MATHTECH The Technical Research and Consulting Division of Mathematica, Inc. BENEFIT AND NET BENEFIT ANALYSIS OF ALTERNATIVE NATIONAL AMBIENT AIR QUALITY STANDARDS FOR PABTICULATE MATTER ------- BBNBF1T AMD HE! K8NKVIT ANALYSIS OF ALTERNATIYB NATIONAL AMBIENr All QUALITY STANDARDS FOR PAETTCULATE VASTER YQUHB 17 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 March 1983 ------- BENEFIT AND NET BENEFIT ANALYSIS OF ALTERNATIVE NATIONAL AMBIENT AIR QUALITY STANDARDS FOR PARTICULATE HATTER 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 Exetutive 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 reconniendations 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. ------- PREFAcE 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: The Benefit Analysis Section 2: The Net Benefit Analysis Volume El Section 3: Health Effects Studies in the Epidemiology Literature Section 4: Health Effects Studies in the Economics Literature Appendix: Valuation of Health Improvements Volume III Section 5: Residential Property Value Studies Section 6: Hedonic Wage Studies Section 7: Economic Benefits of Reduced Soiling Section 8: Benefits of National Visibility Standards Volume IV Section 9: Air Quality Data and Standards Section 10: Selected Methodological Issues Volume V Section 11: Supplementary Tables iv ------- AcL GXENTS While preparing this report, we had the benefit of advice, comments and other assistance from many individuals. Allen Basala, the EPA Project Officer, and lames 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 Renry Thomas, Jeff Cohen, John Bachman, John Names, 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 Rarford, 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 Watsom Other Mathtech staff who assisted us in various ways were Donald Wise, Gary Labovich, and Robert 3. 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. V ------- VOLUIIE IV Section Pane 9. AIR QUALITY DATA AND STANDARDS Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9—1 Sources of Data 9—2 Design Value File 9—2 Pre—Control Air Quality Data File . 9—3 Post—Control Air Quality Data File 9—5 ConsistencyCheckiofData 9—12 Development of Data for Benefits Analysis ... 9—17 Geographical Aggregation 9—19 Temporallnterpolation 9—21 Projections of TSP Annual Arithmetic Means .......... 9—23 Index of Exposure 9—24 Distribution of Air Quality Changes 9—29 Definition of B—Scenario Concentrations ............. 9—35 Benefit Calculations 9—40 TypesofStandards 9—42 Parameters Used in the Benefits Analysis .... 9—42 TabulationofBenefits 9—47 SummaryofAirQualityData............. .... 9—47 References . . 9—49 vi ------- CONTENTS (Continued) Section Page 10. SELECTED METhODOLOGICAL ISSUES Procedures for Calculation of Aggregate Benefits ......... 10—2 Introduction 10—2 Criteria for Assessing Study Quality 10—4 Application of Study Evaluation Criteria 10—6 Alternative Aggregation Procedures 10—13 Results of the Aggregation Procedures 10—17 Applicable Exposure Ranges for Benefit Calculations 10—17 Introduction . . . . . . . 10—17 Concentration Ranges Suggested by CASAC and the EPA/OAQPS Staff Paper 10—19 Applicability of Staff Paper Lower Bounds to the B enef it Analysis . . . . 10—24 Practical Constraints in Applying the Staff Paper Lower Bounds . 10—26 ResultsoftheCalculations 10—29 Chicago Area Case Study . . . . . . . . . . . . . 10—29 Introduction 10—29 Data 10—35 Approach ... . 1036 P indings 1038 Limitations 1039 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1042 vi i ------- FIGURES VOLUME IV Figure No, Page 9—i. Air Quality Data . 9—6 9—2. Air Quality Data With Complete Attainment 9—11 9—3. Time Paths for Air Quality Data 9—22 9—4. EqualMagnitude Improvement 9—31 9—5. EqualPercent lmprovement...................... . . .... .. .. 9—32 9—6. Localizedlmprovement 9—34 9—7. Comparison of A and B Scenarios . 9—36 9—8. DevelopmentofB—ScenarioData................ .. 9—39 9—9. Benefits of Ambient Air Quality Improvements 9—45 10—1. Constrained Estimation of Concentration—Response Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10—28 10—2. EqualMagnitude Improvement 10—31 10—3. EqualPercentimprovement...... .... 10—33 10—4 • Local ized Improvement . . . . . . • . . . . . . . . . . . . , 10—3 4 V Li ]. ------- TABLRS VOLUIB IV Table No. Paae 9—1. AmbientAirQualityStafldardS 9—7 9—2. Counties in Post—Control Files . 9—8 9—3. Swnmary of Consistency Checks for A—Scenario Data 9—18 9—4. Subcounty and Aggregated Observations in Air Quality Data Files . 9—20 9—5. Distribution of SAROAD Monitors by County 9—26 9—6. CountyAverageCorrectionFactOrS................ . ’ .... 9—27 9—7. Review of Pollution Monitors Used by Studies 9—28 9—8. Problem Counties in B—Scenario Data Base . 9—41 9—9. Ambient Air Quality Standards . 943 9—10. Benefit Parameter Values ..... . ...... .... ..... .. .. .. ..... . 9—46 9—11. EPA Federal Administrative Regions by State 9—48 10—1. Application of Study Evaluation Criteria ...... 10—7 10—2. Alternative AggregatiouProcedures ....................... 10—14 10—3. Incremental Benefits for the PM1O (70, 250) Scenario B Standard 10—18 10—4. Summary of Concentration Ranges in EPA/OAOPS Staff Paper . 10—20 10—S. Ratio of Census Division Estimates to Cook County Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10—38 ix ------- TABLES (Continued) Table No. Page 10—6. Benefit Estimates for Two Methods of Air Quality Aggregation and for Benefit Aggregation Procedure B 10—40 10—7. Benefit Estimates for Two Methods of Air Quality Aggregation and for Benefit Aggregation Procedure C 10—40 10—8. Benefit Estimates for Two Methods of Air Quality Aggregation and for Benefit Aggregation Procedure D 10—41 x ------- SECTION 9 AIR QUALITY DATA AND STANDARDS ------- SECTION 9 AIR QUALITY flAIL AND STANDARDS Analyses of the benefits and costs of alternative ambient air quality standards require very different input data. flowever, both analyses use data on ambient concentrations of air pollution. Consequently, if benefits and costs are to be compared, care must be taken to ensure that the air quality data used are consistent across the two analyses. In the analysis of control costs, the focus is on the control of source emissions. The analysis identifies the degree of emission control required to achieve the ambient concentration of the controlled pollutant allowable under the proposed standard. The analysis of benefits is based on an examination of pollutant—related impacts with and without the pro- posed standard. Because the impacts depend primarily on the ambient concentrations of pollution, the appropriate air pollution data for the benefit analysis are ambient air quality concentrations. Thus, data on ambient concentrations provide the linkage between the cost and benefit analyses. Clearly, the interdependence between the benefit and cost analysis data requirements must be recognized if a consistent benefit—cost analysis is to be completed. It is not enough to assume that promulgation of the standard will result in every area being just in compliance. In reality, the limited number and discreteness of available means of controlling emissions can lead to a range of resultant concentration levels across geographical areas. 9 -1 ------- The analysis in this study uses ambient air quality data generated by the contractor performing the cost analysis. As a consequence, both the benefit and cost analyses are based on a common set of air quality data. The purpose of this section is to describe the air quality data used in the benefits analysis. The discussion focuses not only on the scope of available data, but also on the manner in which the data are used in the analysis. In addition, the section includes a description of the alterna- tive air quality standards examined in the study and the conditions under which benefits estimates are derived. If benefits and costs are to be compared, the standards analyzed and the method of reporting dollar values must also be defined consistently across the two analyses. S4RJR S OF DATA This subsection describes the air quality data provided by the cost contractor. The discussion begins with an outline of the source and scope of the historical air quality data used in the analysis. This is followed by a review of the air quality data files generated for future years, both with and without the proposed standards. Finally, the results of a variety of data consistency checks are reported. Additional information on the air quality data base is available in Reference (1). Desian Vali.e File The primary source of air quality data for the analysis is the “design value” file maintained by the U.S. EPA for the RIA analysis. This file contains an index of air quality data by county (or subcounty area) for a variety of averaging times. The index (referred to as the design value) is defined as the concentration reported at the monitor(s) with the highest reading of annual mean and 24—hour observed second high for TSP within a given county. 5 The data used in defining the county design values are * In some cases, design values are adjusted to be consistent with county attainment/nonattainment designations in the 1979 Federal Register or EPA Regional Office classifications. 9 -Z ------- taken from EPA ’s Storage and Retrieval of Aerometric Data System (SAROAD). These data are drawn primarily from 1977 and 1978, but in some cases may reflect conditions from as far back as 1975. For purposes of this study, we view the design value data as nominally representative of air quality conditions in 1978. The specific data available in the design value file include measures of TSP in terms of annual geometric means, annual arithmetic means, annual second highest 24—hour average concentrations observed, and 24—hour expected value concentrations (expected to occur once per year). The latter values are generated by assuming the air quality data fit an exponential distribution. Data for these statistical measures are available for 1,230 county or subcounty areas. Although the scope is limited to about 40 percent of all U.S. counties, the presumption is that the included counties represent those areas that are likely to be in nonattainment with plausible alterna- tive standards. For some county or subcounty entries, one or more of the statistical measures may not be reported. In this case, the cost contractor developed. regression equations to generate the necessary data. These transformations were carried out prior to our acquisition of the data. A discussion of the process for replacing missing values can be found in Reference (1). Pro—Control Air Qmality Data Pu . The design value file is taken to be representative of air quality conditions in 1978. In order to project future air quality, a modified linear rollback model was applied to project future emissions growth. 5 The estimates of future emissions were based on estimates of the growth in new sources, and on the rate of replacement or retirement of sources in exis- tence in 1978. These emission estimates are made under the assumption that $ See Reference (1) for a discussion of rollback models. 9—3 ------- unretired plants are controlled at 1978 control levels, One—half of the replacement sources are controlled at 1978 control levels, while new sources and the other replacement sources are controlled at BACT levels (NSPS and other new source control requirements). However, because this initial (or baseline) situation represents the “without standard” case, we refer to these data as pre-control data. Scope of Pre—Control Data — Data for the pre—control file are provided for a historical year, 1978 (these values correspond to those found in the design value file), and for four future years, 1985, 1987, 1989, and 1995. Four distinct pollution measures are reported for each county and each year.* These measures are: • TSP annual geometric mean. • TSP 24—hour observed second high. • PM1O annual arithmetic mean. • PM1O 24—hour expected value. The designation PM1O refers to particles which are less than 10 microns in diameter. Thus, concentrations of PM1O represent a subset of TSP concentrations. As noted in the Staff Paper for PM (2), there is reason to believe that PM1O concentrations are more relevant for assessing health impacts than the broader measure of TSP. For this reason, many of the alternative standards analyzed in this study are defined in terms of limits on allowable PM1O concentrations. P1(10 Conyer5ions — Since all entries in the design value file are expressed in terms of TSP measures, a conversion factor is required to determine the PM1O $ The geographic scope of the pre—control data file corresponds to that in the design value file. 9-4 ------- concentrations for the pre—control data file. This is accomplished by multiplying the TSP annual arithmetic mean and 24—hour expected value in the design value file by O.55. This yields estimates (in 1978) for PM1O annual arithmetic mean and 24—hour expected value, respectively. Estimates of PM1O for future years are based on proportional relationships discussed later in this section. These conversion calculations were completed prior to our acquisition of the data base.** Post—Control Air Quality Data File With the implementation and enforcement of an ambient standard, air quality can be expected to improve relative to the pre—control case. An example of this situation is portrayed in Figure 9—1. In the figure, the segment AB represents concentrations across time when no ambient standard is in place. After a standard is imposed, concentrations decrease to the level of the standard in year t 1 , but increase over time along the segment CD because of net growth in emissions (and hence higher concentrations of the pollutant). The segment AB represents the pre—control data, while the segment CD represents the post—control data. If all benefits are assumed to occur within the timeframe of t 1 to t 2 . then the benefits to society due to imposition of the standard are given by the health and welfare improve- ments resulting from the air quality change represented by area ABDC. Aabient Standards — The data in the post—control file depend on the particular standard imposed. Table 9—1 lists the six alternative ambient standards considered • This factor was developed through analysis of data collected by EPA’s Inhalable Particulate Network (3). ** One conversion made after we acquired the air quality data involved defining a relationship between TSP and British smoke. A discussion of this conversion procedure is presented in Appendix 3A of Section 3. + The monotonically increasing nature of segment AB in Figure 9—1 is for expository purposes. The segment could also decrease or remain constant through some or all of the interval t 1 to t 2 . 9—5 ------- Figure 9-1. Air Quality Data Nonattairiment Due to Emissions Growth Concentration Pre—Control Data — — Benefits Standard Post-Control Data ti t 2 Time 9-6 ------- Table 9—1 AMBIENT AIR QUALITY STANDARDS Pollutant Level (pg/rn 3 ) Averaging Time Implementation Date Maintenance Date P1 110 70 250 Annual Arithmetic Mean 2 4—Hour Expected Value 1989 1989 1995 1995 P1110 55 Annual Arithmetic Mean 1989 1995 P1110 55 250 Annual Arithmetic Mean 24—Hour Expected Value 1989 1989 1995 1995 P1110 55 150 Annual Arithmetic Mean 24—Hour Expected Value 1989 1989 1995 1995 TSP 75 260 Annual Geometric Mean 24—Hour Second High 1987 1987 1995 1995 TSP 150 24—Hour Second High 1987 1995 ------- in this analysis. The order in the table goes from least stringent (PM1O 70/250) to most stringent (TSP 150). As the standard considered becomes more strict, the post—control concentrations (segment CD in Figure 9—1) would fall. Since the pre—control data are unaltered by the choice of standards, this implies that total benefits become larger the more strict the standard. The data available in each of the six post—control files are defined in a manner consistent with the pre—control data. In particular, data are available for the same measurements, averaging times, and years as in the pre—control file. There is, however, a difference in scope. While the pre—control file contains data for all counties in the design value file, each of the post—control files contain data only for those counties that are not in attainment with the relevant standard in at least one of the years under consideration. As shown in Table 9—1, this timeframe encom- passes the period 1987 to 1995 for TSP standards, and 1989 to 1995 for the PM1O standards. The number of counties included in the analysis for each of the ambient standards is shown in Table 9—2. These observations represent a Table 9—2 COTJNrIES IN POST—CONTROL FILES Standard Number of Counties PM1O (70/250) 93 PM1O (55) 161 PM1O (55/250) 163 PM1O (55/150) 297 TSP (75/260) 282 TSP (150) 499 9-8 ------- subset of total available counties. This occurs because aggregation across county parts was performed and several counties were discarded from the analysis after consistency checks of the data. Both of these procedures are discussed more fully below. *.sidu*1 Nonattainment — In Figure 9—1, point C is located such that the standard is just attained in period t 1 . ilowever, in some cases, it may not be possible to attain the standard in the implementation year. This can occur if all available means of controlling emissions are applied prior to standard attainment. This situation does occur with the data provided us, and a county for which this happens is referred to as being in residual non— attainment. Nonattainment Du. to Growth — Counties may also be in nonattainment with a standard due to growth in emissions across time. For example, in Figure 9—1, segment CD is drawn such that the standard is exceeded in the years after t 1 . In the context of benefits analysis, this implies that the relevant area for benefits calculations is ABDC. If the analysis were to assume that the standard was just met in all years t 1 to t 2 , then the estimates of benefits (relative to control costs) would be overstated by the benefits associated with area DE in Figure 9—1. The air quality modeling effort attempted to account for nonattainment due to growth. This was done by applying all available control strategies in year t, for example, in order to bring concentration levels in year t 2 as close to the standard as possible. In terms of Figure 9—1, this is equivalent to shifting segment CD downward. Note that residual nonattain— ment is still possible because the available control strategies limit the shift in segment CD. Thus, the data provided in the post—control files permit both residual nonattainment and nonattainment due to growth. For purposes of discussion, we will call this situation the “A” scenario. As 9-9 ------- an alternative to this scenario, a “B” scenario was also defined. In this scenario, air quality was reduced to a sufficient degree in the implementa- tion year to ensure maintenance of the standard over the time period analyzed. In order to achieve this degree of control, it was necessary to apply pseudo—controls at a fixed cost per microgram reduction. These procedures are discussed more fully in Reference (1). An example of the B— scenario situation is shown in Figure 9—2. In this figure, if the post— control data are assumed to lie along the segment FE, then benefits are evaluated in terms of the area ABEF. In order to derive the appropriate concentration levels for the B scenario, the post—control files were transformed using relationships described later in this section. These transformations are consistent with those used by the cost contractor in developing B—scenario estimates of control costs. Although it is possible that there will be nonattainment when a particular standard is actually implemented, the maintained assumption in this analysis is that all counties are in attainment throughout the study period. Consequently, benefits estimates róported in subsequent sections of this study represent B—scenario benefits. Benefits estimates from A—scenario post—control data are provided for one of the standards as a sensitivity check. Background Data — In addition to the concentration levels, the post—control data files contain information on background levels of TSP concentrations. The background concentrations are defined to be those concentrations that are not sensitive to control strategies used in the analysis. Thus, the background levels represent natural sources, small sources, and uncon- trolled emission contributions from outside the county. While background * In the benefits analysis, B—scenario refers to the total change from the pre—control data to complete attainment. In the cost analysis, the B— scenario concept is defined as the increment to complete attainment, given A—scenario changes. This difference in definition has been accounted for in comparisons of benefits and costs. 9-10 ------- Post-Control Data “Type A” Post-Control Data “Type B” Figure 9-2. Air Quality Data With Complete Attainment Pre—Control Data B D Standard t 1 t 2 Time 9—11 ------- data are defined for each. county, the data are develo:ped from regional averages (across states). Thus, all counties in a state will likely be assigned the same background values.* Furthermore, all background valties are assumed to remain constant across time. Background data are available in terms of annual arithmetic means and 24—hour second high measures. The corresponding background levels for PM1O arithmetic mean and expected value are derived by multiplying the TSP background values by 0.55. The background value for TSP geometric and arithmetic means are assumed to be equal. Definition of the background levels is an important part of the analysis because it sets a lower bound for the post—control data. This becomes a binding constraint for several counties in the B—scenario (full attainment) analysis when all counties are driven into attainment for all years analyzed. Cons ist.ncv eckz of D.tz Prior to using the pre—control and post—control (A scenario) data files for calculating benefits, a series of consistency checks were carried out to determine if the data met certain reasonable conditions. The condi- tions tested were: • One or more of the standards was violated in each county in the pro—control data file. • Post—control data were less than or equal to pre—control data. • Concentration levels were greater than background. • Arithmetic means were greater than or equal to geometric means. * A small number of counties are assigned background values based on county—specific SIP information. 9—12 ------- • 24—hour second high observations were greater than annual mean measures. • 24—hour PM1O expected values were less than 24—hour TSP second highs. These checks were made only with respect toA—scenario data. Addi- tional consistency checks were undertaken after B—scenario transformations were made. The results of the B—scenario consistency checks are presented in a later subsection. The purpose of these checks is to document the quality of the data base. In some cases, though inconsistencies are present. it is notpossible totake corrective stepsbecause of a lack of supporting information as to what changes are reasonable. Standard Violation — The pre—control file contains usuable data for 1,230 counties or county parts in the United States. Since many of these counties are currently (1978) in attainment with the proposed standards and are expected to remain in attainment through 1995, they are not pertinent for the benefit analysis. 5 The relevant subset of pro—control counties (those counties out of attainment with one or more of the alternative standards) was determined by comparing state and county codes in the pro—control file with state and county codes in a data file defined in terms of a multiple standard of TSP 75 annual geometric mean and 150 24—hour second high. Since this standard represents the binding combination of the two TSP standards considered, it was thought that this comparison would yield a complete list of relevant counties. That is, counties in each of the other post—control files would be a subset of the counties in this “screen” file. Subsequent consistency checks, however, showed this to be false in a few cases. Specifically. $ This presumes that improved air quality in a neighboring county will not load to air quality improvements in the county already in attainment. Since there may be cross—county improvements, this assumption may lead to an underestimate of benefits. 9—13 ------- several counties are in attainment with respect to the TSP standards, but not in attainment with a PM1O standard. This situation is described more fully below. Of the 1,230 counties or county parts available in the pre—control file, 532 areas remain in the sample after comparison with the screen file. Each of these observations was then checked against the remaining consistency criteria. Post—Control Data Versus Pro—Control Data — With the imposition of a control strategy in a given year, emissions should be reduced. As a consequence, post—control data for that year should be less than the pre—control data. This condition is satisfied for all but four counties appearing in the various scenarios. In each case, the negative change (i.e., pre—control concentrations are less than post— control concentrations) occurs for the TSP annual geometric mean and is equal to the difference between the post—control value and the background value. This implies that the pre—control data for the geometric mean are at background and no further controls are possible. However, controls are applied because other averaging times have concentration levels not only above background but above the relevant standard. Given the definition of background, this can only be viewed as an inconsistency in the air quality data set. No corrective action was taken for these counties. Concentration Levels and Backgronnd — Given that the background data represent a lower bound for the post— control files, the post—control data should not fall below the given background levels. To ensure that this condition is not violated, the projections of pre—control and post—control values by the cost contractor were made as if the background levels were the origin. Our check of the data indicated that no county fell belo, background during the 1987 to 1995 * See subsection entitled, “Second Highs Versus Means”. 9-14 ------- time period. Rowever, the geometric mean for one county was at background. No action was thought to be necessary in this case. Note that this test was made for A—scenario data. Transformations for the B scenario do lead to several violations of the background constraint. B—scenario issues are discussed in a later subsection.* Arit]i.etic and G.o.etria Means — By definition, the arithmetic mean of a set of observations should be greater than or equal to the corresponding geometric mean. In the original design value file, however, several counties violate this condition. Since the design value file is the foundation of the other air quality files, this problem is reflected in the pre-control and post—control files. There are ten counties in the various post—control files for which the arithmetic mean is less than the geometric mean. Since no additional information is available as to why the inconsistency exists or whether one of the entries is miscoded, no alteration of the data was attempted. Second High Valuos Versus Moans — In the discussion of the standards violation consistency check, it was noted that a comparison of the pre—control and screen files revealed an unexpected relationship. In particular, there were 13 counties in viola- tion of one or more of the PM1O standards but in compliance with both of the TSP standards. Since PM1O is a fraction (0.55) of TSP, and the TSP standards appear more stringent than the PM1O standards analyzed, this suggested another consistency check. The check involved a comparison of the TSP annual arithmetic mean with the TSP observed 24—hour second high value. Naturally, one would expect a measure of central tendency to be less than an extreme value. * See subsection entitled, “Definition of B—Scenario Concentrations”. 9-15 ------- Recall that neither the pro—control file nor the post—control file contains a measure of TSP annual arithmetic mean.. Instead, this measure is derived (in 1978) by dividing the reported ,PM1O annual arithmetic mean by O.55. When the arithmetic mean of TSP is calculated in this fashion, seven of the 13 counties which violate a PM1O standard, but not a TSP standard, record TSP arithmetic means greater than the observed 24—hour second high for TSP. Thus, it appears that the PM1O data for these counties may be suspect. Since these counties were not on the screen file, they were not included in the benefits analysis. However, given the implicit relationship among the TSP measures, their omission does not seem inappropriate. It should be noted that, to our knowledge, these counties are retained in the cost analysis. Consequently, this is one source of possible inconsistency. A review of the difference between pre—control and post—control data indicates that the impact of this inconsistency on the benefits estimates should be minor. Observed Versus Expected Values — Of the remaining six counties which are in violation of a PM1O standard but in compliance with the TSP standards, three have unusually high PM1O 24—hour expected value readings in 1978. Since these values are derived by multiplying the 1978 TSP expected value in the design value file by 0.55, the implication of this is that the TSP expected value is very much larger than the observed second highest reading. For the three counties mentioned above, the expected value would be more than twice as great as the second high. While this is possible, it does seem unlikely. As with the seven counties described above, these three counties are omitted from the analysis because they were not included in the screen file. Given the data, this may not be inappropriate. However, the basis for omission is weaker than in the previous case. * In a sense, this is a circular calculation since the PM1O mean in the pre—control and post—control files was derived by multiplying the observed TSP arithmetic mean. in the design value file by 0.55. 9-16 ------- Finally, the three remaining counties which did not fail either of the two PM1O consistency checks were also omitted from the analysis because they were not on the screen file. This appears inappropriate and leads to a small inconsistency in benefit and cost comparisons. However, since benefits will be understated relative to costs, the benefit—cost comparison will be overly conservative. S”— ry of Consistency checks — Table 9—3 summarizes the results of the consistency checks described above. Although the omission of 13 counties leads to an inconsistency between the air quality data in the benefit and cost analyses, the direc- tion of inconsistency implies anunderstatement of benefits relative to costs. DEVELoI’1 gwr OF DATA FOl BEN ITS ANALISIS The discussion of the design value, pre—control, and post—control data files focused on the scope and consistency of data provided by the cost contractor. In order to develop benefit estimates, several modifications to the basic data base are required. In this subsection, five specific issues are addressed: • Geographical aggregation • Temporal interpolation • Projections of TSP annual arithmetic means • Index of exposure • Distribution of air quality changes • Definition of Scenario B concentration levels. 9-17 ------- Table 9—3 SUMMARY OF CONSISTENCY CHECKS FOR A-SCENARIO DATA State County Problem Code Disposition Code 02 0380 5 0 04 0480 3 0 04 1100 3 0 04 1680 3 0 04 2000 5 0 04 2600 3 0 06 1520 2 K 07 0565 2 K 07 1155 2 K 11 1980 2 K 18 2520 2 K 18 3280 2 K 23 3600 2 K 23 3980 2 K 24 3400 2 K 37 0360 3 0 37 1440 3 0 44 0300 2 K 45 0540 4 0 45 0860 3 0 45 2160 5 0 45 3440 4 0 45 5550 4 0 50 1580 1 K 51 0400 1 K 51 2600 1 K 51 3280 1 K Problem Codes : 1. Post—control data greater than pie—control data. 2. TSP annual arithmetic mean less than TSP annual geometric mean. 3. TSP 24—hour second high less than TSP annual arithmetic mean. 4. PM1O 24—hour expected value greater than TSP 24—hour second high. 5. Not on screen file; no other problems observed. Disposition Codes : 0 omitted. kept. 9 . 18 ------- Geosraphical Agar.patjon As noted earlier, there are 532 distinct county or subconnty observa- tions available after the pre—control and screen files are compared. Of the 532 observations, 31 represent subcounty data for 12 whole Counties. A list of these counties is shown in Table 9—4(a). Although the subcounty observations are generated to provide a more realistic picture of air quality in a given area, the division into county parts does create some problems for the benefits analysis. In particular, it is difficult to obtain certain economic/demographic information used in the analysis at less than the county level. Thus, if the air quality data are to be merged with economic/demographic data in a consistent manner, an aggregation procedure must be employed for those observations defined in terms of subcounty areas. The aggregation procedure involves selecting the county part (from the multiple parts in a given county) with the highest annual geometric mean of TSP in 1978 as representative of the entire county. This choice is consis- tent with the procedure for selection of design value data, and therefore is consistent with the data reported for whole counties. In addition to the aggregation of subcounty observations, it.was also necessary to disaggregate several observations. Specifically, six counties in Massachusetts are aggregated into two “counties” in the design value, pre—control, and post—control files. Rather than aggregate the economic/demographic data across counties, we disaggregated the air quality data. The disaggregation procedure involved assigning the same air quality levels to each of the county constituents in the two aggregate “counties” of the air quality files. The counties involved in this process are shown in Table 9—4(b). 9—19 ------- Table 9—4 StJBCOIJNTY AND AGGREGATED OBSERVATIONS IN AIR QUALITY DATA FILES a) Subcounties State Code State County Code County 03 Arizona 0620 Pima 14 Illinois 1540 Cook 14 Illinois 4100 La Salle 16 Iowa 0680 Cerro Gordo 16 Iowa 2280 Linn 16 Iowa 3120 Polk 24 Minnesota 3260 St. Louis 29 Nevada 0280 Humboldt 29 Nevada 0540 Washoe 32 New Mexico 0360 Eddy 33 New York 2000 Erie 52 Wyoming 0700 Sweetwater b) Aggregate Counties State Code State Aggregate County Code County Code County 22 22 22 22 22 22 Massachusetts Massachusetts Massachusetts Massachusetts Massachusetts Massachusetts 1798 1798 1798 1291 1291 1291 0691 0801 0815 1325 1555 2205 Franklin Hampden Hampshire Middlesex Norfolk Suffolk 9-20 ------- After these procedures were completed, there were 517 counties avail- able for the analysis.* The distribution of these counties for each of the various standards was shown earlier in Table 9—2. Temporal Interpolation Data in the pre—contro]. and post—control files are available only for five discrete years: 1978, 1985, 1987, 1989 and 1995. In order to estimate benefits over a range of years, it is necessary to make assumptions on air quality levels in intervening years. Figure 9—3 highlights the problem. The format of the figure is similar to the formats shown in Figures 9—1 and 9—2. However, in this case, the smooth segments AS and CD have been replaced by a variety of possible time paths of concentration levels. Clearly, the level of benefits (area ABDC) will be sensitive to the assumptions made with respect to concentration levels in intervening years. There are two straightforward interpolation schemes. One of the approaches is a linear interpolation. This method assumes that concentra- tions fall on a straight line drawn between the two points for which data are available. The other approach is known as geometric interpolation. This method allows concentrations to change by an annual “growth” factor which is computed from the total “growth” observed between the years for which data are available. In the subsequent sections of this study, benefits estimates are derived on the basis of geometric interpolation. The rationale behind this choice can be traced to the manner in which ambient pollution levels are derived from the growth in source emissions. Specifically, emissions are assumed to grow at an approximately constant rate g between two time * was later noted that two of the counties were in compliance with the standards examined in the benefits analysis. Thus, the total number of counties for which benefits are calculated is 515. 9-21 ------- Concentration Standard Figure 9-3. Time Paths for Air Quality Data A C E t 2 TIME 9—22 ------- periods, say 1989 and 1995. The interpolated emissions for 1990 are defined as (1+g) times the observed 1989 emissions. Similarly, emissions in 1991 are (l+g) times the calculated 1990 emissions. Given that future air quality concentrations are based on a proportional relationship with projected emissions, air quality concentrations may be expected to follow a similar pattern. Thus, the geometric interpolation method seems most consistent with assumptions in the cost study. To test the sensitivity of the benefit estimates to this assumption, benefits were also calculated via the linear interpolation method for the various studies reviewed in this report. For some studies, linear interpo- lation resulted in smaller benefits. However, the majority of studies recorded increased benefits with linear interpolation. For all studies, the absolute value of the change was never more than one percent, with the average difference being on the order of one—half of one percent. Projections of TSP Annual Arithmetic Means TSP data in the pre—control and post—control files are limited to measures of 24—hour observed second high and the annual geometric mean. Since many of the studies reviewed in subsequent sections of this report are based on TSP annual arithmetic means, this is also a relevant measure. Earlier, a procedure for determining TSP arithmetic means in 1978 was described. Here, we are concerned with identifying these arithmetic means for future years. The procedure used involves proportional relationships between 1978 values of PM1O arithmetic means and changes in the annual geometric mean of TSP across time. In particular: TSP(AAM)t (PMlo u o 78 /o.55) + (TSP(AGM) — TSP(AGM) 78 ) ( (PMlo . 73 /o.55) — TSP(AAM)b ) (TSPAGM 78 — TSP(AAM)b) 9-23 ------- where TSP(AAM)t is the year t annual arithmetic mean of TSP; PM1O(AAM) is the year t (= 1978) annual arithmetic mean of PM1O; TSP(AGM)t is the annual geometric mean of TSP in year t; and TSP(AAM)b is the TSP annual arithmetic mean background level. The formula for calculating TSP(AAM)t is consistent with the equations used by the cost contractor. Similar proportional relationships are required to determine concentrations for various averaging times after the B—scenario assumptions are imposed. These transformations are discussed in more detail below. Index of Exposure The individual studies reviewed in subsequent sections of this report each employ specific indices of air pollution exposure. If the results of these studies are to be used in the calculation of benefits, the air quality data in the pre—contro]. and post—control files must be defined consistently on a study—by—study basis. In the discussion of the sources and scope of the air quality data, the design value data for a county was defined, in general, as representa- tive of the monitor(s) with the highest readings of TSP (annual mean and 24—hour second high) in the county. Consequently, data in the pre—control and post—control files can be used directly only in those benefit studies which have air quality indices that are defined in an analogous fashion. That is, the studies that used monitors that recorded the maximum values in the geographic area, for a given averaging time and pollutant. Only then will the units of the air pollution coefficient in the benefit equation be compatible with the units of the air pollution variable itself. When the benefit studies use a measure of air quality which differs from a maximum value, adjustments to the pre—control and post—control data are required prior to the calculation of benefits. The studies reviewed in this analysis differ in several ways with respect to the definition of the air quality indices. These differences include: -24 ------- • All but two studies use a measure of TSP; the two remaining studies use British Smoke (BS) and Suspended Particulate Matter (SPM). • All studies use either an annual average concentration, an average 24—hour concentration, or a 24—hour second high concentration. • The geographic areas include Census tracts, cities, counties, and SMSAs (Standard Metropolitan Statistical Areas). • The number of monitors in each county ranges from one to a dozen or more; the monitor recording the highest concentra- tions or an average across monitors was typically used in the original study. Because of these differences in study design, a procedure was developed to ensure a meaningful integration of available air quality data with the benefits equations of the various studies. This procedure involves defining proportional relationships between maximum values for a county (design values) and calculated county averages. County averages were required because this is the form of the pollution variable in several of the studies included in the benefits analysis. Calculation of County Averages — In order to calculate a county average of air pollution, it was necessary to use site—specific information available from EPA’s Storage and Retrieval of Aerometric Data (SAROAD) system. Recall that this is the data base used, in part, to construct the design value file. However because of certain adjustments to the data, more counties are available in the design valu, file than in SAROAD. In fact, of the 515 counties considered in this analysis, only 421 appear in SAROAD for 1978. A distribution of the number of monitors per county is shown in Table 9—5. County averages in 1978 for TSP annual arithmetic mean, annual geometric mean, and 24—hour observed second high were calculated for each county with two or more sites. It was then assumed that the ratio between the county averages and observed design values in 1978 would be maintained 9-25 ------- Table 9—5 DISTRIBUTION OF SAROAD MONITORS BY COUNTY No. of Monitors in County No. of Counties 0 94 1 157 2 65 3 36 4 32 5—10 93 11—20 32 21—30 3 > 30 3 Total 515 for all future years. Since the pre—control and post—control data are generated from design values, the projected equivalent county average value in year t can be calculated as: CAt DVt (CA. 78 1DV 78 ) (9.2) where t is the county average in year t and DVt is the design value (pre— control or post—control data) in year t. This calculation was performed for each of the TSP measures mentioned above, for each county with two or more monitors.’ * Since the design value reading typically represents a maximum value, the term (CA 78 /DV 78 ) was constrained to be less than or equal to one. The constraint was binding in only 16 counties. 9-26 ------- For those counties in SAROAD with no or only one monitor, the factor (CA 78 /DV 78 ) in Equation (9.2) was calculated as the average of this factor across all counties which had two or more monitors. These average values for each of the three pollutant measures are shown in Table 9—6. Numbers in parenthesis represent standard deviations. Given the calculations implied by Equation (9.2). new pre—control and post—control data are defined which can be interpreted in terms of county averages rather than county maximums. These data can then be used in the calculation of benefits for those studies utilizing average pollution readings in the benefit equations. Naturally, the original (design value) pre—control and post—control files are still appropriate for those studies using maximum values. A study—by—study summary of the index used is given in Table 9—7. Sensitivity check — Of the 14 separate studies reviewed in this report, 13 use population or household data in the benefits calculation phase. Because of this, a more conservative approach for estimating county averages would be to use only population—based monitors from the SAROAD data system. Ostensibly, this would yield county averages that are more representative of county Table 9—6 COUNT! AVERAGE CORRECTION FACTORS Pollutant Measure (CA 78 /DV 73 ) A nu*1 Geometric Mean 0.74 (0.16) Annual Arithmetic Mean 0.74 (0.16) 24—Hour Observed Second High 0.70 (0.18) 9-27 ------- Table 9—7 REVIEW OF POLLUTION MONITORS USED BY STUDIES Study Geographic Area Monitor(s) Used 4n c ig inal Study Measure Used for Benefits Malysis - Design Value Avg. Over Monitor in All Monitors County in County Mazi.indar et al. — — County of London Avg. of 7 representative monitors in county x Samet et a!. Area served by hospital in Steubenville, OH Monitor near hospital (not highest value monitor in area) x Saric et al. P ighborhoods in two cities Both indoor and outdoor monitors x Ferris et al. Berlin, NH Avg. of 3 monitors in Berlin x Lave & Seskin SAa Center city monitor in each SMSA (NASN data) X Crocker et a!. Counties County monitor having most complete data between 1967—75 X Ostro Cities Avg. over all 01 monitors x Property Value Studies Census tracts in a city Interpolation of census tract exposure from isopleth or dis- persion modeling of city monitors x Bosen SMSAs Monitor representing worst incidence within SMSA x K. ith SNSA8 Avg. over all monitors within SMSA X Ctunmings et a!. 4 Nzonesn in Phila. SMSA Avg. of monitors in each zone x Watson & Jakach 4 zonesu in Phila. SMSA Avg. of monitors in each zone x Mathtech (Household) SMSAs Highest value monitor in SMSA X tlathtech (Manufacturing) Counties Avg. over all monitors in county x ------- population exposures. Of the 2,983 TSP site monitors in SAROAD in 1978, 2,180 are designated as population surveillance monitors. When county averages are computed from these monitors only, the factor (CA 78 /DV 78 ) in Equation (9.2) is 0.75, 0.75, and 0.69 for the geometric, arithmetic, and second high measures, respectively. Comparing these values with those reported in Table 9—6, there are only minor differences. Thus, the impact on benefits will be minimal. Consequently, benefits estimates for those studies using county average air quality data are reported only for the all—monitor case. Distribution of Air Qma1it kaues The pre— and post—control air quality data identify the air quality changes predicted to occur at a single location in the county. This is the location of the design value monitor. The assumption made in the cost and air quality modeling effort is that all sources contribute proportionally to the concbntrations recorded at the design value monitor. The propor- tionality factor is based on source emissions rates and takes into account differences in stack height. Thus, a given reduction in emissions across all sources is assumed to reduce concentrations at the design value monitor by a known factor. Although certain stack parameters do affect this factor, distance of the source from the design value monitor is not explicitly considered. Consistency of the cost and benefit analyses would require beüeiits estimates to be calculated across all parts of the county, since control of sources is not limited to a specific group of Sources near those monitors that record concentrations higher than the alternative standards. However, an alternative approach would be to assume that Stat. Implementation Plans (SIPs) may in practice require control first on those sources suspected of contributing most to any observed air quality violations. In this case, benefits may arise primarily in those areas near controlled sources. If the population in the county is located in areas away from the controlled 9—29 ------- sources, then their exposure may not change appreciably and few benefits would be generated. 5 The potential for overestimating benefits becomes clearer when the different air quality indices used in the benefits analysis are considered. These are the average of readings at air quality monitors in the county and the reading at the highest valued monitor in the county. Figures 9—4 and 9—5 show how the changes at the design value monitor may affect the exposure of the county population. In Figure 9—4, the pre—control distribution of air quality is labelled as A. For expository purposes, it is assumed that the county is one— dimensional and that population is distributed uniformly. If the air quality index is based on the highest valued monitor, then a reduction in air pollution of I Lg/m 3 at the design value monitor results in an assumed reduction in PM concentrations of I ig/m 3 throughout the county (subject to a possible background constraint). This post—control distribution is labelled as curve B. Benefits would be estimated as the health and welfare effects associated with the shaded area. In Figure 9—5, the pre—control curve is the same (again labelled as A). flowever, it is now assumed that PM concentrations in the county are reduced by the same percentage as concentrations at the design value monitor. Since the average value of PM concentrations in the county is less than at the design value monitor, the shaded area in Figure 9—5 will be less than that in Figure 9—4. Figure 9—5 is probably more representative of reality in view of the dispersion properties of PM and the likelihood that SIPs may focus on specific sources contributing to violations. Thus, Figure 9—5 is used in the benefit analysis. * This is a simplification since population data reflect only where people reside. Since some individuals will work in the vicinity of the controlled sources, they would realize an exposure reduction during the work day. Data available for this study did not permit this distinction. 9-30 ------- pg/rn 3 Standard Figure 9—4. Equal Magnitude Improvement Pre-Contro]. Post-Control A B nitor 2 Design D itor bnitor 3 9—31 ------- 3 igAn Figure 9—5. Equal Percent Improvement Standard Post-Control A B nit r 2 Design Monitor nitor 3 9-32 ------- Depending on actual dispersion properties and the location of controlled sources, it is possible that air quality improvements may be even more narrowly focused than suggested by Figure 9—5. For example, it is possible that air quality may change as shown in Figure 96. In Figure 9—6, curve A again identifies the pre—control distribution. In the post— control case, the distribution of the improvement is limited to the area near the monitor out of compliance. Although the third case is not consistent with the assumptions under- lying the cost and air quality modeling effort, it is possible that air quality improvements may follow this pattern in reality. Thus, it would be informative to identify the extent to which benefits are sensitive to assumptions about the changes in the air quality distribution. In order to provide such a check, it is necessary to have detailed information on air quality dispersion characteristics. An area study that compares the benefit estimates obtained using dispersion modeling results with the estimates calculated assuming air quality improvements proportiona l, to the change at the design value monitor is described in Section 10. Because of time and resource constraints, it is not practical to develop dispersion modeling results in the nationwide study. In the absence of dispersion modeling, the assumption of equal proportional air quality changes in the benefit analysis (Figure 9—5) is the best that can be done with the avail- able information. Note that the lack of dispersion modeling also hinders the use of studies which measure air quality effects at the city or census tract level. For moat appropriate use of this. studios, it would be desirable to have dispersion modeling results for suboounty areas. Alternatively, for studies based on county or SMSA data, using county air quality indices provides a relatively good match with the units of the variables used in the original studies. 9-33 ------- pg/rn 3 Figure 9-6. Localized Improvement Post-c trol nitor 2 Design nitor nitor 3 9-34 ------- Definition of B—Scenario Concentrations The final adjustment made to the pro—control and post—control data provided by the cost contractor involve.s the definition of B—scenario concentrations. As defined earlier in this Section, A—scenario data may leave some counties in nonattainment after imposition of the standard. The data in the pro—control and post—control files are of this type. On the other hand, the B scenario was defined to be the situation where every county would be in attainment with the relevant standard in each year considered. The relationship between A and B scenario data is demonstrated in Figure 9—7. In this subsection, the procedures used to derive the B— scenario concentrations are discussed. It is important to note that the transformations used are consistent with those used by the cost contractor in a corresponding analysis of B—scenario control costs. B-S..nario Calculations — There are two major issues that arise in the calculation of the B— scenario values. These are: • Between implementation of the standard and 1995, which year in the post—control file is furthest from the relevant standard (i.e., what is the controlling year)? • In the case where a standard is stated in terms of two averaging times, which is the most restrictive (binding) averaging time? The rationale behind the B—scenario calculations is that if the data that an, furthest from attainment (i.e., a liven year and averaging time) can be brought into attainment, then all other years and averaging times will also be in attainment. Note that this is just the definition of the B scenario. The procedure used to determine the controlling year and averaging time (standard) is described in detail in the technical appendices to the 9-35 ------- Concentrations Figure 9—7. Pre-Control Data A Scenario Comparison of A and B Scenarios A Standard E Data B Scenario t-Control Data t 2 Time 9—36 ------- cost analysis (1). Only a brief discussion of the major equations will be given here. The first step in defining B—scenario concentrations is to calculate a set of factors of the form: Zjj (x 1 — sj)/(x? — X j) (9.3) where is the A—scenario post—control data in year i for standard j; is the level of standard 3; and X’ j is the pre—control data in year i for standard 3. Z 1 is th. ratio of the additional improvement required to obtain the standard to the improvement obtained with available controls. This proportionality factor is used to alter the A—scenario data so that attainment is guaranteed in all years. The controlling year and standard are determined by computing the maximum of all Zjj. That is, ZB — max (0, ij (9.4) i,j Note that if Zjj is negative for all i, then the standard is attained in all years and B—scenario data are equivalent to the A—scenario data. Given a positive 5 B’ adjustments to the QQntrolljg standard are made by assuming a proportionate shift relative to the A—scenario data and the controlling stAndard level; — (z — x 3 )(z 8 Zjj) (95) where is the B—scenario concentration in year i for controlling standard 3. In the case where 5 B equals Zjji x will be set equal to the standard $j. In all other oases, j will be less than the standard. 9-37 ------- Figure 9—8 portrays the adjustments implied by Equations (9.3) to (9.5). For ease of exposition, assume that there are only two distinct time periods, t 1 and t 2 . Furthermore, let us assume the following numerical values (in tg/m 3 ): x j 100; ‘ 2j 130; x 1 = 65; = 90; and s. 55. Then, by Equation (9.3), 0.286 and 0.875. According to Equation (9.4), this implies that LB e nals z 2 (= 0.875). Finally, Equation (9.5) is used to find the B—scenario concentrations. For i equal to t 2 , ‘ 2j is just equal to the standard s . In year t 1 , X jj is equal to 34.4 pg/ms. These concentrations are indicated as the lower line s egment in Figure 9—8. Knowledge of for the controlling standard permits a straightf or— ward adjustment of the concentration levels in each year of the other standards (in a multiple standard case). Specifically, = — A(x 1 — X j) (9.6) where A = 18,k — xb,k)/(x 8,j — x 1 , ) (9.7) Here, ‘8,k is the 1978 (A scenario) concentration for standard (averaging time) k, and xbk is the background concentration level for averaging time k. Consistency 1secks — Two consistency checks are made as the B—scenario data are con- structed. • In the case of multiple standards, a check is made to determine if one standard controls in one year while the second standard controls in another year. • A comparison is made between the calculated B—scenario values and the background values. 9-38 ------- Concentrations 0 Si_ S — .t]. J I I t 1 t 2 Time Figure 9—8. Development of B—Scenario Data 9-39 ------- The first check is made to determine if zB (the maximum of the Z 13 )’ is sensitive to averaging time. Given that the data are related via propor- tionality factors, it was believed a uriori that a controlling standard in one year would control in all years. This is true for all but three counties in the data base. With ZB computed from all years and averaging times, no correction is required. The most binding standard is always selected for control to the standard level so that complete attainment is ensured. Violation of the background constraint is a relatively serious problem in the construction of the B—scenario data. This is especially true for the stringent TSP standards. Overall, calculation of and Xj from Equations (9.6) and (9.7) led to 12 counties having predicted concentra- tions lower than background in at least one year or averaging time. In fact, because of the linear shift implicit in the equations, seven counties had predicted concentration levels below zero. For all of these counties, post—control B—scenario data were set equal to the appropriate background concentration. Although the concentration levels predicted from the B—scenario transformations are based on very general proportionality assumptions, these data are used as the principal air quality indicators in the benefits analysis. As noted earlier, equivalent assumptions are used in the cost analysis. See Reference (1) for further caveats related to the use of B— scenario data. Table 9—8 reports the counties that failed the two consistency checks for the B—scenario data. Corrections were made as described above. None of these counties are omitted from the benefits analysis. I IT CAL JLATIONS Given the air quality data base described in the previous subsections, we next review the manner in which these data are used in the benefits analysis. Three issues are addressed: 9-40 ------- Table 9—8 PROBLEM COUNTIES IN B-SCENARIO DATA BASE State County Problem Code 01 2400 2 03 0300 2 05 5440 1 10 1800 2 13 0080 2 13 0820 2 13 1420 2 14 1540 1 23. 1340 2 29 0080 2 29 0280 2 32 0640 1 37 2620 2 45 1710 2 52 0700 2 Problem Code : 1. Different standards control in different years. 2. B—scenario concentrations predicted to be less than background. 9-41 ------- • Alternative standards to be analyzed. • Parameters used in reporting the stream of benefits. • The manner in. which benefit estimates are tabulated. Tynos of Standards Table 9—9 shows the alternative standards analyzed in this study. This table is like Table 9—1, except an additional column is added. This column identifies a particular standard as being primary or secondary. In the Clean Air Act, primary standards are set to protect human health with an adequate margin of safety; secondary standards are set to protect the public welfare. 5 Note, in this report, that the reported benefits for the secondary standards represent the benefits that would accrue with a change from current levels to the secondary standard. In order to determine the incremental benefits of the secondary standard relative to a particular primary standard, it is necessary to subtract the primary standard benefits from the reported secondary standard benefits. In this study, the PM1O secondary standard of 55 )ig/m 3 is associated with the PM1O 70/250 primary standard. In addition, the two TSP standards of (75/260) and 150 represent the current primary and secondary standards, respectively. The choice of standards to be analyzed was made by EPA. The rationale for selection of the alternative levels is based on information contained in the Criteria Document (4) and the Staff Paper for PM (2). Paraa.ters Used I a tks Ban.fits Analysis The benefit models described in subsequent sections of this report each calculate benefits in terms of functions of changes in air quality. * The distinction is somewhat unfortunate in that some health effects may be realized at levels below the alternative primary standards, and welfare benefits are generated in the course of attaining alternative primary standards. 9-42 ------- Table 9—9 AMBIENT AIR QUALITY STANDARDS Pollutant Loyal Type Averaging Time Implementation Date Maintenance Date P110 70 250 Primary Primary Annual Arithmetic Mean 24—Hour Expected Value 1989 1989 1995 1995 P110 55 Secondary Annual Arithmetic Mean 1989 1995 P110 55 250 Primary Primary Annual Arithmetic Mean 24—Hour Expected Value 1989 1989 1995 1995 P110 55 150 Primary Primary Annual Arithmetic Mean 24—flour Expected Value 1989 1989 1995 1995 TSP 75 260 Primary Primary Annual Geometric Mean 24—Hour Second High 1987 1987 1995 1995 TSP 150 Secondary 24—Hour Second High 1987 1995 ------- In general, the relationship between benefits and air quality can be expressed as: B = F(AQ 1 ) — F(AQ 0 ) (9.8) where B is the dollar benefit in year i for a particular benefit category; AQ is the air quality in year i (post—control file data); AQ 0 is the pre— control air quality; and F is a function. In this report, the functional form F is assumed to remain constant across years. Diagrammatically, the yearly benefits represented by Equation (9.8) can be shown as in Figure 9—9. This figure is similar to those shown earlier. In Figure 9—9, the air quality associated with the post—control data is pictured as in attainment for all years between t 1 and t 2 , and background levels are assumed tobe binding. This latter feature is for illustrative purposes only. In fact, in most counties, the post—control data for the B scenario were above background. Equation (9.8) indicates that benefits are calculated for each year in the time horizon beingconsidered. This is 1989 to 1995 for PM1O and 1987 to 1995 for TSP. As a consequence, there is a series of anticipated yearly bencf its. In order to provide a consistent summary of the total value of benefits, it is convenient to summarize the stream of benefits in terms of a discounted present value. Disconnt.d Present Vain. — The formula for calculating the discounted present value of benefits is expressed as: 1 N DPV £ B /(1 + r)i (9.9) (1 + r) i1 where DPV is the discounted present value; r is the prescribed rate of discount, and B represents the benefits in year i. Note that all benefit 9 44 ------- Concentrations (i g/ni 3 ) Standard Background Figure 9-9. Benefits of Ambient Air Quality Improvements Pre—Control Data Due to Standard Scenario B Post-Control Data t 2 Time 9-45 ------- estimates are reported in 1980 dollars, and benefits are assumed to accrue at the end of each year the standard is in effect. The parameters for the indices and N, and the assumed value for r are provided in Table 9—10. Given these parameter values, the DPV is interpreted as a discounted present value in 1982, in 1980 dollars, ‘with a 10 percent discount rate. The timeframe assumptions were provided by EPAe The choiCe of discount rate is guided by 0MB requirements.’ Annualized Benefits — Another way to report the stream of benefit estimates is to calculate the constant (levelIzed) annual benefits during the period of the standard. This number is derived by converting the 1982 DPV value forward to a DPV in the implementation year (1987 or 1989) and then multiplying the 1987 or 1989 DPV by an annualizing factor. Because of the different timeframes for PM1O and TSP, the conversion and annualizing factors differ for the two pollutants. For PM1O, the combined factor is: Table 9—10 BENEFIT PARAMETER VALUES Parameter Standard (Pollutant) —— ———— . —— PM1O TSP r N 0.10 7 7 0.10 5 9 * The choice of 10 percent may be high for some of the benefit categories (economic sectors) analyzed in this report. See Reference (5) for a clear discussion of the rationale and problems associated with choosing a discount rate. 9 -46 ------- (1 + r) 7 ] (9.10) For TSP, the combined factor is: (1 + r) 5 r (9.11) 1 — 1/(1+r) Benefit estimates in terms of the DPV and annualized benefits are reported in subsequent sections of this study. Tabm1 tjon of B.nef its Benefits are calculated on a county—by—county basis. This is consis- tent with the level of detail available in the air quality data. For reporting purposes, however, benefits are aggregated to the level of the ten EPA Federal administrative regions. While aggregation hides some detail, this breakdown provides an indication of the distribution of benefits across the country: Table 9—li. shows the assignment of states to the ten EPA regions. S11 ARY OP A QUAIX1 DATL This section has reviewed the source, scope, and use of the sir quality data for benefits analysis. One of the major concerns involved consistency with the companion cost analysis. For the most part, the data used in the two studies are equivalent. For the. small number of counties excluded from the benefits analysis, but retained in the cost analysis, the impact on benefit—cost comparisons should be minimal. In fact, the exclu- sions tend to make the benefits estimates smaller relative to costs. r 9-47 ------- TabiC 9—11 EPA FEDERAL ADMINISTRATIVE REGIONS BY STATE State Code State EPA Region State Code State EPA Region 01 Alabama IV 27 Montana VIII 02 Alaska 1 28 Nebraska VII 03 Arizona I X 29 Nevada IX 04 Arkansas VI 30 New Hampshire I 05 California II 31 New Jersey II 06 Colorado VIII 32 New Mexico VI 07 Connecticut I 33 New York II 08 Delaware III 34 North Carolina IV 09 Washington, DC III 35 North Dakota VIII 10 Florida IV 36 Ohio V 11 Georgia IV 37 Oklahoma VI 12 Hawaii IX 38 Oregon X 13 Idaho I 39 Pennsylvania III 14 Illinois V 40 Rhode Island I 15 Indiana V 41 South Carolina IV 16 Iowa VII 42 South Dakota VIII 17 Kansas VII 43 Tennessee IV 18 Kentucky IV 44 Texas VI 19 Louisiana VI 45 Utah VIII 20 Maine I 46 Vermont I 21 Maryland III 47 Virginia III 22 Massachusetts I 48 Washington I 23 Michigan V 49 West Virginia III 24 Minnesota V 50 Wisconsin V 25 Mississippi IV 51 Wyoming VIII 26 Missouri VII ------- 1. Argonne National Laboratory. Costs and Air Quality Impacts of Alternative National Ambient Air Quality Standards for Particulate Matter, Draft Technical Support Document, June 1982. 2. U.S. Environmental Protection Agency. Review of the National Ambient Air Quality Standards for Particulate Matter. Draft Staff Paper, October 1981. 3. U.S. Environmental Protection Agency. Review of the Relationships of IP1O, IP1S, and TSP. Memo to N. Thomas of EPA, July 1981. 4. U.S. Environmental Protection Agency. Air Quality Criteria for Particulate Matter and Sulfur Oxides — Volume IV. Review Draft No. 2, December 1980. 5. Provenzano, G. j al. Methods for Evaluating the Equity Impacts of Proposed Environmental Regulations. Draft background report of the equity impacts analysis team of the Regulatory Impacts Analysis workgroup, September 1981. 9-49 ------- SECTION 10 SELECTED METHODOLOGICAL ISSUES ------- SEcI’ION 10 S RCI E ODOLOGIChL ISSUES This section discusses several methodological, issues which arise in the benefit analysis. The issues are sufficiently important that they require fairly detailed discussion. They also affect several parts of the benefit analysis and are thus most conveniently considered in one central place rather than at each point where they arise. The first issue concerns how to combine the benefit estimates from individual health and welfare studies and categories. Some combining or aggregation is required in order to obtain a total benefit estimate for comparison with costs. Aggregation is difficult because it cannot be accomplished by simply summing all categories. Simple summation would run the risk of double counting and would not give adequate consideration to study quality. Aggregation procedures are discussed in the first subs e c t ion. The second issue is the question of applicable PM concentration ranges for benefit calculations. Three possible ranges are available. These include: 1) the range of air quality improvements experienced under alternative st.ndards 2) the rang. of air quality included in the data samples of the original health and welfare studies; and 3) the range of concentrat4ons within which the Clean Air Scientific Advisory Committee (CASAC) has concluded that quantitative evidence of health effects exists. The lower bounds of the first two ranges are generally below the lower bound identified by CASAC. The issues this raises for the benefit analysis are discussed in the second subsection. 10-1 ------- The third issue concerns air quality data. The benefit analysis relies on air quality data •developed as part of the cost analysis. Use of common data helps insure that the cost and benefit analyses are as c-onsistent with one another as possible. However, the benefit analysis requires more spatial detail than is available from the cost analysis. Approximations are made to provide this detail. The uncertainties introduced into the benefit estimates by these approximations are the subject of a detailed sensitivity analysis discussed in the third subsection. The sensitivity analysis is based on a case study of air quality dispersion modeling focused on the Chicago area. PZO URE$ FOt CALCULATION OF AGGREGATE B EFITS Introduct ioa Sections 2 through 9 of the report have developed methods for estimating incremental benefits in a variety of health and welfare categories. Those sections do not discuss how to combine the estimates for the individual benefit categories in order to develop an estimate of aggregate incremental benefits. It is the purpose of this section to describe how aggregation is done in the study. Aggregate numbers are required in order for any benefit—cost analysis to proceed. Aggregation requires combining benefit estimates from individual health and welfare effects studies and categories. However, aggregation cannot be accomplished by simply summing all of the categories because this might result in double counting and other problems. One of the most difficult problems is that the available health and welfare effects studies vary widely in scope and quality. Thus, aggregation forces judgments to be made about study scope and quality. Some of the issues which must be considered in choosing an aggregation procedure are: • Does the procedure avoid double counting of benefits? 10-a ------- • Does the procedure provide complete coverage of benefits? • Does the procedure give due consideration to the relative strength of the evidence across different benefit categories? The double counting problem is perhaps best illustrated by the property value studies. It is generally believed that properly value studies measure a combination of effects which may include health, soiling, and aesthetic benefits. However, the proportion of benefits in each category is not generally known. Nor is the relationship to benefits estimated from specific health and soiling studies known. Consequently, it is generally not appropriate to add property value estimates together with estimates from health and soiling studies in developing aggregate benefit numbers. The problem of benefits coverage ii perhaps best illustrated by various health studies from the medical epidemiology literature used in the benefit analysis. Morbidity studies in this category generally measure only the respiratory disease effects of particulate matter. They do not capture other potential health impacts of particulate matter such as possible cardiovascular effects or other forms of acute or chronic illness that might be pollution related. As a consequence, if one relies solely on the medical epidemiology studies for estimates of morbidity—related benefits, benefits that might arise from reductions in non—respiratory diseases would be omitted. It is unfortunately the case that no one aggregation procedure best satisfies all three criteria above. As a consequence, this analysis follows a specific strategy. The first element of the strategy is to use the individual health and welfare effects studies to develop the basic benefit estimates. Second, the property value and hedonic wage studies are used as cross—checks on the basic benefit estimates. Recall that the property value and wage studies measure aggregate effects. Third, the analysis develops several alternative aggregation procedures. These • See Section 5 for additional discussion of property value studies. 10-3 ------- alternatives are described in detail in the paragraphs that follow, Each alternative gives differing weight to the three criteria above. In particular, each reflects different possibilities of double counting versus under—counting. Each reflects different assessments of study quality. Criteria for Assossina Stu4 Qunlitv Study quality is one of the most important issues in the process of aggregation. Therefore, this section sets forth the criteria that are used in assessing the quality of the studies available for developing aggregate benefit estimates. The criteria are similar to those used by the Clean Air Scientific Advisory Committee (CASAC) in evaluating studies for use in the Criteria Document (1). ifowever, there are differences in the two sets of criteria since CASAC was not specifically concerned with ns of studies in a benefit analysis. Study Sample — The first criterion considered is: Did the study use well defined and appropriately selected study populations? At least two issues are important in this regard. First, did the study use microdata on receptors as opposed to aggregate data for cities, counties, or SMSAS. Microdata is preferred because it avoids the problems that arise when using grouped data. One important problem is loss of efficiency in the estimation of statistical parameters. Also, with macrodata, proper statistical controls are loss directly applied. The second issue in terms of sample selection is whether the study sample can be considered ropresentativó of the U.S population. A repre- sentative sample reflects the range of differences in the U.S. in personal characteristics (age, sex, race, occupation, etc.), environmental charac- teristics (climate, methods of home heating, etc.), and other influences on health status. A representative sample provides greater assurance that the estimates of effects developed can be extrapolated to the U.S. population for developing national benefit estimates. A loss representative sample, 10-4 ------- on the other hand, may provide a less valid basis for calculation of national benefits. Study Controls — The second criterion used in assessing study quality is the extent to which the study controlled for potentially confounding or covarying factors. One issue here is the extent to which variables were included in the analysis to control for covarying effects. Example variables include age, sex, race, diet, smoking habits, occupational exposure, and method of home heating. The second issue here is whether the study properly controlled for behavioral adjustments that individuals may make to compensate for pollution effects. As an example, some individuals may stay indoors on particularly polluted days in order to avoid health—related problems. If this is done, and not considered during study design, then the measurement of health effects with respect to levels of pollution could be biased downwards. *easure snt of Pollution Exposure — The third criterion considered in the study assessment is the quality of the measurement of pollution exposure, One attribute considered in this regard is whether the study controls for both particulate matter and sulfur oxides. If only one pollutant was considered or used in the analysis, the potential exists for the estimated effect of that pollutant to be biased. A second characteristic considered is the quality of th. monitoring of pollutiou. This might take two forms. In the case of a large geographic area, a dense monitoring network may be required to provide an accurate estimate of pollution exposure. Or in the case of a relatively small geographic area, a small number of monitors may be adequate for measurement purposes. 10-5 ------- Statistical Techniques — A fourth criterion considered in assessing study quality is the extent to which studies used appropriate and properly performed statistical techniques. This issue includes consideration of adjustments that may have been performed on the raw data, methods used in analyzing the data, and conclusions drawn from the analyses performed. Xessureaent of Effects — The types of effects measured in individual health and welfare studies vary widely. Some are ideally suited for benefit analysis, while others are difficult to apply in a practical way. A good example of the latter is the common practice in medical epidemiology studies to measure health endpoints such as reductions in lung function. Although these measures may be useful for medical diagnostic purposes, they have little practical application for estimating economic benefitg. In some respects then, this criterion helps to separate studies which may be useful for Criteria Document purposes from those which may be useful for benefit analysis purposes. Consist.*cy YIth Otker Research — The final criterion considered is whether the study obtained results which are consistent with other research in the ares. The degree of consistency enables judgments to be made about the degree of confidence that one can have in using the study for benefit calculations. Luolipation of Stidy Evaliatiom Criteria Table 10—1 summarizes how the study evaluation criteria were applied to the individual studies selected for estimating benefits.* The table is * Complete bibliographic citations for the studies can be found in Section 1. 10-6 ------- APPLICATION OF Table 10—1 STUDY EVALUATION CRITERIA Micro Data Repres. S sap 1. Covar. Controls Behavioral Controls Both Pill SO 1 Network or Saall Are a Approp. Stat. Approp. Other Effects Research HEALTh EFFECTS Crocker at *1. Ferris at J. •l• Lave & Seskin/ Lipfe rt Mazuadar at a1 . *+ Ostro Saaet at al + Sane ii. 1 1 0 0 1 0 1 <1 0 1 1 1 0 0 <1 Long it. (1 Longit. <1 Long it. <1 <1 0 0 0 0 0 0 <1 0 1 1 1 0 0 0 1 1 1 1 <1 1 <1 NM 1 0 1 <1 NH 1 1 1 1 1 1 1 (1 1 1 1 <1 1 1 WELFARE EFFECTS Cuaings j. 1 1 0 0 0 1 (1 1 <1 Matbtech Household 0 1 1 1 1 ( 1 1 1 1 Mathtechilanufaot. 0 1 <1 1 1 <1 1 1 <1 Watson&Jaksoh 1 1 0 1 0 1 NM 1 <1 * CASAC “quantitative” study. + EPA staff Paper “quantitative” study. NM Not meaningful. ------- organized as follows. Across the top of the table are the various study evaluation criteria described previously. The rows in. the table correspond to the individual studies available for assessing health and welfare effects. Note that studies have been grouped into separate health and welfare effects categories, and. within those categories, listed alphabeti- cally. Entries in the table are primarily l’s and 0’s. A “1” is assigned if the study met the listed criterion, and a “0” is assigned if the study Was deficient in that criterion. In some cases, a clear judgment was not possible. These cases are generally shown as < 1, not meaningful (NM), or by some other designation. The table also identifies studies which CASAC (2) has indicated provide quantitative evidence of pollution—related effects. These are identified in. the table by an asterisk. They include the studies by Ferris et ii. and Mazumdar j gj. (CASAC also identified other studies providing quantitative evidence, but they could not be adapted for benefit calcula- tions.) Also shown. in the table are the studies judged in the EPA Staff Paper (3) to provide quantitative evidence of concentration—response rela- tionships. These include the Ferris g.. gj.. study, the Mazumdar study, and the study by Sam.t jj. The latter studies are identified by a plus symbol in the table. To highlight the differences between study assessments made for purposes of standard setting and criteria document development, and studies used for benefit analysis, it is useful to compare the quality assessments made for the CASAC “quantitative” studies and some of the other studies available for benefit analysis. This is perhaps best illustrated by comparing the two C.&SAC—approved studies by Ferris j gj,. and Mazunidar j ii. with the study by Ostro. The Ostro study appeared in preliminary form prior to CASAC closure of the Criteria Document, but has not been formally reviewed for purposes of inclusion in the Criteria Document. In the first column of the table, which is concerned with whether the study used microdata, note that the Ferris jtgJ,. study is assigned a “1”. The Mazumdar j jj.. study, which used aggregate data for the County of 10-8 ------- London is assigned a “0”. The Ostro study, which used microdata from the National Health Interview Survey, is assigned a “1”. The second column is concerned with whether the study used a represen- tative sample of the U.S. population. The studies by Ferris j jj. used a small sample of people from a small town, Berlin, New Hampshire. This sample may not be representative in terms of the desired characteristics listed previously. Factors behind this judgment include the small size of the sample, the fact that Berlin is a small rural town, and the fact that the major pollution source in the area is a wood pulp mill, which is atypical of most other areas. This study was therefore assigned a “0”. The Mazumdar j,. study, as mentioned previously, worked with data for the County of London. As a large city and sample size, the Mazuisdar study sample is likely to be more representative of the U.S. conditions. However, as a European city, where methods of home heating are often quite different, it is not completely representative. That is, different methods of home heating can generate indoor and outdoor pollutants that may differ from those experienced by U.S. populations. On balance, the Mazumdar j J.. study was assigned a rating of”< 1”. Finally, the Ostro study makes use of data for 90 mediumsized cities (populations of 100,000 to 600,000). The Ostro study is assigned a “1” on this criteria. The third column is concerned with the extent to which the studies controlled for oovarying factors. The Ferris it ii.. study was a longi- tudinal study which looked at pollution—related respiratory disease at three different points in time (1961, 1967 and 1973) in Berlin, New Hampshire. As a longitudinal study, the Ferris !. gJ.. analysis probably benefits from a high degree of control for such factors as sample popula- tion, cultural differeaces, regional differences, occupational exposure, etc. However, longitudinal studies also suffer from a variety of other problems. For example, characteristics of the sample may change over time due to deaths, aging, migration, changes in life style, and other factors. 10-9 ------- Some of these factors are not easily controlledt In view of this mix of advantages and disadvantages, the Ferris ji,. and Mazumdar studies, which are both longitudinal, are simply denoted as such rather than assigned a specific rating. The Ostro study, in contrast, is cross— sectional. It makes use of data on a large number of cities and attempts to control for covariation by explicit inclusion of other variables. For example, the Ostro study controls for age, race, sex, marital status, general occupational class, family income, population density, weather variables, and smoking habits. The Ostro study does not control for diet or method of home heating and has limited control for exposure to occupa— tiona.l hazards. Kowever, the other controls are sufficiently thorough that on balance the study is assigned a “< 1,’. The fourth column is concerned with control for behavioral adjustments. All three studies are assigned a “0” on this criteria. None of them adequately account for possible behavioral adjustments that individuals may mak. in order to reduce pollution—related health effects. All three studies may thus measure pollution—related effects with a downard bias. The fifth column is concerned with the degree of control for both particulate matter and sulfur oxides. The Ferris j gJ.. study is assigned a “0” on this criteria. The Ferris study considered conditions in Berlin, New Bampshire, at three points in time (1961, 1967 and 1973). Between 1961 and 1967, particulate matter, sulfation, and measured health effects declined. Between 1967 and 1973, particulate matter declined, sulfation increased (yet s ill remained at relatively low levels), and no health effect differences were observed. Because both pollutants were changing during both periods, while health effects changed only during the * For example, in the Ferris j study, 16 percent of the men and 7 percent of the women in the 1967 sample were dead by 1973. Those who died were among those in the 1967 sample with more symptoms and. poorer pulmonary function (4). Comparisons of health status between 1967 and 1973 using the surviving population were thus based on a subsample that was known to be biased in favor of healthier individuals. No correction for this bias was possible. 10-10 ------- first period, at least four data points are required to identify the separate effects of the two pollutants and test for the apparent existence of a no—effects threshold.* Only the three data points are available. The Mazumdar , j. study in contrast used daily data on particulate matter and sulfur dioxide for 14 London winters. Both pollutants were thus measured and in considerable detail, so the study is assigned a “1”. The Ostro study also receives a “1”. It measured both particulate matter (in the form of TSP) and sulfates (SO 4 ) in all 90 cities in the analysis. The next column is concerned with the quality of the pollution measurement. In the case of the Ferris j Al.. study, three monitors were used to represent air pollution conditions in the Berlin, New Hampshire. area. This approach is assigned a “1”. The Mazuadar et j. study used seven monitors in the County of London. This study is also assigned a “1”. The Ostro study used an average of the population—oriented monitors in each of the study cities. The number of monitors in each city varied. This study is also assigned a “1”. The next column is concerned with the application and appropriateness of the statistical techniques used in the studies. In the.csse of the Ferris Al. study, no meaningful assessment can be made since with the limited data available, no concentration—response functions were estimated. The Mazumdar et si. study in contrast used extensive statistical techniques. Unfortunately, one of the early adjustments made to the data was to remove weather—related variation. This is an inappropriate proce- dure because in removing the weather variation in the mortality data, the Mazumdar j Al . study also removes all pollution—related effects that are correlated with weather. Thus, the study is likely to provide a downward— biased estimate of pollution—related effects. The study is thus assigned a “0”. The Ostro study utilizes multiple regression techniques. These techniques appear to have been applied properly and the study is assigned a t Il * The threshold issue is discussed in more detail in a later subsection. 10—11 ------- The next column is concerned ‘with the appropriateness of the effects measured in terms of their use in benefit analysis. As might be expected, all of the studies in this column receive a “1” because they reflect those studies which have passed preliminary screening for use in benefit analysis. It is worth noting, however, that there are still variations in the type of effects measured. For example, the Ferris j aj. study makes measurements of respiratory disease incidence. The Mazamdar j. study utilizes data on total mortality rates. The Ostro study measures work—loss days. Although the latter is not traditionally considered a health effect, and thus may not be as useful for Criteria Document purposes, it is none- theless a usable measure for benefit analysis. The final column is concerned with the availability of other research which provides validation or confirming evidence of the results obtained by the listed studies. As discussed in Section 3, the Ferris et j. study results are confirmed in general by a number of studies. These include studies by Lunn and by Colley and Brasser. The Mazumdar j study is an analysis of London data which has been extensively analyzed by other researchers who have found similar results. Thus,both the Ferris and Mazumdar studies ire assigned “1” on this criteria. The Ostro study is consistent ‘with other general research in the area. However, no specific studies are yet available which confirm the type and magnitude of effects measured. Thus, the Ostro study is assigned a rating of “< 1”, The above discussion compares the studies by Ferris j gj ., Maznmdar j]., and Ostro on the basis of a variety of criteria. A review of the ratings in Table 10—1 suggests that the Ostro study must be judged favorably in terms of its use in a benefit analysis. This is clear because the Ostro study rates “1” or “< 1” on all criteria except behavioral controls, where the Ferris g.. , and Mazuadar j,. studies also rate “0”. The latter studies also rate “0” on other criteria. Thus, the Ostro study is given considerable weight in some of the benefit analysis. In contrast, the Ostro study was available only in preliminary form prior to closure of the Criteria Document and was not formally considered by CASAC. 10-12 ------- Alternative Azarosation Procedures Table 10—2 incorporates the judgments about study quality and study coverage discussed previously. Each of the aggregation procedures shown in the table reflects different tradeoffs between the possibility of double counting versus undercounting. The table also incorporates assessments of study quality. The different procedures are each described below. Procedure A — The first column in the table includes only those studies judged by CASAC to be in the quantitative category (5). This includes the studies by Mázumdar 1 j 1 . and Ferris jt,il. (Other studies were judged by CASAC as providing quantitative evidence, but they could not be adapted for benefit calculations.) It also excludes all non—health studies since the Criteria Document is noncommital regarding the availability of quantitative evidence of non—health effects (6). Procedure A is thus a conservative alternative. It probably understates benefits because of its incomplete coverage of benefit categories. For example, it excludes the possibility of welfare effects, acute morbidity effects, chronic mortality effects, and non— respiratory related diseases. By the same token, it is unlikely that Procedure A involves any double counting of benefits. Procedure B — This procedure is similar to the previous one with the exception that the health study by Samet ii ii. has been included also. This modification is consistent with the conclusions of the EPA Staff Paper (7) concerning studies providing roasonable evidence of concentration—response relation- ships for health effects. The Staff Paper includes the Samet j jj. study in this category, as well as the studies in Procedure A. As with Procedure A, no non—health studies are included. The Staff Paper, like the Criteria Document, is noncommittal regarding the availa- bility of quantitative non—health effects evidence. 10—13 ------- 0 Table 10—2 ALTERNATIVE AGGREGATION PROCEDURES Procedure Benefit Category - - - - - A B C U E F Mortality . Mazumdar at al. Mazuadat at al. Mazumdar et a l. Lave & Seskin; Lipfert Lave & Seskin; Lipfert Col. C +• Col. D Col. B + Col. C Crocker et al. Acute Morbidity —— Samet et al. Ostro Ostro Ostro Chronic Morbidity Ferris ct al. Ferris et al. Ferris at *1. Ferris at al. Crocker et at. household Sector Soiling & Materials —— —— Matbtech Mathtech Geom. mean of Cola. D & P at county level Watson & Jaksch + Cummings et al. Manufacturing Sector Soiling & Materials Mathtech Other Sectors —— — — —— —— — Soiling & Materials Math tech ------- Procedure C — Procedure C brings in the study of acute morbidity by Ostro. As noted previously, the Ostro study compares very favorably with the studies included in Procedures A and B. With the addition of the Ostro study in Procedure C, the study by Samet j gJ,. is omitted to avoid possible double counting of benefits. The Saiset j j,. study is concerned with acute respiratory effectsb which may be a subset of the effects captured in the Ostro study. The Ostro study is also given preference over the acute morbidity study by Saric j j. The latter study provides less complete coverage of diseases (respiratory disease only) and limited concentration—response information (e.g., no separation of PM/SO 2 effects, no information on PM effects below 200 Lg/m 3 annual mean, etc.). Procedure C also incorporates the Mathteoh study of household soiling effects. The Criteria Document and Staff Paper reference this recent study but are noncommittal as to its use for quantitative purposes. The study is included here in order to provide some coverage of soiling effects, in recognition of its strong analytical features, and because of the favorable peer review whic h the study has received. Note that Procedure C excludes possible chronic exposure mortality effects, provides limited coverage of chronic morbidity effects (respira- tory illness only), and offers limited coverage of soiling effects. Pro..dur. D — Procedure D addresses some of the possible incompleteness of coverage provided by Procedure C. It adds the chronic exposure mortality studies by Lave and Seskin and by Lipfert. The Lipfert studies are significant in that they addressed some of the criticisms ascribed to the work of Lays and Seskin (e.g., omission of data on smoking habits), and yet found comparable results. These studies are discussed in more detail in Section 4, 10—15 ------- With the addition of the chronic exposure studies, the acute exposure study by Mazumdar . j,. is dropped from this procedure. This is because of the possibility for overlap between those estimates. In particular, the chronic exposure studies are based on annual mortality rates. Annual mortality rates will include all deaths during the year, including those deaths that may be due to acute exposures. Thus, it is possible for the chronic exposure studies to be capturing the mortality effects of both acute and chronic exposure. The extent to which this may happen is unknown, however. It depends on a variety of factors such as the functional forms for the acute and chronic dose—response relationships, the statistical correlation between the measures of acute and chronic exposure, and so on. To be conservative, the acute exposure study is eliminated from this procedure. Procod res E and 1 — Procedures E and F are most easily considered together. Procedure E addresses the incomplete coverage of the chronic morbidity category and the underestimation of soiling effects. In Procedures A through D, chronic morbidity estimates arà based on the study by Ferris j aj,. which iucludes only respiratory diseases. In Procedure E, the Ferris j j. study is replaced by the Crocker j ,. study which includes more chronic illnesses. In Procedure E, the Mathtech study of soiling and materials damage in parts of the manufacturing sector is included. The coverage of household sector soiling is also expanded. The latter is done by taking into con- sideration the results of studies by Watson and Iaksch and by Cummings al. The sum of these two studies is used in Procedure F, in view of the possibility that together they may overestimate benefits. In contrast, the Matbtech household study probably underestimates benefits. Procedure E thus uses a compromise estimate for household soiling: the geometric mean of the Mathtech estimate and the sum of the estimates based on Cummings j j. and Watson and Jaksch. The geometric mean is used as a conservative measure of the average of the two estimates. 10-16 ------- The remainder of the estimates used in Procedure F also seek to provide more complete coverage of benefits, with the possible risk of some double counting. In particular 1 the benefits for the acute and chronic exposure studies for mortality are added together; the same is also done with the estimates from the acute and chronic exposure studies of acute morbidity. As a result, Procedure F provides the most complete estimate of benefits possible with the available studies. However, it may involve some double counting. The possibility of double counting is not likely to arise with any of the procednzes A through E. Resalts of the Aureaation Procedures Applying the procedures outlined in Table 10—2 leads to the results shown in Table 10—3. The benefit estimates shown in the table are for the PM1O (70, 250) Scenario B standard. This table also appears in Section 1 and is discussed in detail there. Corresponding tables for the other standards can be found in Section 11. APPLICARLR CON l1tATIOt4 RM E$ P01 BEN ?IT C&L LATIGNS Introduction The results of health effects studies can often be expressed in the form of concentratiOn—resPOnSe functions. Such functions represent a quan- titative statement of the relationship between health status and ambient concentrations of an environmental pollutant such as particulate matter. Concentration-respOnse functions are useful in a benefit analysis because they provide a mathematical basis for calculating health improvements associated with air quality changes over different concentration ranges. This subsection is concerned with the ranges over which the concentration— response functions should be applied. EPA’s Staff Paper and EPA’s Clean Air Scientific Advisory Committee (CASAC) addressed the question of concentration ranges over which health effects may occur. These ranges are identified and discussed in the next 10—17 ------- I-. Table INCREMENTAL BENEFITS FOR I1IE PM1O 10—3 (70, 250) SCENARIO B STANDARD* * 1982 discounted present values in billions of 1980 dollars for a 7—year time horizon (1989—95) and a 10 percent discount rate. Individual entries may not sum to subtotals due to independent rounding. Benefit Category A B C D E F Mortality 1.12 1.12 1.12 12.72 12.72 13.84 Acute Morbidity 0.0 1.32 10.65 10.65 10.65 11.97 Chronic Morbidity 0.12 0.12 0.12 0.12 11.40 11.40 Household Sector Soiling & Materials 0.0 0.0 0.73 0.73 3.14 13.85 Subtotal 1 1.24 2.56 12.63 24.24 37.92 51.07 Manufacturing Sector 0.0 0.0 0.0 0.0 1.30 1.30 Subtotal 2 1.24 2.56 12.63 24.24 39.22 52.36 Minimum Midpoint Maiimum Property Value Studies 3.43 6.85 11.42 Hedonic Wage Studies 9.81 19.81 37.21 ------- section. The lower bounds of these concentration ranges will be referred to hereafter as the “Staff Paper lower bounds”. Some of the alternative PM standards under consideration are below the Staff Paper lower bounds, or may result in ambient PM concentrations below them. The below—lower bound concentrations can result for several reasons. One example is that controls required in an area to attain the 24—hour standard may lead to associated reductions in annual average concentrations below the annual standard. The reverse is also possible. Other examples can also arise due to differences between the statistical form of the standard and the statis- tical form of the threshold, due to the timing of control implementation, or due to the dispersion pattern of PM in the vicinity of an emissions source being incrementally controlled. The existence of air quality improvements below the Staff Paper lower bounds poses difficult problems for the benefit analysis. These problems include such issues as the appropriateness of attributing benefits to air quality improvements below the lower bounds, statistical procedures for applying lower bounds in a benefit analysis, and uncertainty about the level of the lower bounds. In view of these uncertainties, calculations and results are included for two cases —— one without consideration of the Staff Paper lower bounds, the other with benefits below the lower bounds excluded. The section begins ‘with a review of the evidence the Staff Paper cited In. determining specific lower bounds. This is followed by a discussion of the applicability of the lower bounds to a benefit analysis in general and to this analysis in particular. Also discussed are other difficulties with restricting the allowable concentration ranges for the benefit calcuja— tions. Finally, the results of the calculations are presented. Concentration Ianus Sua..stsd by and tiii EPA/OAQPS Staff Panor The EPA/OAQPS Staff Paper explicitly addressed the question of concen- tration ranges over which health effects may occur. The Staff Paper identified one set of ranges at or above which health effects are likely to 10—19 ------- occur among sensitive groups in the population; it identified a second, lower set of ranges over which effects may be possible. but for which the evidence and level of risk was less certain. These ranges were derived from the EPA Criteria Document and reflect consideration of the epidemi- ology studies judged tb provide the most reliable quantitative evidence. The numerical values of these ranges are identified in Table 10—4. EPA’s Clean Air Scientific Advisory Committee (CASAC) also addressed the question of concentration ranges over which health effects may occur. Their conclusions were summarized inthe CASAC Chairman s closure letter (8) on the Staff Paper. CASAC concluded that detectable health effects occur at the upper bound of the “Effects Possible” ranges idàntified in the Staff Paper. These ranges are 150 to 350 pig/m 3 for 24—hour PM1O concentrations (150 to 250 ig/m 3 BS) and 55 to 110 ig/m 3 for annual average PM1O concen- trations (110 to 180 g/ra 3 TSP). The closure letter also made it clear that these conclusions were based solely on a review of currently available quantitative evidence from epidemiology studies. That is, CASAC did say that there are no health effects at the lower end or below these Table 10—4 SUMMARY OF CONCENTRATION RANGES IN EPA! OAQPS STAFF PAPERS Short-To rin Long-Term BS PM1O TSP PM1O “Effects Likely” 250—500 350—600 2. 180 90110 “Effects Possible” 150250 150—350 110—180 55110 * All entries are in ig/m 3 . Concentrations are given in both original study units and approximately equivalent PM1O units. Source: EPA/OAQPS Staff Paper, 2.2. iii.’ Tables 1 and 2. 10-20 ------- concentration ranges; it merely said that there was currently no quantita- tive evidence of effects. The evidence for the lower bounds is discussed in. the EPA Staff Paper, which CASAC has reviewed. The evidence is summarized below, first for the 24—hour lower bound and then for the annual lower bound. The 24—Hour Lower Bound — The EPA Staff Paper (9) concludes that there is no strong scientific evidence of health risks of consequence at 24—hour PM1O levels below 150 j ig/rn 3 (approximately 150 iig/m 3 BS). This conclusion is based on the results of Martin and Bradley (10), Ware j j.. (11), Mazumdar al. (12), and Lawther it ii. (13). The first three sets of studies are mortality studies, while Lawther it. .&1. is concerned with morbidity effects. Martin and Bradley and Ware it. ii .. examine mortality data for London for the winter of 1958—59. The lowest average daily pollution level in their data was 150 ig/rn 3 BS. Therefore, the Martin and Bradley and Ware j ],. studies provide no information on effects of BS below 150 118/rn 3 . Mazurndar gJ,. examine London data for 14 winters, including the winter of 1958—59. Their data include daily pollution values as low as 13. BS. Significant effects of BS on mortality are observed for both episodic days (days with BS levels over 500 jig/rn 3 and seven days on each side), and non—episodic days. No formal tests for evidence of a threshold are conducted. Ostro (14) reanalyzes the Mazumdar it . J,. data using specific statis- tical tests for the existence of a threshold. He finds strong evidence of. effects at ES levels below 100 jig/m 3 and some evidence of effects below 75 118/rn 3 , the lowest threshold Ostro considered. Thus, Ostro’s reanalysis of the London mortality data provides no evidence of a threshold at 150 g/m 3 . In fact, Ostro’s analysis indicates that effects may have occurred at BS 10-21 ------- levels of 75 xg/m 3 with effects possible at lower levels. Ostro’s analysis was not yet available at the time the EPA Staff Paper was completed. In a series of studies, Lawther et j. examined the relationship between changes in morbidity and BS levels. Lawther suggests that signifi- cant effects occur at BS levels over 250 tgJm 3 . In the winter of 1967—68, however, a correlation between changes in symptoms and BS levels is observed although smoke levels average 68 Lg/m 3 and exceed 250 g/m 3 only once. In addition, Lawther et perform very limited statistical analysis. Most conclusions are derived from visual inspection of graphs which does not permit systematic examination of the influence of each variable. The Annual Lower Bound — The PM Staff Paper’s analysis of the chronic exposure lower bound is based on the studies of Ferris . j. (1516), Bouhuys e J. (17), and Lunn j gi,. (18). These studies examine differences in health status between a few time periods or areas. As the EPA Staff Paper discusses, When only two geographic areas or time periods are involved in a study, the levels that happened to exist in the more polluted area or time period usually are listed as the ‘effects’ levels. Effects levels thus derived are somewhat arbitrary, but do represent concentrations where effects can be ascribed with some certainty. Because a gradient of effects is often observed for multiple area studies, however, some risk exists that effects decrease but do not necessarily disappear at concentrations substantially below those of the more polluted area. Therefore, ‘effects’ levels derived from two—area (or time) studies cannot be regarded as thresholds or ‘no detectable effects’ levels (19). In fact, effects may continue to occur at concentrations below those in the clean area. For example, even the 97 Lg/m 3 BS level and 40 Lg/m 3 TSP in the clean areas in Lunn et al. and Bouhuys j gJ. cannot be interpreted as threshold or “no effects” levels. 10-22 ------- In Ferris ., decreases in respiratory disease incidence accompany decreases in annual TSP in Berlin, New Uampshire of 180 to 130 j.ig/rn 3 between 1961 and 1967. No decrease in incidence is observed when annual TSP declines from 130 to 80 Lg/m 3 between 1967 and 1973. Based on these results, the Staff Paper selects a “no observed effects” level of 130 TSP. The lack of observed effects, however, may result from the increase in su.lfation or other changing influences between 1967 and 1973. A decrease in disease incidence from the decline in TSP and an increase in incidence from increased sulfation could have the net effect of leaving incidence unchanged. The annual threshold of 110 tg/rn 3 TSP is based largely on a study of two Connecticut towns by Bouhuys, Beck and Schoenberg. The study found evidence of increased respiratory symptoms (cough, phlegm and dyspnea) in the town with dirtier air (Ansonia), but no differences in prevalence rates for chronic bronchitis or lung function. Concentrations of TSP in the cleaner town averaged 40 fLg/m 3 annual mean. Concentrations of TSP in Ansonia averaged 63 fig/rn 3 annual mean during the year of the study and up to 152 ig/m 3 annual mean during the previous seven years. The EPA Staff Paper characterizes pollution in Ansonia by a median value of 110 pg/rn 3 TSP annual mean (20) and this value is the basis for the lower—bound annual concentration (21). Alternative conclusions are also possible from the Bouhuys j study. For example, the reported increase in symptoms could be related to both current and historical pollution levels. In this case, 63 pg/rn 3 TSP rather than 110 ig/rn 3 TSP would be a safer estimate of the lower bound. Also, the presence of symptom differences between the two towns implies effects at concentrations at least as low as in the dirtier town, and possibly lower. Thus, the possibility of symptoms appearing at levels below 63 pg/rn 3 cannot be ruled out. The choice of 110 pg/rn 3 TSP as a threshold level is thus not a uniq ue choice. S Despite the increase, sulfation was still at relatively low levels. 10-23 ------- Snumary of the Evidence — As discussed above and in the EPA Staff Paper (22), there is no clear evidence of thresholds below which effects do not occur. The 24—hour threshold of 150 jig/rn 3 BS (approximately 150 jig/rn 3 PM1O) is contradicted by more recent evidence suggesting effects at BS levels of 75 jig/rn 3 and possibly lower. The annual threshold of 110 jig/rn 3 TSP (approximately 55 jig/rn 3 PM1O) is a choice from the range of 63 jig/rn 3 to 152 jig/rn 3 in the dirtier of two towns. In addition, as noted previously 1 one cannot rule out effects below this range on the basis of a two—area study. Ap- licability of Staff Pacer Lower Bounds to the Benefit £ual sis Applicability of Staff Paper Lower Bounds in General — As the EPA Staff Paper’s use of the Bouhuys j J. study and others suggests, selection of the “Effects Possible” concentration range for the primary standard emphasizes identification of concentrations at which effects of a medically significant or permanent nature are detectable. Effects in the form of increased symptoms by themselves, appear to be less influential in determining that range. Hence, symptomatic effects may still exist at concentrations below the Staff Paper lower bounds. In contrast, the benefit analysis is concerned with the benefits of reducing ambient air pollutiom Benefits are determined by individuals’ willingness to pay for pollution reductions. Individuals may value a reduction in symptoms as well as valuing reductions in more significant health problems. In this case, then, ‘benefits may exist at concentrations below the lower bounds selected-during Staff Paper development. Applica- tion of the lower bounds in the calculation of benefits may thus be inappropriate because of the differing objectives of benefit analysis compared to Staff Paper and Criteria Document development. The distinction between symptoms and significant health effects becomes particularly important if there are separate concentration—response 10-24 ------- functions for the two categories. In this case, the concentration—response functions for the significant health effects could be used to calculate benefits down to the lower bound concentrations prescribed by the Staff Paper. In addition, the concentration—response functions for symptoms could be used to calculate benefits below the lower bound concentrations. The only major constraint on the latter calculations would be those suggested by the sample ranges in the original studies. However, benefit calculations below the Staff Paper lower bounds may be appropriate even when separate concentration—response functions don’t exist. The reason for this is the uncertainty surrounding the level and existence of a no—effects threshold. That is, the location of the thres- hold is not known precisely and is perhaps better characterized by a proba- bility distribution. Or alternatively, the concentration—response functions themselves are perhaps better represented by probability distri- butions. The existence of uncertainty raises the possibility that lay individuals, as well as informed experts, may assess the risks differently, i.e., subscribe to different probability distributions. In this case, the benefit analysis should account for the differing probability assessments. That is, it may not be appropriate to constrain the benefit analysis by Staff Paper and CASAC risk assessment. This is because economic principles prescribe that the benefit analysis should measure the sum of individuals’ willingness to pay for reduced risk. This means that the benefit analysis should reflect lay assessments of risk and risk valuation. These lay assessments may not be substantially different from Staff Paper and CASAC assessments. Nonetheless, restricting the benefit analysis to consistency with their determinations would violate the fundamental principle of benefit—cost analysis —— namely, that benefit estimates should reflect individuals’ preferences. Applicability of Staff Paper Lover sounds to This Analysis — The above discussion summarizes the conceptual problems associated with use of Staff Paper lower bounds in a benefit analysis. Those problems 10-25 ------- arise in connection with this benefit analysis. First, separate concentration—response functions for health effects and symptoms are not generally available or equally amenable to benefit calculations. Thus, it is not possible to treat these two categories differently with respect to restrictions on below—lower bound benefit calculations. Benefits below the lower bounds must be excluded either in both cases or in neither case. Second, PM health risk assessments by individuals in the population are not available. This would require extensive survey work which was not practical for this study. Thus, the analysis must rely either on CASAC and Staff Paper assessment of the health risk (i.e., CASAC and Staff Paper study selection and lower bound determination) or on other interpretations of the health risk suggested by available studies. In view of the above uncertainties, several sets of benefit calcula- tions are developed. One set allows health benefit calculations below the Staff Paper lower bounds, the other excludes health benefits below the lower bounds. Each of these sets is further distinguished by several alternative interpretations of health risk. These alternative interpreta- tions are based on different assessments of the available health risk literature. One assessment is consistent with CASAC’s view, while others provide alternative interpretations. Practical Constraints in Auclvina the Staff Psoer Lower Bounds There are additional practical constraints to imposing the Staff Paper lower bounds on the benefit analysis. These constraints include: 1) data limitations which limit attention to the annual lower bound only; and 2) the possibility of introducing statistical bias by imposing the lower bound without proper adjustments to the concentrationrespOflse function. These problems are discussed further below. Data Liaitations — The Staff Paper identified both 24—hour and annual mean lower bounds. Application of these lower bounds requires pre—control. and post—control air io-26 ------- quality data on both a daily and annual basis. Air quality data available from the control cost and air quality analysis does not include daily data, other than for the second—highest day per year. Thus, the benefit analysis can apply only the annual mean lower bound. The annual mean lower bound is directly applied to those benefit models based on studies that used the annual mean as the pollution exposure index. This includes the studies by Ferris et gj., Ostro, Crocker j j . , and Lipfert/Lave and Seskin. The studies by Mazumdar j gJ. and Samet j . j . both make use of daily pollution data. As explained in Section 3, however, these studies can be used for equivalent or approximately equivalent benefit calculations based on annual data. The annual mean lower bound can be imposed on these calculations. The 24—hour lower bound, however, cannot be applied in the absence of daily data. Statistical Bias — Practical constraints limit the benefit analysis to use of concentration—response functions which in most cases are taken directly from the original research studies. None of these studies imposed the Staff Paper lower bounds at the time the concentration—response functions were statistically estimated, either by the authors or by Mathtech. Estimates of the slopes of the conoentration—response functions reflect the full range of data available in each study. If this range includes a no— effects region. the slope estimatei may be downward—biased estimates of the slope in the effects—present region. The possibility is illustrated in Figure 10—1. In Panel A. linear regression leads to line segment AA as the best fit to the data. In Panel B, the same data are split into two parts around the concentration Q*. Linear regression leads to separate line segments BB and CC in each subset of the data. Line segment BB has approximately zero slope. Line segment CC has a positive slope, and more importantly, a slope that exceeds the slope of AA. 10-27 ------- Panel A Morbidity Rate PM Concentration B C Morbidity Rate Q PM Concentration Figure 10—1. Constrained Estimation of Concentration—ResPOnse Function Panel B B C 10-28 ------- If Q 5 in the figure represents the Staff Paper lower bound, then benefits above the lower bound should be estimated using the concentration— response function given by CC. If AA is used to estimate the above—lower bound benefits, the estimate will be downward biased. Note that the degree of bias is likely to be more severe for linear concentration—response functions. Other functional forms such as the quadratic may be less biased. Correct application of the Staff Paper lower bounds would require: 1) re—estimating all of the concentration—response functions as shown in Panel B (or, at a minimum, all of the linear ones); and 2) using the revised slope estimates, analogous to CC in the figure, to calculate benefits. Re— estimation was not possible as part of this analysis. Thus, our estimates of the health benefits above the Staff Paper lower bounds are likely to be downward biased. R*sults of the Calculations The above discussion summarizes the potential problems with applying the Staff Paper lower bounds to the benefit analysis. Despite these problems, it is useful to know how sensitive the benefit estimates are to the imposition of the lower bounds. For this reason, benefit calculations are developed for all standards, first without any lower bounds imposed, and then with the annual lower bound imposed. The results of these calcu- lations are summarized in Section 1 and reported in more detail in Section 11. mICA O AREA C&SE wuT Introduction The cost and air quality analysis (23) estimates the ambient air quality conditions at the “design value” pollution monitoring site in each county. The design value monitor in a county typically is, the monitor that recorded the highest ambient concentration in 1977 and 1978. The 10-29 ------- concentrations at other monitors in the county are thus generally lower than the design value monitor. Non—monitored areas in a county may experience higher or lower concentrations than the design value monitor, but lower concentrations are more likely. Air quality conditions at these other monitors and locations are not estimated in the cost analysis. Estimates were not developed because it would require detailed dispersion modeling, an impractical task in the context of a national analysis. Benefit calculations require more detailed information about the spatial distribution of the air quality improvement in each county than is available from the cost analysis. Given the limited information available, there are two alternative assumptions that can be made concerning the spatial distribution of the air quality improvement: • The improvement throughout the county is equal in magnitude to the improvement at the design monitor. • The improvement throughout the county is equal in percentage terms to the improvements at the design monitor. The situation with an equal magnitude improvement is illustrated in Figure 10—2. Prior to the implementation of controls, Curve A illustrates the ambient concentration in the county. The county in this case is repre- sented as a one—dimensional county which has three monitors, a design monitor and two others. Prior to implementation of controls, the ambient air quality in the county exceeds the standard by an amount 1. After implementation of controls, air quality at the design monitor is reduced by an amount X and thus brought into compliance. The air quality improvement throughout the rest of the county in this case is also assumed to improve by an amount I. The case with an equal percentage improvement is shown in Figure 10—3. In this case, Curve A again illustrates the ambient concentrations in the county before controls are applied. Conditions at the, design monitor are as in Figure 10—2, namely, exceeding the standard by an amount x. Following implementation of controls, air quality improves to the level 10-30 ------- i -’g/m Standard Figure 10—2. Equal Magnitude Improvement Post-Control A B nitor 2 Design nitor rv cnitor 3 10—3 1 ------- given by Curve B. At the design monitor 1 the improvement is an amount X. Elsewhere in the county, air quality is assumed to improve by the same percentage amount as occurred at the design monitor. The benefit estimates reported in Section 3 through 7 of the report are based on air quality data which incorporate the equal percentage improvement concept illustrated in Figure 10—3. This approach yields more conservative estimates of benefits than would calculations based on the equal absolute improvement shown in Figure 10—2. The conservative approach is used in the benefit analysis because it is believed to be more realistic. That is, the dispersion properties of PM may attenuate the air quality improvement occurring at various distances from the emission sources being incrementally controlled. However, the degree of attenuation cannot be assessed without detailed dispersion modeling of the PM emissions controls applied in each county. As noted previously, such modeling is very time and resource intensive and thus not practical in a nationwide study. There is a possibility that air quality improvements could be even more localized than is suggested by the equal percentage improvement approach. For example, if control sources are concentrated in one part of a county or only selected “hot” sources are targeted for control, changes in air quality may be limited to localized improvements. An example of this possibility is illustrated in Figure 10—4. Note in this case that the improvement at the design monitor is the same as in the previous figures. However, the improvement near the other monitors is substantially reduced compared to the earlier figures. To clarify the implications of possible localized improvement, it was decided that a detailed analysis should be undertaken for one county using detailed dispersion modeling. The county selected for the analysis was Cook County, Illinois. Cook County, which includes the City of Chicago, was selected for two reasons: 1) it accounts for the largest fraction of total national benefits among all counties in the analysis; 2) it is an area that already has undergone detailed dispersion modeling. 10-32 ------- Figure 10—3. Equal Percent Improvement Standard Post-Control A B Monitor 2 Design nitor nitor 3 10—33 ------- pg/rn 3 Standard Figure 10—4. Localized Improvement pre-Control Monitor 2 Design Monitor Monitor 3 10-34 ------- The remainder of this subsection is concerned with the results of the Chicago area case study. Two specific analyses are described: • Benefit estimates based on detailed dispersion modeling results for Cook County. • Benefit calculations based on the same dispersion modeling results but using benefit calculation procedures analogous to those done in the national analysis. In the first analysis, Cook County is divided into 31 subareas corres- ponding to the Census divisions within Cook County. In the second analysis, Cook County as a whole is analyzed using procedures similar to that in the national analysis. The following subsections describe the data, methods, findings, and limitations of these analyses. Dispersion modeling for the Chicago area was done previously by Pechan and Associates (24). The modeling resulted in estimates of TSP concentra- tions for over 1,400 modeled receptors in the Chicago air quality control region (AQcR). Attributes of the data include: • A TSP standard of 150 jig/na 3 observed 24—hour second high is analyzed. • Pro—control (baseline) data are representative of modeled conditions in 1987. • Post—control TSP levels are also assumed to occur in 1987. • Some receptors are not in attainment after available controls are implemented. • The pro— and post—control data measure total concentrations and not just contributions from selected sources. • Estimates are provided for the 24—hour second high measure as ‘well as the equivalent levels for TSP annual arithmetic mean and geometçic mean. • The location of each modeled receptor is defined by UTM coordinates. 10-35 ------- • Pro— and post—control concentration levels are also provided for each monitor site in the Chicago SMSA that meets the EPA summary criteria. Ago ro a oh To test the sinsitivity of the benefit estimates to the level of spatial aggregation in the air quality data, the geographic area of Cook County, Illinois, was considered at two levels of detail. In the aggregate version, Cook County is analyzed as a single area. In the disaggregate approach, Cook County is divided into 31 Census divisions. For each level of aggregation, one file of economic/demographic data (ED) and six files of air quality data are constructed. The attributes of the ED file include: • ED data for Cook County are available from the nationwide PM benefit analysis. • 1980 population and population growth rates by Census division are obtained from (25). No other ED variables are currently available for 1980 for this area. • Data by Census division for the number of households, employment, the size of the work force, and mortality rates are calculated as population—weighted averages from the Cook County estimates. • Data by Census division for average real income, average property values, employment growth rates, and cost of living indices are assumed to be equal to the Cook County values. The six air quality data files for the two levels of aggregation are similar in that they represent information on pre— and post—control levels (two alternatives) and three measurement types (arithmetic, mean, geometric mean. 24—hour second high). However, the data development procedures for the two aggregation schemes are different. In particular: 10-36 ------- • For the Cook County analysis, a single index of TSP for each file is created under the following assumptions: — For each measurement type, the single j or recording the highest pre—control concentration is designated as the “design value” monitor. — A pro—control average index of TSP concentrations is calculated for each measurement typo. The average is based on the available pro—control monitored data. — The percentage change in the average measure of con- centrations is assumed to be equal to the percentage change observed at the design value monitor. This yields the post—control average level of TSP for Cook County. • For the Census division analysis, indices of air quality are created for each division. The assumptions include: — For each measurement type, the modeled receptor recording the highest pro—control concentration in the Census division is identified. — A pro—control average index of TSP concentrations is calculated for the three measurement types in each Census division. The average is based on the avail- able pro—control modeled receptor data. — Thó post—control maximum value is the concentration level occurring at the modeled receptor reporting the highest value in the pro—control case. — A post—control index of average TSP levels in a Census division is calculated from the available post—control modeled receptor data. Given these data files, benefit calculations are carried out according to the procedures described previously in Sections 3 through 7. While the structure of the individual dose—response function and valuation procedures does not change, one change is made in the scenario definition. Specif i— ashy, the time horizon is only one year, 1987, since this is the only year for which air quality data are available. 10-37 ------- FindinKs The results of the analysis indicate that benefits are reduced in the Chicago area when more detailed information on population and air quality distributions is available. This result is relative to a situation in which data from the highest monitor in a county are used to characterize levels and changes in. exposure. Table 10—5 identifies the ratio of benefits calculated for the Census divisions to the Cook County study. The relatively high ratio for Lipfert/Lave and Seskin, Crocker gJ., and Mathtech are due to the fact that the air quality variable in each of these studies is defined in terms of maximum rather than average concentrations. In addition, the small variation around 0.27 for several of the studies is due to nonliuearities in the model dose—response functions. Table 10—5 RATIO OF NSUS DIVISION ESTIMATES TO COOL COlThfl’! ESTIMATES Study Ratio Mazumdar et al. 0.33 Lipfert/Lave and Seskin 0.94 Ssmet jj . 0.27 Saric j. . Ostro 0.27 Ferris j ii.. Crooker 0.80 Property Value 0.28 Hedonic Wage 0.29 Cummings .&.k. 0.27 Watson and Jaksch 0.26 Mathtecb. Household 0.62 * Zero benefits calculated for point estimate. 10-38 ------- Tables 10—6 through 10—8 present the comparative findings for three of the benefit aggregation procedures discussed previously in this section. The three benefit aggregation procedures are designated B, C, and D. The benefit estimates reported represent point estimates, as defined in Section 1. Note that these benefit estimates are based only on individuals residing or working in Cook County. Although air quality improvements also occur in neighboring counties, benefits are not calculated for these areas. Limitations The degree of generality that can be attached to the above findings is limited by several factors. These include: • Limitations with the assumptions implicit in the current analysis. • Limitations in applying these results to the PM NAAQS analysis. With respect to the approach limitations, we note the following attributes: • Air quality improvements occur outside Cook County but are not included in the previous tables. These improvements may be generated in part by controls on sources in Cook County. • Estimates of exposure are still approximate. The daily movements of people between Census divisions are not identified. • Much of the economic/demographic data is attributed. In particular, population—weighted averages are used to define several data series. • Statistical issues may arise when models are used at levels of spatial aggregation that differ from those levels used in the original study. • One of the Census divisions in Cook County accounts for a large percentage of the total county population and is relatively large in area. 10-39 ------- Table 10—6 BENEFIT ESTIMATES FOR TWO METHODS OF AIR QUALITY AGGREGATION AND FOR BENEFIT AGGREGATION PROCEDURE B Study Census Divisions Cook County Mazumdar et al. 0.90 2.74 Samet j ii, . 1.98 7.24 Ferris et j.. 0.00 0.00 Total 2.88 9.98 * Discounted present values in 1982 in millions of 1980 dollars. Table 10—7 BENEFIT ESTIMATES FOR TWO METHODS OF AIR QUALITY AQGREGA ION AND FOR BENEFIT AGGREGATION PROCEDURE C Study Census Divisions Cook County Mazumdar j jj. 0.90 2.74 Ostro 15.88 57.79 Ferris j . ,. 0.00 0.00 Msthtech Household 4.66 7.51 Total 21.45 68.05 * Discounted present values in 1982 in millions of 1980 dollars. 10-40 ------- Table 10—8 BENEFIT ESTIMATES FOR TWO METHODS OF AIR QUALITY AGGREGATION AND BENEFIT AGGREGATION PROCEDURE D 5 Study Census Divisions Cook County Lave and Seskin Ostro Ferris j.. Mathtech Household Total 77.41 15.88 0.00 4.66 97.96 82.40 57.79 0.00 7.51 147.71 * Discounted present values in 1982 in millions of 1980 dollars. With respect to the PM NAAQS analysis, the following points should be considered prior to any comparisons: • The cost—of—control routines differ in the two analyses. In the PM NAAQS study, a modified rollback approach is employed. In the Chicago area case study, a least—cost algorithm is used. • Differences in study design and data files in the cost analysis lead to differences in air quality levels and changes. For example, the Chicago model predicts a county— leve’ change in the design value geometric mean. of 9.78 g/m from a baseline level of 99.17. On the other hand, the PM NAAQS results are based on predicqd changes at the design value of 25.9 (geometric mean) Lg/m from a baseline of 132.2. • The scenarios differ in the two studies. The Chicago area study is based on a one—year benefit stream; the PM NAAQS study is based on a 9—year benefit stream (for TSP). • The air quality and population distributions implicit in the Chicago area analysis are likely to be different in other areas. As a consequence, extrapolation of the results reported here to other areas is tenuous at best. 10-41 ------- • Given the above caveats, if these results are to be used to benchmark the PM NAAQS results, the results of this analysis suggest the following: 1) Detailed dispersion modeling may result in benefits estimates that are between a factor of 2 to 4 below those reported in the PM NAAQS study. 2) Detailed dispersion modeling may also lead to changes in cost of control estimates relative to a county—wide rollback approach. 3) Comparisons of adjusted benefits to costs cannot be made until more complete information is developed with respect to the different cost—of—control approaches and impacts on resultant air quality. 1. 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, North Carolina, December 1981, p. 14—5. 2. Ibid., pp. 14—49 to 14—54. 3. U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards. Rev’iew of the National Ambient Air Quality Standards for Particulate Matter: Assessment of Scientific and Technical Information. OAQPS Staff Paper (EPA—450/5—82—OO 1 ), Research Triangle Park, North Carolina. January 1982, pp. 48 — 63 . 4. Ferris, B. G., Jr., H. Chen, S. Puleo and R. L. H. Murphy, It. Chronic Nonspecific Respiratory Disease in Berlin, New Hampshire, 1967—1973: A Further Follow—Up Study. American Re’ iew of Respiratory Disease, 113:475—485, 1976, p. 480. 5. EPA Criteria Document, . çj ., pp. 14—49 to 1454. 6. Ibid., p. 10—73. 7. EPA Staff Paper, . jt., pp. 48—63. 8. EPA Staff Paper, . ., Appendix E, p. E—4. (Appendix E is the letter from the chairman of CASAC to the EPA Administrator concerning CASAC review and closure of the EPA Staff Paper.) 9. Ibid., pp. xiii—xv. 10-42 ------- 10. 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. ii. Ware, .7., L. A. Thibodeau, F. E. Speizer, S. Colome and B. G. Ferris, 3r. Assessment of the Health Effects of Sulfur Oxides and Particulate Matter: Analysis of the Exposure—Response Relationship. Environ- mental Health Perspectives, 1981 (in press). 12. 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. 13. Lawther, P. 3., R. E. Wailer and M. Henderson. Air Pollution and Exacerbations of Bronchitis. Thorax, 25:525—539, 1970. 14. Ostro, B. D. A Search for a Threshold in the Relationship of Air Pollution to Mortality in London: A Reanalysis of Martin and Bradley. Working Paper, September 21, 1982. 15. Ferris, j iL., .22 Lit. 16. Ferris, B. G., Jr., I. Higgins, M. W. Higgins and I. M. Peters. Chronic Non—Specific Respiratory Disease in Berlin, New Hampshire, 1961—1967: A Follow—Up Study. American Review of Respiratory Disease, 107:110—112, 1973. 17. Bouhuys, A., G. I. Beck and 3. B. Schoenberg. Do Present Levels of Air Pollution Outdoors Affect Respiratory Health? Nature, 276:466— 471, 1978. 18. Lunn. I ., I. Knowelden and A. 3. Handyside. Patterns of Respiratory Illness in Sheffield Infant School Children. British Journal of Preventive Social Medicine, 21:7—16, 1967. 19. EPA Staff Paper, 22. p. 104. 20. Ibid., p. 62. 21. Ibid., p. xvi. 22. Ibid., p. 88. 23. A. E. Smith and L L. Brubaker. Costs and Air Quality Impacts of Alternative National Ambient Air Quality Standards for Particulate Matter. Technical Support Document. Argonne National Laboratory, Argonne, Illinois. October 1982. 10-43 ------- 24. The data were developed for this study by Pechan and Associates, Washington, DC using the AIRMOD model. A description of AIRMOD is available in: Putnam, Hayes and Bartlett, Inc. Analysis of Alternative TSP and PM10 Ambient Standards: A Case Study of the Chicago Metropolitan Area. Draft report prepared for EPA, August 1982. 25. U.S. Bureau of Census. 1980 Census of Population: Illinois. PHC80- 15. 10-44 ------- |