BENEFIT AND MET taauiifii- ANALYSIS OF
ALTERNATIVE NATIONAL AMBIENT AIR aDALTTT
STANDARDS FOR PAiri'Tnn^ATR
'VOLUME II
Prepared for:
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
EPA/450/5-83/004b
March 1983
-------
BENEFIT AND NET BENEFIT ANALYSIS OF ALTERNATIVE
NATIONAL AMBIENT AIR QUALITY STANDARDS FOR
PARTICDLATE MATTER
By:
Ernest H. Manuel, Jr.
Robert L. Horst, Jr.
Kathleen M. Brennan
Jennifer M. Hobart
Carol D. Harvey
Jerome T. Bentley
Marcus C. Duff
Daniel E. Klingler
Judith K. Tapiero
With the Assistance of:
David S. Brookshire
Thomas D. Crocker
Ralph C. d'Arge
A. Myrick Freeman, III
William D. Schulze
James H. Ware
MATHTECH, INC.
P.O. Box 2392
Princeton, New Jersey 08540
EPA Contract Number 68-02-3826
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
March 1983
-------
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.
-------
EPA PERSPECTIVE
There has been growing concern with the effectiveness and burden of
regulations imposed by the Federal government. In order to improve the
process by which regulations are developed, Executive Order 12291 was
issued. The order requires that Federal agencies develop and consider, to
the extent permitted by law, Regulatory Impact Analyses (RIA) for the
proposal and promulgation of regulatory actions which are classified as
major. According to the order, a significant component of the RIA is to be
an economic benefit and benefit-cost analysis of the regulatory alternatives
considered. Under the Clean Air Act, the Administrator of EPA may not
consider economic and technological feasibility in setting National Ambient
Air Quality Standards (NAAQS). Although this precludes consideration of
benefit cost analyses in setting NAAQS, it does not necessarily preclude
consideration of benefit analyses for that purpose.
In full support of the Executive Order, the EPA commissioned Mathtech,
Inc. to accomplish an economic benefit and benefit-cost analysis of some of
the alternatives that were thought likely to be considered in the development
of proposed revisions to the NAAQS for particulate matter (PM). The report,
entitled "Benefit and Net Benefit Analysis of Alternative National Ambient
Air Quality Standards for Particulate Matter," documents the results of the
contractor's study. One of the major objectives of the study was to give a
better understanding of the complex technical issues and the resource
requirements associated with complying with the spirit of the Order for the
NAAQS program. In order to achieve this objective, the contractor was
given a wide range of latitude in the use of data, analytic methods, and
underlying assumptions.
It is important to stress that the benefit analysis portion of the
Mathtech study has not had a role to date in the development of proposed
revisions to the NAAQS for particulate matter. Staff recommendations
currently under consideration are based on the scientific and technical
information contained in two EPA documents. They are the "Air Quality
Criteria for Particulate Matter and Sulfur Oxides" and the "Review of the
National Ambient Air Quality Standards for Particulate Matter: Assessment
of Scientific and Technical Information, OAQPS Staff Paper." These documents
have undergone extensive and rigorous review by the public and the Clean
Air Scientific Advisory Committee in accordance with the Agency's established
scientific review policy. Although the Mathtech study reflects the
"state-of-the-art" in particulate matter benefit analysis, the approach and
results have not been subjected to a comparable extensive peer review
process. In addition, some EPA staff have raised questions regarding the
approach taken in the analysis and the significance of the results for
standard setting purposes under the Act. These circumstances do not
necessarily preclude use of the benefit analysis in some manner after
appropriate peer review and further consideration of the questions that
have been raised.
-------
PKEFACE
This report was prepared for the U.S. Environmental Protection Agency
by Mathtech, Inc. The report is organized into five volumes containing a
total of 11 .sections as follows:
Volume I
Section 1:
Section 2:
The Benefit Analysis
The Net Benefit Analysis
Volume II
Section 3:
Section 4:
Appendix:
Health Effects Studies in the Epidemiology Literature
Health Effects Studies in the Economics Literature
Valuation of Health Improvements
Volume III
Section 5:
Section 6:
Section 7:
Section 8:
Residential Property Value Studies
Hedonic Wage Studies
Economic Benefits of Reduced Soiling
Benefits of National Visibility Standards
Volume IV
Section 9:
Section 10:
Air Quality Data and Standards
Selected Methodological Issues
Volume V
Section 11:
Supplementary Tables
IV
-------
ACKNOWLEDGMENTS
While preparing this report, we had the benefit of advice, comments
and other assistance from many individuals. Allen Basala, the EPA Project
Officer, and James Bain, former Chief of the Economic Analysis Branch
(EAB), were especially helpful. They provided both overall guidance on
project direction as well as technical review and comment on the report.
Others in EAB who assisted us included Thomas Walton, George Duggan, and
John O'Connor, the current Chief of EAB.
Others within EPA/OAQPS who reviewed parts of the report and assisted
in various ways included Henry Thomas, Jeff Cohen, John Bachman, John
Haines, Joseph Padgett, and Bruce Jordan.
Several individuals within EPA/OPA also provided comments' or assis-
tance at various stages of the project. These included Bart Ostro, Alex
Cristofaro, Ralph Luken, Jon Harford, and Paul Stolpman.
Others outside EPA who reviewed parts of the report and provided
comments included V. Kerry Smith, Paul Portney, Lester Lave, Eugene Seskin,
and William Watson. Other Mathtech staff who assisted us in various ways
were Donald Wise, Gary Labovich, and Robert J. Anderson. We also
appreciate the assistance of Al Smith and Ken Brubaker of Argonne National
Laboratory who conducted the parallel analysis of control costs and air
quality impacts.
Naturally, it was not possible to incorporate all comments and
suggestions. Therefore, the individuals listed above do not necessarily
endorse the analyses or conclusions of the report.
The production of a report this length in several draft versions, each
under a tight time constraint, is a job which taxes the patience and sanity
of a secretarial staff. Carol Rossell had this difficult task and managed
ably with the assistance of Deborah Piantoni, Gail Gay, and Sally Webb.
Nadine Vogel and Virginia Wyatt, who share the same burden at EAB, also
assisted us on several occasions.
-------
CONTENTS
VOLUME II
Section Page
3. HEALTH EFFECTS STUDIES IN THE EPIDEMIOLOGY LITERATURE
Summary of Results 3-1
Intr oduct ion 3-5
Purpose 3-5
Scope ~. 3-6
Micro- and Macroepidemiclogy Studies 3-6
Sources 3-7
Selection Criteria 3-7
The Pauc ity of Data 3-8
Conversion Between PM Measurement Techniques 3-9
Sulfur Oxides and Particulate Matter 3-11
Selection of Studies 3-12
Acute Exposure Mortality Studies 3-12
Chronic Exposure Mortality Studies 3-16
Acute Exposure Morbidity Studies 3-16
Chronic Exposure Morbidity Studies 3-19
Summary 3-29
Approach to Benefit Estimation 3-29
PM Standards 3-29
Measure of Exposure 3-32
Summary 3-32
Mortality Risk Effects 3-32
Mortality Risk Effects of Acute Exposure 3-32
Morbidity Effects 3-45
Introduct ion 3-45
Acute Morbidity Effects of Acute Exposure 3-51
VI
-------
CONTENTS (Continued)
Section Page
HEALTH EFFECTS STUDIES IN THE EPIDEMIOLOGY LITERATURE
(Continued)
Acute Morbidity Effects of Chronic Exposure 3-58
Chronic Morbidity Effects of Chronic Exposure 3-69
Summary • 3-77
Benefit Estimation 3-81
Discounted Present Value of Benefits 3-81
Aggregate Benefits 3-81
Estimated Benefits 3-82
Estimates of Physical Effects 3-108
Conclusion 3-113
References 3-114
Appendix 3A: Application of Air Quality Data to
Mazumdar H al 3-119
Appendix 3B: Results of Three Additional Morbidity
Studies 3-125
Appendix 3C: Data Sources 3-136
4. HEALTH EFFECTS STUDIES IN THE ECONOMICS LITERATURE
Summary of Results 4-1
Introduct ion 4-6
Criteria for Selecting Studies 4-8
Measurement of Particulate Matter 4-9
Mortality Studies 4-10
Overview of the Approach 4-10
Literature Review 4-16
Summary 4-48
VII
-------
CONTENTS (Continued)
Section Page
4. HEALTH EFFECTS STUDIES IN THE ECONOMICS LITERATURE
(Continued)
Morbidity Studies 4-56
Overview of Approach 4-56
Literature Review 4-60
Summary 4-85
Approach to Benefit Estimation 4-90
Air Quality Data 4-90
Algorithms 4-96
Mortality Effects of Chronic Exposure 4-97
Morbidity Effects of Chronic Exposure 4-98
Benefit Estimation 4-111
Aggregate Benefits 4-112
Benefits 4-113
Estimates of Physical Effects 4-133
Conclusion 4-137
References 4-142
Appendix 4A: Data Sources 4-146
APPENDIX TO VOLUME II: VALUATION OF HEALTH IMPROVEMENTS
Introduction A-l
Alternative Methods for Valuing Reductions in
Mortality Risk A-2
Surveys A-3
Wage Compensation Studies A-5
Literature Review A-l2
Limitations of the Wage Compensation Method A-21
Application of the Wage Compensation Study
Results A-23
-------
CONTENTS (Continued)
Section Page
APPENDIX TO VOLUME II: VALUATION OF HEALTH IMPROVEMENTS
(Continued)
Methods for Valuing Reductions in Morbidity A-24
Reductions in the Loss of Output A-25
Reductions in Restricted Activity Days (RAD) A-26
Reductions in the Consumption of Medical Services . A-27
Conclusion A-27
References A-28
IX
-------
FIGUXBS
VOLUME II
Figure No. Page
3-1. Calculation of Benefits of Mortality Risk Redactions ... 3-38
3-2. Calculation of Morbidity Estimates (Samet jet .al.) 3-56
3-3. Calculation of Morbidity Estimates (Saric ,e_t al.) 3-64
3-4. Calculation of Morbidity Estimates (Ferris .e_t .al.) 3-74
-------
TABLES
VOLUME II
Table No. Pa2e
3-1. Air Quality Standards 3-1
3-2. Health Benefits Under Alternative Particulate Standards . 3-3
3-3. Mazumdar e_t al. Regression Results Relating Daily
Mortality to Daily Pollution 3-14
3-4. Saric et al. Comparison of Acute Respiratory Disease
Inc idence .• 3-24
3-5. Summary of Selected Studies 3-30
3-6. Alternative Particulate Hatter Standards 3-31
3-7. Exposure Measures for Underlying Studies and Estimates .. 3-33
3-8. Coefficients (Percents) from Mazumdar et al. in mg/m ... 3-35
3-9. Biases in Estimated Based on Mazumdar et al 3-45
3-10. Disease Incidence by Age 3-59
-11. Estimated Per Capita Benefit Per Unit Reduction in TSP .. 3-78
Common Sources of Bias in Morbidity Benefit Estimates ... 3-79
Specific Sources of Bias in the Morbidity Benefit
Estimates 3-80
v.imated Benefits for Mazumdar Acute Mortality Study -
fits Occurring Between 1989 and 1995 - Scenario:
B PM10 - 70 AAM/250 24-hr 3-83
d Benefits for Mazumdar Acute Mortality Study -
Occurring Between 1989 and 1995 - Scenario:
*) - 55 AAM 3-84
-------
TABLES (Continued)
Table No. Page
3-16. Estimated Benefits for Mazumdar Acute Mortality Study -
Benefits Occurring Be tire en 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/250 24-hr 3-85
3-17. Estimated Benefits for Mazumdar Acute Mortality Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/150 24-hr 3-86
3-18. Estimated Benefits for Mazumdar Acute Mortality Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 75 AAM/260 24-hr 3-87
3-19. Estimated Benefits for Mazumdar Acute Mortality Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 150 24-hr 3-88
3-20. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 -.Scenario:
Type B PM10 - 70 AAM/250 24-hr 3-89
*
3-21. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM 3-90
3-22. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/250 24-hr 3-91
3-23. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/150 24-hr 3-92
3-24. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 75 AAM/260 24-hr 3-93
3-25. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 150 24-hr 3-94
3-26. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 70 AAM/250 24-hr 3-95
Xll
-------
TABLES (Continued)
Table No. Pane
3-27. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM 3-96
3-28. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/250 24-hr 3-97
3-29. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/150 24-hr 3-98
3-30. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 75 AAM/260 24-hr 3-99
3-31. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
• Type B TSP - 150 24-hr 3-100
3-32. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 70 AAM/250 24-hr 3-101
3-33. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM 3-102
3-34. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/250 24-hr 3-103
3-35. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type B PM10 - 55 AAM/150 24-hr 3-104
3-36. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 75 AAM/260 24-hr 3-105
3-37. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1987 and 1995 - Scenario:
Type B TSP - 150 24-hr 3-106
X11I
-------
TABLES (Continued)
Table No. Page
3-38. Estimated Benefits for Mazumdar Acute Mortality Study -
Benefits Occurring Between. 1989 and 1995 - Scenario:
Type A PM10 - 70 AAM/250 24-hr 3-109
3-39. Estimated Benefits for Samet Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type A PM10 - 70 AAM/250 24-hr 3-110
3-40. Estimated Benefits for Saric Acute Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type A PM10 - 70 AAM/250 24-hr 3-111
3-41. Estimated Benefits for Ferris Chronic Morbidity Study -
Benefits Occurring Between 1989 and 1995 - Scenario:
Type A PM10 - 70 AAM/250 24-hr 3-112
3B-1. Results of Douglas and Waller 3-126
3B-2. Coefficients Derived from Application of the First
Functional Form 3-128
3B-3. Results from Application of the Second Functional Form .. 3-129
3B-4. Results of Lunn .et jil 3-131
3B-5. Results of Col ley and Brasser 3-132
3B-6. Coefficients Derived for the First Functional Form 3-133
3B-7. Coefficients Derived for the Second Functional Form 3-134
4-1. Health Benefits of Attaining Alternative Particulate
Matter Standards 4-2
4-2. 1960 and 1969 Unadjusted and Age-Sex-Race-Adjusted
Mortality Rate Equations 4-19
4-3. Comparison of TSP Elasticities from Lave and Seskin
and Gre gor 4-28
4-4. Reduced Form Medical Care and Total Mortality
Equations from Crocker et a_l 4-37
xiv
-------
TABLES (Continued)
Table No, Page
4-5. Comparison of Lave and Seskin, and Chappie and Lave
Unadjusted Total Mortality Rate Equations 4-44
4-6. Summary of Macroepidemiological Studies Analyzing the
Chronic Effects of Particulate Matter 4-49
4-7. TSP Levels Used in. Macroepidemiological Studies 4-51
4-8. Comparison of TSP Elasticities Calculated from
Macroepidemiological Mortality Studies 4-53
4-9. Range of Coefficients Measuring the Relationship
Between the Mortal ity Rate and TSP 4-56
4-10. Results from Crocker et. JLi- Morbidity Analysis 4-63
4-11. Definitions of Variables Used in Crocker et al.
Morbidity Analysis 4-65
4-12. Range of Labor Productivity Effects for a 1 ng/m3
Change in TSP from Crocker ojt al 4-73
4-13. Variables Used in Ostro Acute Morbidity Study 4-74
4-14. Estimation of WLD2 for Workers Aged 18-44 4-77
4-15. Estimation of WLD2 for Workers Aged 45-65 ' 4-78
4-16. Estimation of RAD for All Nonworkers 4-81
4-17. Change in WLD and RAD for a 1 ug/m3 Change in TSP
Estimated from Ostro 4-86
4-18. TSP Levels Used in Acute Illness Studies 4-86
4-19. Comparison of Labor Productivity Effects from Acute
Illness Obtained by Crocker .et *1. and Ostro 4-87
4-20. Range of Effects of a 1 ug/m3 Change in TSP on
Acute Illness 4-89
4-21. Effect of a 1 |ig/m3 Change in TSP on the Acute
Illness of Nonworkers 4-89
-------
TABLES (Contiamed)
Table No. Page
4-22. Labor Productivity Effects Resulting from a 1 (ig/m
Change in TSP 4-90
4-23. Alternative Particnlate Matter Standards 4-92
4-24. Air Pollution Monitors Used in Health Studies 4-94
4-25. TSP Levels Used in Health Studies 4-95
4-26. Data Used in Calculating Benefits 4-113
4-27. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Bet-ween. 1989 and
1995 - Scenario: Type B PM10 - 70 AAM/250 24-hr 4-114
4-28. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 55 AAM 4-115
4-29. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 55 AAM/250 24-hr 4-116
4-30. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 55 AAM/150 24-hr 4-117
4-31. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Between 1987 and
1995 - Scenario: Type B TSP - 75 AAM/260 24-hr 4-118
4-32. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Between 1987 and
1995 - Scenario: Type B TSP - 150 24-hr 4-119
4-33. Estimated Benefits for Ostro, Crocker et. al. Acute
Morbidity Studies - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 70 AAM/250 24-hr 4-120
4-34. Estimated Benefits for Ostro, Crocker et al. Acute
Morbidity Studies - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 55 AAM 4-121
xvi
-------
TABLES (Continued)
Table No. Page
4-35. Estimated Benefits for Ostro, Crocker et al. Acute
Morbidity Studies - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 55 AAM/250 24-hr 4-122
4-36. Estimated Benefits for Ostro, Crocker ejb al. Acute
Morbidity Studies - Benefits Occurring Between 1989 and
1995 - Scenario: Type B PM10 - 55 AAM/150 24-hr 4-123
4-37. Estimated Benefits for Ostro, Crocker et al. Acute
Morbidity Studies - Benefits Occurring Between 1987 and
1995 - Scenario: Type B TSP - 75 AAM/260 24-hr 4-124
4-38. Estimated Benefits for Ostro, Crocker et al. Acute
Morbidity Studies - Benefits Occurring Between 1987 and
1995 - Scenario: Type B TSP - 150 24-hr 4-125
4-39. Estimated Benefits for Crocker .et al. Chronic Morbidity
Study - Benefits Occurring Between 1989 and 1995 -
Scenario: Type B PM10 - 70 AAM/250 24-hr 4-126
4-40. Estimated Benefits for Crocker e_t al. Chronic Morbidity
Study - Benefits Occurring Between 1989 and 1995 -
Scenario: Type B PM10 - 55 AAM 4-127
4-41. Estimated Benefits for Crocker et al. Chronic Morbidity
Study - Benefits Occurring Between 1989 and 1995 -
Scenario: Type B PM10 - 55 AAM/250 24-hr 4-128
4-42. Estimated Benefits for Crocker et al. Chronic Morbidity
Study - Benefits Occurring Between 1989 and 1995 -
Scenario: Type B PM10 - 55 AAM/150 24-hr 4-129
4-43. Estimated Benefits for Crocker H ail.. Chronic Morbidity
Study - Benefits Occurring Between 1987 and 1995 -
Scenario: Type B TSP - 75 AAM/260 24-hr 4-130
4-44. Estimated Benefits for Crocker e_t al. Chronic Morbidity
Study - Benefits Occurring Between 1987 and 1995 -
Scenario: Type B TSP - 150 24-hr 4-131
4-45. Estimated Benefits for Lave and Seskin Chronic
Mortality Study - Benefits Occurring Between 1989 and
1995 - Scenario: Type A PM10 - 70 AAM/250 24-hr 4-134
-------
TABLES (Coatlaved)
Table No. Page
4-46. Estimated Benefits for Ostro, Crocker et al. Acute
Morbidity Studies - Benefits Occurring Between 1989 and
1995 - Scenario: Type A PM10 - 70 AAM/250 24-hr 4-135
4-47. Estimated Benefits for Crocker et. al. Chronic Morbidity
Study - Benefits Occurring Between 1989 and 1995 -
Scenario: Type A PM10 - 70 AAM/250 24-hr 4-136
4-48. Summary of Potential Biases in Benefit Calculations 4-139
A-l. Functional Form of Equations Used in Hedonic Wage
Studies A-7
A-2. Summary of Wage Compensation Studies A-l8
A-3. Alternative Estimates of Marginal Risk Valuations A-l9
A-4. Estimates of the Value of a Marginal Reduction in
Death Risk A-21
XVlll
-------
SECTION 3
HEALTH EFFECTS STUDIES IN THE EPIDEMIOLOGY LITERATURE
-------
SECTION 3
HEALTH EWECTS STUDIES IN THE EPIDEMIOLOGY LITERATURE
SUMMARY OF RESULTS
la tliis section, the medical epidemiology literature is used to
develop concentration-response functions relating mortality risk and
morbidity to the level of particulate matter (PM). From these
concentration-response functions, the health effects of six alternative PH
standards shown in Table 3-1 are estimated. These effects are valued using
the methods developed in the Appendix to Volume II.
Table 3-1
AIR QUALITY STANDARDS
Standard
1
2
3
4
5
6
Pollutant
PM10
PM10
PM10
PM10
TSP
TSP
Annual
Mean*
70
55
55
55
75
—
24-Hour
Value**
250
—
250
150
260
150
Implementation
Date
1989
1989
1989
1989
1987
1987
* Annual arithmetic mean for all standards except for No. 5. Annual
geometric mean for standard 5.
** 24—hour reading that is expected to occur once a year for PH standards
and 24-hour second high for TSP standards.
3-1
-------
Column 2 indicates the particle measure on which each standard is
based. PM10 includes particles less than 10 urn in aerodynamic diameter,
while TSP includes total suspended particulates. Column 3 expresses each
standard in terms of the annual average, while Column 4 expresses it in
terms of the 24-hour reading. When the standard is stated in terms of both
the annual average and 24-hour value, the more stringent averaging time is
used, as discussed in Section 9. Column 5 of Table 3-1 lists the implemen-
tation dates for each standard.
For each standard, the total discounted present value of benefits for
the period from the attainment year through 1995 is estimated using a 10
percent rate of discount. A range of PH health effects is compatible with
the results of the epidemiology studies. In consideration of this
ambiguity, a range of benefit estimates is derived for each study.
Benefits are given in 1980 dollars.
The benefits achieved under each of the six standards are shown in
Table 3-2. Benefits for additional standards are presented in Section 11.
Under Standard 1, the mortality risk benefits of reduced acute exposure
range from $0.037 billion to $14.86 billion with a point estimate of $1.12
billion.
In addition to the benefits of reduced mortality risk, the benefits of
reduced morbidity are estimated from the medical epidemiology literature.
The effects of reduced levels of particulate matter on direct medical
expenditures (DME), work-loss days (WLD), and restricted-activity days
(RAD) are valued. Table 3-2 shows the acute morbidity benefits of reducing
acute exposure. Under Standard 1, these benefits range from $0.147 billion
to $11.91 billion with a point estimate of $1.32 billion.
The acute morbidity benefits of reducing chronic exposure are also
shown in Table 3-2. Under Standard 1, these benefits range from $0.0 to
$1.44 billion, with a point estimate of $0.0 billion.
3-2
-------
«
09
H
03
1
B »
M H
H «
Oi o
(N U
I > O
el m
o o o d
\o d oo v*
oo »o vo m
CM 00 A
-------
Finally, the chronic morbidity benefits of reducing chronic exposure
are presented in Table 3-2. Under Standard 1, these, benefits range from
$0.121 billion to $0.128 billion with a point estimate of $0.124 billion.
For all mortality and morbidity categories, the benefits achieved
under the five other standards are also shown in Table 3-2. Benefits
increase with the stringency of the standard.
These benefit estimates have a number of limitations in addition to
the general points discussed in Section 1. First, the mortality and
morbidity studies provide very limited information from which to develop
concentration-response functions. Thus, there is much uncertainty present
in the estimates.
Second, the studies do not consider actions that individuals may take
to offset the effects of particulate matter on their health. If the
relationship between PM and health status in these studies is estimated
after this behavior has occurred, the health benefits in this section may
be underestimates of the actual benefits of PM reductions.
Third, the results for small study samples are generalized to all of
the counties in our analysis. Since the health effects of PH may differ
with the characteristics of the population, exposure measures, and area
considered, application of study results to our analysis may result in an
under- or overestimate of benefits.
Fourth, most of the studies do not control for the effects of
different pollutants. Since, the ambient concentrations of various pollu-
tants may be correlated, attribution of observed health effects to changes
in one PM measure may bias the estimated benefit upwards.
Fifth, much of the data required for benefit calculations often are
not available at the county level. The use of state or national data as a
substitute may affect the results in a variety of unknown ways.
3-4
-------
Sixth, the benefit estimates do not consider the full range of health
effects. Effects of chronic exposure on mortality and effects on non-
respiratory disease are not included. In addition, the full benefits of
the reduced pain and suffering resulting from reduced morbidity may not be
captured by our estimates.
Finally, health studies available for the benefit analysis do not
incorporate particle size information. Benefits shown in the table for the
PM10 standards are based on the TSP change that results. Comparisons
across PH10 and TSP standards thus reflect only differences in relative
stringency in terms of the TSP reduction; they do not reflect differences
in particle size. If PH10 standards lead to proportionately larger
reductions in PM10 relative to TSP, benefits for the PH10 standards may be
underestimated. Data from the cost and air quality analysis suggest
that proportionately larger reductions do not generally occur. However,
approximations in that analysis are such that the comparisons should still
be interpreted with caution, as signified by the line in the table
separating the two groups of standards.
INTRODUCTION
Purpose
The purpose of this section is to estimate the health benefits that
potentially could result from implementation of alternative primary
national ambient air quality standards (PNAAQS) for particulate matter
(PM). The basis for this analysis is a group of existing studies in the
medical epidemiology literature. That is, we have not attempted to collect
new data or postulate original models or methods. Rather, our efforts have
been directed at the identification of relevant studies from the epidemio-
logical literature, the setting of criteria for evaluation of these
studies, the critical review of the studies themselves, the specification
of the concentration-response models obtained from this review, the estima-
tion of changes in human mortality or morbidity that would be realized were
3-5
-------
alternative standards implemented, and the economic valuation of such
health benefits.
Scope
This section and Section 4 share similar goals of benefits analysis.
The major difference between the sections is in the allocation of the
epidemiological literature selected for review. The studies covered here
are those of the more "traditional" sort: medical epidemiology. Reviewed
in Section 4 are those studies typically found in the economics literature.
Our division of effort recognizes the somewhat parallel but independent and
occasionally antagonistic developments of these two bodies of literature.
This division is not based strictly upon data sources, methods, or results.
Micro** ^'"il M>croet>i.deMi.ologT
Ideally, the medical epidemiology studies (hereafter referred to
simply as "epidemiology studies") are microepidemiology studies based upon
individual measurement of possible confounding factors (e.g., smoking,
occupation, age, sex, race, etc.) and health endpoints. That is, the level
of risk to the individual is assessed directly rather than being inferred
from a "population risk". In this way, one avoids the "ecologic fallacy"
— attribution of characteristics of a population to individuals.
The use of disaggregated data, however, has a number of weaknesses.
The costs of data collection for a well-designed and conscientiously
executed epidemiological study may be prohibitive, especially if the effort
is intended to be sufficiently sensitive to both mortality and morbidity
effects in the exposure range of the current primary standard. The macro-
epidemiology, or population health risk, research efforts have the distinct
advantage of being able to exploit (for the most part) existing data
sources. For this reason, the macroepidemiology models may also be updated
or revised with greater ease and frequency and may cover a wider sample
than microepidemiology studies. In addition, the collection of information
from individuals for epidemiology studies may result in inaccurate data if
3-6
-------
participants engage in strategic behavior in providing responses or have
poor recall of the information required for retrospective studies.*
Sources
A variety of sources were consulted in forming the original review
pool of epidemiological studies dealing with particulate air pollution and
human mortality and morbidity. Information came primarily from the
following four reports and papers:
Review of the National Ambient Air Quality Standards for
Particulate Matter: Draft Staff Paper (2).
Epidemiological Studies on the Effects of Sulfur Oxides on
Particulate Hatter and Hunan Health (3).
Holland, Bennett, Cameron, Florey, Leeder, Schilling, Swan
and Waller (4).
Ware, Thibodeau, Speizer, Colome and Ferris (5).
Selection Criteria
The above sources yielded a large number of studies for review. Our
next task was to apply a set of inclusion criteria for selecting the best
candidates, where best in this context meant likely to provide both fruit-
ful and valid benefits estimates. The criteria applied were:
The level of particulate matter must be quantified, or
easily rendered so.
Health effects must be quantifiable.
Relevant variables and confounding factors should be con-
sidered in the risk analysis.
The relationships between levels of particulate matter and
health should be plausible and consistent.
* See Shy (1) for a further discussion of micro- and macroepidemiology
studies.
3-7
-------
A concentration-response curve or equation must be pre-
sented or calculable from the reported results.
In general, the results should be transferrable to our
health valuation data.
Studies which failed to meet one or more of these criteria were
occasionally retained when the results were of particular importance, or
when data in a given health area were limited.
The Pancitv of Data
Three of the preceding criteria are especially crucial to the calcula-
tion of health benefits. Both levels of particulates and health effects
must be expressed numerically, and a quantitative concentration-response
relationship must be available (or calculable) to relate the two. These
three items are basic requirements for estimating changes in mortality
and/or morbidity for concomitant changes in levels of PM.
Unfortunately, simple concentration—response relationships are not
often found in the human epidemiology literature. As Holland .e_t .§_!. (4)
have noted.
Host toxicological investigations have been undertaken on
animals, and the extrapolation of animal experience to man has
many dangers. ... Such evidence as there is from human studies
is difficult to interpret in view of the need to disentangle the
various possible factors influencing mortality and morbidity,
and, in spite of the large number of such studies, only a
minority can be considered scientifically reliable. [Holland et
al- (4>. p. 652]
In addition, Ware (5) and his coworkers, in their review of observational
studies of the effects of TSP and S02, acknowledge that
... the epidemiologic. data base is extremely weak. In particu-
lar, it is insufficient to distinguish between a threshold
hypothesis, that health effects are seen only above certain
concentrations, and a monotonic exposure-response hypothesis,
3-8
-------
that health effects increase (perhaps very slightly) with air
pollution over a very vide range. [Ware .e_t .§_!• (5), p. 61].
In other words, not only the form but also the existence of a
concentration-response relationship at low to moderate levels of PM is in
dispute.
With concensus in the literature only that robust concentration-
response curves for humans cannot be estimated at this time [see Shy (1)
for instance], we were faced with two disparate courses of action in our
efforts at benefits calculations. Lacking complete concentration-response
relationships, we could abandon the whole enterprise or, given EPA's man-
date to conduct a benefit-cost analysis, we could choose to make use of the
available studies, shortcomings and all, for our analysis. We chose a less
extreme version of the latter course. We chose to make use of the studies
in a qualified fashion which makes clear how their shortcomings potentially
affect our results. This approach makes it easier to evaluate both the
magnitude and the uncertainty in our estimates.
Before reviewing the studies which we selected for our benefit calcu-
lations, we will discuss the problems involved in making conversions
between different measurement techniques for particulate matter, as this
issue had a significant impact on the study selection process. In addi-
tion, the virtually interchangeable roles held by PH and SO- for most
studies will be reviewed, not so much because study selection was affected,
but because of the potential mitigating effect of the PM-SO- ambiguity on
any conclusions.
Conversion Between PM Measurement T**i'h'*'*">«
Several methods are popularly used for determination of the level of
suspended particulates in epidemiology studies. Perhaps the most common in
the U.S. is the high volume sampler method, which yields concentrations in
jig/m of total suspended particulates. The coefficient of haze or CoH
technique is based on transmittance of light through filter paper. The
3-9
-------
British. Smoke (BS) filter method has a long history of use throughout
Europe, and results are reported in pg/m of smoke.
The current U.S. primary standards being evaluated in this analysis
are set in terms of level of PH10. As discussed in Section 1, the
availability of TSP concentration data makes it possible to estimate
approximate benefits for these standards using health studies based on TSP.
This approach simply involves estimating the benefits of the TSP reduction
that resulted from PM10 controls.
Section 1 also discussed the fact that smaller particles may be more
significant in producing adverse health effects. Thus, if the particle
size composition of reductions in TSP under PH10 and TSP controls differs,
the benefits for the two types of standards may not be directly comparable.
Preliminary analysis of this issue to date, however, indicates that the
fraction of PM10 does not appreciably change under imposition of either
PH10 or TSP controls! Therefore, it may be appropriate to use TSP health
studies to measure the benefits of PM10 controls by looking at the change
in TSP. However, because of uncertainty concerning the method of
estimating the fraction of PH10 in the air quality analysis, comparisons
between results for TSP and PM10 comparisons should be interpreted with
caution.
To apply studies which examine the effects of BS, the reductions in
TSP will be converted to BS reductions. Unfortunately, the TSP and BS
measurement methods cannot be interrelated in any simple manner. Holland
et al. (4) concluded, "The measurement of suspended particulate matter has
been seen to present a number of unusual problems. Since its physical and
chemical properties are not uniquely defined, and may vary widely from one
locality to another, the method of measurement plays an important part in
characterizing it. The two most widely used methods, the British smoke
filter (BS) and the high volume sampler (HV) do not measure the same
properties, and the results are clearly not directly interchangeable"
[Holland .et al.. (4), p. 552].
3-10
-------
In isolated instances, differing measurement techniques have been used
side-by-side, and approximate conversions have been made from TSP to BS.
Several authors have attempted to generalize the results of such compari-
sons to other localities and pollutant sources. As the comments of Holland
et al. (4) above indicate, this strategy is fraught with peril. The
authors of the Criteria Document object even more vehemently, insisting
that "site specific calibrations ... are therefore necessary in order to
obtain approximate estimates of atmospheric PM concentrations based on the
BS method." [Criteria Document (3), p. 14-9]. The implication is clear:
unless site—specific calibration data are available for converting BS or
CoH to TSP, our primary study selection criteria that the level of PM must
be quantifiable is violated.
The study deemed particularly appropriate for our acute mortality
analysis used London PM exposure levels reported as BS. Luckily, site-
specific London conversions based on Commins and Waller (6) as reported by
Holland ,e_t .§_!. (4) can be used for this study. A large number of research
reports, however, especially those in the chronic morbidity literature,
failed the calibration screen and had to be discarded. It is a major
weakness of the results presented in this section that some of the health
effects could not be related precisely to levels of PM, and thus that
potentially significant information could not be incorporated.
Sulfur Oxides quid Partiaulate Matter
Measurements of sulfur oxides and particulate matter are often highly
positively related. For instance, Maxtin and Bradley (7) reported a corre-
lation of 0.894 for the logarithm of sulfur dioxide atmospheric pollution
and the logarithm of black suspended matter in London in the winter of
1958-59. This high degree of covariation makes the process of inferring
causality for health effects due to PM especially difficult. As Mazumdar
et al. (8) noted (referring to the London studies) "... differentiation of
separate effects of smoke and S0« was found to be impossible because the
two pollutants were so highly correlated (page 1-1)." In fact, level of
SO- may be used as a proxy or index variable for particulates as was done
3-11
-------
in the Glasser and Greenburg (9) study of pollution and weather in New York
City. As a consequence, attribution of health effects to PH alone is often
difficult. Studies for which the PM-SO- issue is a particular problem will
be identified as we proceed.
SELECTION OF STUDIES
This subsection identifies the studies that were ultimately selected
for use in estimating benefits. It also identifies several additional
promising studies which were strongly considered for inclusion, but
eventually rejected based on the previously stated selection criteria. The
studies can be divided into four groups, based on the types of exposures
and effects under study:
• Acute exposure mortality.
• Chronic exposure mortality.
• Acute exposure morbidity (acute or chronic).
• Chronic exposure morbidity (acute or chronic).
Acute and chronic exposure are measured by daily and annual exposure
levels, respectively. Acute morbidity indicates short-term illness such as
pneumonia, while chronic morbidity indicates persistent, long-term illness
such as asthma or chronic bronchitis.
Acute Kypos'i'fe Mortali.tr Studios
The basic study selected in this category is Mazumdar, Schimmel and
Higgins (8). This longitudinal study of daily mortality during 14 London
winters owes much to the earlier works of Martin and Bradley (7), and
Martin (12). The Mazumdar et al. work is attractive for the following
reasons :
The seven air pollution monitoring stations used were those
originally selected by Martin and Bradley as representative
of the air pollution levels in the county of London.
3-12
-------
The mortality data came from a highly reliable source —
the British Office of Population Census and Survey.
"Episodic" (high pollution) and "non-episodic" periods of
air pollution were modeled both separately and jointly.
Both same-day and lagged models for the health effects of
air pollution were used.
The association between daily mortality and daily pollution
levels was examined for "episodic", "non-episodic", and
pooled data.
Site-specific BS-TSP calibration is available for central
London.*
The Mazumdar et al. study contains a quartile analysis that attempts
to isolate the effects of BS and SOj. The effect of changing BS levels for
fixed levels of SO- is analyzed by forming smoke quartiles nested within
SO2 quartiles. The procedure is reversed to find the effect of changing
SO- levels. Based on the results of the nested quartile analysis, Mazumdar
et al. conclude that the association between daily mortality and daily
pollution is principally due to BS.
The Mazumdar et al. study also reports estimated concentration-
response curves, based on regression analysis of daily mortality with daily
pollution. The regression results based on the combined sample of episodic
and non-episodic data are summarized in Table 3-3. The dependent variable
is excess daily winter mortality expressed as a percentage of mean winter
mortality; the alternative independent variables are various combinations
of unlagged daily S02 and smoke measured in mg/m . The S02 and S02~smoke
interaction terms are not significant when a smoke variable is included.
Mazumdar et al. conclude that their results are equally suggestive of
either a linear or quadratic concentration-response function involving only
unlagged smoke. In the linear model, which is consistent with the
* The Criteria Document [(3), pp. 14-18] states that the 1958-1963 mass to
reflectance calibrations in the seven stations "conform reasonably well"
to the calibration in central London. After 1963, calibration shifted
due to changes in chemical composition.
3-13
-------
Table 3-3
MAZUMDAR ET AL. REGRESSION RESULTS RELATING
DAILY MORTALITY TO DAILY POLLUTION*
Equation
1
2
3
4
5
6
Independent Variables (percentages)
so2
17.31**
(1.23)
—
—
—
—
5. 83*
(3.62)
(S02)2
—
5.91**
(0.57)
—
—
—
4.19+
(3.16)
Smoke
__
—
19.14**
(1.44)
—
—
18.83**
(4.74)
(Smoke)2
—
—
—
9.20**
(0.96)
—
-4.60"1"
(3.90)
(S02-Smoke)
—
—
—
—
9.27**
(0.86)
-4.53+
(6.75)
* Results are for the combined episodic and non-episodic data, with pollu-
tion measured in mg/m . Standard errors are in parentheses.
** Significant at p < 0.01.
+ Not significant at p < 0.10.
continuous health effect hypothesis, a 0.019 percent increase in daily
9
mortality is explained by a one ug/m increment in daily smoke.* The
coefficient for the quadratic model is 9.2 x 10 percent. While the
quadratic model is sometimes misleadingly described as a threshold model,
it yields effects at all BS levels.
* Both the linear and quadratic concentration-response functions derived
from Hazumdar et al. have zero intercepts.
3-14
-------
The problems with using this study include conversion from PM10 to
standard smoke, variation in PM composition over the study period, and
generalization to U.S. sites based on London pollution and mortality
figures. For example, the exposure of the London population for a given
monitored level of outdoors British smoke may be higher than the exposure
for the United States population because of a lower degree of sealing in
British buildings. Then the effects of a change in the pollution level
will be overestimated by application of the coefficient derived by Mazumdar
ejt al. to U.S. counties [(3), p. 102].
In addition, the quartile procedure may bias the estimates of pollu-
tant effects.* Therefore, it cannot be definitively concluded from the
quartile analysis that most health effects can be ascribed to particulates.
If SOj influences mortality, its exclusion from the regression equations
will bias the BS coefficient upwards.
The method of removing variations in data due to weather before
analyzing the effect of the pollutants, on the other hand, may produce a
downward bias. Mazumdar gt al. purge their data of weather effects by
regressing the mortality and pollution variables on a set of weather. The
residuals ("corrected values") then are used to estimate the effects of the
pollutants on mortality. The regression coefficients represent only those
effects of SO2 and BS that are not correlated with weather. The greater
the correlation between weather and pollution, the larger is the downward
bias.** Despite this problem and the other difficulties outlined above,
the Mazumdar et al. results will be used because of the lack of alternative
estimates of mortality risk effects that can be employed in benefit calcu-
lations.
The only other study that was strongly considered for inclusion in
this health effects category was Schimmel and Greenburg (15). The primary
reason for consideration was that the mortality data were for a U.S. city
* See Pitcher (13) for more discussion.
** See Goldberger (14) for discussion.
3-15
-------
(New York City), albeit an arguably atypical one. In addition, the authors
made use of several types of corrections for seasonal and temperature
effects, ranging from none to quite elaborate. The effect of these
transformations was to produce a set of pollution-mortality regression
coefficients reflecting various levels of conservatism in attribution of
excess mortality to daily pollution levels. This set of coefficients was
seen as a promising facet in our efforts to produce a range of estimates,
rather than simple point estimates.
Unfortunately, the study suffers from several major problems. Perhaps
the most oft-cited is that only one air pollution monitoring station was
used for the entire city. A second major difficulty confronting the use of
the study here is the problem of PM10 to CoH conversion (the study was done
in terms of CoH). Because of the problem of conversion, the models are
inadequate for present use.
Bxpos^Te Mortality
No studies of either general or disease-specific mortality involving
valid quantitative data were identified from any of our sources. There-
fore, no benefits estimates for chronic exposure mortality were attempted.
Ac ttt e En>os'nTe Morbidity
In addition to the Martin (12) study cited previously, the other
likely candidates in this category appeared to be Lawther (16); Lawther,
Waller and Henderson (17); and Samet et aj. (18). Unfortunately, the Martin
paper is based upon excess total hospital admissions — a crude index of
acute morbidity. More seriously, no concentration-response relationship
was provided or could be derived from the limited data given. For this
reason, the study was not amenable to health benefits valuation.
The Lawther studies employed a diary recording technique for health
effect data collection, yielding a more acceptable accounting of morbidity.
Lawther's results, however, were presented simply as superimposed
3-16
-------
temporally—indexed graphs of morbidity, smoke, and SC^. It is obvious from
inspection that higher levels of particulates (especially the "peaks") are
related to increased morbidity, but no statistical models which could be
used to derive a concentration-response function were presented. There-
fore, this work must also be relegated to the categories of both
interesting and quite convincing, but inappropriate for our needs.
The remaining study found which showed promise was the recent study by
Samet, Speizer, Bishop, Spengler and Ferris (18). The authors abstracted
records of emergency room visits to the major hospital facility in
Steubenville, Ohio during March, April, October and November of 1974-1977.
Daily TSP, SO-, NO-, CO, 0, and meteorologic measurements were obtained
from one site located centrally in the town's valley. Twenty-four hour
means for TSP at a monitor near the hospital ranged from 14 to 696 ug/m
during the study period with a mean of 156. Twenty-four hour means for SO-
and NOj were 90 and 40 ug/m . Adjustments were made for weekly, seasonal,
and yearly cycles in emergency room visits. Pollution measurements were
determined not to be cyclic by visual inspection.
Data in the Samet et al. study were analyzed using two separate
techniques — an analysis of adjusted deviations and regression analysis.
In the first analysis, the mean deviation of emergency room visits for
various disease categories was estimated for strata defined by TSP-level
quartiles and by maximum temperature dichotomized at the monthly mean. The
deviations were not found to be statistically significantly related to
particulate pollution level in any consistent way. A regression model of
daily maximum temperature and unlagged TSP on number of emergency room
visits for respiratory conditions, however, yielded a TSP coefficient which
is significant at p < 0.05.* Essentially similar results were obtained in
a second model containing temperature and SO-. At the sample mean, the TSP
result suggested a 0.03 percent increase in daily emergency room visits for
respiratory conditions per jig/m increase in daily TSP. In addition, it
* Lagged TSP variables were excluded because they did not attain statis-
tical significance in stepwise regressions.
3-17
-------
should be noted that the largest (positive) deviation from average number
of emergency room visits for all causes occurred for the highest pollutant
quartile (TSP >. 202 ug/m3).
This study may be faulted for a variety of reasons. Only one air
quality monitoring site was used. Emergency room visits are a crude index
of acute health effects. Generalization from one hospital in one city is
tenuous. The size of the population served by the hospital is uncertain.
The pollution effect found was not as starkly convincing as perhaps the
(albeit nonstatistical) results of Lawther et al. (16,17) discussed
earlier.
An additional problem with the study for use in a benefits analysis is
that the authors do not report any results for a regression model
containing both TSP and SOj. Rather, both pollutants were analyzed
independently. Thus, one cannot draw any strong conclusions as to the
relative importance of the two pollutants in relation to the observed
effects.
In spite of these weaknesses, we have chosen to include the Samet et
al. study for acute exposure morbidity calculations. As mentioned earlier,
the study provides a concentration-response relationship in the form of a
regression equation. Unrelated but corroborative evidence for presence of
health effects in Steubenville is provided by the work of Dockerv et al.
(19) on spirometry results in children. In addition, the direct availa-
bility of TSP as an indicant of particulate pollution obviates the error-
prone conversions already discussed. Nonetheless, we note that the changes
in admissions estimated using the Samet et al. TSP regression results are
likely to be biased upwards by the omission of SO- from the model.
It should also be noted that the Samet e_t jd. study will be used to
evaluate only acute morbidity effects of acute exposures, as discussed
later. No usable epidemiology studies could be located concerning possible
chronic morbidity effects of acute exposures.
3-18
-------
**T8 Morbidity Studios
Tlie lack of site-specific calibration for both CoH and BS was
especially frustrating in this group of studies. Several papers that had
to be rejected due to PH quantification problems are first reviewed to
provide some feel for the potential benefits data sources that could not be
used. In Appendix 3B, as a cross— check on our benefit estimates, the
magnitude of the effects identified by three of these studies is compared
to the magnitude of the effects identified by the studies used as a basis
for calculations. Following this review, the three studies that did pass
the calibration screen are presented. Unfortunately, even these papers did
not yield tractable concentration— response functions.
Under more favorable circumstances, a primary reference for these
calculations would have been Colley and Brasser (20). This study made use
of data from eight European countries on children at the grade-school
level. It is the best cross-cultural study that we encountered in our
review. In the study, an attempt was made to assess, via questionnaire and
*
clinical examination, the relationship between air pollution and health in
children between the .ages of approximately 8 and 11 years. Children were
chosen for study because of their relative freedoms from the smoking habit
and adverse occupational exposure. Actual age ranges and means differed
from country-to-country. The eight countries participating were:
Czechoslovakia, Denmark, Greece, Netherlands, Poland, Romania, Spain, and
Yugoslavia. The design of the study was quite deliberate, and the actual
protocol very thorough and precise. Levels of air pollution were recovered
at 19 sites throughout the countries. For the most part, particulate
pollution was measured using standard smoke, but no particulate measure-
ments were made in Spain, airborne dust aerosol (ADA) was recorded for
Czechoslovakia, and CoH in Greece. Unfortunately, site-specific calibra-
tion was not available, and the authors used the presumably noncomparable
smoke data from 11 sites to estimate levels of PH. Health effects were
quantified as symptom prevalence rates and peak expiratory flow rate (PEFR)
was the main ventilatory function index used.
3-19
-------
Several statistically significant regression equations for respiratory
symptom response prevalence rate on standard smoke were found despite the
PH quantification problems. A number of the regressions, however, were not
significant. For instance, the prevalence of a history of chronic bronchi-
tis, surprisingly, was unrelated to the level of particulates. Two
explanations of such lack of statistical significance seem plausible.
First, there is the obvious power problem encountered with so few degrees
of freedom. Secondly, the use of noncomparable estimates of levels of PH
doubtlessly inflated the error variance, thus further obscuring any health
effects.* The study can be faulted on the usual bases of lack of informa-
tion on socioeconomic status (SES), dietary habits, passive smoking, and
heating sources and fuels (i.e., indoor exposure levels). Finally, the
reliability of historical data obtained on children by questionnaire is
suspect [see, for example, Lunn $_t ,§_!. (21), Table IV, p. 225].
In addition to the primary work of Colley and Brasser (20), several
other studies were considered for inclusion in the chronic morbidity.esti-
mation phase of our work. They included Douglas and Waller (22), and Lunn,
Know el den. Roe and Handy side (21,23).
The Douglas and Waller research concentrated on a group of British
children born during the first week of March in 1946. Health interviews
were done with the mothers of the children at ages two and four. In
addition, medical examinations were given in the schools at ages 6, 7, 11,
and 15 years, at which time health histories were also taken. Eighty-one
percent of the children either lived at the same address or moved to an
area of similar level of air pollution in the first 11 years. The conclu-
sions of the study were succinctly stated by the authors:
The results are simple and consistent: upper respiratory
tract infections were not related to the amount of air pollu-
tion, but lower respiratory infections were so related. The
* In addition, if the population susceptible to certain respiratory
diseases self-selects to areas of low pollution, a nonsignificant
relationship between the levels of pollution and health may be observed.
3-20
-------
frequency and severity of lower respiratory tract infections
increased with the amount of air pollution. Boys and girls were
similarly affected, and no difference was found between children
in middle class and working class families. An association
between lower respiratory tract infection and air pollution was
found at each age examined and the results of the school
doctors' chest examinations at the age of 15 suggest that it
persists at least until school leaving age. [Douglas et al.
(22), p. 6]
It should be mentioned that the air pollution-lower respiratory tract
infection gradient was statistically significant at p < 0.05.
This study had two major faults. As with most other research of this
kind, possible competing influences on health were only sparsely measured.
More seriously, levels of particulate pollution were not measured directly.
Rather, regions of the country were classified as very low, low, medium, or
high pollution areas based on 1952 domestic coal consumption. This gross
"caricature" of the level of pollutants necessarily led us to exclude the
study. An appendix to the paper yielded some "soft" quantification of
pollution levels in terms of British Smoke, however.
Lunn (21,23) and his coworkers examined patterns of respiratory
illness in children in the town of Sheffield, U.K. Five- and 11-year old
children were studied, and the 5-year olds were re-examined at the age of
nine. Pollutants were monitored in four different areas of the city, with
mean BS levels ranging roughly from 97 to 301 jig/m . In Lunn's first
paper, the incidence of respiratory illness in the 5-year olds was
definitely area-related, with symptom incidence always least in the least
polluted area. History of persistent or frequent cough was the only
symptom related to social class, and was least common in the highest class.
The authors concluded that chronic upper respiratory infections were
influenced by area rather than by social class, number of children in the
house, or sharing of bedrooms. A similar finding was reported for lower
respiratory infections as well.
The three higher pollution areas were merged for analysis in the
second paper because the ameliorative effects of smoke control in Sheffield
3-21
-------
had greatly reduced the differences in levels of particulates among areas
in Sheffield. Upper— and lower—respiratory disease differences in the 11-
year olds were similar to those previously reported for the 5-year olds.
When the 5-year olds were examined at age nine, area differences were
found, but the symptom prevalence rate was lower than it had been four
years earlier. The differences in lower respiratory tract illness were not
significant (p < 0.05). The authors emphasized that the "... most
remarkable occurrence during this study has been the drop in air pollu-
tion." [Lunn ±t al. (23), p. 227.]
The results of Lunn's work seem relatively "clean" and some authors
may accept at face value the health effects claimed. The BS levels
reported must be considered as only crude estimates, however, given some
uncertainty regarding the use of site-specific calibrations in Sheffield
[Criteria Document (3), Table 14-7]. Secondarily, but also of consequence,
was the unreliability of retrospective health history data laid bare in the
1970 paper. When the parents were asked to report presence or absence of a
history of lower respiratory tract illness for their children at age five
and again at age nine, almost half who reported positively initially
responded negatively four years later. From this observation, the authors
suggested that data from retrospective histories of lower respiratory tract
illness may have limitations. We must concur that this observation is
true, at least for children.
Estimates of morbidity effects are estimated for Colley and Brasser,
Douglas and Waller, and the first Lunn study in Appendix 3B. Because of
the lack of calibration of BS and other weaknesses, results for these
studies are not used as a basis for benefit estimates, but as a cross-check
on the estimates from other studies. Even use as a cross-check, however,
is constrained by the quality of the studies.
Three other sets of studies given serious consideration were: Saric,
Fugas and Hrustic (24); Bouhuys, Beck and Schoenberg (26); and Ferris et
aJL. (26-28).
3-22
-------
Saric et al. (24) compared forced expiratory volume (FEV) and inci-
dence of acut'e respiratory disease in 78 second-graders living in a high
pollution area (Zagreb, Yugoslavia) with 70 others living in a clean air
area during the period of November 1977 to March 1978. Mean monitored
smoke concentration for the polluted area was about 70 |ig/m and mean
a
suspended particulate matter (SPM) was 200 (ig/m . SPM was measured by the
high volume sampler method, as is TSP, and thus should be approximately
equivalent to a TSP measurement. The mean smoke concentration for the
cleaner area was approximately 23 jig/m for the study period. Mean SPM was
not recorded, a problem which will be discussed subsequently.
The authors found several indications of pollution-related health
effects. Children from the cleaner area had significantly higher FEV (1
second) values than those from Zagreb. It was concluded that:
..., it is evident that the incidence of acute respiratory
diseases was higher in the families residing in the polluted
area. Pneumonia was recorded only in the families living in the
polluted area. Acute respiratory diseases accompanied by
elevated temperature, which required bed rest and physician
consultation, and diseases of the lower respiratory tract
occurred more frequently in the polluted area, particularly
among second graders, their mothers, brothers, and sisters.
Diseases of the upper respiratory tract occurred more
frequently in the control group of second graders, while in
mothers, brothers, sisters, grandfathers, and grandmothers a
higher incidence was recorded in the families from the polluted
area. [Saric art il- (24), p. 106]
A summary of the findings by Saric et al. concerning the incidence of acute
respiratory diseases is provided in Table 3-4. Entries in the table are
for all respiratory disease categories, by age group, and were taken from
Table 9 of Saric et al. As an example, for a susceptible group such as
"grandfathers and grandmothers", the disease incidence for persons in the
lower pollution area for all categories of respiratory illness is 19 per-
cent lower than that in the polluted area.
It should also be noted that analysis of a number of other possible
confounding factors (e.g., parental smoking, number of minor children in
3-23
-------
Table 3-4
SARIC ET AL. COMPARISON OF ACUTE RESPIRATORY DISEASE INCIDENCE
Affected Individuals
Second graders
Fathers
Mothers
Brothers and
Sisters
Grandfathers and
Grandmothers
Study Area
Polluted
Control
Polluted
Control
Polluted
Control
Polluted
Control
Polluted
Control
Disease Incidence*
144.9
117.2
56.3
47.7
84.0
73.5
151.7
97.3
63.4
51.6
* Number of incidents between November 1977 and April 1978, as a percent of
the total number of persons in particular groups. Since one person may
have more than one incident, the percentages may exceed 100 percent.
the home, household density, heating system) revealed no significant
differences between families in the two areas.
Use of the Saric et al. results in a benefits analysis has many of the
problems encountered previously. For example, SPM was not measured in the
cleaner area, and there was no site-specific calibration for BS. There-
fore, a precise estimate of the difference in SPM associated with the
difference between disease incidence in the clean and polluted areas cannot
be made. The difference in annual SPM, however, cannot exceed 200 jig/m ,
the level in the polluted area, since the clean area by definition has a
lower level of SPM.
Secondly, the Saric et al. analysis does not provide a basis for
separately isolating the acute disease effects of peak exposures from
3-24
-------
chronic exposures, or the effects of BS/SPH from the effects of S02, which
was also monitored and was higher -in the area with higher BS/SPH. One
cannot assess the relative strengths of the association of PM and SOj with
the observed effects. Hence, if we use the study to estimate PM respira-
tory effects, the results must be viewed as upper-bound estimates.
Bouhuys (25) and his coworkers assessed respiratory health for resi-
dents of two Connecticut communities; one rural (Lebanon), the other urban
industrialized (Ansonia). TSP means were approximately 40 jig/m in Lebanon
and 63 in Ansonia during the period of study (1973). Previous concentra-
tions had been considerably higher in Ansonia, ranging from 88 to 152 (ig/m
in the years 1966-1972. Thus, the level of chronic exposure to particulate
matter in Ansonia actually varies in the range of about 60 to 150 {ig/m
TSP, depending on the time lag that one is willing to accept.
Health data were obtained using a questionnaire. Because of a low
number of potential black re spenders in Lebanon, health effects analyses
were limited to white residents. The morbidity findings of Bouhuys et al.
were mixed. Incidence of chronic bronchitis did not differ between the two
groups, while that for history of bronchial asthma was actually higher in
the rural sample. Feeling that the incidence of chronic bronchitis may be
an insensitive index of urban-rural differences due to its scarcity among
nonsmokers, the authors decided to examine concurrent and/or component
symptoms. Prevalence of cough, phlegm, and dyspnoea +1 were significantly
higher for nonsmokers in Ansonia than in Lebanon. The findings led Bouhuys
S_t ajl. to reject the association of chronic bronchitis with particulate
pollution (at least at these levels), but to admit as well the tie with
some lesser degrees of component symptoms in nonsmoking adults.
It is very difficult to value the effects of cough, phlegm and
dyspnoea since there is little or no information on direct medical expendi-
tures or work-loss days that might be associated with these symptoms.
Because of this constraint and the lack of local population data on smoking
habits, we have not developed benefit estimates based on the Bouhuys et al.
study.
3-25
-------
The final set of studies chosen for inclusion because of the presence
of quanti'tative relationships between levels of particulate matter and
health are those of Ferris et al. (26-28). In the original paper, Ferris
and Anderson (27) attempted a cross-sectional study of the interaction of
community air pollution with chronic respiratory disease in Berlin, New
Hampshire during the winter of 1961. For this study and subsequent ones,
chronic respiratory disease was defined as the presence of chronic
bronchitis, bronchial asthma, or irreversible obstructive lung disease.
The area definitions for pollution levels were crude (given only as "less",
"mixed", and "considerable"), and no TSP values were reported. After
adjusting symptom prevalence rates for the effects of smoking, no signifi-
cant differences were found among the areas for any category of respiratory
disease for either sex. The research was wel1—designed and executed
thoroughly, but the authors' admitted the shortcomings of their residence
variables as a valid measure of the effect of air pollution:
It cannot compensate for the racial, social, and occupational
differences between (SIC) areas. Migration probably occurs from
one area to another. [Ferris and Anderson (27), p. 174]
This study takes on importance only in the subsequent work of Ferris et aJL.
Essentially the same study was conducted once again in 1967. The
average of annual levels of TSP at the three study monitors declined over
the 6-year period of 1961-1967 by about 50 jig/m3, from 180 to 130 ng/m3.
(These are 9- to 22-month averages of daily values). Sulfation rates and
dustfall were also observed to decline during that period. In the follow-
up study, no area differences were considered, only the longitudinal
effects of the reduction in TSP. The principal finding was that in
comparison to 1961, the "Prevalence of chronic nonspecific respiratory
disease was less in 1967 after the effects of aging and changes in
cigarette smoking habits were taken into account" [Ferris et. &1. (28), p.
110].
The 1967 sample was followed up in 1973. By this time, the level of
3 O
TSP had fallen by another 50 jig/m to 80 (ig/m , virtually at the primary
3-26
-------
standard. Sulfation rose, however, relatively sharply during this period.
No differences in symptom prevalences were found in comparison to the 1967
data. As Ferris e_t .§_!. [(26), p. 484] concluded, "... either the changes
in the levels of air pollution in Berlin, New Hampshire from 1967 to 1973
are not associated with a beneficial effect on health, or our methods of
assessing an effect are not sufficiently sensitive at these levels".
The results reported by Ferris et al. (26,28) indicate that a health
effect is observable when TSP is reduced from 180 jig/nr to 130 (ig/m3.
Either no beneficial effect .occurs in the 130 to 80 |ig/m range, or the
effect is obscured by the increase in sulfation or other occupational,
personal, etc., influences. Generalization of even the 180 to 130 effect
is threatened slightly by the observations that air pollution in Berlin,
New Hampshire was dominated by the emissions of a wood pulp mill, a situa-
tion atypical of most of the U.S., and that sulfation rates as well as TSP
also declined during that period.
Using the Ferris et al. work, a health effect can be tied to appr'o-
priate aerometry data. No concentration-response functions were reported,
however. As presented by Ferris et al.. results of primary interest to us
take the forms:
(a) Symptom prevalence rates by age by sex for 1961 and 1967.
(b) Age—adjusted morbidity ratios and rates of selected
respiratory symptoms by sex by cigarette smoking category
for 1961 and 1967.
(c) Age-standardized rates of prevalence for chronic non-
specific respiratory disease by sex by cigarette smoking
category for 1961 and 1967.
(d) Age-standardized ratios of all chronic nonspecific respira-
tory disease by cigarette smoking category in 1961 and 1967
by sex.
One confounding factor in assessing changes in simple symptom preva-
lence rates by age group for 1961 and 1967 is that all individuals in the
sample were six years older in 1967. Thus, results such as (a) above are
3-27
-------
of less value. The age-adjusted ratios and rates of selected respiratory
*
symptoms and all chronic nonspecific respiratory disease are of primary
importance, but we do not have smoking habit data by counties, and overall
rates are not provided. For a similar reason, results of form (d) are of
dubious value. The only acceptable strategy seems to be to estimate
overall symptom prevalence rates for the (b) and (c) results.
Unfortunately, the small numbers of observations underlying the symptom-
specific rates render them quite unstable. The only smoking categories
involving good-sized samples are non-smoking women and men who are ex-
smokers. That these categories show by far the largest decreases in
standardized morbidity ratios and in prevalence rates from 1961 to 1967
buttresses the conclusion that the health effects are substantial, but does
not contribute to the estimation of overall figures.
The only course left to us has been to estimate overall (weighted
average across smoking categories) changes in symptom prevalence rates for
all chronic nonspecific respiratory disease in 1961 and 1967. The data for
such estimates come chiefly from Figures 1 and 2 in Ferris et al. (28).
Recalling the numbers of persons in each smoking category and weighting the
changes in age-adjusted rates by those figures, we estimate that the
average absolute decrease in the age-standardized symptom prevalence rate
was about 13.4 percent for adult men and 9.6 percent for adult women.
These declines were associated with the decline in TSP from 180 to 130
fig/m . One problem with using these results for benefit estimates is that
the distribution across smoking categories and age groups of the population
in the Ferris et al. study and in the counties in our analysis may differ.
Uncomfortable as we have been with the approximations and assumptions
needed to apply the results of Saric et al. and Ferris et al., we feel that
we have had some success in developing a basis for qualified benefit
estimates for chronic morbidity effects for TSP.
3-28
-------
Hie various studies selected as a basis for benefit calculations are
summarized in Table 3—5, together with, a listing of additional studies
which provide corroborative, if not directly usable, evidence. As can be
seen, the strict selection criteria greatly limited the number of studies
that could be used. In fact, at most one study per category was selected
and in two categories no usable studies could be found.
In the acute exposure mortality category the basic study to be used is
the Mazumdar et al. (8) analysis of winter daily mortality rates in London.
In the acute exposure morbidity category, the study selected is Samet et
al. (18) concerning emergency hospital admissions for respiratory disease
in Steubenville, Ohio. For chronic exposure, the Saric .e_t .§_!. (24) study
of acute respiratory disease in Yugoslavia and the Ferris et al. (26,28)
studies of chronic respiratory disease in Berlin, New Hampshire will be
used.
APPROACH TO BKNHF1T ESTIMATION
The previous subsection identified the studies on which our benefit
estimates will be based. In the next two subsections, concentration-
response functions showing the effects of changes in PH on mortality risk
and morbidity are developed from the studies selected. These
concentration—response functions are then used to estimate the health
effects of alternative PH standards. The health effects are valued using
the approaches discussed in the Appendix to Volume II.
PM Standard*
The six PH standards that are considered in our analysis are shown in
Table 3-6. Column 2 expresses the standard in terms of the annual arith-
metic average, while column 3 expresses it in terms of the 24-hour average
reading that is expected to occur once a year. When the standard is stated
in terms of both the annual average and 24—hour expected value, the more
3-29
-------
o
u
a
I
M
01
O
a
o
h
U
u
a
e
Oi
M
O
a
X
o 44
••4 -V4
a -o
o -<
W A
<3 S
a
X
••4
» -o
3 A
0 h
X
2
**
O
X
O •**
S3
al
X
•M
S3
»
X
44
•
U
S
**!
4
a
•
U
U
O
Bi
3
Q
••4
09
1
j
1
3
3
44
to
£
71
3
a
M
3
•O
^
O
V
0
01
X
•0
3
CO
M
«rf
U M
<«4
a
o
o
1
X
14
O
•
e>
o
•
a
O -4
< •«
1
1
1
1
5
o
•*4
•o
X
X
0
a
o
Bi 0
• -2
X
•
8
e
H
M
a
*5
IM
Ed
a
o
09
•o
s x
« hi
o
M a*
o o
• **
X o
• a
11
j> «
0.
B a
0 O
••4 (4
•4 «
a •»
Ck •
o
Bl •
a
M 0
4< M
H >
I
,
g
«4
j*
a
a.
a
BI
•4
•
a
a
a
a.
I
•W
O
H
•fl
o
•o
a —
s <•
o s
hi O
o U
a.
12
^j
3
e
a
4
-------
Table 3-6
ALTERNATIVE PAETICDLATE MATTER STANDARDS
(in fig/m3)
Standard
1 PM10
2 PM10
3 PM10
4 PM10
5 TSP
6 TSP
Annual
Mean*
70
55
55
55
75
—
24-Hour
Value**
250
—
250
150
260
150
Implementation
Date
1989
1989
1989
1989
1987
1987
* Annual arithmetic mean for all standards except for No. 5. Annual
geometric mean for standard 5.
** 24-hour reading that is expected to occur once a year for PM standards
and 24-hour second high for TSP standards.
stringently averaging time is used, as discussed in Section 9. Column 4 of
Table 3-6 lists the attainment dates for each standard.
The benefits achieved under each standard will be calculated at the
county level. The change in county pollution under each standard is given
by the difference between the expected level of PM with implementation of
the standard and the baseline level without implementation. The baseline
assumes that some controls are in place, as discussed in Section 1. The
studies used to estimate benefits examine'the relationship between changes
in BS or TSP and health. Therefore, the change in PM under each standard
will be converted to a change in TSP or BS, as 'discussed previously.
3-31
-------
of
Each study analyzes the relationship between one particular measure of
exposure and health. To estimate accurately the health effects of changes
in PM, the measure of exposure for each county in our analysis should be
comparable to that used in the study. Two alternative measures can be
derived from our data: 1) the PH level at the county design value monitor,
and 2) the average of the PM levels for all of the monitors in the county.
Table 3-7 shows the type of monitor used in each study and the exposure
measure selected as more appropriate for our benefit calculations.
The chemical composition and particle size distribution of PH in the
counties in our analysis also may differ from- those of the PH in the health
studies. There are insufficient data, however, to analyze or adjust for
variations in these factors.
Concentration-response functions showing the effect of a change in PH
on health are developed below. To estimate the effects of implementing a
standard, the resulting changes in PH (using the appropriate measure and
conversions) will be estimated. This change will be substituted into the
concentration-response functions to find the effects in each county of
implementing the standard. Effects will only be estimated for the range of
PH levels considered in the original study. Specific restrictions on the
range for each study are discussed below. The general issue of effects
levels is discussed in Section 10.
MOKFALITT KISK EFFECTS
Effects of Acute
Mazumdar et al. (8) relate winter mortality in London to the level of
British smoke (BS). From their results, alternative concentration-response
functions showing the relationship of changes in annual mortality to
3-32
-------
CO
I
M
£
B)
I
03
03
r-
,0
H
09
VI
U 0 >»
0 4-» -U
M O fl 3
fl •* 00
•* CA • S W
W. QQ
•0 .H > -M fl
0 08 < ^H ^H
Vl *» 0
fl -»4 0 fl
(A (H fH -H >,
OBJ 0 0* 4J
0 fl > VI fl
Jg O O 0
ea a -M o
ao -H U
•H fl
M O
0 36
a
fj
*«4
^
•O -0
o fl
M 4-»
B 09
.••» 1-4
M 01
^^ fl
Vl "H
0 60
-M -H
a o
o
as
at
0
VI
«^
o
f-t
««f4
a
at
(4
60
O
0
^
•o
0
4^
03
H
1
at
*t
fl
0
<• fl
O -rt
Vl
Pi «
0 VI
M O
•u
^^ *v4
fl
CM o
o a >.
• 0 fl
60 > fl
> ••« 0
< -w o
fl
o
•o
fl
o
^J
0
VI. -U at
at w
0 0 fl
fl «<3 ••*
60
w *^ w
o ^s o
^•J ^J
fl 0 fl
0 fl 0
x ~ a
M
0
«O «»
O
•0 fl -H
0 -H f-l
VI t-l >
0 « fl
(••WO
f4 ^3
M Pi fl
O M O
VI O *»
•< A 03
H
ol
4^1
01
^
O
a
01
03
X)
1
•W
fl
0
•0
fl *
081 M
VI
VI O
o •«•»
O -»H
•a fl
B 0
.fl VI
•M O
0 0
e
^4
09
•«
0
O ta
W3 O
O *>
1Q **4
ffl U
so
•»* o
l-H
01
•M
0
a
VI
09
03
XI
fl
1-1
M
M
0
•M
•H
fl
O
a
(V)
CM
0 fl
• 1-1
60 Vl
> 0
§
«
fl
-H
1-4
O
ea
•••f
B)
4J
0
45
VI
Vl
0
Cs.
fl
0
•r4
ft
a
A
•a
0
4J
at
a
•H
H
O
M
Pi
Pi
O
•§
0|
M
O
O >H
•»4 0
VI >
«8 0
03 iH
3-33
-------
changes in the daily level of BS will be derived. To apply these functions
to our analysis, a method is developed for estimating changes in daily BS
levels from our PM data. Once the changes in daily BS have been estimated,
the concentration-response function can be used to determine the effect of
a pollution standard on the mortality rate in each of the counties in our
analysis. The change in the mortality rate will be valued using
willingness-to-pay estimates developed in the Appendix to Volume II.
In developing a range of benefit estimates from the study of Hazumdar
et al.. five major factors must be considered. First, the estimates will
depend on the functional form used. Mazumdar et al. found both linear and
quadratic models to be compatible with their data. For the PM levels in
the counties in our analysis, use of the quadratic form results in lower
benefit estimates than use of the linear function.
Second, the coefficients estimated for each functional form depend on
the data set used. Hazumdar et al. estimate the relationship between
mortality and BS for episodic periods (days with BS levels exceeding 500
fig/m and the seven neighboring days on each side), non-episodic periods,
and episodic and non-episodic periods pooled. Aproximately one-fourth of
the sample days were episodic. No episodic periods occurred after the
winter of 1964/65. Data on the current maximum second-high levels of PM in
the counties in our analysis indicate that a very low percentage of the
days would be classified as episodic. The coefficients for the non-
episodic data are 50 to 520 percent higher than the coefficients for the
pooled data, as shown in Table 3-8.*
Third, the data yield a range of values for each coefficient at the 95
percent confidence level. . Table 3-8 shows the standard errors of the
coefficients which measure one component of the uncertainty associated with
the study results.
* There are several alternative explanations for the relative magnitudes of
the two sets of coefficients. Without further information, no final
judgment on the appropriateness of each set can be made.
3-34
-------
Table 3-8
COEFFICIENTS (PERCENTS) FROM MAZUMDAR ET AL. IN MG/M3
(Standard Errors)
Quadratic Form
Linear Form
Pooled Data
Non-Episodic Data
Episodic Data
9.20 (0.96)
48.13 (7.49)
8.58 (1.00)
19.14 (1.44)
27.54 (3.26)
17.16 (1.63)
Fourth, the estimated benefits of mortality reduction depend on the
method used to value reductions in risk of death. As discussed in the
Appendix to Volume II, ire have derived a value of $0.36 to $2.80 for a unit
reduction of 1 x 10~ in annual mortality risk.
Fifth, it is uncertain at what daily BS level mortality effects occur.
The daily BS levels included in the data of Mazumdar .e_t .aJL- range down to
4
approximately 10 fig/m . It could be argued, however, that mortality
effects may not have occurred over the full range. Based on analyses of
the data, the Criteria Document concludes that "Both analyses (linear and
quadratic) indicate that small increases in mortality were associated with
London PH levels in the range of 150-500 jig/m3 BS. ... The findings of
mortality being significantly associated with the lower range of BS values
(150-500 ug/m3) were further confirmed by analyses of mortality rates
a
occurring only on days when BS levels did not exceed 500 |ig/m . [(3), pp.
14-21]
In our benefit calculations, we cannot limit the daily levels of BS
for which benefits are calculated because we do not have data on these
levels. As described below, our calculations will be based on changes in
annual BS and will be calculated for the entire range of BS levels above
the background levels in each county (see Section 9). Because of the small
3-35
-------
size of the coefficient, however, inclusion of mortality effects for
changes in low levels of daily BS will have only a slight impact on the
benefit estimates derived from the quadratic model.* Calculation of
benefits over the full range of BS levels is consistent with the data set
and results of both the linear and quadratic models of Hazumdar et al., the
most comprehensive London mortality study, which identify no lower bound
for effects. The issue of the applicable concentration range is discussed
in detail in Section 10.
To determine the potential range of benefit estimates, we must calcu-
late the lowest and highest values consistent with the study of Hazumdar et
al. Therefore, all the conservative assumptions will be matched for a
minimum estimate. Conversely, all the assumptions that result in higher
estimates will be matched for a maximum estimate. Thus, our minimum
estimate will be based on:
1) The quadratic model of Mazumdar et al.
2) The results for the pooled data.
3) The lower bound of the 95 percent confidence interval
around the coefficient.
4) A value of $0.36 for a unit reduction of 1 x 10 in annual
mortality risk.
The maximum estimate will be based on:
1) The linear model of Hazumdar et al.
2} The results for the non-episodic data.
3) The upper bound of the 95 percent confidence interval
around the coefficient.
4) A value of $2.80 for a unit reduction of 1 x 10 in annual
mortality rise.
* The magnitude of the impact on benefits, however, depends on the size of
the population affected.
3-36
-------
Since both, the linear and quadratic models are compatible with the
data, it cannot be determined which, model is more appropriate for use in a
point estimate. The coefficient for each model and the point estimate of
the value of a unit risk reduction, however, provide the best available
estimates of the effects of PM on mortality and the value of these effects.
Therefore, for each model, an estimate based on the following parameters
will be derived.
1) The results for the pooled data for the quadratic model and
results for the non-episodic data for the linear model.
2) The coefficient for the model.
A value of $1.58
mortality risk.
3) A value of $1.58 for a unit reduction of 1 x 10 in annual
The point estimate of benefits will be given by the geometric mean of these
estimates. Use of the coefficient for pooled data in the estimate for the
quadratic model will result in a lower point estimate than if the coeffi-
cient for non-episodic data, which most closely corresponds to our data,
were used.
Figure 3-1 summarizes the approach used in each estimate. The method
for deriving the minimum and maximum benefit estimates is described below.
Minima Estimate —
Using the lower bound of the 95 percent confidence interval around the
coefficient, the following equation is derived from the quadratic model for
pooled data for the 14 winters analyzed by Hazumdar et al.
EM . - (7.3184 x 10~8)(Bs£.)(M~) (3.1)
where ^wd ~ excess mortality on winter day d (deviation from 15-day
moving average).
MW = mean daily winter mortality.
3-37
-------
at d
a d
ovo "*
i d
0 O 0
P iH -H
at M 0
> e
i-l "O
o
•M
d
o
o
•H
-H
H U M «w
O O 2
01
8
•H
•W
W
BE]
§
a
rt
••H
s
00
e
^
•49-
•M
d
o
-] OS M U
o
•a
o
w
ft
1 fltf
d -M
O at
z o
M
CD
o
d
•H
O
•M
at
8
•W
w
H)
|
^4
M
at
S
d
o
u
ft
•o
o
M
>>
H
O
as
O
w
-------
a
BS d = BS level on winter day d in |ig/m .
The deviation from the moving average and multiplication by the winter mean
are used to control for day-of-week trends and year-to-year variations in
mortality.
Equation (3.1) indicates that the change in daily winter mortality
(change in excess daily winter mortality) for a change in daily BS squared
is:
AMwd - (7.3184 x 10~8)(ABsJd)(M^) (3.2)
where ^wd = change in mortality on winter day d.
ABS^d = change in square of BS on winter day d in |ig/m .
Since the mean daily winter mortality will vary with the population
size, this equation can be used to estimate changes in mortality in areas
with different population sizes.
Equation (3.2) can be used to calculate the change in daily mortality
for a change in daily BS squared for each of the 120 days in the period
November-February, the months covered by the data of Hazumdar et al. For a
set of changes in daily BS over the four-month period, the total change in
mortality can be estimated by summing the daily changes:
120 -8 2 -
AM^ = 2 (7.3184 x 10 8)(ABSjd)(Mw) (3.3)
where AM^ = change in mortality for the winter period.
For our benefit calculations, the change in mortality for a set of
changes in BS over the entire year, not just the winter period, must be
determined. Assuming that the proportional relationship identified by
3-39
-------
Mazumdar ert .§_!. holds for other four-month periods, the change in annual
mortality for a set of changes in daily BS levels is approximated by:
120 -8 2 —
AMA = I (7.3184 x 10 8)(ABSjd>(Mw)
d=l
122 _
I (7.3184 x 10~8) (ABSf.) (M )
d=l
123 » o —
+ 2 (7.3184 x 10~8)(ABS|d)(Mf) (3.4)
d=l
where AM* = change in annual mortality.
= change in BS squared on day d of the period March-June.
This period has 122 days.
M = mean daily mortality for the period March- June.
3 •
j - change in BS squared on day d of the period July-
October. This period has 123 days.
•= mean daily mortality for the period July-October.
This equation, which allows for variation in mortality across seasons, is
equivalent to:
AMA - (7.3184 x 10~8)(120)(ABS2rd)(Mw")
+ (7.3184 x 10~8)(122)(ABS^d)(M^)
+ (7.3184 x 10"8)(123)(ABs|d)(Mf") (3.5)
where ^^wd = »ve*age change in daily BS squared for the period
November-February.
= average change in daily BS squared for the period March-
June.
3-40
-------
ABSi, = average change in daily BS squared for the period July-
October.
Unfortunately, we do not have data on the average levels of daily BS
squared for each four—month period. A lover—bound estimate of the change
in annual mortality can be made, however. In the United States, the mean
daily mortality rate for the period July-October is lower than that for
either of the two other periods. Therefore, a lower-bound estimate of the
change in annual mortality is given by:
AMA - (7.3184 x 10 8)(Mf)[(120)(ABSwd)
j) + (123)(ABSfd)] (3.6)
This equation is equivalent to:
AMA - (7.3184 x 10~8)(M^)(365)(ABSJ) (3.7)
where ABSd = average annual change in daily BS squared (change in
average annual BS squared).
r365 ASS? 365 BS
5- « = A S
1=1 365 dail 365
Mean daily mortality for the period July through October is about 0.26
percent of annual mortality in the United States.
Substituting in this value. Equation (3.7) is equivalent to:
AMA = (7.3184 x 10~8)[(0.0026)(MA)](365)(ABSl) (3.8)
AMA - (6.945 x 10~8)(MA)(ABS^) (3.9)
3-41
-------
For each, county in our analysis, Equation (3.9) can be used as a
minimum estimate of the change in mortality for a change in average daily
BS squared. For each 1 (ig/m change in this value, the percentage change
in annual mortality is greater than 6.945 x 10 . The nonlinear transfor-
mation function used to approximate the change in average BS squared under
each standard for the counties in our analysis is presented in Appendix 3A.
As an example of the values implied by our calculations, assume a
community with annual mortality of 1,000 and population of 200,000. Assume
2 2
a decrease in the average level of British smoke squared from 200 to 150
(ig/m . Our estimate of the change in annual mortality in this community
is:
AMA - (6.945 x 10~8)(1,000)(2002 - 1502) - 1.22 (3.10)
Each individual will experience an average reduction of (1.22/200,000)
= 6.1 x 10 in his or her mortality rate. The lower bound of an indivi-
dual's willingness to pay for this risk reduction is (6.1)($0.36) = $2.20.
The lower bound of the population's total willingness to pay is given by:*
($2.20)(200,000) - $440,000 (3.11)
where 200,000 is the population experiencing risk change.
Maxim* Estimate —
Using the upper bound of the 95 percent confidence interval around the
coefficient, the following equation is derived from the linear model for
non-episodic data.
EMwd = (0.0003393 )(BSwd)(i£) (3.12)
* This calculation is not based on the assumption that each individual will
experience an equal risk reduction. The same result will be yielded by
determining the willingness to pay of each individual and summing for any
distribution of small risk reductions around the mean.
3-42
-------
Equation (3.12) indicates that the change in mortality for a change in BS
on a winter day is given by:
AMwd = (0. 0003393 )(ABSwd)(M^) (3.13)
where ABs = change in BS on winter day d.
Following the procedure outlined for the minimum estimate, the change
in annual mortality for a set of changes in daily BS throughout the year is
given by:
AMA - (0. 0003393 )( 120) (M^)(ABSwd)
+ ( 0.00033 93 )( 122 )(M^)(AlSsd)
+ (0,0003393 ) (123) (M^)(Alsfd) (3.14)
The mean daily mortality rate for the period November through February
is higher than that for either of the other two periods. Therefore, an
upper-bound estimate of the change in annual mortality is given by:
AMA =* (0. 0003393 )(M)[( 120) (ABSwd)
+• (122)(ABS$d) + (123)(ABSfd)] (3.15)
This equation is equivalent to:
AMA = (0. 0003393) (M) (365) (ABSd) (3.16)
where AB^d = average annual change in daily BS (change in average
annual BS)
'365 ABSd 365 BS
= ^365
3-43
-------
Mean daily mortality for the period November through February is about
0.3 percent of annual mortality. Therefore, Equation (3.16) is equivalent
to:
AMA = (0.0003393)[(0.003)(MA)](365)(ABSd) (3.17)
AMA = (0.0003715)(MA)(ABSd) (3.18)
For each county in our analysis, Equation (3.18) can be used as a
maximum estimate of the change in mortality for a change in annual BS. For
each 1 fig/m change in this value, the percentage change in annual
mortality yielded by the model is less than 0.03715 under the assumption of
constant proportional effect for each 4-month period. The nonlinear
transformation function used to approximate annual BS for the counties in
our analysis is presented in Appendix 3A.
As an example of the values implied by our calculations, assume a
community with annual mortality of 1,000 and population of 200,000. For a
a
decrease in the average level of British smoke from 200 to 150 |ig/m , our
estimate of the change in annual mortality in this community is:
AMA - (0.0003725)(1,000)(200 - 150) - 18.6 (3.19)
Each individual will experience an average reduction of (18.6/200,000)
= 93 x 10~° in his or her mortality rate. An individual's willingness to
pay for this risk reduction is estimated to be no more than an upper bound
of (93) ($280) » $260.40. The population's total willingness to pay is
given by:
($260.40)(200.000) - $52,080,000 (3.20)
The potential sources of bias in both our minimum and maximum
estimates derived from Mazumdar et al. are summarized in Table 3-9.
3-44
-------
Table 3-9
BIASES IN ESTIMATES BASED ON MAZTJMDAR ET AL.
Sources of Bias
Direction of Bias*
Generalization of results to counties and
period of our analysis.
Differences in sealing of buildings in
London and the United States.
Use of Commins and Waller pre-1963 data and
nonlinear transformation function to trans-
form our TSP data to BS.
Assumption of a constant proportional effect
on mortality for each four-month period.
Method of accounting for effects of
temperature and humidity
Method of estimating effects on annual
mortality from annual BS data.
Value of risk reduction used
Exclusion of pollutants correlated
with BS and influencing mortality
- (min. est.)
+ (max. est.)
* A plus sign indicates an upward bias, a negative sign indicates a
downward bias, and a question mark indicates that the direction of bias
is uncertain.
MORBIDITY KPVUCTS
Introduction
Concentration-Response Functional Font —
The procedures used in benefit calculations for reduced morbidity
effects are described in this section. The calculations are accomplished
3-45
-------
in a way which is similar to the approach used previously for mortality.
That is, results from the selected studies are used to predict changes in
annual morbidity on a county-by-county basis, corresponding to the changes
in PH under each alternative standard. The economic value of these changes
is then estimated.
Additional problems arise, however, with the morbidity calculations
because the studies selected do not fully identify the shapes of the
concentration-response fuctions. A variety of concentration-response
functions would be compatible with the study results, including linear and
exponential forms. In view of this ambiguity, we estimate morbidity
effects using two alternative functional forms, each fit through the
observed data points. This approach will illustrate how benefit magnitudes
are affected by the particular choice of functional forms.
For the first functional form, we assume that the percentage change in
morbidity is a linear function of the change in PH:
= (pjMAPM) (3.21)
where AMB - change in morbidity.
MB = initial level of morbidity.
APM • change in level of PM.
P! « coefficient relating PM to morbidity.
This functional form allows the effect of a change in PM to vary with base
morbidity. The resulting relationship between disease incidence and PM is
shown below.
3-46
-------
% Change in Morbidity (MB)
Change in PM
For the second form, we assume that the absolute change in morbidity
is a linear function of the change in PM:
MB - (P2)(APM)*
(3.22)
This functional form provides a linear relationship between disease inci-
dence and PM, as shown below. Note that to use this functional form for
benefit estimation, it is necessary to scale the morbidity change to the
size of the affected population. The method for doing this will be
discussed subsequently in the context of each study. For the other
functional form scaling is not required since effects are stated in per-
centage terms.
Change in Morbidity (MB)
Change in PM
When applied to the data in our analysis, the first functional form
yields lower estimates of total benefits than the second. Therefore, the
* This function is equivalent to: AMB/MB = (02/MB)(ATSP)
3-47
-------
first form will be used to derive a minimum estimate and the second to
derive a maximum estimate of total benefits.
Estimation of Base Morbidity —
Another problem encountered in applying the morbidity studies to our
analysis is lack of data on base morbidity in each county. The first
functional form estimates the percentage change in morbidity for a change
in PM. Therefore, for this form, data on the initial levels of morbidity
for each county are required to estimate health effects.
Data on base morbidity are often only available at the national level.
Foe example, data on the base number of respiratory disease emergency
admissions and incidents could only be obtained for the nation. For
benefit calculations at the county level, the number of admissions or
disease incidents in each county are assumed to be proportional to popula-
tion.
This method of determining base morbidity in each county may bias the
estimates of health effects yielded by the first functional form. The
morbidity studies suggest a positive relationship between morbidity and
pollution. Therefore, other factors being equal, the base morbidity per
capita will be higher in an area with high pollution (relative to the
national average in the data year, not level in study area) than in an area
with low pollution. Setting morbidity proportional to population will
result in an underestimate of base morbidity in areas with high pollution
and an overestimate of base morbidity in areas with low pollution.*
The first functional form estimates the percentage change in morbidity
per unit change in PM. If the base level of morbidity is underestimated
for high pollution areas, the change in morbidity per unit change in PM
* The self-selection of ill people to areas with low pollution may occur,
introducing a counteracting factor. Then, the net bias in our estimates
of base morbidity is not clear.
3-48
-------
Till be underestimated for these areas. Conversely, the change in
•
morbidity will be overestimated for low pollution areas. The net effect of
these biases on total benefits cannot be determined.
Valuation of Morbidity Effects —
As discussed in the Appendix to Volume II, a reduction in morbidity
will result in reductions in: 1) direct medical expenditures (DME), 2) the
number of work-loss days (WLD), and 3) the number of restricted activity
days (RAD). These RAD are net of WLD. Therefore, for each study, we
derive estimates of the effects of changes in TSP on DHE, the number of
WLD, and the number of RAD.
To determine the effect of changes in morbidity (as measured by the
number of disease incidents or admissions) on DME, WLD, and RAD, some
assumption concerning their relationship is required. For all calcula-
tions, it is assumed that the ratio of DME, WLD, and RAD to the number of
admissions or disease incidents is a constant. Then, the percentage change
in DME, WLD, and RAD for a change in PM is equal to the percentage reduc-
tion in the relevant morbidity measure. For benefit calculations, the
value of each WLD is set equal to the average county wage. The value of
each RAD is set equal to one-half the average daily wage for the counties
in our analysis.
Depending on the study's health endpoint, the percentage change in
•
morbidity will be applied to base levels of acute or chronic respiratory
disease DME, WLD, and RAD. The National Center for Health Statistics
(38,39), the source of most of our data, defines acute conditions as condi-
tions that have lasted less than 3 months. A condition is defined as
chronic if it has lasted more than 3 months or is a disease that is always
classified as chronic such as asthma. Because an illness is classified as
acute or chronic, there should be no overlap between the estimated benefits
of reduced acute and chronic DME, WLD, and RAD. The estimates, however,
may be slightly biased if the division between acute and chronic morbidity
3-49
-------
in the underlying study from which, the percentage change in derived differs
from that employed in collecting the base level data.
Smnary —
For ea'ch study, we will estimate the percentage change in morbidity
for a unit change in TSP using the first and second functional form.*
Then, we will use these percentages to make alternative estimates of the
effects of a change in TSP on annual DME and the value of WLD and RAD for
each population group considered.
As discussed in the Appendix to Volume II, the total benefit of a
reduction in morbidity will be measured by the sum of the value of reduc-
tions in DME, WLD, and RAD. Thus, for each study, the minimum estimate of
total benefits of alternative standards will be given by the sum of:
i) Reduction in DME (estimated using the first functional
form).
ii) Value of reduction in WLD (estimated using the first
functional form).
iii) Value of reduction in RAD (estimated using the first
functional form).
The maximum estimate of total benefits of alternative standards will be
given by the sum of:
i) Reduction in DHE (estimated using the second functional
form).
ii) Value of reduction in WLD (estimated using the second
functional form).
iii) Value of reduction in RAD (estimated using the second
functional form).
* For purposes of comparison, a percentage reduction is estimated for the
second functional form using the assumption of proportional morbidity.
The per capita base morbidity figure used cancels out in the benefit
calculations for this form, however.
3-50
-------
The geometric mean of the minimum and maximum estimates will be used as the
point estimate.
If there is a choice of assumptions in addition to the appropriate
functional form, the more conservative assumption (assumption yielding
lower benefit estimates) will be used for the minimum estimate of total
benefits while the less conservative assumption will be used for the
maximum estimate. This procedure will yield the largest range of estimates
consistent with the data. For each study, we will discuss any specific
assumptions that are required.
Acute Morbidity Effect* of Acute Exposure
Recall that our estimation of the acute morbidity effects of acute
exposure will be based on the work of Same t e_t .aJL. (18). They relate the
daily number of emergency admissions in a Steubenville, Ohio hospital to
the daily level of TSP. Samet et al. look at deviations from means for the
appropriate day of the week, season, and year. In a linear model with
unlagged TSP and maximum temperature as the independent variables, the
following relationship is observed:
EAd = (0.007)(TSPd) - (0.08) (TEMP) (3.23)
where EAd - number of respiratory disease emergency admissions on day
d.
TSPd - level of TSP |ig/m3 on day d.
TEMP « maximum temperature for the day.
Our benefit calculations are based on this equation. The daily TSP levels
in the study by Samet et al. range from 14 to 696 |ig/m . Since this range
encompasses most of the range of daily levels encountered in our analysis,
we calculate benefits at all levels of TSP above the background level in
each county (see Section 9).
3-51
-------
Application of the First Functional Fora —•
The mean number of daily respiratory admissions in Steubenville is
24.5. Therefore, the change in the number of admissions for each unit
reduction in TSP is 0.02857 percent (100)(0.007/24.5) of the mean.
AEAd = (0.0002857)(ATSPd)(EAa) (3.24)
where AEA. = change in number of respiratory disease emergency admis-
sions on day d.
ATSPd = reduction in TSP level on day d.
EA& - annual arithmetic mean of daily number of respiratory
disease emergency admissions.
To find the annual change in admissions for a set of reductions in daily
TSP levels, we sum the daily changes in admissions:
365
Z
d-1
AEAA - Z (0.0002857)(ATSPd)(EA&) '(3.25)
where AEAA = change in annual number of respiratory disease emergency
admissions.
Equation (3.25) relates daily admissions to daily levels of TSP.
However, we only have data on the annual average of TSP and annual admis-
sions. Therefore, the equation cannot be directly used for our calcula-
tions.
Because data on daily levels of admissions and TSP are not available,
we must convert Equation (3.25) to a relationship between annual TSP and
admissions. Equation (3.25) is equivalent to:
AEAA - (0.0002857)(ATSPa)(EAA) (3.26)
where ATSP& = change in annual arithmetic mean of TSP (change in TSPd).
3-52
-------
Equation (3.26) indicates that for a 1 (ig/nr reduction in annual TSP,
there is a 0.02857 percent reduction in annual respiratory disease
emergency admissions.
Application of tie Second Fractional Form —
The population of Steubenville was 31,000 during the period of the
study. Therefore, assuming this population represents the population
served by the study hospital, there was a change in per capita daily
respiratory disease admissions of about (0.007/31,000) = 2.258 x 10~' for
9
each 1 (ig/m change in TSP.
AEAPd - (2.258 x 10~7)(ATSPd) (3.27)
where AEAP. = change in number of respiratory disease admissions per
person on day d.
The change in the annual number of admissions per capita for a set of
changes in daily TSP levels is given by:
365
AEAPa - Z (2.258 x 10 7)(ATSPd) (3.28)
d=l
(2.258 x 10"~7)(365)(ATSPa) (3.29)
(8.24 x 10~5)(ATSPa) (3.30)
where AEAPa - change in annual number of respiratory disease emergency
admissions per person.
Our national data indicate that there are approximately 0.0044
respiratory disease emergency admissions per person annually. Dividing
3-53
-------
(8.24 x 10 ) by 0.0044, we find a 1.87 percent reduction in admissions for
each 1 fig/in reduction in annual TSP:*
AEAa = (0.0187)(ATSPa)(EAa) (3.31)
a. A A
Benefit Estimation Formulas —
For the two alternative functional forms, we have estimated the per-
centage reduction in respiratory disease emergency admissions for a unit
change in TSP. Under the first functional form, there is a 0.02857 percent
reduction for each 1 ug/m reduction in TSP. Under the second functional
form, there is a 1.87 percent reduction,
These coefficients are based on the change in and base levels of
emergency admissions for both acute and chronic respiratory disease.
However, we will apply the coefficients only to DME, WLD, and RAD for acute
respiratory disease because it is assumed that acute disease incidence will
be more sensitive to acute exposure than chronic disease incidence.** The
bias produced by using the percentage change in total admissions to
represent the percentage change in acute admissions (and incidents) depends
on the relative sensitivities of acute and chronic disease admissions to
PM. For example, if chronic admissions are less sensitive, there will be a
downward bias.
Under our assumptions, the percentages can be applied to the base
level of acute respiratory disease DME, WLD, and RAD in each county to
* Since the 0.007 figure gives the change in admissions in only one of the
two Steubenville hospitals, per capita changes in admissions may be
underestimated. On the other hand, if the Steubenville hospital serves
a larger community than 31,000 (approximately one-third of the county
population), the per capita changes may be overestimated. The 1.87
percent differs from the 0.02857 percent reduction found by Samet e_t al.
because of the difference in the base number of admissions per capita in
Samet et al.'s sample and our national data, and any discrepancy between
31,000 and the size of the population served by the study hospital.
** Our data do not allow us to analyze effects on aggravation of existing
chronic disease.
3-54
-------
determine the effects of a change in TSP. The following benefit estimation
formulas can be used. Figure 3-2 summarizes the approach used for each
estimate.
Benefits of Reduced OMB (Minimum) — The initial level of DME on
respiratory disease emergency admissions in each county is assumed to be
proportional to population, Therefore, the effect of a reduction in TSP on
these DHE in a county is given by:
where AOME^
ATSPa.
population^
population
*
DME.
(0.0002857 )(ATSPai)
population.
population
)(DMEJ
(3.32)
change in DHE on respiratory disease emergency admis-
sions in county i.
change in annual TSP in county i under a standard.
population in county i.
national population.
national DME on respiratory disease emergency admis-
sions.
Benefits of Reduced DUE (Maximum) — The effects of a change in TSP on
emergency admissions only represent a part of the effects of a reduction in
TSP on acute respiratory disease. For the DME component of the minimum
estimate of the total benefits of a reduction in TSP, it was assumed that
the impact was isolated to expenditures on emergency admissions. For the
maximum estimate of total benefits, the impact on total acute respiratory
disease expenditures is considered. For this estimate, the changes in
emergency admissions are related to changes in overall incidence of
disease. The change in total acute respiratory disease expenditures in a
county is estimated by:*
* This equation is equivalent to: ATME^ = (8.24 x 10~5)(ATSPai)(popula-
tioni)(TMEQ/EAJl) where EAn = national respiratory disease emergency
admissions.
3-55
-------
o
•hi
ot
a
•»H
•hi
M
H
§
a
14
H
«
S
a
o
M
0
M
0
•M
14
•O
a
o
ft
H M
« BE]
O
a -H
•H i-*
J -C
O
•hi
A
*v4
•hi
M
w
a
o
a
iH
S3
•«•(
X
M
O
.
« 0
« a
a «
*H ao
tH H
a o
o a
Z M
g
0 0
A n
0
^ •«
S-H
M
o a
•* O
jj rj
** s*
O
5 £9
0 S
ft 0
0
CO
.
M
0
+»
0*
M
•H
ft
M
O
M
o
•U 4>
a a
0 O
< Z
«
M
«S
O
«
•»<
0
b
H
0
•h>
c«
14
•H
ft
M
e
M
0
+» o
a a
o o
< Z
4>
oo
a
48
M
BU
CO
H
- "3
a «
o n
o
M •« .
J_ft *^
** *W
•** w
a a
14 O
J U
^^
c
-1
•"1
«l
4J
O
a
M
N^
QB
O
^^
ctt
a
•«4
•M
W
64
>>
4J
•*
*o
•H
o
VI
0.
eo
f4
tb
3-56
-------
ATMEi = (0.0187)(ATSPai)
/population.^
I KTME )
\population / a
(3.33)
where ATME.
total DME on acute respiratory disease in county i.
TME = national DME on acute respiratory disease.
Benefits of Reduced YLO (Mi.nj.aui) — There are no data on WLO or RAD
associated with emergency admissions alone. Therefore, effects on total
respiratory disease WLD and RAO will be considered. Acute respiratory
disease WLD are assumed to be proportional to employment. The value of
each work-loss day is set equal to the average county wage. Therefore, the
effect of a reduction in TSP on the value of acute respiratory disease WLD
in a county is given by:
where
AVWLD,
AVWLD^
emp.
empn
WLD
Wi
(0.0002857)(ATSPai)
(3.34)
change in value of acute respiratory disease WLD in
county i.
employment in county i.
national employment.
national number of acute respiratory disease WLD.
average daily wage in county i.
Benefits of Reduced WLD (Maximum) — The effect of a change in TSP on
the value of acute respiratory disease WLD in a county is given by:
AVWLD^
(0.0187)(ATSP&i)
fempj
lempT
J(WLD) (W£)
(3.35)
Benefits of Reduced RAD (Minimum) — Acute respiratory disease RAD,
net of WLD, are assumed to be proportional to population. The value of
each RAD is set equal to one-half the daily wage for the counties in our
3-57
-------
analysis* Therefore, the effect of a reduction in TSP on the value of
acute respiratory disease RAD in a county is given by:
AVRADi
where
RAD
HW
(0.0002857)(ATSPai)
'population.
[population
(RADMHW)
(3.36)
- change in value of acute respiratory disease RAD in
county i.
= national number of acute respiratory disease RAD.
3 one-half of the average daily wage for the counties in
our analysis.
Benefits of Reduced RAO (Maxiau) — The effect of a change in TSP on
the value of acute respiratory disease RAD in a county is given by:
AVEAD,
(0.0187)(ATSPai)
S population.
population
(RAD)(HW)
(3.37)
ute Morbid! try Effects of
As discussed previously, we will base our estimate of the acute
morbidity effects of chronic exposure on the work of Saric et al. (24).
They study the effects of chronic exposure to pollution, examining the
incidence of acute respiratory disease in a control area and a polluted
area. Table 3-4, presented earlier, summarizes the study findings. To
apply the results to our analysis, the population must be divided into the
appropriate categories.
The age ranges for the classifications used by Saric e± aji. are not
given. For our analysis, we will apply the results for "brothers and
sisters" to the population under 24, the average of the results for
"mothers" and "fathers" to the population 24 to 55, and results for
"grandmothers and grandfathers" to the population over 55. We obtained age
breakdowns by state.
3-58
-------
Table 3-10 shows the disease incidence in the clean and polluted areas
for our three age categories. Saric et al. estimate the number of
incidents per 100 people. Using Table 3-10, we will estimate the effects
of changes in TSP on acute respiratory disease. It will be assumed that
disease effects occur during the year of exposure. If the larger
differences in TSP in previous years explain differences in disease
incidence in the study year, our coefficients may be overestimated.
For this estimation, the difference in the levels of TSP in the two
areas must be determined. The monitored annual level of TSP in the
polluted area is approximately 200 ug/m. The TSP level is not given for
the clean area. The health effects observed by Saric e_t al. could begin to
occur at annual TSP levels anywhere between 0 and 200 ug/m .
The lower bound level of effect assumed will affect our benefit calcu-
lations in two ways. The estimated incremental effect of a unit change in
Table 3-10
DISEASE INCIDENCE BY AGE
(per 100 people)
Age
Group
0-24
24-54
55+
Acute Respiratory
disease incidents
Novembe r-Apr il
Polluted
Area
151.7
70.2
63.4
Clean
Area
97.3
60.6
51.6
Coefficient
First
Form
0.0018
0.0007
0.0009
Second
Form
0.0036
0.0010
0.0018
Annual No. of
Acute Respir-
atory Disease
Incidents Per
Person in the
U.S.*
1.52
1.0
0.67
* Derived from annual average of acute respiratory disease incidents per
person for age groups 0-24, 24-64, and 65 and over from National Health
Survey (36).
3-59
-------
TSP depends on the level of effect assumed. If the difference in health
observed by Saric et al. is assumed to result from a change in TSP from 130
to 200, the estimated effect of a unit change in TSP will be higher than if
the difference in health is assumed to result from a change in TSP from 100
to 200. The level of effect, however, will also affect the range of TSP
values over which benefits are calculated. If a 130-lower bound level is
assumed, effects will only be estimated for changes in annual TSP over 130
ug/m for the counties in our analysis, while effects for changes between
3 3
100 and 130 ug/m will also be estimated if a level of 100 fig/or is used.
For our benefit calculations, we will estimate the incremental effect
of a change in TSP assuming that the health effects observed by Saric et al
result from a change in TSP from 0 to 200 ug/m . This assumption will
yield the most conservative estimate of the effect and benefits of a unit
change in TSP. For a minimum estimate (using the first functional form) of
the effects of a change in TSP on health for the counties in our analysis,
only changes over an effects level of 200 ug/m will be considered. For a
maximum estimate (using the second functional form), benefits will be
calculated for the entire range of TSP values. This use of alternative
lower-bound level of effect reflects uncertainty concerning the TSP level
at which the acute morbidity effects occurred.*
Application of the First Functional Font —
The percentage change (difference) in acut« respiratory disease inci-
dence between the two locations is a linear function of the change in TSP:
ADI& = (0)(ATSPa)(DIa) (3.38)
* The above approach will yield a conservative range of benefit estimates
consistent with the evidence of Saric .e_t .§_!. The minimum estimate is
below the estimate that would be yielded by the first functional form if
any TSP level below 200 ug/m were selected as the level in the clean
area. Application of the second functional form would yield benefits
above the maximum estimate if several values between 0 and 200 ug/m were
selected for TSP in the clean area.
3-60
-------
where ADI» = difference in annual number of acute respiratory disease
incidents.
ATSPa = difference in annual average TSP.
DI& = base annual number of acute respiratory disease inci-
dents.
P = coefficient relating changes in TSP to changes in disease
incidence.
We can estimate 0 for each age group. As an example, consider the 0
to 24 age group. The disease incidence in the clean area is 54.4 below the
polluted area level of 151.7. It is assumed that this difference occurs
for a reduction in annual TSP from 200 to 0 |ig/m . Substituting into
Equation (3.38), the following equation is obtained:
54.5 - (p) (200)(151.7)
p = 0.0018 (3.39)
For a 1 (ig/m reduction in annual TSP, there is a 0.18 percent decrease in
incidence of acute respiratory disease. Table 3-10 shows the coefficient
derived under this method and the average number of acute respiratory
disease incidents per person for each age group in our analysis.
<9
The effect of a change in annual average TSP over the 200 (ig/m
threshold on acute respiratory expenditures is given by:
0-24 year olds;
ADIa - (0.0018)(ATSPa)(DIla) (3.40)
24-54 year olds:
ADI = (0.0007MATSPHDI2,,) (3.41)
A a ft
55+ year olds;
ADI = (0.0009MATSPHDI3) (3.42)
& ft &
3-61
-------
where ^11 = base annual number of acute respiratory disease incidents
for 0 to 24 year olds.
DI2 = base annual number of acute respiratory disease incidents
for 24 to 54 year olds.
DI3 = base annual number of acute respiratory disease incidents
for 55+ year olds.
Application of tie Second Fractional For* —
The change in disease incidence is a linear function of the change in
TSP:
ADIPa = (0)(ATSPa) (3.43)
where ADIP& = change in annual average number of acute respiratory
disease incidents per person.
P =» coefficient relating TSP to per capita acute respiratory
disease incidents.
We can estimate 0 for each age group. For the 0-24 group, the change
in disease incidents per person is 0.544 for six months. The annual change
in approximately 1.088 (0.544 x 2). This difference is assumed to occur
for a change in the annual TSP level of 200 jig/m3. Substituting these
values in Equation (3.43):
1.088 * (Pi(200)
p - 0.0054 (3.44)
For a 1 (ig/m reduction in annual TSP, the number of acute respiratory
disease incidents per person decreases by 0.0054. To find the percentage
change, we divide 0.0054 by 1.52, the average number of annual acute
respiratory disease incidents per person for the 0-24 year group. There is
a 0.36 percent change in the number of incidents per capita for each 1
Hg/m change in annual TSP. Table 3-10 shows the coefficient derived for
each age group under this method.
3-62
-------
The effect of a change in annual average TSP on acute respiratory
disease expenditures is given by:
0-24 year olds:
ADI, = (0.0036)(ATSPa)(DIl ) (3.45)
& cl It
24-54 year olds;
ADI& - (0.0010)(ATSPa)(DI2a) (3.46)
55+ year olds;
ADIa = (0.0018)(ATSPa)(DI3a) (3.47)
Benefit Estimation Formulas —
For the two alternative functional forms, we have estimated the
percentage reduction in acute respiratory disease incidents for a unit
change in TSP. The results are summarized in Table 3-10. For example, for
the 0-24 year group, the first functional form yields a 0.18 percent reduc-
tion for each 1 jig/m3 reduction in TSP above the level of 200 ug/m . Under
the second functional form, there is a 0.36 percent reduction for this
a
group for each 1 ug/m reduction in TSP.
Under our assumptions, these percentages can be applied to the base
number of acute respiratory disease DME, WLD, and RAD in a county to
determine the effects of a change in TSP. The following estimation
formulas can be developed. Figure 3-3 summarizes the approach used in each
estimate.
Benefits of Reduced DUE (Minimum) — For each age group, DME are
assumed to be proportional to population. Therefore, the effect of a
reduction in TSP over 200 ug/m on acute respiratory disease DME in a
county is given by:
3-63
-------
o
+4
at
a
•
-------
0—24 year olds;
ATME.
24-55 year olds:
(0.0018MATSP .)
a i
rpoplA
- (TMEl )
(3.48)
(0.0007)(ATSPai)
55+ year olds;
ATMEi = (0.0009)(ATSPai)
/pop2 /
=-J(TME2 „)
\pop2
(pop3.
(3.49)
(3.50)
where ATMEi
pop2£
pop3i
ATSPai
TMEln
TME2n
TME3_
change in acute respiratory disease DME in county i.
0-24 year old population in county i.
24-55 year old population in county i.
*
55+ year old population in county i.
0-24 year old national population.
24-55 year old national population.
55+ year old national population.
change in annual TSP.
acute respiratory disease DUE on 0-24 year olds.
acute respiratory disease DME on 24-54 year olds.
acute respiratory disease DME on 55+ year olds.
Benefits of Reduced DUE (Maxiaua) — The effect of a change in TSP on
acute respiratory disease in a county is given by:
3-65
-------
0-24 year olds;
ATME£ = (0.0036)(ATSPai)
-KTMEl )
V
24-54 year olds;
ATMEi = (0.0010)(ATSPai)
POP2A
(TME2n)
54+ year olds;
ATMEj^ = (0.0018)(ATSPai)
pop3 A
r—s- (TME3 >
(3.51)
(3.52)
(3.53)
Benefits of Reduced 1LD (Minima) — For each age group, the number of
acute respiratory disease WLD is assumed to be proportional to employment.
The value of each WLD is set at the average daily county wage. Therefore,
the effect of a reduction in TSP over 200 ng/™ on the value of acute
respiratory disease WLD in a county is given by:
0-24 year olds
AVWLDi - (0.0018)(ATSPai)
empl
|(WLDln)
24-54 year olds;
AVWLDi - (0.0007)(ATSPai)
55+ year olds;
AVWLD;
(0.0009)(ATSPai)
'emp3 i
(3.54)
(3.55)
(3.56)
3-66
-------
where AVWLD
empli
emp3i
empln
emp2a
emp3n
WLDlft
WLD2n
WLD3.
change in value of acute respiratory disease WLD in
county i.
number of 0-24 year olds employed in county i.
number of 24-54 year olds employed in county i.
number of 54+ year olds employed in county i.
national number of 0-24 year olds employed.
national number of 24-54 year olds employed.
national number of 54+ year olds employed.
national acute respiratory disease WLD for 0—24 year
olds.
national acute respiratory disease WLD for 24-54 year
olds.
national acute respiratory disease WLD for 55+ year
olds.
W. = average wage in county i.
Benefits of Reduced WLD (Maximo*) — The effect of a change in TSP on
the value of acute respiratory disease WLD in a county is given by:
0-24 year olds;
AVWLD.
24-54 year olds:
(0.0036)(ATSPai)
em
AVWLDi - (0.0010)(ATSPai)
55+ year olds;
AVWLD. = (0.0018)(ATSPa.)
i **1
/emp3
(3.57)
(3.58)
(3.59)
3-67
-------
Benefits of Reduced RAD (Miniana) — For each age group, acute
respiratory disease RAD, net of ¥LD, are assumed to be proportional to
population. The value of each RAD is set equal to one-half the daily wage
for the counties in our analysis. Therefore, the effect of a reduction in
a
TSP over 200 |ig/m on the value of acute respiratory disease RAD in a
county is given by:
0-24 year olds;
AVRADi = (0.0018)(ATSPai)
(RADln)(HW)
24-54 year olds;
AVRADi - (0.0007)(ATSPai)
Cpop2.
555
55+ year olds;
AVRADi
where AVRAD.
RAD1.
RAD2.
RAD3.
HW
(0,0009)(ATSPai)
r pop3 5
—g- ](HW)
(3.60)
(3.61)
(3.62)
change in value of acute respiratory disease RAD in.
county i.
national acute respiratory disease RAD for 0-24 year
olds.
national acute respiratory disease RAD for 24-54 year
olds.
national acute respiratory disease RAD for 55+ year
olds.
one-half of the average daily wage for the counties in
our analysis.
Benefits of Reduced RAD (Maximum) — The effect of a change in TSP on
the value of chronic respiratory disease RAD in a county is given by:
3-68
-------
0-24 year olds:
AVRAD^
24-54 year olds;
AVRAD.
55+ year olds;
(0.0036)(ATSPai)
(0.0010)(ATSPai)
rpoplj
ipopl.
(RAD1 ) (HW)
s
pop2
(RAD2 J(HW)
(3.63)
(3.64)
AVRADi = (0.0018)(ATSP&i)
Mofbi.di.tY1 Effects of ^*'*'oiii.c Exposii
(HW)
(3.65)
As discussed previously, our estimates of the chronic morbidity
effects of chronic exposure Till be based on the work of Ferris et' al.
(26,28). It should be noted that these estimates will capture effects on
chronic respiratory disease incidence but not aggravation of existing
chronic disease. Ferris et al. relate chronic nonspecific respiratory
disease prevalence in a New Hampshire town for two years to the annual
level of TSP. Figures 1 and 2 in the Ferris et al. study show the change
in the symptom prevalence rates for chronic respiratory disease. The
symptom prevalence rates represent the number of chronic respiratory
disease incidents per 100 people.
The information in. figures 1 and 2 of Ferris will be used to estimate
the effects of changes in TSP on health. The estimation procedure is
similar to that developed for Saric et al. Based on the findings of Ferris
et al.. health effects will be estimated for changes in annual TSP at
3-69
-------
levels of 130 |ig/m and above for both the minimum and maximum estimate.*
Thus, no benefits will be estimated fox most of the range considered by the
study of Bouhuys et al. (25) which found mixed evidence of effects.
Application of the First Functional Font —
Figures 1 and 2 of Ferris et al. (28) indicate that the weighted
average percentage change in the number of chronic respiratory disease
incidents between the two years was approximately 36.5 percent for males
and 46.5 percent for females (derived by using Tables 4, 5, and 10 of
Ferris et al. to weight results for different smoking categories).** The
change in annual average TSP was about 50 ug/m . From these results, the
following equations can be derived.
Male;
ACDI.
(0.0073)(ATSPa)(MCD1&)
(3.66)
Female;
ACDI,
(0.0093)(ATSPa) (FCDla)
(3.67)
where ACDI
MCDI
a
FOWL
change in the number of chronic respiratory disease
incidents.
initial annual number of chronic respiratory disease
incidents for males.
initial annual number of chronic respiratory disease
incidents for females.
* Ferris et al. only looks at the adult population. However, in our
analysis, benefits will be calculated for the entire population. The
results of Saric et al. indicate children's health may be more sensitive
to changes in TSP than adults', in which case total benefits will be
underestimated.
** Because only rates, not actual numbers of incidents, were given, the
percentage change in total incidents cannot be determined. Instead, the
weighted average of the percentage change for each group will be used.
3-70
-------
ATSP& = change in annual TSPug/m3. No change for annual TSP
levels under 130 ug/m is considered.
For a 1 ug/m reduction in annual TSP in the relevant range, there is
a 0.73 (36.5/50) to 0.93 (46.5/50) percent change in disease incidents.
Application of tie Second Functional For* —
Figures 1 and 2 of Ferris (28) indicate that the change in the number
of chronic respiratory incidents per person between 1961 and 1967 is
approximately 0.13 for males and 0.096 for females (derived by using Tables
4, 5, and 10 to weight results for different smoking categories). The
change in annual average TSP is 50 ug/m . Therefore, the following
equations can be derived:
Male:
ACDIPa - (0.0027)(ATSPa) (3.68)
a A
Female;
ACDIPa - (0.0019)(ATSPa) (3.69)
ft &
where ACDIP = change in the annual number of chronic respiratory
disease incidents per person.
9
For a 1 ug/m reduction in TSP, there is a reduction of 0.0027
(0.134/50) to 0.0019 (0.0962/50) in the number of chronic respiratory
disease incidents per person.
Our data indicate that there are an average of 0.247 and 0.281 chronic
respiratory disease incidents annually for males and females, respectively.
Dividing 0.0027 by 0.247 and 0.0019 by 0.281, we find a 1.09 percent
reduction for males and a 0.68 percent change for females in the number of
chronic respiratory disease incidents for each 1 ug/m change in TSP.
3-71
-------
Male;
ACDla = (0.0109)(ATSPa)(MCDla) (3.70)
Female:
ACD1 - (0.0068)(ATSP,)(FCDla) (3.71)
A 41 a
Benefit Estimation Formulas —
For the two alternative functional forms, we have estimated the per-
centage reduction in chronic respiratory disease incidents for a 1 ng/m
reduction in TSP. Under the first functional form, there is a 0.73 percent
reduction for males and a 0.93 percent reduction for females for each 1
a 3
reduction in TSP above the effects level of 130 ug/m . Under the
second function fora, there is a 1.09 percent reduction for males and a
a
0.68 percent reduction for females for each 1 ug/m reduction in TSP over
130 (ig/m3.
The second functional form yields larger changes for males and smaller
changes for females than the first functional form. For the counties in
our analysis, the first functional form yields a lower estimate of total
incremental benefits. Therefore, the first functional form will be applied
for a minimum estimate of total incremental benefits and the second
functional form will be applied for a maximum estimate.
Under our assumptions, these percentages can be applied to the base
number of chronic respiratory disease DME, WLD, and RAD in a county to
*
determine the effects of a change in TSP over 130 ug/m , the level at which
effects were observed by Ferris et al. * The following estimation formulas
* The percentages developed above assume that the differences in disease
incidence in 1961 and 1967 are due to differences in PM levels in the two
years. PM levels preceding each year also may have influenced disease
incidence. Additional information would be required to model a dose-
response function in which morbidity is related to PH levels over a
number of years.
3-72
-------
can be developed. Figure 3-4 summarizes the approach, used in each
estimate.
Benefits of Reduced DME (Minimum) — For each sex, DME on chronic
respiratory disease axe assumed to be proportional to population. There-
fore, the effect of a reduction in TSP on these DHE in a county is given
by:
Male;
ATME.^ = (0-0073) (ATSP&i)
/mpopA
(MTMEJ
\mpop I n
(3.72)
Female ;
where
ATME
ATME
(0.0093)(ATSPai)
/fpopA
(FTMEJ
(3.73)
ATSP
a.
fpopj
fpop
MTME
FTME
change in DME on chronic respiratory disease in county
i.
change in annual TSP over 130 ug/m3 in county i.
male population in county i.
female population in county i.
national male population.
national female population.
national DME on chronic respiratory disease of males
(number .of incidents times the average expenditure per
incident).
national DME on chronic respiratory disease of females.
Benefits of Seduced DME (Maximum) — The effect of a change in TSP on
chronic respiratory disease in a county is given by:
3-73
-------
o
•M
a
•H
•w
W
§
a
-H
H
08
55
H
0*
o
0
-H
o
^•i
e»
a
+4
M
w
I
a
"3
2{
M
at
0
0
v4
(3
0
2
8
o
cs,
r-4
«t
0
O
•H
•M
0
0
ft
Disease
X
M
O
VI
«rt
Ot
M
O
a
•^4
0
0
O
o
M
a
^*
o
4>>
M
O,
M
a
o
^4
0
O
VI
•a
o
VI
o
•o
•m
M
0
O
u
i
^
s
0
0
•**
-------
Male;
ATMEi - (0.0109)(ATSPai)
npop.
(MTME
(3.74)
Female;
ATME,
(0.0068)(ATSPai)
/fpop.
foT ) (F™V
\
(3.75)
Benefits of Reduced YLO (Minimua) — For each sex, the number of
chronic respiratory disease WLD is assumed to be proportional to employ-
ment. The value of each WLD is set equal to the average county wage.
Therefore, the effect of a reduction in TSP on the value of chronic
respiratory disease WLD in a county is given by:
Male;
AVWLD.
(0.0073)(ATSPai)
memp.
(MWLDJ
(3.76)
Female;
AVWLD^
(0.0093)(ATSPai)
(FWLDn)(W.)
(3.77)
There
AVWLD. = change in value of chronic respiratory disease WLD in
county i.
mempj, = male employment in county i.
mempn = national male employment.
female employment in county i.
national female employment.
femp
MWLD = national male chronic respiratory disease WLD.
3-75
-------
FWLD = national female chronic respiratory disease WLO.
W. = average daily wage in county i.
Benefits of Reduced WLD (Maximum) — The effect of a change in TSP on
the value of chronic respiratory diseae WLO in a county is given by:
Male;
AVWLD.
(0.0109)(ATSPai)
1 (MWLDT
«poptty i
i} (Wi>
(3.78)
Female;
AVWLD
(0.0068)(ATSPai)
(FWLDOM
(3.79)
Benefits of Reduced RAD (Miaimm) — For each sex, chronic respiratory
disease RAD, net of WLO, are assumed to be proportional to population. The
value of each RAD is set equal to one-half the daily wage for the counties
in our analysis. Therefore, the effect of a reduction in TSP on the value
of chronic respiratory disease RAD in a county is given by:
Male;
(0.0073)(ATSPai)
fmpop.
i »
(MRADn)(HW)
(3.80)
Female;
AVRAD,
(0.0093)(ATSPai)
rfpopi
(3.81)
There AVRAD^ - change in value of chronic respiratory disease RAD in
county i.
MRAD = national number of male chronic respiratory disease RAD.
3-76
-------
FRAD = national number of female chronic respiratory disease
n RAD.
HW = one—half the average daily wage for the counties in onr
analysis.
Benefits of Reduced RAD (Maximum) — The effect of a change in TSP on
the value of chronic respiratory disease RAD in a county is given by:
Male;
AVRAD£ = (0.0109)(ATSPai)
Female
__
/mpopA
(MRAD )(HW)
^npop J n
(3.82)
AVRAD£ = (0.0068)(ATSPai)
(HW)
(3.83)
For each study. Table 3-11 summarizes the estimated value per person
of reductions in DME, WLD, and RAD for a 1 ng/m3 reduction in TSP.
These estimates of the benefits of reduced morbidity are subject to a
number of limitations. First, they are based on studies that do not
control for the effects of different pollutants. If the major pollutants
in the study samples move together, attribution of all observed effects to
changes in one PM measure may bias our estimates upwards. Second, our
method of valuing reduced morbidity excludes some of the benefits of
reduced pain and suffering, and does not consider some reductions in
activity as discussed in the Appendix to Volume II. Furthermore, effects
on non-respiratory disease are not considered. These omissions will bias
our benefit estimates downwards. Third, the results for very limited
samples are generalized to the counties in our analysis. The effects of PH
on health may differ with the characteristics of the population, exposure
measure, and area considered. Fourth, the two single-equation functional
3-77
-------
*
04
99
H
§
M
g
0
1-1
J3
a a
0 0
•M •)-» *
o Q o *
3 3 3 04
CM ^3 OS *O O9
O0 0 H
at -H pts
W OS i-l
3 at 3<*i 09
^ w a a 3
a ^ a -C a
> ft •< *• fl
o a
•* ^ a
n •*
0 U
04 0
04
a
0
0 -W
>» o
0 r-4 3 *
^H cS *O ^
ft 0 0 04
j s M a
CM Cu a pH
O ^Sco
§o a — «
04 •** ao p
-H a. a
at a M a
> O 0 -H •<
•^4 ^
*> u M a
O O 0 -H
0 » 04
•O
M 3
a 04
3 ^H
a ^.
O ao T+
•* 3. «
•W . 0
w ^ a
«M 0 a
0 -O W <
0 O
Sou ft. a
•^
"« «» W
« ** 2 a
> -« O o
ft -^
01 i-t -M
U •) o
3 0
w a •«
o a 0
04 •< M
*» X
•* M
^W O
o ao
a o
0 -U
BQ *
yj
«M 2
0 0
0 0
ft ft
>> M
H H
X
•0
•w
09
iH
r-
f^
•69°
1
(n
O
O
•69-
^D
^•^
^
<^K
1
O
0
•ea-
tn
•
o
i
00
0
o
o
o
*
o
•69-
>,
O 0
0 -W 9»
•«J «t d
0 M 0
O -H M
•< ft -H
M a
o
M
o
•M
0
"*
4J
O
B
at
09
1-1 p* i«
•* O -H
• • •
o o o
•69- -69- -69-
1 1 1
•H 1C 00
o o o
o o o
•69- -W- «9-
O\ 00 ^*
O O 0
•W--W--W-
1 1 1
3 0 0
• • •
0 O 0
•6* -69- -69-
«S ^C *O
«S O 0
• • *
o o o
1 1 1
iH O 0
• « *
0 O O
•69--W--W-
t« *• ••
M « M
W M W
X X X
^t* ^
«s v> +
I I -4
OS
CO
O 0\
•^
1— 1 1^
d d
l i
o *-.
iH 1-4
d o
•69= -69=
«• ••
0 0
1-1 1-4
•I at
s a
at
£
O O 0
•^ ** w
a «j «
O H 0
(4 *^H M
£« ft •*
O M Q
0
O
"a
o
M
M
•«4
^
Wl
V
PE.
M
1-4
O
0
•O
a
3
o
.0
o
O
a
at
14
O
§5
H
a
M
O
ao
a
a
o
o
CO
o
u
09
O
a
a
at
O
a
(0
o
o
o
M
at
ao
a»
M
0
ed
ao
a
+» -a
o
•o *»
a
I
05
CD
*
*
3-78
-------
forms assumed for our estimates are derived from a small number of observa-
tions. A variety of alternative concentration-response functions are com-
patible with the study results. Furthermore, the studies do not consider
actions that individuals may take to offset the effects of particulate
matter on health. If the relationship between PH and health status in
these studies is estimated after this behavior has occurred, the health
benefits in this section may be underestimates of the actual benefits of PH
reductions. Finally, because of data constraints, several proportionality
assumptions are required to derive county-level data. Similarly, a range
of assumptions concerning the relationship between DME, WLD and RAD, and
the growth of these measures are required. The direction of the bias
introduced by these last three factors is uncertain.
In addition to these common biases, specific sources of bias can be
identified for estimates based on each study. Tables 3-12 and 3-13
summarize these biases.
Table 3-12
COMMON SOURCES OF BIAS IN MORBIDITY BENEFIT ESTIMATES
Source of Bias
Expected Direction of Bias*
Exclusion of other pollutants
Omission of benefits of reduced non-
respiratory disease and pain and
suffering
Generalization of study results to
counties in our analysis
Proportionality assumptions required
to generate county-level data and
benefit estimates
Functional, forms applied
* A positive sign indicates an upward bias, a negative sign indicates a
downward bias, and a question mark indicates that the direction of bias
is uncertain.
3-79
-------
Table 3-13
SPECIFIC SOURCES OF BIAS IN THE MORBIDITY BENEFIT ESTIMATES
Study
Sources of Bias
Expected Direction
of Bias
Samet et al.
Data used for number of res-
piratory disease emergency
admissions (see Appendix 3C)
Data used for average cost
per admission (see Appendix
3C)
Application of results for
total admissions to acute
disease incidents
Estimate of per capita effect
on admissions
Limitation of DME considered
to expenditures associated
with emergency admissions
- (min. est.)
+ (max. est.)
- (min. est.)
- (min. est.)
Saric et al.
Method of est. incremental
effects of TSP & applying
lower bound effects level
Method of dividing study
sample into age groups
Attribution of health effects
to study year differences in
PM without consideration of
PM levels in previous years
Ferris et al.
Application of results for
adults to entire population
Application of results to
populations with different
smoking composition and areas
with different PM composition
Use of single year PM levels
to develop est. of continuous
effects with a 130 jig/m3
lower bound effects level
3-80
-------
iT ESTIMATION
For each county in our analysis, the formulas developed above from the
mortality and morbidity studies are used to estimate the annual benefits of
implementing the reductions in alternative standards. The data sources and
values of the variables used for benefit calculations are presented in
Appendix 3C.
Since the standards will be achieved in future years, the health
benefits must be expressed in discounted present values. To be consistent
with the analysis of the costs of implementing the standards, benefits for
the period between the implementation year and 1996 will be estimated for
each standard using a 10 percent discount rate. The growth rates used to
derive socioeconomic data for future years are described in Appendix 3C.
Therefore, with a 1989 attainment date, the discounted present value in
1980 dollars in 1982 (DPV1582) of the benefits in a county is given by:
1989 Benefits. 1995 Benefits.
DPV1982 , + _ + (3.84)
(1.10)8 (1.10)14
This calculation incorporates the following two conventions used in
the cost analysis:
1) Benefits arising during a particular year are all assumed
to occur on the last day of the year.
2) The discounted present value is calculated at the beginning
of 1982.
Aggregate Benefits
The aggregate benefits resulting from reduced levels of PM under each
standard are found by summing over all the affected counties:
3-81
-------
no. of affected
counties
Aggregate Benefits - Z DPV^82 (3.85)
Benefits
Using the estimation procedures outlined above, the benefits achieved
under each standard are calculated.* These benefits represent the benefits
that would be achieved when all counties included in the analysis are in
compliance with the standard for all years under consideration.** The
total discounted present value of benefits for the period from the attain-
ment year through 1995 are estimated. The results are presented in Tables
3-14 through 3-37.
Tables 3-14 through 3-19 show the acute exposure mortality risk
benefits estimated from the study of Mazumdar et al. As shown in Table 3-
14, the acute exposure mortality risk benefits under Standard 1 range from
$0.037 billion to $14.86 billion, with a point estimate of $1.12 billion.
The acute exposure mortality risk benefits under the five other standards
are presented in Tables 3-15 through 3-19.
The largest share of acute exposure mortality risk benefits under
Standard 1, about 50 percent of the total benefits for the point estimates,
is in Region EL Region V accounts for approximately 20 percent of these
benefits. Because all of the counties are in attainment, there are no
benefits in Region 1. The remaining regions each account for about 1 to 7
percent of total benefits.
* For all benefit estimates, the percentage change in the morbidity
measures in each county is constrained to be below 100 percent. This
constraint is only binding for a few counties for the maximum estimate
of Samet et al.
** In the language of Section 9, these benefits represent "B" scenario
benefits.
3-82
-------
Table 3-14
ESTIMATED BENEFITS FOR: MAZDMDAR ACUTE MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
I
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.0
2.5
1.4
5.2
2.2
0.5
1.5
21.8
1.9
0.0
0.4
79.7
52.4
227.1
80.7
16.7
43.9
549.5
0.0
13.1
1023.7
803.7
4128 . 8
1234.6
241.9
512.9
5982.5
65.9
916.9
Total U.S.
37.1
1116.4 14858.0
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
14.8
446.9
5947.3
3-83
-------
Table 3-15
ESTIMATED BENEFITS FOR: MAZDMDAR ACTTTE MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.2
0.1
3.4
2.3
6.9
2.9
0.9
2.5
28.8
2.1
50.3
10.6
8.0
121.7
95.5
315.3
111.0
36.3
79.9
828.6
75.7
191.6
201.1
1743.1
1649.2
5869.6
1802.0
605.9
1041.4
10302.2
1117.0
1682.5 24523.1
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
20.1
673.5
9816.0
3-84
-------
Table 3-16
ESTIMATED BENEFITS FOR: MAZUMDAR ACUTE MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.2
0.1
3.4
2.3
7.0
3.0
0.9
2.5
28.8
2.2
10.6
8.0
121.7
95.5
317.5
113.6
36.5
79.9
828.6
79.5
191.6
201.1
1743.1
1649.2
5913.6
1854.4
610.0
1041.4
10302.5
1174.8
Total U.S.
50.5
1691.3 24681.8
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
20.2
677.0
9879.6
3-85
-------
Table 3-17
ESTIMATED BENEFITS FOR: MAZUMDAR ACUTE MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.3
0.2
4.1
2.6
7.7
3.2
1.2
3.1
29.7
2.6
18.3
14.3
152.8
113.7
354.7
126.7
50.1
105.0
877.1
104.4
417.3
365.4
2348.7
2056.9
6686.2
2167.8
875.2
1464.2
11240.8
1729.7
Total U.S.
54.7
1917.0 29352.2
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
21.9
767.3 11749.0
3-86
-------
Table 3-18
ESTIMATED BENEFITS FOR: MAZUMDAR ACUTE MORTALITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
Federal Administrative Region Minimum
0.5
0.4
5.9
3.6
11.8
3.5
1.8
3.9
41.9
3.2
76.5
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
26.1
22.0
226.8
162.7
557.1
150.1
81.3
132.0
1312.3
124.5
Maximum
554.5
517.6
3578.8
3060.3
10758.8
2720.1
1469.0
1834.7
17494.5
2000.2
Total U.S.
2794.8 43988.4
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
21.4
781.6 12301.3
3-87
-------
Table 3-19
ESTIMATED BENEFITS FOR: MAZUMDAR ACUTE MORTALITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
0.7
0.6
7.2
4.0
13.3
4.6
2.2
4.3
42.4
3.6
83.0
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
43.6
39.1
296.1
190.3
641.2
192.8
109.8
156.9
1367.9
156.7
Maximum
1148.6
974.7
5065.5
3761.7
12667.2
3518.1
2207.6
2353.4
18898.3
2827.5
Total U.S.
3194.5 53422.5
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
23.2
893.3 14939.6
3-i
-------
Table 3-20
ESTIMATED BENEFITS FOR: SAMET ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximam
0.0
0.2
6.8
8.1
44.3
17.1
2.4
5.7
53.5
8.4
0.0
1.7
61.7
73.5
400.4
154.4
22.1
51.2
483.2
73.0
0.0
15.6
560.7
667.2
3617.3
1391.6
199.9
460.4
4361.2
640.7
Total U.S.
146.6
1321.1 11914.6
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
58.7
528.8
4769.1
3-89
-------
Table 3-21
ESTIMATED BENEFITS FOR: SAMET ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Federal Administrative Region Minimum
1.9
2.4
13.1
18.9
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
.3
,6
65.
27.
6.5
12.1
103.7
10.8
Point
Estimate
17.4
22.2
119.6
171.4
589.6
248.9
59.3
108.9
900.2
94.7
Maximum
158.3
202,
1087,
1552.
5326,
2246,
536.8
982.8
7852.8
835.4
.7
.7
.1
.1
.2
Total U.S.
262.4
2332.2 20781.0
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
105.0
933.5
8318.2
3-90
-------
Table 3-22
ESTIMATED BENEFITS FOR: SAMET ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 199S
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Point
Federal Administrative Region Minimum Estimate Maximum
1.9 17.4 158.3
2.4 22.2 202.7
13.1 119.6 1087.7
18.9 .171.4 1552.1
65.6 592.2 5349.8
28.1 253.8 2288.6
6.6 59.6 539.4
12.1 108.9 982.8
103.7 900.3 7853.4
11.7 102.6 906.8
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
I
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
264.2
2348.0 20921.7
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
105.7
939.8
8374.5
3-91
-------
Table 3-23
ESTIMATED BENEFITS FOR: SAMET ACTTTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
6.7
4.7
19.0
24.8
74.8
33.5
9.4
18.3
117.3
21.0
329.4
Point
Estimate
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
Maximum
549.1
389.1
1571.3
2029.9
6109.5
2724.3
760.3
1472.1
8779.1
1605.1
2923.7 25989.8
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
131.9
1170.3 10403.1
3-92
-------
Table 3-24
ESTIMATED BENEFITS FOR: SAMET ACUTE MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
Federal Administrative Region Minimum
6.4
5.8
29.9
36.7
117.4
45.7
15.8
21.8
183.8
21.9
485.2
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
58.2
53.0
272.9
332.9
1065.3
414.2
142.7
197.7
1544.6
195.3
Maximum
533.0
485.5
2489.1
3023.3
9665.9
3756.7
1290.9
1792.4
13167.6
1744.3
Total U.S.
4276.7 37948.9
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
135.7
1196.0 10612.4
3-93
-------
Table 3-25
ESTIMATED BENEFITS FOR: SAMET ACUTE MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
19.5
11.7
45.4
47.7
141.6
54.0
27.5
31.2
209.5
36.7
624.8
Point
Estimate
178.0
107.0
413.9
432.1
1286.1
490.2
245.9
279.8
1753.4
327.6
Maximum
1625.0
979.6
3702.7
3912.5
11672.1
4450.2
2204.3
2510.6
14911.9
2930.4
5513.8 48899.2
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
174.7
1541.9 13674.7
3-94
-------
Table 3-26
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1.9
65.5
79.5
436.0
170.3
24.2
57.1
524.9
80.1
Total U.S.
0.0
0.0
1439.3
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
0.0
576.1
3-95
-------
Table 3-27
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
18.7
24.0
127.1
185.0
642.0
273.0
64.4
122.2
1016.0
104.3
Total U.S.
0.0
0.0
2576.8
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
0.0
1031.4
3-96
-------
Table 3-28
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between. 1989 and 1995
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION I
REGION II
REGION III
REGION IV
REGION V
REGION VI
REGION VII
REGION VIII
REGION IX
REGION X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
18.7
24.0
127.1
185.0
644.9
278.8
64.8
122.2
1016.1
113.0
Total U.S.
0.0
0.0
2594.6
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
0.0
1038.6
3-97
-------
Table 3-29
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
64.6
46.0
183.3
242.1
737.0
332.5
92.7
183.5
1149.1
202.3
Total U.S.
0.0
0.0
3233.1
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
0.0
1294.1
3-98
-------
Table 3-30
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
63.6
57.7
293.5
363.3
1171.4
459.5
157.4
221 . 5
1824.1
216.5
Total U.S.
0.0
0.0
4828.5
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
0.0
0.0
1350.3
3-99
-------
Table 3-31
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
. 0.0
0.0
192.6
116.3
445.2
473.4
1416.1
547.0
274.8
318.4
2078.8
361.3
Total U.S.
0.0
0.0
6223.8
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
0.0
0.0
1740.5
3-100
-------
Table 3-32
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
120.3
0.6
121.0
Point
Estimate
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
123.5
0.6
124.2
Maximum
127.5
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
48.4
49.7
51.0
3-101
-------
Table 3-33
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Point
Federal Administrative Region Minimum Estimate Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
120.6 123.9 127.2
0.6 0.6 0.7
121.3 124.5 127.9
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
48.5
49.9
51.2
3-102
-------
Table 3-34
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Point
Federal Administrative Region Minimum Estimate Maximum
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
0.0 0.0 0.0
120.6 123.9 127.2
0.6 0.6 0.7
121.3 124.5 127.9
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
48.5
49.9
51.2
3-103
-------
Table 3-35
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
120.6
0.6
121.3
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
123,9
0.6
124.5
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
127.2
0.7
127.9
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
48.5
49.9
51.2
3-104
-------
Table 3-36
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
Federal Administrative Region Minimum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
143.2
0.6
143.8
Point
Estimate
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
147.7
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
150.9
0.7
151.6
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
40.2
41.3
42.4
3-105
-------
Table 3-37
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
143.2
0.6
143.8
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
Maximum
Total U.S.
147.7
151.6
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
40.2
41.3
42.4
3-106
-------
Tables 3-20 through 3-25 show the acute morbidity benefits of reduced
acute exposure as estimated from the study of Samet et al. As shown in
Table 3-20, these acute morbidity benefits under Standard 1 range from
$0.147 billion to $11.91 billion, with a point estimate of $1.32 billion.
The acute morbidity benefits of reduced acute exposure under the five other
standards are presented in Tables 3-21 through 3-25.
The largest share of these acute morbidity benefits under Standard 1,
37 percent of the total benefits for the point estimate, is in Region IX.
Region V accounts for about 30 percent of these benefits. Because all of
the counties are in attainment, there are no benefits in Region 1. The
remaining regions each account for about 0.1 to 12 percent of the benefits.
Tables 3-26 through 3-31 show the acute morbidity benefits of reduced
chronic exposure as estimated from the study of Saric et al. As shown in
Table 3-26, these acute morbidity benefit estimates under Standard 1 range
from $0.0 to $1.44 billion, with a point estimate of $0.0 billion. The
benefits under the five other standards are presented in Tables 3-27
through 3-31.
The zero minimum benefits result from application of the 200 |ig/m
annual TSP level of effect. No counties have initial air quality higher
than 200 ug/m annual TSP. Therefore, there are no benefits of improving
air quality if it is assumed that no health effects occur below this level.
For the maximum estimate, an effects level is not applied and positive
benefits are achieved. Since it is the geometric mean of the maximum
estimate and the zero minimum estimate, the point estimate in each county
is also zero.
Tables 3-32 through 3-37 show the chronic morbidity benefits of
reduced chronic exposure as estimated from the study of Ferris et al. As
shown in Table 3-32, these chronic morbidity benefit estimates under
Standard 1 range from $0.121 billion to $0.128 billion, with a point
estimate of $0.124 billion. The benefits under the five other standards
are presented in Tables 3-33 through 3-37.
3-107
-------
Because the counties have initial air quality levels below the 130
Hg/m annual TSP effects level used in applying the results of Ferris et
al., there are no benefits in Regions II through VII. Over 99 percent of
the point estimate of total benefits under Standard 1 occur in Region IX.
The remaining benefits occur in Region X.
Tables 3-38 through 3-41 show the benefits that accrue under Standard
1 for all four benefits categories when all counties are not in attainment
with the standard throughout the 1989-1995 time horizon.* This can occur
because available control options are exhausted prior to standard attain-
ment. Tables 3-38 through 3-41 can be compared to Tables 3-14, 3-20, 3-26
and 3-32 where all counties were assumed to be in compliance with the same
standard. As expected, the benefits estimates in Tables 3-38 through 3-41
are below those in the other set of tables.
Estimates of Physical Effects
Implicit in the estimates of economic benefits are estimates of
changes in health status. The changes in health status include reduced
risk of mortality or morbidity. For economic valuation purposes, the
physical effects of reduced morbidity risk are further categorized into
fewer work days lost, fewer reduced activity days, and reduced direct
expenditures for medical care. In addition to the aggregate economic
benefits, individual estimates for each physical effect category are
developed for informational purposes. The estimates for each standard and
scenario can be found in the supplementary tables in Section 11 of the
report. The estimates are based on the same methods and data used in
calculating economic benefits except that the final step of economic
valuation is not performed.
* In the language of Section 9, these benefits represent "A" scenario
benefits.
3-108
-------
Table 3-38
ESTIMATED BENEFITS FOR: MAZUMDAR ACUTE MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
0.0
0.0
2.4
1.3
4.5
1.8
0.5
1.5
13.6
0.9
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
0.0
0.4
77.5
46.9
184.
61,
15.9
43.0
316.3
26.4
,5
,1
Maximum
0.0
13.1
989.8
691.7
3112.3
862.7
224.0
496.3
3246.1
313.9
Total U.S.
26.5
772.0
9949.9
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
10.6
309.0
3982.7
3-109
-------
Table 3-39
ESTIMATED BENEFITS FOR: SAMET ACUTE MORBIDITY STUDY
Benefits Occurring Bet-ween 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Point
Federal Administrative Region Minimum Estimate
0.0 0.0
0.2 1.7
6.5 59.4
6.6 59.6
31.9 288.3
10.8 98.1
2.2 19.7
5.3 48.2
28.8 260.8
2.6 23.3
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Maximum
0.0
15.6
540.2
540.4
2603,
887,
178.6
436.8
2365.5
.7
.2
Total U.S.
94.9
859.2
210.1
7778.1
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
38.0
343.9
3113.4
3-110
-------
Table 3-40
ESTIMATED BENEFITS FOR: SARIC ACUTE MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Point
Estimate
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Maximum
0.0
1.9
63.1
64.3
314.0
107.9
21.6
53.5
282.6
25.0
933.8
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
0.0
373.8
3-111
-------
Table 3-41
ESTIMATED BENEFITS FOR: FERRIS CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
103.2
0.6
103.9
Point
Estimate
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
106.0
0.6
106.7
Maximum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
108.8
0.7
109.5
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
41.6
42.7
43.8
3-112
-------
CONCLUSION
In this section, the medical epidemiology literature has been used to
estimate the health benefits that will be achieved under the six
alternative particulate standards shown in Table 3-6. Table 3-2 provided a
summary of the benefits estimated for each standard.
As the table indicates, the largest benefits are associated with
reductions in acute exposure mortality risk. The point estimates of these
benefits range from $1.12 billion for the most lenient standard to $3.19
billion for the strictest.
The next largest group of benefits is associated with the reductions
in acute morbidity resulting from reduced acute exposure. The point
estimates of these benefits range from $1.32 billion for the most lenient
standard to $5.51 billion for the most stringent.
The benefits associated with the reductions in chronic morbidity
.*
resulting from reduced chronic exposure are also shown. The point
estimates of these benefits range from $0.124 billion for the most
stringent standard to $0.148 for the most stringent.
Finally, the benefits associated with the reductions in acute
morbidity resulting from reduced chronic exposure are shown. The point
estimates of these benefits are $0.0 for all standards.
As discussed previously, these estimates are subject to a number of
caveats. First, most of the studies used for benefit calculations do not
isolate the effects of different pollutants. If the major pollutants are
positively correlated in these studies, attribution of all observed health
effects to one measure of PM may bias our estimates upwards. Second, the
results for limited samples are generalized to all of the counties in our
analysis. Since the health effects of PM may differ with specific charac-
teristics of the population, PM composition, exposure measure, and area
3-113
-------
considered, application of study results to our analysis may result in an
under— or over-estimate of benefits.
Third, the benefit estimates are based on the application of a few
concentration-response functions derived from a small number of studies.
If these simple, single-equation concentration—response functions do not
accurately reflect the complex relationship between exposure and health,
our benefit estimates may be biased. Fourth, the data required for benefit
calculations often were not available at the county level. Therefore, a
range of proportionality assumptions were required to estimate benefits for
each county. The effect of these assumptions on our results is uncertain.
In addition to these biases, our estimates are limited in their scope.
Only the reduction in mortality and acute and chronic respiratory disease
DHE, WLD and RAD are considered. Effects on non-respiratory disease and
aggravation of chronic respiratory disease and effects of chronic exposure
on mortality are excluded. Furthermore, the method of valuing reductions
in morbidity may not capture the full benefits of reductions in pain and
suffering and restriction of activity, as discussed in the Appendix to
Volume II. These omissions will result in an underestimate of the total
health benefits achieved under the standards.
REFERENCES
1. Shy, C. H. Epidemiologic Evidence and the United States Air Quality
Standards. American Journal of Epidemiology, 110:661-667, 1979.
2. U.S. Environmental Protection Agency, Office of Air Quality Planning
and Standards. Review of the National Ambient Air Quality Standards
for Particulate Matter: Revised Draft Staff Paper. Research Triangle
Park, North Carolina, December 29, 1981.
3. U.S. Environmental Protection Agency, Office of Research and Develop-
ment. Air Quality Criteria for Particulate Matter and Sulfur Oxides.
External Review Draft No. 3, Research Triangle Park, North Carolina,
October 1981.
3-114
-------
4. Holland, W. W., A. E. Bennett, I. R. Cameron, C. dn V. Florey, S. R.
Leader, R. S. Schilling, A. V. Swan, and R. E. Waller. Health Effects
of Particnlate Pollution: Reappraising the Evidence. American
Journal of Epidemiology, 11:526-659, 1979.
5. Ware, J., L. A. Thibodeau, F. E. Speizer, S. Colome, and B. G. Ferris,
Jr. Assessment of the Health Effects of Sulfur Oxides and Particulate
Matter: Analysis of the Exposure-Response Relationship. Environ-
mental Health Perspectives, 1981 (in press).
6. Commins, B. T. and R. E. Waller. Observations from a Ten-Year Study
of Pollution at a Site in the City of London. Atmospheric Environ-
ment, 1:49-68, 1967.
7. Martin, A. E. and W. H. Bradley. Mortality, Fog and Atmospheric
Pollution — An Investigation During the Winter of 1958-59. Monthly
Bulletin of the Ministry of Health, Laboratory Services, 19:56-73,
1960.
8. Mazumdar, S., H. Schimmel, and I. Higgins. Relation of Air Pollution
to Mortality: An Exploration Using Daily Data for 14 London Winters,
1958-1972. Electric Power Research Institute, Palo Alto, 1980.
9. Glasser, M. and L. Greenburg. Air Pollution and Mortality and
Weather, New York City, 1960-64. Archives of Environmental Health,
22:334-343, 1971.
10. Schimmel, H. and T. J. Murawski. The Relation of Air Pollution to
Mortality. Journal of Occupational Medicine, 18:316-333, 1976.
11. Schimmel, H. Evidence for Possible Acute Health Effects of Ambient
Air Pollution from Time Series Analysis: Methodological Questions and
Some New Results Based on New York City Daily Mortality, 1963-1976.
Bulletin of the New York Academy, 54:1052-1108, 1978.
12. Martin, A. E. Mortality and Morbidity Statistics and Air Pollution.
Proceedings of the Royal Society of Medicine, 57:969-975, 1964.
13. Pitcher, Hugh. Note on the Statistical Methodology Used in Mazumdar,
Schimmel, and Higgens. Unpublished manuscript, 1982.
14. Goldberger, Arthur. Stepwise Least Squares: Residual Analysis and
Specification Error. Journal of the American Statistical Association,
1961. pp. 998-1000.
15. Schimmel, H. and L. Greenburg. A Study of the Relation of Pollution
to Mortality: New York City, 1963-1968. Journal of the Air Pollution
Control Association, 22:607-616, 1972.
16. Lawther, P. J. Climate, Air Pollution, and Chronic Bronchitis. Pro-
ceedings of the Royal Society of Medicine, 51:262-264, 1958.
3-115
-------
17. Lawther, P. J., R. E. Waller, and M. Henderson. Air Pollution and
Exacerbations of Bronchitis. Thorax, 25:525-539, 1970.
18. Samet, J. M., F. E. Speizer, Y. Bishop, J. D. Spongier, and B. G.
Ferris, Jr. The Relationship Between Air Pollution and Emergency Room
Visits in an Industrial Community. Journal of the Air Pollution
Control Association, 31:236-240, 1981.
19. Dociery, D. W., N. R. Cook, B. G. Ferris, Jr., F. E. Speizer, J. D.
Spengler, and J. H. Ware. Change in Pulmonary Function in Children
Associated with Air Pollution Episodes. Presented at the 74th Annual
Meeting of the Air Pollution Control Association, Philadelphia, PA,
June 1981.
20. Colley, J. R. T. and L. J. Brasser. Chronic Respiratory Diseases in
Children in Relation to Air Pollution: Report on a WHO Study. EVRO
Reports and Studies, 28, Regional Office for Europe, Copenhagen, 1980.
21. Lunn, J. E., J. Know el den, and J. W. Roe. Patterns of Respiratory
Illness in Sheffield Junior School Children: A Follow-up Study.
British Journal of Preventive Social Medicine, 24:223-228, 1970.
22. Douglas, J. W. B. and R. E. Waller. Air Pollution and Respiratory
Infection in Children. British Journal of Preventive Social Medicine,
20:1-8, 1966.
23. Lunn, J., J. Knowelden, and A. J. Handyside. Patterns of Respiratory
Illness in Sheffield Infant School Children. British Journal of
Preventive Social Medicine, 21:7-16, 1967.
24. Saric, M., M. Fugas, and 0. Hrustic. Effects of Urban Air Pollution
on School Age Children. Archives of Environmental Health, 36:101-108,
1981.
25. Bouhuys, A., G. J. Beck, and J. B. Schoenberg. Do Present Levels of
Air Pollution Outdoors Affect Respiratory Health? Nature, 276:466-
471, 1978.
26. Ferris, B. G., Jr., H. Chen, S. Puleo, and R. L. H. Murphy, Jr.
Chronic Nonspecific Respiratory Disease in Berlin, New Hampshire,
1967-1973: A Further Follow-up Study. American Review of Respiratory
Disease, 113:475-485, 1976.
27. Ferris, B. G., Jr. and D. 0. Anderson. Prevalence of Chronic Respira-
tory Disease in a New Hampshire Town. American Review of Respiratory
Diseases, 86:165-177, 1962.
28. Ferris, B. G., Jr., I. Higgins, M. W. Higgins, and J. M. Peters.
Chronic Non-Specific Respiratory Disease in Berlin, New Hampshire,
1961-1967: A Follow-up Study. American Review of Respiratory
Disease, 107:110-122, 1973.
3-116
-------
29. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10, No. 107. Hospital Discharges and Length of Stay:
Short-Stay Hospitals. 1972.
30. U.S. Department of Labor. Geographic Profile of Employment and
Unemployment, 1979. December 1980.
31. Cooper, B.S. and D. P. Rice. The Economic Cost of Illness Revisited.
Social Security Bulletin, 39-2:21-35, 1976.
32. U.S. Department of Commerce. 1980 Statistical Abstract of the United
States, Washington, DC, 1980.
33. U.S. Bureau of Labor Statistics. Unpublished employment data, 1980.
34. Rossiter, L. F. and D. C. Walden. Pediatric Care: Charges, Payments
and the Medical Setting. Paper presented at APHA Annual Meeting,
Health Administration Section, New York, NY, November 1979.
35. U.S. Bureau of the Census. Population Report Series P-25 No. 873.
Washington, DC, February 1980.
36. U.S. Department of Health, Education and Welfare. Vital Statistics of
the United States: Mortality. 1978.
37. U.S. Bureau of the Census. County Business Patterns. 1978.
38. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10, No. 136. Current Estimates from the Health Interview
Study: United States 1978, Hyattsville, MD, November 1979.
39. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10, No. 84. Prevalence of Chronic Respiratory Disease
1970, Hyattsville, MD, September 1973.
40. U.S. Bureau of Economic Analysis. Projections of the Population 1976-
2000. Memorandum, March 1981.
41. U.S. Bureau of Economic Analysis. 1980 OBERS BEA Regional Projec-
tions, Vol. 3 - SMSAs. July 1981.
42. U.S. National Center for Health Statistics. Annual Summary of Births,
Deaths, Marriages, and Divorces for the United States. Monthly Vital
Statistics Report 29-13, Hyattsville, MD.
43. U.S. Department of Commerce News. Projections of Personal Income to
the Year 2000. December 9, 1980.
3-117
-------
44. U.S. National Center for Health Statistics. Vital and Health Statis-
tics, Series 10, No. 136. Current Estimates for the Health Interview
Study, United States 1979, Hyattsville, MD, April 1981.
45. 1980 Hospital Record Survey (memorandum).
3-118
-------
APPENDIX 3A
APPLICATION OF AIR QUALITY DATA TO MAZUMDAR ET AL.
TSP-BS TRANSFORMATION
To apply the study of Hazumdar ert a_l., we must convert our PM10 data
to TSP and then convert from TSP to BS. Commins and Waller (6) made side-
by-side readings of BS and TSP at a central London site for the years 1958-
1963. Since the study by Mazumdar e_t ^1. covers this location and these
years, conversions based on Commins and Waller data are appropriate. The
Criteria Document notes that "site-specific calibrations of PM mass (ug/m )
against BS reflectance readings carried out in 1956 at a central London
site, as described by Waller, appear to confirm reasonably well the BS mass
(in jig/m ) to reflectance calibration (D.I.S.R.) curve employed in
estimating mass from reflectance readings at the above seven London sites
[sites used by Hazumdar et al.] in 1958—59 and for several more years until
1963" [(3), pp. 14-181.
Holland et al. report the results of Commins and Waller and find that
3
"At smoke concentrations of the order of 100 ng/m , the corresponding high
volume results (in London) are about double those of the smoke figures,
a a
while around 250 (ig/nr smoke (BS) the ratio is about 4:3, and by 500 ug/nr
smoke (BS) it is approaching unity [(4), p. 549]. A simple nonlinear model
of the following form can be fitted to the Commins and Waller data reported
by Holland £t al.:
BS,
(C2 +
where BSd = daily level of BS.
TSPd = daily level of TSP.
C = constant.
3-119
-------
A value of approximately 180 for C yields the best fit.* Therefore, the
daily level of BS for a given daily level of TSP is given by:
f(BSd) = TSP^/(32,400 + TSPj) (3A.1)
The square of the daily level of BS is given by:**
g(BSd) = TSP^/(32,400 + TSPJ)2 (3A.2)
It must be noted, however, that any transformation function can only
provide a very rough approximation of the BS levels corresponding to TSP
readings.
APPLICATION OF THE TRANSFORMATION FUNCTION TO OUR DATA
Equations (3A.1) and (3A.2) relate daily BS and daily BS squared to
daily TSP. However, we do not have data on pre— and post-standard daily
levels of TSP. The following method for determining daily levels of BS
from the available information was developed.
From our data, we can derive the annual arithmetic mean and the
standard arithmetic deviation of TSP. To estimate the variance of the 24-
hour averages, it is assumed that daily TSP is lognormally distributed (see
Section 2). Then, the variance is given by:
2
V(TSP) - (TSPa)2[e(ln s«d) - l] (3A.3)
* The Commins and Waller data do not cover post-1963 years. However,
Mazumdar et al. do not control for changes in calibration curves after
1963. The net bias in estimates of mortality effects of changes in TSP
resulting from application of a transformation based on Commins and
Waller to the results of Mazumdar et al. is not clear.
** The results of these functions are consistent with the bounds of the
linear transformations in the PM Staff Paper (2). A linear function,
such as BS » TSP - 100 or BS = PM10 - 100, could not be employed for our
calculations because it yields negative BS levels for TSP or PH10 below
100 ug/m . Further, a nonlinear function provides a better fit.
3-120
-------
where V(TSP) = variance around annual mean of 24-hour averages.
sgd = standard geometric deviation of daily averages.
TSP,, = annual arithmetic mean of TSP.
a
Using a Taylor series expanded around the mean to terms of the second
r, the
information.
2
order, the daily levels of BS and BS can be approximated from this
of ABSa
The daily level of BS is approximated by:
BSd = f(TSPa) + f'(TSPa)(TSPd - TSPa)
f"(TSPa)(TSPd - TSPa)2
+ = (3A.4)
Taking the expected value of both sides of Equation (3A.4),
f"(TSPa)V(TSP)
BSd = -f(TSPa) + ^ (3A.5)
Since we know the annual arithmetic mean and variance of daily TSP and
the function transforming daily TSP to daily BS, we can use Equation (3A.3)
to determine the expected value of the daily BS level under alternative
standards. Thus, in our calculations, the expected value of the change in
daily BS (change in expected value of daily BS), ABSd> under a pollution
standard will be approximated by:
f"(TSP ,)
ABSd = f(TSPal) - f(TSPa2) + VCTSP^
f"(TSP -)
j V(TSP2) (3A.6)
3-121
-------
where TSPai = annual arithmetic mean of daily TSP levels before pollu-
tion standard.
= variance of daily TSP before the standard.
TSP 2 = annual arithmetic mean of daily TSP levels after pollu-
tion standard.
V(TSP2) = variance of daily TSP after the standard.
Substituting the transformation functions into our equation for the
expected value of the change in BS, we find:
- TSPil TSp3a2
ABSd - ------- -
32,400 + TSP2^ 32,400 +
(3.14928 x 109)(TSPal)[V(TSPal)]
(32,400 +
(32,400)(TSPal)[V(TSPal)]
(32,400 + TSP2^)3
(3.14928 x 109)(TSPa2)[V(TSPa2>]
(32,400 + TSP32)3
(32,400) (TSP3^) [V(TSP&2) ]
(32,400 + TSP^j)3
(3A.7)
Thus, the mean change in daily BS is approximated from our data by Equation
(3A.7).
3-122
-------
Ap'prox'i<|iati.on of ABSj
The daily level of BS squared is approximated by:
BS? = g(TSP) + g'(TSPJ(TSP. - TSP)
U £L a Q. a
g"(TSPa)(TSPd - TSPa)2
+ = (3A.8)
Taking the expected value of both sides of Equation (3A.8),
g"(TSPa)[V(TSP)]
BSj = g(TSPa) + - - - (3A.9)
The expected value of the change in the square of daily BS (change in
expected value of the square of daily BS), ABS^, under a pollution
standard, will be approximated by:
g"(TSP ,)
ABS| = 8
-------
(129,600) (TSP^) [V(TSPal)]
(32,400 +
(32,400 +
(1.57464 z 1010)(TSP|2)[V(TSPa2)]
(32,400
(129,600)(TSp£2)[V(TSPa2)]
(32,400 + TSP^2)4
(TSP?2)[V(TSPa2)]
(3A.11)
(32,400 + TSP^2)4
Thus, the mean change in daily BS squared can be approximated from our data
by Equation (3A.11).
Application of a. Lower-Bound Effects Level
Lack of data on daily levels of BS or BS squared prevents us from
applying a no-effects level to them. Applying a no-effects level to
approximations of annual means will not yield the same result as applying
it to daily levels and then averaging.
3-124
-------
APPENDIX 3B
RESULTS OF THREE ADDITIONAL MORBIDITY STUDIES
Only one usable study was identified for each morbidity category.
Therefore, no cross-check between the results of different studies could be
made. This lack of cross-check increases the tennousness of our estimates.
Among the reasons for rejecting several morbidity studies was the lack
of site-specific calibration for BS. Because of the lack of calibration,
the concentration—response functions derived from these studies cannot be
transformed to a relationship for TSP. Comparison of results for the
rejected BS studies and included studies is confounded further by differ-
ences in the populations and health endpoints studied.
Despite these difficulties and other weaknesses discussed in the
Literature Review, we will compare the health effects observed for a unit
change in BS for three rejected BS studies to the effects of a unit change
in TSP for the studies selected.* While the precise values yielded by the
two sets of studies are not strictly comparable, comparison of the
magnitude of the effects may provide supplemental information to help judge
the reasonableness of the coefficients used for benefit estimation.
ACUTE MORBIDITY EFFECTS
Douglas and Waller (22) examine the difference between lower respira-
tory disease incidences in areas with different annual levels of BS. Using
the BS levels given in their appendix, the results in Table 3B-1 are
obtained. Table 3B-1 can be used to determine the change in admissions and
incidents for a 1 ug/m reduction in BS for the two functional forms used
for our benefit calculations.
* The results of the BS studies cannot be converted to TSP because there is
no information on the appropriate transformation function to apply to the
uncalibrated BS levels. Because of differences in the health endpoints
studied, the absolute changes identified by the studies, instead of the
percentage changes, will be compared for the second functional form.
3-125
-------
Table 3B-1
•
RESULTS OF DOUGLAS AND WALLER
Annual
BS
Level
67
138
217
281
Incidence of One
or More Lover
Respiratory
Disease Infections
(Percent)
19.4
24.2
30.0
34.1
Incidence of Two
or More Lower
Respiratory
Disease Infections
(Percent)
4.3
7.9
11.2
12.9
Lower Respiratory
Disease Hospital
Admissions
(Percent)
1.1
2.3
2.6
3.1
Following the procedure described for Saric et al., the data in Table
3B-1 can be substituted into an equation of the form:
AMB
where
ABS
MB
(ABSa) (MBa)
(3B.1)
change (difference) in morbidity.
coefficient relating changes (differences) in BS to
changes (differences) in morbidity.
change (difference) in annual BS.
base level of morbidity.
Since data are given for four levels of BS (unlike Saric et al. and Ferris
e_t aJL which only look at two levels), the percentage change in disease
incidence can be estimated for a number of different pairs of BS levels.
3-126
-------
For example, the percentage change in hospital admissions for a change
in BS from 281 to 67 is 64.5 percent [(3.1 - l.D/3.1]. Dividing 64.5 by
214, the change in BS, there is a 0.3 percent reduction in lower respira-
tory disease admissions for each 1 ng/m reduction in BS. This procedure
can be performed for several pairs of BS levels between 281 and 67. The
average of the percentage changes in admissions per unit change in BS
estimated for all pairs of BS levels is 0.33. The average percentage
change in incidence of one or more lower respiratory disease incidents and
two or more incidents is approximately 0.23 and 0.37, respectively, per 1
|ig/m3 change in annual BS.
Application of the Second F^^ctional Fora
Alternatively, the data in Table 3B-1 can be substituted into an
equation of the form:
AMB, = (pHABSJ ' (3B.2)
«L **
The average change in the number of lower respiratory disease hospital
admissions per capita per 1 (ig/m change in annual BS is about 0.00009.
The average change in the per capita incidence of one infection and two or
more infections is 0.00028* and 0.0.0040, respectively.
Comparison of Results
Table 3B-2 compares the coefficients derived from Douglas and Waller,
Samet et al. (18), and Saric et al. (24) for the first functional form.
The Douglas and Waller study shows a 0.23 to 0.37 percent reduction in
incidence of lower respiratory disease in children for a 1 {ig/m reduction
in annual BS. This estimate is slightly above the 0.07 to 0.18 percent
reduction in total respiratory disease incidence per 1 jig/m reduction in
* Estimated by subtracting the incidence of two or more incidents from the
incidence of one or more incidents.
3-127
-------
Table 3B-2
•
COEFFICIENTS DERIVED FROM APPLICATION OF THE FIRST FUNCTIONAL FORM
Study
Douglas & Waller
Douglas & Waller
Samet .ejt al.
Saric et al.
PM
Measure
BS
BS
TSP
TSP
Measure of
Morbidity
Lower respiratory
disease incidents
Lower respiratory
disease admissions
Respiratory disease
emergency admissions
Respiratory disease
incidents
Coefficient
0.0023-0.0037
0.0033
0.0002857
0.0007-0.0018
annual TSP yielded by the Saric .£_t .§_!. study. Douglas and Waller show a
0.33 percent reduction in lower respiratory disease hospital admissions per
1 |ig/m reduction in annual BS. This coefficient is approximately 12 times
larger than the percent reduction in respiratory disease emergency admis-
sions per 1 (ig/m reduction in annual TSP yielded by the study of Samet et
al. If chronic conditions are more sensitive to changes in annual TSP than
acute conditions, total admissions would be more affected by changes in TSP
than emergency admissions which are principally composed of acute cases.
Table 3B-3 compares the coefficients derived for the second functional
form.
The results across studies are consistent with expectations. The
Douglas and Waller study finds a change in the number of lower respiratory
disease incidents per capita per 1 jig/m change in annual BS of 0.0011 or
more (estimated by multiplying the incidence of two or more incidents by
two and adding this product to the incidence of one or more incidents).
The estimate from Douglas and Waller is 31 to 61 percent of the change in
3-128
-------
Table 3B-3
RESULTS FROM APPLICATION OF THE SECOND FUNCTIONAL FORM
Study
Douglas & Waller
Douglas & Waller
Samet et al.
Saric e_t al.
PM
Measure
BS
BS
TSP
TSP
Morbidity Measure
Lower respiratory
disease incidents
Lower respiratory
disease hospital
admissions
Emergency room
respiratory disease
admissions
Respiratory disease
incidents
Change Per Capita
Per 1 |ig/m3
Change in TSP
> 0.0011
0.00009
2.25 i 10~7
0.0010-0.0055
total respiratory disease incidents in children as estimated from Saric et
al. for a 1 |ig/m reduction in annual TSP. Since Douglas and Waller only
look at lower respiratory diseases (which are approximately 85 percent of
the total), and any incidents over two per person are not captured by the
measure from Douglas and Waller, it would be expected that their estimate
of the effect of a given change in PM would be lower than that of Saric et
al.
The Douglas and Waller study finds a change in the number of lower
respiratory disease hospital admissions per capita per 1 (ig/m change in
annual TSP of 0.00009. This figure is 400 times the change in respiratory
disease emergency admissions per capita estimated from Samet ejt al. Samet
et al. look at effects on total respiratory disease admissions instead of
just lower respiratory disease admissions, but do not consider non-
emergency admissions. Since lower respiratory diseases are a large
majority of total respiratory diseases, while emergency admissions are only
3-129
-------
a small component of total hospital admissions, it would be expected that
the Douglas and Waller estimate would exceed the estimate from Samet et al.
However, data on the number of respiratory disease hospital admissions and
emergency admissions indicate that the total number of admissions is less
than 10 times the number of emergency admissions (29). Therefore, even
after adjusting for the difference in the health endpoints examined, the
effect identified by Douglas and Waller greatly exceeds that identified by
Samet et al.
CHRONIC MORBIDITY EFFECTS OF CHRONIC EXPOSURE
The results of two studies of the morbidity effects of exposure to BS
— Lunn et al. (21) and Colley and Brasser (20) — can be compared to the
results of Ferris et al. (28). The Lunn et al. and the Colley and Brasser
results are based on children, limiting their comparability to Ferris et
al. which examines effects on adults. Also, for a very few observations in
Lunn et al. and Colley and Brasser, health effects decrease with the
pollution level. These observations are excluded from our calculations.
Clearly, however, these observations limit the generality and strength of
any conclusions based on the studies.
Results of L*"*" et al.
Lunn et al. compare the incidence of lower respiratory disease, inci-
dence of persistent cough, and incidence of three or more colds in four
locations with different levels of BS. The results obtained are shown in
Table 3B-4.
Application of tic First Functional Fora —
The estimated effect of a change in BS on morbidity will depend on the
pair of BS levels and the morbidity measure used to estimate the change in
disease incidence for a change in BS. The method of estimating p from data
on BS levels and morbidity in any two of the locations studied by Lunn et
al. is directly parallel to the method discussed previously for Douglas and
3-130
-------
Table 3B-4
RESULTS OF LUNN ET AL.
BS
Level
97
230
262
301
One or More Incidents
of Lover Respiratory
Tract Illness (Percent)
23.0
35.9
35.4
30.5
Incidence of
Persistent Cough
(Percent)
22.9
36.1
34.6
50.0
Incidence of
Three or More
Colds (Percent)
34.4
43.8
48.5
46.3
Waller. The average percentage change in incidence of one or more lower
respiratory tract illness is 0.20 per 1 ug/m change in annual BS. The
average percentage change in incidence of persistent cough and frequent
colds is 0.39 and 0.17 per 1 ug/m change in annual BS.
Application of the Second Functional Font —
Alternatively, the second functional form can be fitted to the data in
Table 3B-4 to find the change in morbidity for a change in BS. There is an
average change of 0.0007 in the per capita incidence of one or more lower
respiratory tract illness per 1 ug/m change in annual TSP. There is an
average change of 0.0018 and 0.0008 in the per capita incidence of persis-
tent cough and frequent colds for each 1 ug/m change in annual BS.
Results of Collev and Brasser
The results of Colley and Brasser can also be compared to the results
for Ferris et al. The data collected for the Netherlands, Poland and
Denmark are presented below in Table 3B-5. Following the procedure out-
3-131
-------
lined for Lunn et al., these data will be used to derive estimates of the
effects of TSP on chronic respiratory disease.
Application of the First Functional Form —
Applying the first functional form to the data in Table 3B-5, the
average percentage change in incidence of bronchitis and pneumonia for a 1
(ig/m change in annual BS is: 0.27 for Poland; 0.28 for the Netherlands;
and 0.62 for Denmark. The average across countries is 0.39.
Application of the Second Functional Form —
Applying the second functional form to the data in Table 3B-5, the
average change in the number of bronchitis and pneumonia incidents per
capita per 1 (ig/m3 change in BS is: 0.0013 for Poland; 0.0004 for the
Netherlands; and 0.0015 for Denmark. The average across countries is
0.0011.*
Table 3B-5
RESULTS OF COLLET AND BRASSER
Country
Poland
Netherlands
Denmark
BS Level
53
82
187
9
29
7
29
Incidence of
Bronchitis or Pneumonia
(percent)
42.7
48.0
61.4
13.4
14.2
17.7
20.5
* These results are similar in magnitude to the regression coefficients
reported in Colley and Brasser.
3-132
-------
Comparison of Results
Table 3B-6 compares the coefficients derived from Lunn et al., Colley
and Eraser, and Ferris et al. under the first functional form.
Ferris et al. measure the effect of TSP on total chronic respiratory
disease incidents, while Lunn e_t .aJL, and Colley and Brasser look at more
specific disease categories. The 0.73 to 0.93 percentage change in chronic
3
respiratory disease observed by Ferris et al. for a 1 ug/mj change in
annual TSP over 130
is slightly above the 0.17 to 0.39 percentage
change in chronic respiratory diseases observed by Colley and Brasser and
a
Lunn e_t a_l. for a 1 fig/m change in annual BS.
If the effects of PH vary by specific disease, this comparison only
serves to show that the three studies observe effects of the same magni-
tude.
Table 3B-6
COEFFICIENTS DERIVED FOR THE FIRST FUNCTIONAL FORM
Study
Lunn et al.
Lunn et al.
Lunn et al.
Colley & Brasser
Ferris et al.
PM
Measure
BS
BS
BS
BS
TSP
Type of Incidence
One or more lower
respiratory disease
incidents
Persistent cough
Recurrent cold
Bronchitis and
pneumonia
Chronic respiratory
disease
Coefficient
0.0020
0.0039
0.0017
0.0039
0.0073-0.0093
3-133
-------
Table 3B-7 compares the absolute per capita change in incidents
derived from the three studies.
In accordance with expectations, the change in the number of specific
types of chronic respiratory disease, as estimated from Lann .e_t _a_l. and
Colley and Brasser is slightly less than the change in total chronic
respiratory disease incidents as estimated by Ferris et al. for each unit
change in annual TSP. Bronchitis and pneumonia account for approximately
13 percent of acute respiratory disease incidents. Application of this
share to the results of Colley and Brasser for bronchitis and pneumonia
yields a 0.0085 change in the number of acute respiratory disease incidents
for a 1 ng/m change in BS. This figure is a little above the results from
the study of Ferris et al.
Table 3B-7
COEFFICIENTS DERIVED FOR THE SECOND FUNCTIONAL FORM
Study
Lunn et al.
Lunn et al.
Lunn et al .
Colley &
Brasser
Ferris et al.
PM
Measure
BS
BS
BS
BS
TSP
Type of
Incidence
One or more lower
respiratory
disease incidents
Persistent cough
Recurrent cold'
Bronchitis and
pneumonia
Chronic respira-
tory disease
Change Per Capita Per
1 ug/nr Change in PM
0 . 0007
0.0018
0.0008
0.0011
0.0019-0.0027
3-134
-------
SUMMAST
In this Appendix, we have derived concentration-response functions for
three of the studies rejected for use as a basis for benefit estimates.
The effects implied by these supplemental studies were compared to those
implied by the studies used for benefit estimation.
Comparison of effects was constrained by differences in the popula-
tions, health endpoints, and pollution measures. The rough comparisons
that could be made, however, indicated that the effects identified by the
included studies were generally consistent with the effects identified by
the excluded studies. The results identified by Samet et al. are conserva-
tive in comparison to the results of Douglas and Waller. The consistency
of the results of a number of studies, despite the range of confounding
factors, provides support for the magnitude of the morbidity benefits
estimated in this section. As noted, however, the studies have a number of
weaknesses and differences.
3-135
-------
APPENDIX 3C
DATA SOURCES
(All figures are given for 1980 in 1980 dollars unless otherwise
noted. Figures are for the nation.)
RESPIRATORY DISEASE EMERGENCY ADMISSIONS
Number: 1,015,000.
Source: Hospital Record Survey (45).
Comments: Respiratory disease codes with fewer than 26,000 patients
are excluded. This exclusion will bias the minimum estimate of total
benefits based on Samet et al. downwards and the maximum estimate upwards.
EXPENDITURES ON RESPIRATORY DISEASE EMERGENCY ADMISSIONS
*
Number: $64,980,000.
Source: Rossiter and Walden (34); Hospital Record Survey (33); 1980
Statistical Abstract (32).
Comments: The number of emergency admissions is multiplied by the
average charge per ambulatory visit to the emergency room (inflated to 1980
by the medical CPI). The number of visits is underestimated and any costs
associated with services outside an initial physician visit are not
included. Since an emergency admission could require a hospital stay or a
variety of tests, this figure is a very conservative estimate of expendi-
tures associated with admissions.
NO. OF RESPIRATORY. DISEASE SHORT-STAY HOSPITAL DISCHARGES PER CAPITA
Number: 0.0164.
Source: National Center for Health Statistics (29).
3-136
-------
DISECT MEDICAL EXPENDITURES (DUE)
Number: $187,564,000,000.
Source: 1980 Statistical Abstract (32).
Comments: Expenditures on dentists' services, eyeglasses, administra-
tion, research, construction, government health activities, and other
health services are excluded. The medical CPI and population growth factor
are used to inflate expenditures to 1980.
STfABR OF MEDICAL EXPENDITURES FOR RESPIRATORY DISEASE
Number: 7.9%.
Source: Cooper and Rice (31).
SHARE OF RESPIRATORY DISEASE DME ON ACUTE AND CHRONIC DISEASE
Number: Chronic — 22.8%. Acute — 77.2%.
Source: National Center for Health Statistics (38,39).
Comments: The breakdown of DHE between chronic and acute disease is
based on the percentage of total respiratory disease incidents, restricted-
activity days, and bed-loss days that are accounted for by each category.
Since the costs per day will vary with the type of respiratory disease,
this method is crude. No alternative sources of data were available,
however.
SEX BREAKDOWN OF CHRONIC RESPIRATORY DISEASE INCIDENTS
Number: Male — 45.8%. Female — 54.2%.
Source: National Center for Health Statistics (39).
3-137
-------
DUE OH CHRONIC RESPIRATORY DISEASE
Number: Male — $1,547,000,000. Female — $1,831,000,000.
Source: 1980 Statistical Abstract (36); Cooper and Rice (31);
National Center for Health Statistics (38,39).
Comments: 7.9 percent is applied to total DME to find respiratory
disease DME. 22.8 percent is applied to respiratory disease DME to find
chronic respiratory disease DME. The breakdown of chronic respiratory
disease DME by sex is assumed to be the same as the breakdown of chronic
respiratory disease incidents by sex.
NUMBER OF ACUTE RESPIRATOR! DISEASE INCIDENTS PER CAPITA
Number: 0-24 year olds — 1.52. 25-54 year olds — 1.0. 55+ year
olds — 0.67.
Source: National Center for Health Statistics (38).
NUMBER OF CHRONIC BUSH i tfATOBY DISEASE INCIDENTS **Klf CAPITA
Number: Female — 0.281. Male — 0.247.
Source: National Center for Health Statistics (39).
DME ON ACUTE RESPIRATOR! DISEASE
Number: 0-24 year olds — $5,834,000,000. 24-55 year olds —
$3,889,000,000. 55+ year olds — $1,716,000,000.
Source: 1980 Statistical Abstract (32); Cooper and Rice (35);
National Center for Health Statistics (38,39,44).
Comments: The breakdown by age is based on the breakdown of 1978 and
1979 acute respiratory disease incidents, bed-disability days, and
restricted activity days by age.
3-138
-------
CHRONIC RESPIRATOR! DISEASE YORK-LOSS DATS (WLD)
Number: Male — 16,792,000. Female — 13,595,000.
Source: National Center for Health Statistics (39); 1980 Statistical
Abstract (32).
Comments: The division of chronic respiratory disease WLD (inflated
to 1980 by the employment growth factor) between males and females is
estimated assuming that the ratio of incidents per female worker to inci-
dents per male worker is equal to the ratio of incidents per capita for all
females to incidents per capita for all males. The number of WLD for each
sex is assumed to grow at the same rate as total employment.
ACUTE RESPIRATORY DISEASE WLD
Number: 0-24 year olds — 32,253,000. 24-54 year olds — 87,197,000.
55+ year olds — 23,058,000.
Source: National Center for Health Statistics (38,44).
Comments: Results for 1978 and 1979, adjusted to 1980, are averaged.
The number of WLD in each group is assumed to grow at the rate of total
employment.
CHRONIC RESPIRATORY DISEASE RESTRICTED ACTIVITY DAYS (NET OF WLD)
Number: Male — 150,265,000. Female — 177,824,000.
Source: National Center for Health Statistics (39).
Comments: The breakdown of total RAD between males and females is
assumed to be the same as the breakdown of the number of incidents. RAD
are assumed to grow by the population growth rate. Our measure of the
number of chronic RAD may not include all permanent reductions in activity
since it is based on reductions in activity during a two week reporting
period.
3-139
-------
ACUTE RESPIRATORY DISEASE SAD (NET OF 1LD)
Number: 0-24 year olds — 396,212,000. 24-54 year olds --
214,801,000. 55+ year olds — 143,376,000.
Source: National Center for Health Statistics (38,44).
Comments: Data for 1978 and 1979, adjusted to 1980, are averaged.
RAD are assumed to grow at the same rate as each group's population.
NATIONAL EMPLOYMENT
Number: Male — 58,141,000. Female — 41,326,000. 0-24 year olds —
22,480,00.0. 25-55 year olds — 60,078,000. 55+ year olds — 16,909,000.
Source: 1980 Statistical Abstract (32).
COUNTY POPULATION
Source: Bureau of the Census (35); Bureau of Economic Analysis
(40,41).
Comments: For counties within SMSAs, SMSA population data and projec-
tions are used to estimate growth rates. For rural counties, state-level
data and projections are used.
NATIONAL POPULATION
Number: Male — 112,700,000. Female — 117,300,000. 0-24 year olds
— 95,910,000. 24-55 year olds — 86,480,000. 55+ year olds —
47,610,000.
Source: 1980 Statistical Abstract (32).
COUNTY MORTALITY
Source: National Center for Health Statistics (36).
3-140
-------
Comments: Mortality is assumed to grow at the same rate as popula-
tion. If there are large changes in the age composition of the population
over the period of our analysis, this assumption may result in a slight
bias in our estimates.
MONTHLY MORTALITY
Source: National Center for Health Statistics (42).
COUNTY NOMINAL WAGE
Source: Bureau of the Census (37); Department of Commerce (43).
Comments: Non-government and Federal government payroll information.
Excludes self-employed individuals, railroad employees, farm workers and
domestic service workers, and state and local government employees. The
payroll is divided by the number of employees and 2,080, an estimate of the
number of hours worked each year, to find the hourly wage. The real wage
for the counties in a state is assumed to grow at the rate of personal
income for the state. For all counties, the value of each RAD eliminated
is assumed to grow at the rate of personal income for the United States.
COUNTY EMPLOYMENT
Source: County Business Patterns 1978 (37); 1980 Statistical Abstract
(32); Bureau of Economic Analysis (40,41); Department of Labor (30,33).
Comments: State-level sez and age breakdowns of employment are used.
For rural counties, population projections are used to approximate employ-
ment growth in each group. For counties within SHSAs, SMSA employment
growth rates are used.
POPULATION BY AGE AND SEZ
Source: Bureau of Economic Analysis (40).
3-141
-------
Comments: State-level population data and projections by age and sex
are used to determine the county breakdowns.
CONSUMER PRICE INDEX
Source: 1980 Statistical Abstract (32).
TSP CONCENTRATIONS
Source: U.S. Environmental Protection Agency.
NOTES
The per capita incidence of disease and cost per incident is assumed
to be constant over time. Therefore, the number of admissions, expendi-
tures on admissions, number of incidents, and DME all grow by the appro-
priate population growth rate. The introduction of medical advances could
reduce future morbidity and medical expenditures. On the other hand,
growth of the elderly population and real health costs will increase future
morbidity and expenditures.
All growth rates are derived by fitting an exponential growth function
to the current and projected levels of the variables. The growth rate is
assumed to be continuous over the period of our analysis.
3-142
-------
SECTION
HEALTH EFFECTS STUDIES IN THE ECONOMICS LITERATURE
-------
SECTION 4
HRAT.Tff EFFECTS STUDIES IN THK ECONOMICS LITERATDRE
SUMMARY OF RESULTS
In this section, the health benefits of alternative standards for
particulate matter (PM) are estimated using the results of previous
epidemiological studies that have appeared in the economics literature.
These studies estimate concentration-response relationships between health
status and the ambient level of PM. The concentration-response functions
are used in this section to estimate ranges of health benefits resulting
from reductions in PM. These ranges indicate that uncertainty is inherent
in the benefit estimates and that caution is required in the use of the
point estimates associated with the PM reductions.
The benefits of alternative PM standards are reported in Table 4-1.
The benefits of other standards considered in this report are contained in
Section 11. All of the benefits are expressed in 1980 dollars and in terms
of the discounted present value in 1982 of a stream of benefits ending in
1995. The standards stated in terms of particulates that have an aero-
dynamic particle diameter of up to 10 urn (PM10) assume an attainment date
of 1989, while the standards stated in terms of total suspended particu-
lates (TSP) assume an attainment date of 1987. As indicated in the table,
the range of health benefits under the alternative standards is quite
large.
The range of mortality benefits reported in Table 4-1 is developed
from: 1) macroepidemiological studies that estimate the relationship
between the mortality risk faced by the average individual and the ambient
level of PM; and 2) the value of a marginal reduction in the risk of
4-1
-------
*
CQ
H
99
§
j
87
M M
H al
0« O
a0
&g
H ON
•< H
H
I
(w
O
a>
d
o
%2
M ^-4
Z -n
M ca
•< *-*
s
a.
o
s
M
Ex.
a
n
a
3
8
H
w) at
O
n 04
o
§
a
d
^^
X
*
*
*O
H
ca
•O
d
8j
4J
ca
CM
•
0
^*
*
1— 1
fH
VO
•
•
r-l
(S
p.
•
o
rH
tn
o
o
r-l
{•^
VO
£•«
•
(*^
iH
O
>
2 M
^ M
^3
'"a -i-
**"• CM
ao
3tn
a
o -^
p* ao
a.
1
o
o in
T"4 CS
S
CU o3
(T)
•
CO
^*
to
«
^
cs
\Q
•
•
r-
o
*
Ot
p-
o
*
o
C4
•
0
1-1
r-l
VO
•
C>i|
^s|
O
2
^
^a
^*^
ao
a
in
m
1
O
r-l
S
04
^
•
tn
TT
in
•
Tf
VO
•
m
•
p»
en
^
•
o\
iH
P-
O
*
O
00
*
o
t-l
iH
p-
fs^
^
0
>
2 w
40* (4
f>4
"Vai
-«. es
ao
3cn
8
in *••*.
tn ao
a.
1
oS
rH CS
J2
04 =3
II
II
II
II
VO II P-
• II
«S II II o3
S W II _ M
•< II 3E <
Jfl 11 <
tn 1 II H
S rr n tn Jt
•**. cs H a i
ao n -^. ^r
An II ao
-------
mortality developed in the Appendix to Volume II. Under the most lax PM10
standard of an annual arithmetic mean of 70 (ig/m and a 24-hour expected
value of 250 (ig/m3,* these benefits range from $0 to $62,1 billion. The
point estimate of benefits under this standard is equal to $12.7 billion.
As indicated in the table, the estimated benefits are larger for more
stringent standards. The benefits increase to a range of $0 to $133.3
billion under the most stringent PM10 standard of an annual arithmetic mean
of 55 jig/m and a 24-hour expected value of 150 jig/m . The point estimate
of benefits under this standard is $27.3 billion. The benefits of imple-
menting the current primary standard for total suspended particulates (TSP)
of an annual geometric average of 75 |ig/m and a 24-hour maximum value of
260 ng/m (not to be exceeded more than once a year) are estimated to
result in benefits ranging from $0 to $209.0 billion with a point estimate
of $42.8 billion.
In addition to the benefits of reductions in the risk of mortality,
the benefits of reductions in the incidence of acute illness have also been
estimated in this section. The range of acute morbidity benefits are also
reported in Table 4-1. Based on two studies of the effects of particulate
matter on the acute illness of individuals, the benefits associated with
the most lax PM10 standard range from $0.03 to $21.5 billion. The point
estimate of the benefits under this standard is equal to $10.7 billion.
Benefits are estimated to range from $0.09 to $45.6 billion under the most
stringent PH10 standard and include a point estimate of $23.4 billion.
Benefits under the current primary TSP standard range from $0.14 to $67.2
billion; the point estimate for this range is $35.2 billion.
The benefits of the reductions in chronic illness resulting from the
implementation of alternative particulate matter standards are also
estimated in this section from a longitudinal study of individuals. Table
4-1 lists the range of chronic morbidity benefits associated with these
chronic illness reductions. The most lenient PM10 standard results in
benefits estimates ranging from $2.6 to $20.2 billion and includes a point
* The 24-hour average expected to occur once a year.
4-3
-------
estimate of $11.4 billion. Under the strictest PM10 standard, these
benefits are estimated to increase to a range of $6.8 to $52.6 billion.
The point estimate under this standard is $29.7 billion. Estimates of the
benefits under the current primary TSP standard range from $10.7 to $82.7
billion with a point estimate of $46.7 billion.
The benefits estimated in this section are to be interpreted with the
following qualifications. First, all of the studies used to estimate the
benefits of reductions in particulate matter estimate the relationship
between particulate matter and health status without knowledge of the
"true" model of human health. Neither the functional form of the relation-
ship nor all of the factors influencing health status is known. For
example, most of the averting behavior that individuals may undertake to
offset the effects of particulate matter on their health is not
incorporated into the studies reviewed in this section. If the relation-
ship between particulate matter and health status in these studies is
estimated after this behavior has taken place, the benefits reported in
this section will be underestimates of the actual benefits resulting from
particulate matter reductions. Consequently, the benefits estimated in
this section can only be considered as approximations of the true benefits.
Second, like any epidemiological study, the studies reviewed in this
section are unable to control for all of the factors affecting human
health. For example, the genetic characteristics of the sample populations
are not controlled for in the studies reviewed in this section. If these
omitted factors are correlated with particulate matter, the benefits
reported in this section are under— or overestimates of the true benefits.
Third, many of these studies use particulate matter as a proxy for the
air pollution phenomenon and do not control for all of the other air
pollutants affecting health. If these pollutants are positively correlated
with particulate matter, the benefits reported in Table 4-1 may be over-
estimated. This is particularly relevant for the benefits reported for
chronic illness since these benefits are based on a study that controls
only for particulate matter.
4-4
-------
Fourth, all of the studies used in this section use pollution data
from one or more monitors in a geographic area to represent the' exposure of
all individuals within the geographic area. If the relationship between
monitored air pollution and the population's true exposure to pollution has
changed significantly since the studies were done, the benefits estimated
in this section may only be approximations of the true health benefits of
air pollution control.
Fifth, the morbidity benefits estimated in this section include
estimates of the reductions in medical expenditures associated with
illness. The morbidity studies used in this section do not estimate the
relationship between medical expenditures and illness due to exposure to
particulate matter. For the purposes of this study, it has been assumed
that the percentage reduction in medical expenditures is equal to the
percentage reduction in illness estimated from these morbidity studies. If
medical expenditures go down by more (less) than that indicated by the
percentage change in illness, the benefits associated with reductions in
medical expenditures will be underestimated (overestimated).
Sixth, most of the studies reviewed in this section examine the
effects of particulate matter on urban populations. If the effects of
particulate matter on health differ between urban and rural populations,
for reasons other than the differences in ambient PH concentrations, the
use of these studies' results may under— or overestimate the health
benefits in rural areas.
And finally, the health studies reviewed in this section do not
incorporate particle size information. Benefits shown in Table 4-1 for the
PM10 standards are based on the TSP change that results. Comparisons
across PM10 and TSP standards thus reflect only differences in relative
stringency in terms of the TSP reduction; they do not reflect differences
in particle size. If PM10 standards lead to proportionately larger reduc-
tions in PM10 relative to TSP, benefits for the PM10 standards may be
underestimated. Data from the cost and air quality analysis suggest that
proportionately larger reductions do not generally occur. However,
4-5
-------
approximations in that analysis are such, that the comparisons should still
be interpreted with caution. This is signified by the line in the table
separating the two groups of standards.
INTRODUCTION
The purpose of this section is to provide estimates of the health
benefits associated with alternative ambient air quality standards for
particulate matter. This will be accomplished by critically evaluating
previous studies that have examined the relationship between particulate
matter and human health and using the results of the "best" of these
studies to estimate the health benefits associated with particulate matter
reductions.
As mentioned in Section 3, there are several types of studies that
have been used to analyze the relationship between air pollution and human
health. Human laboratory studies are able to control for most of the
confounding factors that influence health status and, in doing so, can
*
isolate the effect of air pollution on human health. Because air pollution
is hypothesized to be detrimental to human health, ethical considerations
prohibit extensive laboratory experimentation on humans. Another type of
study attempts to infer the susceptibility of humans to air pollution from
animal experiments. The use of animals as proxies for humans in these
experiments is also subject to limitations since animal susceptibility does
not necessarily connote human susceptibility. In the majority of cases
which test the effect of particulate matter on animals, the chemical compo-
sition of particulate matter may differ significantly from the composition
of ambient particulate matter. Differences such as these preclude the
results of laboratory experiments on animals from being directly applied to
humans. The third type of study is called an epidemiological study. An
epidemiological study concentrates on analyzing the effects of air pollu-
tion on humans in their natural environment. Although epidemiological
studies are able to avoid the problems associated with controlled human and
animal laboratory studies, they are subject to their own limitations since
4-6
-------
they are often unable to control for all of the variables that influence
health status.
The studies that will be critically evaluated in this section consist
of a subset of epidemiological studies that has generally appeared in the
economics literature.* These studies have examined both the acute and
chronic health effects of human exposure to ambient air pollution using
regression analysis applied to time series and cross-sectional data. The
economic studies employing time series data attempt to explain whether
differences in the health of people in one geographic area over time are
related to changes in ambient air quality over the same time period.
Cross-sectional analyses, on the other hand, are used to test the
hypothesis that interregional differences in human health at one point in
time can be explained by differences in ambient air quality across these
regions.
The evaluation of these studies will proceed as follows. The next
subsection will list the criteria used in selecting the studies considered
in this section. This will be followed by a brief explanation of the
measures of particulate matter used in the studies in this section.
Following this explanation, the next subsection will evaluate the studies
examining the relationship between particulate matter and human health,
where health is measured in terms of mortality. The following subsection
will concentrate on those studies that measure the impact of ambient parti—
culate matter on morbidity. After the studies have been critiqued,
evaluated and selected, the procedures that will be used to calculate the
health benefits of alternative scenarios for particulate matter will be
explained. Estimated health benefits under these alternative scenarios
will be contained in the following subsection. The final subsection will
contain a summary of the benefit estimates and any necessary qualifications
that are associated with these estimates.
* The remainder of the epidemiological literature has been considered in
Section 3 of this report.
4-7
-------
CRITERIA FOR SELECTING STUDIES
As stated in the Introduction, it is not the purpose of this report to
provide a comprehensive review of all of the studies that have examined the
relationship between human health and particulate matter; rather, it is the
intent to determine which of these studies can be used to develop reason-
able estimates of the benefits resulting from implementation of alternative
ambient air quality standards for particulate matter. In selecting the
studies that can be used for benefit estimation in this section, the
following criteria are used:
The effects of particulate matter on human health are
specifically examined — This criterion is obviously neces-
sary since the purpose of this section is to estimate the
health benefits associated with particulate matter reduc-
tions.
The study is representative — This criterion, for example,
ensures that the estimated benefits of particulate matter
reductions axe based on studies where the levels of parti-
culate matter are representative of ambient conditions.
The study attempts to control for as many factors influ-
encing health status as possible — This criterion is used
in order to minimize the possibility that the estimated
relationship observed between health status and particulate
matter results from the fact that particulate matter is
proxying for an excluded variable that has the "true"
influence on health status.
Results of the study are plausible and consistent — The
relationship observed between health status and the factors
assumed to influence health status are generally in
accordance with .a priori expectations. In addition, these
relationships are relatively consistent across alternative
specifications.
The study results are usable for the purpose of this report
— The results of the study are presented in a manner that
allows estimates of the health benefits of alternative
standards to be calculated.
4-8
-------
MEASDKEMEHT OF PARTICOLATE MATTER
The pollutant called "particulate matter" is composed of many
different elements whose distribution varies with time, region,
meteorology, and source category. Total suspended particulates (TSP), the
measure of particulate matter used in all of the studies critiqued in this
section, is a measurement of particles ranging up to 25 to 45 micrometers
(urn) in diameter without respect to the chemical composition of the
particle. A number of the studies critiqued in this section have measured
the health effects of some of the chemical components of particulate
matter. Because of the availability of data, sulfates (SO^) have been the
chemical component of particulate matter most commonly used in addition to
TSP.* Although sulfates are normally found in fine particles which are
less than 2.5 urn (1) and are therefore a part of TSP, these studies
consider sulfates to have health effects that are separate from TSP.
In this section, the economic studies that specifically use TSP and/or
SO, in order to measure the health effects of particulate matter are
critiqued. Since the control strategies being considered in the cost
analysis for implementation of alternative particulate matter standards are
not expected to affect the ambient level of sulfates, special emphasis will
be given to those studies measuring the health effects of TSP.
Two of the six standards reported in this section are stated in terms
of TSP.** The impact of these standards on human health can be estimated
directly based on the changes in TSP and the information contained in the
studies reviewed in this section. The remainder of the standards being
considered are stated in terms of PM10 (particles less than 10 urn in
aerodynamic diameter). The PM10 information cannot be used directly since
none of the health studies reviewed in this section used a PM10 measure.
* Some of the studies in this section have also considered benzene soluble
organic matter, iron, maganese, and nitrate.
** For an extended explanation of the alternative standards being con-
sidered in this report, see Section 9.
4-9
-------
Fortunately, information on the approximately equivalent levels of TSP that
will result from PM10 controls is available. This allows the health
benefits associated with the PM10 standards under consideration to be
estimated for the "TSP studies".
The application of PM10 controls may reduce both the TSP
concentrations and the fraction of small particles in the TSP that remains.
The EPA Office of Air Quality Planning and Standards (OAQPS) Staff Paper
(2) suggests that the smaller particles are more significant in producing
adverse respiratory effects. If PM10 standards lead to proportionately
larger reductions in PH10 relative to TSP, benefits for the PM10 standards
may be underestimated. Data from the cost and air quality analysis suggest
that proportionately larger reductions do not occur. These data indicate
that a comparison between PM10 and TSP standards may be valid. However,
approximations in the cost and air quality analyses are such that the
comparisons should still be interpreted with caution. (See Section 1 for
further discussion of this issue.)
MORTALITY STUDIES
Overview of the Approach
Before critiquing the individual studies that will be used in this
section to estimate the benefits of reductions in mortality resulting from
alternative particulate matter standards, it is worthwhile to critique the
approach employed throughout the existing mortality studies appearing in
the .economics literature. This critique will assess the advantages and
disadvantages of using the results of these studies to estimate the
mortality effects of exposure to particulate matter.
In general, the economic studies analyzing the effect of ambient
particulate matter on mortality attempt to estimate a concentration-
response function of the following form:
«Bii = f(Pi,Gi,Ei) (4..1)
4-10
-------
where MR. = mortality rate in region i (e.g., deaths per 100,000
people)'.
P. '= vector of personal characteristics of individuals in
region i (dietary habits, smoking patterns, occupation
mix, etc.).
G. = vector of genetic characteristics of individuals in
region i (sex, race, etc.).
E. = vector of environmental characteristics in region i
(weather, ambient air pollution, water pollution, etc.).
The unit of observation in each of these studies is an aggregated
geographic region such as a city or standard metropolitan statistical area
(SHSA). Consequently, the effect of particulate matter on the average
individual within a region is examined in this type of analysis. Because
of the use of aggregate data in these studies, they are often referred to
as macroepidemiological studies.
Both acute and chronic air pollution-induced mortality effects are
measured in these studies. Acute exposure mortality studies estimate the
effects of short-term exposure to air pollution on mortality. These
studies generally use time series data. For example, daily mortality rates
for a particular region over a certain time period are regressed on daily
air pollution measures and other variables such as daily climatological
conditions hypothesized to affect daily mortality rates. Since it is
unlikely that genetic and personal characteristics affect daily variations
in mortality rates, these variables are not included in acute mortality
studies.
Chronic exposure mortality studies estimate the effects of long-term
exposure to air pollution on mortality. These effects are generally
examined using cross—sectional data. In these analyses, the variation in
aggregate mortality rates across regions at a particular point in time
(e.g., annual mortality rates across cities) are assumed to be related to
the variation in the personal, genetic, and environmental characteristics
of the populations across these regions. The variable representing air
pollution exposure in these analyses is usually measured in terms of an
4-11
-------
annual average under the assumption that this average represents the
typical long-term exposure of the population in a region to air pollution.
To the extent that the annual average of air pollution is positively
correlated with daily air pollution levels, it should be mentioned that
these studies may measure some of the effects of acute, as well as chronic,
exposure.
These macroepidemiological studies have several advantages. Since
they examine the effect of air pollution on the general population, these
studies do not encounter the difficulty of extrapolating the results of
laboratory studies to the human population. In addition, since the ambient
level of air pollution is used in estimating these concentration— response
functions, it is not necessary to adjust the health effects observed under
laboratory conditions to those that would be observed under ambient air
quality conditions. Furthermore, the data that are generally used to
estimate "macro" concentration-response equations are available more
readily than those data that would be used to estimate a "micro-level"
concentration-response equation.
A major advantage of these studies is that the effect of a change in
the ambient level of air pollution can be quantified easily from the
concentration— response equations. Assuming that Equation (4.1) is linear,
this effect is equal to:
AMRi = Pi(APMi) (4.2)
where AMRi = the change in the mortality rate in region i resulting
from a change in the ambient level of particulate
matter.
jj^ « the partial derivative of the mortality rate with
respect to a change in the level of particulate matter.
APMj = the change in the ambient level of particulate matter in
region i.
Although the macroepidemiological concentration—response equations can
be used easily to quantify the health effects resulting from the
4-12
-------
implementation of alternative air quality standards, the results of these
studies are subject to numerous qualifications. One of the major
criticisms of any epidemiological study is that it is unable to control for
all the factors that influence mortality rates. Data on all of the genetic
and personal characteristics of the regional populations typically are not
available on an aggregate level and therefore result in incomplete specifi-
cation of the concentration—response equations. If the concentration-
response equations are estimated using Ordinary Least Squares (OLS) and the
excluded characteristics (e.g., smoking, diet, exercise) are not correlated
with the measures of air pollution, the exclusion of these variables will
not affect the estimated relationship between mortality and air pollution.
However, if these excluded variables are correlated with air pollution, the
relationship between mortality rates and air pollution estimated from the
concentration-response equation will be biased. For example, if smoking is
excluded from a mortality rate equation and people in polluted areas tend
to smoke more than people in nonpolluted areas, the coefficient of air
pollution will be "picking up" some of the effects of smoking and will
consequently be biased upward.
Another problem inherent in estimating aggregate concentration-
response equations results from the high degree of correlation that exists
among the variables hypothesized to affect mortality rates. If these
highly correlated variables are included in the concentration-response
equation, the resulting parameter estimates, although unbiased, are not
precise. Since the air pollution variables tend to exhibit a high degree
of correlation among themselves and are also highly correlated with the
"urban" variables (e.g., employment mix, average age of population) that
are usually included in the concentration—response equation, the coeffi-
cients of the air pollution variables will be imprecise. Consequently,
this diminishes the degree of confidence that can be placed in any of the
point estimates of the pollution coefficients.*
* This criticism also applies to micro-level epidemiological studies.
4-13
-------
Another problem inherent in the macroepidemiological approach results
from the type of pollution data that must be used to estimate the
concentration-response relationship. At the present time, it is impossible
to measure accurately the exposure of an aggregated population to air
pollution. Data from a monitoring station or a number of monitoring
stations within a geographic area (e.g., county, SMSA) are used as proxies
for the population's exposure to ambient air pollution, Because a signifi-
cant amount of an individual's time is s£ent indoors, these data do not
represent the 24-hour exposure of the population to air pollution.
Furthermore, these monitors are generally placed in center-city locations
which, although they tend to be in the most polluted area of a region, are
not necessarily in the most heavily populated areas. Use of a "center-
city" monitor may therefore tend to overestimate the population's exposure
to air pollution and consequently underestimate the true effect of air
pollution on the mortality rate.* Conversely, the effect of air pollution
on the mortality rate may be overestimated if the readings from the center-
city pollution monitor are correlated with an omitted urban variable that
has a positive effect on the mortality rate.
In chronic exposure macroepidemiological studies, the annual average
of some measure of pollution is used to approximate the chronic exposure of
a population. If air quality has been consistently improving over time,
the use of the annual average of air pollution in a particular year will
tend to overestimate the true relationship between air pollution and the
mortality rate. If, on the other hand, air quality has been consistently
worsening over time, the use of the current level of air pollution will
result in an underestimate of the effects of air pollution on the mortality
rate. As previously mentioned, the use of the annual average as a proxy
for chronic exposure may result in an overestimate of the true effect of
chronic exposure because the annual average is likely to be positively
* Freeman (3) has shown that although the coefficients of the air pollution
variables may be biased in this case, the regression equation will
predict the "true" change in mortality for a given change in pollution if
the relationship between monitored and true exposure is maintained after
the change in pollution.
4-14
-------
correlated with, shorter term exposures that also have a positive impact on
the mortality rate.*
The functional form of the concentration-response equation that is
used in the majority of the macroepidemiological studies is also subject to
criticism. Host studies estimate a linear concentration—response equation,
while evidence from toxicological studies suggest that the concentration-
response function may be nonlinear. If the relationship between air pollu-
tion and mortality rates is in fact nonlinear, then the estimated effect of
air pollution on the mortality rate is accurate only for the values of the
variables near the means of the sample used to estimate the concentration-
response equation. In this case, estimation of the mortality rate effects
outside the neighborhood of the means may not be accurate. This is
particularly relevant if air pollution beneath a certain level does not
affect mortality rate.
Another criticism levied against these studies is that a single
concentration-response equation is incapable of correctly modeling the
complex relationship between air pollution and mortality rates. Since
individuals may undertake actions — such as moving or seeking medical care
— in order to offset the effects of air pollution on their health, the
relationship between air pollution and the mortality rate that is estimated
by a single equation concentration—response equation may be biased and
inconsistent. If the relationship between particulate matter and health
status is estimated after this behavior has taken place, the effect of air
pollution on health status may be underestimated. Conversely, if the
relationship is estimated before this adjustment has taken place, the
effect of air pollution on health status may be overestimated.**
With the exception of two studies (4,5), macroepidemiological
mortality studies have estimated the relationship between air pollution and
* Micro-level epidemiological studies also suffer from the same problems.
** Again, micro-level studies also encounter this problem.
4-15
-------
mortality rates using a single concentration-response equation, In both of
•
these studies, a medical care variable (e.g., doctors per capita) was
included in the concentration-response equation under the hypothesis that
the mortality rate in a region is influenced by the amount of medical care
in a region. However, this relationship between the mortality rate and
medical care is also likely to work in the opposite direction; that is, the
mortality rate in a region will influence the amount of medical care
provided in a region. Since failure to account for the simultaneous
relationship between the mortality rate and medical care in these equations
will result in biased and inconsistent parameter estimates, these two
studies also estimated a medical care equation. Although these studies
have made a first attempt at estimating the complex relationship between
air pollution and the mortality rate, there is still much to be done in
order to completely model the air pollution-mortality rate relationship.
In summary, although the macroepidemiological approach has several
advantages for estimating the health effects of exposure to .air pollution,
the approach also has a number of disadvantages. Some of these disad-
+
vantages may result in an underestimate of the "true" mortality effects of
air pollution, while other of these disadvantages may result in an over-
estimate of these effects. On balance, it is difficult to determine
whether these disadvantages result in a net under- or overestimate of the
effects of air pollution on the mortality rate. In any event, they
indicate that special care should be taken when using the results of these
studies.
'"^** Rericw
As previously mentioned, epidemiological mortality studies can be
grouped into two categories: 1) acute exposure mortality, and 2) chronic
exposure mortality studies. The acute exposure mortality studies germane
to this study have appeared in the medical epidemiological literature,
4-16
-------
hence they have been discussed in Section 3.* The studies examining the
mortality rate effects of chronic exposure to particulate matter have
generally appeared in the economics literature and are therefore discussed
in this subsection.
Lave and Seskin (6) —
In 1977, Lave and Seskin published a summary of the results of their
extensive research on the effects of chronic exposure to air pollution on
mortality rates. By far, this is one of the most comprehensive macroepi-
demiological studies on air pollution completed to this date. Regression
analysis was used to estimate the relationship between air pollution and
annual mortality rates based on cross-sections of approximately 117
Standard Metropolitan Statistical Areas (SMSA) for 1960, 1961, and 1969.
The air pollution-mortality rate relationship was also estimated using
pooled cross-sectional, time series data from 1960 to 1969 for 26 SMSAs.
Measures of ambient air pollution and socioeconomic variables such as the
percent of the population aged 65 and over (2. 65), the percent poor (POOR),
the percent nonwhite (NWHITE), population (POP), and population density
"2.
(POP/M ) were the major explanatory variables included in the estimated
equations.
The two major pollutants included in the analysis were sulfates (SO^)
and total suspended particulates (TSP). Three measures of each of these
pollutants were available for the analysis — the minimum (MINS, MINP),
arithmetic mean (MEANS, MEANP), and maximum (MAZS, HASP) of 26 biweekly
readings of SO* and TSP, respectively.** These measures of pollution were
* Lave and Seskin (6) have examined the effects of exposure to acute
levels of air pollution on the mortality rates for Chicago, Denver,
Philadelphia, St. Louis, and Washington, D.C. Particulate matter was
not included in any of the concentration-response equations.
** Other researchers have found (7,8) that the sulfate data used by Lave
and Seskin in the 1960 mortality rate equations were primarily based on
data from 1957 to 1959. In addition, approximately 50 percent of this
data was based on quarterly, as opposed to biweekly, readings. With
respect to MINS, the use of quarterly data may tend to overestimate the
value of MINS and consequently underestimate its coefficient.
4-17
-------
included to test for the possibility that the mortality rate effects of a
particular pollutant might differ depending on the type of exposure. For
example, mortality rates might be highest in those SMSAs experiencing
"high" minimum pollution readings (i.e., areas that never experienced a low
level of pollution), high mean pollution readings (i.e., areas with a
relatively high average exposure), or high maximum pollution readings
(i.e., areas with the worst biweekly pollution readings during the year).*
Since these six pollution variables tended to be highly correlated, alter-
native concentration-response equations were also estimated with one
measure of each of these pollutants. MINS and MEANP were the measures used
most frequently because they were the most significant pollution variables
in the concentration-response equations which included all six pollution
variables.
In general, the pollution variables were positive and significantly
related to the SMSA mortality rates. Table 4-2 shows the results of the
unadjusted and age-sex-race-adjusted mortality rate equations estimated for
1960 and 1969.** As can be seen in Equations 4-2.1 and 4-2.3 of the table,
when all six pollution variables are included in the unadjusted regression
equations, the coefficients of these variables are not significant at the 5
percent level. In fact, the signs of some of the pollution variables are
negative. Because of the high correlation among these variables, this is
not surprising. When only HINS and MEANP are included in the total
mortality rate equations, as in Equations 4-2.2 and 4-2.4, they are both
positive and significantly different from zero. It is interesting to note
* There is a question as to whether the inclusion of the minimum pollution
reading was necessary since the mean reading may be indicative of
chronic exposure while the maximum reading may be indicative of acute
exposure.
** The mortality rate equation for 1961 is not reported here since two-
thirds of the sulfate data and one-third of the suspended particulate
data used in the 1961 equation were based on 1960 data. MINS was
insignificant in this equation while MEANP remained significant. The
sum of the elasticities of the pollution variables remained relatively
unchanged from the 1960 equation.
4-18
-------
Table 4-2
1960 AND 1969 UNADJUSTED AND AGE-SEX-RACE-AD JUSTED
MORTALITY RATE EQUATIONS*
R2
Constant
Air pollution variables
MINS
MEANS
MAXS
Sum S elasticities
MIHP
MEAHP
MAX?
Sum P elasticities
Socioeconomic variables
POP/M2
> 65
NWHITE
POOR
Sum SE elasticities
Log POP
Unadjusted
1960
4-2.1
0.831
343.381
4.73
(1.67)
1.73
(0.53)
0.28
(0.25)
0.050
0.199
(0.32)
0.303
(0.71)
-0.018
(-0.19) '
0.044
0.008
(1-54)
68.8
(16.63)
3.96
(3.82)
0.38
(0.26)
0.70
-27.6
(-1.38)
4-2.2
0.828
301.205
6.31
(2.71)
0.033
0.452
(2.67)
0.059
0.008
(1.71)
70.28
(18.09)
4.22
(4.32)
-0.02
(-0.02)
0.71
-21.2
(-1.12)
1969
4-2.3
0.817
386.858
-0.38
(-0.07)
6.33
(1-81)
-0.53
(-0.67)
0.059
0.434
(0.69)
0.056
(0.13)
0.130
(1.83)
0.056
0.013
(2.51)
64.03
(17.11)
2.04
(2.24)
5.11
(2.13)
0.729
-42.7
(-2.22)
4-2.4
0.305
330.647
7.74
(2.11)
0.030
0.818
(3.39)
0.087
0.013
(2.54)
65.68
(18.09)
2.04
(2.27)
5.57
(2.29)
0.75
-36.5
(-1.94)
Age-sex-race-adjusted
1960
4-2.5
0.285
901.196
6.78
(2.29)
0.85
(0.25) '
0.76
(0.66)
0.057
0.199
(0.31)
0.332
(0.75)
-0.030
(-0.29)
0.040
0.005
(0.92)
2.65
(0.61)
1.45
(1.34)
1.39
(0.89)
0.068
-9.3
(-0.44)
4-2.6
0.272
857.665
8.25
(3.38)
0.038
0.465
(2.61)
0.054
0.006
(1.01)
4.11
(1-01)
1.65
(1.61)
0.98
(0.65)
0.076
-2.2
(-0.11)
1969
4-2.7
0.390
975.700
0.49
(0.11)
6.18
(2.05)
-0.77
(-1.15)
0.050
0.329
(0.61)
0.113
(0.31)
0.109
(1.79)
0.050
0.009
(2.08)
1.64
(0.51)
1.07
(1.37)
5.99
(2.89)
0.093
-35.1
(-2.12)
4-2.8
0.348
918.657
7.84
(2.48)
0.028
0.723
(3.48)
0.072
0.009
(2.07)
3.15
(1.00)
1.09
(1.40)
6.40
(3.06)
0.111
-28.3
(-1.74)
* 1960 regressions are based on data for 117 SMSAs, while 1969 regressions are based on data for 112 SMSAs. The
numbers in parentheses below the regression coefficients are t-statistlcs. Mortality rates are expressed in
terms of deaths per 100,000.
Source: Lave and Seskin (6), p. 121. In reporting these results, all
scaling faactors used by Lave and Seskin have been removed. Con-
sequently, the coefficients reported here represent the relation-
ship between the mortality rate and the unsealed variables.
4-19
-------
that although, the elasticity* of HINS remains relatively constant in the
1960 and 1969 regressions, the elasticity of MEAN? is much higher in 1969
than in 1960.
The results of the basic age-sex-race-adjusted mortality rate
equations are reported in Equations 4-2.5 to 4-2.8 of Table 4-2. Age-sex-
race-adjusted mortality rates were used as the dependent variables in these
equations in order to further control for the various factors (age, sex,
race) affecting the SHSA mortality rates.** As Table 4-2 shows, the
elasticities of the pollution variable in the age-sex-race-adjusted
equations are similar to the unadjusted mortality equations indicating that
the unadjusted mortality rate equations appear to sufficiently control for
the age, sex, and racial composition of the population at risk.
In order to test the stability of the air pollution coefficients, Lave
and Seskin estimated many other mortality rate equations. These included
log-linear, quadratic, and linear spline equations' to test for the possi-
bility of a nonlinear relationship between air pollution and mortality
rates; a jackknife analysis to test for the sensitivity of the pollution
coefficients to extreme observations; and dummy variables to look for
systematic effects on the pollution coefficients by regions of the country.
The results obtained for the jackknife analysis were quite similar to the
results obtained for the entire sample, indicating that extreme observa-
tions were not affecting the air pollution coefficients. The inclusion of
the dummy variables reduced the significance of the sulfate variable, but
* The elasticity is a measure of the percentage change in the dependent
variable that can be expected from a percentage change in an independent
variable. In this section, it will be used to represent the percentage
change in the mortality rate resulting from a one percent change in an
independent variable. Since the elasticity is a dimensionless number,
it will be used to compare the health effects of particulate matter that
have been estimated in the studies critiqued in this section.
** The adjusted mortality rate was calculated based on the assumption that
each SMSA had demographic characteristics identical to those of the
entire United States. See Lave and Seskin (6), pp. 346-347 for a
further explanation of this adjustment.
4-20
-------
did not affect the coefficient of TSP. Although the equations that were
estimated to test for the presence of nonlinear!ties between air pollution
and mortality rates (e.g., linear spline, dummy variables, and splitting
the sample based on air pollution levels) did indicate that the linear
specification was not superior to a nonlinear specification. Lave and
Seskin concluded that linearity could not be rejected and chose to estimate
the majority of their mortality rate equations in linear form. However, it
should be noted that these results suggest that the relationship between
mortality rates and air pollution may be nonlinear.
Because the socioeconomic variables included in the basic mortality
rate equations might not have been sufficient to control for the socio-
economic characteristics of the SHSAs, Lave and Seskin also estimated
separate mortality rate equations for specific age, sex, and race
categories for both 1960 and 1969. In these equations, it was found that
both MINS and MEANP were more closely associated with mortality rates among
nonwhites than whites and that the estimated effect of these pollutants
increased with age. The estimated coefficients of the air pollution
variables, however, were not significant in all of the age-sex-race-
specific equations.
The relationship between air pollution and mortality rates was consid-
erably weakened when disease-specific mortality rate equations for 1960 and
1961 were estimated.* Sulfates had a significant effect (i.e., t-statistic
greater than 1.96) on the mortality rates for total cancer, digestive
cancers, endocarditis, cardiovascular, heart, and hypertensive diseases.
TSP was a significant explanatory variable in the disease-specific
mortality rate equations for tuberculosis and asthma, and approached
statistical significance (i.e., t-statistic greater than 1.64) in the
* Disease-specific mortality rate equations were not estimated for 1969.
The disease-specific mortality rates that were examined in the study
were: total cancers and specific types of cancers (buccal, pharyngial.
digestive, respiratory, and breast), total cardiovascular disease, heart
disease, endocarditis, hypertensive disease, respiratory disease, tuber-
culosis, asthma, influenza, pneumonia, and bronchitis.
4-21
-------
mortality fate equations for total cardiovascular disease and endocarditis.
•
Surprisingly, neither one of the pollution variables had a significant
effect on the mortality rates for influenza, pneumonia, and bronchitis.
One reason that was given for the poor performance of the pollution
variables was that the number of deaths in an SMSA from a specific disease
might have been too small to isolate the effect of air pollution on the
disease-specific mortality rate.
Additional socioeconomic variables were added to the basic 1960
mortality rate equations to see if the air pollution variables were
prozying for omitted socioeconomic variables. These variables reflected
the SHSA's occupation mix, climate, and home-heating characteristics. The
addition of the occupation variables tended to reduce greatly the size and
significance of the coefficients of the sulfate variable. The elasticity
of MINS in the mortality rate equations including the occupation variables
was 0.012 as compared to an elasticity of 0.033 in the basic unadjusted
mortality equation (Equation 4-2.2 of Table 4-2). The elasticity of MEANP
remained significant but decreased slightly from 0.059 to 0.041. The
coefficients of the socioeconomic variables also changed when the occupa-
tion mix variables were included. As Lave and Seskin state, these results
are not unexpected since occupation mix is likely to be closely associated
with the socioeconomic structure of an area and hence with air pollution.
However, it does raise the question of whether the sulfate variable can be
considered to be a proxy for the occupation mix variables or vice versa.
The addition of the climate variables to the basic mortality rate
equation did cause MINS to become insignificant but did not affect the
significance level of MEANP. The elasticities of MINS and MEANP in this
equation were 0.021 and 0.049, respectively. When variables reflecting
home-heating fuel were added to the basic mortality rate equation, both
pollution variables were reduced greatly in size and became insignificant.
The elasticities of MINS and MEANP were 0.0095 and 0.022, respectively.
Although this result tends to diminish the degree of confidence that one
can place in the estimated relationship between mortality rates and air
pollution, the lack of significance of the pollution variables may result
4-22
-------
from the correlation that exists bet-ween home heating fuel and air pollu-
tion. It seems likely that the type of fuels used to heat homes may have
an impact on the level of air pollution in a geographic region. Conse-
quently, the variables reflecting home heating fuel may be a better
indicator of the exposure of a population to certain types of ambient or
indoor air pollution.
Besides the general criticisms of the macroepidemiological approach
previously discussed, the Lave and Seskin study has been criticized with
respect to a number of other issues. Both Smith (9) and the Criteria
Document for Sulfur Oxides and Particulate Matter (10) have criticized the
decision rules used by Lave and Seskin (i.e., retaining only those pollu-
tion variables whose coefficients were positive and exceeded their standard
errors with the further requirement that at least one of each pollution
variable be retained). In order to examine the appropriateness of these
decision rules in estimating mortality rate equations, Smith (9) estimated
a mortality rate equation for 50 SMSAs using 1968 and 1969 data. Utilizing
SMS A data on total suspended particulates, percent of population over 65,
income per capita and alternative sets of independent variables (e.g.,
socioeconomic data on percent nonwhite and population density), 36
mortality rate equations were specified. None of the 36 equations passed
the test for normality of the error terms indicating that significance
tests for the coefficients of the variables in these equations were
suspect. In a number of the equations, the presence of heteroskedasticity
(nonconstant variance of the regression's error term) was indicated.
Although adjustments for heteroskedasticity did not appreciably affect the
magnitude of the TSP coefficients, the t-statistics of these coefficients
were diminished, indicating that the relationship between TSP and mortality
rates may not be significant.
The Smith study suggests that specification errors may be a problem in
estimating the relationship between air pollution and mortality rates, and
special care should be taken in selecting specifications based on the
significance of the air pollution coefficients.
4-23
-------
Another criticism of the Lave and Sestin study involves the inclusion
1 •
of atypical SHSAs in the regression equation and the analysis of the
residuals of the mortality rate equations. Thibodeau e_t aj,. (11) tested
the sensitivity of the Lave and Seskin air pollution coefficients to SMSAs
that they considered to be outliers.* They observed that when these SMSAs
were omitted from the basic I960 regression equation with six pollution
variables (Equation 4-2.1), the estimated air pollution coefficients varied
significantly. When these SHSAs were excluded from the mortality rate
equation which included only two pollution variables (Equation 4-2.2), the
coefficients of these pollution variables were relatively unchanged.
Thibodeau et al. also examined the residuals of the basic regression
equation and, in addition to those SMSAs considered to be outliers,
excluded three SMSAs whose residuals were widely separated from the other
SMSAs in the data set. Again, the estimated coefficients of the six air
pollution variables varied significantly when these SMSAs were excluded,
while the pollution coefficients of the equation containing only two pollu-
tion variables were relatively unchanged.
Given the high collinearity among the air pollution variables, the
results obtained by Thibodeau et al. are not unexpected. Since the
collinearity among these variables precludes estimating the parameters with
precision or confidence, one would expect that as alternative equations are
estimated, the point estimates of these parameters may change signifi-
cantly. A better indication of the stability of the air pollution coeffi-
cients is provided by the equations- including the two pollution variables,
MINS and MEANP, as indices of the effect of pollution on the mortality
rate. In these equations, the coefficients of these pollution variables
remain relatively stable when unusual observations are excluded from the
regression equation. This suggests that the coefficients of these
variables are not sensitive to these unusual observations.
* Recall that Lave and Seskin's use of jackknife analysis to test the
sensitivity of the pollution coefficients to extreme observations did not
reveal that the coefficients were sensitive to these observations.
4-24
-------
Additional criticisms of the Lave and Seskin study have centered
around the failure of the study to account for the smoking and dietary
habits of the population at risk, the use of SMSA data as the unit of
observation, and the appropriateness of a single linear concentration-
response equation. Some macroepidemiological studies have attempted to
address these issues. In fact. Chappie and Lave (5) provide a reanalysis
of the Lave and Seskin 1977 analysis that specifically takes these
criticisms into account. That study is discussed below.
The remaining macroepidemiological studies considered in this section
will now be discussed in chronological order. Since the purpose of this
section is to estimate the health effects associated with reductions in the
level of particulate matter, specific attention will be given to the
comparability of the estimated relationship between the measures of parti-
culate matter and mortality rates across all of these studies.
Koshal and Koshal (12) —
Koshal and Koshal estimated log-linear mortality rate equations based
2
on data for 40 cities. Data on the annual arithmetic means in ug/m from
1960 to 1967 for total suspended particulates and benzene soluble organic
matter (a component of particulate matter) were used as explanatory
variables in the mortality rate equation. The other independent variables
included in the specification were the city's population density, percent
of nonwhite population, percent of population aged 65 and over, and the
annual average percentage of days with sunshine.
In the equation where 1967 mortality rates were regressed against the
1967 pollution levels and the other socioeconomic variables, all of the
signs of the coefficients of the independent variables, except benzene
soluble organic matter (BSO), were in accordance with a priori expecta-
tions. Since BSO is a component of particulate matter, high correlation
between these two variables is to be expected [correlation between log(TSP)
and log(BSO) was 0.57]. Hence, the sign of the estimated coefficient of
BSO is not surprising. When Koshal and Koshal re-estimated the mortality
4-25
-------
equation without the benzene variable, the coefficient of TSP was rela-
tively unchanged. The elasticity of the mortality rate with respect to TSP
estimated in this equation was 0.176. This elasticity is significantly
higher than the elasticity (evaluated at the means) of 0.022 to 0.087
estimated by Lave and Seskin.
One possible reason for this discrepancy is that the Lave and Seskin
estimate is based on pollution data that generally represented the worst
air quality in the SMS A. If the distribution of TSP varies significantly
throughout the SMSA, then their use of the worst air quality data may tend
to overestimate the TSP exposure of the population within the SMSA and
hence underestimate the effects of TSP on the mortality rate. If the level
of TSP tends to be uniformly distributed throughout a city, the use of
city-level data may provide a more representative estimate of city popula-
tion exposure. Therefore, the Koshal and Koshal analysis may provide a
better estimate of the effects of TSP on mortality rates. However, not
much confidence can be placed in the Koshal and Koshal results since the
sensitivity of the results to alternative specifications was not investi-
gated.
Gregor (13) —
•Gregor estimated a mortality rate equation based on census tract
information for Allegheny County, Pennsylvania. Although this study has
the advantage that it is able to match closely the population at risk to
the ambient level of air pollution, the use of census tract data seriously
limits the ability of the study to control for migration. Consequently,
the estimated relationship between TSP and mortality rates may be biased
downward if people have moved from "clean" census tracts to "dirty" census
tracts or biased in the opposite direction if the movement has been from
dirty to clean census tracts.
Because the number of deaths in a census tract in any year might be
relatively small or zero, Gregor's dependent variable was a 5-year average
mortality rate from 1968 to 1972. The pollution variables included in the
4-26
-------
analysis were the 5-year annual arithmetic averages of SOj in PFT/24 hours
and TSP in (ig/m . A measure of dustfall was considered initially, but was
excluded from the final specifications due to lack of statistical signifi-
cance. Besides the air pollution variables, the independent variables
controlled for in the final specifications were the percent of adult popu-
lation with a high school education, the number of days with precipitation
exceeding 0.1 inch, the number of days in which the maximum temperature did
not exceed 32 degrees Farenheit, and population density. In order to
control for the different effects that pollution might have on mortality
rates, mortality rate equations were estimated for specific age groups of
white males and females. The mortality rate equations were broken down
further into pollution-related and nonpollution-related deaths.
Using weighted regression analysis in order to correct for heteroske-
dasticity, the results indicated that TSP had a significant effect on
mortality rates. The effects were more pronounced for white men than for
white women and the effect appeared to increase with age. The coefficient
of SO*, although generally plausibly signed, was not significant.
The elasticity of TSP was quite high; ranging from 0.23 to 0.89 in the
pollution-related mortality rate equations. It is interesting to note,
however, that the TSP elasticity estimated from the nonpollution-related
mortality equations ranged from -0.10 (not significant) to 0.53 (signifi-
cant at 0.01 level). For white males and females aged 45 and over, these
elasticities ranged from 0.21 to 0.53 and were significant. This result is
disturbing since it indicates that TSP may be proxying for some excluded
variable that may affect all deaths.
Lave and Seskin have also estimated age-sex-race-specific mortality
rate equations for two age groups that are comparable to Gregor's mortality
rate equations. However, these equations are not broken down by cause of
death. Table 4-3 compares the elasticities obtained for these age-race-
sex-specific mortality rate equations. As can be seen in the table, all of
the elasticities reported by Lave and Seskin are substantially smaller than
those reported by Gregor. In fact, four of the eight elasticities reported
4-27
-------
Table 4-3
COMPARISON OF TSP ELASTICITIES FROM LAVE AND SESKIN (6)
AND GREGOR (13)
Total White Mortality
Rates
Rate
Age : 45-64
Male
Female
Age: 65 and over
Male
Female
Lave and Seskin
1960 MR*
0.041**
0.095
0.026**
0.015**
1969 MR*
0.084
0.051
0.034**
0.067
Gregor
0.40
0.50
0.23
0.31
* MR » mortality rate.
** Not significant at 0.10 level of tiro-tailed test.
from the Lave and Seskin study are not significant at the 0.10 level.* A
number of explanations may be given for the wide discrepancy be tire en the
elasticities reported by Lave and Seskin and Gregor. The first reason
involves the unit of observation used to estimate the relationship between
air pollution and mortality rates. Since census tract data may provide a
better matching of the ambient levels of pollution to the population at
risk, the Gregor analysis may give a better indication of the relationship
between TSP and mortality rates than studies based on larger geographic
areas. On the other hand, the Gregor analysis may tend to overestimate the
effect of TSP on mortality rates since individuals may spend a significant
* Lave and Seskin state that one of the reasons for the relatively poor
performance of the air pollution variables in their age-race-sex-specific
mortality rate equations is that this disaggregation significantly
reduces the size of the population at risk and therefore impairs their
ability to estimate accurately the relationship between air pollution and
mortality rates. Gregor circumvents this problem to some extent by
considering the census tract mortality rate over a 5-year period.
4-28
-------
portion of their time away from the census tract. For example, if indivi-
duals typically reside in "clean" census tracts and work in "dirty" census
tracts, the estimation of a mortality rate equation based on residential
TSP and mortality rates may tend to underestimate exposure and overestimate
the "true" relationship between TSP and mortality rates.
Another reason for the disparity may result from the fact that unlike
Lave and Seskin's analysis, Gregor's analysis does not specifically include
sulfates as an explanatory variable in the mortality rate equation. Since
TSP and sulfates tend to be correlated, the TSP variable in Gregor's
analysis may be capturing some of the effects of sulfates.
The most likely explanation for the disparity between these estimates,
however, is that Gregor's analysis is based on a cross-section of census
tracts within Allegheny County, while the Lave and Seskin study is based on
a cross-section of SMSAs across the United States. Since a significant
amount of the economic activity in Allegheny County is in the coal and
steel industries, the pattern of long-term exposure and the composition of
»
the particulate matter may be significantly different than that experienced
elsewhere in the country. Consequently, the relationship observed between
TSP and mortality rates for Allegheny County may not be representative of
other areas.
Lipfext (14-16) —
To test whether the air pollution-mortality rate relationship would be
better specified with city data, in 1977 Lipfert (14) estimated aggregate
mortality rate equations based on data for 60 cities and their corres-
ponding SHSAs. Air pollution and mortality data from 1969 were used. A
linear mortality rate equation was estimated which controlled for percent
of families living beneath the poverty level, birth rate, and the percent
of housing built before 1950, in addition to the basic socioeconomic
variables considered by Lave and Seskin. The TSP pollution variable con-
sidered in the analysis was measured in terms of the annual geometric mean
a
in jig/m . S02» SO^, iron, manganese, and benzo(a) pyrene were also
4-29
-------
included as pollution, variables in various equations. TSP displayed a
consistently significant and positive effect in the various equations that
were estimated. The minimum level of SO, did not appear to be signifi-
cantly related to the mortality rate and hence was not included in all of
the mortality rate equations. The elasticity of TSP ranged from 0.048 to
0.076 in the SMSA mortality rate equations. The range of elasticities
based on the city mortality rate equations was somewhat closer: 0.063 to
0.068. Both of these groups of elasticities are consistent with the
findings of Lave and Seskin.
In a study prepared for the National Commission on Air Quality,
Lipfert (15) reanalyzed the findings of Lave and Seskin in detail. Because
of the problems with the 1960-1961 sulfate data used by Lave and Seskin,*
this reanalysis concentrated on the mortality rate equations estimated with
1969 and 1970 data. The reanalysis focused on three points: 1) the
replacement of the minimum sulfate variable with a measure of the mean
sulfate level, 2) the effect of including additional socioeconomic and
pollution variables, and 3) the existence of nonlinear relationships
between mortality rates and sulfates and total suspended particulates
(TSP).
The replacement of the minimum level of sulfates with the mean level
resulted in an increase in the significance of the sulfate variable and a
decrease in the magnitude and significance of the TSP variable. The
elasticity of the TSP variable in this specification was equal to 0.05,
while the elasticity of sulfate was equal to 0.06.
In testing the effect of additional socioeconomic and pollution
variables on the relationship between mortality rates and the particulate
and sulfate variables, a stepwise regression strategy was used. This
involved maximizing R at each step by inserting or deleting the explana-
tory variables iteratively. For the unadjusted mortality rate equations,
* Recall that the majority of the 1960 sulfate data were primarily based on
data from 1957 to 1959. In addition, 50 percent of this data were based
on quarterly, as opposed to biweekly, data.
4-30
-------
the independent variables included a proxy for cigarette smoking (measured
by state cigarette sales adjusted for differences in state taxes*), water
quality (measured by dissolved water solids), migration into an SMSA,
residential use of heating fuels that tend to be of a more polluting nature
(i.e., coal and wood), and the basic socioeconomic variables used by Lave
and Seskin (percent of population aged 65 and over, percent nonwhite,
percent poor, and population density). The pollution variables included in
this specification were ozone and TSP.** The elasticity of the smoking
variable was 0.201, while the elasticities of the pollution variables were
0.048 and 0.080, respectively. Sulfates were not included due to lack of
significance. These equations showed that the inclusion of other variables
such as the smoking proxy did not significantly affect the relationship
between TSP and mortality rates.
Age-sex-specific mortality rate equations were also estimated using
the same strategy. "Net" TSP was a significant explanatory variable in the
mortality rate equations for males and females under the age of 65. The
elasticity of TSP in these equations was quite high and equal to approxi-
mately 0.12 for both sexes. TSP did not have a significant effect on the
mortality rate for those persons over age 65. This is in direct contrast
to other studies that have found that the effects of TSP increase with age.
It is interesting to note that, except for the mortality rate of females
over the age of 65, sulfates did not have a significant effect on
mortality rates.
Finally, Lipfert found that a nonlinear relationship between TSP and
mortality rates was suggested. This corroborates with some of the toxico—
logical evidence suggesting that a nonlinear relationship exists between
TSP and mortality rates.
* It is questionable whether st*ate cigarette sales are an appropriate
proxy for SMSA consumption of cigarettes.
** The TSP variable used in this regression was a "net" measure of TSP. It
was obtained by "netting out" the sulfate component of TSP.
4-31
-------
Lipfert's 1980 study (16) was based exclusively on city data.
Measures of cigarette consumption (i.e., state cigarette sales adjusted for
differences in taxes) and education (i.e., percent of population with a
college education) are examples of the types of independent variables
included in the 1980 study. Unlike the 1977 analysis, both 1969 and the
average of 1969 to 1971 pollution and mortality rate data were used to
estimate the mortality rate equations. Total, disease-specific and age-
specific mortality rate equations were estimated.* Cigarette sales was
generally a significant explanatory variable. The coefficient of the
annual average of SO. was inconsistently signed and insignificant. TSP was
significant in the total mortality rate equations (1969 and the average of
1969 to 1971), but was not significant in the two respiratory disease
mortality rate equations that were estimated. Like the Lave and Seskin
study, it is possible that this results from the small number of deaths in
the respiratory disease categories. TSP was also not significant in any of
the estimated age-specific mortality rate equations. Although these
results imply that TSP may not be an important variable in explaining
mortality rates, the insignificance of TSP may result from the inclusion of
manganese, a component of particulate matter, in the mortality rate
equations. Manganese was consistently positive and generally significant
across all of the mortality rate equations that were estimated. Lipfert
has stated that manganese might be a surrogate for occupational exposure.
The elasticities of the TSP variable could not be computed from the
information reported in the study. By assuming, however, that the average
level of TSP and the average mortality rate were similar to those reported
in Lipfert's 1977 study (14), the elasticity of TSP in the total mortality
rate equations appear to be similar to those reported in the 1977 study.
Mendelsohn and Ozeutt (17) —
Mendelsohn and Orcutt estimated linear mortality rate equations based
on data from individual death certificates in 1970, the Public Use Sample,
* In this study, "total" TSP rather than TSP "net" of sulfates was used.
4-32
-------
and 1974 air quality data.* These equations were estimated for county
groups.** This analysis is unique because some of the data used in
estimating the mortality rate equations are based on individual information
and therefore avoids some of the problems of using aggregate data.
Weighted regression analysis was used to estimate the relationship between
illness-related mortality rate for Caucasians and ambient exposure to
sulfates, nitrates, sulfur dioxide, nitrogen dioxide, carbon monoxide,
total suspended particulates, and ozone. Age, income, the average number
of years of schooling, the number of children, and the net migration rate
were some of the socioeconomic variables included in the analysis. Results
were reported for both white males and white females from the ages of 45 to
64 and indicated that the annual average of sulfates had the most signifi-
cant impact on mortality rate. The elasticity of the sulfate variable
ranged from 0.067 to 0.116. Particulates, also measured in terms of the
annual mean, were generally not significant and were sometimes implausibly
signed. The TSP elasticity ranged from -0.105 to 0.036.+ This result is
in direct contrast to Lipfert '(13-15) who found that TSP was more important
than sulfates in explaining variations in the mortality rate.
One possible reason for the lack of significance of the particulate
matter variable is the fact that both urban and rural populations are
included in the Mendelsohn and Orcutt analysis. Since much of the particu-
late matter in rural areas comes from agricultural instead of industrial
operations, the composition of the particulate matter may be significantly
different across these populations, and thus may result in a confounding of
* Mortality rate equations were also estimated using 1970 air quality
data. The results of these regressions were not reported because they
were based on a smaller number of observations due to the lack of air
quality data for 1970. Mendelsohn and Orcutt state that "the coeffi-
cients for the two (sets of) regressions are similar, but the 1974
coefficients are slightly more significant."
** The U.S. Census aggregates the 3,000-odd counties in the United States
into 408 county groups.
+ The coefficients of nitrate, nitrogen dioxide, and ozone were generally
implausibly signed. Nitrate and ozone were frequently negative and
significant.
4-33
-------
the effects. In addition, the inclusion of seven pollution variables in
the mortality rate equation may reduce the chance for statistical
significance if these variables are correlated.
Seneca and Asch (18) —
County-level data were used by Seneca and Asch to estimate the
relationship between total mortality rates and air pollution, as measured
by TSP and SO- (annual geometric averages in ug/m ), in the state of New
Jersey. Both cross-section and cross-section time series were used to
estimate a linear mortality rate equation which controlled for socio-
economic variables such as the percent of county population aged 65 and
over, the percentage of nonwhites in the county, population density, the
percentage of workers employed in manufacturing, and median income.
Questions can be raised regarding the appropriateness of using median
income as an explanatory variable in the mortality rate equation because of
the simultaneous relationship that might exist between income and mortality
rates. Income may be a proxy for the standard of living and thus may have
a negative impact on the mortality rate. On the other hand, the mortality
rate may be a proxy for the incidence of illness and thus may have a
negative effect on income.
In the cross-section equations where both TSP and SO- were entered as
explanatory variables, the annual geometric mean of TSP was positively
related to the county mortality rate. In the majority of the equations
that were specified, the coefficient of TSP was significant. The elasti-
city of TSP in these equations was somewhat higher than that reported by
Lave and Seskin, and ranged from 0.115 to 0.142. The relationship between
SOj and mortality was less stable across alternative specifications and the
elasticity of S02 was significantly lower, ranging from 0.0063 to 0.022.
In order to see whether omitted variables were biasing the air pollu-
tion coefficients, Seneca and Asch also estimated cross-sectional time
series equations. In these equations, the percentage change in the
4-34
-------
mortality rate was regressed against the percentage change in the socio-
•
economic and pollution variables. Under the assumption that it was
unlikely that smoking, dietary, and genetic characteristics would
systematically change over the time period under consideration
(approximately 10 years from 1967 to 1977*), these variables would not
influence the percentage change in mortality rates during this period and
hence were not included in the cross-sectional time series equations.
When both the percentage change in TSP and SO. were included as
explanatory variables in these equations, neither one was significantly
related to the mortality rate. The high correlation between TSP and SO, (r
= 0.59) was given as the reason for the lack of significance of the pollu-
tion variables in these equations and these equations were not reported in
the study. When either TSP or SO- was dropped from the mortality rate
equations, the remaining pollution variable was generally positive and
significant. The elasticity of TSP was higher in the "TSP equations",
ranging from 0.109 to 0.152. However, it should be mentioned that these
elasticities are probably biased upward since the coefficient of TSP in
these equations is probably picking up the effects of SO*.
Crocker .et ml. (4) —
i
A data set of 60 cities was used by Crocker et al. to examine the
relationship between mortality rates and air pollution. The 1970 mortality
rate for each city was regressed against a set of socioeconomic, dietary,
and environmental variables. The socioeconomic variables included in the
mortality rate equation were the 1970 values of median age (MAGE), percent
of the population that was nonwhite (NW), percent of the population living
in homes where there was more than 1.5 persons per room (CROWD) and medical
doctors per capita (a proxy for medical care). Data on 1955 and 1965
dietary patterns were used to develop indices of food consumption.
Variables representing consumption of total protein (PRO), animal fat
* The exact time period considered in this analysis differed from county to
county due to the unavailability of pollution data.
4-35
-------
(FAT), and carbohydrates (CARS) were ultimately used in the concentration-
response equation. Each city's per capita cigarette consumption was
approximated from cigarette tax revenues in the state in which the city was
located. The environmental variables included in the estimation were the
annual geometric means of NOj, SOj, and TSP and the number of days in the
year where the temperature was less than 0 degrees Celsius (COLO).
This study is unique in two respects. First, it is the first macro-
epidemiological study to include dietary variables in the concentration-
response equation. Second, it is also the first study to specifically
account for the fact that individuals may offset the effects of air pollu-
tion on their health by seeking additional medical care.
Because of the simultaneous relationship that exists between the
mortality rate and doctors per capita (i.e., doctors per capita are
expected to have a negative influence on the mortality rate and, at the
same time, the mortality rate, because of its relation to illness, is
expected to have a positive influence on the number of doctors locating in
a particular city), Crocker et al. used a two—stage least-squares estima-
tion technique.* In the first stage, a reduced form medical care equation
was estimated with medical doctors per capita (HD) as- the dependent
variable. In the second stage, the mortality rate equation was estimated
with the predicted value of HD used in place of the actual values of MD.
The results of these estimates are provided in Table 4-4.
As can be seen in the table, the explanatory variables in the
mortality rate equation that have generally not appeared in other macroepi-
demiological studies (i.e., CIG, MD, PRO, FAT, and CARS) appear to play an
important role in explaining the mortality rate. PRO, which was correlated
with the index of cholesterol consumption (r = 0.67), had a significant
* Estimation of this mortality rate model by ordinary least squares (OLS)
would have resulted in biased and inconsistent parameter estimates. The
parameters estimated by the two-stage least-squares estimation technique
are consistent but not necessarily unbiased. For a further explanation
of this technique, see Pindyck and Rubinfeld (19).
4-36
-------
Table 4-4
•
REDUCED FORM MEDICAL CARE AND TOTAL MORTALITY RATE EQUATIONS
FROM CROCKER ET AL. (4)
Reduced Form Equation (dependent variable is medical doctors per capita):
CONSTANT
NW
MAGE
INCOME
EDUCATION
CROWD
COLD
CIG
PRO
GARB
FAT
R2
Coefficient
-1,691.3
50.447
1.351
0.616E-02
1.940
161.53
-0.128
0.458
0.223E-01
0.228E-02
0.240E-01
0.388
t-statistic
-2.712
1.102
0.504
0.867
1.342
0.261
-0.539
1.492
1.414
0.728
1.940
Total Mortality Rate Equation:
CONSTANT
ld>*
NW
MAGE
CROWD
COLD
CIG
FRO
CARB
FAT
NO,
SO-
TSP
R2
Coefficient
-79.296
-0.528E-01
5.628
0.659
31.772
0.144E-01
0.220E-01
0.192E-02
-0.794E-04
0.398E-03
1.646
-0.313E-02
0.107E-02
0.821
t-statistic
-3.512
-4.349
,562
4.
11.540
2.347
.909
.812
.552
-1.361
1.451
0.358
-0.349
0.201
2.
2.
3,
* MD is the value of medical doctors per capita predicted from the reduced
form medical care equation.
4-37
-------
positive impact on the mortality rate. FAT and CARS, although, plausibly
signed, were not significant. CI6 had a positive and significant impact on
A
the mortality rate. The sign of the medical doctors variable, MD, was in
accordance with .a priori expectations and highly significant. It should be
mentioned that when the mortality rate equation was estimated with the
actual values of MD, MD was not significant. This result implies that it
is extremely important to correctly specify the mortality rate equation.
The most interesting outcome of the Crocker ^t jl. analysis is that
the estimated relationships between the air pollution variables and the
total mortality rate were not significantly different from zero. In order
to test the possibility that the pollutants were significantly related to
certain diseases, Crocker et al. also estimated disease—specific mortality
rate equations.* These equations suggested that TSP had a significant
positive impact on the pneumonia and influenza mortality rate, while sulfur
dioxide had a significant impact on the mortality rate for early infant
disease. '
The results obtained by Crocker et al. are extremely disturbing since
they are in direct contrast to the Lave and Seskin study and other macro-
epidemiological studies that have found that 1) particulate matter has a
significant positive impact on the total mortality rate, and 2) particu-
late matter does not appear to be significantly related to pneumonia
mortality rates. Crocker et al. state that one possible reason for the
disparity between their results and the results obtained in other macroepi-
demiological studies may be due to the negative association between medical
doctors per capita and pollution that they observed when they re-estimated
the reduced form equation while including the air pollution variables.
This association implies that medical doctors may choose not to locate in
polluted areas. Consequently, the positive association observed between
air pollution and mortality rates in many studies may result from the
* Disease-specific mortality rate equations were estimated for vascular,
heart, pneumonia and influenza, emphysema and bronchitis, cirrhosis,
kidney, congenital birth defects, early infant, and cancer-related
deaths.
4-38
-------
failure of these studies to include medical care as an explanatory variable
in the mortality equation. In other words, the positive relationship
observed between air pollution and mortality rates may not be the result of
air pollution having a positive impact on mortality rates, but rather, may
result from the fact that medical doctors, who have a negative influence on
mortality rates and choose not to live in polluted areas, are excluded from
the mortality rate equation. Because of the correlation observed between
medical doctors per capita and air pollution, it is difficult to say
whether the insignificance of the pollution variables results from the fact
that medical doctors have the "true" impact on mortality rates or that the
inclusion of the highly correlated medical doctor variable tends to obscure
the relationship between air pollution and the mortality rate.
The results obtained by Crocker e_t jQ,, however, should not be taken
as an unequivocal refutation of the evidence suggesting that air pollution
has a significant positive impact on mortality rates. Although the Crocker
et al. results indicate that the air pollution-mortality rate relationship
is probably more complex than that represented by a single concentration-
response equation, the Crocker et al. model only focuses on one type of
mitigating behavior that individuals may undertake to offset the effects of
pollution. In addition to medical care, Crocker et al. note that a
completely specified model would have to take into account other mitigating
as well as averting behavior such as migration. Consequently, the Crocker
et al. model cannot be considered to be a completely specified model of the
air pollution-mortality rate relationship.
In a recent paper, Atkinson (20) tested the validity of Crocker et
al.'s use of a simultaneous equation framework to 'examine the relationship
between air pollution and mortality rates. When compared to a single
equation concentration-response function, he found that the simultaneous
model resulted in an almost imperceptible change in the coefficients of the
pollution variables. Atkinson therefore concluded that the use of a simul-
taneous equation framework was not justified.
4-39
-------
In addition to the differences in model specification, comparison of
the results obtained by Crocker et al. with the results obtained in other
macroepidemiological studies is not completely straightforward because of
the different units of observations and variables used in each of the
analyses. Crocker et al. used city data, while some studies, such as Lave
and Seskin, have used SMSA data. Although it has been argued that city-
level pollution data are a better indicator of a population's exposure to
air pollution, Lipfert found that the estimated relationship between TSP
and mortality rates did not change significantly when city data were used
in place of SMSA data. Consequently, it is questionable whether differ-
ences in the unit of observation are responsible for the disparity between
the results obtained by Crocker et al. and other macroepidemiological
studies.
Although the inclusion of dietary and smoking variables in the Crocker
e_t .§_!. analysis is a significant improvement in the modeling of the air
pollution-mortality rate relationship, these variables are admittedly crude
surrogates for each city's food and cigarette consumption. Consequently,
it is unclear whether the inclusion of these variables can account for the
insignificant relationship they found between air pollution and mortality
rates. As previously mentioned, the variable for per-capita cigarette
consumption was estimated from the cigarette tax revenues for the state in
which the city was located. This variable obviously reflects the cigarette
consumption patterns outside of the city and thus may be a poor proxy for
the actual cigarette consumption within the city if city consumption syste-
matically differs from rural consumption. In any event, the correlations
between the air pollution and smoking variables are relatively low in the
Crocker e_t a_l. analysis, ranging from -0.08 to 0.23, suggesting that the
omission of smoking from a mortality rate equation may not seriously bias
the air pollution coefficients.
The dietary variables in the Crocker et al. analysis were constructed
based on a U.S. Department of Agriculture Survey of food consumption by
income level for four regions of the country. Each city's food consumption
was determined by a weighted average of the food consumption by the number
4-40
-------
of families in each, income class, and thus may be correlated with income.
Consequently, the relationships observed between mortality rates and these
dietary variables may be proxying for the relationship between mortality
rates and income.
In addition to testing the validity of the simultaneous equation
framework used by Crocker et al., Atkinson (20) tested the appropriateness
of including additional variables in the concentration-response function in
order to minimize the possibility of omitted variable bias. In general, he
found that, with the exception of the variables COLO and CIG, the addi-
tional variables included in the Crocker et al. model add more to the
variance of the estimated coefficients than they do to reduce coefficient
bias. Since the increased variance reduces the possibility of finding
statistical significance, it is not surprising that Crocker et al. fail to
find a significant relationship between air pollution and mortality rates.
In summary, the Crocker et al. results indicate that the air
pollution-mortality rate relationship is an extremely complex one and that
the air pollution coefficients may be extremely sensitive to model specifi-
cation. Their results point out the need for a completely specified
mortality rate-air pollution model and suggest that special care should be
taken in interpreting the results of single equation concentration-response
functions.
Gerkiag mad Sclmlxe (21) —
Using the same data set as Crocker et al,. Gerking and Schulze
estimated the air pollution-mortality rate relationship using ordinary and
two-stage least squares to examine the sensitivity of the coefficients of
the air pollution variables to model specification. They estimated a
single linear concentration-response equation that was similar to the basic
equation estimated by Lave and Seskin and found that the annual geometric
mean of particulate matter was positive and significantly related to the
mortality rate. In addition to adding other explanatory variables to the
"basic" mortality rate equation, a second model was estimated using two-
4-41
-------
stage least squares in order to take account of the simultaneous relation-
ship between, the mortality rate and medical care.* In this model, it was
reported that all of the coefficients of the pollution variables (TSP, S02,
N0~) in the mortality rate equation were negative and significantly related
to the mortality rate.** In the medical care equation, it was observed
that a significant negative association existed between medical doctors per
capita and the annual level of particulate matter.
Like Crocker et al.. Gerking and Schulze stated that one possible
reason for the significant positive association observed between air pollu-
tion and mortality rates in other macroepidemiological studies was due to
the omission of medical care from the concentration-response equation.
Since pollution was shown to be correlated with medical care (i.e., medical-
doctors choose not to live in polluted areas), the coefficients of air
pollution in these studies might be "picking up" some of the effects of the
excluded medical care variable. Hence, air pollution might have appeared
in these studies to have a significant effect on mortality when in fact it
was the doctors, choosing not to live in polluted areas, who had the "true"
effect on mortality rates. Gerking and Schulze, however, could not explain
why the air pollution coefficients in their mortality rate equation were
negative.
Since Gerking and Schulze used the same data base and methodology as
Crocker je_t .§_!., the comments made regarding the Crocker .e_t .§_!. study also
apply to Gerking and Schulze's analysis.
* The reduced form medical care equation (i.e., doctors per capita)
estimated by Gerking and Schulze includes all of the pollution variables
and hence is different from the Crocker et al. reduced-form equation.
** Chappie and Lave (5) state that "the alleged significance arises from a
technical error in (the program used in) the computation of the sampling
variances of the two-stage least squares coefficients. (The program
used by Gerking and Schulze and by Crocker et al.) gives the correct
2SLS (two-stage least squares) coefficients but, using the fitted rather
than the observed values of the endogenous variable in the computations,
gives erroneous values for their sampling variances and t-statistics."
Thus, none of the pollution coefficients were statistically different
from zero in Gerking and Schulze's second stage mortality rate equation.
4-42
-------
Chappie mad Lave (5) —
In response to criticisms raised against the macroepidemiological
approach and the work of Lave and Seskin, Chappie and Lave re—estimated the
relationship between mortality rates and particnlate matter and sulfate
pollution. Using a 1974 data set, they attempted to address the criticisms
regarding omitted variables, aggregation, heteroskedasticity, and simul-
taneous equation bias.*
The basic Lave and Seskin unadjusted total mortality rate equations
for 1960 and 1969 (see Table 4-2) were re-estimated with 1974 data. The
results of this re-estimation are shown in Table 4-5 with Lave and Seskin's
1960 and 1969 mortality rate equations for comparison. Although the
hypothesis of identical coefficients for the independent variables across
alternative years could not be rejected, it is interesting to note the
change in the estimated relationship between TSP and the mortality rate.
As can be seen in Equations 4-5.5 and 4-5.6, the elasticities of the TSP
variables are somewhat smaller in 1974 than they were in the 1960 and 1969
estimations. In addition, the t-tests on the individual coefficients and
the F-tests on the joint contribution of the coefficients in the 1974
estimations indicated that the TSP coefficients were not significantly
different from zero. The sulfate coefficients, on the other hand,
increased in both size and significance over the 1960 and 1969 estimates.
In response to the criticism regarding the omission of relevant
explanatory variables from the mortality rate equation. Chappie and Lave
estimated alternative mortality rate equations with variables representing
alcohol consumption, cigarette consumption, industry mix and occupation
mix. These equations were estimated with the natural mortality rate as the
dependent variable under the assumption that deaths due to accidents,
suicides, homicides, and other external causes were not systematically
related to air pollution. Both alcohol and cigarette consumption, measured
* The 1974 data set consisted of the 3-year averages of mortality rates and
pollution levels from 1973 to 1975.
4-43
-------
Table 4-5
COMPARISON OF LAVE AND SESKIN (6) AND CHAPPIE AND LAVE (5)
UNADJUSTED TOTAL MORTALITY RATE EQUATIONS
R2
Constant
Air pollution variables
MINS
MEANS
MAXS
Sum S elasticities
MINP
MEANP
MAXP
Sum P elasticities
Socioeconomic variables
POP/M2
> 65
NWHITE
POOR
LOG (POP)
Lave and Seskin
1960
4-5.1
0.831
343.23
4.733
(1.67)
1.726
(0.53)
0.279
(0.25)
0.050
0.199
. (0.32)
0.303
(0.71)
-0.018
(-0.19)
0.044
0.0083
(1.54)
68.802
(16.63)
3.J6
(3.82)
0.384
(0.26)
-27.566
(-1.38)
4-5.2
0.828
301.205
6.31
(2.71)
0.033
0.452
(2.67)
0.059
0.0089
(1.71)
70.28
(18.09)
4.22
(4.32)
-0.02
(-0.02)
-21.2
(-1.12)
1969
4-5.3
0.817
387.011
-0.384
(-0.07)
6.329
(1.81)
-0.527
(-0.67)
0.059
0.434
(0.69)
0.056
(0.13)
0.130
(1.83)
0.056
0.013
(2.51)
64.030
(17.11)
2.037
(2.24)
5.113
(2.13)
-42.774
(-2.22)
4-5.4
0.805
330.647
7.74
(2.11)
0.030
0.818
(3.39)
0.087
0.013
(2.54)
65.68
(18.09)
2.04
(2.27)
5.57
(2.29)
-36.5
(-1.94)
Chappie and Lave
1974
4-5.5
0.861
313.342
0.294
(0.04)
16.915
(3.09)
-1.809
(-2.09)
0.132
2.366
(1.32)
-1.386
(-1.51)
0.294
(1.80)
0.006
9.69E-03
(1.92)
64.265
(17.59)
2.000
(1.96)
5.148
(2.15)
-44.594
(-2.80)
4-5.6
0.844
291.505
18.322
(5.40)
0.072
0.434
(1.37)
0.037
0.017
(3.73)
66.182
(18.10)
2.656
(2.67)
4.258
(1.74)
-40.413
(-2.50)
Source: Lave and Seskin (6), p. 121; and Chappie and Lave (5), Table 2.
4-44
-------
in terms of per—capita expenditures on alcohol and cigarettes in each SMSA,
were positively related to the mortality rate. The industry mix variables
indicated that the unemployed and people employed in manufacturing had
higher mortality rates relative to those employed in services and educa-
tion. The sum of the elasticities of the sulfates variables in these
equations was relatively unchanged, ranging from 0.118 to 0.166.* The sum
of the TSP elasticities in this group of equations was insignificant and
close to zero, ranging from -0.015 to -0.030.
Chappie and Lave also estimated the mortality rate equations using
Generalized Least Squares (GLS) in order to correct for the possibility of
unequal variances among the error terms. Use of Ordinary Least Squares
(OLS) under these conditions would result in coefficient estimates that
were unbiased but inefficient (i.e., coefficient estimates would not have
minimum variances). The sums of the sulfate elasticities remained
relatively unchanged when GLS was use~d instead of OLS, while the sums of
the TSP elasticities diminished somewhat (ranging from -0.042 to -0.058)
and were still insignificant. Since the use of GLS did not appear to
affect the significance of the air pollution coefficients, the remainder of
the mortality rate equations were estimated using OLS.
The mortality rate equations were also estimated using counties and
cities, instead of SHSAs, as the unit of observation. Again, the elastici-
ties of the sulfate variables remained relatively constant across these
alternative specifications. The sum of the TSP elasticities, however, did
change depending on the unit of observation. These elasticities, although
insignificant, increased as the unit of observation got smaller. This
result is not surprising, since there may be a better matching of popula-
tion exposure to the ambient level of total suspended particulates as the
unit of observation gets smaller.
* Unlike the Lave and Seskin study. Chappie and Lave used all six measures
of air pollution when they tested the sensitivity of the air pollution
coefficients to alternative specifications. Because of the correlation
that exists among these pollution variables, the possibility that any one
pollution variable will be significant in these equations is reduced.
4-45
-------
Chappie and Lave also tested the sensitivity cf the air pollution
coefficients to a simultaneous equation framework. Following Crocker et
al., a model reflecting the simultaneous relationship between the mortality
rate and medical care was estimated. Medical care was prozied by a
variable representing the number of patient care physicians per capita.
This variable is a better indicator of medical services than the medical
doctors per capita used by Crocker et al. and Gerking and Schulze since it
does not include medical doctors whose primary jobs are teaching and
research.
Estimating the relationship between air pollution and mortality rates
using a simultaneous equation framework did not appear to change signifi-
cantly the estimated elasticities of the pollution variables. The sum of
the sulfate elasticities decreased somewhat but remained positive while the
sum of the TSP elasticities remained negative and insignificant. Re-
estimating the simultaneous model with variables representing dietary
habits also did not affect the air pollution coefficients appreciably.
Several comments can be made regarding the Chappie and Lave study.
Although the study indicates that air pollution has a positive effect on
mortality rates, the specific pollutant having this effect appears to have
changed over time. In Lave and Seskin's 1977 analysis, the coefficients of
TSP were consistently positive and significantly related to mortality
rates, while the coefficients and significance of the sulfate variables
were less stable across alternative specifications. Lave and Chappie's
1981 analysis, however, found that the sulfate variables were more consis-
tently related to mortality rates than the TSP variables. A number of
reasons can be given for this change. The SMSAs included in the Chappie
and Lave analysis had ambient levels of TSP that were significantly lower
than those included in the Lave and Seskin study. The mean of the annual
arithmetic averages of TSP for the SHSAs used in Chappie and Lave's study
was equal to 75.016 ug/m3. The means for the 1960 and 1969 data sets used
by Lave and Seskin were equal to 118.145 ug/m3 and 95.580 ug/m3, respec-
tively. Consequently, the ambient levels of TSP used to estimate the
mortality rate equations in the Chappie and Lave analysis may be beneath
4-46
-------
those where perceptible mortality rate effects occur if the relationship
between air pollution and mortality rates is not linear. The mean sulfate
level, however, did not change significantly during this period. In addi-
tion, the decreased variation in TSP levels in the 1974 data set (the
standard deviation of TSP was 40.942 in the 1960 data set as compared to
20.570 in the 1974 data set) reduces the probability that TSP would have a
statistically significant impact on mortality rates.
The use of three measures of TSP exposure may be another reason why
the relationship beween particulate matter and the mortality rate was
insignificant in the Chappie and Lave analysis. These three measures are
highly correlated and therefore prevents their coefficients from being
estimated with precision. As previously mentioned, the analysis by
Atkinson (20) has suggested that inclusion of highly correlated variables
may do more to increase the variance of the estimated coefficients than to
reduce their bias. Although Chappie and Lave did estimate two mortality
rate equations with only one measure of TSP, they found the coefficients of
these variables to be statistically insignificant. However, it is unclear
whether these coefficients would have remained insignificant across
alternative specifications. In fact, MAXP was plausibly signed and
approached statistical significance (t-statistic greater than 1.64) in 23
of the 38 equations that were estimated.
In summary, the Chappie and Lave results suggest that air pollution.
as measured in terms of the ambient level of sulfates, does have a signifi-
cant impact on mortality rates. TSP, however, does not appear to be signi-
ficantly related to mortality rates. Although many of the criticisms of
the macroepidemiological approach have been addressed in this study,
particular caution should be used in interpreting the results of this study
since the complex relationship between air pollution and mortality has not
been completely modelled.
4-47
-------
In this section, a number of tie macroepidemiological studies that
have examined the relationship between particulate matter and mortality
rates have been reviewed. Although some of these studies have found a
significant positive relationship between TSP and mortality rates, a number
of the studies have failed to find such a relationship. A summary of the
salient characteristics and findings of the macroepidemiological studies
discussed in this subsection can be found in Table 4-6.
As Table 4-6 shows, three of the nine mortality rate studies using
different data bases that have been reviewed in this subsection have not
found a significant relationship between TSP and mortality rates. The
disparate findings of the studies reviewed in this subsection indicate that
estimating the "true" relationship between TSP and the mortality rate is an
extremely difficult task. These studies are beset by a number of difficul-
ties. First, the underlying theoretical structure of how air pollution
affects mortality rates is not known. Neither the functional form nor all
of the variables influencing mortality rates is known. As seen in this
subsection, these models are generally very simple single equation linear
concentration-response equations. Except for the use of medical care,
these studies are unable to reflect how individuals may avert the effects
of exposure to air pollution on the mortality rates in the area where these
individuals reside.
Second, these studies are forced to rely on data that have not been
collected for the specific purpose of uncovering the relationship between
air pollution and mortality rates. Consequently, many of these data are
poor surrogates for the variables that are included in the macro
concentration—response equations. (Consider, for example, the proxies used
in some of these studies for cigarette, food, and alcohol consumption.)
Furthermore, many of the data necessary to completely specify the mortality
rate equations are simply unavailable. As stated before, the omission of
relevant variables from a mortality rate equation that is estimated using
OLS may result in biased air pollution coefficients if the omitted
4-48
-------
H
0
X
ft O,
CO
CO
0)
u
9
AJ
to
41
(0
U
1
C
«
AJ
3
2
.*
3
X
o> e
CO O
tt ft
JQ AJ
m
01 U
fH 01
•85
•H O
U
> 0
e ft
•O 9
S
4>
O.
S
X U
3 S
va
\o
^
00
o
o
o
u
">
o
0
a i
o •*
•H -a
*J T3 y
3 fl) -H
3, !
o s a
°" * u
IM a 4>
O ki 0
41 -H
x»4 o
u *H 0
•HUB
•H 0
Sail a
a 0) co 0)
41 « CO CO
CO M O U
Of CO ft CO
ill
S -I -.
•H x a
e u
1 13 00
i s e
a t>
^» - a •
00 O « CO ON
3V rH
cn ^
00
m
0
"O 1
CO to
& 4)
1 flu 00 O
ft CO 01 AJ
AJ H 1*
SQ
w xS
= l-a2
SSgu
»J J= 01
a a e e
3 O -<
cr-o a
4) 4) u 00
. s y eo
U X 41
a a a\ at
71 ?«*
Isils
fH
§2
§0
i a\
1 «H
~» o
2B '
e a
a •*
•0 U
B i
a S •
4* r*.
SS2
X
4J
fl
r*.
2
i*
•H
03
AJ
H
a
-J-3S-
J— -C r*
co to ON
C
•2 CO
AJ <
9 O
fH O
0 M
a. x
C ft •
x 41 a
» 00 • rt ~.
! u a s 3.
a. 41 — . e
w > oo e c
H a a. o -H
a
u -o i" u
o e .s «
144 a 7 £4
OS ^ 41
£ . , „ .
y fll *Q r^ a
•*4 f^ 41 OX C
S u a T u
S.H 00
c v so x
a. o u cn u
a -H -ic
i " e 3
41 3 O a O
y -H -i o u
U 0 3 -H X
1 OalH (3
X rf
« u C 0 <
sc
oo r*
41 ON
14 *4
00
^0 O
0 0
° 2
o
09 O
O O
O 1
1
l a
1 41
« S
o
0) C
£ O
44 *4
•"
x'o o
4J O.-H
4^ a u
— 41
SU4 a
O A
a a
01 B O
•^2-s
01 y 4i
4>
o
A
0)
n
a
a)
B
a)
CO
c*
o\
m o
y go
f4 0}
(M hi
"Jj g
a —
a x
10 U u
a u* u
•o y ••
.IS
fH a ON
CO 1 f-l
o ao o
H « *J
U <«*K
*! °
aS
^.^
*j
(0
u
s
o c
O ft
CO
0 C
o
1
S CO
c2 §
o 41 e
-1 U O
Q) js a
S-j e
to 4)
>*4 -O 00
•O 3 -0
-i > j: •
§••3 4f1
o s 41 a
0 -1 3 3
0 *O
a e
^ a
a
.. a)
42
90 AJ
= 8
(0
ll
a o •
a cn tn
I 41
1 - 0
BS!
.*
1
t)
4)
ta
i *J
9
Q
41 U
Q.
CO —
I 0
ot ^
-------
variables are correlated with air pollution. A major criticism of many
mortality studies has been that the coefficients of the pollution variables
are probably biased upward if cigarette smoking, a variable that may be
correlated with air pollution, is excluded from the mortality rate
equation. Although both Crocker et al. (4) and Chappie and Lave (5) have
found that their air pollution variables were not highly correlated with
their measures of smoking (the highest correlations between these variables
were 0.23 and 0.17, respectively), this does not rule out the possibility
that air pollution "picks up" some of the effects of smoking when smoking
is not included in these mortality rate equations.
Third, the variables that are included in the mortality rate equations
tend to be highly correlated among themselves, and thus prevent the rela-
tionship between air pollution and mortality rates from being estimated
with precision. This results in a tendency for the estimated coefficients
of the air pollution variables to be insignificant.
Fourth, the actual exposure of a population to air pollution cannot be
measured in these studies and must be approximated by air quality data from
a monitoring station(s) that usually covers a relatively large geographic
area. If these monitoring stations are typically placed in areas where the
worst air pollution occurs, they may overestimate the exposure of the
population and hence may result in air pollution coefficients that are
biased downward.* Conversely, the air pollution coefficients based on
these monitoring stations may be overestimates if the readings from these
monitoring stations are correlated with omitted "urban" effects..
Despite these problems, however, the macroepidemiological mortality
studies do offer several advantages for estimating the mortality effects of
chronic exposure to air pollution. The estimated relationship between air
* As previously mentioned, this may bias the coefficient estimates but will
not bias the changes in mortality rates that result from a change in
pollution measured at the monitor where the worst air pollution occurs if
the relationship between the monitored and true air pollution exposure is
maintained after the change in pollution.
4-50
-------
pollution and mortality rates are based on ambient air quality data and
therefore lend themselves easily to estimating the effect of changes in
ambient air quality on the mortality rate. As can be seen in Table 4-7,
the levels of TSP used by the studies reviewed in this section are repre-
sentative of present levels of TSP.
Given the purpose of this study, the macroepidemiological studies are
useful for providing some information regarding the effects of TSP on the
mortality rates for geographic areas. They are not designed, nor intended,
to measure the mortality effects of chronic exposure to each individual
member of a population. Consequently, the results of the mortality studies
reviewed in this subsection can be used to approximate the health effects
Table 4-7
TSP LEVELS USED IN MACROEPIDEMIOLOGICAL STUDIES
Study
Lave and Seskin (6) 1960 data
1969 data
Koshal and Koshal (12)
Gregor (13)
Lipfert (14)
Lipfert (16)
Mendelsohn and Orcutt (17)
Seneca and Asch (18)
Crocker .et al. (4)
Chappie and Lave (5)
Mean
118.15
95.58
N.R.*
122.57
90.50
N.R.
69.90
66.97
102.30
75.02
Standard
Deviation
40.94
28.64
N.R.
15.38
N.R.
N.R.
19.10
22.37
30.11
20.57
* Not reported.
4-51
-------
of reductions in tie risk of mortality in geographic areas that experience
reductions in the ambient level of TSP under alternative air quality
standards.
Range of Chronic Exposure Mortality Rate Effects —
As evidenced by the studies critiqued in this section, there is much
uncertainty surrounding the mortality rate effects of chronic exposure to
TSP. Since some of the studies reviewed in this section have not found a
significant relationship between mortality rates and TSP, there is even
uncertainty about whether chronic exposure to TSP has any effect on
mortality. Although the "no effects" studies are not an unconditional
refutation of the evidence suggesting that the ambient level of TSP has a
positive effect on mortality rates, they do cast doubt on the validity of
such evidence. In fact, results from the Chappie and Lave analysis suggest
that it maybe sulfates (S04) rather than TSP that have an adverse effect
on mortality rates. In order to reflect this uncertainty in the range of
the mortality effects of chronic exposure to TSP, zero will be chosen as
the minimum of this range.
The maximum of this range will be determined by comparing the elasti-
cities of TSP in those studies finding a significant relationship between
TSP and mortality rates. Table 4-8 reports the elasticities evaluated at
the average of the 1978 mortality rates and the 1978 annual averages of TSP
in those counties* that can be used to calculate the benefits of reductions
in the level of TSP.** As can be seen in the table, these elasticities
range from 1.1E-04 in Seneca and Asch's study to 0.783 in Gregor's
analysis.
* There are 519 counties being considered in this analysis.
** The elasticity of TSP reported in a linear specification is equal to b '
(TSP/MR) where b is the estimated coefficient of TSP in the mortality
rate equation, TSP is the mean level of TSP, and MR is the mean
mortality rate. Therefore, use of the elasticities reported in these
studies would be inappropriate because TSP levels and mortality rates
have changed over time.
4-52
-------
Table 4-8
COMPARISON OF TSP ELASTICITIES CALCULATED FROM
MACROEPIDEMIOLOGICAL MORTALITY RATE STUDIES
Study
Elasticity
Lave and Seskin (6)
Koshal and Koshal (12)
Gregor (13)
Lipfert (14,15)
Lipfert (16)
Seneca and Asch (18)
0.018 to 0.087
0.176
0.027 to 0.783
0.042 to 0.076
0.040 to 0.093
1.1E-04 to 1.35E-04
The elasticities from the Gregor analysis are based on age-sex-race-
specific pollution-related mortality rates and are much higher than the
elasticities for similar equations reported in other studies. As
previously mentioned, the Gregor analysis is based on census tract data
from the heavily industrialized Allegheny County in Pennsylvania where the
long-term levels and chemical composition of TSP are probably not represen-
tative of TSP levels throughout the rest of the country. Consequently, the
Gregor analysis will not be used in determining the range of effects of
chronic exposure to TSP on the mortality rate. It should be kept in mind,
however, that the Gregor results imply that the effects of particulate
matter on mortality rates may be larger than evidence from more aggregate
data suggests. The Seneca and Asch study examined the relationship between
mortality rates and a limited number of socioeconomic variables for only
one state; hence it will not be used for the purposes of this study.
Because Koshal and Koshal did not test the .sensitivity of the coefficient
of TSP to alternative specifications, their study will also not be used in
determining the range of mortality rate effects. Although the maximum
elasticity based on the studies by Lave and Seskin and Lipfert is
4-53
-------
approximately 0.09, the majority of their elasticities fall within the 0.04
to 0.06 range.* The elasticity of O.OSO will therefore be used as the
maximum of the range of chronic mortality rate effects.
Choice of the point estimate will be determined by considering the
evidence from all of the chronic mortality studies. None of the studies
reviewed can be considered perfect; each has its flaws. As previously
mentioned, three of the studies have not found that TSP has a significant
effect on mortality rates. However, because of the previously discussed
limitations within each of these studies, their results should not be taken
as an unequivocal refutation of other evidence that suggests that TSP has
an adverse effect on mortality rates. For example, the lack of signifi-
cance in these studies may be due, in some part, to the inclusion of too
many variables in the mortality rate equations.** As previously mentioned,
this tends to increase the variance of the estimated coefficients and may
reduce the possibility of finding statistical significance.
The six other studies reviewed in this section have found a consis-
tently significant relationship between TSP and mortality rates. Each of
these studies also has limitations. For example, many of the studies have
been criticized for failing to include a sufficient number of socioeconomic
variables in the mortality rate equations.
Because of the conflicting evidence among the mortality studies, the
point estimate will be based on a mortality rate equation that includes a
reasonable number of variables in order to minimize the possibility that
the TSP variable is proxying for an omitted variable. The Gregor, Koshal
and Koshal, and Seneca and Asch studies can be criticized for not testing
* Lipfert's (15,16) inclusion of a proxy for smoking reduces the proba-
bility that these coefficients include the effect of cigarette smoking
on mortality rates.
** Chappie and Lave included six pollution variables in their mortality
rate equations; Mendelsohn and Orcutt included seven pollution
variables; Crocker et al. included 13 explanatory variables.
4-54
-------
the sensitivity of their results to additional socioeconomic variables.
The studies by Mendelsohn and Orcutt, Crocker et al.. and Chappie and Lave,
on the other hand, can be criticized for including too many variables in
their mortality rate equations. Both Lave and Seskin and Lipfert examined
the sensitivity of the relationship between TSP and mortality rates to the
addition of various socioeconomic factors. The Lipfert results, however,
cannot be used for the purposes of this analysis because the TSP variable
used by Lipfert (i.e., TSP net of sulfates) is not comparable to the
particulate matter data that are being used for the calculation of
benefits. Consequently, the Lave and Seskin results will be used in deter-
mining the point estimate.
In order to be conservative, the lowest TSP elasticity obtained from
the Lave and Seskin equations that included these additional socioeconomic
factors will be used as the point estimate. This results in a point
elasticity estimate of 0.018 being chosen from an equation that includes
home-heating fuel characteristics as explanatory variables. Because of the
correlation between home-heating fuel and the ambient level of TSP, the
point elasticity estimate is based on a TSP coefficient that is not signi-
ficantly different from zero. Choice of the point elasticity on the basis
of statistical significance, however, would have resulted in a less conser-
vative point estimate.
The range of elasticities will change depending on the level of TSP;
hence, the actual TSP coefficients associated with these elasticities will
be used to estimate the range of benefits associated with particulate
matter reductions. The coefficients that will be used to estimate the
effects of a change in the ambient level of particulate matter on mortality
rate are given in Table 4-9.
The relatively wide range of coefficients that will be used to
estimate the reductions in mortality risk associated with reductions in
particulate matter reflects the uncertainty attached to these estimates.
Table 4-9 indicates that for a 1 ug/m3 change in TSP, the change in the
4-55
-------
Table 4-9
RANGE OF COEFFICIENTS MEASURING THE RELATIONSHIP BETWEEN
THE MORTALITY RATE AND TSP
Minimum 0.000
Point Estimate 0.171
Maximum 0.471
total mortality rate (deaths per 100,000) is expected to be in the range of
0 to 0.471, with a point estimate of 0.171.
MORBIDITY STUDIES
Overview of Approach
There have been relatively few economic studies that have analyzed the
effect of air pollution on human illness. Data limitations and lack of a
consistent definition of what constitutes an air pollution—induced illness
have prevented extensive research from being done in this area. Early
studies that have analyzed the impact of air pollution on morbidity have
concentrated on hospitalization utilization rates and the incidence of
illness. Other studies have attempted to impute the morbidity effects of
air pollution as some percentage of mortality effects. More recent studies
have examined the labor productivity effects of air pollution exposure.
Studies examining hospitalization utilization rates and the incidence of
illness are being considered in Section 3 and thus will not be reviewed
here. Imputing morbidity costs from mortality studies is considered
inappropriate for the purposes of this report. Consequently, it is the
last of these types of studies that will be critiqued in this subsection.
The labor productivity studies by Crocker e_t al. (4) and Ostro (22)
have both estimated concentration-response equations relating the number of
4-56
-------
days lost from work to the ambient level of particulate matter, as measured
in terms of TSP. * Both estimate their concentration-response equations
based on individual data and thus overcome many of the difficulties of
using aggregate data to estimate the relationship between health status and
air pollution. The data sets used in these studies are rich in detail and
include information on individual habits such as smoking, exercise, and
diet. Like mortality studies, these studies match data on ambient air
quality (measured at the county or city level) to specific individuals and
thus estimate the relationship between ambient air quality and individual
illness.** Consequently, one of the advantages of these studies is that
the implied impact of a change in the ambient level of particulate matter
on illness can be calculated easily.
In simple form, these studies estimate the following relationship
between days lost from work due to illness and air pollution:
f Gt E) (4.3)
where SICK. = the number of days lost from work due to the illness of
the ith individual.
P. = vector of personal characteristics of i (e.g., dietary,
occupation, schooling).
G^ = vector of genetic characteristics of i (e.g., sex,
race) .
E^ = vector of environmental characteristics where one of
the variables in the vector is the ambient level of
particulate matter.
* Ostro also estimates the effects of air pollution on the nonworking
population.
** The disadvantages of using ambient air quality data to represent the
exposure of individuals to air pollution have been discussed in the
review of the macroepidemiological mortality studies. The same comments
apply to the morbidity studies reviewed in this subsection and thus need
not be repeated.
+ This concentration—response equation is for illustrative purposes only
and does not represent the actual concentration-response equations used
in the Crocker et al. and Ostro studies.
4-57
-------
Two types of illness are measured in these studies ~ acute (short-
term) illness and chronic (long-term) illness. Examples of an acute
illness that may be affected by air pollution are influenza and pneumonia.
Examples of chronic illness that may be affected by air pollution are
asthma and chronic bronchitis.
Assuming that Equation (4.3) is linear, the change in individual work-
loss days due to a change in ambient particulate matter measured in terms
of TSP is equal to:
(4.4)
where ASICK- = the change in individual i's work-loss days due to a
change in the ambient level of TSP.
b = the partial derivative of work-loss days with respect
to TSP.
ATSPj = the change in the ambient level of TSP to which indivi-
dual i is exposed.
Both of these studies use the annual average of TSP as a proxy for
long-term exposure, and consequently examine the effects of chronic
exposure to TSP.*
Several of the disadvantages that apply to using macroepidemiological
mortality studies to estimate the health effects of exposure to air pollu-
tion also apply to morbidity studies. Although the morbidity studies
considered in this subsection are able to control for many of the factors
that influence illness, they are unable to control for all of these
factors. Consequently, the models estimated in these studies cannot be
considered to be completely specified. As previously mentioned, incomplete
specification of the health status equation when using OLS will not bias
the estimated relationship between air pollution and health status if the
* As discussed in the review of mortality studies, the coefficient of the
annual average of TSP may "pick up" some of the effects of acute
exposure.
4-58
-------
variables that are excluded are not correlated with the air pollution
variables. If these excluded variables are correlated with the air pollu-
tion variables, however, the relationship between air pollution and health
status will be biased. For example, occupational exposures to hazards and
other environmental pollutants are factors that may influence illness and
may be positively correlated with air pollution. Neither of these factors
are controlled for in the Crocker et al. analysis, and therefore the
relationship between air pollution and illness estimated in these studies
may be overestimated. Ostro does not use OLS to estimate his
concentration-response equation for workers and it is therefore difficult
to determine the effect of these excluded variables on the coefficients of
air pollution in his study.
Both of these studies measure acute illness in terms of the number of
days lost from work.* A problem arises in the use of work-loss days as the
dependent variable in the concentration—response equation since work—loss
days may or may not be related to illness. In addition, there are numerous
types of illnesses that may result in work-loss days that are clearly
unrelated to pollution (e.g., broken bones, job injuries). This error in
the measurement of the dependent variable may result in biased air pollu-
tion coefficients if air pollution is correlated with that portion of
illness that is not pollution-related. In order to minimize this possi-
bility, only the pollution-related component of work-loss days should be
used as the dependent variable in these studies.**
Another problem in these studies is that the theoretical structure
underlying these models is unknown. The studies both estimate simple
models of the relationship between air pollution and illness. Thus, these
models may not account for many of the behavioral adjustments that may
* Crocker also estimates a chronic illness equation using length of
disability as a dependent variable. In addition, Ostro estimates the
effects of air pollution on the illness of the nonworking population
using reduced activity days as a measure of illness.
** Ostro's results are reported for total work-loss days and work-loss days
that can be considered to be.pollution-related.
4-59
-------
affect the air pollution-morbidity relationship. The concentration-
response equations that are estimated in these studies are based primarily
on personal and environmental characteristics. The economic factors that
impact work-loss days are not controlled for adequately. It is highly
probable that the decision to be absent from work is dependent on a number
of other factors besides personal and environmental characteristics. These
factors would include attitude toward work, compensation for time sick,
financial responsibilities, and the characteristics of the job itself.
Again, failure to control adequately for all of the factors that influence
work-loss days may result in biased air pollution coefficients.
The Crocker et al. and Ostro morbidity studies are sufficiently
different that the remaining comments will be addressed in the critique of
each of these studies. At this point, it is evident that their use of
individual data gives them distinct advantages over mortality studies for
measuring the impact of TSP on health, but there are certain disadvantages
of these studies that may attenuate their findings.
It should also be mentioned that the Crocker et al. study only
measures the labor productivity effects of exposure to air pollution. The
Ostro study, on the other hand, measures the effects of air pollution on
the activities of nonworkers in addition to the effects on workers' produc-
tivity. Neither study attempts to measure the medical expenses associated
with these productivity losses, nor do they attempt to impute a value for
residual pain and suffering. Consequently, in this respect, these studies
will underestimate the total effect of a change in particulate matter on
morbidity.
Literature Review
Crocker et »1. (4) —
Using random samples from a University of Michigan longitudinal survey
of approximately 5,000 households from 1968 through 1976 and information on
annual levels of TSP, SOj, and N02, Crocker et .§_!. undertook an
4-60
-------
experimental study to investigate the impacts of air pollution on measures
of acute and chronic illness. They also estimated the impact that changes
in the incidence of acute and chronic illness would have on wages and hours
worked. The structural model they posited was of the following form:
LDSAj = f(Pi, Git Mi, EJ (4.5)
ACDTi = g(LDSAi, PL, Gi, Mj, E^ (4.6)
WAGEi = h(LDSAi, ACllTi, COLi, EXP£, PI, GI) (4.7)
HOURS£ - k(WAGEi( LDSA^ ACUT^, XINCi, W£) (4.8)
where LDSA^ - length of disability of i.
ACDTi » workdays ill of i.
WAGE^ = marginal hourly earning rate of i.
HOURS^ = i's annual hours working for money.
P£ = vector of personal characteristics of i.
G^ = vector of genetic characteristics of i.
M. = purchase of medical care by i.
E. » vector of environmental characteristics in the county
where i resides.
(X)Li = cost of living in the county where i resides.
EXPi = work experience of i.
XINCj = annual transfer income of i.
Wj^ » i's wealth.
i = head of household.
The above model is recursive.
4-61
-------
This model has several unique characteristics. The model is estimated
only for people who have always lived in one state. This is a distinct
advantage over other studies that have tried to measure the health impacts
of chronic exposure to air pollution because migration is specifically
controlled for in the Crocker et al. analysis. Consequently, the model is
better able to measure the effects of long—term exposure to air pollution.
On the other hand, limiting the analysis to people who have always resided
in one state may underestimate the effects of air pollution on the average
individual if people who are more susceptible to air pollution have syste-
matically moved from "dirty" to "clean" states. In addition, the study is
unable to account for migration within the state.
Besides interstate migration, the model is also able to control for
many of the characteristics of the individual that influence health.
Specific information on food and cigarette consumption, exercise habits,
and medical insurance are used to estimate the model.
In addition to estimating the impact of air pollution exposure on
acute and chronic illness, the model goes one step further and estimates
the effect of chronic and acute illness on the wage rate and the number of
hours worked. Unlike other morbidity studies, this ultimately allows the
impact of air pollution on labor productivity to be calculated.
Tables 4-10 and 4-11 report the empirical results and variable defini-
tions, respectively, of the 1970 equations used by Crocker et al. to
estimate the labor productivity benefits of reductions in air pollution.
As can be seen in Table 4-10, the annual geometric mean of TSP has a
significant and positive impact on the incidence of both acute and chronic
illness. The results of these equations indicate that a decrease of 1
|ig/m in the annual geometric mean of TSP would result in a reduction of
4-62
-------
Table 4-10
RESULTS FROM CROCKER, ET AL. (4) MORBIDITY ANALYSIS
Chronic Illness Equation (LDSA)
Constant
DSAB
AGE
EDUC
FEDUC
POOR
RACE
SEI
FOOD
INSR
CHEM
TSP
R2
Coefficient
2.98
0.554
0.005
0.013
-0.044
-0.069
0.072
0.139
-0.902
-0.454
-1.645
0.0028
0.53
t-statistic
*
15.83
1.25
0.45
-1.19
-0.67
0.15
1.22
-0.93
-3.52
-2.86
2.55
Acute Illness Equation (ACDT)
Constant
LDSA
AGE
SEX
CIG
moat
FOOD
RISK
INSR
TSP
R2
Coefficient
165.21
39.52
-1.42
-16.92
-0.09
-78.40
-0.11
-38.84
187.0
0.623
0.20
t-statistic
*
2.96
-1.08
-0.43
-0.73
-1.95
-3.18
-2.93
3.94
1.97
(continued)
4-63
-------
Table 4-10 (Continued)
Wage Equation (WAGE)
Constant
LDSA
EDUC
DSAB
FMSZ
BDALO
LOCC
TARDY
UNION
RACE
R2
Coefficient
-132.32
-25.93
24.07
15.37
26.88
42.38
52.95
-7.16
66.09
47.60
0.41
t-statistic
*
1.80
2.81
0.84
4.42
6.90
2.39
-0.21
1.91
1.39
Hours Worked Equation (HOURS)
Constant
LDSA
WAGE
FMSZ
SEXH
ICTR
BDALO
ACUT
R2
Coefficient
1,266.68
-163.90
0.35
44.26
519.80
-0.27
23.06
0.07
0.55
t-statistic
*
-6.02
2.72
2.65
6.48
-12.36
1.52
2.39
* Not available.
4-64
-------
Table 4-11
DEFINITIONS OF VARIABLES USED IN CROCKER ET AL. (4) MORBIDITY ANALYSIS*
Endogenous Variables
LDSA =» length of disability; <. 2 years = 1; 2-4 years = 2;
5-7 years = 3; >. 8 years = 4; otherwise 0.
ACUT = workdays ill times 16 for the first 8 weeks, and
times 12 thereafter. Only individuals who are
currently employed or unemployed and looking for work
could have positive values for this variable.
WAGE = marginal hourly earning rate in cents.
HOURS = annual hours working for money.
Exogenous Variables
DSAB - degree of disability.
AGE - age in years.
EDUC = educational attainment; 6-8 grades = 2; 9-11 grades =
3; 12 grades = 4; 12 grades plus non-academic
training = 5; college, no degree = 6; college degree
= 7; advanced or professional degree - 8; otherwise
1.
FEDUC = educational attainment of household head's father;
same scaling as EDUC.
POOR = binary variable representing whether household head's
parents were poor (1); otherwise 0.
RACE = 1 if white; otherwise 0.
SEX - 1 if male; otherwise 0.
(continued)
* Unless otherwise stated, all variables refer to household head.
4-65
-------
Table 4-11 (Continued)
Exogenous Variables (continued)
FOOD = family food expenditures relative to a standard
expenditure for food needs.
INSR = 1 if household head has hospital or medical insur-
ance; otherwise 0.
CHEM = 1 if working in chemicals or metals manufacturing
industry; otherwise 0.
TSP = annual 24-hour geometric mean in ug/m .
GIG = annual family expenditures on cigarettes (not indexed
for differences in prices across states).
EXER - 1 if participates in energetic activities; otherwise
0.
RISK = index of risk aversion determined by factors such as
whether the head of household drives a new car and
uses seat belts.
FHSZ = number of persons in the household.
BDALO = cost of living in county of residence.
LOCC = scaled variable reflecting length of present employ-
ment.
TARDY = 1 if late to work once or more a week; otherwise 0.
UNION = 1 if member of labor union; otherwise 0.
ICIR - family income not due to current work effort.
4-66
-------
0.623 annual hours of acute illness and a reduction of 2.33 days of chronic
illness over an 830-day period.*
Turning to the other variables included in the chronic illness
equation, the degree of disability (DSAB), as expected, had the most signi-
ficant impact on the length of disability. AGE, FEDUC, and FOOD, although
conforming to .a priori expectations, were insignificant. INSR was negative
and significant, indicating that the availability of medical insurance
decreases the length of disability.** CHEM, curiously, had a significant
negative impact on LDSA. This result is an anomaly since the equation was
estimated from a sample where only three people were employed in the
chemicals and metals manufacturing sector and none of them had a chronic
disability. The remainder of the variables in the chronic illness equation
were insignificant.
In the acute illness equation, EXER, FOOD, and RISE had a significant
negative impact on the annual number of hours of acute illness. Contrary
to expectations, INSR had a significant positive impact on ACUT. This may
result from the possibility that people who have insurance may tend to
report illness more frequently. LDSA had a significant positive impact on
ACUT and implies that an increase in one discrete unit of LDSA will result
in an increase in the average annual hours of acute illness of approxi-
mately 40 hours. AGE, SEX, and CIG did not appear to have a significant
impact on ACUT. As evidenced by the R , only about 20 percent of the
variation in ACUT is accounted for by the independent variables.
It should be mentioned at this point that because of the high
collinearity among the air pollution variables, Crocker et al. were
* Crocker et al. calculate the change in the number of days in chronic
illness by assuming that each discrete interval of LDSA is slightly more
than two years, or 830 days. Crocker et al. state that the appropriate-
ness of this assumption is questionnable due to the open-ended interval
of LDSA (see variable definition in Table 4-11).
** It is possible that a simultaneous relationship exists between INSR and
LDSA since people who are chronically ill may have difficulty obtaining
insurance.
4-67
-------
hesitant to assign a health effect to a specific pollutant. Rather, they
felt that the effects observed in this study represented the air pollution
phenomenon in general. Consequently, the coefficients of TSP reported in
Table 4-10 are likely to be capturing some of the health effects of other
pollutants that are positively correlated with TSP and not included in the
acute and chronic illness equations. Therefore, the health effects
attributable to TSP alone are likely to be overestimated.
Crocker jit. ^1. estimated other chronic illness equations based on
different samples drawn from the survey population to test the sensitivity
of the relationship between air pollution and illness to different survey
years, socioeconomic variables, and different measures of air pollution
(nitrogen dioxide and sulfur dioxide). At least one of the air pollution
coefficients was statistically significant in five of the eight additional
equations that were estimated. The elasticities of the different measures
of air pollution ranged from 0.268 to 1.143, indicating some sensitivity to
the different samples and air pollution measures that were used. The
majority of the elasticities, however, fell within the 0.2 to 0.4 range.
When the mean level of TSP was included in these equations, it was positive
but not always significant. This insignificance can probably be attributed
to the high correlation between the pollution variables.
Similarly, different air pollution variables and different samples
from the entire survey population were used to examine the sensitivity of
the air pollution coefficients in the acute illness equation. Of the seven
samples that were drawn, air pollution variables were significant in all of
them. The elasticities of the air pollution variables in these equations
ranged from 0.308 to 0.618. With the exception of the 0.618, the majority
of acute illness elasticities were much closer, ranging from 0.308 to
0.544. When both TSP and sulfur dioxide (SOj) were included in the acute
illness equations, TSP generally had a negative sign and was insignificant.
However, when TSP and nitrogen dioxide (N0~) were entered in the same
equation, TSP always had a positive sign. These sign switches of the TSP
variable were probably due to the high collinearity between the pollution
variables (for example, the correlation between the mean levels of TSP and
4-68
-------
SO- in the 1971 sample was 0.80). However, the instability of the TSP
coefficient indicates that the results should be viewed cautiously.
Several specific comments can be made regarding the choice of
variables and estimation of the acute and chronic illness equations.
Although Crocker et al. acknowledged that the relationship between illness
and air pollution is probably nonlinear, both of these equations are based
on the assumption that the relationship between illness and air pollution
is linear. Since there is evidence, however, of a nonlinear relationship
between air pollution and illness, the Crocker et al. illness equations may
not be correctly specified. Consequently, estimation of the effects of TSP
on chronic and acute illness outside of the sample means of these variables
may not be accurate.
A major limitation of the Crocker et al. analysis involves the
measurement of LDSA, the dependent variable in the chronic illness
equation. As shown in Table 4-11, LDSA is not a continuous variable and is
open-ended when the length of disability is greater than eight years. As a
result, interpretation of the regression coefficients in the chronic
illness equation is not straightforward. By assuming that one unit of LDSA
is equal to approximately 830 days, Crocker et al. are able to estimate the
*
impact of a 1 ug/m change in the level of TSP on the number of days of
chronic illness. However, the validity of this approach is admittedly
questionable.
Besides household heads who were members of the labor force, the
sample used by Crocker et al. for the acute and chronic illness equations
also included housewives, students, and retired persons who were heads of
households. These components of the sample were assigned zero hours of
acute illness in estimating the relationship between acute illness and air
pollution. This may tend to underestimate the effects of TSP on acute
illness. Both the chronic and acute effects of exposure to TSP may also be
underestimated since members of the original sample population (i.e., those
in the sample in 1968) who died during the sample period (1968-1976) were
not included in the estimation for the year in which they died.
4-69
-------
As previously mentioned, the system of equations by Crocker et al. was
estimated only for people who had always resided in one state. This may
result in an underestimate of the morbidity effects of pollution exposure
if people who are susceptible to air pollution systematically move from
polluted to non-polluted areas.
On the other hand, the morbidity effects of TSP may be overestimated
since only individuals who resided in areas with valid air quality data
were included in the estimation. Since air quality may tend to be most
consistently monitored in those areas that have air quality problems, this
selection may result in a sample that is biased in favor of highly polluted
areas. Thus, the effects that Crocker et al. observe may not be represen-
tative of the effects that would be observed for the general population.
Another reason why the effects observed in the Crocker et al. analysis
may not be representative of the health effects in the general population
is due to the unrepresentativeness of the University of Michigan survey
itself. Crocker et al. report that a high proportion of the sample is
nonwhite. Therefore, any extrapolation of the Crocker et al. results to
the general population must be viewed in light of this fact.
Turning to the wage and hours worked equations in Table 4-10, both
measures of illness are plausibly signed and significant. Crocker et al.
reported that ACUT did not have a significant effect on WAGE in any of the
wage equations that were estimated and therefore it was not included in the
final specification. Except for the severity of disability variable
(DSAB), the remaining explanatory variables conformed to .a priori expecta-
tions and were significant.
The recursive model estimated by Crocker et al. can be used to
estimate the effects of a change in TSP on the incidence of chronic and
acute illness. This can be easily seen by substituting the LDSA equation
into the WAGE equation and the LDSA, ACTJT, and WAGE equations into the
HOURS equation. Letting the effect of all the other variables besides TSP
4-70
-------
be equal to OQ and a^ in the WAGE and HOURS equations, respectively, the
result of this substitution is equal to:
WAGE - -0.0726(TSP) + aQ (4.9)
HOURS = -0.5389(TSP) + ax (4.10)
From Equations (4.9) and (4.10), the effect of a change in TSP on the WAGE
and HOURS is equal to:
3WAGE/3TSP = -0.0726* (4.11)
3HOURS/3TSP = -0.5389 (4.12)
Equation (4.11) represents the effect of a change in TSP on the wage via
the impact of chronic illness. It states that for a 1 jig/m change in TSP,
the wage will change by approximately 0.07 cents per hour. Note that a
change in acute illness does not affect the wage. Equation (4.12), on the
other hand, represents the effect of a change in the level of TSP on the
number of hours worked. It states that a 1 jig/m change in TSP will result
in a 0.5389 hour change in the annual number of hours worked. It is
comprised of four parts: the effect on HOURS via the impact of chronic
illness (-0.4589), the effect on HOURS via acute illness (-0.0461), the
effect on HOURS via the impact of chronic illness on acute illness
(-0.0082), and the effect on HOURS via the impact of chronic illness on the
wage (-0.0257). Note that acute illness accounts for only a 0.046 hour
reduction in the annual number of hours worked. The remainder of the
effect on hours (0.4928) is due to chronic illness.
From Equations (4.11) and (4.12), the total change in labor produc-
ty of the ith in<
illness is equal to:
tivity of the i individual from a change in TSP via the impact on chronic
* The change in wage is expressed in 1970 dollars.
4-71
-------
ALABOR PRODUCTIVITY = g ' WAGE- + ' HOURS£
= 0.4928 ' WAGE-^ + $0.000726 ' HOURSi (4.13)
where WAGE- = the marginal hourly wage rate of i.
HOURS. = i's annual hours working for money.
Similarly, the total change in labor productivity of the i indivi-
dual from a change in TSP via the impact on acute illness is equal to:
ALABOR PRODUCTIVITY = — ' WAGE- + ' HOURS-
- 0.046 ' WAGE£ + 0 (4.14)
Equations (4.13) and (4.14) can be used as point estimates of the
labor productivity effects of TSP via the impact of chronic and acute
illness, respectively. Minimum and maximum values of the possible ranges
of effects can be obtained from a 95 percent confidence interval for the
TSP coefficients in the chronic and acute illness equations.* The range of
labor productivity effects are given in Table 4-12. (For purposes of the
benefits calculations, the estimated effects on wages axe stated in terms
of 1980 dollars.**) It should be stressed again that not much confidence
can be placed in the range of chronic illness effects because of the
definition of the chronic illness variable. In addition, the range of
effects are based on equations that do not include any air pollution
variables besides TSP. Based on the high positive correlation among the
air pollution variables, the range of effects reported in Table 4-12
probably include the labor productivity effects from exposure to all
pollutants.
* Based on a two-tailed t-test.
** Calculated from the Consumer Price Index for All Items of Wage Earners
and Clerical Workers.
4-72
-------
Table 4-12
RANGE OF LABOR PRODUCTIVITY EFFECTS FOR A 1 (ig/m3 CHANGE IN TSP
FROM CROCKER ET AL. (2)
Chronic illness
HOURS
WAGE*
Acute illness
HOURS
Minimum
0.1126
$0.00035
0.0001
Point Estimate
0.4928
$0.001542
0.046
Maximum
0.8729
$0.002731
0.092
* Based on 1980 dollars.
Ostxo (22) —
Ostro has used the 1976 National Center for Health Statistics Health
Interview Survey (HIS) to estimate the relationship between air pollution,
as measured by TSP and sulfates, and acute illness, measured in terms of
work-loss days and reduced-activity days. Information on individual work-
loss days (¥LD) and reduced-activity days (RAD) was obtained in response to
a survey question asking how many days in the last two weeks did illness or
injury prevent one from working or participating in his usual activities.
Since the recall period was rather short, it is likely that the responses
were accurate indicators of the actual number of days acutely ill. Like
any survey, however, there is the possibility that the respondent may tend
to intentionally bias his answers in favor of (or against) what he thinks
the interviewer wants to hear. Consequently, the respondent may attribute
all work-loss days to illness even if they are non-illness related. This
may be a problem in estimating the relationship between air pollution and
work-loss days based on survey data.
The data set used by Ostro in the estimation of the relationship
between air pollution and acute illness consisted of workers and nonworkers
4-73
-------
in 90 medium-sized cities (populations of 100,000 to 600,000). The
response functions regressed a measure of the individual's acute WLD or RAD
against his personal and demographic characteristics. In addition to the
annual arithmetic means of TSP and SO^ for the city in which the individual
resided, these variables included the individual's age and race, the number
of chronic conditions affecting the individual, whether the individual was
married, whether the individual was a blue-collar worker, the family's
income, the annual mean temperature, annual precipitation, and population
density. The exact definition of these variables are given in Table 4-13.
One of the advantages of this study is that the sensitivity of the
pollution coefficients to a variety of functional forms were tested. Three
Table 4-13
VARIABLES USED IN OSTRO (22) ACUTE MORBIDITY STUDY
TSP = annual arithmetic mean in ug/m in city of residence.
SO. = annual arithmetic mean in ug/m in city of residence.
AGE = age in years.
CHRQN - number of chronic conditions.
RACE - 2 if nonwhite; 1 otherwise.
MARR * 1 if married and living with spouse; 0 otherwise.
BLUE - 1 if blue collar; 0 otherwise.
INC = family income in thousands.
DENS = city population density in thousands.
TEMP = city's annual mean temperature.
PRECIP = city's annual precipitation.
SEX - 2 if male; 1 if female.
CIG = number of cigarettes smoked per day per person.
4-74
-------
different types of concentration-response equations were estimated for
work—loss days (WLD). The first one assumed that the concentration-
response equation was linear. When this equation was estimated for workers
between the ages of 18 and 65, TSP had a significant positive impact on
WLD.* The elasticity of TSP in this equation was 0.44 — within the range
of acute illness elasticities reported by Crocker et al. SO^, however, was
insignificant and implausibly signed.
The second type of concentration-response equation was estimated using
a Tobit model. This model is superior to the linear model since it con-
strains the number of WLD to be non-negative. However, the first and
second derivatives of the Tobit model are positive, implying that the
effect of pollution on WLD increases at an increasing rate, which does not
conform to some of the tozicological evidence on morbidity. In this model,
TSP was positive and significant while SO* remained insignificant.
The third technique used by Ostro was a Logit-linear model. In the
first stage, the logit model transformed the dependent variable into a
probability— the probability of having a WLD during the survey recall
period. This model has two advantages: it constrains the dependent vari-
able to be non-negative and is capable ,of yielding a logistic
concentration-response curve which conforms to some of the existing tozico-
logical evidence. In simple form, the logit concentration-response model
estimated by Ostro was:
(l
-1
(4.15)
where Pw - the probability of having a work-loss day (WLD).
.1 = vector of personal and demographic characteristics
affecting Pw.
b = vector of coefficients of X.
* Significant at the 5 percent level of a two-tailed test.
4-75
-------
In this equation, TSP had a positive impact on the probability of a WLD but
was not significant at the 10 percent level. The coefficient of sulfate
remained insignificant.
After the relationship between the probability of a WLD and air pollu-
tion had been estimated, Ostro then tested whether TSP affected the length
of a work-loss episode. In this equation, Ordinary Least Squares (OLS)
were used to estimate the relationship between air pollution and the number
of WLD, given that the individual had at least one WLD. The general form
of this estimated equation is:
WLD = OQ + Ib (4.16)
where WLD is constrained to be greater than zero. In this equation,
neither TSP nor sulfates affected the length of a work-loss episode.
Because a significant portion of work-loss days may be unrelated to
pollution-induced illness, Ostro also estimated concentration-response
equations for two subsets of work-loss days: 1) work-loss days net of
injuries and illnesses clearly unrelated to air pollution (e.g., diabetes
and gout); and 2) work—loss days associated with circulatory and respira-
tory illnesses. Under the assumption that certain age groups may be more
susceptible to air pollution than the general population, Ostro also
estimated concentration—response equations for specific age groups of
workers. The results of the concentration-response equations using work-
loss days net of injuries and unrelated illnesses (WLD2) for the 18 to 44
and the 45 to 65 age groups are reported in Tables 4-14 and 4-15.*
As can be seen in Table 4-14, neither TSP nor S04 has a significant
impact on the probability of having a WLD2 in the 18 to 44 age group. Of
the other independent variables included in the logit equation, only CHRON,
* The information on the number of work-loss days due to acute circulatory
illness that is necessary to calculate benefits is not available. Conse-
quently, the discussion is limited to work-loss days net of injuries and
non-pollution related illnesses (WLD2).
4-76
-------
Table 4-14
ESTIMATION OF WLD2* FOR WORKERS AGED 18-44
Equation
CONSTANT
TSP
so4
CHRON
AGE
INC
MARR
RACE
TEMP
BLUE
DENS
PRECIP
SEX
CIG
4-14.1
LOGIT
Estimation**
-2.76
-0.00528
-0.027
0.448
-0.0048
-0.0075
0.146
0.032
-0.016
0.105
0.014
0.02
0.361
-0.0088
t-
Statistic
-3.45
-1.47
-0.76
2.99
-0.53
-1.04
1.03
0.16
-1.14
0.74
0.88
2.00
2.69
-0.10
4-14.2
OLS
(WLD2 1 1)
-3.55
0.0188
-0.005
0.48
0.073
0.0148
-0.049
0.37
0.036
-1.6
0.009
0.0074
0.55
0.009
t-
Statistic
-1.64
1.92
-0.06
1.2
3.17
0.87
-0.14
0.74
1.0
-1.2
0.23
0.26
1.53
0.45
X2 27.9
F 2.7
* WLD2 equals the number of work-loss days net of injuries and
nonpollution-related illness.
** The dependent variable in the logit estimation is the probability that
WLD2 is greater than zero.
4-77
-------
Table 4-15
ESTIMATION OF WLD2* FOR WORKERS AGED 45-65
Equation
CONSTANT
TSP
so4
CHEON
AGE
INC
MARR
RACE
TEMP
BLUE
DENS
PRECIP
SEX
CIG
4-15.1
LOGIT
Estimation**
-6.28
0.012
-0.014
0.53
-0.018
-0.018
-0.09
-0.08
0.06
0.53
0.057
-0.017
0.27
-0.0088
t-
Statistic
2.61
-0.28
4.42
1.13
•1.88
0.41
0.28
3.53
2.65
2.48
-1.55
1.42
-0.67
4-15.2
OLS
(WLD2 1 1)
1.76
-0.006
0.187
0.88
0.015
-0.0087
-1.29
0.08
0.066
-0.32
-0.07
-0.002
-0.75
0.03
t-
Statistic
0.33
1.04
1.69
0.26
' 0.24
1.63
0.07
1.0
0.39
0.78
0.05
1.07
0.65
X2 • 54.7
F 0.99
* WLD2 equals the number of work-loss days net of injuries and
nonpo11ution-related illness.
** The dependent variable in the logit estimation is the probability that
WLD2 is greater than zero.
4-78
-------
PRECIP, and SEX had a significant effect on the probability of a work-loss
day. The coefficients of INC, MARK, RACE, BLUE, and DENS conformed to a.
priori expectations but were not significant. CIG did not appear to signi-
ficantly affect the probability of WLD2. In all of the concentration-
response equations that were estimated, CIG never had a significant impact
on the probability of a work-loss episode or the length of an episode.
This curious result may reflect the inability of the smoking variable to
capture the long—term smoking habits of the individual.
Although TSP did not affect the probability of having a work-loss day
for the 18 to 44 age group. Equation 4-14.2 indicates that TSP does affect
the length of a work-loss episode. The elasticity of the length of a work-
loss episode with respect to TSP in this equation is 0.527, indicating that
the length of an incident is somewhat sensitive to the level of TSP. As
illustrated by Equation 4-14.2, the coefficient of SO^ was insignificant.
The equation indicates that older workers have longer work-loss episodes
and blue collar workers have shorter work—loss episodes. The remainder of
the independent variables in this equation, however, were not significantly
different from zero.
Turning to Table 4-15, TSP had a significant effect on the probability
of having a WLD2 in the 45 to 65 age group. Evaluated at the mean levels
of the variables, the elasticity of the probability of WLD2 with respect to
TSP was 0.87.* This is significantly higher than the elasticity of 0.21
reported for the logit equation estimated for all workers using the
broadest definition of work-loss day as the dependent variable. Again, SO,
was implausibly signed and insignificant. The coefficients of CHRON, INC,
TEMP, BLUE, and DENS were significantly related to the probability of WLD2.
The coefficient of AGE, surprisingly, was insignificant.
Equation 4-15.2 of Table 4-15 indicates that neither TSP nor S04
affect the duration of a work-loss episode given that an episode occurs.
* This elasticity increased to 0.91 when the probability of a work-loss day
due to respiratory and circulatory illness was used as the dependent
variable.
4-79
-------
In fact, the F-test for this equation indicated that none of the indepen-
dent variables in this equation were significantly different from zero.
In order to test the effect of TSP on the nonirorking population, Ostro
also estimated a linear concentration-response function using the number of
reduced activity days (RAD) of nonworkers as a dependent variable. The
results of this estimation are reported in Table 4-16. As shown in the
table, TSP has a significant effect on the number of RAD for nonworking
people. The coefficients of CHRON, AGE, INC, HARR, and DENS were in
accordance with ji priori expectations and were significantly different from
zero. The remainder of the independent variables in this equation were not
significant.
The sensitivity of the air pollution variables were tested by con-
sidering various subsamples and other explanatory variables. Ostro
reported that the air pollution coefficients increased for the subsamples
of male non-smokers aged 45 to 65 and for male non-smokers with chronic
illness conditions. The addition of other explanatory variables were
reported not to have an effect on the air pollution coefficients. In order
to test the possibility that TSP was proxying for occupational exposure, a
linear concentration-response equation was estimated for housewives. If
TSP was a proxy for occupational exposure, the TSP coefficient in this
equation would tend to be insignificant. The estimated coefficient of TSP
in the "housewives" equation was significant at the 0.025 level, indicating
that TSP is probably not proxying for occupational exposure.
Several comments can be made regarding the acute illness model
developed by Ostro. Although the model is able to control for many of the
personal and demographic characteristics that affect individual illness, it
is not able to control for all of these characteristics. For example,
other air pollutants besides particulates and sulfates are not specifically
controlled for in the analysis. As mentioned previously, this can be a
problem when the omitted variables are correlated with the included air
pollution variables.
4-80
-------
Table 4-16
ESTIMATION OF RAD FOR ALL NONWORKERS
Equation
CONSTANT
TSP
so4
CHRON
AGE
INC
MARR
RACE
TEMP
DENS
PRECIP
SEX
CI6
4-16.1
Coefficient
-0.48
0.0028
-0.0079
1.241
0.003
-0.007
0.115
0.121
0.004
0.013
0.005
0.067
0.039
t-Statistic
-1.82
2.39
-0.64
33.31
2.90
-2.87
2.01
1.87
0.87
2.09
1.61
1.47
0.37
R2 0.1115
F 137.41
4-81
-------
A problem arises in the use of income as an explanatory variable in
the concentration-response equation since income (INC) and work-loss days
may be determined simultaneously. If this is the case, the resulting
parameter estimates in Ostro's concentration-response equation are biased
and inconsistent. Although Ostro did not examine the possibility of a
simultaneous relationship between INC and WLD, Crocker et al. found that
their measure of work-loss days, ACUT, did not affect the wage rate but had
a small impact on the number of hours worked (HOURS).
Since Ostro examined the effect of air pollution over a 2-week period,
the use of the annual averages of TSP and SO^ may be inappropriate measures
of the individual's air pollution exposure during this time. For example,
the probability of a WLD during a 2-week period may be better explained by
the peak ambient air pollution levels during this period. The use of
annual averages may thus tend to mask the occurrence of these peaks and
underestimate the exposure of the individual. If the annual average and
peak readings of TSP are not correlated, the coefficient of TSP in Ostro's
equations will measure only the effect of chronic TSP exposure on the
probability of having an acute WLD. If these two measures are correlated,
however, the coefficient of the annual average of TSP will pick up some of
the effects of acute exposure to TSP (measured by the peak reading) and
will consequently overestimate the acute WLD effects of the annual mean.
In this case, some of the effects of acute exposure will be attributed to
chronic exposure.
Keeping in mind these qualifications, the results of this study can be
used to estimate the change in acute illness resulting from a change in
TSP. For workers in the 18 to 44 and 45 to 65 year old age groups, the
change in acute illness, as measured by WLD2, will be calculated by taking
the partial derivative of the expected value of WLD2. This derivative is
equal to:
3E(WLD2) 3PW2 . 3N . W2 ,,
3TSP 3TSP 3TSP i*
4-82
-------
where E(WLD2) = expected number of WLD2.
PW2 = probability of a WLD2.
N = the duration, in terms of days, of each WLD2 episode.
As previously illustrated, TSP did not have a significant effect on
the probability of workers between the ages of 18 and 44 having a work-loss
episode (i.e., 3Pf2/3TSP = 0). However, TSP did have a significant
impact on the duration of a WLD2 episode. Consequently, the partial
derivative of the expected value of WLD2 for this age group is equal to:
aB(!LP2) _ _3N_ W2
3TSP ~ ° + 3TSP P
dN/dTSP can be found by taking the partial derivative of Equation
(4.16) with respect to TSP. Based on the estimated results for the 18 to
44 age group, this partial derivative is equal to 0.0188. Based on the
sample's mean probability of a WLD2 episode of 0.0548, this indicates that
for a 1 jig/m change in TSP, the expected value of WLD2 in a 2-week period
will change by 0.001. The minimum and maximum of the range of the changes
in the expected value of WLD2 given a 1 jig/m change in TSP are equal to -
0.00002 to 0.0021.*
In a similar manner, the effect of a change in TSP on the work-loss
days of the 45 to 65 age group can be calculated. Since TSP was not
significantly related to the length of a work-loss episode ON/dTSP = 0)
for this age group, the effect of a change in TSP on the expected value of
WLD2 is equal to:
aE(¥LD2) = IP^. • N + o (4 19)
9TSP 3TSP N ° (4'19)
* Obtained by placing a 95 percent confidence interval around the point
estimate.
4-83
-------
3PW2/3TSP can be calculated by taking tie partial derivative of
•
Equation (4.15) with, respect to TSP. This is equal to:
>W2 ' + -xb^2
' O
>)•
where b^ - the coefficient of TSP in the logit model, and the
other variables are as defined before.
Evaluated at the mean levels of the independent variables. Equation
(4.20) is equal to 0.00048. This means that for a one-unit change from the
mean annual arithmetic average of TSP in Ostro's sample of workers between
the ages of 45 and 65 (78.6 ug/m ), the probability of having a work-loss
day changes by 0.00048. The minimum and maximum values of the possible
changes in the probability of having, a WLD2 will be obtained by using two
standard deviations around the TSP coefficient reported in Equation 4-15.1
of Table 4-15.* Evaluated at the sample mean level of TSP, the minimum and
maximum are equal to 0.00007 and 0.00153, respectively. It should be
mentioned that since the model is nonlinear in the independent variables,
the partial derivative will change as the level of pollution changes.
Evaluated at the mean values of Ostro's logit equation and assuming
that the average number of days lost per acute illness episode of workers
in this age group is equal to 1.75 days,** a 1 ug/m change in the annual
arithmetic average of TSP will result in a change of 0.0008 days worked in
a 2-week period. Based on the range of two standard deviations around the
point estimate, this change ranges from 0.00012 to 0.0027 hours.
Finally, the effect of a change in the annual arithmetic average of
TSP on the acute illness of the nonworking population can be calculated by
* This range is technically not a confidence interval since the covariance
matrix must be used to calculate the confidence interval of a nonlinear
point estimate.
** Calculated from information contained in Reference (23).
4-84
-------
evaluating the partial derivative of the linear concentration-response
equation:
3RAD A .
3TSP = a (
where a = the estimated coefficient of TSP in the concentration-
response equation for RAD.
Based on Equation 4-16.1 in Table 4-16, Equation (4.21) is equal to
0.0028. This indicates that a 1 (ig/m3 change in TSP will result in a
0.0028 change in the number of RAD during a 2-week period. A 95 percent
confidence interval around this point estimate results in a change in RAD
ranging from 0.0005 to 0.0051.
Table 4-17 summarizes the range of the changes in the number of WLD
a
and RAD in a 2-week period resulting from a 1 ug/m change in the annual
arithmetic mean of TSP.
Acute Morbidity Effects of Chronic Exposure to TSP —
In this subsection, two studies estimating the acute illness effects
(i.e., work-loss days and reduced-activity days) of chronic exposure to TSP
have been reviewed. Using different data bases and functional forms, both
of these studies have found that the annual average of TSP has a signifi-
cant impact on acute illness. As shown in Table 4-18, both of these
studies estimate the relationship between acute illness and ambient levels
of TSP that are currently in existence and can therefore be used to
estimate the acute illness effects of changes in the- ambient level of TSP
that are being considered in this analysis.
Table 4-19 compares the labor productivity effects (i.e., the change
in the number of hours worked) obtained by these studies evaluated at the
1978 mean of the annual arithmetic averages of TSP for the counties
4-85
-------
Table 4-17
CHANGE IN WLD AND RAD FOR A 1 jig/m3 CHANGE IN TSP ESTIMATED
FROM OSTRO (22)*
Mini in tun
Point Estimate
Maximum
Workers
Age 18-44
Age 45-65
Nonworkers
-0.00002
0.00012
0.0005
0.001
0.0008
0.0028
0.0021
0.0027
0.0051
Estimates are for a 2-week period.
Table 4-18
TSP LEVELS USED IN ACUTE ILLNESS STUDIES
(in fig/m3)
Crocker et al. (4)
Ostro (22)
Mean
95.53 (AGM)«
77.9 (AAM)**
Standard
Deviation
18.94
42.8 - 150"1
* Annual geometric mean.
** Annual arithmetic mean.
+ Range of TSP levels used in analysis.
4-86
-------
Table 4-19
COMPARISON OF LABOR PRODUCTIVITY EFFECTS FROM ACUTE ILLNESS
OBTAINED BY CROCKER ET AL. (4) AND OSTRO (22)
Study
Crocker it al. (4)
Ostro (22)**
Age 18-44
Age 45-65
Change in Annual Hours Worked*
Minimum
0.0001
-0.004
0.021
Point Estimate
0.046
•
0.206
0.140
Maximum
0.092
0.417
0.437
* Labor productivity effects are calculated based on a 1 jig/m change from
an annual average TSP of 67.064.
** These estimates are based on the assumption that 25 tiro-week periods are
worked in a year.
considered in this analysis. As can be seen in the table, the change in
the annual number of hours worked estimated by Crocker et al. is substan-
tially smaller than that estimated by Ostro. The difference between these
estimates may be due to a number of factors. The model estimated by
Crocker e_t ,§_!. is based on household data that include people who are not
employed. In estimating the acute illness equation, these people are
assigned zero hours of acute illness. Consequently, the relationship
estimated by Crocker et al. may tend to underestimate the true effect of
TSP on the acute illness of people who are employed. Ostro, on the other
hand, includes only working individuals when estimating his acute illness
concentration-response equation. Another reason for the difference may be
that Ostro's measure of acute illness for workers excludes injuries and
illnesses that are unrelated to pollution. The acute illness measure used
by Crocker et al. reflects illness from all causes.
4-87
-------
In addition to the differences in samples _and variable definitions,
each of these studies uses different models and functional forms to
estimate the relationship between TSP and acute illness. Crocker et al.
estimate a recursive system of equations which ultimately examines the
impact of TSP on the supply of labor (measured in terms of the number of
hours worked). One of the equations in this system is a linear
concentration-response equation which indicates that the change in acute
illness is constant over the entire range of TSP. Ostro measures the
effect of TSP on the number of work days lost due to acute illness.
Ostro's model is superior with respect to functional form because it is
consistent with some of the tozicological evidence that suggests that the
relationship between acute illness and TSP may be nonlinear. Unlike the
Crocker et al. model, the Ostro model, however, is not designed to examine
the effect of changes in acute illness on the wage or the effect that
changes in acute illness may have on the incentive to work.
For the purposes of this analysis, both the Crocker et al. and the
Ostro studies will be used to estimate the benefits of reductions in acute
illness resulting from decreases in the annual average of TSP. Based on
the fact that the inclusion of nonworkers in the Crocker et al. acute
illness equation tends to underestimate the effect of TSP on workers, the
acute illness effects estimated by Crocker et al. will be used in this
analysis to calculate the minimum of the range of benefits. In order to
reflect the range of uncertainty inherent in the point estimate of the
relationship between acute illness and TSP, the minimum of the 95 percent
confidence interval around the point estimate of Crocker et al. will be
used as the minimum of the range of benefits. Ostro's point estimates for
the two age groups of workers and the maximum of the 95 percent confidence
intervals around these point estimates will be used, respectively, as the
point and maximum of the range of benefits associated with particulate
matter reductions. The range of acute illness effects on workers that will
be used in this analysis are summarized in Table 4—20.
The Ostro study can also be used to estimate the change in the acute
illness of nonworkers resulting from a change in TSP. Based on a 95
4-88
-------
Table 4-20
RANGE OF EFFECTS OF A 1 (ig/m3 CHANGE IN TSP ON ACUTE ILLNESS
Households
Workers Age 18-44
Workers Age 45-65
Change in Annual Hours Worked*
Minimum
0.0001
Point Estimate
0.206
0.140**
Maximum
0.417
0.437**
* For consistency, Ostro's estimates have been converted from days to
hours.
** The labor productivity effects of a 1 (ig/m change in TSP for the 45 to
65 age group will change as the level of TSP changes.
percent confidence interval, the range of the effect of a 1 ug/m change in
TSP on the acute illness of nonworkers is summarized in Table 4-21.
Table 4-21
EFFECT OF A 1 ug/m3 CHANGE IN TSP ON THE ACUTE ILLNESS OF NONWORKERS*
Change in the No. of
Reduced Activity Days
(RAD) in a 1-Yr. Period
Minimum
0.013
Point Estimate
0.073
Maximum
0.133
* These estimates are based on 26 two-week periods in a year.
4-89
-------
Chronic Morbidity Effects of Chronic Exposure —
•
The chronic morbidity effects estimated by Crocker et al. will be used
to estimate the benefits of reductions in chronic illness resulting from
decreases in the ambient level of TSP. The estimates that will be used in
these calculations are summarized in Table 4-22. The range of effects
reported in the table reflects the confidence interval around the point
estimate.
APPROACH TO BENEFIT ESTIMATION
Air Quality Data
In this subsection, the range of health effects developed in the
preceding subsections will be used to estimate the health benefits of
alternative reductions in the ambient level of particulate matter. An
issue arises, however, in using the results of previous studies to calcu-
late the benefits of alternative particulate matter reductions. This issue
involves the comparability between the measures of particulate matter used
in these health studies and the measures of particulate matter being con-
sidered in evaluating alternative air quality standards. This compara-
bility is comprised of four parts: 1) the type of particulate matter
measured; 2) the averaging time used to measure the ambient level of
Table 4-22
LABOR PRODUCTIVITY EFFECTS RESULTING FROM A 1 ug/m3 CHANGE IN TSP
HOURS
WAGE*
Minimum
0.1126
$0.00035
Point Estimate
0.4928
$0.001542
Maximum
0.8729
$0.002731
* Expressed in 1980 dollars.
4-90
-------
particulate matter; 3) the monitor or monitors used to represent a popula-
tion's exposure to particulate matter; and 4) the relationship between the
ambient air quality levels used in the studies and the levels being con-
sidered in this analysis. Obviously, it is desirable to have the air
pollution values used in the studies and the air quality standards under
consideration be as comparable as possible with respect to each of these
attributes.
With respect to the first attribute, all of the studies reviewed in
this section measure ambient particulate matter in terms of total suspended
particulates (TSP). Some of the ambient particulate matter standards being
considered in this analysis are stated in terms of PM10. PM10 is particu-
late matter whose aerodynamic diameter is less than or equal to 10 urn.
Fortunately, data on an approximately equivalent level of TSP that will
result from PH10 controls are available (see Section 9).
Table 4-23 lists the alternative standards that will be considered in
estimating the health benefits associated with particulate matter'reduc-
tions. Column 1 of the table reflects whether the standard is stated in
terms of PH10 or TSP. Columns 2 and 3 report the averaging times
associated with each of the alternative standards. Column 4 of the table
reports the implementation date associated with each standard. As the
table indicates, only two of the standards are stated in terms of TSP.
Some of the standards being considered in this analysis are stated in
terms of both the annual arithmetic average of particulate matter and the
24-hour expected value of particulate matter that is expected to occur once
a year (for the TSP standard, the 24—hour maximum value not to be exceeded
more than once a year). When the standard is stated in terms of both the
annual average and 24—hour expected or maximum value, the averaging time
that is more stringent will be used to calculate benefits. Since all of
the studies considered in this section examine the relationship between
health status and the annual average of TSP, it is necessary to convert the
4-91
-------
09
H
09
n "o in
p~ •'-'
ol
•u
o o
•* a
w
•w o
O 0
§ -3
0 >
S 5
rt M
a at
4-92
-------
24-hour expected or maximum value to an equivalent annual average* when the
most stringent averaging time is the 24-hour expected or maximum value.
With respect to the second attribute, the health studies that are
being used to calculate benefits in this section use tiro different annual
averaging times in order to measure the health effects of exposure to
particulate matter. Lave and Seskin (6) and Ostro (22) use the annual
arithmetic mean of TSP in their concentration—response equations, while
Crocker et al. (4) use the annual geometric mean to estimate the health
effects of exposure to particulate matter. Consequently, when estimating
the health benefits of PM10 standards based on the Crocker et al. study,
the annual arithmetic mean of PM10 must be converted to an equivalent
annual geometric mean. The details of this conversion are discussed in
Section 9.
The third attribute requires that the monitors used to estimate the
benefits of reductions in particulate matter must be as comparable as
possible to the monitors used in the original studies. Table 4-24 lists
the types and location of the monitors used in the health studies that are
being used in this section. Also contained in this table is a list of the
types of monitors that will be used in conjunction with the results of
these health studies in order to calculate the health benefits associated
with particulate matter reductions.
As Table 4-24 indicates. Lave and Seskin (6) used the center-city
monitors in an SMSA to represent the air pollution exposure of the indivi-
duals residing in an SMSA. In general, a single monitoring station was
used to represent SHSA exposure. These center-city monitors were generally
part of the National Air Surveillance Network (NASN) and tended to repre-
sent the worst air quality in a region. Consequently, when using the Lave
and Seskin results to estimate health benefits, it is appropriate to use
the monitor that represents the worst air quality in a region. The design
value monitor in each county typically records the worst air quality among
See Section 9 for details of this conversion.
4-93
-------
Table 4-24
AIR POLLUTION MONITORS USED IN HEALTH STUDIES
Study
Monitor(s) Used
in Study
Monitor Used to
Calculate Benefits
Lave and Seskin (6)
Crocker et al. (4)
Ostro (22)
Center city monitor(s)
in each SMSA
County monitor having
most complete data
between 1967-1975
Avg. of all popula-
tion oriented monitors
in a city
Design value
Design value
Avg. over monitors
within a county
all monitors in a county (see Section 9). Thus, the design value monitor
will be used to estimate those health benefits based on the Lave and Seskin
results.
Crocker et al. (4) measured county exposure to TSP using the county
monitor having the most complete data between 1967 and 1975. We have
assumed that the NASN center city monitors have the most complete data over
that period. Thus, as mentioned above, the design value monitor will be
used when the results of Crocker et al. are used to estimate the benefits
of particulate matter reductions.
Ostro (22) used the average of all population-oriented monitors within
a city to represent the exposure of the individuals residing in a city.
Approximately 73 percent of all 1978 TSP monitors which meet EPA summary
criteria are population-oriented surveillance monitors. When calculating
health benefits based on Ostro's results, the average of all monitors
within a county will be used.
4-94
-------
Finally, the levels of particulate matter used in the health studies
must be similar to the ambient levels that are being used to calculate
benefits. Table 4-25 lists the mean level and standard deviations of TSP
used in the health studies being used to calculate benefits in this
section.
Because of the wide range of TSP levels used in the Lave and Seskin
study, mortality benefits will be calculated over the entire range of
particulate matter levels found in the counties included in this analysis.
Morbidity benefits based on the Crocker et al. study will be limited
to those TSP levels found in the study. This limitation is based on the
uncertainty expressed by Crocker e_t a_l. regarding the validity of their
results outside of the sample means. The complete range of particulate
Table 4-25
TSP LEVELS USED IN HEALTH STUDIES
(in ug/m3)
Mortality
Lave and Seskin (6)
1960 data
1969 data
Morbidity
Crocker .et al. (4)
Ostro (22)
Mean
118.15 (AAM)*
95.58 (AAM)
95.53 (AGM)**
77.9 (AAM)
Standard
Deviation
40.94
28.64
18.94
42.8 - 150+
* Annual arithmetic mean.
** Annual geometric mean.
+ Range of TSP levels used in analysis.
4-95
-------
matter levels used in the Crocker et al. analysis is not available; hence,
a range that reflects two standard deviations around the reported mean will
be used. Consequently, only counties having annual geometric means of TSP
from 57.6 to 133.4 will be used to calculate morbidity benefits based on
this study.
Because the nonlinear concentration-response equation estimated by
Ostro for workers between the ages of 45 and 65 conforms to some of the
toxicological evidence, the entire range of TSP levels will be used to
calculate benefits for this age group of workers.
The acute illness equations for workers aged 18 to 44 and nonworkers
were estimated as linear relationships. Because the validity of the
estimated relationship outside the sample mean is questionable, only the
range of particulate matter levels considered in the Ostro analysis will be
used in calculating benefits for this category. These levels range from
42.8 to 150 |ig/m3 annual arithmetic mean of TSP.
The unit of observation that will be used to calculate benefits is the
county. Predicted particulate matter levels prior to the controls imple-
mented to attain the alternative standards listed in Table 4-23 will be
used as baseline pollution levels. The change in pollution under each
standard can thus be calculated by taking the difference between the
predicted level of particulate matter before control and after control.
For example, assuming that the annual average under the first standard
listed in Table 4-23 is the more stringent averaging time, and that this
standard can be attained in county i, the change in particulate matter
levels (PM10) in county i under this standard is:*
* Because of the strategies used to control particulates, it is possible
that the ambient level of particulate matter may be reduced to a level
that is beneath the standard. In no case, however, is the ambient level
of particulate matter permitted to go beneath the background level.
4-96
-------
APOLj - POL89. - 70 (4.22)
where APOL- = the change in PH10 in county i.
= the predicted level of PM10 in county i prior to
control.
Calculations for the changes in PM10 under each standard listed in
Table 4-23 for each county in the analysis will be done similarly.*
Given the reduction in particulate matter under these alternative
scenarios, the health benefits of these reductions can be developed from
the range of health effects developed in previous subsections. The calcu-
lations that follow express these health benefits in yearly terms under the
assumption that a given reduction in particulate matter is maintained
throughout the year.**
Mortality Effects of fTtfonic
The effect of a change in the annual arithmetic average of PM10 and
TSP in county i on the annual mortality rate in county i will be calculated
according to:
(4.23)
where AMR^ = the change in the annual mortality rate in county i
(i.e., change in the mortality risk per 100,000 people).
* Only counties that have reliable particulate matter monitoring data are
included in the analysis. Under the most stringent scenario, 499 of the
519 counties considered in this analysis will have changes in the level
of particulate matter.
** Data sources and explanations of the data transformations used in the
benefit calculations are provided in Appendix 4A.
4-97
-------
A
b - the range of coefficients taken from the chronic
mortality studies.*
and APOLi is as defined before.
Equation (4.23) indicates the change in the mortality risk per 100,000
in county i. The dollar benefits of this change Till be approximated using
the estimates of the willingness to pay for a marginal reduction in
mortality risk developed in the Appendix to Volume II. This is equal to:
AMR.^ • VMR (4.24)
where VMR = value of a 1/100,000 change in mortality risk.**
Morbidit Effects of
The morbidity benefits associated with reductions in the ambient level
of particulate matter that will be considered in this section can be broken
down into three categories;
• Reductions in illness occurring during work time (WLD).
• Reductions in illness occurring during leisure time (RAD).
• Reductions in direct medical expenditures associated with
reductions in illness (DHE).
Both the Crocker £i al. (4) and Ostro (22) studies have estimated the
effects of particulate matter on illness that prevents one from working.
In addition, Ostro has estimated the effect of particulate matter on the
illness of nonworkers. Neither of the studies, however, has specifically
* Minimum coefficient = 0.00; point coefficient - 0.171; maximum coeffi-
cient = 0.471.
** The marginal willingness to pay for a one unit change in mortality risk
(i.e., a change in risk of 1/100,000) is equal to: Minimum = $3.60;
Point = $15.80; Maximum = $28.00.
4-98
-------
measured the effects of particulate matter on the illness of workers that
occurs during their leisure time or the effects of particulate matter on
the direct medical expenditures associated with illness.
In order to provide the broadest coverage of the morbidity benefits
resulting from reductions in particulate matter, the effects of particulate
matter on these categories will be approximated from the results of the
Crocker et al. (4) and Ostro (22) studies. For the purposes of this
analysis, it is assumed that a percentage reduction in illness estimated
from these studies will result in the same percentage reduction in the
morbidity categories not considered in these studies. For example, if a 5
percent reduction in acute WLD is estimated to result from a 20 percent
reduction in the ambient level of particulate matter, it will be assumed
that this reduction in particulate matter will also result in a 5 percent
reduction in workers' DUE on acute illness.
The details of these calculations are outlined below.*
Effects on Acute Morbidity —
Miaiaui Estimate — As previously mentioned, the minimum of the range
of acute morbidity effects is based on the analysis by Crocker et al. (4).
These effects consist of the following categories:
• Effects on WLD
- (XAPOLj) • HHj (4.25)
where AHOURS^ = the change in the annual number of hours worked in
the i county.**
* In all cases, the reductions in RAD and DME associated with WLD
reductions are constrained to be less than or equal to 100 percent.
** The change in WLD can be obtained by dividing AHOURS^ by 8.
4-99
-------
A
C - the minimum of the 95 percent confidence interval
of acute illness coefficients (i.e., 0.0001).
HH. = the number of households in county i.
Equation (4.25) estimates the change in the annual number of hours
worked in county i resulting from the impact of a change in particulate
matter on acute illness. Because only household heads are considered in
the Crocker et al. analysis, the change in work hours estimated from
Equation (4.25) is limited to households. This probably results in an
underestimate of the total change in work hours that results from a given
reduction in particulate matter.
The benefits (PROD^) of this change will be estimated by multiplying
the mean hourly wage in county i (WAGE^) by the change in the total number
of hours worked:
PRODi * WAGE^ ' AHODRSi (4.26)
• Effects on RAD
Because the sample used by Crocker et al. includes both workers and
nonworkers, but does not distinguish between them, it is not possible to
develop a separate estimate of the effect of a reduction in particulate
matter on the acute illness of nonworkers. Another reason why it is not
possible to develop morbidity benefit estimates for nonworkers from the
Crocker et al. study is that data on nonworking household heads are not
available.
It is possible, however, to develop estimates of the reductions in
illness occurring during leisure time by assuming that a given reduction in
particulate matter will result in the same percentage decrease in leisure
time illness as that estimated for "work-time" illness. For example, if
the Crocker et al. analysis indicates that a change in particulate matter
will result in a 1 percent reduction in WLD, it is assumed for the purposes
4-100
-------
of this analysis that this change in particulate matter will result in a 1
percent reduction in RAD.
This percentage change in acute illness is equal to:
% A in RADi
(d/16 ' APOLi)/WLDH
(4.27)
where % A in RAD;
A
d
WLDH
the percentage change in the annual number of
leisure days acutely ill of a household in county
i.
the minimum of the 95 percent confidence interval
of the coefficient of the acute illness
concentration-response equation (i.e., 0.00168)*.
the number of annual work—loss days, per household
due to acute illness (i.e., 4.48).
Using Equation (4.27), the effect of a change in particulate matter on the
annual number of RAD in county i can be calculated. This is equal to:
ARADi » (% A in RAD^ • RADH '
(4.28)
where ARAD. = the change in the annual number of leisure days
acutely ill of the households in county i.
RADH » the annual number of leisure days acutely ill per
household.
and the other variables are as defined previously.
The value of a RAD will be assumed to be 0.5 of the average daily wage
over all of the counties being considered in this analysis. This amount is
* This coefficient is divided by 16 because the dependent variable in the
acute illness concentration-response equation is stated in terms of work
days ill times 16 for the first 8 weeks of acute illness and times 12
thereafter. Since there is no information regarding the distribution
between acute illness less than or greater than 8 weeks, the coefficient
is divided by 16 in order to be conservative.
4-101
-------
equal to $27.76 in 1980 dollars. The benefits associated with the reduc-
tion in the number of RADs are therefore equal to:
RAD BENEFIT^ = ARADi * $27.76 (4.29)
• Effects on DME
For the purposes of this analysis, it is assumed that the percentage
reduction in acute illness that results from a reduction in particulate
matter brings about the same percentage reduction in direct medical expen-
ditures associated with acute illness:
j| APOLJ/WLDH (4.30)
where % ADME^ = the percentage change in direct medical expendi-
tures in county i.
and the definitions of the other variables remain unchanged.
The direct medical expenditure benefits in county i (DME BENEFIT^)
resulting from a change in particulate matter are equal to:
DME BENEFIT = % ADMEi • DMEH • HHi (4.31)
where DMEH - direct medical expenditures on acute illness per
household.
In order to be conservative, DMEH will be limited to direct medical
expenditures associated with acute respiratory and circulatory conditions.
In 1980, these direct medical expenditures were equal to $324.08 per house-
hold in the United States (1980 dollars).
Point and Mmxiavp Estimates — The point and maximum of this range of
effects will be based on the results obtained by Ostro (22). Like the
morbidity effects estimated from the Crocker et_ al. analysis, these effects
will consist of three categories:
4-102
-------
• Effects on WLD
•
Because Ostro estimates separate equations for workers from the ages
of 18 to 44, and workers from the ages of 45 to 65, the effects of a change
in particulate matter on WLD must be estimated separately for the workers
in each of these age groups. Ostro's acute morbidity effects for workers
between the ages of 18 and 44 are expressed in terms of the change in the
duration of an acute work—loss episode (net of injuries and nonpollution-
related illnesses) over a 2-week period due to a change in particulate
matter. In order to express this effect in terms of the change in the
annual number of WLD due to a change in acute illness, the effects
estimated by Ostro must be multiplied by a factor of 25 (25 two-week
working periods in a year). Consequently, the effect of a change in parti-
culate matter on the number of WLD will be calculated according to:
AWLD( 18-44) j_ = e • (APOLi) • PW2 • 25 • WOREBR( 18-44) i (4.32)
where AWLD( 18-44). = the change in the number of work-loss days due to
acute illness of workers between the ages of 18 and
44 in county i.
e » the effect of a unit change in TSP on the duration
of a work-loss episode for workers between the ages
of 18 and 44.*
¥2
P * = the probability of a worker aged 18 to 44 having a
work-loss episode during a 2-week period (i.e.,
0.0548 based on the mean of Ostro's sample).
WORKER(18-44). = the number of workers in county i between the ages
of 18 and 44.**
For the 45 to 65 year old age group, Ostro's morbidity effects are
expressed in terms of the change in the probability over a 2-week period of
* Point estimate - 0.0188; maximum estimate = 0.0380.
** Information on the number of workers between the ages of 18 and 44 is
not available. Because the number of workers under age 18 is likely to
be small, the number of workers between the ages of 0 and 44 will be
used.
4-103
-------
an acute illness episode (net of injuries and nonpollution-related
illnesses) that results in days lost from work. Because Ostro's logit
model is nonlinear, the concentration-response function must be evaluated
before and after the change in particulate matter. This calculation is
equal to:
where
AP
,W
i2
il
"
i2
(4.33)
the change in the probability of a work-loss day
(WLO) during a 2-week period of a worker between
the ages of 45 and 65 in county i that is
associated with a change in POL^.
the probability of a WLD after a change in POL^.
the probability of a WLD before a change in POL^.
Except for the values of particulate matter, both P*2 and P*j will be
estimated using the mean values of the independent variables in Ostro's
sample.
The change in the-annual number of WLD by workers between the ages of
45 and 65 in county i [AWLD(45-65)i] will be estimated from:
AWLD(45-65)i = AP* * EPISODE • 25 * WORKER(45-65)i*
(4.34)
where
EPISODE
WORXERi
The mean number of days lost from work per acute
illness episode (i.e., 1.75).
the number of workers in the i county in the 45
to 65 age group.
* It is necessary to multiply the equation by 25 in order to express
benefits in annual terms.
4-104
-------
Like the labor productivity benefits based on the analysis by Crocker
, •
e_t a_l., the change in the annual number of WLD for each age group will be
valued at the average daily wage in county i (DWAGE^):*
For the 18 to 44 age group:
PROD(18-44)i - AWLD(18-44)i • 0WA6Ei (4.35)
For the 45 to 65 age group:
PROD(45-65)i = AWLD(45-65)i ' DWAGEi (4.36)
• Effects on RAP
Because Ostro estimated separate concentration-response equations for
workers and nonworkers, the effect of reductions in particulate matter on
RAD will be estimated separately for each of these groups.
Workers: It is assumed that a change in particulate matter will bring
about the same percentage change in the RAD of workers as the percentage
change in WLD estimated from Ostro's concentration-response equations for
workers.
Based on Equation (4.32) which calculates the change in the number of
WLD of workers between the ages of 18 and 44, the percentage change in the
work-loss days of this age group [WLD(18-44)i] is equal to:
% AWLD(18-44)i - (e • APOL • PW2 • 25)/WLDW(18-44) (4.37)
where WLDW( 18-44) = the annual number of WLD per worker in the 18 to 44
age group (i.e., 2.5657).
* It would be most appropriate to value the change in hours worked at the
average wage for each of these age groups. This information is not
available; therefore, the average wage of all workers in a county will be
used.
4-105
-------
Since it is assumed that the percentage change in WLD is equal to the
percentage change in RAD, Equation (4.37) will be used to calculate the
change in the annual number of RAD of workers between the ages of 18 and 44
[ARAD(18-44)iL This calculation is equal to:
A8AD(18-44)i = % AWLD(18-44)i ' RADWU8-44)
• WORKER ( 18-44 )i (4.38)
where RADW( 18-44) = the annual number of leisure days acutely ill per
worker aged 18 to 44 (i.e., 3.3375).
and the other variables are as defined previously.
For the employed people between the ages of 45 and 65, the percentage
change in the annual number of WLD is based on Equation (4.34) and is equal
to:
% AWLD(45-65)i - (AP? ' EPISODE • 25) j/WLDWX 45-65) (4.39)
where WLDW( 45-65) = the annual number of WLD per worker aged 45 to 65
(i.e., 2.3118).
Based on Equation (4.39), the change in RAD of workers between the
ages of 45 and 65 [ARAD(45-65)il is equal to:
ARAD(45-65)i = % AWLD(45-65)i • RADW(45-65)
• WOREBR( 45-65 ) (4.40)
where RADW(45-65) = the annual number of leisure days acutely ill per
worker aged 45 to 65 (i.e., 3.0052).
The dollar benefits of these effects will be evaluated at one-half of
the average daily wage of all the counties being considered in this
analysis. *
* As previously mentioned, this is equal to $27.76 in 1980 dollars.
4-106
-------
Nonworkers: Since Ostro specifically estimates a concentration-
•
response function for nonworkers, these results can be used to calculate
the effect of a change in particulate matter on the RAD of nonworkers.
This effect will be calculated according to:*
ARADt « f ' APOLi ' 26 • NONWORZERSi (4.41)
where ARAB. = the change in the annual number of days of acute
illness of nonworkers in county i.
A
f = the coefficient of particulate matter in the acute
illness concentration-response equation of non-
workers. **
NONWORKERS= nonworkers in county i.
As previously mentioned, the dollar benefits of the effects of changes
in particulate matter on RAD will be valued at one-half of the average
daily wage.
• Effects on DME
It is assumed that a given percentage reduction in acute illness will
be accompanied by the same percentage reduction in direct medical expendi-
tures on acute illness. Based on the changes in the annual number of WLDs
and RADs for workers and nonworkers for a given change in particulate
matter, the percentage change in acute illness for all individuals in
county i can be obtained from:
% AAOJTEj = [AWLDC 18-44 )£ + AWIJX45-65) £
+ ARAD( 18-44) j. + ARAD(45-65)i
-44) + WLDW(45-65)
+ RADW( 18-44) + RADW(45-65) + RADP] (4.42)
* Based on 26 two-week periods in a year.
** Point estimate = 0.0028; maximum estimate - 0.0051.
4-107
-------
where % AACUTE^ = tlie percentage change in the number of days of
acute illness in county i.
RADP = the annual number of acute illness days per non-
worker.
and the other variables are as defined before.
The change in direct medical expenditures associated with changes in
acute illness will be calculated according to the following algorithm:
ADMEj = % AACDTEi • DMEP ' POPj (4.43)
where DMEP = direct medical expenditures on acute illness per
person.
POP. = population in the i county.
In order to be conservative,. DHEP will be limited to the direct
medical expenditures associated with acute respiratory and circulatory
conditions. In 1980, these direct medical expenditures were equal to
approximately $113.26 per person in the United States.
Effects on Chronic Morbidity —
The effects of reductions in the ambient level of particulate matter
on chronic illness will be estimated using the results of Crocker et al.
These effects consist of the following categories:
• Effects on WLD
The effect of changes in particulate matter on the labor productivity
in county i will be calculated according to the following algorithms:
AHOURSi - g • (APOL..) * HH^
= h • (APOLi) • HHi (4.44)
4-108
-------
where AHOTJRS^ - the change in annual hours worked by a household
head in county i.
g = the range of chronic illness coefficients
reflecting the relationship between POL- and annual
hours worked.*
A WAGE- = the change in the hourly wage resulting from a
change in POL-.
A
h = the range of chronic illness coefficients
reflecting the relationship between POL^ and the
hourly wage.**
and the other variables are as defined previously.
The dollar benefits of these effects will be calculated according to:
PRODi = wAGEi • AHOURSj + HOURSj • AWAGEj (4.45)
where HOURS- => the annual number of hours worked in county i
(assumed to be 2000 * HH.
• Effects on RAD
In the recursive system of equations developed by Crocker et al.. the
equation that represents the relationship between chronic illness and
particulate matter encompasses all types of chronic illness that limits
work or housework that the head of household can do. Because the defini-
tion of chronic illness is not limited to the amount of time lost from
work, but reflects the total length of time a head of household is
chronically ill, it is not appropriate to provide a separate estimate of
the effects of changes in particulate matter on RAD. In other words, the
chronic illness equation captures the effect of particulate matter on
chronic illness occurring during work and leisure time.
* Minimum coefficient = 0.1126, point coefficient = 0.4928; maximum
coefficient = 0.8729.
** Minimum coefficient = $0.00035; point coefficient = $0.00154; maximum
coefficient = $0.0027.
4-109
-------
As previously mentioned, the sample used by Crocker .ejt .§_!. includes
both workers and nonworkers. Consequently, the system of equations
implicitly includes the effects of particulate matter on the chronic
illness of nonworkers. Since this system is limited to identifying the
labor productivity effects of changes in particulate matter, any changes in
the chronic illness of nonworkers that does not result in an increase in
the number of hours worked will not be measured by the Crocker et al.
analysis. Consequently, these estimates will probably be an underestimate
of the total effect of changes in particulate matter on the chronic illness
of nonworkers.
• Effects on DME
Like the other morbidity studies being used in the calculation of
benefits, it will be assumed that a percentage change in chronic illness
brings about the same percentage change in the DHE associated with chronic
illness. Because the dependent variable in the chronic illness
concentration-response equation is an interval variable that includes an
*
open—ended interval as a maximum value, the hours—worked equation will be
used to proxy the percentage reduction in chronic illness. Based on the
hours-worked equation in Table 4-10 and netting out the effect that changes
in chronic illness have on the number of hours worked through the wage
effect, the percentage reduction in the i county is equal to:
% ACHRONICi = (h/8 APOL^/CDAYH (4.46)
where % ACHRONIC^ = percentage change in the annual number of days of
chronic illness in the i county.
A
h = the range of values used to represent the partial
derivative of annual hours worked with respect to a
change in particulate matter (net of wage effect).*
CDA7H = the annual number of days of chronic illness per
household (i.e., 25.41).
* Minimum estimate « 0.10677; point estimate » 0.46711; maximum estimate
0.8268.
4-110
-------
A
Since h is expressed in terms of hours, it is divided by 8 in order to
express the change in hours worked in daily terms.
The change in direct medical expenditures will be calculated according
to the following algorithm:
% ACHRONIC * CDMEH * EB (4.47)
where ACDME, = change in the direct medical expenditures
associated with chronic illness in county i.
CDMEH = per-household direct medical expenditures
associated with chronic illness.
As before, CDHE will be limited to those expenditures associated with
chronic respiratory and circulatory illnesses in order to be conservative.
In 1980, these expenditures were equal to approximately $198.63 per U.S.
household.
BENEFIT ESTIMATION
The algorithms just discussed can be used to estimate the annual
benefits of reductions in the level of particulate matter. Since the
particulate matter standards being considered in this analysis are to be
attained in the future, the estimates of health benefits should be
expressed in terms of discounted present values. In order to be consistent
with the analysis of the control costs associated with these standards, a
stream of benefits ending in 1995 is assumed. For PM10 standards, this
stream is assumed to start in 1989, while for TSP standards, this stream is
assumed to start in 1987. Using a discount rate of 10 percent and a 1989
attainment date, the discounted present value in 1980 dollars in 1982
4-111
-------
1982
j ) of the benefits in county i estimated in this section are equal
to:*
Annual Benefits^ Annual Benefits.
Dpv1982 „ + >B< + (4.48)
(1.10)8 (1.10)14
This calculation incorporates the following two conventions used in the
cost analysis:
1) Benefits arising during a particular year are all assumed
to occur on the last day of the year.
2) The discounted present value is calculated at the beginning
of 1982.
Aggregate Benefits
The aggregate benefits of reductions in the level of particulate
matter in each county will be obtained by summing over all counties
experiencing a change in particulate matter:
Aggregate Benefits = E DPV^982
* For expositional ease, this calculation assumes that there is no growth
in the variables used to estimate annual benefits. In the actual calcu-
lation of benefits, however, growth rates for population, employment,
households, and the real wage have been used. See Appendix 4A.
4-112
-------
The sources of the socioeconomic data used in. the calculation of
benefits are listed in Table 4-26. A detailed explanation of the data
sources appears in Appendix 4A.
Benefits
Tables 4-27 through 4-44 present the health benefits estimated for the
alternative standards listed in Table 4-23. These benefits represent the
benefits that would be achieved when all counties included in the analysis
are in compliance with the standard for all years under consideration.*
The benefits are stated in terms of the discounted present value of
Table 4-26
DATA USED IN CALCULATING BENEFITS
Variable
Source
POPi
HH.
WORKERSi
WAGEi
POPULATION
GROWTH RATE
EMPLOYMENT
GROWTH RATE
REAL INCOME
GROWTH RATE
Current Population Reports Series P-25, No.
873, February 1980 (24).
1980 Census of Population (25).
County Business Patterns 1978 (26).
County Business Patterns 1978 (26).
U.S. Bureau of Economic Analysis, Projections
of the Population 1976-2000 (22)
and 1980 OBERS BEA Regional Projections (23).
U.S. Bureau of Economic Analysis, 1980 OBERS
BEA Regional Projections (23).
U.S. Bureau of Economic Analysis, State
Projections of Personal Income to the Year
2000 (27).
* These benefits represent 'Type B" scenario benefits. See Section 9.
4-113
-------
Table 4-27
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Point
Estimate
0.0
21.3
526.2
696.2
4894.7
1596.9
216.8
457.0
3703.2
610.9
Maximum
0.0
104.2
2568.7
. 3398.3
23892.1
7794.9
1058.3
2230.6
18075.8
2982.1
Total U.S.
12723.3 62104.9
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
5092.9 24859.2
4-114
-------
Table 4-28
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
147.0
263.8
1025.9
1515.3
7648.9
2487.4
519.8
839.8
7284.8
846.4
717.3
1287.8
5007.4
7396.7
37335.7
12141.7
2537.0
4099.4
35558.4
4131.4
Total U.S.
0.0
22579.1 110212.7
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
9037.9 44115.6
4-115
-------
Table 4-29
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
II
X
New England
Jl.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
147.0
263.8
1025.9
1515.3
7669.5
2526.8
521.9
339,8
7285.3
894.3
717.3
1287.8
5007.4
7396.7
37436.4
12334.0
2547.7
4099.4
35560.7
4365.3
Total U.S.
0.0
22689.7 110752.6
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
9082.2 44331.7
4-116
-------
Table 4-30
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Point
Estimate
498.8
456.0
1448.9
1908.3
8330.1
2985.4
721.1
1177.2
8249.9
1530.4
Maximum
2434.8
2225.7
7072.2
9314.7
40660.8
14572.1
3519.9
5746.3
40269.4
7470.3
27306.1 133286.2
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
10930.0 53351.4
4-117
-------
Table 4-31
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Point
Estimate
512.9
584.1
2350.4
2939.9
14204.1
4567.5
1226.3
1495.7
13190.9
1740.2
Maximum
2503.3
2851.3
11472.9
14350.3
69332.9
22294.7
5985.6
7300.7
64387.1
8494.1
42811.9 208972.9
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
0.0
11972.3 58439.2
4-118
-------
Table 4-32
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
1503.1
1071.6
3463.4
3810.2
15581.2
4943.2
2037.5
2078.0
15133.1
2804.6
7336.7
5230.8
16905.5
18598.5
76054.8
24128.8
9945.3
10142.9
73867.4
13689.7
Total U.S.
0.0
52425.8 255900.3
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
0.0
14660.9 71562.5
4-119
-------
Table 4-33
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.1
1.8
2.1
9.4
4.0
0.6
1.4
11.5
1.7
0.0
13.6
519.8
604.7
3191.0
1223.8
174.5
383.6
3956.7
0.0
26.7
1095.8
1253.9
6522.3
2491.3
360.2
811.4
7766.4
585.7
1159.3
Total U.S.
32.5
10653.5 21487.3
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
13.0
4264.4
8600.9
4-120
-------
Table 4-34
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.T.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.5
0.9
3.6
.5.4
19.9
7.3
1.7
2.9
25.2
2.7
140.4
163.4
961.7
1368.0
4686.4
1935.1
462.3
862.0
7655.7
745.4
287.7
323.1
1908.6
2734.2
9484.5
3884.9
930.7
1710.9
14467.8
1466.6
Total U.S.
70.2
18980.4 37199.0
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
28.1
7597.4 14889.9
4-121
-------
Table 4-35
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
H
X
New England
N.I.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.5
0.9
3.6
5.4
20.0
7.3
1.7
2.9
25.2
2.8
70.4
Point
Estimate
140.4
163.4
961.7
1368.0
4707.8
1973.8
464.6
862.0
7656.3
805.6
Maximum
287.7
323.1
1908.6
2734.2
9527.0
3959.7
935.2
1710.9
14469.0
1587.7
19103.5 37443.2
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annnalized Benefits
Between 1989 and 1995
Total U.S.
28.2
7646.7 14987.6
4-122
-------
Table 4-36
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Be tire en 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
n
X
New England
N.T.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
1.3
1.5
5.1
6.9
22.2
8.8
2.4
4.2
28.8
4.3
335.9
682.8
302.1
1362.8
1776.5
5377.7
2312.0
651.1
1315.8
8631.8
1350.3
601.5
2706.6
3525.4
10874.3
4618.5
1286.4
2509.1
16145.8
2615.7
Total U.S.
85.6
23416.1 45566.2
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
34.2
9372.9 18239.1
4-123
-------
Table 4-37
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Between. 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
.Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.T.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
1.9
2.0
8.2
10.3
39.0
13.9
4.1
5.1
45.7
5.4
135.7
Point
Estimate
468.7
382.5
2217.6
2641.5
8511.7
3158.8
1101 . 0
1576.8
13584.1
1526.6
Maximum
945.5
763.3
4258.1
5243.9
17139.2
6277.7
2200.6
3085.1
24248.9
3035.4
35169.2 67197.6
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annnalized Benefits
Between 1987 and 1995
Total U.S.
38.0
9835.1 18791.8
4-124
-------
Table 4-38
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
i
New England
N.T.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
3.6
3.3
11.7
12.1
43.0
14.1
6.0
6.6
49.5
1032.0
711.4
3271.0
3280.1
10268.1
3734.1
1843.8
2174.6
15230.5
2074.3
1424.8
6261.2
6417.4
20647.9
7495.0
3659.9
4104.6
26336.1
7.5
Total U.S.
157.3
2245.6
4389.5
43791.2 82810.7
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
44.0
12246.2 23158.0
4-125
-------
Table 4-39
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STDDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
4.3
136.1
164.8
774.4
316
45
111
915
Total U.S.
138.8
2607.2
Point
Estimate
0.0
18.7
595.0
720.6
3386.6
1383.6
197.9
487.8
4004.1
607.0
Maximum
0.0
33.1
1054.0
1276.4
5998.
2450.
350.6
863.9
7092.1
.4
.7
1075.2
11401.4 20194.3
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
1043.6
4563.7
8083.3
4-126
-------
Table 4-40
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
I
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
39.5
67.1
281.9
416.9
1626.2
588.1
131.6
233.3
1994.1
214.9
5593.6
Point
Estimate
172.8
293.2
1232.8
1823.1
7111.5
2571.7
575.5
1020.3
8720.2
939.7
Maximum
306.1
519.4
2183.5
3229.1
12596.0
4555.1
1019.3
1807.2
15445.3
1664.4
24460.9 43325.3
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
2239.0
9791.1 17342.1
4-127
-------
Table 4-41
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
TV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
39.
67.
281.
416.9
1629.6
592.6
132.
233,
1994.3
221.0
,5
.1
,9
.1
.3
Point
Estimate
172.8
293.2
1232.8
1823.1
7126.4
2591.4
577.8
1020.3
8721 . 0
966.6
Maximum
306.1
.519.4
2183.5
3229.1
12622.4
4590.0
1023.4
1807.2
15446.7
1712.1
Total U.S.
5608.4
24525.5 43439.7
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and '1995
Total U.S.
2244.9
9817.0 17387.9
4-128
-------
Table 4-42
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type B PM10 - 55 AAM/150 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
95.3
107.7
392.8
534.6
1802.4
703.0
184.3
342.1
2280.9
348.6
6791.8
Point
Estimate
416.5
470.9
1717.8
2338.0
7881.9
3074.3
805.8
1496.1
9974.5
1524.7
Max imam
737.8
834.0
3042.6
4141.0
13960.6
5445.3
1427.3
2649.9
17666.9
2700.5
29700.5 52605.8
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
2718.6
11888.4 21056.9
4-129
-------
Table 4-43
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 75 AAM/260 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
133.2
.1
.4
.7
.1
Total U.S.
148.
629,
789.
3143,
1104.3
313.2
406.6
3588.0
425.8
10681.3
Point
Estimate
582.3
647.8
2752.1
3453.1
13745.2
4829.2
1369.4
1777.9
15690.2
1862.1
Maximum
1031.4
1147.3
4874.6
6116.1
24345.7
8553.6
2425.6
3149.1
27790.7
3298.1
46709.4 82732.2
Discounted Present Value in Millions of 1980 Dollars, in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
2987.0
13062.3 23136.0
4-130
-------
Table 4-44
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1987 and 1995
Scenario: Type B TSP - 150 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
I
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
"North Pacific
266.1
236.7
884.4
923.6
3452.2
1098.9
455.7
522.9
3900.2
595.5
1163.8
1034.9
3867.4
4039.0
15096.7
4805.5
1992.5
2286.6
17055.6
2604.3
2061.3
1833.0
6850.0
7153.9
26739.4
8511.6
3529.2
4050.0
30209.0
4612.8
Total U.S.
12336.3
53946.3 95550.2
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1987 and 1995
Total U.S.
3449.8
15086.1 26720.6
4-131
-------
benefits in 1982 in 1980 dollars. Benefits for the alternative PH10
standards are estimated from 1989 to 1995, while benefits for the alterna-
tive TSP standards are estimated from 1987 to 1995.
The dollar benefits of the decreases in mortality risk resulting from
redactions in the ambient level of particnlate matter are listed in Tables
4-27 through 4-32. The point estimates of these alternative standards
range from $12.7 billion under a PM10 standard of an annual arithmetic mean
(AAM) of 70 |ig/m3 and a 24-hour expected value (EV) of 250 |ig/m3 to $52.4
billion under a TSP standard of 150 (ig/m3 24-hour average not to be
exceeded more than once a year.
Table 4-27 indicates that the benefits of the PH10 standard of 70
3 3
ug/nr AAM and 250 ug/ar 24-hour EV range from $0 to $62.1 billion. As can
be seen in the table, the Federal administrative region receiving the most
benefits (38 percent) under this standard is the East North Central Region.
The South Pacific and South Central regions receive, respectively, about 29
and 13 percent of the benefits accruing under this standard. Among the
remaining regions, benefits are divided as follows: North Pacific and
South Atlantic — 5 percent each; Middle Atlantic and Mountain — 4 percent
each; Midwest — 2 percent; and New York-New Jersey — less than 1 percent.
No benefits are expected to accrue to the New England region under this
standard since all counties within this region are predicted to be in
attainment with this standard.
Tables 4-33 through 4-38 report the benefits from the reduction in
acute illness associated with reductions in particulate matter. The point
estimate of these benefits range from $10.7 to $43.8 billion. Attainment
of the most lax PM10 standard reported in Table 4-33 (70 fig/m3 for AAM and
250 ng/m3 for 24-hour EV) results in benefits ranging from $0.03 to $21.5
billion. The South Pacific and East North Central receive about two-thirds
of the total benefits accruing under this standard. Except for the South
Central region, each of the remaining regions receive less than 10 percent
of the total benefits under this standard. Again, the New England region
4-132
-------
is not expected to derive any acute illness health benefits from this
standard because of attainment.
The last category of health benefits estimated in this section are
reported in Tables 4-39 through 4-44. The point estimates of the dollar
benefits associated with the reductions in chronic illness resulting from
the attainment of particulate matter standards range from $11.4 to $53.9
billion. Using the PM10 standard of 70 ng/m3 AAM and 250 ng/m3 24-hour EV
(Table 4-39), the distribution of benefits remains relatively unchanged
from the other categories of health benefits accruing under the same
standard. The East North Central, South Central, and South Pacific account
for approximately 77 percent of the benefits accruing under this standard.
Tables 4-45 through 4-47 show the benefits that accrue under a 70/250
PH10 primary standard when all counties are not in attainment with the
standard throughout the 1989—1995 time horizon. This can occur because
available control options are exhausted prior to standard attainment.*
These tables can be compared, respectively, to Tables 4-27, 4-33, and 4-39
where all counties were assumed to be in compliance with the same 70/250
PM10 standard. Obviously, the benefits accruing when all counties are in
attainment with the standard are greater than the benefits accruing when
all counties are not in compliance with the standard.
of Physical Effects
Implicit in the estimates of economic benefits are estimates of
changes in health status. The changes in health status include reduced
risk of mortality or morbidity. For economic valuation purposes, the
physical effects of reduced morbidity risk are further categorized into
fewer work days lost, fewer reduced activity days, and reduced direct
expenditures for medical care. In addition to the aggregate dollar
benefits that have been reported in Tables 4-27 through 4-44, estimates of
the physical effects associated with these benefits are developed for
* These benefits are referred to as "Type A" benefits. See Section 9.
4-133
-------
Table 4-45
ESTIMATED BENEFITS FOR: LAVE AND SESKIN CHRONIC MORTALITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
II
X
New England
N.Y.-N.J.
Middle Atlantic
South. Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
Total U.S.
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
Point
Estimate
0.0
21.3
507.3
539.1
3406.2
1007.1
189.7
387.5
2014.0
193.8
Maximum
0.0
104.2
2476.2
2631.4
16626.1
4916.0
925.8
1891.6
9830.7
946.1
8266.0 40348.1
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
0.0
3308.7 16150.4
4-134
-------
Table 4-46
ESTIMATED BENEFITS FOR: OSTRO, CROCKER, ET AL. ACUTE MORBIDITY STUDIES
Benefits Occurring Between 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
Point
Estimate
Maximum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
rv
V
VI
VII
VIII
II
I
New England
N.I.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
0.1
1.7
1.6
4.3
2.0
0.5
1.3
5.1
0.4
0.0
13.6
501.2
490.7
2309.7
779.1
157.5
366.9
2057.8
189.7
0.0
26.7
1062.9
1032.8
4762.0
1624.5
329.1
780.3
4213.3
393.1
Total U.S.
16.9
6866.1 14224.7
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
6.8
2748.4
5693.8
4-135
-------
Table 4-47
ESTIMATED BENEFITS FOR: CROCKER, ET AL. CHRONIC MORBIDITY STUDY
Benefits Occurring Between 1989 and 1995
Scenario: Type A PM10 - 70 AAM/250 24-hr.
Federal Administrative Region Minimum
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
REGION
I
II
III
IV
V
VI
VII
VIII
IX
X
New England
N.Y.-N.J.
Middle Atlantic
South Atlantic
E.N. Central
South Central
Midwest
Mountain
South Pacific
North Pacific
0.0
4.3
130.7
121.
360.
158.
37,
101.
404.2
36.5
.5
,5
.7
.9
.3
Point
Estimate
0.0
18.7
571.4
531.4
1576.6
693.9
165.8
443.0
1767.4
159.4
Maximum
0.0
33,
1012.
941.
2792,
1229,
293,
784.6
3130.5
282.3
.1
.1
.2
.6
,1
.6
Total U.S.
1355.5
5927.6 10499.0
Discounted Present Value in Millions of 1980 Dollars in 1982
Using a 10 Percent Rate of Discount.
Annualized Benefits
Between 1989 and 1995
Total U.S.
542.6
2372.7
4202.5
4-136
-------
informational purposes. The estimates for each standard and scenario can
be found in the supplementary tables in Section 11 of this report. The
estimates are based on the same methods and data used in calculating
economic benefits except that the final step of economic valuation is not
performed.
CONCLUSION
In this section, the results of previous epidemiological studies have
been used to estimate some of the health benefits that are expected to
occur under implementation of six alternative particulate matter standards.
Table 4-1 provided a summary of the benefits accruing in each of the
health categories considered in this section. As the table indicated, the
point estimates of the benefits associated with reductions in mortality
risk are consistently the largest health benefits estimated for the
alternative standards being considered in this analysis. Under the
strictest particulate matter standard, mortality benefits range from $0 to
$255.9 billion, while the most lenient standard results in benefits ranging
from $0 to $62.1 billion. The minimum estimate of $0 should indicate to
the reader the wide range of uncertainty inherent in these estimates.
The health benefits associated with reductions in chronic illness rank
second in the categories of health benefits estimated in this section.
Attainment of the most lax particulate matter standard results in benefits
estimated to range from $2.6 to $20.6 billion. Estimated benefits increase
to a range of $12.5 to $96.4 billion under the strictest standard
considered in this analysis.
Table 4-1 indicated that the point estimates of the benefits resulting
from reductions in acute illness are slightly lower than those estimated
for the reductions in chronic illness. The benefits for the acute illness
category range from $0.03 to $20.7 billion under the most lenient standard
to $0.16 to $80.4 billion under the most stringent standard.
4-137
-------
Tie benefits estimated in this section should be interpreted with
respect to the potential biases that are implicit in these calculations.
These potential biases are summarized in Table 4-48. In general, all of
the studies used to estimate the health benefits of reductions in particu-
late matter estimate simple concentration-response equations to measure the
relationship between health status and ambient particulate matter. Because
the relationship between health status and particulate matter is probably
more complex than that estimated in these studies, the benefits reported in
this section can only be considered approximations of the "true" health
benefits resulting from implementation of these standards. For example, if
individuals have avoided the effects of particulate matter through their
actions and the concentration-response equations do not take this into
account, the relationship between health status and particulate matter will
be underestimated. On the other hand, the relationship between health
status and particulate matter may be overestimated if the concentration-
response equation excludes factors that influence health and are correlated
with particulate matter.
Second, in those studies that examine the health effects of chronic
exposure, the annual average of particulate matter in a certain year is
used as a proxy of chronic exposure. The concentration-response functions
estimated in these studies therefore relate .the level of particulate matter
in a certain year to some measure of health status in the same year (e.g.,
mortality rates). Consequently, the effect of previous exposure on current
health status cannot be specifically identified. When using these studies
to estimate the benefits of particulate matter reductions, it is necessary
to assume that the change in particulate matter in one year has an instan-
taneous effect on health status in that year. Since it is possible that
there will be a lagged relationship between the particulate matter reduc-
tions and health status, the discounted present value of the benefits for
the chronic exposure categories may be overestimated.
Third, the studies used in this section generally use particulate
matter as a proxy for other air pollutants affecting health. If these
pollutants are positively correlated with particulate matter, the
4-138
-------
M
O
M
ss
g
oo
0 OS
fH •<
1
M U
o
SO I
0 1
t5
W £Q
.fl I
•** 1
fH
01
0 <
BB
a
o
•H
•W
0
Vl
•H
Q
to
08
•H
ea
e
•*<
+*
ol
O
«•?•)
«M
ft
0
§,
M
X
*
X
X
o
ft
»«M
fH O 0
O fH
M fl ft
-H o a
fH fl 48
•W (A
O O
VI 0 0
at V) A
•H -W
«g
H
at 0 O
0 4 H
a «•• «M
X
X
X
0
Vl
o
a
-w
(A
0
fl
o
•M
+*
08
0
O1
CM
M
fl
O
•H
+»
08
(4
+£
fl
0
o
fl
o
Q
«
«0
fl
iH
w
1
0 1
fl -M
^ -H 0
•^
. .. T3
0 01
« o a
08 -H
rfl 0 -M
fO «
•M 0
fl fH
0 fH fl
a -H o
•W jt -*4
(A -U
*O «H o*
at
at
fl .4
O 0
•* fO
-w
at (4
0 0
0 •»•>
» fl
0 O 0
M fH a
fl fH -V>
O C8 «
ft 0
<••!-» f^
0 O -O
H fl ol
91
*»
«M
0
fl
o
» •
o •d
O 0
08 +•»
fH 08
ft a-
rt ^
4> (0
•M O
<8 H
+•> 0
•O
0 fl
> 0
ffl 0
M
fl fH
0 fl
a s»
s
^
X
X
o
ffl
fl
O
M
•O
0
ft
o
•a
M
08
•^4
0 0
fH
T3 fl ^>
0 O <8
+J -M Vl
•" C8 CB
§^ >
0
M TJ
o Vi o
ffl O -O
•M O 3
^^
«H H O
O 0 M
ffl 0
fi A ffl
at > **
.
BO |
• O
0 fl
«^ 0
0
o
•H -O — »
jo a «
« at 0
fl fH
H fl ,0
al O a]
-M VI
>. 0 «8
U fH >
O fH
•woo
al ft fl
fl a
08 Vl 3
fH 0 fl
ft A 0
M -M 0
0 O 0
(M
O
y
0
0
fH
08
t
0
fl
0
4-139
-------
•a
o
d
d
o
00
^r
i
•«**
M W
O
eo
.2 '
fld
U ffl
08
O •<
B3
d
o
4«*
0
o
M
0
» O M
o3 o -P MH -o o
* >>,a d M o -P
T3 o Pi -P 03
se P -H -* « a
cu +» ** ,0 j» a -H
M **^ a d "^ +4
o o d M o
^ M O O O M
f-l -P -P ,£) o o o ^d d
O -P **H *W ^ 44 ft
r-t ^
^ 03 03 OH iH
1 ao a a >. *
03 d •* •** <« "O
^ -P +* O d «
.O > M W -P -P
,d o o o t-i <• o
•P M MO O
Pi O O > O M -P
CO) O -P
-H o a
d d o
d o -P
d n OB
00) ^3 ^H
o d
CM O
O >» -H
M -P
0 O M
(0 (4 C4
C3 ft ft
w
a
o
•ft
4«
4
P
a
v"4
Itf
CB)
4J
•^
«H
s
o
pp
N
iH
0
N
o
flri
^%
•B
P
«^
M
CM
o
A
o
•*
+}
M
IH
Q
ft
M
M
4.1
^
H
X X M
X MX)
M M
CO)
•P
col 1
a d
M O M
ce) 0 08
o . o
O Ot -P O
d o d
d ^ -o d
W ft r-t • M
-H d «) *H
•P 0
W M 08 -P (0
_« £ d g £
r& rO •»* .O
>» -p
4-1 0
O O CO) O
O O T3 > O
M M O M M
•»4 -rt M ffl -H
«^b M Mk ^^
O P IO Q
o | >, to
id "O 1 "^ J3 O
•p d >. -01-^
03 ^H -U 0 «J -H 0 -O
•p 08 d a to -P -H S
ddd dd ccjMd-p
03 « 03 OO •POOOM
,a to ,a -HO OM M
MOM -P ffl 0 -P J3 (>,
d M P COO MOOtO'P
PI d -P -H •« ,a -H
dO>
•P o <« ft -P 08 +j odoM,a
10 -H O O M 08 "HOMO
•P Pi O*O •PptO (A^a-PiH
«>OOd tfl 0» ffl MCO)08CO)fH
SB ft 43 n| a -a -H •< o
-------
09
e
d
o
o
00
^>
I
43
H
1
1
>» 1
W I 0
0 I
80 |
0 I
*» 1
at |
0 1 CO
1
43 1
at
0 •<
m
i
d
o
o
o
h
•rt
O
M
at
•H
oa
1
d
•g
O
£
M
a
o
4*
§i
f^
0
O
•J
^
41
iw
O
(3
e
0Q
O
«•
M
9
04
^
*o
0
4^
CO
W M M 01O
M ** •** -P >
a es T3 ol -a OOT3
o 2® 2® ®* 'a'0®
. -M at at at d 43 al
0 r* 8 ^S -^B O*»B
O « -H cttiH CB -rt O rt -rt
M ^ ^ ^ ^4 ^ 4^ M CB 4^
•vl OM OM O <« -olOM
Q Ho Ho Ho Q43o
o. d X
43 -a d o 43
MOOt d > *H d •«
d -M o 01 o -H -^ -^
Oalh -H «) 43 -w a at
-rt o O -M *» . oo
+J ^H +J O MO dO «M 8043'
ai'O^a ** d o -u >^ *j o d
•H d d T4'O M43ol,4J >O 43 -W
(4 go O M d Oalf-l M OO
oo d -H o< o< 01 -a -w o
*i M 0 -d 0 43 M o ^ ^ 4J
p4 > d O d OMM MOM >t
d H M -a w M*H d 43 ^ d>H
oodo do doo o « oo
•^ oo A o -u -a od 43 o d -^ *j
-M d -H M -Hold -H ao o ** -o -u on
OlOMOO M-HO) «dd (MOt-H 0,-HM
goo ao do d -H o o a > g> -rt
*» 43
•H
•O .-I
•H at
43 •*
(-( -a
o d
S . o
•M
o o
•H 0,
0
O M
W -H
43
O
d
•8
o
II
o
d
o
43
H
O
80
O
-M
01
O
O
M
O
•fct
01
•»*
•u
M
O
M
ol
d
at
d
•H
O
4-141
-------
relationship between particulate matter and health status may be over-
estimated. This is particularly relevant for the benefits based on the
Crocker .e_t ,a_l. (4) study since this study controls only for particulate
matter.
Fourth, all of the studies used in this section use pollution data
from one or more monitors in a geographic area to represent the exposure of
all individuals within the geographic area. If the relationship between
monitored air pollution and the population's true exposure to pollution has
changed significantly since the studies were done, the benefits estimated
in this section may only be approximations of the true health benefits of
air pollution control.
Finally, the studies used to estimate the health benefits in this
section generally examine the relationship between particulate matter and
health status for urban populations. For the purposes of this report, it
is assumed that these results can be used to approximate the health effects
for the individuals in the counties considered in this analysis. If the
sample populations used in these studies are systematically different than
the populations in the counties being considered in this analysis for
reasons other than exposure to particulate matter, the benefits reported in
this section may be under- or overestimates of the true benefits occurring
in these counties. For example, if the rural counties in this analysis are
systematically more (less) susceptible than urban populations, the benefits
accruing to rural counties will be underestimated (overestimated).
KEFEKENCES
1. Miller, F. I., et al. Size Considerations for Establishing a Standard
for Inhalable Particles. Journal of the Air Pollution Control
Association, 29:612-615. 1979.
2. U.S. Environmental Protection Agency, Office of Air Quality Planning
and Standards. Review of the National Ambient Air Quality Standards
for Particulate Matter: Assessment of Scientific and Technical
Information. OAQPS Staff Paper (EPA-45015-82-001), Research Triangle
Park. NC, January 1982.
4-142
-------
3. Freeman, A. M. The Benefits of Environmental Improvement. Resources
for the Future, Baltimore, Maryland, 1979.
4. Crocker, T. D. et al. Methods Development for Assessing Air Pollution
Control Benefits - Vol. 1: Experiments in the Economics of Air Pollu-
tion Epidemiology. Prepared for the U.S. Environmental Protection
Agency, Laramie, University of Wyoming, February 1979.
5. Chappie, M. J. and L. B. Lave. The Health Effects of Air Pollution:
A Reanalysis. July 1981 (forthcoming in the Journal of Urban
Economics).
6. Lave, L. B. and E. P. Seskin. Air Pollution and Human Health.
Resources for the Future, Baltimore, MD, 1977.
7. Thomas, T. J. An Investigation Into Excess Mortality and Morbidity
Due to Air Pollution. Purdue University, Ph.D. Dissertation, 1973.
8. Viren, J. R. Cross Sectional Estimates of Mortality Due to Fossil
Fuel Pollutants: A Case for Spurious Association. Prepared for U.S.
Department of Energy, Washington, DC, 1978.
9. Smith, V. K. The Economic Consequences of Air Pollution. Ballinger,
Cambridge, MA., 1976.
10. U.S. Environmental Protection Agency, Office of Research and Develop-
ment* Air Quality Criteria for Particulate Matter and Sulfur Oxides:
External Review Draft No. 4. Research Triangle Park, NC, December
1981.
11. Thibodeau, L. A., et al. Air Pollution and Human Health: A Review
and Reanalysis. Environmental Health Perspectives, 34:165-183.
February 1980.
12. Koshal, R. K. and M. Koshal. Environments and Urban Mortality - An
Econometric Approach. Environmental Pollution 4:247-259. June 1973.
13. Gregor, John J. Intra-Urban Mort'ality and Air Quality: An Economic
Analysis of the Costs of Pollution Induced Mortality. Prepared for
the U.S. Environmental Protection Agency, Corvallis, 1977.
14. Lipfert, F. W. The Association of Air Pollution with Human Mortality:
Multiple Regression Results for 136 Cities, 1969. Paper presented at
the 70th Annual Meeting of the Air Pollution Control Association, June
20-24, 1977.
15. Lipfert. F. W. On the Evaluation of Air Pollution Control Benefits.
Prepared for the National Commission on Air Quality, November 1979.
16. Lipfert, F. W. Sulfur Oxides, Particulates, and Human Mortality:
Synopsis of Statistical Correlations. Journal of the Air Pollution
Control Association, 30:366-371. April 1980.
4-143
-------
17. Mendelsohn, R. and G. Or cut t. An Empirical Analysis of Air Pollution
Dose-Response Curves. Journal of Environmental Economics and Manage-
ment, 6:85-106. 1979.
18. Seneca, I. and P. As eh. The Benefits of Air Pollution Control in New
Jersey. Center for Coastal and Environmental Studies, Rutgers Univer-
sity, April 1979.
19. Pindyck, R. S. and D. N. Rubinfeld. Econometric Models and Economic
Forecasts. McGraw Hill, New York, 1976.
20. Gerking, S. and V. Schulze. What Do We Know About the Benefits of
Reduced Mortality From Air Pollution Control? The American Economic
Review, 71:228-234. May 1981.
21. Atkinson, S. E. A Comparative Analysis of Macroepidemiological
Studies. Department of Economics, University of Wyoming (undated).
22. Ostro, B. D. Morbidity, Air Pollution and Health Statistics. Paper
presented at the Joint Statistical Meetings of the American Statis-
tical Association and Biometric Society, Detroit, MI, August 12, 1981.
23. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10 - No. 118. Disability Days: United States 1978.
24. U.S. Bureau of Economic Analysis. Projections of the Population 1976-
2000. Memorandum, March 1981.
25. U.S. Bureau of Economic Analysis. 1980 OBERS BEA Regional Projec-
tions. Vol. 3 SMSAs, July 1981.
26. U.S. Bureau of the Census. Current Population Reports Series P-25,
No. 873, February 1980. Estimates of Populations of Counties and
Metro Areas July 1, 1977-1978.
27. U.S. Bureau of the Census. Population and Households by States and
Counties: 1980 (PC80-S1-2).
28. U.S. Bureau of the Census. County Business Patterns, 1978.
29. U.S. Department of Commerce News, Bureau of Economic Analysis.
Projections of Personal Income to the Tear 2000. December 9, 1980.
30. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10. Current Estimates from the Health Interview Survey:
United States 1977. Hyattsville, MD, September 1978.
31. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10. Current Estimates from the Health Interview Survey:
United States 1978. Hyattsville, MD, November 1979.
4-144
-------
32. U.S. National Center for Health. Statistics. Vital and Health Statis-
tics Series 10. Current Estimates from the Health Interview Survey:
United States 1979. Hyattsville, MD, April 1981.
33. U.S. Department of Commerce. 1980 Statistical Abstract of the United
States. Washington, DC, 1980.
34. Cooper, B. S. and D. P. Rice. The Economic Cost of Illness Revisited.
Social Security Bulletin, 39-2:21-35. 1976.
35. U.S. Department of Labor. Geographic Profile of Employment and
Unemployment, 1979. December 1980.
36. U.S. Bureau of Labor Statistics. Unpublished employment data, 1980.
37. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series 10, No. 133. Selected Health Characteristics by Occupa-
tion: United States 1975-1976. Hyattsville, MD, May 1980.
38. U.S. National Center for Health Statistics. Vital and Health Statis-
tics Series No. 10. Acute Conditions: Incidence and Associated
Disability, United States, July 1975-June 1976. Hyattsville, MD,
January 1978.
4-145
-------
APPENDIX 4A
DATA SOURCES
The purpose of this appendix is to provide a listing of the sources of
the data used to extrapolate the results of the health studies reviewed in
Section 4 to national benefit estimates. Unless otherwise noted, all
dollar amounts are stated in 1980 dollars.
CHRONIC MORTALITY
t
POP; County Population in 1980
Source: Bureau of the Census (26); Bureau of Economic Analysis
(22.23).
Comments: For counties within an SHSA, SHSA projections were used to
estimate growth rates. For rural counties, state-level projections were
*
used.
ACUTE MORBIDITY
HE: Bovsefcolds Pec County in 1980 —
Source: Bureau of the Census (26).
Comments: Projections for household growth based on population
projections and projections in average household size.
1A6E: .Hourly Wage Rate Pec Employee in 1978 —
Source: County Business Patterns (28); Bureau of Economic Analysis
(29).
4-146
-------
Comments: Updated to 1980 using Consumer Price Index for all items.
Non-government and Federal government payroll information. Excludes self-
employed individuals, railroad employees, farm workers, domestic service
workers, and state and local government employees. The payroll is divided
by the number of employees and 2,080, an estimate of the hours worked per
year, to find the hourly wage. The hourly wage is assumed to grow at the
rate of personal income growth for each state. The value of each non-work
sick day is assumed to grow at the rate of personal income growth for the
United States.
1LD8: Work Los* Days Per Household —
Source: Current Estimates From the Health Interview Survey: United
States 1977-1979 (30-32).
Comments: Days lost from work per household due to acute illness.
Data not available on county level; the average of the annual number of
acute work loss days per United States household from 1977 to 1979 was
used.
1ADH: Restricted Activity Days Per Household —
Source: Current Estimates From the Health Interview Survey: United
States 1977-1979 (30-32).
Comments: The number of days (net of work loss days) per household
where activity is limited due to acute illness. Data not available on
county level; the annual average number of restricted activity days per
United States household from 1976 to 1979 was used.
DHEH: Direct Medical Expenditures Per Household —
Source: 1980 Statistical Abstract (33); Cooper and Rice (34); Current
Estimates From the Health Interview Survey (30-32).
4-147
-------
Comments: Information not available on county level; national data
were used. Only direct medical expenditures on acute respiratory and
circulatory disease considered. A crude breakdown of direct medical
expenditures between acute and chronic disease is based on the percentage
of total restricted activity days, bed loss days, and work loss days that
are accounted for by each category. No other data were available to
estimate a finer breakdown between acute and chronic DME. Expenditures on
dentists' services, eyeglasses, administration, research, construction,
government health activities, and other health services are excluded. The
medical consumer price index (CPI) and population growth factor are used to
inflate figures to 1980.
P ^
WAGE: Hourly Wage Rate in 1978 —
Source: 1980 Statistical Abstract (33); Cooper and Rice (34); Current
Estimates From the Health Interview Survey (30-32).
WORKER ( 18-44) : Workers Between the Ages of 18 and 44 —
Source: County Business Patterns 1978 (28); 1980 Statistical Abstract
(33), Bureau of Economic Analysis (25), Department of Labor (35,36).
Comments: State-level percentage breakdowns of employment by age were
used to proxy percentage breakdown of county employment by age. Breakdown
of the percentage employed between the ages of 18 and 44 not available;
percentage of workers under age 45 was used. For rural counties, popula-
tion projections were used to approximate employment growth in the age
group. For counties within SMSAs, SMSA employment growth rates were used.
WORTER(45-64) : Workers Between the Ages of 45 and 65 —
Source: County Business Patterns 1978 (28); 1980 Statistical Abstract
(33), Bureau of Economic Analysis (25), Department of Labor (35,36).
4-148
-------
Comments: Information on workers between the ages of 45 and 65 not
available; state—level percentage breakdown of employees between the ages
of 45 and 64 was used. These state-level percentage breakdowns were
applied to county-level employment to estimate the number of workers
between the ages of 45 and 65 in the county. The growth projections
applied to this age group of workers were the same as those used for
younger workers (see above).
EPISODE: Days Lost Froei York Per Acute Illness Episode —
Source: Current Estimates from the Health Interview Survey: United
States 1977 to 1979 (30-32); Selected Health Characteristics by Occupation:
United States, 1975-1976 (37); Acute Conditions Incidence and Associated
Disability (38).
Comments: Data not available on county level; annual average number
of days lost from work per acute illness episode (net of injuries) of all
workers over the age of 45 were used. Calculated using data from 1977 to
1979.
WLDWU8-44): York Loss Days Per Yorker Aged 18 to 44 —
Source: Current Estimates From the Health Interview Survey: United
States 1977 to 1979 (30-32); Statistical Abstract of the United States
(33).
Comments: The annual number of work days lost due to acute illness
(net of injuries) was only available for the 17 to 44 age group; hence,
this value was used to prory WLDW( 18-44). Annual average calculated from
United States data from 1977 to 1979.
4-149
-------
YLDV(45-65): York Loss Days Per Worker Aged 45 to 65 —
Source: Current Estimates From the Health Interview Survey: United
States 1977 to 1979 (30-32); Statistical Abstract of the United States
(33).
Comments: Because information was only available for the 45 to 64 age
group, data on the annual number of work days lost due to acute illness
(net of injuries) for this age group was used to proxy WLDW(45-65). Annual
average calculated from United States data from 1977 to 1979.
RADf (18-44) : Reduced Activity Days Per Yorker Aged 18 to 44 —
Source: Current Estimates from the Health Interview Survey: United
States 1977 to 1979 (30-32); Statistical Abstract of the United States
(33).
Comments: Annual number of reduced activity days due to acute illness
(net of injuries) per worker that does not result in days lost from work.
RADW of the 17 to 44 age group used to proxy RADVU8-44). Annual average
based on United States data from 1977 to 1979.
KADV(45-65): Reduced Activity Days Per Yorker Aged 45 to 65 —
Source: Current Estimates from the Health Interview Survey: United
States 1977 to 1979 (30-32); Statistical Abstract of the United States
(33).
Comments: Annual number of reduced activity days due to acute illness
(net of injuries) per worker that does not result in days lost from work.
Data on the 45 to 64 year old age group used to represent 45 to 65 age
group.
4-150
-------
NONfOKKERS: Number of Nomrorkers Per County —
Source: Bureau of the Census (26); Bureau of Economic Analysis
(32,24); County Business Patterns 1978 (28).
Comments: The number of nonworkers in a county was obtained by sub-
tracting county employment from county population.
SADP: Reduced Activity Days Per Nomrorker —
Source: Current Estimates From the Health Interview Survey: United
States 1977 to 1979 (30-32).
Comments: Annual number of reduced activity days due to acute illness
per nonworker. Calculated from United States data from 1977 to 1979.
DMEP: Direct Medical Expenditures Per Person —
Source: 1980 Statistical Abstract (33); Cooper and Rice (34); Current
Estimates From the Health Interview Survey (30-32).
Comments: See comments on DMEH. With the exception of using popula-
tion in place of households, the data used to calculate DMEH were used to
estimate DMEP.
CHRONIC MORBIDITY.
Days of ^Tonic Illness Per Household
Source: Current Estimates From the Health Interview Survey: United
States 1976 to 1979 (30-32).
Comments: Annual number of days of chronic illness per United States
household. Annual average calculated using data from 1977 to 1979.
4-151
-------
Direct: Medical RTponditmes Par Household
Source: 1980 Statistical Abstract (33), Cooper and Rice (34), Current
Estimates From the Health Interview Survey (30-32).
Comments: Information not available on county level; national data
were used. Only direct medical expenditures on chronic respiratory and
circulatory disease considered. A crude breakdown of direct medical expen-
ditures between acute and chronic disease is based on the percentage of
total restricted activity days, bed loss days, and work loss days that are
accounted for by each category. No other data were available to estimate a
finer breakdown between acute and chronic DME. Expenditures on dentists'
services, eyeglasses, administration, research, construction, government
health activities, and other health services are excluded. The medical
consumer price index (CPI) and population growth factor are used to inflate
figures to 1980.
4-152
-------
APPENDIX TO VOLUME II
VALUATION OF HEALTH IMPROVEMENTS
-------
-------
APPENDIX TO VOLUME II
VALUATION OF HRAT.TH IMPROVEMENTS
INTRODUCTION
Sections 3 and 4 use the results of epidemiological studies to
estimate the health improvements that result from reduced levels of parti-
culate matter. The epidemiological studies used in these sections examine
the relationship between particulate matter and measures of health status
such as mortality and the incidence of illness. Although these studies
estimate the health effects of exposure to particulate matter, they
generally do not impute an economic value to these effects. In order to
compare the benefits and the costs of attaining alternative reduced levels
of particulate matter, the economic value of the health improvements
measured in these studies must be estimated. The purpose of this section
is to examine alternative methods for valuing health improvements.
It is very difficult to place a monetary value on marginal reductions
in risk of mortality. Since the reduction of pollution uses resources that
could be alternately employed, however, some comparison of the major costs
and benefits of alternative control strategies is an important policy
consideration. To be consistent with the willingness—to—pay criteria used
to calculate other benefits, estimation of the benefits of reduced
mortality risk should reflect affected individuals' own valuation of risk
reduction. Individuals implicitly make tradeoffs between income and risk
in their daily lives, placing a finite value on marginal changes in risk of
death. In this section the willingness-to-pay implied by these tradeoffs
will be estimated and the application of these results to the mortality
reductions identified in Sections 3 and 4 will be discussed.
A-l
-------
After discussing the valuation of mortality risk reductions, the
economic value of reductions in the incidence of illness (morbidity) will
be discussed. The value of this reduction will be based on three
components: 1) work days lost due to illness, 2) non-work days lost due to
illness, and 3) use of health care resources.
ALTERNATIVE METHODS FOR VALUING REDUCTIONS IN MORTALITY RISK
There are several alternative methods for valuing the benefits of
reductions in risk of mortality. One commonly-used method is the human
capital approach which values a person according to his or her contribution
to the output of the economy. In this method, the contribution is measured
by earnings. Therefore, the value of reduced risk of death is equal to the
present value of the expected increase in earnings.
Cooper and Rice (1) present a thorough application of this approach.
They use working life tables and a cross-section of 1972 earnings to calcu-
late the present value of future expected earnings by age, race, and sex.
Other studies [such as Lave and Seskin (2)] have updated Cooper and
Rice's figures and used them to estimate the benefits of mortality reduc-
tions. Liu and Yu (3) and others develop their own figures for produc-
tivity losses using more aggregate data than those of Cooper and Rice.
A number of technical questions can be raised concerning applications
of the human capital approach. The appropriate discount rate to apply,
adjustment of current earnings for expected changes in productivity over
time, and the method for valuing housewives' services are a few of the
issues that arise. More importantly, the theory itself has significant
weaknesses. In this approach, measurement of an individual's worth is
limited to the value of services sold in the labor market. The imputing of
values to housewives' services only partially adjusts for this limitation.
For example, with this approach the value of the elderly is near zero.
Also, any distortions in the labor market are reflected in the values
assigned to individuals. In addition, the approach fails in its goal of
A-2
-------
estimating the net change in the economy's output due to a death since the
responding movements and adjustments in the labor market are not
considered.
Finally, the human capital method does not measure the willingness-to-
pay of an individual or other members of society for a reduction in his or
her probability of death. Individual preferences are ignored. The amount
that a population 'will pay for a reduction in its mortality rate may be
less than or greater than the value of the expected change in earnings.
Conley (4), Linneroth (5), and Usher (6) discuss the potential deviation
between mortality reduction valuations based on willingness—to-pay and
human capital criteria.
Several other methods also fail to measure economic benefits. Estima-
tion of losses in net output (the present value of foregone lifetime
earnings minus consumption) suffers from all the shortcomings of the human
capital approach. Studies of the amounts that individuals spend on life
insurance are not relevant. Insurance payments reflect willingness—to—pay
for a reduction in beneficaries' financial risk, not one's own risk of
death. Lastly, values implied by previous jury or government decisions
will not measure economic benefits unless these decisions themselves were
based on some measure of willingness-to-pay.*
There are two principal methods of measuring individuals' willingness-
to-pay for reductions in mortality risk: 1) surveys and 2) wage compensa-
tion studies. These two methods are reviewed below.
Snrrers
Under the survey approach, implied valuations of risk are derived from
individuals' responses to hypothetical decisions. Acton (8) and Jones-Lee
(9) have conducted surveys of the choices people make when presented with
* According to information from the General Accounting Office, the values
of a reduction of 1 z 10 in annual mortality risk implied by government
decisions range from $0.07 to $624.00 (7).
A-3
-------
situations in which marginal reductions in the probability of death can be
purchased.
Acton asked 36 people how much each would pay for a heart attack
treatment program which would reduce his or her probability of death.
Extrapolating from the mean reported payments, an individual would pay
$0.03 to $0.04 (1973 dollars) for each unit reduction of 1 z 10~6 in annual
mortality risk.
Jones-Lee obtained 30 responses to a questionnaire on the value of
safety. Respondents indicated the premiums or discounts they required to
travel on planes with different safety records. From the range of indivi-
dual values, a value of around $6.00 (1976 dollars) for a unit reduction of
1 z 10 in annual mortality risk can be selected.
The results of the two surveys can be accepted only with major quali-
fications. First, both samples are very limited. Second, respondents may
have difficulty understanding the small probabilities and hypothetical
choices presented. Third, if respondents believe their answers will
influence some public policy decision, they may engage in strategic
behavior and give answers they think will contribute to the outcome they
prefer. Respondents may also attempt to provide responses which will
favorably impress the snrveyer. Finally, responses may vary with specific
details of the situation presented, limiting the transferability of the
results.
The pattern of responses indicates that these problems may be serious.
Not only the results across surveys and individuals, but also responses for
each individual across questions, exhibit wide variation and inconsistency.
Acton notes that his survey reveals ten patterns of willingness-to-pay for
risk reductions. These patterns include positive payments for increased
risk and a constant payment independent of the size of the risk reduction.
Acton suggests that the order of questions and particular risk description
given affect responses. Individuals express different valuations of equal
A-4
-------
risk reductions depending on the formulation and details of the case pre-
•
sented.
The results of Jones-Lee also exhibit inconsistency. The pattern of
payments for different risk reductions follows the smooth plot predicted by
theory for only a few individuals. Several respondents will pay for risk
reductions only over a certain threshold. For others, payment for a given
reduction in risk does not change monotonically with the risk level.
Wage Co«pens«tion Studies
Wage compensation studies infer willingness- to-pay for risk reduction
from the decisions that workers make in the labor market. Implicit
marginal valuations of risk are imputed from the relationship between wages
and risk across jobs. Wage compensation studies are based on the hedonic
wage theory described in Section 6.*
In this subsection, empirical applications of the wage compensation
0
theory will be reviewed. Wage compensation studies attempt to estimate the
curve traced by the wage-risk combinations observed in the labor market.
Wages, or the logarithm of wages, are regressed on worker and job charac-
teristics including the probability of job-related death. The two general
forms of the equation used can be represented by:
w
and
ln(W) = p0 + 0-jR + p2J + p3M (A.2)
where W = annual wage rate
R - annual probability of job-related death in units of 1 z
10~6
* The interested reader is referred to Section 6 for a discussion of the
hedonic technique as applied to the labor market.
A-5
-------
J = vector of job characteristics. Among the characteristics
which may be included are occupation, unionization, stress,
job repetitiveness, and job injury rates.
M = vector of personal characteristics. The variables may
include education, marriage status, race, and tenure.
B. - coefficients of the worker and job characteristics where j
J = 0 to 3.
The implicit marginal valuation of a unit change of 1 x 10 in the annual
probability of death is approximated by the partial derivative of the wage
equation with respect to risk. This implicit marginal valuation is equal
to Pj for the linear form and {^W for the semilog form. If the labor
market is in equilibrium, the implicit marginal risk valuation represents
workers' willingness-to-pay for reductions in the risk of death they, face
on the job. Consequently, wage compensation studies can be used to
estimate the worker's marginal value of a reduction in risk.
Below, some of the specific issues that arise in reviewing the wage
compensation literature are discussed.
Restrictions on. Functional Form —
The estimated wage—risk relationship will be restricted by the
functional form chosen. Host models use the log of earnings as the depen-
dent variable. Under this semilog specification, the loci of equilibrium
wage-risk combinations follows a convex curve. Inclusion of the square of
the probability of job-related death allows for the possibility of worker
self-selection across risk levels and results in a concave loci of equili-
brium points. Table A-l presents the three alternative functional forms
and the derivatives of each.
The correct shape of the curve depends on the distribution of firms
over safety technology and workers over risk preference across risk levels.
The effect of an incorrect specification on the estimate of the coefficient
of death risk depends on the risk level being evaluated. For example, if a
convex curve is specified when a concave curve is appropriate, risk
A-6
-------
0)
•H
en
u
M
a
D
EH
cn
1
§
a
u
cn
D
cn
D
a
u
EH
y
D
0
•H
JJ
(C
TJ
(0
a
«
c?
rH
.•J
•^•t
E
^
^Q
g1
rH
•H
e
^
CO
m
0J
c
•H
|J
CN
OS
CN
ca
+
OS
rH
II
^^
s
"c"
^
OS
rH
ca
II
^>
c.
c
rH
OS
,_^
ca
ii
5
£
rH
(0
C
o
•H
4J
O
C
a
Cu
O
A
S"
CN
ca
CN
+
rH
ca
*
ii
S OS
(O
-------
premiums will be overestimated at high, risk levels and underestimated at
low levels.
Control for Confounding Factors —
In order to isolate the effect of changes in risk on wages, the
regression equations must control for the effects of other job charac-
teristics. Each study includes a range of job variables. However, it is
not possible to capture all the elements that vary across jobs. Omission
of job variables may bias the coefficient of death risk.
For example, most studies omit nonwage compensation. They examine the
relationship between risk and wage compensation alone. If nonwage compen-
sation is positively correlated with risk but negatively with wages, the
risk coefficients will be biased downwards.* Similarly, most studies look
at nominal wages with only crude adjustments for differences in cost-of-
living instead of examining the real income that workers will sacrifice for
reductions in risk. Such specification errors may bias regression coeffi-
cients.**
Accurate estimation of workers' implicit valuation of risk also
requires control of personal and site characteristics which affect wages.
As discussed above, omission of relevant characteristics may bias the
coefficients. For example, workers' skill may be positively correlated
with wages and risk aversion. When a variable for skill is omitted, the
premium workers receive in risky jobs is underestimated. The estimated
coefficient of death risk reflects the premium for risk plus the negative
premium for lower skill.
While including some variables for personal characteristics which
affect wages, most studies assume that implicit risk valuation is indepen-
dent of personal characteristics or organization of workers. It is assumed
* See Dillingham (10), p. 165.
** See V. K. Smith (11).
A-8
-------
that there is only one equilibrium curve and marginal acceptance wage at a
given risk level. Some studies, however, include variables which measure
the interaction between risk and personal characteristics and unionization.
These terms allow risk premiums to vary according to workers' skills or
characteristics. This variation may result from:
Differences in productivity under risk for different
groups. For example, if older or unionized workers work
more efficiently under risk, firms may offer them a higher
risk premium.
Error in measuring risk. Risk within an occupation or
industry may vary according to union status or personal
characteristics such as age and sex. This variation may be
due to differences in risk-handling skills, assignment to
jobs within industries or occupations, or safety control.
Other factors such as job opportunities, information, and
mobility which affect acceptance wage and vary with
personal characteristics.*
In order for any of these factors to lead to differential risk
premiums for a given job, the relevant personal characteristics must vary
across the workers in a j ob.
Quality of Data, Accuracy and Reliability —
Because of data limitations, the matching of wage and personal data to
risk data is often very crude. While most wage data is by individual and
occupation, risk data is usually by industry. To the extent that specific
job risk differs from the industry average, measurement error results. It
can be argued that the risk premium received in an occupation is related to
the riskiness of the dominant job in the industry. Then, the premium may
be related to industry instead of job-specific risk. If this argument is
correct, however, only the premiums in dominant jobs could be used as a
basis for determining implicit risk premiums.
* Dillingham (10), p. 146.
A-9
-------
Measurement error will also result when occupational risk data is used
since risk varies by industry, age, and sex. If the measurement error
introduced by the risk data is random, the coefficient of death risk will
be biased downwards. If measurement errors in occupation or industry risk
data are non-random, the risk coefficient could be biased upwards or
downwards, depending on the direction of the error.
Adaptability of the Results for Benefits Analysis —
Both the wage compensation and mortality studies in Sections 3 and 4
consider reductiuons in annual risk of death. To use the wage compensation
study results to value risk reductions identified by the mortality studies,
the following assumptions are needed:
An individual's tradeoff between income and risk does not
vary according to the source of the risk.
This assumption allows us to equate willingness-to-pay for marginal
reductions in job-related and pollution-related death. As discussed in the
survey section, valuations may be dependent on the particular nature of the
risk.
• The coefficients derived from wage compensation studies are
representative of the risk valuations of the population
experiencing risk reductions in this analysis.
The correspondence between willingness to pay estimates from the wage
compensation studies and willingness to pay of the individuals in the
benefit analysis depends on how closely their risk attitudes are
represented .by the sample groups in the wage studies. Individual
valuations will vary with such factors as nonlabor income, original risk
level,' and cost of risk bearing.
For example, willingness to pay for risk reduction may vary with age.
The human capital approach suggests that there is a lower value to reducing
risk for an elderly person than for a younger worker. However, this result
A-10
-------
may not hold for the wage compensation approach. In a simple model,
*
Freeman [(12), p. 178] finds that "the marginal willingness to pay for job
safety increases, ceteris paribus. with an increase in any component of
risk ... thus one would expect older people to be more conservative about
controllable risk." The marginal value of reduced mortality risk from air
pollution increases as an individual's overall probability of survival
decreases. Then, values derived from studies of the behavior of workers
would underestimate the willingness to pay of the older population.
Data constraints prohibit the identification of the distribution of
relevant factors across the affected population in our analysis and their
effect on valuation of risk. Some judgment on the appropriateness of
different study samples for our analysis is possible, however. Since the
purpose of this section is to estimate the willingness-to-pay of general
populations who involuntarily bear the risks from exposure to pollution,
samples which reflect the risk preferences of average workers are most
appropriate.* The valuations of workers who have self-selected to risky
jobs, voluntarily choosing high risks, represent a lower-bound estimate of
the amount the population in this analysis would be willing to pay for a
reduction in risk.
Marginal willingness to pay for risk reduction is constant
for the small changes in risk considered in wage
compensation studies and our analysis.
From the wage compensation studies, willingness to pay for marginal
changes in risk are derived. For non-marginal changes, income effects
would be observed and some determination of the shape of the demand curve
for risk reduction would have to be made. Since reduction in pollution
levels will result in very small changes in the probability of death,
* Individuals may self-select across locations with varying pollution
levels according to risk preferences. Then, the willingness-to-pay for
reduction of pollution related fatalities may reflect the risk attitudes
of workers in risky occupations and may be lower than the willingness to
pay of the pre-pollution population.
A-ll
-------
however, it is assumed that the marginal willingness-to-pay is constant for
the range of risk changes for individuals in this analysis.*
Literature Review
Considering the above factors, the results of six wage compensation
studies which examine the relationship between wages and probability of job
fatality will be evaluated. Comparison of the full range of relevant
studies allows identification of some of the consequences of using alterna-
tive specifications and data.
Thaler and Rosen (13) take a sample of 900 male workers from the 1967
Survey of Economic Opportunity and match it to risk data from a 1967 study
by the Society of Actuaries. Both linear and semilog specifications are
employed. The results do not provide Thaler and Rosen with sufficient
information to choose between the two specifications. A variable for the
square of death risk is initially included but is dropped because its
coefficient is not significant.
Several risk interaction terms are also included. The terms measuring
the interaction between death risk and marriage and unionization are signi-
ficant and positive. There are two alternative summary measures of
personal characteristics in addition to variables for such characteristics
as age and education: a dummy variable for occupation and an index based
on several socioeconomic status measures. The elasticity equals 0.0290 for
the linear specification and 0.0230 for the semilog specification in
equations using occupation dummies and excluding interaction terms.**
* If large changes in mortality rates were considered to be concentrated
on a few individuals instead of small changes being spread over a large
group, application of the marginal value to the average change in
mortality rate could overestimate willingness-to-pay. Effects on income
would not be incorporated.
** The elasticity is a measure of the percentage change in the dependent
variable that can be expected from a percentage change in an independent
variable. In this section, it will be used to represent the percentage
change in the wage rate resulting from a one percent change in risk.
A-12
-------
The workers in Thaler and Rosen's sample face an average job risk
about 10 times the national average. Therefore, the sample members are
probably less risk averse than the average worker since workers with low
risk aversion will self-select to risky jobs. (Restriction of the sample
to high-risk jobs may explain the insignificance of the squared risk term.
Over small ranges of risk, the risk attitudes of workers will not vary
greatly. Thus, self-selection across risk levels will have a minor effect
on the slope of the curve.) As a result, the implicit valuation of risk of
workers in Thaler and Rosen's sample may underestimate the valuation of the
average worker.
Another factor which may bias the coefficient of death risk downwards
is the method of measuring risk. The actuarial data gives deaths by occu-
pation. Deaths related to chronic work conditions and non-job-related
deaths as well as job accident fatalities are included. Adjustment is made
for variations in non—job deaths that result from differences in the age
distribution across occupations. This adjustment, however, does not
consider variations in risk attitudes. If risk attitudes affect both job
choice and actions in daily life, workers in risky jobs may have a higher
incidence of non-job death than the average worker. Therefore, Thaler and
Rosen will overestimate the job-related deaths for risky jobs and the
implicit valuation of risk will be underestimated.
One factor which may result in an upward bias in the coefficient of
death risk is the exclusion of variables for injury rates or other job
characteristics. If other unpleasant characteristics are positively corre-
lated with death risk, the observed implicit price of risk may include
payment for other negative aspects of the job.
Smith (14) uses a sample of 3,183 white males from the Survey of
Economic Opportunity data used by Thaler and Rosen. Instead of occupation
risk data, Smith employs industry data on risk of fatal accidents from the
Bureau of Labor Statistics. A second sample of 5,458 workers in the manu-
facturing industry was taken from the 1973 Current Population Survey.
Smith argues that use of the more homogeneous sample eliminates some of the
A-13
-------
problems of variation across individuals in job characteristics omitted
from his model.
Smith does not provide information on the average risk level for his
sample so elasticities cannot be estimated. Smith's exclusion of accident
risk variables from the second sample (disability risk variables are
included in the first sample) may lead to an upward bias, for reasons
discussed previously. If measurement error is random, the use of industry
instead of occupation risk data would tend to bias the coefficient of death
risk downwards since occupation data should more closely match individual
jobs. However, studies using industry data consistently yield higher
coefficients than those using occupational risk data. Thaler and Rosen
report that they obtained results similar to Smith's when they used Bureau
of Labor Statistics industry risk data. One interpretation of this result
is that measurement errors caused by industry risk data may bias the death
risk coefficient upwards.*
An alternative explanation is that occupational risk studies use data
for jobs with higher risk than the jobs sampled in studies using industry
risk. Since Smith surveys jobs with a wider range of risk levels and lower
average risk than Thaler and Rosen's samples, the workers in Smith's study
are more risk averse if worker self-selection across risk levels occurs.
This difference may partially explain the higher coefficient estimated by
Smith.
Olson (15) uses a sample of 5,993 full-time workers from the 1973
Current Population Survey and industry risk data from the Bureau of Labor
Statistics. He regresses the logarithm of wages on death risk and the
square of death risk and includes several risk interaction terms.
Variables for accident risk and length of workday loss for each accident
are included. The death risk squared term and union—risk interaction term
* For example, industry statistics include women. Women may be concen-
trated in certain low risk occupations. Then, industry averages would
underestimate risk of male workers. Occupational risk data would not
suffer from this problem.
A-14
-------
are significant at the 0.05 level. Other interaction terms are not signi-
ficant using weekly and hourly wages as the dependent variables.
Olson estimates an elasticity of 0.034 to 0.035. Exclusion of the
death risk squared term results in a 50 percent reduction in the elasticity
of the death risk variable. It should be mentioned again that Olson's
coefficient of risk may be biased due to the use of industry risk data.
Again, the direction of the bias depends on whether the measurement error
is random or positively or negatively correlated with risk.
Viscusi (16) employs a sample of 496 blue collar workers from the
University of Michigan longitudinal study and Bureau of Labor Statistics
industry risk data. He uses both linear and semilog specifications.
Viscusi estimates an elasticity of 0.025 in his linear equation. When
accident variables are excluded, the estimated elasticity increases 21 to
150 percent. This figure indicates the possible magnitude of the upward
bias introduced by Smith and Thaler and Rosen's exclusion of accident
4
variables. The consequences of measurement error from use of industry risk
data are discussed above. Because he uses industry risk averages for all
occupations but then isolates his sample to blue collar jobs, Viscusi may
underestimate risk, biasing the death risk coefficient upwards.
Dillingham (10) employs the most complete risk data of any of the
studies. He matches 1970 New York wage and personal characteristics data
to New York Workmen's Compensation data on risk of fatal accidents given by
occupation, industry, and age for 3,700 New York blue-collar workers in
construction and manufacturing. No union variable is included. A wider
sample of 8,000 male workers yields nonsignificant coefficients.
None of the risk interaction terms that Dillingham includes are signi-
ficant. If interaction terms adjust for errors in risk measurement, they
may be unnecessary when more specific risk data is used. Dillingham
observes multicollinearity between the variables for different types of job
risk. This finding supports the supposition that exclusion of other risk
A-15
-------
variables that are correlated with death risk may bias the coefficients of
the death risk variable.
Dillingham obtains an elasticity of 0.0035 value for the coefficient
of death risk in his semilog equation. Since he uses detailed risk data,
error in risk measurement should be low. In exchange for specific risk
data, Dillingham had to limit his sample to New York. Thus, his sample is
more geographically restricted than those of the other studies.
With the exception of Viscusi, all of the studies that have been
reviewed analyze a cross-section of wages. These cross-section studies
attempt to control for personal characteristics which may affect wages.
Under the assumption that certain personal characteristics such as genetics
and personal habits are unlikely to change over time, Brown (17) uses
longitudinal data to estimate the relationship between wages and risk.
Adjusting for such changes over time as marriage, he evaluates workers'
wage-risk tradeoffs by looking at the combinations of wage and risk each
individual accepts over time. This approach assumes technology and
preferences are constant.
Brown uses a sample of 470 workers from the National Longitudinal
t.
Survey. Occupational risk data is taken from the 1967 Society of Actuaries
study. He specifies a semilog form.
Brown estimates an elasticity of 0.0135. As discussed for Thaler and
Rosen, the risk data used may bias the coefficient. The actuarial data
gives deaths by occupation. Brown assigns a value of zero to jobs not
included. No adjustment is made for variations in nonjob death which may
result from differences in risk preferences across occupations. If workers
in risky jobs experience higher rates of non-job deaths, this risk of job
fatality will be overestimated. Also, the average job risk reported by
Brown is high, suggesting workers in his sample may be less risk averse
than the average worker.
A-16
-------
Because several coefficients are insignificant or have signs which
contradict expectations. Brown rejects the hypothesis that the failure of
some other studies to identify expected wage differentials across risk
levels is a result of omission of personal characteristic variables. Among
the variables that are wrong-signed or insignificant are time spent in
school, job strenuousness, and bad working conditions.
The results of the six studies are summarized in Table A-2.* The wage
and risk data used, functional form, average sample risk, and the elasti-
city of the wage with respect to risk are included. As can be seen in the
table, the average risk of death in these samples varies from 1 per 10,000
to 10 per 10,000. The elasticity of death risk varies from 0.0035 to
0.035.
Table A-3 lists the alternative estimates of marginal risk valuations
from the wage compensation studies discussed in this section. In order to
have a common basis for comparison, these estimates are calculated for the
mean county wage in this analysis and are expressed in 1980 dollars. The
values of a unit reduction of 1 x 10 in annual mortality risk range from
$0.30 to $5.24. The two values derived from the surveys by Jones-Lee and
Acton bound this range with estimates of $0.05 to $8.80 million.
The lower end of our range is consistent with the risk valuation
estimate of $0.47 derived from Blomquist's (19) model of seat-belt use and
Cooper and Rice's (1) human capital estimates of $0.25 to $0.46 for the age
group with the highest discounted earnings. The lower end is also consis-
tent with the value derived by Portney (20) by combining estimates of the
effect of pollution on property values and health risks.
Dillingham's study yields both the lowest elasticity and marginal risk
valuation. A low marginal risk valuation is also estimated by Thaler and
* For a further review, see Smith (18).
A-17
-------
H
os
BE]
OS
O
M
i
w
1
§
OS
£
•H
0
TH
44
(A
01
W
0 0
TH 0
Pi O
4 (0 O
03 -^ TH
OS
(4
eo «
> Pi
•< -~
iH
01
a
o a
•H IH
44 O
0 fj,
a
pj
*
at
44
Of
a
^4
CO*
•H
«
0
M
•H
OS
o
tH
Pi
a
at
OS
• «t
at
44
at
O
O
eo
at
*
^l
•a
0
•M
VI
\f)
(f)
tH
o
•
o
»
O
TH fr>
at H
d 0
O 03
•H
44 fH
at at
Z d
d
*
o
(4
*
<*}
O
0
•
o
f*l
f*.
•
eo
O
TH
TH
8
0
OS
IH
0
Pi <*
M a *o
(4 0 14
QUO
{H o
M 0
jfc * O-i
o d
Z o d
a o
O M TH
p. W 44
as o- at
TH ^
M at
H CM d
000
at 4rf
^ 8 o
<9 0
Z U TH
at 4rf
O »H «»
r- -H d
ON O o
iH O O
1
eo
d
*H
fH
^4
TH
Q
0
•
O
1
•*
to
o
o
00
o
•0
o\
•
0
•> 0
00 TH
O -M
fH at
•H (H
a -o
OS 0
0 at
h -w
0 03
(JQ
f4
m o
t- rf>
ON at
*H J
M
d «
0 ft4
•^ n
•M O
al >
fH
0 0
PI a
OH 44
1
44 fH
'd ••«
0 0
H *M
(H
U >t
4) ff)
r- n o\
<^ ?3 •»
IH « m
a
0
^••J
o
4(
*
»
*
00
O
iH
TH
a
o
OS
o «
o
0 TH
at 44
o J
o d
•H O ••
0 44 d*
d at TH
O 0 fH (4
0 44 00
H TH ft 44
fd o o
CM £ Ok at
O ^w
* CO ^ 0
>« >voo d d
« 44 fH 0 at
> TH . M a
vi d »
tH M 0 00
r- o 0 cn > fH O
wi d o
0 0 OS
OS 44
P> O IA
NO Pi fP
OS Pi O
•H O -^
o3
h
0 d
fH 0
*N Q
H M
w*t
p<
O
•
O
^s^
oo
fH
•
^
h
01
O
d
TH
hj
M
0
O 44
M
0 TH
at 44
0 at
W -M
0 OS
CO
M
as o
NO ft
as at
<— i »J
d
at o
00 0
0
S >>
-ff
^M Cl
O 44
OS
»
•»(* ^"4
TH at NO
d d as
P TH •*
-0
O 0 •«
P- -M M
1 TH at
r~ oo fH
NO d fH
as o o
fH kj O
TH
(0
O
(A
•H
d
o
at
O
TH
CM
•H
O
Pi
(O
ca
TH
.d
eo
d
T4
M
0
O
TH •
•M -O
at O
5. Z
•
CM
o
O
d
d
•M O
•H
d -w
•O V4
o o
ca CM
at d
CQ M
A-18
-------
Table A-3
ALTERNATIVE ESTIMATES OF MABGINAL RISK VALUATIONS
(1980 dollars)*
Study
Value of a Unit Reduction of
1 x 10"6 in Annual Mortality Risk
Brown
Dillingham
Olson**
Smith (1973 sample)
Thaler & Rosen (linear)"*"
Viscusi (linear)
Range:
Mean:
$0.87
$0.30
$5.25
$2.65
$0.43
. $3.26
$0.30 - $5.25
$2.13
* Adjusted to 1980 by CPI, all items. Evaluated at mean wage for the
counties in our analysis.
** Average of results for two alternative specifications.
No interaction terms.
Rosen.* The low valuations in these two studies may be partially explained
by the high level of job risk in the two samples. The workers sampled may
have self-selected to a high level of occupational risk because they have a
low risk aversion. Then, the premium sampled workers demand for acceptance
of marginal risk will be lower than the premium most workers demand. In
addition. Thaler and Rosen's method of measuring j ob risk may bias their
results downwards.
* While their marginal valuation is low. Thaler and Rosen's elasticity is
relatively high. They look at a sample of jobs with very high risk.
However, since the average sample wage is low relative to the level of
risk, the absolute premium for changes in risk is not high.
A-19
-------
Therefore, the estimates of Dillingham and Thaler and Rosen will be
used as lover bound estimates of the marginal risk valuation of the popula-
tion in our analysis.* Averaging the two estimates, a minimum estimate of
$0.36 is obtained.
Olson's study has the highest elasticity and marginal risk valuation.
This study is the only one to use a semilog quadratic form. When the
quadratic term is eliminated, these values decrease by 50 percent.
There is not sufficient information for which to make a definitive
determination of the actual shape of the market equilibrium curve. No
other applications of the semilog quadratic form are available for compari-
son with Olson's findings. Therefore, since his marginal valuation exceeds
the next highest ones by almost $3.00, it is not selected as a maximum
estimate.
A maximum estimate of $2.80 is chosen. This value is between the
results of Viscusi and Smith, the studies with the next highest valuations.
Smith's exclusion of job characteristic variables and the lower levels of
sample risk may partially account for the difference between the valuations
derived by Smith and Viscusi and those of Dillingham and Thaler and Rosen.
The average of the minimum and maximum estimates, $1.58, is selected
as a point estimate. This figure is slightly below the average of the
valuations for all the studies and above the average when Olson's results
are excluded. The three estimates are summarized in Table A-4.**
* Thaler and Rosen's Equation 1 estimate for the linear form is used.
Some of their other specifications yield higher valuations.
** It is assumed that the value of risk reduction is constant in real terms
over the period of our analysis.
A-20
-------
Table A-4
ESTIMATES OF THE VALUE OF A UNIT REDUCTION OF 1 x 10~6
IN ANNUAL MORTALITY RISK
(1980 Dollars)
Minimum
Point
Maximum
$0.36
$1.58
$2.80
Limitations of the Tace CD
ansation Method
In addition to the previously discussed technical problems encountered
in empirical applications of the wage compensation method, the theory
itself has major limitations. First, the wage compensation theory assumes
that workers make an informed choice between different wage-risk combina-
tions offered by firms. It is assumed that the measure of actual risk
selected approximates workers' subjective perception of risk.*
Workers, however, may be ignorant of some of the risks associated with
different jobs. Several wage-compensation studies report positive inter-
action terms for death-risk and unionization. One possible explanation of
this result is that unions have better information than a nonunionized
worker and demand a premium for the risk they identify. Nonunionized
workers, or any uninformed workers, may accept low risk premiums because
they are ignorant of the actual risk. Risk valuations derived from studies
of the behavior of uninformed workers may underestimate valuations under
full knowledge. Furthermore, as discussed in the review of surveys,
workers' choices may be inconsistent even if full risk information is
available.
* See Viscusi (16) for study of accuracy of workers' perceptions.
A-21
-------
Second, the wage compensation theory assumes a competitive job market
with market clearing wages. The observed (wage, risk) combinations only
reflect willingness-to-pay if the market is in competitive equilibrium. An
alternative explanation of the positive union-risk term is that the labor
market may deviate from perfect competition. If workers in risky
industries are highly unionized or unions use their power for negotiations
for risk premiums, premiums may be above the levels which would prevail in
a competitive market. Then, values derived from observation of risk
premiums in unionized industries may overestimate actual willingness-to-pay
for risk reduction.* In addition, discrimination may exist in the labor
market. The jobs and wages available to certain groups may be limited.
One group may only be offered high—risk, low-wage jobs rejected by other
workers. The coefficient of death risk will underestimate the amount that
these workers would require to accept the risk if offered a range of wage-
risk combinations.
Third, .the wage compensation studies only identify the amount a worker
will pay for a reduction in the probability of his or her own death. The
willingness-to-pay of individuals close to the worker for the risk reduc-
tion have not been considered.
Although others' willingness-to-pay may be significant, few studies
examine this component. Needleman (21) examines the incidence of kidney
donation by relatives to measure willingness-to-pay for reduction of a
relative's risk of death. The donation of kidneys to relatives indicates
that there exists a high degree of concern for the survival of others.
However, Needleman's estimates of the exact relationship of willingness-to-
pay for a reduction in a relative's risk and for a reduction in one's own
risk (25 to 100 percent) depend on a number of strong assumptions. These
assumptions include the number of operations an individual would undertake
to reduce his or her own risk.
* See Olson (IS), page 183.
A-22
-------
Even though the exact magnitude of others' willingness- to-pay cannot
be calculated, it is expected that the sum of individuals' willingnesses-
to-pay for a reduction in the probability of his or her death will
underestimate total willingness-to-pay for risk reduction.
Fourth, in addition to affecting the utility of individuals' whose
life expectancy is prolonged and the people close to them, changes in
mortality will affect the welfare of society as a whole. (The human
capital and net output approaches look at the effect on society's output
only and do not directly consider the utility of affected individuals.)
Arthur (22) argues that the effects on both the utility affected indivi-
duals derive from prolonged life and the output of the economy should be
evaluated.
For each individual, he subtracts the net consumption society foregoes
to lengthen life from the individual's expected utility of extra years.
The net consumption cost is measured by the value of expected increases in
labor years and reproduction of children minus increased consumption
support costs. The total value of mortality reduction depends on the
weights attached to the utility of living versus pure consumption available
to society. The weight attached to enjoyment of life relative to
consumption should increase with the affluence of society. In our
analysis, it is assumed that the value of living dominates the pure
consumption effect. The general equilibrium effects of the increase in
consumption support costs that accompanies decreased mortality are ignored.
Application, of the Taga Co»oan
-------
The wage compensation studies find that individuals are willing to pay
for small reductions in risk. From these studies, estimates of the value
of marginal risk reductions have been derived in this section. These
values can be applied to the risk reductions in our analysis.
A numerical example will help illustrate this result. Because of an
air quality improvement, the mortality rate in a county with a population
wJ«
of 100,000 is reduced by 2 z 10 . Wage compensation studies yield a value
of $1.58 for a unit reduction of 1 x 10 in annual mortality risk. Thus,
each individual will be willing to pay ($1.58) (0.2) = $3.16 for the
mortality rate reduction. The total willingness to pay of the population
is ($3.16)(100,000) = $316,000.
The above discussion shows how the results of the wage compensation
studies can be directly applied to benefit calculations. The health
studies in Sections 3 and 4 will be used to calculate the reduction in
annual mortality risk under each standard. The estimates of the value of
marginal risk reduction will be used to measure the amount each individual
will pay for the reduction in his or her mortality risk.
METHODS FOR VALUING DEDUCTIONS IN MORBIDITY
In addition to reductions in mortality. Sections 3 and 4 identify
reductions in morbidity resulting from reduced levels of particulate
matter. Our estimate of the benefit of mortality reduction is based on
willingness to pay for reductions in employment-related deaths. Most wage
compensation studies only consider accident fatalities, not deaths related
to chronic work conditions. Accidents usually involve pre-death pain,
suffering, and inconvenience of short duration. Thus, the pre-death
illness accompanying pollution-related fatalities is considered in the
morbidity benefits categories which includes reductions in both morbidity
preceding mortality and morbidity associated with non-fatal incidents.
Following the approach used for measurement of benefits of mortality
reductions, it would seem logical also to estimate the willingness-to-pay
A-24
-------
for the decrease in morbidity. However, there are no comparable studies of
the implicit willingness-to-pay for changes in morbidity. Some wage com-
pensation studies estimate a coefficient for nonfatal accidents. Extrapo-
lation from payment for reduction of job accidents to valuation of
morbidity reduction, however, requires unacceptably gross assumptions.
Since there is not sufficient information on which to base a measure
of willingness—to—pay, an alternative method to value reductions in
morbidity must be applied. The effect of a decrease in morbidity on the
economy will be evaluated in three parts: 1) reductions in the loss of
output in the workplace due to illness (labor productivity benefits), 2)
reductions in non-work days due to illness, and 3) reductions in the
consumption of medical services.
Reductions in the Loss of Output
In Sections 3 and 4, the effect of a change in the ambient level of
particulate matter on the number of work days lost has been estimated. As
the number of work loss days decreases, output will increase. Assuming
that the wage rate is equal to the marginal revenue product of each worker,
this increase in output resulting from fewer work loss days will be valued
at the average daily wage. Since the benefit estimates in this analysis
are based on county level population data, county level wage data will be
used to estimate the labor productivity benefits. Consequently, the labor
productivity benefits in the i county are equal to:
PRODi = (AWLD^ ' WAGEi (A.3)
where PROD. = labor productivity benefits in county i
AWLD. = change in total work loss days resulting from a change
in the ambient level of particulate matter in county i
WAGE^ - average daily wage in county i.
The value of the increased output resulting from reduced work loss
days underestimates the willingness-to-pay for a reduction in morbidity for
A-25
-------
a number of reasons. First, available data on WLD only consider the number
of days on which more than one-half day of work is lost. Therefore,
reductions in productivity resulting from illness that do not cause a
worker to miss over one-half day of work will not be measured in our
estimates. Second, the reduction in pain, suffering, and inconvenience
that results from reduced morbidity is not measured by the change in
output.
Reductions in Restricted Activity Davs
A decrease in morbidity will result in a reduction in non-work days on
which activity is restricted because of illness (RAD) as well as in work-
loss days. No increase in output in the workplace is associated with
reduced RAD. However, productivity in activities outside the workplace
will increase, and pain and suffering will decrease. For our calculations,
we assume that the benefits from each RAD eliminated are given by one-half
of the average daily wage for the counties in our analysis. Consequently,
the benefits of a reduction in RAD in the i county are:
(A.4)
where Bei = benefit of RAD.
= change in number of RAD resulting from a change in the
ambient level of PM in county i.
Wage, = average daily wage.
Because of data constraints, the benefits resulting from reductions in non-
work days on which an individual restricts his or her activity for only
part of the day are not measured by this calculation. Furthermore, the
full benefits of reduced productivity, pain, suffering, and inconvenience
may not be captured by this method of valuation.
A-26
-------
;ion of Medical Serrices
In addition to a reduction in both work-loss and reduced-activity
days, the benefits of a reduction in-morbidity will include decreased
medical expenditures. Because information on medical expenditures is not
available at the county level, national expenditure data are used in
Sections 3 and 4. The expenditure figures used in benefit calculations are
detailed in these sections.
Equation (A.S) provides an example of the type of calculation that
will be used in Section 3 to estimate the morbidity benefits in teras of
reduced medical expenditures in the i county on acute respiratory
disease, of reductions in particulate matter.
AEZP.
(AINCi) ' ACi
(A.S)
where
AEIP.
AINC,
AC,
change in direct medical expenditures on acute respira-
tory disease in the i county.
change in the number of acute respiratory disease inci-
dents in the i county.
average direct medical expenditure per acute respiratory
disease incident.
CONCLUSION
In this section, the valuation of health improvements resulting from
reduced levels of particulate matter has been discussed. A range of
estimates of the willingness-to-pay for marginal reductions in risk of
death has been developed. These estimates can be used to value the
mortality reductions identified in Sections 3 and 4.
As discussed in the section, wage compensation theory and empirical
studies of wage differentials suffer from a number of weaknesses. Even if
these shortcomings are ignored, application of study results to our calcu-
lations will yield only approximate estimates of benefits because of the
A-27
-------
following factors: 1) the results reflect willingness-to-pay for reduction
of risk for job fatality, not risk of death from other sources; 2) the
estimates exclude the valuation of other individuals in society; and 3) the
estimates only apply to marginal risk reductions.
In addition, the results are based on the willingness-to-pay for
voluntarily assumed risk of sample groups whose selection is restricted
according to personal and job characteristics. As discussed previously,
individuals' risk valuations will vary with such factors as nonlabor
income, initial risk level, and cost of risk bearing. For example, other
factors being equal, a worker has a lower level of uncontrollable risk, and
consequently a lower risk valuation, than an elderly person. Without
detailed examination of all factors affecting risk valuation and their
distribution across the populations in the studies and our analysis,
further conclusions on the bias introduced by application of results for
workers to children and the elderly cannot be made. Despite these weak-
nesses, the figures derived in this section provide an indication of the
magnitude of the economic benefits of mortality reduction.
A method to calculate the benefits of reductions in morbidity has also
been developed. The effect of this reduction on medical resource use and
output is considered. While this resource cost-based approach fails to
measure total willingness to pay for changes in morbidity, it provides a
rough, lower-bound estimate of the impact of morbidity changes on the
economic resources available.
REFERENCES
1. Cooper, B. S. and D. P. Rice. The Economic Cost of Illness Revisited.
Social Security Bulletin, 39:21-36, 1976.
2. Lave, L. B. and E. P. Seskin. Air Pollution and Human Health. Johns
Hopkins University Press, Baltimore, MD, 1977.
3. Lui, B. and E. S. Yu. Physical and Economic Damage Functions for Air
Pollutants by Receptor. U.S. Environmental Protection Agency,
Corvallis, Oregon. 1976.
A-28
-------
4. Conley, B. C. The Value of Human Life in the Demand for Safety.
American Economic Review. 66:45-55, 1976.
5. Linneroth, J. The Value of Human Life: A Review of the Models.
Economic Inquiry. 17:52-74, 1979.
6. Usher, D. An Imputation to the Measure of Economic Growth for Changes
in Life Expectancy. In: The Measurement of Economic and Social
Performance, Milton Moss, ed. NBER, New York, 1973.
7. U.S. General Accounting Office. Approaches Toward Valuation of Human
Life by Certain Federal Agencies. PAD-82-21, November 9, 1981.
8. Acton, J. P. Evaluating Public Programs to Save Lives: The Case of
Heart Attacks. Rand Corporation, Santa Monica, 1973.
9. Jones-Lee, M. W. The Value of Life: An Economic Analysis. Univer-
sity of Chicago Press, Chicago, 1976.
10. Dillingham, A. E. The Injury Risk Structure of Occupation and Wages.
Ph.D. Thesis, Cornell University, Cornell, New York, 1979.
11. Smith, V. K. The Role of Site and Job Characteristics in Hedonic
Wage Models. Unpublished paper, University of North Carolina, Chapel
Hill, 1981.
12. Freeman, A. M., III. The Benefits of Air and Water Pollution Control:
A Review and Synthesis of Recent Estimates. Report for Council on
Environmental Quality, December, 1979.
13. Thaler, R. H. and S. Rosen. The Value of Saving A Life: Evidence
from the Labor Market. Household Production and Consumption. N. E.
Terleckyj, ed. Columbia University Press, New York, 1976.
14. Smith, R. S. The Occupational Safety and Health Act: Its Goals and
Its Achievements. American Enterprise Institute for Public Policy
Research, Washington, D.C., 1976.
15. Olson, C. A., An Analysis of Wage Differentials Received by Workers on
Dangerous Jobs. Journal of Human Resources, XVI:167-185, 1981.
16. Viscusi, W. K. Labor Market Valuations of Life and Limb: Empirical
Evidence and Policy Implications. Public Policy, 26:359-386, 1976.
17. Brown, C. Equalizing Differences in the Labor Market, Quarterly
Journal of Economics, XCIV:113-134, 1980.
18. Smith, R. S. Compensating Wage Differentials and Public Policy. A
Review. Industrial and Labor Relations Review, Vol. 32:339-352, 1979.
19. Blomquist, G. Value of Life Saving: Implications of Consumption
Activity, Journal of Political Economy, 87:540-558, 1979.
A-29
-------
20. Portney, P. R. Housing Prices, Health Effects, and Valuing Reductions
in Risk of Death. Journal of Environmental Economics and Management,
8:72-78, 1981.
21. Needleman, L. Valuing Other People's Lives. Manchester School of
Economic and Social Studies, 44:309-342, 1976.
22. Arthur, W. B. The Economics of Risk to Life. American Economics
Review 71:54-64, 1981.
A-30
------- |