CASE STUDY: Evaluating Environmental Equity in Allegheny County

Community Risk Profiles: A Tool to Improve Environment and Community Health

• Summary
• Background Paper
• Evaluating Environmental Equity in Allegheny County – Theodore S. Glickman
• Comparing Apples and Supercomputers: Evaluating Environmental Risk in Silicon Valley – Lenny Siegel
• List of Participants

THEODORE S. GLICKMAN
Resources for the Future, Washington, DC.

INTRODUCTION

No environmental issue seems to be attracting more attention in Washington these days than environmental justice, as evidenced by the executive order requiring due consideration to be given to the environmental equity impacts of federal actions and by the establishment of a special office at the U.S. Environmental Protection Agency (EPA) for that purpose. These developments have been motivated by public concerns over situations in which disadvantaged minorities and the poor are discriminated against by having to shoulder a disproportionate share of the impacts of environmental hazards.

In my view, to remedy existing or potential environmental inequities, they must first be identified and measured. While this process goes on, the policies and practices that have permitted or encouraged such problems to develop in the first place must begin to be changed. The concern here, however, is with deciding which cases of environmental inequity are the most serious ones and which ones should be given the highest priority. To support these decisions, equity determinations need to be made by evaluating the difference between the risk to the parties of concern and the risk to the rest of the population. Then priorities need to be set by jointly considering the magnitudes of the disparities and the costs of reducing them.

Fortunately, the environmental justice issue has emerged at a time when a new information technology has made it easier to measure environmental impacts in a spatial context. This technology–known as geographical information systems (GIS)–makes use of software that was originally developed for the purpose of weaving together multiple layers of spatial data, such as the topographical features of a region, its environmental characteristics, and the land uses within it. Of course, as with any software, the use of GIS requires appropriate databases, but much of the data needed to assess environmental impacts in terms of risk are already available, including population counts, levels of environmental hazards such as air pollution, and dose*-response relationships. GIS software simplifies the tasks of specifying the geographical areas of concern, combining data layers within the areas, calculating area-based statistics, tabulating the results, and, finally, displaying the results on a map, perhaps in conjunction with other physical attributes such as the locations of pollution sources and other hazards. The result is that environmental equity can be evaluated quantitatively, thereby injecting an essential degree of objectivity into environmental justice debates.

The study we are currently conducting in the Center for Risk Management at Resources for the Future (RFF) to generate part of a community risk pro file uses GIS to evaluate environmental equity with respect to industrial hazards in Allegheny County, Pennsylvania. Allegheny County includes the city of Pittsburgh, which is well-known for its industrial activities and its diverse population. This study differs in four important respects from related efforts conducted elsewhere:

• First, it constructs impact zones according to the nature and magnitude of the hazard (circles centered at facilities or plumes positioned according to wind directions) rather than relying on spatial units such as counties or zip code areas, which are more convenient but not as meaningful.
• Second, in addition to chronic hazards in the form of air pollution from industrial facilities, it also addresses acute hazards in the form of accidents involving the airborne release of toxic chemicals from facilities where they are stored.
• Third, it measures equity based not only on people’s proximity to hazards but also according to the separate and combined risks associated with the various hazards.
• Fourth, for selected facilities, it also traces the evolution of the current state of environmental equity using historical data on hazard levels, land use, property values, demographics, and agents or indicators of change, be they legal, political, or economic.

Studies such as this one often serve only to confirm the obvious and to imply a level of precision in the results that belies the underlying uncertainties. But our study, whereas not necessarily immune to such failings, also reveals some interesting subtleties related to the need to look beyond aggregate results when making equity determinations and the need to be cautious when using worst-case assumptions along the way. Somewhat surprisingly, it also shows that when a facility is very hazardous there are good reasons why the majority of the victims of environmental injustice may not be members of the most disadvantaged segments of the population.

BACKGROUND

Until relatively recently, the civil rights and environmental movements remained surprisingly separate in the United States. Most academic researchers, environmental groups, and federal, state, and local environmental agencies gave little attention to the idea that environmental risks might be distributed inequitably, although such inequities have been documented for more than two decades (Freeman 1972; Zupan 1973; Kruvant 1975; Berry 1977; Asch and Seneca 1978).

The tide began to turn in 1983, when the General Accounting Office conducted a study of hazardous waste landfills in eight southeastern states (GAO 1983). The study was requested by congressional representative Walter E. Fauntroy, who along with 500 other citizens had been arrested for demonstrating in opposition to the siting of a polychlorinated biphenyl (PCB) disposal landfill in the predominantly black and poor Warren County, North Carolina (Bullard 1990). The GAO found that blacks formed the majority of the population in the counties of three of the four offsite hazardous waste sites in the region. (We use the term “blacks” as used in the census data, rather than the presently preferred term “African Americans.”)

The GAO report was limited by its regional scope and did not claim that its findings reflected a national pattern. In 1986, however, the United Church of Christ Commission for Racial Justice embarked on a study that was intended to fill this void (UCC 1987). It found that the mean percentage of minorities in communities with one or more operating commercial hazardous waste treatment, storage, or disposal (TSD) facilities was twice that in communities with no such facilities (24 versus 12 percent). In communities with two or more operating TSD facilities or with one of the five largest landfills, the mean percentage of minorities was more than three times that in communities with no such facilities (38 versus 12 percent). Somewhat surprisingly, the study also found that race was a stronger explanatory variable than socioeconomic status. In other words, relatively affluent minority communities were more likely to host TSD facilities or large landfills than similarly affluent white communities.

(The minority segment of the population was defined as consisting of blacks not of Spanish origin; Asian and Pacific Islanders, American Indians, and Eskimo and Aleut populations not of Spanish origin; other nonwhite populations not of Spanish origin; and Hispanics.)

In 1987, the release of the UCC’s Toxic Wastes and Race in the United States brought the issue of “environmental racism” to national attention and set the environmental justice movement on its way. Building on this momentum, the University of Michigan held the Conference on Race and the Incidence of Environmental Hazards in January 1990 (Mohai and Bryant 1992a). Following this conference, a small group of social scientists and civil rights leaders formed the Michigan Coalition expressly to lobby the EPA. The coalition’s lobbying efforts were instrumental in influencing the EPA administrator to form an environmental equity workgroup. Two of the workgroup’s principal tasks were to examine the evidence that racial minorities and low-income groups do indeed bear a disproportionate burden of risk, as is commonly assumed, and to review EPA’s performance in addressing environmental inequities.

Released in June 1992, the environmental equity workgroup’s report, Environmental Equity: Reducing Risk for All Communities, concluded that racial minorities and low-income groups suffer higher than average potential exposures to selected air pollutants, hazardous waste facilities, and pesticides in the workplace (EPA 1992a).* The report also noted that whereas there are clear differences between racial groups in terms of disease and death rates, environmental and health data are not routinely collected and analyzed by income or race. The workgroup recommended that EPA give greater attention to issues of environmental equity and revise risk assessment procedures to improve the “characterization of risk across populations, communities or geographic areas” (EPA 1992a, 4), although it did not address what units of analysis might be most appropriate. The workgroup report and EPA’s other efforts to come to grips with the issues of environmental equity have been vehemently criticized by representatives of the minority community for being tardy and incomplete and offering little new information or insights (see Mohai and Bryant 1992b and critical comments in the supplement to the report [EPA 1992b, 81-121]). Nevertheless, the release of the report has put the issues of equity and environmental justice firmly on the policy agenda and has stimulated additional research.

(The report took great care to point out that minorities and the poor are potentially exposed to higher than average levels of these pollutants, recognizing that in the absence of actual measurements it is not possible to state categorically that such groups are exposed to higher levels.)

WHY ALLEGHENY COUNTY?

As shown in Figure 1, Allegheny County, in southwestern Pennsylvania, contains the city of Pittsburgh. In 1990, Allegheny County was the 19th most populous county in the U.S., with a total population of 1.34 million. Pittsburgh is the 40th largest city in the U.S., with a population of 370,000. The larger metropolitan area, known as the Pittsburgh-Beaver Valley Consolidated Metropolitan Statistical Area, has a total population of 2.2 million. Initially, the project was intended to focus only on the city, but it rapidly became apparent that we also needed to consider the nearby area beyond the city limits. Environmental hazards do not respect political boundaries, and many of the facilities of concern were established or had relocated beyond the city limits for a variety of reasons. At the same time, the metropolitan statistical area is a creation of the Census Bureau that has little cultural or political meaning on the ground. Because one goal of the project is to provide information that may be helpful in setting environmental priorities, we extended the geographic scope of the analysis to encompass all of Allegheny County.

Figure 1. Pittsburgh and Allegheny County

This area was chosen as the study site for a variety of reasons. Pittsburgh is by no means the most severely polluted nor racially divided city in the U.S., but it is broadly representative of a “traditional” industrial town and its social and economic history has been well documented. With the development of the steel and coal industries, it suffered severe environmental problems through the late 19th and early 20th centuries. Through a unique public/private partnership, major socioeconomic and infrastructural changes since World War II–the “Pittsburgh Renaissance”–have vastly improved the region’s environmental quality, although some problems remain as part of the legacy of that era (Tarr 1989). The city has a broad racial mix (with approximately 72 percent white, 26 percent black, and 2 percent Asian) and a diversified industrial base. The county is 88 percent white, 11 percent black, and 1 percent Asian.

Unlike many other cities in the northeastern U.S., Pittsburgh has no other metropolitan areas in its vicinity. This distinction lends conceptual and methodological clarity to the study and simplifies data collection and analysis. The city itself maintains relatively few environmental databases, but because it is wholly within Allegheny County, several of the relevant databases are centralized at the county level. For example, vital statistics (on births, deaths, infectious diseases, etc.), air pollution monitoring data, and other environmental data are available through the county Department of Health.

COMPUTING CAPABILITIES AND CENSUS FILES

The analyses in this study were conducted with AtlasGIS (Version 2.1) on a 486 Personal Computer. In addition, we used TIGER Boundary Translator (Version 2.7) to build the geographic base layers from the Census Bureau’s TIGER files. These files include the following geographical information: census block and tract boundaries; city, municipality, and county boundaries; the names, locations, and address ranges for individual streets; lakes and rivers; and road and rail networks. Socioeconomic and demographic data were drawn from the Census of Population’s Summary Tape Files (STF) 1A and 3A. STF1A, which draws on a survey of the entire U.S. population, includes information on the general characteristics of the population (e.g., age, race, sex, Hispanic origin), households (e.g., marital status, family size), and housing (e.g., value, rent, number of rooms). STF3A is drawn from sample data (approximately 15 percent of the population in the case of Allegheny County) and includes additional information on the population (place of birth, education, etc.), employment (occupation, income, etc.), and housing (year moved to residence, year structure built, etc.). Both files are available on CD-ROM down to the block-group level. Also used was STF1B Extract, which has population and housing data for selected categories down to the block level.

As illustrated in Figure 2, Allegheny County consists of 130 municipalities, of which the city of Pittsburgh is one. Each municipality has a number of census tracts and each census tract is divided into block groups. There are 499 census tracts and 1,491 block groups in Allegheny County. Census tracts are small, relatively permanent statistical subdivisions. They usually have between 2,500 and 8,000 persons and, when first delineated, are designed to be relatively homogenous with respect to population characteristics, economic status, and living conditions. Thus, their areas vary widely depending on population density. Over time they may be split due to population growth or combined as a result of substantial population decline. Block groups are clusters of census blocks, which are the smallest units of analysis in the census data. Census blocks are small areas bounded on all sides by visible features such as roads, streams, and railroad tracks and by invisible boundaries such as city, town, and county limits. Block groups generally contain between 250 and 550 housing units.

Figure 2. Census Geography of Pittsburgh and Allegheny County

HAZARDOUS FACILITY DATA

Information on releases from manufacturing facilities was taken from the EPA’s Toxic Release Inventory (TRI) database for 1990. Several limitations of the TRI should be noted. First, the inventory does not include all facilities releasing toxic materials to the environment but only those facilities in SIC codes 22*-39 that are actively engaged in manufacturing activities. Other potentially large polluters, such as coal-fired power plants, solid and hazardous waste incinerators, mining operations, and federal facilities (e.g., weapons manufacturing facilities) are not required to report under the Act. Any facilities that cease operations are also not part of the inventory. Second, approximately 80 percent of all manufacturing facilities are not required to report, either because they employ less than 10 full-time employees or they use less than the threshold amounts of the 328 listed TRI chemicals. According to a recent report (GAO 1991), 95 percent of the toxic chemicals released in the U.S. are not included in the TRI. Third, the law requires facilities to report only those accidental spills or leaks that exceed the threshold amounts specified for each chemical. Fourth, routine releases are estimated, rather than measured, using standard EPA methods. Finally, all of the data are self- reported, which leaves considerable room for both interpretation and error, especially because the EPA has a limited capability to verify the reports.

Despite these limitations, the TRI is one of the most complete, comprehensive, and accessible databases available from EPA. Each year the inventory is analyzed extensively by industry, government, and environmental groups. Whereas some observers believe that the publicity focused on the “bad actors” has resulted in substantial reductions in toxic releases, others question the true magnitude of the impact of TRI and believe the apparent reductions can be attributed to changes in estimation methods, corrections of errors in previous reporting years, and a loophole in the law that allows companies not to report wastes shipped off-site to waste recyclers (Citizens Fund 1991).

While many of the problems with the inventory were ironed out during the first couple of years, the data still contain many errors and inaccuracies. When using the data in a GIS, a major problem is the inaccuracy or complete absence of latitude and longitude coordinates for facility locations (some coordinate pairs in the database even put the facilities in the wrong hemisphere!). Hence, substantial correction and verification of the coordinates had to be done before we could proceed with the analysis. The built-in address- matching function of AtlasGIS was of little help, not because of the limitations of the software but rather those of the data (i.e., inaccuracies in both the reported address data and in the street names and address ranges in the basic census files). Approximately half of the 88 TRI facilities in Allegheny County were successfully matched. The remaining ones had incomplete addresses or addresses that could not be matched with the basic census data (e.g., no such street was listed or the street number was outside the listed range). Unfortunately, even the successful address matches were of little use, because each match gave us only an approximate location of the facility on a particular segment of a street rather than accurate coordinates.

Our only recourse was to contact each establishment by telephone to verify the facility name and address. Obviously, this is a major limitation of the use of the TRI database at the national level, where such extensive verification would be impossible. Using the GIS in conjunction with a CD-ROM product called Street Atlas USA and local street maps, each facility was located on the computer screen by “walking” with the facility contact through a set of local landmarks (e.g., “the facility is located 200 yards from the corner of Smith Avenue and Charles Street”). Then AtlasGIS automatically assigned coordinates to each facility. This process was tedious, but there was little alternative because one cannot examine the relationship between facility location and the socioeconomic and demographic characteristics of the neighboring population without accurate locational data.

Information on the storage of acutely hazardous chemicals was taken from the 1992 Section 312 (Tier II) reports for Allegheny County. These reports, obtained from the Pennsylvania Department of Labor and Industry, Bureau of Right to Know, reveal that a total of 867 facilities stored a total of approximately 1,750 chemicals. Of these facilities, 176 reported at least one extremely hazardous substance (EHS). EPA considers these chemicals to be the ones most likely to have severe toxic effects on human beings exposed to an accidental release. On closer inspection, we found that some of the chemicals reported by these facilities were incorrectly marked as EHSs. After eliminating them, we ended up with only 128 facilities reporting a total of approximately 120 different EHSs.

PROXIMITY-BASED MEASUREMENTS

Calls for environmental justice usually center on racial minorities and the poor, who in many cases are the same people. They often live in areas that have more undesirable facilities than areas where other people live. And, unfortunately, their communities are more apt to have a disproportionate number of very young children and elderly people, who are especially susceptible to the health effects of pollution. The observation that racial minorities and the poor tend to live in closer proximity to environmental hazards than other people can be tested on a case-by-case basis in three steps as follows: (1) calculate their percentage of the population in the census areas that contain the facilities of concern, (2) do the same for the census areas that do not contain such facilities, and (3) compare the results.

There are several problems, however, with this simple approach. For one thing, no distinction is drawn between areas having only one facility and ones that have multiple facilities. Another problem is that the facility or facilities may be so close to the edge of the area that the neighboring area is affected as much, if not more, than the host area. And perhaps most important is the fact alluded to earlier that areas such as census tracts and counties, while convenient to use, are statistical or administrative constructs that do not generally represent either the affected neighborhood or the range of the hazard associated with a facility. A circle centered at the facility is more sensible, although the question of how large the radius should be is open to question. In the case of facilities in urban areas, a distance of a few miles seems a reasonable choice, because neighborhoods and serious environmental hazards do not usually extend any farther than that. The Appendix demonstrates in detail how the selection of different spatial units affects the evaluation of environmental equity.

We constructed circles with radii of a half-mile, one mile, and two miles around each of the facilities associated with the chronic (TRI) and acute (EHS) hazards of concern in Allegheny County. For each type of facility, we divided Allegheny County into two parts, one being the area formed by the circles and their overlapping portions and the other being the rest of the county. In each case, the combined area inside the circles, which is not entirely contiguous, is the “region” where people live in close proximity to the facilities. It is assumed implicitly that, for a given choice of radius, the region is homogeneous with regard to proximity effects, that is, it makes no difference which facility you are close to or how close you are to it as long as you live within the region.

For chronic hazards, Figure 3 shows the locations of the 88 facilities that submitted TRI reports in 1990. Each facility is centered within a half-mile radius circle, and the inset illustrates how these circles intersect the census block groups. Figures 4 and 5 show the same circles superimposed on the areas formed by the quartile of block groups that had the highest density of black residents and the highest density of poor people, that is, residents living below the poverty line. The proportions of black residents inside and outside the close-proximity region turn out to be 14 and 11 percent, respectively. These numbers increase to approximately 15 and 12 percent for all nonwhites. The proportions of poor residents inside and outside the close-proximity region turn out to be 16 and 11 percent, respectively.

Figure 3. TRI Facilities in Allegheny County, with Half-Mile Radius Circles

Figure 4. TRI Facilities and Areas with High Densities of Blacks

Figure 5. TRI Facilities and Areas with High Densities of Poor People

For acute hazards, Figure 6 shows the locations of the 62 facilities (out of 128) that stored more than a minimal quantity of EHSs in 1992, along with the one-mile radius circles centered at those locations and superimposed on the areas formed by the quartile of census block groups that had the highest density of nonwhites in 1990. Figure 7 does the same for the areas formed by the quartile of census block groups that had the highest density of poor people. We calculated the proportion of nonwhite residents inside and outside the close-proximity region to be 16 and 11 percent, respectively, and we calculated the proportion of poor residents inside and outside the region to be 16 and 10 percent, respectively. Thus, the percentage of nonwhites and the percentage of poor people among everyone who lives close to the facilities are higher than the corresponding percentages elsewhere in the county, indicating the existence of inequities. The comparable inside versus outside numbers when one-mile radius circles are centered at the TRI facilities are 15 and 11 percent for nonwhites and 15 and 10 percent for poor people. In other words, with respect to proximity to industrial facilities in Allegheny County, approximately the same levels of inequity exist for chronic hazards as for acute hazards.

Figure 6. EHS Facilities and Areas with High Densities of Nonwhites

Figure 7. EHS Facilities and Areas with High Densities of Poor People

RISK-BASED MEASUREMENTS

Risk-based equity measurements take much more information into account than proximity-based measurements. The risk depends not only on the proximity but also on these other major factors: the probability of a release accident; the size of the area impacted by the release (which depends in turn on the substance, the quantity released, the release rate, and the meteorological conditions); the wind direction at the time of release; the choice of a toxicity criterion and its critical value for the substance released; and the exposure of the population segment of concern. We defined risk as the expected annual number of persons exposed to release accidents, and we developed a risk assessment procedure that takes all the preceding factors into account, using a formula that multiplies the probability of an accidental release by the size of the impact area (often referred to as the “footprint of the plume”) and the population density in that area. This procedure takes into consideration the possibility that any person might be exposed to several such accidents in a year, thereby contributing several “person-exposures” to the annual total.

Because population exposure varies by time of day, so does risk. To estimate the nighttime population, we simply used residential census statistics. To estimate the daytime population, we adjusted the residential number in each block group using “journey-to-work” data that reflect the weekday comings and goings of commuters. We then used these results to calculate first the nighttime and daytime risks for nonwhites and the poor due to each EHS facility and then the total risk to nonwhites for each facility alone and for all facilities taken together. The same procedure was followed for the poor to obtain their total risks by facility and for the county as a whole. Any nonwhite people in the overlap between two areas were thus counted twice, which is appropriate since the total risk to any such person is essentially the sum of the two risks. The weighted average of the nighttime risks, which only take the residents of each area into account, and the daytime risks, which take into account the working population and the nonworking residential population in each area (both of which were determined from journey-to-work data), is the daily average risk. The same calculations were made for poor people.

Based on these measurements, equity for nonwhites (or the poor) is said to exist if their percentage of the total risk, that is, the annual number of nonwhite (or poor) persons exposed divided by the total number of persons exposed, is the same as the respective percentage of nonwhites (or the poor) among the entire county population. The percentage of nonwhites and poor people at risk from chemical release accidents are 9 and 8 percent, respectively. The corresponding percentages of nonwhites and poor people in the county population are 13 and 12 percent, indicating that these people actually bear proportionately less of the risk than they would if equity existed. In other words, in this county, for this environmental hazard, the inequity in risk works in favor of nonwhites and poor people.

At first, this finding comes as a surprise because environmental inequities are generally expected to be favorable to the white, more affluent majority, as demonstrated by our proximity-based estimates. Upon reflection, however, the reasons why the risk-based estimates go the other way are clear. First of all, this is an aggregate result obtained by combining the results for all the facilities where an EHS is stored. On a facility-by-facility basis, the direction of the inequity varies, sometimes working in favor of these population segments and sometimes against them. The aggregate result shows that, on balance, it worked against them more than it worked for them. Second, because the radius of the impact area of a release accident often exceeds one mile (ranging between 1.1 and 9.2 miles by night for 27 facilities and between 1.4 and 6.8 miles by day for 18 facilities), the demographics of the area tend to be different than the demographics of smaller areas. Thus, because nonwhites and poor people tend to live closest to such facilities, more whites and more relatively wealthier people will be impacted at larger radii.

In general, when the most disadvantaged segments of the population live closest to hazardous facilities, small hazards will tend to be most inequitable to those people, while larger hazards will tend to affect other people more. This phenomenon calls for caution in risk assessment, because it is commonly assumed that, in the name of conservatism and simplicity of analysis, worst- case assumptions should be used. However, this can introduce a bias when the population is distributed in this way, because the larger the hazard area, the greater is the impact on communities with relatively more whites or relatively higher incomes. The result is that the analysis tips the hazard burden balance in the direction of these communities because their share of the risk is larger than it would be under average-case assumptions.

Figure 8 gives an indication of the distribution of the nighttime and daytime risks attributable to the EHS facilities, where the shaded areas are formed by the 5 percent of the block groups that had the highest risks in each of the two periods. The differences between the night and day results show the importance of accounting separately for both periods, rather than for the nighttime alone. Common practice is to consider the nighttime only because, in that case, only residential census data are needed.

Figure 8. Highest Risk Areas due to EHS Facilities by Night and Day

We are currently still in the process of generating risk-based measurements of equity for the chronic hazards associated with air pollution from the TRI facilities in Allegheny County. This is a more time-consuming process because it requires that concentration contours representing county-wide pollution patterns first be modelled for criteria pollutants and air toxics, based on plant locations, stack heights, emission rates, pollutant characteristics, meteorological conditions, and other parameters. When imported to the GIS as a set of data layers, one for each pollutant, the concentration levels will be combined with the aforementioned estimates of population exposure levels to assess the associated risks to the population segments of concern based on the appropriate dose*-response relationship for each pollutant (i.e., the toxicity of the pollutant as a function of its concentration). These risk estimates, which will be expressed not just as person-exposures but as cases of cancer or other diseases, will be used in turn to evaluate equity. Equity will also be evaluated, to the extent possible, on the basis of the combined risks of accidents and air pollution, which means that the acute impacts of accidental injury or fatality and the chronic health effects of pollution exposure will have to be measured on a common scale, such as the total expected reduction in life expectancy.

FUTURE DEVELOPMENTS

While there is still a great deal of experimentation to be done and practice needed in using GIS to evaluate environmental equity, certain observations are safe to make. Given the widespread availability and relative ease of using census data and TRI data, and the increasing availability of user-friendly GIS packages, the capability to produce proximity-based estimates of industrial air pollution hazards is within the reach of many interested parties. Naturally, such estimates should not be considered the “last word” on environmental equity, because TRI data are self-reported (as are EHS data), TRI facilities are but one source of air pollution, and proximity is not a surrogate for risk. Other pollution sources and any environmental hazards of a nonpolluting nature or not related to health effects can readily be subjected to a proximity-based analysis as well, providing that the data are available, complete, and “clean” (GIS may be a new technology, but the oldest maxim in computing–“garbage in, garbage out”–still applies).

Risk-based analysis is another matter entirely for several reasons: risk assessment is still the province of technical specialists, it requires much more data than proximity measurement, and the results are more difficult to interpret. However, as more research of the kind we are conducting is done, as better risk assessment software becomes available, and as risk education and communication improve in general, these obstacles will become less formidable. In the meantime, research is also needed on combining risks, especially those that are difficult to measure in common units, such as carcinogenic and noncarcinogenic risks, and those that do not merely sum when accumulated, that is, health risks that are disproportionately exacerbated in the presence of certain other health risks. Evaluating the uncertainty in risk estimates is, as usual, another area where more research is needed.

When completed, the study we are conducting is expected to benefit a wide audience, ranging from community groups to professional peers, by demonstrating how to assemble the available data, how to analyze it using moderately-priced software, and how to interpret the results. The use of GIS to evaluate environmental equity will contribute objective information to local and national debates about environmental justice, helping to set priorities by identifying where the inequities are greatest, which subpopulations are most affected, and what hazard sources are most responsible. The approach we have used is most useful in measuring what been referred to as “outcome inequity,” that is, a situation where one socioeconomic group bears more of the hazard burden than another. In contrast, the part of our study that deals with the use of historical data to examine the evolution of inequities may also prove useful in the consideration of “procedural inequity,” wherein the institutions in society act in such as way as to be discriminatory from the viewpoint of environmental justice.

In the near future, the principal benefit of using GIS to measure environmental equity will probably be twofold. One is its value as a screening tool, that is, a capability for concerned parties such as public interest groups or government agencies to evaluate a region, as we did for Allegheny County, and determine which facility or facilities are detracting the most from a state of acceptable environmental equity for the region. The other is the contribution that GIS can make to the process of facility siting, where, ideally, all the stakeholders, whether industrial, governmental, or community-based, would participate in the process of identifying and evaluating the candidate sites for locating an undesirable facility with the assistance of a GIS. If at some future point in time an inventory of risk estimates could be developed for such regions, the facilities considered in the screening or siting process could be evaluated not only in terms of the absolute risk they pose to each population segment of concern, but also according to their relative contribution to the overall risk burden of the various socioeconomic groups involved.

Raising our sights a good deal higher, we observe that there is an overarching policy issue that should be confronted in the not too distant future: is it ultimately better for all parties concerned to spread out the region’s environmental hazards to achieve short-term equity–which would be the outcome of making piecemeal, relatively quick improvements in the status quo– or to concentrate them in one or more “hazard zones” and induce longer-term equity by reducing the associated risks and putting programs in place, if necessary, to enable nearby residents to relocate over time? This issue goes well beyond using GIS to evaluate environmental equity, although GIS could potentially be of use in conducting such a policy analysis.

ACKNOWLEDGEMENTS

The author is grateful to Robert Hersh of the Center for Risk Management (and a native of Pittsburgh) for his tireless participation in this research and to Dominic Golding, now of Clark University, for his invaluable contributions in the earlier stages of this project. Support for this effort has been provided by the Pew Charitable Trust, the Richard King Mellon Foundation, and the U.S. Environmental Protection Agency.

REFERENCES

Asch, P., and Seneca, J.J. 1978. Some Evidence on the Distribution of Air Quality. Land Economics 54:278-297.

Berry, B.J., ed. 1977. The Social Burdens of Environmental Pollution: A Comparative Metropolitan Data Source. Cambridge, MA: Ballinger.

Bullard, R.D. 1990. Dumping in Dixie: Race, Class, and Environmental Quality. Boulder, CO: Westview.

Citizens Fund. 1991. Poisons in Our Neighborhoods: Toxic Pollution in Pennsylvania. Washington, DC.

Freeman, A.M. 1972. The Distribution of Environmental Quality. In A.V. Kneese and R.M. Bower, eds., Environmental Quality Analysis. Baltimore, MD: Johns Hopkins Press/Resources for the Future. Kruvant, W.J. 1976. People, Energy, and Pollution. In D.K. Newman and D. Day, eds., The American Energy Consumer. Cambridge, MA: Ballinger. Mohai, P., and Bryant, B. 1992a. Race and the Incidence of Environmental Hazards: A Time for Discourse. Boulder, CO: Westview Press.

Mohai, P., and Bryant, B. 1992b. Race, Poverty, and the Environment: The Disadvantaged Face Greater Risks. EPA Journal 18(1):6-8.

Tarr, J.A. 1989. Infrastructure and City-Building in the Nineteenth and Twentieth Centuries. Pp. 213-264 in S.P. Hays, ed., City at the Point: Essays on the Social History of Pittsburgh. Pittsburgh, PA: University of Pittsburgh Press.

United Church of Christ. 1987. Toxic Waste and Race in the United States. New York, NY: UCC Commission for Racial Justice.

U.S. Environmental Protection Agency. 1992a. Environmental Equity: Reducing Risk for All Communities. Volume 1: Working Group Report to the Administrator. Document No. 230-DR-92-008. Washington, DC.

U.S. Environmental Protection Agency. 1992b. Environmental Equity: Reducing Risk for All Communities. Volume 2: Supporting Document. Document No. 230-R- 92-008A. Washington, DC.

U.S. General Accounting Office. 1983. Siting of Hazardous Waste Landfills and Their Correlation with Racial and Economic Status of Surrounding Communities. Washington, DC.

Zupan, J.M. 1973. The Distribution of Air Quality in the New York Region. Baltimore, MD: Johns Hopkins Press/Resources for the Future.

APPENDIX

A Comparison of the Results of Equity Evaluations Using Different Units of Analysis

(This material was adapted from the chapter “GIS-Based Environmental Equity Analysis,” in the forthcoming book Computer Supported Risk Management , edited by W.A. Wallace and E.G. Beroggi.)

INTRODUCTION

Many studies have demonstrated the existence of inequities in the distribution of noxious, toxic, or otherwise unwanted facilities, and many more studies are ongoing. Most of these studies have been conducted at a national or regional scale and involve comparing the socioeconomic and demographic characteristics of host and nonhost communities. The distribution of facilities is considered inequitable if the socioeconomic status is disproportionately lower or the percentage of minorities is disproportionately higher in the host communities.

Census tracts and counties are the most commonly used unit of analysis because data are readily available in these forms from the U.S. Census Bureau. Relatively little attention has been paid, however, to the problem that using data in such an aggregated form may mask much that is significant in analyses of environmental equity. Similarly, little has been done to determine what the most appropriate units of analysis are for these kinds of studies. By comparison, this investigation illustrates that the unit of analysis chosen can have profound effects on the results of an equity analysis and its interpretation. It also demonstrates that geographical information systems (GIS) offer a unique, flexible way to examine such effects and to devise more appropriate units of analysis.

BACKGROUND

Table A.1 summarizes the topics addressed, the scope, and the unit of analysis employed in several previous equity studies. Most of them focus on the sociodemographic distribution of one particular environmental problem, such as the siting of hazardous waste facilities (e.g., GAO 1983; UCC 1987; National Law Journal 1992), but some have tried to address a more diverse array of hazards (Goldman 1992; Mohai and Bryant 1992a; Nieves 1992). Many ongoing projects are mining the EPA’s Toxic Release Inventory (TRI) database, as we do here, and some are trying to tackle the problem of multiple hazards. Invariably, however, these studies use only one unit of analysis, such as census tracts, zip codes, or counties, to examine the problem at a local, regional, or national scale.

(The 1986 Emergency Planning and Community Right-to-Know Act requires certain manufacturers to report to the EPA the amounts of over 300 toxic chemicals that they release directly to air, water, or land or that they transport to off-site facilities. The law also requires EPA to compile these reports into an annual inventory of releases and transfers and make the inventory available in a computerized database. This compilation is called the Toxic Release Inventory.)

Relatively few studies have examined the question of what unit of analysis should be used in a given situation and what impact the choice of a particular unit of analysis has on the results of the analysis. As Zimmerman (1993, 652) notes, “The scale of analysis chosen is often dictated more by expediency, determined by how existing data bases are aggregated and which level of aggregation provides the most data at the smallest geographic scale.” The unit of analysis chosen, therefore, may bear little relation to the actual community affected and may severely distort the outcome of an equity analysis. The appropriate choice of a unit of analysis depends on the nature of the harm, ranging from risk narrowly conceived as potential human health effects to more nebulous community concerns such as stigmatization. Even if we con fine our attention to potential health risks, different kinds of hazards (e.g., waste sites, TRI facilities, incinerators, etc.) will affect populations at different distances. The point here is not to identify a single most appropriate unit of analysis for all hazards but rather to illustrate that the unit chosen can have a dramatic impact on the results of the analysis. As Zimmerman goes on to say, “[t]he potential variation in results at different geographic levels suggests a need to at least explore a number of scales simultaneously, and to conduct sensitivity analyses to ensure the implications for equity at different scales are not wildly different.” She then suggests that GIS can help address these problems by allowing the creation of user-defined units of analysis from basic census data.

This is precisely what we do in this study by exploring the impact of using different units of analysis to examine the relationship between (1) the location of manufacturing facilities releasing toxins into the air and (2) the socioeconomic and demographic characteristics of the host and nonhost communities in Allegheny County, Pennsylvania, which contains the city of Pittsburgh. The socioeconomic and demographic data are drawn from the 1990 Census of Population. The analysis is part of a larger project that uses GIS to examine the spatial, social, and economic distributions of a variety of environmental risks in Allegheny County.

EQUITY ANALYSIS

To explore the importance of the unit of analysis in this environmental equity study, we were only concerned about the location of TRI facilities in relation to the neighboring population characteristics. Therefore, we did not need to examine levels of toxicity, geographic dispersion of releases, or corresponding populations exposed, all of which are considerations that will be tackled in subsequent research.

Our primary concern was to determine whether there are higher proportions of poor and minorities in the communities that are host to TRI facilities compared with the communities in the same area that do not have such facilities. We used four different units of analysis to define “community”: block groups, census tracts, municipalities, and circles of half-mile radius centered on each TRI facility. The first three units of analysis are standard geographical divisions; the fourth was created using the GIS. We did not use census block data, the finest unit of analysis for census data, because several important variables (e.g., household income) are not available at that scale.

Because TRI facilities are not necessarily located in the geographic centers of block groups, census tracts, or municipalities, defining these spatial units as the communities likely to be affected or concerned may be misleading. Census units are often bounded by very real geographical features, such as roads, rails, and rivers, and the sociodemographic characteristics of these units may change substantially on crossing these boundaries. Thus, by using a circle of standard size around each facility we avoided the problem that arises when a facility is located within a predominantly white community but is near the border and actually closer to one or more adjacent black communities. The circular shape is also appropriate for a proximity analysis such as this one, which differs from a risk analysis in which a plume might be more appropriate because wind direction would be a factor. In the absence of additional information, we assume that the people most likely to be affected by and concerned about a facility are those who live in its immediate vicinity and that everyone in that vicinity is equally likely to be affected.

The choice of a half-mile radius results in a circular area of 0.8 sq. mi., which approximates the average area of a census tract (1.5 sq. mi.) or a block group (0.5 sq. mi.) in Allegheny County. Half a mile is a distance that intuitively connotes close proximity. Although it could be argued that a smaller radius, say a quarter of a mile, might better distinguish one neighborhood from another, this would introduce complications associated with such small circles. In some cases, the plant boundary would extend beyond the circle and in others, most of the circle would be occupied by a river. Choosing instead a larger radius than a half-mile would result in a circular area that is more consistent with the average size of a municipality in the county, which is 5.2 sq. mi. (not counting the 55 sq. mi. occupied by Pittsburgh). A circle with a 1.3-mile radius has about the same area. Each block group, census tract, and municipality was designated either as a TRI community, if it hosted one or more TRI facilities, or as a non-TRI community if it contained none. Table A.2 indicates the total number of TRI and non-TRI communities using each of the different units of analysis, and the respective values for a variety of socioeconomic and demographic variables. The values were generated by aggregating the separate numbers for each unit of analysis, not by taking the mean of their values. For example, we calculated the aggregated proportion of blacks in the TRI block groups by dividing the total number of blacks in the 67 individual TRI block groups by the total population of all races in those block groups. Thus, the value of 10 percent represents the true percentage of blacks, rather than the mean of the block group percentages. All the values in the Table were obtained in this fashion, except for household income. Calculating mean household income by dividing the aggregate household income for any set of block groups, census tracts, or municipalities by the corresponding number of households could be misleading because outliers would skew the result. Instead, we generated the household income frequency distributions in each case from the census data and then determined the median household income for TRI and non-TRI communities associated with block groups, census tracts, and municipalities.

TABLE A.2. Comparison of the Characteristics of TRI and Non-TRI Communities for Different Units of Analysis

Deriving values for the half-mile radius circles was quite a different kind of effort, involving the full use of the GIS. Circles of this radius were drawn around each of the 88 TRI facilities and the boundaries of overlapping circles were “dissolved” to form a single layer representing the combined TRI communities. (As indicated in Figure 3, not all the circles overlap with other circles, and several remain disconnected areas (i.e., the union of all the circles is spatially discontinuous)). These areas are still incorporated in the aggregated numbers.The residual non-TRI area then refers to all of Allegheny County outside the union of the circles. To calculate the proportion of blacks, for example, living within the combined TRI communities, the numbers of blacks living in each whole or partial block group were aggregated. The numbers for block groups falling partially within the union of the circles were weighted according to the relative proportion of each block group’s area inside the union. These weighted values were then added to the values for all the whole block groups to obtain the total number of blacks living within the union of the circles. A similar computation was performed for the total population. The mean percentage of blacks is thus the total number of blacks divided by the total population of all races. Similarly, in calculating median income, the number of households in the block group or portion of the block group within the union of the circles was used as a weight which was then applied to the median household income for that block group. The values given in the shaded part of Table A.2 refer to the aggregated area or union of all the circles for all variables. Because the values in the Table are based on subsets, not on samples, of the entire population of Allegheny County, there was no need to conduct tests of significance. The differences in the percentages are by definition statistically significant. The set of variables was chosen to allow us to explore whether TRI communities are more likely to comprise a higher proportion of minorities and poor people than non-TRI communities. The variables chosen also allowed us to examine the notion of sensitive populations. For example, it is often assumed that the elderly (defined here as 65 years old or older) and the young (defined here as 5 years old or younger) are biologically more susceptible to certain pollutants or are likely to be more exposed through inactivity (e.g., older people being confined to polluted neighborhoods for extended periods) or specific activities (e.g., children playing outside).

RESULTS OF THE ANALYSIS

Table A.2 demonstrates that the choice of unit of analysis will affect even the most basic findings of an environmental equity study. Had we used only block groups to define “community,” we would have found contrary to expectations that in TRI communities the proportion of blacks (10 percent) and minorities (11 percent) is slightly lower than in non-TRI communities (11 percent for blacks and 13 percent for minorities).+ Similar results hold for census tracts. This pattern is reversed, however, when we look at the proportions for the combined half-mile radius circles around TRI facilities versus the area beyond the circles (14 versus 11 percent for blacks and 15 versus 12 percent for minorities). We also see that the proportion of blacks and minorities is substantially higher in municipalities with TRI facilities than in those without such facilities (15 versus 6 percent for blacks and 18 versus 7 percent for minorities), a result that is heavily influenced by the inclusion of Pittsburgh, which has a large number of TRI facilities and a large population of blacks and minorities. Comparing the TRI municipalities other than Pittsburgh to the non-TRI municipalities demonstrates the relatively small size of the black and other minority populations beyond the city limits.

The U.S. Census of Population classifies people as white; black; American Indian, Eskimo, Aleut; Asian or Pacific Islander; and other. Hispanic is an ethnic, not racial, classification. Thus, there are white Hispanics, black Hispanics, etc. Consistent with other equity studies (e.g., UCC 1987), and in order to include Hispanics as minorities, we defined minorities to be the summation of the following population groups: Blacks not of Spanish origin; Asian and Pacific Islanders, American Indians, Eskimos, and Aleuts not of Spanish origin; other nonwhite populations not of Spanish origins; and Hispanics. Given the particular concern about blacks in previous equity studies, and the fact that there are few other minorities in Pittsburgh, we also assessed blacks as a separate category for analysis.

We have several direct and indirect measures of poverty in the next two sections of Table A.2. The direct measures of poverty are median household income and the number of persons with family incomes below the poverty level. We find that median household income is lower in TRI communities than in non-TRI communities across the board, regardless of the unit of analysis chosen. This is in accordance with most previous equity analyses but contrary to some studies that have found higher income levels in proximity to industrial facilities (e.g., Napton and Day 1992). Median black household income is also lower in TRI communities than in non-TRI communities, whether they are based on block groups ($14,746 versus $14,811, an insubstantial difference), census tracts ($11,469 versus $15,105) or circles ($13,043 versus $15,077). Median black household income is considerably lower in TRI municipalities than in non-TRI municipalities ($12,883 versus $23,909), a very substantial difference which is probably due to the fact that there are relatively few blacks living in the more affluent, less industrial non-TRI municipalities and that those who do are likely to be more affluent middle-class blacks.

The proportion of “poor” persons, defined as the number of persons living below the poverty level ($12,674 for a family of four) divided by the number of persons in the “adjusted total population,” is higher in TRI communities when three of the four units of analysis are used but lower in TRI block groups (10 percent) than in non-TRI block groups (12 percent). (The “adjusted total population” is the total population less those individuals held in institutions such as prisons and psychiatric hospitals.) Using municipalities as the unit of analysis gives the largest differential (15 versus 7 percent), but this result is heavily influenced by the large proportion of Pittsburgh residents living below the poverty level (21 percent). The pattern is similar when we look at the proportion of poor whites and the proportion of poor blacks (i.e., the number of whites or blacks living below the poverty level as a proportion of the adjusted total population in the county). Curiously, the pattern changes for census tracts when we examine the number of poor blacks as a proportion of all poor people (33 percent for TRI versus 35 percent for non-TRI), and it changes for block groups when we examine the number of poor blacks as a proportion of all blacks (41 percent for TRI versus 36 percent for non-TRI). These changes reflect an instability associated with using block groups or census tracts as the unit of analysis. Whichever unit of analysis is used, we see that among all blacks a greater proportion live in poverty in TRI communities than in non-TRI communities.

The remainder of the Table deals with other socioeconomic and demographic variables. The use of circles as the unit of analysis shows consistently that compared to the non-TRI area, the combined TRI area had (1) a higher proportion (20 versus 17 percent) of elderly residents, who are generally considered to be more sensitive to pollutants than the general population; (2) higher unemployment in total (8 versus 6 percent), both for whites alone and for blacks alone; (3) a higher incidence of single parents (10 versus 7 percent) and black single parents; and (4) a higher proportion of vacant housing units (9 versus 7 percent).

Part (a) of Figure A.1 displays in chart form selected results from the half-mile circle columns in Table A.2. The purpose is to show more clearly the magnitudes of the inequities, according to race, income, age, and race and income combined. For instance, it presents the percentage of blacks in the combined TRI communities and in the non-TRI area as solid squares separated by a dotted gap. The larger the gap, the greater the inequity. Based on this unit of analysis, it is evident that the inequity is greatest for the poor. Note that when a gap of a given length is situated further to the right, the associated inequity is less serious in a relative sense. For example, whereas the entries for percent poor whites and percent over age 65 span the same length, the relative difference is (10 – 7)/7½ 43 percent for the former but only (20 – 17)/17 = 18 percent for the latter. Part (b) of Figure A.1, which is based on one-mile radius circles, generally shows the same kinds of inequities as for half-mile radius circles, except that the absolute magnitudes of the differences tend to be larger. In other words, the two sets of communities (TRI and non-TRI) are less similar when the larger circles are used. This means that there is more inequity if the undesirable effects of the TRI facilities are assumed to extend out to one-mile than if they only extend out only half as far, because there is more of a difference in the social and economic characteristics of the communities inside of and outside of the area formed by the larger circles than there is when the area is formed by the smaller circles. The largest discrepancy between the half-mile and one-mile radius results is in the difference between the TRI and non-TRI percentages of poor blacks among all blacks. Using a half-mile radius reveals no inequity for this segment of population, whereas with a one-mile radius, we observe an inequity of magnitude 38 – 31 = 7 percent between the TRI and non-TRI communities.

CONCLUSIONS

The foregoing discussion clearly demonstrates the importance of choosing the spatial unit of analysis very carefully. In past studies of environmental equity and related issues, expediency has often determined the choice–select the smallest, most easily obtained and easiest to use unit of analysis for which data are readily available. Obviously, data availability and ease of use are important selection criteria, especially when the analysis involves large amounts of regional or national data. Nevertheless, with the advent of relatively inexpensive and user-friendly GIS software, the forms in which data are ordinarily available need no longer drive the analytical design. It is now possible for the analyst to create user-defined units of analysis that are more appropriate to the problem at hand, rather than be constrained by the standard census divisions. In doing so, however, the analyst must also recognize that this manipulation necessarily entails tradeoffs in accuracy and generates its own uncertainties (Zimmerman 1993).

Figure A.1 Equity-Related Differences between TRI and Non-TRI Communities

The appropriateness of a particular unit of analysis will vary according to the nature of the problem. In examining issues of environmental equity, we envisage four principal types of units of analysis. First, the unit of analysis could be based on health risk criteria, in which case its size might depend on the distance at which a particular toxic release poses a significant risk to the surrounding population. Unfortunately, there are many different kinds and sizes of releases, depending on the different chemicals and related factors, and there are various exposure pathways (air, food, drinking water, etc.). Each different kind of release might require a different unit of analysis, which could easily become unmanageable. Alternatively, one could use a single unit of analysis as a compromise for all kinds of releases (e.g., the area around a facility that might be affected by a worst-case release). Second, the unit of analysis could be based on “perceived” risk rather than “objective” risk. Based on survey data, it might be possible to identify the levels of concern among the public at different distances from particular facilities and establish units of analysis accordingly. Again, this poses analytical problems given the multiplicity of different kinds of facilities and public attitudes about them, even in a relatively small region such as Allegheny County. Third, local political jurisdictions may be an appropriate unit of analysis if the concern is how to deal with environmental problems, because it is likely that local government will be responsible for monitoring and regulating the offending facilities. Fourth, it could be argued that the unit of analysis should reflect more closely the boundaries of individual communities or neighborhoods as defined in social and economic terms.

It is obviously easier to devise workable units of analysis for some of these criteria than others. For most of these criteria, the data do not presently exist to allow the construction of operational units of analysis. The point, however, is to recognize that existing census geography may be poorly suited to the analysis of many problems and that GIS offers a way to overcome some of the limitations. Even in the absence of the kinds of data noted above, it is still possible to tailor the unit of analysis to fit the problem more appropriately.

Finally, we would like to summarize the implications of this research for those interested in environmental equity. First, examine closely the selection and use of the units of analysis in any equity study, because these can have far-reaching impacts on the results. Second, when conducting your own study choose the units of analysis carefully, bearing in mind the caveats above. Third, use a variety of units of analysis and conduct sensitivity analyses (as we intend to continue to do) to examine the impact of the choice on the results. Such sensitivity analyses may show that the choice of the unit of analysis has little impact on the results and that any of the selected units may be appropriate. If the sensitivity analysis shows large differences according to the unit chosen, then further analysis is necessary to examine why and to determine the most appropriate unit or units given the nature of the problem at hand. Overall, in this analysis we found that the use of block groups appears to give the greatest number of potentially misleading results, a finding that may not be generalizable. Because our primary concern here was proximity, we believe that the circle was generally the best unit of analysis. Not addressed by this study are the fundamental questions of which of the calculated differences represent the most important inequities, how they could be eliminated, and whether these results are representative of other metropolitan areas.

ADDITIONAL REFERENCES FOR THE APPENDIX

Goldman, B.A. 1992. The Truth About Where You Live: An Atlas for Action on Toxins and Mortality. New York, NY: Times Books/Random House.

Napton, M.L., and Day, F.A. 1992. Polluted Neighborhoods in Texas: Who Lives There? Environment and Behavior 24(4):508-526.

National Law Journal. 1992. Unequal Protection: The Racial Divide in Environmental Law. National Law Journal, Monday, September 21:S1-S11.

Nieves, L.A. 1992. Not in Whose Backyard? Minority Population Concentrations and Noxious Facility Sites. Presented at the American Academy for the Advancement of Science Meetings, Chicago, IL, February 9, 1992.

Zimmerman, R. 1993. Social Equity and Environmental Risk. Risk Analysis 13(6): 649-666.