Toxic Releases and Demography in Minneapolis/St.Paul: A GIS Exploration

This study compared race, age, and income in areas that were near or far from airborne toxic releases (using EPA Toxic Release Inventory data). Since toxins disperse variously according to what toxin is being releases, how much of the toxin is releases, and the atmospheric conditions present at the time of release, a variety of buffers were used to measure nearness. Buffers were either fixed distances or variable according to the amount of toxin released. A GIS overlay was performed that intersected demographic data with the buffers and aggregated that data inside and outside the buffers. Statistical tests were performed that calculated whether or not the numbers of toxic releases were significantly different according to whether people lived inside or outside the buffers. Various maps were created to help communicate results and explain spatial relationships. Exploring the relationships between toxic releases and waterways, railroads, and highways helps explain why toxic releases occur where they do and suggests that an historical development of industrial patterns is necessary to explain spatial patterns between toxic releases and race, age, and income.

Objective

This research addresses the issues of environmental equity in the Minneapolis/St. Paul metropolitan area by investigating whether environmental quality varies with race, age, and income. Past research has used a variety of data and methods to address these issues in locations other than the Twin Cities, and has usually found significant relationships between environmental quality and race; sometimes between environmental quality and income; and age seems not to have been widely studied. This study uses the Environmental Protection Agency's Toxic Release Inventory (TRI) data as a measure of environmental quality. It uses census block groups for demographic data. This study uses both traditional statistical methods and those of GIS, geographic information systems, to analyze relationships between environmental quality and demography.

The Problem

Are minorities, the poor, the young or the old exposed to more toxins and pollutants than the general public? Such questions are termed "environmental equity" or "environmental justice." There are several dimensions to these problems. There is a geographic dimension that investigates relationships among the spatial distributions of demography and environmental quality. Do minorities, the poor, the young, or old live nearer to pollutants and toxins than the general public? There is a medical and toxicological dimension. Even if such demographic groups live near to pollutants and toxins, is there evidence that their health is impaired and that the cause is proximity to pollutants? There is a legal dimension to problems of environmental equity. Does environmental legislation protect all groups equally, for example, does it protect wilderness more than the urban environment to which minorities and the poor are exposed? Is the workplace fairly protected, or are low-income groups more likely to have a job that has environmental hazards (Lazarus, 1993)? There are important historical, political, and economic dimensions to problems of environmental equity. Is a hazardous facility located in a certain area because neighboring groups are poor, powerless, and want the jobs despite the health risks? Or did the location of a cluster of industries eventually lead higher income groups to leave an area, being replaced at a later time by minorities and lower income groups (Westra and Wenz, 1995)? Problems of environmental equity are complex and multi-faceted.

Previous Research

Previous research has used various ways to measure environmental quality, such as proximity to waste incinerators, toxic waste storage sites, and toxic releases. The research has studied various parts of the country, used different statistical methods and employed various geographical units for demography, such as tracts, groups of tracts, zip codes, and counties. While most studies have found significant relations among environmental quality and race and income, with such a variety of data, methods, and places studied, it is perhaps not surprising that all of these studies do not yield consistent results.

Several studies have as their chief focus the question of whether or not there are relationships among environmental quality and race and income, especially addressing the question of whether race or income is the more important factor. The U.S. General Accounting Office (GAO, 1983) performed an early study in the southeastern U.S. using landfill sites as a measure of environmental quality, finding that the majority of population surrounding three of the four sites was black. Bullard (1990) used waste incinerators to measure environmental quality in Texas, finding that the locations of waste incinerators are related to minority and low income people. In West Virginia, Louisiana, and Alabama positive correlations were again found between environmental quality and demography (Bullard, 1990; Wigley, 1995). Bullard repeatedly finds that race is more important than income in issues of environmental equity (Bullard, 1990, 1993, 1995). He states that "an abundance of documentation shows blacks, lower-income groups, and working-class persons are subjected to a disproportionally large amount of pollution (1990)." The United Church of Christ used toxic waste storage sites (i.e. TSDF, or treatment, storage, and disposal facilities) to measure environmental quality in a national study and again found race to be more significant than income, (UCC, 1987; Lee, 1992). Discriminant analysis in that study showed that the race was a stronger predictor of the level of TSDF activity than income, home value, the number of toxic waste sites, or the estimated amount of waste generated by industry, even when controlled for differences in regions and urbanization. Some researchers have not found race to be more significant (e.g. Burke, studying Los Angeles, 1996). Glickman (1994) used the presence of a TRI site as being in or not-in census tracts and found that race and income had about the same relationship to toxic releases. When he used data on storage of large amounts of hazardous substances (Extremely Hazardous Substances) in relation to census tracts, Glickman found that non-white and poor people bear less health risk from hazardous substances than white and more affluent people. Mohai and Bryant (1992) compared results of 15 studies analyzing the relationship between environmental hazards, on the one hand, and income and race on the other. These studies used a variety of measures of environmental quality (predominantly air pollution) in a variety of regions, mostly urban. Almost all found that environmental hazards correlated with low income and race. Most studies found race to be the better predictor.

Some recent studies have addressed the general questions of environmental equity and also investigated questions of appropriate units of geographic analysis. Anderton (1994) conducted an environmental equity study using 32,000 census tracts that had a commercial TSDF present in or near the tract. His conclusions were that results of any study depend on the geographic unit of analysis used, whether it is census blocks, tracts, or counties. In studies investigating the relationships between commercial TSDFs and demography, significant race differences are found when larger spatial units are used. The study also found that the distance of toxic diffusion is important and affects statistical results. Anderton suggested using TRI data in future studies. Bowen, et. al. (1995) recently used TRI data in a study for Ohio and Cleveland. For the Ohio study, they used counties as geographic units, and for Cleveland they used census tracts. The study analyzed how TRI data is related to population density, minority, and income. TRI data is used both as pounds and weighted for toxicity (by Threshold Limit Values). Relations are sometimes more significant for pounds than toxicity-weighted releases. Bowen's results vary with geographic units used, sometimes finding significant relationships with environmental quality and minorities and income, and sometimes not. McMaster (1990) suggests using block level data and some way to account for various toxicity of different materials and the fact that different materials disperse differently. While there is no obviously correct way to determine spatial units (Openshaw, 1983), the general insight is that smaller spatial units are more desirable because they have less data aggregation.

Overview of This Study

This study compared race, age, and income in areas that were near or far from toxic releases in the Twin Cities metropolitan area. Census block groups were chosen as the spatial unit of analysis. Air releases of toxins were used as the measure of environmental quality. GIS was employed to find areas near to toxic releases and buffers were created that were sized either by fixed distances or with a size that varied according to the amount of toxic releases (air releases, in lbs.). A GIS polygon-on-polygon overlay was performed to calculate demographic data according to whether or not it was inside or outside the buffers. Maps were made to show the spatial relationships between toxic releases and race, age, and income. Descriptive statistics were calculated to compare demography inside and outside the buffers. The demographic data was analyzed inside and outside the buffers, using the Kruskal-Wallis test. Both descriptive statistics and statistical analysis were calculated for each of a variety of buffer distances.

In studying this problem of environmental justice, no inferences are to be made about individuals from aggregate or grouped geographic data. It would be unwarranted to infer from this study that any individual living near toxic releases has health problems, nor is it warranted to infer that chemical releases cause any individual's health problems. These would be ecological fallacies (Langbein, et. al. 1978; Martin, 1991).

Data

Demographic data was as follows. Census block groups were chosen for this project because in general, smaller geographic units have less aggregation so should more accurately reflect people's characteristics. Demographic variables of race, age, and income were used. Race was computed as 100 - the percent of non-white. Age was calculated for young people as the percent of people aged 0 to 5, and for old people as the percent aged 65 or more. Median household income was used as the measure of income.

For measures of environmental quality, the study used the Environmental Protection Agency's Toxic Release Inventory data (TRI) for 1990. TRI data is collected as toxins released into the air, water, soil, or transferred to another site. Only toxins released into the atmosphere were used in this study, since releasing a toxin into a water waste pipe or transferring a toxin to another site may have no effect on the surrounding population, or an effect quite different than an air release.

The accuracy of toxic release locations is a major concern. Each facility that releases a toxin is required to report the latitude and longitude of that release to the E.P.A. Much of this data is inaccurate or missing. A study of New York state TRI data revealed the need for correcting the locations of toxic releases (New York State Parks Management and Research Institute, 1993). Table 1 shows how extensively the locations in New York had to be changed. The study states that "risk assessment using a GIS using unverified or inaccurate facility locations could be severe" (op. cit, p. 1).

Analysis

The data was analyzed in the following way. The general idea was to create buffers around the toxic releases, intersect the buffers and the census block groups, then aggregate the data inside and outside the buffers and compare the results.

However, real problems arise in estimating how big to make the buffers. How far are people affected by toxic releases? How do the diffusion of various chemicals differ? How do GIS buffers differentiate between one toxin released at one site and 43 toxins released at a different site? Ideally, empirically-based toxicological diffusion models would be applied for each toxin or group of toxins, that specified meaningful distances within which real health threats exist. For example, given a VOC like toluene and a certain wind direction and velocity, a certain probability function approximates empirically-measured diffusion, and toxicological knowledge shows that people's health is affected at such- and-such distances. Of course, such models would be different for a heavy metal like selenium, so many such models would be needed. Without such models, one might try various buffer distances and compare results. If the resultant demography differed radically according to buffer distances, little could be inferred in the absence of empirical diffusion models. However, if results were similar for different buffer distances, one might have more confidence in the results. That approach was taken here. Four buffers were created (see Table 3).

Following buffer creation, a polygon-on-polygon overlay was performed which intersected census block groups with the buffers and aggregated the data inside and outside the buffers. Descriptive statistics and statistical tests were performed that calculated whether or not the numbers of toxic releases were significantly different according to whether people lived inside or outside the buffers. Maps were created to show the spatial relations of toxic releases and demography.

Results

The results help answer questions of whether or not environmental quality varies with race, age, or income in the Twin Cities metropolitan area. Figure 2 shows the general distribution of toxic releases in the seven-county Twin City area (and the fixed 500 m buffers surrounding the toxic releases). Figure 3 shows the spatial relationship of toxic releases and percent minority in a sample area of North Minneapolis, where the relationship of toxic releases and percent minority is strong. (Figure 3 also shows the small, variable-distance buffers). Figure 4 maps toxic releases and income in a sample area of St. Paul, where that relationship is quite evident. (Figure 4 also portrays the large, variable-distance buffers).

Descriptive statistics are given in the top half of Table 4 and inferential statistics in Table 5. An appropriate statistical test is the Kruskal-Wallis H test because in- buffer and out-of-buffer values are binary and can be used to divide the other variables (percent minority, percent young people, percent old, and median household income) into two groups. A non-parametric test is appropriate because both the percent minority and the income variables are skewed. The null hypotheses assumed by these tests is that no differences exist in percent minority, percent young, percent old, and median household income according to their location of being in-buffer or out-of-buffer. The H statistic is distributed as Chi-squared . The significance of Chi-squared gives the probability of there being a real difference between being in-buffer and out-of-buffer, for example, in the percent minority, when in fact the null hypotheses are true. That is, the significance of Chi-squared gives the probability that the values shown in Table 4 could happen by accident, an artifice of the data, a random event. Table 5 shows statistical results.

Table 5 offers the following conclusions. Examining the descriptive statistics, it is apparent that the median percent minority population and the median household income are always higher inside the buffers, by a sizable margin, no matter what size the buffers are. This result is reinforced by the inferential statistics because those two variables are the only consistently significant ones, regardless of buffer size.

The median percent of young people (age under five) is always lower near to toxic releases, regardless of buffer size. This, combined with the fact that population is generally low near to toxic releases, results in the fortunate fact that large numbers of young people are not in close proximity to toxic releases. The statistical significance of the percent of young people varies with buffer size in no systematic way for this data set. The percent old (over age 65) is usually higher near to toxic releases, but is significant in only one of four buffer sizes.

What does explain the location of toxic releases? A finding of Anderton (1994) is that the percentage of employment in manufacturing and industry is significant. This idea seems reinforced by Figure 5, in which a very close relationship between toxic releases and railroads is quite apparent. (Figure 5 also explains the linear patterns of toxic releases that is apparent in Figure 2). A finding of Westra and Wenz (1995) addresses the need for an historical interpretation. Understanding the relationship of environmental quality and demography no doubt requires an economic history and is only suggested here. Industry first located in the Twin Cities along rivers. As rail transportation grew in importance, industry relocated there and many industries remain in those locations. As the highway system rose in prominence, industry increasingly found a new place to locate, beyond the edge of current suburbs where land was inexpensive, the workforce was nearby, highways provided adequate transportation, and there was little environmental liability from past land use. In the cities, as workers living near industries began to build wealth, they moved to suburbs, leaving behind housing that was affordable for lower-income people.

The important results of this study are that minorities and low-income people are closer to toxic releases. The percent of young people near toxic releases is somewhat significant, and relationship between old people and proximity to toxic releases is not usually significant. Buffer sizes have little effect on the statistical significance of variables when relationships between toxic releases and demographic variables are strong. In other words, for the percent minority and income variables, the data in this and other studies shows such a strong relationship to environmental quality that buffer size does not affect statistical significance. For variables whose relationship is no so obvious, such as age variables, buffer size does affect the statistical significance of results.

Limitations and recommendations

The data used in this study is true of census block groups, so no conclusions can be drawn about individuals. It would be a mistake to infer that a person is ill simply because they live near a toxic release.

This study has no implications for causal connections between toxins and human health. A medical/toxicological assessment would be necessary to make a case that proximity to toxins causes health problems. The assumption that proximity to toxins directly relates to absorption of those toxins is certainly not established here.

Variable-distance buffering at least addresses the problem that a facility that releases 1 lb. of a toxin is different from a facility that releases 1,200,000 lbs, yet problems with these buffers remain. That a heavy metal disperses differently than a volatile organic compound is not accounted for by the buffers used in this study. Buffers should represent an empirically-based diffusion model that incorporates the type of chemical and its dispersion rates. (Waller, et. al., (1995) have suggested Bayes methods for weighting distances from toxic releases). Yearly wind direction and velocity averages for a particular area could also be incorporated into variable distance buffers that were weighted by toxicity.

Yet another assumption of this and other studies that use TRI data is that release of toxins is consistent over time. The release of a toxin all at once could poison and kill thousands of nearby residents but the gradual release of the same quantity over the whole year may have no ill health effects at all.

It is imperative for future studies to correct the locations of TRI data. As shown in the New York study and this one, the accuracy of TRI locations is poor and analysis using the raw location data is unlikely to be reliable.

An economic history is necessary to understand the locations of toxic releases. A one-time shapshot gives valuable information but is enhanced by understanding how industries located where they did, how they changed, and what demographic groups were present over time. Measures of industrial production or industrial employment in certain industries (e.g. metals and oil refining, chemicals, leather and textiles, other heavy and precision manufacturing) might reveal quite a lot about the locations of toxic releases.

Footnote

acknowledgements

The author thanks the Wallace Atwood family and the Association of American Geographers for their assistance in supporting this research, through the Wallace Atwood Research Fund, and the National Science Foundation, grant DUE9451413.

The author also thanks Jennifer Lee, Carlson Companies, for her work correcting TRI locations, and Heather Petry, Minnesota DNR, for her cartographic assistance.

References

Anderton, Douglas L., Anderson, Andy B., Oakes, John Michael, and Fraser, Michael R. 1994. "Environmental Equity: The Demographics of Dumping". Demography, Vol. 31, No. 2; 229-248.

Bowen, William M. , Salling, Mark J. Haynes, Kingsley E. and Cyran, Ellen J. 1995 "Toward Environmental Justice: Spatial Equity in Ohio and Cleveland. Annals of the Association of American Geographers, Vol. 85, No. 4, 641-643.

Bullard, Robert D. 1995. "Decision Making." pg. 3-28 in Laura Westra and Peter S. Wenz (eds) Faces of Environmental Racism: Confronting Issues of Global Justice. Rowman & Littlefield Publishers, Inc., Lanham, Maryland.

Bullard, Robert D. 1990. Dumping in Dixie: Race, Class, and Environmental Quality. Westview Press, Boulder, Colorado.

Bullard, Robert D. 1994. "Grassroots Flowering: The Environmental Justice Movement Comes of Age". The Amicus Journal, Spring: 32-37.

Burke, Lauretta M. 1996. "Race and Environmental Equity: A Geographic Analysis". Geo Info Systems, vol 3 no. 9, pp. 44-50.

Commission for Racial Justice, United Church of Christ. 1987. Toxic Wastes and Race in the United States: A National Report on the Racial and Socio-Economic Characteristics of Communities with Hazardous Waste Sites. New York.

Glickman, Theodore S. 1994. "Measuring Environmental Equity with Geographical Information Systems". Resources, Summer 1994, pp. 1-6.

Langbein, Laura Irwin, and Lichtman, Allan J. 1978. Ecological Inference. Sage Publications, Beverly Hills.

Lee, Charles. 1992. "Toxic Waste and Race in the United States", in Race and the Incidence of Environmental Hazards: A Time for Discourse. Bryant, Bunyan and Mohai, Paul (eds). Westveiw Press, Boulder.

Lazarus, Richard J. 1993. "Pursuing "Environmental Justice": The Distributional Effects of Environmental Protection", Northwestern University Law Review, vol. 87, no. 3. pp. 787-855.

Martin, David. 1991. Geographic Information Systems and their Socioeconomic Applications. Routledge, New York.

McMaster, Robert B. 1990. "Modeling community vulnerability to hazardous materials using geographic information systems", in Introductory Readings in Geographic Information Systems, Peuquet, Donna J. and Marble, Duane, F, eds. Taylor and Francis. New York.

Mohai, Paul, and Bryant, Bunyan. 1992. "Environmental Racism: Reviewing the Evidence", in Race and the Incidence of Environmental Hazards: A Time for Discourse. Bryant, Bunyan and Mohai, Paul (eds). Westveiw Press, Boulder.

New York State Parks Management and Research Institute. 1993. "Locational Verification and Correction of 1989 New York State TRI Facilities", Final Report, a report prepared for the U.S. Environmental Protection Agency, Office of Toxic Substances.

Openshaw, Stan. 1983. "The Modifiable Areal Unit Problem". In Concepts and Techniques in Modern Geography, no. 38. Geo Books, Norwich, England.

Waller, Lance A, Louis, Thomas A. and Carlin, Bradley P. "Bayes and Empirical Bayes Methods to Assess Environmental Justice", a paper presented at the annual meeting of the American Statistical Association, Orlando, Florida, August 13-17, 1995.

Wenz, Peter S. 1988. Environmental Justice. State University of New York Press, Albany, New York.

Westra, Laura, and Wenz, Peter S. (eds). 1995. Faces of Environmental Racism: Confronting Issues of Global Justice. Rowman and Littlefield, Lanham, Maryland.

Wigley, Daniel C. and Shrader-Frechette. 1995. "Consent, Equity, and Environmental Justice: A Louisiana Case Study." pg. 135-159 in Laura Westra and Peter S. Wenz (eds) Faces of Environmental Racism: Confronting Issues of Global Justice. Rowman & Littlefield Publishers, Inc., Lanham, Maryland.