
= 0.0088
= .83876
Interval]
.1626956
-.005193
.0104681
3.308838
12444
90.71
= 0.0000
= 0.0186
= .78693
Interval )
.1353323
-.0146553
.0167039
2.035351
12444
53.76
= 0.0000
= 0.0108
= .48095
Interval]
.1948699
.0191829
-.0027013
.1547596
12444
= 164.33
-------
341
1
knowflr I
Idisw |
trend I
Idiswy |
cons 1
5.1.8 Floors
Regression with
floors I
Idisw I
trend 1
Idiswy I
cons I
5.1.9 Lotsize
Regression with
I
lotsize I
Idisw I
trend I
Idiswy I
cons I
Coef .
.0950693
.0333765
-.0152776
.2355526
Robust
Std. Err. t
.0228831 4.15
.0023747 14.27
.001719 -8.89
.0317171 7.43
robust standard errors
Coef.
.1395111
.0040315
-.0051249
:. 5663 37
Robust
Std. Err. t
.0439333 3.18
.0046354 0.37
.0032299 -1.59
.0633544 24.71
robust standard errors
Coe f .
.2690467
-.003972
-.0026604
.7361747
Robust
Std. Err. t
.0336682 7.99
.0034553 -1.15
.0025702 -1.04
.0456018 16.14
Prob > F
R-s qua re i
Root MSE
P>|tl [95% Conf.
0.000 .0502149
0.000 .0292218
0.000 -.0186371
0.000 .1733322
Number of obs
F( 3, S566)
Prob > F
R-squared
Root MSE
P>lti [95* :onf.
0.001 .0535176
0.384 -.0050553
0.113 -.0114565
0.000 1.441191
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
?>|t| [95* Conf.
0.000 .2030519
0.250 -.0107449
0.301 -.0076984
0.000 .6467382
= o.ooco
= 0.0324
= .49114
Interval]
.1399238
.0385313
-.0119081
.2977229
6570
13.04
= 0.0000
= 0.0063
= .54328
Interval I
.2257646
.0131183
.0012067
1.689582
12444
= 168.25
= 0.0000
- 0.0373
= .63668
Interval |
.3350415
.0028008
.0023775
.8255612
5.2 Census tract attributes
5.2.1 Females
Regression with robust standard errors
Nuihber of obs = 12444
F( 3, 12440) = 61.19
Prob > F = 0.0000
-------
342
pfemales 1
Idisw |
trend |
Idiswy |
cons |
Coef.
-.0007029
.0002497
.000025
.5138941
Robust
Std. Err.
.0005885
.000063
.0000437
.0008457
r
-1.19
3.96
0.57
607.67
P>|t|
0
0
0
0
.232
.000
.568
.000
R-squared
Root MSE
[95% Conf.
-.0018564
.0001262
-.0000608
.5122364
= 0.0128
- .01249
Interval]
.0004506
.0003731
.0001107
.5155518
- pfemales_78
-pfemales_B7
• pfemales_82
- pfemales_92
53-
2
Q-
.5-
8.4
distwel
Fitted pfemales by distance from nearest Woburn site (km)
5.2.2 Whites
Regression with robust standard errors
Number of obs = 12444
F{ 3, 12440) = 651.79
Prob > F = 0.0000
R-squared = 0.0668
Root MSE = .06899
I
pwh its |
Coef.
Robust
Std. Err.
P>|t I
[95% Conf. Interval)
Idisw I -.0426479 .0025869
trend I -.0033329 .00021
idiswy I .0009262 .0001799
cons | 1.036867 .0028473
-16.49 0.000
-15.37 0.000
5.15 0.000
364.15 0.000
-.0477185 -.0375773
-.0037444 -.0029213
.0005736 .0012783
1.031286 1.042449
-------
343
78
-pvstiite 82
1 05 J
2
a.
.91-
6
distwel
Fitted pwhite by distance from nearest Wobum site (km)
5.2.3 Blacks
Regression with robust standard errors
I
pblack I
Idisw I
trend I
Idiswy |
cons I
Coef .
.0028297
.0005395
-.0002135
.0004607
Robust
Std. Err.
.0002498
.0000266
.0000192
.0003257
t
11.33
20,25
-11.12
1.41
P>|t|
0,
0,
0,
0 ,
.000
.000
.000
,157
Nurrber of obs = 12444
F( 3, 12440) = 332.76
Prcb > F = 0.0000
R-squared = 0.0400
Roct MSE = .0066
[95% Conf.
.0023399
.0004873
-.0002511
-.0001777
, Interval]
.0033194
.0005917
-.0001759
.0010991
-pblack 78
- pblacO?
-pblack 82
- pblack~92
.015-
-.003364 -
distwel
84
Fitted pblack by distance from nearest Woburn site (km)
-------
344
5.2.4 Other ethnic groups
Regression with
1
pother !
Idisw |
trend |
idiswy i
cons |
robust standard errors
Coef .
.0542575
.0043839
-.0012235
-.0510659
Robust
Std. Err.
.0034674
.0002694
.0002377
.0037505
15,
16.
-5,
-13,
t
.65
.28
.15
.62
P>|t |
0.000
0.000
0 . 000
0.000
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
[95% Conf.
.0474608
.0038559
-.0016894
-.0584174
12444
= 673.37
= 0.0000
= 0.0594
= .09349
Interval]
.0510542
.0049119
-.0007576
-.0437144
- pottier_78
- pother_87
- pother_82
- pother_92
.11 -
a
s
a
-.07-
distwel
84
Fitted pother by distance from nearest Woburn site (km)
5.2.5 Children under 5
Regression with
page
underS 1
Idisw |
trend !
Idiswy |
cons 1
robust standard errors
Coef .
.0016573
.0002899
.0002793
.0563436
Robust
Std. Err.
.0006504
.0000513
.0000457
.0007246
t
2.55
5.66
6.11
77 .76
P>|t |
0,
0,
0.
0,
.011
.000
.000
.000
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
195% Conf.
.0003824
.0001894
.0001897
.0549233
12444
= 547.26
= 0.0000
- 0.0856
= .01342
Interval]
.0029322
.0003903
.000369
.0577639
-------
345
- page_under5_7fl
- page_undeiS_87
-page under5 82
- page_under5_92
.08-
S
Q.
.05-
0
distwel
8.4
:itted page_under5 by distance torn nearest Wobum site (km
5.2.6 Persons between 5 and 29
Regression with robust standard errors
Number of obs = 12444
F( 3, 12440) = 4739.40
Prob > F = 0.0000
R-.squared = 0.4156
Root MSE = .03848
1
page_5_29 1
Idigw |
trend I
Idiswy I
cons 1
Cosf .
.0172771
-.0047735
-.0013235
.4028142
Robust
Std. Err.
.0017827
.0001676
.0001303
.0022685
t
9.69
-26.49
-10.16
177.57
P>|t|
0.000
O.ODO
0.000
0.000
[95% Conf.
.0137827
-.OO.H02
-.001r)789
.3993676
Interval]
.0207714
-.0044451
-.0010681
.4072606
- page 5 29 78
-pase_5_29_87
- page_5_29_82
-page_5_29_92
.45-
.28-
distwel
8.4
Fitted page_5_29 by distance from nearest Wobum site (km)
-------
346
5.2.7 Persons between 30 and 64
Regression with
robust standard errors
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
1
page 30 54 1
Idisw I
trend I
Idiswy |
cons 1
Coef .
.0058642
.002485
.0006111
.4135398
Robust
Std. Err.
.0013361
.0001163
.0000905
.0016954
4.
21
6
245.
t
.39
.36
.75
.37
P>|t I
0.000
0.000
0.000
0.000
[95% Conf.
.0032452
.002257
.0004337
.4102362
12444
= 2582.91
= 0.0000
= 0.3111
= .02587
Interval]
.0084832
.0027131
.0007885
.4168435
-page 30_64_7B
- pagOO_64_87
-page_30_64_B2
-page_30_64_92
.9
I
4-
0
8.4
distwel
Fitted page_30_64 by distance from nearest Woburn site (km)
5.2.8 Persons 65 and older
Regression with robust standard errors
Number of obs = 12444
F( 3, 12440) = 988.59
Prob > F = 0.0000
R-squared = 0.1826
Root MSE = .03314
page 65 up I
Coef.
Robust
Std. Err.
P>l 11
[95% Conf. Interval]
Idisw I -.0216667 .0015527
trend I .0020491 .0001788
Idiswy I .0004368 .0001241
cons I .1236413 .0022441
-13.95 0.000
11.46 0.000
3.52 0.000
55.10 0.000
-.0247101 -.0186232
.0016986 .0023996
.0001935 .00068
.1192425 .12804
-------
347
-page_65_up_7B
•page~65_up~87
-paae_65 up_82
-paae_65_up_92
18-
.07-
distwel
84
Fitted page_65_up by distance from nearest Wobum site'(km)
5.2.9 Married heads of household
Regression with robust standard errors
Number of obs = 12444
F( 3, 12440) = 1283.77
Prob > F = 0.0000
R-;-;quar<=d = 0.2347
Root MSE. = .06565
1
pmarhh chd 1
Idiaw I
trend I
Idiswy I
_cons 1
Coef .
.0410018
-.0058504
-.0001361
.3268933
Robust
Std. Err.
.0034139
.0003632
.0002568
.0048437
12,
-16,
-0,
67,
t
.01
.11
.53
.49
P>
0.
0.
0.
0.
1 1 1
000
000
596
000
(95% Conf.
.034:il01
-.006!. 623
-.OOOiJ396
.317:5989
Interval]
.0476936
-.0051385
.0003673
.3363877
- pmarhh_chd_78
-pmarhh~chd_87
-pmarhh_chd 82
- pmartm_chdl92
.43-
2
a.
.18-
6
dlstvtel
84
Fitted pmarhh_chd by distance from nearest Woburn site (km)
-------
348
5.2.10 Male-headed of household with children
Regression with
1
pmhh child I
Idisw !
trend I
Idiswy |
cons I
robust standard errors
Cos f .
.0026316
,0002315
-.0001315
.0045081
Robust
Std. Err.
.0002132
.0000177
.0000153
.0002412
t P> 1 t I
12.34 0
13.09 0
-8.51 0
18.69 0
.000
.000
.000
.000
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
[95% Conf.
.0022137
.0001968
-.0001614
.0040354
12444
- 129.82
= 0.0000
= 0.0144
- .00534
Interval ]
.0030496
.0002662
-.0001015
.0049808
- pmhh_child_78
- pmhh_child_87
- pmhh_ctllld_82
-pmhh_child_92
02-
-.01-
T
8.4
distwel
Fitted pmhh_child by distance from nearest Woburn site (km)
5.2.11 Female-headed households with children
Regression with
1
pfhh_child 1
Idisw I
t rend 1
Idiswy i
cons 1
robust standard errors
Coef .
.0158892
.0012692
-.0010325
.026495
Robust
Std. Err.
.0012568
.0001026
.0000861
.0014844
t
12.64
12.38
-11 .99
17.85
P> 1 1 I
0.000
0.000
0 .000
0.000
Number of obs
F( 3, 12440)
. Prob > F
R— squared
Root MSE
[95% Conf.
.0134256
.0010681
-.0012013
.0235853
12444
57.43
= 0.0000
= 0.0089
= .03134
Interval ]
.0183528
.0014702
-.0008637
.0294047
-------
349
-pfhh child_78
-pfhh~child_87
-pfhh child 82
- pf hhlchild_92
.07-
02-
6
distwel
6.4
Fitted pftih_child by distance from nearest Wobum site (km)
5.2.12 Owner-occupancy
Regression with
1
powner occ I
Idiaw I
trend 1
Idiswy |
cons i
robust standard errors
Coef .
.0387542
-.0058857
.0022567
.7266904
Robust
Std. Err.
.0067619
.0007026
.0004911
.0095303
t
5.73
-8.38
4.60
76.25
P>|t |
0.000
0.000
0.000
0.000
Numbe r of obs
F( 3, 12440)
Prob > F
R-squared
Root MSB',
[95% Conf.
.0254998
-.007:!629
.001294
.708)095
12444
= 247.27
- 0.0000
= 0.0548
= .16376
Interval!
.0520086
-.0045085
.0032193
.7453713
- powner_occ_78
- powner_occ_8?
- povwter_occ_82
- powner_occ_92
.796959-
•e
a
s
Q.
.54-
distwel
Fitted powner_occ by distance from nearest Woburn site (km)
-------
350
5.2.13 Renter-occupancy
Regression with
prenter occ 1
Idisw I
trend !
Idiswy I
cons 1
robust standard errors
Coe f .
-.0474617
.005599
-.0022182
.25679
Robust
Std. Err.
.0061438
.0006564
.0004497
.0088694
-7
8
-4
26
t
.73
.53
.93
.35
P> 1 1 1
0.000
0.000
0.000
0.000
Number of obs
F{ 3, 12440)
Prob > F
R-squared
Root MSE
[95% Conf.
-.0595044
.0043124
-.0030996
.2394047
12444
= 368.05
= 0.0000
= 0.0800
= .13685
Interval]
-.035419
.0068856
-.0013368
.2741754
- prenter occ_78
- prenterocc_87
- prenter occ_82
- prenter_occ_92
432238-
o
Q.
0-
T
distwel
84
Fitted prenter_occ by distance from nearest Woburn site (km)
5.2.14 Vacancy rates
Regression with
pvacant 1
Idisw |
trend 1
Idiswy I
cons I
robust standard errors
Coef .
.00766
.0002288
.000031
.0173909
Robust
Std. Err.
.000804
.0000664
.000056
.0009295
t
9.53
3.44
0.55
13.71
P>lt 1
0.000
0.001
0.580
0 .000
Number of obs
F( 3, 12440)
Prob > F
R-squared
Root MSE
195% Conf.
.006084
.0000986
-.0000788
.0155689
12444
= 156.39
= 0.0000
- 0.0395
= .02059
Interval]
.0092359
.000359
.0001408
.0192129
-------
351
»pvacant_78
-pvacant_87
- pvacant_82
-pvacant~92
.05-
2
a.
0-
0
distwel
e.4
Fitted p vac ant by distance from nearest Wobum site (km)
Chapter 6 Complete regression results - No lot size interactions
6.1 Just structural characteristics and year dummies
Regression with robust standard errors
I
Isprice I Cos 5.
notold
age
sqft
sqft2
bedrms
bthrtns
sqftbed
sqftbth
fplace
knowf Ir
floors
lotsize
Idisw78
Idisw79
IdiswSO
IdiswSl
Idisw82
Idisw83
Idisw84
IdiswSB
Idisw86
- Idisw87
IdiswSS
.0489693
.0010374
-.0000112
.2479951
-.0170873
-.0122332
.1598301
.0156497
-.012486
.1584083
-.1482357
.0315333
.0365814
-.0285236
-.1322543
-.0005082
-.0533517
.0077582
-.0978117
.0443689
.1042162
-.0087994
-.011426
.0430796
Robust
Std. Err.
.0186774
.0006339
7.92e-06
.0306219
.0100014
.0157373
.0216087
.0077049
.0113685
.0144585
.0251731
-0122127
.0070445
.0383282
.0741523
.0618726
.0843768
.053554
.0542451
.0377947
.0440051
.0439745
.0331795
.0279399
t P> 1 t I
2.62
1.64
-1.42
8.10
-1.71
-0.78
7.40
2.03
10.96
-5.B9
2.58
5. IS
-0.74
-1.78
-0.01
-0.63
0.14
-l.SO
1.17
2.37
-0.20
-0.34
1.54
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.009
.102
.157
.000
.088
.437
.000
.042
.272
.000
.000
.010
.ooc
.457
.075
.993
.527
.885
.071
.240
.018
.841
.731
.123
Nurrtber of obs = 12444
F( 52, :2391) = 248.46
Prob > F = 0.0000
R-squar
-------
352
Idisw89 I
Idisw90 I
Idisw91 I
Idisw92 I
Idisw93 1
Idisw94 I
Idisw95 I
Idisw96 !
Idisw97 I
year79 I
yearSO 1
yearSl I
year32 1
year 93 •
year34 i
year35 I
year86 I
year87 j
year39 i
year89 I
year90 I
year91 1
year92 i
year 93 I
ye a r 9 4 I
year95 1
ye a r 9 6 I
year97 I
cons 1
.0371748
.0521827
.050278
.0737088
.0793316
.0416848
.1094682
.1031653
.0463116
.2503722
.1730845
.333171
.3156999
.5386808
.5379653
.6560054
1.114607
1.201649
1.164397
1.175654
1.120914
1.037364
1.002075
1.024214
1.039912
1.077126
1.14618
1.216637
10.11873
.027407
.0270853
.0285997
.0279899
.0220493
.024991
.0248348
.0182737
.0362596
.1048783
.1085707
.122066
.0983298
.093381
.0752274
.0852122
.0795717
.0753569
.0695533
.0699542
.0684935
.0695909
.0698281
.0659142
.066143
.0672051
.0631965
.0716661
.0730562
1
1
1
2
3
1
4
5
1
2
1
2
3
5
7
7
14
15
16
16
16
14
14
15
16
16
18
16
138
.36
.93
.76
.63
.60
.67
.41
.65
.28
.39
,59
.73
,21
,77
.15
.70
.01
.95
.99
.81
.37
.91
.35
.54
.48
.03
.14
.98
.51
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
c
0
c
0
0
0
0
0
.175
.054
.079
.008
.000
.095
.000
.000
.202
.017
.111
.006
.001
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
.000
-.0165471
-.0009087
-.0057818
.0188443
.0361106
-.0073014
.0607882
.0673459
-.0247629
.0447945
-.0397309
.0939027
.1229582
.3556396
.3905079
.4889762
.9586336
1.053938
1.030022
1.038533
.9866558
.9009554
.865201
.8950119
.9602613
.9453934
1.022305
1.07616
9.975527
.0908968
.1052741
.1063379
.1285734
.1225527
.090671
.1581433
.1389846
.1173861
.4559499
.3858999
.5724393
.5084415
.7217221
.6854228
.8230345
1.27058
1.349361
1.298772
1.312775
1.255172
1.173773
1.138949
1.153416
1.219562
1.208858
1.270055
1.357114
10.26193
All
All
All
Hypothesis P-VJ
ofF
due Reject @
-test 5% level?
structural attribute slopes simultaneously zero o . o o o o
year-specific slopes on LDIST simultaneously zero o . oooo
year-specific slope on LDIST the same o . 0010
6.2 Including Census tract attributes
Regression with
i
Isprice 1
notolc 1
age 1
robust standard errors
Coef .
.1250637
-.0014915
Robust
Std. Err. t
.0186467 6.71
.0006219 -2.40
Number of obs
F( 63, 12380!
Prob > F
R-squared
Root MSE
P>lt! [95% Conf.
0.000 .0885133
0.016 -.0027106
12444
= 253.09
= 0.0000
= 0.5265
= .41375
Interval ]
.161614
-.0002724
-------
353
age2
sqft
sqft2
bedrms
bthrms
sqftbed
sqftbth
fplaca
knowf Ir
floors
lotsize
Idisw78
Idisw79
IdiswSO
IdiswSl
Idisw82
Idisw83
Idisw84
Idisw85
Idisw86
IdiswB?
IdiswSB
IdiswSS
Idisw90
Idisw91
Idisw92
Idisw93
Idisw94
Idisw95
Idisw96
Idisw97
pfemalas
pblack
pother
page underB
page 5 29
page 65 up
pma rhh c hd
pmhh child
pfhh child
pvacant
prenter occ
year?9
yearSO
yearBl
year82
year83
year84
year85
yearSG
year87
year88
year89
year90
year91
year92
year93
year94
year95
year96 1
year 97 I
5.09e-06
.1971115
-.0165331
.002461
.1053631
.014"4?8
-.0088652
.0818046
-.0506454
.012164
.0399528
-.C751494
-.1632537
-.0203892
-.1036433
-.0416704
-.1412171
-.1057222
-.0239091
-.1010463
-.130573
-.0772486
-.083541
-.0732827
-.0624389
-.0390139
-.0293196
-.088817
-.010883
-.0126859
-.0572987
5.006331
-1.663042
2.064265
-2.822898
-6.98893
-3.650558
1.497224
3.084206
1.664087
4.66123
.0094355
.1964707
.0964912
.2823778
.2492813
.4342925
.5763723
.6411623
.9610217
1.069124
.9772057
. 9572482
.9056303
.7918292
.7353282
.7145283
.7645767
.6921696
.7058066
.725549
7.65e-06
.0296864
.0094491
.0152669
.0207582
.0074223
.0108892
.0138499
.0274446
.0117416
.0071743
.0374318
.072774
.0607769
.0833055
.0538218
.0514768
.0395021
.0418568
.0423291
.0340084
.0292287
.0289126
.0279338
.0284187
.0295106
.0232881
.0255264
.0253123
.019582
.0359245
.8314503
1.580983
.4791806
1.00667
.5387391
.4198468
.3315558
2.517625
.856072
.7967777
.1820208
.1018905
.1054726
.1204158
.0977224
.0893982
.0754589
.0813667
.0779536
.0783105
.0725633
.0740356
.0727857
.0735363
.0742926
.0715639
.0724445
.0755341
.0742792
.0861535
0.67
6.64
-1.79
0.16
5.08
1.99
-0.81
5.91
-1.85
1.04
5.57
-2.01
-2.24
-C.34
-1 .24
-0.77
-2.74
-2.68
-0.69
-2.39
-3.84
-2.64
-2.89
-2.62
-2.20
-1.32
-1.26
-3.48
-0.43
-0.65
-1.59
6.02
-1.05
4.31
-2.80
-11.87
-8.69
4.49
1.23
1.94
5.85
0.05
1.93
0.91
2.35
2.55
4.86
7.64
7.88
12.33
13.65
13.47
12.93
12.44
10.77
9.90
9.98
10.55
9.16
9.50
8.42
0.505
0.000
0.073
0.872
0.000
0.04T
0.416
0 .000
0.065
0.300
0.000
0.045
0.025
0.737
0.213
0.439
0.006
0.007
0.490
0.017
0.000
0.008
0.004
0.009
0.028
0.186
0.208
0.001
0.667
0.517
0.111
0.000
0.293
0.000
0.005
0.000
0.000
0.000
0.221
0.052
0.000
0.958
0.054
0.360
0.019
O.C11
0.000
0.000
0.000
o.occ
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
-?i.90e-06
.1389216
-.0354E9S
-.02746.45
.0646189
,COC1'-S9
-.Q302C78
.0546!. 66
-.:o44<;ii
-.0108.'. 15
.0258891
-.1435.215
-.3059021
-.1395-15
-.266335
-.1471696
-.2421196
-.1831525
-.110955
-.1840178
-.1972348
-.1345415
-.1402143
-.123C373
-.1181439
-.0966593
-.0745-679
-.1386.527
-.06CK991
-.051(>696
~.127'-163
3.37''059
-4.76:.'014
1.12-1996
-4.79 5127
-8.1 1295
-4.47 3523
.8373234
-1.850731
-.0133479
3.099471
-.3473036
-.0032504
-.1102515
.0463441
.0577303
.2590581
.4264611
.4E1671
.8062204
.9156238
.834. 9704
.8U1269
.7629591
. 64" 6366
.5397032
.57.52519
.62;'.' 57 4 2
.5-14111
.5602079
.5556747
.0000201
.2553015
.0015835
.0323865
.1460573
.0292966
.0124774
.1089526
.0031504
.0351795
.0540165
-.0017773
-.0206053
.098743
.0596485
.0638288
-.0403147
-.0282918
.0531368
-.0180748
-.0639113
-.0199557
-.0269677
-.018528
-.006734
.0188315
.0163287
-.0387814
.038733
.0256978
.013119
6.636603
1.43593
3.003533
-.9496692
-5.83491
-2.327593
2.137125
8.019144
3.342121
6.223088
.3662745
.3961919
.3032339
.5184115
.4408323
.6095268
.7242835
.8006536
1.113823
1.222625
1.119441
1.10237
1.048302
.9359718
.3809532
.8548046
.9065791
.8402281
.8514053
.3944233
-------
354
cons I 10.36435 .4859749
21.33 0.000
9.411766
11.31694
AH
All
All
All
Hypothesis P-v
ofF
alue Reject @
-test 5% level?
structural attribute slopes simultaneously zero ° • 000°
year-specific slopes on LDIST simultaneously zero o . oooo
year-specific slope on LDIST the same 0.0590
Census tract characteristic effects simultaneously zero o . oooo
6.3 Including other distances
Regression with robust standard errors
Isprice
not old
age
age2
sqft
sqft2
bedrms
bthrms
sqftbed
sqftbth
fplaca
knowf Ir
floors
lotsize
Idisw79
Idisw79
IdiswSO
IdiswSl
Idisw82
Idisw83
Idisw84
IdiswSS
Idisw86
Idisw37
IdiswSB
Idisw89
IdiswSC
Idisw91
Idisw92
Idisw93
Idisw94
Coef .
.1569426
-.0023662
.0000164
.1811277
-.0121392
.0081811
.1071896
.0115474
-.0125214
.0874144
-.3360679
.0150786
.0441894
-.0316017
-.1858336
-.0535433
-.1366499
-.0940644
-.1710932
-.1121743
-.0180357
-.1164692
-.1179319
-.0808622
-.0813797
-.0872954
-.0853543
-.054765
-.0341513
-.1061577
Robust
Std. Err.
.0196394
.0006303
7.72e-06
.0295521
.0093027
.0150692
.0205156
.007288
.0107615
.0139716
.0329574
.012128
.0073256
.042338
.0752901
.0637115
.0864024
.0565769
.055943
.0425505
.0467954
.047105
.0379481
.0338971
.033095
.0328177
.0338196
.0355136
.023733
.029993
7
-4
2
6
-1
0
5
1
-1
6
-10
1
6
-1
-2
-C
-1
-1
_-5
-2
-0
-2
-3
-2
-2
—2
-2
•" 1
-1
-3
t
.39
.55
.12
.13
.30
.54
.22
.58
.16
.26
.20
.24
.03
.93
.47
.84
.58
.66
.06
.64
.39
.47
.11
.39
.46
.66
.52
.54
.IS
.54
P> 1 1 I
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.000
.000
.034
.000
.192
.587
.000
.113
.245
.000
.000
.214
.000
.054
.014
.401
.114
.096
.002
.008
.700
.013
.002
.017
.014
.008
.012
.123
.235
.000
Number of obs = 12444
F( 75, 12368) = 226.64
Prob > F = 0.0000
R-squared = 0.5346
Root MSE = .41041
(95* Conf
.1184463
-.0041017
1.26e-06
.1232009
-.0303739
-.021357
.0669759
-.0027381
-.0336157
.0600279
-.4006695
-.0086942
.0298301
-.1645907
-.3334139
-.1784283
-.3060121
-.2049639
-.2807503
-.1955798
-.109762
-.2088023
-.1923161
-.1472862
-.146251
-.1516232
-.151646
-.1243771
-.0904729
-.1649585
Interval )
.195439
-.0016306
.0000315
.2390545
.0060955
.0377192
.1474034
.0258329
.0085728
.1148009
-.2714663
.0388514
.0585487
.0013873
-.0382532
.0713406
.0327123
.0168351
-.0614362
-.0287687
.0736907
-.024136
-.0435476
-.0144381
-.0165084
-.0229675
-.0190626
.0148471
.0221694
-.047357
-------
355
Idisw95
Idisw96
Idisw97
Id summits
Id school
Id retail
Id hospital
Id church
Id cemetery
Id railroad
Id prinarte
Id othpriro
Id ma roads
Id 195
Id_i93
Id fp tewma
Id fp milit
Id fp logan
Id fp bevtnu
Id parks
Id mjwater
Id cclubs
Id tewmae
Id military
Id logan
Id bevmuni
year79
yearBQ
yearBl
year82
yearS3
year84
year85
year96
year87
year88
year89
yearSO
yearSl
year92
year£)3
year94
yearSS
year96
year97
cons
-.C355342
1 -.C4663C5
-.1076644
-.012E029
-.0308617
-1.445565
.0944295
.4647825
-.0417229
.0172871
-.0236997
.0220981
.0068839
-.0190925
.0377832
-.0228639
-.0081251
-.0130995
.0101698
-.0188613
-.0383665
-.0183595
.1830531
1.124274
1.681858
1.783187
.2456496
.1790832
.3597841
.3684415
.5783117
.7248305
.8107106
1.235729
1.343715
1.322354
1.320379
1.305581
1.207136
1.161401
1.167479
1.277792
1.266398
1.333023
1.413227
-26.33767
.0299655
.0254737
.0385314
.0070101
.006655
.1539512
.017747
.1418528
.0088408
.0059608
.OOS10S7
.0044153
.0028982
.0114279
.007746
.006108
.0057943
.0094828
.0075557
.0075207
.0106903
.00524
.0993552
.134809
.1706983
.2342924
.0992993
.1033019
.1195308
.0936482
.0856281
.0720866
.0795955
.0747583
.0717565
.0646663
.0676762
.0633565
.0649972
.0648116
.0626225
.0618888
.062091
.0581002
.0663079
5.550022
-1 .19
-1.93
-2.79
-1.33
-4 .64
-9.39
5.32
3.29
-4.72
2 .90
-2.92
5.00
2.38
-1 .58
4 .88
-3.74
-1.40
-1 .38
1 .35
-2,51
-3.59
-3.50
1.84
8.34
9.85
7.63
2.47
1.73
3.01
3.93
6.75
10.06
10.19
16.53
18.73
20.45
13.51
2C.61
18, 57
17 .92
18.64
;:o.65
20.40
22 . 94
:>;.3i
-4.75
0.235
0.067
0.005
0.063
0.000
C.OOO
0.000
0.001
0.000
0 .004
0.003
0.000
0.018
0.113
0.000
0.000
0.161
0.167
0.178
0.012
0.000
0.000
0.065
0.000
0.000
0.000
0.013
0.085
0.003
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0 .000
0.000
0.000
-.0943213
-.096663
-.1831919
-.0265439
--0439C65
-:.. 7471-34
.0596426
.1S67;:S9
-.D590S21
.0056029
-.0395'J39
.0134.134
.0012)29
-.0404 329
.0225399
-.0348-416
-.0194328
-.0316373
-.0046415
-.0336029
-.0593212
-.0286307
-. 0116985
.S60C276
1.347262
1.32E937
.051C074
-.024J847
.1254851
.1848763
.410.',673
.583!>296
.654 J909
i. oenai
1.205061
1 .19.J593
1 .187723
1.181392
1 .079731
1.03436
1.044729
1.15648
1.14469
1 .219138
1.283254
-37.21657
.0231529
.0033019
-.032137
.000938
-.0178163
-1.143797
.1292164
.7428361
-.0243935
.0289713
-.0073054
.0307528
.0125649
.004308
.0529665
-.0108963
.0032327
.0054883
.0249791
-.0041196
-.0174118
-.0080882
.3778047
1.338521
2.016453
2.247436
.4402918
.3825512
.5940831
.5520066
.7461561
.8661314
.9667302
1.382267
1.484369
1.44911
1.453035
1.42977
1.33454
1.288442
1.290229
1.399103
1.388106
1.446908
1.543201
-15.45876
Hypothesis
All structural attribute slopes simultaneously zero
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
P-value
ofF-test
0.0000
0.0098
0.1207
Reject @
5% level?
NO
-------
356
All other distance effects simultaneously zero
0.0000
6.4 Including both other distances and tract attributes
Regression with robust standard errors Number of obs = 12444
F( 86, 12357) = 208.24
Prob > F = 0.0000
R-squared = 0.5418
Root MSE = .40739
Isprice
notold
age
age2
sqft
sqft2
bedrms
bthrms
sqftbed
sqftbth
fplace
knowf 1 r
floors
lot size
Idisw78
Idisw79
IdiswSQ
IdiswSl
Idisw82
Idisw83
Idisw84
Idisw35
Idisw86
Idisw87
Idisw83
idisw33
Idisw90
Idisw91
Idisw92
Idisw93
Idisw94
Idisw95
Idisw96
Idisw97
Id summits
Id school
Id^retail
Id hospital
Id church
Id cemetery
Id railroad
Id prinr-irt.e
Coe f .
.1644526
-.0031931
.0000183
.1750219
-.0129308
.0081091
.0974569
.0121049
-.0119339
.0640722
-.2486657
.0122557
.047623
-.1183986
-.2127232
-.0720667
-.1564728
-.1156535
-.1990855
-.1771748
-.0842971
-.1607007
-.1774609
-.1327165
-.1365158
-.1376485
-.1279449
-.0981975
-.0819654
-.1525042
-.0810414
-.0921013
-.1447539
-.0195731
-.031723
-.8727287
.074898
-.0182594
-.0066797
.0154953
-.0268746
Robust
Std. Err.
.0195503
.0006327
7.72e-06
.0293972
.0092274
.0150001
.0203827
.0072525
.0106868
.0136916
.039298
.0120643
.0073779
.0442745
.0773112
.0650719
.0874353
.0587284
.0559997
.0452931
.0469693
.0480024
.0404264
.0364416
.0355327
.0362797
.0368466
.0384945
.0311738
.0318486
.0326235
.0286108
.0409226
.0075343
.0063727
.1640059
.0184074
.1565671
.0094331
.0061249
.0083477
8
-5
2
c;
-1
0
4
1
-1
4
-6
1
6
-2
-2
-1
-1
-1
~- "^
-3
-1
-3
-4
_3
_3
-3
-3
-2
_2
-4
-2
-3
-3
-2
-4
-5
4
-0
-0
2
-3
t
.41
.05
.37
.95
.40
.54
.73
.67
.12
.68
.33
.02
.45
.67
.75
.11
.79
. 97
.56
.91
.79
.35
.39
.64
.84
.79
.47
.55
.63
.79
.48
.22
.54
.60
. 62
32
.07
. 12
.71
. 53
.22
P> 1 1 1
C .000
c.ooo
0.018
0.000
0.161
0.589
0.000
0.095
0.264
0.000
0.000
0.310
0.000
0.008
0.006
0.268
0.074
0.049
0.000
0.000
0.073
0.001
0.000
0.000
0.000
0.000
0.001
0.011
0.009
0.000
0.013
0.001
0.000
0.009
0.000
0.000
0.000
0.907
0.479
0.011
0.001
[95* Conf
.12613
-.0044332
3.13e-06
.1173987
-.031018
-.0212933
.0575037
-.0021112
-.0328817
.0372345
-.3256959
-.0113921
.0331612
-.2051836
-.3642652
-.1996178
-.3278596
-.2307703
-.3088536
-.2659564
-.1763642
-.2547929
-.2567029
-.2041478
-.2062635
-.2087622
-.20017
-.1736527
-.143071
-.2149324
-.1449885
-.1481829
-.2249686
-.0343414
-.0451996
-1.194206
.0388167
-.3251553
-.02517
.0034894
-.0432374
Interval]
.2027752
-.001953
.0000334
.232645
.0051564
.0375115
.1374102
.026321
.0090138
.0909099
-.1716354
.0359036
.0620847
-.0316137
-.0611812
.0554844
.0149141
-.0005367
-.0893174
-.0883933
.00777
-.0666085
-.098219
-.0612852
-.066768
-.0665347
-.0557198
-.0227423
-.0208593
-.0900759
-.0170942
-.0360197
-.0645392
-.0048047
-.0182564
-.5512516
.1109793
.2886366
.0118107
.0275011
-.0105119
-------
357
Id othpriro j .0172256 .0044938 3. S3
Id ma roads .005349 .0029124 1.34
Id 195
Id_i93
Id fp tewma
Id fp mi lit
Id fp logan
Id fp bevrau
Id parks
Id mjwater
Id cclubs
Id tewroac
Id military
Id logan
Id bevmuni
pfemales
pblack
pother
page underS
page 5 29
page 65 up
pmarhh chd
pmhh child
pfhh_child
pvacant
prenter occ
year79
year90
year91
year82
year83
year84
yearSS
year86
year87
yearSS
year 8 9
yearSO
yearSl
year92
year 93
year 9 4
year95
year96
year 97
cons
-.0115119 .011621 -0.99
.0228202 .0086358 2.63
-.0038737 .0062309 -1.42
.0061882 .0059402 1.04
-.0010185 .0097232 -0.10
.0087479 .0076922 1.14
-.020464 .0077396 -2.64
-.0425965 .0108006 -3.94
-.0091896 .0056494 -1.63
.1C59918 .1216317 C.87
.4444311 .1671331 2.66
1.161737 .2011811 5.77
.6839151 .2874421 2.33
3.422701 .9781858 3.50
.7426583 1.829367 (1.41
.1730552 .5284539 0.33
-1.922992 1.15022 -1.67
-3.006824 .7212157 -4.17
-.2182788 .5168144 -0.42
1.889583 .3605468 5.24
9.428491 3.218717 2.62
.692235 1.005446 0.69
3.04204 .8719766 3.49
.2323277 .1902791 1.22
.1979666 .0994367 1.99
.1091657 .104416 1.05
.2954158 .1209502 2.44
.3025366 .0951893 3.18
.510794 .0860329 5.94
.7186159 .0741993 9.68
.7976214 .0806764 9.89
1.16927 .0777086 15.05
1.295141 .0789244 16.41
1.249451 .0748807 36.69
1.242527 .C773055 16.07
1.220508 .0755873 16.15
1.109959 .0766219 14.49
1.054543 .0773724 13.63
1.047361 .0760047 13.78
1.137708 .0774669 14.69
1.10167 .0795231 13.85
1.148644 .0802195 14.32
1.199211 .0919314 13.04
-5.593676 7.165651 -0.78
0.000
0.066
0.322
0.009
0.154
C.298
0.917
0.255
0.008
0.000
0.104
C.3S4
0.008
0.000
0.017
0.000
0.685
0.743
0.095
0.000
0.673
0.000
0.009
0.491
0.000
0.222
0.047
0.296
0.015
0.001
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.436
.0064172 .0260341
-.0003E98 .0110577
-.0342S-08 .0112671
.0357<4"> .0398457
-.0210?'72 .0033393
-.0054556 .0178319
-.0200'74 .0180404
-.C063;:93 .0233258
-.0356349 -.0052931
-. 0637673 -.0214256
-.0202534 .0018841
-.1324254 .344409
.1168.242 .772038
.7673903 1.556083
.1204339 1.247346
1.505305 5.340098
-2.843186 4.328503
-.8627967 1.208907
-4.1776C2 .3316187
-4.420519 -1.593129
-1.231316 .7947579
1.182855 2.59631
2.119303 14.73768
-1. 278595 2.663065
1.3:283 4.75125
-.140f471 .6053025
.003(1552 .392878
-.095;'058 .3138373
.0583347 .532497
.116(1008 .4891725
.342';.561 .6794319
.573:737 .864058
.639.1831 .9557598
1.01695 1.321591
1.14)437 1.449845
1.102673 1.396229
1.090996 1.394058
1.072346 1.368671
.9597683 1.26015
.9028858 1.20621
.8983803 1.196342
.9858611 1.289555
.9457925 1.257548
.9914009 1.305886
1.019011 1.379411
-19.^2947 8.46211S
Hypothesis
All structural attribute slopes simultaneously zero
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All other distance effects simultaneously zero
All Census tract characteristic effects simultaneously
zero
P-value Reject @
ofF-test 5% level?
0.0000
0.0002
0.1518 NQ
o.ooco
0.0000
-------
358
Chapter 7 Complete Regression Results - Models exploring absolute
directional effects
7.1 Including latitude and longitude linear shifters
Regression with robust standard errors
Isprice
PiOtold
age
age2
sqft
sqft2
bedrrns
bthrms
sqftbed
sqftbth
fplace
knowf Ir
floors
lotsize
Idisw79
Idisw79
IdiswSO
IdiswSl
Idisw52
Idisw83
ldiswS4
Idisw85
Idisw86
Idisw87
IdiswSS
Idisw89
idisw90
Idisw91
Idisw92
Idisw93
Idisw94
Idisw95
Idisw96
Idisw97
latl 78
latl 79
latl 80
latl 81
latl 82
Coef .
.1678627
-.0032195
.000018
.172859
-.0120145
.0086916
.0986794
.0117746
-.0127978
.0628759
-.2792558
.0156098
.0480764
-.0865772
-.1705382
-.0316755
-.0613865
- .0615999
-.1353376
-.1910096
-.0655846
-.1230856
-.1801952
-.1482044
-.1766546
-.1567063
-.1166325
-.071312
-.0713309
-.1246955
-.0718703
-.071406
-.1210283
-9.99526B
-10.90935
-11.52204
-10.50789
-10.35135
latl" S3 ! -10.53996
Robust
Std. Err.
.0195833
.0006334
7.72e-Q6
.0292987
.0091985
.0149955
.0200734
.0072313
.0105102
.0136873
.0419747
.0120067
.007399
.051932
.0837022
.0770966
.0820256
.0729772
.0691343
.0466838
.0510681
.0501139
.0430095
.0410086
.0390364
.0387646
.039175
.0408806
.0330497
.0331211
.0341138
.0302158
.0391171
3.344724
3.402743
3.395527
3.516509
3.405715
3.41709
9
-5
2
t;
-1
0
4
1
-1
4
-6
1
6
-1
-2
-0
-0
-0
-1
-4
-1
-2
-4
_3
-4
-4
_ '}
-1
— ?
-3
-2
-2
_ -i
-2
-3
-3
_2
-3
-3
-.
.57
.08
.33
.90
.31
.58
.92
.63
.22
.59
.65
.30
.50
.67
.04
.41
.75
.84
.96
.09
.28
.46
.19
.61
.53
.04
.06
.74
.16
.76
.11
.36
.09
.51
.21
.39
.99
.04
.08
Number of obs = 12444
F(107, 12317) =
Prob > F
R-squared = 0.5457
Root MSE = .40634
P>l t I
0 .000
0.000
0.020
0 .000
0.192
0.562
0 .000
C.103
C.223
0.000
0.000
0 .194
0.000
0.095
0.042
0.681
0.454
0.399
0.050
0.000
0.199
0.014
0.000
0.000
0.000
0.000
0.002
0.081
0.031
0.000
0.035
0.018
0.002
0.003
0 .001
0.001
0.003
0.002
0.002
195% Conf
.1294764
-.0044611
2.85e-06
.1154485
-.0300451
-.020702
.0593323
-.0023998
-.0333993
.0360467
-.3615327
-.0079252
.0335732
-.188176
-.3346076
-.1827969
-.2221696
-.2046456
-.2708516 ,
-.2825269
-.165686
-.2213167
-.2645005
-.2285877
-.253172
-.2326909
-.1914615
-.1514445
-.1361115
-.189618
-.1387393
-.1306336
-.1977039
-16.55145
-17.57926
-18.17781
-17.40079
-17.02709
-17.23699
Interval )
.2062491
-.0019778
.0000331
.2302695
.0060161
.0380852
.1380265
.025949
.0078038
.0897052
-.1969788
.0391448
.0625797
.0150216
-.0064687
.1194459
.0993965
.0814479
.0001764
-.0994922
.0345169
-.0248546
-.0958898
-.0678211
-.1001373
-.0807216
-.0418035
.0088204
-.0065504
-.0597729
-.0050024
-.0121784
-.0443528
-3.439085
-4 .239439
-4.866278
-3.614977
-3.67562
-3.940927
-------
359
latl 84
latl 85
latl 36
latl 87
latl 88
latl 89
latl 90
latl 91
latl 92
latl 93
latl 94
latl 95
latl 9S
latl_97
longl 78
longl_79
longl 80
longl 81
longl 82
longl 83
longl 84
longl 85
longl 86
longl 87
longl 88
longl 89
longl 90
longl 91
longl 92
longl_93
longl 94
longl_95
longl 96
longl 97
Id summits
Id school
ld_retail
Id hospital
Id church
Id cemetery
Id railroad
Id prinarte
Id othpriro
Id ma roads
Id i95
Id_i93
Id fp tewma
Id~fp3nilit
Id fp logan
Id fp bevjnu
Id parks
Id mjwater
Id cclubs
Id tewmac
Id military
Id logan
Id bevmuni
1 -10.6434
-9.367841
-10.16724
-11.01324
-10.97404
-10.80833
-10.22694
-10.15712
-10.22132
-10.33204
-9.943703
-11.05756
-1C. 6092
-11.4889
17.07279
17.22039
17.17941
19.01076
17.53915
17.71589
15.34552
16. 94392
17.2224
15.41344
14.97575
13.75165
15.26085
16.30977
16.88479
16.28274
16.98656
16.21568
16.78047
16. £5555
-.0118507
-.026592
-.6543589
.096675
.1025613
-.0032977
.0208018
-.0237586
.0173061
.0047173
-.0067621
.0287577
-.0046972
-007C074
-.0148987
.004843
-.0244976
-.0415827
-.0083934
-.1159348
.5655227
3.225574
3.196433
3.449506
3.373491
3.386186
3.393283
3.382271
3.413203
3.399833
3.393799
3.433291
3.382554
3.395069
3.429388
3.402489
3.473529
4 .506995
4.580658
4.596946
4.703044
4.605401
4.550759
4.489626
4.50633
4.583381
4.550221
4.488746
4.531638
4.528317
4.545593
4.463558
4.502637
4 .48894
4 .509991
4.505312
4.597943
.0079568
.0069503
.1651116
.0188616
.1682832
.0094851
.0062375
.0083555
.004495
.0029284
.0117185
.0100969
.0063882
.0059584
.010372
.0077896
.0077814
.0103703
.0058236
.1399692
.2230279
.5657666
.7770825
-3.09
-2.78
-3 .00
-3.25
-3.24
-3.17
-3.01
-2.99
-2.93
-3.05
-2.90
-3.22
-3.12
-3.31
2.79
3.76
3.74
4 .04
3.82
3.89
3.42
3.76
3.76
3.39
3.34
3.03
3.37
3.59
3.78
3.62
3.78
3.60
3.72
3.67
-1.49
-3.83
-3.96
5.13
0.61
-0.35
3.33
-2.84
3.85
1.61
-G.58
2.85
-0.74
1.18
-1.44
0.62
-3.15
-3.83
-1.44
-0.83
2.54
5.70
4.10
0.002
0.005
0.003
0.001
o.oo:
0.002
0.003
0.003
0.003
0.002
0.004
0.001
0.002
0.001
0 .000
0.000
0.000
0.000
0.000
0.000
0.001
0.000
0.000
0.001
0.001
0.002
0.001
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.136
0.000
0.000
0.000
0.542
0.729
0 .001
0.004
0.000
0.107
0.564
0.004
0.462
0.240
0.151
0.534
0.002
0.000
0.150
0.408
0.011
0.000
0.000
-17.40997
-15.98C41
-:.6.80<;69
»17.6(.46
-:.7.60?82
-:L7.49?-74
-16.89:24
-16.8095
-16.95111
-L6.96;:37
-16. 50457
-17.77'S69
-17.27:361
-18.29756
9.238395
9.242081
8.16368
9.792D58
S. 561344
8.795691
6.545148
8.110809
8.238261
6.494297
6.177106
4.866926
6.384636
7 .399698
8.13E515
7.456.872
8.18^533
7.375596
7.949354
7.843366
-.027-J472
-.04o;:i56
-.9780035
.0597033
-.2273002
-.02188
.008 J754
-.0401366
.0084953
-.0010228
-.0297323
.0089663
-.0172191
-.0046721
-.0352295
-.0104259
-.0397504
-.0628903
-.0198086
-.3902963
.128353
2.116583
1.663229
-3.88683
-2.75527
-3.529786
-4.361872
-4.344262
-4 .117918
-3.562637
-3.504739
-3.491534
-3.701702
-3.194836
-4 .33542
-3.939795
-4.680236
25.90719
26.1997
26.19015
28.22946
26.61646
26.63609
24.14589
25.77703
26.20655
24.33259
23.77439
22.63437
24.13706
25.21985
25.63406
25.10862
25.78558
25.05617
25.61159
25.86924
.0037458
-.0129684
-.3307142
.1336467
.4324228
.0153046
.0330283
-.0073805
.0261169
.0104573
.016208
.0485491
.0078247
.0186869
.005432
.0201118
-.0092448
-.0202751
.0030218
.1584267
1.002692
4 .334565
4.709636
pfemales I 3.533212 1.010342
3.55
0.000
1.6C2782
5.563641
-------
360
pblack
pother
page under 5
page 5 29
page 65 up
pmarhh chd
pmhh child
pfhh_child
pvacant
prenter occ
year? 9
yearSO
year 31
year82
year33
year34
yearSS
year 96
year87
ysarSS
year89
year90
year91
year 52
year93
year 94
year95
ye a r 9 6
year 97
cons
2.300747 1.926858 1.19
-.2035338 .6165694 -0.33
-2.379441 1.228092 -1.94
-3.303959 .7604302 -4.34
-.487004 .5296284 -0.92
2.057904 .4039131 5.09
5.675075 3.46519 1.64
1.413925 1.11722 1.27
2.749521 .9685295 2.84
.3964303 .2032783 1.95
49.58279 97.17507 0.51
72.59571 97.0445 0.75
159.8673 113.6422 1.41
52.14509 88.58604 0.59
69.33264 85.05546 0.82
-94.32473 70.07437 -1.35
-35.00272 73.86917 -0.47
19.11967 79.69364 0.24
-73.41773 64.69125 -1.13
-106.2531 65.93876 -1.61
-200.3425 63.34656 -3.16
-117.7529 62.71998 -1.38
-46.24665 64.9046 -0.71
-2.685334 56.25792 -0.05
-40.79697 57.63423 -0.71
-11.15381 59.04563 -0.19
-14.6607 57.22312 -0.26
6.478051 55.69291 0.12
49.32666 76.23972 0.65
1586.383 381.6292 4.16
0.232
0.741
0.053
0.000
0.358
0.000
0.102
0.206
0.005
0.051
0.610
0.454
0.160
0.556
0.415
0.17S
0.636
0.808
0.256
0.107
0.002
0.060
0.476
C.962
0.479
0.850
0.798
0.907
0.518
0.000
-1.476196 6.07769
-1.412106 1.005039
-4.766693 .0278117
-4.794522 -1.813397
-1.525159 .5511506
1.266171 2.849637
-1.117241 12.46739
-.7760016 3.603852
.8510516 4.647991
-.002028 .7948886
-140.8956 240.0611
-117.6267 262.8181
-62.8893 382.6239
-121.4974 225.7876
-97.38936 236.0547
-231.6815 43.032
-179.7979 109.7924
-135.1326 173.3719
-200.2227 53.38726
-235.5034 22.99716
-324.5117 -76.17332
-240.6939 5.188062
-173.4698 80.97653
-112.9597 107.589
-153.7691 72.17514
-126.8975 104.5799
-126.827 97.50557
-102.6888 115.6449
-100.1132 198.7665
838.3296 2334.436
Hypothesis
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All slopes to north (LAT) simultaneously zero
All slopes to north (LAT) equal
All slopes to east (LONG) simultaneously zero
All slopes to east (LONG) equal
All other distance effects simultaneously zero
All Census tract characteristic effects simultaneously
zero
P-value Reject @
of F-test 5% level?
0.0002
0.0236
0.0162
0.0764 NO
0.0000
0.0000
0.0000
0.0000
7.2 Including latitude and longitude linearly and as site distance interactions
t
Regression with robust standard errors
Number of obs =
F(130, 12277) =
12444
-------
361
Isprice
notold
age
age2
sqft
sqf t2
bedrms
bthrms
sqftbed
sqftbth
fplace
knowf Ir
floors
lotsize
Idisw78
Idisw79
IdiswSO
IdiswSl
Idisw82
Idisw83
Idisw84
Idisw85
Idisw86
Idisw87
Idisw88
Idisw89
IdiswSO
Idisw91
Idisw92
Idisw93
Idisw94
Idisw95
Idisw96
Idisw97
latl 78
latl 79
latl 80
latl 81
latl 82
latl 83
latl 84
latl 85
latl 86
latl 81
latl~88
latl 89
latl 90
latl*91
latl 92
latl 93
latl 94
latl~95
1
i C:3e f .
.1693701
-.0032359
.0003183
.1725774
-.0121996
.0089724
.0953513
.0114375
-.0116592
.0653969
-.2684747
.0161969
.0476245
.235701
-.0167963
.0799902
.0835096
.1011022
.23759
-.142172
.2567009
.0162311
.003416
.3317733
.0189916
.0566923
-.103528
.0375697
.0566083
-.0456962
-.1295729
.0873205
.1048136
-7.854626
-4.54332
-3.906506
-20.31844
-1.71C448
-2.123659
-10.09993
4.02576
-3.560855
-3.543371
.9749088
-5.225542
-6.751576
-7.362134
-5.453587
-4.296164
-2.785327
-3. 920755
Robust
Std. Err.
.0195426
.0006353
7.75e-06
.0294303
.009235
.0150539
.0201763
.0072604
.0105611
.0136906
.0471905
.0120352
.0073996
.1859465
.2196915
.2585054
.2606316
.1803859
.1807092
.1961219
.156646
.1743658
.2015805
.1660725
.1252626
.1627727
.1125498
.1109438
.1405055
.0987638
.1072435
.1022196
.1584129
8.268107
10.27744
8.753795
10.45338
9.854894
9.706452
8.154696
7.822169
7.852957
7.77029
7.107195
7.317663
6.915312
6.761536
6.676022
6.787563
6.839482
6.72515
-
8 . 67
-5.09
2.36
5.86
-1.32
0.60
4.73
1.58
-1.10
4.78
-5.69.
1.35
6.44
1.27
-0.09
0.31
0.32
0.56
1.31
-0.72
1.64
0.09
0.02
2.00
0.15
C. 35
-0.92
0.34
0.40
-0.46
-1.21
0.85
0.66
-0.95
-0.44
-1.02
-1.94
-0.17
-0.22
-1.24
0.51
-0.45
-0.46
0.14
-0.71
-0.98
-1.09
-0.82
-0.63
-0.41
-0.53
P>it 1
0.000
0 . 000
0.018
0.000
0.187
0.551
O.OCO
0.115
0.270
0.000
0.000
0.176
0.000
0.205
0.932
0.757
0.749
0.575
0.189
0.469
0.101
0.926
0.986
O.C46
0.879
0.728
0.358
0.735
0.687
0.644
0.227
0.393
0.508
0.342
0 . 656
0.309
0.052
0.362
0.827
0.216
0.607
0.650
0.648
0.391
0.475
0.325
0.276
0.414
0.527
0.684
0.560
?rob > F
R-square
-------
362
t
latl 96 |
latl_97 I
longl 78 |
longl 79 |
longl 30 |
longl 31 |
longl 32 |
longl 83 |
longl 84 |
longl 35 |
longl_86 I
longl 37 |
longl 88 I
longl 89 |
longl 90 1
longl_91 |
longl_92 1
longl 93 1
longl 94 1
Iongi_95 1
longl 96 1
longl 97 |
Iatldisw78 I
Iatldisw79 i
latldiswSO :
latldiswBl I
latldiswS2 1
Iatldisw83 I
Iatldisw84 |
latldiswSS I
Iatldisw86 |
Iatldisw87 I
latldiswSS I
Iatldisw89 I
iatldisw90 I
Iatldi3w91 |
Iatldisw92 I
Iatldisw93 I
Iatldisw94 |
Iatldisw95 I
Iatldisw96 |
Iatldisw97 I
Iongldisw78 I
Iongldisw79 !
longldiswSO I
Iongldisw83. 1
Iongldisw32 1
Iongldisw83 I
Iongldisw34 i
Iongldisw85 1
Iongldisw86 I
Ior.gldisw87 |
Iongldisw88 1
iongldiswSg 1
Iongldisw90 1
Iongldisw91 1
Iongldisw92 1
Iongldisw93 1
-4 .485706
-5.590626
18.51479
22.03932
19.8913
14.09608
23.70465
22.75367
17.49563
24.95424
22.29309
20.75179
21.70761
18.01021
18.38307
19.77447
21.05556
21.066
22.51121
22.22124
21.51437
21.25979
.4577654
-2.096774
-.195017
6.205866
-3.248244
-2.751189
.719245
-5.41568
-2.253397
-2.593667
-4.31729
-1.584891
-.4521447
-.5373405
-1.357564
-1.986987
-2.624828
-2.963697
-1.994049
-1.757131
.2818642
-1.248972
-.1127681
3.714421
-1.937044
-I .635528
.4316473
-3.228576
-1.34293
-1.545273
-2.569335
-.9422424
-.2651618
-.3200687
-.8081924
-1.183859
6.673725
7.382297
6.471835
7.146128
6.927386
7.723434
7.210277
7.294275
6.416465
6.302712
6.140378
6.175532
5.989776
6.170309
5.88568
5. 966141
5. 974474
5.888981
5.821359
5.795697
5.803518
5.824644
2.872228
4.421206
3.515421
4.423378
3.958501
4.034986
3.044232
2.780299
2.730719
2.579635
2.261097
2.36769
2.008208
1.921805
2.038669
1.9532
1. 931393
1.970275
1.780974
2.209255
1.714691
2.641708
2.100523
2.642447
2.365293
2. 411234
1.817419
1.660013
1.630995
1.539562
1.349887
1.413379
1.198751
1.147467
1.217396
1.127336
-0.67
-0.76
2.86
3.08
2.37
1.33
3.29
3.12
2.73
3.36
3.63
3.36
3.62
2.92
3.12
3.31
3.52
3.58
3.87
3.33
3.71
3.65
0. 16
-0.47
-0.06
1 .40
-0.82
-0.68
0.24
-1.95
-0.83
-1.01
-1.91
-0. 67
-C.23
-0.28
-0.67
-1.05
-1.36
-1.S1
-1.12
-0.80
0.16
-0.47
-0.05
1.41
-0.82
-0.68
0.24
-1.94
-0.82
-l.OC
-1.90
-0.67
-0.22
-0.28
-0.66
-1.05
0.502
0.449
0.004
0.002
0 .004
0.068
0.001
0.002
0.006
o.oco
0.000
0.001
0.000
0.004
0.002
0.001
c.ooo
0.000
0.000
0.000
0.000
0.000
0.873
0.635
0.956
0.161
0.412
0.495
0.813
0.051
0.409
0.315
0.056
0.503
0.822
0.780
0.505
0.293
0.174
0.132
0.263
0.426
0.869
0.636
0 . 957
0.160
0.413
0.498
0.812
0.052
0.410
0.316
0.057
0.505
0.825
0.780
0.507
0.294
-17.56725
-20.06109
5.828974
8.031788
6.312534
-1.043064
9.571372
8.455744
4.918345
12.59994
10.25699
8.646775
9.966709
5.91543
6.84621
8.079898
9.344649
9.522668
11.10044
10.86076
10.13856
9.842573
-5.172252
-10.76303
-7.085794
-2.46465
-11.00753
-10.6604
-5.247928
-10.8655
-7.606036
-7.650156
-8.749395
-6.225935
-4.388549
-4.304381
-5.353676
-5.688156
-6.410661
-6.830747
-5.485038
-6.087617
-3.079199
-6.427135
-4.230123
-1.46519
-6.57339
-6.361925
-3.13078
-6.482463
-4.539938
-4.563056
-5.215326
-3.712687
-2.614902
-2.569285
-3.19448
-3.393615
8.595844
8.879836
31.2006
36.04686
33.47007
29.23522
37.83792
37.0516
30.07291
37.30855
34.3292
32.8568
33.44851
30.10498
29.91993
31.46905
32.76646
32.60932
33. 92199
33.58171
32.89018
32.67701
6.087783
6.569485
6.69576
14.87638
4.51104
5.158018
6.686418
.0341435
3.099242
2.462822
.1148148
3.056153
3.484259
3.2297
2.638547
1.714181
1.161006
.8933526
1.49694
2.573356
3.642928
3.92919
4.004587
8.894033
2.699302
3.09087
3.994075
.0253107
1.854077
1.472509
.076656
1.828202
2.0S4579
1.929147
1.578095
1.025898
-------
363
Iongldisw94
Iongldisw95
Iongldisw96
Iongldisw97
Id summits
Id school
Id retail
Id hospital
Id church
Id cemetery
Id railroad
Id prinarte
Id othpriro
Id ma roads
Id i95
Id_i93
Id fp tewrna
Id fp mi lit
Id fp logan
Id fp bevmu
Id parks
Id irijwater
Id cclubs
Id tewmac
Id military
Id logan
Id bevmuni
pfemales
pblack
pother
page underS
page 5 29
page 65 up
pmarhh chd
pmhh child
pfhh child
pvacant
prenter occ
year79
yearSO
yearSl
year82
year83
year84
year85
year86
year87
yearBB
year89
year 90
year91
year 92
year93
year94
year95
year 96
year97
cons
-1.565912
-1.774002
-1.187466
-1.044624
-.0076867
-.0256489
-.8358928
.0961899
.2C21624
-.0020529
.0202007
-.0282984
.0187431
.0043521
.0193873
.0359392
-.0030909
.0075982
-.0150445
.0025637
-.0220602
-.0473648
-.0029476
-.1664644
.4892863
3.402231
3.715331
3.971701
1.805294
-.1301623
-2.669591
-3.40653
-.7613174
1.914783
4.569843
1.706427
3.377611
.3366676
109.8418
142.4608
215.6738
107 . 9564
58.34367
23.28113
-46.18921
87.07214
-23.06799
-146.8431
-146.5822
-55.24889
69.32137
79.37276
30.95689
69.57374
97.05387
71.03207
99.99672
1593.208
1.153302
1 .17629
1.063341
1.319C3S
.0081585
.0069644
.1881123
.0223765
.1734784
.0100655
.0063095
.009108
.0046133
.0029496
.0135562
.0105477
.0066062
.0059751
.0105632
.0078115
.0082728
.0113291
.0060042
.1406579
.2390684
.5302234
.3655053
1.0217
1.979133
.6632735
1.323864
.8013081
.5453215
.4488914
3.473691
1.129584
1.102132
.2238031
139.0156
132.3538
165.234
123.357
110.3102
93.96239
101.0093
109.683
96.67206
92.31466
91.06275
88.11312
82.55524
75.56147
79.37931
82.25522
76.72732
74 .44973
105.9543
390.3143
-1.36
-1 .51
-1.12
-0.79
-0.94
-3.68
-4.44
4.30
1.17
-C.20
3 .20
-3.11
i. .06
1.48
1.47
3.41
-0.47
1.27
-1.42
0.33
-2 . 67
-3.18
-0.49
-1.18
2.05
5.86
4.29
3.99
0.91
-0.20
-2.02
-4.25
-1.40
4.27
1.32
1.51
3.06
1.50
0.79
1.03
1.31
0.88
0.53
0.25
-0.46
0.79
-0.24
-1.58
-1.61
-0.63
0.84
1.05
0.39
0.85
1.26
0.95
0.9-4
4.08
0.175
0.132
0.264
C.429
0.346
0.000
0.000
0.000
0.244
0.838
0 .001
0.002
0 .000
0 .140
0.142
0.001
0.640
0.204
0.154
0.743
0.008
0.000
0.623
0.237
0.041
0.000
0.000
0.000
0.362
0.844
0.044
0.000
0.163
0.000
0.188
0.131
0.002
0.133
0.429
0.282
0.192
0.382
0.597
0.804
0.647
.0.427
0.811
0.114
0.107
0.531
0.401
0 ,294
0.697
0.398
0.206
0.340
0.345
0.000
-3.326564
-4.073715
-3.271781
-3.630^ 44
-.0236" 86
-.0393002
-1.204(523
.0523;!84
-.1378824
-.0217-329
.0078>32
-.0461515
.C097003
-.0014275
-.006685
.0152641
-.0160401
-.0041139
-.0357501
-.012748
-.0382761
-.069E717
-.0147168
-.442176
.020(747
2. 26'. 902
2.019305
1.96')OOS
-2.074118
-1.430233
-5.26-1572
-4.97722
-1.83 J233
1.03 1886
-2.239133
-.5077341
1.217259
-.1020218
-162.6507
-116.9735
-108.2108
-133.8428
-157.8817
-160.8999
-244.1833
-127.9237
-212.5604
-328.7745
-325. 0795
-227. 9644
-92.49988
-68.7396
-12.1 .639
-91. '55942
-53.34375
-74. '30111
-107.6903
828.1306
.6947402
.5317113
.8963488
1.540995
.0033052
-.0119976
-.4671631
.1400513
.5422073
.0176771
.0325683
-.0104452
.0277959
.0101317
.0464596
.0566143
.0098584
.0193103
.005661
.0178755
-.0058442
-.0251579
.0088216
.1092471
.9578978
4.53956
5.412358
5.974395
5.684706
1.169958
-.0746094
-1.83584
.3375985
2.794681
11.37892
3.920588
5.537963
.7753569
382.3344
401.8951
539.5583
349.7556
274.569
207.4622
151.8049
302.068
166.4244
35.08818
31.9151
117.4667
231.1426
227.4851
186.5528
230.8069
247.4515
216.9652
307.6838
2358.285
t
f
-------
364
t
Hypothesis
P-value
ofF-test
Reject @
5% level?
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All slopes to north (LAT) simultaneously zero
AH slopes to north (LAT) equal
All slopes to east (LONG) simultaneously zero
All slopes to east (LONG) equal
All LAT*LDIST simultaneously zero
All LAT*LDIST equal
All LONG* LDIST simultaneously zero
All LONG* LDIST equal
All other distance effects simultaneously zero
All Census tract characteristic effects simultaneously zero
0.7466
0.7004
0.7272
0.6967
0.0071
0.3276
0.6290
0.6915
0.6289
0.6910
0.0000
0.0000
NO
NO
NO
NO
NO
NO
NO
NO
NO
-------
365
Appendix D - Eagle Mine Site
Contents:
1 CRITERIA FOR EXCLUSION FROM RAW SAMPLE 366
2 ANNUAL COUNTS IN SAMPLE 366
3 DESCRIPTIVE STATISTICS 367
3.1 Housing prices and distances from site 367
3.2 Structural variables 368
3.2.1 Changing distance profiles of house age over time 369
3.2.2 Changing distance profiles of framing material overtime 369
3.3 Census tract attributes 370
3.4 Other distances 370
4 COLL1NEARITIES 372
4.1 Time patterns in average site distances in< sample 372
4.2 Time trend in average lot sizes 372
4.3 Distance to site vs. structural variables 373
4.4 Distance to site vs. Census tract attributes 373
4.5 Distance to site vs. other distances 373
5 COMPLETE REGRESSION RESULTS - No LOT SIZE INTERACTIONS 374
5.1 Just structural characteristics and year dummies 374
5.2 Including other distances 376
6 COMPLETE REGRESSION RESULTS - WITH LOT SIZE INTERACTIONS 377
6.1 Just structural characteristics and year dummies 377
6.2 Including other distances 380
t
365
-------
I
366
Chapter 1 Criteria for exclusion from raw sample
Condominium units are retained in the Eagle Mine sample because of the shortage of dwellings
within a radius of the Superfund site that could plausibly be directly affected by proximity to the
site. Both single-family detached dwellings and condos are included in our sample of owner-
occupied units. Given the high proportion of rental properties in the population of dwellings,
this sample is systematically different from that for our other three Superfund examples.
Observations are excluded from the Eagle Mine sample if:
• lot size is zero
• lot size is greater than 30000 square feet (e.g. 100 by 300 ft)
• less than 6 kilometers from the nearest portion of the Eagle Mine site (affects only nine
dwellings (none condos); recorded selling prices for these nine houses vary widely, from
23,900 to 238,000).
• further than 13.5 kilometers from the site (excludes 86 dwellings, but property sales at
these distances appear only in the time interval between 1984 and 1999)
• dwelling is older man 50 years (affects two outliers)
• a couple of obvious outliers that stand apart from the rest of the data: in 1978 with selling
price lower than $8100; in 1999 with selling prices lower than $6000.
S
Chapter 2 Annual counts in sample
YEAR I
Total
Freq.
Percent
1087
100.00
Cum.
76
77
78
79
SO
81
32
83
34
85
36
87
88
89
90
91
92
93
94
95
96
97
99
99
^ •
31
12
17
27
39
28
13
15
13
16
25
26
50
50
55
62
62
68
86
51
c; G
76
7 9
127
2
1
1
2
3
2
1
1
1
1
2
2
4
4
5
5
5
6
7
4
5
6
7
11
.85
.10
.56
.48
.59
.58
.20
.38
.20
.47
.30
.39
.60
.60
.06
.70
.70
.26
.91
.69
.43
. 99
.27
.68
2.
3
5.
8.
11.
14.
15.
16.
17.
19.
21.
24.
28.
33.
38.
44.
49.
56.
63.
68.
74 .
SI.
33.
100.
85
96
52
00
59
17
36
74
94
41
71
10
70
30
36
07
77
03
94
63
06
05
32
00
-------
367
Chapter 3 Descriptive statistics
3.1 Housing prices and distances from site
Variable I Obs Mean Std. Dev. Min
t
dist
sprice
Max
1087 9.423684
108"' 215365.3
1.929098 6.093462
12403-1.4 6400
13.05585
503000
.135235-
0-
aist
13,0559
Marginal distribution of distances: Eagle Mine
s
.048758-
C
.2
s
U.
1 h-
rtf'
ifl
_
1
1
,
liJIk
6400 500000
SPRICE
Marginal distribution of house prices: Eagle Mine
-------
t
368
500000 -
400000 -
g 300000-
200000-
100000-
" * mi' "' 1 *
' •./. '.. >* 't .'• ;' . !.'
: .-. ' *; ' •': . '
=v ^•V;:-i|ri • :• *$$/.•
v ^L^'^Vi- '.'v'1*-' ^ ;
^ ;?-*;:$ f ' ?' '
6 7 8 9 10 11 12 13
dist
Distance from nearest Eagle Mine site (km)
t
500000 -
400000 -
<*>
"c
1 300000
u
OJ
•H
§
o
x
200000 -
100000-
I 5 5
dist
11
13
14
Distance from nearest Eagle Mine site (km)
3.2 Structural variables
Variable I Obs Mean
Std. Dev.
Min
Max
sfd I
age I
aqe'2 I
bedrras i
bthrms I
notwdfrarae I
heatelec I
constgood I
constfair I
lotsLze I
1087
1087
1087
1087
1087
1087
1087
1087
1087
1087
.3965041
13.43054
258.5529
2. 614535
2.49494
.1269549
.4912603
.2842686
.2529899
1
.4893965
3.845643
253.1993
. SS83439
1.044023
.3330757
.5001537
.451273
.4349253
.9819864
0
0
0
0
1
0
0
0
0
.0310415
1
36
1296
6
6
1
1
1
1
5.789237
The only structural characteristics that have exhibited different trends downstream from the
Eagle mine site and nearer, as opposed to farther, from the site are the ages of dwellings and the
proportion which are not wood frame structures.
-------
369
3.2.1 Changing distance profiles of house age over time
In our sample, the age of houses at their last time of sale ranges only from zero to 36 years.
t
Regression with robust standard errors
Numoer o": obs = 1C87
F( 5, ".081) = 237.35
Prco > F = 0.0000
R-squared = 0.4137
Root MSE = 6.769
1
age 1
downstream 1
Idist i
trend 1
downstreamy i
Idisty !
cons I
-5
-4
-.
Coef.
.117111
.566435
4063416
-.019133
.5600853
20.99599
Robust
Std. Err.
.
1
2
3821816
.019259
3277674
0585814
1471051
.268103
-13.
-4.
-1 .
-0,
3.
9.
t
.39
,48
,24
,33
,81
,26
P>
0.
0.
0.
0.
0.
0.
Itl
000
000
215
744
000
000
i
-5
-6
-1
95% Ccnf.
.867013
.566385
.049474
I34C792
2714415
16.E456
Interval]
-4.367209
-2.566484
.2367908
.0958132
.8487291
25.44637
Controlling for distance from the Eagle Mine site, age at time of sale for houses downstream of
the Eagle Mine site (as opposed to those located ori Gore Creek) has not changed over time.
However, at the beginning of the sample period, houses being sold closer to the mine site are
older than houses being sold at a greater distance. By the end of the sample period, many more
newer houses are being sold closer to the site. Over time, the housing stock closer to the site
seems to be getting newer (on average) more quickly than the housing stock in the rest of the
area
3.2.2 Changing distance profiles of framing material over time
S
Regression with robust standard errors
1
no twd frame 1
downstream I
Idist I
trend I
downstreamy 1
Idigty 1
cons I
Coef.
-.1360607
-.1804713
-.0524
.0138754
.0199838
.5716354
Robust
Std. Err.
.0172675
.0463737
.0127841
.0023731
.005445
.1107124
-7
-3
-4
4
3
5
t
.88
.89
.10
.83
.67
.16
P>
0.
0.
0.
0.
0.
0.
Itl
000
000
000
000
000
000
Number of obs
F( 5, 1081)
Prob > F
R-squared
Rcot MSE
195^ Conf.
-.1699422
-.27:4639
-.07":4844
.OOH2379
.O0'»2997
.35'13993
1067
16.57
= 0.0000
= 0.0391
= .32725
Interval]
-.1021791
-.0894797
-.0273155
.0195129
.0306678
.788871
At the beginning of the sample period, the proportion of non-wood-frame houses was lower
downstream of the mine and declined with distance from the Eagle Mine site. Over time, there
was a relative increase in the proportion of non-wood-frame houses downstream of the mine and
-------
t
t
370
nearer the site. However, the low R-squared statistic on this model suggests that these results
are not very robust. Non- wood-frame houses average only about 13% of the stock.
3.3 Census tract attributes
Census tract attributes are not employed in this analysis, since there are too few tracts in the
affected area
3.4 Other distances
Data were collected for the full set of other amenities/disamenities that have elsewhere been
found to play some role in explaining housing prices. In the estimated models that we report,
however, collinearity problems are so severe that we are forced to shorten the list of variables
used. We eliminate variables first if the feature is located at such a great distance that it does not
seem plausible that it should have any detectible effect on housing prices in this sample. This
leads us to drop rivers, hospitals, and churches. Other variables are highly collinear due to the
topology of the area. We are left with only the distance to the Vail ski area, the distance to
Interstate 70, the distance to the nearest river, and the distance to the nearest recreational area
(golf course or country club). Apparent distance effects relative to the Eagle Mine Superfund
site will be unavoidably confounded with the effects of distance to the nearest cemetery, so we
exclude the cemetery variable.
Distance variable
d_summits
d_rivers
d_school
d_retail
d_hospital
d_church
Description
Distance from the nearest summit of land. Of course, there are
many of these in Eagle County. Approximately 20 different named
summits are within the Vail zip code, within which the Eagle Mine
site is more or less centered.
Distance to the nearest river. Most houses in the sample lie close to
the main branch of either the Eagle River, which flows north past
the site and then west after its confluence with Gore Creek, which
runs from east to west through Vail.
Distance from the nearest school. The Minturn Middle School is
about 2.5 miles northwest along Eagle River from the mine site.
Battle Mountain high school lies about 7.5 miles northwest of the
site, near the confluence of Eagle River and Gore Creek. Lake
Creek School lies further downstream on Eagle River, near the
boundary of the Vail zip code area.
Distance to the nearest retail center. Unlikely to be relevant for this
sample of houses. The nearest shopping mall appears to be West
Glenwood Mall, about 52 miles to the west of the site.
Distance to the nearest hospital. Unlikely to be relevant for this
sample. Nearest hospital, Mercy Hospital, lies about 28 miles ENE
of the site.
Distance from the nearest church. There appear to be no major
churches anywhere with the Vail zip code area. The Saint Benedict
Monastery lies about 38 miles southwest of the mine site.
-------
d_cemetery
d railroad
d i70
d cords
d_mj water
d_airport
d recareas
d locale
Distance to the nearest cemetery, The River View cemetery lies
about 3.5 miles downstream (NNW) of the mine site. The Gold
Park Cemetery is about 9 mile's SSW of the site, and is unlikely to
be relevant to explaining housing prices in our sample.
Railroads in the sample area follow the route of the Eagle River.
The line appears to belong to the Chicago and Northwestern
Railway Company (CNW), although ownership of a number of
segments of lines just outside our sample area is not recorded in the
CIS dataset.
Distance from Interstate 70, [an east-west] freeway that runs
through the northern third of the Vail zip code area. The coefficient
on this variable is a proximity1 effect in addition to proximity from
the nearest main roads, d cords.
Distance from the closest main roads. This includes 170 if it
happens to be the nearest main road. The only other major roads are
US Highway 6, which runs alongside 170 to the west of the
confluence of the Eagle River and Gore Creek were it flow in from
the direction of the Vail settlement, and US Highway 24, which
runs alongside the Eagle Riv$rto its junction with 170 and US6 near
the confluence of the Eagle River and Gore Creek.
iRi\
iody
Distance from the nearest body of water. Four significant reservoirs
are located within a radius of! 18-25 miles of the mine site, but none
of these is likely to have any bearing on housing prices in our
sample.
Distance from the nearest airport. Three airports lie between 29 and
24 miles of the mine site, but there is unlikely to be much of a
discernible effect of proximity to these airports on housing prices in
our sample.
Distance to the nearest Golf Club. There are three golf clubs in the
Vail zip code area. All three lie in close proximity to 170. One is
centered in the Vail census tract, one is slightly west of the
confluence of the two rivers and the junction of US24 with the 170
freeway. The third is near trie western boundary of the Vail zip
code area, close to the Lake Creek School.
Distance to different miscellaneous points of interest, such as
campgrounds, ranger stations, etc. Nearest entities in this class are
likely to be highly heterogeneous, so distances will not be expected
to have common systematic effect on housing prices.
371
f
t
Variable
Obs
Mean
Std. Dev.
Min
Max
-------
t
372
Chapter 4 Co I linearities
4.1 Time patterns in average site distances in sample
t
Regression with robust standard errors
Idist
year77
year? 8
year79
yearSO
year 81
year 82
year83
year 84
yearSS
year86
year37
year83
yearS9
year90
year91
year92
year 93
ye a r S 4
ve a r 9 5
ye a r 9 6
year 97
year 93
ye a r 9 9
cons
Coef.
.0087313
-.015558
.0266883
.1051202
.2294165
.0497185
.0579519
,1120218
.0665697
.0461149
.0490009
,0749078
,0752998
.1068947
.0040524
.067759
.0525805
.0647503
.1068004
-.000675
.0109184
.0125766
.0590857
2.167601
Robust
Std. Err.
.0415166
.0588126
.029099
.0352026
.0348561
.0516654
.0535447
.0436912
.0479459
.0423219
.0495231
.0334853
.0337427
.0315633
.031803
.0340214
.0306113
.0294466
.0321517
.0356166
.0313955
.0324308
.0263302
.0184332 1
t P> 1 t I
0
-0
0
2
6
0
1
2
1
1
0
2
2
3
0
1
1
2
3
-0
0
0
2
17
.21
.26
.92
.99
.58
. 96
.06
.56
.39
.09
.99
.24
.23
.39
.13
.99
.72
.20
. 32
.02
.34
.39
.24
.59
0.
0.
0.
0.
0.
0.
C.
0.
0.
0.
0.
0.
0.
0.
0 .
0.
0.
0.
0.
0.
0 .
0.
0.
0.
833
791
359
003
000
336
279
010
165
276
323
025
026
001
899
047
036
028
001
965
732
698
025
000
Number of obs
F( 23, 1063)
Prob > F
R-squared
Root MSE
[95% Conf.
-.0727326
-.1309601
-.0304097
.0360458
.161022
-.0516593
-.0471133
.0262911
-.0275096
-.0369289
-.0431734
.009203
.00909
.0449613
-.0583513
.0010023
-.007485
.0069703
.0437124
-.0705618
-.0516669
-.051059
.0073226
2.131431
1087
3.28
= 0.0000
= 0.0473
= .20432
Interval]
.0901952
.0998441
.0837864
.1741946
.297811
.1510963
.1630172
.1977526
.160649
.1291588
.1461751
.1406127
.1415096
.1688281
.0664562
.1345156
.112646
.1225303
.1698334
.0692118
.0735036
.0762122
.1108489
2.20377
4.2 Time trend in average lot sizes
Regression with robust standard errors
1
lotsize I Coef.
year7T | .317916
year78 1 .5209934
year79 1 .1706109
year30 t .0681403
vearSl ! .1000136
Robust
Std. Err.
.26954
.3025645
.2103721
.1365142
.1497379
1
1
0
0
0
t
.13
.72
.81
.50
.6','
p>
0.
0 .
0 .
0.
0.
itl
233
085
418
618
504
Number of obs
F( 23, 1063)
Prob > F
R-squared
Root MSE
[95* Conf.
-.2109749
-.072698
-.2421808
-.1997275
-.1938018
1087
3.38
= 0.0000
= 0.0711
.95661
Interval]
.846807
1.114685
.5834025
.3360081
.393829
-------
373
year92
year83
year84
ye a r 8 5
year86
year87
year88
year89
year 90
year91
year92
year93
year94
year95
year96
year97
year98
year99
cone
.7517196
.579502
. 3066324
.3028399
1 .035695
1 .139514
.6645393
.3791137
.568S317
.4324467
.8707075
.6066884
.719057
.300751
.3009981
. 3053314
.5701365
.2245463
.5374871
4.3 Distance to site vs.
.2941222
.3099456
.2881732
.1810967
.2723908
.2989708
.2037596
.1513063
.1654391
.1755239
.1940101
.1750321
.1609307
.1672993
.1558229
.1440659
.1665174
.1223279
.1111585
structural
2.56
1.87 •
1.06
1.67
3 .80
3.81
3.19
2.51
3.44
2.46
4 .49
j .47
4 .47
1.80
1 ,93
2.12
3.42
:. .84
4.84
variables
0.011
0.062
C.238
0.095
0.000
0.000
0.001
C.012
0.001
0.014
0.000
0.001
0.000
0.073
0.054
0.034
0.001
3 .067
0.000
Regression with robust standard errors
Idist
sfd
age
age2
taedrms
bthrms
notwdframe
heateiec
constgood
const fair
lotsize
cons
Cosf .
.062327
-.0038515
.CQC0434
-.0413243
.0078928
-.0371402
.0600777
.0326282
.0336852
.0135193
2.210167
Robust
Std. Err.
.0183986
.0022568
.0000825
.0100509
.0099483
.0141457
.0135586
.0148927
.0180831
.0104568
.0307442
t
3.39
-1.71
0.53
-4.11
C.79
-2.63
4.43
2.19
1.86
1.29
''3.84
P>|t i
0.001
0.088
0.599
0.000
0.428
0.009
0.000
0.029
0.063
0.196
0.000
,1745<-35
-.0286727
-.258S;.05
-.0525(176
.5012:03
.5528745
.2553-.21
.0822207
.2439074
.0880341
.4900 J12
.2632 JOT
.403279
-.0275235
-.0047573
.0223456
.2433964
- .0154853
.3193721
Number of obs
F( 10, 1076)
Prob > F
R-:;quare-d
Root MSE;
(95* Conf.
.026. -25S
-.O08.>798
-.0001186
-.0610463
-.0116275
-.0649965
.0334733
.0034061
-.0017969
-.0069568
2.209841
1.328846
1.187677
.8720953
. 6581874
1.57018
1.726154
1.074568
.6760067
.8931559
.7768593
1.251394
. 9501362
1.034835
.6290254
.6067535
.5877173
.8968766
.4645778
.7556022
1087
7.97
= 0.0000
= 0.0627
= .20143
Interval]
.0934282
.0005768
.C002053
-.0216032
.027413
-.009334
.086682
.0618503
.0691672
.0340373
2.330492
f
t
4.4 Distance to site vs. Census tract attributes
There are no census tract characteristics for this data set (insufficient numbers of tracts).
4.5 Distance to site vs. other distances
Regression with robust standard errors
Number of obs = 1087
F( 4, 1062) = 1990.96
Prob > F = 0.0000
R-squa::ed = 0.7795
Root M:;E = .09743
-------
t
Id
Id
Id
1
Idist |
vail ski I
recareas [
railroad 1
Id rivers 1
cons !
Coef .
.4150413
-.0741603
.1467133
.0014111
1.433252
Robust
Std. Err.
.008733
.0052133
.0033063
.0028557
.0173293
47
-14
44
0
82
t;
.25
.23
.37
.49
.71
P> 1 1 I
0.000
0.000
0.000
0.621
0.000
[95% Conf.
.3978076
-.0843896
.1402259
-.0041923
1.399251
Interval]
.432275
-.0639309
.1532007
.0070145
1.467253
374
t
Chapter 5 Complete regression results - No lot size interactions
5.1 Just structural characteristics and year dummies
Note that the time-differentiated "downstream" dummy variable (downstrX) and and time-
differentiated log(dist) variables (IdisX) and the interactions between the downstream dummies
and the log(dist) variables (downldistX) are summed across subsets of years. There was
insufficient data on sales in many individual years to permit a full complement of 23 distinct
yearlyy coefficients. The labeling of the combined years corresponds to the last year in the
interval..
Regression with robust standard errors
Isprice
sfd
age
age2
bedrtr.s
bthrms
notwdf rame
heatelec
constgood
constfair
lotsize
downstr79
downstr82
downstr85
downstr83
downstr91
downstr 94
downstr97
downstr99
Idis79
Idis82
Idis85
Coef .
.087723
-.002036
.0000844
.0911766
.2452747
.2857618
-.1120886
.2404679
-.2633383
-.0180865
-14.6678
-2.451152
2.915498
-4.156908
1.102785
1.29416
-1.838408
-1.447165
-.1810061
-.3655425
-.0074668
Robust
Std. Err.
.0339329
.0057189
.0002022
.0198941
.0204907
.0503166
.0299623
.0377774
.0327453
.0218644
2. 620287
1.868129
1.881971
1.B69033
1.418707
.8353652
1.394555
. 945762
.3159642
.271496
.3969009
t P> 1 1 I
2
-0
0
4
11
c;
-3
6
-8
-0
-5
-1
1
"~ £.
0
1
-1
-1
-0
-1
-0
.£9
.36
.42
.58
.97
.68
.74
.37
.04
.93
.60
.31
.55
.22
.73
c; ^
.32
.53
.57
.35
.02
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.010
.715
.676
.000
.000
.000
.000
.000
.000
.408
.000
.190
.122
.026
.437
.122
.188
.126
.567
.178
.985
Number of obs = 1087
F( 56, 1029) =
Prob > F
R-squared = 0.6846
Root MSE = .38696
(95% Ccnf
.0211374
-.013308
-.0003123
.0521388
.2050665
.1870269
-.1708828
.1663384
-.3275934
-.0609904
-19.80951
-6.116929
-.7774421
-7.824459
-1.681104
-.3450539
-4.574904
-3.303008
-.8010139
-.8982916
-.7862942
Interval )
.1543086
.0091361
.0004812
.1302143
.285483
.3844967
-.0532944
.3145973
-.1990831
.0248174
-9.526082
1.214625
6.608437
-.4893577
3.886674
2.933374
.898088
.4096768
.4390017
.1672065
.7713607
-------
375
Idis88 ; -.1457056 .2072288 -0.70 0.482
Idis91
Idis94
Idis97
Idis99
downldist79
downldist82
downldistSS
downldistSS
downldist91
downldist94
downldist97
downldist99
year77
year78
year79
yearSO
yearSl
year82
year83
year84
yearSB
year86
year87
year88
year89
year 90
year91
year92
year93
year94
year 95
year96
year97
year98
year99
cons
.1181851 .1689896 0.70 . 0.484
.4585914 .1722099 2.66 0.008
.0562526 .0913914 0.62 0.533
.1029265 .0861261 1.20 0.232
6.480981 1.164892 5.56 0.000
.9652991 .7914221 1.22 C.223
-1.419509 .8196063 -1.73 • 0.084
1.715963 .9129617 2.11 0.035
-.5503184 .6185063 -0.89 0.374
-.7252007 .3532013 -2.05 0.040
.6997834 .6091435 1.15 0.251
.5511937 .416007S 1.32 0.185
.OC99S76 .1170336 C.09 0.932
-.4767802 .178538 -2.67 0.008
.5140408 .1410527 3.64 0.000
1.085522 .9378863 1.16 0.247
1.106994 .946807 1.17 0.243
1.082301 .9894484 ].09 0.274
.3706013 1.163475 0.32 0.750
.5333833 1.135127 0.47 0.639
.405868 1.105695 0.37 0.714
.5661312 .831686 0.68 0.496
.4394625 .8371651 0.52 0.600
.323816 .82593 0.39 0.695
.0047135 .7938476 0.01 0.995
-.0944347 .7852646 -0.12 0.904
-.0058193 .7881901 -0.01 0.994
-.6572875 .7956385 -0.83 0.409
-.6032985 .7900044 -3.76 0.445
-.4135891 .7940377 -0.52 0.603
.6792759 .7207184 0.94 0.346
.6282686 .718934 0.87 0.382
.6077811 .7197148 0.84 0.399
.5657678 .719305 0.79 0.432
.7156931 .7223124 0.99 0.322
10.83916 .7015253 15.45 0.000
-.5523449 .2609338
-.2134195 .4497887
.12066-37 .7965142
-.1230E21 .2355373
-.0660765 .2719294
d. 195145 8.766816
-.5876<<64 2.518285
-3.0;:78 .1887814
.1207:.53 3.31122
-1.763996 .6633592
-i.4ie;:79 -.0321227
-.4955:!18 1.895089
-.2651268 1.357514
-.219664 .2396393
-.8271203 -.1264402
.237257 .7908246
-.7548665 2.92591
-.7508984 2.964887
-.B592656 3.023368
-1.912454 2.653657
-1.694344 2.760511
-1.7638C6 2.575542
-1.065863 2.198125
-1.203283 2.082203
-1.296883 1.944515
-1.553031 1.562458
-1.635338 1.446468
-1.552463 1.540824
-2.218547 .9039717
-2.15J492 .9469151
-1. 971707 1.144529
-.734?-697 2.093521
-.7824756 2.039013
-.8044952 2.020057
~.845"043 1.97724
-.7016803 2.133067
9.462572 12.21574
Hypothesis
All structural attribute slopes simultaneously zero
All year-specific slopes on DOWNSTR simultaneously zero
All year-specific slope on DOWNSTR the same
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All year-specific slopes on DOWNSTR*LDIST sim. zero
All year-specific slope on DOWNSTR* LDIST the same
P-value Reject?
ofF-test
0.0000
0.0000
0.0000
0.1463 NO
0.1995 NQ
0.0000
0.0000
t
t
-------
t
376
5.2 Including other distances
t
Regression with
Isprice I
sfd 1
age 1
age2 1
taedrms I
bthrms 1
notwdframe 1
heatelec 1
constgood 1
constfair I
lotsize 1
downstr79 1
downstr82 1
downstr85 I
downstrSS 1
downstr91 1
downstr94 1
dcwnst r97 |
downstrS9 1
Idis79 I
Idis82 1
IdisSS I
Idis33 1
Idis91 !
Idis94 I
Idis97 I
Idis99 I
downldist79 I
downldist82 I
dowr,ldist85 I
downldist88 I
downldist91 1
downldist94 1
downldist97 |
=ownldist99 !
Id vail ski I
Id recareas 1
Id railroad I
Id river? 1
year77 |
year78 1
year79 1
robust standard errors
Coef .
.1599711
.0008054
-.0001994
.1099823
.2132256
.1667254
-.1032714
.1519202
-.181157
-.0097934
-11.77349
-1.828606
3.119721
-5.236217
-.1198468
-.0094171
-2.291998
-2.639503
-.3848692
-.4108056
.1480834
-.2160061
.0026152
.3132996
-.181086
-.0486977
5.701665
1.171806
-1.083623
2.598629
.4107277
.2592562
1.326131
1.489794
-.6410497
-.0649664
.2146156
-.0287666
.033184
-.313459
.5574546
Robust
Std. Err.
.0318435
.0056284
.00019
.0192905
.0205788
.0463674
.0288212
.0363367
.032681
.0245693
2.833724
1.903158
2.065909
1.966144
1.3S7133
1.036293
1.216899
1.067854
.4002505
.2797502
.4044314
.2329473
.2214624
.2093543
.1707862
.178804
1.246576
.798117B
.8962712
.8641985
.8277091
. 444089
.5286867
.4788069
.093648
.0316718
.046235
.0112226
.1034472
.1864471
.1449649
5
0
-1
5
10
3
-3
4
-5
-0
-4
-0
1
-2
i"i
U
-0
-1
_o
-0
-1
0
-0
0
1
-1
-0
4
1
-1
3
0
0
2
3
-6
-2
4
-2
0
-1
3
t
.02
.14
.05
.70
.36
.60
.58
.'-8
.54
.40
.15
.96
.51
.66
,C6
.01
.68
.47
.96
.47
.37
.93
.01
.50
.06
.27
.57
.47
.21
.01
.50
.58
.51
. 1 1
.85
.05
.64
.56
.8C
.65
.85
P> 1 1 1
0.000
0.886
0 .294
0.000
0.000
0.000
0.000
0.000
0 . 000
C.690
0.000
0.337
0.131
0.008
0.949
0. 993
0.060
0.014
0.336
0.142
0.714
0.354
0.991
0.135
0.289
0.785
0.000
0.142
0.227
0.003
0.620
0.559
0.012
0.002
0.000
0.040
0.000
C.011
0.422
0.093
0.000
Number of obs
F( 61, 1025)
Prob > F
R-squared
Root MSE
[95% Conf.
.0974852
-.0102392
-.0005723
.0721288
.1728441
.0757395
-.1598267
.0806173
-.2452864
-.0580053
-17.33405
-5.563136
-.9341736
-9.094345
-3.822931
-2.042916
-4.679886
-4.734933
-1.170273
-.9597541
-.6455246
-.6729179
-.4319563
-.0975124
-.5162165
-.3995614
3.255534
-.394325
-2.842359
.9028285
-1.21347
-.6121712
.2886987
.5492399
-.8248134
-.1271153
.1238895
-.0507886
-.1198085
-.6793205
.2729927
1087
79.20
= 0.0000
= 0.7234
= .36307
Interval]
.2224571
.0118499
.0001735
.1478357
.2536071
.2577113
-.0467161
.2232231
-.1170276
.0384184
-6.212924
1.905924
7.173616
-1.378089
3.583238
2.024082
.09591
-.5440732
.4005347
.1381428
.9416914
.2409058
.4371867
.7241117
.1540444
.302166
8.147797
2.737938
.675113
4.294429
2.034926
1.130684
2.363563
2.428347
-.4572861
-.0028175
.3053418
-.0067447
.2861766
.0524026
.8419166
-------
377
yearSO i .9247319 .9558066 0.97 0.334
yearSl .9375531 .9601498 0.38 0.329
yearS2 .9351966 .9796211 0.95 0.340
year83 -.2396555 1.155135 -0.21 0.936
year94 -.0543507 1.133581 -0.05 0.962
year85 -.2392129 1.102717 -0.22 0.823
year36 .4137392 .9975717 0.46 C . 645
year37 .373063 .8990918 0.42 0.674
year88 .2666087 .8914042 0.30 0.765
year99 -.0249319 .8844667 -0.03 0.978
year90 -.0754031 .8765698 -0.09 0.931
year91 .0135561 .8810495 0.02 0.988
year92 -.5351233 .8854444 -C.60 0.546
year93 -.495669 .881808 -C.56 0.574
year94 -.2744524 .8945733 -C.31 0.756
year95 .9902915 .8329522 1.19 0.235
year96 .9723515 .8320823 1.17 0.243
year97 .990077= .8321521 1.19 0.234
year98 .775383 .8417312 0.92 0.357
year99 .8966721 .8443634 1.06 0.289
_cons 11.72799 .8490379 13.81 3.000
Hypothesis
All structural attribute slopes simultaneously zero
All year-specific slopes on DOWNSTR simultaneously zero
All year-specific slope on DOWNSTR the same
All year-specific slopes on LDIST simultaneously zero
All year-specific slope on LDIST the same
All year-specific slopes on DOWNSTR* LDIST sim. zero
All year-specific slope on DOWNSTR* LDIST the same
All other distance effects simultaneously zero
-.9508294 2.800293
-.9465207 2.821647
-.9870954 2.857489
-J:.506j55 2.027044
-2.278755 2.170054
-2.403CS3 1.924627
-i. 347499 2.175077
-1.386208 2.142334
-1, 482577 2.015794
-1.760504 1.71064
-1.795479 1.644673
-1.715311 1.742423
-2.272514 1.202367
-2.226324 1.234686
-2.010244 1.461339
-.6441949 2.624778
-.6604279 2.605131
-.6428384 2.622994
-.8763302 2.427096
-.7602061 2.55355
10.06194 13.39404
P-value Reject?
ofF-test
0.0000
0.0000
0.0001
0.1176 NQ
0.0796 jsJO
0.0000
0.0000
0.0000
I
t
Chapter 6 Complete regression results - With lot size interactions
6.1 Just structural characteristics and year dummies
Regression with robust standard errors Number of obs =
F( 75, 1007) =
Prob > F
R-squa :ed =
Raot M:3E
1087
0.7016
.38044
-------
378
1
Isprice |
sfd [
age I
age2 I
bedrms I
bthrms I
notwdframe I
heatelec 1
constgood I
constfair I
iotsise I
downstr"?9 I
downstr82 1
downs tr 85 I
downstrSS I
downstr91 1
downstr94 1
downstr97 I
down5tr99 I
Idis79 I
Idis82 I
Idis85 I
]dis88 I
Idis91 I
Idis94 i
Idis97 !
Idis99 I
Jownldist_79 I
downldist82 I
dowrildist35 I
downldistSS I
downldist91 1
downldist94 i
downldist97 1
d
0
-0
-0
-1
-1
-0
-1
0
0
-0
-0
_^_
-0
-1
4
'•>
-0
0
o
-6
-1
1
4
2
1
0
-0
2
2
2
1
1
2
2
2
6
1
t
.58
.92
.34
.74
.59
.51
.00
.83
.43
.73
.39
.63
.07
.40
.90
.17
.44
.98
.50
.09
.28
.91
.90
.72
.09
.41
.36
.57
.03
.68
.79
.20
.47
.25
.34
.73
.24
.04
.11
.60
.14
.23
.20
.84
.69
.91
.39
.92
.4"1
.24
P> 1 1 1
0
0
0
0
0
0
0
0
0
0
0
0
c
c
c
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.010
.357
.298
.000
.000
.000
.003
.000
.000
.007
.695
.000
.039
.686
.368
.668
.151
.047
.618
.278
.779
.365
.370
.469
.276
.680
.176
.000
.043
.498
.428
.841
.000
.211
.180
.000
.025
.301
.910
.550
.002
.026
.028
.066
.091
.004
.017
.004
.000
.217
[95% Conf,
.0227952
-.0163292
-.00018
.05707
.2061133
.1819271
-.149623
.1501537
-.3158016
-.6765503
-9.141448
-26. 9124
-24.76393
-1.700092
-6.328848
-5.309375
-1.138388
-1.091067
-1.035478
-.7650507
-.3436798
-.2038597
-.3211604
-.2659009
-.2157954
-5.90234
-8.867168
4.697605
.1774492
-1.271559
-1.147511
-1.95454
-15.89669
-15.53254
-35.57244
4.828538
1.459632
-.4800922
-2.966395
-4.348972
.0879591
.0199859
.018961
-.0030234
-.018576
.0597272
.0302266
.0646219
3.709053
-1.483945
, Interval)
.1668249
.0058904
.0005876
.1378014
.2901729
.3833815
-.0313846
.3025347
-.1837903
-.1102977
13.71619
-10.87841
-.6709796
2.582133
2.348552
4.479628
.1753176
-.0062411
.6158955
.2200542
.4582411
.5540692
.1198002
.1224833
.0615945
3.850462
1.623177
11.77184
10.71244
.6191578
2.706357
2.401043
-8.499616
3.432179
189.3365
11.68375
21.87629
1.552822
3.327685
2.317975
.3799897
.3156533
.331526
.2539795
.2518613
.3068443
.3086808
.3290635
6.940507
6.539795
-------
379
vdownldist 85
vdownldist 88
vdownldist 91
vdownldist 94
vdownldist 97
vdownldist 99
year77
year78
year79
yearSO
yearSl
year 32
year83
year34
yearSS
year86
year87
yearSS
yearSS
year90
year91
year92
year93
year94
year95
year96
year97
year98
year99
cons
-31.07023
-3.718982
-5.038796
-.2542936
-.0886993
.3232028
-.0279239
-.555388
.4340608
.922831
.9341329
.6939822
.2306941
.4085505
.2530232
.4008101
.3222346
.1157686
-.3423158
-.4179741
-.3361387
-.661575
-.5863717
-.4200524
.4363901
.3972187
.3655801
.3212503
.4601599
11.399
23.12852
.8020938
2.253747
.2341911
.7097366
.768631
.1078797
.1803076
.1432401
.9737513
. 979137
1.050655
1.177666
1.163137
1.132406
.8529452
.8573298
.8485191
.8283121
.8192932
.8230389
.8281353
.8227413
.8256306
.7452147
.7441496
.7450614
.7389294
.7408714
.7401734
-1
-4
-2
-1
-0
0
-0
-3
3
0
0
C
C
C
C
0
(I
0
-0
-0
-0
-0
-0
-0
0
0
0
0
0
15
.34
.64
.24
.09
.12
.42
.26
.08
.38
.95
.95
.85
.20
.35
.22
.47
.33
.14
.41
.51
.41
.80
.71
.51
.59
.53
.49
.43
.62
.40
0
0
0
C
0
0
0
0
0
C
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.179
.000
.026
.273
.901
.674
.796
.002
.001
.344
.340
.395
.845
.725
.823
.639
.707
.892
.679
.610
.683
.425
.476
.611
.558
.594
.624
. 664
.535
.000
-">
-••
-
_ '1
-1
_ .
-.
-.
-.
-1
-2
-1
-1
-1
-1
-1
-2
-1
-2
-2
-2
-1
-1
-1
-1
-.
9
6.45585
.292S-49
.46i:-76
.713J'S2
.481431
.185: 97
2396186
9097096
2 02 9",' 7 6
9879U32
9872-197
.167'M2
.080266
-1.8739
.969122
.272944
.360123
.549299
.967731
.025692
.9512C6
.286643
.200856
.040206
.025962
.063043
.096471
.12676S
9936688
.946539
1
-2
-.
1
1
-.
2
2
2
2
2
2
2
2
1
1
1
.
1
1
1
1
I
1
1
4.31538
.145015
6162172
2052643
.304033
.831603
1837709
2020664
.765144
.833645
.855515
.955706
.541654
.691001
.475168
.074564
.004592
.780837
1.2831
.189743
.278929
9634934
.028112
.200101
.898742
1.85743
.827631
.771268
.913989
2.85146
Hypothesis
P-value
ofF-test
Reject?
All structural attribute slopes simultaneously zero
All lotsize-independent year-specific slopes on DOWNSTR
simultaneously zero
All lotsize-ihdependent year-specific slope on DOWNSTR the
same
All lotsize-independent year-specific slopes on LDIST
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same
All lotsize-dependent year-specific slopes on DOWNSTR
0.0000
O.C002
0.0002
0.3611
0.3490
All lotsize-independent year-specific slopes on DOWNSTR*LDIST ° • °003
simultaneously zero
All lotsize-independent year-specific slope on DOWNSTR*LDIST 0.0002
the same
0.0000
-------
380
simultaneously zero
All lotsize-dependent year-specific slope on DOWNSTR the same o . oooo
All lotsize-dependent year-specific slopes on LD1ST ° • °003
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on o . oo 1 6
vX Idist variables)
All lotsize-dependent year-specific slopes on DOWNSTR*LDIST ° • 000°
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on DOWNSTR*LDIST ° • 000°
the same (on vX Idist variables)
6.2 Including other distances
Regression with robust standard errors
Isprice | Coef.
sfd .1561376
age
age2
bedrms
bthrms
notwdf rams
heatelec
const good
const fair
lotoize
down s t r7 9
downst r82
downstrSS
downst r83
downstr91
downst r 94
downst r 97
downst r 93
Idis79
Idis82
IdisSt
Idis88
IdisSl
Idis94
Idi39~!
Idis99
-.0016907
-.0001244
.116823
.2148243
.134415
-.0824512
.1145337
-.1638627
-.5241375
(dropped)
4.150832
(dropped)
-19.77913
-18.88275
-1.161994
-2.807719
-4.892306
-.6136583
-.4277661
.0235332
-.3167741
.0119798
.1284788
-.2289993
-.1487755
Robust
Std. Err.
.0342351
.0056049
.0001868
.0197209
.0210301
.0431922
.0286962
.0359167
.0328503
.5693396
4.76774
4.022621
6.378922
1. 940248
2.226972
3.04658
.4881272
.4383145
.5397356
.3957669
.3772201
.3748758
.3460179
.3472802
4
-0
-0
5
10
1
-2
3
-4
-0
0
-4
-2
-0
-1
-1
_i_
-0
u
-0
0
0
-0
-0
t
.56
.30
.67
.92
.22
.79
.87
.19
.99
.92
.87
.92
.75
.60
.26
.61
.26
.98
.04
.80
.03
.34
.66
.43
Number of obs = 1087
F( 83, 999) =
Prob > F
R-squared = 0.7475
Root MSE = .35141
P>|t 1
0.000
0.763
0.506
0.000
0.000
0.005
0.004
0.001
0.000
0.357
0.384
0.000
0.006
0.549
0.208
0.109
0.209
0.329
0.965
0.424
0 . 97 5
0.732
0.503
0.666
[95% Conf
.0889566
-.0126893
-.000491
.0781239
.17355S
.0398454
-.1387629
.0440529
-.2283261
-1.641424
-5.205102
-27.67289
-32.38154
-4.969424
-7.177799
-10.87074
-1.571531
-1.287889
-1.03566
-1.093404
-.7282548
-.5071555
-. 9080045
-.8302578
Interval]
.2233187
.0093079
.0002422
.1555222
.2560926
.2289845
-.0261394
.1850144
-.0993992
.5930491
13.50677
-11.88538
-5.383952
2.645436
1.562361
1.086125
.3442139
.4323565
1.082827
.4598557
.7522144
.8641132
.4500059
.5327069
-------
381
downldist79
ctownldist82
downldist85
downldist 88
downldist91
downldist 94
downldist 97
downldist 99
vdownstr79
vdownstr82
vdownstrBS
vdownstr88
vdownstrSl
vdownst r94
vdowristr97
vdownstr99
vldis79
vldis82
vldisSS
vldis88
vldis91
vldis94
vldis97
vldis99
vdownldist79
vdownldist82
vdownldist85
vdownldistSS
vdownldistSl
vdownldistl>4
vdownldist97
vdownldistBS
Id vail ski
Id recareas
Id railroad
Id rivers
vld vail ski
vld recareas
vld railroad
vld rivers
year77
year78
year79
yearBO
yearSl
year82
year83
yearS4
year85
year86
year87
year88
year89
year90
year9i
year92
.6424392
-1.229116
-.1.558664
9.119049
3.620668
.8690293
1.664035
2.645432
-12.69702
-7.860077
108.1734
S. 154826
15.51394
.7983145
.0597452
1.370786
.1342792
.0713173
.088456
.0196382
.0110167
.0669103
.061494
.0676226
5.310019
3.147268
-43.82669
-3.771125
-6,798681
-.4721381
-.1531307
-.7944184
-.8855005
-.1447152
.2314952
-.0528532
.2696188
.0580336
-.0256867
.0267628
.0879751
-.3021502
.5380431
.6160871
.6C49962
.5508142
-.40708
-.1249217
-.3689251
.3645976
.3830043
.2085509
-.300348
-.3361528
-.2379597
-.4955765
.1175245
2.036985
2.4115
1.776809
2. 991758
.8645906
. 9699183
1.358699
2.128615
3.915284
52.02991
2.126575
6.106109
1.444248
1.573182
2.341305
.4226906
.4173672
.425311
.4218423
.4251677
.4199073
.422089
.4217778
.9562161
1.649982
21.01229
.9906189
2.638362
.6994937
.7049515
1.097581
.1257912
.0632044
.0887395
.0147172
.1472277
.0311546
.1088857
.016317
.0902128
.1947759
.1521684
.9896216
.9944454
1.030322
1.156329
1.154315
1.112378
.8911179
.8937187
.8858359
.8858321
.8783079
.3819083
.8883239
5.47
-0 . 60
-1 .39
5.13
;:.88
:. .01
:. .72
1.95
-5.96
-2.01
2.08
3.83
2.54
D.S5
3.04
0.59
0.32
0.17
0.21
0.05
0.03
0.16
0.15
C.16
5.55
1.91
-2.09
-3.81
-2.58
-0.67
-0.22
-0.72
-7.04
-2.29
2.61
-3.59
:.83
1.86
-0.24
1.64
0.98
-1.55
3.54
3.62
0.61
0.53
-0.36
-0.11
-0.33
0.41
0.43
0.24
-0.34
-0.38
-0.27
-0.56
o.oco
0.546
0.059
0.000
0.004
0.315
O.OS7
0 . 052
O.OCO
0.045
0.038
0.000
0.011
0.581
0.970
0.558
0.751
0.864
0.836
0.963
0.979
0.873
0.884
0.873
O.ODO
0.057
0.037
0.000
0.010
0.500
0.828
0.469
0.000
0.022
0.009
0.000
0.067
0.063
0.814
0.101
0.330
0.121
0.000
0.534
0.543
0.593
0.725
0.914
0.740
0 .683
0.668
0.314
0.735
0.702
0.787
0.577
.4116161
-5.226376
-9.290349
5.632342
2.749318
-.8275927
-. 2392759
-.02C8
-16.87409
-15.5432
6.072946
3. 98176
3.531668
-2.035792
-3.027376
-3.223655
-.6951841
-.7476997
-.7473267
-.808:.605
-.8233076
-.7570911
-.7667888
-.7600494
3.433597
-.093561
-85.05993
-5.715057
-11.97605
-1.844784
-1.536486
-2.943247
-1.132346
-.2687437
.0573581
-.0817333
-.01S2923
-.0031023
-.2393576
-.0052566
-.0850533
-.6E'4367
.2394368
-1.3:5883
-1.346445
-1.47103
-2.6".-' 61 93
-2. 35)0082
-2. 5!. 1791
-1 .38408
-1.3V0777
-1.5:>9761
-2.033653
-2.059693
-1.958565
-2.238771
.3730623
2.768145
.1735218
12.60576
14.49152
2.565651
3.567346
5.311664
-8.519948
-.1769526
210.2739
12.32789
27.49621
3.632421
3.146866
5.965226
.9637425
.8903343
.9242388
.8474369
.845341
.8909118
.8897767
.3952946
7.136442
6.385097
-2.593399
-1.827192
-1.621313
.9005073
1.230225
1.35941
-.6386553
-.0206866
.4056324
-.0239731
.558529B
.1191695
.1879843
.0587823
.2650034
.0800667
.8366495
2.558063
2.556438
2.572659
1.862033
2.140238
1.813941
2.113275
2.136785
1.946863
1.437957
1.387387
1.492645
1.247618
-------
382
year93 1 -.4383989 .8846305 -0.50 0.620 -2.174346
year94 1 -.2521909 .8871148 -0.28 0.776 -1.993013
year95 I .7265304 .8293872 0.98 0.381 -.9010104
year96 I .7370473 .8285022 0.89 0.374 -.8887569
ysar97 1 .7486399 .8289987 0.90 0,367 -.8781336
year98 I .6435452 .8349206 0.77 0.441 -.9948541
year99 I .760182 .83654 0.91 0.364 -.8813951
_cons I 12.45681 .9428994 13.21 0.000 10.60652
Hypothesis P-VJ
ofF
1.297548
1.488631
2.354071
2.362351
2.375418
2.281945
2.401759
14.3071
ilue Reject?
•test
All structural attribute slopes simultaneously zero o . o o o o
All lotsize-independent year-specific slopes on DOWNSTR o . o o o o
simultaneously zero
Ail lotsize-independent year-specific slope on DOWNSTR the o . oooo
same
All lotsize-independent year-specific slopes on LDIST 0.3535
simultaneously zero
All lotsize-independent year-specific slope on LDIST the same 0.2775
All lotsize-independent year-specific slopes on DOWNSTR*LDIST ° • 000°
simultaneously zero
All lotsize-independent year-specific slope on DOWNSTR* LDIST ° • 000°
the same
All lotsize-independent other distance effects simultaneously zero o - oooo
All lotsize-dependent year-specific slopes on DOWNSTR o . oooo
simultaneously zero
All lotsize-dependent year-specific slope on DOWNSTR the same o . oooo
All lotsize-dependent year-specific slopes on LDIST o . ooee
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on LDIST the same (on o . oo 4 1
vX Idist variables)
All lotsize-dependent year-specific slopes on DOWNSTR*LDIST o . oooo
simultaneously zero (on vX Idist variables)
All lotsize-dependent year-specific slope on DOWNSTR*LDIST o . oooo
the same (on vX Idist variables)
All lotsize-dependent other distance effects simultaneously zero (on ° • ° ° 1 2
vX "other distance" variables)
-------
f
-------
t
ECONOMIC VALUATION OF MORTALITY RISK
REDUCTION
Volume II
THE EFFECTS OF AGE AND FAMILY STATUS ON THE
VALUE OF STATISTICAL LIFE - EVIDENCE FROM THE
AUTOMOBILE MARKET AND A NATIONAL SURVEY OF
AUTOMOBILE USE
William Schulze
Project Director
Cornell University
Ithaca, NY
Prepared by:
Timothy Mount, William Schulze
Weifeng Weng, and Ning Zhang
Department of Applied Economics and Management
Cornell University
Ithaca, NY 14853
Laurie Chestnut
Stratus Consulting
Prepared for:
U.S. ENVIRONMENTAL PROTECTION AGENCY
C.R. 824393-01-0
November 2004
Project Officer
Dr. Alan Carlin
National Center for Environmental Economics
Office of Policy, Economics and Innovation
U.S. Environmental Protection Agency
Washington, DC 204060
This research was supported by United States Environmental Protection Agency Cooperative Agreement Number CR824393-01-1. We would
like to thank Margaret French for her assistance in preparing the manuscript. All conclusions and remaining errors are the sole responsibility of
the authors.
-------
TABLE OF CONTENTS
TABLES .3
ABSTRACT 4
SECTION 1 INTRODUCTION 5
SECTION 2 THEORETICAL ISSUES 7
SECTION 3 SURVEY DESIGN AND IMPLEMENTATION 18
SECTION 4 ECONOMETRIC ANALYSIS — 23
4.1 HEDONIC PRICE AND FUEL EFFICIENT MODELS 25
43. ESTIMATING THE VSL FOR THE DIFFERENT TYPES OF FAMILIES 26
43 VSL FOR FAMILIES WITH MULTIPLE MEMBERS AND MULTIPLE VEHICLES 27
4.4 ESTIMATING THE COMPONENTS OF A VSL MODEL 29
4.4.1 Estimates of Risk by Vehicle...?. 29
4.4.2 Household Types in the Survey Data 31
4.4.3 Estimating How Vehicles Are Used 35
SECTION S CONCLUSIONS: ESTIMATES OF AVERAGE VSL BY GROUP AND
INCOME LEVEL 41
REFERENCES 46
APPENDIX A 48
APPENDIX B ~ 53
APPENDIX C 64
APPENDIX D: AUTO SAFETY SURVEY (FULL SCALE) 80
f
-------
TABLES
Table 3.1 Disposition 20
Table 3.2 Response Rate Data For Follow-Up Survey 21
«
Table 3.3 Detailed Response Rate Information 22
Table 4.1 Distribution of Six Types of Household (HH) by the Number of Vehicles Owned 32
Table 4.2 Demographic Characteristics of Representative Households 33
Table 4.3 Total Annual Riding Miles of the Family in Each Vehicle (TPM) 33
Table 4.4 Total Annual Miles Driven per Vehicle (TVM) 33
Table 4.5 Household Characteristics 34
Table 4.6 Parameter Estimates for Mileage and Occupancy 36
Table 4.7 Parameter Estimates for Occupancy by Kids 36
Table 4.8 Parameter Estimates for Allocating TPM, TVM and KM for a 2-Vehicle Household 38
Table 4.9 Parameter Estimates for Allocating TPM and KM to the First Vehicle in a 3-Vehicle
Household 40
Table 5.1 Estimated VSL for Families 42
Table 5.2 Income Elasticity Estimates 43
Table 5.3 Fragility Adjusted VSL (Smillion) by Family Group 45
-------
Abstract
This study reports on a new national survey of individual automobile usage designed to
provide information on automobile safety expenditures by family status and age. Noting that, for
a family, the safety of an automobile is a public good, these data, when combined with an
analysis of the PARS data set on fatal automobile accidents and a hedonic price function for
automobiles, allows estimation of the value of statistical life for individual family members over
their lifetime. The research also attempts to resolve1 the problem of an anomalous sign on the
coefficient on fuel consumption in prior hedonic price studies of automobile safety. The principal
result is that the value of statistical life remains relatively constant over the lifetime for all family
members with the exception of parents with children living in the household, who have a lower
value. Adults without children do not show a similar decrease in the value of statistical life.
Estimates of income elasticity are also presented and a theoretical explanation for the results is
provided.
t
t
-------
Section 1
Introduction
Little work has been done either theoretically or empirically to value morbidity and
mortality either for children or retired adults (for exceptions see Blomquist, et al., 1996, and
Jenkins, et al. 1999), This paper attempts to address both of these issues by first presenting a
theoretical model of how families value risk and then examining family automobile purchases.
In particular, using a standard model of family decision-making, we show that parents may value
risks to their children's lives (the model assumes two altruistic parents) through Nash
cooperative bargaining to determine how much money to invest in the safety of their children. To
allow empirical estimation of values, automobile safety is then shown to be a family public good,
where the marginal cost of purchasing and operating a safer automobile is set equal to the usage-
weighted sum of the values of statistical life (VSL) of family members. We use data on
automobile purchases to estimate how much families with children spend on automobile safety,
how much families with retired members and no children spend on automobile safety, and how
much families without children or retired members spend on automobile safety. This not only
allows estimation of an average value of a statistical life (VSL) for each type of family, but also
allows estimation of an average value of a statistical life (VSL) for different age groups
(children, adults and seniors) by family type and income level.
The research reported here combines primary data on automobile usage by family
members with secondary data from both the automobile market and the PARS data set on
automobile accidents. This allows calculation of the VSL for different family members from
choices made concerning the type of vehicle and usage pattern by family members. An important
issue that has clouded the potential reliability of the VSL obtained from estimated hedonic price
functions for automobiles (that include risk of death) is that prior studies have shown what
appears to be a positive correlation between fuel consumption and the price of automobiles rather
than the expected negative correlation (people should be willing to pay less for cars with poor
fuel economy). Our theoretical work in the next section provides a possible explanation that also
suggests a revised estimation procedure.
-------
The paper is organized as follows: Section 2\ presents a theoretical model of family
automobile purchase decisions focusing on safety, fuel usage, how safety values for each
individual are determined in a family setting, and proposes a methodology for estimating the
VSL of family members of different ages. Section 3 describes the survey methodology used to
obtain new data on automobile usage by children, adults, and seniors. Section 4 addresses the
problem of driver characteristics affecting estimate? of the inherent risk of fatality of different
automobiles and develops a procedure for identifying the driver independent level of risk,
summarizes our empirical work estimating a hedonjc price function for automobiles showing a
negative correlation between risk of fatal accident and price and fuel costs, and addresses issues
which arise with multiple vehicle families. Section 5 presents estimates of average implied
values of life for different family groups and income levels by age as well as estimated income
elasticities.
6
-------
Section 2
Theoretical Issues
How willingness to pay (WTP) for health and safety may vary with the age of the person
at risk is a very important policy question for which we have little well-established empirical
data Cropper and Freeman (1991) address this question with a life-cycle consumption-saving
model that they apply with a quantitative example to examine how WTP for a risk reduction in
the current time period can be theoretically expected to change over a person's lifetime. This
model is based on the premise that a person makes consumption and saving decisions over time
to maximize personal utility. Because this model is based on the premise that utility is a function
of consumption, the authors note that, if there is additional utility derived from survival per se,
then the life-cycle model provides a lower bound estimate of WTP. The quantitative example
depends on assumptions regarding a lifetime pattern of earnings, endowed wealth, the rate of
individual time preference, and other parameters of the model. These will all vary for different
individuals, and uncertainty exists empirically about population averages for many of these
factors. However, using reasonable values to calibrate the model is illustrative. Cropper and
Freeman note that if consumption is constrained by income early in life, the model predicts that
VSL increases with age until age 40 to 45, and declines thereafter. Shepard and Zeckhauser
(1982) also illustrate this point with numerical examples for the life-cycle model. When they
estimate the model with reasonably realistic parameters and assume no ability to borrow against
future earnings or to purchase insurance, they find a distinct hump in the VSL function with a
peak at around 40 years and dropping to about 50% of the peak by 60 years. When they allow
more ability to borrow against future earnings and to purchase insurance, the function flattens
and at 60 years drops only to 72% of the VSL at age 40. However, the hump shape to the VSL
over a person's lifetime remains.
The conclusions reached by these theoretical analyses of the effect of age on WTP for
mortality risk reduction using the life-cycle model are somewhat consistent with the empirical
findings obtained by Jones-Lee et al. (1985). However, the empirical findings show that WTP
varies with age much less than would be predicted by the life-cycle models. In this stated
preference study, respondents gave WTP estimates for reductions in highway accident mortality
-------
risk and the answers showed a fairly flat hump-shaped relationship between VSL and age,
peaking at about age 40. Although the directions of the changes in WTP with age are consistent
with what the life-cycle models predict, the magnitudes of the changes are smaller. The Jones-
Lee et al. results show that at age 65 the VSL is about 90% of the VSL of a 40-year-old person.
It is often suggested that WTP will be lower for the elderly than for the average adult
because expected remaining years of life are fewer. This expectation is based on the presumption
that WTP for one's own safety declines in proportion to the remaining life expectancy. Some
analysts have suggested that effects of age on WTP might be introduced by dividing average
WTP per statistical life by average expected years 6f life remaining (either discounted or not) to
obtain WTP per year of life (Moore and Viscusi, 1988; Miller, 1989; Harrison and Nichols,
1990). Such a calculation implies very strong assumptions about the relationship between life
expectancy and the utility a person derives from life; namely, that utility is a linear function of
life expectancy and that the value of life year remains constant.
Determining appropriate WTP values for changes in mortality risks to children poses
some particular analytical challenges. Children are not the economic decision makers whose
preferences can be analyzed to determine an efficient allocation of society's resources regarding
their own health and safety, so both revealed and stated preference approaches must rely on
parental decisions to show what WTP for children's health and safety might be. Based on the
expected relationship between WTP and expected life-years lost, it may be reasonable to assume
that reductions in risks to children are valued equal to or greater than risks to adults. Blomquist,
et al. (1996) support this view in their analysis of seat belt use for children. On the other hand,
the life-cycle consumption-saving models show increasing WTP for risk reductions between the
ages of 20 and 40, reflecting the typical pattern of increasing income and productivity during this
stage of life. Extending this to children might suggest lower WTP for reducing risks to children,
however, this pushes beyond the theoretical constructs of the life-cycle model regarding an
individual as an economic decision maker. The only theoretical model that addresses these
concerns, with respect to dependent children, has been developed by Chestnut and Schulze
(1998). Their work treats the case of a family with non-paternalistic altruistic parents who
-------
engage in Nash cooperative bargaining to determine health and safety expenditures on their
children and the implied VSL. We use this model as a starting point for our analysis.1
As indicated in the introduction, a secondary theoretical issue is that fuel consumption
appears to have the wrong sign in existing hedonic price functions for automobiles that have
been used to estimate the VSL (Atkinson and Halvorsen, 1990, and Dreyfus and Viscusi,1995).
Atkinson and Halvorsen (1990) use the data for 112 models of new 1978 automobiles to obtain
estimates of the VSL. Since the available fatality data is a function of both the inherent risk of
the vehicle and the driver's characteristics, the drivers' characteristics are included in the
regression as control variables. Their estimated VSL for the sample as a whole, based on
willingness to pay, is $3.357 million 1986 dollars. The data used in Dreyfus and Viscusi (1995)
differ from those used in earlier studies in that they reflect actual consumer automobile holdings.
Dreyfus and Viscusi (1995) use the 1988 Residential Transportation Energy Consumption
Survey together with data from industry sources. They generalize the standard hedonic models to
recognize the role of discounting on fuel efficiency and safety. Their estimates of the implicit
value of life range from $2.6 to $3.7 million. Both studies show a positive correlation between
automobile price and fuel consumption.
Given the state of existing research, our first task is to develop a model that can
potentially explain the positive correlation between automobile price and fuel consumption. The
second task is to develop a model of the behavior of households with dependent children. This
model is developed in the context of automobile safety to allow empirical estimation of the VSL
for family members by age group, family status, and income group. The existing theoretical
literature only considers individuals rather than families, with the exception of the work by
Chestnut and Schulze mentioned above.
1 It should be pointed out that some interesting revealed preference empirical approaches based
on a household production function framework to analyze household expenditure decisions as they
relate to children's health have been attempted {Ague and Crocker, 1996; Joyce et al. 1989). These analyses
infer implicit WTP for changes in children's health as revealed by expenditure decisions of the household.
Limitations in available data and analytical difficulties in properly specifying and verifying modeled
relationships pose challenges for this approach; however, its basis in actual household decisions and
behavior is an important strength. Estimates of WTP for changes in mortality risk for children are not
directly available from these two studies, but similar approaches might be applied to obtain such WTP
estimates.
-------
ft
To begin, we address the problem of fuel consumption by considering the case of a single
individual with no family who may, or may not, survive for a single period. The following
notation will be useful:
c = consumption,
w = wage income,
r = risk of a fatal automobile accident per mile driven,
n = probability of survival without automobile fatality risk,
fl-r <= probability of survival with automobile fatality risk,
m = total miles driven
a = level of some other positive automobile attribute (e.g., acceleration)
P(r,a) - automobile price per mile driven (decreasing in r and increasing in a)
F*(r,a)) = fuel consumption per mile (increasing in r and a)
G = price of fuel
U(c,a,m) = strictly concave utility function.
Note: subscripts or primes denote derivatives where appropriate.
Note that we propose that the individual realizes that the fuel consumption of the car is
itself a function of the attributes of the automobile. We will justify this proposal when we
consider the manufacturer's decision below. Also, to abstract from the life cycle issues of
owning and financing an automobile, we analyze the problem in terms of the annualized price
per mile of owning the vehicle, P, without loss. In this setting, the individual must make four
choices. First, the individual chooses the level of consumption, c. Second, this is traded off
against the choice of automobile safety (how risky per mile a car to purchase, r) taking into
account that lower r implies that both the price of ;the car itself over the m miles driven each year,
P(r,a)m, and total cost for fuel with price per gallon G and fuel consumption F* driven m miles
per year, GF*(r,a)m, are greater for a safer car since Pt, Fr*<0. Third, the individual chooses the
other characteristic of the car, a, realizing, for example, that increased acceleration will both
increase the price of the car and increase fuel consumption since Pa, ¥»*>$. Fourth, the individual
will choose how many miles to drive, m. The individual is assumed to maximize expected utility,
10
-------
(n-rm)U(c,a,m), (1)
where it is assumed that the death state provides no utility because the individual has no family,
subject to the budget constraint,
(n-rm)(w-c) - P(r,a)m - GF*(r,a)m = 0. (2)
This budget constraint assumes that costless insurance (priced at expected value) is available
both to cover the purchase price and operating costs of the automobile. Most car loans, in fact,
carry life insurance for the amount of the loan, and life insurance could presumably cover other
costs. The optimal choice for r, risk per mile, is determined by
VSL = -(Pr + GFr*), (3)
where
= (U/Uc) + w-c. (4)
Equation (4) sets the marginal increase in cost for purchasing and operating a safer car
per mile equal to the VSL. The VSL is defined in (4) for the case of perfect insurance markets
and is equal to the monetized value of utility, (U/UC), which is lost in death, plus the excess of
earnings over consumption. The interpretation of this relationship is much clearer in the family
setting that we treat below, so we will defer discussion.
The optimal choice of the attribute, a, is determined by
(5)
which sets the marginal willingness to pay for the attribute (acceleration) equal to the
incremental total cost.
The total miles driven, m, is determined by
11
-------
Um/Uc - rVSL - GF* = P
(6)
so that the marginal willingness to pay for an additional mile driven. Um/Uc, net of the risk cost
of driving an additional mile, net of the cost of fuel for an additional mile, GF*, is set equal to
the per mile capital cost of the car, P. It is this last condition that helps explain the peculiar result
obtained in prior estimates of the hedontc price function for automobiles. All buyers have the
same marginal value for improved fuel economy equal to G, the price of fuel.
Competitive automobile manufactures should attempt to minimize the cost per mile of
driving their automobiles including both the capital and fuel cost per mile of automobile life
given the choice of other characteristics (r and a). Thus, for any given vector of automobile
characteristics, manufacturers optimize fuel economy at the fixed marginal value of G. There is
no hedonic market for fuel economy per se because for any vector of attributes, there is only one
optimal level of fuel economy, because all buyers have the same marginal valuation of fuel
economy. This is unlike an other attributes, a, such as acceleration, where, for the same safety
level, there are a variety of marginal values for different buyers for acceleration depending on
tastes. For these attributes, makers respond by offering a variety of vehicles with the same level
of risk but different levels of acceleration. In contrast, the marginal value for fuel economy is
always G, so no hedonic market exists. Clearly, fuel consumption itself then becomes a function
of other car attributes. This can be shown by considering the design problem of a particular
manufacturer with a cost of production per mile of life for the cars that they offer of C(a,r,F).
Given a particular choice of a and r by a buyer, the maker is forced by competitive pressure to
minimize the total cost per mile to buyers,
C(a,r,F) + GF.
(7)
The condition for optimal fuel consumption in the engineering design of the vehicle is then
(8)
12
-------
•f
This implies that there is an optimum fuel consumption F*(a,r) for any choice by consumers of a
and r and the cost function relevant for the hedonic price solution for profit maximization over a
and r by the maker is C*(a,r,F*(a,r)). The maker faces a hedonic price function only defined in a
and r, P(a,r), not fuel consumption which is optimized in the engineering design of the vehicle,
and maximizes profits P(a,r) - C*(a,r,F*(a,r)) with respect to a. implying
Pa = Ca*, (9)
and with respect to r, implying
Pr = Cr*. (10)
So, a particular maker will pick a and r by setting marginal costs equal to the slope of the
hedonic price function for r, given a, and for a, given r, implying a mix of cars with different
levels of a and r available to consumers from different makers with different cost functions.
In summary, given G, the price of fuel, the choice of F will be made by the automobile
maker and becomes a function of r and a, since fuel usage will be optimized by makers for any
combination of these attributes chosen by consumers. Consumers and makers are faced with a
hedonic price function P(r,a) which is the envelope curve of the cost tradeoffs for makers and
value tradeoffs for consumers between attributes. Buyers face a pre-optimized choice of fuel
consumption, F*(r,a), for each level of attributes that they choose in their purchase decision.
If these arguments are correct, then adding fuel economy as an explanatory variable in
the estimated hedonic price function results in a mis-specification of the model. This mis-
specification could easily result in an anomalous sign on the coefficient for fuel economy.
Rather, the appropriate procedure may be to estimate F*(r,a) and P(r,a) and use (3) above to
estimate the VSL for the individual from these relationships and the price of gasoline, G.
The mode! developed above can readily be extended to a family setting by using the Nash
cooperative bargaining between parents approach employed by McElroy and Horney (1981).
Following our previous work (Chestnut and Schulze, 1998), we modify the notation used above,
again considering a single car family, as follows:
13
-------
n = the size of the family,
i = 1,2,...,,n denotes individual family members,
i = 1 denotes the mother,
i = 2 denotes the father,
i = k = 3, ,n denotes children,
Ci = consumption of the ith family member,
w; = wage of family member i,
r= automobile fatality risk per mile driven, the same for all family members,
flj =* probability of survival, excluding automobile fatality risk, of i,
m = total vehicle miles driven
m;=total miles of driving for family member i
P(r,a) = automobile price per mile driven,
F*(r, a) = fuel consumption per mile driven,
Uk (cic,a,nik) = child's utility function,
U'( Ci;...., mi,a,(nk-r)Uk(cit,mk),....) = parent's utility function (i = 1, 2), and
E'=individual expected utility in separation (i = 1, 2).
The family must decide how much to allocate to each family member for consumption,
on the risk level of the single automobile they purchase for all, attribute a. and the number of
miles driven for the car itself and each person who rides in the car. The hedonic price and fuel
consumption functions for the automobile are the same as in the previous model. Utility
functions of both the father and mother are assumed to depend not only on their own
consumption, driving and car attribute, but also on the expected utilities of each of their children.
The children's utility is assumed to be a function of their own consumption, the car attribute, and
the miles they ride in the car.
Investment in the safety of their children is a public good to the parents, which is the
subject of negotiation, as is the level of consumption of each. The Nash cooperative bargaining
model assumes that the solution maximizes the multiplication of the increase in the expected
utility of the outcome over the threat point of expected utility in separation for the mother and
14
-------
t
s
the father. The threat points, El, are assumed, in models of the family, to be a function of divorce
laws, job opportunities, etc. Thus, in the Nash cooperative bargaining solution,
[(fl ,-rmOU1 - E1] [(n2-rm2)U2 - E\ (1 1)
is maximized with respect to Ci, r, a, m, and mi, subject to the budget constraint,
(ni-rmi)(wi-ci)-(P-GF*)m = 0, (12)
and constraints on the use of the car such as,
m-m;>0 i = l, ..... ,n
so that no individual family member can ride more miles than the car itself travels, and
so that no child can ride more miles than the parents can collectively drive the child. Note that, to
avoid pointless complication of the model, mi2 is taken to be a constant number of miles that the
parents ride together, where it is assumed that mi, m2 > mn.
The resulting conditions for choosing the level of automobile risk and miles driven imply
that the individual VSLs of family members all take the form:
(13)
The remarkable fact is, that, in spite of the complicated structure of the problem specified
above, the implied VSL; for each family member shown in (1 3) is identical in form to that for the
single individual shown in (4) above. The interpretation of the VSL; can be illustrated with the
following examples. Imagine that the mother is the sole breadwinner with a stay-at-home father.
15
-------
In this case, assuming that the children are young, Wj - c; <0 for the other family members and wm
- cm >0 for the mother. Thus, if the mother were to die, this would be a severe financial blow to
the rest of the family and the mother's VSL would reflect this relative to the VSL of other family
members. For young children it is clear that Wk- CK <0 in the short run. However, in the inter-
temporal version of the model, wk~ Ck is replaced by its discounted present value, which may be
positive, U'/U'c depends primarily on qin the single period model and on the lifetime
consumption pattern in the full inter-temporal model. The important point is that the child's
consumption depends in youth on the parents' income and wealth. Further, if parents find the
value of their child's smile to be high enough, the child's consumption will be maintained by
them, at a high level, leading to a high VSL. A young child's utility and the utility they derive
from that happiness may also be large in the parent's view from relatively small levels of money
consumption, also leading to a high VSL. These arguments suggest that the VSL of children is a
purely empirical question and depends not only on their own life cycle wealth but also on their
family's wealth and the beliefs of the parents regarding their children's utility.
The choice of automobile risk, r, is determined by
(14)
t
where usage weights for the vehicle for each family member are defined as ki = m;/m. Thus, the
safety of the shared family vehicle is determined by a public good condition that sets the sum of
the usage weighted VSLs of individual family members equal to the marginal cost of obtaining a
safer automobile. The marginal cost of a safer vehicle is the slope of the hedonic price function
for automobile safety, -Pr, plus the marginal fuel cost penalty, -GFr, which, by (14), is set equal
to sum of the usage weighted VSL; for the family, kjVSL;, to determine the choice of per
1=1
mile automobile risk, r.
Thus, if we obtain predicted values for the marginal cost of reduced risk per mile (Pr +
GFr*) and the share of automobile use, k;, for each family member by age group for different
households, we can use equation (14) to obtain estimates of the VSLj. Note that equation (14) is
a single equation embedded in a system of simultaneous FOC equations. To each FOC equation,
16
I
-------
t
s
we appended an additive error term. Assume that each of these error terms is independently,
identically distributed over families around a mean of zero. Because mj and m are endogenous
variables in the simultaneous FOC equations, consistent estimates will not be obtained by using
the method of lest squares if m; and m are correlated with the disturbance term in equation (14).
A two-stage procedure is required to obtain the consistent estimates. In the first stage, reduced-
form equations for m; and m will be estimated using appropriate exogenous variables which
reflect the family characteristics. The predicted m; and m that are uncorrelated with the residuals
in equation (14) will be used as the instrumental variables for m, and m. In the second stage,
expression (14) will estimated by least squares using predicted mi and m (which provide
predicted kj) to obtain consistent estimates of the VSL for adults, children and seniors.
The next section describes the survey methodology used to collect the necessary primary
data to employ the proposed methodology.
17
-------
Section 3
Survey Design and Implementation
Secondary data describing the detailed usage of vehicles by family members has been
unavailable, Since such data are necessary to implement the methodology proposed in the last
section for measuring the VSL of family members, a national survey was undertaken to collect
data on how families choose and use automobiles, as well as on their attitudes and beliefs
regarding automobile safety. This survey consisted of two parts, a telephone screening survey
used to develop an appropriate sample and collect information on usage, followed by a mail
survey. Both the telephone and mail surveys were extensively pre-tested and revised prior to
implementing a pilot aimed at 80 households to formally test the telephone/mail survey
methodology. Only small changes were made to the survey instruments following this final test.
Both surveys can be found in Appendix D.
The purpose of the telephone survey was to identify appropriate households and to obtain
data on automobile usage that was judged too difficult for respondents to fill out themselves in a
mail survey. Note that the mail surveys were customized for each respondent and included
respondent specific information on automobile make, model, and purchase price. Both the
telephone and the mail survey were developed following Donald Dillman's Tailored Design
Method (1999).
The telephone survey begins by indicating that the interviewer is calling on behalf of
Cornell University. The first five questions determine if the interviewer and household meets the
requirements for the sampling. Question 6 asks fqr detailed information on automobiles owned
or leased by the household while Question 7 elicits information on the residents' ages and
relationships. Question 8 elicits the percentage of miles that each member of the household rides
in each of the three most driven cars. Needless to say, this is a difficult question and necessitated
a personal telephone interview with trained interviewers. Question 9 attempts to find out
whether household members typically ride in the front or back seat of the three most driven
vehicles. Questions 10 to 18 collect information on the reasons and distances to various
destinations that people drive their cars to help in explaining driving patterns. Question 19
t
s
18
-------
s
recruits respondents for the follow up mail survey. Questions 20 to 27 collect socioeconomic
data on respondents including income.
The cover of the mail survey booklet is titled "WHAT ARE YOUR VIEWS ON AUTO
SAFETY," shows a picture of a family next to a Ford Windstar (thanks to Ford for granting
permission to use the photo), and has indicates that the survey is being conducted for Cornell
University in the lower left hand corner. The first page thanks the respondent for "agreeing to
complete this important survey on automobile safety," and repeats the information on the most,
second most and third most driven automobiles taken from the telephone survey and asks the
respondent to correct any errors. Question Ml asks if the respondent has read or heard about
automobile safety in the last six months. Questions M2-M6 ask about insurance and repair costs
and features of each of the vehicles. The mail survey was necessary to allow collection of
subjective risk information from respondents that required use of a risk ladder as a visual aid.
Thus, M7 asks for a subjective risk assessment of having a fatal accident (compared to the
average driver in the same type of automobile) for the respondent. M8 asks for a subjective
assessment of a child's risk of dying relative to an adult's risk in a serious automobile accident
The next questions ask the respondent for their perceived risk of the safety of the vehicles that
they drive. The last two pages ask a Contingent Valuation question on the value of improved
automobile safety for comparison to the hedonic price estimates of the VSL to be obtained from
the study.
A random digit-dialing sample of 8519 telephone numbers was obtained from Sample
Survey Inc., a well-known and respected source of survey information. Although the target
number of completed mail surveys was only 600, past experience has shown that random digit
dialing produces a large number of non-household, disconnected, or ineligible numbers for
household surveys. The telephone screening survey was implemented between July 1 and
August 5, 2001 and employed a minimum of 13 attempts to reach each telephone number. The
completed telephone surveys averaged 14 minutes in length. After screening out businesses and
other non-household phone numbers, ineligible households such as those with more than 5
people or three automobiles (it proved impossible to design a manageable survey for such
households), those with no car, etc, but including those households which were unreachable after
13 or more tries, the overall response rate was about 40% for the telephone survey as shown in
19
-------
Table 3,1. This produced 1,235 completed interviews. Of these, 926 or 75% agreed to participate
in the mail survey.
Table 3. 1 Disposition
Total Cases (T)
Known non-household Ineligible (A)
Final Disconnect
Final Computer Tone
Business/Government
Non-Residential Number
Known Household Ineligible (B)
Ineligible - > 5 people, 0 autos, > 3 autos, employer vehicle, gift vehicle,
don't know make
Language Barrier
Known Household Eligible (C)
I
NC
Completed Interview
Non-Contact - Respondent not available for duration of
study
Refusals ( R)
R
R
SCR-Soft Mid-Interview Terminate
SCR-hard Mid-Interview Terminate
Unknown Household Status (D)
UH
UH
No Answer/Phone Busy
Initial Disconnect/Computer Tone
UO NON HUDI
UO
UO
UO
UO
Non-HUDI
Non-HUDI
Non-HUDI
Non-HUDI
UO HUDI
UO
UO
HUDI
HUDI
Total Dialed
Answering Machine
Remainder Respondent not available
Interviewer Reject
Scheduled Callback
Soft Refusal
Hard Refusal, Don't know/Refuse
Ql or Q2
Known non-household Ineligible (A)
Known household (KH) = ( I + P + R + INC + B)
Unknown Household Status (D)
Working numbers (WN) = KH + D)
Working % (WKG) = (WN / T)
Non-household % = (A / T)
Known household Ineligible (B)
Household Eligibility Rate (NEI) = (KH - B) / KH
Final
8,519
2,712
1,302
423
897
90
1,093
679
414
1.235
45
168
0
168
3,266
1,204
9
450
308
86
18
38
1,603
72
1,531
8,519
2,712
2,541
3,266
5,807
68.17%
31.83%
1,093
56.99%
s
s
20
-------
Table 3.1 (Continued)
Completed recruitment Survey (AAPOR RR4*)
**AAPOR RR4 = 11 [I + R + NC + (WKG*NEI*UH) + (NEI *
UO HUDI))1
39.98%
(UO_NON_HUDI +
The mail survey was sent in waves from July 6, 2001 to August 6, 2001. The survey packet
included a letter from Cornell University describing the importance of their response and the
nature of the study, a $5 cash incentive, the 12-page survey booklet, and a post-paid return
envelope. A reminder post card was mailed 7 days after each survey packet was sent thanking
those who had returned their survey and reminding those that had not to please complete the
survey or ask for a replacement. Two weeks after each survey packet was sent, follow-up phone
calls were made to non-respondent households with more than 6 attempts, if necessary. Table 3.2
presents the response data for the telephone follow up survey. The overall response rate for
completed follow-up phone calls was 78%.
I
t
Table 3.2 Response Rate Data For Follow-Up Survey
*
Starting Sample
Nonworking Numbers
Disconnected
Computer Tone
Ineligibles
More than 5 people
No autos
Adjusted Sample
Refusals (R)
More than 6 attempts
Active sample
Completed Reminder Call
(completes/adjusted sample)
Will return survey
Needs survey
Won't return survey
Already returned survey
Completed survey over the
phone
Count
394
7
1
13
5
368
4
55
0
309
140
45
9
114
1
Percent of Starting Sample
1.78%
0.3%
3.3%
1.3%
1.0%
34.7%
0.0%
78.4%
45.3%
14.6%
2.9%
36.9%
0.3%
Note: Response rate includes pretest calling
21
-------
The detailed response rate information for each wave of the mail survey by date mailed is
presented in Table 3.3. The overall response rate for the mail survey was 74% with 625
completed surveys, exceeding the initial target of 600.
Table 3.3 Detailed Response Rate Information
Filename
Pretest
Iist7-5f.xls
Iist7-9f.xls
Iist7-llf.xls
Iist7-13fxls
Iist7-16f.xls
Iist7-18f.xls
Iist7-20.xls
Iist7-23f.xls
Iist7-25f.xls
Iist7-27f.xls
Hst7-30f.xls
Iist8-lf.xls
Iist8-3f.xls
Iist8-6f.xls
TOTALS
Total
Quantity
80
98
242
70
42
58
30
36
49
53
72
22
12
42
20
846
Caseid range
1001-1080
2001-2098
3001-3242
4001-4070
5001-5042
6001-6058
7001-7030
8001-8037
9001-9049
10001-10053
11001-11072
12001-12022
13001-13012
14001-14042
15001-15020
Date
Survey
Mailed
7/2/01
7/6/01
7/9/01
7/1 1/01
7/13/01
7/16/01
7/18/01
7/20/01
7/23/01
7/25/01
7/27/01
7/30/01
8/1/01
8/3/01
8/6/01
Date,
Postqard
Mailed
7/9/01
7/13^01
7/16/01
7/18/01
7/20/01
7/23/01
7/25/01
7/25V01
7/27'/01
7/3Q/01
8/1/01
8/3/01
8/6/01
8/9/01
8/13/01
Date
Reminder
Calls
Began
7/20/01
7/25/01
7/25/01
7/31/01
7/31/01
7/31/01
8/3/01
8/3/01
8/3/01
8/9/01
8/9/01
8/15/01
8/15/01
8/17/01
8/17/01
Response
Rate
Before
reminder
Calls
Began
65%
65%
49%
64%
62%
40%
40%
57%
49%
51%
42%
41%
33%
67%
40%
Number of
Completed
Mailed
Survey
64
74
180
51
31
45
18
27
37
39
53
11
6
36
17
625
Final
Response
Rate
80.0%
75.5%
74.4%
72.9%
73.8%
77.6%
60.0%
75.0%
75.5%
73.6%
73.6%
50.0%
50.0%
85.7%
85.0%
73.9%
t
t
22
-------
t
Section 4
Econometric Analysis
For purposes of describing the econometric analysis, we use the following notation: Let
TPM = total annual personal riding miles of the family in the automobile,
TVM=total annual driving miles of the family in the automobile, which is generally less than
TPM,
M ,.=total annual personal riding miles of the ith family member,
MM= total annual mother riding miles in the automobile,
FM = total annual father riding miles in the automobile,
KM = total annual children riding miles in the automobile,
(TPM= E, M, =MM+FM+KM)
r = average automobile inherent fatality risk per driving mile per occupant (the same for all
family members),
P(r) = automobile price or capital cost per driving miles (decreasing in r),
P(r) xTVM= annual automobile price or capital cost per family
F(r)= automobile fuel consumption expenses per driving mile (decreasing in r), and
F(r) xTVM=anmml fuel consumption expenses per family.
Using this notation, the approach used in the study to obtain the VSLs of family members
requires estimation of
-[P'(r)+F'(r)]=Sf=1 M,VSL,/TVM (15)
Thus, the safety of the shared family vehicle is determined by a public good condition
that sets the marginal cost of obtaining a safer vehicle for the family equals to the usage-
23
-------
weighted average of individual family member's VSL where the weights are each family
member's relative use. The marginal cost of a safer vehicle for an each occupant -P'(r) and -F'(r)
can be derived from the slope of the hedonic price fjunction HP(r,O) and hedonic fuel efficiency
function HF(r,O) for automobile safety, r is still the'automobile fatality risk per driving mile per
occupant and O is the other automobile characteristics.
Each family will select the available automobile risk-price and risk-fuel efficiency
combination that yields the maximum expected utiljty for the whole family. This is obtained
where P(r) is tangent to the hedonic price function HP(r,O) and F(r) is tangent to the hedonic fuel
efficiency function HF(r,O).The equilibrium obtains when P' (r)= HP, and F'(r)=- HFr.
Hence, we can use the slope of the hedonic price arid fuel efficiency functions with respect to r to
get the marginal cost of obtaining a safer vehicle for each family.
By (15), the marginal cost of obtaining a safer vehicle for each family is set equal to the
usage-weighted average VSL for the family, X "_t M, VSL,. /TVM, to determine the choice of
automobile risk, r. If we use hedonic functions to represent the left hand side of equation (15), it
reflects vehicle characteristics. The right hand side of equation (15) reflects the family
characteristics. Based on this equation, the VSL for each family member can be estimated using
the expected driving habits of individual family members.
t
24
-------
f
t
4.1 Hedonic Price and Fuel Efficient Models
The first step is to obtain the marginal cost of a safer vehicle for a family owning vehicle
j using hedonic models:
(16)
The right hand side of equation (15) is the marginal cost of purchasing and operating a
safer vehicle j, and it can also be regarded as the standard VSL for vehicle j (StCVSL,)).
Obviously, it should only depend on vehicle characteristics such as the make, model and year of
a vehicle. We standardize the total annual driving mileage in each vehicle to 14000 miles. With
the data on vehicle characteristics and average risk of a fatality per riding mile per occupant for
different types of automobile, we can estimate hedonic indices of the purchase price and fuel
efficiency for each vehicle. The standard expression for determining the marginal cost
(StCVSL,. )) for any make, model and year of vehicle j from the hedonic price and fuel efficiency
models is:
St(VSL,)=-pmxP./[ryx 14000x2^, (-L-)< ] + aj (^ xfe ^city,) (17)
Where P m is the regression coefficient for inherent vehicle risk in the hedonic price model,
P^ is the purchase price of vehicle j,
TJ is automobile fatality risk per driving mile per occupant,
i is the discount rate, set to 10 percent,
L j is the expected vehicle life, set to Max{ l,10-(purchase year ^ -model year, )}, which
standardizes the age effect of a vehicle on its price,
fe_city j is the fuel efficiency in miles per dollar of gasoline for vehicle] in year 2001,
(ignoring the difference in gasoline price at different locations)
25
-------
am is the regression coefficient for inherent vehicle risk in hedonic fuel efficiency model.
Now, given information on the characteristics of each vehicle j, and estimates of pm and
am from the hedonic functions, we can calculate St(VSL}) by Equation (17). From Equation
(15), VSLj a _(F+F% = ^"^MfVSLf/TPMj. If we divide people mto three groups
according to age: adults (16-64), seniors (>=65) anjj kids (0-15), the VSL for each age group in
the jth vehicle is VSL aj, VSL SJ and VSL ^, respectively.
t
Equation (15) becomes:
StCVSL,.)
=(AMy
KM
^+SM,. xVSL^./TVM,)
(18)
where AMy, KM . and SM, are the total riding miles of adults, kids and seniors in the jth,
vehicle respectively.
4.2 Estimating the VSL for the Different Types of Families
If we assume that the VSLs of adults, kjids and seniors are constant across different
I
families, then VSLa, VSLk and VSLS can be treated as parameters and estimated from Equation
(18) directly. However, the VSL in different types of households will almost certainly vary. For
example, an adult in a rich family is likely to have, a higher VSL than one in a poor family. Thus,
to estimate the VSL for different income groups the sample will be split into families with low,
medium and high incomes.
26
t
-------
Another way to remove the influence of family characteristics on VSL is to express
y, FSL^.and VSL^ as functions of family characteristics. Among all the family
characteristics that might affect the VSL, income is the most important, and VSL is almost
certainly positively related to the average income of a household. If EY is the average income
fj-v
per adult equivalent in household i, we assume VSL(i)=/7, + (3EY log(^=) where £T is the
EY
average equivalent income for all households. $ is the purified VSL for a household, and (18)
becomes:
, *AM/TVM+/7/, *KM/TVM + /?S*SM/TVM
where TPM=KM+SM+AM.
FY
4^)*TPMATVM] (19)
The VSL for adults, seniors and kids can be estimated directly by using OLS regression.
Estimated parameters pA, PK and fis correspond to the average VSL for adults, seniors and
kids for a household with average equivalent income EY, respectively. PEY measures the
income effect on VSL and it can be used to calculate the income elasticity.
4.3 VSL for Families with Multiple Members and Multiple Vehicles
Assume that a multiple-vehicle family bought all vehicles owed by the family, step by
step, rather than simultaneously. For example, there were no other vehicles owned by a two-
vehicle family when it determined the optimal risk-usage-price-fuel efficiency combination for
the first vehicle. The expected utility maximization problem faced by the family for the choice of
the first vehicle is not different from the problem faced by a one-vehicle family. We can derive
the same formula for VSL associated with the first vehicle as Equation (15):
27
-------
=i;=1 Ml,VSL,/TVM1
(20)
t
where the number "1" attached to variables represents the corresponding variables for the first
vehicle. When the family determined to buy the secjond vehicle, the first vehicle's condition and
all physical variables related with the first vehicle have been fixed and could be regarded as
exogenous variables. The family chose the optimal risk-usage-price-fuel efficiency combination
for the second vehicle conditional on the existing first vehicle. The expected utility maximization
problem faced by the family for the second vehicle.is:
[(nm - r\MM\ - r2MM2)UM (cm (nk - rlKMl -r2KM2)Uk (ck,KM\
MM\+MM2)-Em]x[(nf-r\FM\-r2FM2)Uf(cf(Kk -r\KM\-r2KM2)
is maximized with respect to M2 , , TVM2, c , , TI , , and r2, subject to the budget constraint,
S;=1(7t/-rlxMli-r2xM2).)x(W).-ci.)-P(rl)xTVMl-F(rl)xTVMl
- P(r2) x TVM2 -F(r2) x TVM2- H(« , , . . . ,n , )=0
The FOCs for this problem give a result similar to Equation 15) for the second vehicle.
-[P'(r2)+F'(r2)] =1^ M2,VSL(/T1VM2
(21)
This implies that if we can obtain vehicle characteristics and the family usage variables for
an individual vehicle, the same procedure for estimating VSL for a one-vehicle family can be
applied to a,multiple-vehicle family. In our empirical work, if the family has multiple vehicles,
firstly we estimate the TPM and TVM in all vehicles owned by the family. Secondly, we allocate
the TPM and TVM to each vehicle j to get TPM _, and TVM,. Thirdly, we decompose TPM_,.
28
-------
t
into AM j, KMy and SM^. Finally, the VSL for adults, kids and seniors can be estimated using
Equation (18) for each vehicle.
4,4 Estimating the Components of a VSL Model
In order to get a consistent estimate of VSL aj from Equation (18), we need to get the
appropriate measures of TVM,, AM i, KM y. and SM, accounting for the fact that these
variables are determined by the family (i.e. endogenous). Decisions to purchase a vehicle are
made on expectations about how the vehicle will be used. The new data set collects enough
information on family characteristics and how vehicles are used, to estimate the mileage
variables associated with each vehicle. In addition, the risk of having a fatality, r, must be
determined for each type of vehicle, and used to estimate hedonic models for the purchase price
and fuel efficiency.
4.4.1 Estimates of Risk by Vehicle
When a family makes a decision to buy a new or used vehicle, the selection is based on
expectations about how the vehicle will be used. The most important factors considered for the
analysis are how far the vehicle is driven each year and what is the typical occupancy rate. The
price of the vehicle and the fuel efficiency, the two primary economic costs to the family, will be
determined by the vehicle's physical characteristics. These characteristics include the size,
power, and quality of the vehicle, and most importantly for the analysis, the safety of the vehicle.
The safety ratings of each type of vehicle were estimated from an earlier analysis of data on
traffic fatalities (Fatal Accident Reporting Service, PARS) and vehicle ownership (National
Personal Transportation Survey, NPTS). This analysis has been presented in full in a report to
the EPA (Environmental Protection Agency) and a research paper.
The safety rating of a vehicle was determined by estimating the probability per thousand
miles traveled of having a fatality in an accident. This safety rating was determined by the
probabilities of having different types of accidents (one-vehicle, two-vehicle and multi-vehicle),
29
-------
and the probabilities that the occupants will survive jin these accidents. .All of these probabilities
are functions of the vehicle's characteristics and the characteristics of the driver and the
occupants. For example, heavy vehicles are relatively safe in a two-vehicle accident, but may
have a relatively high probability of having a one-vehicle accident. Wearing a seatbelt is more
important for survival in an accident than having an1 airbag. The statistical framework for the
different models underlying the safety rating of a vehicle is described in Appendix A, and the
estimated models and definitions of the explanatory variables are presented in Appendix B.
The safety rating of each type of vehicle is computed using the same set of characteristics
for the driver and the occupants. The rationale is to standardize the effects of driving behavior.
Some types of vehicle, for example, have higher probabilities of accidents because the drivers
are more likely to fail tests for sobriety. Similarly, very young drivers have higher probabilities
of having an accident. In general, the overall probability of having a fatality in a vehicle is
proportional to the number of miles driven and the total number of occupants. The safety rating
used in the hedonic models for each type of vehicle was computed under the assumption that
there are two adults in each vehicle who drive 14,OpO miles in a year. The effect of making this
assumption is that some vehicles, which have high-observed rates of fatalities, such as pickup
trucks, have lower predicted rates of fatalities. Theireason is that the specified occupants are
more safety conscious (e.g. by wearing seat belts) than the typical behavior of the actual
occupants in the fatality data.
Using a standardized set of characteristics for the occupants is an important
distinguishing feature of this analysis compared to other studies in the literature. A discussion of
other studies and the estimated safety ratings from our analysis are presented in Appendix C. The
safety rating of a vehicle measures the probability of having a fatality for a specified number of
miles driven. This measures the value of r in the hedonic models for the price and the fuel
efficiency of each type of vehicle (make, model and year). The two estimated hedonic models
are also presented in Appendix C. The estimated elasticities for r used to compute the standard
VSL in Equation (11) are am= 0.0258 for the fuel efficiency and pm = -0.069 for the price of the
vehicle.
t
30
-------
8
4.4.2 Household Types in the Survey Data
The 2001 National Auto Safety Survey (Full Scale) includes two parts: a recruitment
survey and a mail survey. It obtains information on household characteristics related with the
choice of automobiles and the use of automobiles, and vehicle characteristics such as the make,
model, year, price and perceived risk of a fatality (i.e. safety factor).
The main characteristics of the survey data are:
• It merges information about household characteristics and vehicle characteristics into the
same data set;
• It provides detailed information on the usage of different vehicles by individuals in a
family. Hence, the expected total personal riding miles of a family in each vehicle
(TPM), total vehicle driving miles in each vehicle (TVM) and the riding miles of each
age group of family members, such as AM, SM and KM can all be estimated.
• It includes a risk ladder of different types of vehicles. The ladder assumes that each type
of automobile is driven an average of 14,000 miles per year by someone with average
driving ability. Since the drivers' characteristics are standardized by average driving
ability, the effects of drivers' characteristics are removed and this risk ladder reflects the
inherent risk associated with each type of vehicle. The risk is measured by the number of
fatalities occurring in each year for every 100,000 automobiles per occupant using the
models described in Section 3.1. Therefore, we can derive the automobile fatality risk per
driving mile per occupant by using the formula: [risk value/(l 4,000 x 100,000)].
The survey covers 1147 sampled households, with no more than five family members,
owning 1 vehicle, 2 vehicles or 3 vehicles. For each household, there are 349 variables, each one
corresponding to a question. Only 623 households completed both surveys, and only 487
households answered all of the important questions about family member's age, total riding
miles, each person's riding percentage, and the cost of gasoline. There were five households that
reported at least one vehicle driven over 80,000 miles per year (the average miles driven per year
for the vehicle with the highest VMT per family is less than 16,000 miles), and three households
31
-------
reported a vehicle with zero miles driven. These households were regarded as outliers and
deleted. Therefore, the final sample had complete information about 479 families and 791
vehicles, and a description of this data set follows.
!
Table 4.1 Distribution of Six Types of Household (HH) by the Number of Vehicles Owned
ll -vehicle [2-vehicle
Type of HH HH BH i
PAHH 75 |l27
AKHH
SKHH
ASKHH
SAHH
PSHH
Total
29
0
1
6
25
136
si i
i
0
11
23
259:
3-vehicle
HH
48
29
0
1
2
4
84
Total
250
155
1
2
19
52
479
Kid: 0<=age<=15
Adult: 16<=age<=64
Senior: 65<=age
PA HH: every family member is an adult
AK HH: household is composed of adults and kids
SK HH: household is composed of seniors and kids
ASK HH: household is composed of adults, seniors and kids
SA HH: household is composed of seniors and adults with at least one member younger
than 60
PS HH: all family members are no less than 60 years old, and at least one member is
a senior
From the Table 4.1, we can define three types of representative household that have a
relatively lange number of families in the sample. Tthese are:
(1) 2-vehicle PA HH: Pure adults household with 2 vehicles;
(2) 2-vehicle AK HH: 2-vehicle household with both adults and kids;
(3) 1-vehicle PS HH: 1 -vehicle household with every family member no less than 60
and at least one member is a senior.
32
-------
The basic demographic characteristics of the three representative households in the
survey data are listed in Table 4.2.
Table 4.2 Demographic Characteristics of Representative Households
[Number
Type of HH bfHH
2-vehicle PA
HH
2-vehicle AK
HH
1 -vehicle PS
HH
127
97
25
Total [Total
Adults [Kids
260
206
2
0
176
o
Total
seniors
0
0
30
Table 4.3 Total Annual Riding Miles of the Family in Each Vehicle (TPM)
Type of HH
1 -vehicle HH
2-vehicle HH
3-vehicle HH
allHH
Number
ofHH
136
259
84
479
AVG(TPM1)
15256
23516
27589
21885
AVG(TPM2)
0
14756
15156
14854
AVG(TPM3)
0
0
6599
6599
AVG(TPM)
15256
19136
16448
17806
AVG(TPMj) average TPM in the jth vehicle owned by the household (j=l, 2, 3)
AVG(TPM) average TPM in all vehicles owned by the household
Table 44 Total Annual Miles Driven per Vehicle (TVM)
Type of HH
1 -vehicle HH
2-vehicle HH
3-vehicle HH
allHH
number
ofHH
136
259
84
479
AVG(TVM1)
12055
16038
18976
15422
AVG(TVM2)
0
10182
11510
10507
AVG(TVM3)
0
0
5768
5768
AVG(TVM)
12055
13110
12085
12666
AVG(TVMj) average TVM in the jth vehicle owned
by the household(j=l, 2, 3)
AVG(TVM) average TVM in all vehicles owned by the household
The data in Tables 4.3 and 4.4 summarize how the vehicles are used. The first vehicle in
each type of household is driven more in households with more vehicles. The same relationship
33
-------
holds for the miles ridden and driven in the second fehicle between 2-vehicle and 3-vehicle
households. This implies that one reason for buying another vehicle is to use at least one of the
vehicles more intensively. However, the AVG (TPM) (average TPM per vehicle) and
AVG(TVM) (average TVM per vehicle) are similar for households with 1, 2 or 3 vehicles. In
other words, the total distance ridden and driven by a household is roughly proportional to the
number of vehicles owned. Nevertheless, the distribution of the TPM and TVM among the first,
second and third vehicles is not even. The first vehicle is always the vehicle ridden and driven
most by the family. This illustrates how important the survey data were for determining how to
allocate TPM and TVM to each vehicle in multi-vehicle households.
The annual average driving miles for all vehicles is 12,666 in our sample. This is slightly
smaller than the average miles per year used in the risk ladder (14,000 miles per year). The ratio
of AVG (TPM)/ AVG (TVM) is 1.4, implying that vehicles have a driver only for at least 60
percent of the miles driven.
Table 4.5 Household Characteristics
Type of
HH
[-vehicle
HH
2-vehicle
HH
3-vehicle
HH
allHH
Number of
HH
136
259
84
479
average
household
size
1.75
2.76
3.02
2.52
average
number
of adults
1.103
1.873
2.393
1.745
average number
of adult
equivalents
i
1.226
1.541
1.629
1.467
average
household
income ($)
46213.24
67432.43
87738. 1
64926.47
average income
per adult
equivalent EY($)
39225.59
46149.25
56813.37
46053.57
The demographic and income characteristics for each type of household are summarized
in Table 4.5. For households with more than one member, household income is converted to
income per adult equivalent using the standard weights adopted by the U.S. Bureau of the
Census. The equivalence scale is based on the official weighted average poverty thresholds for
1992 (Data Source: Bureau of the Census (1993: Table A)), following the Table 3-1 of Citro and
Michael (1995). The values of the equivalence scafes are 1,1.279,1.566,2.007, 2.323,2.679,
3.023, 3.367 and 4.024 for family size 1, 2, 3, 4, 5, 6, 7, 8 and 9 or more, respectively. Using this
34
-------
measure, the income per adult equivalent is over $39 thousand, $46 thousand and $57 thousand
for 1-vehicle, 2-vehicle and 3-vehicle households, respectively, and the overall average is $46
thousand.
4.4.3 Estimating How Vehicles Are Used
Given estimates of the hedonic models for the price of a vehicle and the fuel efficiency
presented in Appendix C, the final component of the VSL model in Equation (18) is to estimate
the mileage (TVM) and the occupancy (TPM) for each vehicle. These estimates are treated as the
expected levels of use of a vehicle when it is purchased, and are, therefore, the appropriate levels
to use when estimating the VSL for adults, seniors and kids. The summary of the survey data in
the previous section shows strong positive relationships between the number of vehicles owned
and household income and household size (see Table 4.5). In addition, the total mileage and
occupancy for a household are roughly proportional to the number of vehicles owned, and the
composition of a family is also a potential factor in determining how vehicles are used.
The first models of how vehicles are used by each household determine the mileage in all
vehicles (TVM), the total occupancy in all the vehicles (TPM), and the proportion of miles
traveled by kids. These variables are determined by the economic and demographic
characteristics of each household and the number of vehicles owned. The estimated equations
have the following form:
t
log
-------
Table 4.6 Parameter Estimates fqr Mileage and Occupancy
(1) Parameter Estimates for log(TPM)
Variable
Intercept
Seniorratio
InEY
InT
InN
D
Dratio
Kinverse
Square(Kinverse)
Parameter
Estimate
6.53983
-0.54969
0.25296
0.27909
1.00904
-0.11621
1.05343
3.13611
-4.11556
t Value
12.48
-1.94
5.16
2.58
9.43
-0.38
2.48
3.08
-2.06
(2) Parameter Estimates
Variable
Intercept
Siniorratio
In^Y
InT
InN
D
Dratio
Parameter
Estimate
6.5161
-0.46013
0.25614
0.06833
1.03166
-0.28313
1.07935
for log(TVM)
t Value
12.63
-1.66
5.31
0.87
10.36
-0.94
2.57
Table 4.7 Parameter Estimates Ijor Occupancy by Kids
Parameter Estimates for log(KM/(TPM-KM))
Parameter
Variable Estimate
Intercept -1.43278
Inkidnonkidratio 0.8454
InEY 0.52692
InN 0.04631
t Value
-1.08
6.22
1.58
0.19
where: Seniorratio=(seniors)/(total family size)
lnEY=log(Average equivalent income)
lnT=log(number of household members)
lnN=log(number of vehicles owned)
D: Dummy variable for a senior household
(D=l if the household is a senior household)
Dralio=D/(the age of the oldest person in the household"64)
Kinverse=l/(l+the age of the youngest kid in a household)
if the household has at least 1 kid
Square(Kinverse)= Square of Kinverse
lnkidratio=log((K/(T-K)), and K is the number of kids
f
36
-------
t
The estimates in Table 4.6 show that both TVM and TPM are almost proportional to the numbers
of vehicles (N), as expected. In contrast, the effect of household size is much smaller,
particularly on TVM. Given a high enough income, most adults (and seniors) would like to have
their own vehicle, and total mileage is proportional to the number of vehicles. The effect of
income is inelastic, but it is clearly statistically significant.
The effects of the composition of a household require some further explanation. For total
occupancy, (TPM), the positive coefficient for Kinverse (3.13611) and the negative coefficient
for Square(Kinverse) (-4.11556) implies that the TPM for kids increases until the youngest kid is
2 and then decreases. The survey data indicate that average KM decreases with age. Since the
TPM for "young seniors" should not drop a lot compared to adults, and the TPM for "old
seniors" drops dramatically in the survey data, we include three variables: Seniorratio, D and
Dratio. The coefficient for Seniorratio and D are both negative (-0.54969 and -0.11621) and the
coefficient for Dratio is relatively large and positive (1.05343). Hence, if the household is a
senior household (everyone is older than 60 and at least one member is a senior) and the oldest
person is only 65 years old, then the total senior effect will be (-0.54969-0.11621+1.05343) > 0.
In other words, for a senior household with young seniors, the TPM will be higher than it is in an
adult household. Nevertheless, when a senior household is composed of "old seniors", Dratio
will decrease and the TPM will also decrease as age increases. This is exactly the type of
behavior observed in the survey data. The TPM for seniors does not drop in one step. It drops at
ages above 65, slowly for "young seniors" and then more rapidly. One reason for the implied
increase in TPM at age 65 is that these people typically have more free time for travel and are
still healthy. The effects of seniors dominate the effect of household composition on TVM, and
the effects of kids were not significant. In general, older seniors have lower values of both TPM
and TVM, as expected.
Given predictions of TPM and TVM, it is necessary to allocate these values among
adults, seniors and kids. For all adult and all senior households, there are no problems with this
allocation. For mixed households with adults and seniors, the estimates in Table 3.6 imply that
SM = Exp (-0.56) AM = 0.57 AM (i.e. when Seniorratio = 1), and consequently, TPM can be
allocated between adults and seniors. For kids, a separate equation is estimated (see Table 4.7)
37
-------
for occupancy by kids. For a given household, KM increases with the number of kids and with
!
income, but has only a small positive relationship with the number of vehicles. For a given
income, the total mileage traveled by kids is not afffcted by the number of vehicles. For adult
households, however, the total mileage traveled is twice as large in a two-vehicle household than
a one-vehicle household.
The next step is to allocate TPM, TVM and KM to individual vehicles for households
that own more than one vehicle. When a household makes a choice about which vehicle to drive
given the estimated TVM, the main concerns affecting the choice are the vehicle's
characteristics, such as its size and level of safety. Hence, the explanatory variables for the
allocation will only reflect the vehicle's characteristics. We assume that the optimal vehicle
characteristics for each vehicle were chosen when tr^e vehicle was purchased. Hence the
vehicle's characteristics are predetermined explanatory variables for the observed data in the
survey.
For a 2-vehicle household, the dependent variables for allocating TPM, TVM and KM
between the first and the second vehicle are log odds ratios:
JnTPMratio2—log(TPM in the first vehicle/TPM in the second vehicle),
lnTVMratio2—log(TVM in the first vehicle/TVM in the second vehicle),
lnKMratio2 —log(KM in the first vehicle/KM in the second vehicle).
The explanatory variables are:
InVehicleagel—log(the model year of the first vehicle),
lnVehicleage2—log(the model year of the second vehicle).
The least square estimates are presented in Table 4.8.
Table 48 Parameter Estimates for Allocating TPM. TVM and KM for a 2-Vehicle Household
Variable
Intercept
InVehicleagel
InVehicleageZ
TPM
Parameter
Estimate t ratio
367.06428 1.21
82.48443 2.31
-130.71572 -5.21
TVM
Parameter
Estimate t ratio
493.4^121 2.01
25.33822 0.88
-90.21!446 -4.46
KM
Parameter
Estimate t ratio
1114.50482 0.82
359.34027 2.38
-506.03159 -4.87
t
38
-------
I
In all three models, the coefficients have the expected signs. The proportions of TPM,
TVM and KM driven in the first vehicle are higher in newer vehicles and lower if the second
vehicle is newer. Given these predicted proportions, it is possible to determine TPM, TVM and
KM in the first vehicle and second vehicle respectively using the observed model year of the
vehicles owned by each 2-vehicle household. In other words, we can get TPM t , TVM ; and
KM; for a 2-vehicle household. The corresponding values of SM ; are determined by the
following rule for households with adults and seniors.
SM j = [0.57*(TPM j -KM , )*total seniors numberl/(0.57*total seniors + total adults) If
the seniors are in a senior household, then SM ; =TPM y . Note that the difference in safety
between the two vehicles was not statistically significant in these models, and this variable is not
reported in Table 4.8.
For a 3-vehicle household, the final model allocated the miles between the first vehicle
and the other two vehicles. TPM ; , TVM ; and KM ; for the first vehicle. Efforts to model the
allocation between the second and third vehicle were not successful.
lnTPMratio3— log(TPM in the first vehicle/TPM in the other two vehicles),
lnTVMratio3 — log(TVM in the first vehicle/TVM in the other two vehicles),
InKMratioS — log(KM in the first vehicle/KM in the other two vehicles).
The logarithm of model year of each vehicle are explanatory variables and a new
variable, which is Inriskratio — log(minimum risk rate among the three vehicles/the risk rate of
the first vehicle). This new variable measures the relative risk of the first vehicle when one of the
other vehicles is safer (ratio < 1). A bigger ratio implies a safer first vehicle. Therefore, positive
coefficients for Inriskratio are expected. The least square estimates of the models are given in
Table 4.9.
t
39
-------
Table 4.9 Parameter Estimates fur Allocating TPM and |KM to the First Vehicle in a 3- Vehicle Household
Variable
Intercept
InVehicleagel
InVehicleage2
InVehicleageS
Inriskratio
TPM
Parameter
Estimate t ratio
557.82765 1.72
43.30447 1.16
-79.85588 -3.65
-36.84872 -1.92
0.38974 1.97
1
TVM
Parameter
i Estimate t ratio
279.86044 0.91
70.2771 1.99
:-80.21175 -3.87
-26.92953 -1.49
'0.31179 1.67
KM
Parameter
Estimate t ratio
552.07303 0.31
606.17145 2.62
-557.17158 -2.96
-121.61599 -0.9
3.1269 2.71
I
The positive coefficients for the model year of the first vehicle and the negative coefficients for
the other two vehicles are consistent with our expectations. In addition, the risk coefficients have
the expected positive sign. This model allocates TPM, TVM and KM between the first vehicle
and the other two vehicles.
There are no formal models for explaining the allocation of miles between the second and
third vehicles. From the Tables 4.3 and 4.4, the average TPM in the third vehicle is
approximately 44% of that in the second vehicle, arid the average TVM in the third vehicle is
approximately 50% of that in the second vehicle. Therefore, the following rule is used to allocate
the miles between the second and third vehicles:
TPM in the third vehicle =0.44*TPM in the second vehicle
TVM in the third vehicle =0.5*TVM in the second vehicle
KM in the third vehicle = 0.44*KM in the second vehicle
Finally, the same rule for determining the allocation between AM and SM described for
two vehicle households is used for households with both adults and seniors, and three vehicles.
Combining all of the vehicles in this section gives estimates of the mileage traveled by adults,
seniors and kids in each vehicle, and TVM for each vehicle.
40
-------
t
0
t
Section 5
Conclusions: Estimates of Average YSL By Group and Income Level
Three typical family groups own most of the total 783 vehicles:
1) PA: pure adults family (424 vehicles);
2) AK: family with both kids and adults (267 vehicles);
3) PS: pure senior family (57 vehicles).
To address possible income effects on the VSL, we divide each type of family into three
types according to per capita income, low income, middle income and high income. Specifically,
income type is defined as:
Low income family: Per Capita Income<=$15000;
Middle income family: $15000$37500.
Three no intercept OLS regressions were run, one for each of three family groups. (If we
run regressions with intercepts, the intercepts are insignificant). The estimated results without
intercepts are shown in the following table.
41
-------
Table 5.1 Estimated VSL for Families
Family
Type
PA
Income
Type
Low
Middle
Nigh
AK
^ow
Middle
High
PS Low
Middle
High
Sample
Size
67
188
169
133
120
14
9
31
17
Person Type*
Adult low
Adult middle
Adult high
Adult low
Kid low
Adult middle
Kid middle
Adult high
Kid high ;
Senior low
Senior middle
Senior high
VSL
(million)
6.81
6.07
7.27
3.36
2.54
3.79
5.12
-
-
7.67
8.42
8.25
; value
9.37
13.63
14.88
3.36
?.64
8.96
6.46
-
-
4.60
5.85
3.35
Note:
1. *Person Type is Defined as:
Adult low: adults from low-income families;
Adult middle: adults from middle-income families;
Adult high: adults from high-income families;
Kid low: kids from low-income families;
Kid middle: kids from middle-income families;
Kid high: kids from high-income families;
Senior low: seniors from low-income families;
Senior middle: seniors from middle-income families;
Senior high: seniors from high-income families.
2. * means insufficient sample size to obtain reliable estimates.
t
Since the average ages for adults, seniors and kids in our data set are 39.8, 74.2 and 7.8
respectively, the VSL for each group can be interpreted as the VSL for that group at the
average group age.
The estimated results are inconsistent with discounted present value of life-year model.
Seniors are more valuable than adults in all families for any of the income levels. For families
with both adults and kids, the VSL of kids is higher than that of adults in the middle-income
family, but lower in low income families.
An alternative procedure is to estimate VSL in a pooled model that assumes identical
VSLs across family types by age group.
From our theoretical model, the VSL can be estimated according to equation:
-[P'(r)+F'(r)]*TVM = [VSL(A)*AM+VSL(K)*KM+VSL,(S)*SM]+(y-y) *TP
42
-------
s
Using this approach, (which also allows calculation of income elasticities) the VSL for different
age groups for individuals of average income and driving miles is shown in the following table.
Table 5.2 Income Elasticity Estimates
Family
Type
PA
AK
AK
PS
Sample
Size
424
267
267
57
Per Capita
Income
40776
18709
18709
26462
Person
Adult
Adult
Kid
Senior
VSL*
(million)
6.67
3.59
3.64
8.18
t value
22.28
12.25
6.80
8.97
BEY
18.19
0.62
65.08
7.97
lvalue
2.05
-0.02
1.14
0.14
elasticity
0.111
-0.003
0.335
0.026
0
Comparing the estimated VSL between tables 5.2 and 5.1, the difference is not
surprising. The VSL estimated here (pool model) is for people from a standardized household
with average driving miles, occupancy and income. The VSL estimated previously (three group
model) refers to people with average income only and differing average driving miles and
occupancy.
The average estimated income elasticity for the VSL for each group is:
£jncome(forkids)=0.13,and
These results provide estimates of income elasticity for each age group that is smaller
than Blomquist's estimates of about 0.3.
The estimated VSL from both the group and pooled model show that seniors have the
highest value among all age groups, given the same income. Moreover, the relative value of
kids' VSL compared to adults depends on income class in the first model and is slightly
higher in the second estimate than adults' VSL. The overall pattern is somewhat inconsistent
with the discounted present value of life-year model, which suggests that VSL at age t is
equal to the value of a life-year times the discounted present value of remaining years of life
at age t. Because the average ages for adults, seniors and kids in our data set is 39.8, 74.2 and
7.8 respectively, we would expect the VSL for kids is to be somewhat lower than for adults.
Similarly, the VSL for seniors should also be lower than for adults according to discounted
present value of life-year model. However, since the VSL for kids depends on parents' tastes
43
-------
and preferences and may not be stationary over the life cycle, the discounted present value of
life year model may well be misleading.
However, it should be noted that the analysis so far has omitted an important effect
that has not previously been considered, fragility. Fipr estimating the hedonic models, we use
a standardized or inherent automobile risk ladder fo • each occupant in the vehicle that
removes the effects of drivers' and occupants' characteristics on the risk. In other words, we
assume that each occupant in the same vehicle has the same risk rate. However, seniors are,
on average, more fragile than adults and kids are, on average, less fragile than adults. The
effect of perceived fragility is that seniors will regard themselves more risky than adults in
the same vehicle and will be induced to buy a less risky (more expensive) vehicle even if
their VSL is the identical. Therefore, the fragility unadjusted VSL estimates obtained above
may well over-estimate the actual VSL for seniors. The same logic implies that the fragility
unadjusted VSL obtained above underestimates the actual VSL for kids since they are less
fragile than adults in accidents (except for infants), If we express fragility unadjusted VSL as
VSL1 and fragility adjusted VSL as VSL2, then the following relationship holds between
VSL1 and VSL2 for people from PA and PS households:
VSL2(a)PA=VSLl (a)PA*r/r(a)
VSL2(s)PS=VSLl(s)PS*r/r(s)
Where r is the average driver value used in Ihe standardized automobile risk ladder we
used in hedonic models, r(a) is the risk for adults and r(s) is the risk for seniors, if we assume an
adult at average age is an average driver, then r is equal to r(a). Hence, VSL2(a)PA equals
VSLl(a)PA, i.e. the fragility adjusted VSL for adults from PA household is the same as the
fragility unadjusted value. From the survey data, people's perception of the likelihood of a 70-
year-old person dying compared to an average adult when involved in a serious accident is about
39% higher. For households earning the average income, VSLl(a)PA and VSLl(s)PS are 6.62
and 8.44 respectively. If we consider the fragility effect on VSL, fragility adjusted VSL (VSL2)
of seniors from pure senior households will be less than that of adults from pure adults
households by 8.3%. Because the fragility adjusted VSL of adults from PA household,
t
t
44
-------
s
t
VSL2(a)PA, is 6.62, the fragility adjusted VSL of seniors from PS household, VSL2(s)PS, is
6.07.
For the AK family, it is more complicated to adjust VSL by fragility because the
household's real risk is an appropriate weighed average risk with both kids and adults. To
simplify problem, we still assume VSL2(k)AK=VSLl(k)AK*r/r(k). For children, the survey data
shows that the perception of the likelihood of a 8-year-old child dying compared to an average
adult when involved in a serious accident is about 12% lower. VSLl(k)AK is 3.63, therefore the
fragility adjusted VSL for kids, VSL2(k)AK, is 4.13.
The fragility adjusted VSL shows that, for the average household in the sample, kids are
more valuable than adults in a family with both kids and adults. We compare the VSL for seniors
from pure senior family with adults from pure adults family in order to remove the overestimated
income effect from adults in AK family. Seniors' VSL is 6.07 which is less than adults' VSL
6.62. Thus, our results are now much more consistent with the simple discounted present value
of life years approach when we include the effect of fragility on the VSL. However, parents have
a relatively low VSL which may simply reflect imperfect capital markets and the cost of
children, factors not considered in the discounted present value of life years approach.
Table 5.3 lists both the fragility unadjusted and adjusted VSLs for people from different
family groups.
Table 5.3 Fragility Adjusted VSL (Smillion) by Family Group
Age Group
Kids(AK)
Adults(AK)
Adults(PA)
Seniors(PS)
Fragility
Unadjusted VSL
3.63
3.72
6.62
8.44
Fragility
Adjusted VSL
4.13
3.72
6.62
6.07
t
Similarly we can adjust for fragility in estimating the VSL using the pooled model. The
fragility-adjusted VSLs obtained from the pooled model are consistent with the discounted
present value of life years model.
45
-------
References
t
1. Agee, M.D... and T.D. Crocker. (1996). "Parental Altruism and Child Lead Exposure:
Inferences from the Demand for Chelation Therapy." The Journal of Human Resources
31:677-691.
2. Atkinson, Scott E and Robert Halvorsen! (1990). "The Valuation of Risks to Life:
Evidence from the Market for Automobiles." The Review of Economics and Statistics 72
(1): 133-136.
3. Becker, G. S. (1974) "A Theory of Social jInteractions," Journal of Political Economy,
82,1095-1117.
4. Becker, G. S. (1991) A Treatise on the Family, Harvard University Press, enlarged edn.
5. Bergstrom, T. C. (1996) "Economics in a Family Way," Journal of Economic Literature,
34(4), 1903-1934.
6. Blomquist, Glenn C., (1979). "Value of Life Saving: Implications of Consumption
Activity," The Journal of Political Economy 87(3): 540-558.
7. Blomquist, Glenn C., David Levy and Ted R. Miller (1996) "Values of Risk Reduction
Implied by Motorist Use of Protection Equipment: New Evidence from Different
Populations," Journal of Transport Economics and Policy 30 (January 1996): 55-66.
8. Chestnut, L. and W. Schulze. (1998). "Valuing the Long Term Health Risks From
Wartime Toxic Exposures," Proceedings of the First International Conference on
Addressing Environmental Consequences of War.
9. Citro, Constance F. and Robert T. Michael. |( 1995), Measuring Poverty: a New Approach.
National Academy Press.
10. Cropper, M.L. and A.M. Freeman III. (1991). "Environmental Health Effects."
11. Measuring the Demand for Environmental! Quality. J.B. Braden and C.D. Kolstad (ed.)
North-Holland. New York.
12. Dillman, Don A, Mail and Internet Surveys : The Tailored Design Method, John Wiley
& Sons; 2nd edition (November 1999).
13. Dreyfus, Mark K. and W. Kip Viscusi (1995). "Rates of Time Preference and Consumer
Valuations of Automobile Safety and Fuel Efficiency." Journal of Law and Economics,
volXXXVIIl (April 1995): 79-105.
t
t
46
-------
s
0
14. Greene, William H. (1997). Econometric Analysis. 3rd Edition. Prentice-Hall, Inc.
15. Harrison, D. and A.L. Nichols. (1990). Benefits of the 1989 Air Quality Management
Plan for the South Coast Air Basin: A Reassessment. Prepared for California Council for
Environmental and Economic Balance by National Economic Research Associates, Inc.,
Cambridge, Massachusetts.
16. Jenkins, Robin, Nicole Owens, and Lanelle Bembenek Wiggins (1999) "The Value of a
Statistical Child's Life: The Case of Bicycle Helmets." USEPA Office of Economy and
Environment.
17. Jones-Lee, M.W., M. Hammerton, and P.R. Philips. (1985). "The Value of Safety:
Results of a National Sample Survey." The Economic Journal 95(March):49-72.
18. Joyce, T.J., M. Grossman, and F. Goldman. (1989). "An Assessment of the Benefits of
Air Pollution Control: The Case of Infant Health." Journal of Urban Economics 25:32-51.
19. McElroy, M. and M. Homey (1981) "Nash-bargained decisions: Toward a Generalization
of the Theory of Demand/' International Economic Review 22:333-349.
20. Miller, T.R. (1989). "Willingness to Pay Comes of Age: Will the System Survive?"
Northwestern University Law Review 83:876-907.
21. Moore, M. J. and W. K. Viscusi (1988). "The Quantity-Adjusted Value of Life."
Economic Inquiry 26:369-388.
22. NADA official used car guide (1995) Vol. 62. McLean, Va., National Automobile
Dealers Used Car Guide Co.
23. Powell, James L. (1986). "Symmetrically Trimmed Least Square Estimation for Tobit
Models." Econometnca 54 (6): 1435-1460.
24. Rosen, Sherwin (1974). "Hedonic Prices and Implicit Markets: Product Differentiation in
Pure Competition." Journal of Political Economy 82 (1): 34-55.
25. Shepard, D.S. and R.J. Zeckhauser. (1982). "Life-Cycle Consumption and Willingness to
Pay for Increased Survival." in M.W. Jones-Lee (ed.). The Value of Life and Safety. New
York: North-Holland.
26. U.S. Bureau of the Census (1993). Current Population Reports, Series P60, No. 185,
Poverty in the United States: 1992. U.S. Department of Commerce, Washington, D.C.
27. Ward's Automotive Yearbook (1990-1996). Vol. 52-58.
47
-------
Appendix A
The Statistical Framework for Modeling Automobile Fatalities
The basic data on fatalities in automobile accidents provide a census of accidents with at
least one fatality. Hence, the probability of an accident being included in the data set depends on
the number of individuals involved in an accident as well as the characteristics of the vehicles
and driving behavior (e.g. the use of seat belts). This can be illustrated by the following
examples for a one-vehicle and a two-vehicle accident For a one-vehicle accident, assume that
the driver and one passenger have the same probability of survival P" = /'{survival} = .5. The
four possible events are illustrated below, and in thi$ example, each event has the same
probability of occurring of 0.52 = .25.
s
Passenger
Fatality
Survives
Driver
Fatality
Survives
Accidents in which both the driver and the passenger survive (shaded) are not included in the
data set. Hence, the probability of either the driver or the passenger surviving in an accident
with a fatality corresponds to the probability of one of three possible events with a probability of
P = P{survival | at least one fatality} = 0.25 / (1 - 0,25) = 0.33. The observed probability of
survival in the data set, P, is much lower than the unconditional probability, P . The observed
probabilities of survival, P, are 0, 0.33,0.43 and 0.47 for 1, 2, 3 and 4 occupants, respectively,
and the values ofP increase and get closer to P as the number of occupants increases.
In the one-vehicle accident with two occupants and P' = 0.5, the expected number of
fatalities is one (the modal type, corresponding to 91 % of one-vehicle accidents in the data set).
In a two-vehicle accident with two occupants in each vehicle, the same expected number of
fatalities would occur if P* = 0.25 (for multiple-vehicle accidents, 54% of vehicles have no
fatalities, and 40% have one fatality). The probability of an accident having at least one fatality,
and being in the data set, is (1 - 0.754) = 0.68. There are 16 possible permutations of survival /
fatality for the four individuals and 15 of them are in the data set. For any selected individual, 7
of the 15 observed events correspond to surviving with a probability P = 0.63. While this is
lower than the unconditional probability of survival P* = 0.75, it is much larger than the
corresponding probability for the one-vehicle accident P = 0.33. Setting the severity of the two
types of accident at the same level (E [number of fatalities] = 1) makes the probability of a
specific individual surviving in a fatal accident almlost twice as large in the two-vehicle accident
as in the one-vehicle accident. The reason is simple, for any unconditional probability of
t
48
-------
s
survival P* , the expected number of fatalities is P* x number of individuals in the accident.
Since the data set includes all accidents in which at least one fatality occurs, a fatality is more
likely to occur if more people are involved.
In reality, the unconditional probabilities of survival for individuals differ by individual
characteristics such as age, whether or not a seat belt was used and the location of the seat in a
vehicle. In addition, these probabilities differ by the type of vehicle, and for two-vehicle
accidents by the relative size and type of the other vehicle. For an individual / riding in vehicle/,
the unconditional probability of survival in a two-vehicle accident, for example, can be written:
where x, are the characteristics of individual i
vn are the characteristics of individual i's vehicle (/'=!)
v,2 are the characteristics of the other vehicle (/' = 2)
ztj is the vector of all explanatory variables
The probability of observing at least one fatality in the accident is
where «,- is the number of individuals in vehicle;.
If P*. - /(zy ) is specified as a logistic function, then it can be written:
where P is a vector of unknown parameters that are the same for all individuals and vehicles.
Using this form, it would be possible to recover the unconditional probabilities of survival using
the available data on accidents with at least one fatality. In the simplest case with one individual
in each vehicle, for example, the probability of observing two fatalities in the data set would be:
1
and the unconditional probability of two fatalities would be:
1
The unconditional probability of the individual in vehicle 1 surviving would be:
PI, = ;—;— = ;—
i ZttS 2\tB (*M1 "*" "\1 jf* 1 i a^\\r
I i /? 11~ _|_ ff ••** _l_ f>^ '*/' | -f- g '**
An equivalent expression for P*i2 can be derived in exactly the same way. Since P could
be estimated from the available data on fatal accidents, the unconditional probabilities of survival
could be calculated.
The parameters in p can be estimated by maximum likelihood, estimation. The
likelihood function for the probability of survival in two-vehicle accidents, for example, can be
specified as:
49
-------
_ .
n -
U
t
where K = 1 , . . . , m, number of accidents;
njkis the number of individuals in vehicle./, accident k;
Yjjk = 1 if individual i survived, else (p.
The basic structure of the model of the risk of having a fatality in an accident is to
distinguish between one-vehicle, two-vehicle and multiple-vehicle accidents. The expectation is
that the characteristics of drivers contribute more to the probability of having a one-vehicle
accident than to a two- or multiple-vehicle accident. On the other hand, vehicle characteristics,
particularly the weight relative to the weight of the other vehicle, will affect the survival rate in
two-vehicle accidents but may be less important for one-vehicle accidents. In addition, the
earlier discussion of why survival rates are likely to i differ systematically between one-vehicle
and two-vehicle accidents provides another reason for modeling one-vehicle and two-vehicle
accidents separately. The justification for separating multiple- vehicle accidents from two-
vehicle accidents is that it is impossible to identify the "other" vehicle from the data for multiple-
vehicle accidents.
If r is the overall fatality rate, then the model's components can be written as follows:
where r is the annual fatality rate per occupant;
P {VI} is the probability of having a one-vehicle accident per 10,000 miles;
P{V2) is the probability of having a two-vehicle accident per 10,000 miles;
P{Vm) is the probability of having a multiple (three or more) vehicle accident per 10,000
miles;
PI* is the probability of surviving in a one-\iehicle accident;
PI* is the probability of surviving in a two-vehicle accident;
PS* is the probability of surviving in a multiple-vehicle (three or more) accident;
M is the average annual mileage traveled (13.989 miles from the NPTS).
The units for r, P{V\] , P{V2} and /'{Pmjare all standardized to measure the probability of
having a fatal accident per 1000 vehicles.
Conceptually, all six components of the observed values of r may be functions of the
characteristics of the driver (and the passengers) and the vehicle driven (and the other vehicle for
two-vehicle accidents). For computing a hedonic price index, the characteristics of an average
driver and passenger are used to predict r for different types of vehicle (make, model and year),
and each type of vehicle is assumed to have an accident with a typical other vehicle in a two-
vehicle accident. Hence, the effects of drivers' characteristics are removed prior to estimating the
hedonic price equation. The effect of standardizing} the other vehicle in a two-vehicle accident is
relatively small because the observed combinations of vehicles in two-vehicle accidents are
approximately random. Standardizing drivers' characteristics, however, matters a lot for the
probabilities of being in a fatal accident. It is the primary reason for the difference between our
estimated value of a statistical life compared to a conventional model in which drivers'
characteristics are added as additional regressors in the hedonic price equation.
The structure of the equations for the six components of r can be written as follows:
50
t
-------
t
where V\ are the characteristics of a selected vehicle.
D\ are the average driver's characteristics for the selected vehicle and include factors
such as the use of seat belts and whether alcohol was a factor.
O\ are the characteristics of the occupants of the selected vehicle, including the
driver.
F2 are the characteristics of the other vehicle, its weight relative to the weight of the
selected vehicle being the most important
Since all six dependent variables are probabilities, appropriate statistical models for
limited dependent variables are used. PI*, ?2* are specified as logistic functions and estimated by
maximum likelihood in GAUSS. For Pm , we assume the unconditional probability Pm is the
same as the observed probability Pm, and Pm is specified as a regular logit model and estimated in
SAS. P\V\], P{V2\ and P{Vm} are determined by a censored regression model to allow for a
probability mass at zero. Note that Pm is determined by the characteristics of the own-vehicle
only because it is not possible to identify the "other" vehicle in a multiple-car accident.
The complete econometric analysis for determining the fatality rate, r, for a specified type of
vehicle, consists of the following three steps:
Step 1. Augment the PARS data on observed fatal accidents with additional characteristics
about the vehicles (e.g. weight and safety features), and use these data to estimate equations for
the unconditional probabilities of survival in one-vehicle, two-vehicle and multiple-vehicle
accidents (Pi*, P-i and Pm)- Derive the estimated numbers of serious accidents (including
accidents with no fatalities) for one-vehicle, two-vehicle and multiple-vehicle accidents.
Step 2. Calculate the average drivers' characteristics in fatal accidents from the PARS data by
make, model and year of the vehicle driven, and combine with survey data on the composition of
the fleet of vehicles. Use these data to estimate equations for the probabilities of having one-
vehicle, two-vehicle and multiple-vehicle accidents by the make, model and year of vehicle
}, P{V2}andP{Vm}).
51
-------
Step 3. Use the average drivers' characteristics frofn the PARS data, and the average other
vehicle in two-vehicle accidents, to standardize the unconditional probability of a driver and/or
passenger being killed in a fatal accident by make, nfodel and year of the vehicle.
t
52
-------
Appendix B
The Estimated Models Used to Determine the Fatality Rates for Different Types of Vehicle
The inherent fatality rate for an individual in a vehicle can be decomposed as follows:
where r is the annual fatality rate per capita;
P{V1} is the probability of having a one-vehicle accident per 10000 miles;
P{V2} is the probability of having a two vehicle accident per 10000 miles;
P{Vm) is the probability of having a multiple (three or more) vehicle accident per 10000
miles;
PI* is the probability of surviving in a one-vehicle accident;
?2* is the probability of surviving in a two-vehicle accident;
Pm is the probability of surviving in a multiple-vehicle (three or more) accident;
M is the average annual mileage traveled (13989 miles from the NPTS).
The likelihood function for the probability of survival can be specified as:
One-car accidents:
L = fl—'-^—
Two-car accidents:
Multiple-car accidents:
where i = 1 , . . . , n, individuals;
j = 1 , . . . , m, vehicle;
k= 1, ..., K, accidents;
= 1 if survived, else 0.
I
Survival rates PI* , ?2* and Pm are specified as logit functions and estimated by
maximum likelihood in GAUSS using data from the Fatality Analysis Reporting System (PARS)
augmented with additional data about vehicle characteristics (step 1 in Appendix A). The
53
-------
explanatory variables are summarized in Table Bl, find the estimated equations are shown in
Tables B2 , B3 and B4, respectively. For the survivs 1 rate in a one-vehicle accident, the effect of
using a restraint (seat belt or car seat) is very important and clearly positive, but the effect of an
airbag was not significant. The number of occupanljs is significant, but without a clear
explanation. The survival rate is relatively high in pickup trucks (Class 7). A very inexperienced
driver, 16 years or younger, has a strong negative effect on the survival rate.
The equation for P2* in Table B3 implies that the weight ratio is the most important
explanatory variable. Being in a larger vehicle increases the chance of survival and visa-versa.
Weight also has a positive effect on survival in multiple car accidents (see Table B4). The
number of occupants is also important. The positive; effect of using a restraint (seat belt or car
seat) is substantially larger than the effect of airbags. In general, the effects of the class of
vehicle are consistent with the effect of the weight ntio. Seating in a small vehicle (Classl)
reduces the probability of survival, while hitting a small vehicle increases the probability of
survival.
The equation for P{V1}. P{V2} and P{Vm} are specified as censored regression models
to allow for a point mass at zero (18% and 27% of the vehicle types having no recorded fatalities
for one-vehicle and two- vehicle accidents, respectively). This specification worked much better
than a linear probability model. The data used corresponds to observations of make, model and
year augmented by average driving characteristics from the PARS. Since the observed
probabilities pf having a fatal accident per 1000 vehicles are very small, it was unnecessary to
impose an explicit upper limit of one on the dependent variable. The equations were estimated
in SAS.
In order to be consistent with the unconditional probability of survival, each fatal
accident is scaled by the inverse of the probability of observing the accident, i.e. at least one
fatality occurred. The scaling is very easy for one-vehicle accidents. But for two-vehicle
accidents, we need to know the characteristics, e.g. weight, of both vehicles. Among the 25126
two-vehicle accidents that occurred in 1995-1997 involving at least one of the vehicles we
studied, there are 8282 accidents having complete information for both vehicles' characteristics.
Thus, only one-third of the accidents have complete information about both vehicles'
characteristiQs. There are two possible solutions: oAe is to find out the complete information of
the other vehicle, the other is to scale the accidents with unknown characteristics of the other
vehicle by the same scalar used to scale accidents with both vehicles' characteristics known. If
the pattern of hitting the other vehicle is the same for each make/model/year vehicle whether the
characteristics of the other vehicle is known or not, then the second way is a reasonable
approximation.
A goodness-of-fit test is used to test whether the pattern of accidents is the same or not.
The probability of having a two-vehicle accident is calculated by each make/model/year, but due
to the limited number of observations, accidents for each make/model/year were aggregated to
23 types of vehicle. The overall x2 test is rejected, but when we only consider the first 21 types,
the x2 test cahnot be rejected. The remaining two tipes are small and large pick-up trucks. After
comparing the distribution of the other vehicles hit py the 21 types, and by small and large pick-
ups, pick-up trucks were found to hit a high proportion of old vehicles, whose characteristics are
not included in this study. Since old and new vehicles are similar in weight, and the age of
vehicle isn't a significant factor determining the probability of survival, all accidents by
I
54
-------
make/model/year were inflated by the scalar derived from the subset with complete vehicle
characteristics.
The remaining part of the fatality rate model is to estimate the probabilities for multiple
vehicle accidents P {Vm} and Pra. Unlike two-vehicle accidents, the pattern of collision is very
hard to identify in multiple-vehicle accidents. Some of the vehicles may have no direct impact on
each other. Therefore, the model for Pm is more like the model for a one-vehicle accident, i.e. no
information of the other vehicles is included. In addition, we assume that all multiple-vehicle
accidents are observed. Since the total number of vehicle occupants involved in a multiple-
vehicle accident could be quite large (at least 3), this is a reasonable approximation. Also, the
fatalities in multiple-vehicle accidents are only 8.5% of the total fatalities that occurred in 1995-
1997. The equation for the survival rate Pm is specified as a regular logit model and estimated by
maximum likelihood in SAS. The equation for P{Vm} is specified as a censored regression
model to allow for a point mass at zero (24% of the vehicle type) and estimated by SAS.
Explanatory variables in the censored models for PI, P2> Pm that are not listed in Table Bl
are described in Table B5. The basic differences are that additional subdivisions of the classes of
vehicles are made, for example, to identify sports cars from non-sports cars for one-vehicle
accidents. In addition, variables such as styling ((length plus width/height) are included to
provide more information about the type of vehicle.
Before estimating the censored regression of P{V1}, P{V2}and P{Vm), 12 of the total of
1261 vehicle types were dropped because they had sales less than 500 vehicles. With a very
small number of vehicles on the road, even one fatal accident for that make/model/year will
count as a big probability of having a fatal accident. The increase in the number of subclasses of
vehicle for P (VI} was prompted by inspection of the raw data. The effects of variables such as
alcohol and previous convictions are partly responsible for the high rates of accidents for some
types of vehicles. For P (VI}, P {V2} and P{Vm}, shown in Tables B6, B7 and B8, the accident
rate increases for young drivers, for older drivers and, surprisingly, for female drivers, Accidents
are more likely to occur at highway speeds (Sp), and for all three types of accidents, powerful
vehicles (Acceleration) are more likely to have accidents, especially for one-vehicle accidents.
The use of alcohol and previous convictions increases P{V1}, P{V2} and P{Vm}. The overall
conclusion is that driving behavior does matter and affects the probabilities of having a fatal
accident for different types of vehicle.
t
55
-------
Table B.1: Variable Definitions for Estimating the Probability of Survival
VARIABLE
NAME
Restraint
AgeO_5
Age 15
Age21
Age24
Age_o
female
Occupants
Number
ClassX
Weight
Weight; Ratio
Acceleration
Vehicle Age
O_classX
Female Driver
Driver 16
Young'Driver
Older Driver
Alcohol
Late N;ght
No Previous
Offenses
Sp_lim|it
Seatfp
Seatb
airbag
Definition
CODED AS 1 IF THE PASSENGER USED RESTRAINT, 0
OTHERWISE.
Coded as 1 if the passenger a|[e
Coded as 1 if the passenger a);e
Coded as 1 if the passenger aj;e
Coded as 1 if the passenger a|;e
Coded as 1 if the passenger a|;e
Coded as 1 if the passenger is female
logarithm of number of occup ants.
is <5,0 otherwise.
is >6 but <15, 0 otherwise.
is >16 but <21, 0 otherwise.
is >22 but <24, 0 otherwise.
is >65, 0 otherwise.
, 0 otherwise.
Discrete variables coded as 1 for the appropriate class. Classl to class?
represent small, middle, krge, luxury, SUV, van, and pick-up truck,
respectively, class40, clas:i41 represents luxury non-sports and luxury
sports, respectively. ;
Weight of the vehicle (lOOOltf).
Weight ratio of the vehicle to (the other vehicle in a two-vehicle accident.
Horsepower to weight ratio.
The age of the vehicle when t le accident happened.
The class code for the other vehicle.
Code as 1 if the driver is ferns le.
Code as 1 if the driver is < 16.
Coded as 1 if the driver is Sl<> but <24, 0 otherwise.
Coded as 1 if the driver is 65 or older.
Coded as 1 if the alcohol invc Ivement is reported
Code as 1 if the accident occi rred between 12:00am to 5:59am.
Code as 1 if the driver had no previous offenses.
Speed limit (10 miles).
Coded as 1 for front seat non-jdriver passenger.
Coded as 1 for back seat pass mger.
Coded as 1 for airbag in that iieat position.
t
56
-------
t
Table B2: The Probability of Survival in a One-Vehicle Accident
Parameters
Constant
Restraint
AgeO 5
Age 15
AgeZl
Age24
Age_o
female
Occupants Number
Weight
Acceleration
Vehicle Age
Class2
Class3
Class40
Class41
ClassS
Class6
Class?
Spjimit
airbag
Seatfp
Seatb
Driver 16
Young Driver
Older Driver
Female Driver
Alcohol
No Previous Offenses
Late Night
Estimates
1.485
1.0943
0.2011
0.6061
0.4501
0.3464
-1.0999
-0.2026
0.3961
-0.0737
-8.913
0.0025
-0.0018
-0.0028
-0.1111
0.2465
0.4951
0.2715
0.62
-0.0627
-0.0054
-0.0612
0.1249
-0.4153
-0.2345
0.5054
0.2059
-0.0607
-0.1916
-0.1014
t ratio
5.067
25.028
2.407
9.552
8.547
5.701
-10.49
-5.948
5.999
-1.176
-2.539
0.154
-0.022
-0.016
-0.739
0.924
3.601
1.992
5.024
-2.792
-0.109
-1.854
2.535
-4.046
-3.42
3.625
3.248
-0.965
-3.417
-1.672
Prob.
0
0
0.0161
0
0
0
0
0
0
0.2395
0.0111
0.8777
0.9821
0.9875
0.4602
0.3555
0.0003
0.0463
0
0.0052
0.9132
0.0637
0.0112
0.0001
0.0006
0.0003
0.0012
0.3347
0.0006
0.0945
57
-------
Table B3: The Probability of Survival in a Two-vehicle Accident
Parameters
Constant
Restraint
AgeO 5
Age 15
Age21
Age24
Age_o
female
Occupants Number
Weight
Weight ratio
Vehicle Age
Class!
Class3
Class40
Class41
ClassS
Class6
Class?
O_class2
O_class3
O_class4
O_class41
O_class5
0_class6
0_class7
Spjimit
airbag
Seatfp
Seatb
Driver 16
Young Driver
Older Driver
Female Driver
Alcohol
No Previous Offenses
Late Night
Acceleration
Estimates
2.0436
0.9234
0.0402
0.4342
0.4615
0.4571
-1.5055
-0.1744
0.2572
0.1566
1.3538
-0.0018
0.0684
0.0407
-0.28
-0.4624
0.4543
0.4554
0.6685
-0.1202
-0.0789
-0.289
-1.7245
-0.3271
-0.3155
-0.4214
-0.447
0.1316
-0.114
0.2585
-0.613
-0.0165
0.172
-0.0423
-0.606p>
-0.0599
-0.6222
-2.2826
t ratio
6.544
18.053
0.297
3.81
4.758
4.14
-13.25
-3.467
4.65
1.431
6.626
-1.02
0.735
0.254
-1.773
-1.136
2.837
3.184
5.06
-1.167
-0.47
-1.792
-5.119
-2.064
-2.067
-3.12
12.714
2.395
-1.306
2.418
-4
-0.177
1.445
-0.696
-7.997
-1.15
-6.02
-0.594
Prob.
0
0
0.7663
0.0001
0
0
0
0.0005
0
0.1523
0
0.3077
0.4624
0.7993
0.0762
0.2561
0.0046
0.0015
0
0.2431
0.6387
0.0732
0
0.039
0.0387
0.0018
0
0.0166
0.1915
0.0156
0.0001
0.8597
0.1484
0.4862
0
0.2501
0
0.5523
t
58
t
-------
Table B4: The Probability of Survival in a Multiple-vehicle Accident
Parameters
Constant
Restraint
AgeO 5
AgelS
Age21
Age24
Age_o
female
Occupants Number
Weight
Acceleration
Vehicle Age
Class2
Class3
Class40
Class41
ClassS
Class6
Class?
Sp_limit
airbag
Seatfp
Seatb
Driver 16
Young Driver
Older Driver
Female Driver
Alcohol
No Previous Offenses
Late Night
Estimates
0.0075
0.9570
-0,0337
0.2592
0.2673
0.4072
-1.5429
-0.1232
0.5399
0.3223
7.7231
-0.0072
0.2409
0.4135
0.2486
-0.0749
0.7535
0.6679
0.7384
-0.2297
0.1788
-0.0683
0.2892
-0.3106
-0.0581
0.2642
0.0268
-0.8750
0.0627
0.0192
Waldx2
0.001
437.001
0.066
5.312
6.274
10.560
177.689
4.486
100.273
35.925
4.251
0.216
11.420
11.050
3.611
0.042
33.524
36.716
52.197
116.802
8.380
1.050
8.380
2.337
0.354
4.784
0.199
108.279
1.844
0.045
Prob.
0.9774
0.0001
0.7978
0.0212
0.0123
0.0012
0.0001
0.0342
0.0001
0.0001
0.0392
0.6424
0.0007
0.0009
0.0574
0.8380
0.0001
0.0001
0.0001
0.0001
0.0038
0.3055
0.0038
0.1264
0.5518
0.0287
0.6557
0.0001
0.1745
0.8316
59
-------
Table B5: Variable Definitions for Estimating the Probability of Having an Accident
Variable Name
Definition
TypeXX Coded as 1 for the appropriate type. Typel to Type23 represent lower,
upper small, small specialty, lower, upper middle, middle specialty,
large, large specialty, lower, middle, upper luxury, luxury specialty,
luxury sport, small, middle, large, luxury suv, small, middle, large,
luxury van, small, large pickup, respectively.
Alcohol Proportion of accidents in this make/model/year vehicle in which the
alcohol involvement was reported.
No Previous Proportion of accidents in this make/model/year vehicle in which the
Offenses driver had no previous offense.
Late Night Proportion of accidents in this make/model/year vehicle which occurrec
between 12:00amto 5:59am.
Driver 16 Proportion of accidents in this make/model/year vehicle in which the
driver is 16 or younger.
Young Driver Proportion of accidents in this make/model/year vehicle in which the
driver is younger than 25 years, but older than 16..
Older Driver Proportion of accidents in this make/model/year vehicle in which the
driver is 65 or older.
Female Driver Proportion of accidents in this make/model/year vehicle in which the
driver was female.
Sp Proportion of accidents at highway speed.
Acceleration The horsepower-to-weight i atio.
Traditional Length plus width divided by height.
Styling
D_airt)ag Coded as 1 for the driver-side airbag.
P airbag Coded as 1 for the passenger-side airbag.
I
60
-------
Table B6: Censored Regression for the Probability of Having a One-Vehicle Accident
PARAMETER
Estimate std. Error ChiSquare
t
s
constant
Alcohol
No Previous Offenses
Late Night
Driver 16
Young Driver
Older Driver
Female Driver
Sp
Acceleration
Traditional Styling
Weight
D_airbag
P_airbag
Type2
Type3
Type4
TypeS
Type6
Type?
TypeS
Type9
Type 10
Type 11
Type 12
Type 13
Type 14
Type 15
Type 16
Typel?
Type 18
Type 19
Type20
Type21
Type22
Type23
Sigma
-0.2231
0.0925
-0.1255
-0.0056
0.1232
0.1814
0.1038
0.1018
0.1247
3.9825
0.0280
-0.0059
-0.0360
-0.0159
-0.0702
-0.0604
-0.0965
-0.0805
-0.0632
-0.0969
-0.0996
-0.0985
-0.1095
-0.1558
-0.0797
0.0725
0.1634
0.1194
0.0569
0.1454
-0.0232
-0.0491
0.0110
-0.0366
0.1359
0.0932
0.1526
0.097
0.019
0.017
0.022
0.048
0.022
0.028
0.019
0.017
0.654
0.025
0.018
0.012
0.013
0.023
0.029
0.027
0.028
0.031
0.038
0.054
0.034
0.033
0.041
0.043
0.038
0.044
0.045
0.060
0.057
0.040
0.054
0.062
0.065
0.032
0.052
0.003
5.25
22.69
51.53
0.07
6.69
69.42
13.44
28.28
53.40
37.08
1.24
0.11
8.74
1.42
9.11
4.21
12.94
8.17
4.23
6.44
3.40
8.45
10.70
14.47
3.40
3.64
13.98
7.07
0.91
6.58
0.33
0.82
0.03
0.31
18.32
3.23
61
-------
T>ble B7: Censored Regression for the Probability of Having a Two Vehicle Accident
parameter
constant
Alcohol
No Previous Offenses
Late Night
Driver 16
Young Driver
Older Driver
Female Driver
Sp
Acceleration
Traditional Styling
Weight
D_airbag
P_airbag
Type2
Type3
Type4
Type5
Type6
Type?
Type8
Type9
TypelO
Type 11
Type 12
Type 13
Type 14
Typel5
Typel6
Typel?
TypelS
Type 19
Type20
Type21
Type22
Type23
Sigma
Estimate
0.0197
0.0063
-0.1561
0.0448
0.1449
0.0980
0.1098
0.1102
0.1067
0.0858
0.0074
0.0261
-0.0119
-0.0062
-0.0258
-0.0377
-0.0630
-0.0476
-0.0549
-0.0577
-0.0887
-0.1069
-0.0972
-0.1146
-0.0909
-0.0939
-0.0438
-0.0274
-0.0196
-0.0679
-0.0469
-0.0627
-0.0193
-0.0590
0.0338
0.0218
0.0851
std. Error
0.061
0.023
0.015
0.024
0.051
0.018
0.019
0.014
0.014
0.428
0.015
0.010
0.007
0.008
0.013
0.017
0.015
0.016
0.019
0.022
0.031
0.020
0.021
0.026
0.027
0.026
0.026
0.026
0.035
0.034
0.023
0.031
0.036
0.036
0.018
0.030
0.002
Chi Square
0.11
0.08
109.11
3.54
7.94
30.26
32.47
61.16
58.72
0.04
0.23
6.22
2.88
0.61
3.84
4.75
16.87
8.36
8.73
6.68
8.28
28.49
22.06
19.86
11.46
13.39
2.75
1.09
0.31
3.98
4.03
3.98
0.29
2.64
3.35
0.52
t
62
-------
Table B8: Censored Regression for the Probability of Having a Multiple (three or more) Vehicle Accident
parameter
Estimate
std. Error ChiSquare
f
constant
Alcohol
No Previous Offenses
Late Night
Driver 16
Young Driver
Older Driver
Female Driver
Sp
Acceleration
Traditional Styling
Weight
D_airbag
P_airbag
Type2
Type3
Type4
Type5
Type6
Type?
TypeS
Type9
Type 10
Type 11
Type 12
Type 13
Type 14
Type 15
Type 16
Type 17
Type 18
Type 19
Type20
Type21
Type22
Type23
Sigma
-0.0067
0.0051
-0.0192
0.0118
0.0291
0.0130
0.0180
0.0213
0.0214
0.1539
0.0012
0.0038
0.0006
0.0004
-0.0100
-0.0133
-0.0128
-0.0078
-0.0124
-0.0099
-0.0139
-0.0164
-0.0163
-0.0203
-0.0020
-0.0173
-0.0005
-0.0006
-0.0035
-0.0136
-0.0086
-0.0049
-0.0036
-0.0031
0.0030
-0.0018
0.0200
0.013
0.005
0.002
0.005
0.008
0.003
0.004
0.002
0.002
0.092
0.003
0.002
0.002
0.002
0.003
0.004
0.004
0.004
0.004
0.005
0.007
0.005
0.004
0.006
0.006
0.005
0.006
0.006
0.008
0.008
0.005
0.007
0.008
0.008
0.004
0.007
0.000
0.25
1.10
68.94
6.58
11.76
17.97
26.31
93.38
103.77
2.79
0.13
2.58
0.12
0.06
10.86
11.30
13.05
4.30
9.00
3.80
3.77
12.73
13.13
12.85
0.12
10.03
0.01
0.01
0.20
3.13
2.65
0.49
0.20
0.16
0.50
0.07
63
-------
Appendix C
f
The Hedonic Price and Fuel Efficiency Models
The econometric model used for the hedonic price of a vehicle is based on the work of
Rosen (1974), Atkinson and Halvorsen (1990), and Dreyfus and Viscusi (1995) on hedonic
pricing. Atkinson and Halvorsen (1990) use the dalja for 112 models of new 1978 automobiles to
obtain estimates of the VSL. Since the available falality data is a function of both the inherent
risk of the vehicle and the driver's characteristics, the drivers' characteristics are included in the
regression as control variables. Their estimated VSL for the sample as a whole, based on
willingness to pay, is $3.357 million 1986 dollars.
The data used in Dreyfus and Viscusi (1995) differ from those used in earlier studies in
that they reflect actual consumer automobile holdings. Dreyfus and Viscusi (1995) use the 1988
Residential Transportation Energy Consumption Survey together with data from industry
sources. They generalize the standard hedonic models to recognize the role of discounting on
fuel efficiency and safety. The estimates of the implicit value of life range from $2.6 to $3.7
million and the estimates of the discount rate range
The hedonic price equation for automobiles
from 11 to 17 percent
;can be written, following Atkinson and
Halvorsen (1990), as follows:
Pauto = f(R, A),
where Pauto is the price of an automobile, R is the inherent risk of mortality (a similar measure for
injury could also be included) associated with the automobile, and A is a vector of other
characteristics. The available mortality rate, F, is a function of both R and a vector of the
involved driver's characteristics D. Assuming that F is monotonic in PS the above equation can
also be written as:
The standard functional form used for the estimation of a hedonic price equation is:
where Xt is a representative measured regressor (e.g. horsepower to weight ratio), D; is a dummy
variable for vehicle type, yt , /?# are the corresponding parameters and e is an unobserved
residual.
A different approach was used in this research, and it involves predicting the inherent
mortality rate using standardized driver's characteristics. In other words, the unobserved values
of R are predicted directly. Since the specified number of occupants of a vehicle is two, the
observed mortality rate F is twice the size of the average mortality rate per occupant. The
corresponding value of R should also reflect the fact that there are two occupants on average.
Consequently, the predicted value R =/\ + r2 (i = 1 is the driver and i =: 2 is the passenger),
where r, is the predicted probability of a fatality for an individual, defined in the previous
section. The standardized inherent mortality rates for two male occupants for year 1995
automobiles: are summarized by type of vehicle in figure Cl. The minimum, average and
maximum risks of mortality for each type of vehicle are illustrated. Figure C2 provides the
corresponding scales for the raw (unadjusted) mortality data based on 1996-1997 PARS data.
Comparing the two figures, the relative ranking among different types of vehicle are quite
This research was supported by United States Environmental Protection Agency Cooperative Agreement Number CR824393-01-1. We would
like to thank Margaret French for her assistance in preparing the manuscript. All conclusions and remaining errors are the sole responsibility of
the authors.
-------
t
consistent, but the standardizing procedure significantly reduces the ranges of the risk of
mortality.
One might be surprised by the implication from Figures Cl and C2 that large sports
utility vehicles (SUVs) are not safer than small sedans and wagons. From Table Cl, the average
standardized and risks of mortality show that large SUVs are safer in two-vehicle and multiple-
vehicle accidents (1.6+0.7=2.3 compared to 4.4+1.4=5.8 for small sedans). However, they are
much less safe in one-vehicle accidents (7.1 compared to 3.4 for small sedans) because the
probability of having an accident is higher. This point can be further illustrated by the
information in Table C3. For two-vehicle accidents, large SUVs have the lowest observed
mortality rate per occupant (0.186) among all types of vehicle, which is about a third of the rate
for small sedans (0.512). However, the observed accident rates for large SUVs and small sedans
are the same (0.193). The impression that large SUVs are safer than other vehicles comes from
observing that occupants in a large SUV are more likely to survive in a fatal accident with
another vehicle than the occupants of other types of vehicle.
Another cost associated with reducing the risk of mortality and injury is buying more fuel
because heavier vehicles are safer but have lower fuel efficiencies. Consequently, ahedonic
model of fuel efficiency augments the standard hedonic model of the purchase price in our
model. In this model, the cost of additional safety has a capital component and an operating
component. In the latter case, the cost penalty corresponds to the reduced fuel efficiency when a
heavier vehicle is purchased. The hedonic model of fuel efficiency has the same form as the
hedonic model of the purchase price, and it can be written:
log(/e _city) = a0
1 + e
where fe_city is the rated miles per gallon for city driving, Xk is a representative measured
regressor, D; is a dummy variable for vehicle type, 5;, and o^ are the corresponding parameters
and e is an unobserved residual.
f
The primary source of the data for estimating the hedonic price and fuel efficiency models was
the 1995 National Personal Transportation Survey (NPTS). This data was used to obtain
information on each household's choice of automobiles. The 1995 NPTS was conducted by the
Research Triangle Institute (RTI) under the sponsorship of the U.S. Department of
Transportation (DOT). The survey covers 42,033 sampled households. A sub-data set of 4036
one-car households holding a 1990-1995 model year vehicle were merged with vehicle attribute
data collected from industry and other sources for the same years. The vehicle price data were
gathered from NADA Official Used Car Guide, and other attribute data were collected from
NADA Official Used Car Guide, Ward's Automotive Yearbook, and Consumer Reports. The
mortality rate is measured by the number of fatalities occurring in each make/model/year vehicle
per 1000 vehicles sold. The number of fatalities is based on the models described in Appendix B.
Since the observed mortality rate is jointly determined by the inherent risk associated with the
type of automobile and the driver's characteristics and behavior, driver's characteristics were
also collected from the data on fatal accidents for each make, model and year to provide control
variables.
65
-------
In addition to the risk of mortality, a second safety measure, injury rate, is introduced.
The injury rate by make and model of vehicle is published annually by the Highway Loss Data
Institute. It is measured by the frequency of insurance claims filed under Personal Injury
Protection coverages. The raw injury rates are adjusted by the same factors used to standardize
raw mortality rates. The implicit assumption is that the "bad" driving characteristics that
contribute to fatal accidents also affect injuries.
The variables used in the hedonic price equation are summarized in Table C4, and Table
C5 shows the descriptive statistics of selected vehic le attributes. The selection of vehicle
attributes and driver's characteristics is similar to Dreyfus and Viscusi (1995) and Atkinson and
Halvorsen (1990). It should be noted that the observed mean mortality rate is higher than the
standardized mean and the observed standard deviation is also higher. The reason is that the
standardized mortality is based on one average male driver and one average male passenger.
Even though average values of the other regressors are used, the elimination of young drivers,
for example,, results in lower average mortality rates. The effect of standardizing drivers'
characteristics to predict the inherent mortality rate {has the effect, as expected, of reducing the
variability of mortality among vehicles.
The Estimated Hedonic Models
Least square estimates of the hedonic price model and the fuel efficiency model are
presented in Table C6. Model A is the hedonic equation of fuel efficiency, using the
standardized mortality rate. Model B is the hedonic equation of capital cost, using the
standardized! mortality rate. In Model A and B, variables with small t ratios and perverse signs
have been dropped.
The most important parameter for computing the VSL is the coefficient for the mortality
rate, and the values in Model A and B have the right signs and are both significant. In other
hedonic price models, fuel efficiency is included as a regressor in Model B, but it often has a
large t ratio and a perverse negative sign (fuel efficiency is a positive attribute). Hence, some
explanation is needed to explain why fuel efficiency is omitted in Model B. The implication of
Model A is tjiat fuel efficiency is a dependent variable, like the price, and is a function of the
vehicle's characteristics. The model corresponds to a simplified reduced form for a system of
two equations. If the predicted fuel efficiency from Model A is used as a regressor in Model B,
the coefficient has a logical positive sign. The overall effect on the estimated VSL is small,
however, if the direct effects of mortality on price and fuel efficiency are combined with the
indirect effect on the price through fuel efficiency. This is not really surprising because the
model presented in Table C6 is equivalent to a solved reduced form for a structural model which
has fuel efficiency as a regressor in the hedonic prij;e equation (the equation for fuel efficiency
remains the Same).
t
66
-------
t
t
Table Cl: The Standardized Risk of Mortality by Vehicle and Type of Accident
Vehicle Type Total
Risk
small sedans & wagons 9.2
middle sedans & 6.9
wagons
large sedans & wagons 6.5
luxury sedans & 7.2
wagons
small & mid. 9.5
specialties
luxury sports 25.3
small suv 17.1
large suv 9.4
van (minivan) 5.0
small pickup 12.4
large pickup 8.6
One-Car
Accidents
3.4
3.3
3.5
4.7
5.6
21.8
12.0
7.1
2.7
7.7
5.8
Two-Car
Accidents
4.4
2.5
2.1
1.7
3.0
2.6
3.6
1.6
1.5
3.5
2.0
Multiple-Car
Accidents
1.4
1.0
0.8
0.8
1.0
0.9
1.6
0.7
0.8
1.2
0.8
This research was supported by United States Environmental Protection Agency Cooperative Agreement Number CR824393-01-
1. We would like to thank Margaret French for her assistance in preparing the manuscript. All conclusions and remaining errors
are the sole responsibility of the authors.
-------
t
Table C2: The Observed Risk of Mortality by Vehicle and Type of Accident (Year 1996-1997
Average)
Vehicle Type
Total
Risk
Onls-Car
Accidents
Two-Car
Accidents
Multiple-Car
Accidents
small sedans & wagons 30.8 12.2
middle sedans & 22.5 12.1
wagons
large sedans & wagons 17.7 5.4
luxury sedans & 9.3 4.9
wagons
small & mid. 33.8 20.5
specialties
luxury sports 26.2 23.6
small suv 53.4 26.6
large suv 21.1 16.2
van(minivan) 24.6 12.8
small pickup 26.1 17.6
large pickup 17.6 11.6
14.1
8.8
11.2
3.2
10.3
2.6
25.6
3.5
8.8
7.2
4.8
4.6
1.6
1.2
1.2
3.0
0.0
1.2
1.4
3.0
1.4
1.2
f
68
-------
Table C3: The Observed Mortality Rates Per Occupant and Accident Rates per 1000 Vehicles for
Fatal Two-vehicle Accidents (Average 1996-1997)
Vehicle Type Mortality Accident
Rate Rate
small sedan & wagons 0.512 0.193
middle sedan & wagons 0.435 0.169
large sedan & wagons 0.370 0.159
luxury sedan & wagons 0.369 0.113
small & mid. specialties 0.429 0.171
luxury sports 0.329 0.109
small suv 0.430 0.189
large suv 0.186 0.193
van (minivan) 0.218 0.221
small pickup 0.368 0.188
large pickup 0.201 0.244
69
-------
Table C4: Variable Definitions
Variable Name
Definition
Price
Value Retained
Mortality Rate.
Observed
Mortality Rate.
Standardized
Injury Rate
CityFuel efficiency
CityFuel efficiency
Predicted
Reliability Rating
Acceleration
Trajditionat Styling
ClassX
YearXX
Young Driver
Older Driver
Alcohol
Gender of Driver
Seat Belt
Previous Offenses
Late Night
One-car Accident
Ford, GM, Chrysler,
Germany, Japan
MB
Vehicle price as of end-of-year 1995.
Original sales value retained, as of end-of-year 1995.
Number of fatalities occurring in that make/model/year vehicle
per 1000 of that vehicle sold.
Predicted number of fatalities in that make/model/year vehicle
per 1000 of that vehicle sold with average 2 occupants.
An Index based on the frequency of insurance claims. The
lower, the safer,
Miles per gallon in city area.
Predicted Miles per gallon in city area.
A discrete variable coded from 1 to 5, 5 is the highest while 1 is
the lowest.
The horsepower-to-weight ratio.
Length plus width
Discrete variables
divided by height.
coded as 1 for the appropriate class. Class 1
to class? represent small, middle, large, luxury, SUV, van,
and pick-up truck, respectively.
Discrete variables coded as 1 for the vehicle model year.
Proportion of fatalities in this make/model/year vehicle in
which the driver was younger than 25 years.
Proportion of fatalities in this make/model/year vehicle in
which the driver was 65 or older.
Proportion of fatalities in this make/model/year vehicle in
which the alcohol involvement was reported.
Proportion of fatalities in this make/model/year vehicle in
which the driver was male.
Proportion of fatalities in this make/model/year vehicle in
which the driver was wearing a seat belt.
Proportion of fatalities in this make/model/year vehicle in
which the driver had no previous offense.
Proportion of fatalities in this make/model/year vehicle which
occurred between 12:00am to 5:59am.
Proportion of fatalities in this make/model/year vehicle in
which only one vehicle was involved.
Discrete variables coded as 1 for the manufacturer and 0
otherwise.
Dummy variable coded as 1 for Mercedes Benz, 0 otherwise.
t
70
-------
Table CS: Summary Statistics of Selected Variables
Variable
Price
Value Retained
Mortality Rate, Observed
Mortality Rate, Standardized
Injury Rate
City Fuel-efficiency
Reliability Rating
Acceleration
Traditional Styling
Mean
15703.53
0.7720
0.1345
0.0939
73.72
20.26
3.019
0.0475
4.451
Standard Deviation
9371.57
0.1753
0.0994
0.0401
42.12
4.82
1.321
0.0102
0.519
71
-------
Table C6: Parameter Estimates for the Hedonic Equations
Variable
Dependent
Constant
Value Retained
Mortality Rate
Injury Rate
Reliability Rating
Acceleration
Traditional Styling
Class2
Class3
Class4
Class5
Class6
Class?
Yqar91
Year92
Year93
Year94
Year95
Ford
GM
Chrysler
Germany
Japan
MB
R2
Model A
Estimated
Coefficient
Fe_city
2.5689
0.0549
0.0258
0.0330
0.0170
-0.2290
-0.2786
-0.1873
-0.2751
-0.2852
-0.6397
-0.4846
-0.4352
0.0347
0.0334
0.0196
-0.0562
0.0470
-0.0078
0.7626
t ratio
14.13
3.35
1.99
4.01
5.05
-8.04
-5.21
-16.56
-14.69
-19.29
-37.47
-24.84
-27.49
1.90
1.94
1.12
-2.84
2.73
-0.33
Model B
Estimated
Coefficient
Pauto
7.7174
0.45S>4
-0.0690
-0.0161
0.06] 7
0.6014
0.6035
0.2426
0.3734
0.6752
0.8127
0.6558
0.3398
0.1137
0.2100
0.2977
0.3880
0.4474
-0.0972
-O.OK79
-0.1148
0.1489
-0.0430
0.5237
0.8996
t ratio
25.45
11.10
-3.53
-1.31
5.23
13.99
7.56
14.34
13.28
29.76
31.94
22.67
14.31
6.31
10.53
13.16
15.30
16.14
-3.58
-3.44
-4.43
5.05
-1.71
14.89
t
t
72
-------
o
M
Jj
•8
1
vi
00
VI
IP
en
'T
»i
t wagons
=o
st
•a
a
(A
&
|
small & mid.
\p
^o
VI
tS
t<
| specialties
1
•^
r*i
•£}
^1
TT
m
"™*
W
r
§.
(A
t
3
X
3
oc
^
1f\
v%
>•
3
M
13
E
(A
VI
vi
t
e<
p^
>
s
M
4>
ff
2
p
r^
O
^S
0
>»
van (minivan)
r~~
^t
•q;
ri
=~
P
&.
5
U
0.
13
fi
00
—
>c
ac
en
S
large pickup
Q.
e
M
X
I
I
£
00
r-
i
I
e
£
s
CM
01
X
-------
I
90
U
I
£
1
1
iH C
r* T
O
ffi
00
-
<*
in
m
IN
*•*
i— i
8
s
O
f-.
s
>*J
O
in
0
S
o
rs
o
0
Tf
s
00
mm
o
small sedan &
wagons
\§
in
•S
H
middle sedan &
wagons
i
t-s
ti
ts
oc
S
?.
large sedan & waj
*
«
o
luxury sedan &
wagons
i
CO
cp
r- '
small & mid.
specialties
i^
«
IN
S
Q
}
i
CO
•w
$
i
o\
R
*
O
1- 1
^3
09
in
1-H
ON
•S
1
CO
in
S
1-H
CN
T— *
1
'On
A
^
N
O
i-i
O
r- »
large pickup
8,
o
•3
•a
o
8 J5
"S -a
2 B
D 2;
> $
3
1
ON
•a S
si
•8
O
"8
.3
•3
«
o
-------
Appendix D: Auto Safety Survey (Full scale)
t
worn
L Variable names ace in bold type
2, A eode of systeot aaissiaf (*) means &e question w«s ao I ?ppHpaljl<
3. Questions were
4 (A) afl»F variable .aa«te tatucates a
5, F^r c^ej wfeer^ttje aftswei^ were no* oiesi;
*ni
a
iu $504^0 om M3Ji so w&
All cases with usclear ffliswers or ranges are defied atong wilh &eir casetds
n
Sample fafwatation
RESPOND Unique identification number for completed recruitment screeners
CASEID Unique case identification number for mailed surveys
DATE_C Date recruit completed
DATE_R Date mail survey returned
VERSION Version number of survey mailed
1 Version 1
2 Version 2
STATUS Level of mail survey completion
0 Refused to be recruited for mail survey
1 Agreed to mail, but did not return survey
5 Partial mail complete
10 Full recruit and mail complete
STATE (A) State where respondent resides
ZIP (A) Zip code where respondent resides
FIPSTATE Numeric FIPS state code
FIPCONTY Numeric FIPS county code
REMAIL Whether or not respondent received follow-up survey mailing
80
-------
0 No reminder survey mailing (either because had returned completed
survey or said during follow-up call that Ihey would return the
survey they had)
1 Sent follow-up survey mailing
t
81
-------
Recruitment Smrey
t
Hello, my name is (FILL IN) and I am calling on behalf of Cornell University. May I
speak With an adult head of your household?
My name is (FILL IN) and I am calling from Discovery Research We are conducting a
Soy™ Cornell Univasity of people's use of automobiles and then- opmiom about
automobile safety. I'd like to assure you that this .s a research project and I am not trymg
to sell you anything.
(IFNEEDED): Youarepart of a small group of U.S. residents who have been
sdentifically selected for this study. Your opinions will represent other people hke you.
This should take about 15 minutes.
For this survey, when 1 say automobiles, I mean all types of passenger automobiles
including station wagons/sport utility automob.les (SUVs), mim-vans, and light trucks
h w rf £Ulor leased by someone in your household and are used regular!iy by
someone in your household. Do not include motorcycles, or automobiles, such as antique
cars, that are rarely driven,
(VERIFY THAT RESPONDENT IS AN ADULT HEAD OF HOUSEHOLD.)
Ql
Q2
Including yourself, how many people live in your household? Please count
yourself and any family members or partners who live with you. Do not include
unrelated adult roommates who make their own automobile purchase decmons.
(INTERVIEWER NOTE: WE WANT HOUSEHOLDS THAT ARE JOINT
SlON^NG UNITS, SO THIS SHOULD INCLUDE PARTNERS^
NOT UNRELATED ADULTS WHO ARE JUST ROOM^TES AND MAKE
THEIR OWN SEPARATE CAR PURCHASE DECISIONS. IFTHELATTER,
THEN SAY TO RESPONDENT: Tor the remainder of this survey, I only want
you to tell me about yourself and anyone else in this household who is part of
your decision making for automobiles purchases.",)
People (including respondent)
(NOTE: IFQ1 GREATER THAN 5, THANK AND TERMINATE.)
57 Households with more than 5people.
How many automobiles does your household currently own or lease?
82
-------
(IFNEEDED): When I say automobile, I mean all types of passenger
automobiles including station wagons, sport utility automobiles (SUVs), mini-
vans, and light trucks that were purchased or leased by someone in your
household. Do not include motorcycles, tyr automobiles that are not driven more
than 500 miles per year.
automobiles
(NOTE: IF Q2 EQUALS 0, OR Q2 GREA TER THAN 3, THANK AND
TERMINATE.)
235 Households -with 0 autos.
181 Households with more than 3 autos.
Q3 Does anyone in your household regularly use an automobile that was purchased
by his or her employer?
1 Yes
TERMINATE.)
2 No
(NOTE: IF YES, THANK AND
118 Households had an employer-
purchased auto.
Q4 Does anyone in your household regularly use an automobile that was given to
them by someone outside your household?
1 Yes
TERMINATE.)
2 No
(NOTE: IF YES, THANK AND
74 Households had a gift auto.
Q5 Do you drive any of your household's automobiles?
(ASK TO SPEAK WITH AN ADULT
1 Yes
2 No
DRIVER)
83
-------
t
Q6 I'd like you to tell me about each of the automobiles that are owned or leased by
members or your household. Starting with the automobile that is driven the most,
Characteristics
a. What make or brand is this automobile? (e.g.,
Chevrolet, Mercedes)
b. What model is it? (e.g., Camaro, S-Class) (If don't
know, please ask respondent to check.)
c. Type of model? (e.g., LE, LX, SE) (If don't know,
please ask respondent to check.)
d. What model year is the automobile?
e. What year did you purchase or lease the
automobile?
f. What was the approximate purchase price of this
automobile/equivalent price that was used in
calculating your lease payments?
g. Have you had this automobile for less than 1 2
months?
h. About how many miles (was this automobile
driven over the last 12 months//F HAD FOR LESS
THAN 12 MONTHS: will this automobile be
driven in a 12-month period?)
Range of miles (ifQ6h isDK)
Most Driven
Automobile
(in terms of miles/year)
Q6al (A)
Q6bl (A)
Q6cl (A)
Q6dl
-8 Don't Know
Q6el
-8 Don't Know
$Q6fl
-8 Don't Know
-9 Refused
Q6gl
1 Yes 2 No
Q6hl
-8 Don't Know
(If Don't Know,
prompt with
categories; see Qdhlf
below.)
-9 Refused
Q6hlf
1 Under 3,000 miles
2 3,000 to 5,999
3 6,000 to 8,999
4 9,000 to 11, 999
5 12,000 to 14,999
6 15,000 to 17,999
7 18,000 to 20,999
8 21,000+ miles
-8 Don't Know
• NA
2nd Most Driven
Automobile
Q6a2(A)
Q6b2(A)
Q6c2 (A)
Q6d2
-8 Don't Know
Q6e2
-8 Don't Know
$Q6f2
-8 Don't Know
-9 Refused
Q«g2
1 Yes 2 No
Q6h2
-8 Don't Know
(If Don't Know,
prompt with
categories; see Q6hlf
below.)
-9 Refused
Q6h2f
1 Under 3,000 miles
2 3,000 to 5,999
3 6,000 to 8,999
4 9,000 to 11, 999
5 12,000 to 14,999
6 15,000 to 17,999
7 18,000 to 20,999
8 21,000+ miles
-8 Don't Know
• NA
3rd Most Driven
Automobile
Q6a3(A)
Q6b3 (A)
Q6c3(A)
Q6d3
-8 Don't Know
Q6e3
-8 Don't Know
SQ613
-8 Don't Know
-9 Refused
Q6g3
1 Yes 2 No
Q6h3
-8 Don't Know
(If Don't Know,
prompt with
categories; see Qfhlf
below.)
-9 Refused
Q6h3f
1 Under 3,000 miles
2 3,000 to 5,999
3 6,000 to 8,999
4 9,000 to 11,999
5 12,000 to 14,999
6 15,000 to 17,999
7 18,000 to 20,999
8 21,000+ miles
-8 Don't Know
• NA
(PROGRAMMING NOTE: FROMNOWON, FILL "MOST DRIVEN AUTOMOBILE," "2™ MOST
DRIVEN VEHICLE", "f° MOST DRIVE AUTOMOBILE" WITH ACTUAL MAKE AND MODEL AND
MODEL YEAR OF AUTOMOBILE.")
(PROGRAMMING NOTE: FROMNOWON, IF AN AUTOMOBILE HAS BEEN IN THE HOUSEHOLD
LESS THAN 12 MONTHS, USE THE SECOND PHRASE IN THE PARENTHESIS WHICH REFERS TO
THE AMOUNT THE VEHICLE WILL BE USED IN A 12-MONTH PERIOD.)
84
-------
Recruit:
t
Q7 To understand auto safety, we need to know how your household uses your
automobiles, I'd like to get the first namq or initial, age and gender of all the
people living in your household. We will be referring to these family members in
the next set of questions, so it will be easier if we have a way to identify each
individual. You can use nicknames if you prefer. First, whal is your first name?
Person
Person A
(Yourself)
Person B
Person C
Person D
Person E
First Name
NA
Q7bname (A)
Q7cname (A)
Q7dname (A)
QTename (A)
Age
-9 Refused
• NA
Q7aage
I
Q7bage
Q7cage
Q7dage
Q7eage
Gender
I Male
2 Female
• NA
Q7amf
Q7bmf
Q7craf
Q7dmf
Q7emf
Relationship with Respondent
1 Spouse
2 Partner
3 Son
4 Daughter
5 Mother
6 Father
7 Other relative
8 Friend
. NA
NA
Q7brelat
Q7crelat
Q7drelat
Q7erelat
(PROGRAMMING NOTE: FROM NOW ON, FILL "PERSONA" WITH "YOU, "AND "PERSON B-E"
WITH THE FIRST NAME).
85
-------
t
Q8
In this next question, I would like to know how much each person in your
household (was in each automobile during the past 12 monfhs/ZFHAD FOR LESS
THAN ]2 MONTHS: will be in this automobile in a 12-month period), either as a
driver or a passenger For the [FILL WITH AUTOMOBILE MAKE AND
MODEL], you mentioned it (was/will be) driven about [FILL WITH ANNUAL
MILEAGE] miles in a 12-month period. What percentage of these miles (did/will)
[FILL WITH PERSON NAME] drive or ride in this automobile? We know that
this is a difficult question, and it is okay to give approximate answers. (PROBE):
For example, did this person drive or ride in this automobile for all the miles (it
was driven/it will be driven), for about 50% of the miles, for about 25% of the
miles, or some other amount?
For Q8a_l to Q8e_3:
-8
-9
Don't Know
Refused
NA
Most Driven Auto:
Approximate percentage of annual miles each person rides in the Most Driven Car
(either as driver or passenger)
A - Yourself
Q8a_l
Person B
Q8b_l
Person C
Q8c_l
Person D
Q8d_l
Person E
Q8e_l
•>nd
2 Most Driven Auto:
Approximate percentage of annual miles each person rides in the 2nd Most Driven Car
(either as driver or passenger)
A - Yourself
Q8a_2
Person B
Q8b_2
Person C
Q8c_2
Person D
Q8d_2
Person E
Q8e_2
3rd Most Driven Auto:
Approximate percentage of annual miles each person rides in the 3"1 Most Driven Car
(either as driver or passenger)
A - Yourself
Q8a_3
Person B
Q8b_3
Person C
Q8c_3
Person D
Q8d_3
Person E
Q8c_3
(ADDITIONAL CLARIFICATION): The following example may help you think
through the question. Suppose that the "Most Driven Car" in your household is
your car that you drive alone to work most days. Also, most household outings
and longer vacation trips are in that car. Thus, you may estimate that of the total
annual miles that the "Most Driven Car" is driven, about 70% you are driving
alone, and about 30% are family trips when you are all in the car. Thus, you may
estimate that you are in the "Most Driven Car" 100% (70% + 30%) of the miles it
is driven and your spouse and each of your children are in the car 30% of the
86
-------
miles it is driven. In this case you will fill in the box of the 'Most Driven Car,"
under A-yourself, 100%, and under persons B (your spouse), C and D (your
children) 30%.
s
87
-------
Q9
t
You said that [FILL WITH PERSON NAME] (rode in or drove/will ride in or
drive) [FILL WITH AUTOMOBILE MAKE AND MODEL] about [FILL WITH
PERCENT OF MILES] of the miles it is driven in a 12-month period What
percent of this time (did they/will they) occupy the front seat and the back seat?
For Q9a_fl to Q9e_b3: %
-8 Don't Know
-9 Refused
• NA
(NOTE: THIS MUST TOTAL 100% IF THEY RODE AT ALL IN THE
AUTOMOBILE.)
Most Driven Auto:
A - Yourself
Seat %,of
miles
Front
Back
Total
Q9a_fl
Q9a_bl
100%
Person B
c * %of
Seat ..
miles
Front
Back
Total
Q9b_fl
Q9b b
1
100%
Person C
„ . %of
Seat .,
miles
Front
Back
Total
Q9c_H
Q9c_bl
100%
Person D
Seat %,°f
miles
Front
Back
Total
Q9d_fl
Q9d_b
1
100%
Person E
Seat %,of
miles
Front
Back
Total
Q9e_H
Q9e_bl
100%
-,nd
2 Most Driven Auto:
A - Yourself
_ % of
Seat ..
miles
Front
Back
Total
Q9a_f2
Q9a_b2
100%
Person B
c * %of
Seat ..
miles
Front
Back
Total
Q9bJ2
Q9b_b
2
100%
Person C
C 4 %0f
Seat .,
miles
Front
Back
Total
Q9c_f2
Q9c_b2
100%
Person D
c * %of
Seat ..
miles
Front
Back
Total
Q9d_f2
Q9d b
2
100%
Person E
C 4 %0f
Seat ..
miles
Front
Back
Total
Q9e_f2
Q9e_b2
100%
3rd Most Driven Auto:
A - Yourself
c * %°f
Seat .,
miles
Front
Back
Total
Q9aJ3
Q9a_b3
100%
Person B
c 4 °/»°f
Seat miles
Front
Back
Total
Q9b_f3
Q9b_b
3
100%
Person C
o . Mot
Seat ..
miles
Front
Back
Total
Q9c_O
Q9c_b3
100%
Person D
c 4. y» °f
Seat .,
miles
Front
Back
Total
Q9d_f3
Q9d b
3
100%
Person E
0 4 %°f
Seat ..
miles
Front
Back
Total
Q9e_f5
Q9e_b3
100%
88
-------
(ADDITIONAL CLARIFICA TION): The following example may help you think
through the question. Suppose that of the\ total miles that you ride in the "Most
Driven Car, " 90% of these miles you drive and 10% you occupy a hack seat while
another family member drives. In this cake you -will fill in the box of the "Most
Driven Car," under A-yourself, 90% Front and 10% Back.
Q10 In the next set of questions, I am interested in learning the main reasons why you
and other members of the household use your automobiles, either as a passenger
or a driver. (Not asked if person was younger than 14. If younger than 14,
QJOa=No.)
Use of Automobiles for
Work
a. Do/Does [FILL WITH
NAME] work outside the
home?
b. Do/Does [FILL WITH
NAMH] travel to work in
one of your household's
automobiles?
c. What automobile do
you/they use most often
to get to work?
d. How many miles is it
(one-1ivay) from your
house to the workplace?
(INTERVIEWER NOTE:
THE PRIMARY
WORKPLACE IF MORE
THAN ONE.)
e. Who usually drives
you/fFILL WITH
NAME] to work?
Person A -
Yourself
QlOa a
1 Yes
2 No (SKIP TO
Q10a_0
QlOa b
1 Yes
2 No (SKIP TO
QlOa f)
. NA
Q10a_c
1 Most Driven
Auto
2 2nd most
driven
3 3rtmost
driven
• NA
Q10a_d
-8 Don't Know
-9 Refused
• NA
QlOa e
1 Yourself
2 PersonB
3 Person C
4 Person D
5 Person E
6 Other
person
• NA
Person B
QlOb a
1 Yes
2 No (SKIP TO
QlOb f)
• NA
QlOb b
1 Yes
2 No (SKIP TO
QlOb f)
- NA
Q10b_t
1 Most Driven
Auto
2 2nd most
driven
3 3tdrrtost
driven
. NA
Q10b_d
-8 Don't Know
-9 Refused
• NA
QlOb e
1 Yourself
2 Person B
3 Person C
4 Person D
5 Person E
6 Other
person
• NA
Person C
QlOc a
1 Yes
2 No (SKIP TO
QlOc f)
• NA
QlOc b
1 Yes
2 No (SKIP TO
QlOc f)
• NA
Q10c_c
1 Most Driven
Auto
2 2nd most
driven
3 3rd most
driven
• NA
Q10c_d
-8 Don't Know
-9 Refused
• NA
QlOc e
1 Yourself
2 Person B
3 Person C
4 Person D
5 Person E
6 Other
person
• NA
Person D
QlOd a
1 Yes
2 No (SKIP TO
QlOd 1)
• NA
QlOd b
1 Yes
2 No (SKIP TO
QlOd 0
• NA
Q10d_c
1 Most Driven
Auto
2 2nd most
driven
3 3ldmost
driven
- NA
Q10d_d
-8 Don't Know
-9 Refused
• NA
QlOd e
1 Yourself
2 PersonB
3 Person C
4 Person D
5 Person E
6 Other
person
• NA
Person E
QlOe a
1 Yes
2 No (SKIP TO
QlOe t)
. NA
QlOe b
1 Yes
2 No (SKIP TO
QlOe f)
• NA
Q10c_c
1 Most Driven
Auto
2 2nd most
driven
3 3rd most
driven
• NA
Q10e_d
-8 Don't Know
-9 Refused
• NA
QlOe e
1 Yourself
2 Person B
3 Person C
4 Person D
5 Person E
6 Other
person
. NA
89
-------
I
t
Use of Automobiles for
School/Day Care
a. Do/Does [FILL WITH
NAME] go to school or
day care?
b. Do/Does [FILL WITH
NAME] travel today
care or school in one of
your household's
automobiles?
c. Which automobile is
used most often to get to
school or day care?
d. How many miles is it
(one-way) from your
house to school or day
care?
e. Who usually drives
you//F/LL WITH
NAME] to school or day
care?
Person A -
Yourself
QlOa f
1 Yes
2 No (GO TO
QlOb a)
- NA
QlOa g
1 Yes
2 No (GO TO
Q10b_a)
• NA
Q10a_h
1 Most Driven
Auto
2 2nd most
driven
3 3rd most
driven
. NA
Q10a_I
-8 Don't Know
-9 Refused
• NA
QlOaJ
1 Yourself
2 Person B
3 Person C
4 Person D
5 Person E
6 Other person
• NA
Person B
QlOb f
1 Yes
2 No (GO TO
QlOc a)
• NA
QlOb g
1 Yes"
2 No (GO TO
Q10c_a)
• NA
QlOb h
1 Most Driven
Auto
2 2nd most
driven
3 3ldmost
driven
• NA
Q10b_I
-8 Don't Know
-9 Refused
• NA
QlObJ
1 Yourself
2 PersonB
3 Person C
4 Person D
5 Person E
6 Other person
• NA
Person C
QlOc f
1 Yes
2 No (GO TO
QlOd a)
- NA
QlOc g
1 Yes
2 No (GO TO
QlOd a)
• NA
Q10c_h
1 Most Driven
Auto
2 2"* most
driven
3 3rd most
driven
• NA
QlOcJ
-8 Don't Know
-9 Refused
• NA
QlOcJ
1 Yourself
2 PersonB
3 Person C
4 Person D
5 Person E
6 Other person
. NA
Person D
QlOd f
1 Yes
2 No (GO TO
QlOe a)
• NA
Q10d_g
1 Yes
2 No (GO TO
QlOe a)
« NA
Q10d_h
1 Most Driven
Auto
2 2nd most
driven
3 3rfmost
driven
• NA
Q10d_l
-8 Don't Know
-9 Refused
• NA
QlOdJ
1 Yourself
2 Persons
3 Person C
4 Person D
5 Person E
6 Other person
• NA
Person E
QlOe f
1 Yes
2 No (GO TO
Qll)
• NA
QlOe g
1 Yes
2 No (GO TO
Qll)
• NA
Q10e_h
1 Most Driven
Auto
2 2nd most
driven
3 3rd most
driven
• NA
QlOe I
-8 Don't Know
-9 Refused
• NA
QlOeJ
1 Yourself
2 Person B
3 Person C
4 Person D
5 Person E
6 Other person
• NA
Qll What is the approximate distance, one way in miles, from your house to the
grocery store where you most often do your grocery shopping?
miles (NOTE: less than I mile = I)
-8 Don't know
-9 Refused
QHa Does your household use your automobile(s) for grocery shopping?
(SKIP TO Ql 2)
1 Yes
2 No
90
-------
(IF ONLY 1 AUTOMOBILE, SKIP TO Q12. NOTE: IF ONLY I AUTOMOBILE, THE
INFORMATION FOR THAT AUTOMOBILE WAS RECORDED DURING DATA
CLEANING INTO Qllb.)
Qllb Which of your automobiles is used most often for grocery shopping?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
• NA
Qllc Which of your automobiles is used second most often for grocery shopping?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
4 No second car is ever used
• NA
(SKIP TO Ql2)
(IFNO THIRD AUTOMOBILE, SKIP TO Q12)
Qlld Which of your automobiles is used third most often for grocery shopping?
1
2
3
4
Most driven automobile
Second most driven automobile
Third most driven automobile
No second car is ever used
NA
Q12 What is the approximate distance, one way in miles, from y our house to the
shopping centers, malls or stores where you go most often to do your other
shopping?
-8
-9
miles (NOTE: less than I mile = 1)
Don't know
Refused
Q12a What is the approximate distance, one way in miles, from your house to the
shopping centers, malls or stores where you go second most often to do your other
shopping?
91
-------
s
t
t
miles (NOTE: less than 1 mile = I)
-7 No other place (SKIP TO Ql 3)
-8 Don't know
-9 Refused
QlZb What is the approximate distance, one way in miles, from your house to the
shopping centers, malls or stores where you go third most often to do your other
shopping?
miles (NOTE: less than I mile = I)
-7 No other place (SKIP TO Q13)
-8 Don't know
-9 Refused
• NA
Q13 Does your household use your automobile(s) for getting to the shopping centers,
malls, or stores?
1 Yes
2 No (SKIP TO Q14)
(IF ONLY 1A UTOMOBILE, SKIP TO Q14. NOTE: IF ONLY IA VTOMOBILE, THE
INFORMATION FOR THAT AUTOMOBILE WAS RECORDED DURING DATA
CLEANING INTO Q13a.)
Q13a Which of your automobiles is used most often for going to shopping centers,
malls or stores?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
-7 Not asked
• NA
Q13b Which of your automobiles is used second most often for going to shopping
centers, malls or stores?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
92
-------
4 No second car is ever used
-7 Not asked
• NA
(SKIP TO Q14)
t
(IF NO THIRD AUTOMOBILE, SKIP TO Q14)
QI3c Which of your automobiles is used third most often for going to shopping centers,
malls or stores?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
4 No second car is ever used;
-7 Not asked
• NA
(SKIP TO Q14)
Q14 What is the approximate distance, one way in miles, from your house to the
theater where your household most often watches movies?
miles (NOTE: less than I mile = 1)
-7 No other place
-8 Don't know
-9 Refused
(SKIP TO Q15)
t
Q14a Does your household use your automobile(s) to get to the theater?
1 Yes
2 No (SKIP TO Q15)
• NA
(IFONLY I AUTOMOBILE, SKIP TO Q15. NOTE: IF ONLY 1 AUTOMOBILE, THE
INFORMATION FOR THAT AUTOMOBILE $AS RECORDED DURING DATA
CLEANING INTO Ql4b.)
93
-------
s
t
Q14b Which of your automobiles is used most often for going to the theater?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
-7 Not asked
• NA
Q14c Which of your automobiles is used second most often for going to the theater?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
4 No second car is ever used (SKIP TO Ql 5)
-7 Not asked
• NA
(IF NO THIRD AUTOMOBILE, SKIP TO Q15)
QI4d Which of your automobiles is used third most often for going to the theater?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
4 No third car is ever used (SKIP TO Ql 5)
-7 Not asked
• NA
Q15 Does anyone in your household use one of your automobiles on the job for more
than just commuting to and from work?
1 Yes (IF YES, GO TO Q15_R)
2 No (SKIP TO Ql6)
• NA
ForQ15_RtoQ15_C: 0 Not mentioned
1 Mentioned
• NA
Q15_R = Respondent uses auto on the job (Ask Q15al and Q15bl)
Q1S_B = Person B uses auto on the job (Ask Q15a2 and Q15b2)
94
-------
Q15_C = Person C uses auto on the job (Ask Q15a3 and Ql 5b3)
t
QlSal Which of your automobiles is used by (Respondent) on the job?
i
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
-7 Not asked
• NA
QlSbl About how many miles did (Respondent) travel in this automobile in the last 12
months for the job?
miles
-8 Don't know
• NA
Q15a2 Which of your automobiles is used by (Person B) on the job?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
• NA
Q15b2 About how many miles did (Person B) travel in this automobile in the last 12
months for the job?
miles
-8 Don't know
• NA
Q15a3 Which of your automobiles is used by (Person C) on the job?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
• NA
Q15b3 About how many miles did (Person C) travel in this automobile in the last 12
months for the job?
miles
-8 Don't know
• NA
95
-------
s
Recruit: Household Travel
Q16 Some households have certain places that they often drive to on weekends such as
a lake, a park, or relative's home. In this question, I would like to learn about your
household's travel as a group on weekend trips. Please think about how far it is
(one way) from your home to the three farthest places you went on weekends over
a typical year, and how many times you visited.
(NOTE: RESPONDENTS OFTEN DID NOT ANSWER THIS QUESTION IN THE
ORDER INTENDED. MANY WERE LIKELY TO ANSWER IN THE ORDER OF
HOW OFTEN THEY MADE THESE TRIPS.)
Q16a What is the approximate distance, one way in miles, from your house to the
farthest place you go on weekends over a typical year?
miles (NOTE: less than 1 mile = 1)
-7 Don't go on weekend trips (SKIP TO Q17)
-8 Don't know
-9 Refused
Q16aa About how many times do you go to this place each year?
miles (NOTE: less than 1 mile = 7,)
-8 Don't know
• NA
Q16b What is the approximate distance, one way in miles, from your house to the
second farthest place you go on weekends over a typical year?
miles (NOTE: less than 1 mile - 1)
-7 No other place (SKIP TO Ql7)
-8 Don't know
-9 Refused
• NA
Q16bb About how many times do you go to this place each year?
times
-8 Don't know
• NA
96
-------
Q16c What is the approximate distance, one wjay in miles, from your house to the third
farthest place you go on weekends over a typical year?
miles (NOTE: less than 1 mile = 1)
-1 No other place (SKIP TO Ql 7)
-8 Don't know
-9 Refused
• NA
s
Q16cc About how many times do you go to this place each year?
times
-8 Don't know
• NA
Q17 Does your household use your automobille(s) on these weekend trips?
(SKIP TO Ql 8)
1 Yes
2 No
• NA
(IF ONLY I AUTOMOBILE, SKIP TO Q18. NOTE: IF ONLY 1A UTOMOE1LE, THE
INFORMATION FOR THAT AUTO WAS RECORDED DURING DATA CLEANING
INTOQl7a.)
Q17a Which of your automobiles is used or will be used most often for these weekend
trips?
1 Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
-7 Not asked
• NA
Q18 In a typical year, do you drive your automobile(s) on any longer vacations?
1 Yes
2 No (IF NO, SKIP TO Ql 9)
97
-------
I
Q18a What is the approximate distance, one way in miles, from your house to the
farthest place your household drove in a typical year on a longer vacation?
miles (NOTE: less than 1 mile = I)
-8 Don't know
• NA
QlSaa About how many times do you drive to this place each year?
times (.5 = every other year)
*
-8 Don't know
• NA
(IF ONLY 1 AUTOMOBILE, SKIP TO Q19. NOTE: IF ONLY 1 AUTOMOBILE, THE
INFORMATION FOR THAT AUTOMOBILE WAS RECORDED DURING DATA
CLEANING INTO Q18b.)
Q18b Which of your automobiles is used or will be used most often for longer vacation
trips?
I Most driven automobile
2 Second most driven automobile
3 Third most driven automobile
-7 Not asked
• NA
Recruit: R«crai
-------
complete the survey, but we will include; a token of our appreciation along with
the survey. Would you be willing to help us out?
t
1 Yes (IF YES, PLEASE SPECIFY):
2 No
name
address
city
state
zip
(IF NO, GO TO Q20)
Q20 I just have a couple final questions that will help our researchers better understand
automobile use by different types of households. What is your present marital
status? Are you. . .? (READ LIST)
1 Single, never married
2 Married
3 Separated
4 Divorced
5 Widowed
7 Other (PLEASE SPECIFY) (SEE Q20oth BELOW)
-9 Refused
Q20oth Other marital status
1 Domestic partner
2 Engaged
3 Living with someone
4 Widowed and divorced
• NA
(FROM Q20 ABOVE)
Q21 Are you presently . . .? (READ LIST)
1 Employed full-time
2 Employed part-time
3 A full-time homemaker
4 Unemployed
5 Retired
6 A student
7 Other (PLEASE SPECIFY)
-9 Refused
(SKIP TO Q23)
(SKIP TO Q23)
(SKIP TO Q23)
(SKIP TO Q23)
(SEEQ2loth below)
t
99
-------
Q21oth Other employment (FROM Q21 ABOVE)
1 Disabled
2 Self-employed
• NA
Q22 Which one of the following occupational categories most closely reflects the type
of work you do in your job? (If you had more than one job in 2001, we only need
to know about your main job.)
1 Service worker, such as retail sales or hair stylist
2 Transportation operator, such as taxi, bus, train, or limo driver
3 Equipment operator
4 Craft worker, such as plumber or electrician
5 Traveling salesperson
6 Farm worker
7 Clerical worker
8 Laborer
9 Manager or administrator
10 Professional or technical
11 Other (PLEASE SPECIFY) (SEE Q22oth below)
-9 Refused
• NA
Q22oth Other occupations (FROM Q22 ABOVE)
1 Actor
2 Child Care
3 Communications
4 Correctional Officer
j 5 Custodian
7 Disabled
9 Hotel
10 Mail Carrier
11 Manufacturing
12 Meat Department
13 Parent Liaison
15 Self-employed
16 Teacher Aid
17 Truck Driver
18 U.S. Military
19 Works with Handicapped
20 YMCA
• NA
100
-------
Q23 What is the highest grade or year of schojol that you have completed?
1 No school
2 Grade school (1-8 years):
3 Some high school (9-11 years)
4 Completed high school (12 years)
5 Some college, but no degree (13-15 years)
6 Associate degree
7 Bachelor's degree
8 Post graduate
-9 Refused
Q24 What was your approximate gross household income from Eill sources (before
taxes and other deductions) in 2000?
1 Under $10,000
2 $10,000 to $19,999
3 $20,000 to $29,999
4 $30,000 to $39,999
5 $40,000 to $49,999
6 $50,000 to $59,999
7 $60,000 to $69,999
8 $70,000 to $79,999
9 $80,000 to $89,999
10 $90,000
11 $100,000
12 $120,000
13 $140,000
14 $180,000
15 $22.0,000
to $99,999
to $119,999
to $139,999
to $179,999
to $219,999
to $259,999
16 $260,000 to $300,000
17 More than $300,000
-9 Choose Not To Answer
(IFNEEDED FOR Q25 - Q27): We are studying household choices and automobile
safety so it is important for us to know if there is a parent outside the household who
provides'financial support for one or more of the children.
Q25 Does any of your household's income come from child support?
1 Yes
2 No
-9 Refused
(SKIP TO CLOSING)
(SKIP TO CLOSING)
Q26 How many children receive support?
Q27 How regularly are the full support payments received? Would you say ...?
(READ LIST)
101
-------
1 All of the time
2 Most of the time
3 Some of the time
4 None of the time
-9 Refused
* NA
CLOSING:
(IFRECRUITED, READ): You will be receiving your opinion survey by priority mail
within the next few days. I would appreciate it if you could return it as soon as possible.
It is very important that we collect the opinion data in this mail survey so we can use it to
better understand usage and opinions.
I'd like to thank you for taking the time to help me and Cornell University out with this
important study.
I
102
-------
Mail Survey
iVIaHt What Are Your! Vtews on Auto Safety?
Important Informatiop Before You Start
Thank you for agreeing to complete this important survey on automobile safety. When
talking with you on the telephone we asked several questions about the automobiles that
your household owns or leases.
Please look over the information below and fill jn anything that is incomplete as best you
can. Cross out any incorrect information and write in correct information as best you can.
Please continue to answer the remaining questions about the automobiles you had at the
time of the phone survey, even if there have be^n changes since then.
Characteristics
Make or brand
Model
Type of model
Model year
Year you purchased or leased the automobile
Approximate purchase price or equivalent price used
in calculating lease payments
Miles this automobile driven over the last 12 months.
(If you've: had this auto for less than 12 months, please
estimate the miles it will be driven in a 12-month
period.)
Most Driven
! Automobile
make_fl (A)*
make_cl (A)**
model_fl (A)
model" d (A)
type_f 1 (A)
type_cl (A)
yearjl
; year_cl
purch_fl
i purch_cl
price_fl
price_cl
miles_fl
miles_cl
2nd Most Driven
Automobile
make_f2 (A)
make_c2 (A)
model_f2 (A)
model_c2 (A)
type_f2 (A)
type_c2(A)
year_f2
year_c2
purch_f2
purch_c2
price_f2
price_c2
miles_f2
miles_c2
3ri Most Driven
Automobile
make_f3 (A)
make_c3 (A)
modeI_O (A)
modef c3 (A)
typeJ3 (A)
type_c3(A)
year_f3
year_c3
purch_f3
purch_c3
price_f3
price_c3
mile$_f3
miles_c3
* The _f# variables contain the information provided in the recruit screenur (series Q6al to Q6h3f)
that were used to customize the mail survey. Respondents could change or update this information if
necessary.
t
103
-------
** The _c# variables contain any new data that was added by the respondent. Sometimes this included
the addition of another automobile.
MaUt Automobile Safety and Your Hoaseheid
M1 The purpose of this survey is to find out how important automobile safety is to you.
About how often have you seen, heard, or read about automobile safety from TV,
radio, newspapers, or magazines in the past 6 months? (Please circle one
number.)
1 Never
2 A few times (1 to 4)
3 Several times (5 to 10)
4 Many times (11 to 20)
5 Very many times (More than 20)
-9 Missing
M2 Below is a list of factors that might affect your decision when purchasing or leasing
an automobile for use by yourself and your household. For each factor, please
indicate how important that factor would be to you when selecting an automobile
for purchase or lease. Circle the number that most closely corresponds to your
answer, where 1 = not at all important, and 7 = extremely important. (Please circle
one number for each factor.)
Not at all
Important
M2a
M2b
M2c
M2d
M2e
M2f
M2g
M2h
Passenger capacity 1
Cargo space 1
Comfort 1
Fuel economy 1
Four wheel drive 1
Engine power 1
Price 1
Safety 1
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
6
6
6
6
6
6
6
6
Extremely Missing
Important
7
7
7
7
7
7
7
7
-9
-9
-9
-9
-9
-9
-9
-9
t
104
-------
M3 Please tell us the approximate monthly costs for gas for each of your automobiles.
(Please fill in the dollar amounts.)
ForM3 ltoM3 3: $ >
-9 Missing
• NA
MONTHLY COST
Gas
Most Driven Auto
M3_l
2nd Most Driven Auto
M3_2
3rd Most Driven Auto
M3_3
M4
PJease tell us the approximate annual (yearly) insurance and maintenance and
repair costs for each of your automobiles. (Please fill in the dollar amounts.)
For M4a 1 to M4a 3 and M4b 1 to M4b 3: $
-9
Missing
NA
ANNUAL COST
Insurance
Maintenance and repair
Most
Driven Auto
M4a_l
M4b_l
i 2nd Most
Driven Auto
M4a_2
M4b_2
3rd Most
Driven Auto
M4a_3
M4b_3
Totals
M4a_t
M4b_t
105
-------
M5
Please tell us if your automobiles are equipped with the following features: (For
each automobile, please circle either "Yes" or "No" for your answer for each
feature.)
ForM5a_l toM5p_3:
1 Yes
2 No
-9 Missing
• NA
DO YOU HAVE THESE FEATURES ON YOUR
AUTOMOBILES?
Automatic Transmission
Sun roof/Moon roof
Air Conditioning
Compact Disc Player
Driver Side Air Bag
Passenger Side Air Bag
Side Door Air Bag
Anti-Lock Brakes
Two Doors (i.e., not 4 door)
Wagon
Convertible
Anti-theft/Recovery System
Cruise Control
Alloy Wheels
Leather Seats
Special Package (Sport, Limited, GTS, etc.)
Most
Driven Auto
M5a_l
M5b_l
M5c_l
M5d_l
M5e_l
M5f_l
M5&.1
M5h_l
M5i_l
M5j_l
M5k_l
M5i_l
M5m_l
M5n_l
M5o_l
M5p_l
2nd Most
Driven Auto
M5a_2
M5b_2
M5c_2
M5d_2
M5e_2
M5f_2
M5&.2
M5h_2
M5i_2
M5j_2
M5k_2
M51_2
M5m_2
M5n_2
M5o_2
M5p_2
3rd Most
Driven Auto
M5a_3
M5b_3
M5c_3
M5d_3
M5e_3
M5f_3
MSgJJ
M5h_3
M5i_3
M5j_3
M5k_3
M51_3
M5m_3
M5n_3
M5o_3
M5p_3
t
M6 Has anyone in your household bought any roadside assistance packages such as
AAA, in the last five years? (Circle the number of your answer.)
M6a
1
2
-9
Yes
No
Missing
NA
(If Yes, ASK M6a)
Approximately how much did/do you pay per year?
I paid/pay $ __ per year.
-8 Don't know
106
-------
-9
Missing
NA
I
107
-------
1VI7 We are interested in your perception of the likelihood of you having a fatal accident
while driving, compared to the average driver in the same type of automobile. On
the scale below, the likelihood that an average driver will have a fatal accident is
equal to 1.0.
Please compare yourself with the average driver and tell us how likely you think it
is that you will have a fatal accident compared to the average driver. (Please circle
the number in the middle of the ladder that best reflects your opinion. -8 indicates
Don't Know and -9 indicates Missing data.)
I
108
-------
Your Chances of Having a Fatal Auto
Accident Compared to the Average Driver
TWinr AS I IVFT V »
Ofl°/L fufnrr I ilrrlv ^
Sn0/n M/ifY* 1 ilrrlv *-
1ft%Mntr T itrlv >•
I
2n
i i
1.8
1 7
1 £
1 C
1.4'
1.3
1 f
1 i
i n
On
0.8'
0.7
0.6
OS
OJ
01
0-7
01 _
Ofl
^ AVFRAnir nufvim
^ lO0/. I t-CT T ilr-lir
« <;n% I rcc I ikrlv
^. _.. Ofl°
-------
Using the scale below, where the chance that an adult will die in a serious
accident is equal to 1.0, tell us how likely you think it is that an 8-year-old child
would die in an equally serious automobile accident compared to an average
adult. (Please circle the number in the middle of the ladder that best reflects your
opinion. -8 indicates Don't Know and-9 indicates Missing data.)
Chances of an 8 Year Old Child
Dying in an Auto Accident
Compared to the Average Adult
TWICF AS I IKTI Y »-
Oflfofi Mnrr T iVrlv ^
*KWn Mnrf- T itrlv fe
1 ft
— 1 *)
1 ft
1 T
1 £
1 S
\ A
1 -»
1 1
11
1 0
A O
0.8
n i
0.6
Oe
0.4
0.3
0.2
01
n fl
^ AVFR AHF ADTTI T
a 1 fin/. I ncc T il-i»lw
- *in% I r-cn T il'r*I%r
+ . OfVW, 1 rr-s; T iVrlv
M^-fc-r- T Tf l?T^7 AT* A T ¥
I
1VI9 Next, we are interested in your perception of the likelihood of a 70-year-old
person dying compared to an average adult when involved in a serious accident.
110
-------
Assume that they are riding as a passenger in the front seat of the same type of
automobile.
Using the scale below, tell us how likely you think it is that a 70-year-old person
would die in an equally serious automobile accident compared to an average
adult. (Please circle the number in the rhiddle of the ladder that best reflects your
opinion. -8 indicates Don't Know and -9 indicates Missing data.)
Mail: How Safe U Y
-------
M10 2
M10 3
Driven
Auto
2nd Most
Driven
Auto
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
I I I
2345
"1 \ 1 1 1 1 1 1
9 10 11 12 13 14 15
3'" Most ' "• ' ' '
Driven 123456
Auto
n i i i n i I I
8 9 10 11 12 13 14 15
Mll_vl (Version I):
Several promising safety features are being developed that would improve
automobile safety. Experts estimate that these features can reduce average fatality
risk per occupant by 1 or 2 steps on the risk ladder on the previous page for all
types of automobiles.
If safety features were added to your household's most driven automobile that
reduced the risk by 1 step, what would be the new fatality risk for that automobile?
Please indicate the step number that is 1 step below your answer to Q10 for your
household's most driven automobile. (Please mark an X on the line indicating the
new value for your most driven automobile.
ForMll_vl: -8 Don't know
-9 Missing
• NA
Most
Driven
Auto
10
11
12 13 14 15
Mll_v2 (Version 2):
Several promising safety features are being developed that would improve
automobile safety. Experts estimate that these features can reduce average fatality
112
-------
risk per occupant by 1 or 2 steps on the risk ladder on the previous page for all
types of automobiles.
If safety features were added to your household's most driven automobile that
reduced the risk by 2 steps, what would be the new fatality risk for that automobile?
Please indicate the step number that is t steps below your answer to Q10 for your
household's most driven automobile. (Please mark anXon the line indicating the
new value for your most driven automobile.)
For Mil v2:
-8 Don't know
-9 Missing
• NA
Most
Driven
Auto
10
11
I
12
I
13
I
14
I
15
t
113
-------
M12_vl (Version I):
How important to you would it be to have the fatality risk for your [MOST
DRIVEN AUTOMOBILE] reduced by 1 step on the risk ladder? (Circle the
number of your answer.)
Not at all Extremely Don't Missing NA
Important Important Know
12 34567-8-9*
Ml2_v2 (Version 2):
How important to you would it be to have the fatality risk for your [MOST
DRIVEN AUTOMOBILE] reduced by 2 steps on the risk ladder? (Circle the
number of your answer).
Not at all Extremely Don't Missing NA
Important Important Know
12 34567-8-9.
M13_vl (Version I):
Please imagine that when you purchased or leased your [MOST DRIVEN
AUTOMOBILE] you could have selected an automobile with additional safety
features but otherwise exactly the same. The annual fatality risk per occupant
would be decreased by 1 or 2 steps on the risk ladder, depending on the safety
features you selected.
What is the most extra you would have been willing to pay on the price of the
automobile to have the safety features that reduce the fatality risk by 1 step on
the ladder? (Please do your best to give a dollar amount; approximate answers
are fine. If you wouldn't be willing to pay anything extra, write $0.)
I WOULD HAVE BEEN WILLING TO PAY $ EXTRA FOR 1 STEP.
-8 Don't know
-9 Missing
• NA
M13a_vl (A) Other comments on payment amount.
t
114
-------
M13_v2 (Version 2):
Please imagine that when you purchased or leased your [MOST DRIVEN
AUTOMOBILE] you could have selected an automobile with additional safety
features but otherwise exactly the same. The annual fatality risk per occupant
would be decreased by 1 or 2 steps on the risk ladder, depending on the safety
features you selected.
What is the most extra you would have been willing to pay on the price of the
automobile to have the safety features that reduce the fatality risk by 2 steps on
the ladder? (Please do your best to give a dollar amount; approximate answers
are fine. If you wouldn't be willing to pay anything extra, write $0.)
I WOULD HAVE BEEN WILLING TO PAY $
-8
-9
EXTRA FOR 2 STEPS.
Don't know
Missing
NA
M13a_v2 (A) Other comments on payment amount.
M14_v1 (Version 1):
What is the most extra you would have been willing to pay on the price of the
automobile to have the safety features that reduce the fatality risk by 2 steps on
the ladder? (Please do your best to givq a dollar amount; approximate answers
are fine. If you wouldn't be willing to pay anything extra, write $0.)
I WOULD HAVE BEEN WILLING TO PAY $
-8
-9
EXTRA FOR 2 STEPS.
Don't know
Missing
NA
M14a_vl (A) Other comments on payment amount.
M14_v2(Vere/on2J:
What is the most extra you would have been willing to pay on the price of the
automobile to have the safety features 'that reduce the fatality risk by 1 step on
ttie ladder? (Please do your best to give a dollar amount; approximate answers
are fine. If you wouldn't be willing to pay anything extra, write $0.)
I WOULD HAVE BEEN WILLING TO PAY $
-8
-9
EXTRA FOR 1 STEP.
Don't know
Missing
NA
t
115
-------
M14a_v2 (A) Other comments on payment amount.
I
116
-------
M15 Below are some reasons why people choose the amounts they do when answering
Questions 13 and 14. Please read each statement and indicate whether you agree
or disagree. If you agree with the statement, please then indicate how much it
influenced your answer of how much you would be willing to pay. (Circle agree
or disagree for each statement, and then, if you agree, circle the number of your
answer.)
NOTE: This section was data entered as it was answered. There were many respondents
who filled in a rating but who did not indicated whether they agreed or disagreed. We
made no assumptions about what their answers Should have been because some
respondents who circled Disagree also circled a rating. Therefore, we entered only what
was circled.
MISa a
M15a
b
M15b a
M15b
b
MISc a
MISc
b
I could not afford to pay more for my automobile....
1 Disagree (SKIP TO M15b_a)
2 Agree
• NA
I could not afford to pay more for my automobile ....
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
• NA
I believe it is important to increase automobile safety.. ..
1 Disagree (SKIP TO M15c_a)
2 Agree
• NA
I believe it is important to increase automobile safety....
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
* NA
I don't believe that the safety features would actually save lives ....
1 Disagree (SKIP TO M15d_a)
2 Agree
• NA
I don't believe that the safety features would actually save lives ....
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
• NA
t
117
-------
M15d a
M15d
b
MISe a
M15e
b
I don't believe it is my responsibility to pay for automobile safety improvements. . ..
1 Disagree (SKIP TO Ml 5e_a)
2 Agree
• NA
I don't believe it is my responsibility to pay for automobile safety improvements ....
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
• NA
I was thinking more about the cost of the safety features than about the reductions in fatality risk
1 Disagree (SKIP TO M15f_a)
2 Agree
• NA
I was thinking more about the cost of the safety features than about the reductions in fatality risk
t
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
• NA
M15f_a I need more information before committing any money ....
1 Disagree (SKIP TO M15g_a)
2 Agree
• NA
M15f b I need more information before committing any money ....
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
• NA
MlSg^a Automobile safety is good enough now - improvements are not necessary ....
1 Disagree (SKIP TO Ml 6)
2 Agree
• NA
Automobile safety is good enough now - improvements are not necessary ....
1 Did not influence my answer at all
2 Moderately influenced my answer
3 Greatly influenced my answer
• NA
118
-------
Ml6 Is there anything we have overlooked? Please use this space for additional
comments you would like to make.
i
(Verbatim comments are located in a separate section of the User Guide.)
YOUR PARTICIPATION IS GREATLY APPRECIATED!
Please return your completed survey in the enclosed envelope or return to:
William S
-------
|