Authors: John R. Hipp, Yuqing Wang, Nisha Gottfredson O’Shea, Robert W. Faris, Dorothy L. Espelage, Alberto Valido, Tamara Taggart
Categories: Article, Structural racism and discrimination, spatial scale, measurement, neighborhoods, life expectancy
Source: Social science & medicine (1982)
Authors: John R. Hipp, Yuqing Wang, Nisha Gottfredson O’Shea, Robert W. Faris, Dorothy L. Espelage, Alberto Valido, Tamara Taggart
Understanding the role of structural racism and discrimination is crucial for understanding persistent social disparities across many domains, but a key concern how does it operate at different spatial scales, and does that affect how it should be measured? We explore both empirical and theoretical differences between structural racism measures at meso and macro scales in the United States. We construct two different measures of structural racism at the meso-level (that is, neighborhoods), one capturing absolute measures of neighborhood blight, the other relative indicators of Black-White inequality. We test four different scales of these meso-level measures, each defined by a buffer centered on a block with four different radii. We simultaneously measure relative racial inequality at the macro level (that is, the county).
We compared the relationship between structural racism measures on life expectancy at varying meso-scales. The structural racism measures include traditional indicators like residential segregation, as well as novel ones capturing neighborhood blight, such as eviction rates, housing vacancies, home loan denials, and proximity to toxic waste, fringe banks, liquor stores, and bars. We estimate multilevel linear models in which the outcome measure was the average life expectancy at the meso-level.
We found consistent evidence that neighborhood- and county-level structural racism are associated with lower average life expectancy in neighborhoods. For our combined neighborhood structural racism measure (Black/White inequality and neighborhood blight), we detected a nonlinear relationship in which life expectancy is even more sharply lower for higher concentrations of multidimensional structural racism. The novel neighborhood blight measure had stronger relationships with life expectancy than did Black/White inequality, and they were stronger in high concentration Black neighborhoods.
Understanding how structural racism operates at different spatial scales is important for exploring its mechanisms and informing targeted multilevel interventions to address health disparities.
Studies examining the effects of structural racism and discrimination on health, particularly their impact on life expectancy, have been limited by two fundamental measurement (a) identifying observable indicators of structural racism and (b) identifying the appropriate spatial scales for their aggregation. A key question is whether structural racism manifests differently at different geographic scales, and thus requires scale-specific measurement approaches. In other words, what is structural racism made of, and how is it distributed? How similar, with respect to structural racism, is one block to the next, one neighborhood to the next, or one county to the next?
Our investigation is premised on the assumption that structural racism is unevenly geographically it is not simply the air we breathe, uniform and consistent. Some places are worse than others. But how large are those places? Counties? Cities? Neighborhoods? Blocks? We suggest that the appropriate scale depends on structural racism’s underlying components, and point to two of its distinct manifestations that, we hypothesize, have differing radii of influence. The first, racial inequality, is central to the concept of structural racism. We suggest that it is more readily observed over larger areas, because residential segregation along racial and socioeconomic strata reduces block- and neighborhood-level diversity. The second dimension, exposure to blight and other material circumstances that limit life chances, operates on a much smaller geographic scale. Home values, for example, are generally unaffected by sales beyond a small radius. In contrast, even air quality can vary significantly over the span of feet, rather than miles (Manavi et al., 2025).
Extant evidence has relied heavily on residential segregation patterns and redlining maps to construct community-level measures of structural racism (Groos et al., 2018; Riley, 2020). However, recent research has expanded to incorporate additional dimensions of structural racism, including examining disparities in educational opportunities, economic investments, and criminal justice practices. In this study, we attempt to capture structural racism at a more proximate meso-level, and introduce additional novel measures to capture neighborhood blight, such as the neighborhood eviction rate, the presence of fringe banks, and toxic waste exposure. Despite this progress in broadening the scope and reach of structural racism indicators, systematic attention to various possible geographic scales of measurement has lagged. Many studies continue to rely on convenient or census-defined units (Taggart et al., 2023). Although structural racism’s pernicious effects—including negative consequences for health and reduced life expectancy— are well-documented in existing literature, the hypothesized mechanisms through which structural racism operates (e.g., barriers to accessing health care, exposure to environmental hazards, socioeconomic status disparities) may not manifest equally across different geographic scales. Greater attention to various geographic scales is therefore needed.
Although much has been written on the effects of internalized and interpersonal forms of racism on health and health inequities (i.e., the micro level) there remains a gap in the investigations of more upstream levels of racism (Bailey et al., 2017; C. P. Jones, 2000). In this study, we focus on structural racism, defined as “the totality of ways in which societies foster [racial] discrimination, via mutually reinforcing [inequitable] systems (e.g., in housing, education, employment, earnings, benefits, credit, media, health care, criminal justice, etc.) that in turn reinforce discriminatory beliefs, values, and distribution of resources” (Bailey et al., 2017; Krieger, 2014). Structural racism merits particular attention as it represents the foundational mechanisms through which racial inequities are produced and maintained across the life course, making it critical to understanding persistent health disparities that have resisted intervention at the individual and interpersonal levels.
The earlier literature generally focused on the macro-scale, using larger units of analysis such as cities (Siegel et al., 2023b), counties (Dougherty et al., 2020; Tan et al., 2022), and even U.S. states (Brown & Homan, 2024; Homan et al., 2021). This body of work consistently found that structural racism helps explain racial and ethnic health disparities in the U.S., as higher levels of structural racism are associated with increased mortality rates, elevated rates of chronic diseases, and adverse birth outcomes (Bailey et al., 2017; Gee & Ford, 2011; Williams & Mohammed, 2013). Furthermore, this burgeoning literature has measured structural racism using a variety of indicators, including indices of residential segregation, racial income gaps, mortgage lending disparities, and inequities in criminal justice outcomes (Groos et al., 2018; Krieger et al., 2017).
Whereas the implications of structural racism can be observed at the macro-scale, arguably, many of its consequences play out at the meso-scale in what we would refer to as “neighborhoods.” Despite the challenge of defining “neighborhoods,” a large body of research in recent years has explored structural racism at the meso-scale, using census units such as tracts or block groups to define the neighborhood (Beyer et al., 2016; Jacoby et al., 2018; Dara D. Mendez et al., 2011). However, evidence from the social determinants of health literature shows that using an even more geographically specific spatial scale, rather than census units, may yield stronger relationships (Chaix et al., 2009; Cummins et al., 2007; Lovasi et al., 2008). Thus, it may be necessary for scholarship to measure “neighborhoods” based on either activity spaces that capture residents’ actual experiences or on precise geographic buffers as an approximation to better capture these spatial processes, and we adopt this strategy here (Taggart et al., 2023). This implies that these buffers may be a superior meso-level aggregation than traditional census defined units such as tracts or block groups.
In the present study, we use data from both meso- and macro-scales to better capture how structural racism may contribute to lowered average life expectancy for residents. Rather than measuring neighborhoods based on census-defined units such as census tracts, we measure them using buffers around specific blocks, previously referred to as “egohoods” to reference its social network analogue, the ego network (Hipp & Boessen, 2013). This approach builds on literature that focuses on the spatial buffer around a person’s residence, under the presumption that doing so better captures the environment the person experiences (Chaix et al., 2009; Lovasi et al., 2008). We test whether a single meso-level variable combining our novel measure of neighborhood blight with a traditional measure of racial disparity is associated with reduced life expectancy, and also whether this operates in a nonlinear fashion. We then disaggregate the meso-level neighborhood blight and racial disparity measures to assess their independent relationships with neighborhood life expectancy, and then assess whether the neighborhood blight measure has a stronger relationship with life expectancy in neighborhoods with a larger concentration of Black residents. Given the empirical importance of neighborhood blight, we then test whether the specific components of this measure are all related to neighborhood life expectancy. The final test assesses whether these specific components have stronger relations in neighborhoods with a larger concentration of Black residents. In all models, we also included a macro-level measure of structural racism for each U.S. county to assess its simultaneous relationship with neighborhood life expectancy. By modeling structural racism across multiple scales, we test whether this enhances our understanding of the consequences for life expectancy. It is possible that failing to account for structural racism across multiple scales may underestimate the actual impact of structural racism, which may lead to incomplete or ineffective intervention strategies, limiting the strategies’ potential impact on mitigating structural racism and health inequities.
A large body of research has focused on how to measure structural racism and then assess how it relates to various health outcomes for racial minorities, particularly for Black Americans. Much of the earliest work in this area focused on macro-level geographic units at larger geographic scales, such as cities, counties, or even states (Dougherty et al., 2020; Homan et al., 2021; Siegel et al., 2023a). A common methodological strategy was to create a measure of racial residential segregation between Black and White residents in neighborhoods across the macro-scale, such as the county (Anderson et al., 2023; Siegel et al., 2023a; Tan et al., 2022). This racial segregation was then posited to capture structural racism across neighborhoods within these larger geographic units, with the presumption that counties in which Black residents experienced greater segregation by neighborhoods would experience more structural racism. Subsequent research has measured structural racism by capturing the relative differences between Black and White residents along various social dimensions such as housing, employment, or the criminal justice system (Chantarat et al., 2021; Dougherty et al., 2020; Siegel et al., 2023a). These studies typically include measures capturing each of these various social dimensions simultaneously in the models. Subsequent scholars have argued that structural racism is instead a single construct rather than each of these separate dimensions, and have, therefore, combined separate Black/White structural racism dimensions into a single scale (Homan et al., 2021) or a latent variable (Brown & Homan, 2024; Dougherty et al., 2020; Siegel et al., 2023b).
Although studies measuring structural racism at larger geographic scales have provided key insights, one could argue that despite its pervasiveness, some neighborhoods are worse than others due to contextual and historical factors. Therefore, recent research has focused on meso-units when measuring structural racism, typically using neighborhoods based on census-defined geographic units, such as block groups (Mohottige et al., 2023) or tracts (Dyer et al., 2023; Jacoby et al., 2018; Mendez et al., 2011). Prior research takes geographic units for granted, based on data availability, and are not a focal concern. This is the first investigation of structural racism to systematically consider its spatial scale, asking whether the context of the egohood centered on a block can be expected to meaningfully differ from the egohood centered on a nearby block. The answer depends, however, on how structural racism is measured.
Most prior studies at the meso-level measure structural racism simply as housing segregation, based on historical maps of redlined neighborhoods (Collin et al., 2021; Jacoby et al., 2018; Krieger et al., 2020). Another common strategy is to measure structural racism as contemporary housing discrimination, using home mortgage denial rates from the Home Mortgage Disclosure Act (Beyer et al., 2016; Mendez et al., 2011; Mendez et al., 2014 ). A less common strategy is to measure structural racism through indices of economic inequality (L. Dyer et al., 2022; Krieger et al., 2017). We are aware of only one neighborhood study that uses multiple dimensions of structural racism (Dyer et al., 2023).
A smaller body of literature has instead measured economic disadvantage as a proxy for structural racism (Krivo et al., 2009). In this literature, rather than focusing on racial inequality, the strategy is to measure economic disadvantage with indicators sometimes referred to as “race-neutral” measures, under the presumption that structural racism generates economic disadvantage for racial minorities (Mohebbi et al., 2024). A limitation of this strategy is that it assumes that any economic inequality observed is due entirely to structural racism. Given that economic inequality is inherent in capitalist societies, this assumption may not be plausible.
Our strategy differs from these earlier approaches in important ways. First, given that we are measuring structural racism at meso- and macro-scales, it is important to consider theoretically how it manifests at these different levels. Thus, whereas we measure county-level structural racism in a manner similar to that of existing research based on racial disparities between White and Black residents along several sociodemographic dimensions, structural racism may operate differently at the neighborhood level; structural racism that occurs more locally may affect an individual’s outcomes more directly than does structural racism measured over a broader geographic region.
Second, we measured neighborhood structural racism along two broad racial disparity and neighborhood blight. Although racial disparities at the neighborhood level likely operate differently than at the broader macro-scale, we still test them here. We believe it is more likely that macro Black/White structural racism operates by pushing residents into disadvantaged neighborhoods, and therefore macro structural racism operates at the neighborhood level through neighborhood blight. We present our conceptual model in Figure 1. As shown there, we conceptualize racial disparities at the macro level (measured in counties in the present study) as evidence of structural racism, and this impacts the two meso-level dimensions we racial disparity and neighborhood blight. These two dimensions are then expected to negatively impact life expectancy in these neighborhoods. The dotted line in the figure shows that county-level racial disparity may also impact life expectancy through other meso-level mechanisms that we have not measured here. We highlight that a challenge for the structural racism literature is the potential conflation of the structures of racism and discrimination with the outcomes of structural racism. We face the same challenge, as one could argue that Black–White inequality is merely a symptom, not the disease itself.
Our measure of neighborhood blight is impacted by these broader racial disparities across various dimensions. One dimension of our measure of neighborhood blight is based on features that are at least in part a function of a racially unequal housing market, such as the neighborhood eviction rate, the vacancy rate, and the home loan denial rate. Reduced loan availability disproportionately affects racial minority neighborhoods by reducing capital inflow through investment (Immergluck & Smith, 2005; Velez et al., 2017), and a further consequence is greater vacancies in such neighborhoods and the negative consequences of these vacancies for general health (Boessen & Chamberlain, 2017; Krivo & Peterson, 1996). Furthermore, inequities in the housing market result in lower-quality apartment buildings in some neighborhoods, which may lead to reduced services and increased eviction rates (Desmond, 2016). Another dimension is that macro Black/White structural racism arguably can affect the placement of toxic facilities, and we measure the presence of toxic waste facilities, given the long literature showing disproportionate placement of such facilities in racial minority neighborhoods (Bullard et al., 2007; Pastor et al., 2001). We measure fringe banks—that is, check cashers, payday lenders, and pawnshops—given their high-cost, high-dependency business model (Stegman & Faris, 2003) and disproportionate location in predominantly racial minority neighborhoods (Kubrin & Hipp, 2016); we also measure bars and liquor stores, given the extensive literature showing the disproportionate placement of such facilities in racial minority neighborhoods (Lipton & Gruenewald, 2002; Treno et al., 2001).
These measures of neighborhood blight negatively impact residents’ life expectancy through several interconnected mechanisms. First is through chronic stress, and the negative consequences it has for health (Lupien et al., 2018; McEwen, 2008). Housing instability through evictions and economic challenges due to high interest rates from predatory financial institutions can create such chronic stress. Second is through negative lifestyle patterns, which can be exacerbated by the presence of liquor stores and bars, as well as reduced recreational physical activity due to safety concerns in neighborhoods with many bars and vacancies. Likewise, housing instability can disrupt access to healthcare, which can result in long-term negative health consequences. A third mechanism is the direct negative health consequences from environmental hazards, particularly prolonged exposure to effluence from nearby toxic waste facilities. We argue that all of these neighborhood features are driven, at least in part, by macro structural racism forces.
Whereas existing research almost always measures structural racism at a single geographic scale, our strategy measures structural racism simultaneously at both macro and meso geographic scales. Although we anticipate that macro-, or county-level, structural racism will be associated with reduced life expectancy, we further expect that meso-level structural racism, at the scale of blocks and their immediate surroundings, will also be associated with reduced life expectancy. Our strategy is to construct measures at both of these spatial scales, given our contention that it is important to measure the meso and macro contexts simultaneously. Studies have consistently shown evidence that in larger units—such as counties (Chambers et al., 2018; Dougherty et al., 2020; Tan et al., 2022) or states (Brown & Homan, 2024; Homan et al., 2021)—residents, particularly Black residents, living in macro units with greater levels of structural racism experience greater negative health consequences. These macro results highlight the importance of accounting for this larger context.
Although accounting for the meso-level context is also important, a challenge encountered by scholars is determining what geographic units are most appropriate for capturing such meso-level features. Most commonly, studies measuring meso-level structural racism have used census geographic units (tracts or block groups) as proxies for “neighborhood”, which may not be the ideal way to capture how the social environment affects individuals (Perchoux et al., 2013). This contention follows from a large body of literature focusing on how various features of the environment can affect the health of persons more generally. This literature typically measures the context based on a spatial buffer around a person’s residence, under the presumption that doing so better captures the social and physical environment the person experiences (Chaix et al., 2009; Lovasi et al., 2008). The logic is based on evidence that persons tend to spend most of their time in the environment around their home location, with a declining likelihood to spend time further away from their home (M. Jones & Pebley, 2014). That is, a spatial buffer. This buffer more accurately captures spatial mobility than census-defined geographic units do; for example persons living near the boundary of a tract may spend more time outside of their tract than inside it. Indeed, there is evidence that even adolescents spend more of their waking hours outside of their own tract than inside it (Browning et al., 2021). As described in the methods, rather than simply measuring the context based on a census tract, we define it based on a buffer centered on a census block. Such a definition arguably more accurately characterizes the local context for residents and is consistent with numerous studies measuring how the environment influences health (Chaix et al., 2009; Lovasi et al., 2008).
We expect structural racism to be manifested at both the macro- and meso-scale, but the meso-scale may be particularly proximate. Therefore, to better understand how structural racism operates, it is necessary to simultaneously measure both the macro and meso contexts to determine how structural racism negatively impacts residents’ estimated life expectancy. Furthermore, given that there is no “proper” sized buffer to capture spatial mobility, we will compare neighborhood structural racism effects on estimated life expectancy across four different meso-scales defined below. Given the importance of how structural racism operates for individuals, measuring the meso context is necessary, while also accounting for the macro context in which they are situated.
Data from the entire country were used for the analyses. We combined data from several sources to create our unique measures of structural racism at meso- and macro-spatial the 2010 U.S. Census, the American Community Survey 5-year estimates (2008–2012), the Reference USA Historical Business data (Infogroup, 2015), the Princeton Eviction Lab, the Environmental Protection Agency’s Toxic Release Inventory data, and the Home Mortgage Disclosure Act tract-level data. As we described earlier, the structural racism measures included in the study were based on prior literature and theoretical guidance on how we expected structural racism to operate in neighborhoods.
Given that there is no a priori proper meso-scale, we tested four different aggregation scales for our meso-level structural racism measures. For variables that we had at the block level, we constructed measures with four different-sized we chose values for radii of ¼ mile (approximately the size of a census block group in an urban area), ½ mile (approximately the size of a census tract), ¾ mile (approximately the size of two census tracts, which some scholars have used as neighborhoods) (Sampson et al., 1997), and 1 mile, each chosen to give a measure of the context experienced by the residents living in the block at the center of the buffer.
Ideally, we would aggregate our outcome variable—period life expectancy—to egohoods as well. However, given that this measure is only aggregated to census tracts, we adopted the following aggregation strategy. Based on the previous paragraph, we assigned the value of a variable in an egohood to the focal (central) this gives us the context for a specific block. We then computed the average for a measure across all the blocks in a tract to get an estimate of the typical context experienced by residents in the tract. This strategy was utilized for each of the variables constructed, and for each of the various buffer sizes we employed. This method provides a better approximation of the social environment experienced by the residents of a block (for a description of how such a strategy can be more effective in capturing the environment—even when the outcome variable is aggregated to tracts—see the simulation described in Hipp, 2020). For example, Figure 2 shows an example tract in Charlotte in North Carolina and three random blocks in the tract (the solid red areas) with ¼ mile context around each block (the dotted circle). This ¼ mile buffer captures the context of the specific block, and for each of these blocks we computed the value of a particular measure based on all the blocks contained in this focal block’s buffer, and assigned this value to the focal block. We performed the same computations for all of the other blocks in this tract, and then computed the average value across all these buffers—this is the value that is then assigned to the tract.
A limitation is that some of the variables were available only at block group or tract scales. Therefore, for these variables we used the following four geographic the block group or tract; a buffer that contains the centroids of all units within a radius of ½ mile; a buffer of all centroids within 1 mile; and a buffer of all centroids within 2 miles. Given that we are using different aggregations of the independent variables, but the same aggregation of the dependent variable, this does not raise the modifiable areal unit problem (MAUP) and subsequent uncertainty. Whereas there was only minimal missing data, we addressed it by using the multiple imputation strategy implemented in Stata (multiple imputation of chained equations), and used Rubin’s techniques to correct the standard errors (Rubin, 1987).^1^
The dependent variable in our analyses is a measure of period life expectancy within a census tract, sourced from the U.S. Small-area Life Expectancy Estimates Project (USALEEP) provided by the National Center for Health Statistics, which provides age-adjusted information on life expectancy for residents in each census tract for the years 2010–2015 (National Center for Health Statistics, 2018). These data were constructed through a multi-stage process. The NCHS researchers obtained death records of all U.S. residents for deaths between 2010–2015 from the National Vital Statistics System registration areas and the National Center for Health Statistics, and geocoded the decedents’ residential addresses to the census tract. They then constructed population estimates based on the 2010 decennial census and the 2011–2015 American Community Survey 5-year estimates. They addressed small population sizes for age-specific death counts through standard demographic techniques and statistical modeling, and adjusted for sample error with standard, abridged life table methods. For most tracts, the estimate is based entirely on information about tract residents. However, for approximately 13% of tracts—those with small populations—estimates are constructed by combining statistical modeling with demographic techniques (Arias et al., 2018). To verify the robustness of our findings, we also estimated ancillary models by excluding these small tracts. Our results remained unchanged.
We constructed a measure of neighborhood blight based on several dimensions using egohoods, which do not use a spatial decay (in which nearby areas are weighted more heavily) (Hipp & Boessen, 2013). First, we constructed a measure of Bars/liquor/convenience store employees as a percentage of total employees. The Reference USA data source provides us with the latitude and longitude of each business establishment in the United States, and we are able to aggregate the employees in these businesses to the blocks within the buffer of a particular block. Using the six-digit North American Industry Classification System (NAICS) code, we determined establishments that provide alcohol, are bars, or are convenience stores; summed the employees in them; and then divided by the total number of employees in businesses in the egohood. Second, we created a measure of fringe bank employees as a percentage of total employees in which we defined fringe banks as check cashers, payday lenders, and pawnshops, following the existing literature (Kubrin & Hipp, 2016). We created a measure of waste site toxicity by determining any toxic waste sites in the surrounding egohood (based on the Toxic Release Inventory data) and then weighting them by the toxicity in their exhaust (noncancer on-site air release; cancer on-site air release; cancer on-site water release). These toxicity estimates come from the Environmental Protection Agency’s Risk-Screening Environmental Indicators (Environmental Protection Agency, 2015). We measured the eviction rate using data from the Princeton Eviction Lab (Gromis et al., 2022). The vacancy rate was measured with data from the American Community Survey 5-year estimates from the U.S. Census. We created a measure of the denial rate of home purchase loans with data from the Home Mortgage Disclosure Act.
We also constructed seven variables capturing Black/White disparity. These measures are based on those in the existing literature that capture group differences across several dimensions, including housing, education, the economy, employment, and segregation (Arias et al., 2018; Chantarat et al., 2021; Siegel et al., 2023a). These variables are Black/White income inequality (difference between Black and White average income), Black/White education inequality (difference between Black and White average years of education), Black/White owner inequality (difference between Black and White homeownership rates), Black/White labor force inequality (combined measure of difference between Black and White unemployment rate [reversed] and labor force participation rate), Black/White mortgage inequality (difference between Black and White average mortgage purchase amount), Black/White mortgage denial inequality (difference between Black and White mortgage denial rates), and Black resident concentration (difference between percent Black in block group and in county). Because of data limitations, the two mortgage variables are constructed at the tract level whereas the others are constructed at the block-group level. For units with a Black population of less than 250, this is set to a missing value. The spatial scales of each meso-level structural racism measure we constructed are shown in Table 1.
For each of the aforementioned variables, we constructed measures that focused on the presence of each structural racism measure. Thus, we hypothesize that although the presence of high values of a structural racism measure has negative consequences, low values are not more advantageous than mid-range values. Therefore, rather than including these as continuous variables in our models, we created three-category ordinal variables for each. We chose the cutoffs for each variable such that approximately 10% of tracts were in the highest category, 20% of tracts were in the next highest category, and the remaining 70% of tracts had values of zero on these categorized variables. Thus, we focus on the presence of these various forms of disadvantage, with an extra weighting for tracts with very high values on a measure.
Although we estimated models including each of these categorized measures separately to estimate their distinct effects, we also summed the six categorized variables that capture neighborhood blight. Additionally, we summed the seven categorized variables capturing Black/White disparity. To further analyze the overall impact, we summed these two aggregated measures into a combined structural racism scale. Furthermore, to test for a nonlinear effect of this latter measure, we created a quadratic version of it. Given that these were categorized variables, each unit increase in these variables effectively indicates that one more structural racism measure is for example, an increase of vacancies from 0 to 1 would result in a one unit increase in the overall measure.
We also created a county-level measure of Black/White racial disparity. To construct this measure, we first calculated the ratios of Black to White across several household income, average years of education, ownership rate, labor force participation, mortgage purchase amount, and Black/White segregation (based on the index of dissimilarity). We then standardized each of these ratios and computed their mean (to avoid losing observations due to missing values in any of the component variables). As a check, we also conducted a principal components analysis of these variables and extracted a single factor score (Eigenvalue = 2.42 with an average loading of .61). We found that it correlated 0.99 with our mean of z-scores.
To minimize the possibility of obtaining spurious results due to confounding, we also constructed several variables that might be associated with structural racism and neighborhood life expectancy but that are not direct indicators of structural racism. We constructed a measure of economic disadvantage based on a factor score from a confirmatory factor analysis, which combined the percentage in poverty, percentage of single-parent households, average household income, and percentage with at least a bachelor’s degree.^2^ This economic disadvantage measure is commonly included in similar studies, allowing us to assess whether our novel measure of neighborhood blight has unique the correlation of .50 between the two measures suggests that, despite some similarities, these measures do capture distinct constructs.
Additionally, we created a measure of racial/ethnic heterogeneity based on a sum of squares of proportions of racial/ethnic groups (Black, White, Latino, Asian, and other races), and subtracted this value from 1 to make this a measure of heterogeneity. We also created measures of the percentage immigrants and the percentage of homeowners, as well as a measure of population density. These measures were all constructed based on egohoods with a particular radius around each block and then averaged within the tract.
At the county level, we constructed measures of percentage owners, percentage immigrants, average household income, percentage ages 22–29, percentage ages 30–44, percentage ages 45–64, percentage ages 65 and up, population density, and income inequality (measured using the Gini coefficient).
The summary statistics are displayed in Table 2. We display the correlations among our meso-level structural racism measures in Table 3.
We estimated multilevel linear models to account for the nesting of census tracts in counties. Generally, the models can be expressed as (1)yij=β0j+β1jx1ij+β2jx2ij+eij, (2)β0j=γ00+γ01w1j+μ0j,and (3)β1j=γ10+γ11w1j+μ1j, where yij is the outcome of period life expectancy in tract i in county j; β0j is a random intercept across counties; β1j and β2j are potentially random coefficients for variables x1 and x2 in tract i in county j, eij is an error term at the tract level; w1 is the vector of county-level control variables and our structural racism measure that can affect the random intercept β0j or any random coefficients β1j; the γ’s are county-level intercepts; and the μ’s are disturbances at the county level. Our level 2 equations are specified the same across all models, but level 1 is modified in each model.
More specifically, in the first model the level 1 equation (1a)yij=β0j+β1jx1ij+ω1jN2ij+eij, where x1 is our combined measure of structural racism, and N2ij is a vector of our neighbourhood-level control variables that are included in all models.
In the second model, we test for a nonlinear relationship between our combined measure of structural racism and life expectancy, modifying the level 1 equation (1b)yij=β0j+β1jx1ij+β2jx21ij+ω1jN2ij+eij, where x21 is the quadratic version of our combined measure of structural racism.
In model 3, we disaggregate the combined structural racism measure into the separate component measures of neighborhood blight and racial (1c)yij=β0j+β1jx1ij+β2jx2ij+ω1jN2ij+eij, where x1 is now our measure of neighborhood blight, and x2 is our measure of neighborhood racial disparity.
In the fourth model, we account for whether neighborhood blight shows a stronger impact as the percent Black in the neighborhood increases by adding an interaction of these two variables in the level 1 (1d)yij=β0j+β1jx1ij+β2jx2ij+β3jx2ijx3ij+ω1jN2ij+eij, where x1 and x2 are still our measures of neighborhood blight and racial disparity, and x1 x3 is the interaction between neighborhood blight and percent Black.
In the fifth model, we disaggregate the neighborhood blight measure by replacing it with the six separate measures that constitute (1e)yij=β0j+∑βqjxqij+β2jx2ij+ω1jN2ij+eij, where ∑βqjxqij indicates that we are including each of the six (q) neighborhood blight measures separately.
In the final model, we test whether each of these neighborhood blight indicators has a stronger impact on life expectancy as the percent Black in the neighborhood (1f)yij=β0j+∑βqjxqij+β2jx2ij+∑βqjxqijx3ij+ω1jN2ij+eij, where we extend model 1e by including ∑βqjxqijx3ij, which indicates the interactions between each of the neighborhood blight measures and the percent Black in the neighborhood.^3^
Across all models, we found that the optimal specification used the egohoods in the smallest buffers (comparing models based on the Bayesian Information Criterion (BIC) values). Therefore, the models presented are those that use these variables in the smallest buffers. In the initial model, we included our total neighborhood structural racism measure and our county-level Black/White racial disparity measure. All models also accounted for all neighborhood- and county-level control variables.^4^
In Model 1 of Table 4, we observe a strong negative association between our combined neighborhood structural racism measure and period life expectancy across U.S. census tracts (β = −.099, p-value < .001). Specifically, a 1-SD increase in the neighborhood structural racism scale is associated with a reduction of .232 years in life expectancy, or a decrease of .058-SD in period life expectancy. Note that this model controls for other neighborhood measures, including economic disadvantage and racial/ethnic heterogeneity, which are both negatively associated with average life expectancy, and the percentage of homeowners, which is positively associated. The model is also controlling for various measures of the county context. Notably, we see that the county-level Black/White racial disparity measure is also associated with lowered life expectancy, as a 1-SD change is associated with −.031 lower life expectancy.
In Model 2, we tested and found a nonlinear relation between the combined neighborhood structural racism measure and period life expectancy. The graph of this association in Figure 3A makes this finding explicit, and we see that whereas a 3-point change in this structural racism scale for a neighborhood with a value of 0 is associated with −.23 years life expectancy, this same change for a neighborhood with a value of 7 is associated with −.40 years life expectancy, and the same change for a neighborhood with a value of 13 is associated with −.54 years life expectancy.^5^ Thus, we see evidence of a compounding pattern in which increasing structural racism on top of already high structural racism levels at the neighborhood level has particularly strong negative consequences for period life expectancy in these neighborhoods.
Models 3 and 4 split the general neighborhood structural racism measure into its constituent Black/White disparity and neighborhood blight. Model 3 shows that, although both sub-measures are negatively associated with life expectancy at the neighborhood level, it is neighborhood blight that exhibits the strongest relationship. Specifically, a 1-SD increase in Black/White disparity is associated with just −.04 years less life expectancy, whereas a similar change in neighborhood blight is associated with −.46 years less life expectancy. In Model 4, to assess whether neighborhood blight has a stronger negative relationship in neighborhoods with a higher percentage of Black residents, we tested an interaction between neighborhood blight and percentage Black residents in the tract, and we detected a negative interaction term. The neighborhood Black/White disparity measure is now half as large in this model. Comparing a neighborhood with 40% Black residents to one with no Black residents, when neighborhood blight has a score of 1, neighborhood with more Black residents has −.11 years life expectancy, when the score is 2 the gap is −.22 years, and when the score is 3 the gap is −.32 years for life expectancy (see Figure 3B). Thus, neighborhood blight has a stronger negative relationship with life expectancy in neighborhoods with a higher concentration of Black residents.
In Table 5 we disaggregated the neighborhood blight measure into its constituent components (model 1). All six dimensions are separately and significantly related to lower life expectancy in neighborhoods. The strongest relationship occurs for the measure of the eviction rate, as a 1-SD increase in this measure is associated with a −.27 year decrease in life expectancy. A similar increase in bars and liquor stores is associated with a −.26 year decrease, and life expectancy also decreases with higher denial rates of home conventional loans (−.14), fringe banks (−.11), vacant units (−.07), and toxic waste sites (−.07). However, this disaggregation does not improve model fit based on the BIC compared to Models 3 or 4 in Table 4.
Model 2 tests interactions between the disaggregated neighborhood blight measures and the percentage of Black residents in the neighborhood. Three of the measures show stronger negative relationships with life expectancy when accompanied by a higher percentage of Black residents in the bars and liquor stores, the eviction rate, and vacant units. Only toxic waste sites show a weaker relationship in neighborhoods with a higher concentration of Black residents.^6^ This model has the best model fit of all estimated models based on the BIC values.
We tested several ancillary models to assess the robustness of our results. We present these results in the Supplemental Materials. We aggregated our structural racism measures to tracts rather than egohoods, and the model fit was always inferior when aggregated to tracts, and the size of the effects was smaller. Likewise, if we used the continuous versions of our variables, rather than constructing three-level ordinal variables that emphasize the presence of blight or racial disparity, the results were again consistently weaker. We also compared our strategy to the common one using the Black/White mortgage denial ratio variable, and found that this variable never had a significant negative relationship with life expectancy when added to our models. Finally, we assessed and consistently found that our measures have stronger relations with life expectancy in neighborhoods with a larger Black presence. In all cases, the optimal model was the final model we presented in Table 5, which disaggregated the neighborhood blight measure and allowed these measures to interact with the percent Black.
This study explored the consequences of structural racism measured at both macro- and meso-scales for period life expectancy. Whereas studies sometimes measure neighborhood structural racism based on a single dimension, our approach combined many sub-dimensions of structural racism into two broader dimensions—racial disparity and neighborhood blight—to create richer, more comprehensive, and multi-dimensional measures of neighborhood structural racism. Furthermore, moving beyond studies that measure neighborhood structural racism simply based on aggregations to census geographic units such as tracts, we created structural racism measures based on egohoods around every block in a tract. We found that our more spatially precise approach provided stronger results than did simply aggregating to census tracts and assuming that tracts capture the relevant context. This finding highlights the need to consider how structural racism might operate at the meso-scale when creating measures, and that failing to do so can result in misunderstanding the scope of structural racism or lead to biased assessments of intervention strategies. Simultaneously, we included a measure of Black/White disparity at the county level in the models and found that it was also consistently associated with lower neighborhood life expectancy in these counties.
One key finding was that a more comprehensive measure of neighborhood structural racism provides stronger results relating to reduced life expectancy. Rather than measuring a single dimension, our measure combined several dimensions. We captured neighborhood blight with measures capturing the eviction rate, the presence of toxic sites nearby, bars and liquor stores, fringe banks, vacancies, and a high denial rate of home loans. These various dimensions combined into a robust measure predicting lower life expectancy in neighborhoods. Furthermore, when we included all of these dimensions simultaneously in the model, we found that all dimensions were significantly related to lower life expectancy. Thus, it appears that these various dimensions are all important.
Relatedly, we found that a combined index of all dimensions of our meso-level structural racism measures not only exhibited a strong negative association with life expectancy but also demonstrated an accelerating negative relation as structural racism increases. This suggests that greater concentrations of these various dimensions of structural racism have increasingly negative consequences for life expectancy. The implication is that these dimensions do not simply provide an additive effect, but rather that presence of many dimensions of structural racism has particularly deleterious consequences. This finding underscores the need for public health strategies to simultaneously prioritize comprehensive neighborhood development initiatives that tackle multiple dimensions of structural racism to enhance overall health outcomes. For example, the Purpose Built Communities and Promise Neighborhoods initiatives prioritize developing interventions that are community engaged; are multilevel; and that target upstream factors such as housing conditions, education quality, and socioeconomic mobility. These interventions have shown positive effects on violent crime, education, employment, and housing outcomes, potentially reducing associated health inequities (Bailey et al., 2017).
We also highlight that our strategy for measuring meso-level structural racism used a more spatially explicit approach. Rather than simply aggregating to census tracts—and assuming that the tract is the relevant context for residents—we created buffers around every block to provide a better measure of the social context of residents. Even though we would have ideally had life expectancy also measured in egohoods, our strategy nonetheless yielded stronger results. In additional tests, we found that the smallest-sized buffers (1/4 mile radius) showed the strongest relation with life expectancy. This observation implies that the relatively nearby spatial environment has the most pronounced consequences for residents. Therefore, targeting interventions at the immediate neighborhood level may be the most effective strategy for improving life expectancy, as residents are profoundly affected by the conditions in their nearest surroundings.
In our conceptual model, we posited that macro-level structural racism might also impact neighborhood life expectancy through meso-level mechanisms not in our model. We tested this with a measure of county-level Black/White disparity, and found that it had a negative association with life expectancy. These results are consistent with our earlier comments on the need for scholars to consider the various spatial scales at which structural racism operates. Whereas neighborhood Black/White disparity was weakly related to general life expectancy in our models, one could argue that this spatial scale is too small to capture such processes effectively. As one example, a county with a strong discrepancy in Black/White denials of home loans likely would be indicative of structural racism. However, a neighborhood with such a strong discrepancy could instead indicate a White-dominant neighborhood where the loan applications of potential Black residents are systematically denied to prevent the in-movement of this group. Thus, such a measure has a very different meaning depending on the spatial scale (Mendez et al., 2013). Indeed, we found that Black/White disparity had the strongest relationship with life expectancy when measured at the macro-scale of counties. As one final point, we also tested whether neighborhoods high in neighborhood blight in counties with high Black/White disparity experienced more pronounced consequences for average life expectancy. However, we found no such effects in ancillary cross-level interaction models.
We acknowledge two limitations to this study. First, the cross-sectional nature of our data precludes causal interpretations of our results. Second, we would have preferred to have race-specific outcome measures, but data limitations prevented this. Thus, we must be careful in our interpretations of the relationship between our structural racism measures and the average life expectancy outcome. There may be racial differences in life expectancy at the neighborhood level in locations with higher levels of neighborhood racial inequality, but we were unable to test this more generally. Third, we were not able to measure our outcome measure of period life expectancy in egohoods, which would have been preferable. We believe that this limitation simply weakened our detected results. Finally, some of our meso-level racial disparity measures were only available in block groups or tracts, limiting our ability to aggregate them to egohoods. Again, we suspect that this may have weakened the size of these effects.
In conclusion, we have highlighted the importance of accounting for how structural racism can operate at multiple geographic scales. Although existing studies using macro units have clearly provided considerable insights, our results highlight that the consequences of structural racism do not manifest uniformly across the spatial landscape. Indeed, macro studies do not assume this, as studies using racial segregation as a measure of structural racism presume that the consequences will be felt more strongly in neighborhoods predominantly occupied by Black residents. Our results, however, highlight that there are differences among these meso-level units as well. In part, this is due to the level of neighborhood blight that these locations experience, which is likely at least in part a function of larger macro processes. Furthermore, this is also due to spatial processes that do not conform to simple neighborhood definitions, and therefore arguably exist based on a more spatially precise pattern in which spatial buffers may better capture such effects, as we found.
Indeed, we have demonstrated that structural racism has negative consequences for period life expectancy in neighborhoods. We measured structural racism at both the meso-level (using egohoods) and the macro level (using counties). A key insight is that structural racism based on Black/White disparity appeared to have negative consequences for life expectancy when measured at the macro level, but less so at the meso-level. Instead, it was neighborhood blight, measured across six dimensions, that was negatively associated with average life expectancy. Thus, although structural racism has important consequences, scholars need to carefully conceptualize their measures of this construct across various geographic scales.