The Evolution of Empirical Environmental Justice Research Methods: A Call for Greater Use of Geographically Weighted Regression

Introduction

Evidence from hundreds of studies suggests that minority and low-income communities suffer disproportionate exposures to, and disparate health effects from, environmental hazards. ¹ The methods and tools environmental justice (EJ) researchers have used to investigate claims of environmental injustices have evolved over time. This article explores the evolution of empirical methods for identifying potential environmental justice communities (EJCs) and describes an opportunity to use geographically weighted regression (GWR) to supplement first-level screenings for potential EJCs.

Many early studies demonstrated that minority and low-income populations were disproportionately exposed to environmental hazards.^2,3,4 These studies took a unit-hazard coincidence approach to identify potential EJCs with the hypothesis that environmental hazards were more likely to be located in minority neighborhoods. With this method, the demographics of the population within the affected geographical “unit” (e.g., zip code and census tract) are compared with the demographics of other units that do not contain the hazard. In his 1983 investigation of hazardous facility locations in Houston, Robert D. Bullard used unit-hazard coincidence techniques to investigate the racial/ethnic composition of neighborhoods near hazardous waste disposal sites using U.S. Census Bureau demographic data. He found that waste incinerators and landfills were overwhelmingly located in Black communities or near predominately Black schools. ⁵

Unit-hazard coincidence was also the chief methodology in another important and early study: the 1983 U.S. General Accounting Office ⁶ report Siting of Hazardous Waste Landfills and their Correlation with Racial and Economic Status of Surrounding Communities. The GAO report used unit-hazard coincidence techniques to compare the demographics of four communities (approximated by census tracts) containing major hazardous waste facilities in the South to nearby communities. All communities containing the hazardous waste facilities were disproportionately Black, and three of the four were predominately Black. Among these communities, 26% to 42% of families lived below the poverty line, and 90% to 100% of them were Black.

Other early studies applied more sophisticated statistical methodologies, including ordinary least-squares (OLS) regression and other advanced multivariate statistical approaches. The United Church of Christ Commission for Racial Justice's 1987 report Toxic Wastes and Race in the United States used U.S. Census Bureau demographic data and multivariate statistical techniques that controlled for income, home values, waste generation, and abandoned waste sites. ⁷ They found that zip codes across the United States with one or more hazardous waste facilities contained disproportionately high percentages of racial and ethnic minorities compared to those with no hazardous waste facilities. The study concluded that there was a consistent national pattern, and that race was the most significant factor in determining the location of commercial hazardous waste facilities in the United States. An update to this classic study was completed 20 years later, and it found that the racial disparities in hazardous waste siting were even greater than reported in 1987. ⁸

Contemporary EJ researchers have used a number of increasingly sophisticated analytical techniques. For example, Morello-Frosch et al. ⁹ used correlations and multivariate regression to determine that race helps explain lifetime cancer risks from exposure to air toxics in southern California. However, both correlation and regression assume that variables are independent and that the relationship between model parameters is spatially homogeneous. ¹⁰ That is to say, correlation and regression assume that the value of one variable is unrelated to that of any other variable and that the relationship between model parameters is stable across a given study area—two assumptions unlikely to be true when dealing with geospatially distributed data.

To overcome the independence assumption, researchers^11,12 have applied simultaneous autoregressive spatial regression models that account for spatial dependence by assuming the autoregressive process is occurring in the error term (spatial error model), the dependent variable (spatial lag model), or both (spatial mixed model). ¹³ However, spatial regression models are global and do not account for heterogeneity in relationships across space ¹⁴ —a key limitation that is often overlooked. As discussed later, this limitation can be addressed using GWR.

Given the substantial and growing evidence that vulnerable and susceptible groups are disproportionately impacted by environmental hazards, government agencies, including the U.S. Environmental Protection Agency (EPA), have sought to develop means to screen for potential EJCs. Another factor driving the development of these screening tools is the availability of data and the increasing ease of performing spatial and computational analyses. However, developing an EJ screening tool is no simple matter. Key challenges include deciding which hazard exposures are of concern; which sociodemographic indicators (e.g., race, ethnicity, income, and age) are potentially important; and which geographic scale (e.g., county, zip code, and census tract) represents an appropriate unit of analysis. Furthermore, missing data or data limitations may prevent the inclusion of necessary parameters in the analysis.

One effort at a nationally consistent EJ screening tool was EPA's Environmental Justice Strategic Enforcement Screening Tool (EJSEAT). EJSEAT used geographic information systems (GIS) and other analytical tools to identify potential EJCs in an effort to focus enforcement and compliance activities in overburdened communities. ¹⁵ EJSEAT used 18 indicators in four categories: environmental, human health, compliance, and social/demographic. ¹⁶ However, EJSEATs methodology was criticized by the National Environmental Justice Advisory Council (NEJAC) ¹⁷ and others ¹⁸ as inappropriate for identifying EJCs. One problem they ¹⁹ identified was that, despite having a different number of variables, with different consequences for EJ, the four categories of indicators receive equal weight when calculating each census tract's EJSEAT score. An additional limitation was that users were unable to compare scores between states or tribal territories because each census tract's EJSEAT score was normalized based on data from other census tracts in that state, rather than data from census tracts across the United States as a whole. EJSEAT did not advance beyond the “draft” form.

In response, the EPA developed a new online EJ screening tool, EJSCREEN, in 2015. EJSCREEN, like EJSEAT, combines a number of metrics with GIS to try to identify potential EJCs, although the EPA did state that it was not currently using the tool to formally identify an area as an EJC. ²⁰ As described in EPA ²¹ technical documentation, EJSCREEN attempts to help users identify potential EJCs by creating 12 “EJ Indices” for each U.S. Census block group that combine an environmental indicator (e.g., air pollution concentration; traffic proximity; National Priorities List site proximity; treatment, storage, and disposal proximity; and cancer risk) with a demographic index that accounts for the percentage of minority and low-income residents. The value of each block group's EJ index can be compared against others in the immediate area, within the state, and within the entire United States, using color-coded maps and charts.

While EJSCREEN is a useful tool to get a sense of the relationship between minority and low-income communities and exposure or proximity to environmental hazards, the tool itself is not sufficient to identify an EJC, with the EPA noting that “[EJSCREEN] should not be used to identify or label an area as an ‘EJ Community.’ Instead, EJSCREEN is designed as a starting point, to highlight the extent to which certain locations may be candidates for further review or outreach.” ²² The EPA states that supplemental and local information is needed to identify an EJC, saying, “EJSCREEN's initial results should be supplemented with additional information and local knowledge, whenever appropriate, for a more complete picture of the location.” ²³

Few researchers have attempted to develop EJ screening tools. However, noting the challenges the EPA has faced with developing screening tools—EJSEAT in particular—Sadd et al. ²⁴ developed an Environmental Justice Screening Method (EJSM) that creates a relative ranking of cumulative impacts and social vulnerability (a “CI score”) based on 23 indicator metrics organized into three categories: hazard proximity and land use; air pollution exposure and estimated health risk; and social and health vulnerability. The EJSM is promising but has several limitations that may hamper its use as a national EJ screening tool. One key limitation is that the 23 indicator metrics are weighted equally in developing the CI score, ignoring the relative importance of each metric in driving environmental hazard exposure and health risks. The authors hold that weighted metrics could help address this, should evidence indicate that it were necessary. However, even if a weighting scheme were developed, the EJSM would not account for local variability in the metric weights.

Despite their limitations, these screening tools and methods can provide useful ways of expanding attention to and focus on environmental justice issues, separate from explicit designations and definitions that are associated with the needs of managers administering governmental programs. These tools and methods can be supplemented by other analytical approaches that explicitly account for local variability in the relationship between exposure to environmental hazards and sociodemographic factors, such as GWR, which produces a unique set of regression parameters for each spatial unit of analysis.

Discussion

Empirical EJ research results, paired with direct EJ activism, helped secure EJ on the U.S. research and regulatory agendas and led to the development of federal EJ screening tools and methods, including EJSEAT and EJSCREEN, and researchers have developed similar screening approaches (e.g., EJSM). These screening tools and methods can be valuable as a first-level exploration of potential environmental injustices in the United States. These tools and methods can be supplemented by in-depth local analyses that explore the local variability in the relationship between exposure to environmental hazards, sociodemographic factors, and social processes.

One way to supplement first-level EJ screening approaches is to leverage geospatial statistical methods that embrace the computational and data visualization power of GIS, such as GWR. While not without its limitations, the key benefit of GWR is that it allows the relationship between model parameters to vary across space. That is, it accounts for spatial dependence and heterogeneity in the relationship between model parameters, overcoming critical limiting assumptions of other approaches such as correlation and OLS regression that have been used to identify EJCs in the past. Therefore, GWR may be particularly well suited to supplement EJ screening tools. Alternatively, GWR can be used on its own to screen for potential EJCs, provided an analyst suspects a relationship between exposure to an environmental hazard and sociodemographic characteristics. However, due to the relative novelty of GWR compared to other statistical techniques, and the small number of EJ studies incorporating the method, more research is needed to fully explore its potential to identify potential EJCs.

While this article has focused on the uses of GWR relative to environmental hazards, we recognize that GWR could also be used in the other facet of environmental justice work: exploring disparities in access to environmental amenities, parks, and so on. Most research in environmental justice has examined burdens but another thread of work has focused on benefits,^38,39 and regulators acknowledge the importance of examining the distribution of benefits. In fact, as of 2016, the EPA modified EJSCREEN by adding parks and greenspace layer. A group of EJ advocates ⁴⁰ have celebrated this addition because, in their words, park access and health are a “significant national civil rights and environmental justice challenge.” GWR could be used to assess the relationship between sociodemographic factors and the distribution of environmental benefits.

It should be mentioned that an EJC screening tool, even one augmented by follow-on use of GWR, is neither necessary nor sufficient to identify EJCs. Communities have successfully advocated for attention and policy change to work toward environmental justice in their neighborhoods without first being identified as a potential EJC by a regulatory agency. Also, it appears that using an EJC screening tool to identify potential EJCs is not sufficient to confirm that environmental injustices are occurring. However, GWR can be a powerful screening tool for identifying areas that deserve further scrutiny as potential EJCs.

Also, it is important to note that, while well intentioned, identifying a community as an “EJC” has consequences. While EJCs may receive more attention and care from regulatory agencies, especially as it relates to permitting, enforcement, and compliance, there can be a stigma associated with being labeled an EJC. An in-depth discussion of the literature on stigma associated with living in a “dirty” neighborhood was beyond the scope of this article. However, suffice to say that stigma can produce substantial negative economic, social, and psychological harm to residents living in, or near, an EJC. ⁴¹ Thus, it is crucial to avoid any approach that seems to pathologize a community, and it is essential that community representatives participate in ongoing efforts to identify and protect EJCs.

With these caveats in mind, the value of identifying potential EJCs lies in recognizing that there are those in our society who bear an unfair and disproportionate burden from environmental hazards. These groups are often economically, politically, or socially disadvantaged and they deserve to play an active role in decisions that directly impact their well-being and quality of life. Identifying EJCs is one step toward environmental justice, and methods and tools that take advantage of powerful statistical techniques and geospatial analytical tools—such as GWR—can play an important role in this process.

Conclusion

GWR deserves further consideration as a means to help identify potential EJCs. The method can help identify areas where potential environmental injustices are occurring, including some that other methods may miss. GWR overcomes many of the simplifying assumptions of other EJ research methods, including OLS regression. As an additional analytical tool, GWR can be useful for EJ researchers, community members, regulators, and other stakeholders.