United States Office of Air Quality EPA-450/5-83-001f
Environmental Protection Planning and Standards August 1982
Agency Research Triangle Park NC 27711
Air
S-EPA Benefit Analysis
of Alternative
Secondary
National Ambient
Air Quality
Standards
for Sulfur Dioxide
and Total
Suspended
Particulates
Volume VI
-------
-------
FINAL ANALYSIS
BENEFITS ANALYSIS OF ALTERNATIVE SECONDARY
NATIONAL AMBIENT AIR QUALITY STANDARDS FOR
SULFUR DIOXIDE AND TOTAL SUSPENDED PARTICULATES
VOLUME VI
BENEFITS ANALYSIS PROGRAM
ECONOMIC ANALYSIS BRANCH
STRATEGIES AND AIR STANDARDS DIVISION
OFFICE OF AIR QUALITY PLANNING AND STANDARDS
U-S- ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK
NORTH CAROLINA 27711
U.S. Environmental Protection Agency
Region V, -,-.v
-. > • , r '• ' f"~ ' - 1 v • • '-" •*• *-
2"") '.--oat > •- - ''-""' f
AUGUST 1982 Chicago, Illinois 60604
-------
\J,S
. Environmental Protection
-------
FINAL ANALYSIS
BENEFITS ANALYSIS OF ALTERNATIVE SECONDARY
NATIONAL AMBIENT AIR QUALITY STANDARDS FOR
SULFUR DIOXIDE AND TOTAL SUSPENDED PARTICULATES
By:
Ernest H. Manuel, Jr. William N. Lanen
Robert L. Horst, Jr. Marcus C. Duff
Kathleen M. Brennan Judith K. Tapiero
With the Assistance of:
Richard M. Adams A. Myrick Freeman, III
David S. Brookshire Shelby D. Gerking
Thomas D. Crocker Edwin S. Mills
Ralph C. d'Arge William D. Schulze
MATHTECH, Inc.
P.O. Box 2392
Princeton, New Jersey 08540
EPA Contract Number 68-02-3392
Project Officer:
Allen C. Basala
Economic Analysis Branch
Strategies and Air Standards Division
Office of Air Quality Planning and Standards
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina 27711
August 1982
-------
PREFACE
This report was prepared for the U.S. Environmental Protection
Agency by MATHTECH, Inc. The report is organized into six volumes
containing a total of 14 sections as follows:
Volume I
Section 1:
Section 2:
Section 3:
Volume II
Section 4:
Section 5:
Sect ion 6:
Volume III
Sect ion 7:
Section 8:
Volume IV
Section 9:
Volume V
Section 10:
Section 11:
Volume VI
Section 12:
Section 13:
Section 14:
Executive Summary
Theory, Methods and Organization
Air Quality and Meteorological Data
Household Sector
Residential Property Market
Labor Services Market
Manufacturing Sector
Electric Utility Sector
Agricultural Sector
Extrapolations
Bibliography
Summary of the Public Meeting
Analysis of Pollutant Correlations
Summary of Manufacturing Sector Review
The analysis and conclusions presented in this report are those
of the authors and should not be interpreted as necessarily reflecting
the official policies of the U.S. Environmental Protection Agency.
11
-------
ACKNOWLEDGMENTS
This report and the underlying analyses profited considerably
from the efforts of Allen Basala, who served as EPA Project Officer,
and V. Kerry Smith, who served as a reviewer for EPA. Allen provided
the initiative and on-going support to conduct an applied benefits
analysis. Kerry's technical insights and suggestions are reflected in
nearly every section of the report.
James Bain and Tom Walton of EPA, and Jan Laarman and Ray
Palmquist, who served as reviewers for EPA, also contributed
substantially to individual report sections through their advice and
comments during the course of the project. Also providing helpful
comments and assistance were Don Gillette, Fred Haynie, Neil Frank and
Larry Zaragosa, all with EPA.
Several other members of the Mathtech staff contributed to the
project during various stages of the work. They included Robert J.
Anderson, Jr., Neil Swan, John Keith, Donald Wise, Yaw Ansu, Gary
Labovich, and Janet Stotsky.
The production of the report was ably managed by Carol Rossell,
whose patience remained intact through countless drafts and deadlines.
Carol was assisted by Sally Webb, Gail Gay, and Deborah Piantoni.
Finally, we extend our appreciation to the many dozens of
individuals, too numerous to list here, who provided advice,
suggestions, and data during the course of the project.
111
-------
CONTENTS
Section Page
12 SUMMARY OF THE PUBLIC MEETING
Introduction 12-1
Summary of Public Meeting 12-1
Household Sector 12-5
Electric Utility Sector 12-7
Agricultural Sector 12-7
Manufacturing Sector 12-9
Concluding Remarks 12-11
Recommendations 12-12
Appendix A: Federal Register Notice, Agenda for
Public Meeting, ~ List of Participants 12-13
Appendix B: Comments from Panel of Experts in the
Field of Environmental Benefits
Analysis 12-22
Appendix C: Comments from the American Iron and
Steel Institute 12-56
13 ANALYSIS OF POLLUTANT CORRELATIONS
Introduction 13-1
Estimation of Correlation Coefficients 13-3
Potential for Bias in Mathtech Study 13-19
Summary and Conclusions 13-28
References 13-31
IV
-------
CONTENTS (Continued)
Section Page
14 SUMMARY OF MANUFACTURING SECTOR REVIEW
Introduction 14-1
Informal Plant Interviews 14-2
Statistical Reanalysis 14-8
-------
TABLES
Number Page
13-1 Count of Available Site Observations for
Correlation Analysis 13-8
13-2 Correlation Coefficients for 1978 Annual Arithmetic
Means of TSP and Other Pollutants for Different
Levels of Spatial Aggregation 13-11
13-3 Correlation Coefficients for 1978 Annual Arithmetic
Means of S02 and Other Pollutants for Different
Levels of Spatial Aggregation 13-11
13-4 Significance of Correlation Coefficients for 1978
County Annual Arithmetic Means 13-13
13-5 Correlation Coefficients for Annual Arithmetic Means
of County Levels of TSP and Other Pollutants for
Three Years 13-15
13-6 Correlation Coefficients for Annual Arithmetic Means
of County Levels of S02 and Other Pollutants for
Three Years 13-15
13-7 Correlation Coefficients for 1978 County-Level Data
of TSP and Other Pollutants for Different Statistical
Measures 13-18
13-8 Correlation Coefficients for 1978 County-Level Data
of S02 and Other Pollutants for Different Statistical
Measures 13-18
13-9 Correlation of S02/N02 for Household Sector Study,
1972 and 1973 13-22
13-10 Correlations for S02/N02 and TSP/N02 for
Manufacturing Sector Study, 1973 13-24
13-11 Correlations of S02/03 for Agricultural Sector
Study — Quarterly Data 1974-76 13-27
VI
-------
TABLES (Continued)
Dumber Page
14-1 Counties Used in the Draft Analysis for SIC 344 14-4
14-2 Possible Effects and Responses to Particulate
Matter in SIC 344 14-5
14-3 Comparison of Results for SIC 344 14-9
VII
-------
SECTION 12
SUMMARY OF THE PUBLIC MEETING
-------
SECTION 12
SUMMARY OF THE PUBLIC MEETING
INTRODUCTION
On July 27 and 28, 1981, a public meeting was held at the
Sheraton Crabtree Inn in Raleigh, North Carolina. The purpose of this
meeting was to provide an opportunity for public comment on the study
and to receive comments from a panel of experts in the field of
environmental benefits analysis. In addition to this panel of
experts, the major participants at the meeting included staff and
consultants from the Office of Air Quality Planning and Standards
(OAQPS) for the Environmental Protection Agency (EPA), and staff and
consultants from Mathtech, Inc. Other interested individuals also
particpated. A copy of the notice for this meeting as it appeared in
the Federal Register, the agenda for the meeting, and a list of those
attending the meeting are contained in Appendix A of this volume.
SUMMARY OF PUBLIC MEETING
The meeting was opened by James Bain, Chief of the Economic
Analysis Branch of OAQPS who stated that the purpose of the meeting
12-1
-------
was to present, discuss, and receive feedback regarding the benefits
analysis done by Mathtech.
An overview of the general approach used for estimating the
benefits of alternative secondary national ambient air quality
standards (SNAAQS) and a summary of the results of the study were
given in a slide presentation by Ernest Manuel of Mathtech. This was
followed by individual slide presentations by members of the Mathtech
staff for each of the economic sectors analyzed in the study.
Comments and questions from the panel of environmental experts were
made after each slide presentation. Following the panel's comments,
comments and questions were received from members of the audience.
General comments and recommendations regarding the study were made at
the conclusion of the two-day meeting.
With respect to the overall study, three general observations
were made by the panel of environmental experts. The first observa-
tion was related to the study's overall quality. Each member of the
panel and various members of the audience agreed that the Mathtech
study was a high quality piece of research and represented "state-of-
the-art" economics. It was emphasized that the methodology used to
estimate the economic benefits of alternative SNAAQS was based on
conventional economic theory. In addition, several comments were made
commending the use of sound and sophisticated econometric techniques
throughout the analysis. In fact, several members of the panel
12-2
-------
suggested that shortened versions of some of the sector analyses
should be submitted to economic journals for publication.
The second general observation was concerned with the quantity
and quality of the data used in the empirical estimation of the sector
models developed in the study. A few members of the panel pointed out
that although the models developed were theoretically sound, less than
optimal data were available for the estimation of some of the models.
It was acknowledged, however, that this was not a problem specific to
the Mathtech analysis, but rather a problem that beset all analyses of
this type. Specific comments were raised regarding the appropriate-
ness of using data from air quality monitoring stations as proxies for
ambient exposure to certain pollutants and the omission of pollutants
such as ozone and nitrogen dioxide from the estimated models. The
general concern of the panel was directed toward how these data limi-
tations and omissions might affect the conclusions that could be drawn
from the study. It was suggested that the implications of these data
limitations should be examined with respect to the Mathtech study and
with respect to the needs of future research.
The discussion of the data limitations encountered in the study
prompted a third general comment from the panel members — the plausi-
bility of the benefit estimates. For the types of benefits measured
in the study, the panel felt that the benefit estimates presented at
the meeting were the best estimates currently available. The panel
members and others attending the meeting acknowledged that because of
12-3
-------
the conservative assumptions employed throughout the study, the esti-
mates reported in the study were most likely lower-bound estimates of
the benefits of attaining alternative SNAAQS in the economic sectors
analyzed. It was also pointed out that the estimated benefits did not
reflect all of the benefits of attaining SNAAQS since the estimation
of other welfare benefits were not part of the Mathtech analysis
(e.g., aesthetics, recreation, other economic sectors). In addition,
it was mentioned that any health benefits that might occur as a result
of the implementation of a secondary standard were not included in the
benefit estimates presented at the public meeting.
Several members of the panel expressed concern that only the
point estimates of the benefits were reported in the analysis. It was
suggested that it would have been more appropriate to report a range
of benefit estimates based on the confidence intervals that are
associated with these point estimates. It was explained, however,
that because of the complexity of the econometric models used in the
benefit estimation, it might not be possible to develop rigorously
such confidence intervals.
The remainder of the comments were specific to each economic
sector and will be discussed according to the agenda followed at the
public meeting. It should be mentioned that since copies of the
written comments made by the panel of environmental experts are given
in Appendix B of this volume, only the major comments regarding these
sectors will be discussed in this section. Those interested in
12-4
-------
obtaining further information regarding the comments made at the
public meeting should refer to Appendix B.
HOUSEHOLD SECTOR
The comments specific to the household sector analysis focused on
two major points: 1) the coverage of the economic benefits accruing
to households that was provided by the household expenditure model,
and 2) the comparability between the benefits estimated by the house-
hold expenditure model and the benefits estimated by the residential
property value and wage studies. With respect to the first point,
several panel members praised the innovative approach taken in the
household sector analysis. It was felt that since many of the actions
that households might undertake as a means of mitigating the effects
of air pollution were capable of being reflected in the household
expenditure model. Consequently, the model provided more realistic
coverage of the economic benefits accruing in this sector. This was
considered to be a significant improvement in the field of environ-
mental benefits analysis.
A few panel members noted, however, that the effects of all air
pollutants were not measured in this model. It was also mentioned
that all of the mitigative actions available to households were not
reflected in the model. For example, it was noted that neither house-
hold location decisions nor labor-leisure decisions were reflected in
the consumer expenditure model. In addition, the possibility that the
12-5
-------
household may have a "pure" utility increase as a result of reduced
air pollution exposure was not captured by the model. One panel
member explained that any effects of air pollution that do not result
in a change in the household's market behavior were not "picked up" in
the consumer expenditure model. This panel member further explained
that the omission of non-market adjustments and some of the mitigative
actions available to the household would result in an underestimate of
the benefits of reductions in air pollution. There was a general
concurrence among those attending the meeting that the household
sector estimates were lower-bound estimates of the household sector
benefits of attaining SNAAQS. Robert Horst of Mathtech acknowledged
that Mathtech was aware of this issue and hoped to examine how the
model could be modified to incorporate other mitigative actions.
The second comment regarding the comparability between the
benefits estimated by the household expenditure model and the benefits
estimated from the property value and wage studies was directly
related to the comment about the amount of benefit coverage provided
by the household expenditure model. It was observed by several panel
members that the benefits estimated from the household expenditure
model were much less than those estimated from the property value and
wage studies. One reason for the disparity between these estimates
was the fact that the household expenditure model was not designed to
estimate the health and aesthetic benefits that may result from imple-
mentation of a secondary standard. Another reason for the differences
12-6
-------
between these estimates was the exclusion of the previously mentioned
mitigative actions from the household expenditure model.
ELECTRIC UTILITY SECTOR
The major comment that was raised in reference to this sector
concerned the multicollinearity among the variables used in estimating
the cost functions for these utilities and how this collinearity would
affect the estimated relationship between air pollution and the utili-
ties' costs. It was mentioned that variables such as age of plant,
stack height, sulfur content of fuel, and air quality that were
included in the cost functions were likely to be highly correlated
with one another. It was noted that although this collinearity would
increase the standard errors of the estimated coefficients for these
variables, it would not bias the coefficient estimates.
AGRICULTURAL SECTOR
The one comment that seemed to pervade the panel's discussion of
the agriculture sector analysis concerned the lack of data. Kathleen
Brennan of Mathtech mentioned that estimation of the crop yield
equation was extremely difficult since county level data on the use of
many of the farm inputs used to produce specific crops could not be
obtained from the agriculture sector analysis. Because of the lack of
data, the question was raised whether the results of controlled field
experiments could be used in place of economic studies in order to
12-7
-------
approximate the economic benefits of reductions in sulfur dioxide.
Reservations about using the results of controlled field studies were
expressed because these studies do not replicate the actual conditions
under which agricultural crops are grown. It was felt that the agri-
cultural analysis emphasized the need for the collection of better and
more data.
Several panel members expressed concern about the omission of
relevant pollution variables such as ozone and nitrogen dioxide from
the agricultural crop yield equation. It was suggested that the
impact that these omissions might have on the estimated relationship
between agricultural crop yield and sulfur dioxide be investigated.
Another comment raised by the panel focused on the inability of
the model, as currently developed, to take into account some of the
actions that a producer might undertake to offset the effect of sulfur
dioxide on agricultural crops. Cultivar and crop substitution were
two mitigative actions specifically mentioned. It was acknowledged
that because producers had probably adjusted to the effects of air
pollution through cultivar and crop substitution and because these
adjustments were not incorporated into the model, the estimated
benefits were likely to be lower-bound estimates of the economic
benefits of SNAAQS.
12-8
-------
MANUFACTURING SECTOR
The format followed for the manufacturing sector discussion
differed somewhat from that published in the agenda. After Ernest
Manuel of Mathtech completed the manufacturing sector slide presenta-
tion, comments were received from representatives of the American Iron
and Steel Institute (AISI). (Written copies of the comments provided
by the AISI representatives can be found in Appendix C.) Comments and
questions were taken from the panel of environmental experts and
general audience following Mr. Manuel's reply to AISI's comments.
The major comment from the representatives of AISI was that the
study's finding of a statistically significant relationship between
air pollution and manufacturing costs could not be construed to mean
that air pollution caused manufacturing costs to be higher. The
representatives cited several urban factors that they felt were the
"true" causes of higher production costs: age of plant, wage rate,
tax rates, and older labor force. Since these factors tend to be
correlated with air pollution, the representatives asserted that the
relationship observed between air pollution and manufacturing costs
was due to correlation and not causation. Mr. Manuel replied that
variables reflecting age of plant and the labor costs were included in
the manufacturing model and hence the effects of these factors could
not be attributed to air pollution. Mr. Manuel felt that the problem
of other omitted urban variables such as the age of the labor force
and tax rates was minimized somewhat since the majority of the
12-9
-------
manufacturing data were for establishments located in relatively urban
areas. He stated further that although no statistical analysis out-
side of controlled laboratory conditions could ever "prove" a cause-
and-effect relationship, he felt that the manufacturing sector
analysis adequately controlled for the other factors influencing pro-
duction cost so that the relationship observed between air pollution
and production cost in the study suggested a cause-and-effeet rela-
tionship.
The panel members also commented about the omission of "urban"
variables from the production cost functions that were estimated for
the industries in the manufacturing sector. It was mentioned that the
omission of these variables might bias the coefficients of the air
pollution variables because these excluded variables tend to be corre-
lated with air pollution.
The other comment raised by the panel of environmental experts
dealt with the benefits estimated for the fabricated structural metal
products sector. Questions were raised as to why the benefits
estimated for this particular industry were so large relative to the
benefits estimated for the other manufacturing sectors. The sensi-
tivity of the inventories and welding operations in the fabricated
structural metal products industry to dust, and therefore to particu-
late matter, were given as two possible reasons for the magnitude of
these benefits. It was suggested, however, that the reasons for the
12-10
-------
magnitude of the benefits estimated for this industry be investigated
further.
In general, the panel members praised the manufacturing sector
analysis. Both the careful and sophisticated research undertaken in
this sector were highly commended. The panel members acknowledged
that based on the conservative assumptions employed throughout the
analysis and the fact that only 4 percent of the manufacturing sector
was covered in the analysis, the estimates reported for this sector
were clearly lower-bound estimates of the benefits of implementing
SNAAQS.
CONCLUDING REMARKS
During the general comment period, James Bain of OAQPS asked the
panel members to comment on the potential usefulness of the Mathtech
study in the standard setting process. Everyone concurred that the
study could be extremely useful in helping policy makers make environ-
mental decisions. Because of the conservative assumptions and the
limited number of sectors covered in the study, it was stressed that
the benefits estimated in the study were lower-bound estimates of
alternative SNAAQS. It was mentioned, however, that the current
regulatory process required that the study appear in the Criteria
Document before it could be considered in the standard setting
process.
12-11
-------
It was emphasized that although the study was a significant
advancement in the field of environmental benefits analysis, there
were still other benefit categories, not considered in the Mathtech
study, that have not been adequately examined. It was suggested that
in order to comply with Executive Order 12291, the study's estimates
should be combined with the "best available" estimates of the benefits
accruing in other economic sectors in order to approximate the total
benefits of SNAAQS implementation.
RECOMMENDATIONS
At the close of the meeting, two recommendations were made
regarding the Mathtech study. The first one was to examine the corre-
lations between the excluded air pollution variables and total
suspended particulates (TSP) and sulfur dioxide (802). This was
suggested in order to find out whether the exclusion of these
variables might bias the coefficients of TSP and S02.
The second recommendation was to investigate whether it is
feasible to report a range of benefit estimates based on the confi-
dence intervals associated with the point estimates reported in the
study.
12-12
-------
APPENDIX A
Federal Register Notice, Agenda for Public Meeting, and
List of Participants
12-13
-------
r
33626
Federal Register / Vol. 46, No. 125 / Tuesday, June 30, 1981 / Notices
The above notices of determination
were received from the indicated
jurisdictional agencies by the Federal
Energy Regulatory Commission pursuant
to the Natural Gas Policy Act of 1978
and 18 CFR 274.104. Negative
determinations are indicated by a "D"
before the section code. Estimated
annual production (PROD) is in million
cubic feet (MMCF). An (*) before the
Control (JD) number denotes additional
purchasers listed at the end of the
notice.
The applications for determination are
available for inspection except to the
extent such material is confidential
under 18 CFR 275.206, at the
Commission's Division of Public
Information, Room 1000, 825 North
Capitol St., Washington, D.C. Persons
objecting to any of these determinations
may, in accordance with 18 CFR 275.203
and 275.204, file a protest with the
Commission on or before July 15,1981."
Categories within each NGPA section
are indicated by the following codes: '
Section 102-1: New DCS lease
102-2: New well (2.5 mile rule)
102-3: New well (1000 ft rule)
102-4: New onshore reservoir
102-5: New reservoir on old OCS lease
Section 107-DP: 15,000 feet or deeper
107-GB: Geopressured brine
107-CS: Coal seams
107-DV: Devonian shale
107-PE: Production enhancement
107-TF: New tight formation
107-RT: Recompletion tight formation
Section 108: Stripper well
108-SA: Seasonally affected
108-ER: Enhanced recovery
lOft-PB: Pressure buildup
Kenneth F. Plumb,
Secretary.
[FR Doc. 81-19223 Filed 6-29-81: 8:45 am]
BIUJNO CODE 64SO-«S-H
ENVIRONMENTAL PROTECTION
AGENCY
[AD-FRL-1869-8J
Benefits Analysis of National Ambient
Air Quality Standards for Sulfur Oxides
and Particular Matter, Meeting
AGENCY: Environment Protection
Agency (EPA).
ACTION: Notice of public meeting.
SUMMARY: EPA's Office of Air Quality
Planning and Standards is conducting a
public meeting to solicit input on a
contractor's technical report containing
a benefits analysis methodology for
National Ambient Air Quality Standards
for sulfur oxides and particulate matter
under Section 109 of the Clean Air Act,
42 U.S.C. 7409. Aj>anel of experts in the
field of environmental benefits analysis
will critically discuss the analysis of
various economic sectors. Questions
and comments from the general public
will be also be discussed. The meeting
will be from 9:00 a.m. to 5:00 p.m. on
both Monday, July 27, and Tuesday, July
28,1981. It will be held in the Governor's
Room of the Sheraton-Crabtree Inn, U.S.
70 West, Raleigh, NC 27612.
FOR FURTHER INFORMATION CONTACT:
Janet Scheid (919) 541-5611/(FTS 629-
5611) of the Ecnomic Analysis Branch,
Strategies and Air Standards Division,
Office of Air Quality Planning and
Standards. The mailing address is:
Economic Analysis Branch (MD-12),
U.S. Environmental Protection Agency,
Research Triangle Park, 27711.
Dated: June 24,1981.
Edward Tuetk
Acting Assistant Administrator for Air, Noise.
and Radiation.
[FR Doc. 81-19162 Piled 6-29-81; 8:45 am]
MUJNG CODE 65M-26-M
[AMS-FRL-1871-2]
Fuel Economy Retrofit Devices;
Announcement of Fuel Economy
Retrofit Device Evaluation for "FUEL-
MAX"
AGENCY: Environmental Protection
Agency (EPA).
ACTION: Notice of Fuel Economy Retrofit
Device Evaluation.
SUMMARY: This document announces the
conclusions of the EPA evaluation of the
"FUEL-MAX" device under provisions
of-Section 511 of the Motor Vehicle
Information and Cost Savings Act
Background Information
Section 511(b)(l) and Section 511(cj of
the Motor Vehicle-Information and Cost
Savings Act (15 U.S.C. 2011(b)) requires
that
(b)(l) "Upon application of any '
manufacturer of a retrofit device (or
prototype thereof), upon the request of
the Federal Trade Commission pursuant
to subsection (a), or upon his own
motion, the EPA Administrator shall
evaluate, in accordance with rules
prescribed under subsection (d), any
retrofit device to determine whether the
retrofit device increases fuel economy
and to determine whether the
representations (if any) made with
respect to such retrofit devices are
accurate,"
(c) "The EPA Administrator shall
publish in the Federal Register a
summary of the results of all tests
conducted under this section, together
with the EPA Administrator's
conclusions as to—
(1) the effect of any retrofit device on
fuel economy;
(2) the effect of any such device on
emissions of air pollutants; and
(3) any other information which the
Administrator determines to be relevant
in evaluating such device."
EPA published final regulations
establishing procedures for conducting
fuel economy retrofit device evaluations
on March 23,1979 [44 FR 17946).
Origin of Request for Evaluation
On January 18,1980, the EPA received
a request from FID CO, Fuel Injection
Development Corporation, for
evaluation of a fuel saving device
termed "FUEL-MAX." This device is an
air bleed device that replaces the EGR
valve. It is claimed to conserve fuel.
Availability of Evaluation Report
' An evaluation has been made and the
results, are described completely in a
report entitled: "EPA Evaluation of the
FUEL-MAX Device Under Section 511 of
the Motor Vehicle Information and Cost
Savings Act." This entire report is
contained in two volumes. The
discussions, conclusions and list of all
attachments are listed in EPA-AA-TEB-
511-81-10A, which consists of 18 pages.
The attachments are contained in EPA-
AA-TEB-511-81-10B, which consists of
120 pages. The attachments include
correspondence between the Applicant
and EPA, all documents submitted in
support of the Application and the "EPA
testing of the device.
As a part of its evaluation EPA has
actually tested the FUEL-MAX device.
The EPA testing is described completely
in the report "Emissions and Fuel
Economy of FUEL-MAX, a Retrofit
Device," EPA-AA-TEB-81-15,
consisting of 8 pages. This report is
contained in the preceding FUEL-MAX
511 Evaluation as an attachment and
can be obtained separately or as part of
the attachment package.
Copies of these reports may be
obtained from the National Technical
Information Service by using the above
report numbers. Address requests to:
National Technical Information Service,
U.S. Department of Commerce,
Springfield, VA 22161, Phone: Federal
Telecommunications System (FTS) 737-
4650, Commercial 703-J87-4650.
Summary of Evaluation
EPA fully considered all of the
information submitted by the Device
manufacturer in the Application. The
evaluation of the "FUEL-MAX" device
12-14
-------
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Public Meeting
Benefits Analysis of Secondary National Ambient Air Quality Standards
for S02 and TSP
July 27-28, 1981
AGENDA
Monday, July 27
9:00 a.m. Introduction
Jim Bain, Chief
Economic Analysis Branch
U.S. EPA
9:15 a.m. Project Overview
Ernest Manuel, Project Manager
Mathtech, Inc.
9:45 a.m.
Household Sector
Analysis
Robert Horst
Mathtech, Inc.
10:00 a.m. Household Sector
Discussion and Comments
12:00 noon Lunch
1:00 p.m.
Utility Sector
Analysis
Ernest Manuel, Project Manager
Mathtech, Inc.
1:45 p.m. Utility Sector
Discussion.and Comments
5:00 p.m. Adjournment
12-15
-------
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Public Meeting
Benefits Analysis of Secondary National Ambient Air Quality Standards
for S02 and TSP
July 27-28, 1981
AGENDA
Tuesday, July 28
9:00 a.m. Introduction
Jim Bain, Chief
Economic Analysis Branch
U.S. EPA
9:15 a.m. Agricultural Sector
Analysis
Kathleen Brennan
Mathtech, Inc.
9:45 a.m. Agricultural Sector
Discussion and Comments
11:00 a.m. Manufacturing Sector
Analysis
Ernest Manuel, Project Manager
Mathtech, Inc.
12:00 noon Lunch
1:00 p.m. Manufacturing Sector
Discussion and Comments
3:00 p.m. Overall Conclusions &
Recommendations
Jim Bain, Moderator
5:00 p.m. Public Meeting Adjournment
12-16
-------
EPA/OAQPS
Joseph Padgett
James Bain
Allen Basala
John Raines
Jan Laarman
Raymond Palmquist
V. Kerry Smith
Thomas Walton
MATHTECH, INC.
Ernest Manuel
MAJOR PARTICIPANTS
•
Director, Strategies & Air Standards Div.; EPA
Chief, Economic Analysis Branch; EPA
Project Director & Chief, Methods Development
Section; EPA
Chief, Section B, Ambient Standards Branch; EPA
Ass't Prof, of Forestry Economics; N.C. State Univ.
Ass't Professor or Economics; N.C. State Univ.
Professor of Economics; Univ. of N.C., Chapel Hill
Economist; EPA
Project Director & Mgr. Economic Analysis Section;
Mathtech
Robert Horst Senior Economist; Mathtech
Kathleen Brennan Economist; Mathtech
Professor of Economics; Univ. of Wyoming
Professor of Economics; Univ. of Wyoming
Thomas Crocker
Ralph d'Arge
A. Myrick Freeman Professor of Economics, Bowdoin College
PANEL OF EXPERTS
Gardner Brown
Anthony Fisher
Robert Haveman
Allen Kneese
Bart Ostro
Prof, of Economics; Univ. of Washington, Seattle
Prof, of Economics; Univ. of California, Berkeley
Professor of Economics; Univ. of Wisconsin, Madison
Professor of Economics; Univ. of New Mexico
Economist; EPA/OPE
George Provenzano Economist; EPA/ORD
12-17
-------
ATTENDEES
Name
Dick Atherton
John Bachmann
Lisa Baci
Jim Bain
David Barrett
Allen Basala
Tayler H. Bingham
Judith Bohan
Frank Bunyard
Vandy Bradow
Chuck Bray
Kathleen Brennan
Gardner Brown
Ed Burrows
Vega George
Jeff Cohen
Tom Crocker
Ralph d' Arge
Milt David
John M. DeMeester
Dale Denny
Bill Desvousge
George Duggan
Affiliation
EPA, Ambient Standards Branch
EPA, Ambient Standards Branch
GCA Corporation
EPA, Economic Analysis Branch
EPA, Source Receptor Analysis Branch
EPA, Economic Analysis Branch
RTI
N.C. Department of Natural Resources
EPA, Economic Analysis Branch
EPA, Environmental Criteria &
Assessment Office
Occidental Oil Shale, Inc.
Mathtech, Inc.
University of Washington
Public Interest Economics
N.C. Department of Natural Resources
EPA, Ambient Standards Branch
University of Wyoming
University of Wyoming
DPRA
Dow Chemical
Environmental Research & Technology
RTI Economics
EPA, Economic Analysis Branch
12-18
-------
Name
Affiliation
Mike Dusetzina
Anthony Fisher
Rick Freeman
Barry Gold
Linda Gree
John Haines
Terry Hammond
Geneva Hammeker
Debra Harper
Robert Haveman
Fred Haynie
Robert Horst
Richard E. Jenkins
Pam Johnson
Allen Kneese
Jan Laarman
Maureen Lennon
Tom Link
justice A. Manning
Ernest H. Manuel, Jr.
Chuck Marshall
Kenneth McCarthy
Tom McCurdy
David MeLamb
EPA, Pollutant Assessment Branch
University of California, Berkeley
Bowdoin College
MRI
MRI
EPA, Ambient Standards Branch
Environment Reporter
DPRA
EPA, Economic Analysis Branch
University of Wisconsin
EPA, Environmental Sciences Research
Laboratory
Mathtech, Inc.
EPA, Economics Analysis Branch
EPA, Ambient Standards Branch
University of New Mexico
North Carolina State University
American Petroleun Institute
EPA, Economic Analysis Branch
TVA
Mathtech, Inc.
JACA
Duke University
EPA, Ambient Standards Branch
EPA, Economic Analysis Branch
12-19
-------
Name
Affiliation
John C. Morrissey
Calvin Ogburn
Nancy Olah
Bart Ostro
Tom Pace
joe Padgett
Ray Palmguist
G. E. Pashel
Johnnie Pearson
Paul R. Portney
George Provenzano
Harvey Richmond
John Robson
Hugh Rollins
Sue Schechter
Janet Scheid
Souad Shehata
Robert Short
Roger D. Shull
V. Kerry Smith
Elaine Smolko
David H. StonefieId
Mike Teague
Henry Thomas
Yale/American Enterprise Institute
Carolina Power & Light Company
Korf Industries, Inc.
EPA, Office of Planning & Evaluation
EPA, Air Management Technology Branch
EPA, Strategies & Air Standards
Division
North Carolina State University
Bethlehem Steel Corporation
EPA, Standards Implementation Branch
Resources for the Future
EPA, Office of Research & Development
EPA, Ambient Standards Branch
EPA, Economic Analysis Branch
GCA Technology
EEA
EPA, Economic Analysis Branch
EPA, Economic Analysis Branch
EPA, Economic Analysis Branch
U.S. Department of Energy
University of North Carolina
Duke University Medical Center
EPA, Standards Implementation Branch
H & W
EPA, Ambient Standards Branch
12-20
-------
Name
C. Van Sehayk
Vincent Smith
Bill Vatavuk
George Viconivic
Tom Walton
Marianne Webb
Al Wehe
John Yocom
Earle F. Young, Jr,
Larry Zaragosa
Affiliation
MVMA
North Carolina State University
EPA, Economics Analysis Branch
GCA Technology
EPA, Economic Analysis Branch
N.C. Department of Natural Resources
EPA, Economic Analysis Branch
TRC
AISI
EPA, Ambient Standards Branch
12-21
-------
APPENDIX B
Comments from Panel of Experts in the Field of
Environmental Benefits Analysis
12-22
-------
Comments Prepared On:
BENEFITS ANALYSIS OF ALTERNATIVE SECONDARY NATIONAL
AMBIENT AIR QUALITY STANDARDS FOR SULFUR DIOXIDE
v AND TOTAL SUSPENDED PARTICULATES
For Review
July 27-28, 1981
Raleigh, North Carolina
Gardner M. Brown, Jr.
University of Washington
"Benefits Analysis of Alternative Secondary National Ambient
Air Quality Standards for Sulfur Dioxide and Total Suspended Particu-
lates," is an exceptional piece of research. The theoretical under-
pinnings of the applied economic analysis are at the forefront of the
profession. The analytical model used by Mathtech appeared in the best
economics journals only six years ago.
The general presentation is articulate, careful and quite complete.
The authors identify the plausible alternative research strategies
available at each juncture of the analysis. The explanations given for
the models adopted, the behavioral responses captured—and not cap-
tured—by the models, the variables included and excluded and the
alternative assumptions which might have been made are professionally
treated. There is a welcome tone of critical evaluation present,
surprisingly more than one finds in professional journals. The results
are presented clearly and the statistical features are capably and
critically treated as well. The repeated checks for the sensibility
of the results are laudable. ,
12-23
-------
It pays to allocate different levels of effort to different
tasks depending on the economic importance of the sectors. Such dis-
crimination is evident in the volumes reviewed. As a result the house-
hold and manufacturing reflect the highest quality of analysis.
It is apparent from reading the study and from the discussion
which ensued at the meetings in Raleigh, at which the study was re-
viewed, that there are serious data limitations which limit benefit
analysis and bring into question the precision of the estimates ob-
tained. This fact prompts a recommendation and a query. The recom-
mendation is for Mathtech to identify the missing data which in its
estimation presented the greatest bottlenecks toward reaching accurate
answers. The "holes" should be ranked and a brief explanation should
be given about the gains to be made if the four or five most highly
ranked gaps in data were filled.
In view of the fragile data base would not it have been better
to have adopted other modes of analysis, particularly those with less
voracious appetites for data? My uncomfortable judgement is no.
One alternative is to have done nothing except choose the best esti-
mates gleaned from a literature review. These estimates could not have
come from a better data base—it does not exist—and their conceptual
foundations are not conceptually superior to the ones adopted, in
my judgement.
A second alternative is to have conducted different studies with
the existing data base. I do not think the credibility of the re-
sults of the alternative studies necessarily would have been better
12-24
-------
and it well might have been much worse. Other approaches, for example,
gain simplicity simply by making even more assumptions than appear in
this study. For instance, to assume firms are governed by a Cobb-
Douglas production or cost function is to rule out the possibility of
interactions, which the Mathtech analysis shows to be statistically
significant in their analysis in the manufacturing sector. In short,
neither the alternative of doing no further analysis nor the option of
doing another type of study has unequivocal attractiveness.
Another perspective is gained by the reader considering the
scientific credibility of studies which supported earlier environmen-
tal decisions of the same order of magnitude as the secondary standards
in question. Is it superior to the credibility of the present study?
I think not, although there may be one or two exceptions. Thus, if
there must be a. benefit analysis of the secondary standards and if
there is insufficient time to enhance the data base much beyond the
present level, then my conclusion is that the overall benefit estimate
found in this study is reasonable. It appears that the authors have
taken precautions, perhaps too many, to make their estimate of bene-
fits lower than what reasonably might be expected. One exception to
this general conclusion, maybe SIC 344, fabricated structural netal
products, in the manufacturing analysis, for reasons given at the
meeting in Raleigh.
The comments which follow are directed to particular sections of
the study and are more detailed in character.
12-25
-------
Volume 1
3-3 If a firm was sensitive to pollution and had to choose a loca-
tion in within a county, what is the probability it would
locate where pollution is measured? I am concerned here and
elsewhere in the study that firms will locate where pollution
is low, not high, ceteris paribus, and pollution is not being
monitored in the low pollution areas, on average. What bins
does this create? The study cites Freeman's discussion of bias
but I did not find this issue addressed in Freeman.
3-11 Please add a figure exhibiting the discussion on this page.
3-21 What is the correlation of Temp with Sechigas?
Volume 2
4-12 We only have to have perfect information about welfare effects
if we want to have an error-free model. A weaker assumption
which permits well behaved errors will do.
4-20 Are damage functions related to changes in concentration? Why
not add "marginal" damage?
4-24 Can you ask Waddell what he meant rather than assuming what he
meant?
4-78 It would be useful to explain what "consistent" means since it
pays such an important role in choosing an aggregator.
12-26
-------
5-30 The explanations on this page seem strained. Households cannot
consistently lag in adjusting to air quality, continuously
changing in one direction. Arbitrage prevents this in a com-
petitive framework.
5-31 If there is discounting of environmental improvement, this is
testable using lagged relationships.
5-35 When prices are linear in a market, a dcr.nnd function i ^ rsti-
niated by fitting a function from prices obtained in different
markets. The cross-section provides the price variability
necessary to estimate the price slope for quality. This is
explained in G. Brown and R. Mendelsohn, "Hedonic Travel Cost
Method," University of Washington.
5-36 Explanation does not make sense. You cannot use the price
function to get infra-marginal values.
5-37 This discussion holds when the Polinsky-Shavell assumptions are
met. I suggest citing them here.
4-8ff The treatment of methodological issues Is helpful. The study
points out that people can vote with their feet in response to
pollution, moving to cleaner areas, incurring some cost, not
necessarily captured by wage or property value differentials.
The treatment of separability is very good on the whole.
I think it is a strong assumption to separate consumption from
leisure and here is a place to be more candid. Moreover, it is
(
possible to test for separability within groups and
12-27
-------
across groups. Some of the more vulnerable divisions might
be tested. Groups 2 and 3 might be aggregated or groups 2 and
4. Drycleaning might be tested for separation from clothes.
In all the sectors, there should be tests for stability
at the mean and at the observation points.
4-152 I am troubled that separability and rules of aggregation can
lead to a price increase, i.e., produce a result counter to
our expectations. Maybe this suggests that the model has
problems.
4-160 It is assumed that the primary standard is achieved in 1985
and the secondary standard met two years later. Assumed also
that any SMSA below the primary standard in 1978 remains at
the 1978 levels to 1985. Cities at secondary levels do not
receive any benefits. But what if quality would have deteriorated
in the absence? Thus this assumption understates benefits.
4-162 Are the prices implied in the income projections consistent
with the prices imbedded in the assumption that relative prices
do not change?
Volume 3
7-61 It would have been good to have a bit more explanation about the
importance or unimportance of rented capital.
7-61 Is it of any concern that the study use an opportunity cost of
7 percent for the manufacturing sector and a discount rate of <
10 percent?
12-28
-------
It would be useful to exhibit standard errors and estimated
coefficients in the TC and MC functions, at least for SIC 344
which is so critical.
7-46+ Assumption 2 assumes a doubling of input prices doubles the out-
7-98
put price. The estimated output price equation indicates that a.
doubling of input prices is not associated with a doubling of
the output price for six of the seven industry classifications.
The assumptions and facts are eternally inconsistent.
With respect to the discussion of missing data, I suggest an
experiment to get some feeling for using sample means for the
missing data. Why not remove some observations in Sector 344,
say, and put the sample mean in instead? One could then see the
effect the change has on the basic results obtained.
7-128ff Why are Ernie's elasticity of substitution estimates with re-
spect to capital, so much smaller than were found in this study?
7-126 Small shares definitely is a problem. I expect it has produced
results equivalent to concave (misbehaved) isoquants. There are
standard tests for stability which should be made.
7-178 Regional shares of output are assumed to remain constant. Is
this assumption the usual one made by federal agencies predicting
future regional output and income, and so forth? If not, your
study is internally inconsistent.
Section 8—Utilities
r
This section does not seem to be a vitally important one and I
have the sense that it was not done as painstakingly as other sections.
12-29
-------
8
For example, we are treated to a most sophisticated measure of capital
»
in the manufacturing sector, but here the measure of capital is quite
simple: capacity, augmented by capacity utilization. Age of plant
never is significant and that seems a little odd. One sentence will do
about why the elaborate approach is not necessary in this section.
8-17 Sulphur is aggregated across plants in a linear fashion if I
understand the statement on 8-17, but the aggregate outcome is
entered non-linearly in the analysis. If the stakes were high
the aggregation rule should be more carefully constructed. Also
if stack heights really mattered, I would think more carefully
about the linear rule .used. Stacks' height surely is not a con-
stant marginal cost item. Besides it too enters linearly in the
regression.
8-44 Investigators used restricted Cobb-Douglas functional form be-
cause of too many variables. They could have excluded some
variables as was done in other sections. For example, age and
wage rate were never significant so could be omitted on a trial
basis. Rain and sulphur content also could be removed. Regional
benefits assume no relocation of population forever. That is
probably a strong assumption. Note that benefits are zero in
the Pacific region and positive in the Central regions. If
people continue to move to the Pacific region, benefits will
have been overestimated.
12-30
-------
Volume 4
Agriculture
I have a strong temptation to regard this as a marginal piece
of work. If, in response, I was challenged to come up with a better
estimate of the benefits to pieces of agriculture from changing SCL
levels, I doubt I could do it. At least I am not confident I could.
Thus, judged against the terribly important standard of doing the best,
given reasonable constraints, this is a good study. Suppose, however,
I was asked to review an edited version of Section 9 for a journal.
I would reject it.
What troubles me most is the inadequacy of the data base. I
would have thought yield/acre depended on the level of capital (e.g.,
machinery) inputs. It is omitted. I would have thought that the mar-
ginal produce of labor diminished and depended on the level of other
inputs. It does not—not in the equation used for benefit estimation
(Tables 9-10, equation 2). Using regional observations for labor in
the production of soybeans (used also to produce flax seed) may be
the closest one can approximate a representative farmer's decision.
It seems like an heroic assumption to me. Regressing yield per acre
on labor per region is also a bit curious.
Data for the years 1975, 1976, 1977 are assumed to be drawn from
the same population in the yield equations. Is there adequate variability
of SCL over these years, holding region constant?
Why aren't the estimated costs to Texas cotton and soybean farmers
from decreasing SO entered into the analysis and net benefits computed?
12-31
-------
10
9-45 Why were past prices used as a proxy for expected prices when
futures prices exist for cotton and soybeans?
9-24 Would not the acreage planted be a function of costs as well as
prices, unless of course the cost of producing the crop and its
substitute are the same? (See 9-24).
Given the present data base and the results of this study,
should EPA finance studies of other crops? Or wh.it kind of
priority should such studies have in EPA's ranking of studies?
12-32
-------
Comments on "Benefits Analysis of Alternative
Secondary National Ambient Air Quality Standards for
Sulfur Dioxide and Total Suspended Particulates".
rr
by
Anthony C. Fisher
University of California, Berkeley
Presented at a public meeting put on by the
Environmental Protection Agency in Research
Triangle Park, North Carolina
July 27-28, 1981
12-33
-------
Comments on "Benefits Analysis of Alternative
Secondary National Ambient Air Quality Standards
for Sulfur Dioxide and Total Suspended Particu-
lates"
by
Anthony C. Fisher
This five-volume study represents a very significant advance in the theory
.T
and method of estimating the benefits from air pollution control.
For some time I have been presenting this topic as involving two basic
approaches: 1) estimating (the reduction in) physical damages as a function
of a change in air quality and then imputing a value, and 2) estimating in one
step the relationship between a change in air quality and a market value, such
as property value. To these one could add a third, less widely used: asking
individuals what they are willing to pay for designated improvements in air
quality. The first is the approach generally taken by non-economists, and as
the authors of the present study argue, is likely to be inconsistent with a
theoretically preferred approach to benefit estimation (one that gets at
»
willingness-to-pay for a change). Further, it is beset by severe data problems.
The second has been used by economists, including the authors of this study. It
is consistent with theory, but is perhaps not well suited to the task at hand
in this case. The difficulty is that in principle it captures health as well as
non-health benefits, and the charge of the present study is to estimate only
the benefits of moving from primary to secondary standards, i.e., only the non-
health benefits. The third approach has the obvious drawback of depending on
response to hypothetical questions, rather than observed behavior.
The great merit I see in the main body of the present work is its develop-
ment and application of a new approach, one that is consistent with the theory
12-34
-------
-2-
of benefit estimation in that it does get at willingness-to-pay for air quality,
discriminates among classes of benefits in such a way that it (in principle)
captures only the benefits of moving from primary to secondary standards, and
depends on observed behavior in markets for conventional goods and services.
In the more detailed remarks that follow I shall raise a number of questions
concerning still (to me) unresolved theoretical issues, and the relationship of ,
rr
the new, "sector" approach to the existing approaches to benefit estimation.
The focus is mainly, though not exclusively, on the household sector, because
this is where I see the more challenging remaining issues. Also, much of the
previous work by economists, such as property value studies, has been in this
area, so questions about the relationship of the present work to previous
approaches are specially relevant for the household sector.
I want to emphasize, before proceeding, that the questions and comments
are offered in a spirit of constructive criticism of a study I regard as
making a significant contribution to the field of environmental economics.
One set of questions has to do with the modes of adjustment to pollution
*
by those who suffer from it, and the ways in which adjustment behavior is
reflected in the theoretical and empirical analysis. The authors point to their
analysis here as important and innovative. I agree, and feel this gives special
urgency to unresolved questions.
A first such question I would raise, then, is the following. In treating
(as on pp. 2-27 and 2-28) the value of a reduction in pollution as the change in
(consumers' and producers') surplus, how do the authors capture the effect of a
switch to a different product or, more likely perhaps, a different crop? Figure
2-6, for example, is clearly valid if there is no switch. But if there is,
shouldn't one look at the net change in profits, or site rents, either of which
12-35
-------
might capture the higher value of the new crop? As far as I can tell, this
case is*not explicitly treated either here (chapter 2) or in the empirical
analysis of the agricultural sector (chapter 9, pp. 34-77).
My suggestion, then, is to try the property value approach, relating ambient
concentrations of particulates or sulfur oxides to agricultural property values.
In the agricultural sector, as opposed to the household sector, comparative
rf
property values are not likely to reflect (human) health impacts, and so are
appropriate for estimating ..the benefits of moving from primary to secondary
standards.
A related question concerns the ability of the model to pick up benefits
in a situation where there is np_ adjustment by the pollution receiver. Sup-
pose a decrease in sulfur dioxide or total suspended particulates does not
lead to a decrease in household demand for laundry and cleaning products;
people just live in cleaner surroundings. Does the model pick this up as an
increase in utility in the household sector, as it would an analagous increase
in output in the agricultural or manufacturing sectors? Or do the authors per-
haps underestimate benefit to the household sector for this reason? Another way
of putting the issue here is to note that we are getting into esthetics, which
the model is not designed to pick up. Of course esthetics might be picked
up, but only if people change their market behavior in some fashion.
A related issue is the following, raised on p. 4-11 and perhaps elsewhere.
The authors note there that "failure by the analyst to consider the range of ad-
justment opportunities...can lead to an overstatement of benefits". I believe
the outcome is just the reverse. The value of pollution abatement: is greater
if the receiver is free to make adjustments than it would appear to be if the
potential for making these adjustments is ignored. On the other hand, the damages
12-36
-------
from an Increase in pollution would be overstated, to the extent that the esti-
mation does not take account of possibilities for mitigating them by defensive
actions.
Both effects are easily demonstrated graphically. On Figure 1, which de-
scribes a reduction in ambient pollution concentration, the implicit price of
cleanliness is reduced from pQ to pr If no adjustments, or substitutions, are
possible, the demand for cleanliness is inelastic and benefits are estimated as
ABCD. If, on the other hand, adjustments are possible, demand is relatively
elastic, and benefits are estimated as ABCE. The difference, DCE, is the amount
by which benefits are understated if adjustment possibilities are ignored. A
similar analysis in the case of an increase in pollution levels shows the damages
to be overstated by the amount DCE on Figure Z. For a somewhat different and
more detailed exposition, see Zeckhauser and Fisher (1976).
Figure 1. Reduction in Ambient
Concentration
Figure 2. Increase in Ambient
Concentration
12-37
-------
Still another related point: on p. 4-12 the authors note four ways in
which the "null response" may arise. I think there is a fifth, which is likely
to be important in practice: the costs of adjustment exceed the perceived damage
This point is implicit in a discussion (p. 4-13) of the household's budget con-
straint, but even with a perfect capital market, a response would not be ex-
pected, or observed, if costs exceeded benefits. This may be a point of some
(
general relevance, to the property value and wage studies as well. If there are
costs of movement, or of course other barriers to movement, neither property
values nor wages will fully reflect the benefits of an air quality improvement.
Or have I missed something?
A second set of questions, or comments, has to do specifically with the re-
lationships among benefit estimates based on (a) the "sector" approach, (b) the
property value approach, and (c) the wage rate approach. First, as between
the existing approaches, (b) and (c). I understand these are not the authors'
main concern,but a number of loose or vague references scattered through the
text might be further considered. Freeman (1979, pp. 118-121) suggests that
•
under plausible conditions, such as qualitative differences in natural resource
endowments or non-constant returns to scale, factor price (wage) equalization
among regions will fail, and benefits of an air quality improvement are then the
sum of the absolute values of the changes in wage payments and residential land
rents. I get the impression, though, again, the treatment of this question in
the text is rather loose, that the authors see property value and wage changes
as somehow separate (and unequal?) estimates of the same thing. Perhaps they
do not, but at a minimum the discussion ought to be made more precise.
With respect to the relationship between the sector and property value, or
property value-plus-wage approaches, the latter is an upper bound to the former
12-38
-------
only if health effects and benefits are accurately perceived, and then fully
•'^O
reflected in property/wage values. I have suggested they may not be due to
cost or other barriers to mobility. Further, it is my impression, shared I
believe by many people in the field, that health effects are not in fact
accurately perceived, particularly where they result from chronic, low-level
exposure to a pollutant. It is for this reason I have argued elsewhere for
a separate estimate of the health benefits of pollution control.
Now let me raise a number of questions relating to the empirical pro-
cedures or findings of the study. Do the authors find it disturbing, as I
do, that the benefits of attaining the secondary standards as they calculate
them on the basis of either property value or wage studies are orders of mag-
nitude greater than the same benefits calculated on the basis of the (house-
hold) sector approach? (The disparity is of course still greater if, as I
have suggested, it is the sum of property value and wage changes that measures
the benefits.) Again, as I have suggested, these alternative techniques may
fail to capture the health benefits also not captured by the sector approach,
•
and may underestimate such benefits as they do capture. Further, even if health
benefits are captured in property value or wage studies, they are presumably
exhausted by the move to primary standards. Does the disparity between house-
hold sector and property value/wage estimates suggest the existence of very
substantial remaining health benefits from the move to secondary standards?
Or is it all esthetics? Or could the sector approach be substantially under-
estimating benefits, perhaps for the reasons I have suggested? My guess is that
all three effects are present, and should be noted as explanations of the wide
disparity in results.
A quite unrelated question I have is about discounting procedure. It is
12-39
-------
not clear to me whether the estimates presented represent the present value,
in 1980 dollars, of annual benefits of moving from primary to secondary standards,
•
as of (I think) 1987, or rather the present value, in 1980 dollars, of a stream
of annual benefits beginning in 1987—and if the latter, how long a stream.
Also, if the latter, why? I believe that the other studies, to whose results
the authors compare their own, present only annual benefits. It is not clear
rf
to me how to interpret, for example, the comparison with Freeman on p. 4-173,
until this question is clarified.
A related comment is that a 10% real rate of discount, the one emphasized
in the text and (especially) the tables, seems much too high, even in today's
fevered financial conditions, and certainly (one hopes) in the longer run.
Results based on the 2% and 4% rates used for purposes of sensitivity analysis
ought to be at least as prominently displayed.
Finally, a comment suggested by the study concerning directions of future
research. Probably one of the most profitable would be more extensive physical
monitoring of ambient concentrations. I vaguely recall hearing recently of
•
some promising new low-cost techniques. And in the absence of monitoring, the
ability of environmental economists to build on the work represented in this
study is greatly impaired. For example, the authors note (p. 2-45) that sulfur
dioxide, the more important pollutant in terms of effects on vegetation, is
monitored in only 10% of U.S. counties, thereby precluding analysis of the
forestry and fishery sectors.
* * *
Concluding Remarks
The study is to be commended for its introduction of new and sophisticated
methods for estimating the benefits of air quality, and perhaps equally, for its
12-40
-------
-c-
readable discussion of these methods. The question can be raised as to whether
the theoretical development is worthwhile, or appropriate, since data limitations
permitted application only to fractions of the several economic sectors dis-
tinguished by the authors. My answer is that the development is both worthwhile
and appropriate. It leads us to identify gaps in the data and coverage. As
these gaps are filled, the benefit estimates that can be generated will be
increasingly plausible, and defensible.
A second, related point, is that the conservative assumptions employed
at various places in the study, and the other sources of under-estimation of
benefits identified in my comments, combine to suggest that the..resulting esti-
mates are lower bounds. If, as I learned informally at the close of the public
meeting, costs of achieving the secondary standards are still lower, we can be
confident that the standards are justified.
References
Freeman, A.M. The Benefits of Environmental Improvement: Theory and Practice.
Baltimore: Johns Hopkins University Press, 1979.
Zeckhauser, R.J. and A.C. Fisher. "Averting Behavior and External Diseconomies",
Kennedy School Discussion Paper 41D, 1976.
12-41
-------
Comments on "Benefits Analysis of Alternative
Secondary National Ambient Air Quality Standards
For Sulfur Dioxide and Total Suspended Particulates"
July 27-28, 1981
Robert H. Haveman
University of Wisconsin-Madison
A question of longstanding scientific importance concerns the
basis on which to appraise research which purports to add to the stock
of knowledge. This question is what this conference is about; these
comments w'ill venture an answer as well.
Several bases for appraising research — especially, applied
microeconomic research — come immediately to mind. They include:
(1) Do the empirical models rest on a solid theoretical
base? Does the economic theory used represent that at
the frontier of the discipline? Is the theory
interpreted correctly and is it employed appropriately
in specifying the empirical models?
(2) Are sound econometric methods used in estimating the
models? Is the basis for choosing the techniques used
rather than alternative approaches fully explained?
(3) Are appropriate data used in the estimation? Was
superior data available but not employed, and, if so,
why not? Are the data sufficient to the task?
(4) Were the many judgments and decisions necessary in
empirical research made judiciously, sensibly, and
with good reason?
(5) Were the empirical results interpreted correctly? Was
the sensitivity of the results to alternative
assumptions tested? Are the limitations of the
estimates — and their dependence on particular
assumptions — spelled out clearly?
(6) Are the results plausible? Is their relationship to
other scientific research explored? Are they used
appropriately in discussing policy implications?
12-42
-------
In my comments, I will appraise the study on each of these
criteria. To begin, let me offer a bottom-line: The MATHTECH Report
is a very solid piece of work. While the findings are clearly less
than optimal for making a policy judgment, they are the most
comprehensive and easily the most defensible ones in existence. They
rest solidly on economic theory and derive from the application of
appropriate modern econometric methods. While they are plausible, I
would find it more difficult to say that they are believable.
However, if I had to believe something about the economic benefits
stemming from reduced soiling and materials and crop damages of
attaining secondary standards, I would believe these estimates.
Let us consider the study in terms of each of the criteria noted
above: First, the microeconomic theoretical basis for this study is
precisely on target. Measuring the welfare gains from policy actions
is a tricky business. It requires the careful specification of
utility functions and the precise definition of the measures of gains
and losses to be employed. These welfare gains are not observed, and
must be inferred from underlying behavioral responses to economic
stimuli. The authors of this study have done a very careful job with
this component — they lay out clearly the underlying theory, state
openly the implied assumptions and constraints required to develop
empirical estimates of the theoretical concepts, and specify the
limitations of the estimates even if the empirical work were to
conform to theoretical requirements. They deserve an A on this
criterion.
12-43
-------
As with the first criterion, the second — concerning appropriate
econometric methods — is also well-fulfilled. The methods used are
advanced and appropriate to the task. And, by and large, they are
used artfully. Especially in the household analysis, the approach is
sophisticated and "state of the art". In terms of both theoretical
underpinnings and econometric sophistication, this analysis should
find its way into the professional literature. Econometric methods
again deserve an A.
The third criterion concerns data and its use. As the criterion
is phrased, the judgment regarding the absolute quality of the data
must be separated from the judgment on its use. In my view, the
authors have employed the best — and in many cases — the only data
relevant to the estimates they require. Indeed, in a few instances,
the search for data, the creation of proxies where data did not exist,
and their construction of data bases assembled from information from a
variety of sources was most creative. The development of input prices
and capital inputs and values in the estimates for the manufacturing
sector is a case in point.
Judgment on the overall quality of the data, however, is quite a
different story. In several instances, the comprehensiveness and the
quality of the estimates have been severely compromised by the
availability of relevant data. For example, in estimating the
corrosion and other materials damage costs in manufacturing
industries, the lack of appropriate data has sorely constrained the
number of industries which could be analyzed and the significance of
the industries included. Some of the sectors most likely to be
12-44
-------
adversely affected by air pollution — for example, the steel industry
— were excluded for data reasons. In terms of the collection and
creation of appropriate data, the researchers get high grades.
However, data constraints which they faced — in no way their fault —
does impose serious limits on the reliability and comprehensiveness of
their estimates.
The constraints imposed by the data raise a quite fundamental
question regarding the research strategy chosen for this sort of
study. Would one have chosen the empirical approach suggested by
frontier economic theory if one knew ex_ ante that the data constraints
relevant to that research option would severely limit industry
coverage, such that all but about 4 percent of the manufacturing
sector (including some of the most pertinent industries) would have to
be excluded from the analysis and that extrapolation based on heroic
assumptions would be required to build an estimate of benefits for but
25 percent of the manufacturing sector? Is it possible that a less
sophisticated, more straightforward empirical approach, but one
without the theoretical underpinnings of that chosen, could yield
estimates which are at least as plausible as the final, extrapolated
estimates of the approach chosen? Clearly, some trade-off exists
between theoretical and econometric sophistication requiring poor or
unavailable data and a less pure, less sophisticated, less
theoretically sound analysis which has fewer data needs. I have some
doubts as to whether the study authors chose efficiently along this
continum.
12-45
-------
This comment is a natural lead-in to an appraisal of the numerous
judgments and decisions made by the researchers in executing the
project — the fourth criterion. By and large, I have little trouble
on this score. The numerous operating decisions taken by the
researchers appear to me to be based on clear thinking, common sense,
and a recognized need to press through to a bottom line estimate.
Having undertaken the research and developed the estimates, the
next question concerns their interpretation. Are the uncertainties
and potential biases created by the numerous assumptions and data
limitations accounted for in the interpretation of results? Or do the
researchers come to believe in their final estimates once they have
been placed on paper? It is with respect to this issue that the
report can be moderately faulted. Perhaps my main objection on this
score concerns the tendency to believe in the point estimate on the
pollution variables if statistical significance is found. My concern
here rests upon several considerations — the classical omitted
variables can serve as an illustration. The bulk of the benefit
estimates in the report can be traced back to some regression in which
some dependent variable (e.g., industry costs) is related to some air
pollution measure. The issue is: Are there phenomena which may
affect industry costs which are not included in the regression and
which may be correlated with air pollution levels? If such phenomena
do exist, the coefficient on the pollution variable will be capturing
their effect as well as that of air pollution.
Consider, for example, the measured effect of air pollution (TSP)
on total cost in the fabricated structural metal products industry
12-46
-------
(Table 1-25). For the primary model estimated, the TSP variable is
significant at about the 0.04 level. In this model, TSP enters, with
the S02 variable omitted. The other explanatory variables are the
prices of labor, capital, and materials inputs, and climatological
variables. The regression is fit over 57 observations which are
counties with available industry output and input prices, and air
pollution and climatological data. Most of these counties represent
urban areas. The question is: Are there phenomena associated with
industry costs which are not captured by the included variables (e.g.,
input prices or air pollution)? Some reflection suggests that there
may well be. Urban problems — air pollution, crime, high poverty
incidence, low property tax base, high urban infrastructure costs —
tend to come in bunches and are often associated with the size and
nature of the metropolitan area. All of these phenomena are likely to
(1) adversely affect industry costs (if in no other way than through
local taxation levels) and (2) be correlated with air pollution
levels. All are uniformly excluded from the regression. As a result,
the magnitude of the coefficient on the air pollution variable will
reflect the cost impact of these phenomena as well. To the extent
that it does, the effects of TSP on industry costs will be biased (and
probably overstated) — and the benefits of reducing TSP levels are
likewise biased.
In the report, little recognition is given to this very real
possibility, and — after some caveats — typically the coefficient is
accepted at its face value. This, I would suggest does not represent
12-47
-------
a careful and thoughtful interpretation of results. On this fifth
criterion, then, I can award them no better than a B-.
The final criterion is, in some sense, the nub of the issue: Are
the results plausible and believable? Answering this question is
difficult, as plausibility is an elusive concept. What is it that
causes one to believe or not believe research results? Some of these
factors are clear, others are not. Common sense, institutional
knowledge, comparison of estimates with known related values,
comparison of results with those of other studies, an assessment of
the limitations and biases imposed by assumptions and constraints
adopted in the research, and an ability to trace and verify the
various steps in the analysis are all elements in establishing
plausibility. At one level, I am inclined to view the overall results
of the study as plausible. The magnitudes are not enormous, the
researchers have done an admirable job in comparing their findings
with those of other studies, the methodological basis is solid, and by
and large, most of the separate assumptions and decisions are
reasonable. At another level, however, I am troubled. The deeper one
probes into some of the estimates, the more one is impressed by the
tenuous nature of the terrain.
Consider, as an example, the major component of the benefits of
achieving the secondary standards. The benefits from achieving the
secondary TSP standard in the fabricated structural metal products
sector totaled $3.7 billion of the total of $4.5 billion of estimated
total manufacturing sector benefits; $11.4 billion of $15.9 billion of
total extrapolated manufacturing sector benefits for the TSP secondary
12-48
-------
standard, and $11.4 billion of $22.0 billion of total benefits for the
household, agricultural, manufacturing and electric utilities
industries for both the SC>2 and TSP standards. That the estimate for
this small industry, with $6.7 billion of annual 1972 value added —
0.6 percent of GNP — should, when extrapolated, account for over one-
half of the total estimated national benefits of the secondary
standards seems somewhat remarkable.
The methodology applied in obtaining this estimate involves a
regression estimate of the effect of TSP on production costs in this
industry, after accounting for input prices and quantities and
climatological factors. The following is a partial listing of the
constraints imposed and assumptions made in deriving this estimate:
• The production function in the industry is assumed to
be weakly separable.
• Prices of inputs and level of output in the industry
is assumed to be exogenously determined.*
• Climate and air quality are assumed exogenous to the
firms.
• The production function is taken to be a transcen-
dental logarithmic function.
• Firms in the industry are assumed to be cost-
minimizers.
• Symmetry (values of cross partial derivatives
independent of the ordering) is imposed.
• Input prices for all variable inputs are homogeneous
of degree 1.
* Note that independent evidence cited in the report indicates that
wage rates are a negative function of air pollution levels.
12-49
-------
• Elasticity of total cost with respect to TSP and S02
is assumed to be independent of pollutant concentra-
tion.
• Variations in rainfall are assumed to be independent
of input cost shares.
• The effects of temperature and rainfall are con-
strained not to interact with each other.
• The effects of the fixed factors are assumed to be
independent of firm size.
• The cost function estimated is taken to represent the
fictional "average" establishment in the county.
• All non-production labor inputs have been excluded
from the analysis.
• Supplemental labor costs (fringe benefits) have been
excluded from the analysis.
• Capital stocks and capital service prices for each
industry and county are defined and measured so as to
(1) exclude land, (2) exclude rented assets, (3) meet
the constraints of a particular perpetual inventory
formula, (4) accommodate inter-censal data gaps, (5)
accommodate missing capital expenditure data, (6)
presume that the benchmark year of capital accumula-
tion is 1953, and (7) fit a particular concept of the
price of capital services, which assumes the cost of
money equals the average yield on industrial bonds in
1972, average tax life of assets is 20 years, the
average effective corporate income tax rate is con-
stant over industries, the depreciation rate is the
average of that used in two other studies, and
regional asset prices are estimated crudely from
partial components of the Producer Price Index.
• Materials inputs and prices across industries and
times were constructed from less than ideal data and
(1) assume national industry cost shares apply to all
regions, (2) price indices for 38 percent of materials
inputs are assumed to hold for all inputs, and (3) the
index for 1958 to 1972 rests on regional prices only
through 1963.
• Equilibrium output prices are estimated as a Cobb-
Douglas unit cost function using components of the
Producer Price Index which crudely match the specific
industry, a variety of functional forms, and input
prices and time as independent variables. Regional
12-50
-------
variation was not introduced into the output price
estimates.
Air pollutant and climatological variables, by county/f
were used, after correcting for missing data, by
inserting mean values. Only the second average high
measurement is used for TSP.
Quadratic and interaction terms involving air
pollution and climate variables were excluded in the
cost estimations with stable partial derivatives of
total cost.
Benefits are calculated for each county for each year
from 1978 to 2050 discounted at a 10 percent discount
rate, and aggregated up to obtain a national total.
The reduced estimated cost function without the SC>2
variable was employed and output forecasts for the
larger SIC industry of which this industry is a part
were obtained from the Wharton model.
Given this set of assumptions and imposed procedures, the
question of the plausibility of this dominant estimate of pollution
control benefits must be answered. While, in some sense, it is not an
extraordinary figure, there are substantial reasons for doubting its
accuracy — its magnitude relative to that of other industries, the
fact that it applies only to corrosion and soiling damages (in a
context in which S02 was found to be statistically insignificant), and
the fact that the benefits of achieving the secondary standards are
2.6 percent of production costs when total expenditures in this
industry on maintenance and repair equal 2.9 percent of total payroll.
While this estimate appears to be among the most implausible of those
reported in the study, it is also the largest and the dominant
contributor to the overall assessment of the benefits of achieving an
improvement in air quality. However, having said this, it would not
be surprising if many of the same kinds of concerns would surround
12-51
-------
other components of the estimates were they to be given close
scrutiny. As to plausibility, then, I assign a B-, but with a large
standard error.
After all this, then, what can be concluded? The report is an
ambitious effort to tackle a question which may be empirically
intractable given the current state of data. It rests on a firm
theoretical underpinning, and data collection and modeling has been
vigorously guided by appropriate, state of the art econometric
techniques. The data has been gathered carefully and analyzed
vigorously. The interpretation of results is somewhat unsettling in
that more confidence is placed in estimates than they appear to
warrant. More sensitivity analysis should have been done and
displayed. And, while the reported results are not unreasonable, it
is difficult to say that they are reliable point estimates. A
doubling of some of them or a halving of all of them, or the placement
of values where none were assigned would not make them more or less
plausible. They are, nevertheless, the most comprehensive available
— and their dependence on state-of-the-art theory and empirical
techniques makes them uniquely defensible.
12-52
-------
Comments by Allen V. Kneese on
Benefits Analysis of Alternative Secondary National Ambient
Air Quality Standards for Sulfur Dioxide and
Total Suspended Particulates
EPA, July 1981
My general impression is that this document reports on a very
competent piece of research. In general, its important strong points
are:
(a) The analyses take account of the fact that man can
adapt to pollution situations. Most other studies do
not do this.
(b) The analysis is based on appropriate, carefully
developed, economic models.
(c) The empirical work is technically sophisticated and
sound.
(d) The report is very well written.
I do, however, have a number of questions and doubts (these are
not in order of priority).
(a) The study accepts the legalistic fiction that primary
standards completely protect against health effects,
This is most likely not true and starting from this
premise could seriously bias the interpretation of the
study's results.
(b) It counts no benefit for maintenance of high quality
air in areas which already have air as good as, or
better than, the secondary standards. This could be a
substantial source of downward bias, especially in
rapidly developing areas of the west.
(c) This is not a fault of the research under review but,
it should be noted, it excludes categories of benefits
that may be much larger than the ones included —
these may include health, aesthetic, and nonuser
benefits.
12-53
-------
(d) Despite the claim that it does not include acid
effects, the study did implicitly do so. The
statement is frequently made in the report that SC>2
does this or does that, where in fact, it is not S02r
but the actual damage often results from 803 and
H3S04.
(e) This leads to an instance of what may be a rather
general deficiency in the report — inadequate
attention to the use of prior knowledge in the
specification of estimation equations. For example,
S02 and N02 may be highly correlated precursors of
113804 and HNC>3. In some areas HNC>3 may be a large
source of acid damage. If HNC>2 is not included, some
of its actual effects might be picked up by the sulfur
variable. I understand that there are data problems,
but specification issues are potentially so serious
that I think they need greater attention in the
report.
(f) Another general matter is aggregation of variables to
the county level. I think more systematic treatment
of what this means both conceptually and empirically
is needed. For example, in the industry studies, can
the county level data take adequate account of such
potentially important factors as technological design
of equipment, plant layout, degree of vertical
integration, etc.? Given the aggregation and crudity
of data, can much credence be given to the small
effects found, even though in the formal statistical
sense they are "significant"?
A few more specific comments:
(a) On the household studies •— others have concluded from
surveys that people do not alter their cleaning
behavior when there is more pollution, but simply live
in dirtier surroundings. Since the disutility
experienced because of this cannot be captured by the
method used in the study, a large benefit may be
missed.
Second, since the available measures of atmospheric
pollution are often poor reflections of actual ambient
conditions, and since the exposure of home furnishings
and equipment occurs indoors, it is hard to see how
there could be any real connection between those
measures and damage to household items.
12-54
-------
(b) On the manufacturing study, I have already mentioned
problems that might be associated with county-level
aggregation. But cost differentials may be more
complex than the study appears to recognize. Hoch has
found systematic wage differentials between large and
small cities for what he took to be all sorts of
environmental reasons (e.g./ traffic congestion, long
commuting times, crime, high levels of pollution). If
S02 correlates highly with city size, might not these
analyses be attributing cost differentials to S02 than
actually related to other things also? I realize that
pollution and wages were controlled for, but it may be
that other "urban phenomena" also affect costs. This
might be tested by including some sort of an urbaniza-
tion variable.
(c) Utilities — one might think they would be more
affected by their own pollution than by air ambient
condition possibly measured some distance away.
12-55
-------
APPENDIX C
Comments from the American Iron and Steel Institute
12-56
-------
STATEMENT
OF
EARLE F. YOUNG, JR.
VICE PRESIDENT
ENERGY AND ENVIRONMENT
AMERICAN IRON AND STEEL INSTITUTE
AT
EPA PUBLIC MEETING
ON
"BENEFITS ANALYSIS OF ALTERNATIVE SECONDARY
NATIONAL AMBIENT AIR QUALITY STANDARDS
FOR SULFUR DIOXIDE AND TOTAL
SUSPENDED PARTICULATES"
BY
MATHTECH INC.
JULY 28, 1981
12-57
-------
My name is Earle Young. I am Vice President, Energy and Environ-
ment for the American Iron and Steel Institute. AISI is a trade associa-
tion whose 65 domestic member companies account for 91% of the nation's
steel producing capability. I am extremely gratified to have this oppor-
tunity to present some comments relative to a technical report containing
benefit analysis methodology related to the National Ambient Air Quality
Standards.
AISI has long been an advocate of the use of cost benefit analy-
sis in the area of environmental regulations. In our major policy state-
ment, "Steel at the Crossroads: The American Steel Industry in the 1980s,"
published in January 1980, AISI said in discussing the Clean Air Act:
"All welfare related requirements should, at a minimum, be subject to
rigorous cost benefit analyses and local options." Therefore, the report
being discussed today, "Benefits Analysis of Alternative Secondary National
Ambient Air Quality Standards for Sulfur Dioxide and Total Suspended Par-
ticulates," is of great interest to us.
The steel industry is very familiar with the cost side of the cost
benefit considerations in the steel industry. Just today, we have released
a report by Arthur D. Little, Inc., entitled "Environmental Policy for the
1980's: Impact on the American Steel Industry" which shows that the steel
industry has already in place $5 billion dollars worth of equipment (in
1980 dollars) to control air pollution and will in the next few years be
spending another $1.3 billion essentially to achieve the present primary
standards. That report also shows that if the industry is forced to make
additional expenditures directed at attainment of the secondary standards
it will have to spend by the end of this decade another $3.8 billion.
Thus, we do have what we consider is a good estimate of the costs of achiev-
ing the secondary standard for particulate.
The other half of the cost benefit consideration is evaluation of
the benefits. The steel industry has long recognized that it is much more
difficult to evaluate the benefits of achieving air quality standards than
to estimate the costs. Therefore, we welcome efforts to develop a quanti-
tative methodology which will put those benefits on some sort of a sound,
realistic basis.
12-58
-------
Since the primary pollutant emitted by the steel industry and the
one we have spent and will spend the greatest amount of money to control
is total suspended particulates , we were particularly interested in that
portion of this report dealing with the particulate standard.
A quick analysis of the report showed that it indicates a total
benefit value of about $18 billion dollars for achieving the secondary
standard for total suspended particulate, of which $16 billion can be at-
tributed to benefits in reduced costs to the manufacturing sector. By a
complex mathematical analysis, the contractor has developed a relationship
which indicates that manufacturing costs , particularly in the metal fabri-
cating and machining industries, are higher where suspended particulates
in the air are higher.
The mathematics involved are quite sophisticated. AISI does not
have the expertise internally to comment on them. Therefore, we have re-
quested an outside expert to review and comment on the methodology and
mathematics. Following my presentation, Dr. Dale Denny of ERT, Inc., will
present a more detailed analysis of the methodology.
A major point that I would like to make, however, is that the con-
*
tractor ' s report makes no attempt to distinguish between correlation and
causation. It is not particularly surprising to me that manufacturing
costs are somewhat higher in areas where the total suspended particulate
content in the air is somewhat higher. Generally speaking, you can expect
higher particulate levels in the air in concentrated urban areas, where
there are greater concentrations of people and traffic, where there are
older manufacturing facilities, where there tends to be an older, more
unionized labor force, and where there tends to be a higher cost of living
and higher tax rates. These are exactly the same factors which one would
expect to result in higher costs of production. The older plants located
in these more polluted areas are precisely the plants which do tend to
have higher production costs. Therefore, I repeat, I don't find it sur-
prising that some relationship may be indicated or calculated between pro-
duction costs and the average level of air pollution in the areas. On
the other hand, I see nothing in that sort of correlation which implies
causation, nothing that indicates it is the air pollution which leads to
these higher production costs. The authors point out deep in Chapter 7
12-59
-------
of their text that their findings, "do not prove the existence of a cause
and effect relationship."
A well known statistical correlation exists between the stork
population and the human birth rate in Rothenstead, England; yet, I haven't
believed for a long time that there is a cause and effect relation between
stork population and human birth rate. In his classic paper, "Use and
Abuse of Regression," George E. P. Box discusses the phenomenon of "non-
sense correlation" and discusses "regression variables and latent varia-
bles" and points out that regression coefficients can be "utterly mislead-
ing. "
Yet, the entire statement of benefits, the attribution of a value
of $16 billion as a benefit attributable to the attainment of secondary
standards, depends on the existence of such a cause and effect relation-
ship.
Thus, we are seriously concerned that the methodology used in this
study has resulted merely in an apparent correlation and that, in the in-
terpretation of the study, that correlation will be construed as causation.
Benefits could be attributed to the secondary standards which, in truth,
do not in any factual sense relate to the attainment of those standards.
We ask that you give very serious consideration to this question
of correlation versus causation before any use is made of the potentially
misleading results generated in this study.
I would now like to introduce Dr. Dale Denny of ERT who will pre-
sent a critique of the study and its methodology.
12-60
-------
TESTIMONY: A CRITIQUE OF EPA
COST BENEFIT ANALYSIS ON
SECONDARY AIR QUALITY STANDARDS
ERT Document No. P-B129
Prepared for
American Iron and Steel Institute
1000 16th Street, N.W.
Washington, D.C. 20036
Environmental Research & Technology, Inc.
601 Grant Street, Pittsburgh, PA 15219
July 28, 1981
12-61
-------
Table of Contents
o Introduction
o Contractor's Premise and Choice of Data Base
o Contractor's Regression Analysis of the Data
o Statistical Significance of Contractor's Final Results
o Appendix: Specific Models for SIC344
12-62
-------
INTRODUCTION
My name is Dr. Dale A. Denny. I am Senior Program Manager with the
Environmental Engineering Group of Environmental Research and Technology,
Inc. (ERT). I was formerly a technical manager in the U.S. Environmental
Protection Agency, Office of Research and Development. I and my
colleagues at ERT have reviewed the subject contractor report, "Benefits
Analysis of Alternative Secondary National Ambient Air Quality Standards
for Sulfur Dioxide and Total Suspended Particulates" and wish to offer
comment on the study and its conclusions. The focus of our comments are
on the parts of the report that address information from the Manufacturing
Sector, Volume 3, Section 7 and Volume 5, Section 10. Six 3-digit SIC
categories are evaluated in the Manufacturing Sector, but our emphasis
addresses SIC 344 (Fabricated structural metal parts).
We appreciate the difficult task that faced MATHTECH, Inc. in its
assignment; we recognize that they have tried to make use of state-of-the-
art econometric models, but we seriously question whether or not this
modeling approach properly fits into this benefits analysis. The
contractor has used a multiple regression technique with the premise that
there is a correlation between air quality and costs of production in the
manufacturing sector. There is no evidence to display this relationship
on a cause-and-effeet basis, and we believe the contractor is aware of
this lack. However, the report concludes that, in spite of a lack of
knowledge of cause-and-effeet, there is a positive correlation between
production costs and air quality and then assigns dollar benefits to the
air standard on the basis of this correlation. We do not believe that the
contractor has shown the confidence that one can assign to this
correlation, nor has he shown the uncertainty that accompanies the use of
his production cost model in calculating marginal costs due to incremental
changes in air quality. Until these state-of-the-art, unproven
econometric models have been otherwise verified, we do not think that the
models can be used and extrapolated to calculate benefits on a national
scale. Fundamentally we also have serious problems with (1) the
contractor's choice and manipulation of data bases for the study and
12-63
-------
(2) the contractor's efforts to achieve a solvable regression and to
demonstrate the mathematical significance of his results. Unless and
until the EPA and its contractor can resolve adequately these
reservations, we question the technical significance of MATHTECH's report
as a basis for assigning any quantitative measure of benefits to
attainment of the secondary air quality standards. We believe that when
the EPA and its contractor take into account fully the variability in the
data as employed in the subject study and express accurately the standard
errors in final study results, no statistical significance will be evident.
What follows is a discussion of specific problems we believe cast
serious question on the usefulness of the study results. Remarks focus
first on the contractor's premise and choice of data bases. Comments are
then directed to the contractor's regression modeling. Finally, comments
are made regarding the statistical significance of the contractor's
findings.
CONTRACTOR'S PREMISE
AND CHOICE OF DATA BASE
The contractor's basic premise is that (1) a statistically significant
portion of the variability in national statistics on SIC 344 manufacturing
costs can be explained by variations in ambient S0_ and TSP air quality
(2) such a relationship can be shown to exist at a county-level of
aggregation of air quality and manufacturing cost data. There is a large
set of factors which is known to contribute to wide variability in
measured local air quality levels, e.g., rural vs. urban station readings,
proximity of station to industrial vs. other sources (power plants, road
dust, cars). There is also a large set of factors which the industry
knows contributes to observed variability in member company SIC 344
manufacturing costs, e.g., cost of labor, cost of materials, cost of
capital, age of facility and its various process units, geographical
location. It was the contractor's challenge either to address these
12-64
-------
factors and,their interpendencies directly, or to demonstrate a procedure
for filtering out these dependencies to permit a direct observation of the
degree of impact of variable air quality levels on manufacturing costs.
We believe that the contractor has inadequately tried the second approach,
and has not displayed in his report the data that would demonstrate the
many simplifications, and assumptions, that were necessary to make the
problem mathematically tractable. In most cases it appears to us that the
contractor's efforts addressed questions regarding how to make the
mathematics fit the problem rather than the question of the
appropriateness of the model and the assumptions for the problem.
Air Quality Data Base
Serious analysts of air pollution and its sources appreciate the local
nature of emission impacts on air qualiaty. They also appreciate the
difficulty of allocating cause for both the levels and the variability in
levels of local air quality to specific sources. Without specific
evidence of justification, the contractor has dismissed all of these
interactions. He has chosen to aggregate available local air quality data
to a countywide level. In so doing, he has largely decoupled the local
relationships between emission sources and air quality. The result is an
estimate of the average air quality in each county having a SIC 344
facility. The contractor's report does not contain information to
evaluate either the representativeness of these averages, or the size of
the standard deviations in estimated values.
Manufacturing Cost Data Base
Having chosen the county level for the display of data, the contractor
found it necessary to aggregate large amounts of site-specific
manufacturing statistics to produce estimates of the manufacturing costs
12-65
-------
for an average SIC facility. The contractor's view of the difficulties
inherent in such a exercise speak for themselves:
" the total cost function is based on the assumption of
cost-minimizing behavior on the part of individual firms. It
is therefore more appropriately estimated using data on
individual firms. However, the data available from the -
Census of Manufacturing are county totals for all of the
firms in each county. To reduce the potential problems
introduced by the aggregation of data across firms, we
divided total cost, C, and total quantity of output, Q, in
each county by the number of establishments, N, in the
county. This does not completely solve the problem of
aggregation but may reduce its importance the cost
function is hypothesized to represent the production
structure for the "average" establishment in a county."
(P7-113, Text)
Again, the contractor provides neither specific indication of the
representativeness of his averages for each SIC 344-containing county nor
the size of the standard error in his estimates.
CONTRACTOR'S REGRESSION ANALYSIS
OF THE DATA
Contractor selection of the county-level as the level of data
aggregation yielded a maximum candidate sample size of 124 sets of data
(economic and environmental) for regression analysis (Text Table 7-12).
When corrected for missing economic data, the sample was reduced to only
57 sets of data. To achieve even this sample size, missing SO- and TSP
data had to be estimated for 21 and three data sets, respectively.
In its most general form, the contractor's regression model contains
54 coefficients all of which must be estimated on the basis of only 57
sets of data. Even the contractor admits to the impossibility of
12-66
-------
meaningful analysis with such a mismatch in model complexity and data
availability. Through a series of largely unsubstantiated assumptions
regarding the actual behaviour of the SIC 344 air quality/cost
relationship, the contractor attempts to reduce the complexity of his
regression model to one compatible with his meager data base. The result
is a regression model with 17 coefficients (Model 4; SIC 344). While most
analysts would expect to use many more data sets to reliably estimate such
a large number of coefficients, the contractor estimated his coefficients
using his meager sample size of 57 data sets.
When fitted with its estimated coefficients, the contractor regression
model perports to correlate manufacturing costs with a series of variables
including air quality. The cost model, by itself, is not central to the
benefit analysis. The partial derivitive, a quantity derived by
mathematical manipulation of the cost model, with respect to air quality
is the critical term upon which further benefit estimates and
extrapolations are made. Thus, the contractor must take the mathematical
partial derivative of his regression model to arrive at his final model
for estimating benefits from air quality improvements. In the case of
TSP, the result is an equation for marginal total cost of productioin due
to changing TSP level (MTCTSP) with only three coefficients. Although we
appreciate the mathematical accuracy of the step, the AISI seriously
questions whether the available data base justifies such a mathematical
manipulation. At a minimum, we note somewhat skeptically that the widely
accepted complex problem of air quality benefits analysis has in the end
been reduced by the contractor to numerical analysis of a three parameter
modelJ These particular parameters are identified by the contractor as
parameters which describe interactions between TSP and labor costs, TSP
and material costs, and TSP alone on the MTCTSP.
12-67
-------
STATISTICAL SIGNIFICANCE
OF THE CONTRACTOR'S FINAL RESULTS
The contractor argues the significance of his results in two ways,
(1) by displaying range in his SIC 344 results, e.g., S240 to $410
improvement per microgram of TSP per cubic meter per establishment and
(2) by conducting tests on the statistical significance of individual
coefficient estimates and on the collection of coefficients in his model.
Neither of these procedures represents a full test of the statistical
significance of the partial differential or marginal cost portion of his
regression model results which are the basis of his calculation on which
is based the study conclusions. The range of data for MTCTSP merely
represent the results of two modeling calculations made with different
assumptions. The statistical testing applies only to individual
coefficient data in the cost model rather than the more important marginal
cost model. No test, statistical or otherwise, is proposed to give a
quantitative measure of the uncertainty in the individual estimates of
marginal costs (MTCTSP) for each of the models. A quantitative measure of
uncertainty in individual estimates of marginal costs is more relevant for
a decision maker forced to make a value judgment on benefits than a
display of a range of results from different models.
While the above problems are troublesome, our greatest concern centers
on the contractor's procedures, or, more accurately, his lack of
procedures for identifying and dealing with uncertainty in his data
bases. We are dubious of the merits of any air quality-related analysis
which is predicated on county-wide averages of data. However, if the EPA
and its contractor wish to postulate such a basis for their data, it is
incumbent upon them (1) to provide reasonable estimates of the standard
errors inherent in their county-based average values and (2) to propagate
these errors to their logical impact on the statistical significance of
final study results. We believe that, when properly corrected for the
full impact of data base variability, contractor results will show no
statistical significance. While the contractor performs no such analysis
12-68
-------
of significance, his own words seem to suggest his bias toward the
possible outcome:
" total cost of production in certain industries are
positively associated with local ambient TSP/S02
concentrations, after accounting for other sources of cost
variation (e.g., input prices, in-place capital, and
climate). These findings are, of course, contingent upon the
assumptions made in the analysis and do not prove the
existence of a cause-and-effeet relationship."
(p. 7-202 Text).
12-69
-------
APPENDIX
In critiquing the contractors report we found it helpful to write out
the multiple regression models in expanded form and to show explicity the
reduced form of the equations for Model 4, SIC 344.
The general regression model in (Equation 7.9 Text)
log C = aQ + a..^ log P£ + 1/2 a., log P£ log P,.
+ b0 log Q + 1/2 b00(log Q)2 + bQ1 log Q log P£
+ c._j log P.^ log Zj H- d£ log Z£
+ 1/2 d£j log a£ log Zj + eQi log Q log Z.,
The 54 Coefficients to be evaluated may be grouped as:
(a) Economic coefficients (18 numbers)
aO» al» a2« a3> a4 and
all» a!2» a!3
a21» a22> a23
a31» a32> a33 and
bO> bOO» b01> b02> b03 where
1 = (labor), 2 = (capital, 3 = (materials)
(b) Economic-environmental coefficients (12 numbers)
Cll» C12' C13> C14
C21> C22> C23' C24
C31» C32» C33» C34 where cij
i = (1, 2, 3) = (labor, capital, materials) and
j = (1, 2, 3, 4) = (S02> xsp, TEMP, RAIN)
12-70
-------
(c) Environmental coefficients (20 numbers)
dl, d2» d3« d4 and
dll» d!2» d!3' d!4
d21' d22' d23' d24
d31» d32> d33» d34 where dij
i,j = (1,2,3,4) = (S02> TSP, TEMP, RAIN)
(d) Size of Firm - Environment coefficients (4 numbers)
C o A o w n fi IT G
(1 = S02, 2 = TSO, 3 = TEMP, 4 = RAIN)
The important equation for marginal total cost due to incremental
charge in TSP is of the form
MTCTSP = (C/TSP) (£ ci2 log P± + d2 + , d2- log Z- + eQ2 log Q)
which has 9 coefficients to be evaluated, namely:
C12 = (labor» TSp) coefficient
C22 = (capital, TSP) coefficient
C32 = (materials, TSP) coefficient
d2 = (TSP) coefficient
^21 = (TSP> labor) coefficient
d22 = (TSP, TSP) coefficient
d23 = (TSP, TEMP) coefficient
d24 = (TSP, RAIN) coefficient
602 = (s^26 °f firm, TSP) coefficient
12-71
-------
For SIC 344, Model 4, the marginal cost equation reduces to a three term
expression using data reported in the text (p. 7-214).
fc)
MTCTSP = ' [0-020111 10S Plabor ' °-026114 mat'ls
+ 0.069585)
Using as a definition of uncertainty the ratio of the coefficient standard
error divided by its value (data from table, p. 7-214 Text), we may
estimate that in the above equation, the first coefficient (number) has an
uncertainty of about 102 percent, the second coefficient - 90 percent
uncertainty, and the third coefficient - 80 percent uncertainty. The
contractor does not demonstrate the influence of these uncertainties on
the estimation of MTCTSP.
12-72
-------
SECTION 13
ANALYSIS OF POLLUTANT CORRELATIONS
-------
SECTION 13
ANALYSIS OF POLLUTANT CORRELATIONS
INTRODUCTION
The sector models estimated in Volumes II through IV of this
report include only two measures of ambient air pollution — total
suspended particulates (TSP) and sulfur dioxide (S02). Other pollu-
tants, however, may also contribute to the physical effects or
behavioral responses analyzed in the models. For example, ozone
concentrations are believed to be an important factor in explaining
reductions in yield for various agricultural crops [Criteria Document
(1)]. Unfortunately, because of data limitations, it was not possible
to test whether estimated coefficients for other pollutant measures
were significantly different from zero in the model specifications.
In many cases, the number of observations would have been reduced to
less than ten if data on additional, plausibly significant, environ-
mental variables were employed.
If other pollutants are indeed relevant explanatory variables for
the postulated models, then omission of these variables results in a
specification ecror. This has implications for the statistical
integrity of the estimated models. In fact, in the case where there
13-1
-------
is a non-zero correlation between an omitted variable and an included
variable, the estimator of the included variable will be biased and
inconsistent. The direction of bias in such circumstances depends on:
1) the sign of the correlation between the omitted and included
explanatory variables; and 2) the sign of the coefficient of the
excluded variable when the "true" regression model is estimated. If
these two factors work in the same direction, the bias imparted to the
coefficient of the included variable will be positive; otherwise, it
will be negative.
Another way to view the problem of multiple pollutants affecting
behavioral responses is in terms of proxy variables. That is,
inclusion of a single pollutant measure in a specification serves as a
proxy for the many pollutants that may affect behavioral decisions.
In this case, the statistical issues are better analyzed as an errors-
in-variables problem rather than as an omitted variables problem. The
implication of adopting such a perspective (i.e., proxy variable) is
that the relevant explanatory variable becomes "air quality" and not a
specific pollutant. Given that the objective of this study is to
determine the benefits associated with particular pollutant standards,
individual pollutants and not the generic "air quality" are the
desired variables for the various specifications. In fact, the
results of physical damage function studies provide support for
including specific pollutants in the developed models. As a
consequence, empirical estimation of the specifications best proceeds
in terms of the pollutants. Thus, we feel it is more appropriate to
13-2
-------
examine the issue of multiple pollutants as an omitted variables
problem.
This section of the report addresses the omitted variables
problem by analyzing pairwise correlations between various pollutants.
Under the presumption that pollutants other than TSP and SC>2 are
relevant explanatory variables in the sector models, the analysis of
correlations provides a partial indication of the direction and
severity of bias that may be present due to specification error.
ESTIMATION OF CORRELATION COEFFICIENTS
Correlations among the two pollutants included in the sector
models of this report (TSP, S02), and five pollutants not included in
the sector analysis — nitrogen dioxide (N02); carbon monoxide (CO);
ozone (03); total oxidants (TOX); and total hydrocarbons (THC) — are
examined in this subsection. The discussion of bias due to an
omitted, relevant explanatory variable is based on results derived
explicitly for single equation, linear multiple regression models.
Although such models are more restrictive than the nonlinear systems
methods employed in some sections of this report, similar conclusions
with respect to bias can be made. In fact, the problem may be made
worse, since as Theil (2) notes, a disadvantage of full information
systems estimation is that any biasedness and inconsistency problems
may be propagated to all equations in the system. Thus, even if only
13-3
-------
one equation in the system is mis specified, statistical problems need
not be limited to the single equation.
Our analysis begins with a general overview of the statistical
problems that arise with the omission of a relevant explanatory
variable from a regression specification. This is followed by a brief
discussion of the source of the air quality data. Finally, results
are reported for pairwise correlations between air pollution variables
when the variables are defined in three distinct ways. Specifically,
correlations are reported for alternative assumptions with respect to
the spatial, temporal, and pollutant measurement characteristics of
the data. A major conclusion of the analysis is that the degree and
direction of bias introduced because of the omission of a relevant air
quality variable is sensitive to the forms of the variables used in
the analysis. This result highlights the importance of selecting the
"correct" measure of air quality for use in benefits analyses.
Statistical Background*
The problem examined here involves the omission of a relevant
explanatory variable from a linear, multiple regression model. Assume
that the correct specification of an equation is:
Yi = PI + '32xi2 + /33xi3 + Ei
* The discussion in this subsection is taken from Kmenta (3)
13-4
-------
but because of lack of dataf it is not possible to develop a data
series for Xi3. Thus, the equation that is estimated is of the form
(13.2)
In this case, it can be shown that the expected value of the
coefficient £2 in Equation (13.2) is equal to
E(/82) * #2 + /33d32 (13.3)
where
2(Xi2 - X2) (Xi3 - X3)
•r / v -_ ^ \ *"
2(Xi2 - X2)
In the case where Equation (13.1) is estimated, the expected value of
£2 is equal to /S2. This is the statistical property of unbiasedness.
As can be seen from Equation (13.3), estimation of the misspecified
equation (13.2) will lead to a biased estimate of £2 with the bias
equal to /83d32. The purpose of the present analysis is to attempt to
define at least the sign of £3d32 so that it is possible to identify
the direction of bias.
There are two components to the term ^2<^22' The firstf £3?
represents the marginal effect on Y^ of a small change in X^3. If we
think of X^3 as some measure of air pollution and YJ as a variable
13-5
-------
describing an output that may be affected by air quality improvements
(e.g., agricultural yield), then we might generally take /?3 as
positive. However, because Equation (13.1) has not been estimated, it
is not possible, without outside information, to determine the
magnitude of £3.
The second component of interest is d32. As shown in Kmenta (3),
this term represents the coefficient in the linear specification
relating the excluded and included explanatory variables. In
particular,
Xi3 = ^31 + d32xi2 + residual (13.5)
Note that the coefficient d32 in a regression specification like
(13.5) can also be stated in terms of the correlation between X^2 an(^
Xi3:
sxi3
d32 = r (13.6)
sxi2
Here, r is the simple correlation between X^2 and Xj_3, SX^3 and SX^2
are the standard errors of X^3 and X^2, respectively.
In view of Equation (13.6), if £3 is different from zero, then
estimation of the misspecified (13.2) will lead to an estimator of £2
that is biased unless d32 is zero. That is, unless Xj_2 and X^3 are
13-6
-------
uncorrelated. If ySj and dj2 ^ave the same sign, then the bias will be
positive; otherwise it will be negative. Note that additional
information on SXi3 and SX^2 would be required before one could assess
the magnitude of dj2« However, knowledge of the simple correlation
may help provide a partial indication of the severity of the bias.
All else equal, a small correlation will lead to less bias than a
relatively high correlation.*
Source of Data
The data used in the calculation of correlations are taken from
the SAROAD data base maintained by EPA. The measures of TSP and S02
included in our sector analyses also came from SAROAD. Although
measures of concentration can be obtained on a daily basis, the data
used here represent annual measures. This is consistent with the
frequency assumed in the estimation of the sector models. In addi-
tion, only those sites meeting EPA's summary criteria for reporting
annual averages are included in the correlation analysis of this
section. Finally, because of a possible bias in S02 data collected by
the gas bubbler method, only continuous monitor S02 data are included
in the correlations reported below. Similarly, no N02 data collected
via the biased Jacobs-Hochheiser method are included in this analysis.
* Statistical problems can also arise with a zero correlation between
an included and an excluded variable. With zero correlation, the
estimator of /32 is no longer biased. However, the estimator of the
variance of £? in tne misspecified equation will contain an upward
bias, so that tests of statistical significance will be overly
conservative.
13-7
-------
Data were available for the seven pollutants mentioned earlier
for each of seven years, 1972 through 1978. The total number of valid
site observations by pollutant and year are shown in Table 13-1.
Although it would be possible to analyze pairwise correlations as a
time series across the seven years, all correlations reported here are
based on cross-sectional data.
Spatial Variation
For this analysis, we focus on three factors that help define the
air quality variable. These included the attributes of time,
Table 13-1
COUNT OF AVAILABLE SITE OBSERVATIONS FOR CORRELATION ANALYSIS
Year
1972
1973
1974
1975
1976
1977
1978
TSP
2,172
2,294
2,634
2,702
3,105
3,133
3,003
so2
1,356
1,874
2,824
3,138
3,644
3,482
2,498
N02
53
95
685
964
1,162
1,000
783
Pollutant
CO
323
429
606
676
733
734
715
°3
1
19
81
138
207
232
247
TOX
62
67
68
57
23
26
18
THC
40
50
21
59
67
59
50
13-8
-------
location, and measurement type. In this section, we assume that the
time and pollutant measurement factors are fixed, and analyze the
impact of changes in spatial definition on the pairwise correlations.
The locational identifiers examined include sites, counties, and
Standard Metropolitan Statistical Areas (SMSAs). For this analysis,
the time period is fixed at 1978 and all correlations are reported for
annual arithmetic means.
The data reported in Table 13-1 are site data. This represents
the most disaggregate data in terms of location currently available.
Each site is identified by a 12-digit identification code. Observa-
tions are included in the analysis of pairwise correlations only if
data are available for both pollutants at a given site.
The second level of spatial aggregation is the county. Air
pollution statistics at the county level were evaluated in three ways.
First, the arithmetic average of readings from all sites in the county
were calculated. Second, the maximum across all sites in the county
was recorded as the indicator of county air quality. Finally, the
minimum across all site readings in the county was taken as the air
quality index. Although correlations were calculated for each of
these ways of forming an index of county air pollution, only the
correlations involving the arithmetic means are reported here. Note
that the number of observations at the county level can exceed the
number of observations in the analysis of site data since different
sites in a county may monitor different pollutants.
13-9
-------
The final level of aggregation was to the SMSA level. Since
SMSAs are groups of contiguous counties, the SMSA air quality index
was formed in a manner equivalent to that described for the site-to-
county aggregation. In particular, indices reflecting average,
maximum, and minimum of county readings were developed. As with the
county data, the SMSA data reported here represents the arithmetic
average of county parts.
Note that correlations at the SMSA level are reported only for
the 24 SMSAs used in the household sector analysis of Volume II. This
limitation was imposed only in the interest of reducing the scope of
the analysis.
Results of the correlation analysis for the three levels of
spatial aggregation are shown in Tables 13-2 and 13-3. The tables
show the correlations between TSP and the other pollutants, and SC>2
and the other pollutants, respectively.
It is difficult to draw specific conclusions from the correla-
tions reported in these tables. First, the correlation coefficient by
itself does not provide sufficient information to determine the
possible magnitude of bias due to an omitted variable. Second, no
information is provided in the tables as to the statistical signifi-
cance of the reported correlations.
13-10
-------
Table 13-2
CORRELATION COEFFICIENTS FOR 1978 ANNUAL ARITHMETIC MEANS OF TSP
AND OTHER POLLUTANTS FOR DIFFERENT LEVELS OF SPATIAL AGGREGATION
Area
Site
County
SMSA
so2
0.521
0.117
0.020*
N02
0.450
0.373
0.432*
Polli
°3
-0.037
-0.075
0.141*
it ant
CO
__
0.082
0.586*
TOX
0.486+
0.657+
—
THC
0.648
0.611*
0.291"1"
Table 13-3
CORRELATION COEFFICIENTS FOR 1978 ANNUAL ARITHMETIC MEANS OF
S02 AND OTHER POLLUTANTS FOR DIFFERENT LEVELS OF SPATIAL AGGREGATION
Area
Site
County
SMSA
Pollutant
TSP
0.521
0.117
0.020*
N02
0.554
0.285
-0.266*
°3
—
-0.084
0.048*
CO
--
-0.058
-0.108*
TOX
—
0.991+
—
THC
—
0.046*
-0.693+
* 15 to 30 observations.
+ 3 to 14 observations.
— Less than 3 observations.
13-11
-------
With respect to the latter observation, note that if the
pollution measures are normally distributed random variables, a test
statistic of the form
t* = T- „ . (13-7)
follows a student-t distribution with (n-2) degrees of freedom.* In
Equation (13-7), r-^ represents the pairwise correlation and n is the
number of observations. For t greater than the critical value (given
a significance level), the null hypothesis of no correlation would be
rejected.
A plot of the frequency of concentration intervals indicates that
the pollution data tend to follow a log-normal distribution.
Consequently, the test statistic in Equation (13-7) is more properly
applied to a logarithmic (base e) transformation of the original data.
This transformation was made for the 1978 county annual arithmetic
mean data and correlations computed. The results are recorded in
Table 13-4. Except for the logarithmic transformation, the definition
of the variables in Table 13-4 is equivalent to the middle rows of
Tables 13-2 and 13-3. Table 13-4 shows that at the 5 percent level of
significance (two-tail test), we fail to reject the null hypothesis of
* See Neter and Wasserman (4), p. 405.
13-12
-------
Table 13-4
SIGNIFICANCE OF CORRELATION COEFFICIENTS FOR
1978 COUNTY ANNUAL ARITHMETIC MEANS
TSP With:
rij
n
t(n-2)
S02 With:
rij
n
fc(n-2)
S02
0.168
221
2.52*
0.168
221
2.52*
N02
0.423
335
8.52*
0.425
124
5.19*
°3
-0.074
156
-0.92
0.025
89
0.233
CO
0.130
144
1.56
-0.127
99
-1.26
TOX
0.490
7
1.26
0.997
3
12.88*
THC
0.587
29
3.77*
0.214
20
0.929
* Significantly different from zero at the 95 percent level of two-
tailed t-test.
no difference for 1) TSP with CO, TOX, and 03; and 2) for S02 with CO,
THC, and 0^.
A comparison of the middle-row correlations in Tables 13-2 and
13-3 versus the correlations in Table 13-4 illustrates that the
mathematical form of the data represents another way in which the
variables may be characterized. Additional application of the
13-13
-------
significance test of the correlation coefficient will be presented in
the discussion of the specific Mathtech sector models.
A qualitative observation can also be made with respect to the
variation in the reported correlations. From the tables, it is clear
that both the magnitude and sign of the correlation coefficient are
sensitive to the level of spatial aggregation. For example, in one
case, the correlation is positive at the site and county level, but
negative at the SMSA level. Thus, for any given study, the definition
of geographical unit may be an important element that influences the
magnitude and direction of bias introduced because of a misspecified
equation.
Temporal Variation
Tables 13-5 and 13-6 show the pairwise correlations found at the
county level for three distinct years. The statistical measure for
each pollutant is the annual arithmetic mean. Table 13-5 reports the
correlations between TSP and the other six pollutants, while Table 13-
6 records the correlations between S02 and the other pollutants. The
calculation of correlations for different years was done to check the
stability of the measure across time. This can be important
information for time series, cross-section models or in those cases
where there is some flexibility in the choice of data from a specific
year.
13-14
-------
Table 13-5
CORRELATION COEFFICIENTS FOR ANNUAL ARITHMETIC MEANS OF
COUNTY LEVELS OF TSP AND OTHER POLLUTANTS FOR THREE YEARS
Year
1972
1975
1978
so2
0.023
0.388
0.117
N02
0.206
0.351
0.373
Polli
°3
—
0.134
-0.075
itant
CO
0.058
0.175
0.082
TOX
0.341
0.728*
0.657*
THC
0.175
0.609
0.611*
Table 13-6
CORRELATION COEFFICIENTS FOR ANNUAL ARITHMETIC MEANS OF
COUNTY LEVELS OF S02 AND OTHER POLLUTANTS FOR THREE YEARS
1972
1975
1978
TSP
0.023
0.388
0.117
N02
0.339+
0.289
0.285
Polll
°3
—
0.037
-0.084
itant
CO
0.201
0.161
-0.058
TOX
0.219+
0.143+
0.991+
THC
-0.396+
0.115*
0.046*
* 15 to 30 observations.
+ 3 to 14 observations.
— Less than 3 observations.
13-15
-------
As in the tables reporting correlations for different location
aggregates, Tables 13-5 and 13-6 show that the correlation
coefficients vary across the different years. In general, the
correlations are not large in magnitude. However, as noted earlier,
this is not sufficient information to determine the magnitude of bias.
From the information reported in the tables, it is only possible to
address the issue of direction of bias. If the correlation
coefficient and the (unestimated) coefficient of the relevant excluded
variable have the same sign, the bias is positive; otherwise it is
negative.
It is important to reiterate that this discussion of bias depends
on the assumption that an excluded variable is a relevant explanatory
factor. In the economic models estimated in this report, there are
typically no sound economic reasons for believing that one or more
pollutants belong in a particular specification. As a consequence,
there must be reliance on literature such as that summarized in the
Criteria Document (1) to guide the selection of environmental
variables. For the soiling, materials damage, and yield reduction
effects analyzed in this report, there is evidence that both N02 and
0^ may be relevant explanatory factors. For example, N02 has been
implicated in the fading and yellowing of textiles, while 03 has been
shown to have a deleterious impact on a variety of agricultural crops.
Note that if other likely impacts of air pollution, such as health
effects and aesthetic impairments, are considered, then each of the
13-16
-------
seven pollutants looked at here may be expected to be relevant
variables in a benefits analysis.
Variations in Pollution Measurement
A third way in which the correlation of air pollution data can be
reported is by the statistical measure used to summarize the data.
Three such measures are analyzed in Tables 13-7 and 13-8. These are
the arithmetic mean, the geometric mean, and the second highest
readings. These statistical measures were chosen because they are
consistent with the forms typically used in the definition of the
ambient air quality standards.
The correlations reported in Tables 13-7 and 13-8 were calculated
for combinations of pollutants defined for the same statistical
measure. That is, the correlations shown for "arithmetic means"
represent the case where both pollutants were measured in this way.
Correlations that mixed the statistical measures were calculated but
they are not reported here.
As a point of clarification, the arithmetic mean used earlier in
defining a county index from site data is a different measure than the
arithmetic mean reported in Tables 13-7 and 13-8. In these tables,
since the data are defined at the county level, the row labelled
arithmetic mean should be interpreted as an arithmetic mean of site
data in a county, where the site data are recorded in-terms of the
13-17
-------
Table 13-7
CORRELATION COEFFICIENTS FOR 1978 COUNTY-LEVEL DATA OF TSP AND
OTHER POLLUTANTS FOR DIFFERENT STATISTICAL MEASURES
DLaulSulCcLL
Measure
Arithmetic mean
Geometric mean
Second high
S02
0.117
0.117
0.000
N02
0.373
0.383
0.214
Pollut
°3
-0.075
-0.233
0.043
:ant
CO
0.082
0.097
0.109
TOX
0.6S74
0.324+
0.576+
THC
0.611*
0.553*
0.348*
Table 13-8
CORRELATION COEFFICIENTS FOR 1978 COUNTY LEVEL DATA OF SO, AND
OTHER POLLUTANTS FOR DIFFERENT STATISTICAL MEASURES
Statistical
Measure
Arithmetic mean
Geometric mean
Second high
Pollutant
TSP
0.117
0.117
0.000
N02
0.285
0.309
-0.085
°3
-0.084
-0.278
-0.085
CO
-0.058
-0.071
0.166
TOX
0.991Hr
0.955"1"
0.751"h
THC
0.046*
-0.120*
0.288*
* 15 to 30 observations.
+ 3 to 14 observations.
13-18
-------
annual arithmetic mean. Similarly, for the row labelled geometric
mean, the interpretation is that the data represents the arithmetic
mean of site data in a county, where the site data are recorded in
terms of the annual geometric mean.
As might be expected, the magnitude and signs of the correlations
for the arithmetic and geometric means are similar. Also of interest
is the fact that the correlations for the extreme measure, the second
high, appear to be generally lower in absolute magnitude. As before,
since there is no knowledge of the magnitude of the (unestimated)
coefficient for the relevant omitted variable, it is not possible to
assess whether this lower correlation translates into a lower bias
relative to annual mean measures.
POTENTIAL FOR BIAS IN MATHTECH STUDY
Tables 13-2 through 13-8 report correlations for general
assumptions related to time, location, and pollutant measurement
alternatives. This was done to facilitate comparisons of correlations
when any one of these factors is varied. Clearly, there are many
other ways in which each of these factors could be defined and
combined to analyze selected pairwise correlations.
In this subsection, the three factors are defined in ways that
best represent the underlying structure of the three major economic
sector models estimated in this study — the household, manufacturing,
13-19
-------
and agricultural models. Correlations are then computed for pairs of
pollutants. One of the pollutants selected will be a variable
included in the economic model under discussion, while the other
pollutant represents an excluded variable for which there is an a.
priori belief that it may be a relevant explanatory variable. As
before, the correlations help to identify the likely direction of
bias, but more information is required if the magnitude of the bias is
to be ascertained. Finally, the statistical significance of the
computed correlation coefficients is determined for the manufacturing
sector results. This sector was chosen because the air quality data
used in the original study (Section 7) are defined in logarithmic
terms. Consequently, the transformed data more closely follow a
normal distribution and the test statistic in Equation (13-7) can be
applied.
Household Sector
The estimation of demand systems in the household sector model is
based on SMSA-level data from 1972 and 1973. The measures of TSP and
S02 included in the reported specifications are maximum second high
values.
One of the demand equations for which S02 is a relevant variable
is the demand for household textiles. Evidence in the Criteria
Document (1) indicates that this is to be expected since acids derived
from S02 can lead to the discoloration and deterioration of fabrics.
13-20
-------
In addition, the revised draft staff paper for N02 (5) notes that N02
has been demonstrated as having deleterious effects on textile dyes,
natural and synthetic fibers, metals, and various rubber products. As
a consequence, there is reason to consider N02 as a plausible explana-
tory variable in the demand equation for textiles.
Correlations for both 1972 and 1973 are calculated. Because both
S00 and NO-? observations are limited to data collected by current
£ &
pollutant reference methods, less than ten observations are available
in each case. Note that the correlation analysis described here is
appropriate for single equation, multiple regression models only.
Since the household sector analysis estimates systems of nonlinear
demand relations, more complex methods are required to determine the
actual impact of omitting a relevant variable from the demand specifi-
cations.
Table 13-9 lists the correlations between S02 and N02 in 1972 and
1973. The geographic unit is the SMSA. Since the household sector
model used the maximum of the second highs for all sites in an SMSA,
this is the measurement unit of S02 in the correlation analysis. This
measure is then correlated with the maximums of the arithmetic mean,
the geometric mean, and the second highs for N02.
In the context of the omitted variables problem, the correlations
reported in Table 13-9 indicate that if N02 is a relevant explanatory
variable, then omission of the variable would likely impart a positive
13-21
-------
Table 13-9
CORRELATION OF S02/N02 FOR HOUSEHOLD SECTOR STUDY, 1972 AND 1973
N02 Arithmetic Mean
N02 Geometric Mean
N02 Second High
S02 - Maximum Second High
1972*
0.320
0.350
-0.087
1973**
0.530
0.480
0.375
* Correlations calculated from five observations.
** Correlations calculated from seven observations.
bias to the estimator of the coefficient for S02 in the textile demand
equation. This occurs because the correlation coefficient; is positive
in all but one case. In turn, this implies that the benefits
associated with specified changes in S02 would be overstated. This
result is independent of the three measurement types shown in Table
13-9 for 1973. However, if the appropriate N02 measure is the maximum
of second high readings, then the data indicate that in 1973 there
would be an underestimate of benefits. Thus, the direction of bias is
sensitive to the measurement method and year of data assumed.
Although it is possible to deduce the direction of bias, it is
not possible to assess the magnitude of bias without additional
13-22
-------
information. It should be stressed that the discussion of bias rests
on the assumption that N02 is a relevant explanatory variable.
Manufacturing Sector
The various cost equations estimated in the manufacturing sector
were based on county-level data for 1972. The basis for including
measures of air pollution in the cost specifications was to test the
hypothesis that ambient levels of air pollution increase costs of
production for specific industries. On the basis of the analysis
conducted in Section 7, measures of TSP or SC>2 were relevant explana-
tory variables in several of the 3-digit SICs that were analyzed.
As noted above, N02 can have a damaging effect on a variety of
materials that may be instrumental in production processes. In parti-
cular, N02-related damage to metal products may be an important
element in increased production costs. Thus, there is reason to
believe that N02 may be a relevant variable in the cost equations.
Table 13-10 reports county-level correlations for 1973 between
S02 and N02 as well as between TSP and N02. Observations were drawn
from the valid SAROAD data and not limited to those counties covered
by the manufacturing sector analysis. In the manufacturing sector
analysis, the TSP and S02 measures were defined as a logarithmic
transform of the arithmetic averages of the second hi-gh readings
across sites in a county. This is the pollutant measurement index
13-23
-------
Table 13-10
CORRELATIONS FOR S02/N02 AND TSP/N02 FOR
MANUFACTURING SECTOR STUDY, 1973
S02 - Second High
Correlation
Observations
TSP - Second High
Correlation
Observations
Arithmetic
Mean
-0.249
22
-1.15
0.408
54
3.22*
NO2
Geometric
Mean
-0.179
22
-0.81
0.375
54
2.92*
Second
High
-0.082
22
-0.37
0.438
54
3.51*
* Significantly different from zero at the 5 percent level of the two-
tailed t-test.
used in the correlation calculations shown in Table 13-10. Because
the logarithmic transformation is approximately normally distributed,
the test statistic of Equation (13-7) is used to ascertain the
statistical significance of the correlations in Table 13-10. The
results of this test are also shown in the table.
Each of the correlations between S02 and N02 are negative.
•
However, the significance test indicates that we cannot reject the
13-24
-------
null hypothesis that these correlations are zero. Thus, one can infer
that omission of N02 as an explanatory variable in the manufacturing
sector analysis may not lead to bias in the estimated coefficient for
S02. There are three caveats to this conclusion. First, the correla-
tions are computed from a sample that differs from the one used in the
manufacturing sector analysis. Second, prior information and
empirical testing must be undertaken before a specific form for NC>2
can be considered as the appropriate variable in the "true" regres-
sion. Third, the correlation test is really only appropriate for
single equation multiple regression models.
With respect to TSP, the positive correlations observed in Table
13-10 for all averaging times indicate that the estimator of the
coefficient for TSP may be biased upward if NC^ concentrations contri-
bute to an increase in production costs. Thus, benefits for a speci-
fic reduction in TSP could be overestimated. In this case, the
statistical tests of significance indicate that the correlations are
significantly different from zero. However, the caveats mentioned
above apply here as well.
Agricultural Sector
The estimation of yield equations for soybeans and cotton in
Section 9 was based on county-level agricultural, economic, and air
quality data from 1974 to 1976. The air quality data used in the
analysis were data from the second quarter, rather than annual data.
13-25
-------
This was cone to better match the air quality readings to the time
period when the crops were likely to be most susceptible to air
pollution-related damage. Because of data limitations, only measures
of S02 were included in the yield specifications reported in Section
9.*
There exists a good deal of evidence from laboratory and field
studies that elevated levels of ozone may be a significant contri-
buting factor to damage for a variety of crops — including soybeans
and cotton. Thus, exclusion of ozone from the yield equations may
lead to a specification error.
An analysis of the correlation between ozone and the SC>2 data
used in the agricultural study required the collection of air quality
data different from that used earlier. In particular, quarterly data
.were required for ozone to be consistent with the measure of S02 used
in the agricultural study. These data were obtained from the SAROAD
data base for the second quarters of 1974-76 for a variety of
averaging times. The sample was limited in that data were collected
only for 47 counties included in the agricultural analysis of cotton.
Time and resource constraints precluded the acquisition of a more
general data base for this correlation analysis. Even with the
restricted sample, not all counties had data available for both
pollutants in each of the years. However, because three years (one
* TSP was not .considered to be an appropriate explanatory variable for
the yield equations. However, other pollutants may effect yield.
13-26
-------
quarter per year) of data were available, the sample size for the
correlations is greater than 30.
Table 3-11 reports the correlations between ozone and S02. The
diversity of signs and magnitudes for the various pollutant measure-
ments shows the sensitivity of results to the choice of pollutant
measurement. This reinforces the conclusion that caution must be
exercised in an analysis of the omitted variables problem. The sensi-
tivity of the correlation coefficients to specific assumptions about
the form of the excluded variable cannot be ignored.
Table 3-11
CORRELATIONS OF S02/03 FOR AGRICULTURAL SECTOR STUDY —
QUARTERLY DATA 1974-76
so2
Average Second High
Maximum Second High
Avg. Arithmetic Mean
Max. Arithmetic Mean
Average
Second
High
-0.201
-0.380
0.250
0.188
Maximum
Second
High
-0.176
-0.315
0.312
0.233
Ozone
Average
Arithmetic
Mean
-0.085
-0.328
0.291
0.204
Maximum
Arithmetic
Mean
-0.160
-0.375
0.295
0.194
13-27
-------
SUMMARY AND CONCLUSIONS
A possible source of bias in the benefits estimates reported in
this study relates to statistical problems associated with misspecifi-
cation of the economic models. The particular issue addressed in this
section involves the extent to which omission of potentially relevant
environmental variables in the models may lead to such a bias. Under
the assumption that pollutants other than TSP and S02 are relevant
explanatory variables in the economic models, the relevance of the
bias issue is evaluated through an analysis of pairwise correlations.
Such an analysis is limited in its usefulness. Although a non-
zero correlation implies that some bias is introduced by omitting a
relevant variable, it is not possible to identify either the magnitude
or the direction of bias without additional information. In particu-
lar, knowledge is required of the expected sign and magnitude of the
coefficient of the excluded variable in the (unestimated) "true"
regression. In most cases, it is possible to specify the sign of the
coefficient with some confidence so that the direction of bias can be
identified for specific assumptions about the definition of the
included and excluded variables. As the analysis in this section
shows, the definition of the variables is a crucial element that
cannot be overlooked.
The key question considered in this section is: Are the benefit
estimates developed in Volumes II through IV of the study biased
13-28
-------
because only measures of TSP and SC>2 are considered in the estimation
process? With currently available information, it is not possible to
say with any confidence whether the bias will be positive or negative,
large or small. Any combination of these outcomes is a possibility.
However, we have shown that in several instances it is not possible to
reject the null hypothesis that the correlations are zero. In such
cases, the possibility of bias introduced through the omission of a
relevant explanatory variable is reduced.
•
Despite this negative conclusion, there are several positive
aspects to the analysis in this section. First, the analysis makes
apparent the sensitivity of the correlations to assumptions about the
implied temporal, spatial, and pollution measurement definitions for a
given variable. With respect to studies involving environmental
variables, this result has not always been explicity recognized.
The second area where valuable information has been generated
relates to the size of the correlation coefficients reported in this
section. The conventional wisdom is that very high correlations exist
between pollutant types so that inclusion of more than one pollution
variable in a regression specification will likely result in a high
degree of multicollinearity.* If multicollinearity is present, this
* Note that a high level of correlation between two explanatory
variables is a sufficient but not necessary condition for the
presence of a high degree of multicollinearity when the number of
explanatory variables in the regression model exceeds two. See
Kmenta (3), p. 384.
13-29
-------
can lead to inflated standard errors, and hence, overly conservative
tests of statistical significance.
In general, the correlations reported in this section tend to be
lower than expected, with most being less than 0.50. This can be
contrasted with correlation estimates in the benefits literature of
0.7 to 0.8 between TSP and S02 [see, for example, Crocker (6)].
However, extreme caution must be exercised in making comparisons
across studies. Again, the magnitude and sign of the correlations is
very sensitive to the definition of the variables. For example, Table
13-2 shows that the correlation between TSP and S02 is 0.521 (annual
arithmetic mean) when computed across sites in 1978. This is close to
the values reported elsewhere in the literature. On the other hand,
if the geographical unit is the county or the SMSA, the correlation
between TSP and S02 falls to 0.117 and 0.02, respectively. Thus, much
like the case of omitted variables, multicollinearity may or may not
be a serious statistical problem, depending on the units of the
variables included in the specification. This implies that emphasis
should be given to the identification of the most appropriate form of
the explanatory variables, with the choice based on a combination of
theoretical and empirical criteria.
13-30
-------
REFERENCES
U.S. Environmental Protection Agency. Air Quality Criteria for
Particulate Matter and Sulfur Oxides - Volume III: Welfare
Effects. Environmental Criteria and Assessment Office, Research
Triangle Park, North Carolina, 1981.
v
Theil, H. Principles of Econometrics. John Wiley and Sons,
Inc., New York, 1971.
Kmenta, J. Elements of Econometrics. Macmillan Publishing Co.,
Inc., New York, 1971.
Neter, J. and W. Wasserman. Applied Linear Statistical Models.
Richard D. Irwin, Inc., Homewood, Illinois, 1974.
U.S. Environmental Protection Agency. Preliminary Assessment of
Health and Welfare Effects Associated with Nitrogen Oxides for
Standard-Setting Purposes. Revised Draft Staff Paper, Office of
Air Quality Planning and Standards, Research Triangle Park, NC,
October 1981.
Crocker, T. D., gt al. Methods Development for Assessing Air
Pollution Control Benefits - Vol. 1: Experiments in the
Economics of Air Pollution Epidemiology. Report prepared for the
U.S. Environmental Protection Agency, University of Wyoming,
February 1979.
13-31
-------
SECTION 14
SUMMARY OF MANUFACTURING SECTOR REVIEW
-------
SECTION 14
SUMMARY OF MANUFACTURING SECTOR REVIEW
INTRODUCTION
In July 1981, a public meeting was held to review a draft version
of Volumes I through V of this report. Participants in the meeting
included a panel of experts in environmental benefits analysis,
assembled to critically review the report. In the review of the
manufacturing sector analysis in Volume III, the general conclusion at
the meeting was that the analysis was a careful and sophisticated
piece of research. Specific positive and negative features of the
study were identified. Section 12 contains a summary of the comments
from the panel and audience.
One suggestion of the review panel was that additional investiga-
tion of one of the manufacturing industries be undertaken. The
industry of interest was SIC 344, the Fabricated Structural Metal
Products industry. The draft analysis had suggested that production
costs in that industry were adversely affected by high levels of
ambient particulate matter (PM) concentrations. The analysis also
suggested that reducing PM concentrations could lead to significant
14-1
-------
economic benefits for this industry ($3.7 billion discounted present
value for attainment of the TSP secondary standard). The panel recom-
mended that further effort be made to evaluate the plausibility of
this result. Two specific tasks were undertaken in response to this
suggestion. These included interviews with some plant managers in the
industry and re-evaluation of the draft analysis. The results are
summarized in this section.
INFORMAL PLANT INTERVIEWS
During the course of the study reported in Section 7,, telephone
contact was made with a small number of companies to see if additional
evidence could be obtained concerning the plausibility of the results
for SIC 344. As summarized in Section 7 (see pp. 7-190 through 7-
192), two metal fabricating companies and one air filtration system
manufacturer were contacted. They suggested several ways in which
dust or particles might affect metal fabricating operations. These
included contamination of welding operations, power sources, and
electronic equipment; and contamination or corrosion of metal inven-
tories. These were suggested as possibilities, but no specific docu-
mentation or quantitative evidence was obtained.
Following the public meeting, a small number of additional
companies in SIC 344 were contacted. As with the previous contacts,
the interviews were informal and unstructured. However, some of the
interviews were conducted in person, as well as by telephone, and
14-2
-------
three geographic areas were included to provide some variation in
ambient TSP concentrations. The three geographic areas included
Denver, Seattle, and Philadelphia. The ambient TSP concentrations in
these three areas and the other areas included in the draft analysis
are identified in Table 14-1.
There were originally three objectives to the additional inter-
views. These included: 1) obtaining qualitative evidence of damage
and/or behavioral adjustments due to PM deposition; 2) determining
whether plant managers perceived ambient PM deposition as affecting
either their production processes or their production costs; and 3)
estimating the potential impact of PM deposition on production costs.
The third objective proved very difficult to accomplish from inter-
views and thus effort was focused primarily on the first two.
Questions asked during the interviews sought to identify how
plants in SIC 344 might be affected by particulate matter. A list of
possible effects compiled prior to the interviews is provided in Table
14-2. The first column in the table identifies some of the possible
physical effects which PM may have on SIC 344 plants. The second
column suggests some of the ways in which plant managers may respond
to the physical effects (in addition to the option of not responding
in any way). The third column identifies the economic consequences of
the different responses. Questions asked during the interviews
attempted to determine whether any of the listed physical effects had
14-3
-------
Table 14-1
COUNTIES USED IN THE DRAFT ANALYSIS FOR SIC 344
State
County
TSP*
Alabama
California
Colorado
Connecticut
Connecticut
District of Columbia
Georgia
Illinois
Louisiana
Louisiana
Maryland
Maryland
Massachusetts
Massachusetts
Massachusetts
Massachusetts
Michigan
Minnesota
New Jersey
New Jersey
New York
New York
New York
New York
New York
Ohio
Ohio
Pennsylvania
Pennsylvania
Pennsylvania
Pennsylvania
Rhode Island
Texas
Texas
Washington
Wisconsin
Jefferson
San Francisco
Denver
Fairfield
New Haven
Fulton
Cook
E. Baton Rouge
Orleans
Baltimore (City)
Baltimore (County)
Bristol
Hampden
Middlesex
Worcester
Wayne
Hennepin
Bergen
Union
Bronx
Erie
Queens
Suffolk
Westchester
Hamilton
Stark
Allegheny
Berks
Lackawanna
Philadelphia
Providence
Dallas
Harris
King
Milwaukee
336.8
109.0
325.2
132.2
176.1
325.0
134.0
213.3
130.0
122.5
210
154
102
142.3
160.0
257.3
231.1
168.0
181.8
161.3
188.1
166.0
174.5
117.5
124.5
168.3
144.8
235.0
178.6
251.5
206.0
138.3
196
159
129
195.0
* Second highest 24-hour concentration in 1972, average across all
monitors in county.
Source: See Section 2.
14-4
-------
Table 14-2
POSSIBLE EFFECTS AND RESPONSES TO PAETICDLATE MATTER IN SIC 344
Physical Effect
Plant Response(s)
Economic Effect ($)
Contamination of welding
joints
Contamination of metal
inventories
Contamination of finished
ox semifinished products
Contamination of other
equipment (e.g., optical
or magnetic tape readers,
transformers, etc.)
Contamination of freshly
painted surfaces
Install dust control systems
Clean metal before welding
Store inventories in covered
or indoor areas
Maintain smaller inventories
Purchase coated metals
Clean metal before use
Store produce inventories in
covered or indoor areas
Maintain smaller inventories
Apply paints or coatings to
products
Increased maintenance
frequency
Use of closed or sealed
equipment
Install dust control systems
Utilize clean rooms
Cost of installing & maintaining
dust control equipment
Cost of installing & operating
metal cleaning equipment
Cost of storage facilities
Loss of quantity discounts on
purchased metal due to smaller
quantities purchased
Production bottlenecks
Higher metal cost
Cost of metal cleaning
Cost of storage facilities
Production bottlenecks
Loss of sales due to "out of
stock"
Cost of painting & coating
Higher maintenance cost
Higher equipment cost
Cost of installing & main-
taining dust control equipment
Cost of constructing & main-
taining clean rooms
14-5
-------
occurred and what the plant responses had been. As noted earlier,
data on economic factors proved more elusive.
Few of the plant managers perceived PM as producing the physical
effects listed in the first column of Table 14-2. This included no
reported PM effects on welding operations or painting operations, and
little or no physical effect on metal inventories. One plant indi-
cated its product underwent a final cleaning and finishing operation
to remove PM and other contaminants acquired during processing or
while in inventory. However, it was not clear how much of the PM came
from ambient air as opposed to internal plant sources. One other
plant indicated a prior problem with dust and soiling caused by
ambient PM generated from a nearby highway. This problem was eventu-
ally solved by installing double-glazed, non-opening windows and addi-
tional air filters.
The minimal perception of PM physical effects by plant managers
raised the question of whether PM effects could be perceived even if
present. That is, the draft analysis predicted PM would cause small
changes in cost and thus PM effects may be present but not observed.
For example, the effects of temperature and moisture may be larger
than those of PM. Yet, it was found that managers did not respond
differently in cities with quite different levels of humidity and
temperature (i.e., Denver and Seattle). One advantage of the econo-
metric methods used in the draft analysis is that they can capture
both perceived and unperceived effects of PM in the industry (see pp.
14-6
-------
7-16 to 7-17). The interviews are likely to reflect only perceived
effects.
In addition to the problem of perception, there is also the
possibility that prior adjustments made to prevent contamination or
corrosion may already have occurred. These adjustments would reduce
the extent to which PM effects might currently be observed. For
example, several of the plants indicated that they practiced one or
more of the activities listed in the second column of Table 14-2.
These included the use of coatings on exposed metal surfaces, indoor
storage of metal inventories, and surface cleaning before painting or
welding. However, it was also indicated that these activities were
undertaken for a variety of reasons (e.g., indoor inventory storage
for prevention of rust or vandalism). The activities were not attri-
buted exclusively, or even primarily, to ambient PM.
On balance, the overall results of the interviews were inconclu-
sive. There was little evidence found suggesting that PM is perceived
as affecting plants in this industry. There was little evidence of
actual effects. At the same time, however, there was an indication
that perception might be difficult even if effects are present. And
some plants undertake various activities that may have been an earlier
response to PM or that would at least reduce the extent to which PM
may currently have an effect.
14-7
-------
STATISTICAL REANALYSIS
After the plant interviews proved inconclusive, it was decided to
undertake further analysis of the econometric model for SIC 344. The
analysis included additional testing, sensitivity analysis and data
evaluation. The process identified two problems with the data for the
District of Columbia (DC). First, the data used for DC were found to
be an unintended mixture of data for DC and for the DC SMSA. When the
data were corrected, the model was re-estimated and the results were
as reported in Section 7 of this report (pp. 7-149 through 7-158). A
comparison of these results with those appearing in the July 1981
draft analysis report shows a decrease in the magnitude of the implied
effect of TSP. However, the estimated coefficients for TSP became
even more statistically significant.
A second feature observed with the DC data is that the calculated
wage rate for DC was found to be 4.4 standard deviations from the mean
wage rate among all counties in the sample listed in Table 14-1. As
discussed in Section 7 of this report, some of the problem may be due
to the limited number of significant digits used by the Census Bureau
in reporting the data. In any case, a test was made to see how this
particular observation may have influenced the results. The model was
therefore re-estimated after excluding the DC data. In this version,
the implied effect of TSP was found to increase but the TSP coeffi-
cients were not as statistically significant.
14-8
-------
The various results from the statistical reanalysis are compared
in Table 14-3. As can be seen, the implied effect of TSP varies from
$247 to $410 per microgram. Both versions with the DC data are
statistically significant at the 5 percent level. The version without
the DC data is not as significant. The implication of these differ-
ences is that the model for SIC 344 is somewhat sensitive to sample
composition. Thus, some caution is warranted in interpreting benefit
estimates developed using the econometric model for this industry.
For reasons discussed in Section 7, benefit calculations for this
industry in the final analysis are based on the model estimated with
the DC data excluded.
Table 14-3
COMPARISON OF RESULTS FOR SIC 344
MTCTSP+ ($)
Level of
Significance (%)
Original
DC Data*
410
3.8
Dataset
Corrected
DC Data**
247
0.8
Without
DC Data**
339
10.6
* As reported in July 1981 Draft Analysis Report.
** As reported in Section 7 of this report.
+ Dollar increase in total production cost per unit increase in
ambient TSP.
14-9
-------
U.S. Environmental Protection Agency
Region V, L;'a;r"y
230 South LV^-.r^n Sliest
Chicago, Iliinois 60604
------- |