Method: Apportioning Temporal Changes in Environmental and Social Justice Metrics by Geographical Layers

INTRODUCTION

In the late 1980s, the United Church of Christ (UCC) set a high standard for applying quantitative methods in the evolving field of environmental justice (EJ). ¹ Reflecting on the infamous case of the Afton, North Carolina PCB landfill, the director of the UCC project, Charles Lee, was convinced there were many other hazardous waste EJ cases. Lee and colleagues examined the location of 451 commercial hazardous waste landfills and uncontrolled toxic waste sites across the nation. Using simple statistical tests like difference of means and more complicated ones like discriminant analysis, the group verified that their results did not depend on a single statistical test. The UCC report, Toxic Waste and Race, ² made a strong case that commercial waste sites were clustered in relatively few of the tens of thousands of U.S. zip code areas, and race and income were strong discriminating variables in their findings. That study was repeated two decades later with equally distressing findings. ³

Other researchers examined the locations of selected toxic waste sites and most reported similar findings, i.e., a disproportionate concentration of locally unwanted land uses (LULUs) in areas occupied by poor people of color.4^,5^,6^,7^,8 Interest in EJ has expanded beyond studying toxic waste sites to include additional LULUs (e.g., highways, ports, industrial sites, mines, trash incinerators, and affordable housing), natural hazards (e.g., floods, hurricanes, and tornadoes), soft power sites (e.g., museums, zoos, and parks), and some social injustices. Methods of analysis in EJ have also expanded to include historical analysis, legal analysis, meta-analysis, and more.9^,10^,11^,12^,13^,14^,15^,16^,17^,18^,19^,20^,21 Similarly, the evolution of GIS now allows the curious to explore data in interesting ways, ²² and new datasets have become available for public use.

Given this brief context, this article has two objectives: 1.

Present a method for examining temporal changes in local-scale environmental and social justice metrics and filtering those changes through trends at larger scales.

We view the method as a step toward understanding the evolution of EJ across time and space. The method begins by using publicly available databases to calculate a baseline of how a local place of interest currently compares with its surrounding area and host state. Having prepared the baseline, we use a spatiotemporal layering method to extend the baseline to include recent trends.

We illustrate the importance of adding time with an example. Suppose the proportion of public high school graduates in County X located in State Y was 75 percent in the initial year and 90 percent at year’s end year, an impressive 20 percent increase. Is this increase truly impressive? If the surrounding areas and the host state increased 12 and 15 percent, respectively, and if they had lower overall rates than County X, we would consider it truly impressive. However, if the surrounding areas and state improved more than County X, there would likely be a call for aggressive action to address the county’s lagging graduation rate. Posing a moral argument to move policy change forward may not be sufficient. Presenting empirical data to demonstrate that conditions are not improving relative to adjacent areas is harder for decision-makers to ignore.

METHOD

Our suggested approach uses a multiscale spatial analysis that combines influences on EJ at different spatial scales to help understand local processes. With new applications, advanced GIS, satellite imagery, and new datasets, multiscale analysis is already used in other fields of inquiry. For example, Ghimirc et al. ²³ studied multiple spatial scales looking for spaces where it was once possible to reuse animal manure but was now difficult because of land scarcity. The authors used remote sensing and mapping tools to find places that would accommodate and use the waste. Danson et al. ²⁴ used multiple spatial scales to investigate the spread of a parasite from foxes, coyotes, and dogs to human populations by following environmental assets, climate variables, and the presence of susceptible animals and people. Cain et al. ²⁵ used a multiscale analysis with remote sensing data to determine the persistence of land classifications. Similarly, we assume that changes within a local area are influenced by what occurs at larger scales. We begin by calculating the change in a metric of interest (the graduation rate) for layer 1 (a county) during a given study period. We then calculate the portion of that increase that can be assumed to be apportioned to layers 3 (the state) and then layer 2 (the surrounding region).

Equations 1 and 2 use this information to find the effect of layer 3 (state) change on the change in layer 1 (county). Equations 3 and 4 cover the calculations for the effect of layer 2 (region) on the change in layer 1 (county). Attributed& State& Change ij = & Initial& Rate ij × & Change is(1)where

Initial Rate_ij = the initial measurement of metric i in county j (75%)

Change_Is = the change in metric i in the state during the study period (13.7%)

Attributed State Change_ij = the component of change of metric i in county j attributed to the state during the study period (calculated as 75 × 0.137 = 10.28%) Unattributed& Local& Change ij = & Total& Change ij – & Attributed& to& State& Change ij(2)where

Total Change_ij = total change in metric i in county j during the study period (90% − 75% = 15%)

Unattributed Local Change_ij = the difference between total county change in metric i in county j and the portion attributed to the host state (15% − 10.28% = 4.72%). Attributed& Regional& Change& & = & Initial& Rate ij × & Change ir(3)where

Change_ir = the change of the metric i in Layer 2 (region) during the study period (8%)

Attributed Regional Change_ij = the component of change of metric i in county j attributed to the region during the study period (calculated as 75 × 0.08 = 6%) Unattributed& Local& Change ij = & Total& Change ij – & Attributed& to& Region& Change ir(4)

Where

Unattributed Local Change_ij = the difference between total county change in metric i in county j and the portion attributed to the region (15% − 6% = 9%).

In this hypothetical case, the county increased more rapidly than both the state and region. The result should generate discussions about the differences and why they might exist.

MORE READILY ACCESSIBLE DATA

During the last decade, the U.S. federal government and not-for-profits built and released multiple national-scale databases for public use that provide information about environmental, social, demographic, health, and economic metrics. County Health Ranking and Roadmaps ²⁶ provides county, state, and national data on health outcomes and behaviors, along with some demographic, social, and environmental indicators. This is the primary database used to illustrate the method suggested by this paper.

We also examined several other databases. First released in 2015, the U.S. Environmental Protection Agency’s (EPA) EJScreen ²⁷ provides data by census tracts, local governments, and comparable data for states and the nation as a whole. Its GIS tool allows users to construct collection areas with a variety of polygon options. Some historical data are also available. The EPA warns users that current numbers are not directly comparable to earlier datasets because the agency has modified its calculation of many of the metrics and changed the locations of monitoring locations. With assistance from the EPA, we were able to find some of the earlier data and compare results, especially for fine particulate data (2.5 ppm).

The Climate and Economic Justice Screening Tool, ²⁸ first published in 2022 by the Council on Environmental Quality, is used by the Justice40 ²⁹ program along with other data to distribute federal funds to distressed localities. The Agency for Toxic Substances and Disease Registry (ATSDR) published a risk database in 2023. ³⁰ The Center for Disease Control and Prevention publishes PLACES, ³¹ which provides details data about health outcomes and behaviors for cities, counties, and census tracts. Historical data for PLACES are available, but if the user changes to the census tract scale, they must first confirm that the historical boundaries are comparable. FEMA’s hazard database ³² allows users to examine estimated vulnerability for agriculture, property, and population, as well as human health vulnerability.

These databases and their updating led us to hope that more current and historical data will become available at multiple scales, along with methods to standardize between them. Indeed, historical data are essential to look for trends. We note that County Health Rankings and Roadmaps ³³ provide annual data at the county level back to the year 2010.

A TWO-COUNTY FIVE-METRIC ILLUSTRATION

To demonstrate the method, we chose two counties and five metrics. The two are markedly different in several ways. Allendale has a population of < 7,000, and the Bronx’s population is 1.3 million. In other ways, the counties are similar. Both have proportions of persons of color exceeding 75 percent. About 29 percent of the population in both counties is classified as living in poverty. Similarly, both have been classified as in “persistent poverty” (20 percent or more in poverty for at least 30 years.) ³⁴ Allendale, located along the Savannah River in South Carolina, illustrates rural poverty. The Bronx, located between Manhattan and Westchester counties, represents urban poverty. County Health Rankings and Roadmaps rank both as last in health in their respective states. Selected simply to illustrate the method, we do not assume these two counties represent the variety of places worthy of EJ examination.

We chose five of the metrics used in EJ studies to test the method. The first, high school graduation rate, is the number graduating divided by the number entering ninth grade four years earlier. Cohort data are used because students often transfer in or out. While the rate has been increasing nationwide, several authorities suggest that school boards feeling political pressure are graduating students who have not learned the requisite skills. ^{35 , 36}

Children under 18 years who live in poverty are the second metric. These children have an increased risk of injury as a result of unsafe environments. They are also more likely to be susceptible to asthma, anxiety, behavioral problems, diabetes, obesity, and other problems compared with children living in more affluent circumstances. ^{37 , 38}

Fine particles, our third metric, represent a national ambient air quality standard that is arguably the nation’s most important air quality measure. We used the National Environmental Public Health Tracking Network to find comparable historical data. ^{39 , 40}

Premature death rates are powerful indicators of distress. Years of life lost before age 75, our fourth metric, is widely used because it emphasizes the deaths of younger people, especially for drug and alcohol abuse by persons experiencing a sense of hopelessness. ^{41 , 42}

The fifth metric is the proportion of persons of color. The baseline for all five indicators appears in Table 1. The starting and ending dates are not the same because (1) the datasets are updated at different times, and (2) we wanted to avoid single-year reporting because of limited numbers of people in some local areas.

Although both counties demonstrate increased graduation rates, Allendale’s increased more than that of the Bronx. Allendale’s proportion of children in poverty increased while that of the Bronx decreased. Both counties show substantial decreases in fine particulates; both have clear pluralities of persons of color. Of special interest is that the premature death rates are markedly different. Allendale’s are higher and increased over time, whereas those for the Bronx were lower and decreased. In other words, despite their similarity in poverty and people of color, the metrics show they are distinctly different places.

Table 2 provides insights into why the counties have different metrics. Allendale has about the same high school graduation rates as its state, but notably worse results for children in poverty, 2.5 ppm, and premature deaths. Its population of persons of color exceeds 75 percent and is increasing. These results are informed by comparisons with its adjacent rural counties. Allendale’s results for high school graduation are worse and its poverty rates are slightly better than those of its surrounding region. The fine particulate results are about the same. The premature mortality results are the most interesting because Allendale’s high and increasing rates are also found among its neighbors. In short, Allendale resembles its poor and underserved minority neighboring counties more than it does its host state. While it has been underserved and is highly vulnerable to natural hazard events, it is not without assets. Part of the county is occupied by the Savannah River nuclear defense site and is included in the Justice40 program. However, Allendale and its neighbors still need improvements in infrastructure, especially broadband access and public schools.

The Bronx does not resemble its neighbors except for the 2.5 ppm metric where it is similar. It is closer to the state than to its surrounding region for the other metrics. Historically, the Bronx provided apartment dwellings for middle-class commuters to New York City. It is now a major source of affordable housing. Westchester County, the Bronx’s northern neighbor, represents suburban affluence and the border between the counties is marked with sharp differences in indicators of social justice. ⁴³ The most obvious of these is in persons of color. The Bronx’s proportion is almost 86 percent. While the proportion has declined slightly, the Bronx remains marked by poverty associated with race and ethnicity. Persons of color also increased in the New York State as well as in the counties surrounding the Bronx.

The time series data in Table 2 show limited evidence of narrowing the cumulative EJ burden on the Bronx. For example, it had among the highest COVID-19 death rates and hospitalization rates of any urban county in the United States. Despite having among the strongest social and environmental communities in the nation, the Bronx needs even more support from local officials and state representatives.

DISCUSSION AND PATHS FORWARD

This method has limitations. First, it should only be applied after completing an initial layering analysis using one or more of the databases described earlier. The baseline should be informed by testing the region with spatial autocorrelation tools and/or by obtaining community input for defining the regional layer. In our experience, County Health Rankings and Roadmaps and EJScreen are particularly easy to use for this task. The other four databases bring additional assets to the table.

Second, although there are over 100 variables to draw upon in national databases, they are marked by large gaps in data, especially in rural areas and for drinking water quality. Much of the data incorporated into the national databases come from work done by states. For example, Maryland’s screening tool ⁴⁴ added information relevant to agriculture areas, on infrastructure (e.g., railroad lines, food access, and more), as well as a cumulative burden score. California ⁴⁵ added indicators on water quality and industrial zoning, produced interesting new visualizations of data, and information on racial distinctions. These gaps were issues for the first version of CEQ’s CEJST database ⁴⁶ identified by Holifield. ⁴⁷ Mullen et al. ⁴⁸ note that large databases do not adequately represent the concerns of Indigenous populations, pointing to the need for data about extractive industries, cultural assets, and the extent of public engagement.

Another limitation of this paper, although not of the method, is how we defined region (counties that share a land border with the county of interest). How a region is defined depends upon the goal of the user and their knowledge of the local area. Indeed, quantitative decision-support databases require qualitative decisions about appropriate geographic levels and may miss critical information that can only be obtained by direct communications with community groups.

This method provides a simple way of identifying spatiotemporal patterns in EJ indicators. We suggest using existing methods such as evaluating clusters to check the spatial-temporal results. The layers are used here to demonstrate that the method will not satisfy those who want to use census tract data. We note that small-scale comparisons present challenges and that census tract boundaries may have changed over time. In other words, users must evaluate small-area data with caution.

Historical data are another limitation. If boundaries have remained the same or if transition estimates are provided, social and demographic variables may be appropriate to compare. Environmental variables, however, pose a challenge because monitoring locations change as do the methods for calculating values. There is no obvious way of overcoming this problem at the national scale although some states have records that should be accessible.

A final limitation is logistics. Performing these calculations for a handful of areas is tedious but doable. Performing them for hundreds of local areas is a task for computers. Better still is using artificial intelligence to engage the data. Ideally, if the method proves useful, a federal agency will take responsibility for performing the calculations and issuing periodic reports, perhaps even incorporating it into one or more of the databases for public access.

This method needs further testing with a major database, such as all the counties in a state or a large metropolitan area. Census tracts should also be tested, being aware of the issues raised above. If census tracts are layer 1, we suggest that the municipality or county be layer 3 and the surrounding tracts be layer 2.

Notwithstanding the limitations of the method, there are good reasons for separating local changes from larger-scale elements. It allows for a deeper understanding of what is happening locally. Similarly, examining geographically apportioned EJ concerns over time should lead to fruitful strategy sessions about EJ remedies.