FINAL REPORT
ESTABLISHING AND VALUING THE EFFECTS OF IMPROVED
VISIBILITY IN EASTERN UNITED STATES
by
George Tolley
Alan Randall
Glenn Blomquist
Robert Fabian
Gideon Fishelson
Alan Frankel
John Hoehn
Ronald Krumm
Ed Mensah
Terry Smith
The University of Chicago
USEPA Grant #807768-01-0
PROJECT OFFICER: Dr. Alan Carlin
Office of Health and Ecological Effects
Office of Research and Development
U.S. Environmental Protection Agency
Washington, D.C. 20460
March 1984
-------
DISCLAIMER
Although prepared under EPA Cooperative Agreement #CR807768-01,
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
Page
I. Section 1: Introduction
1.1 Summary of Project Objectives 1
1.2 Economic Effects on Visibility 5
1.2.1 Economic Effects: Introduction 5
1.2.2 Visibility in the Eastern United
States Since World War II 5
1.3 Definition and Measurement of Visibility 17
1.4 Outline of the Report 21
II. Section 2: Expressed Willingness to Pay for Visibility
2.1 Overview of Section 2 25
2.2 Alternative Contingent Valuation Approaches 26
2.2.1 Overview of Section 2.2 26
2.2.2 The Process by Which Atmospheric Visibility
Acquires Economic Value 27
2.2.3 Strengths and Weaknesses of Contingent
Valuation 55
2.2.4 Conceptual Framwork for Contingent Valu-
ation 65
2.2.5 Structure of Contingent Valuation Instru-
77
ments
2.2.6 A Contingent Valuation Experiment 84
2.2.7 Conclusion 94
2.3 Alternative Econometric Approaches 103
2.3.1 Overview of Section 2.3 103
2.3.2 Tobit Estimation 104
2.3.3 Comparison of Empirical Results 114
2.4 Visibility Value Function ]_31
2.4.1 Overview to Section 2.4 131
2.4.2 Visibility in Household Production
2.4.3 Basic Properties of Visibility Valuation ^35
2.4.4 The Visibility Value Function 138
2.4.5 Empirical Estimation of Visibility Value ]_5]_
Function
-------
(ii)
Page
III. Section 3: Secondary Data Analysis of Visibility Valuation
3.1 Overview of Section 3 157
3.2 Outdoor Recreation 159
3.2.1 Swimming 159
3.2.2 Television Viewing 171
3.2.3 Baseball 175
3.3 Hancock Tower Valuation 179
3.3.1 Demand-Based and Contingent Valuation 179
3.3.2 The General-Choice Model 190
3.3.3 The Contingent Valuation Experiment 196
3.4 View-Oriented Residences 204
3.4.1 Contingent Values for View-Oriented
Residences 205
3.4.2 Estimates of the Values of Views and
View Characteristics
3.5 Auto and Air Traffic ^
3.5.1 Visibility and Air Traffic ^10
3.5.2 A Model of Air Traffic Responses to
Lowered Visibility 212
3.5.3 Visibility and Traffic Accidents 229
3.5.4 Analysis of Highway Casualties in DuPage
and Cook Counties 235
3.5.5 Summary and Conclusions 249
3.6 Effects of a One Mile Change in Visibility:
Comparisons of Willingness to Pay and Secondary
Data Results ^
IV. Section 4: Use of Results to Estimate Benefits for the
Eastern United States
4.1 Evaluation of Policy Effects on Visual
Range 255
4.2 Illustration of Method 256
4.2.1 Outline and Summary 256
4.2.2 Step A: Establish Hypothetical Policy Scenarios and
Estimate Visiblity Effects 25/
-------
(iii)
Page
4.2.3 Step B: Forecast Emersions Under the
Hypothetical Policy Scenarios 259
4.2.4 Step C: Forecast Spatial Distribution of
Ambient Air Quality 261
4.2.5 Step D: Estimate Visibility Effects of
Scenarios 258
4.2.6 Step E: Estimate the Value of Visibility
Benefits of Hypothetical Pollution Control
Strategies 263
4.3 Benefits of Hypothetical Policy Scenarios 266
4.3.1 Measurement of Physical Effects and
Willingness to Pay for Improvement 266
4.3.2 Aggregation of Physical Effects in
the Eastern United States 266
4.3.3 Aggregate Willingness to Pay for Visibility
Improvements in the Eastern United States, 1990—
Preliminary Estimates Subject to Revision
4.4 Summary of Project Approach to Visibility
Policy Analysis 27
271
-------
Section 1
INTRODUCTION
-------
1.1 SUMMARY OF PROJECT OBJECTIVES
While visibility is receiving increasing attention, it is still relatively
neglected as an attribute of the environment whose worth is important. Visibility
is a pervasive and inescapable phenomenon which is subject to both general and
periodic deterioration. The effects are significant to the individuals affected,
and extremely large numbers of people are affected. The relative neglect of
visibility as a subject of investigation appears to be due not to its lack of
importance, but rather to the fact that it is more difficult to value than many
other environmental attributes. Visibility is not explicitly bought and sold,
and the consequences of poor visibility are not as overt as illness and death.
Yet visibility affects the quality of life and is potentially important to well-
being.
Valuing visibility raises methodological questions to which recent contri-
butions have been made. The present effort utilizes and develops these contribu-
tions, enhancing their validity and accuracy. Previous work on visibility has
concentrated on sparsely populated areas of the West. The present research,
concerned with visibility in the Eastern United States, deals with larger numbers
of people under a wider variety of circumstances. People in urban and rural areas
are affected in the course of daily living, and a variety of special activities
centering on recreation and related activities are particularly sensitive to
visibility conditions.
Three major objectives have been accomplished by the research contained in
this Report. The first and most important result is the establishment of a visi-
bility value function. This function is the Project's basic contribution to the
analysis of visibility policy effects. Research was directed not at measuring
1
-------
2
the value of current visibility or any other specific value, but rather at
estimating the value of policy-induced changes in visibility. The generality
of the visibility value function permits estimation and comparison of benefits
from any set of policy alternatives.
The benefits of a visibility policy depend upon the extent of improvement,
on initial visibility conditions and their geographic distribution, and upon
social and economic characterisrics of people in various regions. Benefits are
a function of these variables in the visibility value function. Changes in
socioeconomic characteristics of the population will occur over time as well
as policy-induced visibility changes. The visibility value function accounts
for the separate and joint effects on benefits of changes in these variables
over time.
The second major objective was to identify particular activities likely
to be influenced by visibility and to measure the value of visibility to house-
holds in producing these activities. Recreational swimming and enjoyment of
residential views are among the wide range of activities investigated. Visi-
bility value functions for individual activities were derived. The individual
activity functions compliment the aggregate function in several important ways.
Theoretically, they are based upon information derived from transactions in
ordinary markets or from activity in implied markets. An important result is
that these studies corroborate the findings from the aggregate function, which
is based upon hypothetical behavior in contingent markets. First, the activity
functions consistently establish positive values for improved visibility in
individual markets. One example is that property values are observed to increase
with improved visibility. Secondly, the magnitudes of benefits in individual
-------
3
markets are plausible in relation to aggregate benefits.
The third major contribution of Project research was to establish a rigorous
and operational method of aggregating visibility policy benefits over the entire
Eastern U.S. From the beginning it was recognized that the visibility value
function, based upon contingent valuation, would be the basis for measuring
aggregate policy benefits. This is because it was not feasible to develop
individual value functions for all markets in which visibility is important.
The basic problem was to use a limited amount of information obtained from
contingent markets in six cities to measure visibility valuation in the entire
eastern U.S. Approximately 800 expressions of willingness to pay were obtained
for five visibility programs. Each program covered a specific geographic area
and offered a specific change in visual range.
An early empirical approach was to estimate a separate willingness to pay
function for each program in each city. Several aggregation problems resulted.
First, there was only one eastern U.S. policy program to use (along with the
endowment point) to fit the eastern bid curve. This was inadequate. Secondly,
there was no satisfactory way to estimate willingness to pay for improvements at
different distances from the bidder. One would have to resort to an expedient
like "average improvement over all eastern states" as an argument of a city's
eastern U.S. bid function. Thirdly, estimation of policy benefits required add-
ing values derived from local bid functions and values derived from eastern U.S.
bid functions. This was rather arbitrary in that local visibility improvements
and distant visibility improvements were treated as separate goods, rather than
as a single good which yields different service flows at different distances.
These difficulties were overcome by pooling all observations and estimating
a single function directly applicable to all bids, both local and region-wide.
-------
4
The resulting visibility value function permits direct aggregation of all policy
benefits based upon parameter values derived from a quite limited but carefully
chosen set of contingent market observations.
The spatial index is the feature of the visibility value function that
produces direct aggregation of policy benefits. The index expresses willingness
to pay for visibility in any location as directly related to the number of
square miles of improvements and inversly related to distance. Thus, the benefits
of a policy in a state in a particular year are a function of policy-induced
improvements in all states that year. Estimates of policy benefits take account
not only of the size but also of the complicated and changing spatial distribu-
tion of visibility improvement over time.
This report is a summary of a 32-month effort aimed at arriving at estimates
of the value of improved visibility for the Eastern United States. The project
was carried out under a Cooperative Agreement with the Environmental Protection
Agency, with active day to day participation by the staff of the Resource Analysis
Group of the Committee on Public Policy Studies of the University of Chicago and
the staff of the EPA, including Dr. Alan Carlin and others.
Austin Kelly of the University of Chicago and James Ciecka of DePaul University
served as consultants to the project.
The project was completed in two phases. The basic phase ran from Month 1
through Month 17, during which time detailed methodology was developed and visi-
bility situations examined for the Chicago area. The supplementary phase of the
project, running from Month 8 through Month 32, was devoted to examining six addi-
tional metropolitan areas and six non-urban cases.
-------
5
1.2 ECONOMIC EFFECTS ON VISIBILITY
1.2.1 Economic Effects: Introduction
The history of visual air quality in the eastern United States is essen-
tially a history of economic development of the region. The relationship be-
tween economic development and visibility has changed over the years in response
to changing technology, energy prices and other factors. A requirement of effec-
tive visibility policy is to alter the direction of these occurrences optimally.
Measurement of policy effects requires a knowledge of historical trends.
Policy evaluation requires that regulatory rules be modelled in proper relation-
ship to other factors, so that their partial effect on visibility may be isolated.
1.2.2 Visibility in the Eastern United States Since World War II
Examination of the path of visibility in the twentieth century provides
many insights into the short and long term factors which influence pollution and
visibility in the eastern United States.
Visibility trend data were initially used in the scenario-setting of the
contingent valuation (CV) portion of this study. Examination of the data
-------
6
immediately raised a difficult question: Just what is typical visibility
in these urban areas: Median visibility over the last four years was used,
buy a satisfactory answer to the question still requires some knowledge of
the history of visibility and its determinants in these cities.
Fig. 1-1 shows a seasonally-adjusted time-series of visibility in Chicago.
The vertical scale represents the difference between the month's median visi-
bility and the average median for the particular month over the entire series.
While this method is flawed, in that seasonal shifts have occurred in the pat-
tern of visibility, it is nevertheless useful in showing the distinguishing
features of the trend line, which has been smoothed somewhat using a modified
spline routine. Fig. 1-2 through 1-4 repeat the exercise for Atlanta, Boston,
and Cincinnati. Fig. 1-5 presents all four cities simultaneously, to aid in
regional comparisons. The major features are presented below. In Fig. 1-5
the vertical, broken lines occur at the midpoints of business troughs, while
the first solid vertical line occurs at the time of the OPEC oil price hikes
of 1973-1974. The second solid vertical line occurs at the Iranian Revolution,
which was accompanied by another round of oil cutbacks and price hikes. It
is important to note at this point that substitute fuels respond to oil price
hikes, as demand for them increases. Fig. l-6a shows a deflated (1972
dollars) schedule of several fuel prices, in energy equivalents, as well as a
quantity-weighted composite of all mineral fuel prices in the United States
since 1950. It is clear that economic activity and relative factor prices in-
fluence pollution and visibility. Any projections of future trends should
carefully consider these effects. As an example, Fig. 1-7 shows the trend of
visibility at O'Hare Airport in Chicago. This series is interesting in that more
-------
7
FIGURE
Med i an V i s i b i I i ty
Difference From
1-1
at Chicago-Midway
Sample Mean
i i » 1 ' 1
4 71..5 32.3 37.3 52-3 57 ¦ 3 72-3 77. z
Source: National Climate Center
Bureau of Labor Statistics
Note: Recessions are drawn at local troughs
-------
8
FIGURE 1-2
Difference From Mean Monthly Vis.
City = Atlanta
.4
'49
Recessi on
' 54
Recassi on
' 61
Reces s ion
Note.: From 1950 to 1978
total man-hrs. employed
in Atlanta increased 133
(in manufacturing)
'75
Recession
Iranian
Revo 1u ci on
Oil
Embargo
¦o—
4Q
45
50
30
5G
YEAR
73
-------
9
FIGURE 1-3
Median Visibility in Boston
Difference From Mean
-1
IS:
I
' 54
Recession
' 49
Recess i on
'75
Recession
'45
Rece.ssi on
I rani an
Revo 1uti on
_ '52.
Recession
Oil
Embargo
Note: Setv/een 1 550 and 1 978, total
Man-Hours employed in rnanufact-r-
in Boston declined by over 12Z
-7-1
40
45
30
o-
70
-------
10
FIGURE 1-4
Monthly Median Visibility in Cincinnati:
Difference from Month's Sample Mean
-------
11
FIGURE 1-5
'EAR
LEGEND: Broken vertical lines are U.S. Recessions. First solid line
occurs at oil Embarco. Second occurs at Iranian Revolution.
-------
12
FIGURE l-6a
Wcw P* Oom«*bealty Preducsd Mkrmnt Fum
M ibv U 0.J nuum fntm 1930 u»uutf I9H. San 1VTJ, i&ta poor
taiMWrtatrwar aipnn^^.
Tte nfM" nM «*oa>Mac srad* oJ one* awmirt 1.2 won 9m y*r
>«¦!¦ tt40 tad 1973. UWB roao« J4J stow u» JTU. >mac ta*
1 fn-\9T1 P*VM.a*ema* am err*mu AMui j*mof t^ fnif
t«iMMipa9a«MM(nA IttQttraac* IWhiwiWU
pgcfM yj taawa. la 1*7* tats pnee m £3.* pares*, —a tooi fro«
!773 tafvofs IfH taamtrt u, m mm nu of
Hoi bHiuuni aaJ anco f«3 «a aw«m« «f parem 9* yar
MM 19)0 AM t*n, «JM1 wemurt U.? jwroct a 1974. f«* 1775
Cbfe««a )9t7,iMiBMMpp«»rM<«rw«MMCMlgrai««49
FIGURE l-6b
Energy Consumption by P*frn«»Y S««rsy Typ*
-------
13
FIGURE 1-7
Monthly Median Visibility at Chicago-01 Hare:
Difference from Sample Mean, by Month, 1958-1981
5S S3 50 32 54 55 53 'C ?2 74 75 73
year
Sources: National Climatic Center
Bureau of Labor Statistics
Note: Recessions are drawn at the I oca I troughs
-------
14
recent levels are available, and have been added to the plot. The recession of
1975-1976 increased visibility. Following this is the recovery into 1978, when
visibility fell once again. In 1979, the oil price hikes again increased visi-
bility, and the 1980 recession followed soon thereafter. The quick recovery
from this recession is seen at the end (September 1981) of the series, and we
are confident that additional data would again reflect the business downturn
beginning in the final quarter of 1981.
This kind of historical analysis is primarily intended to explain the
short-run peaks and valleys of the observed series, but the method is equally
valid for longer time periods. As an illustration, the plot of median visi-
bility in Atlanta should be compared with the plot of employment in manufacturing
industries for the same city (Fig. 1-8). Atlanta was chosen because of its
dramatic pattern of growth. During episodes of rapid growth in the 1950's, and
again in the early 1970's, Atlanta's visibility declined appreciably. No doubt
this was also influenced by regional growth in general as well as local growth.
In almost all cases, a decline in employment was matched closely with an increase
in visibility. More precise econometric estimates of the effects of legislation,
fuel prices, and business cycles will aid in the prediction of policy benefits,
especially as more refined estimates of future fuel prices are developed. The
effects of legislation on visibility, and pollution in general, are difficult
to measure, as the 1970's also saw so much economic turmoil. Persons should be
cautioned against the indiscriminant use of two-year comparisons of pollutant
levels, as a look at these graphs clearly shows that the choice of end points
can be made to produce almost any trendline of pollution.
The best that can be said of typical visibility is that it is the level
of visibility which exists with a typical level and rate of growth of economic
-------
15
FIGURE 1-8
ATLANTA MAN-HOURS (THOUSANDS)
Total Manufacturing
manhrs
575
550
525
500
475
450
425
400
375
350
325
300
275
250
225
J
47-5
52 • 5
37.5
5 2-3
57.
72-5
77-5
-------
16
activity, typical fuel prices, wages, and prices of other production inputs,
and typical weather conditions. It is clear that it is neither valid nor in-
formative to base policy oriented pollution projections on trend data assembled
from spot readings taken several years apart. It is hoped that more reliable
projections will be made through careful econometric estimation procedures.
-------
17
1.3 DEFINITION AND MEASUREMENT OF VISIBILITY
Visibility is rooted in human perception. As atmospheric conditions change,
the human perception of distance, clarity, color, texture and contrast change.
An adequate notion of visibility, as related to atmospheric quality, involves
(1) relationships between atmospheric conditions and those atmospheric quality
attributes which are objectively measurable with scientific instruments, and
(2) relationships between measurable quality attributes and human perceptions
of visual quality.
Visibility traditionally has been defined as the relative distance at which
an object can be seen under the prevailing conditions; i.e., as the visual range.
Husar et.al. (1979) define visibility as the maximum distance at which an ob-
server can discern the outline of a black object. According to Trijonis and Yuan
(1978) the procedure commonly used to determine visibility is to observe markers
against the horizon sky, e.g., buildings or mountains during the daytime and
unfocused, moderately intense light sources at night. Markers are chosen whose
distance from the observation point is known. Prevailing visibility is the
greatest visibility that is met or exceeded around at least 50 percent of the
horizon circle. The procedure has two limitations. The measurement of visibility
is affected by the visual acuity of the observer and the quality of objects ob-
served. The latter leads to a systematic underestimation of daytime visibility
because the objects are rarely black as required by the definition. There is an
even greater problem with measurement of nighttime visibility because of the
variation in intensity of the light sources. This lack of standardization makes
accurate comparisons of visibility among different sites difficult, especially
for nighttime visibility. There seems to be reasonable confidence in comparison
of daytime visibility among sites probably because less variation in the charac-
-------
18
teristics of target objects is suspected. Visibility is the good that indivi-
duals value, measured in this Report in miles.
Natural scientists who are concerned with the relationship between visi-
bility and pollutants have found it convenient to study the "bad"—haziness or
lack of visibility. Haziness is increased by the presence of light scattering
and absorbing aerosals and gases and is proportional to their concentration in
the air. Trijonis and Yuan measure haziness by the extinction coefficient (B),
which is inversely proportional to visibility (V) in the following way:
(1-1) B = 24.3/V ,
where 24.3 is the Koschmieder constant, V is measured in miles and B has
the units (10^ meters) The relationship means that in a uniform atmos-
4 -1
phere with extinction coefficient equal to x(10 meters) , a b]_ac]<; ob-
ject against the horizon sky will be reduced to the threshold level of
contrast for the human eye at a distance of 24.3/x miles. It is the ex-
tinction coefficient that is used to determine the causes of haziness.
Both the extinction coefficient and visibility are used to describe air
quality patterns and trends.
In addition to visual range, important components of human percep-
tion of atmospheric visual quality include color and texture. These con-
cepts can be measured objectively as contrast, color and lightness, using
scientific instruments. Formulae have been developed to combine these con-
cepts into a single parameter called color contrast (Malm, Leiker, and
Molenar). Research in which personal interview subjects rated carefully
calibrated color slides and actual scenes for visual quality has established
that the relationship between color contrast and perceived visual quality is
linear and statistically significant. Other factors such as scenic beauty
serve as shifters, leaving the essential linear relationship between color
contrast and perceived visual quality intact.
-------
19
Several prominant patterns and trends are reported by Trijonis and Yuan.
First, visibility is rather low in the Northeast, ranging from 8 to 14 miles
typically. In the Southwest, visibility ranges from 30 to 80 miles. Second,
visibility is fairly uniform throughout the Northeast in that visibility is
only 2 or 3 miles less in urban than nonurban areas. Third, there is a sea-
sonal pattern in that visibility is now typically 2 to 3 miles lower in
the summer quarter than the rest of the year, especially for non-metropolitan
(urban/suburban and nonurban) locations. Fourth, over the period 1953 to
1972, visibility declined in the Northeast, -2 percent for metropolitan areas.
It appears most of the decline occurred early in the period.
Trijonis and Yuan explain the deterioration in visibility by an increase
in sulfates in the atmosphere. Sulfates tend to occur in the particle size
range of 0.1 to 1 micron, which is the size range that is optically most im-
portant. Despite the fact that sulfates comprise only 15 percent of the aerosal
mass, they account for approximately 50 percent of the reduction in visibility
in the Northeast. Through multivariate analysis of the extinction coefficient
Trijonis and Yuan find contributions to total extinction as follows:
Component
Contribution
Sulfates
TSP*
Blue-sky scatter
49
16
(background)
Nitrates
Unaccounted for
5
2
28'
*TSP is total suspended particulate other than sulfates and nitrates.
-------
20
The conclusion that sulfates are the primary cause of visibility reduction is
robust with respect to six different data sets and linear and nonlinear specifi-
cations. Physical modeling which relates sulfate reductions in one area of
the Northeast to visibility in the other areas of the Northeast—a distributional
concern—has been supplied by D.M. Rote of ANL, and is used in the policy
simulation chapter of this report.
-------
21
L-q- OUTLINE OF THE REPORT
Section 2 is "Expressed Willingness to Pay for Visibility." This is the first
major empirical part of the Report. Analysis is based upon data drawn directly
from contingent markets in six eastern cities.
The most important literature on contingent valuation is reviewed in 2.1.
Important extensions of this literature are made in design, reported here, of a
contingent valuation research project carried out in Chicago. The project made
a fundamental contribution to the main results of this Report.
In 2.2 it is argued that geographically dispersed visibility improvements
are substitutes. Empirical support provided for the theoretical argument. This
work was fundamental to the development of the contingent valuation instrument
and the visibility value function, which are the key elements of Section 2 research.
Alternative econometric approaches to estimating the parameters of the visi-
bility value function are discussed in 2.3. Tobit estimation, discussed in 2.3.2,
is applied to a contingent valuation study at Indiana Dunes State Park. Tobit
and probit specifications are compared with ordinary least squares in 2.3.3, in an
application to National Park Service data.
The visibility value function is presented and analyzed in 2.4. Drawing upon
the theory of household production, it is an empirical statement which summarizes
the information gathered from the contingent valuation work. Aggregate policy
benefits by state are derived by substituting mean state values for each of the
variables in the function.
Section 3 is the second major empirical part of the Report: "Secondary Data
Analysis of Visibility Valuation." "Secondary Data" includes information such as
prices and quantities determined in ordinary markets. The term also denotes infor-
-------
22
mation about behavior in implicit markets, such as increased probability of acci-
dent while driving at a slower speed under reduced visibility conditions. This
can be interpreted as in increased price of safety.
A brief description of each topic, and corresponding empirical results, are
given in 3.1. Section 3.2.1 analyzes visibility effects on outdoor swimming. A
theoretical model of visibility demand is developed and tested by means of several
regression specifications. In 3.2.2 and 3.2.3 the effects of changing visibility
on television viewing and baseball attendance are analyzed. The theoretical
foundation of these studies is the idea of visibility as a productive input which
households use to produce services that yield satisfaction. Relevant theory is
developed in the Conceptual Appendix.
Section 3.3.1 reports the development of statistical procedures for analyzing
Hancock Tower visitation, and estimates of consumer surplus from improved visibility
The Hancock analysis is continued in 3.3.2. Results of contingent valuation and
analysis of secondary data from the Tower are found to be in close agreement with
contingent valuation results of the kind reported in Section 2. This comparison
greatly strenghtens confidence that can be placed on both types of analysis employed
in this Report. In this study of the value of residential view quality and atmos-
pheric visibility, property value and contingent valuation estimates of visibility
were found to be compatible. Benefits estimates of improved view quality and
visibility are reported.
A model of consumer behavior under visibility constraints on air travel is
developed in 3.5.1, and a framework is provided for measuring the net costs of
lowered visibility on air travel in 3.5.2. The relationship between visibility
and highway accidents on metropolitan Chicago is examined in 3.5.3. Underlying
the quantitative estimates is a behavioral theory of choice in which drivers are
assumed to balance the risks of injury or death against travel objectives. Consumer
surplus estimates of visibility benefits are reported.
-------
23
Section 4, "Use of Results to Estimate Benefits for the Eastern United States,"
shows how the visibility value function can be used to derive dollar estimates of
policy benefits. Four alternative illustrative policies are analyzed. Each
policy produces a set of state-by-state visibility improvements to the year 2000,
as determined by the Argonne long range transport model. More stringent policies
produce greater visibility improvements, which are distributed unequally among
the states. The benefits received by a state are seen to depend not only upon
local improvements but also importantly upon improvements in all other states as
well. Benefit estimates for each eastern state in 1990 under the four hypothetical
scenarios are presented.
-------
Section 2
EXPRESSED WILLINGNESS TO PAY FOR VISIBILITY
-------
25
2.1 OVERVIEW OF SECTION 2
The major objective of Section 2 is to formalize an aggregate visibility
value function. This function is the central contribution of Project research
to the measurement of region-wide visibility policy benefits.
In Section 2.2, a general theoretical framework of visibility valuation
is developed. It pertains both to the contingent valuation work of Section 2
and the analysis of secondary data in Section 3. The theory and practice of
contingent valuation are then reviewed. Project contributions to this litera-
ture are explained in detail. The empirical data used in the Project were
gathered in conformity to the framework established in the section.
Section 2.3 is an investigation of econometric approaches to data analysis.
Section 2.4 presents the visibility value function and its underlying rationale.
-------
26
2.2 ALTERNATIVE CONTINGENT VALUATION APPROACHES
2.2.1 Overview of Section 2.2
The basic problem addressed in this Section is the gathering of reliable
data on maximum willingness to pay for visibility improvements by the contingent
valuation (CV) approach. Sections 2.2.2 and 2.2.3 give a critique of the
current state of CV literature, stressing issues that need special care in
visibility valuation. This is followed by a general theoretical model of
household production of visibility services, 2.2.4, in which visual air
quality and purchased goods are productive inputs. The household pro-
duction model and regional economic theory—spatial economics—underlie the
content of the CV instrument. Section 2.2, therefore, addresses the two
basic issues: what information is needed and how most effectively to obtain it.
-------
27
2.2.2 The Process bv Which Atmospheric Visibility Acquires Economic Value
2.2.2.1 The Conceptual Model
Atmospheric visibility is desired by households not so much as a com-
modity for direct consumption but rather as an input into the production of
things (variously called "commodities" or "activities") which yield satis-
faction. Thus, the "new" demand theory of Lancaster and the household
production approach of Becker are both relevant. Stoll, building on the
work of Lancaster and Becker, developed a conceptual model of the process
by which environmental resources yield satisfaction, and applied it to the
analysis of wildlife-related outdoor recreation. The following is a
modification of Stoll1s approach, specifically designed to recognize the
nonrival character of the good, atmospheric visibility.
Assume that the household seeks to maximize the satisfaction it derives
from the characteristics provided by the activities it produces. Activities
are produced by combining time with exclusive, priced goods, and nonexclusive
and/or nonrival goods. Thus both time and goods serve as inputs into activity
production. The process of producing activities is constrained by the house-
hold's activity production function (a mathematical depiction of its
consumption or household production technology) and by constraints on avail-
able time and income. Assuming, as does Becker, that time may be traded for
wages, these two constraints may be combined into a "full income constraint."
-------
28
Symbolically, the process may be depicted as one in which the household
maximizes
(2-1)
Subject to
U(c^ ..., c^,..., c^)
B N
(2-2) Z < S p * +T ?s+1) S S
j»l n-1
(2-3) ^ k = 1,2,... ,L
(2-4) z. -
w., ,E) j = 1,2,B,... ,J
J n = 1,2,...,N
(2-5) Cm ** Cm^Zj ,Wjk^ m-l,2,...,M
(2-6)
z. > 0
3 -
where c are characteristics; z. is an activity; z1( •••' 7 are nonwork
m ] 1 B
activities, and zB+L , are work activities; x n is a purchased input
whose unit price is Pn? is a nonrival good; r3+L is the unit wage rate for
the highest-marginal-wage work activity available; S is full income; is
the total initial endowment of nonrival good; and E is a vector of deter-
minants of the household's activity production technology at a given point
in time.
Constraint (2-2) is the full income constraint; (2-3) is a constraint
on availability of nonrival goods; (2-4) is a household activity production
function; and (2-5) is a characteristic production function depicting how
activities yield characteristics. To repeat, it is characteristics which
provide satisfaction. Note that enters both e^S- (2-4) and (2-5) . In
(2-4)the important point is whether is present in at least the threshold
quantity necessary to permit production of z^, (2-5), is recognized
-------
29
that, given that a is produced, the amount of characteristics it provides
depends upon the quantity of w available for use in its production.
]k
The level of satisfaction that the household enjoys may vary with full
income, prices of purchased goods, wage rates, production technology, and
the endowment of nonrival goods. Activity production technology in the form
of human capital may be acquired by the household and may depreciate over
time. The endowment of nonrival goods, e.g., atmospheric visibility, at
any location is determined jointly by background conditions and the aggregate
activities of mankind and thus may be influenced by public policy. By choice
of location, the household may influence the endowment of nonrival goods
available to itself.
Solution of the household's maximization problem yields implicit
prices (or opportunity costs), ir , for the various characteristics, cm.
Since these it depend on a particular household's activity production
function and full income constraints, they are , in principle, differ-
ent for each household. Furthermore, the it are affected by those
m
factors that influence the household's activity production technology and
its full income, the endowment of nonrival goods, and the price of purchased
goods.
The conceptual model of the consumption process has a number of
interesting attributes.
1. It recognizes both the role of time in the consumption pro-
cess (eq. (2-4)) and the consumer's choice in allo-
eating marginal units of time between work and non-
work activities (eq. (2-2)).
2. The role of activity production technology (eq. (2-3))
permits explanation of changes in consumption
bundles in the absence of changes in tastes, prices
of purchased goods, or endowments of nonrival goods.
-------
30
A change in activity production technology (e.g., the
acquisition of some specialized consumption or leisure
skill) may be sufficient to change the c , and
Indicators of household activity production tech-
nology would be expected to prove useful in explaining
variation in the WTP for (e.g., atmospheric visibility)
across households.
3. The two-step relationship between goods, activities and
characteristics (eq. (2-4) and (2-5)) permits more com-
plete understanding of the relationship between goods
which are substitutes or complements in consumption,
and the reasons why goods enter and exit the marketplace
(Lancaster.) If it is charactertistics which are demanded,
if various activities produce different (but, in some
cases, overlapping) vectors of characteristics, and
if changes in activity production technology change
the amounts of the activities which may be produced
from given quantities of purchased and nonrival goods, then
the process by which changes in prices or activity pro-
duction technology lead to substitution among activities
and perhaps the total elimination of some activities
may be completely understood. A set of general hypotheses
may be developed along these lines, testable in specific
natural resource and environmental contexts.
Thus, the model incorporates the possibility of
substitutes and complements for visibility. In the pro-
duction of safety characteristics for aviation, navi-
gation instruments may be excellent substitutes. In
the production of view characteristics for valued vistas,
the only available substitute, photographs taken by
another at a time when visibility was better, may be
quite poor substitutes.
4. These concepts may be used to more precisely
define activity value, expected activity value, option
-------
31
value, the expected activity value for the non-risk-
neutral individual and existence value, In our context,
if one or more valued characteristics may be derived
from one or more activities which are produced using
only w^, their value is the pure existence value for w ,
This model of the process through which the household derives satis-
faction from a non-rival endowment such as ambient visibility is useful for
several purposes:
-it permits the derivation of welfare impacts, in con-
sumer's surplus terms, of changes in the endowment of
a non-rival good, ambient visibility;
-in so doing, it provides a conceptual linkage between
contingent valuation methods, analyses of behavioral
choices, and valuation methods which use observations
from the markets in goods whose demands are systemati-
cally related to the demand for visibility;
-it identifies the relevant categories of variables for
use in bid equations to explain variation in individual
WTP for improvements in ambient visibility, thus in-
creasing the likelihood that regularities in WTP can
be documented;
-with its focus on the role of nonrival endowments in
the production of activities which yield satisfaction,
it provides a conceptual focus for a major section of
our research effort: analysis of the relationship be-
tween ambient visibility and the observed activity
production behavior of individials. This research is
a major, original contribution of our project. Previous
projects have, for the most part, confined their atten-
tion to contingent valuation and the analysis of rela-
tionships between property values and ambient air
quality (of which visibility is one characteristic).
-------
32
2.2.2.2 Welfare Impact and Consumer's Surplus
The following model derives expressions for the consumer's surplus
value of the welfare impacts of changes in the endowment of environmental
goods. These expressions are conceptually straightforward but quite lengthy.
So, for expository purposes, we will revert to a simpler model in this section
in which utility is a function of the endowed level of nonrival amenity (ambient
visibility) and a vector, _X, of ordinary, priced goods,
(2-7) U - U(W,X)
From this point, the valuation methods may be devised by either of two ap-
proaches .
1. The Income Compensation Function Approach
Define Y as the numeraire value of _X. The utility function, implicit
in prices, _P_, may then be represented as
(2_2) U - U(W,Y) = U[?(W,Y)] ,
where W is taken as initially fixed to the individual.
Using the income compensation function, u(w|w*,Y), which represents the
least amount of the numeraire the individual would require with W to achieve
the same level of utility as with W* and Y, a system of partial differential
equations may be derived for various reference levels of W,
(2_9) 9u(wh/,Y) _ ?[W,u(Wjw*,Y)] •
sw
-------
33
For a change in visibility from W' to W", where U(W ,Y) < U(W",Y), the Hicksian
compensating measure of the welfare impact for the individual's willingness to
pay (WTP), is
(2-10) WTP - Jw, P[W,uCW|W,Y)]dW.
An equivalent measure, the individual's willingness to accept (WTA), is
/•W"
(2-H) WTA -Jw, P[W,y(W|W",Y)]dW.
That is, both WTP and WTA are defined as areas under (different) Hicksian
compensated demand curves for W. WTP and WTA may be directly observed using
any technique which permits estimation of the respective indifference surfaces
passing through
(2-12) u' (W' ,Y) = U'(W",Y - WTP), for WTP, and
U"(W",Y) = U"CW',Y + WTA), for WTA.
Most contingent valuation (CV) methods, (including direct questions,
checklist questions, iterative bidding, and various experimental formats) are
designed to estimate (2-12). The theory is direct, undemanding in terms of the
analytical assumptions needed, and easily applied. The most serious challenge
in empirical application concerns data quality. Most CV methods are in principle
susceptible to some kind of strategic behavior. WTP and WTA data may also be
disturbed by outside influences. The principal challenge in implementation
of CV methods is to minimize (1) opportunities for strategic behavior and (2)
the incidence of noise in the data set.
-------
34
2.2.2.3 The Expenditure Function Approach
An alternative formulation of the same problem posits the utility func-
tion (2-7), in which X is a vector (x^, .. . ,x^, . . .x ) of ordinary, private
(i.e., exclusive, divisible, and nonrival) goods. Maximizing (2-7) subject to
a budget constraint, II p.x.= Y°, generates a set of Marshallian demand
i
functions,
(2-13) xi = xi CP,W,Y°).
The possibility that W is an argument in the demand for private goods (c.f.
eq. (2-4) and (2-5)) suggests that market data, prices and quantities taken,
for x^ may be used to reveal the welfare impact of changes in W. Let us
explore this possibility. First, we establish the theoretical equivalence
of the expenditure function and income compensation function approaches.
Then, we consider the implementation of the expenditure function approach.
The utility maximization problem yields ordinary demand equation (2-13).
The dual of the same problem minimizes expenditure, I p.x., subject to the
i
constraint that utility must be at least equal to some specified level, U.
Solution to the problem
niin £ p.x.
i r
s.t. U « U(X,W)
yields the expenditure function. Considering a proposed change in the avail-
ability of a nonrival good from W' to W", where U'(X,W) < U"(X,W")» the relevant
expenditure functions are, respectively,
-------
35
(2-14) E'CP^U'} and
E"CP,W,U") .
The derivative of any expenditure function with respect to any price,
p^,yields a Hicksian compensated demand function for x^. For the expenditure
functions (14), the compensated demand functions are:
ih-f = 3 E'/ 3 p£' E^ (P, W, U') and
(2-15) x _ _
' p
x^" - 3 ?'/ 3 p. - E|j
The inverse Hicksian compensated demand curves for W are given by
(2-16) -3E'/&w«E^ (P,W,U*) and
-3 E"/3^ = E" (P, W, U").
w —
Thus, the compensating and equivalent measures of the welfare impact of
the proposed change are respectively,
f2-17:
f W"
WTP= - Jw, E^ (P,W,U')dW, and
,W"
(2-18
W"
WTA= -Jw, E^ (?,W,U")dW.
Eq. (2-17) is, of course, equivalent to eq. (2-10) and similarly
eq.(2—18) is equivalent to (2-11). This alternative formulation, however, offers
the prospect of empirically estimating WTP and WTA without directly observing
(relevant points on) indifference surfaces expressed in (W,Y) space. Instead,
-------
36
under favorable conditions, it should be possible to estimate WTP and WTA via
appropriate manipulation of readily accessible market data for private goods,
expressed in forms suitable, initially, for estimating (2-13) . A number of
techniques have been developed to use this approach. Examples include methods
which analyze travel costs , property values, and hedonic prices.
Let us now consider the conditions under which these various approaches
may be effective.
2.2.2.4 Comparison of Approaches
a) Separable utility functions. If the utility functions is strongly
separable in W, i.e.,
(2-19). U(X,W) = Ux(X) + Uw(W) ,
then the demand functions for x, will all be of the form
i
(2-20) xi"xi(?,Y),
that is, completely independent of the level of W. Certain commonly used func-
tional forms for utility functions (e.g., the Cobb-Douglas and CES forms) have
this property, and Freeman (1979) argues that some important classes of environ-
mental amenities may in fact be separable. In such cases valuation methods based
on the expenditure function approach are without prospects, and valuation will be
performed with CV methods or not at all.
b) Nonseparability of x^ and W. In many cases, demands for may not be
spearable from W, as in eq. (2-13) . If such a system of demand equations has
been estimated and it satisfies the Slutsky conditions for integrability, it may
be possible to solve for the underlying expenditure function. If it is, eq. (2-17)
eq. (2-18) can be estimated and the value of W at the margin, of the welfare
-------
37
impact of a nomarginal change from W' to W", can be estimated by implicit
pricing methods. However, it is generally necessary to impose additional
conditions on the problem in order to solve the system completely (Maler,
1974). Two, often benign, assumptions that are useful are (1) weak com-
plementarity and (2) the existence of a perfect substitute.
Weak complementarity occurs if when the quantity of demanded
is zero, the marginal utility of W is zero (Maler, 1974). In such cases,
when W increases the demand for x, shifts out, and the value of W" - W' is
i
approximated by the integral between and This valua-
tion approach can be operationalized as long as demand curves approximates
the integral between Hicksian compensated demand curves (Willig, 1976;
Randall and Stoll, 1980).
The assumption of weak complementarity provides the basis for the
travel cost method of valuing recreation amenities (Clawson and Knetsch,
1966; Stevens, 1966) and the land value method of valuing increments in air
quality, view quality, and other residential amenities (Freeman, 1974; Brown
and Pollakowski, 1977). It should be noted, however, that Maler (1977)
expresses doubts as to whether the weak complementarity assumption is satis-
fied in the housing market or (by extension) in other markets frequently
used for implicit valuation of non-marketed goods.
A second approach is operational if we can suppose that some good x.
is a perfect substitute for W. If some and W are perfect substitutes,
while W and X"' Gs^ is not in X*') are independent in the utility and demand
functions, the marginal demand price of W reduces to the price of multi-
plied by the substitution ratio between and W (Maler, 1974; Freeman, 1979.
-------
38
This idea suggests that if there exist some x^ which counteract the effects
of pollution so that x^ are perfect substitutes for improvements in W, ex-
penditures on x^ provide evidence of the value of W. If the elasticity of
substitution between x, and W is less than infinite, this method would
1
underestimate the value of W. While this method has promise, we have yet
to find published studies demonstrating its successful application in empiri-
cal research.
c) Hedonic Prices.
Assume first that x^ and W are not separable in the utility
function. Second, assume that x, can be defined in terms of a vector of
1
characteristics C, = Cc., , • • •, c. ). Third, assume that a purchaser, j,
—i xx in
of good x^ can vary C^ by choosing a particular unit, x^j• That is, x^
is not the usual homogeneous good but a bundle of attributes as are houses
and automobiles. Finally, suppose that one of the characteristics in is
ciw> amount ® enjoyed along with :x. Therefore, as the consumer
selects, for example, a given house or car, the amount of residential air
quality he enjoys along with his house or the amount of safety he enjoys
along with his car is also determined. For any unit of x^, say c^, its
price,px , is
ij
<2"21) px.. = 9x. (cijl' 'cijw' 'Cijn^'
where p is the hedonic price function for If Px can be estimated from
xi 1 i
observations of the prices and the characteristics of different
"ij
x then the price of any xik, , can be calculated from a knowledge of
-------
39
its characteristics. The implicit price of the characteristic, c.. , for
r r 1]W
individual j can be found by differentiation:
(2-22) P = 3 p /3c...
1 ' c.. x. ijw
Ijv 1
Under favorable conditions, it is possible to use information in the
implicit price function to identify the demand for c. , that is, the demand
1W
for W if W is enjoyed only as a characteristic of x^. Assume the individual
purchases only one unit of (or, if more than one unit, only identical units)
and the utility function is spearable in x^ and (x^ is not in X"') so that
the marginal rate of substitution between any pair of x^ is independent of
X^ . Then, depending on the form of the characteristic demand function (Rosen,
1974), it is possible to estimate the inverse demand curves for W. In such a
case, the integral between the inverse demand curves for W' and W" would
approximate the intetral between the appropriate Hicksian compensated demand
curves (Willig, 1976; Randall and Stoll, 1980) .
In the brief period since publication of Rosen (1974), many attempts to
use hedonic prices to value nonmarketed goods have been initiated. Applications
have included many aspects of residential amenities (e.g., airport noise,
Abelson, 1979), and work place safety (Thaler and Rosen, 1975) . An literature
is emerging to identify and catalog the analytical difficulties this approach
encounters.
The priminary advantage of methods which use the expenditure function
apporach is data quality. Such methods use data sets of actual transactions.
CV methods, by definition, will never enjoy that advantage. However, that does
-------
40
not mean that the estimated values for W derived from expenditure function
approaches are necessarily valid or, for that matter, superior to estimates
using CV methods. When X. and W are strongly separable in the utility function,
these methods cannot be used. When (nonseparable) relationships between _X and
W are not of the most simple kinds, the analytical assumptions will be violated
to a greater or lesser degree, with corresponding deleterious effects on the
validity of the value estimates for W. Thus, while the data base is, in a sense,
real, the stringent analytical assumptions necessary to derive the value of
W from observations in the market for_X provide more than enough opportunities
for bias or noise to intrude. Our empirical research plan, therefore, pro-
vided opportunities for replication of value estimates with both CV methods
and methods which use various expenditure function approaches.
2.2.2.5 Econometric Specification of the Model
Herein, let us explore the implications of the above model for the
specification of econometric equations to explain individual WTP for
7^. The model implies that the satisfaction derived from a change in
the ambient level of visibility will be influenced by:
(1)—the array of activities produced using visibility; the charac-
teristics these activities provide; and the array of activities which do
not use visibility as an input, but which provide (some of) the charac-
teristics provided by visibility-using activities.
(2)—the prices of purchased inputs used in production of the activi-
ties discussed immediately above. Taking a long time horizon, one would
also be concerned with the availability at a particular time of pur-
chased inputs which may enter and/or exit the marketplace and with
-------
41
changes in input quality. In the static time frame, these would not be
considerations.
(3)—in a cross-section of households spatially arrayed across the
land surface, the array of Y^, endowments of nonrival goods, would be
expected to vary; and this variation will influence the productivity
of the activity production process. This suggests a focus on nonrival
goods, in addition to air quality, which are used in production of
visibility-using and nonvisibility-using activities which provide (some
of) the same characteristics.
(4)—the marginal opportunity cost of time to the household.
(5)—the household's activity production technology in general and
in particular as it applies to visibility-using activities and, non-
visibility-using activities which provide (some of) the same characteristics.
Technology can be expected to vary across households and one important
subset of technology, the things that contribute to visual acuity, may
vary within the household. In general, activity production technology
may be acquired and many depreciate, which is important in a longitudinal
time frame, but not in the static time frame.
(6)—the household's preferences across characteristics.
Economics has made little headway in using information about preferences
to explain individual household demand for purchased goods, or household
valuation of nonrival goods. The revealed preference approach by-passed
the fundamental question by taking it as axiomatic that purchases reveal
preferences. Time-series analyses of demand often resort to the use of
crude trend variables which are presumed to correct for secular changes
-------
42
in tastes (and anything else which may not be properly accounted by the
other, more precisely defined, independent variables). One could argue
that a significant trend variable should lead to the rejection of
the hypothesis that the model is adequately specified.
Becker has shown that, under certain plausible assumptions about
caring within the household, the household acts as though it is seeking
to maximize a single preference function. Stigler and Becker have argued
that, since economics has made such poor positive use of the notion of
preference (for the most part, being satisfied with negative uses
such as using it as an all-purpose copout to explain away otherwise
inexplicable results), progress might best be sought by assuming that
preferences are constant across households and across time periods, thus
ascribing behavioral differences to differences in opportunity sets and
activity production technology.
If the above-mentioned factors influence the satisfaction derived from
changes in the level of atmospheric visibility, WTP for these changes is
influenced, in addition, by
(7)—household full income.
(8)—the competing demands within the household, which may influence
the marginal and total WTP for characteristics that may or not be provided
by visibility-using activities versus WTP for characteristics always pro-
vided by non-visibility using activities. If this latter group of char-
acteristics is treated as a numeraire, then we are speaking of those things
that influence the marginal rate of substitution between the numeraire
and the group of characteristics that may or may not be provided by
visibility-using characteristics.
-------
43
In summary, eight cataegories of variables which may influence WTP
have been identified. Of these, we may a priori assign low priorities
to categories (2) and (6) : (2) on the grounds that unit prices of
homogenous purchased goods used along with visibility to produce char-
acteristics are unlikely to experience much variation in a static cross-
section; and (6) on the basis of the Stigler-Becker argument which
suggests an emphasis on inter-household variations in activity production
technology rather than preferences.
In the light of the preceding conceptual analysis, let us now con-
sider the variables traditionally used to explain variations in individual
WTP. To what extent do these variables capture precisely the kinds of
factors thought to influence WTP? Are the traditional variables addressed
to a single factor or to multiple factors. If to a single factor, is the
underlying relationship clear, unambiguous and fully specified? If to
multiple factors, are the various underlying relationships between these
factors and WTP unidirectional. (If not, a priori expectations will be
unclear, and the interpretation of results will be ambiguous.) Are there
variables and relationships that the conceptual model suggests are likely
of importance, but which are ignored by the traditional variables?
Below, the traditional variables are listed and for each, its in-
terpretation in terms of the factors identified by the conceptual model
is explored.
-------
44
Traditional Variable Category of Factors Influencing WTP
Income —(7), i.e., income addresses the notion
of "full income," but incompetely, since
it ignores the relationships between cur-
rent income, work and wealth.
Education —(5), presumably, better education
assists the acquisition of activity pro-
duction technology (APT), but this re-
lationship is unclear. Formal education
may be of little use in the acquisition
of outdoor APT's, and the time spent gaining
it may have come at the cost of time which
would otherwise be spent acquiring outdoor
APT's.
— Education may be a better indicator of
acquired technology useful in handling CV
exercises.
Age —(5), presumably. However, advancing age
implies the depreciation of certain APTs
while it may permit the acquisition of
others. For specific APT's, the relationship
between age and technology has yet to be
conceptualized.
— if the program (e.g., to improve visual air
-------
45
quality) is seen as one which requires the
passage of time, in order to achieve its full
effectiveness, advancing age may indicate
shorter time horizons (a problem our model
does not explicitly address) or pessimism
about the speed and effectiveness of program
implementation.
Race/Ethnicity —(5), if R/E or Sex determines propensity
Sex to acquire certain APT's. Does it? Which
ones?
— (1), if overt or subtle descrimination
removes some x's or z's from opportunity
sets.
Household Size —to some extent, an indicator of (8) .
Unemployed —(4), if it indicates a temporary change
in the marginal opportunity cost of time.
If unemployment is voluntary, it indicates
something more permanent about the res-
pondent's MOC of time.
—(7), temporary change in full income.
—(5), if unemployment frees up time for
the acquisition of APT's.
Rural/Urban
— (3), a crude indicator.
— (5), if R/U residence indicates something
-------
46
Location of residence
about opportunities to acquire APT's. In
this context R/U for the first two decades
of life may be a better APT indicator than
current R/U residence.
—(1), perhaps some Xs are available
in R but not U, as vice-versa.
*—Unfortunately, R/U may indicate different
beliefs about the state of nature with
respect to markets in environmental goods:
R may feel environmental goods should be
free and available in virtually unlimited
quantities, while U may not object to paying
for restricted quantitives.
— (3), perhaps a little better indicator
than R/U. However, location is unlikely to
identify all of the respondents enjoying a
particular Y^.
—(5), e.g., Florida residence increases
the travel component in the activity produc-
tion function for downhill skiing.
— (1) . Maybe some x's are unavailable in
some localities.
*These are considerations of how effectively a respondent uses a CV instrument
to reveal his true WTP, not the value of his true WTP.
-------
47
Water/Fish/Swim/Boat
(From RFF water quality
instrument)
— (5). However, it is crude, since it
fails to distinguish among e.g. different
fishing APTs. (A sociologist has iden-
tified 5 classes of trout fishermen;
perhaps he means people possessing 5 cate-
gories of trout fishing APTs.)
Walk along the Ridge?
(From U.C. Indiana Dunes
instrument)
—(5); but, which APT's?
--(4), maybe: Marginal opportunity cost of
time is low enough to permit walking.
Binoculars?
(From U.C. Indiana Dunes
instrument)
— (5)? Actually, it indicates the decision
to purchase a specific x.
Environmentalist
— (6), an "attitude" to the sociologist.
—(5), to a Stigler-Becker economist.
But which APT's do respondents associate
with the word "environmentalist? (After
all, it is self-reported?)
To summarize, these traditional variables provide the following
qualities of information in each of the 8 categories:
(1) Almost nothing. Every variable which may be interpreted in terms of
(1) has at least one other interpretation. None is yet specific to
any particular category of x's, z's, or c's.
(2) Nothing about input prices, but in a static, cross-sectional varia-
tion in input prices may not be especially significant.
-------
48
(3) Very little. Only R/U and Location address this issue, and both are
very blunt proxies.
(4) Very little. Only Unemployment and "Walk along Ridge?" address this
issue. The Latter, especially, is blunt.
(5) Several variables may address APT, but none is capable of addressing
specific categories of APT's precisely and to the exclusion of other
APT's.
(6) If you believe Stigler-Becker, (6) is a dead-end street, anyway.
(7) Income is addressed in money terms, but not full income terms.
(8) Only Household Size addresses (8), but it is a blunt indicator.
Further, many of the variables lack any clear a priori expectation
as to the sign or magnitude of the coefficient, and any clear interpretation
of empirical results in term of the conceptual model. This occurs in the
cases of variables which say address two or more of the categories, and
variables which address, e.g. category (5), but in no clearly-conceived
my (e.g. Education, Age, R/E, R/U).
2.2.2.6 Review and Summary
The discsussion thus far suggests that many previous CV exercises
may have encountered at least some of the following problems (or, at
least, may have been suspected of being susceptible to some of them):
1. Strategic bias: There is agreement that scope for strategic bias
exists but little evidence to suggest that strategic behavior is prevalent.
-------
49
2. Conservative/cautious initial response. That is, the kind of
unsure and unconfident initial reaction to new and radically different
hypothetical markets which may be the cause of WTP understatements noted by
Bishop and Heberlein.
3. Unsatisfactory bid equations.
a. small samples.
b. bids, themselves, may be poor quality data.
(i) the good being bid for may be incompletely perceived,
or perceived differently across respondents.
(ii) respondents may have difficulties arriving at what is,
for them, the optimal bid.
c. poor specification of bid equations.
(i) independent variables poorly defined.
(ii) independent variables imprecisely measured.
(iii) poor selection of independent variables, resulting
from inadequate conceptualization of the process
through which environmental goods acquire value.
-------
50
Of the 8 categories of variables which the conceptual model
suggests as likely to influence WTP for atmospheric visibility, five
seem especially important. Let us consider these five cate-
gories of variables, attempting to identify and define variables
appropriate for observation and use in WTP equations.
Full Income (7): Annual value of household consumption is important,
i.e., annual household disposable income corrected for saving or dis-
saving. However, gross annual household income is most readily observed.
Also important is net worth, since especially in higher age groups,
consumption is financed in part by dissaving.
Marginal Opportunity for Cost of Time (4): The expected wage rate
for one additional hour of work weekly is important. The question
must be worded carefully, to ensure that respondent does not inter-
pret it to mean "the reservation price for an additional hour of work."
-------
51
Competing Demands on the Household Budget (8): Household size is impor-
tant. It is also desirable to know the life cycle stage of the
household (young children, college students, aged dependents, etc.).
Endowments of Nonrival Goods (3) : Of particular importance is the definition of
bundle of nonrival goods available for consumption jointly with atmos-
pheric visibility.
a. big city/town/rural non-farm/farm.
b. coastal/mountains, hills/flatlands.
c. some indication of the variety and aesthetic quality of the vistas
encountered in the course of normal activity (at home, at work,
commuting, shopping, local recreation). Secondary evaluation
based on, say, zipcode, is not good enough, since within a lo-
cality different residential addresses, workplaces, and patterns
of activity will lead to different view exposures. More satisfying
than secondary evaluation is the self-reported subjective evaluation,
e.g. "in course of a typical week, would you say that the most attrac-
tive view to which you are regularly exposed are: spectacular?
more pleasant views than most folks get to see regularly? ordinary
views? worse than ordinary?
In a study-region-wide sample, it is useful to know whether
the respondent is concerned primarily with his own locality, or whether
his concern is geographically broader.
d. Do you expect to live here for the indefinite future?
or, do you expect you might move to a place selected because, among
other reasons, it is scenically attractive?
or, do you expect you might move, but the decision would be unrelated
to scenic concerns?
-------
52
e. Do you usually vacation
—at home?
—at a place where
—you spend most of the time indoors?
— outdoors, urban?
— outdoors, rural?
— outdoors at a place chosen.
among other reasons,for its scenic vistas?
Seasonal aspects of WTP for visibility, climatic
aspects (temperature, cloud cover, snowfall, etc.—secondary data) are
of interest in analyzing a broad cross-sectional sample.
Activity Production Technology (5) : Activity production technology may,
in concept, be observed directly,or indirectly via observation of purchased
goods used (x's), activities produced (z's), or characteristics enjoyed (c's).
a. Direct observation of APT's.
—visual acuity (is it "too much" to ask respondent to submit to a
simple eyesight test?).
--powers of observation: in the evening, if asked, do you think you
could accurately describe visibility conditions during the pre-
ceding daylight hours?
—knowledge of what is being viewed:--identification of features of
scenes, e.g. animal/bird/plant species, distant objects, geological
formations, etc.
—identification of location of U.S. scenes represented in photographs.
-------
53
—health and physical fitness (self-reported? enumerator evalua-
ted? ). Presumably this is a major element in APT" s for vigorous
outdoor activities which use visibility as an input.
—acquired skills: do you hold a pilot's license? have you ever
been recognized (e.g. by winning a prize or selling your work)
for landscape painting or photography? do you feel confident
doing the following things: rock climbing or mountaineering;
hiking through the back country; taking a good landscape photo-
graph; walking/running/bicycling long distances; cross-country
skiing?
b. z's produced
—list them all (data overload)
—indicate if you regularly engage in any activities in the
following categories:
strenuous outdoor—rural scenic (examples: hiking, biking,
backpack) .
—urban scenic.
—non-scenic (examples: tennis, team sports).
other outdoor —rural scenic (examples: picnicking, sunbath-
ing, flying, driving to enjoy
scenery) .
—urban scenic.
—non-scenic.
indoor view-
oriented —looking out the window.
—looking at collections of landscape
photography.
-------
54
c. x's bought
—binoculars, cameras with telescopic lenses.
—equipment for activities which use visibility as an input (it
could be a long list).
d. c's provided: Probably not much of value can be gained by
getting a list of the visibility related characteristics from which
respondents, derive satisfaction.
Visual, characteristics probably serve two purposes: (1)
a source of aesthetic pleaure, and (2) an indicator of the health and
comfort related aspects of air quality. Since it is important to isolate
the visibility affects from the health and comfort affects, it may be
useful to ask: indicate on this list the things you associate with
atmosphere conditions depicted in the (worst case) set of photographs
(list includes respiratory distress, poor color contrast, eye irritation,
poor long distance visibility, poor ventilation in homes, etc. in addition
to "placebo" and "decoy" items).
-------
55
2.2.3 Strengths and Weaknesses of Contingent Valuation
For more than a decade contingent markets have been used to elicit
individual valuations of unpriced (usually, nonrival and/or nonexclusive)
goods and services. The basic idea is that the researcher constructs a
model market in considerable detail and, in a survey or experimental set-
ting, communicates the dimensions and characteristics of that market to
the subject. The researcher specifies an increment (or decrement) in
some good or service and invites the subject to make a conditional dollar-
valued offer to buy (sell) the increment (decrement). The conditional
offer is contingent on the existence of the model market as structured
and communicated to the respondent; hence, the term contingent valuation.
However, the exercise does not involve the actual exchange of goods and
services for money.
Contingent valuation has several advantages, which seem likely to
encourage its more general use. (1) Contingent markets may be inexpen-
sively constructed and used by subjects (see, e.g., the argument of
Brookshire and Crocker, 1981) . Market structure and rules, and the quan-
tity and quality dimensions of the good or service involved, may easily
be manipulated in a conscious experimental design strategy; and such
manipulations need not be limited to the currently observed range of
market rules and quantities/qualities. (2) Contingent market data are
generated in forms consistent with the theory of welfare change maeasure-
ment (Bradford, 1970; Randall, Ives and Eastman, 1974; Brookshire, Randall
and Stoll, 1980). (3) Contingent markets do not rely on the actual delivery
of goods and services. Thus, their use is not limited to cases in which
delivery is feasible and convenient to the researcher.
-------
56
Other candidate techniques for valuation of unpriced goods do not
enjoy all of these advantages. Indirect methods of inferring value data
by observing actual markets in related goods (e.g., the travel cost, land
value, and hedonic methods) have considerable failings with respect to
points (1) and (2) above. The theoretical difficulties implicit in the
restricitive assumptions required to yield value estimates from these kinds
of observations should not be underestimated. Experiments with actual
markets for exclusive but not customarily marketed goods may sometimes
be contrived (Bishop and Heberlein, 1979) . Perhaps more opportunities
exist for incentive-compatible (Groves and Ledyard, 1977) laboratory
experiments in which groups of subjects contribute toward the purchase
of collective (i.e., nonexclusive and often, nontival) goods. However,
these kinds of methods are adaptable for value-revealing purposes (as
opposed to work with induced preferences, see Smith, 1977 and 1980) only
in cases when the direct and side payments can be actually collected and
the collective goods actually delivered--a restrictive condition.
The discussion thus far suggests that, if contingent valuation methods
were generally accepted as accurate, there would be little reason to use
other kinds of valuation methods in benefit cost analyses of programs that
provide unpriced goods. However, it has generally been assumed from the
outset that the accuracy and reliability of contingent valuation methods
is minimal. Two blanket criticisms were raised: (1) "everybody knows"
that hypothetical questions rarely enjoy accurate responses; and (2) "every-
body knows" that where nonexclusiveness or nonrivalry are involved, strategic
behavior is general, and the data collected are nothing but the pooled pro-
ducts of individual attempts to mislead the researcher.
-------
57
In spite of the pervasive skepticism engendered by these sweeping
criticisms, there has accumulated a body of evidence to the effect that
considerable real information can be generated in contingent markets.
In early applications, Davis (1963) and Randall, Ives and Eastman (1974)
obtained results which were plausible and which did not fail certain
(rather minimal) validation tests. The results of the last-mentioned
study were later replicated by Brookshire, Ives and Schulze (1976) and
Rowe, d'Arge and Brookshire (1980). Starting with Knetsch and Davis
(1966) recreation demand analysts have consistently demonstrated compara-
bility between the results of contingent valuation and travel cost methods.
More recently, Brookshire et al. (1982) have demonstrated considerable
consistency between results of hedonic analysis and contingent valuation.
Individual willingness to pay for nonexclusive or nonrival goods,
as revealed in contingent markets, exhibits some regularities. Many re-
searchers have found the theoretically expected relationships between
individual bid and income (among others, Brookshire, Randall and Stoll, 1980;
Mitchell and Carson), quantity of the good offered (Brookshire, Randall and
Stoll, 1980) and the availability of substitute goods (Majid, Sinden and
Randall) . Socio-demongraphic and attitudinal variables are sometimes signi-
ficantly related to bid (Brookshire, Randall and Stoll, 1980; Mitchell and
Carson) . These variables seldom account for a large proportion of the variance
in individual bids. However, when individual observations are grouped in
some way, to reduce the influence of outlying observations, much of the variance
1
m bids across groups can be explained (Brookshire, Randall and Stoll, 1980).
Nevertheless, some reasonable doubts about the accuracy and reliability
of contingent valuation persist. (1) The possibility has been raised that
-------
58
contingent markets in general, or in particular formats may be susceptible
to various biases. This line of thinking leads to a cataloging of potential
biases and empirical testing to determine the presence if any of the identi-
2
fied biases in particular data sets (Brookshire, Ives and Schulze, 1976;
Rowe, d'Arge and Brookshire, 1980; Schulze, d'Arge and Brookshire, 1981).
Some of these biases are merely problems to which all survey research is
susceptible, and sound research procedures are routinely available for their
avoidance (e.g., sampling and interviewer biases). Others are more inte-
resting: "strategic bias," "hypothetic bias," "starting point bias," and
"information bias." However, there is nothing compelling about the taxonomy
developed by Brookshire and his associates. Grether and Plott (1979) develop
a quite different taxonomy, in an attempt to explain apparent preference
reversal; and Mitchell and Carson quarrel with several aspects of the Brook-
shire _et _al_- discussion.
"Strategic bias" is fairly clear. It provides the basis for the main-
stream economic analysis of nonexclusiveness and nonrivalry; and it is
strategic bias the incentive-compatible mechanisms (Groves and Ledyard,
1977) are designed to thwart. The basic idea is that when the consequences
of truth-telling are more costly to the individual than those of some pre-
varicating stretegy, truth-telling inevitably gives way to strategizing.
Since most contingent markets provide disincentives for free-riding, the
most likely strategy is for an individual to bid in a way which exaggerates
the difference between his true bid and his expectation of the sample mean
bid, so as to move the sample mean bid toward his true bid. Pervasive behavior
of this kind would increase the variance of a sample of bids, in the extreme
producing a bimodal distribution. Given a minimum acceptable bid of zero (for
-------
59
an increment in a positive-valued good) but no a_ priori maximum limit,
such behavior would bias sample mean bids in an upward direction.
"Information," "starting point" and "hypothetic" biases are not so
clear. In the hands of Grether and Plott (1979) these concepts merge to
3
become Theory 8: the notion that, in the absence of good reasons to
care about the consequence of their responses, subjects minimize invest-
ment in information processing and decision making by clutching at any
"anchor" provided in the question format. As it turned out, Grether and
Grather and Plott experimentally rejected Theory 8 by finding that intro-
ducing real incentives (reasons to care about consequences) did not diminish
apparent preference reversal. In contingent valuation, there is little
evidence of the general occurrence of "information" and "starting point"
bias. Rowe, d'Arge and Brookshire (1980) claim to have found both kinds
of bias in a single data set, but that finding appears to be the exception
rather than the rule. The interpretation of "information" bias is contro-
versial, since significant changes in the information provided to respon-
dents must change the quantity/quality definition of the good being offered
or the structure of the contingent market. Thus, a finding that changes in
information generate changes in bids can seldom be unambiguously interpreted
as a finding of bias. Often, it shows a rational response to a change in
the situation posited, and provides more reason for comfort than alarm.
While Schulze, d'Arge and Brookshire (1981) argue that "hypothetic"
and "strategic" biases are opposite sides of the same coin--contingent mar-
kets which give subjects less reason to care are susceptible to "hypothetic"
bias while those that offer more reason to care are susceptible to strategic
influences--Mitchell and Carson attempt a more subtle distinction. They
-------
60
suggest that both kinds of bias can be simultaneously minimized by constructing
realistic contingent markets but reassuring subjects that actual bids will not
be collected during the experiment.
"Hypothetical bias," if it occurred, would increase the variance of bids.
Given a lower limit of zero for acceptable bids but no upper limit, its in-
fluence would also be in the direction of overestimating true sample mean bid.
(2) A second attack on the efficacy of contingent markets focuses directly
on the size of the value estimates obtained. Mitchell and Carson appear to
be stating that conventional wisdom when they claim that contingent markets
generally overestimate the true sample mean value of the nonexclusive and/or
nonrival good under consideration. However, there is surprisingly little evi-
dence to support this position. Bohm (1972) found a small upward bias when
payemnts were hypothetical, but Mitchell and Carson question his interpre-
tation of the evidence. Babb and Scherr (1975) found no evidence of bias in
either direction. Brookshire_et _al (1982), in a comparison of hedonic and
contingent valuation results, found good correspondence. A close examination
of their analysis suggests that, if the contingent valuation results deviate
at all from the true values, that deviation is almost surely on the downward
side. Bishop and Heberlein (1979) compared contingent valuation results with
those of a willingness-to-sell experiment in which actual exchange was consumated.
They reaffirmed that contingent willingness to sell (in situations where selling
is not customary or morally acceptable in the real world) leads to substan-
tially larger value estimates than contingent willingness to pay--a well-
established finding. Of more interest, they also found that contingent willing-
ness to pay yielded considerably lower value estimates than actual willingness
to sell--a finding which they interpret as showing that contingent WTP
4
substantially underestimates true value.
-------
61
The evidence seems to suggest that the conventional wisdom is
unsupportable. There is almost no evidence that contingent WTP over-
estimates true value, but there is some evidence to suggest underesti-
5
mation.
(3) A third source of doubts about the efficacy of contingent markets
focuses not on mean sample bid but on the frequency of extreme bids.
Starting with Randall, Ives and Eastman (1974) researchers routinely
separate "protest bids" (that is, those zero WTP on infinite willingness
to accept, WTA, bids which the subject identifies as a protest against
some aspect of the contingent market structure) from the sample of bids
prior to calculating the sample mean value estimate. The frequency of
protest bids in various contingent markets has ranged from less than ten
percent of all bids to more than fifty percent (Mitchell and Carson); so,
it appears that the structure of contingent markets influences the quality
of data obtained. While the literature contains less discussion of "high
bids," most researchers find a few scattered respondents bidding a substantial
fraction of annual income for increments in a single nonexclusive or nonrival
good. While there exists no perfect test for strategic bids, most researchers
take one of the following two courses: reject all bids above some arbitrary
maximum, expressed as a dollar amount or a fraction of annual income; or
reduce all high bids to the arbitrary maximum. The first approach arbitrarily
treats all high bidders as dissemblers. The second grants some plausibility
to high bids and, rather than disenfranchising high bidders, seeks to limit
their influence on the sample mean bid. While we can be less certain that
high bids are poor-quality data than we can about protest bids, contingent
valuation researchers tend to treat both kinds of bids as unreliable and focus
-------
62
their analysis on those bids which are identified as neither protest bids
nor "too high."
This approach, incidently, parallels Smith's (1980) discussion of
his experiments, in which he treats zero-bidders as free-riders and
endowment bidders as anti-free-riders (p. 396).
Let us attempt a very brief summary of what is now known about
contingent markets.
1. Contingent markets are not incentive-compatible, but strategic
behavior does not seem to be pervasive among human beings asked to con-
tribute toward providing collective goods (Marwell and Ames, 1974; Smith,
1980; Sweeney, 1973). That does not mean that strategic behavior never
occurs, just that there appears to be a substantial class of decision con-
texts in which a good many people do not behave strategically.
2. Contingent markets do not deliver the goods and collect the payments,
but that does not necessarily render them wildly unreliable. The data sets
collected via contingent valuation have, for the most part, performed fairly
well in those quality tests which have been applied to them. This finding is
consistent with the result of Grether and Plott (1979), who found that the
introduction of real consequenses for their subjects did little to change
decisions those subjects made in experimental contexts.
3. Contingent markets collect some "junk data": protest bids, for
sure, and presumably some of the high bids. However, they appear to collect
a solid core of serviceable value data. These findings are entirely consis-
tent with Smith's (1980) experimental results.
4. Analyzing this solid core of serviceable data, we find no evidence
that it consistently overestimates true value. If anything, the evidence
-------
63
points to underestimation. In addition, individual bids are to some
extent regular and predictable. In short, the solid core of data generated
via contingent markets is neither fanciful nor random.
5. The structure of contingent markets does appear to have some
(perhaps limited) influence on the value data generated. This ought not
be surprising in principle--the performance of real-world and actual-
experimental markets is influenced by their structure--but it is an appro-
priate subject for further investigation.
The remainder of this section reports some preliminary results of an
experiment designed to explore two aspects of market structure: (1) the
number of distinguishable commodities offered for bid and the sequence in
which offered, and (2) the process in which bid data rea collected.
An extensive contingent valuation pilot study for the visibility
project was consciously designed to permit, inter alia, experimental testing
of the effect of contingent market structure on the characteristics of the
bids generated. The general objective was to empirically explore the two
apects of contingent market structure identified in the preceding paragraph.
We proceed as follows. A conceptual framework is developed and specific
empirically testable hypotheses are generated there from. Data collection
procedures are briefly described. Analytical procedures consistent with the
conceptual framework are introduced and used in hypothesis testing. Some
preliminary results are presented and briefly discussed.
-------
64
2.2.4 Conceptual Framework for Contingent Valuation
Consider a household which at any time is producing a simple activity
selected from the vector Its activity production function is
(2-23) Yi = Yi(xi» qi5 a), 0 = y^O, a),
where x- is a vector of priced goods with prices p. , q, is an unpriced
l —l l
nonrival good and a is the household's activity production technology.
If is the probability that the household is producing y^, and i is
limited for convenience to the values 1, 2 and 3, and y refers to other
goods, the indirect utility function is
(2-24) v(p, q, it, a, m) = max uOj ' ^\ + ^2 ' y2 + ^3 ' Y3' ^
3
subj .to y + £ p. X; =
i = l
and Yi^xi' qi' = Yi •
Using duality and the expected utility property,
3
(2-25) c(p, q, it, u) = min y + I p. x-
i=l ~1 —L
3
subj. to I -t. qi, a) , y] .
i = 1
< CO
*i = 0 '
there may exist prices p^ at which the household would choose to set x^ and
Y_^ equal to zero.
With the expenditure function defined, consider a change in the level
of provision of nonrival good^ q^.
Letting the utility function be specified such that 3u/3x^
-------
65
(2-26) 9e/3qi = -yiTi3u/9Yi ' 3T±/3qi
= 5i(p, qj it, a, u) £ 0 .
While the conceptual framework for contingent valuation is often
derived via an income compensation function approach (section 2.2.2.2),
it is possible to proceed via the expenditure function. For the moment,
suppress a (which is used below in the empirical analysis) and it (which
is of more interest in analyses explicitly directly toward option price,
(see Schmalensee, 1972, and Graham, 1981), so that
e(p, q, u) = e(p, q, ir, a, u) .
At an initial situation (p° , q°) , the household requires m° =
e(p°, q°, u°) to attain u°. If the level or provision of a single environ-
mental good q^ changed to q^, the minimum expenditure to attain u° would be
m' = e(p°, q', u°)
The welfare impact of that change, in compensating surplus terms
(Randall and Stoll, 1980) is
(2-27) CS(q?, q') = e(p°, q\ u°) - e(p°, q°, u°)
= e(p°, q', u°) - m° .
Locating e(p°, q, u°) in the real plane with (p°, m°) as the origin,
e(p°, Qr u°) describes the indirect version of the familiar Bradford (1970)
bid curve.
Now consider in all three nonrival environmental goods, i = 1, 2, 3.
For clarity, we express q = (q , q , q ) as
-------
66
e(p, q, u) = e(p, q^ q2> q3> u) .
For a change from q° = (q°, q^> Qp to " = ( qp '
(2-23) CS(q°, q") = e(p°, q", u°) - e(p\ q°, u°)
= /q 3e(p°, qx, q2, q3> u°)|3q dq ,
q°
c Cq)
where C(q) denotes some path from q° to q".
Choosing a particular rectangular path from q° to q", say (q^, Qp
to (q^, q^, qp to (q^, q^> Qp to (qj^ q2> Qp » the line integral (2-28) can
be transformed to the sum of several ordinary integrals,
(2-29) CS(q°, q") 5 e(p°, q", u°) - e(p°, q°, u°)
(2-29.1) = /ql 3e(p°, q^ qq3> u°)/3q1dq1
qi
(2-29.2) + /q2 3e(p°, q[, q2> q|, u°)/3q2dq2
q2
(2-29.3) + /q3 3e(p°, q'v q£, q3> u°)/3q5dq3 .
q3
An alternate rectangular path from (q^» q2> Qp to (q^> Q2> Qp
to (q^> q2> Qp to (Q^> ^2' Qp results in the same aggregate valuation
as in (2-29):
(2-30) CS(q°, q") = e(p°, q", u°) - e(p°, q°, u°)
(2-30.1) fql 3e(p°, qr q^3 q£, lO/aq^
ql
-------
67
(2-30.2) "+ {q2 3e(p°, q°, q2> q£, u°)/3q2dq2
q2
.(2-30.3) + /q3 3e(p°, q°, q?, q3, u°)/3q3dq3
q3
However, unless 3^e/3q^3q= Of (249.1) ^ (2-30.1), (2-29.2) f (2-30.2)
and (2-29.3) j- (2-30.3). Thus, we have
Proposition 1: The contribution of an increment in a single to the
value of an increment in the q vector from q° to q" varies with the
2
sequence of valuation, unless 3 e/3c^_.9cK = 0.
2 2
Further, if 3 e/3q23q1 > 0 and a ^e/3q33q1 > 0 (i.e. and and
7
and q^ are substitutes ) the contribution of q^ to the value of an
increment in the q vector will be greater, the earlier q^ appears in
the valuation sequence.
Identities (2-29) and (2-30) suggest that, in general, it is erroneous
to value a change from q? to q!^ and a change from q? to Rj independently
and then calculate the value of a simultaneous change from [q^, q! ] by
simple addition. Suppose q^, q^, an^ ^3 are substitutes. If we were
to proceed as if the valuations of the individual changes were independent,
we would measure
(2-31) V(q°, q")
(2-31.1) = e(P°> ^2' q3' U°'i " e(P°' ql* q2' q3' u°3
(2-31.2) + ql' q2' q3' U°^ " eCp°' ql' q2' q3' U°^
(2-31.3) + e(p°, qj, q°, u°) - e(p°, qj, q°, q°, u°) .
A well-conceived valuation would recognize the non-independence of
q^, q2 and q^ select a policy path (for example, the path in eq.
(2.29)), and obtain
-------
68
(2-32) CS(q°, q") = e(p0, q", u°) - e(p°, q°, u°)
(2-32.1) = e(P°> qj, q^ u°) - e(p°, q°, q°, q°, u°)
(2-32.2) + e(p°, q^, qq°, u°) - e(p°, q^, q°, q°, u°)
(2-32.3) + eCp°, qj, q^, q^, u°) - e(p°, q', q£, q°, u°) .
In (2-31) and (2-32), only lines (2-31.1) and (2-32.1) are equal. In the
case of substitutes, (2-31.2) is larger in absolute value than (2-32.2) and
(2-31.3) is larger in absolute value than (2-32.3). Thus we have
2
Proposition 2 : If 8 e/Sq^Sq^ £ 0, the value of a change in the vector q
is not equal to the sum of the independently estimated vales of the changes
in the elements of the vector.
2
Further, if 3 e/3q^9qj > 0 for all i t j, the value of a change in the
vector q is less than the sum of the independently estimated values of the
independently estimated values of the changes in its elements.
By identifying appropriate valuation and aggregation procedures, (2-29),
(2-30) and (2-32) provide important restricitons on the design of contingent
8
valuation exercises. In addition, they provide an explanation for pheno-
mena observed but not well explained in previously reported contingent valua-
tion studies (e.g., Schulze et_ al., 1981b; and Walsh _et _al., 1978) . In
these studies, authors report with some surprise that environmental goods
valued later in a valuation sequence are not valued as highly as had been
predicted.
-------
69
Competitive and complementary relationships arising from price changes
are frequently observed. It is important to consider the possibility that
competitive and complimentary effects are absent or weak for changes in
non-rival goods. A possibility is the case where non-rival goods are addi-
tively separable in the utility function. In this case, Proposition 1
applies. Let preferences of an individual be represented by an additively
separable utility function,
I K
u mY_ '
'!=1 k=l
where x]^=(x]^g ) i-s a G-dimensional vector of market goods, q^= (q^)
is an H-dimensional vector of non-rival goods, ke{l,...,K} indexes
subcategories of market and non-rival goods used in v^, the v., are each
increasing and strictly concave with non-negative second-order cross
partial derivatives, and for k^fe {1,. . . ,K) . Let
=min px ^
el
¦ • j." i\
x
u =zZ •
I-l k=l
Then the following properties hold:
(1) For non-rival goods in different subcategories (k f) the
substitution relationship is competitive ( kh r>0' all h and r)
(2) For non-rival goods in the same subcategory the substitution
relationship may be either competitive, independent, or complementary
0, all h and r) . ^
Proposition 1 demonstrates that independence in valuation does not
arise from additive separability. Indeed, the case of additive separability
between non-rival goods results in unambiguous competitive effects.
-------
0
Where additive separability cannot be assumed, competitive and comple-
mentary effects are both possible. Complementary effects may outweigh
competitive effects. Less likely is the case where competitive and comple-
mentary effects just cancel and result in independence in valuation.
Given the implications of Proposition 1 it is useful to consider the
empirical circumstances that may justify additive separability between
non-rival tools. Below, we examine two possible cases: the first where an
individual enjoys equivalent activities each affected by different sets of
non-rival goods and the second where future use is uncertain. These illu-
strative cases are easily linked to common benefit cost contexts. Thus
interpreted, Proposition 1 provides an a priori prediction of competitive
effects.
Consider the first case where the household production technology for
activity i is not specific to a particular site or region k. Market goods
x^, and non-rival goods q^, available at site or region k, enter as inputs
into the production technology and (x]<-' 3^) Within a given time
period total activity production of type is a simple summation over all
k
visited sites or regions k, a. . • If preferences are defined
i=l
over a similar time period (say, a month or a year) utility can be written
(2-33)
u = u[a., a(x,u)]
K
i
-------
1
where a(.) is a vector of other activities, x is a vector of market goods,
and 0) is a vector of non-rival goods specific to a (. ) • If activities a_,
are broadly defined and do not directly and strongly affect the enjoyment of
other activities (32u/3a3a.=r (a constant)), then utility is approximated by
K
(2-34) u == + ;'a(x>u) '
1=1
where , is a vector of ones conformable to a(x,u)). On grounds of convenience,
additive separability as in eg. (2-34) is a common assertion in both
economic theory and econonometrics (Deaton and Muelbauer). Moreover, in
this case of equivalent activities over different sites or regions, additive
separability has strong intuitive appeal. For instance, enjoyment
of slack-water recreation at site k is not likely to be directly affected by
water guality at site m; snowskiing activities at site n are not likely to be
10
directly affected by the slopes available at site p.
A second source of dominating additivity comes from the rationale underlying
option demand and option price. Consider a simple case where an individual faces
the future possibility of either recreating within the region of residence or
visiting one of two unique but distant recreation areas. By unique we mean that
activity production technology is peculiar to the recreation itself. For an
easterner, candidate areas might be the Grand Canyon National Park or Yellowstone
National Park; for a westerner, the Maine coast or the Florida everglades. If
the areas are indeed distant and quite costly to visit relative to home region
alternatives, the probability of future use is likely to be small and dominated
by exogenous random elements rather than explicit individual choice. With
-------
2
probabilities of visitation parametric to the individual at the time of valuation,
the expected utility model can be meaningfully applied.^ Supposing the con-
ventional additive utility structure over time, expected utility in future
period t is
3
(2-35) UrUt^*tk° 2tk(xtk'V .
3 t=l
where denotes a lottery over the three described possibilities, k=l, 2, 3,
t=l
and it ^ is the probability that in time period t recreational activity z^ is
chosen. For simplicity, suppose there is only one future period and that we
can therefore suppress the notation t. Using the expected utility property,
3
(2-36) u = ^
k=l
J
= Y_ Vk'Vk' ¦
k=l
3
where ^ denotes arithmetic summation. Thus, the case of parametric
k=l
uncertainty leads to additive independence between activities and
respective non-rival goods by a fairly direct route.
Proposition 1 is straightforwardly translated into the two valuation
contexts detailed above. In the context of equivalent activites at
different sites or in different regions, let v^(.)=a^ (.) and let the
v^(.) , ...,Vj(.) equal the respective 1-1 elements of a(x,w). Subcategory
indexes conform to the site-or region-specific indexes of the market and
-------
73
"ition 1, then, non-rival goods used
erent regions are competitive in
me activity at the same site or
or complementary. To translate
n visitation, let K=3 , .),
The subcategories index services specific
j ivalent activities, non-rival goods
ions but may be either competitive,
thin the same region,
sties of a given choice context can lead to
activities and categories of non-rival goods in the
additive separability between activities in the utility
independence in valuation. Quite the contrary. Given a
ilevel of some non-rival good, an individual maintains
educed expenditure by shifting activity production
more productive activities and away from the relatively
thout direct complementary effects, activities
rival goods become relatively less productive. As individuals
y from these less productive activities, the value of associated
s declines. Thus, where non-rival goods are additively separable
onstrained expenditure minimization imposes strictly competitive
cross-qu^ ty valuation effects.
Propositions 1 and 2 provide the basis for a major empirical hypothesis
to be tested in the experiment reported below. Nonindependence and the
associated question of valuation sequence constitute one of the questions
of contingent market structure. The other question concerns the process in
-------
4
which value data (individual bids) are collected.
The literature reports a variety of ways to collect bids. Published
studies have used devices ranging from a single direct question (e.g., Hammack
and Brown, 1974), iterative bidding routines (e.g., Randall, Ives and Eastman,
1974), checklists (e.g., Schulze et al. 1981b) and payment cards (e.g. Mitchell
and Carson) . Considering this array of devices, we identify two important
dimensions of the value data collection process: (1) the extent to which it
provides the opportunity to iterate toward the maximum WTP (i.e., the points
of indifference between paying WTP and obtaining the good, and doing neither) ;
and (2) the amount of value-relevant or price-relevant information provided in
the format. The payment card device (Mitchell and Carson) provides information
on the cost per typical household of various public programs now in effect. A
modified payment card developed by the authors provides additional information
on typical annual expenditures for various market goods. Considering these two
dimensions of the value data collection process, we propose the set of hypotheses
2, below.
The experiment reported below was designed to test the following hypotheses.
Hypothesis 1: The estimated value to Chicago residents of a specified atmos-
pheric visibility program for the Grand Canyon is greater if measured independently
than if measured last in a sequence which first considers programs for Chicago
and all of the U.S. east of the Mississippi.
This hypothesis is derived from proposition 1.
Hypothesis 2: (a) The quality of value data is improved by the use of devices
which permit more opportunities to iterate toward maximum WTP.
(b) The quality of value data is improved by the use of devices which
provide a greater quantity of value-relevant (or price-relevant) information
to assist the respondent in decision making.
-------
5
We offer no hypothesis concerning the trade off between opportunity to
iterate and the provision of value-relevant information.
To operationalize hypotheses 2(a) and (b), measures of value data
quality must be defined. We propose the following measures:
(i) The larger the solid core of serviceable value data in a data set,
the higher its quality. That is, the higher the frequency of protest bids
and "too high" bids, the lower the data quality.
(ii) Since strategic and hypothetical influences both seem likely to
increase the variance of a value data set, lower variance in individual bids
is taken as an indicator of a better data set.
(iii) Increased regularity and predictability of a value data set is taken
as an indicator of better quality. Thus, data sets which yield better bid
equations are taken to be of higher qualtiy.
(iv) Since the evidence appears to tilt toward the conclusion that contin-
gent markets underestimate sample mean values, any data set which exhibits
unusually low mean bid (relative to the other data sets) is taken as of poor
quality.
-------
6
2.2.5 Structure of Contingent Valuation Instruments
As described above, both region-wide and special, geographically
limited contingent valuation studies were carried out. The region-wide
or general study instruments were of modular design to facilitate pre-
testing and the coordination of the general and special studies. There
are seven basic modules to the general study instrument.
Module 1: Area Context Module
The area over which visibility improvements were offered were required
to be clearly comprehended by each individual. For the research to provide,
among other things, guidance as to sub-regional allocation of resources for
air quality improvement, it was important to collect WTP data for improve-
ments in visibility (i) in the individual's home sub-region, and (ii) in
the whole study region. Thus, for different purposes, the area context
differed increasing the burden of communicating the area context to subjects.
-------
Since the eastern region is larger than the customary territorial range
of individuals, a map card as well as a portfolio of photographs were used
to convey the size and diversity of the region over which visibility is
valued.
Module 2: Visibility Module
The nature of alternative levels of visibility can best be communicated
via color photographs. This required a set of scenes representative of the
area over which visibility changes were to be valued. For each level of visi-
bility a set of the same scenes , with only the visibility different, was used.
Some purely factual verbal material (on cards, and delivered orally) was used
to quantify the visual range represented in each photo set. In order for WTP
for visibility improvements in both the home sub-region and the whole study
region to be elicited separately, separate photo sets were needed to repre-
sent both the sub-region and the entire East.
Module 3: Activity Module
Since we conceptualize V. (w ) as the value of visibility as an input
1 jk
in the production of it had to be hypothesized that = f(z_^...). To
test that hypothesis, it was necessary to know the following:
1) the activities produced in the household,
2) the inputs, other than visibility, used in activity production,
3) the activity production technology used, and
4) whether visual air quality is the only air quality input used
and, if not, whether visual air quality is used by the subject
as an indicator of other aspects of air quality, For example,
the individual may avoid strenuous outdoor sports on days of poor
-------
8
visibility, not because visibility per se is an important
input, but because he treats poor visual air quality as an
indicator of high pollutant concentrations which treatening
respiratory stress.
The activity module was vital to the estimation of equation (3) . In addition,
the module served to sensitize the individual to the full variety of activi-
ties in which he might value visibility, thus eliminating possible sources of
underestimation of V.. A complete breakdown of all relevant activities would
have been time-consuming and would have generated more data than could
effectively be used in statistical analyses. Therefore, at the pre-test stage,
considerable effort was allocated to devising and testing ways to more effici-
ently serve the basic purposes of this module.
Module 4: The Market Module
Contingent valuation established a hypothetical market and encouraged
individuals to reveal their WTP by using that hypothetical market. Thus,
the structure of hypothetical market was a major influence on the quality
of WTP data. Major elements of this module described what was being purchased
through the bid and the market rules regulating payment for and receipt of
the good in question. To describe the good available for purchase, the general
level of visibility as well as possible increments and decrements in visibili-
ty were portrayed in both photographs and narratives. Market rules provided
assurance that the increment in visibility would be delivered if and only if
the respondent was willing to pay. At the pre-test stage, alternative versions
of the market module were developed and tested for their effect on bidding
behavior.
-------
9
Module 5: The WTP Data Collection Module
This module presented the fundamental WTP questions. In the Chicago
research, questions were structured in several different ways. The first
simply asked for a statement of WTP for some given improvement in visibility,
the second used checklists of possible values from which a number representing
maximum WTP was selected. The third used an iterative bidding format (e.g.,
Randall, et al, 1974). The fourth format presented information on relative
tax prices of other public sector goods and then called for a statement of
WTP for an increment in visibility. In this approach, the relative prices
of other public programs served as reference points for the respondent.
Intensive pre-testing of WTP modules context was carried out. New WTP
module designs were developed and tested. The most important modification to be
introduced during the pre-test was the marginal bid question. Respondents
bid first on local improvement, and then were asked how much they would add
to their local bid to extend the improvement to the East and then to the entire
U.S.
Module 6: Post-Bid Probing
With certain market rules and WTP formats, some individuals recorded
a zero WTP which, in further questioning, turned out to be a protest against
some aspect of the format rather than an accurate reflection of the value
of the good offered. Probing of zero WTP's was, therefore, a routine element
of the data collection schedule.
Even with protest bids eliminated, it has recently been shown that
WTP data generated by individuals who are in some way uneasy with the market
rules and WTP format exert a highly significant downward influence on mean WTP
-------
80
(Brookshire, Ransdall, and Stoll). Thus, it was necessary to provide op-
portunities for subject to confidentially evaluate the WTP instrument for
credibility/plausibility and their own responses as valid WTP indicators.
These evaluations were taken into account in developing the CV instrument
used in the six eastern cities.
Module 7: Socio-Demographic Data
This module collected an array of socio-demographic data used to
estimate equation (3) . It has been argued (Second Quarterly Progress Report,
Exhibit C) that full income concepts are highly relevant to the processes
through which individuals demand and hence value, visibility. Thus, questions
have been included in the CV instrument to capture the concept of full in-
come and collect the appropriate data.
Implementation of Contingent Valuation
Following completion of those special studies which were designed to
serve as pre-tests and pilot studies for the general study, the general study
instrument was finalized. A region-wide data set was assembled during the
winter of 1981 and analysis was completed during by January 1983.
Special studies address key issues in the design of effective contingent
valuation devices. Two objectives were served: (1) the selection of thoroughly
tested contingent valuation devices for use in the general study; and (2) the
generation of experimental data sets which permitted formal comparison of the
effectiveness of contingent valuation devices under consideration for use
in the general study and additional devices used in previous research. Thus,
this phase of the research design was intended to permit advances in the
implementation of contingent valuation.
-------
81
Formal experiments compared alternative systems of disincentives for
strategic and hypothetical biases, and alternatives WTP data collection for-
mats. The latter effort tested the four basic formats identified above, a
fifth format combining formats (3) and (4), and two experimental formats new-
ly devised during the current research. The two new formats were, resepec-
tively, an "interative bidding with budget breakdown and reiteration" format,
and group decision format utilizing linked computer consoles.
This work permitted (1) the first rigorous test of hypotheses about
the efficieincy of a wide variety of WTP formats, (2) the selection of one,
well-validated, WTP format for use in the general study, and (3) by selecting
for study some visibility values in specific markets, also examined via
secondary data analyses, the completion of test for corroboration and repli-
cation of CV results with behavioral data.
In addition to formal experiments, a series of informal studies using
open-ended questioning, content analysis, and similar techniques were used
to explore a series of important issues in instrument design for the general
study. The purpose of these informal studies was to gain an understanding
of citizen's perceptions in order to permit more effective communication
with the general study subjects, and to develop more effective ways of obtain-
ing important and/or sensitive information. Informal studies explored:
how citizens conceptualize visibility, and the effectiveness of
color photographs in communicating visibility to them.
whether visibility is best presented in typical or in frequency terms.
-------
82
the activities Z... , for which visibility is an input; in what sense
ijit
is it an input, i.e., in what ways does poor visibility hinder activity
production; is it a major or minor input; is visibility used by citi-
zens as an indicator of other air-pollution-related problems, e.g.,
respiratory stress; in order to reduce data collection time and data
overload, can meaningful categories of activities be developed?
are there effective ways to gather information about activity pro-
duction technologies (e.g., acquired outdoor skills) and complementary
inputs (especially, specialized consumer durables), again without data
overload.
particular versions of the wording of modules 4 and 5 can be examined
for effectiveness of communication and comprehension.
can the notion of full income (which includes income, the marginal
wage rate, and wealth] be implemented without an unacceptable number
of refusals to answer particular questions?
-------
83
2.2.6 The Chicago Contingent Valuation Experiment
2.2.6.1 Basic Contingent Valuation Structure
Following a small-scale pretest, a major pilot study was conducted to
generate contingent valuation estimates of the value of atmospheric visibility.
This pilot study was conducted by personal interview in the city of Chicago
and suburban Cook and DuPage counties. The basic instrument contained sections
for collection of the following data:
--Indicators of attitudes toward environmental quality.
--Activities of respondent (categorized as indoor-outdoor, strenuous or
otherwise, etc.); identification of activities for which the respondent had
invested in acquiring specialized skills or knowledge; identification of
activities which are avoided for health, etc. reasons; and identification of
activities the respondent was more likely to do on days when visibility was
unusually good, and those he was less likely to do on poor visibility days.
--Ownership of or access to, equipment which could be used in activities
which also use visibility (e.g., cameras with telescopic lens, binoculars, etc.).
--Contingent valuation modules that describe three alternative levels of
visibility in the immediate Chicago region; one alternative level in the much
broader east-of-the-Mississippi region; and one alternative level at the Grand
Canyon. Verbal descriptions and color photographs were provided. Visual
range in miles were stated and contingent market rules were defined. Respondents
were given the opportunity to re-examine all 5 bids and adjust any or all of
them. Protesters were identified—for example, respondents who objected to
citizens bearing the costs of environmental clean-up. Six interchangable CV
modules were used, each differeing only in the process by which bids were
collected.
-------
84
--Time horizon, with respect to expected length of residence near
Chicago or east-of-the-Mississippi.
--Homeowner or renter status, estimated rental value of home, and
rental income from other residential real estate owned.
--Quality of view from the place of residence.
--Socio demographic information about respondent and other household
members, including income, wealth, average and marginal wage, and income
expectations, as well as age, sex, education, race, ethnicity, etc.
A randomized cluster sampling design was developed, with a cluster
size of six and specific instructions that each CV module be used once and
once only within a cluster. Sixty starting locations were randomly selected
using a computer routine which (after eliminating high density neighborhoods
where interviewers would have trouble gaining access to apartments) gave every
citizen in the region an equal chance of having his residence selected as a
starting location. Thus, the target sample size was a maximum of 60 (and a
minimum of 50) interviews with each CV module, for a total of at least 300
and no more than 360 interviewers.
2.2.6.2 Alternative Formats
The six contingent valuation formats used varied only in the process via
which WTP bids were collected. They were:
A^ directly asked respondents to report their maximum WTP, as Hammack
and Brown (1974) had done in a mail survey.
A^ stated an amount, invited acceptance or rejection of the program at
that price, and then asked maximum WTP. This format duplicated the procedure
used by Bishop and Herberlein (1979) to collect contingent WTP.
A^ was an iterative bidding routine similar to those previously used by
Randall, Ives and Eastman (1974) and Brookshire, Randall and Stoll (1980),
among others.
-------
85
B allowed respondents to indicate their maximum WTP by checking
the appropriate number on a checklist of possible numbers. This format
had been used by Schulze et al (1981b).
provided a payment card, as developed and used by Mitchell and
Carson.
expanded the payment card concept to include typical annual household
expenditures (by income group) on several categories of goods purchased in the
private sector, as well as typical annual household costs of public programs.
As one progresses from A^ to A , there is successively more opportunity
to iterate toward the point of indifference between (1) paying the amount
stated and taking the good and (2) paying nothing and foregoing the good.
Formats and C0 provide information on the current levels of household expen-
diture on other goods and public programs; Cprovides a greater array of such
information than C . Format B has been promoted by Schulze et al (1981b) as
speeding-up the data collection process relative to, say, A^ and eliminating
the possibility of starting point bias.
2.2.6.3 Results
A data tape containing results of 273 completed interviews was used. While
the target was 300 to 360 interviews, a few aborted interviews had to be dis-
carded and a few stragglers had not been completed, coded and added to the data
set. All analyses reported below are based on this set of 273 observations.
Let us look first at the effect of value data collection format. Hypo-
thesis 2(a) suggests that formats A , A and A are expected to generate value
data of highest, medium and lowest quality, respectively. hypothesis 2(b)
suggests that formats C0, and B are expected to generate data of highest,
medium and lowest quality, respectively. There is no a priori hypothesis about
relative value data quality across the two sets of formats.
-------
86
All three A formats and format B generated noticeably more protest
bids than the C formats (Ta.2-1). The differences in generation of high
bids were not so noticeable. However, the C formats clearly generated a
larger solid core of serviceable value data than the A and B formats. Examining
this solid core (the 4 rightmost columns of Ta.2-1), we notice that formats A^
and B produced notably lower sample mean bids, and C2 produced notably higher
sample mean bids than the others. Within the solid core, there is little to
be observed with respect to dispersion of bids. If one considers for example
the mean bid relative to its standard error, the formats do not perform very
differently.
Since the format subsamples are small (fewer than 50 bids in every case,
and as few as 31 solid core bids in the case of A^), it is important to control
for differences in household characteristics across the sub-samples OLS regres-
12
sion analysis was used for this purpose. Two regression specifications
suggest themselves for estimation: the familiar linear-in-levels specification
(2-37)and an alternative specification (13) developed below.
The linear-in-levels specification posits
(2-37) WTP (q ?, q'.), = b + Ib.Z., + e
j J k o x xk
where k=l, ..., K refers to individual households; is a vector of
descriptors of the household's endowments, consumption technology, etc.;
are estimated parameters; and e is the error term.
Since one would suspect that (2-27) is likely to be non-linear, an alter-
native non-linear specification was developed. Rearranging (2-27) and entering
the vector of human capital endowments a, we obtain
(m + CS)/m = 3(p, q', a, u)/e(p, q°, a, u)
If u can be approximated by a homothetic direct utility function, the above
-------
TABLE 2-1
Value Data, Atmospheric Visibility, Chicago 1981, by Format.
Format
Sample
Size
(n)
Zero
All
(% of n)
bids
Protest
(% of n)
High
Bids
(% of n)
Mean Annual Willingness tc
Full Sample
WT P 9 C WTP10d WTP116
Pay per Household (Stand. Error of Mean)
Solid Core
n WTP 9° WTP 10d WTP11
A1
47
15
15
21
278
300
380
37
250
250
236
i
(191)
(116)
(145)
(51)
(50)
(50)
A2
45
24
18
11
140
136
157
35
156
147
171
X.
(26)
(22)
(24)
(30)
(22)
(24)
A3
45
22
18
18
312
299
329
31
222
210
240
(133)
(132)
(133)
(37)
(38)
(39)
B
46
22
15
24
9 8
8 8
150
36
121
109
152
(21)
(18)
(34)
(25)
(22)
(29)
C1
45
8
2
13
296
250
322
42
210
186
234
JL
(66)
(61)
(74)
(44 )
(35)
(53)
C2
45
4
0
16
425
446
560
42
283
324
456
-------
88
equation can be approximated by a normalized version e,
(2-37) (m + CS)/m = e(p, q', a, 2.|q°) ,
which describes the proportional reduction in minimum expenditures due
to the change in q as a function of prices, subsequent q', household
characteristics and an error term I —all conditional on the reference
level of q, q°. If (2-37) can be further approximated by a multiplicative
form, the following log linear form can be specified:
(2-38) ln(l + CS/m), = b irZ.^i exp(Zb.d.)e ,
k. o l j ]
where are dummy variables.
Results Of estimating models (2-36) and (2-38) for WTP11 are presented
(Ta.2-2 and 2-3, respectively).
Household standard of living, respondent's age, a grade 12 or lower
education, and the environmental index clearly influenced WTP11 in the
expected directions (Ta.2-2) . Using format A3 as a basis for comparison,
only format appeared to generate significantly different solid core bids.
Turning to the non-linear specification (Ta.2-3), we find the numbers of
adults in the household and the wage rate exerting significant influence,
along with several of the same variables which were influential in (2-36) . How-
ever, no format generated a sample of bids significantly different fromA^.
Our conclusion is that, for the most part, the choice of format seems to exert
13
statistically insignificant influence on the solid core bids.
In summary, it is clear that formats and C2 elicited fewer protest
bids than the other formats. Beyond that, little else is yet clear with respect
to hypotheses 2(a) and (b) and the performance of the alternative formats.
-------
TABLE 2-2
Estimated Bid Equation, WTP11, Using Specification (11)
Dependent Variable:
WTP11a f RATIO 3.04
DFE 180 PROS>F 0.0007
R-SQUARE 0.168 4
PARAMETER STANDARD
VARIABLE DF ESTIMATE ERROR T RATIO PROB>|T|
INTERCEPT
SOL
RYOUNG
RSENIOR
QHIGHS
QGRAD
ENVIR
CITPAY
A1
Ao
1 172.4
1 3.9
1 -90.1
1 -90.0
1 -82.2
1 -40.6
1 7.9
1 74.4
1 11.0
1 -52.2
1 -51.0
1 4.4
1 170.0
Independent Variables
112.7
2.4
63.9
79.2
59.9
81.1
3.0
55.1
96.3
91.8
97.3
91.6
91.2
1.52
1.58
-1.41
-1.13
-1.37
-0.50
2.58
1.34
0.11
-0.56
-0.52
0.04
1.86
0.127
0.115
0.160
0.257
0.171
0.616
0.010
0.178
0.908
0.570
0. 601
0.961
0.064
SOL
Annual household income divided by the Lazear - Michael
(1980) index of standard of living.
RYOUTH ~ 1 if age of respondent < 35 years.
0 otherwise.
RSENIOR = 1 if age of respondent >_ year.
0 otherwise.
QHIGHS = 1 if highest level of education of respondent, head, or
spouse of head of household is a high school diploma or less
0 otherwise.
QGRAD
1 if highest level of education of respondent, head, or
spouse of head of household is one or more years beyond a
bachelor's degree.
0 otherwise.
ENVIR
an environmental attitude index estimated for each individual
on the basis of observations obtained in section 1 of the
interview.
CITPAY
1 if respondent stated that citizens should pay the cost of
environmental improvement.
0 otherwise.
-------
90
TABLE 2-2. Continued
B, c , C. = 1 if an observation from a given format.
1
= 0 otherwise.
aWTPll is willingness to pay for an improvement in visibility from
9 to 30 miles. Sample includes solid core responses only.
-------
91
TABLE 2-3
Estimated Bid Equation, WTP11, Using Specification (13)
Dependent Variable:
Percent*
VARIABLE
DF
DFE
PARAMETER
ESTIMATE
159
STANDARD
ERROR
F RATIO
PROD F
R-SQUARE
T RATIO
2.53
0.0014
0.2126
PROB T
INTERCEPT
LNWAGE
RYOUNG
RSENIOR
QHIGHS
QGRAD
ENVIR
CITPAY
HA?
HAo
HCX
HCo
hc3
A1
A0
1
4.5771
0.00594
770.10
0.001
1
0.0031215
0.00166
1.87
0.063
1
0.0018667
0.00251
0.74
0.459
1
0.0074137
0.00354
2.09
0.037
1
-0.003111
0.00239
-1.30
0.195
1
0.0012065
0.00310
0.38
0.698
1
-0.0002132
0.000129
-1.64
0.101
1
-0.0003975
0.00220
-0.40
0.684
1
0.0105
0.00305
3.45
0.001
1
0.0103
0.00369
2.81
0.005
1
0.0076425
0.00325
2.42
0.016
1
-0.001909
0.0029
-0.65
0.516
1
0.0625697
0.00330
0.77
0.437
1
-0.0009201
0.00386
-0.23
0.812
1
0.0035101
0.0037
0.93
0.349
1
0.0037209
0.00395
0.94
0.348
1
-0.001814
0.00371
CO
0
1
0.626
1
0.00065131
0.00364
0.17
0.858
2
LNWAGE = Natural log of the respondent's marginal wage.
HA-
= 1 if household includes two members whose age is greater than or
equal to 18 years.
0 otherwise.
HA^ = 1 if household includes three or more members whose age is greater
than or equal to 18 years.
0 otherwise.
HCi
1 if the household includes one member of less than 18 years of age.
0 otherwise.
HC2, HC3 are similarly defined for households with 2, and 3 or more members
less than 18 years of age.
See Table 2 for definitions of other included variables.
WTP11
^Percent is the natural lof of (m - - )(100)
m
-------
92
Now we consider the valuation sequence. Question 10 considered an
increment in Chicago-area visibility from a typical level of 9 miles to
18 miles. Q 12 considered a similar visibility improvement over the whole
east-of-the-Mississippi region. Q 13 considered the visibility program offered
in Q 12 plus a program to prevent a threatened visibility decline at the Grand
Canyon. In the previous year, the authors had collected in Chicago 128 bids
. . . . 14
to prevent the decline m Grand Canyon visibility, using formats A3 and B.
Adjusting for one-year's inflation, these two data sets permit a test of
Hypothesis 1. Thus, we hypothesize that WTP to prevent the visibility decline
at the Grand Canyon when measured independently is greater than when measured
third in a sequence of three visibility programs.
Given a Chicago-eastern region-Grand Canyon valuation sequence, the Grand
Canyon program was valued by Chicago residents at a little more than 10 percent
of the value of a Chicago program (Ta.2-4). More interesting, a direct
-------
93
comparison of the independently measured value of the Grand Canyon program
(GCBid, Ta.2-5) with the value of the same program considered third in a
three-program sequence (WTP13 - WTP12, Ta.2-5) shows the mean value of the
former was more than five times the mean value of the latter. A linear
regression analysis (Ta.2-6) shows that GCBid and WTP13 - WTP12 are different,
at a very high level of significance. Thus, the null version of Hypothesis 1
is emphatically rejected.
2.2.7 Conclusion
Our experiment permits a clear conclusion with respect to Hypothesis 1:
the null version is rejected. In the light of Propositions 1 and 2, this
indicates that to the individual, visibility programs in Chicago, the east-
of-the-Mississippi region and the Grand Canyon are substitutes: not perfect
substitutes, but substitutes nevertheless.
If the real world of policy is characterized by the simultaneous augmen-
tation of several collective goods in one or more policy packages or programs,
our conceptual Propositions 1 and 2 and our empirical test of Hypothesis 1
suggest the following conjecture. If these several collective goods are each
valued independently and the independent values then summed to determine the
value of the program, the value of the program is inevitably overestimated
(except in the special case where the program elements are strong complements).
This conjecture would seem to apply when q = (q_,, qq^) is defined so that
i, j and k are regions (as in our experiment) or goods with different charac-
teristics, e.g., visibility, health-related air quality, and water quality.
All that is needed is substitute relationships among the elements of the q vector.
We have much less to say about the effect of value data collection format.
It is clear that the payment cards were helpful in reducing the incidence of
protest bids. Eyeball evaluation of mean bids suggests that formats and B
-------
TABLE 2-4
Incremental Mean Value (and Standard Error) of Regional and Canyon Visibility Programs
Format Sample Sizea WTP10
(a) ($/year) Regional Program; Grand Canyon Program;
WTP12 - WTP10 WTP13 - WTP12
($/year) ($/year)
29 382 161 30
(183) (72) (21)
A 31 139 14 9
(23) (6) <6)
A3 27 375 29 12
(217) (12) (6)
B 32 103 26 20
(24) (8) (11)
C1 29 251 21 39
(86) (9) (28)
26 608 354 83
(206) (181) (76)
Total 174 298 95 31
(58) (31) (13)
Protest bids eliminated; erratic bids (e.g., those which bid more for a less-preferred program)
eliminated; "high" bids neither eliminated nor reduced.
-------
95
TABLE 2-5
The Value of a Grand Canyon Program to Chicago Residents.
Format3 GCBid 1980b WTP13 - WTP12
(adjusted) 19 81c
n Mean SE n Mean SE
A 57 69.02 13.84 27 12.00 5.58
3
B 73 105.64 24.91 32 19.88 10.892
A^ and B
pooled 130 89.58 15.28 59 16.27 8.942
aSince the GCBid 1980 survey used only the A^ and R formats,
only the A^ and B format results for WTP13 - WTP12 are shown.
b
GCBid 1980 is an independent valuation.
c
WTP13 - WTP12 is a valuation of the same program, obtained third
in a three-program valuation sequence.
-------
96
TABLE 2-6
Willingness to Pay for the Grand Canyon Program:
versus Sequential Programs.
Independent
Dependent variable:
Annual WTP to avoid visibility DFE: 152
decline at Grand Canyon
PARAMETER STANDARD
VARIABLE D F
ESTIMATE
ERROR
F RATIO
PROB>F
R-SQUARE
T RATIO
4.41
0. 0002
0.1689
PROB>|T|
INTERCEPT 1
SOL 1
RYONG 1
RSENIOR 1
RHIGH 1
RGRAD 1
Z1 1
CITPAY 1
26.8
1.3
-3.5
-65. 8
52 .5
63.6
-74.7
54.0
27.5
0.9
21.8
31.5
24.0
36.2
23.5
19.2
0. 97
1.43
-0 .16
-2.08
2.18
1.75
-3. 17
2.80
0.331
0. 152
0. 871
0.038
0. 030
0.081
0.001
0.005
Variables are defined as before, except for Zl, which is defined as
Z1 = 1 if WTP13 - WTP12 (i.e., third in a three-program valuation sequence)
= 0 if GCBid 1980 (i.e., independent valuation)
-------
97
seem to generate lower mean bids in the solid core, and seems to generate
higher mean bids, than the other formats.
More generally, we believe the effect of data collection format is a
useful subject for further study. We suspect that, within the set of well-
designed contingent markets, format makes some limited difference. However,
we would be hesitant to casually apply some lable (such as "information bias")
to this effect. In real-world and actual-experiment markets, market structure
has some influence, and logic suggests that it should. That same kind of logic
should be applied to contingent markets.
Contingent markets generate a solid core of serviceable value data, but
a persistent fringe of protest bids and suspiciously high bids require and
have received close examination. We perceive substantial convergence between
the kinds of results we obtained in this and previous studies and the results
of, e.g., Smith (1980).
The research agenda has shifted from "contingent valuation (CV) must be
assumed useless because it is not incentive-compatible" to "CV must have some
15
merit because its results are consistent with those of hedonic methods"
(Brookshire, et al., 1982). On the immediate horizon, in recent CV and experi-
mental work (Smith, 1980) we see some indication that CV may have merit simply
because many people really do try to tell the truth much of the time. The
stage now appears set for a further shift in the research agenda toward pains-
taking study of the effects of contingent market structure on the quality of
value data generated. In this process, we might expect a further convergence
of survey and experimental methods.
We can expect however that there are limits to truth-telling. While income
tax liability is self-reported, the IRS finds the need to employ auditors, inspec-
tors and systematic reporting procedures. The possibility must be entertained
-------
98
that if CV were widely and routinely used to gather data which directly
influenced many public programs, and "everyone" knew it, more people would
invest in strategic efforts to influence its results.
-------
99
FOOTNOTES
1. This seems to be a typical finding when cross-sectional data are
used. For example, changes in the aggregate level of consumer
confidence have predicted the onset of the last six recessions and
the onset of each subsequent recovery. However, individual con-
sumption and saving decisions are not predictable on the basis
of individual consumer confidence (Katona, 1980) .
2. We find much of the discussion of "biases" in contingent valuation
imprecise and not especially perceptive. It seems to us that a bias
is a systematic influence, predictable in its occurrence and the
direction of its impact on results. Many of the "biases" identified
in the lieterature cited as merely possible sources of [a priori
undetermined) observation error.
3. We wish they had used the term, conjecture.
4. We believe their experiment was subject to certain influences which
would lead to overestimating the difference between contingent WTP
and true value. Nevertheless, we believe these influences were
insufficient to account for all of the observed differences
between contingent WTP and actual WTS. Thus, it is our
-------
100
position that Bishop and Heberlein's result may overstate the
difference between contingent WTP and true value, but is unlikely
to have misidentified its sign.
5. Why underestimation? We do not know for sure, but we conjecture
that contingent markets may take basically unprepared subjects by
surprise. While their instinct in such circumstances is probably
to tell the truth, their unpreparedness and inexperience with such
markets leads to a cautious and conservative response: in WTP
markets, to "sit pat" (i.e. bid zero) or to bid conservatively.
This conjecture is also consistent with the observed high bidding be-
havior of many respondents in contingent WTS markets. In that circum-
stance, the cautious response is to refuse to sell or to announce
a high selling price.
Since Bishop and Heberlein's (197 9) experimental WTS market was
highly unusual and new to its participants, we suspect that it was
subject to the influence conjectured above. If so, that would
account for some portion of the observed difference between ex-
perimental WTS and contingent WTP.
Small and Rosen (1981) address the difficulty introduced by lack of
smoothness in the expenditure function when x_. (p, q, ii, a, u) approaches
Jm
zero.
7. Substitute relationships are more likely to occur than complementary
relationships, although both kinds of relationships are possible.
8. In a working paper, the authors show that these restrictions are
not peculiar to contingent valuation but apply also to those procedures
which seek to infer the value of by analyzing the demand for
(see Freeman, 1979).
-------
101
9. Proof of Proposition 1 follows from the comparative static properties
of the additively separable utility function. a full proof is given
in Hoehn.
10. In a similar context Domenich and McFadden characterize additive separa-
bility as a "good general working hypothesis" (p.40).
11. The context described corresponds fairly closely to Malinvaud's case
of individual risks. Graham argues that in this case option price is
a lower bound on the correct BC measure of value.
12. Subsequent analyses will use methods more appropriate to the
distribution of WTP observations. Some analysts have successfully
used tobit (e.g., Adams _et _aj_. , 1980). we propose to use censored
sample correction methods (see Gronau, 1974; Heckman, 1976 and 1979)
to more closely analyze protest bids, "high" bids and "solid core"
bids.
13. It happens that the subsample which used format had (by pure
chance, so far as we know) mean household income some $5,000 higher
than the whole sample. One hypothesis for further investigation is
that the non-linear specification (13) better accounted for a
possible non-linear relationship between income and bid.
14. This survey was a contribution to work, reported by Schulze et al
(1981b) .
13. This position is logically supportable only if we accept the (un-
testable) premise that hedonic methods reveal true value.
-------
102
2.3 ALTERNATIVE ECONOMETRIC SPECIFICATIONS
2.3.1 Overview of Section 2.3
Section 2.3 reports the results of early CV experiments in Chicago on
Grand Canyon National Park. The main purpose of these experiments was to
investigate the solution to an important econometric problem--the presence
of a substantial number of zero valuations of visibility improvements in the
DV data. Ordinary least squares regression estimates, frequently employed
in econometric analysis, can bias the results when a limiting value (zero in
this case) occurs in the data set. Accordingly, tobit and logit specifications
were investigated.
The conclusion was that the empirical results were consistent with concep-
tual reasons for employing tobit analysis. Tobit analysis is designed for use
in models in which the dependent variable takes on a limiting value (zero) or
a non-limiting value of some specific (positive) amount.
-------
103
2.3.2 Tobit Estimation
2.3.2.1 Estimation When the Dependent Variable is Truncated
In the bidding game, an individual i's bid b., is elicited on the
basis of some increment or decrement in visibility. Analytically then, the
*
bid function becomes b, = $ + B'-X. + £. , where x, is a vector of individual
i oil i i
attributes including the represented level of visibility, and is a normally
distributed random error term. As the increment of visibility
x^j > approaches zero, the distribution of the error term causes more and
more of the b^ to fall on the negative side of the abscissa. With bids
limited to the positive guadrant (no one pays a negative amount to get more
visibility), the error term causes an accumulation of zero bids. The effect
of such a limit causes the distribution to be truncated at zero. With trun-
cation, ordinary least sguares (OLS) estimators result in the regression
x.), the dotted line in Fig. 2-1. OLS tends to bias the estimation
line E(b?
of 8 and 3, and, in the illustrated case, cause to be greater than 8
o 1 o ^ o
and 8^ to be less than 8^. Because of OLS bias the statistical significance of
8 is reduced and the effect of an increase or decrease in the variable x, ,
i]
is underestimated. Truncation may therefore contribute to the usual problem
of insignificant income effects or the underestimation of the rate at which
bids increase with increments in visibility.
-------
104
FIGURE 2-1
The Tobit Model with Lower Limit L = 0
-------
105
To deal with the problem of truncation, tobit analysis was used. Tobit
analysis uses the distribution of the error tern, , and the number of zero
bids as information in the estimation process. Depending upon the seriousness
of the truncation problem, tobit analysis will improve estimates of the coefficients
Bq and in the bid function.
2.3.2.2 Tobit Analysis of Three National Parkland Study Experiments
This section presents results of the National Parkland Study's (NPS) valuation
of visibility. Previous analysis of the Chicago resident data were discouraging
in that selected independent variables did not show a significant and systematic
effect on individual bids. Bid functions estimated using ordinary least squares
fit the Chicago data poorly. Because the independent variables of interest were
consistently shown to be insignificantly related to the bids, tests of hypotheses
regarding instrument design were impeded.
Results of a review of the concepts suggested tobit analysis as a potentially
superior means of explicitly accounting for zero valuations. Reported below are
the output of a tobit analysis.
The collaborative effort with NPS offered an opportunity for a contingent
valuation experiment. Three different questionnaires were used: The AAA check-
list, the AAA bidding game, and the CCC bidding game. The three CV formats were
combined with a photographic display. The photographs represented five different
levels of visibility, ranging from very poor at level A through intermediate
levels B, C, and D to very good visibility at level E. Each of the three CV
formats described level C as the current level of visibility. The CCC format
elicited valuations directly from level C. Five CCC bids were elicited; (1) to
improve Grand Canyon visibility from the current level C to level E, (2) to prevent
a decline in Grand Canyon visibility from level C to level B, (3) to prevent a
-------
106
decline in Grand Canyon visibility from level C to level A, to improve regional
visibility from level C to level E, and (5) to prevent a decline in regional
visibility from level C to level A. The AAA formats described a decline in
visibility to level C and elicited all bids as bids for improvements from
level A. For visibility at the Grand Canyon, the AAA formats elicited three
bids: bids for the improvements from A to b, A to C, and A to E. For regional
visibility, the AAA format elicited bids for improvements from A to C and from
A to E.
The bid function specified for the tobit analysis differed little from that
used earlier in the ordinary least squares estimates. The variables in the bid
function were:
ED _ The number of years of schooling completed by the respondent.
A2534 _ A zero/one dummy variable. Equals one if the respondent's
age is from 25 to 34 years and zero otherwise.
A3544 - A zero/one dummy variable. Equals one if the respondent's
age is from 35 to 44 years and zero otherwise.
A4554 - A zero/one dummy variable. Equals one if the respondent's
age is from 45 to 54 years and zero otherwise.
A55+ - A zero/one dummy variable. Equals one if the respondent's
age is 55 or more and zero otherwise.
INC - Income in thousands of dollars.
USTGC - A zero/one dummy variable to indicate whether or not
the individual has plans to visit the Grand Canyon,
Equals one if yes, has plans, and zero otherwise,
PSTGC - A zero/one dummy variable to indicate whether or not
the individual has visited the Grand Canyon. Equals
one if yes and zero otherwise.
SEX - A zero/one dummy variable to indicate whether or not
the sex of an individual. Equals one if male and zero
otherwise.
PRIM - A zero/one dummy variable to indicate whether or not
the respondent is the primary income earner in house-
hold. Equals one if yes and zero otherwise.
CITPAY - A zero/one dummy variable. Equals one if respondent
believes that all citizens of U.S. should pay the cost
of visibility impairment and zero otherwise.
USTPAY
- A zero/one dummy variable. Equals one if respondent
believes that visitors to National Parks should pay
the cost of preventing visibility impairment and zero otherwise.
-------
107
POLPAY - A zero/one dummy variable. Equals one if respondent
believes that polluters should pay the cost preventing
visibility impairment. Equals one if yes and zero other-
wise .
A priori notions regarding the sign attached to variables in the estimated
bid equation were much the same as with the OLS test. ED, INC, and USTGC were
expected to affect valuations positively. The effect of respondents' age, given the
N.P.S. results, was expected to be negative. Age was entered as a dummy
variable in order to test for non-linear effects of increasing years and to
more accurately represent the actual responses elicited from respondents. No
a priori notions were held regarding the estimated signs of PSTGC, SEX,
PRIM, CITPAY, USTPAY, and POLPAY.
Dependent variables in the estimated bid functions are the five valuations
elicited in each question. A valuation is identified by a four letter code
(see Ta.2-7, 2-2 and 2-9). The first two letters indicate the area or region that
could be affected by the bid; GC indicates the Grand Canyon and RE
indicates the regional parks as a whole. The second two letters indicate
the increment in visibility for which a bid was elicited. For instance,
AB indicates a program that would shift visibility from level A to level B.
Bid functions estimated on the three sets of data are presented in
Tables 1, 2, and 3, Examining the results overall, note first that the
number of observations was similar in each case. Second, the number of
zero bids tends to decline as the increment in visibility is increased.
This tendency of zero bids is consistent with the conceptual framework
justifying a tobit analysis. Third, average bids (E (Y x="x) ) tend to
increase as the increment in visibility increases. This trend in
-------
108
TABLE 2-7
AAA Checklist Results
([t| values in parentheses)
Dependent Variable
# of OBS
# of Zero Bids
ED
A2 534
A3 54 4
A4554
A55 +
INC
USTGC
PSTGC
SEX
PRIM
CITPAY
USTPAY
POLPAY
Constant
Va
?v(Y>0j^=x
GCAB
57
18
-00962
(.87)
-.4801
(.94)
-.3346
(.66)
-1.402
(2.33)
-1.174
(2.30)
-.0014
(.10)
-.0164
(.04)
-.4327
(1.27)
-.0962
(.24)
.4740
(1.16)
.8418
(2.57)
.4157
(.86)
-.1670
(.47)
2.142
(1.31)
.1602
.603
GCAC
57
16
-.0518
(.46)
-.0738
(.15)
.1243
(.25)
-.5974
(1.04)
-.8504
(1.67)
.0091
(.62)
.0160
(.04)
-.3482
(1.02)
-.0995
(.24)
.0983
(.24)
1.059
(3.16)
.6953
(1.52)
-.2971
(.84)
.8227
(.48)
.0837
.579
GCAE
57
11
-.0159
(.14)
.1346
(.27)
.6961
(1.40)
-.2721
(.47)
-.3812
(.79)
.0086
(.61)
-.0641
(.18)
-.0084
(.03)
-.0220
(.05)
-.4299
(1.05)
1.126
(3.38)
.9206
(1.99)
-.3720
(1.07)
.1212
(.07)
0.569
.644
REAC
57
15
.0111
(.10)
.0854
(.17)
.1021
(.21)
-.5737
(1.00)
-.7858
(1.58)
.0003
(.02)
-.1483
(.41)
-.1218
(.36)
-.0583
(.14)
-.0065
(.02)
.9943
(3.00)
.8151
(1.77)
-.3801
(1.08)
.2006
(.12)
.1262
.628
REAE
57
11
.1593
(1.39)
-.3505
(.70)
.5452
(1.10)
-.3461
(.60)
-.5949
(1.21)
.0001
(.01)
-.2761
(.78)
.4593
(1.37)
-1. 640
(.40)
-.3068
(.74)
1.228
(3.61)
.8951
(1.94)
-.2712
(.78)
-2.095
(1.19)
.0630
.649
E(Y |
RZ
3.39
.376
6.05
.365
10.72
.400
4 . 62
.350
9.84
.454
-------
109
TABLE 2-8
AAA Bidding Game Results
(|t| values in parentheses)
Dependent Variable
GCAB
GCAC
GCAE
REAC
REAE
# of OBS
50
50
50
50
50
# of Zero Bids
7
6
6
4
3
ED
-.0069
-.0880
-.033-
-.1334
-.1128
(.09)
(1.10)
(.42)
(1.67)
(1.43)
A2534
-.4590
-.7812
-.6631
-.7476
-.9802
(.84)
(1.42)
(1.23)
(1.37)
(1.83)
A3544
-.1252
-.3248
-.1437
-.4026
-.5212
(.21)
(.54)
(.24)
(.67)
(.88)
A4554
-.4361
-.4460
-.3113
-.5435
-.7563
(.63)
(.65)
(.46)
(.80)
(1.13)
A55 +
-.3076
-.4968
-.4270
-.3441
-.5962
(.57)
(.92)
(.80)
(.64)
(1.13)
INC
-.0042
-.0020
-.0009
-.0039
-.0000
(.31)
(.14)
(.07)
(.29)
(.00)
USTGC
.3171
.5507
.5613
.4882
.4256
(.91)
(1.58)
(1.61)
(1.42)
(1.24)
PSTGC
.5567
.3284
.2877
.3650
.3563
(1.37)
(.79)
(.69)
(.89)
(.87)
SEX
.0184
-.1421
-.0739
-.1465
-.1596
(.04)
(.34)
(.18)
(.35)
(.39)
PRIM
-.1231
.5664
.4318
.6525
.5848
(.28)
(1.28)
(.98)
(1.50)
(1.35)
CITPAY
.8005
.7876
.7452
.7649
.6896
(2.31)
(2.29)
(2.17)
(2.24)
(2.03)
USTPAY
-.2291
-.3464
-.3689
-.3836
-.3351
(.48)
(.74)
(.80)
(.82)
(.72)
POLPAY
.5675
.8425
1.044
.9927
1.012
(1.41)
(2.07)
(2.53)
(2.42)
(2.47)
CONSTANT
.0241
1.353
.2194
2.063
1.949
(.02)
(.85)
(.14)
(1.30)
(1.23)
1[3
.2205
.2012
.1708
.1863
.1689
P (T>0|:c-x
.721
.768
.766
.795
.809
E(Y |x»X
3.44
4.31
5.04
5.05
5.81
R2
.254
.381
.336
.420
.389
-------
110
TABLE 2-9
CCC Bidding Game Results
(|t| values in parentheses)
Dependent Variable
GCBC
GCAC
GCCE
RE AC
RECE
# of OBS
53
53
53
53
53
# of Zero Bids
9
7
12
7
9
ED
.2548
.2188
.2307
.2741
.3103
(2.53)
(2.23)
(2.23)
(2.76)
(3.04)
A2534
.1269
.0455
-.1982
.1219
-.0268
(.22)
(.08)
(.33)
(.22)
(.05)
A3544
-.4698
-.3478
-.4378
-.3902
-.4532
(.79)
(.59)
(.74)
(.66)
(.77)
A4554
-.1444
.3124
-.4201
.0377
-.2329
(.24)
(.52)
( . 69)
(.62)
(.38)
A55 +
.0480
-.0223
-.1085
.0492
-.0593
(.08)
(.04)
(.17)
(.08)
(.09)
INC
.0191
.0207
.0203
.0257
.0244
(1.93)
(2.10)
(2.04)
(2.58)
(2.43)
USTGC
.5742
.1107
.7405
.5266
.6131
(1.40)
(.27)
(1.79)
(1.29)
(1.49)
PSTGC
.1842
.1413
.3795
.2283
.2933
(.45)
(.34)
(.92)
(.56)
(.71)
SEX
-.9014
-.4648
-1.063
-.9284
-.8774
(2.09)
(1.11)
(2.42)
(2.17)
(2.05)
PRIM
1.197
.8802
1.315
1.179
1.240
(2.67)
(2.00)
(2.87)
(2.64)
(2.76)
CITPAY
.5292
.3928
.4160
.4651
.3737
(1.42)
(1.07)
(1.12)
(1.26)
(1.01)
USTPAY
.7941
-.8523
.8444
.8193
.9124
(2.15)
(2.35)
(2.26)
(2.26)
(2.47)
POLPAY
.4938
.5590
.4092
.4685
.5294
(1.45)
(1.66)
(1.20)
(1.39)
(1.56)
CONSTANT
-4.309
-4.222
-3.639
-4.663
-5.244
(2.45)
(2.46)
(2.01)
(2.69)
(2.93)
L/a
.1367
.0744
.1302
.1110
.1084
?v(T>0|jc-x
.615
.611
.610
.659
.638
E
-------
Ill
valuations indicates an internal consistency among bids; on the average, people
2
will pay more to get more. Finally, note that the R xlOO, the percentage of
explained variation, ranges from a low of 25.4% on the GCAB bid of the AAA bidding
game to 52.0% on the CCC bidding game. Relative to the OLS, tobit estimators
seem to attain a better fit to the data. For the AAA checklist, tobit analysis
does not appear to have improved our ability to discern significant decision
variables. Results of the AAA bidding game appear rather similar to the checklist
results. Results for the CCC bidding game (Ta.2-9) are substantially different
from the other bid functions. Each of the a priori expectations regarding the
positive effects of variables is confirmed. Education (ED), income (INC), and
planned visits (USTGC) each affect valuations positively and very significantly.
Expectation regarding the age variables are not confirmed. With regard to the
shift (dummy) variables,(CITPAY) retains a positive sign and is consistent
across all three data sets. USTPAY is again significant and demostrates the
same positive effect that it had on the AAA checklist bids. POLPAY is also
significant and positively related to bids as it was in the AAA bidding game.
Finally, a respondent's sex (SEX) and whether or not the respondent was the
primary income earner (PRIM) both appear to affect valuation--a result unique
to the CCC bidding game.
Two propositions may be stated. First, tobit estimators
appear to utilize the information contained within zero valuations more effectively
and therefore result in superior estimation of bid function parameters. OLS
failed to discern any systematic relationships in the CCC data whereas the tobit
analysis uncovered several significant relations between dependent and decision
variables. The effectiveness of tobit is also noticeable in the rather sizeable
2
R"'s. Second, if only an average bid is of concern, then the method of eliciting
bids, whether bidding game or checklist, may not significantly affect results.
However, a contingent valuation design that accurately describes the decision :
-------
112
as well as forcing careful consideration of valuation will be more sensitive to
individual variations. Such a design, therefore, may be more likely to permit
discernment of systematic relation between individual dependent variables and
individual decision variables.
The tobit procedure can glean information from some of the O's. Tobit
corrects biases that result from truncation of the dependent variable, but
does nothing to solve the problem of individuals systematically refusing to
participate in the bidding scheme. Thus, some of the O's in the sample are
informative, and some represent noise. Finding the right set of "Why 0 bid"
questions is necessary to decide which observations should be deleted from
the sample, and which O's should be left in for the tobit estimation. A
lower proportion of protesters among the 0 bids might explain why the tobit
procedure was more successful than OLS in analyzing some sets of data.
2.3.3 Comparison of Empirical Results
2.3.3.1 Grand Canyon and Regional Park Visibility Programs
In the sections below the results of analyzing WTP data obtained by the
Wyoming group for the NPS are presented. After removing invalid observations,
about 85 percent of the NPS observations were left,* Of these, about 25 Percent
were at the limit of the dependent variable (0 bids). Thus, a tobit model was
chosen as the appropriate model for explaining the bid behavior. In a second
stage, probit and OLS analyses were used.
The data for Albuquerque, Los Angeles and Denver were provided by the
Wyoming group headed by William D. Schulze. The Chicago data were collected
by us using methods identical to those used by the Wyoming group. The theo-
retical background for the survey and the results obtained by the Wyoming
group can be found in Schulze, W. D. et. al. "The Benefits of Preserving
Visibility in the National Parklands of the Southwest", Office of Exploratory
Research, U.S. EPA, Washington, D.C. (1981).
-------
113
Ta.2-10, 2-2 and 2-12 are the most general relationships. potentially
relevant variables are included. We also allowed for non-linearities in
income, age, education, and the electric bill. income per family was
restricted to a minimum of $5,000.
The common characteristics of the three tables are:
1) The "why zero" coefficient is negative as expected,but only the
one that stands for "polluter should pay" and "other" is significant.
2) The non-white coefficient is negative but only barely significant.
3) Household size is mainly negative but is nowhere significant.
4) The quantitative variables which are assumed to have non-linear
effects and are introduced by a linear and a quadratic term do exhibit non-
linearity but mainly the coefficients are insignificant. Also the signs on
the linear and quadratic terms are inconsistant across cities.
The possible combinations of coefficient and the implied effect are
described below.
FIGURE 2-2
Linear Quadratic Shape
1) + +
2)
3) +
4) +
-------
TABLE 2-10
Grand Canyon Visibility Value-Tobit
Dependent Variable-The Grand Canyon Bid
(' 11f values in parentheses) Page
CITY
LA
DEN
ALB
CHC
ALL
Total Ob.
127
110
115
98
450
Valid Ob.
118
103
99
68
388
Limit Ob.
19
33
24
16
92
Urban Dummy
..0452
(.14)
-.1334
(.42)
-.4539
(1.70)
-.0243
(.08)
-.0727
(.53)
Female Dummy
.4442
(2.03)
-.0324
(.13)
.0403
(.17)
.2607
(.86)
.2029
(1.74)
NonWhite Dummy
.2605
(.97)
-.5969
(1.52)
-.3099 '
(.99)
-.0676
(.20)
-.1477
(1.04)
Why O-Not Significant
Difference.
-3.439
.(.2)
-6.658
(.01)
-.5214
(.75)
-.1162
(.10)
-1.054
(2.43)
Why O-Other
-1.205
(4.27)
-2.633
(6.00)
-1.352
(4.22)
-1.490
(3.76)
-1.448
(8.90)
Education
1.162
(2.01)
-.2510
(.39)
0.8872
(1.46)
-.3522
(.48)
-.1319
(.45)
(Edu)^
-.0370
(1.89)
.0064
(.29)
.0351
(1.63)
.0103
(.40)
.0450
• (.45)
Age
..0515
(.83)
-.0867
(1.24)
-.1329
(1.74)
.0950
(1.00)
.0135
(.39)
(Age) 2
-.0007
(.98)
.0010
(1.15)
.0015
(1.64)
-.0012
(1.07)
-.0003
(.64)
Household Size
.0788
(1.26)
-.0784
(.82)
-.0003
(.003)
-.0916
(1.08)
-.0548
(1.50)
Income
-.0612
(3.09)
-.0044
(.20)
0.759
(1.45)
-.0040
(.11)
-.0054
(.48)
0
L
(Income)
.0009
(3.94)
.0000
(.10)
-.0018
(1.66)
.0002
(.36)
.0001
(1.06)
Electric Bill
.0619
(1.20)
.0076
(.73)
-.0311
(1.57)
-.0062
(.29)
.0008
(.14)
2
(Electric Bill)
-.0002
(1.62)
-.0000
(.57)
.0003
(1.98)
.0000
(.33)
.0000
(.07)
Constant
-3.751
(1.62)
5.261
(1.04)
3.718
(1.S0)
1.390
(.33)
1.674
(.74)
D(Y< Q | x - x)
E(Y)[ x = x
LLF
.608
5.59
-337
.486
2.13
0224
.495
4.53
-297
" .476
9.87
-249
.498
5.95
-1026
s2
.319
.297
.267
,096
.075
LA
-.194
(1.1)
Den
Alb
-.480
(2.67)
-.223
/I 171
-------
TABLE 2-11
CITY
Total Ob.
Valid Ob.
Limit Ob.
(D) Urban
(D) Female
(D) NonWhite
Air Quality N.S.
Other
Education
(Edu)2
Age
2
(Age)
Household Size
Income
n
L
(Income)
Elec. B.
¦k
(Elec. B)
Constant
?C£ > 0|x - x)
E(Y) !x- x
LLF
D2
LA
Den
Grand-Canyon Visibility Study
Dependent Variable-The Regional Park Bid
( |t| values in parentheses)
Page 115
LA
127
118
23
.2434
(.70)
.3690
(1.65)
-.1237
(.40)
-3.451
(.21)
-1.402
(4.73)
.4351
(.78)
-.0129
(.69)
.0921
(1.48)
-.0012
(1.66)
-.0349
• (.56)
-.0467
(2.32)
.0007
(2.95)
.0267
(1.86)
-.0002
(1.97)
-4.719
(1.12)
.573
5.086
-364
.320
DEN
no
103
39
-.3096
(.93)
-.3506
(1.31)
-.2854
(.72)
-6.458
(.01)
-3.013
(5.88)
-.9061
(1.35)
.0298
(1.30)
.0512
(.72)
-.0007
(.85)
-.0488
(.50)
.0171
(.75)
-.0002
(.92)
.0082
(.77)
-.0000
(.63)
7.009
(1.33)
.438
1.413
-194
.350
ALB
115
99
9Q
-.8654
(2.94)
.0394
(.15)
-.2455
(.70)
-5.801
(.01)
-3.116
(3.78)
-1.871
(2.80)
.0667
(2.83)
-.1500
(1.86)
.0016
(1.65)
-.0846
(1.03)
.1150
(1.38)
-.0026
(2.05)
-.0417
(.89)
.0005
(2.35)
16.762
(3.13)
.240
1.756
-273
.495
CHC
98
68
21
.2747
(.83)
.1031
(.30)
-.3008
(.77)
-5.679
(.00)
-10.152
(.03)
.2297
(.26)
-.0102
(.34)
..2228
(1.98)
-.0027
(2.01)
-.0648
(.67)
-.0881
(1.22)
.0010
(1.69)
-.0270
(1.05)
.0003
(1.41)
-3.646
(.52)
.009
.035
-1S8
.463
ALL
400
388
113
-.1600
(1.11)
.1513
(1.23)
-.3037
(2.04)
-6.350
(.01)
-1.998
(9.65)
-.7083
(2.36)
.0242
(2.35)
-.0180
(.51)
-.0004
(.85)
-.0121
(.34)
-.0080
(.69)
.0002
(1.23)
.0020
(.32)
.0000
(.70)
5.510
(2.37)
.393
3 .458
-783
.146
.0903
(.49)
-.3580
(1.38)
-.0352
-------
TABLE 2-12
Grand Canyon Visibility Value-Tobit
Dependent Variable-The Plume Bids
Page 116
LA
Den
Alb
( 111
values
inparenthe
ses)
CITY
LA
DEN
ALB
CXCH
ALL
Total Ob.
127
110
115
98
450
Valid Ob.
118
103
99
68
388
Limit Ob.
35
37
36
23
131
Urban (D)
-.0110
-.3126
-.4935
-.0189
-.2239
(.03)
(.98)
(.81)
(.061)
(1.60)
Female (D)
-.2236
.1147
.0448
.0820
.1229
(1.57)
(.44)
(.201)
(.201)
(1.03)
NonWhite (D)
-.2236
-.9724
-.2670
-.2388
-.3313
(.82)
(2.25)
(.81)
(.70)
(2.22)
Air Quality N.S.
-3.296
-6.468
-.1515
-3.481
-.7502
(.38)
(.08)
(.22)
(.26)
(1.68)
Other
-1.363
-2.335
-1.292
-1.569
-1.474
(4.7)
(.57)
(.375)
(3.72)
(8.86)
Education
-.6434
-1.298
-1.375
-.7034
-.8716
2
(Education)
(1.12)
(1.97)
(2.18)
(.94)
(2.98)
.0201
.0445
.052 8
.0217
.0300
(1.04)
(1.97)
(2.38)
(.93)
(2.97)
Age
.0511
-.1040
-.1279
.0074
-.0339
(.82)
(1.47)
(1.61)
(.08)
(.30)
(Age) 2
-.0008
.0011
.0015
-.0001
.0003
(1.03)
(1.34)
(1.54)
(.10)
(.67)
Household Size
.0 691
-.0378
.0030
-.1147
-.0197
(1.09)
(.34)
(.04)
(1.29)
(.55)
Income
.0282
.0136
.0800
.0096
.0141
2
(Income)
(1.40)
(.64)
(1.45)
(.27)
(1.22)
-.0004
-.0001
-.0018
.0000
-.0002
(1.64)
(.57)
(1.60)
(.10)
(1.32)
Electric Bill
.0046
-.0010
-.0426
-.0039
-.0060
(Electric Bill) ^
(.35)
(.10)
(2.03)
(.17)
(.99)
-.0000
.0001
.0004
.0001
.0001
(.2)
(.14)
(2.27)
(.43)
(1.30)
Constant
4.207
12.305
11.846
5.938
7.587
(.98)
(5.21)
(2.38)
(1.01)
(3.32)
? (Y > 01 x = ic)
.602
.435
.416
.463
.473
ECO |x- X
2.580
1.579
3.345
3.041
3.239
LLF
267.7
2 0 6.9
2 63.4
109.3
980.3
R2
.216
.255
.309
.129
.103
-.0524
(.46)
-.1547
(1.37)
.0454
-------
117
Cases 1) and 2) never occurred. We consider the permissible range for case 3)
to be to the left of the dividing line and_a priori do not have expectations
for case 4) . Note that the turning points are at values of the independent
¦* A
variables that are 3/^^ where a is the estimated coefficient of the linear
A
tern and b of the quadratic term. Given the range of the variables, which
is representative of the U.S. population, the estimated turning points in
many cases are outside the range. The common conclusions for the three tables
are related to the relevant range:
a) Education effect on the bid is positive although there might be a cut-
off point (e.g. Ta. 2-9, Albuquerque 12 years).
b) Age effect is negative. It might be pronounced for ages above the
cutting point. Thus for age the common picture is the right side of 3) and the
left side of 4) in Fig.2-1.
c) Income has a similar effect as education.
d) The electric bill has a similar effect as income.
The final conclusion is related to the, question whether the observed
behavior is the same in the four cities. The similarity is related only to the
marginal propensities of the explanatory variables (city effects are accounted
for by a city dummy variable). The answer is negative*. Searching for reasons
for the insignificance of coefficients led to the possibility of multi-
colinearity. This might arise due to the inclusion of both linear and
quadratic terms and also due to potential expected (although non-linear)
relationships between income on one side and education, age and race on the
other side. One would also expect a positive relationship between income
and the electric bill.
-Jr
Based upon an F test on the residuals sum of squares (the Chow test).
-------
118
Concerning city and variable results, we find that they are con-
sistant. The consistancy is exhibited in the each city equation for
each bid. The results are similar in nature. One might argue that this
is to be expected since the explanatory variables are the same. While this is
a fact,the consistancy of the estimated coefficients would not hold if the
bids were not consistant. Hence, the three bids are not independent. Although
each is expressed one at a time, they are motivated by the same reasons and
affected by the same random errors. Thus, from the econometric point of
view a "seemingly unrelated tobit model" is the appropriate model (does not exist).
2.3.3.2 Analysis of User Valuations
The analysis of user data is limited to those that visited or planned to
visit the Grand Canyon. Thus, one expects them to be capable of better evaluating
visibility in the western parks. The model and method of analysis are the
same as the cities results reported above. The explained bid is for a specific
improvement of visibility.
The various results presented in Ta. 2-13, 2-14 and 2-15 are strikingly consis-
tent with this pattern of insignificance in the coefficient of "planned days
at the Grand Canyon"; the coefficients of this variable are significant in
almost all runs. Furthermore, the log likelihood ratio indicates that none
1
of the probit runs is significant at the .05 level.
Reviewing the probit analysis, neither rural residence, sex, nor race
of the respondent is significantly related to the probability of a positive
bid. Metropolitan location, specifically residence in Los Angeles, did in
some cases affect the probability of a positive bid relative to residence in
1
Albuquerque. The coefficient for Denver (dummy) is always insignificant.
Neither age nor education is significantly related to positive bids although.
2
The log likelihood ration m each probit runs is less than the critical y
-------
119
TABLE 2-13
Coefficients of the Model Explaining Positive Bids
(Probit Analysis)
\ Dep.(3)
Indep
1
GCAB (14)
GCAC (10)
GCAD (9)
GCAE (7)
RPBC (17)
GCPL (11
Rural(D)
2.223
2 . 780
2 . 635
2 . 530
2 .660
2 . 358
(5.44) 2
(8.20)
(9.06)
(9.57)
(5.40)
(5.42)
Female (D)
. 0738
-0.0459
0.0536
0 . 6032
-0.0042
0 . 6353
(0.36)
(0.42)
(0.44)
(0.54)
(0.34)
)0 . 43 )
Non-White (D)
.3705
0.1558
-0.0094
0 .1440
0.8500
0.5359
(0.45)
(0.48)
(0.49)
(0.58)
(0.52)
(0.55)
Los Angeles (D)
1.229
1 . 073
0.9095
0.3987
0.8072
0.9781
(0.53)
(0.57)
(0.58)
(0.61)
(0.47)
(0.55)
Denver (D)
.1866
-0.2148
-0.3338
-0.6158
-0.2898
-0.097
(0.37)
(0.44)
(0.44)
(0.51)
(0.37)
(0.39)
Education (Yrs.)
.0055
-0 . 0077
-0.0033
-0.0190
0.0701
-0 . 0082
(0.007)
(0.08)
(0.08)
(0.09)
(0.071
(0.08)
Age (Yrs.)
-0.0049
-0.0013
-0.0063
-0 . 0043
0 . 0054
0.0042
(0.01)
(0.02)
(0.02)
(0.02)
(0.01)
(0.01)
Income ($1000.00)
-0 . 0118
-0.0148
-0 . 00141
-0 . 0163
-0.0173
-0.0164
(0.01)
(0.0;)
(0.01)
(0.01)
(0.01)
(0.0.)
Days Visited
0.0578
0.2917
0 . 3036
0 .2235
0.1069
G.C. (#)
(0.05)
(0.16)
(0.17)
(0.161
(0.08)
Planned Days To
0.0983
0.1169
0.0950
0.0560
0.0713
visit G.C. (#)
(.07)
(0.08)
(0.09)
(0.08)
(0.07)
Constant
0.8025
0.7458
1.394
2.021
0.1978
0.3135
(1.30)
(1.58)
(1.65)
(1.70)
(1.14)
(1.4)
-2LLR
18 . 0
16.9
15.4
13.6
17.5
17 .76
1
Number in parentheses Indicates number of zero bids out of 147 cases.
2
Standard errors noted in parentheses underlying estimated coefficients.
GCAB = Improving the value of visibility in the Grand Canyon from level A to level B.
GCAC = As above frati level A to level C.
GCAD = As above from level A to level D.
GCAE = As above from level A to level E.
RPBC = As above but for the regional parks from level B to level C.
GCPL = As above but for the Grand Canyon removing the plume.
-------
TABLE 2-14
Bid Analysis Coefficients for Positive Bids
( OLlS )
\Dep.
Indep.
GCAB
GCAC
GCAD
GCAE
RPBC
GCPL
Rural (D)
0.4131
0.4180
0.645
0.1337
0.1189
-0.0892
(0.79)
(1.25)
(1.65)
(2.51)
(1.81)
(1.92)
Female (D)
-0 .2600
-0.6547
-1.058
-1.514
-0.0119
0.1142
(0.31)
(0.49)
(0.65)
(0.99)
(0.68)
(0.75)
Non-White (D)
-0.4432
-0.8512
-0.9147
1.487
-0.9846
-0 . 8794
(0.37)
(0.58)
(0.77)
(1.17)
(0.80)
(0.88)
Los Angeles(D)
0.2001
0.4361
0 . 5371
0.5029
0.8181
0.2889
(0.35)
(0.55)
(0.72)
(1.1)
(0.761)
(0.83)
Denver (D)
-0.0135
0.1511
0.5096
-0 .2747
-0.7337
0.8039
(0.65)
(0.86)
(1.31)
(0.92)
(0.99)
Education (Yrs.)
-0 . 0405
-0 . 0228
-0.0716
-0 .1041
0.0040
0.0604
(0.07)
(0.12)
(0.15)
(0.23)
(0.16)
(0.18)
Age (Yrs.)
-0.0098
-0.0249
-0.0361
-0.0761
-0.0251
-0.0554
(0.01)
(0.02)
(0.02)
(0.04)
(0.02)
(0.03)
Income ($1000.00)
0.0076
0.0136
0 . 0240
0.0251
-0.0365
-0 . 0254
(0.01)
(0.01)
(0.02)
(0.03)
(0.02)
(0.02)
Days Visited G.C. (#)
0.0216
-0.0356
-0.0788
0.0171
0 . 0282
(0.04)
(0.06)
(0.081)
(0.12)
(0.09)
Planned Days To
0.0315
0.1042
0.2079
0.2816
0.2027
Visit G.C. (#)
(0.04)
(0.06)
(0.07)
(0.11)
(0.09)
Constant
2.534
3.611
5.213
9.041
5.042
4.587
(1.09)
(1.73)
(2.27)
(3.46)
(2.39)
(2.62)
R2
0.064
0.103
0.142
0.161
0.238
0.162
1 See notes to Table 7.
-------
TABLE 2-15
Coefficients of the Normalized Index of Bids
(Tobit Analysis)
V Dep.'11
IndepNv
GCAB(14)
GCAC (10)
GCAD (9)
GCAE (7)
RPBC (17)
GCPL (11)
Rural (D)
0.3617U)
0.2614
0.2713
0.0899
0.3300
.0885
(0.47)
(0.47)
(0.47)
(0.47)
(0.50)
(.47)
Female (D)
-0.1186
-0.2266
-0.2708
-0.2375
-0.0257
.0981
(0.17)
(0.18)
(0.18)
(0.17)
(0.18)
(.17)
Non-White (D)
-0.1209
-0.2224
-0.1835
-0.1444
0.0428
-.1101
(0.21)
(0.20)
(0.20)
(0.21)
(0.21)
(.21)
Los Angeles (D)
0.3345
0.3444
0.3069
0.2051
0.4325
.2124
(0.20)
(0.20)
(0.20)
(0.20)
(0.20)
(.20)
Denver (D)
0.0045
0.1181
0.1467
-0.0360
-0.2019
.2065
(0.22)
(0.22)
(0.22)
(0.22)
(0.24)
(.22)
Education (Yrs.)
-0.0214
-0.0025
0.0073
-0.0089
0.0257
.0120
(0.04)
(0.04)
(0.04)
(0.04)
(0.04)
(.04)
Age (Yrs.)
-0.0065
-0.0092
-0.0112
-0.0136
-0.0036
-.0124
(0.006)
(0.006)
(0.006)
(0.006)
(0.006)
(.016)
Income ($1000.00)
-0.0013
0.0011
0.0023
-0.0001
-0.0156
.0096
(0.005)
(0.005)
(0.005)
(0.005)
(0.005)
(.001)
Days Visited
0.0178
-0.0058
-0.0131
0.0077
.0111
g:c. (#)
(0.02)
(0.02)
(0.02)
(0.02)
(.02)
Planned Days To
0.0324
0.0487
0.0657
0.0630
.0625
Visit G.C. (#)
(0.02)
(0.02)
(0.02)
(0.02)
(.02)
Constant
1.171
1.082
1.211
1.427
0.734
.9384
(0.54)
(0.59)
(0.61)
(0. 1)
(.57)
(.61)
i
0.5965
0.3881
0.2948
0.1966
0.2344
.2588
O
(0.037
(0.024)
(0.018)
(0.012)
(.019)
(.016)
P(Y>0 X - X)
0.833
0.043
.861
.865
.808
.816
E(Y X - X)
1.77
2.80
3.92
5.95
2.49
3.867
R2
0.073
0.112
0.148
.160
.143
.177
*See notes to Table 7.
9 n
Coefficients estimated are —
o
-------
122
the age coefficient is at least consistently negative. The income coefficient
is also consistently negative though insignificant. The number of days
a respondent has spent at the Grand Canyon is close to being significantly
related to positive bids. The number of days to be spent at the Grand Canyon
in the future is not significantly related to a postive bid.
The OLS analysis attempts to estimate the behavioral structure of bids
for those who bid a positive amount. Coefficients for the rural, race, metro-
politan area, education, age, income, and days visited variables are consis-
tently insignificant. The age coefficient, though insignificant, is again
consistently negative. Planned days to be spent at the Grand Canyon is, how-
ever, significantly related to the magnitude of the bid. For each day planned,
the bid on AC rises by 10c, that on AD by 21c, that on AE by 28c and that on
2
the plume by 20c. In each case, R 's are very small.
Results of the tobit analysis are only slightly more revealing. As with
the OLS, most coefficients remain insignificant. Age, however, is significantly
negative with respect to the magnitude of bids. The income coefficient, where
significant, is negative. Planned days to be spent at the Grand Canyon is in
three out of four cases highly significant. Considering the equation as a
2
whole, the R As again tend to be low. However, the predicted bids conditioned
upon mean values for the independent variables are consistently increasing, as
the conceptual structure of the bid curve would suggest. This consistency
suggests that the bids were determined by a systematic method. Furthermore,
predicted probabilities of a positive bid, conditioned upon mean values, tend
Albuquerque is defined to be the base city.
-------
123
to correspond well with actual sample results. Thus, while the significance
of the coefficients may not be very satisfying, the equations do seem to pre-
dict fairly well at average levels.
The Regional Parks tobit equation was also estimated for the case where
the sum of past visits and sum of planned future visits to all Western Parks
were the explanatory variables. The variable means are correspondingly 7.5
9.9 and they range from 0 to 80 and 0 to 60. The tobit equation does not
change compared to the previous one. Also, the coefficient of the sum of past
visits tends to be insignificant while that of future planned visits is pos-
tive significant. (-.0061 (.009) and .0223 (.008) respectively)
P(y>0|x-x) - .776
ECy) | (x»x) - 3.347
R2 -.137
In the corresponding probit equation the visit variables have coefficients
below their standard errors. The -2LLR is 14.9 with 10 D.F., which implies that
the equation is not significant.
When analyzing the user survey we also looked at a model in which the
answers for "Why a zero bid" were explicitly included as explanatory variables
The coefficients of these variables (dummies) are always significant
2
and negative. Thus obviously the R is higher than m analyses without these
variables. The explanation by other variables, mainly age and income, is some
what better, although income never emerges as an important variable. The other
socio-economic variables, including city effects,do not become more pronounced.
The only exception is race. in several cases, being non-white results in
significantly lower indexes (the tobit normalized coefficient); the coeffi-
cient of being non-white (dummy) is negative and significant.
-------
124
The final run of the users survey data was an attempt to directly construct
a bid curve. The variables to be explained are the differences in the bids,
i.e., the vertical differences along the indifference curve in Fig. 2-3.
Future visits are important, although not always significant. The
consistently significant variable is the height of the starting level of
the bid. This is another clue for the consistency of the valuation of
visibility.
Age is significantly negative while income has no effect. The same
holds for education. City dummy variables and sex, race, rural-urban dummy
variables have unstable coefficients. In most cases their standard error of
estimate is larger than the corresponding coefficient.
Overall, two observations can be made. First, the coefficient of the
explanatory variables, with only an occasional exception, are insignificant.
Second, predicted bids across increments in visibility are consistent. The
implications that can be drawn are that the knowledge and perception of the
population affected the quality of their answers. Those that have not been
in the western parks and do not intend to be there in the future are likely
to have less information about them than those that have either visitied or
plan to visit.
Deficient information does not relate only to what one expects to see
but mainly to the costs involved in getting there, the time required, the
effort and effect of the weather on enjoyment. Those that have less infor-
mation make decisions under greater uncertaintly where the distribution of
perceptions they are drawing from is not stable.
The amount of information available differs depending upon whether they
have already visited or plan to visit. The idea that these differences will
cause their bids to change was tested by estimating separate relationships
-------
125
FIGURE 2-3
The Bid Curve (AK) *
*
In the analysis, the vertical segments FN, GT and HR are the explained variables.
-------
126
for each group (Ta.2—16). The disadvantage with this approach is that the
sample sizes are small, which is important given that we employ a maximum
likelihood estimation procedure. Note the distance effect for Chicago. Hence,
everything else the same, the information is low and the expected variance in
the bids large (row 4 of Ta.2-16). On the other hand, comparison of means
and variances of other population characteristics indicates considerable
similarities (e.g., income, the last two rows of Ta. 2-16).
-------
127
TABLE 2-16
Distribution of Bidders by Status
w.r.t. Visits to the Grand Canyon
(percent)
Visited
LA
28.8
Denver
31.4
Alb.
41.4
Che.
21.7
All
31.4
Plan to
Visit*
1.5
71.4
74.7
1.1
74.4
Mean Bid
4.!
3.79
3.78
7.64
4.83
Std. Dev.
10.9
5.4
11.5
25.5
13.8
Mean Income 29.0
32.0
20.7
30.0
28.0
Std. Dev.
20.1
20.2
10.5
17.5
18.2
*
Contains also those that visited in the past.
-------
131
2.4 VISIBILITY VALUE FUNCTION
2.4.1 Overview to Section 2.4
The visibility value function was the concern of all of Section 2
research. The function embodies important results of this research and
extends them in significant ways. The theory of household production,
fundamental to the development of the CV instrument, was equally important
to the development of the visibility value function. The importance of
regional, or spatial economics was recognized from the beginning of the
Project. However, the spatial dimension receives its most complete formu-
lation in the work of Section 2.4.
The spatial problem was how best to use evidence from six cities to
measure the value of visibility improvement in the entire eastern U.S. The
earliest solution to the problem, as reported in Section 2.2 for example, was
to regress measures of willingness to pay for each separate program on social
and demographic variables . This would lead to a regression equation for each
CV program in each city. For example, willingness to pay (WTP) for a ten
mile improvement in Atlanta would be estimated separately from WTP for a
twenty mile improvement in Atlanta. Similarly, there was no hypothesis about
what a ten mile improvement in Atlanta would be worth to residents of Mobile,
as distinct from Chicago's WTP for the Atlanta improvement. WTP statements
were modelled as if people regarded the East as a spatially undifferentiated
area.
Spatial differentiation is introduced by the visibility value function
in Section 2.4. It modelled WPT for regional improvements as directly propor-
tional to the area of improvement in square miles and inversely proportional
to distance from the improvement. This specification permitted valuations of
-------
132
different hypothetical programs in the CV exercise to be treated as
data underlying a single demand curve. The implication for policy appli-
cation in Section 4 was that a regional visibility policy, which produces
numerous geographically dispersed improvements, can be evaluated by means
of a single visibility value function. The spatial aspects of behavior and
the substitute nature of visual air quality in different locations established
in Section 2.1.4, were explicitly modelled. In addition, by pooling the data
and estimating a single equation, more precise parameter estimates were
obtained.
We have seen in the previous section that households were willing to pay
less for visibility-improving program when presented at the end of a
series of similar programs then when presented alone to the respondents. In
this section a model is developed which accounts for this behavior and allows
the construction of a general visibility value function which can be used to
estimate aggregate benefits of a wide variety of policy scenarios.
A central feature of the model is its direct incorporation of spatial
relationships into the empirical specification. In order to make meaningful
statements about these spatial relationships an expanded data sample was
gathered from the metropolitan areas in and around six major cities in the
eastern United States. The iterative bidding game technique was again used
for this purpose, although it was somewhat modified to reduce confusion found
among some respondents. As before, a large amount of socioeconomic data and
data on household participation in liesure activities were also gathered.
More complete description of this dataset follows later in this section.
First, we will develop more fully the conceptual framework that is used to
analyze the problem at hand.
2.4.2 Visibility in Household Production
Visibility is primarily a spatially-distributed public intermediate good
in the framework of household production and consumption, although there may
be important effects from the direct entry of visibility into the utility
functions of individuals as an amenity. In the household production analy-
-------
133
sis, visibility is combined with other factors of production such as scenery,
eyeglasses, telescopes, and other human and physical capital such as astro-
nomy classes or picture windows, to produce a service or "commodity" which
enters into the utility function of the individuals.
The individual's demand for visibility is, in this framework, formed by
the vertical summation of the derived demand curves for visibility from each
commodity. The market demand is the vertical summation over individuals of
these demand curves, thus representing a second level of aggregation.
For the remainder of this analysis, the first level of aggregation, that
of each individual over the array of utility producing commodities, will be
summarized under the heading "visual services." Our goal is to explain
variation in household demand for visual range (VR) based on the household's
stock of other inputs of production of visual services (VS), income, and
current consumption of VS. This latter variable is important since the demand
being measured is the marginal or net demand, given an initial endowment of
VS and other goods and services.
To make sense of a household's demand for increments in visibility we
need to establish some way of quantifying VS which is consistent with eco-
nomic theory. For our purposes it is not sufficient to say that a certain
person in Chicago consumes visibility of, say, twelve miles, for this state-
ment would ignore altogether how the value of these twelve miles might differ
for, as an example, a poor-sighted individual in a basement apartment and a
keen-sighted owner of a high-rise condominium with a spectacular view and a
telescope mounted on the balcony. In addition, using local VR as a measure of
a household's consumption of VS would ignore completely the value of non-
local visibility, which we have seen and will see again in this section has
value to households as they have expressed by their willingness to pay for
increments in nonlocal VR. This latter effect is of critical importance in
the analysis of the social value of visibility improvements because sometimes
-------
134
areas receiving visibility protection might have few if any permanent inhabi-
tants, and so a measure of VS which did not allow for nonlocal effects would
place a zero value on these areas when our common sense tells us otherwise.
To get a better understanding of the spatial nature of VS we will draw an
analogy from a more commonplace example of the same kind of economic struc-
ture, that of urban parks. If we require an estimate of the social value of
an additional lakefront park in the City of Chicago, for instance, we would
want to know where the park would be located, where the population is loca-
ted, the current distribution of parks and park facilities, and lastly any
unique site-specific features of the new park. We can abstract somewhat and
think of each household as facing an array of parks distributed on a two-
dimentional plane with the household at the origin. Each park has a certain
amount of facilities and scenery, which can be thought of as a measure of
quality, and each park has some unique characteristics. We should expect some
basic properties to hold in this framework. First, it is reasonable to sup-
pose that for a given park there are diminishing returns to quality. Second,
the value of a given park to a given household will be negatively related to
the distance between the residence and the park. Lastly, the value of the new
park would be lower for households already in close proximity to parks than
for households very distant from all parks, controlling for the other
characteristics.
A measure of park consumption would then need to add all available park
acreage, but only after weighting in some way each park according to its
distance from the household and its quality. Similarly, a measure of visi-
bility consumption should add together visibility in all places, but weight-
ing each place's contribution by its distance, scenery, and quality. In
particlular we define a function relating VS to these variables as
(2-39)
VS .= = £VR'ai SM^D^3 SCM ,
J i i i i i
-------
135
where VS. is household j's consumption of VS, VR^ is visual range in state i,
SIYL is the area of state i in square miles, D., , is the distance between
household j and the center of state i, and SC.. is a measure of scenery in
state i. The summation is done for the "continental" United States, including
the District of Columbia. the own-state distance is approximated by
half the radius of a circle which would have area SM., or
1
(2-40)
D = • / SM.
IX i
Although it might be possible to construct a proxy for SC , no such proxy is
both convincing and readily available. Therefore, for the remainder of this
analysis SC will be set equal to one for each state, equivilent to the as-
sumption that each state has an equal amount of unique scenery. In addition,
the following simplifications will be used:
1. All states west of the Mississippi River are combined into a
single "super-state" centered near Denver.
2. The paramenters ct^and a2 from eq. (2-39) will each be fixed at
unity.
The value of the remaining parameter <*3,the exponent on distance, will be
estimated jointly with the vector of household characteristic parameters, as
will be discussed below.
The current distribution of visibility as calculated by Trijonis is shown
in Fig. 2-4. The isopleth map represents lines of equal VR at nonurban loca-
tions. Based on the data contained in this map, each state is assigned an
initial level of VR. For additional information on this data and application
of this distribution to the estimate of actual program benefits see the
expanded discussion in Section 4 of this report.
2.4.3 Basic Properties of Visibility Valuation
Each household is assumed to have a well-defined, continuous, and mono-
-------
FIGURE 2-5. Median yearly visibilities and visibility isopleths for suburban/nonurban areas.
Source: Trijonis and Shapland, 1979
-------
137
tonic increasing total benefit curve for VS. In Fig. 2-6a such a carve is
shown. For a given household at a given moment, VS is fixed exogenously at vst
The total benefit at this level of VS is also shown in Fig. 2-6a. These two
quantities determine the "endowment point" of VS and all other goods which we
are measuring in dollar bundles along with the benefits of VS. These two
lines become the axis for the marginal bid curve merely by rescaling the old
axis. The only non-trivial point is that we do not know the original scale or
the total benefit curve. All we can observe is the benefit from changing visi-
bility from its present level as Fig. 2-6b for any individual.
Being a simple transformation of the total benefit curve, the marginal
benefit curve, or bid curve, has the following properties:
Property 1: BID(0)=0
Property 2: BID'(AVS)_>0
Property 3: BID"(AVS )<0
Property 4: Limit BID'(AVS)=0 as avS->-<»
It is important to note that some individuals will be at a point on their
total benefit curve such that the slope of the bid curve is not significantly
different from zero over the range of VS which is encountered by the respon-
dent during the iterative bidding procedure. This does not imply, of course,
that the individual does not value visibility, just that total benefits are
some arbitrary constant over the relavent range.
As we have seen, for a given individual the marginal value of visibility
(or VS) declines as total consumption increases. We might therefore expect
that households in high VS cities bid less for increments in VS than do
households in low VS cities, controlling for income and all the other fac-
-------
138
FIGURE 2-6a
Total Benefit Function
FIGURE 2-6b
Benefits of Changing Visibility from Present Level
0
-------
139
tors. Such an expectation cannot be sustained, however, as long as the
population is not homogeneous with respect to household demand for VS.
Once we acknowlege a heterogeneous population we must recognize that
there will be some tendency of individuals to sort among the cities according
to their demands for VS (and other amenities, of course). Thus, at the margin
an extra mile of VR might be worth more to the average household in the
high-VS city than the corresponding household in the low-VS city. This effect
is reinforced by the additional tendency of households in low-VS cities to
specialize their human and physical liesure capital in activities not
visibility-intensive, such as indoor recreational facilities and training.
Households in these areas might also spend resources on other factors of
production, such as a residence with a glorious view of a nearby park or
garden, as opposed to a household in a high-VS area investing in a residence
with a view of a distant vista. Thus, even if the marginal product of VR is
higher when the initial level of VR is low, it may be the case that the value
of this marginal product may be rather low, especially in the short-run when
households are even less able to adjust some other factors of production.
Since we will be examining a cross section of only six cities any esti-
mate of this reduced-form effect of the level of initial visibility should be
treated with some caution, although it remains an interesting and important
parameter in the bid function.
2.4.4 The Visibility Value Function
We now turn to the empirical specification and estimation of the visi-
bility value function (WF) . We require for this a functional form consistent
with Properties 1-4 and capable of handling both continuous and discrete
explanatory variables. This is not a simple matter. A normal OLS regression,
even without an intercept term, will violate Property 1 if simple dummy
-------
140
variables are used. Also, a dummy variable for a discrete effect will not be
correctly specified, since we know from Fig. 2-6b that a variable which tends
to increase bids for positive changes in visibility will neccessarily tend to
decrease (increase in absolute value) bids for negative increments in
visibility.
What is needed is a functional form which has Properties 1-4 and which
allows the bid curve to pivot around the origin with changes in the vector of
explanatory variables while preserving these properties. Such a form is
suggested by the "negative exponential growth" function, which we adapt as
(2-41) BID=[ l-exp(-yAVS ) ] ,
which is monotonic increasing, passes through the origin, and has an upper
limit of +1 (for all positive values of Y). This gives us our prototype bid
function. We now need to include a rotational vector of household character-
istics H, where
(2_42) H=(a+ieiZij<-Uj ) ,
so that H is a linear combination of these characteristics Z, and there is an
unexplained household-specific rotational parameter u .
Our complete empirical bid curve is then given by the product of these
two terms to form
(2_43) BID, = [1-exp(-yAVS .) ] [(ct+Zg .Z ,+u ) ]
J J i 13 3
where VS is given by eq. (2-44) below and BIDj is the willingness-to-pay (WTP)
of household j. VS is given by changes in eq. (2-44) due to the program; a is
a common intercept term (of rotation, not level of bid); Z is the vector of
-------
141
household characteristics with parameters B; is the household-specific rotation
of the bid curve.
TO demonstate the properties of this function, a bid curve was estimated
through each city's mean bids for the five programs. The non-linear regres-
sion was run once for each city, estimating only the a and the y parameters.
The hypothetical visibility programs are presented in Ta.2-17. The scenarios
are the same in each city, but a given scenario represents different values
of VS, depending on the other factors in eq. (2-39) . (the parameters of
which were estimated from preliminary maximum-likelihood regressions). In
Ta.2-18 the initial value of VS, the value of VS for each program, and the
mean bids for each program are presented for each city in the sample. The
formula used to calculate VS for the empirical analysis is
(2-44) vsj = EVR.^SM.^D."1-5
where the exponent on the distance variable was estimated by a ML method
jointly with the vector of household characteristics and the parameter Y, as
discussed below. An important result of the derivation of VS is that some
cities with very good local visibility conditions appear to have very poor
quantities of VS since they have rather poor proximity to the other parts of
the country. This is most notable in New England, where VR is the highest in
the eastern U.S. but VS is calculated to be among the lowest. Since, in the
eastern U.S., centrally located areas tend to have the lowest VR and the
peripheral areas have the highest VR the estimated effect of initial VS will
tend to be of opposite sign of that of the effect of local VR. If one be-
lieves that eq. (2-44) inadequately weights local effects then this will be
the direction of change due to increasing this weight.
-------
FIGURE 2-7
Marginal Bid Curves, by City
Washington
DVIS
See text for derivation of bid curves
-------
143
In Figure 3 the mean bids are plotted against VS as calculated in (6)
for each of the six cities. For each set of points, a non-linear regression
is fit of the form
(2-45) BID =»[l-exp(-YAVS) ]*a+e .
The figure shows the plot of the regression lines for each city. It should
be emphasized that these city results are illustrative only. The visibility
value function finally estimated applied a maximum likelihood approach to
eq. (2-43) in which all cities were included in one regression, as will be
discussed below.
We now turn our attention to the members of Z, their definitions, and the
economic implications of each. Summary statistics of each of these variables
can be found in Ta.2-19 for those observations which were used in the final
regression i.e. excluding those households which did not report BID or one
of the explanatory variables, usually income, and those who identified them-
selves as protesting the bid framework as strategic bidders. In addition, 21
persons who did not voluntarily identify themselves as one of these were
dropped by the investigators for bidding substantially more than their
available income, or for inconsistent answers coupled by interviewer reports
of confusion.
The first variable we will consider has already been discussed at some
length. This is VISENDOW, the initial level of VS as calculated in (2-44) above
and reported in Ta.2-18. As discussed above, this variable will capture the
net effect of the combination of the pure endowment effect from diminishing
marginal utility, the sorting effect, the substitution effect, and the other
complications discussed.
The second characteristic to be considered is that of income. A quad-
ratic form is used to estimate the income effect, with a first order variable
INCOME, in thousands of dollars, and a second-order term INC0ME2, which is
equal to INCOME squared. The parameter estimates on these variables (along
with INCAGE discussed below) will be used to calculate a point estimate of
-------
144
TABLE 2-17
Hypothetical Visibility Programs
as Presented to Survey Respondents
Change in Area of
Program Visual Range Coverage
1 -5 Miles Local*
2 10 Miles Local
3 20 Miles Local
4 10 Miles Eastern U.S.
5 10 Miles All U.S.
* Note: Local is defined as all land.area within 75-mile radius of the city
center. East U.S. includes all land area east of Mississippi River.
All U.S. includes all states except Alaska and Hawaii, and includes
District of Columbia.
the income elasticity of demand for VS. This estimate is of interest because
most researchers report or suggest that the income elaticity for environ-
mental goods is greater than unity. This data provides a check on this
hypothesis.
The number of persons in the household, HSLDSIZ, is important for two reasons
having opposite expected signs, making the net effect ambiguous. The first effect is
the public good effect within the household itself of the increments in VS.
The respondent is asked to accept or reject a program at a given cost to the
entire household. Since the good is non-rival, the respondent will sum as
accurately as he can the marginal benefit functions of each household member
to arrive at the household benefit function.
The other effect, however, works in the opposite direction. The actual
disposable income available to the household for the programs is probably
calculated by subtracting certain fixed or very inelastic costs from total
-------
145
TABLE 2-13
Initial Levels of VS and Proposed Changes,
by City with City Mean Rids
1980
Endowment
¦k
AVSi
avs2
&VS3
AVS
AyS5
BID1
bid2
bid3
bid4
bid5
Atlanta
4.34
-0.02
-0.02
0.21
0.26
0.21
-195.92
188.39
286.21
281.42
352.81
Boston
4.20
-0.11
0.05
0.24
0.41
0.17
-144.59
138.94
170.56
188.79
224.22
Cincinnati
4.51
-0.11
0.11
0.34
0.56
0 .22
-57.48
56.94
63.64
73.53
79.72
Miami
3.51
-0.01
0.01
0.11
0.14
0.16
-98.69
88.47
104.04
115.53
113.34
Mobile
4.59
-0.03
0.02
0.15
0.20
0.21
-156.40
168.00
196.68
214.52
238.48
Washington
4.66
-0.04
0.15
0.35
0.57
0.22
-231.70
238.36
302.97
358.14
421.93
*Change from 1990 Base Case value.
-------
146
income. These costs, such as food, clothing, etc. are likely to be correlated
with household size, so that for a given money income the actual disposable
income is reduced as household size increases. Thus the net effect is
ambiguous.
Education, HOHED affects BID in two ways, although in this case the two
act in the same positive direction. The variable is defined as the number of
years of schooling of the head of household. The direct way that education
affects BID is through the household production functions for various activities.
In the human capital model, education enters the production function as an input.
As long as education has a positive marginal product in production of these acti-
vities it will positively influence BID.
The other way that education affects BID is through its effect on household
permanent income. So far we have looked at current income only. The now classic
treatment by Milton Friedman of consumption as a function of transitory and perma-
nent income gives us some guide to the effect of some of the explanatory variables.
For a given level of current income, the more educated person will tend to have a
higher permanent income, given quantities of other human and nonhuman capital.
Thus we would expect BID to be positively affected by HOHED.
Age is a variable that combines permanent income and human capital effects.
For many outdoor activities, youthfulness can be considered as an input in produc-
tion, or at least as a cost-reducing factor. Thus, the direct effect of age would
be to reduce the value of increments in visibility.
The permanent income effect also works in this direction. For a given money
income, a middle-aged person will tend to have a lower permanent income than a
young person, given the usual age-wage profile. Again, if the person is consuming
out of permanent income then, in this example, the young person will have a higher
WTP.
-------
147
It is likely that the effects of income and age are not independent. In
particular, the marginal propensity to consume VS out of money income may
vary with age, aside from the independent effect of age on BID. To capture
this effect an additional variable, INCAGE, is introduced which is equal to
the product of INCOME and HOHAGE. This variable is included in the calculation
of the income elasticity of demand along with the independent income terms.
Two additional variables enter the vector Z which arise partially out of
permanent income considerations. These are race and sex. It has been shown
that race and sex enter significantly into the earnings function of indivi-
duals. Nonwhites tend to earn less, even after controlling for other human
capital variables; and the same is true for women. A special problem exists
for female-headed households when children are present, especially among
poorer households.
In the case of nonwhites, there is often a geographical separation from
whites, and often the division is along central city/outlying area grounds.
It is not clear what the net effects will be of these variables, but we can
guess that the effects will be negative, based on the permanent income analy-
sis. The variable FEM is a dummy for female-headed households (it should be
noted that this includes households where both husband and wife are present
and the wife responded and listed herself as "head of household"). The
variable NONWHITE, also a dummy variable, represents any of the following
groups: Blacks, Latinos, Asians, and Native Americans.
We have said that the household's stock of human and physical capital
influences BID by increasing the marginal product of VIS, but that VS may be
high already because of the capital that BID is lower in households with
large stock of these inputs. One item on the questionnaire asked the respon-
dent to indicate whether or not the household owned or had access to such
-------
148
things as a private plane, binoculars, telescope, and others. To get a large
enough sample to allow estimation of the effect of the physical capital
ownership, these responses were pooled so that ownership of any of these
specialized capital goods caused the dummy variable EQUIP to be set equal to one
Otherwise this variable equals zero.
The view quality from the residence is treated as a special case of
physical capital ownership. EXVIEW is a dummy variable which equals one if
the respondent believes their view to be excellent or especially attractive,
zero otherwise. Aside from the ambiguity resulting from the effect discussed in the
preceeding paragraph, view quality is sugject to an additional caveat. A respon-
dent who reports an excellent view might bid a low amount because VS consump-
tion is already very high, or because they are insensitive to VS to begin with,
and thus report a good view where other might not. Both of these possibilities
are consistent with low WTP. Like EQUIP, EXVIEW cannot be signed a priori
Just as household size is important for the intra-household public good
effect, so too will the number of activities participated in by the household be
important to the household's WTP for the visibility programs. The variable
ACT is a crude measure of the household's participation in various activities
throughout the year. The respondent was handed a checklist of activities and
asked to indicate those which the household takes part in during a normal
year. The excercise was motivated both by the recognition of this intra-
household and intra-individual public good effect across activities, and
also for its usefulness in getting the respondent to think carefully about
the various ways in which visibility entered into their household activities.
Presumably, this aided in the accurate revelation of WTP's for the various
programs. The variable ACT is just a count of the number of activites checked
by the respondent on the list, each receiving equal weight.
-------
149
One aspect of human capital which closely parallels the discussion of
physical capital is the quality of eyesight. If we take extremes, a blind
person will likely find changes in VS to be worthless, except insofar as
they have indirect benefits such as saftey on commercial airplanes or
crossing the street. On the other hand, a person with highly acute vision
may find the marginal product of VR to be high in producing more VS, but
can see so well already that the increase is of little value. The variable
POOREYES is a dummy variable indicating an admission of poor eyesight on the
part of the respondent.
The next set of variables addresses the ownership of residential pro-
perty. The wording of the questionnaire emphasized that the BID would reflect
the total cost of getting the program enacted. We recognize,however, that
some individuals will not quite appreciate the meaning we are attatching to
the word "all" and might believe that their property values might change if a
local amenity changes the desireability of living in their city, or they might
think that controlling pollution makes life in their city less profitable, thereby
reducing property values. We could not be more explicit in steering any such
persons away from these ideas, since the very suggestion might well have led
to even more suspicion on the part of persons to whom the idea hadn't
occurred.
Aside from this potential flaw in the reported WTP's, the ownership of
property may well indicate real differences in economic value of visibility.
If an owner-occupied home provides better opportunities for indoor substi-
tutes for outdoor activities than does a rented apartment, then we should see
such households bidding less. Also, if one own income-earning property, then the
increase in tenant's WTP may be partially collected by the owner. Thus, for a
given change in visibility the property owner would be willing to pay more,
reflecting someone else's increased welfare. We do not, however, have to worry
about double-couting of a single gain. To the extent that this indirect gain
-------
150
is important, the tenant will subtract an amount equal to the extra rent payments
in the new equilibrium, so it is a pure transfer and will not affect the aqqreqate
benefits as calculated in Section 4 of this report. The variable OWN siqnifies
ownership of the housinq unit occupied by the household, and the variable PROP
indicates ownership of other residential property in the eastern U.S.
Finally, some qeoqraphic identifier dummy variables enter the analysis.
The first of these is a dummy which equals one if the household is located in
a rural area, named RURAL. There are several possible effects of a rural
location on the bid function. First, a rural household miqht receive less
benefits from an improvement in air quality centered in the middle of the
city. Second, the qeneral view quality may be hiqher in the rural area;
havinq the effects discussed for EXVIEW. Third, cost-of-livinq differentials
may result in a dollar buyinq more of other qoods in rural areas than in the
city, thus reducinq BID for a qiven increase in welfare. This latter effect
will also be important in the city-specific effects discussed below. The first and
third of these effects tend to reduce bids while the second is ambiquous. Our
hunch is that the neqative effects will prevail.
In addition to the urban/rural dummy variable a set of four city-specific
dummy variables will be used to help account for unexplained differences
between cities. Only four can be used since one of the six city deqrees of
freedom has already been used up by the variable VISENDOW and the intercept
uses another. The four cities with dummies are Atlanta, Cincinnati, Miami and
Washinqton, with variable names A, C, M, and W respectively. Boston and
Mobile remain as the base. Ta.2-19 qives the variable means for observations
used in the reqressions reported in section 2.4.5.
2.4.5 Empirical Estimation of Visibility Value Function
Eq. 2-43. has been estimated usinq a modified Gauss-Newton non-linear
-------
151
TABLE 2-19
Variable Means for Observations
Used in Regression
Variable Mean
BID 108.704
DVIS 0.852
VISENDOW 3.754
INCOME 23.195
INCOME2 837.070
HSLDSIZ 3.177
HOHED 13.066
HOHAGE 45.391
INCAGE 1027.709
FEMHOH 0.395
NONWHITE 0.323
EQUIP 0.539
ACT 11.919
OWN 0.663
PROP 0.136
EXVIEW 0.491
POOREYES 0.226
RURAL 0.114
A 0.173
C 0.179
M 0.089
W 0.166
-------
152
regression routine. Overall, between one-half and two-thirds of the variation of BID
is accounted for by the explanatory variables, a high amount for cross-sectional
survey data of this type. A point-estimate of the income elasticity of 0.539
is computed, holding all non-income variables at their means. This does not support
the hypothesis that visibility is a luxury good, but rather that it is in the range
of a normal good between zero and one. The first-order effect of income on BID is
strongly positive as expected, but the negative second-order effect and the negative
income-age interaction effect were somewhat larger than expected (although the
direction was correctly forecasted). The negative interaction term confims the
hypothesis that the marginal propensity to consume visibility does indeed decrease
with age.
The above analysis takes account only of current money income, but as dis-
cussed above, stocks of human and nonhuman capital alter expected future income,
thus having an effect on current consumption through the permanent income model.
Turning to the human capital variables, we find an unexpected result. The estimate
of the education parameter is negative, so that more educated person tend to bid
less, holding the other variables constant. The explanation for this could be that
education can have the same negative property discussed for the case of a good view,
so that education, being more or less fixed as far as the individual is concerned,
has already increased the productivity of leisure time so much that additions of
VR have little additional value.
The variable HOHAGE must be considered jointly with the variable INCAGE. For
very low income households, age actually increases WTP for VS, but as this declines
until about an income of $9,000 per year the net effect becomes negative. This is
not difficult to explain. As age increases, leisure time tends to increase,
-------
153
especially when one or more household members retire from the labor market. This
reduction in the opportunity cost of time will shift out the demand curve for
visibility and other leisure inputs. However, there will exist a negative corre-
lation between income of these households and the amount of leisure time available.
Thus, an older couple still working full time have a lower demand than if they
retired, even though measured income is higher.
Nonwhites bid significantly less than whites, and females bid more than makes.
We have no good explanation for the latter finding other than the possibility that
women are less suspicious and conservative in responding to the (typcially female)
interviewers than were men, although there doesn't seem to be any way for us to
test this hypothesis.
Poor eyesight and ownership of specialized capital equipment did not have a
clear effect, perhaps confirming our notion of the two underlying and opposing
effects discussed earlier. As expected, participation in activities has a positive
influence on bids, reflecting the non-rivalness of visibility within the household.
One of the dramatic results is the negative influence of view quality on bids.
As discussed previously, it could be the result of diminishing marginal utility
comgined with a fixed factor (view). Alternatively, the correlation could be spurious,
reflecting the fact that people who are very satisfied with their present view are
the ones who will not bid much. Thus, we may in part be measuring the same thing
in two different ways. Both of these effects are probably important here.
The property ownership variables were of rather large magnitude, with home
ownership having a negative impact and the ownership of other residential property
having a positive effect. See the previous discussion of these variables for some
possible interpretations of these results.
The package used to estimate the parameters in Ta.2-20 does not provide a
confidence interval for estimated bids. It seems likely that Gamma and Alpha have
-------
154
TABLE 2-20
Non-Linear Least Squares Summary Statistics
Dependent Variable BID
SOURCE DF SUM OF SQUARES MEAN SQUARE
REGRESSION 22 130303017.02030957 5911864.41001407
RESIDUAL 140479409.60049038 44996.60781566
UNCORRECTED TOTAL 270782426.62079995
(CORRECTED TOTAL) 3143 233630610.1008546
PARAMETER ESTIMATE
(VARIABLE)
GAMMA 0.700
ALPHA -472.606
VISENDOW 155.757
INCOME 14.797
INCOME2 -0.029
INCAGE -0.172
HSLDSIZ 5.327
HOHED -2.011
HOHAGE 1.586
EQUIP 4.417
EXVIEW -67.139
BADEYES 12.065
ACT 5.175
PROP 97.183
FEMHOH 50.684
OWN -138.736
RURAL -41.049
NONWHITE -78.691
A 139.928
C -187.137
M 112.550
W -17.078
-------
155
a high degree of correlation, and errors in the Gamma estimate are largely offset
by corresponding errors in Alpha. Standard errors are almost irrelevant in this
case, as they are only assymptotically valid, and the function is degenerate for
values of Gamma near 0. Because of this degeneracy, a direct test of the hypothesis
Gamma = 0 is not possible; however, an indirect test of the hypothesis was carried
by constraining the estimate of Gamma to be less than 0, and re-estimating the
function.
The parameter estimates complete the specification of Equation (5)--the visi-
bility value function. For an example of the uses of this function to estimate
aggregate policy benefits see Section 4 of this report.
-------
Section 3
SECONDARY DATA ANALYSIS OF VISIBILITY EFFECTS
-------
157
3.1 OVERVIEW of SECTION 3
Section 3 is a related group of studies of the role of visual air quality
in particular household activities. Swimming, Hancock Tower visitation, and
baseball attendance represent active and passive outdoor recreation. Studies
of view-oriented residences explore the relationship between view and visual
air quality at the household residence. Auto and air traffic studies inves-
tigate the importance of visual air quality in basically non-recreational outdoor
activities. Finally, the study of TV viewing establishes the role of visual
air quality in influencing the choice between indoor and outdoor recreation.
These studies complement the contigent valuation work of Section 2 in
several ways. First of all, the studies of Section 3 all pertain to parti-
cular markets, such as baseball attendance or TV viewing, whereas contingent
valuation estimates total visibility value irrespective of the uses to which
they are put. In each case the individual market studies demonstrated that
people reveal an implicit willingness to pay for visibility improvement.
Ideally, aggregate visibility benefits would be determined by both methods
and compared in order to validate the results. While this is not feasible,
nevertheless a judgment can be made concerning the plausability of the
partial comparison that is possible.
Secondly, the value of visiblity improvements in these papers are esti-
mated from historical records of completed activities. For example, the value
of a one mile average improvement in visual range is estimated to be worth
about 3 cents per person in attendance, including approximately 10,000 addi-
tional persons who would attend under the better visibility conditions. This
result is derived from recorded time series information on attendance along
with visibility and a number of other variables that effect attendance. People
reveal the dollar value of their preference for visibility by their behavior in
the face of actual visibility change .
-------
158
Thirdly, the underlying theory of visibility valuation is the same for
the market studies of Section 3 and the CV work of Section 2. The modeling
and empirical estimation are quite different. Nevertheless, the common theo-
retical basis makes the two empirical approaches complimentary. Evidence
that results are consistent strengthens our confidence in the results as well
as the methods that have been developed to obtain them. The Hancock Tower
study in 3.3 provides important directly comparable evidence concerning
the two empirical approaches. The conclusion is that the hypothesis of a
statistically significant difference between them is rejected.
-------
159
3.2 OUTDOOOR RECREATION
3.2.1 Swimming
Swimming is one of the major summertime recreational activities
available in the Chicago metropolitan area. with numerous beaches and
over one hundred pools, the Chicago Park District alone has an annual
attendance of many millions. Unfortunately for this analysis, admission
to Chicago facilities is without charge, and no accurate records are
kept of attendance as a result. Data for both beach and pool attendance
were provided by the Wilmette Park District, which operates one of each
type of facility just north of Chicago.
Visibility affects the demand for swimming in at least three ways.
Consider the simple utility function:
Up = U (H, Q, C, T) ,
where U is the utility generated by a pool visit, H is the perceived health
P
benefits from swimming, Q is a measure of environmental quality, C is the
level of thermal discomfort faced during the day, and T is the time spent
at the pool. It is clear that all of these parameters are interrelated to
some extent. For example, a hot day may cause an increase in photochemical
smog, which may induce an individual to spend less time outdoors due to
the decreased health benefits as perceived by the individual. The simple
function is useful because it illustrates the mechanisms by which visibility
may enter into the demand equation. The first of these mechanisms is the
"pure-visibility" effect, and represents the amenity value of visibility in
determining the overall utility generated simply by enjoying a nice day.
The second is the "indicator" effect, which reflects the use made by indivi-
duals of visibility as an indicator of the presence of unhealthy air-pollutants.
The indicator effect may be quite important in the Chicago area, as the public
-------
160
receives many warnings in the summer to avoid physical activity during periods
of high ozone levels. These warnings may come to be associated with days in
which visibility is poor, so that poor visibility may deter swimming for health
reasons, even if the poor visibility is caused by harmless natural conditions.
The third way visibility enters the demand equation is through its effect
on the transmission of ultraviolet radiation, which is responsible for tanning
(and burning) the skin. Since many swimmers spend a great deal of time and
money to get a tan (i.e., special lotions, etc.), any decrease in the ability
to get a tan represents a real loss in utility.
To identify these effects from raw attendance figures requires an accurate
treatment of thermal comfort. A precise, absolute definition of comfort is
not possible, as it is a subjective evaluation which differs greatly among indivi-
duals. Auliciems (1) showed that four factors influence human comfort, that is,
the proportion of individuals who respond negatively to the question, "Are you
comfortable?". These four factors are temperature, humidity, air movement,
and thermal radiation, such as the infrared radiation from the sun. These fac-
tors interact with each other to yield a level of comfort: which is particular
to the individual. The National Weather Service reports two indices which
attempt to integrate these factors into a more useful measure than simply using
temperature. These are the temperature-humidity index (THI) and the wind-chill
index (WCI). Neither is particularly suited to this analysis for several reasons.
The THI neglects the effect of the wind, since it was developed primarily to
monitor factory conditions, and it does not respond to human comfort in a
linear way. A THI reading of 65 implies that everybody is comfortable, while
a reading of 70 corresponds to discomfort in 10% of the population, 75 corre-
sponds to 50%, and 80 to virtually 100% discomfort. The WCI does not take
-------
161
into account humidity, as this factor is almost always negligible when compared
to the wind effect outdoors in the winter. Also, the published formulas are
inappropriate because they assume a normal amount of skin exposure and moisture,
while in swimming the entire body is wet with most of the skin exposed to the
wind. To account for temperature, humidity, and wind, a set of interaction
terms is included in the regression, as well as the terms' independent effects.
The fourth comfort-related factor, radiant energy, is assumed to be a simple
linear function of cloud cover and visibility.
It is important to keep in mind that the true marginal decision variable
is how much time to spend at the pool, or in the aggregate, how many person-
hours are spent, and not how many people attend in a day, which is what we
have data for here. At best, we can make some crude assumptions about average
time spent at the pool and the average value of time of those who attend. Even
so, it is questionable whether any reasonably accurate dollar value can be
assigned to visibility in this particular case. What can be established, how-
ever, is the extent to which visibility plays a role, consciously or not, in
the consumption decision of individuals. A decrease in attendance due to
reduced visibility implies a decreased opportunity set and a reduction in
utility to those who no longer attend as well as those who continue to attend.
Assigning a dollar value based entirely on the reduction in attendance may
also prove unsound due to the substitution into other, less visibility-elastic
activities or even into more work and less leisure as the quality of leisure
time is decreased.
3.2.1.1 Empirical Model
Two models are estimated using Wilmette data and surface weather observations
at O'Hare Airport for the years 1977-1979. Swimming data are also available for
-------
162
1980, and are used for prediction-verification. Due to the lack of data on
certain important variables, such as wave height, water temperature, and
pollution levels in the lake, the beach data are not used in this analysis.
Rather, the emphasis is placed on the pool, which is a controlled environment
not subject to closing unrelated to the weather.
The first model to be estimated assumes a simple, readily interpretable
linear relationship. The relationship is of the form
P = a + exV il2Bx
where P is daily pool attendance, V is visibility, and are other factors
which effect attendance. Unbiased estimates could be achieved for the esti-
mated parameters by taking first differences of all the variables, 364 days
apart. However, with the limited dataset and the subtle quality of the effects
being measured, first-differencing is highly undesireable. To account for
purely temporal effects, a comprehensive set of dummy variables and functions
are employed on a portion of the data, the results of which are compared with
those obtained using first differences. In addition, the data are analyzed for
each year separately in addition to the pooled regression to check for struc-
tural stability between years. Data for the year 1980 are included as an
additional check on the parameter estimates.
A simple plot of attendance by date indicates a tendency for the attendance
to fall in clusters. It is determined whether this is due to a simple clustering
of days similar meteorologically, or whether there is a lagged relation among
the data. The disturbances are examined for autocorrelation to see whether
General Least Squares methods would be more appropriate than OLS estimators.
-------
163
In addition to the linear model, a second model is used,
of the form.
n m
LOG(P) = a + Z S.LOGCx.) + - S.X,
v . i=1 1 i i=n+i i i
where the are expressed in log form, if continuous, or else left in levels
if the relationship is best described by an exponential function, or if the
variables are discrete. This model has the advantage that elasticities are
estimated directly, but is not as straightforward and simple as the linear
model.
3.2.1.2 Regression Results
Ta. 3-1 shows the results of the first regression model. The important
points which led to this final regression are:
1. Day-of-week effects were minimal and not statistically
significant. This includes a simple weekend/weekday
dummy variable, which was also tried.
2. The linear model is not structurally stable. The values
for the coefficients differ significantly for each of
the three years in question. (F-ratio of 3.978.
Separate year results are not reported here.)
The pooled regression using all three years can
be looked at as an "average" representation of the effects.
3. Lagged exogenous variables were not statistically significant,
though their signs and relative magnitudes were as expected.
In addition, the data showed no significant autocorrelation,
using the Durbin-Watson method.
-------
TABLE 3-1
Pool Attendance:
VARIABLE (units)
PARAMETER
ESTIMATE
STANDARD
ERROR
INTERCEPT
RAIN (% Of Day)
FOG (% of Day)
TEMP (°F)
WIND (MPH/10)
HUMIDITY (%)
CLOUD-COVER (%)
VISIBILITY (Mi./lO)
vTwind
TEMP-WIND **
'l'l'MI' JViiM[)
HUMIDITY-WIND **
TEMP.-HUMIDITY **
COS(T) ***
SIN (T) ***
TTREND ***
464633.7
-1 . 061104
-0.051259
543 .921259
-292.932312
57.678240
-4.782367
1.852527
6511.505
3 . 943894
-84.489434
-0.192682
-0.434404
3364.711
-3488.21
-78,873748
350765.7
2 .273052
2.489467
164.347770
117.645255
39.192380
1. 209490
0.853752
2526 . 044
1. 500730
32.034411
0.066548
0.494560
1648.974
2921.867
54.698816
* One-tailed test
** Comfort - Interaction Terms
*** Time-Effect Terms
SSE
Deg. of Freedom
MSE
32258740
220
146630.6
Model 1
T-RATIO
1.3246
-0.4668
-0.0206
3.3096
-2.4900
1.4717
-3 . 9540
2.1699
2.5777
2.6280
-2 . 6375
-2.8954
-0.8784
2.0405
-1.1938
-1.4420
PROB > T
0.1867
0.3206
0.4618
0.0006
0.0068
0.0713
0.0001
0 .0156
0.0068
0.0092
0.0089
0.0042
0.3807
0 . 0425
0 . 2338
0.1507
F-Ratio
Prob> F
R-Square
25 . 51
0.0001
0 . 6349
-------
168
The results of the final regression can be summarized thus:
1. Rain and fog effects are not well accounted for in a
linear model. This is perhaps due to the discrete nature
of these variables as they exist in our data set.
2. The model accounts extremely well for comfort-related
effects, both independent and interaction terms are
significant with the proper signs.
3. Visibility has a significant effect on attendance. The
effect is not stable between years, but ranges between
1.24 and 3.73 persons per tenth-of-a-mile increase in
visibility. When the data are pooled, an estimate of 1.85
is arrived at. The high of 3.73 was achieved in 1979, the
year the model best fit the data.
The second model which was estimated was the log-log relationship. On
the whole, this model was a disappointment, as some of the variables' effects
were masked, or were not well accounted for in multiplicative relationships.
Results from this regression are listed in Ta. 3-2.
While the log-log relationship expressed rain and fog effects in exponential
form, which was found most appropriate, it seems to have been an inappropriate
functional form for other variables. Temperature and wind have the anticipated
effects, but cloud cover, humidity, and visibility have no significant effect.
This model also has less overall explanatory power than the linear model
2
(R = .5717), and so the conclusions for this investigation rely heavily on
the first model.
-------
VARIABLE
PARAMETER
ESTIMATE
TABLE 3-2
Pool Attendance (Log)
STANDARD
ERROR
INTERCEPT
RAIN
FOG
LOG(TEMP)
LOG(HUMIDITY)
LOG(WIND)
LOG(CLOUD-COV.)
LOG(VISIBILITY)
LOG(TTREND)
COS(T)
SIN(T)
1338.153
-0.040805
-0.021650
15 . 991371
-0.561598
-0.663739
-0.00686768
0 . 025559
¦158 . 950272
3.453727
0.203768
10907.83
0.007502444
0.008816437
1.486479
0.594286
0.293846
0.051006
0.252146
1244 .464
5.731853
10 . 422159
* One-Tailed Test
SSE
DEG. OF FREEDOM
MSE
435 . 025664
225
1. 933447
Model 2
T-RATIO
0 . 1227
-5.4389
-2 .4556
10.7579
-0 . 9450
-2 .2588
-0 . 1346
0.1014
-0 . 1277
0.6025
0.0196
P R O B > T
0.9025
0 .0001
0.0074
0.0001
0 . 1728
0.0125
0.4465
0.4597
0.8985
0.5474
0.9844
F-RATIO
PROB> F
R-SQUARE
30 . 04
0.0101
0.5717
-------
170
3.2.1.3 Conclusions
1. An increase in ambient visibility levels of one mile will increase
attendance from three to five percent. This represents an annual
increase in attendance of between 1728 and 2880 persons.
2. The lack of day-of-week effects suggests a population consisting
mainly of children and younger adults with a correspondingly low
employment rate. Since environmental amenities are usually income-
elastic, this would tend to yield a site-specific estimate which was
below the average valuation over the entire population.
3. A large portion of the variation remains unexplained in the models
used here. There is likely a large random element, due to reasons
cited in number 1 above, but in addition, it appears that the inter-
relation between the variables is a rather complex function, which
can only be approximated by a linear relationship.
The remainder of the chapter presents the results of an investigation
into the effects of visibility on common recreational and other activities.
For the most part, we examine activities for which the relevant demand
elasticities are unknown, and so benefit estimates of visibility changes are
not possible. However, in the case of major league baseball attendance,
estimates of demand elasticities have been made, for example, by Noll
and Demmert.
General models of activity choice with visibility as an input into
household production functions have already been presented in this report.
For this reason, none are presented here. Instead, regression models are
introduced, and the variables described. Following each are the results of
-------
171
one or more regression analysis with a brief discussion of the results.
All of the activities measured were in the Chicago Metropolitan Area.
3.2.2 Television Viewing
With the aid of A.C. Nielsen's "Nielsen Television Index"* a dataset
consisting of the total number of households using television at the hours
of 1:00 P.M., 2:00 P.M., and 3:00 P.M., for each day during calendar years
1978 and 1979 was assembled. In addition, the number of households watching
Chicago Cubs home games was determined. Due to the lack of lights at the
stadium, all games take place between noon and around 4:00 P.M. These data
are useful in the discussion of baseball attendance below.
Many factors undoubtably influence the number of television viewers.
One for which we have little independent data is program quality. The choice
of the early afternoon hours is partly an attempt to control for program
quality, as there are relatively few changes in scheduling in this time
period. Also, it enabled the comparison of the game and non-game days of
the Cubs, as described above.
To examine the influence of visibility on television audiences, we sepa-
rated its effects from other meteorological and temporal factors. The
regression results are given in Ta. 3-3. The intercept, 31.86, represented
an average Wednesday in May, meaning 31.86% of the 3 million households
watching T.V. The effect of visibility is given by the two variables
VIS15 and WKNDVIS. The effects of a one mile increase in visibility, assuming
-Jr
Thanks are due to Maureen Gorman of NTI for her kind assistance in providing
these data.
-------
172
TABLE 3-3
Percent of Households Using Television, 1978-79
VARIABLE
INTERCEPT
RA15
SN15
WIN15
TCL15
VIS 15
TEM15
FOOTBLSA
FOOTBLSU
FT8LH0L
CUSHOME
CUBAWAY
BLIZZARO
M
T
R
F
S
su
M1
M2
M3
M4
M6
M7
M8
M9
M10
M1 1
M12
WKNOVIS
WKNOTEM
WKNDRA
WKNOSN
OF
PARAMETER
STANOARO
ESTIMATE
ERROR
T RATIO
PR08>|T|
31.862655
1.407201
22.6426
0.0001
0.019619
0
.005993813
3.2732
0.0011
-0.00618701
0
.007097675
-0.8717
0.3837
0.008701367
0
.003107462
2.8002
0.0053
0.016687
(
0.00402075
4.1501
0.0001
-0.013373
0
.003915276
-3.4157
0.0007
-0.081347
0.014113
-5.764 1
0.0001
1 .617240
0.764520
2.1154
0.0348
6.678667
0.763180
8.751 1
0.0001
5.454071
1.765358
3.0895
0.0021
2.56230S
0.534636
4.7922
O.OOOI
0.716211
0.530855
1.3492
0. 1777
5.241333
1.123943
4.6633
0.0001
0.918224
0.471162
1 .9489
0.0517
-0.320498
0.465496
-0.6885
0. 4914
-0.073249
0.470864
-0.1556
0.8764
-0.283101
0.467240
-0.6059
0.5448
6.847284
1.241751
5.5142
0.0001
12.25906 1
1.247545
9.8265
O.0001
4.850261
1.004 174
4.8301
0.0001
2.067644
0.952657
2.1704
0.0303
2.955393
0.806152
3.6660
0.0003
1.445582
0.639560
2.2603
0.0241
1.800524
0.620328
2.9025
0.0038
2.639546
0.628826
4. 1976
0.0001
3.760193
0.627449
5.9928
0.0001
2.744425
0.645459
4.2519
0.0001
3.327091
0.739155
4.5012
0.0001
2.894163
0.792583
3.6516
0.0003
3.107789
0.854282
3.6379
0.0003
-0.00134655
0
.007019629
-0. 1918
0.8479
-0. 104334
0.012805
-8.1432
0.0001
0.017010
0.014474
1.1752
0.2403
0.015358
0.015645
0.98 17
0.3266
SSE
~ FE
MSE
7584. 145
639
11.007467
F RATIO
PR08>F
R-SOUARE
49.41
0.0001
0.7030
Source: A. C. Nielsen Co.
-------
173
local linearity, is -.0134, meaning .134% of the 3 million households stop
watching T.V. or around 4,000 households. The effect if that increase happens
on a weekend is a further reduction of 400 households. The prime effect is
very well estimated, with a t-statistic of -3.42, while the second is not,
with a t-statistic of only -0.19. Overall, television appears to be highly
seasonal, with a peak in January and a trough in the base month of May.
The day-of-week dummies acted as expected, with a large weekend increase.
The weather variables also behaved as expected, with higher temperature and
visibility causing less television watching, as people shift to outdoor
activities, and with wind, clouds, and rain driving people indoors to the
T.V. Snow had a negative effect, but was not precisely estimated.
In a further attempt to abstract from mere seasonal variation, 7-day
first differences were calculated. The new regression is presented in
Ta. 3-4. The variables prefixed with the letter D are the same as the pre-
vious regression, only having undergone first-differencing.
The results for visibility are still negative, but the effect is less
precisely estimated, with only a 1.06 t-statistic.
-------
174
TABLE 3-4
Percent of Households Using Television at 2:00 P.M. 1978-79:
7-Day First-Differences
VARIABLE
INTERCEPT
07RA
D7SN
D7WIN
07TCL
D7VIS
D7TEM
07FTBLSA
D7FT8LSU
07CU3H0M
D7CUBAWA
07H0L
07FTBLHL
D7BLIZZ
OF
PARAMETER
ESTIMATE
STANDARD
ERROR
T RATIO
PROB> T
0.063820
0.024133
-0.0001S3412
0.008463473
0.024703
-0.00419665
-0.090849
-1.629750
-0.562519
2.S15677
0.695619
1.373662
13.847987
4.136090
0. 174468
0.007834551
0.00737355S
'0.003586003
O.0O4452127
O.00394 3219
0.015345
2.752302
2.791929
0. 523036
0.520638
0.633384
1 . 597012
1 .068155
0.3658
3.0367
-0.0222
2.3601
5.5486
-1.0643
-5.9206
-0.5921
-0.2015
4.9536
1.3361
2. 1688
8.6712
3.8722
0.7146
0.0021
O.9823
0.0185
0.0001
0.2876
0.0001
0.5539
0.8404
0.0001
0. 1819
0.0304
0.0001
0.0001
SSE 15576.95 F RATIO 25.94
OFE 709 PROB>F 0.0001
MSE 21.970307 R-SCUARE 0.3223
-------
175
3.2.3 Baseball
Two analyses were performed on baseball data. The first is an analysis
of attendance data and relevant team information published for the Chicago
Cubs during the 1978 and 1979 seasons. The second was an analysis of tele-
vision viewing of the Cubs during the same two seasons. For both the same
explanatory variables will be used.
The variables are all briefly described in Ta.3-5 with the results of
the regression of attendance data. The results in Ta.3-6 are for the per-
cent of Chicago metropolitan area households watching WGN Television at 2:00 P.M.
during each game. Many similar and highly correlated variables were included
in the regression. These include mainly statistics on team performance during
the season, and opposing team characteristics. These results were not examined
in detail. Instead, we merely noted the effects of visibility on attendance.
An increase in visibility of one mile increases gate attendance by
approximately 125 people, although the effect is not precisely estimated.
Interestingly, the effect of the same increase in visibility is to increase
television watching of the Cubs by about 3,000 households, even though the
total effect on television watching of all types is to decrease viewing by
about 4,000 households. Perhaps picture quality is enhanced with the improved
visibility. Whatever the case, both attendance and television increase.
Noll provided an estimate of the effect of ticket prices on attendance
for an SMSA of population of around 3.5 million. Since Chicago has an SMSA
of approximately 7 million, the effect is doubled, yielding a reduction in
attendance of 380,000 persons per year for a one dollar increase in ticket
price. Our measured visibility effect of 125 persons per game, multiplied
by 81 games yields a total of 10,125 additional persons per year in gate
-------
TABLE 3-5
VARIABLE
INTERCEPT
M
T
i
F
s
su
M4
M6
M7
M8
M9
DATE
LASTHOME
DOUBLE
RA09
RA12
RA15
TEM12
WINDOUT
VIS1 2
SOXPCT
SOXPLAY
CHIFEST
IN RACE
CUBPCT
HMGMBK
SAMEDIV
CPTCHERA
VSSTAN
VPTCH500
EQUALITY
EQUALSD
KINGMAN
YEAR79
CUBWIN10
DF
Chic
ago Cubs Total
In-Person
Attendance
PARAMETER
STANDARD
ESTIMATE
ERROR
T RATIO
PROB> |T|
19137.39
60316800888
0.0000
1 .0000
1892.86
2421.542
0.7817
0.4362
-2010.47
1881.489
-1 .0685
0.2878
1438.35
1948.707
0.7381
0.4622
-398.466013
2093.582
-0.1 903
0.8494
1 0936.1 1
1880.054
5.8169
0.0001
13464.3
1916.078
7.0270
0.0001
-1 0060.6
3186.865
-3.1 569
0.0021
5966.58
2168.68
2.7512
0.0070
7907.502
3011.217
2.6260
0.0100
10158.55
3905.221
2.6013
0.0107
2512.577
4325.281
0.5809
0.5626
-0.81 0883
38.412070
-0.2294
0.8190
141.073569
167.978923
0.8398
0.4030
3161.818
1845.086
1.7136
0.0897
-33.978231
22.961630
-1 .4798
0.1 420
-25.077909
30.1 91 844
-0.8306
0.4081
1 5.8981 1 5
26.908620
0.5908
0.5560
214.071109
82.563972
2.5928
0.0109
1730.691
1 503.1 1 1
1.1514
0.2523
12.487959
1 4.521 299
0.8600
0.3918
-13109.2
17161.29
-0.7639
0.4467
58050.9
60316800889
0.0000
1 .0000
-2027.1 3
3221 .1 81
-0.6293
0.5306
3999.039
231 7.1 96
1 .7250
0.0874
-1 9223.8
16608.63
-1.1 575
0.2498
-935.843864
312.870576
-2.9912
0.0035
-1 6637.5
14290.28
-1.1 643
0.2471
680.1 58836
405.725853
1 .4003
0.1 645
-998.0821 56
405.395244
-2.0562
0.0423
179.609536
176.324238
1.0186
0.3108
-11718.5
13620.63
-0.8604
0.3916
24302.1 3
15857.92
1 .5325
0.1 285
-3335.01
1 724.91 5
-1 .9334
0.0560
8823.667
13533.4
0.6520
0.5159
1059.82
560.594588
1 .8639
0.0652
1978-79
VARIABLE
LABEL
MONDAY
TUESDAY
WEDNESDAY
FRIDAY
SATURDAY
SUNDAY
APRIL
JUNE
JULY
AUGUST
SEPTEMBER
LINEAR TIME TREND
DAYS SINCE LAST HOME GAME
DOUBLE HEADER
RAIN AT 9 AM
RAIN AT 12 NOON
RAIN AT 3 PM
TEMPERATURE AT NOON
DUMMY, EQUALS 1 WHEN WIND BLOWS OUT
VISIBILITY AT NOON IN TENTHS OF A MIL
SOX WINNING PCT
ZERO-ONE DUMMY
DUMMY FOR CHICAGOFEST
DUMMY, ONE WHEN TEAM IN PENNANT RACE
CUBS WINNING PCT
GAMES BEHIND LEADER (CUBS)
1 WHEN OPPONENT IN SAME DIVISION
CUB PITCHERS ERA
VISITORS STANDING IN DIVISION
VISITING PITCHERS GAMES ABOVE 5
DIFFERENCE IN WINNING PCT
EQUALITY X SAMEDIV
DUMMY, ONE WHEN KINGMAN PLAYED
YEAR DUMMY
NO. OF GAMES WON OF LAST TEN
SSE
DFE
MSE
2G10887601
101
25850372
F RATIO
PROB>F
R-SQUARE
12.76
0.0001
0.8155
-------
TABLE 3-6
Chicago Cubs Television Audience, 1978-79:
Percent of
Households
PARAMETER
STANDARD
VARIABLE
VARIABLE
DF ESTIMATE
ERROR
T RATIO
PROB> |T|
LABEL
INTERCEPT
1 28.310590
27804381
0.0000
1.0000
M
1 1.508206
1.116264
1.3511
0.1797
MONDAY
T
1 -0.333530
0.867315
-0.3846
0.7014
TUESDAY
W
1 0.336566
0.898300
0.3747
0.7087
WEDNESDAY
F
\ 0.895605
0.965083
0.9280
0.3556
FRIDAY
S
1 4.545163
0.866653
5.2445
0.0001
SATURDAY
su
1 5.355864
0.883259
6.0638
0.0001
SUNDAY
M4
1 -1.992947
1.469057
-1.3566
0.1779
APRIL
M6
1 2.428024
0.999702
2.4287
0.0169
JUNE
M7
1 3.579786
1.388088
2.5789
0.0114
JULY
M8
t 6.405515
1.800199
3.5582
0.0006
AUGUST
M9
1 5.339600
1.993835
2.6781
0.0086
SEPTEMBER
DATE
1 -0.018761
0.017707
-1.0595
0.2919
LINEAR TIME TREND
LASTHOME
1 -0.066878
0.077434
-0.8637
0.3898
DAYS SINCE LAST HOME GAME
DOUBLE
1 0.364654
0.850534
0.4287
0.6690
DOUBLE HEADER
RA09
t 0.001897492
0.010585
0.1793
0.8581
RAIN AT 9 AM
RA12
t 0.032381
0.013918
2.3266
0.0220
RAIN AT 12 NOON
RA15
1 -0.010960
0.012404
-0.8836
0.3790
RAIN AT 3 PM
KM 12
i 0.042599
0.038060
1.1193
0.2657
TEMPERATURE AT NOON
WINDOUT
1 0.370211
0.692893
0.5343
0.5943
DUMMY, EQUALS 1 WHEN WIND BLOWS OUT
VIS12
1 0.010100
0.006693918
1.5089
0.1344
VISIBILITY AT NOON IN TENTHS OF A MILE
SOXPCT
1 12.036824
7.910881
1.5216
0.1312
SOX WINNING POT
SOXPLAY
1 110.357756
27804381
0.0000
1.0000
ZERO-ONE DUMMY
CHIFEST
1 -2.988367
1.484876
-2.0125
0.0468
DUMMY FOR CHICAGOFEST
IN RACE
t -0.115474
1.068163
-0.1081
0.9141
DUMMY, ONE WHEN TEAM IN PENNANT RACE
CUBPCT
1 -16.721749
7.656122
-2.1841
0.0313
CUBS WINNING POT
HMGMBK
1 -0.520589
0.144225
-3.6096
0.0005
GAMES BEHIND LEADER (CUBS)
SAMEDIV
1 -7.081642
6.587425
-1.0750
0.2849
1 WHEN OPPONENT IN SAME DIVISION
CPTCHERA
1 -0.279615
0.223906
-1.2488
0.2146
CUB PITCHERS ERA
VSSTAN
1 -0.081824
0.223754
-0.3657
0.7154
VISITORS STANDING IN DIVISION
VPTCH500
1 -0.034274
0.081281
-0.4217
0.6742
VISITING PITCHERS GAMES ABOVE 5
EQUALITY
1 -10.780878
6.278732
-1.7170
0.0890
DIFFERENCE IN WINNING PCT
EQUALSD
1 9.484610
7.310063
1.2975
0.1974
EQUALITY X SAMEDIV
KINGMAN
» 0.592985
0.795138
0.7458
0.4575
DUMMY, ONE WHEN KINGMAN PLAYED
YEAR79
1 9.447361
6.238523
1.5144
0.1331
YEAR DUMMY
CUBWIN10
1 0.599823
0.262106
2.2885
0.0242
NO. OF GAMES WON OF LAST TEN
SSE
554.802019
F RATIO
7.18
DFE
101
PROB>F
0.0001
MSE
5.493089
R-SQUARE
0.7134
-------
178
attendance per mile increase in visibility. Thus, the change in consumer's
surplus associated with increase in visibility is at least 2.7 cents per
person in attendance, or approximately $30,000 for a typical season's
attendance. This benefit of a one mile visibility improvement represents
somewhat less than one million dollars per year for baseball attendance in
the entire U.S., assuming a homogeneous population.
-------
180
three stand out. In the earliest study, Davis and Knetsch (DK) compared
willingness to pay elicited in contingent valuation with a valuation derived
through a travel cost model of demand. DK found the two estimates to be
strikingly similar in magnitude. However, later work by Bishop and Heberlein
(BH) suggested that the similarity found by DK might be misleading. Three
of the BH results are relevant. First, travel cost valuations computed by
BH were found to vary widely depending upon the choice of elements included
in the cost of travel index that serves as price. Thus, a single travel cost
estimate may be unreliable as a datum. Second, when compared to a range of
travel cost estimates, the contingent valuation estimate lay close to the mean
of the travel cost valuations. Third, both contingent and travel cost valuations
tended to underestimate the BH datum of true value. In a third and most recent
comparative study, Brookshire et al. found, in a manner consistent with a theory
of individual versus market valuations, that valuations of visual air quality
based on contingent valuation tended to lie below those based upon a rent
gradient estimated on residential property prices. In light of the results
of previous studies, two tentative conclusions can be drawn. First, contingent
valuation performs at least as reliably as the operational, alternative
valuation techniques. Results presented below tend to corroborate previous
research.
-------
181
3.3.1.1 Early Analysis of Hancock Tower Visitation
The Hancock Tower offered an unusual opportunity to determine the
effects of visibility on the demand for view services. The view offered
by the Tower is particularly sensitive to changes in visual range. Since
an explicit price is charged and attendance is recorded it was possible
to provide an estimate of the demand for Hancock Tower view services as a
function of admission price, visibility, and a set of demand shifters.
A mean per person consumer surplus of $2.12 in 1981 prices was computed
from the aggregate demand estimate. Extrapolating this benefit estimate
to cover the entire eastern United States is equivalent to assuming that
identical viewing opportunities (as the Chicago urban landscape and skyline)
exist in the entire eastern region. Assuming that similar experiences are
obtainable in other areas of the region, then, given a homogeneous population,
the aggregate consumer surplus is 275 million dollars in 1981 prices.
Early empirical analysis of Hancock Tower visitation completed four
objectives. First, the error structures resulting from previously specified
models were examined for non-random patterns and remedial estimation pro-
cedures employed where appropriate. Second, having selected appropriate
estimation procedures, lagged groups of independent variables were tested
for explanatory power. Third, the functional form of the specified equa-
tion was evaluated. Fourth, preliminary estimates of consumer surplus and
revenue were computed for changes in visibility at the site.
The empirical analyses began with a demand equation specified in inverse
exponential [IE] form. Such a functional form appeared most consistent with
the color contrast results of Malm and Leiker. An examination of the error
structure resulting from estimation in the IE form revealed a clearly non-random
-------
182
pattern. To remedy this difficulty, two steps were taken. First, the model
was respecified in a simple linear form. The linear form was chosen since
it can be viewed as a first-order approximation to more complex functional
relationships. Second, a modified Cochrane - Orcutt (C-0)^ procedure was
used to allow for serial correlation errors and their effect on estimation.
Combining the linear form with the C-0 procedure resulted in an error structure
approximating an i.i.d. process and, thus, appropriate for the computation
of covariance statistics.
The second step in the empirical analysis was to check the explanatory
power of lagged groups of variables. Conceptually, lagged variables could be
important for two reasons. First, if the visiting population is fairly con-
stant, extremely favorable visibility and weather conditions on a given day
would tend to deplete the visitor stock for the nest. Within this context,
lagged variables would tend to carry signs opposite to those of the respective
comtemporaneous variable. Second, individuals may form expectations on the
basis of past realizations of visibility and weather variable. In this
context, the signs of lagged variables would depend upon the particular
processes used to form expectations. Given this ambiguity, the net effect
on the signs and significance of lagged variables cannot be determined a priori
To determine the empirical effect of lagged independent variables, F
statistics (Chow type test) were computed to test several hypotheses. The
basic form of the null hypothesis was : *d - the lags x,y, and z do not
contribute to variation in visitation. The set of variables lagged were
VS1, VS2, RA, SN, CL, WIN, TEMP, and FG (see Ta. 3-7 for variable description).
^ee SAS AUTOREG procedure, SAS Institution, 1980.
-------
183
TABLE 3-7
Statistic and Variable Descriptions
for Visitation, Weather and Visibility
VARIABLE
NAME
MEAN
STANDARD
DEVIATION
DESCRIPTION
VST
VS1
VS2
RP
RPI
M,TU,W,
F,S,SU
955.12
12 .55
16.28
0.76?
916.91
0.14
710.77
13.94
15.42
0.07659
9.23
0.35
Daily Ticket sales at
Hancock Tower
Visibility in miles from
H.T., 1st reading
Visibility in miles from
H.T., 2nd reading
Admission price divided
by C.P.I.
Personal Income (National)
divided by C.P.I.
Day of week dummy
variables
TIME
SNX
CSX
RA
SN
CL
WIN
TEMP
270.50
0.2169
.01215
0.0700
0.0719
0.4727
10.82
50.72
151.41
0.6896
0.6922
0.1950
0.2145
0.3262
3 . 983
22.09
Linear trend variable
runs from 1 to 524
SINE Values with period
of 365 days. Intended to
pick up seasonal cycle
COSINE Values with period
of 365 days. Intended to
pick up seasonal cycle
Proportion of days with
rainfall
Proportion of days with
snowfall
Average cloud cover
measured from 0 to 1.
Average windspeed in Knots
Temperature in degrees
Fahrenheit
FG
0.08715
0.2418
Proportion of days with fog
^"Observations are for the period Iron 1/9/81 to 6/15/81.
Weather observations are for O'Hare Int. Airport.
-------
184
The lags tested were lags 1,2,3,7,8 versus lags 1,2,7; lags 1,2,7 versus lags
1,7; lags 1,7 against lag 1; and lag 1 against an equation with no lags. The
statistic used for testing was
where SSE is the sum of squared errors resulting from the regression without
lags x,y, and z; ""~s degrees °f freedom associated with SSE^ ; and
SSEg and are analogous quantities for the regression with lags x,y,z
included.
Ta. 3-8a and 3-8b exhibit the results of regressions computed with various
sets of lagged variables. At the 5 percent level, Chow test computed from
the given statistics failed to reject any of the null hypotheses involving
lagged groups of variables. Hence, none of the lagged groups of variables
are shown to contribute to the variation in visitation. Additionally, inspec-
tion of Ta. 3-8a and 3-8b shows that the lagged variables contribute little
to the long run effects on visitation. For example, the combined effect of
VS1 and VS2 in the regression with no lags differs little from the long run
effects when lags are included. Similar results are apparent for other
variables such as RP and PR1. With their effects neither statistically nor
absolutely significant, lagged effects are provisionally rejected in favor
of the more parsimonious contemporaneous equation.
With a satisfactory specification of demand for Hancock Tower visitation,
consumer surplus and revenue changes were estimated for various
percentage changes in mean visibility. Results appear in Ta. 3-9. For these
-------
185
TABLE 3-8 a
LAGGED VARIABLES AND THEIR LONG RUN EFFECT ON VISITATION
' tOHC RUN COEFFICIENTS1
LACS
EXPLANATORY
VARIABLES
1,2,3,7,3
1,2,7
1,
,7
1
HOME
HONE
(HS1 DROP!
SVS1(2>
-4.38
-3.90
-4,
.13
-1.77
2.49
—
(1.49)
1732
11.13
13.29
12.
.60
12.05
7.10
8.49
(4.63)
(7.17)
:sa
-445.83
-527.45
-462,
.84
-403.52
-535.89
-541.86
(-5.87)
(-5.94)
ESS
-188.07
-127.25
-69.
.83
-125.31
-175.33
-183.07
(-2.07)
(-2.16)
icl
-143.02
-221.92
173.
.64
-226.86
-169.03
-174.33
(-3.05)
(-3.17)
ETOI
6.19
-U.92
6,
.32
1.92
2.26
2.00
(0.52)
(0.46)
utemp
1.81
2.08
0.
.81
3.26
5.70
5.16
(2.75)
(2.50)
IFG
-283.03
-271.74
-457,
.33
-317.19
-316.38
-317.97
(-3.90)
(-3.92)
R2
-1613.83
-1752.92
-1908.
.37
-1360.13
-1492.49
-1376.04
(-2.00)
(-2.23)
(-2.
.30)
(-1.79)
(-2.05)
(-1.35)
SSI
23.04
23.34
26.
.77
23.33
24.21
23.76
(1.98)
(2.05)
(2.
.38)
(2.29)
(2.21)
(2.17)
M
-6.44
0.41
7,
.68
-9.30
-13.59
-11.90
(-0.09)
(0.00)
co.
.11)
(-0.13)
(-0.19)
(-0.17)
TO
-66.53
-64.82
-66.
.71
-74.58
-64.75
-63.62
(-0.93)
(-0.92)
(-0.
.96)
(-1.08)
(0.94)
(-0.93)
W
-29.26
-37.70
-43
.20
-67.97
-60,14
-56.92
(-0.47)
(.-0.62)
(-0.
.72)
1-1.14)
C-i.au
(-0.97)
?
311.55
302.03
311.
95
292.21
295.83
299.01
(4.95)
(4.91)
(5,
.15)
(4.92)
(4.98)
(5.06)
S
1071.55
1070.22
1074.
,62
1058.59
1363.66
1072.23
(14.90)
(15.18)
(13.
,40)
(15.35)
(15.43)
(15.62)
so
319.21
313.91
320.
,55
314.54
315.99
321.99
(4.30)
(4.31)
(4.
.45)
(4.39)
(4.41)
(4.51)
TIME
1.73
1.77
1.
.99
1.69
1.71
1.61
(2.57)
(2.70)
(3.
.09)
(2.68)
(2.30)
(2.50)
ssx
-10.22
16.38
14.
.16
-4.30
29.20
14.31
(-0.12)
(0.21)
(0.
19)
(-0.06)
(0.42)
(0.21)
CSX
-407.94
-389.01
-437.
.34
-359.99
-303.30
-311.59
(-2.99)
(-3.24)
(-3.
36)
(-3.99)
(-3.38)
(-4.00)
nrr
-19638.55
-19777.28
-22361.
00
-22073.96
-21036.09
-20638.35
(-1.30)
(-1.35)
(-2.
.16)
(2.12)
(-2.04)
(-2.31)
EVS1+EVS2
6. "5
9.39
3.
, 45
10.28
9.59
3.49
t values given la parentheses
'Coefficients estimated usint* the SAS AUTOREG procedure with mtoccrrelation
coefficients estiaated at las?g3 I and 7.
Indicates the sum of the coefficients of both contemnoraneaus and lagged
values of the particular explanatory variable. For exaraole, if i. and
7 are included, :.VS2 ^ives The sum of the coefficients cat i.*nnc»»d on che
contemporaneous value of r-"S2 and che values of 7S2 at ia^s I and *.
-------
186
TABLE 3-8b
1
Statistics for Regressions
LAGGED EXPLANATORY VARIABLES
REGRESSION
WITH LAGS
SSE
D.F.
R2
P1
?7
1,2,3,7,8
62693407
464
.65
.31
.14
(7.54)
(3.39
1,2,7
63477889
480
.64
.32
.15
(7.66)
(3.55
1,7
64670558
488
.64
.32
.14
(7.72)
(3.35
1
65825254
496
.62
.32
.13
(7.72)
(3.28
NONE
67334226
504
.63
.32
.13
(7.66)
(3.26)
NONE
67518458
505
.62
.32
.13
(VS1 DROPPED)
(7.66)
(3.16!
t values in parentheses
1 ...
Autoregressions estimated with autocorrelation
coefficients estimated at lag l(p^) and lag 7 (p )
-------
187
TABLE 3-9
Consumer Surplus and Revenue Estimates
1
Derived from Linear Demand Function
AVERAGE DAILY CHANGE2
CHANGE IN MEAN
VISIBILITY CONSUMER REVENUE TOTAL TOTAL
(VS2 = 16.28) SURPLUS
10%
26
28
54
19710
20%
52
57
109
39785
30%
78
85
163
59495
40%
105
113
218
79570
50%
133
115
248
90520
1
Estimated from regression without interaction
term as reported in Table 4. In dollars.
2 .
Adjusted to current dollars using April 1981,
C.P.I, of 266.8.
-------
188
computations the regression "None (VSI Dropped)" of Ta. 3-8a was used along
with the mean variable values given in Ta. 3-7. Revenue changes were included
since, at this point, it is assumed that additional visitors are admitted to
the Tower at close to zero marginal cost.
Caution must be taken against placing too much weight on the estimates
Of Ta. 3-9. As Ta. 3-10 demonstrates, the response of individuals to changes
in visibility is very likely non-linear. Ta. 3-10 gives results for two
regressions. The first regression, "No Interaction," is entirely linear in
the coefficients of all included variables. Note that the coefficient on
visibility is rather small. The second regression, "With Interaction Term,"
includes two terms for visibility. The first is simply VS2. The second is
VST2 > 10 = VST x D ,
where
D = 1 if VST2 > 10 miles ,
= 0 otherwise.
The regression "With Interaction" clearly demonstrates a differential response
to different ranges of visibility. When visibility is less than 10 miles the
response in visitation to a one mile change in visibility is 23.91 versus the
8.49 person response of "No Interaction." When visibility is initially greater
than 10 miles, the response to a one mile change in visibility is 9.6
(=23.91 - 14.31) and still greater than the 8.49 person response of "No
Interaction." From these results, two implications can be drawn. First,
non-linear forms should he explored for fit to the Hancock data; second,
consumer surplus and revenue simulations performed with the "With Interaction"
regression or other non-linear forms are likely to result in significantly
larger estimates.
-------
189
TABLE 3-10
TESTING FOR NON-LINEAR RESPONSE TO VISIBILITY
REGRESSION RESULTS
EXPLANATORY
VARIABLE
NO INTERACTION
TERM FOR US2
WECa INTERACTION
TERM
Err
VS2
VS2 > 10
RP
RPI
M
TO
W
?
S
SO
TIME
SMC
CSX
RA
SN
cc
WIN
TEiP
TG
a2
SSE
OF
-20638.85
(-2.010)
8.49
(7.17)
-1376.04
(-1.85)
23.76
(2.17)
-11.90
(71.80)
-63.62
(-0.93)
-56.92
(-0.97)
299.01
(5.06)
1072.28
(15.62)
321.99
(.4-51)
1.61
(2.60)
14.81
(0.21)
-311.59
(-4.01)
-541.86
(-5.94)
-183.01
(-2.16)
-174.83
(-3.17)
2.00
(0.46)
5.16
(2.50)
-317.97
(-3.92)
o.6:
67518458
505.
-20640.20
C-2.00)
23.91
(3.47)
-14.31
(-2.27)
-.1416.66
(-1.90)
23.68
(2.17)
-12.68
(-0.18)
-59.36
(-0.86)
-45.23
9-0.76)
313.52
(5.26)
1095.40
(15.74)
344.40
(4.77)
1.61
(2.60)
21.20
(0.30)
-310.95
(-4.00)
-540.29
(-5.95)
-171.16
(-2.03)
-183.05
(-3.33)
1.64
(0.38)
5.24
(2.55)
-307.39
(-3.81)
. 52
66831464
5G4.
t values in parentheses
-------
190
3.3.2 The General-Choice Model
The activity or action of record at HTO is not the enjoyment of viewing
services but the number of individuals purchasing access to the viewing site.
At any particular admission price, the quantity of access supplied is assumed
to be perfectly elastic within the range of realized visitation. Given this
perfect elasticity of supply, a demand function can be estimated through simple
regression techniques and without reference to problems of simultaneity.
The demand for access to HTO may be thought of as derived from an
individual's use of access in producing viewing services given the characteristics
of the observatory, the city skyline, and environmental conditions including
visibility. The most notable aspect of demand is that, at the individual level,
it is discrete: an individual either accesses Tower services or does not.
Borrowing from the relevant literate on discrete choice (Domencich and McFadden),
aggregate demand can be represented by
(3-1) VST,. = N tt
L t
where VSTfc is total visits on day t, is a pool of potential visitors on day
t, and tt is the probability that an individual in N^_ visits the HTO. More
specifically, ~ is the probability that the utility gained by an individual
through a set of activities that includes an HTO visit is greater than the
utility of all sets of activities that do not include a visit to HTO.
-------
191
Variables relevant to the determination of N^_ and t can be identified
by considering the abbreviated "decision tree" (Domencich and McFadden)
given in Fig. 3-1. On any particular day one can imagine that individuals
sort themselves out over mutually exclusive activities as indicated by the
direction of the arrows in Fig. 3-1. However, as the literature on discrete
choice points out, the flow of information and choice is just the reverse of the
sequence of actions. That is, individual choice begins at Branch 4 in
Fig. 3-1. To make the Branch 3 decision between downtown activities and
other alternatives, the individual must first select the optimal package of
downtown activities. The decision at Branch 3 can then be made optimally by
comparing the utility gained from the best set of downtown activities with
the utility gained from the best set of alternative activities.
To identify variables relevant to choice, decisions represented in
Fig. 3-1 are partitioned into those made in the longer run and those made in
the short run. For example, choices above Branch 3 are likely to require
major commitments of personal resources and be relatively fixed by long term
contracts. For these long run decisions, the most important variables to the
HTO visit choice are likely to be time series variables. Clearly, for the
individual, relative prices contemporaneous to the long run decision may be
inportant indicators of future relative prices. However, in the research
problem at hand, this portion of the the individual's information set remains
unobservable and must be relegated to an error term. Time series variables,
however, are observable and are likely to be quite pertinent to long run
individual planning. For instance, seasonal merchandizing sales and weather
conditions are probably best judged by seasonal or other time series variables
-------
192
FIGURE 3-1
Decision Tree for Choice of Activities
Branch 1
Location
Branch 2
Branch 3
3ranch 4
Metropolitan Chicago
N/
Leisure (N^)
Downtown Activities
Other (Nq)
Work (>
Other
Activity Sets Including HTO
Activity Sets
Excluding HTO
-------
193
Specifically, for purposes of long run decisions, an individual can expect
prices at downtown shopping areas to be relatively high in December but low
in January; it is likely to be cooler in January than in July but whether
January 1 or January 7 is colder is largely a matter of random occurrence.
In addition, day of week effects may enter due to conventions of a 40 hour
workweek and work scheduling. For the long run decisions of location and
work/leisure choice, the information (potentially observable by the
researcher) passed back up the decision tree therefore depends largely upon seasonal
and other time series considerations. Thus, if decisions above Branch 3
are primarily long run decisions, we can write the pool of potential HTO
visitors on day t at Branch 3 as a function
(3"2) NtML =NtML(s'd'e:) '
where s is a vector of time series variables, d is a vector of day of week
dummy variables, and e is an error term introduced for unknown price
information used by individuals.
For individuals within N , a decision regarding the day's excursion
t M L
must be made. Assuming that the choice between downtown and other activities
is fairly decisive and that variables specific to HTO contribute rather little
1 . . .
to choice at Branch 3 , the only variables affecting choice at Branch 3 that
are also potentially observable by the researcher are local weather conditions.
Entering these local weather conditions as a determinant of the visitor pool,
1 ...
The assumption is not entirely unreasonable. Of the individuals sampled at
HTO, 75 percent indicated that their visit HTO was only a sidetrip and
apparently not crucial to their visit downtown.
-------
194
we can write
(3-3) NtMLD = *\MLD^s, u - z • ) ,
h v n m o oi J '
2
Small and Rosen have suggested the conditional maximization process in dealing
with discrete choice.
-------
195
where Z, • and z , are the respective deviations of individual utility from
ru 01
the utility of the typical individual. Eq. (3-1) can now be written
(3_/) VSTt = NtMLD\
= NtitjLDCs,dfw,e)irhCp,w,m,ii,ph3.
Assuming that and are extreme value or Weibull distributed, ^
can be written in terms of the cumulative logistic distribution (Domencich
and McFadden):
(3-8) VSTt = vh(m-nph)/( v^m-np^ + vqb )) ,
where
To proceed further with specification, specific funtional forms must
be applied to N , v, , and v . For present purposes the most tractable
^ tMLD h' o
functional form is the general Cobb-Douglas (CD) form, xaexp (b + cy+e) where
x is a continuous variable, y is a dummy variable, e is a log-normally
distributed error term, and a, b , and c are the coefficients of interest.
Applying this general CD form to the aggregate demand equation in eq. (3-8)
an estimable form is
(3-9) InVST = lnA(s,d,w,p,e) + InCm-np^) + lnCv^(m-np^)+vom) ,
where A(.) is of the form X exp(b+cy+e) . Because we have no information on
the typical excursion budget or group size of individuals in X , the log
t M L D
terms which include m are replaced by first order Taylor series approximations.
The approximation to be estimated is
(3-10) InVST _ = a, + lnA(s,d,w,) + b,p, + Ins »
L 1 X il
where again A(.) is of the general CD form, is a constant term, and p^
enters the equation in level form with coefficient b .
-------
196
Given an estimate of eq. (3-10), it can be shown by direct intergretion
that approximate total surplus is defined by estimated visits, "/ST , divided
by the coefficient of admission price, b^. Thus, approximate average or
expected surplus obtained per person visiting HTO is
(3-11) AVCS = (VST/b^VST
= l/b1 •
A
Because the error bounds on b^ are straightforwardly calculated, AVCS is
selected as the basis of contrasting demand-based valuation with contingent
valuation in the HTO case.
333 The Contingent Valuation Experiment
During the Spring of 1981, a contingent valuation instrument was designed
3
that would elicit the maximum willingness to pay (MWTP) for access to HTO
During the summer of 1981, contingent valuations of visiting groups at HTO
were recorded. Valuations were obtained under a variety of environmental
conditions and, by the end of the summer, 319 usuable observations had been
recorded.
Ta. 3-11 displays the results of the contengent valuation experiment at
HTO. MWTP is the maximum willingness to pay elicited. ADMCOST gives the
average actual cost of admission. Average SURPLUS per group is MWTP minus
ADMCOST or an average of 3.93 dollars. Finally, average GROUPSIZE was 2.67
for groups during the summer of 1981.
-------
197
TABLE 3-11
Results of the 1981 Contingent Valuation Experiment
at the Hancock Tower Observatory
1
Variable Sample Mean Standard Error
MWTP 9.43 .428
ADMCOST 5.50 .199
SURPLUS 3.93 .314
GROUPSIZE 2.67 .115
1
Number of respondent groups was 319. Means m this Table are
computed for groups, not individual persons. Covariance between
SURPLUS and GROUPSIZE is 4.59.
-------
198
During the Spring of 1981, the HTO management apparently decided to
experiment with well-publisized price variations in order to determine the
relationship between price and attendance. For the purpose of estimating
demand, the price variation was sufficient enough for a statistically
significant estimate of the coefficient on admission price as shown in
Ta. 3-13. By using the variable definitions given in Ta. 3-14, it is
clear that the overall specification of the estimated equation (Ta. 3-14)
paralleled the identification given in eq. 3-10. Relevant statistics
for the secondary data are given in Ta. 3-15.
The coefficient of central interest is the coefficient on admission
price, the variable PP. By inverting the coefficient and using the
approximation formulas given in Mood, Graybill, and Boes (p. 181) for quotients
of random variables, average surplus, AVCS, was computed and is presented in
3-16. jn same Table and computed using the same approximation formulas,
the average from contingent valuation (AVCV) is also given. Given the fairly
large sample sizes, a z statistic was computed for the difference between
AVCS and AVCV and is also given in Ta. 3-16. Quite clearly, the z statistic
indicates no statistically significant difference between the two means at
conventional levels of significance.
The Hancock Tower Observatory in Chicago offered conditions suitable
estimates of both a demand based valuation of access to the Observatory and
a contingent valuation of access. Given the functional form developed for
aggregate demand, average consumer surplus per person-visit to the Tower
-------
TABLE 3-13
Regression
Estimates of an
Aggregate Demand
for Access
to HTO, March
15 to May 31, 1981
PARAMETER
STANDARD
VARIABLE DF
EST I MATE
ERROR
T RAT I 0
PFiOEO- i T i
INTERCEPT 1
-33 . 47981 6
1 4 . 5981 37
-2 . 2934
0 .
0253
LNVIS 1
0.139551
0 . 054726
2.5500
0 .
0133
PP 1
-0.532835
0 . 1 92970
-2.7612
0 .
0076
MAR 1
0 . 327406
0 . 1 95630
1.6736
0 .
0994
MAY 1
-0.334280
0.125514
- 2.6633
0 .
0099
M 1
-0.171819
0.181041
-0.9491
0 .
3464
T U 1
-0.348115
0.159548
-2.1819
0 .
0330
W 1
-0.126686
0.158907
-0.7972
0 .
4285
F 1
0.375736
0.158148
2.3758
0 .
0207
S 1
0 . 786929
0 . 1 58722
4.9579
0 .
0001
S U 1
0.271636
0 . 161977
1 .6770
0 .
0987
RA I N 1
-0.926709
0.215838
-4 . 2935
0 .
0001
TSC 1
-0.00239967
0.001542321
- 1 .5559
0 .
1 250
FOG 1
-2.295919
0.297832
-7 . 7088
0 .
0001
LNWIN 1
0.034347
0 . 1 28057
0.2682
0 .
7895
LNTMK 1
7.136954
2 . 61 2609
2.7317
0 .
0083
LNT 1
0.232934
0.116005
2.0080
0 .
0492
HAZE 1
-0.090610
0 . 395829
-0.2289
0 .
8197
SSE
7.601226
F RAT I 0
" <;. >>\)
DFE
60
P RO B > F
0
. 0001
DEP VAR: LNTVST
MS E
0 . 1 26687
R-SQUARE
0
.8759
-------
200
TABLE 3-14
Definitions of Variables Used in Estimating
Aggregate Demand
Variablel
Definition
LNVIS
PP
MAR
MAY
M, TU, W,
F, S, SU
RAIN
TSC
FOG
LNWIN
LNTMK
LNT
HAZE
Log of visibility where visibility is measured
in miles.
Price of admission to HTO in dollars.
Month of March dummy variable (March=l, 0 otherwise)
Month of May dummy variable (May=l, 0 otherwise) .
Day of week dummy variables (No dummy variable
entered for Thursday).
Proportion of day in which rain fell.
Total sky cover in percent.
Proportion of day with fog.
Log of wind speed where wind speed is measured
in mph/10.
Log of temperature where temperature is in degrees
Kelvin.
Log of a time series variable beginning with 1 on
March 15 and running consecutively through the
intergers to 78 on March 31.
Proportion of day with haze.
1
All weather observations except visibility were recorded at O'Hare
International Airport in Chicago. Visibility was recorded at HTO.
-------
201
TABLE 3-15
Sample Statistics for Variables Used in
Estimating Aggregate Demand, March 15 to May 31, 1981
VARIABLE MEAN + STANDARD
DEVIATION
LNTVST
6.58799580
0 . 89175811
LNVIS
2.56384683
1 . 12190785
PP
2.1314102 6
0.28411505
MAR
0 .2 17 94872
0 . 41552 458
MAY
0 .3 97435 90
0 . 49253502
M
0 . 141025 64
0.3503007 6
TU
0 . 141025 64
0 . 3503007 6
W
0 . 141025 64
0.3503007 6
F
0 . 141025 64
0 . 3503007 6
S
0 . 141025 64
0.3503007 6
SU
0 . 1538 4615
0 . 36313652
RAIN
0 . 11111111
0.2557 6565
TSC
69.35897436
32.98544737
FOG
0.06410256
0 . 20142 130
LNWIN
2.40314246
0.37150081
LNTMK
5.65218864
0 . 02217227
LNT
3 .39643141
0.91573362
HAZE
0.04273504
0.12436244
TVST*
931.61538462
567.76436101
VISB1**
2 0 . 2 65338 62
15 .42756495
¦k
Total daily visits recorded at HTO.
£ £
Visibility in miles recorded at HTO
* Number of observations equals 78.
-------
202
TABLE 3-16
Estimates of Mean Per Person Consumer Surplus
Obtained by Access to the HTO
Mean per person surplus from aggregate
demand estimate (AVCS): $2.12
Variance: -46
Mean per person surplus from contingent
valuation estimates (AVCV) : $1.47
Variance: -01
Test statistic: z - ( 2.12 - 1.47
Conclustion: Do not reject null hypothesis of no
significant difference between AVCS
and AVCV.
-------
203
embodied the most desirable statistical properties. On the basis of a
comparison of average estimated surpluses, the hypothesis of a statistically
significant difference between demand-based and contingent valuation was
rejected. Thus, consistent with the results of other researchers,
contingent valuation is shown to perform at least as well as the next best
operational alternative in valuation.
-------
204
3.4 VIEW-ORIENTED RESIDENCES
Clean air and attractive vistas are firmly established as valuable
dimensions of environmetnal quality. Analysis shows that there are substan-
tial benefits derived from clean air and that it is a valuable resource in-
deed. Typical is the housing market analysis of Bender et al_. (1980) which
shows that for a uniform 20 percent reduction in particulate concentration
in Chicago the average household is willing to pay approximately $600 per
year. Using a survey approach Brookshire et al. (1982) estimate that the
typical household is willing to pay approximately $310 per year for a 30
percent reduction in pollutant concentrations in Los Angeles. Further
analysis shows that attractive views yield benefits to which approximately
9 percent of some house prices in Sydney (Abelson, 1979) and 15 percent of
some rents in Chicago (Pollard, 1977) can be attributed. Rowe et al.
(1980) find that people will bid approximately $100 per year for clear,
unpolluted vistas in the Grand Canyon National Park Area.
This study takes as its point of departure an earlier paper, "Visibility,
Views and the Housing Market" which suggests that intensive
analysis of view-oriented submarkets of the residential housing market
would be productive. The objectives of this research are: (1) to measure
the values of views and view characteristics including visibility using
a survey instrument which establishes a contingent market for each; (2)
to measure the values of views and view characteristics using a hedonic-
demand analysis of housing consumption for the same group surveyed and
(3) compare the contingent values from the survey and the implicit values
from the housing market for individuals dwelling in view-oriented residences.
-------
205
To insure comparability, a survey was conducted among Chicago
residents of high-rise buildings along Lake Michigan. The survey
instrument was designed to elicit contingent values for views, view
characteristics and visibility and to get from the same individuals
sufficient information to estimate the values of some of the same
amenities from their housing consumption. An abbreviated bidding game
was used to obtain contingent values. During the period May through
September 1981, a team of interviewers collected 208 responses from
residents of 10 high-rise buildings located mostly north of Chicago's
Loop. Although further verification was warranted, the integrity of
the data was well enough established that some results can be reported.
3.4.1 Contingent Values for View-Oriented Residences
3.4.1.1 Willingness to Accept Payment for No View
Residents of units with relatively unobstructed views of the lake
and/or Loop were asked how much their monthly housing payments would
have to be reduced for them to choose a unit with no views. Of those
who responded, 92 percent replied that the amount would have to be
greater than $50; only 8 percent replied that they would choose a
viewless unit for a $50 reduction. The mean of the responses to the
query about the minimum amount individuals would be willing to accept
-------
206
for loss of view is $169.39. It should be noted that this is average for
only 40 percent of the sample and does not incorporate the 6 0 percent who
bid zero, an infinite amount or did not respond.
3.4.1.2 Willingness to Pay for Lake View
Residents who do not have an unobstructed view of the lake were asked
how much their monthly housing payment could be increased if they got a
good lake view. Of those who responded, 52 percent replied that the amount
could be more than $30; 48 percent replied that they would choose their
current unit without a lake view if the amount was $30 or more. The mean
of the responses to the query about the maximum amount individuals would
be willing to pay for a lake view is $43.06.
3.4.1.3 Willingness to Pay for a Unit which Is Ten Floors Higher
All residents were asked how much their monthly housing payments could
be increased if they got otherwise identical units 10 floors higher than
their current units. Of those who responded 73 percent replied that the
amount would have to be less than $30; 27 percent replied that they would
choose the higher unit even if the payments increased by $30. The mean of
the responses to the query about the maximum amount individuals would be
willing to pay for the higher unit is $25.32. The average is based on
responses from 79 percent of the 2 08 people surveyed.
3.4.1.4 Willingness to Pay for Better Visibility
All residents were asked how much their monthly housing payments could
be increased if they got more days with better atmospheric visibility. This
improvement in visibility was described by showing residents 9 color photographs
-------
207
which depict three Chicago lakefront vistas under visibility conditions of
3 miles, 13 miles and 3 0 miles. These ranges occur throughout the year and
under current conditions there may be 12 consecutive days of 3 mile visibility.
The specified improvement would reduce to four the number of consecutive days
with only three mile visibility. All people surveyed responded and 65 percent
replied that the amount their monthly payments could increase would be $10 or
more; 35 percent rplied thay they would choose current visibility conditions
if they were to pay $10 per month. The mean of the responses to the query
concerning the maximum amount individuals would be willing to pay for the
improvement in visibility is $14.27. The average is based on responses from
99 percent of the 208 people surveyed.
3.4.1.5 Implicit Value from the Housing Market
Using the same survey instrument containing the contingent valuation experi-
ments, data on housing consumption and consumer characteristics were collected.
Some tentative estimated can be made from a housing hedonic equation for
renters. The housing hedonic equation is
(3-12) RENT = 100.96 + 28.950 TOTROOMS + 83.918 BATES + 0.0816 AREA
(2.90) (3.77) (1.98) (1.75)
+ 41.995 CARPET + 19.994 DISHWASH + 2.6219 FLOOR
(3.31) (0.72) (2.67)
+ 0.0139 WARUN + 0.21135 LWARA
(0.09) (1.53)
2
R = .8537 F = 28.44 n = 48
where RENT is monthly rent in dollars, TOTROOMS is total rooms, BATHS is
number of bathrooms, DISHWASH is 1 if the apartment comes furnished with a
-------
208
dishwasher and 0 if not, FLOOR is the number of floors up the apartment is
in the building, WARUN us square feet of total window area with unobstructed
view, and LWARA is square feet of window area with an unobstructed view of
Lake Michigan. Of the view-related characteristics, FLOOR is significant
at the 2 percent level, LWARA is significant at the 14 percent level, but
WARUN is not significant at any reasonable level.
Estimates based on this housing hedonic equation may be biased and
imprecise since (1) relevant housing characteristics may have been omitted,
(2) the functional form of the hedonic housing equation may be nonlinear,
(3) the benefits might have to be estimated from demand equations and not
directly from the average hedonic prices, (4) the remaining 160 residents
may differ from the 48 in the sample, and (5) data errors may remain.
3.4.1.6 Implicit Value of a Unit which Is Ten Floors Higher
The value of height and the associated breadth of view is obtained
by multiplying the coefficient of FLOOR by the 10 floor change in height.
The value of the increase in height is (2.6219) (10) = $26.22 per month.
This value is remarkably close to the contingent value of $25.32 from the
bidding experiment.
3.4.1.7 Implicit Value of a View
The value of a lake or Loop view would be obtained by adding the products
of the coefficients of WARUN and LWARA with their respective changes in window
area. Performing the calculation gives an implicit value which is appromi-
mately one-third of the average contingent value. However, the difference
could be easily due to 44 percent of the contingents bids being excluded
from the sample and the (perhaps overly) restrictive definition of WARUN.
-------
209
3.4.2 Estimates of the Values of Views and View Characteristics
The similarity of the contingent and implicit values for height (10
floors up) , the high response rate on the bidding experiment and the highly
significant coefficients in the renters' housing hedonic equation are favorable
to the use of contingent value of better visibility for policy analysis.
Aggregation of individual values over the population residency in the view-
oriented submarket would be straightforward, but it must be recognized that
this subgroup has high annual incomes (the average is $33,000) and is well-
educated (the average is some graduate work). Values of views and visibility
from this submarket must be considered in the social value of improved air
quality, but they are likely to be higher than those values of the entire
population which is less oriented to views, view characteristics and visibility.
-------
210
3.5 AIR AND AUTO TRAFFIC
3.5.1 Visibility and Air Traffic
Lowered visibility imposes costs on air travelers in many ways. If
visibility falls below three miles, all traffic must operate under Instrument
Flight Rules (IFR). All general aviation for flight training or recreation
which is not IFR rated must terminate. The people engaged in general aviation
lose the benefits gained from flying, aircraft rental operators lose revenue,
and airports also lose revenue from landing fees. Those still engaging in
aviation experience losses in waiting time since aircraft must maintain
greater increments between each other under IFR conditions. Not only do
travelers experience time costs in queuing, but also may miss connecting
flights or appointments. Under lowered visibility, the probability or air
accidents also increases. If visibility is poor enough to cause an in-flight
diversion, the traveler's involved and airlines suffer losses. The nature
of these costs are discussed in detail, and a formal economic model developed
later in this section. This model captures consumer behavior under visibility
constraints on air travel and provides a framework for measuring the net cost
or benefits of lowered visibility on air travel.
In the next section, a generally used method of measuring the
cost/benefit structure is outlined and critiqued. A formal model of utility
maximization is presented. Finally, empirical estimates of visibility effects
on total take-offs and landings at three Chicago area airports are presented
and discussed within the context of the economic model.
One procedure used in estimating net benefits is to regress the affected
variable on a vector of independent variables. In this case, air traffic
-------
211
counts would be regressed on visibility (possibly current and lagged), and
a vector of other weather variables. The equation would resemble
(3-13)
C.^. = a + a, V\ + a W.. + e..
it 0 1 it —z —it it
whereC.^ is traffic counts at the ith airport in period t. W. and V,. are
it _lt lt
vectors of airport-specific weather and visibility variables in time t, and
the stochastic error term, is taken to be the effect of changes in
visibility on traffic counts. In log form is the elasticity of traffic
counts with respect to visibility. Then an average value for a traffic
count is determined and if = 10%, then a one percent change in would
imply a 10 percent decrease in counts. So the number of counts lost times
the average value is the cost of decreased visibility.
When presented in this way, several important points emerge. Besides
the obvious problem is assessing the value of a count lost, ct^ is neither
a supply nor a demand elasticity. It is an amalgam of supply and demand
effects. Consider the simple supply and demand structure:
(3-14)
(3-15)
c = Tl Vt «• y2 Wt . Y3 Pc
C = e V + 3, w + 3, P
it —^ —t 3 c
S D
Setting counts supplied (C ) equal to counts demanded (C ) yields a reduced
form equation for the equilibrium counts (C ):
h
(3-16)
3 3
-1
Y1 S1
c— - —J vc + c-p - ~ 1 *
3 C r3 ®3
L '3
i
-------
212
If eq. (3-13) and (3-14) were the true underlying structure of supply and
of counts demanded, is the price elasticity of supply, y^ is the visi-
bility elasticity of demand and is the visibility elasticity of supply,
Clearly, interpreting ct^ as an elasticity is incorrect. In fact, cannot
be shown to be an upper or lower limit of the true underlying elasticities
Even if could be shown to be a limiting case of the underlying
parameters, just multiplying cs^ times the count value does not give a true
social cost. The count value chosen is usually an aircraft rental fee, or
a plane ticket price. These are at best lower bound estimates of the true
cost of the delays. They do not include the social cost due to inefficient
allocation of resources.
In this section, the problems of infering social cost estimates from
reduced form equations with no underlying structural model have been dis-
cussed. The importance of structural models in interpreting reduced form
coefficients was shown.
3.5.2 A Model of Air Traffic Responses to Lowered Visibility
Air transportation is an input to a demand for location change. Y, or
location changes, is the produced good directly entering the utility function.
In meeting the demand for a Y, the individual choses the lowest cost combina-
tion of productive inputs. Among the possible combinations is air travel,
either purchasing a ticket on a commercial airline or chartering a flight.
where y^ is the price elasticity
•Y3 b3
-------
213
There is also a time input involved which is the trip to the airport, the
time of the trip itself, and waiting time. Visibility affects the time
componenet of air transportation by increasing the landing or takeoff queue.
Consequently, the magnitude and direction of the visibility effects on pur-
chased inputs can be analyzed. The purchased input on which the analysis
focuses, in the aggregate, is the number of take-offs and landings per day
in Chicago area airports. The model presented below develops a method of
estimating the true social cost of visibility changes on Y by analyzing
effects in the input, or counts, market.
Following Tolley (1972), the demand curve for Y is
(3-17) Py = F(Y) ,
where Y is produced according to
(3-18) Y = Y(z,v) ,
v is the level of visibility which acts as a cost shifter. That is, changes
in v affect the amounts of x needed to produce the same level of Y. From this
framework the marginal cost of Y can be derived:
(3"19) py - pz
The right hand side of (3) is the marginal cost of producing Y, and y is the
marginal productivity of z in the production of Y.
The question to address is what are the costs associated with a decrease
in visibility in the framework presented by eq. (3-17) and (3-19). Fig. 3-2
reproduced from the Tolley paper, shows that a decrease in visibility shifts
the cost curve back, while leaving demand for Y unaffected. The social cost
associated with this shift is the shaded area. The analytic solution of the
area is
-------
FIGURE 3-2
I'K fCii
Social Cost Associated with
Meeting Location Demand
due to Decreased Visibility
yv
c (V )
y o
t-i
H
-P-
-------
215
Y
(3-20) C Cv) = J P Y dY ,
y J y zv '
Y
o
where is the effect on the marginal productivity of z of a change in v.
In order to view this cost in the framework of a model for counts, this
area must be transformed.
By substituting eq. (3-19) into eq. (3-20), this area is
Z1 Y
(3-21) c (y}=f p -|I dz .
z • J z Y
z
z
o
Y
zv
P is the supply curve for z, and —^ can be viewed as the percentage
z z
change in z's marginal productivity resulting from the change in visibility.
The graphical analog to (3-21) is shown in Fig.3-3. p is an upward
sloping supply curve for 2. Dz'v0' is demand for z derived from the
demand for Y under visibility v . D (v.) is the demand for z at the lower
1 o z 1
visibility level The cast associated with this fall in demand is the
shaded area in Fig.3-3. So, if Pz were invariant to changes in visibility,
the area ABC would be the associated social cost.
Now, consider the problem of a shift in P due to a change in visi-
bility. The supply curve P can be viewed as the standard supply curve of
an exhaustible resource. Fig. 3-4 presents the supply of counts curve for
an airport. As p*, the landing fee associated with this particular airport,
the supply of counts is completely elastic up to z, the technological or
legal bound on the number of counts which can be supplied per period. The
effect of decreased visibility is to add queuing time due to in-air stack ups
-------
216
T3
O
<+4 4-1
•r"1
-C +J
co o
cn
i
X Jh
rn
—t C3
C,2
w
«
Ef *->
S
CO 3
O
CL,
i-l
<4-1 C
0 H-<
+4 0)
,-c:
O +4
u
o
—< 4->
ss
~pH
o
o
c/3
-------
FIGURE 3-4
Supply Curve for Air Traffic Counts
-------
218
and take-off delays. Thus, at some point z*, the supply curve begins to
slope upward reflecting this increased true cost. The effect of visibility
& —
changes is to shift z across the interval (0,z) and thus shift the upward
sloping portion of the supply curve.
The cost associated only with a shift in the supply of counts due to
visibility changes is, as in the prior case of changes in costs of Y, the
area between the two cost curves. Fig.3-5's shaded area is the cost asso-
ciated with a shift of supply only. The complete cost is derived from a
shift in the supply and demand for counts--which means combining the shaded areas.
Using the theoretical model constructed in the previous section, a frame-
work for estimation can be developed. Consider the simple structural model below.
(3-22) CDU = c0 + ^ + a2 V.t +
(3-23) C®t - Y„ + y1 P®t + V2 V,. + 3,,
-------
FIGURE 3-5
Cost Associated with a Supply Shift Only
-------
FIGURE 3-6
Social Cost Associated with Demand and
Supply Shifts due to Visibility Variation
P (V.)
Z 1
-------
221
Eq. (3-22) is the demand curve for counts. Counts demanded are specified
as a function of landing fee and time costs C?^t), visibility (V^ ), and a
vector of other weather - related variables Q^t) at airport i for time t.
Counts supplied are also expected to be a different function of the same
variables. Some of these parameters can be signed a priori, ct^ is expected
to be negative since an increase in price decreases demand, expected to
be positive since visibility decreases lower counts demanded by increasing
time costs. y^ is the standard positive effect in supply of price increases.
Y^ is expected to be positive since decreases in visibility decreases the
amount of counts supplied.
The reduced form equation for counts is
(3-24) r =
it
arYi
a Y, B-
C— - + (— - —)v. + C
. va y1 ct y xt a
-6
1 a2 Y9
The reduced form parameter associated with visibility, (. 1 C ), is
arYi ai Yi
expected to be positive in sign, but the underlying structural parameters are
unidentified. By making some assumptions about relative magnitudes of and
a range of values for a2'Y2 can esta^lished for the cost-benefit analysis
discussed in the previous section.
Ta.3-17 presents the results from a regression of total daily traffic counts
at Aurora Airport on a vector of weather variables. Ta.3-18 defines each of
the regression variables. All continuous variables are in logarithm. One
drawback of the data is that weather conditions are available only for O'Hare
-------
222
TABLE 3-17
Classical Least Squares Regression Estimates
of Total Traffic Counts for Aurora Airport
DEPENDENT VARIABLE: LTOTO
SSE 374.402890 F RATION 2279.71
DFE 645. PROB > F 0.0001
MSE 0.580470 R SQUARE 0.9815
PARAMETER STANDARD
VARIABLE DF ESTIMATE ERROR T-RATIO PROB > T
LVIS
1
0.413987
0.077050
5.3730
0.0001
LCL
1
-0.104677
0.044098
-2.3737
0.0179
LWS
1
-0.282124
0.085868
-3.2856
0.0011
LWD
1
0.006086538
0.037512
0.1623
0.8712
RA
1
-0.00882506
0.001742717
-5.0640
0.0001
SN
1
-0.00699878
0.001800427
-3.8873
0.0001
FG
1
-0.014861
0.001654214
-8.4838
0.0001
LTEM
1
0.398944
0.050810
7.8517
0.0001
M
1
3.923506
0.570428
6.8782
0.0001
T
1
3.994875
0.560049
7.1331
0.0001
W
1
4.033440
0.566187
7.1239
0.0001
R
1
4.077325
0.559592
7.2862
0.0001
F
1
4.125296
0.571374
7.2200
0.0001
S
1
3.862951
0.571230
6.7625
0.0001
su
1
3.739265
0.568384
6.5788
0.0001
-------
223
TABLE 3-18
Regression Variable Definitions
LVIS Visibility at O'Hare International Airport (in Logarithms)
LCL Ceiling at O'Hare International Airport (in Logarithms)
LWS Wind Speed at O'Hare International Airport (in Logarithms)
LWD Wind Direction at O'Hare International Airport (in Logarithms)
RA Discrete Variable indicating presence of rain at O'Hare
SN Discrete Variable indicating presence of snow at O'Hare
FG Discrete Variable indicating presence of fog at O'Hare
LTEM Temperature in degrees Fahrenheit at O'Hare (in Logarithms)
M Monday dummy for day of week effects
T Tuesday dummy for day of week effects
w Wednesday dummy for day of week effects
R Thursday dummy for day of week effects
F Friday dummy for day of week effects
S Saturday dummy for day of week effects
SU Sunday dummy for day of week effects
-------
224
International Airport. Thus, to the extent that weather conditions vary across
airports, this analysis will be in error. However, all airports fall within a
20 mile radius of the Chicago Loop area, so major weather changes are unlikely.
Landing fees over the sample are also unavilable. The regression equation
estimate is
(3"25) Git " ao + aiL7Ist + a2LCLc + a3Lysc + a4UClt + "s^t +
a6SNt + + «3LIEHc + i 2t + sc ,
where ^ is a vector of day of week dummies and e is the white noise error
term. The high value of the F-statistic and R-squared in Table 3 inidcates
that the regression has high explanatory power over the sample. The visi-
bility parameter is positive, as expected and quite precisely estimated. All
parameters are of the expected sign except for that associated with LCL. The
negative value indicates that as the ceiling increases, traffic counts fall.
Wind direction effects are small and imprecisely estimated. However, it is
included in the regression to capture differential runway capacity effects at
multiple runway airports.
Ta.3-19 presents the estimates for DuPage County Airport. Again,
the visibility coefficient is positive in sign and precisely estimated. Its
value of .392 is quite close to the visibility coefficient at Aurora of .413.
The negative effect of ceiling height again occurs, and the effect of wind
direction is larger than at Aurora but is imprecisely estimated. The high
F-statistic and R-squared values again indicate a good fit.
-------
225
TABLE 3-19
Classical Least Squares Regression Estimates
of Total Traffic Counts at DuPage County Airport
DEPENDENT
VARIABLE:
LTOTO
SSE
90.172072
F PATIO
3270.19
DFE
319.
PROB > F
0.0001
MSE
0.282671
R-SQUARE
0.9935
PARAMETER
STANDARD
VARIABLE
DF
ESTIMATE
ERROR
T-RATIO
PROB > T
LVIS
1
0.391728
0.076608
5.1134
0.0001
LCL
1
-0.104518
0.043144
-2.4225
0.0160
LWS
1
-0.485604
0.084391
-5.7542
0.0001
LWD
1
-0.037855
0.036887
-1.0263
0.3055
RA
1
-0.00582789
0.001709277
-3.4096
0.0007
SN
1
-0.012183
0.001735787
-7.0189
0.0001
FG
1
-0.012260
0.001619163
-7.5715
0.0001
LTEM
1
0.299262
0.049938
5.9927
0.0001
M
1
6.328694
0.562298
11.2550
0.0001
T
1
6.443391
0.551889
11.6751
0.0001
W
1
6.393385
0.557940
11.4589
0.0001
R
1
6.498858
0.5500934
11.7961
0.0001
F
6.499807
0.562287
11.5596
0.001
S
1
6.615916
0.563341
11.7441
0.0001
SU
1
6.526664
0.560167
11.6513
0.001
-------
226
Ta.3-20 reports the regression coefficients for Chicago's Meigs Field.
The visibility effect is positive as before, but is smaller at .25 than the
other airports where it was around .4. Ceiling effects are still negative,
but wind direction effects, while small, are more precisely estimated
than at other airports. Again, all other signs are as expected.
This section has reported on the estimated effects of visibility for
three airports in the Chicago area. All of the regression equations have
2
very good explanatory power as indicated by their R and
F-statistic values. Visibility effects are strongly positive, and precisely
estimated at all sites. The next section attempts to bound the range of
supply and demand elasticities of visibility by referring to the structural
model presented at the beginning of the section.
As eq.3-24 showed, the parameter associated with visibility in
the reduced form regressions is an amalgam of prior elasticities and the
true underlying elasticities of visibility. This section examines the
values of these visibilities under several polar assumptions in order to
determine a reasonable range for the true visibility elasticities.
Ta.3-21 presents the values of ct^, the demand elasticity of visibility,
and ^2> the supply elasticity of visibility at the three airports under
alternative assumptions about the relative price elasticities. As Ta.3-21
shows, if the demand and supply curves are unitary price elastic or price
inelastic, then the visibility elasticities are on the order of .4 or below.
That is, a one percent decrease in visibility would yield at most a .4
-------
227
TABLE 3-2 0
Classical Least Squares Regression Results of
Total Traffic Counts for Meigs Field
DEPENDENT
VARIABLE: LTOTO
SSE
127.117252
F RATIO
1491.54
DFE
316.
Prob > F
0.0001
MSE
0.402270
R-SQUARE
0.9861
PARAMETER
STANDARD
T-RATIO
PROB > T
VARIABLE DF
ESTIMATE
ERROR
LVIS 1
0.250323
0.089207
2 8061
0.0053
LCL 1
-0.096790
0.051904
-1.8648
0.0631
LWS 1
-0.055751
0.100681
-0.5537
0.5801
LWD 1
0.063096
0.044101
1.4307
0.1535
RA 1
-0.00825438
0.002051089
-4.0244
0.0001
SN L
-0.00495015
0.002105944
-2.3506
0.0194
FG 1
-0.012995
0.00194284
-6.6889
0.0001
LTEM 1
0.273146
0.059633
4.5805
0.0001
M 1
3.716479
0.671756
5.5325
0.0001
T 1
3.866213
0.659868
5.8591
0.0001
W 1
3.885791
0.667383
5.8224
0.0001
R 1
3.835062
0.659811
5.8124
0.0001
F 1
3.930859
0.673699
5.8347
0.0001
S 1
3.274191
0.672222
4.8707
0.0001
SU 1
3.159501
0.669603
4.7185
0.0001
-------
228
TABLE 3-21
Sensitivity of Visibility Elasticity
Estimates to Price Elasticity Assumptions
PRICE ELASTICITY ASSUMPTIONS
a2 = Y2 Y2 = 2a2 a2 = Y2 Y2 = 2a2 Y2 = a2 Y2 = 2a2
a = -.1 = -.1 = -.1 a1 = -1 = -10 = -10
AIRPORT y1 = .2 yx = .2 r1 = ~-9 r1 = 1.9 yl = H H
AURORA ct2=.083 a2=.Q5 a2=»4 a2=t25 a2=45,5 ct2=29_, 39.
Y 2=.083 y 2= * Y2=' ^ Y2=' ^
DUPAGE a2=.08 a2=.05 a2=.35 a2=.24 37=43.12 ct9=28.3
Y2SS'08 y2='1 Y2=-35 Y2=-48 y2=43.12 y2=56-6
MEIGS ao=.05 a2=.03 a2=.23 a2=.15 a2=27.5 a?=18.0
Y2=.05 ^2=*06 Y2=,23 Y2='30 y^=21.5 y2=36.0
-------
229
percent decrease in traffic counts demanded or supplied. However, if price
elasticities are very large in absolute value, then the visibility elasticities
are also quite large. For the type of traffic at these airports, one would
expect to find a price elasticity which was quite small, thus implying small
visibility effects. However, notice that by eq. 3-24, whatever the price
elasticity is, given these results, visibility effects will be large in absolute value.
3.5.3 Visibility and Traffic Accidents
The automobile has become a way of life in industrialized societies, and
closely associated with this fact is the annual increase in reported highway
casualties in the major cities. The Department of Transportation (1981) reports
there were 45,212 fatal accidents and 51,083 fatalities due to roadway usage in
the U.S. in 1979. The number of motor vehicles involved was 64,754 and the
accident rate was 3.35 fatalities per 100 million vehicle miles. For Illinois
there were 2,017 fatalities and the accidnet rate was 3.2.
The number of accidents is affected by those factors which determine
travel demand and travel behavior as well as by driving conditions. Several
studies of traffic accidents exist which consider accidents to be the result
of the demand and supply of motor vehicle travel under various conditions.
Peltzman (1975) developed a model of driver behavior and analyzed fatal accident
rates to estimate the impact of national highway safety policy in the U.S. The
time series analysis of national data covered the period 1937-1972 and his cross-
section analysis of state data covered 1962, 1965, 1967, and 1970. He explicitly
recognized drivers' utility maximizing use of safety inputs including those supplied
exogenously. Peltzman incorporated into his study the earlier research by safety
scientists who focused almost exclusively on driving conditions for the effect of
-------
230
traffic density and the like. Ghosh, Lees, and Seal (1975) modeled drivers as
trading off safety and low fuel consumption rates for savings of time in choosing
their utility maximizing speed of travel. As part of their analysis they estimated
a production function for casualties on British motorways using monthly data for
the period Januaryl972 to March 1974. The evidence indicates that relevant
factors include driver characteristics and driving conditions including weather.
In addition to the research which centers on driver behavior, there is consi-
derable research on the contributions of vehicle and roadway design, and driving
conditions to traffic accidents. In Blomquist (1977), a search to identify factors
affecting seat-belt productivity found that vehicle speed, alcohol consumption,
week-end and night driving, small cars, and high-speed travel on non-interstate
highways each tend to increase the probability of a fatal accident.
Fatal Accident Reporting System 1979 gives facts and figures which quantify
the gross (as opposed to partial) effects of these and other factors on the number
of fatal accidnets. One of the relevant characteristics of the 1979 fatality pro-
file is that an overwhelming majority of fatalities occured during clear weather
conditions. According to the Department of Transportation (1981), only fourteen
percent of the fatalities were associated with inclement conditions. With rain-
slick or ice-slick roads being the worst weather conditions, one would not expect
atmospheric visibility to be dominant. However, it is identifiable and measurable.
Measuring the benefits of better visibility can be accomplished by: (1) esti-
mating the physical damage caused by poor visibility, and (2) placing a dollar value
on that damage. Our analyses showed that while improvements in visibility lead
to decreases in nonfatal accidents, it also resulted in an increase in the probability
of fatal accidents. It was also found that a unit improvement in visibility resulted
in cost saving of 9.45 million dollars (1980 prices).
-------
231
In this study we examine the effects of weather (rain, snow, ice, fog),
visual range (visibility) and the seasonal variables on highway accidents in
Cook and DuPage counties in the Chicago SMSA. The data utilized in the analysis
covered the period from January 1978 to June 1980 and the highway casualties
are classified into two categories: fatal and non-fatal accidents. First is
provided a theoretical examination of the effects of visibility on traffic
accidents based on the assumption that travel cost minimization is the main
driving force behind the choice of vehicles, speed, direction of travel or
route in making a trip between given destinations. It is shown that while the
partial effect of improvements in visibility on highway accidents is positive,
the total effect is ambiguous. Next are provided some econometric estimates
of the relationships between highway accidents - fatal as well as non-fatal -
and visibility, weather conditions and seasonal variables for Cook and DuPage
counties. it is important to note that only one dimension of benefits from
visibility improvements has been estimated--reduction in traffic accidents.
Other important benefits, such as increases in speed and volume of traffic have
not been addressed. Thus, the benefits estimated in this section represent a
lower bound of visibility improvement benefits.
In this section, we attempt to find out whether there is an unambiguous
relationship between improvements in visibility and accident rates, assuming
that cost minimization is the major driving force behind drivers' travel
decisions. Assume two urban communities of the same socio-economic charac-
teristics, highway design conditions and population size. At first thought,
most observers would agree that the community with very poor visibility
conditions will be less safe (in terms of highway accident reductions) com-
pared to the community with good visibility conditions, even thought poor
visibility might lead to a slow down of speed and a decrease in the volume
of traffic.
-------
232
Let us define an improvement in safety as a change in climatic conditions,
visibility, traffic volume, speed etc., which reduces the rate of traffic
accidents. In this respect, we are more concerned with traffic volume, speed,
environmental conditions and visibility, while holding vehicle designs, road
conditions (e.g., potholes), highway design and other engineering characteris-
tics of the highway constant. Economic efficiency requires that the cost of
achieving a given level of safety be minimized. Let us assume that the
consumer computes the price of travel as a solution to the problem of mini-
mizing the cost of travel to his or her destination where the cost of travel
is made up of vehicles operating cost and the cost of accidents (measured in
terms of what consumers will be willing to pay to avoid accidents). The
value of the motorists' time, although positive, is not explicitly included in
the model. Let us further assume that decisions concerning choice
of vehicle type and direction of travel have already been made by the motorist,
Then the most relevant variable under the control of the motorist is speed.
The motorist has no control over highway conditions such as traffic volume
and the behavior of other motorists as well as the weather and visibility,
but all these variables do affect his cost of travel. If we assume that the
safety of a trip depends on speed, weather conditions, visibility, traffic
volume for given highway design characteristics, mechanical conditions of
the automobile, age of driver, blood alcohol level etc., then the accident
rate AR = AR(VIS, RC, SP, TV, 0) , where
VIS = visibility (e.g., visual range in miles) ,
RC = road conditions e.g., inches of rain, snow, ice etc.,
SP = speed,
TV = traffic volume in vehicle miles per highway mile,
0 = other relevant variables.
-------
233
For simplicity, let us assume that travel cost
(3-26a) TC = AC(sp) AR(VIS, RC, SP, TV, 0) + OC(sp) ,
where AC(sp) = average cost per accident. It is assumed that accidents
which occur at higher speeds are more costly in terms of the damages done
to life and property than accidents which occur at lower speeds
OC(sp) represents the operating cost per mile. This may include the
value of the motorists' time. It is also assumed that, up to the relevant
speed limit, the marginal cost of a vehicle mile decreases as speed increases,
Without considering other environmental variables and visibility condi-
tions, the choice of speed to minimize travel cost, TC, requires that
Eq. (3-26c) requires the motorist to equate the marginal increase in
accident cost per mile (LHS) to the marginal savings in operating cost per
mile. For the extreme point to be a minimum, the second derivative of the
TC function, represented by Z, must be positive.
Our present task is to find the effect of improvement in visibility on
accident rates. To obtain the solution to this problem, we totally differen-
ce.
(3 - 2 6 c)
-------
234
tiate the accident rate AR with respect to visibility,
From eq.(3-26a) the total effect of improvement in visibility on
accident rates,
(3-26d) _ 3AR d(sp) , 3AR 3AR dTV
dCVIS) 3 Csp) dCVIS} 3 CVIS) 3 CCYl ' d(yiSl '
C+) C-l C+l
Let us assume that the partial effect of improvement in visibility on
accident rates, 3aR , is negative and 3AB. , which measure the partial
3(VIS) 3Csp)
effect of speed on accident rates, is positive. The third term, 3AR . dTV
3(fV). '
measures the effect of visibility on accident rates through its influence on
highway congestion, TV. The partial effect of highway congestion on AR, 3AR
3 cm'
is assumed to be positive i.e., more accidents occur on congested urban highways
than on rural highways. For simplicity, let us assume that the effect of visi-
bility on traffic volume is small and positive. The total effect of improve-
ment in visibility on accident rates then depends on dCsol i.e., the total
d CVIS).
effect of improvement in visibility on speed.
Totally differentiating eq.(3-26b) holding RC,TV, and 0 constant, we
obtain
(3 - 2 6e) »
d(VIS)
"AC32AR ^ 3AC 3AR
C+) 3 Csp)3 CVIS) ' 3Csp) 3 CVIS)
C-) M C-) J
„-l
X L
where Z represents the second derivative of cost per mile with respect to
speed. This is positive.
The average cost of an accident, AC, is positive and 3-aR which
3 Csp 13 CIS 1'
measures the effect of an improvement in visibility on the rate at which accident
-------
235
rates change with respect to speed, is assumed to be negative, i.e., accidents
are more likely to increase less, for given speeds, following improvements in
visibility. Since accident costs are more likely to increase with speed, 3AC
3(sp)
is positive, which makes the bracketed term in eq.(3-26d) negative. Thus d(sp)
-devis)
is positive i.e. improvements in visibility encourage higher speed levels,
Substituting d(sp) > q into eq. (3-26d) the sign of dAR , the total
dCVIS) dCVIS)
effect of an improvement in visibility on accident rates, becomes ambiguous.
3.5.4 Analysis of Highway Casualties in DuPage and Cook Counties
3.5.4.1 Empirical Analysis
Data on the number of fatal and non-fatal accidents have been collected
for Cook and DuPage counties from January 1978 to June 1980 on daily basis.
Visibility data, measured in terms of miles of visual range, have also been
assembled from the O'Hare airport. In addition to the above information,
weather data have also been collected from the O'Hare weather station on the
occurence of snow, fog and rain as well as daily recording of the dry bulb
temperature in degrees F. The data do not include information on traffic
volume and speed in these two counties. Given the quality of data available,
the best one can do is to attempt to estimate an econometric relationship be-
tween traffic accidents and visibility, weather and the day or season in which
the accident occured. These relationships were estimated for DuPage and Cook
counties for non-fatal and fatal accidents separately. The following general
equation was estimated separately for both counties:
(3-27a) Zc = aQ + + cu WrCRt + + a4S?Rfc + a5VISt + c^VIS^
+ a_DV13 + agVWTR + 3gVSPRt + a^VSL^ + + a12SNt + ai3rG:
+ a14VTZMt + a15VSAt + V t-1,2 912
-------
236
Variables definitions are as follows:
Z = Number of non-fatal accidents per day in DuPage county (DPNONFATI
or Number of non-fatal accidents per day in Cook county (£KNQNTATL,
DD equals 1 if the accident occured on weekends and equals 0 otherwise,
WNTR equals 1 in winter time and 0 otherwise,
SUMR equals 1 in spring and 0 otherwise,
VIS represents visibility measured in miles,
DVD represents the interaction between visibility and day of occurence
of the accident, while VWTR, VSPR AND VSUM measure the interactions between
visibility and the seasons (winter, spring and summer). RA equals 1 if there was
an occurence of any of the following phenomena on the day the accident occured -
rain, rain showers, freezing rain, rain squals, drizzle or freezing drizzle, and
0 otherwise. SN is a 1/0 dummy variable indicating the occurence/non-occurence
of any of the following phenomena on the day the accident occured - snow, snow
pellets, ice crystals, snow showers etc. FG is also a 1/0 dummy variable in-
dicating the occurence/non-occurence of either fog, ice fog, ground fog, etc.
TEM represents temperature in degrees F., while VTEM,VRA, VSN measure the effects
of the interaction between temperature, rain and snow, respectively, on traffic
accidents.
Ta.3-22 presents the results of a linear regression model for non-fatal
accidents in DuPage county. The low R obtained can be partly attributable to
the absence of such variables as speed and traffic volume from the model. The
parameter estimates indicate that the number of non-fatal accidents increases by
almost 8 units per day on weekends compared to weekdays. The coefficient for
-------
237
TABLE 3-22
DuPage County Non-Fatal Accidents Regression Results
Dependent Variable: DPNONFAT
VARIABLE
PARAMETER
T RATIO
ESTIMATE
Intercept
69.088
8.065
DD
7.844
3.159
WNTR
15.187
3.154
SUMR
7.069
1.343
SPR
15.137
3.254
VIS
-3.445
-3.250
VIS
0.046
1.265
DVD
-0.064
-0.293
VWTR
0.907
2.123
VSPR
0.791
2.001
VSUM
0.424
0.955
RA
7.463
2.406
SN
13.451
3.621
FG
0.140
0.086
VTEM
0.022
2.133
VRA
0.086
0.242
VSN
-1.273
-2.86
TEM
-0.405
-3.49
PR > F = 0.0001
R2 = 0.323
DW =1.46
-------
238
visibility shows that an improvement in visibility by one mile decreases the
number of non-fatal accidents by 3.4 per day. This result is consistent
with a priori expectations concerning the partial effects of an improvement
in visibility on highway casualties. The results also show that seasonal co-
efficients for winter and spring are precisely estimated. The number of non-
fatal accidents increases by 1.5 units per day in winter and spring compared to
the base season (fall) . But summer shows an increase of only 7 per day
above the base season. The summer coefficient is, however, imprecisely estimated.
The interactions between visibility improvement and the seasons show that a unit
increase in visibility increases the number of non-fatal accidents by almost one
unit per day each in winter and spring, while the coefficient of the interaction
between visibility and SUMR is imprecisely estimated.
The sign of the coefficients for the weather variables are consistent with
a. priori expectations. The occurence of rain increases the number of non-fatal
accidents by 7.5 per day while the presence of snow increases the number
of non-fatal accidents by 13.5. Thus, the number of non-fatal accidents
which occur in the presence of snow can be expected to exceed the non-fatal
accident which occur in the rainy season. The coefficient for fog is, however,
imprecisely estimated. An increase in temperature by 10 degrees F., decreases
the number of non-fatalities in DuPage county by 4 per day. This is probably
due to the fact that people are more likely to engage themselves in other outdoor
activities when the temperature increases.
The interactions between visibility improvements and the weather variables
for DuPage county indicate that, although the number of non-fatal accidents in-
creases by 13.5 per day in the presence of snow, a unit improvement in
visibility In the presence of snow decreases the number of non-fatal accidents
-------
239
by 1.3 per day. An improvement in visibility by one unit on a snowy
weekend at an average winter temperature of 30°F can be computed for DuPage
county by evaluating the following expression:
Eq. (3-27b) is obtained by taking the first derivative of the equation
presented in Ta.3-22 with respect to visibility. Evaluating the expression
obtained at SN=1, DD=1, WNTR=1, VIS = average visibility = 10.3 miles, TEM =
average winter temperature = 30°F provides the required result, Ta.3-23 presents
the average values of some of the variables used in the analysis. Substituting
these values into eq.3-27b it is realized that a unit improvement in
visibility on a snowy weekend leads to a decrease in the number of non-fatal
accidents by 2.28 per day in DuPage county. The effect of an improvement
in visibility on the number of non-fatal accidents occuring on a rainy day
can also be obtained by evaluating the following expression at the average
values of the variables:
Inserting the relevant average values of the variables into eq. (3-27c)
shows that on a rainy weekend, a unit improvement in visibility leads
to a decrease in the number of non-fatal accidents by 1.35 per day,
compared to a decrease of 1.28 on a rainy weekday.
(3-27b) HDPNONFAT)
3(VIS)
= -3.455 + 2xQ.Q46VI£ - Q.0&4DD r .WEilR
+ Q.022TEM - 1.273SN
(3 - 27c)
3(DPNONFAT) _ -3.445 + 2^0.Q46VIS -0.064DD + CL022TEM
3(VIS) " +0.QS6RA
-------
240
TABLE 3-23
Statistics on Some Variables
Included in the Regression Analysis
VARIABLE*
NUMBER OF
MEAN
MINIMUM
MAXIMUM
RANGE
OBSERVATIONS
VALUE
VALUE
DPNONFAT
1035
28.98341
5.00000
118.00000
113.00000
CKNONFAT
1035
194.29372
72.00000
729.00000
657.00000
CKFATAL
1035
0.41836
0.00000
1.00000
1.00000
DPFATAL
1035
0.10725
0.00000
1.00000
1.00000
SN
912
0.11952
0.00000
1.00000
1.00000
TEM
912
51.27412
-8.33333
89.33333
97.66667
VLS
912
10.31060
0.31250
16.66667
16.35417
VARIABLE DEFINITIONS:
DPNONFAT = Number of non-fatal accidents in DuPage County
CKNONFAT = Number of non-fatal accidents in Cook County
CKFATAL = Number of fatal accidents in Cook County
DPFATAL = Number of fatal accidents in DuPage County
SN = Snow (dummy variable)
TEM = Temperature (°F)
VIS = Visibility in miles
-------
241
Ta.3-24 presents the non-fatal accidents regression results for Cook
Country. By comparison with Ta.3-22, almost all the coefficients have the
same signs as obtained from the DuPage County regression results, except the
FG coefficient. In Cook County, the presence of fog decreases the number of
non-fatal accidents by 10.9 while it virtually has no effect in DuPage County.
The magnitudes of the effects the explanatory variables in the Cook County
regression results exceed those obtained for DuPage County.
In Cook County the number of non-fatal accidents increases by 48 at
weekends compared to weekdays. All the seasonal coefficients are precisely
estimated except the coefficient for summer. The results show that the number
of non-fatal accidents increases by 60 per day in winter compared to fall.
During the spring season, non-fatal accidents increase by 56.72 per day compared
to fall base season. As in DuPage County, a one mile improvement in visibility
in Cook County leads to a reduction in the number of non-fatal accidents but
the decrease is almost by 16 per day compared to 3 per day for DuPage County.
This effect does not include the interaction terms of visibility and the other
variables. The coefficients of the weather variables also show that the number
of non-fatal accidents increases by 46.7 per day in the presence of rain while
the effect of an occurence of snow increases the number of non-fatal accidents
by 63 per day in Cook County.
Considering the interaction terms between visibility and the other explana-
tory variables, an improvement in visibility by one mile on a snowy weekend or
weekday at an average winter temperature of about 30°F can be computed by
evaluating the following expression:
-------
242
TABLE 3-24
Cook County Non-Fatal Accidents Regression Results
VARIABLE
Dependent Variable:
CKNONFAT
PARAMETER
ESTIMATE
T RATIO
Intercept
387.
55
9.
47
DD
48.
.27
4.
. 18
WNTR
60.
.37
2 .
,48
SUMR
22 .
.77
0.
87
SPR
56.
.72
2.
.44
VIS.
-15.
.63
-3 .
25
VIS2
0.
.026
0.
,16
DVD
-0.
.72
-0
.71
VWTR
4
.82
2.
.36
VSPR
2
.96
1.
.57
VSUM
2
.17
1.
.02
RA
46
.73
3.
.33
SN
63.
.15
3.
.84
FG
-10.
.88
-1.
.15
VTEM
0
.148
3.
. 06
VRA
-0
.027
-0.
. 02
VSN
-4
.11
-2 .
.07
TEM
-2
.35
-4.
.17
PR > F = 0.0001
R2 = 0.35
DW =1.39
-------
243
(3-2 7d)
3CCKN0NPAT1 -15.63_+_2 CO • Q261VIS -Q..72DD + 4,82WSTR .
3 CVISl ~ +0.148TEM -4.11SN
Eq. (3-27d) is obtained by taking the first derivative of the re-
gression equation presented in Ta.3-24 with respect to visibility. An
evaluation of eq.(3—27d) at the mean values of the relevant variables
and an average winter temperature of 3 0 0 F shows that an improvement in
visibility by one mile on a snowy weekend leads to a decrease in the
number of non-fatal accidents by 10.7 per day. it is observed from
Ta.3-24 that the effect of an improvement in visibility alone, without
considering the interaction terms, is to decrease the number of non-fatal
accidents by about 15 per day. But when the interaction terms are
considered, the effect of the interaction between an improvement in visi-
bility and winter season is to increase the number of non-fatal accidents
in Cook County by 4.82 per day.
The effect of an improvement in visibility on the number of non-fatal
accidents occuring on a rainy day can be computed by evaluating the follow-
ing expression at the average values of the relevant variables:
Inserting the relevant average values of the variables into eq.(3—27e)
shows that on a rainy weekend, an improvement in visibility by one
mile leads to a decrease in the number of non-fatal accidents by 8.3
per day.
(3-2 7e)
3(CKNONFAT) _ -15.63 + 2CO.026)VIS -0.72DD + Q.148TEM
3(VIS)
-0.027RA
-------
244
3.5.4.2 Linear Probability Models of Traffic Fatalities
The average number of non-fatal accidents reported for DuPage County
during the period for which the accident data were collected was 28.98.
while the average for Cook County was 194.3 non-fatal accidents per day.
Very few fatalities were recorded. In fact an average of 0.42 fatalities
Per day was recorded for Cook County compared to an average of 0.11 fatali-
ties per day for DuPage County. This means that most of the elements under
the dependent variable column in the regression model are zeroes and ones.
Very few fatal accidents greater than one were recorded for both counties.
Therefore, it was decided to use a qualitative choice model in which
the dependent variable is 0 when the accident is non-fatal and 1 when the
accident was fatal.
The simplest specification of a qualitative choice model is the linear
probability model, where it is assumed for the purpose of this analysis
that the probability of occurence or non-occurence of a fatal accident on
any given day is a linear function of the explanatory variables listed in
Ta.3-22 and 3-24.
Let FATAL = a + a,DDfc + a.WNTR + a.SUMR. + a,S?R + a-VIS
+ cigDVD + a7VWNTRt + a8VS?Xt + agVSUMt +
+ + ai5VSH
I, if fatal accident
vas recorded,
0, otherwise.
-------
245
f 14 if fatal accident
For Cook County, FATAL= CKFATAL= ) was recorded
V.0, otherwise
Thus, the regression coefficients may be interpreted as the effects
of unit changes in the explanatory variables on the probability of occurence
of fatal accidents. The above model was estimated by Ordinary Least-Squares
procedure for DuPage and Cook Counties and the results are presented in
Ta.3-25. The very low R suggests that a good deal of variance in the model
is unexplained. Nonetheless, it is our belief that, with the availability of
data on relevant variables such as vehicle speed and traffic volume, there would
be an improvement in the fit of the Linear Probability Model.
The results show that an improvement in visibility by one mile leads to an
increase in the probability of fatalities by 0.005 in DuPage County, compared
to an increase of 0.02 in Cook County. This result does not include the inter-
actions between visibility and the other explanatory variables. If we consider
the interaction between visibility and the day of week effect (DVD), an improve-
ment in visibility leads to an increase in the probability of fatalities by
0.009 in Cook County and a decrease in the probability of fatalities by 0.014
in DuPage County during the weekends. The DuPage County estimate of the inter-
action between visibility and the day of week effect is, however, more precisely
estimated than the Cook County estimate. The effect of the interaction between
visibility and the seasons is to decrease the probability of occurence of
fatalities in winter and spring in Cook County by 0.022 and 0.020 respec-
tively. An improvement in visibility in summer time leads to an increase
in the probability of occurence of fatal accidents by 0.003 in Cook County.
-------
246
TABLE 3-25
Linear Probability Models of Traffic
Fatalities in Cook and DuPage Counties
Cook County Results DuPage County Results
VARIABLE PARAMETER T RATIO PARAMETER T RATIO
ESTIMATE ESTIMATE
Intercept
-0.059
-0.215
0.095
0.545
DD
0.026
0.289
0.137
2.372
WNTR
0.258
1.473
-0.037
-0.334
SUMR
-0.062
-0.319
0.000
0.002
SPR
0.180
1.041
0.049
0.447
VIS
0.023
0.979
0.005
0.318
DVD
0.009
1.080
-0.014
-2.688
VWTR
-0.022
-1.417
-0.002
-0.166
VSPR
-0.020
-1.353
-0.007
-0.764
VSUM
0.003
0.181
-0.002
-0.176
RA
0.008
0.075
-0.001
-0.016
SN
0.037
0.289
0.026
-0.331
FG
-0.047
-0.801
0.0363
0.977
VTEM
-0.0002
-0.659
0.000
0.112
VRA
1
O
o
-0.147
-0.007
-0.865
VSN
0.004
0.250
0.006
0.593
TEM
0.006
1.534
-0.0004
-0.147
PR > F = 0.0059
R2 = 9.0367
DW = 1.932
PR > F = 0.5997
R2 = 0.0154
DW = 2.098
-------
247
The coefficients of the interaction terms between visibility and winter, and
spring (VSPR) are more precisely estimated than the summer interaction term
in the Cook County model. The DuPage County results show that the effect
of interactions between visibility and the seasons is to decrease the pro-
bability of occurrence of fatal accidents, but these coefficients are impre-
cisely estimated.
3.5.4.3 Monetary Value of Benefits
The results of the Cook County linear probability model parameter estimates
for the occurrence of fatal accidents shows that an improvement in visibility by
one mile increased the probability of occurrence of daily accidents by 0.023.
The daily fatal accidents rate for Cook County is 0.42. Thus the expected number
of fatal accidents occurring in Cook County per day due to a mile improvement
is 0.01. This represents 3.65 traffic fatalities per annum. The loss in human
lives represents a cost to society, largely resulting from risks voluntarily
incurred. This cost partly offsets the gains obtained by the great majority of
motorists because of time saved. Ignoring the net affects of traffic fatalities
contributes to a conservative estimate of the benefits of improved visibility.
Professor Sherwin Rosen's risk-compensating wage differential estimates (1976)
produce an average statistical value of life of 494,000 dollars (1980). The
3.65 traffic fatalities which occur due to an improvement in visibility by one
mile in Cook County represents a cost of 1.80 million dollars (1980) in human
life. A simple linear extrapolation of this value to cover the entire eastern
United States yields a benefit of 204 million (1980) dollars.
In valuing the reduction in nonfatal accidents we make use of the nonfatal
injury costs estimated by Faigan (1975) and the Proceedings. Ta. 3-26 presents
the breakdown of the injury costs in 1972 dollars. The average nonfatal injury
loss which can be aboided is $3000 per accident in 1972 dollars. Using the
-------
248
estimate of the annual reduction in traffic accidents due to a one mile improve-
ment in visibility, a rough estimate of the annual benefits from a one mile
improvement in visibility is 17 million dollars in Cook County. This translates
into 35 million 1980 dollars, using the 1980 consumer index. A simple linear
extrapolation to the entire U.S. yields an annual benefit of about $750 million
(1980) .
TABLE 3-2 6
Non-Fatal Injury Accident Costs*
TYPE OF COST
Labor Productivity Low
Medical
Pain and Suffering
Property Damage
Legal
InsuranceAdministration
Other
Total
COST IN 1972 DOLLARS
850
350
100
700
150
800
50
3000
*Source: G. Blomquist "Value of Life: Implications of Automobile
Seat Belt Use" p. 47
-------
249
3.5.5 Summary and Conclusions
A conceptual model of the relationship between travel cost, accident
rates, weather conditions, improvement in visibility, vehicle speed, and
traffic congestion has been developed. Based on the assumption that travel
cost minimization is the main driving force behind drivers' choice of vehicle
speed and direction of travel when vehicle and highway designs, road condi-
tions and other engineering characteristics of highways are held constant,
it is shown that the total effect of an improvement in visibility on acci-
dent rates depends crucially on the effect of improvements in visibility on
vehicle speed. It has been demonstrated that improvements in visibility
encourage higher speed levels, for a given traffic volume and road condition,
thus leading to the conclusion that the total effect of improvements in visi-
bility on traffic casualties is ambiguous.
The empirical estimations of the relationship between improvements in
visibility, weather variables and traffic casualties show that visibility
improvements lead to significant reductions in non-fatal accidents in both
Cook and DuPage Counties. This result is consistent with the partial effect
of improvements in visibility on highway casualties. While the occurence
of rain and/or snow lead to an increase in the number of non-fatal accidents
in Cook and DuPage Counties, the empirical results also show that an improve-
ment in visibility in the presence of snow leads to a decrease in the number
of non-fatal accidents in both counties. Empirical estimates of benefits
from increased speed and traffic volume have not been made.
Results of linear probability models in analyzing the traffic
fatalities show that an improvement in visibility during the weekends leads
-------
250
to an increase in the probability of occurrence of fatal accidents in Cook
and DuPage Counties. Visibility improvements in winter and spring, however,
lead to decreases in the probability of occurrence of fatal accidents in both
counties, although these coefficients are not very precisely estimated. An
improvement in visibility in Cook County by one mile leads to an estimated
benefit of 35 million dollars as a result of reductions in traffic casualties.
This translates into an annual benefit of about $750 million for the entire
eastern U.S.
3.6 Effects of a One Mile Change in Visibility: Comparisons of Willingness to
Pay and Secondary Data Results
Estimated willingness to pay for a uniform one mile visibility improvement
in the eastern U.S. is given in Ta.3-27. The one mile improvement scenario is
suitable for comparison with benefits derived from analyses of secondary data.
Scenario benefits in Ta.3-27 are derived from the six-city eastern survey, using
the visibility value function from section 2 aggregates according to the method
explained in section 4. Aggregate 1990 benefits are about $10 billion for the
hypothetical argument on visibility of one mile. It should be emphasized that
tje one mile improvement does not refer to any real program and is used here only
for purposes of comparing the contingent valuation and secondary ratio
estimates.
Reduction of nonfatal traffic accidents is responsible for the largest
visibility improvement benefit among the Project's secondary data analyses.
Based upon the Cook County, Illinois results, eastern U.S. benefits from a
one mile uniform visibility improvement would be about 0.75 billion in 1980
dollars. The $10 billion aggregate benefit reported in Ta.3-27 comprises all
visibility benefits, whether they be aesthetic, safety-related or derived from
a multitude of other goods to which visibility contributes.
-------
251
TABLE 3-27
BENEFITS OF ONE MILE VISIBILITY IMPROVEMENT
IN THE EASTERN U.S. 1990 (1983 dollars)
Benefits
Total Benefits
per household
($000)
Alabama
167
233666
Connecticut
144
182760
Delaware
141
34578
District of Columbia
209
60670
Florida
116
514983
Georgia
179
380602
Illinois
206
902688
Indiana
220
464536
Kentucky
199
269036
Maine
117
51153
Maryland
230
413287
Massachusetts
149
339302
Michigan
194
706202
Mississippi
144
124967
New Hampshire
160
58592
New Jersey
157
465041
New York
163
1120832
North Carolina
171
390607
Ohio
201
848300
Pennsylvania
179
799842
Rhode Island
111
42780
South Carolina
193
220656
Tennessee
194
333294
Vermont
154
31456
Virginia
233
495 369
West Virginia
198
132774
Wisconsin
169
314799
TOTAL 9,932,774
Note: A detailed discussion of visibility scenarios is
in section 4.
-------
252
Two conclusions are suggested by this comparison. The first is that
improved traffic safety is one of the major benefits of visibility improvement--
about 7% of the total. A plausible conjecture is that there are several such
major areas of bnefit, plus a great number of areas where much smaller benefits
are derived. One such example is the benefit to spectators of major league
baseball in the entire U. Ssomewhat less than $1 million annually resulting
from the hypothetical one mile improvement, or less than one ten-thousandth
of the total. This is not a big part of the overall picture, but it undoubtedly
has importance to some people. (See section 3.2.3.)
The second and more important conclusion is that the secondary-data and
willingness-to-pay results appear to be consistent. While we cannot be certain
that a far more exhaustive secondary-data study would confirm the survey results
by adding up to the same total, nevertheless these results are plausibly related
to each other. Thus the evidence from the two approaches gives reason to have
confidence in both as a means of valuing this elusive non-market good.
Section 3 contains controlled experiments that directly compared secondary-
data and contingent valuation results in well defined situations. These results
corroborate our conclusions about the one mile improvement experiments. In section
3.4, a contingent market in visibility for view-oriented residences among high-
rise residents along Lake Michigan in Chicago was established. A hedonic
demand analysis was carried out for the same group of subjects. The similarity
of results confirmed the reliability of each approach for policy analysis. A
similar study of demand-based and contingent valuation in section 3.3.2 of
Hancock Tower visitation rejected the hypothesis that different results are
obtained from the two analytic approaches.
In future work, the findings of significant effects of visibility on the
other activities that have been considered in this section (section 3)--namely,
-------
253
air traffic and recreation in addition to baseball attendance--could be used
to develop benefit estimates to compare with the contingent valuation
estimates.
-------
SECTION 4
Use of Results to Estimate Benefits
for the Eastern United States
-------
255
4.1 EVALUATION OF POLICY EFFECTS ON VISUAL RANGE
This chapter provides a detailed illustration of the application of
the visibility value function developed in Section 2 to analysis of policy
benefits. The visibility value function indicates how people's expressed
willingness to pay to enjoy visibility improvements or to prevent visibility
deterioration depends on their personal characteristics and on prevailing
visibility conditions where they live. This function is general in that it
can be used to estimate visibiltiy benefits associated with any amount of
pollution reduction. The benfits are obtained by summing over affected areas
taking account of willingness to pay for the change in visibility that will be
brought about in each area by the pollution policy.
Forecasting visibility policy effects requires comparing a without-
policy or base-case scenario with one or more scenarios of regulatory stringency.
In this chapter, the visibility value function is applied to four policy
hypothetical or illustrative policy scenario for electric and utility pollution control
relative to a base-case scenario. Benefits connected with these illustrative
scenarios are estimated for the year 1990. Specifically, per-household and
aggregate benefits are estimated for each eastern state and the eastern
United States.
A method is needed which relates reductions in pollution emissions from
the scenarios to visibility improvements. In the present chapter, the relation
between emissions and visibility is provided by results from research at
Argonne National Laboratory. The major task of the chapter is to estimate
visibility benefits using the visibility value function.
-------
256
4.2 ILLUSTRATION OF METHOD
4.2.1 Outline and Summary
Step A in the analysis of visibility regulation was to establish policy
alternatives. Alternative policies produce different patterns of visibility
improvements whose effects need to be evaluated in order to make a policy
choice. Four such policies were considered. In addition to the policy scenarios
a without-policy or base-case scenario was formulated. The base-case scenario
is a judgement as to the most likely regulatory climate in the absence of a
visibility policy. It provides the standard against which the benefits of the
policy scenarios are measured.
Step B was to forecast emissions under the base-case and hypothetical-
policy scenarios by type of emitter, season and amount of pollution. These
forecasts depended in part on the technical requirements of pollution abatement.
To an even greater extent the emissions forecasts depended upon forecasts of
future levels of economic activity.
Step C was to forecast the spatial distribution of ambient air quality.
The relationship between emissions and ambient air quality depends upon the way
emissions are dispersed geographically and the chemical transformations that
occur during dispersion. This step was performed for each of the scenarios
by means of the Argonne long-range-transport model. [Rote, 1982]
Step D was to measure the effects of ambient air quality on visibility resulting
from each hypothetical scenario. The solution to this problem, also supplied
by Argonne [Rote, 1982b] , provides a set of predictions as to the course of
visual air quality on a state by state basis in the future.
Step E was to use the visibility value function to establish values
associated with alternative pollution control strategies. Each hypothetical
-------
257
scenario produced a set of improvements in visual range for each state in future
years. The function estimated the value of these improvements to a state as
the sum of the value of the local component and value of improvements in other
parts of the region due to existence and option values. Non-local improvements
are less valuable to the state depending upon their distance from the state. The
value of visibility improvements is the sum of all local and non-local improvements
for all states in a given year. The visibility value function is used to evaluate
improvements for each state in 1990 for each of the four hypothetical policy
scenarios.
4.2.2 Step A: Establish Hypothetical Policy Scenarios and Estimate Visibility
Effects
In this step, a base case and four illustrative policy scenarios are consi-
dered. [Rote, 1982b] The base case the three hypothetical policies that
yield improvements are summarized in Ta.4-1. They are as follows:
4.2.2.1 Base Case: Scenario 2
This scenario assumes that all electric utilities governed by State Imple-
mentation Plans (SIP) meet promulgated regulations by 1985. Compliance is
determined by comparing annual emissions with specified SIP regulations.
For industrial emitters that burn coal, the base-case scenario assumes that
large units burn low sulfur coal, and medium and small units comply with SIP
regulations. For oil-fired industrial emitters, the base case assumes that large
units burn medium- or low-sulfur coal, and small units comply with SIP regulations.
These industrial assumptions are maintained for all of the scenarios. All other
emitters are assumed to continue emitting at the 1979 rate in the base-case
scenario. This assumption about other emitters is also used in each of the other
scenarios.
-------
258
This scenario is crucial to policy analysis because it measures without-
policy or base-case conditions against which policy effects are measured. It
provides the basis for an estimate of future pollution by type of emitter in
the absence of the policy being evaluated.
4.2.2.2 Hypothetical Control Scenarios
The state of completion of the Argonne study necessitated limiting the
analysis to illustrative policies in which utilities are controlled more
stringently than in the base case, but emissions for other sources remain as
in the base case. No implication is intended that this combination of controls
would be chosen.
The scenarios are numbered according to increasing stringency of control.
Remembering that Scenario 2 is the base case, and shows some improvement over
1979, the control scenarios are as follows:
TABLE 4-1
Scenario 1 (1979 status quo).
All utility units continue to emit SC^ at the 1979 rate. Units with
operating scrubbers keep them; units with planned scrubbers install
them.
Scenario 3 (First level of increased stringency for utilities) .
All unility units covered by SIP regulations are required to meet
promulgated regulations by 1985. No such unit is allowed to exceed
4 pounds SO^ emissions per millions BTU's from fuel used to produce
electricity.
Scenario 4 (Second level of increased stringency for utilities) .
All utility units covered by SIP regulations are required to meet
promulgated regulations by 1985. No such unit is allowed to exceed
2 pounds SO^ emissions per million BTU's from fuel used to produce
electricity.
Scenario 5 (Third level of increased stringency for utilities) .
All utility units covered by SIP regulations are required to achieve
a 50 Percent reduction in SC^ emissions beyond SIP compliance levels
by flue gas desulfurization retrofitting where retrofitting is most
cost effective.
-------
259
4.2.3 Step B: Forecast Emissions Under the Hypothetical Policy Scenarios
Sulfur dioxide is the emitted pollutant of central importance to the
analysis because it is a precursor of ambient air constituents that cause
the greatest extinction of visual range. Argonne obtained the scenarios
underlying forecasts of future emissions from electric utilities from
Technekron, Inc., and those underlying the industrial emissions forecasts
from ICF, Inc.
Emissions estimates are made for the base-case and the four hypothetical-policy
scenarios to the year 2000. The model requires that the conditions under
which emissions take place be specified in detail. These conditions include
type of emitter (utility, industrial, other), stack height (short, medium,
tall), season (summer, winter), and fuel type (coal and oil of various
grades). The symbol specifying the amount of emissions from a
type under a given control scenario is where
Q is emissions of S02 in kilotons per year;
m is the scenario (m = 1, . . ., 5 as described under Step A;
j is the state from which emissions originate. All emissions are
aggregated and assumed to originate from the geographic center of the state;
k stands for the other conditions under which emissions occur: type
of emitter, stack height, season, fuel type, k = 1, ..., n for each of
these conditions;
t is the year, t = 1980, ..., 2000. Hereafter, t will be understood
to be present but not written down.
4.2.4 Step C: Forecast Spatial Distribution of Ambient Air Quality
Forecasting pollution is a regional problem because there are many source
regions, defined as states, and many receptor states. Each state is both a
source and a receptor, and the source-receptor relationship is a complicated
one. The Argonne long-range-transport model accounts for the processes by
which pollutant emissions are transported and transformed into ambient pollution
within a regional framework [Rote, 1982a] . All of the states in the
present project study area are represented (eastern United States).
-------
260
Based upon the pollution emissions variable, » an equation can be
written down which expresses the key relationships of the ambient air forecast:
(4-1) X™ = tiE < * ' where
is ambient pollution in state i under scenario m , measured
, 3
m ug/m of SC>4 ;
e. . is the amount of emissions from state i reaching state i ,
i]
per kiloton of emissions in state j;
t, is the amount of ambient pollution in state i resulting from
a kiloton of emissions of SC^ arriving in the state.
cm-j
Eq. (4-1) may be explained as follows. To solve for , first
Cm)
sum emissions Q.^ , over the k source types in state j , where Qjj.
k (m)
is obtained from Step A. Multiply the resulting £Qjk emissions by
e , , to obtain emissions from state j arriving in state i . Sum over
i]
all states j to obtain total emissions arriving in state i , and multiply
by t, to obtain the state's ambient pollution.
In the Argonne model, air-quality variables estimated on a state-by-state
basis are as follows:
Model-predicted sulfate ion concentrations;
Estimated sulfate ion concentrations computed by adjusting the
model-predicted values with regression parameters;
Fine particle (FP) concentrations computed from sulfate ion
concentrations estimated with regression equations;
FP concentrations computed from an alternative theoretical/
empirical relationship between FP mass and other constituents;
-------
261
Controllable sulfate mass concentrations computed from a theoretical
relationship between sulfate ions and other FP constituents;
Estimated first and second 24-hour maximum FP mass concentrations;
Model-predicted sulfate ion wet and dry deposition rates [Rote, 1982a] .
Several qualifications are noted in the Argonne report which affect the
applicability of the results discussed in this chapter. First, emissions
from each source state are assumed to emanate from a single point at the
geographic center of the state. Second, modeling results need more comparisons
with actual visibility measurements. Available comparisons show a good
correspondence; however, adjustments have been made to model-generated
visibility endowments in estimating benefits in the Report. Third, the
Argonne Report questions the validity of the base-case industrial scenario
as representative of likely economic trends between 1980 and the year 2000.
4.2.5 Step D: Estimate Visibility Effects of Scenarios
Predictions of visibility levels for 1990 for the base case and policy
scenarios are given in Ta.4-2 for each state considered in this study.
Estimates of actual visibility in 1980 are also given.
The analysis of visibility effects may be represented by the following
equation,representing the approach used in the Argonne study:
(4-2)
.where
AV^11^ is the improvement in visual range in miles in the
i state caused by policy scenario m. It is computed from
a theoretical-empirical relationship involving sulfate ion con-
centration and other factors in / defined below;
-------
TABLE 4-2
Visibility Projections in Miles
for Base Case and Three Control Scenarios, 1990
STATE
1980
Base Case
Scenario 2
Scenario 3
Scenario 4
Scenario 5
Actual
Visibility
SIP Compliance
by 1985
SIP SO2 Emission
Limits 41bs. per
million BTU
SIP SO2 Emission
Limits 2lbs. per
million BTU
S0„ Emissions
504 below SIP
Compliance Levels
Alabama 14.3 13.7
Connecticut 9.9 9.9
Delaware 10.6 9.9
D.C. 10.6 10.6
Florida 14.9 14.3
Georgia 13.7 13.0
Illinois 13.0 13.0
Indiana 9.9 10.6
Kentucky 10.6 11.8
Maine 13.7 13.7
Maryland 10.6 9.9
Massachusetts 10.6 9.9
Michigan 13.0 13.0
Mississippi 15.5 14.3
New Hampshire 11.8 11.8
New Jersey 10.6 9.9
New York 10.6 10.6
North Carolina 13.0 12.4
Ohio 8.7 9.3
Pennsylvania 8.7 8.7
Rhode Island 10.6 9.9
South Carolina 13.7 13.0
Tennessee 11.8 11.8
Vermont 11.8 11.8
Virginia 10.6 10.6
West Virginia 9.9 9.9
Wisconsin 14.9 14.3
13.7 14.3 14.3
9.9 10.6 11.2
10.6 11.2 11.8
10.6 11.8 12.4
14.3 14.9 14.9
13.0 14.3 14.3
13.0 14.3 14.3
11.2 11.8 13.0
11.8 13.0 13.7
13.7 14.3 14.3
10.6 11.2 11.8
9.9 10.6 11.2
13.0 13.7 14.3
14.3 14.9 14.0
11.8 13.0 13.0
10.6 11.2 11.8
11.2 11.8 13.0
13.0 13.0 13.7
9.9 11.2 12.4
9.3 9.9 11.3
9.9 10.6 11.2
13.0 13.7 13.7
11.8 13.0 13.7
11.8 12.4 13.0
11.2 11.8 12.4
10.6 11.2 12.4
14.9 14.9 15.5
-------
263
XW is ambient pollution as defined and calculated in Step C,
1 <• <1
equation (1) ; X. is ambient pollution in state i under scenario
CO) .
m; is base case ambient pollution m state i;
Y_^ are variables such as humidity and fine particle constituents
other than sulfate ion which affect the relationship between ambient
air quality and visual range;
Eq. (4-2) is a summary of a study of the determinants of visual
range in the eastern United States by D. M. Rote. [ Rote, 1982a]
4.2.6 Step E: Estimate the Value of Visibility Benefits of Hypothetical
Pollution Control Strategies
In this step the visibility value function is applied to the visibility
effects obtained in Step D. Visibility improvement attributable to a policy
equals the difference between visibility under a policy scenario and base-case
visibility. The value of visibility improvement depends upon the size of
the improvement, the characteristics of the people enjoying it, and the
prevailing level of visibility. The value of an extra mile of visual range
depends upon the income of a household, for example, and the number and ages
of household members. An extra mile of visibility is valued more when
prevailing visibility is low than when it is high.
The relationship between the expressed valuations and the influential
factors, or predictor variables, was specified according to economic theory
and measured econometrically in Section 2 of this study. The resulting
relationship is the visibility value function. By using the visibility
improvements and the predictor variables, a predicted value for visibility
improvement was calculated for each state in the eastern United States.
-------
264
In symbols, the use of the visibility value function in benefit estimation
can be expressed as follows:
/_v j i
(4-3) B = Z[1 - exp(-YAVS. )](a + Zg.X.,)N. , where
L r jm J x 13 3
B is aggregate dollar benefits of scenario m over the base case;
AVS. is change in visibility services from the m^ scenario over
the base case in the state as calculated using eq. (2-43) in
Section 2.4;
X.. is the value of the household characteristic in the
i] J
state;
N is the number of households in the j ^ state; and
j
the parameters y , a and the B^'s are as given in Ta.2-20 of
Section 2.4.
Regarding the values of the household characteristics (X^'s), for the
following variables, samplewide means were used: respondent believed he had
an excellent view (EXVIEW) , female head of household (FEMHOH), equipment index
(EQUIP), bad eyesight (POOREYES), rural residence (RURAL), activity index (ACT),
ownership of other residential property in eastern U.S. (PROP), and ownership of
occupied unit (OWN). For other variables, state-specific values were used.
These are household income (INCOME), income squared (INC0ME2), age of household
head (HOHAGE) , education of household head (HOHED), household size (HSLDSIZ),
visibility endowment (VISENDOW), percent nonwhite (NONWHITE), dummies for
Atlanta (A) , Cincinatti (C) , Miami (M) , and Washington, DC (W) .
In summary, the preceeding steps summarize the entire analytic framework
underlying the estimates of benefits that begins with the statement of policy
alternatives and ends with a dollar estimate of the benefits of these policies.
While the policy scenarios examined here are illustrative, the established
-------
2 65
framework has been shown to be entirely general and capable of analyzing any
set of policy alternatives that are of regulatory interest.
The following sections explain in more detail how the visibility value
function is applied, and present benefits estimates for hypothetical policy
scenarios for the year 1990.
-------
266
4.3 BENEFITS OF HYPOTHETICAL POLICY SCENARIOS.
In this section, calculations for two states are described to explain
how the visibility value function is used to derive benefits estimates. The
calculations illustrate the spatial nature of regional visibility effects.
Benefits for each state and for the eastern United States as a whole for the
hypothetical policy scenarios are presented.
4.3.1 Measurement of Physical Effects and Willingness to Pay for Improvements
4.3.1.1 Forecast Emissions under Scenario 5 in Georgia and Ohio (Step B)
Using Argonne scenario simulations, this section illustrates the
policy analysis process described in Section 4.2. For illustrative purposes
we consider two eastern states, Ohio and Georgia, and trace through the effects
of scenario 5 implementation in terms of the five steps previously outlined.
Ta.4-3, base-case emissions in the two states are given by the row "SC^
emissions" in kilotonnes per year. In the absence of visibility policy, ambient
SO2 emissions in Georgia would increase from 630 kilotonnes in 1980 to 873 kilo-
tonnes in 1990 and 1026 kilotonnes in 2000.
Under scenario 3, on the other hand, Georgia's SO2 emissions would be
554 kilotonnes in 1990 instead of 873, and 567 kilotonnes instead of 1026 in 2000. Thus
scenario 3 produces a 36 percent reduction in emissions in Georgia during the
1980's and a 15 percent reduction during the 1990's compared with the base case
projection. In Ohio the emissions pattern is quite different. Ohio's 1980
emissions are about four times higher than Georgia's—2748 kilotonnes vs 630
kilotonnes. However, Ohio's emissions are forecasted to decline between 1980
and 2000, even under the base-case forecast. Furthermore, policy effects in
Ohio are even greater than in Georgia. In Ohio, scenario 3 produces a 58
-------
TABLE 4-3
1
Policy Effects in Two States
GEORGIA
Base
Case: Scenario 2
Policy Scenario 5
Policy
Amount %
4
Effects
Amount I
1980
1990
2000
1990 2000
1990
2000
1
S0„ emissions
\ ¦ 2
Ambient SO
630.0
873.0
1026.0
554.0 567.0
-319.0 -36.0
-459.0 -45.0
7.3
9.8
11.7
7.1 8.2
- 2.7 -28.0
- 3.5 -30.0
3
Visibility
13.7
13.0
13.0
14.3 13.7
1.3 10.0
.7 5.4
Aggregate benefits
(per household)^
OHIO
365
(168)
Base
Case:
Scenario 2
Policy Scenario 5
Policy
Amount %
Effects
Amount %
1980
1990
2000
1990 2000
1990
2000
SO2 emissions
2748.0
2300.0
2207.0
964.0 1056.0
-1336.0. -58.0
-115.0 -52.0
Ambient S02
37.0
32.8
32.8
17.8 19.8
- 15.0 -46.0
- 13.0 -40.0
Visibility
Aggregate benefits
(per household)
8.7
9.3
9.3
12.4 11.8
3.1 33.0
1516
(360)
2.4 27.0
Kilotonnes per year
Micrograms per cubic meter
Miles
Physical effects are drawn from simulations provided by D.M. Rote of Argonne [Rote, 1982a, 1982b]
Aggregate benefits in millions of dollars per year; household benefits in (dollars per year). From Ta.4-6.
-------
268
percent emissions reduction during the 1980's. and a 52 percent reduction
during the 1990's. The combined effect of trends and policy effects in the
two states therefore, is that Ohio emissions in 1980 are over four times greater
than Georgia emissions, whereas by 2000 Ohio emissions are less than twice as
large as Georgia's.
4.3.2.1 Forecast Ambient Air Quality under Scenario 5 in Georgia and Ohio (Step C)
Ambient air quality is given by the row "Ambient SO2" in micro-
3
grams per cubic meter (yg/m ) in Ta.4-3. In 1980, ambient air quality is over
3
five times worse in Ohio than in Georgia by the SO2 criterion—37.0 yg/m in
Ohio vs 7.3 yg/m^ in Georgia. As in the case of emissions, air quality in Ohio
is projected to improve in the base case (from 37.0 yg/m"^ in 1980 to 32.8 yg/m^
in 2000) and to deteriorate in Georgia (from 7.3 yg/m"^ in 1980 to 11.7 yg/m^
in 2000). As for the policy effects of scenario 5 in the
two states, both states experience improvements in 1990 and
2000, compared with the without-policy or base-case scenario. However, taking
account of both trends and policy effects in the two states, Georgia experiences
3 3
a net deterioration in ambient air quality by 2000 (from 7.3 yg/m to 8.2 yg/m ) ,
while Ohio experiences a net improvement by 2000 (from 37.0 yg/m^ to 19.8 yg/m^) .
4.3.1.3 Forecast Visibility Effects of Scenario 5 in Georgia and Ohio (Step D)
Visibility effects of scenario 5 are given by the row labeled "Visibility"
for each state. In the absence of a visibility policy, Georgia is forecasted
to experience a reduction in visibility—from 13.7 miles in 1980 to 13.0 miles
in 2000. Ohio visibility improves from 8.7 to 9.3 miles over the same period in
-------
269
the base forecast. The effect of scenario 5 is to convert deteriorating
visibility in Georgia into improved visibility in 1990 (14.3 miles vs 13.0
miles). By 2000, visibility under scenario 5 has fallen back to its 1980
level of 13.7 miles, but it is still better than it would have been in the
absence of the policy—13.0 miles. The policy gains in Georgia are 1.3 miles
during the 1980's and 0.7 miles in the 1990's. In Ohio, visibility would
have improved even in the absence of a visibility policy—from 8.7 miles in
1980 to 9.3 miles in 1990 and 2000. But the policy effect is to produce an
even greater improvement—to 12.4 miles in 1990 and 11.8 miles in 2000. The
policy gains in Ohio are 3.1 miles in the 1980's and 2.4 miles in the 1990's.
4.3.1.4 Forecast Willingness to Pay for Visibility Improvements from Scenario 5
in Georgia and Ohio (Step E)
Monetary values of visibility improvements for each state are derived by
substituting appropriate values for each variable into the visibility value
function. The result is an estimate of the state population's maximum willingness
to pay for improved visibility in a given year. For example, from Ta.2-20, Section
2.4.5, the contribution of changes in visual range to the estimate of Ohio's
willingness to pay for the policy improvement is equal to 155.844 times (5.14
minus 4.57)(times 1.229)—the parameter estimate of VISENDOW times Ohio's 1990
visibility index change under scenario 5 times 8. The sum of similar calculations
over all the function variables in eq. (2-43), Section 2.4.4 equals Ohio's
policy benefit.
-------
270
Total benefits are estimated to be about $1.5 billion in Ohio and
$350 million in Georgia in 1990 under scenario 5. On a per-household basis,
Ohio benefits are about $360 and Georgia benefits about $170. These values
correspond to a 3.1 mile visibility-policy improvement in Ohio and a 1.3 mile
visibility-policy improvement in Georgia.
Ohio derives larger policy benefits than Georgia for a variety of reasons.
First, Ohio's population is larger. While household benefits in Ohio are
about 1.5 times greater than in Georgia, aggregate Ohio benefits are over
four times greater than aggregate Georgia benefits. Second, the policy effect
is almost two miles greater in Ohio than in Georgia, largely because of the
much greater emissions reduction required by Ohio. By dividing the percentage
change in visibility by the percentage change in emissions, we obtain a number
that measures the relationship between local benefit and local clean-up effort.
This may be done using numbers in Ta.4-3 for each state in 1990 and 2000.
The result is that the ratio is one fourth to one half as large in Ohio as in
Georgia. One of the main reasons for this result is that local visual range
is affected by distant sources of pollution as well as local sources. Hence
under scenario 5, Ohio derives visibility benefits from out-of-state emissions
reductions to a greater extent than Georgia.
The third reason is that Ohio citizens derive greater benefits from visi-
bility improvements in other states than do people living in Georgia. This is
because Ohio is more centrally located than Georgia with respect to regional
visibility improvements. According to the visibility value function, visibility
improvements in other eastern states are worth more to the citizens of Ohio
than they are to the citizens of Georgia.
-------
271
4.3.2 Aggregation of Physical Effects in the Eastern United States (Step C)
Ta.4-2 summarized the results of each of the alternative policies in
miles of local visibility by state. Comparison of scenarios 3, 4, and 5 with
the base case demonstrates the rather complex geographic distribution of local
visibility improvements that results from alternative policy standards.
Effects of policy on local visibility, as recorded in Ta. 4-2, do not
however describe the entire policy effect of relevance to the local area.
As explained in Part 2, distant visibility conditions are part of local endow-
ment. In other words, the entire column of improvements associated with each
regulatory strategy is relevant to the measurement of benefits in each state,
because they are all part of each state's visibility endowment.
Ta.4-4 gives measures of visibility sources for each state. The
measure of visibility services is a weighted contribution of visibility in
all states to the state in question, as obtained from eq.2-43 in Section 2.4.
Ta.4-4 was derived by using projected policy improvements for all states to
calculate visibility services for each state. Ta.4-5 gives an idea of the
relationship between the visibility services measure and local visibility in
miles for each state. States are ordered from highest to lowest on the endow-
ment index for 1980. The corresponding visibility in miles in each state does
not follow the same order. Florida, for example, has relatively high local
visibility, yet ranks last on the index scale because of its geographic remote-
ness from the rest of the coutry. Visibility in other areas contributes rela-
tively little to Florida's endowment. Fig.4-1 illustrates the visibility
endowment index for 1980.
4.3.3 Aggregation of Scenario Benefits in the Eastern United States, 1990--
Preliminary Estimates Subject to Revision
Ta.4-6 presents 1990 policy benefits for the three improvement scenarios.
Total program benefits for the three illustrative scenarios in the year 1990
-------
272
TABLE 4-4
Measure of Visibility Services (VS)
STATE Base Case Policy Scenarios, 1990
1980
1990
3
4
5
Alabama
4.59
4.52
4.53
4.67
4.72
Connecticut
3.72
3.70
3.75
3.90
4.06
D.C.
4.66
4.59
4.74
4.94
5.16
Delaware
3.73
3.67
3.78
3.92
4.08
Florida
3.51
3.44
3.46
3.56
3.58
Georgia
4.34
4.26
4.28
4.47
4.52
Illinois
5.52
5.52
5.56
5.73
5.81
Indiana
5.12
5.19
5.28
5.46
5.66
Kentucky
5.01
5.11
5.16
5.40
5.55
Maine
4.93
4.92
4.94
5.13
5.18
Maryland
4.71
4.63
4.80
5.00
5.24
Massachusetts
4.20
4.12
4.17
4.36
4.53
Michigan
4.94
4.94
4.98
5.12
5.26
Mississippi
4.94
4.83
4.84
4.95
4.99
New Hampshire
5.02
5.00
5.04
5.34
5.46
New Jersey
3.91
3.84
3.96
4.11
4.29
New York
4.36
4.34
4.48
4.65
4.93
North Carolina
4.53
4.44
4.56
4.66
4.80
Ohio
4.51
4.57
4.68
4.91
5.14
Pennsylvania
4.51
4.50
4.66
4.85
5.15
Rhode Island
3.70
3.64
3.68
3.84
3.98
South Carolina
4.54
4.46
4.52
4.67
4.76
Tennessee
5.11
5.11
5.14
5.37
5.49
Vermont
4.90
4.89
4.94
5.17
5.35
Virginia
4.87
4.84
4.99
5.17
5.37
West Virginia
4.69
4.69
4.82
5.01
5.25
Wisconsin
5.58
5.51
5.59
5.64
5.75
Source: Explained in text.
-------
273
TABLE 4-5
Ranking of States by 1980 Visibility Endowment
State
Visibility
Endowment Index
Visibility
in Miles
Wisconsin
5.
58
14.9
Illinois
5.
52
13.0
Indiana
5.
12
9.9
Tennessee
5.
11
11.8
New Hamsphire
5.
02
11.8
Kentucky
5.
01
10.6
Mississippi
4.
94
15.5
Michigan
4.
94
13.0
Maine
4.
93
13.7
Vermont
4.
90
11.8
Virginia
4.
87
10.6
Maryland
4.
,71
10.6
West Virgina
4.
69
9.9
District of Columbia
4.
,66
10.6
Alabama
4.
,59
14.3
South Carolina
4.
,54
13.7
North Carolina
4,
.53
13.0
Ohio
4.
,51
8.7
Pennsylvania
4
.51
8.7
New York
4,
.36
10.6
Georgia
4,
.34
13.7
Massachusetts
4,
.20
10.6
New Jersey
3
.91
10.6
Delaware
3
.73
10. 6
Connecticut
3
.72
9.9
Rhode Island
3
.70
10.6
Florida
3
.51
14.9
-------
-------
275
TABLE 4-6
Annual Household Benefits and Total State Benefits
Relative
to Base Case,
1990
Scenario
3
Scenario
4
Scenario
5
State
Benefits
State
Benefits
State
Benefj
Benefits
per
Benefits
per
Benefits
per
($ millions)
Household
($ millions)
Household
($ millions)
Housel
($)
($)
($)
New York
397
58
1111
162
2394
350
Pennsylvania
315
71
820
184
1725
386
Ohio
224
53
773
184
1516
360
Virginia
163
1 1
418
197
785
370
New Jersey
152
52
430
146
862
292
Maryland
150
84
388
216
756
421
North Carolina
111
49
244
107
492
216
Indiana
107
51
359
171
714
339
Illinois
93
21
634
145
1029
236
Wisconsin
89
48
174
93
368
198
Michigan
78
21
421
116
904
249
Massachusetts
48
21
282
124
588
260
West Virginia
39
59
109
163
219
328
Kentucky
30
22
211
157
380
282
South Carolina
30
26
126
110
217
190
Connecticut
28
22
211
157
380
282
Tennessee
24
14
244
142
427
249
Georgia
22
10
230
109
355
168
D.C.
20
70
56
192
107
371
Florida
20
4
214
48
342
1 1
Alabama
11
8
110
79
176
12 6
Delaware
11
44
30
123
61
248
New Hampshire
8
21
72
197
114
311
Mississippi
6
6
57
66
88
102
Rhode Island
6
14
33
87
72
187
Vermont
5
23
31
153
59
289
Maine
4
10
49
113
73
167
TOTAL
2,193
7,766
15,134
-------
276
range from about two billion dollars (scenario 3) to about fifteen billion
dollars (scenario 5) .
New York, Pennsylvania, Ohio, Illinois, Michigan, and New Jersey are
the six leading beneficiaries of scenarios 4 and 5 in 1990. New York,
Pennsylvania and Ohio lead in scenario 3 as well. These six states account
for between 50 and 60 percent of eastern benefits under all three scenarios.
New York, Pennsylvania and Ohio receive between 35 and 45 percent of eastern
benefits under all three scenarios. The pattern of benefits is a little
different on a per household basis. Still, it is the highly-populated and
industrialized Northern states that place the highest value on improved
visibility. While individual state rankings are somewhat sensitive to the
specification of the endowment index and the aggregation pattern based upon
contingent valuation, nevertheless the basic pattern is rather striking.
Figure 4-2 illustrates the geographic distribution of benefits derived from
scenario 3 relative to the base case.
-------
FIGURE 4-2
Benefits per Household of Scenario 3
Relative to Base Case, 1990
If.
LEGEND: HSBEN
] 10
Y77777A 2E
!!!!!!.'•!
4G
Ei.niitMiiiniri -
ifflSiM SE
N>
-------
278
4.4 SUMMARY OF PROJECT APPROACH TO VISIBILITY POLICY ANALYSIS
The monetary values of visibility policy benefits obtained in this
chapter for alternative hypothetical policy scenarios illustrate the accom-
plishment of the major project objective, which was to develop a method of
converting the physical effects on visual range of any proposed policy into
values of benefits indicated by people's willingness to pay in the eastern
United States. In this chapter we have described how policy scenarios that
affect SO^ emissions in the entire region can be translated into sets of
effects on visual range in each eastern state. This phase of the work was
completed by Argonne researchers, who simulated the visibility effects of
several regional policy scenarios which control SC^ emissions. The present
chapter also describes how the resulting geographical changes in visual
range are valued by the people of each state. This is accomplished by the
visibility value function, which is the most improtant output of this study
and is the expression that converts visibility changes into dollar values,
based upon the personal characteristics of the resident population, and the
goegraphic distribution and size of changes in visual range. Further work
could include a more refined investigation of the effect of distance on
valuation of visibility improvement. Additional econometric work could
investigate estimations in view of truncation of the dependent variable.
This work would extend the work reported on in Section 2.3. The importance
of unique eastern views to willingness to pay for eastern visibility improve-
ments could be studied in further contingent valuation survey work. These
CV results would extend the analysis of the six-city survey in this report,
which did not focus on existence of particular unique or spectacular scenic
eastern views. The secondary-data analysis of section 3 could be refined and
-------
279
additional work on attaching monetary values performed. The further unique-
view and secondary-data analysis could make possible a corroberation and
refinement of the six-city survey results that would be more extensive than
the one presently reported in Section 3.6 of this report. Further work along
the lines discussed in this paragraph is being undertaken in a follow-up study
now under way.
In closing, it should be emphasized that estimates of the visibility
valuation function are the best we have at this time, but are subject to
considerable refinement and investigation of reliability. The aggregate
benefits estimates have been presented only for purposes of illustrating
aggregation methodology. Care should be exercised that the results not be
used out of context. The policy scenarios are for various kinds of utility
controls and are not to be taken as indicating that these policies are actually
being contemplated or should be enacted. A major point in illustrating the
aggregation method is to emphasize there is no one unique value of increased
visibility, but rather the benefits of a program affecting visibility depends
on how much visibility is improved in different places, and on the numbers and
characteristics of people in the places affected. It would defeat a major
purpose of this study if the numbers in this chapter were applied out of context
to other programs. The use of the results of this study should be to estimate
differential improvements in visibility that would be brought about by a program
and then to use the visibility function to obtain benefits in different areas
which would then be summed. The purpose of this study has been to develop
operational tools. The tools can be applied for actual policy purposes, but
they have not been so applied in this study.
-------
REFERENCES
Abelson, Peter W., "Property Prices and the Value of Amenities," Journal
of Environmental Economics and Management, Vol. 6 (March 1979), pp. 11-28.
Auliciems, A., The Atmospheric Environment, Toronto, The University of
Toronto Press, 1972.
Becker, G. S., "The Theory of Allocation of Time," Economic Journal, Vol. 75,
1965, pp. 493-517.
Bender, Bruce, Timothy J. Gronberg and Hal-Shin Hwang, "Choice of Functional
Form and the Demand for Air Quality," Review of Economics and Statistics,
Vol. 63 (November 1980), pp.638-643.
Bishop, R. and T. Heberlein, "Measuring Values of Extramarket Goods: Are
Indirect Measures Biased?" American Journal of Agricultural Economics,
Volume 61 (1979), pp.926-939.
Bishop, R. and T. Heberlein, "Simulated Markets, Hypothetical Markets, and
Travel Cost Analysis," Department of Agricultural Economics Staff Paper
#187, University of Wisconsin, December 1980.
Blomguist, Glenn, "The Economic Value of Life: Implications of Automobile
Seatbelt Use," Urban Report No. 50, University of Chicago, February 1976.
Blomguist, Glenn, "Economics of Safety and Seat Belt Use," Journal of Safety
Research, Vol. 9 (December 1977), pp.179-189.
Blomguist, Glenn, "Estimating the Value of Life and Safety: Recent Developments,"
Paper for the Conference on the Value of Life and Safety held at the
University of Geneva, Geneva, Switzerland, on March 30-April 1, 1981.
Blomguist, Glenn, "The Role of Behavioral Response in Traffic Safety Regulation,"
Working Paper in Public Administration, University of Kentucky, Lexington,
October 1983.
Blomquist, Glenn, "Visibility, Views and the Housing Market," Department of
Economics, University of Kentucky, Lexington, December 1980.
Blomquist, Glenn and Lawrence Worley, "Hedonic Prices, Demand for Urban Housing
Amenities, and Benefit Estimates," Journal of Urban Economics, 1981.
Blomquist, Glenn and Lawrence Worley, "Specification of Demands for Housing
Characteristics" in G. S. Tolley and D. B. Diamond, eds., Amenities
and the Residence Site Choice, New York: Academic Press, Inc., 1982.
Bohm, P., "Estimating Demand for Public Goods: An Experiment," European
Economic Review, Vol. 3 (1972), pp.111-130.
-------
Brookshire, Davids., Ralph C. d'Arge, William D. Schulze, and Mark A. Thayer,
Vol. II of Experiments in Valuing Non-Market Goods: A Case Study of
Alternative Benefit Measures of Air Pollution Control in the South Coast
Air Basin of Southern California, EPA-600/6-7 9-0016, February 197 9.
Brookshire, D. S., B. Ires and W. Schulze, "The Valuation of Aesthetic
Preference," Journal of Environmental Economic Management, Vol. 3
(1978) , pp. 325-326.
Brookshire, D. S., A. Randall and J. R. Stoll, "Valuing Increments and Decre-
ments in National Resource Service Flow," American Journal of Agricultural
Economics, Vol. 62 (1980), pp.478-488.
Brookshire, D. S., Mark A. Thayer, William D. Schulze, and Ralph C. d'Arge,
"Valuing Public Goods: A Comparison of Survey and Hedonic Approaches,"
American Economic Review, Vol. 72(1) (March 1982), pp.165-177.
Brown, Gardner, M., Jr., and Henry 0. Pollakowski, "Economic Valuations of
Shoreline," Review of Economics and Statistics," Vol. 59 (August 1977),
pp.272-278.
Burt, Oscar R. and Durward Brewer, "Estimation of Net Social Benefits from
Outdoor Recreation," Econometrica, Vol. 39 (1971), pp.813-827.
Cicchetti, C. and V. K. Smith, Congestion, Quality, Deterioration and Optimal
Use: Wilderness Recreation in the Spanish Peaks Primitive Area,
Washington: Resources for the Future, Inc., 1973.
Cocheba, D. J. and W. A. Langford, "Wild Life Valuation: The Collective Good
Aspect of Hunting," Land Economics, Vol. 54 (1978), pp.490-504.
Deaton, Angus and John Muelbauer, Economics and Consumer Behavior, Cambridge
University Press: Cambridge, 1980.
Demmert, H. G., The Economics of Professional Tennis Sports, Lexington, 1973.
Domencich, Thomas A. and Daniel McFadden, Urban Travel Demand, North-Holland:
Amsterdam, 1975.
Eastman, C., P. Hoffer and A. Randall, Socioeconomic Analysis of Environmental
Concern: The Case of the Four Comers Electric Power Complex. Bulletin #626.
Agricutural Experiment Station, New Mexico State University, 1974.
Faigin, Barbara M., "Societal Costs of Motor Vehicle Accidents for Benefit-Cost
Analysis, " In Proceedings of the Fourth International Congress in Auto-
motive Safety, Washington, D.C., U.S. Department of Transportation,
National Highway Safety Administration, July 14-16, 1975.
Freeman, A. Myrick, III, "Approaches to Measuring Public Goods Demands,"
American Journal of Agricultural Economics, Vol. 61 (December 1979 a),
pp.915-920.
Freeman, A. Myrick, III, The Benefits of Environmental Improvement, Baltimore:
Johns Hopkins University Press for RFF, 1979 b.
-------
Freeman, A. Myrick, III, "Hedonic Press, Property Values and Measuring
Environmental Benefits: A Survey of the Issues," Scandinavian Journal
of Economics, 1979c, pp.154-173.
Ghosh, Debapriya, Dennis Lees and William Seal, "Optimal Motorway Speed and
Some Valuations of Time and Life," Manchester School of Economics and
Social Studies, Vol. 43 (June 1975), pp.677-726.
Goodman, Allen C., "Hedonic Prices, Price Indices and Housing Markets," Journal
of Urban Economics, Vol. 5 (October 1978), pp.471-484.
Haefner, Lonnie E. and Edward K. J. Morlok, "Optimal Geometric Design Decisions
for Highway Safety," Highway Research Record No. 371, Washington, D.C.
Highway Research Board, 1971, pp.12-23.
Hammack, J. and G. Brown, Waterfowl and Wetlands: Towards Bioeconomic Analysis,
Baltimore: Johns Hopkins University Press, 1974.
Harrison, David, Jr., and Daniel L. Rubinfeld, "Hedonic Housing Prices and the
Demand for Clean Air," Journal of Environmental Economics and Management,
Vol. 5 (March 1980), pp.81-102.
Hausman, J. A., "Specification Tests in Econometrics," Econometrica, Vol. 46
(November 1979), pp.1251-1270.
Hicks, J. R., "The Four Consumer's Surpluses," Review of Economic Studies,
Vol. 11 (1947) , pp. -41.
Knetsch, Jack L., "Displaced Facilities and Benefit Calculations," Land
Economics, Vol. 53 (1977), pp.123-129.
Knetsch, Jack L. and Robert K. Davis, "Comparisons of Methods for Recreation
Evaluation," in Allen V. Kneese and Stephen C. Smith, eds., Water Research.
Baltimore: Johns Hopkins University Press, 1966.
Krumm, Ronald J., "Neighborhood Amenities: An Economic Analysis, " Journal
of Urban Economics, Vol. 7 (March 1980), pp.208-224.
Krumm, Ronald J., "The Non-independent of Household Adoption of Home Space
Heating and Fuel Input," University of Chicago, Department of Economics,
Memo, 1980.
Lancaster, K., "A New Approach to Consumer Theory," Journal of Political
Economy, Vol. 74(2) (1966), pp.132-157.
Lancaster, K., Consumer Demand: A New Approach, New York: Columbia University
Press, 1971.
Linneman, Peter, "An Analysis of the Demand for Residence Site Characteristics,"
Ph.D. Dissertation, University of Chicago, 1977.
Linneman, Peter, "Some Empirical Results on the Nature of the Hedonic Price
Function for the Urban Housing Market," Journal of Urban Economics. Vol. 8
(July 1980), pp.47-68.
-------
List, R. J., Smithsonian Meteorological Tables, Washington, D.C., The
Smithsonian Institution, 1951.
Majid, I., J. Sinden and A. Randall, "Benefit Evaluation of Increments to
Existing Systems of Public Facilities," Land Economics.
Malm, William C., Karen K. Leiker and John V. Molenar, "Human Perception
of Air Quality," Journal of the Air Pollution Control Association,
Vol. 30 (February 1980), pp.122-131.
Manski, C. F., L. Sherman and J. R. Ginn, "An Empricial Analysis of Household
Choice Among Motor Vehicles," Cambridge Systematics, Inc., Cambridge, MA,
November 1978.
Middleton, W. E. Knowles, Visibility in Meteorology, Toronto, The University
of Toronto Press, 1941.
Mishan, E. J., "Evaluation of Life and Limb: A Theoretical Approach, " JPE;,
Vol. 79 (July/August 1971), pp.687-705.
Mood, Alexander M., Franklin A. Graybill and Duane C. Boes, Introduction to
the Theory of Statistics, McGraw-Hill: New York, 1974.
Morey, Edward R., "The Demand for Site-Specific Recreational Activities: A
CharacteristicApproach," Journal of Environmental Economics and Management,
Vol. 8 (1981), pp.355-371.
National Weather Service, Temperature-Humidity Index, Wind Chill Index.
Nelson, Jon, "Residential Choice, Hedonic Prices and the Demand for Urban Air
Quality," Journal of Urban Economics, Vol. 5 (July 1978), pp.357-369.
Nielson Co., A. C., Nielson Television Index.
Peltzman, Sam, "The Effects of Automobile Safety Regululation," Journal of
Political Economy, Vol. 83 (August 1975), pp.677-726.
Pollard, Robert, "Topographic Amenities and Building Height in an Urban Housing
Model," unpublished Ph.D. dissertation, University of Chicago, 1977.
Randall, A., B. Ives and C. Eastman, "Bidding Games for Valuation of Aesthetic
Environmental Improvements, " Journal of Environmental Economic Management,
Vol. 1 (1974), pp.132-149.
Randall, A. and J. R. Stoll, "Consumer's Surplus in Commodity Space," American
Economic Review, Vol. 70 (1980), pp.449-455.
Randall, A. and J. R. Stoll, Economic Surplus and Benefit Cost Analysis.
Agricultural Economic Research Report #135, University of Kentucky, 1980.
Ridker, R. and J. Henning, "The Determinants of Residential Property Values with
Special Reference to Air Pollution," Review of Economics and Statistics,
Vol. 49 (May 1967), pp.246-257.
-------
Rosen, Sherwin, "Hedonic Prices and Implicit Markets," Journal of Political
Economy, Vol. 82 (January/February 1974), pp.34-55.
Rote, D. M., Methodology for Analysis of Alternative Ambient Fine Particulate
Air Quality Standards, Draft Report for Planning and Standards Branch of
Air Quality Planning and Standards, USEPA, Argonne National Laboratory,
January 22, 1982a.
Rote, D. M., Visibility Simulations of Argonne Long Range Transport Model,
Personal Communication, January 18, 1982b.
Rowe, Robert D., Ralph C. d'Arge and David S. Brookshire, "An Experiment on
the Economic Value of Visibility," Journal of Environmental Economics and
Management, Vol. 7 (March 1980), pp. 1-19.
Schulze, William D., David S. Brookshire, Eric Walther, and Karen Kelley, "The
Value of Visibility in the National Parklands of the Southwest," Preliminary
Final Report to EPA (August 1980) .
Small, Kenneth A. and Harvey S. Rosen, "Applied Welfare Analysis with Discrete
Choice Models," Econometrica, Vol. 49 (1981), pp.105-130.
Smith, Barton, "Demand for Housing: A Reexamination of Methodological Issues,"
paper presented at the Conference of Urban Economics held June 24-25, 1980
at the University of Chicago, Chicago, IL.
Smith, Barton and Douglas B. Diamond, "Analysis of the Demand for Housing
Characteristics and Spatial Goods," paper presented at the AEA/AREUEA
meetings held in Denver on September 5-7, 1980.
Smith, V. L., "Experiments with a Decentralized Mechanism for Public Good
Decisions," American Economic Review, Vol. 70 (1980), pp.584-599.
Stoll, J., "Valuation of Hunting Related Amenities : A Conceptual and Empirical
Approach," unpublished Ph.D. dissertation, University of Kentucky, 1980.
Thaler, R. and Sherwin Rosen, "The Value of Saving a Life: Evidence from the
Labor Market," in N. Terleckyj, ed., Household Production and Consumption,
NBER studies in Income and Wealth No. 40, New York: Columbia University
Press, 1976, pp.265-298.
Tideman, T. N. and G. Tullock, "A New and Superior Process for Making Public
Choice, Journal of Political Economy, Vol. 84 (1976), pp. 1145-1159.
Tolley, George S. and Douglas B. Diamond, eds., Amenities and the Residence
Site Choice, New York: Academic Press, Inc., 1982.
Trijonas, J. and K. Yuan, "Visibility in the Northeast: Long-Term Visibility
Trends and Visibility/Pollutant Relationships," EPA Report 600/3-78-075,
August 1978.
U.S. Council of Economic Advisors, "Economic Indicators, January 1980,"
Washington, D-C.: U.S.G.P.O., 1930.
-------
U.S. Department of Transportation, National Highway Traffic Safety Administra-
tion, Fatal Accident Reporting System, 1979, Report DOT HS 805 570,
January 1981.
U.S. Water Resources Council, "Procedures for Evaluation of National Economic
Development Benefits and Costs in Water Resources Planning: Final Rule,"
Federal Register, Vol. 44(242) (December 14, 1979), pp.72892-72976.
Willig, R. D., "Consumer Surplus Without Apology," American Economic Review,
Vol. 66 (1976), pp.589-597.
Witte, Ann D., Howard J. Sumka and Homer Erekson, "An Estimate of a Structural
Hedonic Model of the Housing Market: An Application of Rosen's Theory
of Implicit Markets, " Econometrica, Vol. 47 (September 1979), pp.1151-1173.
-------
EXECUTIVE SUMMARY
Visibility is a pervasive and inescapable phenomenon, subject to both
general and periodic deterioration, which affects extremely large numbers
of people. The relative neglect of visibility as a subject of investigation
appears to be due not to its lack of importance, but rather to the fact that
it is more difficult to value than many other environmental attributes.
Previous work on visibility has concentrated on sparsely populated
areas of the West. The present research, concerned with visibility in the
eastern United States, deals with larger numbers of people under a wider
variety of circumstances. People in urban and rural areas are affected in
the course of daily living, and a variety of special activities centering
on recreation and related activities are sensitive to visibility conditions.
Four major objectives have been accomplished by the research. The
first objective was to use the contingent valuation (CV) approach to obtain
information on values attached to visibility in the eastern United States.
A major conceptual effort to extend and refine the CV technique preceeded
data gathering. Several different CV formats were pre-tested in Chicago,
followed by a six-city eastern survey.
The second objective was to define and estimate a visibility value
function. The benefits of a visibility policy depend upon the extent of
visibility improvement, on initial visibility conditions and their geographic
distribution, and upon social and economic characteristics of people in various
regions. Benefits are related to these variables in the visibility value
function.
The third major objective was to identify particular activities likely
to be influenced by visibility and to measure the effects of visibility on
these activities using secondary data. Activities investigated were swimming.
-------
ii
television viewing, baseball attendance, Hancock Tower visitation, fatal
and non-fatal traffic accidents, and air traffic counts. An important
result of these studies is to corroborate findings from the aggregate func-
tion based on the contingent value (CV) approach.
The fourth major objective of project research was to establish a
rigorous and operational method of aggregating visibility policy benefits
over the entire eastern U.S.
OBJECTIVE ONE: CONTINGENT VALUATION SURVEY
The theory of household production was used in the development and use
of a contingent valuation (CV) survey questionnaire. There are seven basic
modules to the CV instrument.
Module 1: Area Context Module
The area over which visibility improvements were offered had to be
clearly comprehended by each individual. For the research to provide results
on regional differences in air quality improvement, it was important to collect
willingness-to-pay (WTP) data for improvements in visibility (i) in the indi-
vidual's home sub-region, and (ii) in the whole study region. A map card and
a portfolio of photographs were used to convey the size and diversity of the
region over which visiblity is valued.
Module 2: Visibility Module
The nature of alternative levels of visibility was communicated via
color photographs. This required a set of scenes representative of the area
over which visibility changes were to be valued. For each level of visibility
a set of the same scenes, with only the visibility different, was used. Some
factual verbal material was used to quantify the visual range represented in
-------
iii
each photo set. Separate photo sets were used to represent the sub-region,
the entire East, and the West.
Module 3: Activity Module
To employ the household production model, it was necessary to know
the following:
• the activities produced in the household,
• the inputs, other than visibility, used in activity production,
• the activity production technology used, and
• whether visual air quality is the only air quality input used
and, if not, whether visual air quality is used by the subject
as an indicator of other aspects of air quality. For example,
the individual may avoid strenuous outdoor sports on days of
poor visibility, not because visibility per se is an important
input, but because he treats poor visual air-quality as an
indicator of high pollutant concentrations which threaten
respiratory stress.
The module served to sensitize the individual to the variety of activities
in which he might value visibility.
Module 4 : The Market Module
Contingent valuation established a hypothetical market and encouraged
individuals to reveal their WTP by using that market. Major elements of this
module described what was being purchased through the bid and the market rules
regulating payment for and receipt of the good in question. To describe the
good available for purchase, the general level of visibility as well as possible
increments and decrements in visibility were portrayed in both photographs
and narratives. Market rules provided assurance that the increment in visi-
bility would be delivered if and only if the respondent was willing to pay.
Module 5: The WTP Data Collection Module
This module presented the fundamental WTP questions. Respondents bid
first on local improvement, and then were asked how much they would add to
their local bid to extend the improvement to the East and then to the entire U.S.
-------
iv
Module 6: Post-Bid Probing
With certain market rules and WTP formats, some individuals recorded
a zero WTP which, in further questioning, turned out to be a protest against
some aspect of the format rather than an accurate reflection of the value
of the good offered. Probing of zero WTP's was an important element of the
data-collection schedule.
Module 7: Socio-Demographic Data
This module collected an array of socio-demographic data, including
full income concepts relevant to the processes through which individuals
demand and hence value, visibility.
OBJECTIVE TWO: VISIBILITY VALUE FUNCTION
The objective of the contingent valuation research was to define and
estimate a visibility value function. The theory of household production,
fundamental to the development of the CV questionnaire, was equally impor-
tant to the development of the visibility value function. The importance
of regional or spatial economics was recognized and receives its most com-
plete formulation in the visibility value function.
Central to the development of the visibility value function is the
concept of visibility services. Visibility services are aggregates
of visibility in different places, weighting each place's
contribution by its distance, scenary, and quality. Accordingly, there is
a production function relating visual services to these variables. Speci-
fically the production function for visual services (VS) is
(1) VS. = IVR^SM^D?3 SC?1"
1 ' j i i l ij l '
where VS is household j's consumption of VS , VR, is visual range in
i 1
state i , SM. is the area of state i in square miles, D, , is the distance
1 i]
-------
V
between household i and the center of state i , and SC, is a measure
1
of scenery in state i .
It was reasoned that the marginal benefit curve, or bid curve for a
change in visibility services, should have the following properties:
Property 1: BID(O) = 0
Property 2:- BID'(AVS)
Property 3: 3ID"(AVS) _< 0
Property 4: Limit BID'(&VS) =0 as AVS -»¦ 00
A functional form was required that would be consistent with Properties 1-4
and capable of handling both continuous and discrete explanatory variables.
Furthermore a functional form was needed which allows the bid curve to
pivot around the origin with changes in the vector of explanatory variables
while preserving these properties. The following negative exponential func-
tion was found to fulfill their requirements:
(2) BID = [1 - exp(-yAVS)] ,
which is monotonic increasing, passes through the origin, and has an upper
limit of +1 for all positive values of y • This gives the prototype
bid function. A rotational vector of household characteristics H , is
included:
(3) H = (a + E6.Z. . + u.)
[ ' i ij j
so that H is a linear combination of household characteristics Z , and
there is an unobserved household-specific rotational parameter u .
The empirical bid curve is given by the product of (2) and (3) or
(4) BIDj = [1 - exp (-yAVS^ ) ] [a + + u. ) ] ,
-------
vi
where VS is given by (5), below, BID. is the willingness-to-pay
]
of household j , AVS is given by changes in equation (1) due to the
program; a is a common intercept term (of rotation, not level of bid);
Z is the vector of household characteristics with parameters 6 ; and u^
is the household-specific rotation of the bid curve.
The formula used to calculate VS for the empirical analysis is
i -1.5
(R) VS. » ZVR *SH. *D.
. J ill
where the exponent on the distance variable was estimated by a maximum
likelihood method jointly with the vector of household characteristics and
the parameter y .
The estimation results for the visibility function are shown in Table 1.
Overall, between one-half and two-thirds of the variation of BID is accounted
for by the explanatory variables. The positive effect of a change in visibility
on BID is reflected in coefficient of 0.700 for GAMMA. The common constant term
ALPHA added to the individual estimated effects of household characteristics
in determining rotation of the bid curve, is negative.
The first variable in H, rotating the bid curve is VISENDOW, the
initial level of VS as calculated in (5) above. This variables has a posi-
tive effect and captures the net result of a pure endowment effect from
diminishing marginal utility, a sorting effect and a substitution effect.
A point estimate of the income elasticity of rotation is 0.539 is
computed, holding all non-income variables at their means. The first-order
effect of income (INCOME) on BID is positive, and the second-order effect
(INCOME SQUARED) and the income-age interaction effect (INCAGE) are negative.
-------
vii
TABLE 1
Non-Linear Least Squares Summary Statistics
Dependent Variable Bid
SOURCE
DF
SUM OF SQUARES
MEAN SQUARE
REGRESSION 22
RESIDUAL 3122
UNCORRECTED TOTAL 3144
130303017.02030957
140479409.60049038
270782426.62079995
5922864.41001407
44996.60781566
(CORRECTED TOTAL)
3143
233630610.10008546
PARAMETER
(VARIABLE)
ESTIMATE
GAMMA
ALPHA
VISENDOW
INCOME
INCOME2
INCAGE
HSLDSIZ
HOHED
HOHAGE
EQUIP
EXVIEW
BADEYES
ACT
PROP
FEMHOH
OWN
RURAL
NONWHITE
A
C
M
W
0.700
-472.606
155.757
14.797
-0.029
-0.172
5.327
-2.011
1.586
4.417
-67.139
12.065
5.175
97.183
50.684
-138.736
-41.049
-78.691
139.928
-187.137
112.550
-17.078
-------
viii
The negative interaction term confirms the hypothesis that the marginal
propensity to consume visibility decreases with age.
Turning to the human capital variables, the estimate of the education
parameter (HOHED) is negative, SO that more educated persons tend to bid
less, holding the other variables constant.
The age variable HOHAGE must be considered jointly with the variable
INCAGE. For very low income households, age actually increases WTP for VS,
but at an income of about $9,000 per year the net effect becomes negative.
Nonwhites (NONWHITE) bid significantly less than whites, while females
(FEMHOH) bid more than males.
Poor eyesight (BADEYES) and ownership of specialized capital equipment
(EQUIP) did not have a clear effect. As expected, participation in activities
(ACT) has a positive influence on bids,, reflecting the non-rivalness of visi-
bility within the household. There is a negative influence of view quality
(EXVIEW) on bids, which could be the result of diminishing marginal utility
combined with a fixed factor (view).
With regard to the property ownership variables, home ownership (OWN)
had a negative impact and the ownership of other residential property (PROP)
had a positive effect.
In addition to the urban/rural dummy variable a set of four city-specific
dummy variables were used to help account for unexplained differences between
cities. Only four were used since one of the six city degrees of freedom
has already been used up by the variable VISENDOW and the intercept terms uses
another. The four cities with dummies are Atlanta, Cincinnati, Miami, and
Washington, with variables names A, C, M, and W respectively. Boston and
Mobile remain as the base.
-------
ix
OBJECTIVE THREE: EFFECTS OF VISIBILITY ON BEHAVIOR
To complement the contingent valuation work and the visibility value
function based on it, a series of studies of the effects of visibility on
particular activities was carried out. Evidence that the CV and behavioral
results are consistent strengthens confidence in the results as well as
the methods that have been developed to obtain them.
Swimming
The swimming model assumes a linear relationship of the form
P = a + 0iV+.|2 f^X.
where P is daily pool attendance, V is visibility, and are other
factors which effect attendance. Visibility was found to have a significant
effect on attendance. The effect differs between years and ranges between
1.24 and 3.73 persons per tenth-of-a-mile increase in visibility. A one
mile increase in visibility increases attendance from three to five percent.
Television and Baseball
Similar analyses were performed on afternoon television viewing and
on Chicago Cubs baseball attendance. The effect of a one mile increase in
visibility on afternoon viewing is that 0.134% of 3 million households stop
watching T.V., or about 4000 households. Weekend viewing is reduced by an
additional 400 households. An increase in visibility of one mile increases
Cubs gate attendance by approximately 125 people. The change in consumer's
surplus associated with increase in visibility is at least 2.7 cents per
person in attendance, or approximately $30,000 for a typical season's attendance
The benefit of a one mile visibility improvement represents somewhat less than
one million dollars per year for baseball attendance in the entire U.S.
-------
X
Hancock Tower Recreation
The Chicago Hancock Tower offered an opportunity to determine the
effects of visibility on the demand for view services. Using visitation
data, it was possible to estimate the demand for Hancock Tower view
services as a function of admission price, visibility, and a set of demand
shifters. A mean per person consumer surplus of $2.12 in 1981 prices was
computed from the demand estimate. Assuming that similar experiences are
obtainable in other areas of the region, aggregate consumer surplus would
be $275 million in 1981 prices.
Contingent valuation responses were also obtained at the Tower. The
results indicate no significant difference between demand-based estimates and
contingent valuation bids.
View-OrientedResidences
An analysis of view-oriented submarkets of the residential housing
market was undertaken. The objectives were: (1) to measure the values of
views and view characteristics including visibility using a survey instru-
ment which establishes a contingent market for each; (2) to measure the values
of views and view characteristics using a hedonic-demand analysis of housing
consumption for the same group surveyed and (3) compare the contingent values
from the survey and the implicit values from the housing market for indivi-
duals dwelling in view-oriented residences.
The similarity of the contingent and implicit values for height (10 floors
up) , the high response rate on the bidding experiment and the significant
coefficients in the renters' housing hedonic equation suggested that contin-
gent value and market values are similar.
Air Traffic
To investigate the effects of visibility on air traffic, empirical
-------
estimates were made of visibility effects on take-offs and landings at
three Chicago-area airports. The effects of visibility on the air traffic
counts were found to be positive and highly significant in all areas.
The elasticities of traffic counts with respect to miles of visibility
were 0.415, and 0.392 and 0.250 at Aurora, DuPage and Meigs Field airports
respectively. The other variables in the regressions, including rainfall,
snow, fog, temperature, wind speed, wind direction, and day of the week
were in almost all cases of expected sign and significant.
Auto Traffic
A model of the relationship between travel cost, accident rates,
weather conditions, improvement in visibility, vehicle speed, and traffic
congestion was developed. It was shown that the total effect of an improve-
ment in visibility on accident rates depends crucially on the effect of
improvements in visibility on vehicle speed.
The empirical estimations of the relationship between improvements in
visibility, weather variables and traffic casualties show that visibility
improvements lead to significant reductions in non-fatal accidents in both
Cook and DuPage Counties, in the Chicago SMSA. This result is consistent
with the partial effect of improvements in visibility on highway casualities
While the occurrence of rain and/or snow leads to an increase in the number
non-fatal accidents in Cook and DuPage Counties, the results also show that
an improvement in visibility in the presence of snow leads to a decrease in
the number of non-fatal accidents in both counties.
Results of linear probability models in analyzing traffic fatalities
show that an improvement in visibility during the weekends leads to an
increase in the probability of occurrence of fatal accidents in Cook and
DuPage Counties. Visibility improvements in winter and spring, however,
-------
xii
lead to decreases in the probability of occurrence of fatal accidents in
both counties, although these coefficients are not very precisely estimates.
An improvement in visibility in Cook Counry by one mile leads to an estimated
benefit of 9.45 million dollars as a result of reduction in traffic casualties.
OBJECTIVE FOUR: EVALUATION OF POLICY EFFECTS ON VISUAL RANGE
A detailed illustration of the application of the visibility value
function to analysis of policy benefits was developed. Forecasting visi-
bility policy effects requires comparing a without-policy or base-case
scenario with one or more regulatory scenarios. The visibility
value function was applied to four hypothetical or illustrative policy
scenarios for electric utility pollution control relative to a base--case
scenario. Benefits connected with these purely illustrative scenarios were
estimated for the year 1990. Specifically, aggregate and per-household benefits
were estimated for each eastern state and the eastern United States.
A method was needed which relates reductions in pollution emissions
from the scenarios to visibility improvements. The relation between emissions
and visibility was provided by results from research at Argonne National
Laboratory.
Illustration of Method
Step A in the analysis of visibility regulation was to establish policy
alternatives. Alternative policies produce different patterns of visibility
improvement whose effects need to be evaluated in order to make a policy
choice. Three such policies were considered. In addition to the policy
scenarios a without-policy or base-case scenario was formulated. The base-
case scenario is a judgement as to the most likely regulatory climate in the
-------
xiii
absence of a visibility policy. it provides the standard against which
the benefits of the policy scenarios are measured.
Step B was to forecast emissions under the base-case and policy
scenarios by type of emittor, season and amount of pollution. These
forecasts depended in part on the technical requirements of pollution
abatement. To an even greater extent the emissions forecasts depended
upon forecasts of future levels of economic activity.
Step C was to forecast the spatial distribution of ambient air quality.
The relationship between emissions and ambient air quality depends upon
the way emissions are dispersed geographically and the chemical transformations
that occur during dispersion. This step was performed for each of the
scenarios by means of the Argonne long range transport model. [Rote, 1982a]
Step D was to measure the effects on visibility of ambient air quality
resulting from each scenario. The solution to this problem, also supplied
by Argonne [Rote, 1982b] provided a set of predictions as to the course
of visual air quality on a state by state basis in the future.
Step E was to use the visibility value function to establish values
associated with alternative pollution control strategies. Each scenario
produced a set of improvements in visual range for each state in future years.
The function estimated the value of these improvements to a state as the
sum of the value of the local component and value of improvements in other
parts of the region due to existence and option values. Non-local improvements
are less valuable to the state depending upon their distance from the state.
The value of visibility improvements is the sum of all local and non-local
improvements for all states in a given year. The visibility value function
evaluated improvements for each state in all years for each of the four
policy scenarios.
-------
xiv
Aggregation of Illustrative Scenario Benefits in the Eastern United States, 1990
Table 2 presents 1990 policy benefits for the three illustrative
improvement scenarios. Total program benefits for the three illustrative
scenarios in the year 1990 range from about two billion dollars (scenario 3)
to about fifteen billion dollars (scenario 5).
New York, Pennsylvania, Ohio, Illinois, Michigan, and New Jersey are the
six leading beneficiaries of scenarios 4 and 5 in 1990. New York, Pennsylvania
and Ohio lead in scenario 3 as well. These six states account for between 50
and 60 percent of eastern benefits under all three scenarios. New York.,
Pennsylvania and Ohio receive between 35 and 45 percent of eastern benefits
under all three scenarios. The pattern of benefits is a little different on
a per-household basis. Still, it is the highly populated and industrialized
Northern states where the highest values of improved visibility occur. While
individual state rankings are somewhat sensitive to the specification of the
endowment index and the aggregation pattern based upon contingent valuation,
nevertheless the basic pattern is rather striking.
Estimates of the visibility valuation function are the best we have
at this time, but are subject to considerable refinement and investigation
of reliability. The aggregate benefits estimates have been presented only for
purposes of illustrating aggregation methodology. Care should be exercised that
the results not be used out of context. The policy scenarios are for various
kinds of utility controls and are not to be taken as indicating that these
policies are actually being contemplated or should be enacted. A major point
in illustrating the aggregation method is to emphasize that there is no one
-------
($)
350
386
360
370
292
421
216
339
198
249
260
328
282
190
244
249
168
371
126
248
311
102
187
289
167
xv
TABLE 2
Annual Household Benefits and Total State Benefits
Relative to Base Case, 1990
Scenario 3
State
Benefits
($ millions)
Benefits
per
Household
Scenario 4
State
Benefits
($ millions)
Benefits
per
Household
Scenario
State
Benefits
($ millions)
397
58
1111
162
315
71
820
184
224
53
773
184
163
11
418
197
152
52
430
146
150
84
388
216
111
49
244
107
107
51
359
171
89
48
174
93
78
21
421
116
48
21
282
124
39
59
109
163
30
22
211
157
30
26
126
110
28
22
137
109
24
14
244
142
22
10
230
109
20
70
56
192
20
4
214
48
11
8
110
79
11
44
30
123
8
21
72
197
6
6
57
66
6
14
33
87
5
23
31
153
4
10
49
113
2,193
7,766
2394
1725
1516
785
862
756
492
714
368
904
588
219
380
217
308
427
355
107
342
176
61
114
88
72
59
73
15,134
-------
xvi
unique value of increased visibility, but rather the benefits of a program
affecting visibility depend on how much visibility is improved in different
places, and on the numbers and characteristics of people in the places
affected. It would defeat a major purpose of this study if the numbers in
this study were applied out of context to other programs. The use of the
results of this study should be to estimate differential improvements in
visibility that would be brought about by a program and then to use the
visibility function to obtain benefits in different states which would then
be summed. The purpose of this study has been to develop operational tools.
The tools can be applied for actual policy purposes, but they have not been
so applied in this study. Further work is being undertaken to extend and
refine the results of this report.
-------
APPENDIX A: SURVEY INSTRUMENT
This Appendix contains the Contingent Valuation instrument used in
the Eastern survey. It contains the modules discussed in detail in the
main report. The same survey was used in all six cities, within some ci
specific modifications, as on page 3.
-------
Form# A
IntPr";
- /?¥
& '¦
[Check One]
City_
ATLANTA
Center City_
Suburban
Rural
EASTERN U.S. RESIDENTS
e
c
h
;A
the University of Chicago. We are
( as part of a research study about
e are talking with a scientifically
dents, the viewpoint of your house-
la. Are you the male/female head of household?
YES (Go to statement at bottom of page)
NO (Ask Ib-)
lb. Is the male or female head of household at home?
YES (Ask to speak with head of household. Start Over.)
NO (Thank respondent and terminate.)
Fine. I have a few questions that I would like to ask you.
It will take about 20 minutes, and your answers will kept
confidential.
-------
FORM NUMBER A-174
ACTIVITY SHEET
GROUP 1
Walk to Work
Drive to Work
Eat Lunch Outdoors
Leave Place of Work
for Lunch
Take a Vacation Day
Outdoor Work Around House
Employed in Outdoor Job
GROUP 2
Jogging/Running/Bicycling
Swimming/Sailing
Tennis(outdoor)/Golf
Outdoor Team Sports
GROUP 3
Sightseeing(Rural or Urban)
Photography (Outdoor)
Drive in the Country
Flying/Gliding/Hang Gliding
GROUP 4
Stroll in the Park
Walk the Dog
Sunbathe
Go to Outdoor Fair/Concert
Play Catch/Frisbee
GROUP 5
Indoor Tennis/Racketball/
Basketball/Volleyball
Work Out at the Gym
Bowling
Other Stenuous Indoor Activities
GROUP 6
Go to Shopping Mall
Go to Museum
Go to Movies
Other Indoor Activities
Away From Home
Group 7
Stay at Home
GROUP 8
Nature Study/Bird Watching
Fishing/Hunting
Hiking/Trail Riding
Camping/Backpacking
Attend College or Pro Ballgame
Sightseeing Outside Local Area
Visit Friends in East U.S.
Visit Friends in West U.S.
Visit State/National Park
Other Activities Away
From Local Area
-------
SKETCH OF
PHOTOGRAPH DISPLAY BOARD FOR
LOCAL VISIBILITY IN THE EAST
Apartments
and
Skyline
Apartments
and
Skyline
Apartments
and
Skyline
Poor Visibility
Medium Visibility
Excellent Visibility
t"1
1
H
1
L - I - 2
L - I - 3
Outer Drive
Outer Drive
Outer Drive
Poor Visibility
Medium Visibility
Excellent Visibility
L - II - 1
L - II - 2
L - II - 3
Urban Shoreline
from High Floor
Urban Shoreline
from High Floor
Urban Shoreline
from High Floor
Poor Visibility
Medium Visibility
Excellent Visibility
L - III - 1
L - III - 1
L - III - 3
-------
3
SKETCH OF
PHOTOGRAPH DISPLAY BOARD FOR
VISIBILITY IN THE EASTERN REGION AS A WHOLE
Great Smokies
Poor Visibility
E - 1
Great Smokies
Medium Visibility
E - 2
Great Smokies
Excellent Visibility
E - 3
-------
4
SKETCH OF
PHOTOGRAPH DISPLAY BOARD FOR
VISIBILITY IN THE WEST
Grand Canyon
Poor Visibility
W - 1
Grand
Canyon
Medium
Visibility
W
- 2
Grand Canyon
Excellent Visibility
W - 3
-------
;T •
' * •
' v • i
• ** - >
f-3
¦-» - -'Si-fV
,-' t. ;Jif ^
.¦-f ; - vJ
x- - ' . ¦ i "* ~
• -X' J. i ' !
.; .-\ . •; i ;
-* - -,f#- i ¦ " -2 ' ---
¦•; ' '•• •!
"fc-V
1
-------
f«. " **
; uf. ' "r*vs'
, l -"V .;•. -iv -;, ,.
! mmm
& " - jhb*^'
\M - -?
VJ-V
IV- \
-------
-------
L-jl -3
L-TT-V
L -ii-|
-------
-------
-2-
1. [Hand respondent Activity Sheet]
Please look at this sheet. it lists some of the things
people do with their time. Place an X beside each activity that
you do in the course of an ordinary year. if there are any other
activities that you do, check the spaces marked 'other'.
[Pause, for respondent to complete Activity Sheet]
2. Do you own or have the use of the following items?
[Check For Yes]
Binoculars
A light plane, glider, hang glider, or
hot air balloon
A birdwatcher's guide
A recreation vehicle, camper, or motor home
A guidebook for amateur astronomers
A camera with telephoto lens
Backpacking equipment
A vacation home or cabin
-------
-3-
[Present photograph set]
3. Now, please look at these photographs. Each row shows the
same scene, only with different visibility. [point to photos] The
pictures on the left show a visibility of 4 miles. The ones in the
center show 13 miles, and the ones on the right show 30 miles,
Notice that when visibility increases you can see farther, and the
things you do see become sharper and more distinct. [PAUSE]
a) [Present card A] This card shows the relationship between
the photos and visibility. If you had to guess how many
miles would you think you could see on a typical Atlanta day?
It doesn't have to be one of these photos, they are just there
to help you.
Enter Guess (In Miles)
Records show that typical visibility in the Atlanta area
is actually about 10 miles.
Please look again at the activity sheet.
b) Are there any activities which you would do on a day with 30
miles visibility, which you wouldn't do with 13 miles? Which ones?
c) Are there any activities which you would do on a day with 13
miles which you wouldn't do with 4 miles? Which ones?
-------
-4-
In the following questions, we would like you to answer for
your entire household, that is, any one who contributes to, or is
supported by, household income. To understand your answers, we
need to know how many people are in your household. How many are
there?
Enter # in household
4. Let us return to the photographs.
Visibility is affected by both natural and man-made causes.
In particular, there are a number of man-made things in the air
which do not affect health but .do affect visibility. We can do
something to affect these things, but this costs all of us money,
since it makes the things we buy more expensive. The following
questions are designed to help us find out how much visibility is
worth to you.
[Present Expenditure Card, and then read slowly]
I'd like you to look at this card. It shows how much a
typical household with the indicated income spends each month f°r
various things. Included are expenses for ordinary goods, like
groceries and housing. Also, it shows how much is paid, through
taxes and higher prices, for various public programs. Some of
these expenses are quite small, like for toothpaste and the space
program, while others are quite large, like for housing and
national defense.
[Pause, to allow respondent to examine card.]
You may look at this card if you wish to help answer the next
few questions.
-------
-5-
[Present Card B]
4a.. Typical visibility in the Atlanta area is 10 miles. Consider
what would happen if typical visibility in Atlanta fell to 5
miles. A program could be set up to prevent the decline. if the
total cost of the program to you/ [your household] was $13 a month,
would you accept the program or reject it?
Accept.
Re j ect_
[Check One]
Now, assume the program would cost $
accept the program or reject it?
./month. Would you
[ Follow Bidding Instructions. If respondent bids zero, ask
QUESTION 4b. Otherwise, enter BID4 and go on to question 5)
[Enter maximum amount ACCEPTED.]
./month [BID4]
kkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkk
4b. ONLY THOSE WHOSE FINAL RESPONSE WAS $ZERO FOR QUESTION 4a.
[Present Card C]
Did you reject the program which would spend your money to
maintain visibility because:
[Check Only One]
_Visibility is not worth anything to
you (or, it wouldn't matter even if
visibility declined to 5 miles).
_You would appreciate [or value] improved
visibility, but you think someone else
should be made to pay for it.
Some other reason:
* [If respondent says someone else should pay, then say: ]
Later, you will get a chance to say who should pay. For now,
we are interested in finding out how much it is worth to you.
Let's say that you could buy visibility, and there was no one else
to pay or enjoy the benefits. Then, would you be willing to Pay
something?
YES (Go back to 4a.) NO [Go on to Q 5.]
-------
-6-
[Present Card D]
5. Now let's go back to the our starting point, where typical
visibility is 10 miles. A program could be set up to improve
it to 20 miles. Suppose the total cost of the program to you
would be $13 a month. Would you accept the program, or
reject it? (Point out change on Card D)
Accept
[Check One]
Rej ect
What if it cost $ * /month. Would you accept the
program or reject it and stay at 10 miles?
*(Bidding as for Q.4)
/month [BID5: Remember this amount]
-------
-7-
Present Card E]
For the next question:
If BID5 is GREATER THAN ZERO, say the words in (). If BID5
was ZERO, sav the words in < >.
6. Now, what if the program improved visibility all the way to
30 miles?
Would you accept the 32 mile program if it cost
($10 more, for a total of $ TBID5 + 101 per month?)
[OR]
<$13 a month?>
Accept
[Check One]
Rej ect
What if it cost $ * /month (more, for a total of
$ (BID5 + *1 ?) Would you accept the program or reject it?
(Bidding as for Q.4)
Enter both BID5, the additional amount bid for Q.6,
and BID6. in the three answer blanks provided.
ENTER: $ + $ MORE = $ /month
(BID5) (BID6)
-------
-8-
[Present Card F and Eastern U.S. Photo Set]
7. Now let's consider a program which would improve visibility
in Atlanta by ten miles, AND ALSO improve visibility in the
rest of the Eastern section of the United States by ten
miles. The shaded area on this map shows the area to be
covered by this program. [BE SURE RESPONDENT UNDERSTANDS
THAT THE ATLANTA AREA IS INCLUDED!]
(Before, you accepted a ten mile improvement in Atlanta alone
when it cost $ [BID5]/month. )
If this program cost you/your household
($10 a month more, for a total of $ TBID5 + 10]
<$13 a month>
would you accept the program or reject it?
Accept
[Check One]
Rej ect
What if it cost $ ] /month (more, for a total of
$ [BID5 + *1 ?) Would you accept the program or reject it?
* [Bidding as for Q.6]
FILL IN ALL BLANKS:
ENTER: $ + $ MORE = $ /month
(BID5) (BID7)
-------
-9-
[Present Card G]
8. One last program. [Show WEST picture set] This row of photos
shows a scene from the western United States.
Now, consider a program
visibility by ten miles over
Visibility in Atlanta would
places in the country would
program cost your household
which would improve typical
the entire country, [show Map]
go to 20 miles, and all other
get similar improvements. jf -(-he
(an additional $10, for a total of $ (BID7 + 10) )
[OR]
<$13 a month>
would you accept the program or reject it?
Accept
[Check One]
Reject
What if it cost $ ^ /month (more, for a total of
$ [BID7 + *] ?) Would you accept the program or reject it?
* [Bidding as for Q.6]
FILL IN ALL BLANKS:
ENTER: $ + $ MORE = $ /month
(BID7) (BID8)
-------
-10-
10a. Who should pay the costs of pollution control?
[You may check more than one]
Ordinary Citizens
The Polluters
The Government
[Present Card H]
10b. For some years now, government and industry have been spending
money to control pollution and improve the environment. Which of
the following three statements best expresses your views about this?
[Check One]
Current levels of spending will eventually
balance environmental quality and economic
goals.
It is time to cut back on spending for
environmental purposes.
We need to spend more, to achieve the
kind of environment we want.
Now, a few more questions.
11. Do you own or rent the residence you live in?
[Check One]
Own (go to 12a)
Rent (go to 12c)
Other (go to 12d)
-------
-11-
12a. OWN: If, for some reason, you wanted to rent out your
residence, how much rent would you expect to receive? (or: what
would a residence like this bring on the rental market?)
/month
b. [IF DOES NOT KNOW] Perhaps it might be easier to think about
the sale price. If you needed to sell your residence within 2
months and the buyer would have to arrange his/her own financing,
how much do you think it would sell for?
$.
[sale price)
c. RENT: How much
(house,apartment)?
do
you pay per month to rent this
/month
d. OTHER: If you had to rent a house or an apartment like this
on the rental market, how much do you think you'd have to pay?
/month
13a Do you have any definite plans to move your residence in the
next five years?
Yes
No .
b [If a: Yes] when you move, do you expect to settle west
of the Mississippi River?
Yes
No
Don't know
c. Do you expect to retire somewhere near Atlanta?
Yes
Currently retired
No.
Don't know
(go to d)
d. [If c:No] Then, do you expect to retire:
(Check One)
Somewhere east of the Mississippi River
Somewhere west of the Mississippi River
Other
-------
-12-
14a Do you own any residential property(houses, apartments),
other than the place you are living in?
No
Yes [Continue]
b. Is this property located:
In or near Atlanta (Check All That
Elsewhere in the eastern U.S. Apply)
Other
c. How much do you receive in monthly rents from residential
property:
In or near Atlanta? $ /month
Elsewhere in the eastern U.S.? $ /month
15. [Show Card I] Please choose the best description of the view
you have from your residence, and give me the number.
Number from card.
SOCIODEMOGRAPHIC DATA
So that we can analyze the responses we get from different
people, we need to ask you a few questions about your household.
Your answers will be completely confidential.
16. Of the people who usually live in your household, how many
are children, 18 years or younger?
-------
-13-
17a. For those who are not children, please fill in the table.
[The following notes are for the interviewer's guidance]
#: Each person is assigned a #, 1,2,3, etc.. The head of the
household is always #1. Circle the # which represents the
Respondent.
Relationship to Head: Indicate the customary family
relationships (spouse, son, grandmother, etc.). For
non-family relationships, just write "friend".
Education: What is the highest grade or year in school completed?
NONE 0
ELEMENTARY 1 2 3 4 5 6 7 8
HIGH SCHOOL 9 10 11 12
COLLEGE 13 14 15 16
SOME GRADUATE SCHOOL...17 18
GRADUATE OR
PROFESSIONAL DEGREE 20
Is ...currently attending a School, College or University
FULL TIME?
Does ...usually work [or seek employment] outside the
household?
IF NO, go to next person.
IF YES, continue.
How many months did ...work in 1981?
How many hours/week did ...usually work in 1981?
[record either HOURLY, WEEKLY, OR MONTHLY WAGE]
SCHOOL:
WORK:
MONTHS:
HOURS:
WAGE:
17b. Do you have any of the following?
[Check those that apply]
Poor eyesight
Allergies (e.g., hay fever, asthma)
Any chronic respiratory ailment [e.g. T.B., emphysema, etc.)
-------
-14-
PERSON
1.
HEAD OF
HOUSE-
HOLD
2.
3.
4.
5.
6.
AGE
RELATION
TO HEAD
X X
X X
X
X X
X X
SEX
(M/F)
EDUCATION
IN
SCHOOL
(YES/NO)
WORK
1981
(YES/NO)
MONTHS
WORKED
1981
HOURS
WORKED
PER WEEK
1 981
HOURLY
WAGE
TORI
WEEKLY
WAGE
TORI
MONTHLY
WAGE
-------
-15-
18. [Race/ethnic group, of respondent. Interviewer Check One].
Asian
Black
Hispanic
White
Other
19. In your household, do you: [Check One]
a . share or pool your incomes, as a family or couple might do.
b. live alone, or keep your personal incomes separate, as
friends sharing a house/apartment might do.
20. [Present Card J] Please look at this card. Tell me which letter
best describes your [household if 19a; or personal if 19b] income
before taxes in 1981. Include income from all sources, including
work, investments, business profits, interest on savings, pen-
sions social security, support from relatives, and any other
benefits.
[Letter]
[Refused, or didn't know and refused
to guess].
2 1 . Was your personal income in 1981 [Check One]
about the same as other recent years?
much higher than in other recent years?
much lower than in other recent years?
22. Would you expect your income, corrected for inflation [Or
your purchasing power, Or your standard of living] in five years'
time to be:
about the same as in 1981?
much higher than in 1981?
much lower than in 1981?
-------
-16-
23. [Does your household if 19a; Do you if 19b]
[Check One]
manage to save or invest a little?
just get by on current income?
have to dip into savings or
investments just to make ends meet?
24. If you wanted to work a few more [or "a few" for non-income
earners] hours a week,
Do you think you could find work? Yes No
[If Yes] How much do you think you'd be paid? $ /HOUR
[Present Card K]
25. NET WORTH means the value of things you own (personal
property, automobiles, equity in a residence, investments, savings
etc.) MINUS the total amount you owe to others (loans, mortgages,
balance owing on credit cards and installment purchases, etc) .
Please look at the card and tell me which letter best describes
your [household's if 19a; personal if 19b] net worth at the end of
1981.
[Letter]
[Refused, or didn't know and refused
to guess] .
26. May 1 please have your name and phone number in case my
supervisor wishes to check that I completed this interview.
Thank you very much. You have been very helpful.
-------
INTERVIEWER EVALUATION
Record any comments which might help us understand the an-
swers given by the respondent, especially those who protest during
the bidding questions.
-------
APPENDIX B: SAMPLING RATIONALE AND PROCEDURES
To obtain contingent valuation responses, 792 households in the Eastern
United States were questioned about the value of preserving or improving visi-
bility in the United States. This survey represented the opinions of about
100 million people living in the Eastern U.S. It provided the basic information
for a monetary estimate of the value that people in the Eastern U.S. would place
on alternative degrees of visibility improvement in their area. Indirectly it
provided some clues about how much people in the West might value improved
visibility in the Eastern U.S.
In order to enable the 7 92 households to give us the information we sought
from them, it was essential that they be made representative of the population
from which they were drawn. Stratified-cluster random sampling was used. There
are several reasons for this approach. First of all there is a great deal of
diversity in annual average visibility in the area. (See Map A.) Also, there
is substantial social diversity among the eastern regions, and they may differ
from one another in important ways in their valuation of visibility. Economic
theory thells us that geographic and socio-economic differences are important
and should be included in the analysis. To make it highly likely that a simple
random sample would cover those categories would require a much larger sample
than is feasible within the project budget.
The creation of sampling sub-regions was desirable for policy purposes.
Pollution control is the means by which visibility can be altered in any region
by human choice. However, pollution levels differ substantially from one region
to the next. Consequently, any change in ambient air quality standards will
affect visibility in different regions differently. Regions that already meet
the standard will experience no change in visibility; regions the farthest from
-------
compliance will experience the greatest visibility improvement. A sample
design that does not permit the analysis of separate regions would not
answer the requirements of policy analysis.
To implement the sampling plan, six city areas in the Eastern U.S., in
addition to Chicago, were chosen to represent each level of average annual
visibility in geographically dispersed areas of the Eastern U.S. The cities
were Atlanta, Boston, Cincinnati, Miami, Mobile, and Washington, D.C. Selection
of city and rural areas outside the cities created sub-populations within the
Eastern U.S. The second major aspect of the sampling plan was to apply random
sampling within each urban and rural area. The urban sample in each city area
was drawn using 1970 census tract maps and census statistical tables. First,
all of the n census tracts in the urban portion of the metropolitan area
were assigned numbers one through n . Then twenty numbers between one and n
were drawn from a table of random numbers and matched with the corresponding
census tracts. Eight interviews were to be taken within each tract, in the
order drawn, until 120 interviews were obtained. (The extra tracts were drawn
in case eight interviews could not be obtained in some of the tracts. However,
the sampling order of the random draw had to be followed; no interviewer discre-
tion was allowed in tract choice.)
Random selection of household within each tract was achieved in a similar
way. Every block within each selected tract was assigned a number between one
and m , which was matched with the corresponding block number assigned by
Block Housing Statistics. A random number between one and m was chosen to
determine the block where interviewing started. Additional blocks were
determined by the going to the next higher numbered block, using the block
numbers given in Block Housing Statistics (returning to the lowest numbered
block if necessary).
-------
The interviewer's starting point on each block and the direction to proceed
around the block were uniformally specified in advance for all interviewers.
The procedure continued until eight interviews were obtained within a tract.
Interviews were conducted in two rural areas outside the metropolitan areas
of each city. Maps, interviewing routes and procedures for each area were
worked out between the field supervisors and the survey coordinator at the
University of Chicago.
Xerox copies of census tract maps and lists of tract orders were provided
to all interviewers, with starting blocks clearly indicated. Field supervisors
in each city worked closely with interviewers, and monitored their work. The
field supervisors all attended a training meeting in Chicago before field work
began, and remained in close contact with the U of C survey coordinator during
the entire survey period.
Of the 792 households from which guestionnaires were obtained, results
from 538 were used in the regression analysis for the visibility value function.
As indicated in Section 2.4, the major reason for not being able to use all the
guestionnaires was the refusal of some households to give income and wealth
information. Some guestionnaires were not used because respondents bid zero
for reasons other than how much visibility was worth to them (for example, they
said the pollutant rather than the respondent should be expected to pay) or in
a few cases unreasonably high bids were given.
-------
This folio explains the visual material used in the contingent
valuation survey under USEPA Cooperative Agreement #8077 68-01-0.
The folio contains exact copies of the photographs used. Identifi-
cation is given on the back of each photograph. The sketches of the
Photograph Display Board indicate how the photographs were set up
and shown to respondents.
-------
APPENDIX C : BACKGROUND PAPERS
ESTABLISHING AND VALUING THE EFFECTS OF IMPROVED
VISIBILITY IN EASTERN UNITED STATES
by
George Tolley
Alan Randall
Glenn Blomquist
Robert Fabian
Gideon Fishelson
Alan Frankel
John Hoehn
Ronald Krumm
Ed Mensah
Terry Smith
The University of Chicago
USEPA Grant #807768-01-0
PROJECT OFFICER: Dr. Alan Carlin
Office of Health and Ecological Effects
Office of Research and Development
U.S. Environmental Protection Agency
Washington, D.C. 20460
March 1984
for quotation. Empirical results subject to change
-------
APPENDIX C: BACKGROUND PAPERS
Contents
Page
A.l Theoretical Approach to Valuing Visibility A-l
A.2 Atmospheric Visibility and Contingent Valuation Exercises A-26
A.3 An Early Contingent Valuation Excercise A-31
A.4 Economics of Visibility - An Input Approach A-37
A. 5 On the Evaluation of the Social Benefits of Improving A-47
Visibility
A.6 Visibility and Its Evaluation A-54
A.7 Visibility and Outdoor Recreation Activities: A Research
Framework A-63
A.8 The Demand for Visibility Services A-68
A.9 The Effects of Visibility on Aviation in Chicago A-83
A. 10 View Primary Recreation, The Hancock Tower A-90
A.11 Visibility, Views and the Housing Market
A. 12
A.13
A.14
-------
Introduction to Appendix C
This appendix contains papers which represent the conceptual development
during the research effort. Numerous contributions to current economic theory
and empirical practice are found in these papers. They represent an exploration
of the fundamental issues involved in the visibility project and were necessary
in attaining the focus achieved in the final product.
-------
A-l
A-l THEORETICAL APPROACH TO VALUING VISIBILITY
General Framework:
Atmospheric visibility is most effectively conceptualized as a matrix
of services provided by atmospheric resources. In order to place the value
of atmospheric visibility in perspective , consider the following conceptual
model for valuation of atmospheric resources in a benefit-cost context.
In accordance with the potential Pareto-improvement criterion (the
generally accepted criterion for benefit-cost analysis--see, for example,
Mishan, 1976), an existing environmental resource is valued at the seller's
reservation price for a capital good. The capital value of a given en-
vironmental resource, for example, "atmospheric resources" (A) which pro-
duce a stream of visibility services, is the net present value to the seller
of the stream of services in each time period, S , where t = 0, 1, 2,
and the present time period is defined at t = 0. Thus,
(1)
00
v(st)
P.V. (A) = I -
t=0 (1+r)
where V(S ) = the net value, at time t, of the bundle of services produced
by A resources in time t, and r = the discount rate.
The bundle of services, S , provided by A resources is a vector of
n types of atmospheric services, , where i = 1, ... , n, including
those services associated with visibility. Thus,
n
(2)
v(st) -
-------
A-2
Now, let us consider, first, the production of atmospheric services,
and, then, the value of those services. The supply of an atmospheric
service, s_, (i,. ..,n), in any time period is a function, uniquely deter-
mined by geological, hydrological and ecological relationships, of the
attributes, (k = l,...,m), of the atmospheric resources. Thus, for
all services in i = 1,...,n, we have
Sn " Val am)
Man enters the production system as a modifier of atmospheric resource
attributes. He may do this directly, e.g., by generating residuals and
permitting their release as pollutants into the atmosphere. He may also
modify atmospheric resources as a side effect (expected or unexpected) of
some other decision pertaining to, e.g., the management of solid wastes or
water pollutants, or of those resources which influence the capacity of
the atmosphere to absorb wastes. For each kind of atmospheric resource
attribute in k = 1, . ..,m, we have
U r S UN
- hi
-------
A- 3
5
where n = a vector of "natural systems inputs", i.e., the inputs
which would determine atmospheric quality in the absence
of man's technology, and
xU = a vector of inputs controlled by man, e.g., anthropogenic
pollutants, and any efforts on the part of man to improve
the quality of atmospheric resources.
s u
Both n and x are subject to scarcity; and the attribute production
functions are determined by the laws which govern natural systems and by
man's technology. The production system is now complete. It is entirely
possible that the levels of production of some kinds of services, s_,, in-
fluence the level of some attributes, a , by a feedback mechanism wherein
s_^ alters the level of some man-controlled inputs in xU . For example,
the attempt to enjoy high levels of waste assimilation services involves
high level of pollution inputs, which may directly or indirectly modify
environment attributes.
Now, consider the value of atmospheric services. Each individual, j,
enjoys utility in each time period, t:
(5) U = f (Sg zY yZ )
]t ] t t t
g ....
where s = a vector of atmospheric services, which are directly en-
joyed for their amenity value, including those which con-
tribute to directly enjoyed atmospheric visibility,
y ...
z = a vector of goods and services for which atmospheric ser-
vices are inputs, such as outdoor recreation services, and
-------
A-4
yz = a vector of goods and services which are produced in pro-
cesses bearing no immediate relationship to environmental
services.
Each individual makes decisions in the initial time period, and subject to
his initial budget constraint, in order to maximize the present value of
expected lifetime utility.
By minimizing his expenditures, subject to the constraint that his
utility must always be equal to the utility he enjoys with the existing
level of atmospheric resources, his Hicksian income compensated demand
curves [see Hicks, Mishan, Currie, _et al.; Willig; and Randall and Stall]
for atmospheric services may be derived. From this, the Hicksian compensa-
ting measure of the value of the loss which the individual would incur in
time t, should the quality of atmospheric resources be degraded--or the
value of the gain the individual would enjoy in time t, should the quality
of atmospheric resources be improved--can be calculated. The total social
loss from a degradation of atmospheric resources—or the benefits from an
improvement in atmospheric resources—may be calculated by summing the
Hicksian compensating measures of welfare change across individuals and
across time periods.
To adapt this general model to the study of the economic value of
atmospheric visibility in the eastern United States, account must be taken
of several specific factors.
a) Due to the relatively rapid recovery, under favorable circumstances,
of atmospheric resources from assaults by pollutants (compared to,
-------
A-5
say, land and water resources, and complex ecosystems) intertemporal
relationships, while significant, may be less important than in
the cases of some other kinds of resources.
b) Due to the dominant west-to-east (or southwest-to-northeast) trans-
portation pattern of atmospheric pollutants, welfare impacts (i.e.
social costs or benefits) of visibility change in one part of the
study area are attributable to antropogenic pollutants generated in
other parts of the study area. Analysis by D. M. Rote of ANL
long range transport model incorporates these effects.
c) The Primary emphasis of the research on atmospheric visibility has
required that considerable subtlety and discernment be applied to the
task of differentiating between those welfare effects due to visi-
bility change and those due to other effects of atmospheric pollution
(e.g. plant, animal and human health effects). For example, outdoor
recreation activities may be adversely affected by visibility degra-
dation, but also by damage to plant communities and fish from acid
precipitation; the market value of residential property may be ad-
versely affected by poor visibility conditions, but also by exposure
to human health hazards and property damage.
It is also important to note that the same anthropogenic pollutants,
interacting with natural atmospheric conditions,
responsible for effects on visibility and, e.g., the health of plant
communities and human beings.
-------
A-6
d) While consistent with the conceptual framework developed here,
the research in this report concentrated upon empirical estimation
of the relationships expressed in equations (1), (2), (3), and (5),
that is, the relationships between changes in atmospheric resource
attributes (i.e., various relevant measures of ambient quality) and
the value of visibility services provided.
The estimation of the relationships expressed in equation (4)--
i.e., the relationships between natural atmospheric conditions,
anthropogenic emissions and ambient air quality--will not be a
primary focus of the research proposed herein. However, the re-
search is designed to be compatible with estimates of the (4)
relationships, which are provided by ANL. In this way, the re-
search makes a major contribution to the understanding of rela-
tionships between atmospheric emissions, ambient air quality and
the economic value of changes in atmospheric visibility in the
eastern United States.
e) The particular atmospheric visibility services which are
foci of the proposed research are: (1) Those which contribute to
the satisfactions enjoyed by owners and occupants of urban and
suburban residential property; (2) those which contribute to the
satisfactions of recreationists in urban, mountain, and coastal
-------
A-7
environments; and (3) those which influence the safety of users of
ground and air transportation services (given the hypothesis that
atmospheric visibility influences the flow of traffic and the
freguency of accidents).
Extended Framework
In this section we expand upon the conceptual framework
by further developing the relationships between atmospheric visibility ser-
vices and utility [equation (5)] and the value of service flows [equation
(2)].
There is now general agreement that the change in consumers' surplus
is the proper measure of the economic value of a change in the level of
provision of a good, service, or amenity [Currie, Murphy and Schmitz; Dwyer
et al.; Harberger; Hicks, 1940-41; Hicks, 1943; Hicks, 1945-46; Mishan,
1947-48; Mishan, 1976; Mishan, May 1976; Randall and Stoll; Willig] .
The conceptual framework presented below provides a general basis for
estimating changes in consumers' surplus resulting from changes in the
provision of goods, services and amenities--in this case, those associated
with atmospheric visibility--including the marketed and the non-marketed,
the divisible and the indivisible, and the exclusive and the non-exclusive
[Brookshire, Randall and Stoll] . Consider Figure 1. The origin is at^>
which represents the consumer's initial holdings of the atmospheric visibil
service in question, Q, and "income" (or, more precisely, the "all other
goods" numeraire). As one moves to the right on the horizontal axis, the
quantity of Q increases; as one moves to the left, Q decreases. As one
-------
A-8
moves upward, on the vertical axis, "income" decreases; as one moves down-
ward, "income" increases. The total value curve, or willingness to pay
curve, passes from the lower left quadrant through the origin and into the
upper right quadrant. For an increment in the service from to Q+, the
individual is willing to pay the amount Y^ - Y , which is a positive amount.
After having paid his willingness to pay (WTP) and receiving the increment
q+ - the individual is exactly as well off as he was at the origin.
For a decrement in the level of provision of the service to Q , the indi-
vidual is willing to pay the amount Y^ - Y and, hating paid that amount
and received the decrement, is exactly as well off as he was at the origin.
Observe that Y+ is greater than Y^. Thus, the individual's WTP for the
decrement is a negative number. In other words, the individual is willing
to accept (WTA) some positive amount of additional income, along with the
decrement in the level of provision of the service.
The total value curve measures the net change in consumer surplus
resulting from increments or decrements in the level of provision to the
individual of the service in question. if the service is unpriced, the
change in consumers' surplus is exactly equal to the value of the incre-
ment or decrement [Brookshire, Randall, and Stoll] .
This value model is applicable to goods and services which are un-
priced, divisible or indivisible in consumption, and lumpy in production
being available only in quantities Q , Q^, and Q+. If the good in question
was divisible in consumption, infinitesimally divisible in production, and
available in infinitely large, frictionless markets at a competitive price,
the total value curve could be replaced with the broken price line (which
-------
A-9
Figure 1. The Total Value Curve
-------
A-10
is tangent to the total value curve at the origin) . in such a case, the
absolute value of WTP for an increment would be exactly equal to the abso-
lute value of WTA for an equal sized decrement, and both are equal to P*AQ
(i.e., the unit price multiplied by the quantity change). Observe that, in
cases where the total value curve (rather than the price line) is relevant,
WTP for an increment in Q is smaller in absolute value than WTA for a similar
sized decrement. Theoretical analyses have developed formulae for the
empirical estimation of the difference in absolute value between WTP and WTA
in this circumstance [Randall and Stoll; Willig].
The above conceptual framework is entirely general, and develops the
relationships between consumer surplus, WTP (and WTA, the counterpart of WTP
in the case of decrements in the good), and market price. Where some de-
finable population, e.g., the residents of a given community or the users of
a given recreation site, experience the same increment or decrement in the
availability, the aggregate value of the change, in benefit-cost terms,
is equal to the sum of the individiual values [Bradford, Dwver et a 1.1.
The value of increments or decrements in atmospheric visibility ser-
vices (the v^ of equation 2) were estimated, using various techniques,
but always in a manner consistent with the above conceptual framework. In
those cases where competitive markets exist for atmospheric visibility
services, market observations were analyzed in order to permit esti-
mation of the value (i.e., price) of visibility services. Where at-
mospheric visibility services are not directly marketed, two general
classes of analytical techniques for value estimation are available.
-------
A-11
a) Hedonic methods utilize observations from markets in goods or ser-
vices which bear some relationship to visibility services (e.g. are jointly
consumed with visibility services, or are produced in processes which re-
quire visibility services as inputs) in order to estimate implicit prices
or values for visibility services. This class of techniques includes the
land value method of valuing environmental amenities [Abelson; Anderson and
Cracker; Brown and Pollakowski; Maler]; the hedonic and household production
function methods [Deyak and Smith; Muellbauer; Pollak and Wachter; Rosen],
which have been applied to valuation of a wide variety of non-market goods
including human health and safety; and the travel cost method which has been
widely applied in the economic valuation of outdoor recreation amenities
[Brown, Singh, add Castle; Cesario and Knetsch; Clawson and Knetsch; Gum
and Martin; Knetsch].
b) Contingent valuation (CV) methods approach the valuation of non-market
goods directly by creating hypothetical markets and treating the decisions
of respondents or experimental subjects using these hypothetical markets
as values which exist, contingent on the existence of hypothetical markets
[Brookshire, Ives and Schultze; Bishop and Heberlein; Brookshire, Randall
and Stoll; Davis; Hammack and Brown; Randall, Ives and Eastman; Randall
et al.; Smith].
-------
A-12
Overview
To estimate the change in aggregate consumer's surplus resulting
from changes in average or typical visibility situations were identified
that are affected by changes in the level of services rendered by visi-
bility. A major consideration in the research design was to include situ-
ations where visibility effects are likely to be most pronounced where
they are likely to have significant influence on benefits due to the num-
bers of people or the value of property affected. With situations identi-
fied, an appropriate valuation method was selected and the change in con-
sumer's surplus estimated. Table 1 presents the results of such an identi-
fication process for Chicago. Examining Table 1, the first column gives
a taxonomy of situations that are, to a greater or lesser extent, hypothe-
sized as being affected by the level of visibility. Columns adjacent to
the first in Table 1 match at least one valuation technique to each cate-
gory of identified situations. Wherever possible, more than one approach
is matched to a situation so that valuation results may be replicated and
compared. Both the taxonomy of situations and also the data required for
the valuation of effects are discussed.
Using the contingent method, visibility levels for a given situation
were described in both narrative and photographs. By carefully structured
questioning, an individual's valuation of a given increment of visibility
was then elicited. The method was contingent because valuations were con-
tingent upon an individual's behavior in a hypothetical choice situation.
The contingent method was administered directly to individuals. The
-------
Table 1. Situations Affected by Visibility and
Methods of Valuation for Chicago
SITUATI (
VALUATION METHOD
Cont i ngent ReveaIed
Hedonic Demand Cost of Inputs
Aesthetic or View Related
A, Urban Visibility Services
1, Res i dent i a I
a, Lakeshore residences
b. Non-Lakeshore city
c Metropol itan suburbs
2, Non-Residential
a Workplace
x
x
X
i. Loop area (First National x
Bid., Stan. Oil Bid., etc.)
ii. City, non-loop (Oakbrook) x
x
X
-------
Tab Ie 1, cont i nued
SITUATI ON
VALUATION METHOD
Cont i ngent ReveaIed
Hedonic Demand Cost of Inputs
b. Commuting and othen intra-
unban tnavel
Expressways (Kennedy, x
Eisenhower, etc.)
ii. Bridges (Chicago Skyway) x
c Recreation
View Primary
a Elancock Tower x (Consent) x
b Sears Tower x
i i . View Secondary
a. Spectator Activities x
b Participatory Activities x
i i i . Substitutes x
-------
Tab Ie 1, cont i nued
SITUATION
B Rural Visibility Services
1, Res i dent i a I
a Michigan City, Indiana
2. Recreation
b Indiana Dunes State Park
Non-View Related
A Effect on Traffic Flows
1. General Aviation
a DeI ays
b. Cancel I at ions
2, Commercial Aviation
a DeI ays
b Cancel I at ions
VALUATION METHOD
Cont i ngent ReveaIed
Hedonic Demand Cost of inputs
x
x
x
x
X
-------
Tab Ie 1, cont i nued
SITUATI (
VALUATION METHOD
Cont i ngent ReveaIed
Hedonic Demand Cost of Inputs
Safety Related
1 Air Traff i c
a Single plane accidents
b MuIti-plane accidents
c Near-misses
2 Ground Traffic
a Highway accidents and col I isons
III. Option and Existence Value of Visibility
A. National Landmarks
1, Wash i ngton Monument
2, Statue of Liberty
3, National Parks
-------
A-17
revealed behavior methods relied upon an individual's actual behavior
for evidence in valuation. Because actual behavior may be only indirectly
related to visibility, revealed behavior approaches confronted both conceptual
and statistical difficulties on application. Of the revealed behavior
methods, the hedonic technique values visibility as a characteristic
of property. Property values as well as supplementary information on
housing and view characteristics were required for valuation. The demand
method measured the effect of visibility on demand for acti-
vities such as outdoor recreation. To apply the demand method, only
secondary data on attendance was required in most cases considered below,
inally, the opportunity cost-of-inputs method was applied to situations
or events that occur only sporadically and thus did not generate suf-
ficient data for any of the other techniques.
Examining Table 1 once again, the broadest distinction of the types
of situations affected by visibility is between those situations in which
visibility affects aesthetic appreciation and those situations where the
effect is not directly aesthetic. The aesthetic or view-related effect was
further distinguished by demographic area: by urban and non-urban or rural
visibility services. Using the contingent valuation technique, both urban
and rural visibility services were valued directly by observing residents
in both urban and rural areas. In the Chicago area, urban visibility ser-
vices were valued directly. Three strata correspond to the three divisions
under residential urban visibility services: lakeshore residents, non-
lakeshore city residents, and residents of the metropolitan suburbs. The
approach had three purposes. First, using a set of photographs and
-------
A-18
the contingent technique, a valuation of visibility increments over the en-
tire urban area was elicited. This first valuation was for urban visi-
bility services as a whole. Second, the CV instrument elicited
information on housing and view characteristics. This information was
required for the hedonic approach to valuation. Third, the CV instrument
inquired about recreational activities. Such participation
data were essential to population estimates for the non-residential ef-
fects of urban visibility services and their aggregation.
The third major effect of visibility within the metropolitan area
is on urban recreation. Two types of affected recreation activities can
be distinguished. The first is recreation that focuses on the enjoyment
of specific views. The second is recreation in which a view and associated
visibility level are only secondary, used mainly as a background. Within
Chicago, the two major view primary sites are Hancock Tower Observatory and
the Sears Tower Skydeck Observatory. Each of these locations offers
views of Chicago at various levels of visibility to approximately one mil-
lion visitors a year. Hancock Tower cooperated with our demand approach
to valuation by sharing attendance records. Attendance records were analyzed
along with airport visibility and weather data to determine the effect
of visibility on visitations. Finally, a contingent valuation of visi-
bility was conducted at the Hancock Tower. To elicit a valuation
of increments or decrements of visibility at the Hancock Tower, a special
CV instrument was constructed for those who visit the Tower.
Valuation of the effect of visibility on the enjoyment of spectator
sports was made by the demand method. Fist, attendance data was regressed
-------
A-19
on weather, visbility, and other secondary data to determine the effect
of visibility. The effect of visibility was shown to be significant in
preliminary analysis and a more complete demand model was specified
for the valuation of its effect. This more complete demand model included
equations for local substitutes to outdoor recreation, such as museum
and aquarium attendance.
The non-aesthetic effect of visibility on general aviation and highway
accidents were also examined for the Chicago area. These are discussed
in the chapter on secondary data analysis.
To extend the valuation of visibility beyond the Chicago region and
thus permit a benefit estimate for the eastern United States as a whole
a valuation of visibility services were made for six other population
areas. The same basic approach used for the Chicago area also was used
for these six additional population areas. That is, both contingent and
revealed behavior methods were applied to value the effect of visibility in
each of the situations outlined in Table 2. The six additional population
areas chosen for investigation were selected on the basis of experience
regarding the prevailing visibility conditions over different zones
within the eastern United States, and the requirements of a systematic
aggregation procedure.
Selection of the areas entailed references to median
yearly visibility . Over the eastern United States
there exist several distinct visibility zones. Except for the Mississippi
delta area and the Ohio River basin, median visibility from the Appalachian
-------
A-20
Mountains to the plains states is approximated by that of Chicago.
By sampling from urban and rural areas near Cincinnati, for example,
information was obtained regarding the value of visibility for an in-
land area of generally poor visibility. By sampling from urban and rural
areas in and near Boston, information was obtained regarding the value of
visibility for a coastal area of generally good visibility. A sample
from the area of Atlanta provided information regarding the value of visi-
bility by residents of a median range visibility zone for an inland
city of the south.
-------
A-21
Benefits as Measured in Housing Markets
Housing markets can yield useful information about the
demand for goods such as clean air and visibility which are not traded
in their own explicit markets. Analysis of markets, whether they be explicit
or implicit, has great appeal relative to non-market benefit measures because
it is based on observable behavior where preferences are revealed through
some monetary expenditure rather than through an imaginary response to a
hypothetical situation. Nonetheless, since the Ridker and Henning (1967)
and Anderson and Cracker (1971) studies of residential property values and
air pollution doubt has arisen as to exactly what information is contained
in a regression of property values on characteristics of housing, i.e., a
hedonic regression. Maler (1977) points out the value of any estimates
based on analysis of property values is limited by potential malfunctions
in the housing market which might be caused by lack of information about
the costs of air pollution, in particular, or all factors which cause the
market to be in a state which differs from equilibrium attained under ideal
conditions of zero information, transactions and adjustment costs, in
general. Such criticism depicts the trade-off inherent in the alternative
methods of benefit estimation, market and non-market, and suggests the im-
portance of using them as complementary inputs into benefit estimation.
While criticism of housing market studies remains, considerable pro-
gress has been made. Due largely to contributions by Freeman (1971) and
Rosen (1974), it is clear that a hedonic regression does not yield a use-
ful measure of benefits—at least directly. Rosen's conceptual framework
for analysis of implicit markets shows that a hedonic regression is a mar-
ket clearing function yielding only hedonic prices which then must be used
along with other determinants of demand to estimate the demand for traits
-------
A-22
implicitly traded in the housing market.
Using Rosen's approach housing is viewed as a package of traits made
up of both structural characteristics and neighborhood amenities. House-
holds respond to the configuration of traits in addition to the traits them-
selves since the traits are not easily repackaged. Since households de-
mand housing, not land, they consider various structures in various neigh-
bodhoods and choose housing packages which must suit them. As such, house-
hold utility depends on housing, market goods and tasts or:
(1) U = U (Z, X: T)
where U is household utility, Z is a vector of housing traits, X is a vec-
tor of market goods and T is a vector of taste variables. Household utility
maximization is constrained by the available money income:
(2) I = X + P(Z: I,U,T)
where I is household money income, X is the numeraire, and P(Z: I,U,T) is
the household's total valuation of housing traits which depends on the
housing traits, income, utility level and tasts, respectively. The valua-
tion function gives an indifference map depicting the willingness of the
household to trade off units of market goods, X, for incremental additions
of any housing trait, Z, given income, utility and tastes. As Rosen shows
the valuation function has the properties that it is increasing at a de-
2 2
creasing rate with trait consumption, i.e., 3P/3Z > 0 and 3 P/3Z < '0, and
that the ratio of marginal valuations of traits equals the ratio of marginal
utilities of traits for each pair of traits, i.e., P./P. = U./U. where P,
1 2 1 3 i
is the marginal valuation of trait i and IL is the marginal utility of trait
i, etc.
-------
A-23
The household faces a market equilibrium price function, P, which
indicates the amount of market goods which much be paid for additional
housing traits. If consumers have approximately zero market weights and
the market clearing price function is exogenous to the household this price
function for packages of housing traits is:
(3) P = P(Z)
where P is the price of the factor of traits, Z. The partial derivative of
the market price function with respect to a trait, P^, gives the equili-
brium marginal price of which is often called the hedonic or implicit
price.
Given that households maximize utility in a way similar to that when
they face a linear budget constraint, the first order conditions yield de-
mand function for housing site traits:
(4) zj = Z*(Plt . .., P±, ¦ ¦¦, Pn, I, T)
where the quantity demanded of trait i depends on its own marginal price,
p , the marginal prices of complementary and substitute traits, P.for J = 1,
i ]
... , n and J 7s i, household income and tastes.
To estimate the demand for visibility, or clean air, we first estimate
the price of clean air. The price is implicit in the hedonic regression in
that is is the partial derivative of housing price with respect to clean
air. If the true functional form of the hedonic regression is nonlinear,
then the marginal price of clean air will vary across sites. Second, we
use price of clean air along with the prices of complements and substitutes
income and taste variables as well as whatever else is necessary to identify
demand to estimate the demand for clean air in the usual manner.
-------
A-24
Recent empirical studies demonstrate that the theoretically-preferred
approach is feasible and that it does yield benefit estimates which differ
from those based only on the hedonic regression, Harrison and Rubinfeld
(1978), Nelson (1978), Brookshire et. al. (1979), and Bender et. al.
(forthcoming) all estimate the demand for clean air applying Rosen's model.
Linneman (1977), Blomquist and Worley (1978) and Witte et. al. (1979) es-
timate the demands for housing traits other than clean air. A pattern
which emerges is that the estimates from a hedonic - demand, i.e., two-
step, approach differs from the simple hedonic estimates. Harrison and
Rubinfeld find that the simple linear hedonic overestimates the benefits of
cleaner air by approximately 42% while Brookshire et. al. find the linear
hedonic overestimates the benefits by approximately 1594. Bender et. al.
also find that linear hedonic is quite misleading, but, in contrast, it
underestimates the benefits by approximately 60%. Blomquist and Worley
find that the linear hedonic overestimates benefits for some housing traits
and underestimates benefits for others. While each of the four studies in-
dicates the superiority of a Rosen approach, the last two emphasize the im-
portance of a systematic search for the best functional form of the hedonic
equation, e.g., using a Box-Cox maximum likelihood procedure for searching
transformations of variables in the hedonic equation. These recent contri-
butions were carefully considered in our estimation of the demand for
visibility.
Our estimates of benefits of greater visibility more fully exploit
the gains of the Rosen procedure by paying particular attention to the es-
timation of total social benefits from the demand equations. Previous bene-
fit estimates have been made by simply multiplying the benefit for the
typical household times the number of households benefiting from the im-
provement. This estimation is appropriate for marginal or nonmarginal changes
-------
A-25
for the typical households. However, this does not yield true benefits
for all if those consuming some amount other than the average (typical)
amount of clean air (or any other trait) do not have demands symetrically
distributed about the demand for the typical household. For example,
those with higher incomes will value the cleaner air more than those with
average income and those with lower incomes will value the cleaner air
less than those with average incomes. The values of higher income house-
holds are unbounded, but those of lower income households are bounded be-
low by zero. In this case, simple aggregation can lead to an overestimate
of total benefits. Harrison and Rubinfeld do consider three income sub-
groups and find that indeed the total benefits are less than those estimated
by simple aggregation based on average income. We used distribu-
tions of demand shifters, such as income, representative of the eastern
portion of the United States to aggregate household benefits. This not on-
ly includes the valuations of these households not observed at the margin
consuming the average amount of clean air, but adjusts for any differenes
between particular areas studies and the entire region.
-------
A-26
A. 2 ATMOSPHERIC VISIBILITY AND CONTINGENT VALUATION EXERCISES
A decade has passed since the initiation of the research which provided
the data base for the first contingent valuation study of aesthetic aspects
of air quality to gain respectability among economists (Randall, Ives and
Eastman) . In that time, the theoretical basis of contingent valuation has
been clarified (see Brookshire, Randall and Stoll for an exposition of
current theory, and Randall, 1980 manuscript, for the theoretical relation-
ship between contingent valuation total cost, property value, markets
in substitutes, and hedonic methods of valuation); contingent valuation for-
mats have been classified, codified, and accepted for use in benefit cost
analysis of federal water projects (U.S. Water Resources Council); and a
growing number of studies applying various contingent valuation formats to
a wide variety of nonmarketed goods have been completed and published.
Contingent valuation (CV) methods have always encountered some skep-
ticism from economists, since the basic data used are not generated by
actual transactions in near-perfect markets. Nevertheless, opposition to
the use of such techniques--or, perhaps, to the attribution of respectability
to them--has noticeably softened in recent years (see, e.g., Freeman).
Skepticism seems to have been undermined by several developments: the
above-mentioned work in developing the theoretical relationship between
consumers' surplus concepts, non-exclusive and nonrival goods, and contingent
-------
A-27
valuation methods; the fairly precise replication of earlier CV results
in later exercises (Rowe, d'Arge and Brookshire); and the fairly general
finding of similar results when CV methods are compared with travel cost
(Knetsch and Davis) and property values (Brookshire, d'Arge, Schules and
Thayer) methods.
Nevertheless, some doubts remain. (1) The generally accepted theory
of "public goods" (Samuelson) indicates scope for strategic behavior, in
which individuals avoid revealing their true valuations of such goods in
order to maximize their surplus, i.e., the difference between the value they
enjoy and the contribution they make. For some economists, the scope for
such behavior is prima facie evidence of its prevalence; hence, a general
refusal to take seriously the results of any CV method which fails to elimi-
nate that scope. The search for "incentive compatible demand-revealing
mechanisms" is in part a response to the "scope proves prevalence" argu-
ment. For others, the prevalence of such behavior is much more problematical:
while no country seems to rely on voluntary taxation, many "public goods"
are, in fact, voluntarily provided in substantial (but not necessarily
efficient)quantities. Smith assembles impressive experimental evidence that,
at least in the kinds of circumstances he and others he cites have studied,
strategic behavior is simply not a significant influence on aggregate valuations.
(2) In an interesting recent experiment, Bishop and Heberlein created
an experimental market in which they actually purchased goose hunting per-
mits from permittees, effectively establishing in real transactions the WTA of
hunters to forego the hunting season. In a mail survey conducted at about
the same time, WTP for hunting permits was established via single (i.e. non-
iterative) questions asking respondents to nominate a dollar amount which
-------
A-2 8
represents their maximum WTP. It turned out that WTA established in actual
transactions was about three times WTP generated in the survey, a difference
far greater than can be explained by income effects (Randall and Stoll,
1980a and b) . There are good reasons to suspect the Bishop-
Heberlein WTA experiment of upward bias, while their WTP survey used a format
which I consider inferior to the iterative bidding routine (Randall, 1980
manuscripts). Nevertheless, the various possible biases are probably not suffi-
cient to account for all of the observed differences. Tentatively, it can be
concluded that WTP surveys such as that conducted by Bishop and Heberlein
may typically generate understimates of the "true" value of the good con-
cerned. The temptation to overstate the WTP knowing that one is unlikely
to be forced to actually pay the stated amount (the "strategic bias" most
commonly attributed by economists to this kind of CV exercise) seems to be
more than counterbalanced by a tendency to respond ultra-conservatively to
the suggestion that one may be expected to pay for goods which are customarily
non-marketed (or to pay substantially more for goods which are customarily
underpriced by public institutions) . The conclusions stated immediately
above are tentative; a firmer conclusion is that the Bishop-Heberlein
experiment raises, in a dramatic way, some serious questions about the
quality of data generated in direct question CV exercises.
(3) Those researchers who have attempted to estimate statistical re-
lationships which use various economic, social and demographic variables
to explain the individual WTP bids generated in CV exercises have typically
been disappointed by the results (Cicchetti and Smith; Eastman, Hoffer and
Randall; Brookshire, d'Arge, Schulze and Thayer). The recent work by the
University of Chicago and the University of Wyoming teams in this and a
closely related study has encountered similar frustrations.
-------
A-2 9
While there is abundant and convincing evidence that individual WTP
bids are not merely random numbers, researchers have not been notably suc-
cessful in finding relationships between individual bids and variables de-
scribing the individual's economic, social and demographic condition,
2
In estimated equations, the adjusted R" is often low and few
variables are related to individual bid in a statistically significant way.
Sometimes, even the relationship between individual bid and individual in-
come is not significant. These kinds of results are unsettling to those who
believe that, if individual bids are in fact "good" economic data, they
should be related in systematic ways to the kinds of variables are related
to individual bid in a statistically significant way. Sometimes, even the
relationship between individual bid and individual income is not significant.
These kinds of results are unsettling to those who believe that, if indi-
vidual bids are in fact "good" economic data, they should be related in
systematic ways to the kinds of variables which often successfully explain
demand and/or value data for marketed goods.
This issue has several vantage points.
(a) Perhaps it is unreasonable to expect to be able to obtain strong
statistical relationships, using individual observations obtained from small
samples. After all, most demand studies use observations of broad aggregates
(time series of aggregate sales and/or cross-sections of total sales by
state, SMSA, etc.). Surely, the explanation of individual variables is
a task of quite a different order.
It has been observed that demand analyses using individual data gen-
erated from panel studies have generally yielded more robust statistical
relationships than have WTP exercises. But, these studies typically
-------
A-30
use much larger panels than most WTP survey samples, and (2) they typically
deal with fairly broad categories of regularly purchased foods (e.g. "food"
or "meat") whereas WTP studies often deal with highly specific goods
(atmospheric visibility at some specific place, elk hunting in a particular
kind of terrain in a given state or sub-state region).
Brookshire, Randall and Stoll report obtaining considerably more
2
robust equations--not merely higher R , but also highly significant income
relationships—when they grouped their sample of 58 respondents into
4 classes, according to household income, prior to the analysis. This
procedure suppresses within-group variation (presumably diminishing the in-
fluence of a few "extreme" observations in a small sample). Statistically, the
apparently improved estimates and lower mean square error were obtained at
the cost of higher principal diagonal (X'X) \ Thus, their procedure may
not necessarily be viewed as attractive
(b) Perhaps WTP vids, viewed as cardinal indicators of dollar valuations,
are not especially reliable. Different individuals probably perceive
the offered good (e.g., a given increment in atmospheric visibility)
differently. On this front, progress has been made (as Freeman acknowledged)
via the use of standardized photographs and devices to improve uniformity
of perception. Nevertheless, problems remain. In the case of atmospheric
visibility, no amount of effort in standardizing the verbal and visual
information provided to respondents can overcome different perceptions
due to individual differences in visual acuity.
-------
A-31
A. 3 AN EARLY CONTINGENT VALUATION EXERCISE
1. Pretest: Chicago Residents
In order to pretest the basic instrument for subsequent contingent
valuation exercises and to explicitly field test certain innovations in
C.V. instrument design, a C.V. exercise was conducted in Chicago and sur-
burbs. Sixty-eight households participated. After rejecting 15 observa-
tions (apparent enumerator bias), 2 (outliers) and 8 (self-identified pro-
test bids) all subsequent analyses were based on 43 observations.
The basic instrument tested included the following elements:
_ questions designed to test the efficacy of color photographs in
in representing visibility levels.
_ alternative methods of defining and representing visibility levels.
_ a listing of activities in which the household participates.
_ questions exploring whether visibility conditions influence choice
of activities and, if so, in what ways.
_ questions to determine whether the household owned certain equip-
ment used in producing activities for which visibility is an input.
_ WTP questions
_ follow-up questions to identify protest bidders and obtain partici-
pant's evaluation of the C.V. exercise.
_ home ownership v s. rental.
_ view quality at the home.
-------
A-3 2
-expected period of residence in Chicago SMSA(i.e., short-term,
. . . , through retirement).
-demographic information
-questions to probe the notions of life cycle consumption, per-
manent income, and marginal wage-cost of leisure-time.
All of these elements were serious candidates for inclusion in sub-
sequent C.V. work.,
Four kinds of innovations in C.V. instrument design were explicitly
tested:
a). WTP Instrument
Earlier C.V. work under this project and published
research suggested that the iterative bidding format is more effective
than single question formats which ask the participant to simply state
his/her WTP or to select from an array of numbers that which best repre-
sents WTP.
Recent work at Resources for the Future (Mitchell and Carson, draft
report) used a payment card, on which typical household annual costs--$ in
taxes and higher prices -- for various public programs were stated. Parti-
cipants were asked to examine the data provided and then state their WTP for
improvements in water quality. Mitchell (personal communication, and draft
report) reports that he considers the payment card device sucessful.
For the pretest, we developed a "modified payment card and rebid" format.
The payment card was modified to include typical expenditures for both public
programs and private goods. About ten minutes after the payment card was used
-------
A-33
to obtain WTP, the participant was asked "if the program to improve visi-
bility actually cost (stated WTP plus $25), would you accept or reject the
program?" This question was re-iterated with sucessively higher cost amounts
until a "reject" response was given,
The two WTP instruments tested were:
-iterative bidding ($/month)
-modified payment card and re-bid ($ annually).
On an annual basis, the predicted household bid was $109 higher with
the "modified payment card and re-bid" device than with the iterative monthly
bid (Table 1, model 1). Only about $20 of the difference was attributable
to the re-bid. It was notable that "zero" bids were much less frequent with
the "modified payment card and re-bid" device - 7% of all bids as opposed to
39 percent with the iterative bid (Table 2). This explains much of the dif-
ference in predicted household bids.
b) . Definition of Visibility Levels
Previous work has used color photographs depicting various visibility
levels, and defined visibility programs as improving typical visibility from,
e.g., the level shown in photo set D to, e.g., the level shown in photo set A.
The notion of typical visibility is easy to communicate, but may be an overly
simplistic specification of visibility.
Within any year, emissions and background visibility exhibit considerable
day-to-day and week-to-week variability. Thus, the relative freguency of good,
moderate and poor visibility days may be a more realistic way to specify visi-
bility conditions. A program to improve visibility would increase the relative
frequency of good visibility days while reducing that of poor days.
The worst visibility days tend to come clustered together, as ambient pol-
-------
A-3 4
lutants accumulate during periods of air stagnation. Conceptualized in these
terms, a program to improve visibility would reduce the length of the longest
run of consecutive poor visibility days in a typical year.
The pretest was designed to examane the effectiveness of these alternative
ways of communicating visibility conditions. Three specifications of visi-
bility improving programs were used:
-typical visibility would be improved from level B (about 12 miles) to
level C (about 30 miles): VISTYP.
-the frequency of various visibility levels would change from 30 percent
A (about 4 miles, 40 percent B and 30 percent C to 10 percent A, 30 percent
B and 60 percent C: VISFREQ.
-the length of the longest run of consecutive days like A in a typical
year would be reduced from 12 days to 4 days: VISRUN.
The predicted annual household WTP was lower with VISFREQ and VISRUN than
with VISTYP, but the differences were not statistically significant. VISRUN
generated a greater proportion of zero bids than VISTYP.
These findings suggest that, while all three visibility specifications
seemed to communicate effectively, VISFREQ and VISRUN offered little advan-
tage over VISTYP. Since VISTYP was more readily related to existing data
series on observed visibility, VISTYP was used in subsequent C.V. work.
c) . Income Concepts
It is expected on conceptual grounds that WTP bears a positive and signifi-
-------
A-3 5
cant relationship to household income. This expectation has been borne out
in previous published reports, although some small-sample studies have re-
ported insignificant income coefficients.
In this pretest, we took the opportunity to explore ways to improve the
specification of income concepts, as follows:
-the notion of standard of living, SOL, which adjusts household income
for household size to permit comparability of standard of living across
households of varying sizes (Lazear and Michael, American Economic
Review. 1980)
-permanent income notions, which were implemented by identifying those
households which had recently experienced significant changes in in-
come level, and those which expected to experience such changes within
the next five years.
-the notion that for some life-cycle stages annual consumption is more
representative of standard of living than annual income. For example,
some households of retired persons may consistently dissave or disinvest
in order to maintain current consumption.
-the marginal wage-cost of leisure-time, which is an important vari-
able when the demand for visibility is modeled in a household pro-
duction function framework.
No difficulties were encountered in obtaining the necessary data to
specify these various concepts. SOL proved an effective specification of
household Income (Table 1). Preliminary analyses (not presented) suggested
that permanent income concepts are significant with a larger sample of
households. The pretest sample included very few cases of dissaving, thus
-------
A- 3 6
providing no opportunity to examine the usefulness of this concept in
statistical estimation of bid equations.
d). Activities
The household production function framework conceptualizes visibility
as a non-rival input in the production of activities which provide utility-
gneerating characteristics. To implement that framework, it is necessary to
identify:
-the activities which households produce,
-the role of visibility in the production of those activities, and
-the purchased inputs, e.g., equipment, which are used along with
visibility in activity production: ACTEQ.
No difficulties were encountered in obtaining data on activities pro-
duced and ACTEQ. We were less successful in obtaining data to help specify
the role of visibility in activity production. Enumerators and participants
reported that section of the instrument was tedious. ACTEQ is an important
variable in WTP equations.
Pretest Result
Predicted annual household WTP for visibility improvements in the
Chicago region ranged from $125 (with MIB, VISFREO instrument) to $325
(with a AMPCR, VISTYP instrument).
-------
A-37
A. 4 ECONOMICS OF VISIBILITY - AN INPUT APPROACH
Several recent studies have dealt with both the theory and empirical
results of the issue of the value of visibility. Particularly notewor-
thy are Brookshire et al [1979] and the references cited there, and
Rowe et al [1980] and the references cited there. Indeed, Brookshire
et al contains a solid theoretical basis for valuing visibility using
the concept of the willingness to pay approach. In this section
we first discuss the consumer surplus-equivalent variation and compen-
sating variation issues. We then go on to critically evaluate the wil-
lingness to pay approach, arguing that it results in values of both vi-
sibility and vistas, since they are used simultaneously as inputs in the
production of consumerable service.
The Model
Let's assume the existence of a vista, located at a particular site
in the city. It can be located either offshore on the lake, or be the
lake itself. We define visibility as the possibility of being able to
see this site. We define a product, immediately consumed by the viewer,
as a function of the site, the conditions which allow it to be viewed,
and personnal inputs. Hence,
- f(V »»«• ™
where is the quantity of viewing services obtained per unit of time
g
at location 1, hour h and time t, when viewing site S^ . j stands for
site j ad includes its particular characteristics such as its height,
shape, and colors. are the viewing conditions at location 1, hour
h and time t. Note that 1 embodies the height of the observation point,
-------
A-3 8
distance from the site, direction to the site and other characteristics
one of which might be the existence of buildings located between the
viewer and site j which, by obstructing the view, pushes W^ht zero"
The traditional assumptions,
f(0, PI) - fCS, 0, ?I) - f(S, 0) - 0
*1 > °» ^11 < ^2 * f22< 0
hold for this production function. As already noted, v^htj "'~S consumec^
and produced simultaneously (the only way to transfer it from one time
to another is by using the storage device known as memory which often
has limited capacity). If stored, the quantity of services retrieved
from storage (memory) declines by a rate of s per unit of time. Thus,
if retrieved at t, the maximum of services retrieved are given by the
equation:
Uato e
Furthermore, discounting future utility by a rate p, the present value
o
f producing and inventorying visibility services of quantity n j-s
where >0, 0^ < 0.
Co
The above discussion suggests that the particular nature of the
product "viewing services" is of the form of a durable with a relativly
long life span (as, for example, "I visited the Grand Canyon only once,
but I still remember 'every' detail"), although some might depreciate
1
rapidly. Also, there is still the need for proof (although not by ec-
1 .
This depreciation is frequently supplemented by taking pictures of a
particular site or scene. The "quality" of the picture, as does the
quantity of viewing services, depends upon the conditions of visibili-
W , , ,
ty, 1th (Another supplement is picture taking by a different individ-
ual, however, this won't be discussed here).
-------
A-3 9
onomists) that affects the durability of the product, i.e.
« - s(Sr ?t>
and again,
?I) " s
-------
A-4 0
Vistas are consumerable goods. We also assume that they are nor-
mal goods. Thus, if visibility conditions are a non-inferior input,
their derived demand curve is downward sloping (demand for an input, i.e.
their marginal value product). We distinguish between two types of de-
mand curves - both extracted from consumer behavior. One is the regular
Marshallian demand curve, along which full income is kept constant but
utility is allowed to vary; and the Hicksian income compensated demand
curve along which full income varies but utility is held constant. Us-
ually, this distinction is made for a good that is explicit in the utili-
ty function. We argue legitimacy for the case of visibility given that
the producer is the consumer, i.e. the simultaneity of activities and
identity of quantities both produced and consumed.
We apply similar reasoning in the case of the quantity of visibility
services, W, and the price (implicit) of visibility services, P . Accor-
dingly, in Figure 1, we have drawn three demand curves (following Willig
[1976]): AA is the Marshallian curve, BB is the income compensated demand
curve at utility level UO , and CC is simply BB for a different utility
1 10
level, U , such that U > U (see also Appendix A). Let M denote money
0 1.
income. Then in Figure 1, the area P P ac is the conventional measure
0 1 ...
of consumer surplus, A; P P be measures the compensation variation, C,
for U(P^ , M^) ; and, P^P^ad measures the equivalent variation, E, for
10 ¦
U(P , M ). Again following Willig, we assume W to be a non-inferior pur-
> >
chased input, such that the inequality, C ¦ A * E, holds. Hence, if a
market for W existed, and prices varied between and changes in con-
sumer surplus can be calculated. The more pertinent issue, however, is
how to handle non-market inputs. In addition to being a public good, the
quantity of viewing services, not price, is fixed exogenously for a given pro-
-------
A-41
Figure 1
-------
A-4 2
ducer. Furthermore, these quantities may be noncontinuous. In the
following section the traditional consumer surplus equivalent variation
and compensation variation concepts are applied to exogenous changes of
the quantity. If one could find the price (shadow price) the consumer
would be willing to pay per unit of visibility directly (whether by
questionnaire or by market observations), then the consumer surplus
could be approximated. However, this approach is usually not feasible
and one has to resort to other methods. (In the last section, we dis-
cuss, with some skepticism, the success of the presumably correct wil-
lingness to pay method).
BB in Figure 2 is a derived demand curve. When the quantity of visibil
services, given free of charge, increases from to W, the area under
0 111..
the curve increases by W a d w , which is the measure of the equiva-
lent variation, E, at the utility level represnted by BB,
It is easy to show that the area under the Marshallian demand curve be-
For BB parallel to CC, and for AA, BB and CC linear, the convention-
al consumer surplus is the average of the above defined compensating
and equivalent variations.
Another interesting comparison is between the following pairs:
tion variation for the CC curve, such that,
new1, H°), i.e. new1, M°) - UCW°, M° +¦ C)
P°aW
P°« c1?1 and
-------
A-4 3
pVc1?1 and AVw1.
1 1
The paired relations have a common triangular shape (the first is fa d ).
Thus, the difference (using the BB income compensated curve) is OP^a^W^ minus
2 11.. . .0011
OP d W , which m conventional demand terms is P Q - P Q . This difference
depends upon the demand elasticity:
P°Q° - 0 as n • 1.
Hence, the approximation of consumer surplus by the ares under the income
compensated demand curve, BB, better approximates the equivalent variation
measure of consumer surplus the closer is its elasticity to 1. The CC
curve is of about the same elasticity as the BB curve. However, for normal
goods the Marshallian curve, AA is definitly more elastic. Thus, the fow-
lowing cases are noted; the difference for the Marshallian curve is the
same or lower when the elasticity of BB and CC is less than unity while it
is higher when the elasticity is above unity. If we assume that the pol-
icy maker is interested in the welfare implications of changing the quan-
tity of visibility services (e.g. by improving air quality), he may regard
the willingness to pay, defined by the Marshallian consumer surplus, as an
approximation to true consumer surplus (compensating or equivalent varia-
tion) .
The Demand for Visibility Services
If W is determined exogenously then its marginal product times the
marginal utility of the vista's services (MP x MU) is its shadow price.
If W is endogenous, its quantity is determined by equating its marginal
costs with the product MP x MU, (MUP).
As conventially noted, at equilibrium along the demand for W, the
-------
A-4 4
consumer surplus is the rent to the fixed factor - the existing site j .
For a given demand for viewing conditions, the lower the marginal cost of
visibility services, the more viewing conditions are purchased (e.g. tra-
vel until you find the "right" angle to view the rock). The rock's rent,
then, is also larger. Hence the point of maximum willingness to pay for
visibility, will be determined by the specific site. The maximum sum
that a consumer is willing to pay for a particular site is the consumer
surplus. The maximum amount the consumer is willing to pay for an addi-
tional unit of viewing conditions, W, is its marginal utility value,
.... ... 01.
If visibility conditions improve from W to W ma given site, the
and declines by W B BW when conditions are worsened. The size of area
OABW^ is unknown. If one suggests an improvement in visibility from
to w\ then the amount the consumer is willing to pay for the improved
111 02.
visibility is OAB W , M ; if a change from W to W is suggested, the value
2221222113 ¦¦ ¦¦
is OAB W,M.M - M = WBBW = M . The willingness to pay for visi-
0 3
bility conditions at W is approximated by M /2.
-------
A-45
Conclusions
The visibility valuations found in previous studies are biased upward
with respect to the marginal value product since they are totals and em-
body the rents for the various sites that the interviewee is viewing.
The experiment that we suggest would subtract out these rents. The willingness
to pay experiments, themselves, would not change except that each time an
initial will be chosen explicitly. Willingness to pay is indicated
for different changes from the initial W^. In this manner, the proper
3 3 0
M /2 can be calculated. We expect that M /2 will decline as W is
increased for a given site.
In addition, the difference between valuations for increasing and de-
creasing W ought to diverge further as the change between visibility levels
becomes larger. Large changes, however, might be necessary if the demand
is relatively inelastic. Since this is not apriori known, a conclusion of
no value might be reached although the consumer's surplus is large (re-
call the discussion on the relation between the "true" consumer surplus
and the one discussed in the previous section).
-------
A4 6
APPENDIX A
The consumer surplus function is the income compensation function
denoted by M(W|W^, M^) . The function denotes the least income required
by the consumer when no more than W units of visibility are available,
10
while he is (promised) to enjoy the same utility level as at W , M .
Hence,
u
-------
A-47
A. 5 ON THE EVALUATION OF THE SOCIAL BENEFITS
FROM IMPROVING VISIBILITY
The following paragraphs contain several thoughts on the evaluation
of the social benefits from improving visibility. Information on the re-
action of the public to improved visibility came in two ways. One was via
personal interviews out of which the willingness to pay for improvement were
found. The second was the result of analyzing aggregate behavior and parti-
cipation in specific activities (secondary data).
Analysis of willingness to pay data explains differences in
the magnitudes of bids (given the same "objective" improvement in visibility)
submitted by different people. The explanatory variables are thus specific
to the individual's socio-economic characteristics. Actually in order to
find the total value of visibility (improvements) to the population of a cer-
tain geographic area the product of the mean bid by the population (or if the
bid is per household by the number of households) is a good approximation for
it. The parameters of the bid function are needed for a more accurate evalua-
tion, given that either the distribution of the relevant population by the
variables that affect the magnitude of the bid is non-symmetric or that the
effects of these variables on the magnitude of the bid are non-linear. The
two issues of non-symmetric distribution and non-linear effects required
ground preparation of sampling a sufficiently large number of observations,
a sufficiently wide spread of socio-economic characteristics and well defined
-------
A-4 8
representative areas for which the distributions of the population by the
various characteristics are known. These requirements have been taken
care of in the planning stage.
Analysis of secondary data usually uses environmental variables, in-
cluding weather and visibility, to explain variation in the participation
rate in a certain activity either over time or space or both. Analysis of
these data yields the sensitivity of participation or the intensity of the
relevant activity to changes in visibility. The following question is how
to transform this information into a monetary evaluation of visibility. The
present note is aimed at answering this question.
The Evaluation
The analysis of participation in an activity is aimed at explaining
observed differences in participation over time i.e., between one day and
another. One of the explanatory variables is visibility. If one agrees to
the concept of a standard quality unit of the activity and that visibility
is one of the components of the vector of characteristics of the quality
then, ceteris paribus, a change in visibility changes the quality of a unit
of activity, which implies a change in the number of standard units per unit
of activity. Formally let a standard unit of activity j be defined by
/v0 v0 o o«
where the X s are the quantities of each attribute of the
standard (for simplicity we disregard the possibility of substitution).
Let attribute n be visibility. Thus, if
3(Quality of activity j)_ ,
3X
n
i.e., the quality denoted by (X^, x° +1) is 1+B larger than standard
ized quality we interpret it as if it is equivalent to 1+B standard units
of activity j.
The use of demand and supply framework to describe different market
equilibria requires that the product (service) be homogeneous. Thus, when
-------
A-4 9
analysing observed participation in activity j the activity has to be trans-
formed into homogeneous units - each at the quality level of the standard.
If we assume that the activities people are involved in are not Giffen goods,
then, aggregate demand for each activity is downward sloping in the quantity
(of standard units)-price per units of standard quality plane. Furthermore,
as long as socio-economic characteristics and population size are constant,
demand is stable.
Assuming that visibility is a positive attribute and that the quality -
quantity transformation into units of standard quality is at a one to one
ratio (as formulated above) then a change in visibility can be viewed as a
change in the average cost of supplying standard units of activity j. Hence,
if for the relevant range of participation in activity j the average cost
of supply is assumed to equal the marginal cost of supply, i.e., they are
identical and horizontal in the quantity price plane, an improvement in
visibility implies a downward paralled shift of the supply curve (Figure 1).
FIGURE 1
$/Unit of
Standard j
Standard Units of j
-------
A-50
Let the elasticity of demand for activity j be rj. then, due to improved
visibility from level V° to V"*" if the observed change in consumption of stan-
AVQi
dard units was AQ, the implied decline in cost of production is AP /p = —¦'—J ,
J j j n.
The social gains due to the improved visibility equal the area At
this stage two problems are encountered. The first is that the observed Q is
i
not in terms of standard units but in units which are unadjusted for quality.
Thus, if we use changes in participation rates due to improved visibility as
a measure for the change in standarized quality units, AQ is underestimated
i
and also AP/P is underestimated. Secondly, the average cost of production of
a standard unit at different levels of visibility is unknown and likewise the
demand elasticity for standarized units is usually unknown. To overcome the
second difficulty, studies on the demand for various activities can be con-
sulted. However, none of the estimated elasticities is for a standarized units
of activity. Thus, in the following an approximation is suggested. The out-
come is obviously an underestimation of the social value of improved visibility.
Hence, when defending it, or similarly, advocating public action to improve
visibility we are on the safe side.
Let's return to Figure 1. Consider a demand elasticity of unity and re-
gard observed changes in participation rate as changes in quality-adjusted
units of activity j . Thus,
AP/P = AQ/Q and AP = — . P;
where Q refers to calculated participation at average annual visibility.
-------
A-51
One can calculate the value of P when a "regular" (non-standard) unit of
activity j is purchased (e.g., value of travel time, automobile costs,
parking costs, entrance fee). The social benefits of improving visibility
from V° to are approximated by
CQ° + qj) AP/2
A very conservative value would be just . AP, and an inbetween value
(f Q° + J Qj) AP/2.
Note that the values of Q° and to be used are those calculated from the
equation for participation in activity j, i.e., they are the predicted values
^ o * 1
(Qj, Qj)* Using the variance covariance matrix of the estimated coefficient,
3 A o 1A1
the variance of the sum (— Qj + can be calculated and confidence intervals
constracted for measurement of the social benefits.
Generalization
Figure 1 can be augmented by adding to it the distribution of visibility
over the relevant period of the year (e.g., for swimming May-Sept.)
-------
A- 5 2
FIGURE 2
$/unit
Prob(V)
\V visi-
* bility
Define an improvement in visibility as the shift of the distribution of visi-
bility 1 unit (or 1 percent if the analysis of participation was done in a log-log
model) to the right. The social benefits due to this improvement are equal to the
sum of the areas of type Pj°ABPj in Figure 1 weighted by the corresponding pro-
bability distribution of visibility. In a discrete formulation it is
-------
A-5 3
1
2
Qj CV^lMj CV£1 . APCQ^lI * Pro&cv^,
where i denotes a level of visibility (m levels are assumed). Also recall that
where AP is calculated only once, at the average V.
Summary
The note suggests a common procedure for the evaluation of social benefits
due to improved visibility when information on the effects of visibility on
behavior is derived from activity participation rates. The method is based
on various approximations. This is its weakness but also its advantage. It is
relatively easy to apply it to various activities. In addition to the estimation
of the participation function only the calculation of average cost per unit of
activity is needed. The final outcome is already an aggregate value for the
corresponding geographic area for which the participation was measured. We also
argue that the various approximations lead to an underestimation of social benefits.
Thus, they would not be refuted by more careful and sophisticated estimation-
calculation technigues.
m
Z Prob(V )-l.
i-1 1
As an approximation one can assume
APCQjCV^) - AP (Qj (V^1)) ,
-------
A-5 4
A. 6 VISIBILITY AND ITS EVALUATION
In the following we discuss the concept of visibility, explain how
different persons conceptualize visibility, and attempt to explain why dif-
ferent people bid different amounts of money for what is "objectively" the
same change in visibility.
Visibility
The dictionary defines visibility in general terms:
a) The quality or state of being visible
(the visibility of a navigational light)
b) The degree or extent to which something is visible,
as by the clearness of the atmosphere
c) Capability of being readily noticed
d) Capability of being distinguished
e) Capability of affording an unobstructed view
The term visible is defined similarly:
a) capable of being seen
b) perceptible by vision
c) easily seen, impressive to the viewer
The conclusion one can draw from these definitions is that visibility
is a subjective property assigned by the human mind via the eyes with or with-
out the usage of visual aids (e.g., binoculars) to various
-------
A-5 5
capabilities all of which are related to vision. The capabilities usually
emphasized are: the identification of objects at different distances at
different levels of clearness, preciseness and brightness, the capability
of distinguishing between different objects and between definite colors.
With regard to colors a comparison with an "ideal" color takes place where
the ideal is a subjective standard the individual has acquired and con-
structed given past experiences of viewing various objects under various
environmental and topographical conditions.
Hence, the declaration that visibility is good or bad, improving or
getting worse reflects differences between perceived visibility at a
specific site, of a specific object, at a specific time of day and environ-
mental conditions and the ideal visibility one has in mind as the numeraire.
We might consider ideal visibility to be a constant for each individual but
different for different individuals. Then experimentation with the same in-
dividual will yield a set of values all refering to the same base. On the
other hand, experimentation with many individuals on one scene yields many
values which however, are non-comparable, The reason is that they refer to
different bases and different subjective perceptions of the same view by
different people. Furthermore, differences between people's "ideals" and
differences in subjective perception are not necessarily perfectly correlated,
given the host of factors that affect perceived visibility and which affect
different people differently. Thus, attempting to adjust for the unknown
ideal base by using background socio-economic variables related to indivi-
duals does not necessarily transform statements of perceived visibility
to a common base. On the top of this is the question whether we know what
are the relevant variables that determine the standard of ideal visibility.
-------
A-5 6
Following the various definitions and expectations from visibility it
seems reasonable to conclude that visibility is not single dimensioned. It
is composed of a set of characteristics or functions it fulfills. Hence,
7- [vrv2,...,vj
where v, is the level of achievement of the aimed at function i. When an
l
individual is shown a picture or is asked to compare two pictures from
their visibility point of view we hypothesize that he is capable of classi-
fying the difference for each i. Now let's experiment with him.
Show the individual a picture and ask him to rank the level of visi-
bility it displays on a scale from 1 to 10. Then ask him to give it the
rank he thinks the majority in the society would rank it. This first ex-
periment would indicate whether the questioned individual has any particular
attitude towards visibility that is different (and knows about it because
of previous experience) from the average in the society. Then show the
individual at least three sets of three pictures each and ask him to rank
visibility within each set on the 1 to 10 scale. The purpose of this
ranking is to quantify the perceived n dimensional vector into a single
dimensional vector. (See reservation below.) An interesting test of the
hypothesis that each individual has a different perception of visibility
would focus on the distribution of the ranks given to the same picture by
different individuals. Similar tests for different perceptions could be
done on the differences in ranks given to two pictures.
For each set of pictures, following the order they were ranked from
top down, ask the individual about his WTP per year in order to avoid deterio-
ration of visibility from that ranked at top to that ranked second and then
from that ranked second to that ranked third, and so on. So far, attempts to
explain WTP data have employed conventional socio-economic characteristics
and variables revealing an individual's attitudes towards the environment,
recreation habits and intention to migrate. We hypothesize that the ex-
-------
A-57
planation of WTP data would be improved if the analysis also included as
variables the absolute difference in the ranks given by the subject to the
pictures, the rank given to the "best" picture, and the difference in
rank for the picture evaluated by the subject for himself and for society.
To be more explicit we postulate that the absolute difference in
ranking affects WTP positively (it quantifies the difference in visibility).
The rank given to the "best" picture captures the particular evaluation of
the entire set. (If the best already ranks low there is little to expect
to be paid for avoiding further deterioration - no use, or, maybe high
payment - increasing marginal disutility.). We suggest that the ranking
of visibility on a 1 to 10 scale be part of the questionnaire and the ranks
be used in explaining the bids.
More on Ranking and Valuation
When the individual is asked to rank visibility on the 1 to 10 scale
we actually ask him to apply his personal weights to each of the n attributes
in the visibility vector. Hence the rank by individual j is:
n
s - Z w v s - 1,. .
j ^ « «
Given the idea of an individual ideal standard
v » V - V
ij ij ij
V. . . . .
where ij is the ideal, and V^,, the perceived. The final rank assigned is
-------
A-5 8
thus a weighted average of the difference between the ideal and the per-
ceived. If we could be sure that the individual is consistent with regard
to the weights he uses, the experiment suggested above would permit the
explanation of WTP for visibility. However we doubt this consistency. In
particular it is uncertain whether the w are constant for individual j or
ij
are a function of the circumstances of the experiment i.e.
"yt" "u
-------
A-5 9
in the market. The individual takes these given quantities and employs
them in the production of the consumed good or service (e.g., watch a boat
race on the lake) . In the production process other inputs, some which are
tradeable, can be employed as substitutes or complements to the visibility
attributes or human eye whose characteristics are not good enough (e.g.,
glasses, binoculars, standing on a high building). For different activities
(production of consumable services), different attributes of visibility are
needed to a different extent. E.G., if one is watching boats on the lake the
distance attribute is most important and next to it the capability of dis-
tinguishing among colors. When visiting the Brayce National Park color
contrast is more important than the capability to see a long distance. I
am using the term important to stand for the economic term MUP = MP * MU --
the marginal utility product. (Recall the similarity to MRP -- marginal
revenue product, which is the product of MR and MP.) The units of the mar-
& units of service x
ginal utility product are of utility (MPy. = L unit of attribute of visibility i'
A units of utility Hence, MUP . = A units of utility ,
MU = A unit of x ' V1 A unit of attribute of visibility i .
In the process of producing service x, more than one attribute of visi-
bility is employed. (It may be that attribute i + 1 improves the quality of x
that is produced using attribute i. This change in quality affects utility
and thus can be expressed similarly.) Thus, the weights the individual assigns
to the various utility attributes when we ask him to evaluate a certain visi-
bility on the 1 to 10 scale are the MUP's that are particular to the view we
show him and circumstances at which he sees it. Thus, the same individual will
assign different per unit of attribute i under different circumstances.
-------
A- 6 0
Furthermore for the presumably same view different people will assign dif-
ferent w, per unit of v, simply because their personal production function
differs and utility differs; thus their MUP , differ.
VI
When an individual is asked to rank visibility he calculates the
values
n k k n 1 1
Z *7 VT and Z w? V*
1-1 1-1
where 1 and k are the same picture at two different levels of visibility.
We traditionally assume that
k 1 . ,
I * 1,... ,n
Thus the difference on a one dimensional scale is
is - Sw.CvJ - vjl - Z w S7.
i 1. L liBi
Thus when asked about WTP the relation is
OTP - f(£S,yl
where y is all other variables affecting WTP.
Two different individuals would thus bid differently even if their
preceived are the same if their differ. I suggest that by asking the
-------
A-61
individual to scale various picture on the 1 to 10 scale we get a good
approximation for his A3 and thus our explanation for the WTP would improve.
A difficulty arises if = f^(some elements contained in y) . This can
be checked by relating AS (and also the scale he assigned the best picture
we showed him) to all the elements we consider to constitute y. (A multiple
regression would do this job.).
Conclusions and Preliminary Remarks for the Eastern U.S. Study.
The main argument put forward in the discussion is that visibility
is multi-dimensional; that the importance of each dimension depends on the
specific scenery; that judgment of changes in visibility depends among
other things on the standards people get use to and to what each vector of
visibility attributes is compared to.
In order to better understand the WTP declared by people (without
currently reflecting or suggesting changes in the various questions in the
questionnaires) we have to get a better idea of the quantification of
perceived changes in visibility. One simple reason for that need is that
declared WTP is a second stage quantification of visibility after
applying to differences in attributes weights that are dependent upon the
process of producing viewing services and output in the individual's sub-
jective utility function. Without knowing the basic information how could
we explain the outcome?
The issues raised above are magnified once the area over which the
planned improvement of visibility is widened to the extent that the individuals
questioned are not familiar with all available views. The possible extention
-------
A- 62
carried out by individuals can be in either of two directions. The first
is a mere extrapolation i.e., given that the extended area is k times the
ot
area previously questioned, willingness to pay is K times the previous
payment where OS a S 1. Another way is more sophisticated and can be ex-
pected only from people that are familiar with the area. They attempt to
apply specific weights to various scenes and then aggregate over the
scenes. Both procedures are probably inadequate, implying that any extra-
polation is likely to yield WTP which would be difficult to explain. Thus,
the alternative of sampling different people at different locations for
different vistas and then aggregating over them seems to be the preferable
way.
-------
A- 63
A. 7 VISIBILITY AND OUTDOOR RECREATION ACTIVITIES:
A RESEARCH FRAMEWORK
In this study we attempt to outline the value of visibility in out-
door recreation activities. The underlying idea is that there is an al-
ternative cost in addition to the direct cost and that these costs and
visibility are the inputs in a production function that provides the con-
sumable commodity - the Becker approach (1965) . This approach is com-
patible with that in which the "production" phase is by-passed and the
utility function contains two arguments that are related to the recreation
activity: a quantity measure which is a function of the cost and a quality
measure which is a function of visibility. The two are substitutes in the
sense that one can compensate for the other along an indifference curve. Yet
we emphasis the assumed assistance in increasing utility by letting the
second cross derivative of the utility function be positive. This second
approach is in line with Maler (1974), but is somewhat more general
since it does not necessarily require the quantity of the recreation
activity to take either of the two values 0, 1.
Visibility Value One Activity
Assume that the expenditure on the recreation activity, R, is
variable and positively related to the quantity of services obtained (seat
in the stadium, length of stay on the tennis court or golf course) . There is
another consumption good which we refer to as income. Visibility affects
only the utility from the recreation activity. Visibility does not have
an explicit market price and it is a public good. If we could have a three
dimensional space, an indifference curve map would represent the tradeoffs
between income, quantity of recreation and quality of recreation. We use
-------
A- 64
a two dimensional space. Thus over each indifference curve both the level
of visibility and of utility are constant. Individuals' total income is Y.
The observed relationships are
u(Y - R°, R°, V°) - u(Y° - R° - AY°, R°, V1)
or
u(Y - R°, R°+AR°, V°) = u(Y - R° , R° , V^")
Hence 6Y° is the compensating variation - while &R is the equivalent varia-
tion. Also both &Y0 and AR° might vary with Y,R° and V0(V^"=V0+AV,AV= Constant.)
Similarly MRS at A is not necessarily equal to that at B. They are equal
y/r
if MU is independent of visibility (R=R°) . The assumption that XU is
Y R
independent of visibility is more difficult to grasp. One would expect it
to increase with visibility. Hence given that the MRS ^ is MU^/MU^ one
/ "d \ (CM
would expect MRS > MRS
-------
A- 6 5
Empirical Implications
The purpose of the study is to get a quantitative measure of the values
of AY and AR. If the two are obtained independently and one might expect the
corresponding MRS to be aobut 1.0 (both are measured in dollars) then a check
for consistency is at hand. Yet before approaching this task one should be
aware of the fact that there are several recreation activities and they
may be close substitutes. The individual behaves such that his utility
from the allocation of the budget (full income) is maximized. Hence under
unfavorable visibility conditions that affect the derived utility from a dollar
spent on activity A by more than the utility of a dollar spent on activity B
we might observe a corner solution with respect to A. This is more likely
to happen if the cost per activity is of the form of a two-part tariff
(fixed plus variable) . Hence the "market" observations on the effect of
visibility take two forms. One is the number of participants, the second
is the intensity of participation. The situation is confounded if we
realize that due to the time consuming input that each activity requires,
participation is feasible in only one out of the set of available activities.
Usually the length of time needed for consumption is disregarded in empiri-
cal demand analysis. Becker (1965) emphasizes its economic role by gene-
rating the full price, full income concepts. However the physical limit of
time - two activities cannot be performed simultaneously-does not bear its
importance in the Becker analysis. For an individual, this constraint leads
to a bang-bang solution (either A or B). For the aggregate we expect to get
different distributions of participates by activity for different visibilities
given that the "reservation" visibilities differ for different persons.
-------
A- 6 6
For empirical investigation we collected data on one outdoor spec-
tator activity - baseball - and one participating outdoor activity - swim-
ming. For each activity the data needed were the attendance rates and the
distribution of attendance by length or intensity. The intensity variable
can be proxied by the guality of seat, which is positively related to the
ticket price. Hence, following the model presented in the first section, one
expects that the worse the visibility the better is the purchased seat. Yet
several difficulties must be realized.
a) Seats are sold in advance. Thus the purchase is done under un-
certainty with respect to the visibility at the day of the game. The larger
the variance of visibility the higher the mean of the guality of seats sold.
Given the seasonality of each of the games, unless cross-sections-over-cities
data are collected the variance effect is undetected.
b) The individual decision making model does not account for exter-
nalities. In the framework of our discussion these will be reflected in
congestion and by "all seats of guality 9 are sold" which are due to capacity
limits of spectators recreation locations. Thus, if capacity is reached the
distribution by guality of seats is invariant to visibility.
c) For spectator activities the demand for attendance and the distri-
bution of seats are not independent from the competing teams. While one of
the teams is always the home team the other team varies. Data for more than
one season are needed in order to estimate an unbiased effect of visibilities
on attendance and seat distribution.
-------
A- 67
The data referred to above are the "macro" data. In order to estimate
the effects of the socio-economic characteristics of the population on the
corresponding compensating variations and equivalent variation "micro"
data are needed. At this stage, we do not discuss the specific contingent valu-
ation instrument but would like to raise one point: the ex ante vs. the ex post values.
Ex ante refers to before the game and thus before the actual effect of visi-
bility on the utility derived from the game is observed. Ex post refers to
the after-observing-and-experiencing effect of visibility. In the ex post
<\ A
case more information is available and thus the AY, AR are better representa-
tives of the CV and EV. Yet the whole experiment of valuing visibility
has an ex ante nature.
-------
A- 6 8
A.8 THE DEMAND FOR VISIBILITY SERVICES
In this section we measure the economic value of an aesthetic charac-
teristic of the environment as revealed through the demand for a private
and priced service. Specifically, we estimate a site specific valuation
of visual air quality by estimating the demand for access to views at a
major observation deck in Chicago. Unlike alternative methods for the
Valuation of environmental services, the method examined requires no
extensive primary data collection. Day to day variation in vistation
and visibility permit an estimate of aggregate demand.
The salient unorthodox feature of the demand analysis is
that neither an explicit price of the service, nor income nor wealth of
the demanders are explicit variables in the model. For the price of the
service we substitute a variable that is presumed to be perfectly corre-
lated with the true price variable. Because the time period examined is
so brief, income can be assumed to remain constant. While the outcome
is but partial valuation of visibility, we suggest that such analyses of
observed behavior offer important corroboration to values derived through
less conventional methods.
The Demand for Visibility
The purpose of this section is to describe the quantitative response
at the observation deck to changes in visibility conditions. We thus defer
theoretical considerations of utility and indirect utility functions which
-------
A- 6 9
are a usual starting point for demand analysis. Instead, we specify the
general aggregate demand function for that activity as a function of its
price, income and the prices of substitutes and complements:
(1) q - fCPd, I. PI,...,Pul
Insofar as q measures a quantity - visitation in a given time period — the
variables specified in (1) are defined somewhat differently from those in a
conventional, demand study. Also on theoretical grounds, it is possible to find
better definitions than the ones used here. However, the empirical orientation
of the analysis leads to practical and observable definitions. For example,
a more precise quantity variable would be the number of man hours per day
spent observing. Correspondingly, an ideal price measure would be marginal
cost per unit of time spent viewing, including relevant direct and indirect
costs. Unfortunately, however, these two measures are not available. In-
stead, the quantity variable is represented by the number of people partici-
pating in viewing while the price variable is assumed to be the sum of
all costs divided by the quantity of visibility services. These total costs
are assumed to be constant across all users. The quantity of visibility
services is the pivotal point of the theoretical model developed below.
For reasons of simplicity, assume that viewing from the tower observa-
tory is in all directions and that the density of vistas is equal per unit
of area regardless of the distance from the tower. A major input for pro-
ducing visible objects is the visual air quality. This input can be measured
by different dimensions, all of which are convertible to "distance of visibi-
lity." Eyes, too, are a necessary element in the viewing process. The
-------
A-70
natural characteristic of eyes are such that the further away is the ob-
ject on which the eye focuses, the less clear is that object. Hence, ad-
justing the quantity of objects viewed by the quality of the view (simi-
lar to a discounting procedure except in this case with respect to dis-
tance) yields a measure of standardized visible objects, denoted VO, where,
V R 2 r 1
VO - / 2TYSe"P - irP l-e"p (Vpfl)
T L J
where V represents the viewing distance allowed by air quality. Clearly,
3V0/ 3V > 0 and 3^0/ 3V2 < 0.
The sum of the entrance price charged by the observatory tower, the
value of traveling time, and travel costs is assumed independent of visi-
bility and is denoted TP hence, the average per unit of view is p = TP/VO,
which is negatively related to VO. Given the above relation between VO and
V, the figure below relates P to V.
-------
A-71
We now rank the potential customers of viewing services by their
reservation price per unit viewed. If this distribution is stable, then
the lower the price per unit of view, the greater the number of people
whose reservation price would exceed the actual price. Hence, visitation
would rise and more would consume the services of the observation tower.
H is the measure of the quantity demanded the number of visitors per unit of time.
Heaee, - •
5 " _/"M(P)dP, such Chat 3M/3P < 0.
P
The remaining elements in the demand function are the prices of sub-
stitures and complementary goods which are not built into the reservation
price. Substitutes as a group would be comprised of all other recreational
activities. We argue here that either the prices of alternative activities
are constant over the analyzed period, i.e. are unaffected by changes in
visual air quality, as for example, museums; or that changes in visibility
affect their effective prices to a lesser extent than they affect the effec-
tive price of the services rendered by an observation tower. (This is another
-------
A-7 2
difficulty with valuing visibility in an urban setting compared to a
National Park where only visual air quality at the time of visitation may
be important.) Obviously, it is less costly to postpone or forego a trip
than changing or canceling plans for activities that are highly time
Intensive. Effective competition comes only from other towers in the area.
Assuming that increments in visibility affects VO uniformly, the
relative price between towers for visibility services is independent of
the level of visibility. This implies a constant distribution of the con-
sumers of observation tower services over the various observation towers.
Hence, changes in visibility conditions leads to equi-proportional changes
in the demand for each of their services.
1
Model, Data and Results
The basic model that has emerged from the previous section relates
the number of visitors per unit time to air visual quality at the time.
In order to get this "net" relation, the gross figures of visitation have
to be adjusted for other variables that determine or cause variation in
visitation. These variables include day of the week, season of the year,
special events, holidays, and meteorological conditions other than visual
air quality. The unit of time for which the participation rate is explained
is: once a morning; once an afternoon; and once the entire day (which in
some sense accounts for substitution among activities during the day),
Substitution over time may take another form - that of substitution
1-We are grateful to the management of the John Hancock Tower for providing
us with the visitation rate by day for the last year and a half. For unknown
reasons, the management of Sears Tower refused to provide us with comparable
data.
-------
A-7 3
between visiting days. This for of substitution is particularly likely
to be found among visitors to the area. Normally, visitors plan to consume
a bundle of services over their period of stay in Chicago. The exact timing
of consumption of a particular service does not change the utility derived
from the entire bundle nor from any particular service. Thus, not only will
there be substitution between periods in a day, but also between days them-
selves. This implies that a relatively high demand might be observed in
spite of poor visual air quality, if this day is the second or third in a
row of poor visibility conditions. Along this line of reasoning, we see that
consumers may indeed hasten their consumption of observatory services on
days when air quality is high because of uncertainty about the quality of
visibility over the next day or two.
These substitution effects, both forced and planned, obscure the inter-
pretation of the coefficient of visibility in the demand relationship from
the point of view of the calculation of the social costs of low visibility
in an urban area.
The estimated model is that of a linear least squares regression, where
specific attention is paid to its the series nature. The model is
Model 1: Y - X + X8+2 Y+e
t c c
Yodel 2: Yt - X® e ^ + Zzy +
Y = number of visits to Hancock Tower on visit day t, t=l,...,N
A visit day may be defined in the following ways:
-------
A-7 4
Y , = number of visits in A.M. hours
cl
Y = number of visits in P.M. hours
Y^ = number of adult tickets sold during A.M. and P.M.
periods combined
Y£^ = number of student tickets sold during day t
Y ^ = total number of visits by all groups during day t
Explanatory Variables:
x,,, = visibility services during time period t,
til 1
Visibility services will take either one of two
alternative measures. The first will be simply visual
range at the Tower. The second will be defined as the
area of a circle determined with visual range as the
0
radius discounted by the R maximizing rate. That is,
Xtill " V in milaS
*til2 " ^T" [ 1 ' _
(in log form the will be dropped)
2c
In addition, two lagged visibility variables will be
included; the first will be the appropriate V from
the previous period and the second from two periods
earlier.
-------
A-7 5
Finally not introduced
Price of substitute
x * P,/P where P is a price index and P is the price
Ci2 l a I e
of admission to the observation deck
ci3 - ,c-lt... ,N = a time trend variable.
= tourists in Chicago (conventions)
= percent of sky covered at 9:00 AM.
= rain (a zero/one-dummy variable)
= cloud cover height in feet.
= Temperature in degree Celsius (This effect
might be non-linear)
= a day of week dummy, either weekday/weekend
or a dummy for each day of week.
= holiday/ non-holiday, dummy variables
= month or season, dummy variable. Eleven
dummies or 3 for groups:
1) Dec., Jan., Feb.
2) Mar., April, May
3) June, July, August
L14
'til
*ci2
*ci3
Zci4
*ci5
*ei6
"ci7
-------
A-7 6
Z^g = special events dummy variable.
As described above, the model can be estimated in both levels and
on a log-log transformation where the estimated coefficients can be inter-
preted directly as elasticities. The VO variable is entered
1/ ......
as vo and the coefficient is invariant with regard to
*
fixed costs and total costs TP. Hence the true coefficient is 3 • * (3)(TP),
where S is the estimated coefficient. In the log-log regression, TP can
2
be disregarded as well as (they become part of the constant) . The
estimated coefficient can be, however, interpreted directly as the elasticity
of visitation with regard to price.
Current atmospheric conditions may affect visitation due to changes in
visibility or through more direct effects on the costs or comforts and safety
of urban travel. Past atmospheric conditions may alter current visitation
through effects such as snow and ice accumulations. The degree of cloudiness
or sunshine may also effect the pleasantness or unpleasantness of outdoor
travel or recreation.
On first trial all the mentioned atmospheric variables were introduced
into the estimated eguation. Given that both visibility and atmospheric
conditions are introduced with lagged values, multicollinearity is likely
to show up. If one uses the rule of thumb definition of multicollinearity,
that is, "correlation among the independent variable," then it is very possibly
present in our study as such responsible for the relatively high standard errors
of estimated coefficients.
-------
A-7 7
As is apparent, the variable of greatest interest is visibility
services, VO. Denoting the coefficients of t y y a
and ^2 , a program that stabilized visibility at a steady state
implies elasticity of visibility with respect to visitation of a a
6°, si,
V
Deducing the Value of Visibility
The models estimated above quantify the response of visitation with
regard to visibility services and other independent variables. Evaluating
the visitation response equations in the admission price/total visitation
plane, one can examine the demand for admission to the Tower.
Visibility services resemble a pure public good where
consumption by one individual leaves unaffected the amount of service re-
maining for the consumption by another. Hence, to value visibility services,
a total value equation is of interest.
The total value equation is estimated by evaluating the visitation
response equation at mean values of independent variables and then multi-
plying the result by the Tower admission price (Figure 1). Total value curve
(1) results from evaluating estimated equation (1) at various levels of
visibility and mean values of other independent variables. Total value
curve (2) results from evaluating estimated response curve (2) in the same
manner. As shown in Figure 1, the non-linear total value relation yeilds a
slightly higher value of Tower services at current visibility level V . To
estimate the daily value of a change in visibility services at the Tower,
one need simply calculate the change in total value. For example, if policy
-------
A-7 8
Figure 1
Visibility and the Value of Visitation
Y^ - S(TV|uf, X)
Yx - E(TV|uf, X}-
Yq =¦ E(TV|uf, X)- -
Yq 0 E(TV] uf, X)--
Mlles
-------
A-79
is presumed to shift typical visibility from to V , then the value of
this shift in terms of services at the Tower would be Y^ - YQ in the case
of the non-linear total value curve or Y. - Y in the case of the linear
1 0
total value curve.
In terms of a total valuation of a policy change, present value
estimates are biased downward. First and perhaps most obviously, the present
value estimates are site specific and only consider the change in value
due to services viewed from a single site. To approximate a site valuation
total, a study would identify all important sites within the area affected
by policy and then total the effects of a policy induced change over all
sites.
A second important reason for undervaluation conceptual. As
visibility rises, an individual's reservation price is also likely to rise.
However, admission price does not change and individual's already viewing
Tower services at the initial level of visibility would realize an un-
measured gain in utility. In Figure 2, this gain is demonstrated. At
visibility level V and income level Z , an individual realizes a utility
c o
level by paying price p^ and visiting the Tower. However, if visibility
rises to V , the same individual by paying the same price can realize
a utility level u0. Given an initial situation (Z ,p ), the individual
would be willing to pay up to $8.00 to realize this gain. Hence, the
estimated total value functions overlook 6 for each individual who would
pay at visibility level and estimate only the value due to additional
patronage. For either increments or decrements in visibility from V, then,
the total value curves will tend to underestimate willingness to pay.
-------
A-80
A third reason the valuation of visibility may be downwardly biased is
due to the definition of the dependent variable. As simply the aggregate
visits to the Tower, the dependent variable does not account for variations
in the amount of time an individual may spend at the Tower. If each in-
dividual spends the same amount of time at the Tower regardless of visibility
then obviously this specification error is not a problem. However, if time
spent at the observation is positively related to visibility, then by dis-
regarding this relation, the total value specified as above may tend to
underestimate the effect of visibility.
Depending on the precise relation between visibility and time spent
viewing, the effect on the valuation procedure may be minimal. For example,
let price be defined as a function of time spent viewing. Specifically, let
the relevant price be the price per unit of time spent viewing and let this
price therefore be calculated as total costs including opportunity costs
divided by the time spent viewing at the Tower. Given that time spent viewing
at the Tower is presumed to be increasing, then we might assign the following
relation:
h -h Va
o
where h is time spent viewing, hQ is some minimum input of time, and V is
visibility. Then the price of viewing per unit of viewing time is:
wb. + c
-------
A-81
FIGURE 2
-------
A-8 2
If another leisure activity and not work is the alternative to visiting
the observation deck, then w equals zero and the coefficient of V in
the estimated equation (1) is an estimate of a. In so far as the func-
tional form chosen for f(V) seems general enough as an approximation,
estimates of total value with respect to V do not seem to be seriously
affected by the present specification of dependent variables.
-------
A-83
A. 9 THE EFFECTS OF VISIBILITY ON AVIATION IN CHICAGO
Visibility affects the flow of air traffic in many ways. First, if
visibility falls below 1 mile, all traffice must be under Instrument Flight Rules
(IFR) . This stops some general aviation activity for both flight training or
recreation. Depending on the aircrafts equipment and landing systems at certain
airports, operations may be legally continued down to 200 yards of visi-
bility.
Another effect of lowered visibility is the delay of take-offs (TO)
and landings. At low levels of visibility, a spacing of at least 1 mile
must be maintained between aircraft. This greater spacing reduces the
numbers of TO and landings that can be made. For instance, suppose that
greater spacing delays each aircraft by one minute at O'Hare International
Airport. Assuming that approximately 60 take-off's and landings are handled per
peak hour of traffic, total operations are delayed overall by one hour.
Decreased visibility can also lead to accidents or near-misses by
contributing to either pilot or air controller error. Lowered visibility
can cause incoming flights to divert to other destinations causing delays
to those on board and imposing additional aviation and ground transporta-
tion costs.
Economic Modeling
The object of this section is to provide a framework for valuing visi-
bility. First consider the effects of visibility on TO or landing opera-
tions at a given airport. For commercial air carriers the effect of visi-
bility on the actual number of flights is expected to be quite low. This
is because they generally operate at the best equipped airports and with
-------
A-8 4
the most sophisticated equipment. The effects of diminished visibility
on general aviation are not so clear. First, when the visibility falls
below 1 mile, all VFR flights stop. Prospective flyers must then decide
whether they wish to fly IFR or postpone their trip. If IFR is chosen,
pilots must be IFR rated and have properly equipped aircraft. Given these
observations, it is an a priori expectation that lowered visibility would
decrease the number of flights. However, this a priori notion may be ob-
viated by the fact that flights is may not be cancelled but merely postponed
until the visibility increases. Weather forecasts are available to pilots
from which they can make decisions on postponement or cancellation. If
early morning visibility is expected to improve within a short time, de-
parture may only be delayed within a day and hence within the period of
observation.
The flexibility of departure tine form the basis for an intertemporal
optimization-of-utility model. The pilot/traveler decides when to leave given
visibility, general weather conditions and expectations of future weather in
order to maximize utility gained from the trip. By the nature of the inter-
temporal trade-off the value of a trip declines as it is put off, but the
increased visibility gained by waiting may add more present value than the
cost of waiting. Consider the folowing intertemporal choice model under
perfect foresight:
Choose t so as to maximize
uact) ,xtt) lv7t+r• ¦ • ^ •
U is utility which is a function of the trip X, which varies in value
over t (hence X(t)). is a vector of quantities of other goods. V^,
-------
A-8 5
V VT are the known future visibility values and V. is a vector of
t+1 ' • • •' N J -i
weather related factors other than visibility. Now, consider the
function
jx(t) IVC>VC+1. • • • '
The value of X(tQ) is 1 when t is optimal, where optimal is defined by
weighting the discounted values of X^_ is 0 for t ^ t . From
this, a demand system can be derived.
Another model of visibility's effect on air travel considers the time
delay caused by restricted visibility. As visibility is reduced, the space
between aircraft must be increased, creating time delays. This line Of attack
could allow a dollar value to be placed on visibility effects. Consider the
following technical relationship:
TDC - GCKUVt, saiw^cpaio^l •
Time Delay (TD) is a function of some lag function of visibility,
a lag function of weather and a lag function of some other factors
such as mechanical breakdowns. The lag functions are included because
J_G
these delays accumulate over time. From this eguation, 3 V shows the
effect of a marginal visibility change on the time delay. By making some
assumptions on the value of passengers, a lower bound cost of visibility
changes can be calculated.
Empirical Modeling
Consider estimating the first conceptual model of the effect of
-------
A- 8 6
visibility on the number of flights. The currently available data consist
of counts, the total number of takeoffs and landings by day at six local
airports by class of aircraft. Weather data are also available. The equa-
tion to be estimated is
CI] log C » v- D + aLogV + SLog a + 5 P + e .
c c c t t
is the count of total take-offs and landings at O'Hare. This variable's
meaning is somewhat ambiguous. First, it cannot he determined how many
aircraft left and returned on the same day, so the number of take-offs
cannot be distinguished from landings. Another even more important
problem is involved with determining the degree of intertemporal trade-
offs. Since the data are for a twenty-four hour period, we cannot determine
if decisions to depart were put off for periods less than twenty-four
hours due to weather expectations. That is, after adjusting for seasonal
and day of week effects, there may be little variation in counts attri-
butable to visibility because all put off effects are very short run,
The vector D is a set of dummies to capture day of week effects.
After viewing the data, differencing may be necessary to filter seasonal
effects. is visibility on day t and is cloud height on day t, and
is a 0-1 variable for whether or not precipitation was present.
A
From this specification, a is the estimated percentage change in
counts for a one percent change in visibility. In order to place a dollar
value on this effect, the average one hour rental fee in Chicago, for a
Cessna 310, a small twin engine aircraft, may used. A lower bound estimate
for the daily cost of a one percent decrease in visibility is ct multiplied
by the average count per day multiplied by the average aircraft cost. This
-------
A-8 7
represents the average cost of increased visibility to someone planning to
take a trip and cancelling or postponing. Clearly, this represents a lower
bound for the actual cost incurred.
The other method of deriving a value on visibility uses time delay
data. By estimating the technical relationship,
LogCTDtl - ID + *ailogVt + SailogHc + s Pc+Dt;,
the relationship between and TD^ can be found. Again, is the per-
centage change in TD induced by a one percent change in V, Two pieces of
data are now needed. First, the mean number of passengers effected by a
time delay and, the value of each passenger's time. By assuming reasonable
values for these two factors a lower bound for the cost of time delays due
to decreased visibility can be estimated.
Another method of deriving the value of visibility deals with the
idea of diverted flights. As was previously mentioned, if flights are
diverted due to low visibility, the aircraft passengers have a cost im-
posed on them. Also, the original destination loses revenue from landing
fees, hanger and fuel charges and, the city of destination loses the
revenue the passengers would have spent. One way to derive this cost is
to look at flight plans filed with the FAA. The number of diverted flights
due to low visibility can he found, as well as the number of flights
diverting to Chicago due to low visibility elsewhere. This can also be done
for flights going to different Chicago airports. If Meigs is socked in by
low visibility then incoming flights may divert to Midway, which means that
Midway then benefits from Meig's loss. The problem with this analysis is
mostly in the expense of gathering data.
-------
A-8 8
At this point, it seems relevant to discuss, relationships across
airports. Each airport has a different schedule of landing fee rates.
There are also non-pecuniary costs differences across airports due to
varying congestion levels. Each airport offers a different bundle of
services. There are two major services to be considered. First con-
sider an airport's location to be an input to producing final services;
i.e., that of getting the passengers to their final destination. An
airport will be chosen so as to minimize transportation costs from the
passenger's point of origin to their final destination. A second service
or set of services acts as a constraint to this decision. This constraint
is in the form of having a runway long enough for the aircraft chosen and
the proper landing system given the prevailing weather.
In choosing which airport to fly into, the passenger or pilot chooses
that which is most easily accessible to the final destination given that
it can be used in the current weather. If Meigs is closed, the flight
may divert to Midway. When viewed in this manner, at least for general
aviation, the substitutability of airports is evident, as is the fact that
the degree of substitutability is a function of the current weather. The
third factor in determining the degree of substitutability is of course
the prevailing landing rate structure.
A similar route selection decision may be made by passengers of scheduled
air carriers. Clearly, for non-pilots and those who do not own aircrafts,
the least cost alternative is usually a scheduled commercial flight. How-
ever, if time cost savings are substantial, the possibility of aircraft
charter enchance the range of substititability. Such charter and non-
-------
A-8 9
scheduled flights may be particularly important at Meigs Field near
down-town Chicago. However, at other airports and/or most commercial
passengers, the cost of charter is likely to outweigh time savings.
Extensions
This section suggests how to extend analysis in ways which add
precision to the estimates for visibility costs in aviation. First,
consider the model for counts. As weather data for each airport lo-
cation are collected, six separate equations can be developed in the form
of (1) . Estimating the six equations jointly adds information to the
estimation procedure. The method of seemingly unrelated regression
provides a straightforward way to proceed. Consider the following
equation system
lo® CL,t ™ —i. aL lQ8 7it + SL lo® *£t + ^ t.lf _ >N
t»l,...,6
This gives us sixa^'s, one for each airport, each of which is estima-
ted more precisely than in the six regressions run separately. So, a
lower bound cost can be estimated for each airport and these costs can
be aggregated to derive a lower bound visibility value for the entire
area.
The other extension applies to the time delay model. Again, the
residuals from the six separate regressions are correlated. By applying
the seemingly unrelated regression procedure to that system of equations,
a more precise time delay elasticity of visibility is estimated for
each airport, and as before, more precise estimates of the cost of
visibility are made.
-------
A- 90
A.10 VIEW PRIMARY RECREATION, THE HANCOCK TOWER
An urban resident or visitor is presented with a large number of
opportunities to view the urban landscape and skyline. A great many
of these viewing opportunities carry a price insofar as one must gain
access to a private viewing site to enjoy a special vista. However,
in very few of these situations is view-use recorded. For several
reasons, urban observation points such as Hancock Tower offered an unu-
sual opportunity to determine the effects of visibility on the demand
for viewing services. First, the panoramic view offered by the Tower is
particularly sensitive to changes in either visual range or color
contrast. Second, an explicit price is charged for access. Finally
attendance is recorded on a daily basis.
Various quarterly reports have described intital findings regar-
ding the behavioral and revenue effects of visibility at Hancock
Tower. Behavioral equations were refined and progress was made toward
a site-specific valuation of visibility. This section provides an
overview of the valuation strategy and presents some demand estimates
for Hancock Tower services as a function of admission price, visibility
and a set of additional demand shifters.
Unlike the common demand analysis which considers goods as divisi-
ble or at least capable of repackaging, a visit to Hancock Tower is more
readily modeled as a discrete choice. That is, the utility maximizing
individual purchases entrance to the Tower if the marginal quality weighted
gains meets or exceeds the marginal cost or entrance price. The maximum
an individual would pay, a reservation price p*, can be modeled analytical-
1
ly and is, for the individual a function of view quality (q), income
-------
A-91
(y), other goods prices (p) , and visit cost shifters such as inclement
weather conditions. That is,
Pi " Pi^i.y.p.vL
In this reservation price context the individual chooses to visit the
site if p* meets or exceeds the price of admission, p°. Hence, the
individual demand for admission to the site is a zero-one valued
choice index it ,,
1'
ffi"
Furthermore, we hypothesize that reservation price rises with an increase
in quality. For the individual whose initialp*(q°, y, p,w) exceeds the
market price, Figure 2 illustrates the gain in consumers' surplus (CS)
1 ...
due the quality change to q . Clearly, an individual who does not visit
either before or after the quality change gains no consumer surplus due
to the view quality change at the site.
$
p*Cq1,q»P.w)
p*(q°,y>P>w).
0 1 T
Figure 2
Hifhen income is included, we are discussing the Marshallian demands.
However, It can be shown that as the budget snare of a commodity approaches
zero, as is likely in the present case, the Marshallian demands approximate
the compensated demands.
-------
A- 92
Aggregate demand for access to the view at Hancock Tower is
simple sum of individual demands. Hencs aggregate demand is considered a func-
tion of current Tower price, (p) , view quality (q), income levels, other goods
price, and the same weather variables (w) that affect individual choice.
For given values of these variables, aggregate demand yields an attendance
count. A particularly convenient functional foot for approximating aggregate
demand is a modified Cobb-Douglas,
„ al a2 a3 a4
VST. = Ap q q t a
where VST is the recorded number of visits for a particular day, A is a yet
to be specified function of shifters, y is aggregate income, t is a time
trend variable, and a is a lognomial error term. As steps prior to estimation,
admission price charged at the Tower is deflated by a montly cost of
living index and monthly real personal income for the U.S. proxies
, , , , , , 2
individual variations m income . Other goods prices are not included
explicitly in the analysis.
The shifter, A is specified as an exponential function of weather
and time related variables such as day of week and seasonal cycles:
A ¦ A(w,d,s)
»exp(w>3l
where d are day or week dummy variables. The seasonal vector, s, may be
specified as either zero-one dummy for south or as sine and cosine functions
of period 365. In the current case with daily observations, the sine and
cosine functions are better suited to fit the likely smooth day to day
change of a seasonal cycle.
0
Both the cost of living index (CPI) and personal income are referenced
in"Economic Indicator, January, 1980" and economic Indicators. Nov., 1980"
prepared by the U.S. council of Economic Advisers.
-------
A- 93
For an initial specification of view quality, we reference recent
work by Malm, et al., that seeks to develop tentative conceptual and
empirical linkage between physical measurements and perceived view quality.
The findings of Malm, et al., suggest that the relationship between
perceived view quality, q, and color contrast, C , is linear:
q - ACr
where A is a function of shift variables such as cloud cover, snow in scene,
and time of day. Due to the tentative nature of the Malm, et al., view
quality/color contrast relationship, it is convenient to allow a more
general form. The function is generalized only slightly:
q - AC8
^ r
where the relationship is linear if 3*1.
Malm, et al., go on to note that
-rb " ,
C - C e. 8X6
r o
where C is the inherent color contrast of a viewed object, r is the
o
observer's distance from that object, and b is a monochromatic or wavelength
ext
weighted, spacially averaged estinction coefficient. Furthermore, the ex-
tinction coefficient is related to visibility, v , by
v - 3.912/b
ext
Hence, the initial relationship between color contrast and view quality
can be transformed to one between quality, object distance, and visibility
-------
A-9 4
or visual range,
q ¦ AC exp(-5r(3.912)/v)
n o
or in log form,
lag - luACo - 8r(3.912/v)
For a given site such as Hancock Tower, it may he considered a weighted
average of viewed object distances. Such a transformation for view quality
is particularly convenient for in the log - log form of the VST equation,
visibility enters as
where 03 InA becomes either a component of the intercept or is added to
the effect of demand shifters such as snowfall and cloud cover.
Once final estimates of the VST equation are completed, consumers
surplus due to view quality change or visibility change at the site can be
easily calculated as long as aj, + l
-------
A- 95
Once CS is calculated it may be accepted as an approximation to compensating
variation or transformed to compensating variation by well documented methods.
Estimates of VST were obtained using a log-log transformation and
ordinary sguares. Suggestive results appear in Table 1. The
dependent variable is the log of total duly attendance and includes all but
one day from the period from January, 1979, through June, 1980. In considering
these results, one may keep in mind that average daily attendance is approxi-
mately 950 persons and the average deflated adult price of admission is about
$0.79 in 1967 dollars. View guality variables are specified in a manner con-
sistent with the Malm, et al, results. IVISB1 and IVTSB2 are simply the
first (VISB1) and second (VISB2) visibility readings (miles) at the Tower,
inverted and multiplied by the constant 3.912. Average VISB1 is about 12
miles and average VISB2 is about 16 miles for the period considered.
Weather observations are for O'Hare International Airport and were obtained
from the National Climatic Center. Independent variables other than IVISB1
and IVISB2 are:
= Log of deflated Tower admission price,
PI = Log of deflated personal income,
LT = Log of time trend variable,
RA = Proportion of weather observations per day recording
rainfall,
SN = Proportion of weather observations per day recording
snowfall,
CL = Proportion of sky covered in clouds,
3
IVISIB1 = 3.912/VISIB1 and IVISB2 = 3.912/VISB2
-------
A-9 7
HTCL = Height of lowest layer clouds in hundreds of
feet,
WIN = Average day windspeed in knots,
TEMP = Average daily temperature in degree fahrenheit,
M, Tu,W,
F, S,Su, = Day of week zero/one dummy variable and
SNX
CSX = Sine and cosine transformations of period 365.
2
Examining the statistical results of Table 1, both the F value and R
are adequate. Estimated coefficients tend to have expected signs. The
price coefficient is very significant, has the expected sign, and indicate
the elasticity of visitation with respect to a price change. The income
variable, RPI, has neither the expected sign nor is it statistically signi-
ficant. Rainfall, snow, and cloud cover are each statistically significant,
have expected signs, and are quite substantial in effect. For example,
ceteris paribus, a full day of rain reduces visitation to about one third
of what if otherwise would have been (exp (-1. 035) =. 35) . Both of the visi-
bility related view quality variables IVISB1 and IVISB2 are statistically
very significant and each having the expected signs; that is, as visibility
increases, extinction coefficients (IVISB1 and IVISB2) decline. As the
extinction coefficient declines, view quality increases and visitation rises.
Hence, the coefficients or IVISB1 and IVISB2 are negative. Coefficients on
day of week variables indicate that visitation an Friday and weekends differs
significantly from visitation on weekdays. Seasonal variables indicate a
strong seasonal cycle with a peak in mid-summer and a trough in early January.
-------
A- 9 8
A.11 VISIBILITY, VIEWS AND THE HOUSING MARKET
Freeman (1979a) identifies three major approaches which can be used
to estimate the demand for a public good such as visibility. These
approaches are: (1) analyze market transactions for something related to
the public good to estimate the implicit demand for the public good itself,
(2) collect individuals' stated values revealed through a contingent market
for the public good and (3) analyze jurisdictional provision of public
goods, taxes and constituency characteristics. Some important contributions
on the aesthetic value of cleaner air have been made using the second
approach, contingent valuation, with Rowe et. al. (1980), Schulze et. al.
(1980) and Tolley_et. al. (1980), focusing specifically on visibility.
As Rowe et. _al. and Freeman argue, the demand estimates based on contingent
values are useful, but they are hardly definitive because of at least some
concern about strategic and induced biases. While Brookshire et. al. (1979)
maintain that these potential biases are practically negligible and that
contingent valuation is reliable, some doubts remain. There is no question
that our understanding can be improved by exploring other approaches.
-------
A-9 9
The purpose of this section is to consider the prospects of using the
implicit market approach to estimate the value of improved visibility
through analysis of the housing market. This section is organized in the
following way. The next part provides the theoretical basis for
estimating the demands for housing amenities through the analysis of im-
plicit markets for amenities. Part III reviews the relevant housing
studies of the demand for amenities related to visibility. The concluding
part deals with what further insights can be expected from studies of
the housing market and suggests a way of obtaining that additional in-
formation on the value of improved visibility.
II. The Implicit Market for Housing Characteristics
Even casual observation suggests that housing is heterogeneous com-
modity composed of various important features other than structural
characteristics alone. These non-structural housing characteristics are
sometimes categorized as: (1) publically-provided services which include
schools, fire protection and garbage collection and (2) neighborhood
amenities which include such characteristics as accessibility, serenity
and air quality. The substantial contribution of neighborhood amenities
to the total price of a house has been established by numerous studies
including that by Krumm (1980) . Tolley and Diamond (1982)
is devoted entirely to the role played by amenities in residence site
choice. Currently estimation of the demand for housing amenities related
to air quality follows some variant of the implicit market approach sug-
gested by Rosen (1974) .
Housing is viewed as a bundle of traits consisting of not only
structural characteristics but neighborhood characteristics and services
as well. Households respond to the traits themselves and, if they cannot
-------
A-100
rearrange or repackage them to exactly suit their tastes, the configura-
tion of traits as well. Households choose a bundle of housing located at
a particular site having only incidental dealings in the market for land.
Utility is maximized over housing and other goods subject to an income
constraint, and an exogenous, through not necessarily linear, price func-
tion for housing. As described by Blomquist and Worley (1981), such a
process yields demand equations for each of the housing traits where
own-price, the prices of complementary and substitutable traits, income,
and tastes are determinants of trait demand. Given that the housing
hedonic function (the market price of housing as a function of the quan-
tities of the various housing characteristics) is interestingly non-linear,
the demand for any particular characteristic is not directly obtainable
in that the housing hedonic equation is a market clearing function in-
fluenced by supply as well as demand conditions. See Freeman (1979b) . In
order to get trait demand, we must estimate the market clearing function,
calculate the marginal trait (hedonic) prices, and use these prices along
with income, other demand shifters, and whatever is necessary to identify
trait demand, see Witt et_. al. (197 9) . By finding the area under the
estimated demand curve, we can estimate the benefits of amenity provision.
This housing market approach, while not without the limitations noted by
Freeman (1979b) and Smith and Diamond (1980), provides useful information
on the value of improved amenities. These estimates can be compared to
that obtained by contingent valuation.
Ill. Housing Studies of Amenities Related to Visibility
A great deal of effort is being devoted to measuring the demands for
clean air and pleasing views -- two housing amenities related to visibility.
Clean Air -- Recent representative studies of the demand for clean
air are those by Harrison and Rubinfeld (1978) who use Boston census
-------
A-101
tract housing and household data to measure the benefits of reduced con-
centrations of nitrogen oxide and particulate, Nelson (1978) who uses
Washington DC census tract and household data to measure the benefits
of reduced concentrations of particulate and oxidants, Brookshire et. al.
(197 9) who use household-specific Los Angeles area data to measure the
benefits of reduced concentrations of nitrogen oxides and particulates,
and Bender et_. a_L_. (1980) who use household-specific Chicago data to
measure the benefits of reduced concentrations of particulate. Table 1
shows the benefits per household of improved air quality as estimated by
Harrison and Rubinfeld, Brookshire et. a_U and Bender et. al. Given that
these measurements are accurate, the estimated benefits of cleaner air
are an upper bound on the value of improved neighborhood visibility to
the resident households. Benefits of improved visibility outside the
neighborhoods and benefits of improved neighborhood visibility to non-
residents are not captured.
Shoreline -- Further information on the upper bound on the value of
improved visibility comes from the study of pleasant views. Brown and
Pollakowski (1977) use the housing market approach to estimate the value
of shoreline. The value of shoreline property would reflect the
desirability of quick access to water-related activities and also the
desirability of views associated with water-related open space. Using
house-specific data for sale price and housing characteristics, they
estimate the value of shoreline in Seattle, Washington. They find that
a house located in an area near a 200 foot-wide setback area will sell
for about $2100 more than a comparable dwelling near a 100 foot-wide
setback and that a house near a 300 foot-wide setback will sell for
about $3336 more than a 700 foot-wide setback (again using the CPI to
convert to June 1980 dollars). This estimated value of shoreline is
-------
A-102
TABLE 1
The Benefits of Cleaner Air
Study
Area
Dependent
Variable
Pollutants
Average Annual
Benefits per Household
Harrison &
Rubinfeld
Boston
Median property
values from cen-
sus tract data
Nitrogen Oxides
and Particulate
$187 for reductions from
auto emission controls
(90% reduction in tail-
pipe emissions)
Brookshire
et. al.
Los Angeles
Sale prices of
individual
houses
Nitrogen Oxides
and Particulate
$68 6 for combined reduc-
tion of about 30% in
average ambient levels
Bender et.
al.
Chicago
Sale prices of
individual
houses
Particulate
$593 for a uniform 20%
reduction.
benefits are converted to June 1980 dollars using the Consumer Price Index (CPI).
The estimates shown are the best point estimates, but each study should be con-
sulted for ranges and gualifications.
b
A 10% discount rate is used to convert the estimate to an annual value.
Source: Calculated from Harrison and Rubinfeld (1978, p. 92), Brookshire et. al.
(1979, p. 131) and Bender _et. _aJ_. (1980, Table IV) .
-------
A-103
relevant, but of limited usefulness for two reasons. The first is that
the value of visibility and viewing cannot be separated from that of access
to water and park-related activity. The second is that the methodology
fails to estimate the demand for shoreline unless we make the heroic
assumption that the housing hedonic equation reveals the demand directly.
Harrison and Rubinfled (1978), Bender et. al. (1980) and Blomquist and
Worley (1981) all find, with different data sets, that there can be
great differences between any benefits estimated directly from the he-
donic and those estimated more appropriately using a two-step procedure.
Pleasing Views -- Abelson (1979) provides more specific information
on the value of visibility-related amenities. In his analysis of housing
prices in the Rockdale section of Sydney, Australia, he considers two
environmental amenities of interest: (1) view, which is measured sub-
jectively as good, average or poor and (2) block level, which indicates
whether or not the house is either on the top side of sloping street or
built well above street level. Abelson relates that some houses have
views overlooking the Pacific Ocean and that views vary greatly in
quality. For all houses in the sample, the value of a good view over
an average view is 1.7% of the average house price, and the value of a
good view over a poor view is 3.5% of the average house price. The value
of a house built on a high block level is 5.5% of the average house price.
If Abelson's specification is correct, then a house with a good view built
on a high level is worth more than a house with a poor view built on a
non-high level by 9% (or 2160 Australian dollars in 1972-73). This
substantial percentage of the total house price suggests that view-related
amenities are important and that even though the value of visibility is
less than that of the view, it may still be non-negligible. Another of
-------
A-104
Abelson's findings indicates that the values of view and visibility in-
crease with income. For the sample with only houses priced above the
average, he finds the values of good views over average views, good views
over poor views and a house built on a high block level all to be approx-
imately twice those for the entire sample. Thus, visibility-related
amenities make up approximately 17% of the total value for higher-priced
houses. This finding is substantiated by the positive simple correlations
between good view and social status (.271) and between good view and ex-
ternal house condition (.156). As with the benefits of shoreline, these
for viewing are estimated directly from the housing hedonic equation
which reflects supply as well as demand conditions and consequently are
subject to unknown bias.
The most exhaustive analysis of view-oriented residences is by
Pollard (1977) who explores the implications of topographical amenities
in an urban housing model. According to Pollard, visual amenities are
a function of the breath (scope) of view which he measures by building
height (floors) and the composition of the view. Since the data are com-
posed of 232 Chicago apartments north of the Loop along Lake Michigan,
dummy variables are created for each loopview and lakeview. Estimating
a rental expenditure function and a building height function which he
derives from a modified Muthian model, Pollard finds that the view affects
both rents and building height. As shown in Table 2, the value of the
views is approximately 14%-17% of average rental payments with values for
lakeview and breadth of veiw based on significant regression coefficients
and loopview on an insignificant coefficient. Given Pollard's estimate
of total monthly rent in the study area is correct, the additional total
rental premium paid for visual amenities is approximately $113 million in
-------
A-105
TABLE 2
The Value of Loop and Lake Views in Chicago
VALUE OF VISUAL AMENITIES
Visual Amenity
Value of Amenity
Lakeview
n o,
/ 0
$332
Loopview^
3%
$142
Q
Breadth
7%
$332
Total
14%-17%
$664-806
EXAMPLE OF A LOOP APARTMENT
Description of Apartment
Premium
for Visual Amenity
Share of Rent of June 1980 Dollars per Year
Apartment with View
1st floor, no special
view
10th floor, no special
view
14%
$791
b
10th floor, Loopview
17%
$957
10th floor, Lakeview
20%
$1177
10th floor, Loopview
22%
51343
and Lakeview
aValues for 1975 are converted to June 1980 dollars using the CPI.
b
The coefficient on which this estimate is based has a t-value of only 0.8.
Q
Since proximity to Lake Michigan increases building heights and hence the
breadth of view, part of the value of breadth is due to a lakeview.
Pollard finds that lakeview apartment buildings are 7 6% taller than non-
lakeview buildings. The value of lakeview implied by taller buildings is
4.3% of average rent (.067 x 64 = 4.3 where 64 = 1.77 x 36). The value
of breadth without the lake height effect is 2.4% of average height (.067
x 36 = 2.4).
Source: Pollard (1977)
-------
A-106
1980 dollars (43.8 x 12 x .14 x 1.533 = 112.8 where 247.1/161.2 = 1.533).
While we must again remember that these values come directly from the
hedonic equation and not from the demands for visual amenities, Pollard's
research clearly indicates their substantial impact on view-oriented
residences and that dimensions of viewing can be successfully considered.
IV. Further Work Based on View-Oriented Residences
Conceptually, the value of any perceived housing characteristic
(including area visibility) can be found through analysis of the implicit
market for the characteristic. As described above, several studies have
estimated the demand for clean air. However, no such study has been done
for visibility, and given the extreme data requirements, it is quite un-
likely that one will ever be done especially for a housing market as large
as a Standard Metropolitan Statistical Area. In marked contrast is the
excellent prospect for learning more about the value of views and compo-
nents of views. We have seen that in view-oriented submarkets, there is
some indication that viewing can be worth as much as 20% of total housing
expenditures -- an effect readily detectable by statistical hedonic price-
trait demand analysis with average quality data. We now address what
such a study might entail.
Let us assume that households maximize their utility which is separ-
able and depends on housing and a composite good excluding housing.
Housing, which is a vector of housing characteristics, can be considered
as having view-related characteristics such as breadth and composition as
well as characteristics unrelated to viewing. Following the theory and
methodology described in part II, we would estimate the hedonic housing
function which includes the view-related characteristics estimate the
-------
A-107
the demands for these special characteristics, and aggregate to get the
value of views.
For a submarket like Pollard's where view-oriented residences are
prominent, the hedonic housing function would specify rent as a function
of structural characteristics such as floor space, rooms, baths, age,
fireplaces, central air conditioning, central heating, units in building,
floors in building, garage, separate storage area, building elevator;
payment characteristics such as whether or not rent includes utilities,
heating, air conditioning, garbage collection, parking; neighborhood
characteristics such as access to employment and shopping, school quality,
crime rate, street conditions, litter, noise, abandoned buildings; and
view characteristics such as height of the apartment in floors, percen-
tage of horizon which can be viewed from the apartment, a dummy for
Lakeview, a dummy for Loopview, a dummy for ability to view to the hori-
zon, and a dummy for extraordinary window space. (The hedonic equation
can accommodate condominiums with adjustments for property taxes, and the
annual flow of housing services similar to those found in Linneman (1980)).
The best functional form for the hedonic function can be determined by
using a quadratic Box-Cox procedure similar to that used by Bender _et. al.
(1980) .
Estimating the demand for view characteristics will make use of the
hedonic prices for housing characteristics and household characteristics
such as income, family size, age structure and education. The proper
specification of the demand equation can be determined through a series
of tests for the superiority of alternatives following Blomquist and Worley
(1982) and Harrison and Rubinfield (1978).
By coordinating the housing market and contingent valuation approaches
-------
A-108
to estimating the value of improved visibility progress can be made in
critical areas of benefit estimation. First, sturctural and neighborhood
housing characteristics obtained from cooperative building managers can
be supplemented and matched with view and household characteristics ob-
tained through the contingent valuation survey. This merger would permit
estimating benefits from the demands for view characteristics, not the
hedonic housing equation. Second, by carrying out a contingent valuation
study for views (in addition to a study for visibility) we can compare
the estimates of the value of views obtained from the housing (implicit)
market and contingent market studies. Such a comparison is crucial to
understanding the usefulness of contingent values of environmental
amenities such as visibility which are not easily estimated by alterna-
tive approaches.
------- |