Spatiotemporal Changes in the Slavery–Inequality Relationship: The Diffusion of the Legacy of Slavery

Abstract

Despite the persistence of relationships between historical racist violence and contemporary Black–White inequality, research indicates, in broad strokes, that the slavery–inequality relationship in the United States has changed over time. Identifying the timing of such change across states can offer insights into the underlying processes that generate Black–White inequality. In this study, we use integrated nested Laplace approximation models to simultaneously account for spatial and temporal features of panel data for Southern counties during the period spanning 1900 to 2018, in combination with data on the concentration of enslaved people from the 1860 census. Results provide the first evidence on the timing of changes in the slavery–economic inequality relationship and how changes differ across states. We find a region-wide decline in the magnitude of the slavery–inequality relationship by 1930, with declines traversing the South in a northeasterly-to-southwesterly pattern over the study period. Different paces in declines in the relationship across states suggest the expansion of institutionalized racism first in places with the longest-standing overt systems of slavery. Results provide guidance for further identifying intervening mechanisms—most centrally, the maturity of racial hierarchies and the associated diffusion of racial oppression across institutions, and how they affect the legacy of slavery in the United States.

Introduction

Black–White inequality persists throughout the United States in part because of its foundations in slavery. Research has repeatedly demonstrated a geographic link between historical reliance on enslaved labor and subsequent outcomes—namely, greater White advantage and greater Black disadvantage (e.g., Gabriel et al. 2021; Gottlieb and Flynn 2021; Price et al. 2008; Reece 2020; Reece and O'Connell 2016; Ward 2023); however, there is evidence that this place-based link has eroded over time (O'Connell et al. 2020; also see Gabriel and Tolnay 2017). The concentration of higher Black–White inequality in places with strong historical attachments to slavery has diminished. We test hypotheses concerning the temporal and spatial patterns of the changing slavery–inequality relationship to generate empirical evidence that advances theories on the long reach of historical forces (for a recent call to attend to the mechanisms perpetuating contemporary legacies of historical racism, see Cunningham et al. 2021). Two interlinked questions focusing on explaining county-level economic vulnerability (i.e., sharecropping and poverty) guide our research: when did the legacy of slavery relationship change in the U.S. South, and what was the spatiotemporal pattern of change among the Southern states.

To answer these questions, we examine county-level data from 1900–1930 and 1970–2018 and employ integrated nested Laplace approximation (INLA) models that simultaneously account for spatial and temporal dynamics (see also Curtis et al. 2019). Our outcomes reflect economic vulnerability relevant to the selected study periods: sharecropping (1900–1930) and poverty (1970–2018). We treat these outcomes as indirect measures that are indicative of institutionalized racism. Our INLA models allow the county-level slavery–inequality relationship to vary across decades and states, which provides empirical evidence on when and where the relationship changed among counties within the U.S. South. Results provide new knowledge about the historically rooted institutionalized dimensions of Black–White inequality by demonstrating that the slavery–inequality relationship declines in magnitude at different paces depending on the duration of slavery. The spatial pattern of temporal changes suggests that the decoupling of institutionalized racism from historical concentrations of enslaved labor occurred first where overt systems of slavery had existed the longest. We draw on these findings to advance theory on the persistence of Black–White inequality in the United States and to encourage more engagement with the spatial and temporal dynamics of demographic processes.

The Historical, Institutionalized Roots of Black–White Inequality

We conceptualize the legacy of slavery as a contemporary feature of places that manifests as institutionalized racism rooted in the system of chattel slavery. Institutionalized racism is a type of structural racism that manifests across myriad social institutions (e.g., education, the criminal legal system, and politics/government) (see also Baker and O'Connell 2022). Our attention to institutionalized racism emphasizes the centrality of laws and policies that explicitly or implicitly advantage Whites and discriminate against Blacks when explaining Black–White inequality. Examples include the creation of “White flight academies,” or private schools catering to Southern Whites who wanted to escape desegregation in public schools (Andrews 2002; Porter et al. 2014), different laws for crack and powder cocaine (Sklansky 1995), and voter disenfranchisement (Alexander 2010).

Institutional racism does not exclusively stem from the legacy of slavery. Indeed, there are other events that have spurred on the institutionalization of racism (e.g., the Great Migration). However, there are elements of contemporary U.S. institutions that can be traced back to the institution of slavery, thereby mirroring and upholding White supremacy across numerous social domains. The ideology used to justify racial slavery relied on ideas about the “natural” or biological inferiority of specific racialized groups and the superiority of specific others, which gained public support through pseudoscience (Chase 1997) and was implemented into public policy (Pascoe 2010). Slavery was a legalized system of exploitation that necessitated the institutionalization of a racist ideology based on an assumed White superiority and Black inferiority (Du Bois 1903; Omi and Winant 2014). History is embedded in place through policy and social organizations and perpetuated over time by the repeated reliance on the past to build the future (Molotch et al. 2000).

Research provides consistent evidence of the link between historical slavery and contemporary outcomes, especially when considering different aspects of Black–White inequality (e.g., Gottlieb and Flynn 2021; Kramer et al. 2017; O'Connell 2012; Reece 2020; Reece and O'Connell 2016; Ruef and Fletcher 2003; Vandiver et al. 2006). Across the board, results suggest worse outcomes for the Black population and/or better outcomes for Whites in places with stronger connections to historical slavery. We observe similar evidence from studies that substitute slavery with subsequent forms of oppression (e.g., lynching, anti-miscegenation laws) (Bratter and O'Connell 2017; Gabriel and Tolnay 2017; Jacobs et al. 2005; King et al. 2009; O'Connell et al. 2022) and that combine multiple forms of racist oppression into a single measure of the “historical racial regime” (Baker 2022). Critically, in-depth studies of Black culture (Johnson 1934) and poverty (Duncan 1999) in communities with strong historical ties to slavery support the theoretical perspectives developed in large-scale legacy of slavery studies. Despite the preponderance of evidence, the local connection between institutionalized racism and inequality identified in legacy of slavery research is not necessarily static, permanent, or deterministic. In the following sections, we articulate the theoretical foundations we use to motivate our analysis of temporal and spatial change in the strength of the county-level, slavery–inequality relationship.

Changes in the Legacy of Slavery

Temporal Dimensions of Change

Gabriel and Tolnay (2017) argued that historical legacies are more or less likely to persist depending on the extent of local social “resistance” to their transmission over time. Conditions that reinforce race-based thinking, such as support for explicitly racist political candidates, represent low resistance and may facilitate the transmission of racist legacies over time. In contrast, they argue that population processes, such as migration and instability, disrupt the link between historical and contemporary contexts.

The disruption resulting from population dynamics would sever place-based ties between the historical institution of slavery and contemporary inequality, thereby weakening local connections over time. Yet, concurrently, population dynamics could geographically spread the legacy of slavery by diffusing White supremacist ideology via migration. In a recent study, O'Connell et al. (2020) found that the slavery–inequality relationship is weaker in counties that experienced White population increase in the early 1900s than in other counties. Observed levels of Black–White poverty inequality suggested that the system of institutionalized racism did not break down among Southern places within which enslaved populations were concentrated. Rather, the legacy of slavery grew as it detached from the centers of enslaved labor exploited by Whites. Relatedly, Baker (2022) suggested that the institutionalized racism associated with slavery expanded beyond this singular racist institution to create and maintain local “historical racial regimes” (for a broader argument regarding shifts in structural racism, see Bonilla-Silva 1997).

These recent studies suggest that the expansion of institutionalized racism occurs across institutions and places (Baker 2022; O'Connell et al. 2020). We draw on this perspective to inform our investigation of the spatiotemporal dynamics of the slavery–inequality relationship. A key theoretical piece on which we draw is the role of migration in changes to the slavery–inequality relationship (e.g., Gabriel and Tolnay 2017; O'Connell et al. 2020). We are unable to empirically examine migration given current data limitations. Yet, we rely on previous research to suggest that the diffusion of institutionalized racism via migration weakens the intensity of the slavery–inequality relationship within local contexts while expanding the reach of slavery's ideology across a broad sweep of institutions and places.

Spatial Dimensions of Change

Spatial variation in legacy effects is not clearly established in prior research. Although O'Connell et al. (2020) alluded to the importance of space by linking their evidence to a diffusion explanation, they do not empirically pursue this possibility, in part because of data limitations that preclude a direct assessment of migration processes. No other legacy studies have engaged the possibility of spatial variation in its temporal decline. However, given the fundamental role of space and place in shaping inequality-generating processes (e.g., Chetty et al. 2014; Connor et al. 2020; Curtis et al. 2012), we argue that identifying the timing and spatial patterning of change will aid in isolating the mechanisms contributing to changes in the legacy of slavery and, subsequently, contemporary Black–White inequality.

In this effort, we develop a theory of entrenchment on the basis of what would tie racial hierarchy to reliance on enslaved labor and, therefore, root it more deeply in economic institutions and place. Places where the linkage is less specific to the local economy would provide an opportunity for the ideology of White supremacy to pervade other, noneconomic institutions and diffuse across places (e.g., counties). We focus on state variation in temporal changes to the county-level slavery–inequality relationship. We expect the slavery–inequality relationship to be most persistent in states where historical conditions concentrated the local development of racialized boundaries within economic institutions. Conversely, we expect earlier declines in the strength of the slavery–inequality relationship in states where the racial hierarchy had already been integrated across institutions. We develop specific hypotheses by considering the broader state context.

Hypotheses

When Change Occurs

Drawing on previous research, we expect the slavery–inequality relationship to decline over time (see, especially, O'Connell et al. 2020). Moreover, we expect that significant declines will manifest after the 1950s and proceed gradually. We refer to this as the late decline hypothesis. This expectation is consistent with research focusing on Confederate monuments (O'Connell 2022) and with historical accounts of demographic and social processes occurring in the South. Specifically, the 1950s mark the beginning of a more visible and successful movement against racism in the United States (however, for arguments regarding the importance of earlier efforts, see Brown 2003; Jones 2010). This movement may have inspired greater and more widespread reliance on (subtle) institutionalized racism (e.g., increased reliance on private schools; see, especially, Andrews 2002; Brown 2010). As a result, racialized ideas that migrated with (White) people in the early decades of the twentieth century may have been reemphasized in later decades when (White) residents, as a collective, deemed it necessary to reassert racial dominance. Anti-racist movements initiated in the 1950s may have sparked racist countermovements, new legislation, and migration trends that ultimately coincide with the spread of the legacy of slavery (also see Andrews 1997; Santoro 2008).

An alternative perspective on the timing of the declining relationship has been developed in research examining the intergenerational transmission of socioeconomic status (e.g., literacy, occupation) among persons who were considered free and enslaved prior to emancipation. Sacerdote (2005) suggested a much earlier onset of decline in the legacy of slavery relationship after finding that the status distinction between “free” and “enslaved”—a key boundary maintained during the period of legalized slavery—disappears by the 1920s. We refer to this as the early decline hypothesis. The disappearance of the distinction between free and enslaved status indicates that some aspects of the historical institution declined faster than others, particularly when considering intraracial inequality. However, given our focus on Black–White inequality, the boundary between Black and White statuses is much more relevant, and the existing legacy of slavery literature suggests that this boundary is more persistent than the free/enslaved boundary. Consequently, we anticipate that the legacy of slavery will remain significant until recent decades despite initial declines midcentury.

Where Change Occurs

Our second set of hypotheses concern spatial variation in when the legacy of slavery changes over time. Within the theoretical framework we develop on entrenchment, two relevant forces stand out: the dominant cash crop and the maturity of the system of White supremacy (for a history on the differing development of slavery within the U.S. South, see Berlin 1998). We refer to these as the cotton-dominance hypothesis and maturity hypothesis, respectively. We note that although the characteristics we discuss (i.e., cotton dominance and institutional maturity) vary at the county level, we view the relevant dynamics as playing out within broader, state-level policy and economic regimes (e.g., the legality of slavery).

Cotton was among the most lucrative crops grown in the South and was concentrated in Alabama, Arkansas, Georgia, Mississippi, and Tennessee, as well as parts of Florida, Louisiana, and Texas. As a result of slavery's centrality to the cotton economy and the associated financial incentive to maintain enslaved labor, we expect the institutionalized legacy of slavery to be most entrenched and directly connected to the local concentration of enslaved people in states reliant on cotton. Consequently, according to the dominant crop hypothesis, we anticipate that the slavery–inequality relationship would first decline in non-cotton-dominant states (e.g., Virginia, North Carolina, and South Carolina), followed by middle-ground states (e.g., Florida, Louisiana, and Texas), and then finally reach the core of the cotton belt (e.g., Alabama, Arkansas, Georgia, Mississippi, and Tennessee).

However, the relative maturity of the systems of White supremacy may, instead, explain the spatial patterning of the change in the slavery–inequality relationship, where maturity refers to the duration of systems of White supremacy, which is distinct from economic dependence on enslaved labor (e.g., specific areas of Texas were highly dependent on slavery, but state-level structures were relatively new). In terms of a specific spatial pattern associated with the maturity hypothesis, we know slavery developed primarily in a northeast to southwest transverse as it spread throughout the U.S. South. Consequently, we might anticipate more sophisticated systems of oppression that span institutions in states like Delaware, Florida, Maryland, South Carolina, and Virginia, where recorded dates of institutionalized enslavement are the earliest among the Southern states (see Table A1, shown in the online appendix, along with all other tables and figures designated with an “A”).¹ The corresponding spatial pattern would show declines in the slavery–inequality relationship starting in the northeastern part of the South and moving gradually to the states where Whites established systems of slavery much later (e.g., Alabama, Mississippi, and Texas). We suspect that there are two notable exceptions to this broad spatial pattern on the basis of state-specific histories. First, although Louisiana is geographically part of the Mississippi Valley, it was first colonized by the French, who developed formal systems of coercive control prior to its neighbors (i.e., Mississippi and Texas). Second, Florida was first colonized by the Spanish and did not become a state until 1830, yet slavery existed in some areas long before statehood (see Berlin 1998).

In considering these spatial hypotheses, our methodological approach does not require strict definitions of the different subregions, because we focus on relationship variation across states (discussed further below). Instead, we use the historical factors discussed above as general guides that allow for inevitable variation within the identified subregions. Moreover, although the overall patterns suggested by the dominant cash crop and relative maturity of slavery are similar, we pay particular attention to the timing of change in the relationship for Tennessee and Texas because they occupy contrasting positions. We would expect Tennessee to experience a late decline on the basis of the cotton-dominance argument, but an early decline on the basis of the maturity argument. The opposite applies to Texas: we would expect a midperiod decline to align with the cotton-dominance argument, and a late decline on the basis of the maturity argument.

Data

Census Data

To examine changes in the slavery–inequality relationship, we draw on county-level decennial census data from 1860 and all decades with available data from 1900 until 2000, plus five-year period estimates for 2014–2018 from the American Community Survey (ACS). Data on relevant outcomes by race are unavailable for 1880, 1890, 1940, 1950, and 1960.² We retrieve data from the IPUMS National Historical Geographic Information System unless otherwise noted (Manson et al. 2020). We adjust the 1860 data to reflect the county boundaries used in each of our analyses (see Reece and O'Connell 2016; Slez et al. 2017).³

We focus on counties in the U.S. South (excluding Oklahoma because of data availability in 1860, prior to its statehood). The county is an appropriate unit of analysis because we focus on the persistence of local institutional manifestations of the legacy of slavery and subcounty data for specific towns and cities are not available. Moreover, the county is commonly used in studies of racist legacies (e.g., Gabriel and Tolnay 2017; O'Connell 2012) and this geographic scale is linked to social institutions that are relevant to our analysis, including local governments and school districts, particularly when focusing on the U.S. South (Irwin 2007; see also Petersen and Ward 2015; Reece and O'Connell 2016). The availability of historical population data at the county level and the links between counties and policy implementation make it the best available option for the local dimension of our study. As alluded to previously, we use states as a second scale to examine spatial variation linked to state-level policy and economic development.

Variables

There are no consistent outcomes available for all decades in our study period that are also reported by race. This data limitation generates some challenges when interpreting the overall trend in how the slavery–inequality relationship changes over time. We examine conceptually comparable measures to overcome data constraints, a strategy consistent with a robust literature examining trends between or over historical periods (e.g., see Curtis White 2005; Curtis White et al. 2005; Eichenlaub et al. 2010; Eriksson 2019). Our primary focus centers on economic vulnerability.

We capture economic vulnerability from 1900 to 1930 using farm tenancy data to indicate whether a farmer is an owner, partial owner, or tenant. We estimate the “share tenants” rate among Black farmers by dividing the total number of cash-croppers and sharecroppers by the number of Black farmers in a county;⁴ we do the same for White farmers. Share tenants were people who agreed to work land owned by someone else for a share of the profits and/or crop yield. This was the most tenuous position a person could occupy, both economically and socially (see Johnson 1934). Moreover, this economic arrangement was part of the sharecropping system that some argue served as an institutional replacement for (i.e., legacy of) slavery (see, especially, Blackmon 2008). Consequently, Black–White differences in the concentration of “share tenants” is a particularly important outcome for this study.

We use official poverty estimates from the census once they become available by race at the county level in 1970. The official poverty threshold is adjusted each decade for inflation, (some) changes in the cost of living, and the number of members in a household, which makes it comparable over time.⁵ We derive poverty rates by dividing the number of individuals living in a household with an income below the official poverty threshold by the total number of noninstitutionalized persons for whom poverty status is determined.⁶ We estimate two rates: one for the Black population and one for the non-Hispanic White population.⁷

In supplemental analyses, we examine illiteracy by race for 1900 to 1930, and educational attainment starting in 1970. We construct illiteracy rates for the Black and White populations aged 10 or older for the first period and create estimates of “low” educational attainment using decade-specific, regional average attainment levels as our threshold. We identify the average level of educational attainment using the category into which the median case falls. This approach to estimating educational attainment provides a substantively comparable value while acknowledging changes over time in an “employable” level of education. For 1970–2000, the average level of educational attainment is a high school degree; for 2014–2018, the median falls within the “some college” category. As a result, to reflect “low” educational attainment in 1970 through 2000, we focus on those with less than a high school degree, whereas in 2014–2018, we focus on those with a high school degree or less. The universe for the educational attainment variables is the population aged 25 or older. We use these supplemental results to confirm that the general pattern identified using economic vulnerability (i.e., tenant farming and poverty) holds for another indicator of socioeconomic disadvantage.

We measure Black–White inequality using the difference between the rates for the two groups, Black minus White (see also O'Connell et al. 2020). We conduct sensitivity analyses using a ratio measure of the Black poverty rate divided by the White rate to assess the robustness of our findings. The substantive conclusions regarding temporal and spatial variation in changes to slavery–inequality relationship are similar based on those results. All inequality outcomes are measured so that higher values reflect greater Black–White inequality to the advantage of Whites.

We approximate local historical attachments to slavery using the concentration of enslaved people in 1860 (e.g., see also Acharya et al. 2016; Reece and O'Connell 2016; Ward 2023). Larger proportions of people who were enslaved within the total population reflect a greater reliance on coerced labor and, arguably, a place's stronger investment in the social institutions that supported slavery. More invested places are expected to have stronger institutional legacies because the racist ideology of slavery would have been a defining socioeconomic foundation. We depict the concentration of slavery among Southern counties in Figure 1 using contemporary boundaries. It is evident from this figure that reliance on enslaved labor in 1860 was concentrated in “Deep South” states (e.g., South Carolina, Georgia, Alabama, Mississippi, and Louisiana), but also in Virginia and North Carolina. More importantly, there is substantial variation at the county level even within these states. County-level variation in the concentration of the enslaved population suggests a local institutional imprint and not a regional- or even state-level phenomenon. We emphasize that the legacy of slavery plays out across local areas but may still be structured by processes occurring at the state level (e.g., state law), which motivates our investigation of spatial variation in the changing slavery–inequality relationship among counties across southern states.

We omit control variables from our analysis because we aim to identify changes in the slavery–inequality relationship and not to isolate a legacy effect exclusive of other forces. The unique contribution of the legacy of slavery to Black–White inequality has been well established in the literature (e.g., Acharya et al. 2016; Kramer et al. 2017; O'Connell 2012; Reece 2020; Reece and O'Connell 2016).⁸ Our approach advances research in this area by leveraging sophisticated modeling strategies to empirically identify the spatiotemporal pattern of the slavery–inequality relationship. Specifically, we establish where, when, and how the relationship changed across place and over time.

Methods

We begin with a time trend reflecting the regional county average to provide a summary measure of Black–White inequality in the U.S. South. Additionally, we use several maps to demonstrate descriptively how the spatial patterning of Black–White inequality in the region changed over the study period. We use 1900, 1970, and 2014–2018 to exemplify the changes. Maps of share tenant/poverty inequality for all decades are available at our project website: https://apl.wisc.edu/shared/changes-in-legacy.

Our primary statistical strategy is INLA models. These models are designed to simultaneously account for spatial and temporal autocorrelation, a statistical necessity when examining spatial panel data and a conceptual match with our research questions (Curtis et al. 2019; for an alternative modeling strategy, see Curtis et al. 2013). To test our first set of hypotheses regarding temporal change in the slavery–inequality relationship, we allow the legacy of slavery relationship to vary over time while testing for statistical differences in the magnitude of the coefficients. This approach permits us to assess when, on average, the magnitude of the local relationship begins to change in the region. To test our second set of hypotheses concerning spatial differences in the temporal change, we allow for simultaneous temporal and spatial variation in the relationship to examine the spatial patterning of changes in the magnitude of the slavery–inequality relationship. This approach enables us to identify where the magnitude of the local relationship changes.

We provide additional details on each model and the multiple pairwise comparison tests we conduct in the following subsections; however, we offer a brief discussion of our use of statistical significance first. Given that INLA is a Bayesian model, we assess the significance of the parameters by examining whether the zero value falls within a Bayesian credible interval. For simplicity of presentation, however, we borrowed the relationship between a p value and a confidence interval from the frequentist perspective. In our study, we use “p value < α” as shorthand for “the zero value falls outside a (1 – α) × 100% credible interval,” where α = .05, .01, or .001.

Model 1: Legacy of Slavery Over Time

We begin by assessing when the magnitude of the slavery–inequality relationship begins to change. For spatial unit $i=1,...,N$ and time point $t=1,...,T$⁠, let $Yit$ denote the response variable at the $i$th spatial unit and the $t$th time point. The tenant farming concentration data consist of 920 counties (⁠$N$ = 920) in 15 states (S = AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, SC, TN, TX, VA, and WV) for four decades (⁠$T$ = 1900, 1910, 1920, and 1930); $Yit$ is the racialized difference in tenant farming concentration at $(i,t)$⁠, $t=1$ is year 1900, $t=2$ is year 1910, and so on. The poverty rate data consist of 1,077 counties (⁠$N$ = 1,077) in 15 states (S = AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, SC, TN, TX, VA, and WV) for five “decades” (⁠$T$ = 1970, 1980, 1990, 2000, and 2014–2018); $Yit$ is the racialized difference in poverty rates at $(i,t)$⁠, $t=1$ is year 1970, $t=2$ is year 1980, and so on.

To address the spatial and temporal dependencies inherent in these data, we consider the Besag model with a second-order queen neighborhood structure for spatial autocorrelation, and the AR(1) model with time lag 1 (10 years) for temporal autocorrelation with a spatiotemporal model as follows:

where $xi$ denotes the concentration of enslaved people (in 1860), and $αt$ and $βt$ are the corresponding intercept and regression coefficient at time t, respectively. $τt$ and $ϵi$ are the AR(1) temporal autoregressive effect and the Besag spatial autoregressive effect, respectively, and $γit$ is the remaining statistical noise.

Model 2: Legacy of Slavery Across States Over Time

Next, we assess where the pace of change in the magnitude of the local relationship between slavery and inequality differs. For all counties, we derive state-specific estimates of the slavery–inequality relationship based on the model as follows:

where $αtS$ and $βtS$ are the intercept and regression coefficient for the state S at time t, respectively, and 𝟙${⋅}$ is the indicator function. For example, to estimate the focal relationship for counties in Alabama at t = 1 (year 1970), we draw from the following model:

We approximate the posterior means of the parameters in (1) and (2) using INLA (Rue et al. 2009), which is readily available for use in R software (R-INLA).

Multiple Pairwise Comparison

In Eq. (2), we can estimate the slavery–inequality relationship for counties during a specific time and located in a specific state from the posterior distributions of $βtS$⁠. Furthermore, we are interested in comparisons of the county-level relationship in two states for a given time and between two decades for counties in a certain state. For example, we can consider a total of six pairwise comparisons of the slavery–inequality relationship between years in Alabama as:

• Comparison 1: $β1AL$ vs. $β2AL$⁠,

•  . . . 

• Comparison 5: $β2AL$ vs. $β4AL$⁠, and

• Comparison 6: $β3AL$ vs. $β4AL.$

Similarly, a total of 105 pairwise comparisons of the slavery–inequality relationship between states in year 1900 can be analyzed as:

• Comparison 1: $β1AL$ vs. $β1AR$⁠,

•  . . . 

• Comparison 104: $β1TX$ vs. $β1WV$⁠, and

• Comparison 105: $β1VA$ vs. $β1WV.$

For the pairwise comparison, we approximate the posterior distribution of the difference between two slavery relationships (e.g., $β1AL−β2AL$⁠, $β1AL−β1AR$⁠) estimated in INLA and apply the Bonferroni correction for multiple comparisons (Dunn 1961).

Results

Baseline Patterns of Inequality

Black–White inequality in economic vulnerability declined somewhat in the early period, but more noticeably between 1970 and 2014–2018 (Figure 2).⁹ The average U.S. Southern county had a level of Black–White inequality in 1900 of 23, indicating a Black disadvantage of 23 percentage points in the percentage of residents working as tenant farmers. This decreased to 17 by 1930, representing a 6-percentage-point decline. Average poverty inequality for the more recent period registers higher at a 27-percentage-point difference to the advantage of Whites in 1970 and declines much farther to 15 in the 2014–2018 period, for a decrease of 12 percentage points. The standard deviations (reflected by the dashed lines) suggest substantial variation in the extent of county-level Black–White inequality in the South. We are centrally concerned with the spatial variation in these temporal changes.

We visualize the associated spatial variation in Figures 3–5, showing county-level inequality using decade-specific quantiles. In 1900, inequality was lowest in counties on the fringes of the Southern region, as indicated by the lightest gray shading. Inequality progressively intensifies deeper into the region, as indicated by darker shading. This spatial pattern closely overlaps with the concentration of enslaved people shown in Figure 1.

The distribution of Black–White inequality in roughly the middle of the study period (1970) is remarkably similar to that seen for 1900 (Figure 4).¹⁰ We note a slight shift in the intensity of inequality between the early and middle periods farther south, where the highest levels of inequality in the decade concentrate (shown in darker gray). Still, the core pattern remains the same throughout the early and middle periods. Critically, the block of high-inequality counties breaks apart by the 2014–2018 period (Figure 5). Pockets of high inequality in North Carolina, South Carolina, and Georgia are diminished, and in the case of Texas, they are dispersed to the west.

Results show a clear change in the spatial distribution of the highest levels of Black–White inequality even as the average level of inequality declined (see Figure 2). Equally clear, some states more noticeably experienced this change than others. We turn to our empirical models to assess whether changes in inequality over time translate into changes in the slavery–inequality relationship (Hypothesis 1) and whether there is state-level variation in that temporal trend (Hypothesis 2).

Temporal Variation in the Slavery–Inequality Relationship

In Figure 6, we provide the information to evaluate our first hypothesis regarding the timing of change. We show the value of the coefficient for tenant farming in the left-hand panel of the figure and poverty on the right. We indicate the 95% credibility/confidence interval in gray shading to indicate the statistical significance of the relationship in each decade. Recall that gradual declines in the slavery–inequality relationship starting after the 1950s would support the late decline hypothesis, whereas declines beginning around the 1920s would be consistent with the early decline hypothesis.

Consistent with our general expectations, we find a near continuous decline in the magnitude of the relationship between historical slavery and later Black–White inequality among Southern counties. Findings show a statistically significant relationship throughout the period, yet the magnitude of the association is nearly zero by 2014–2018. However, consistent with the early decline hypothesis, we find a significant shift in the magnitude of the slavery–inequality relationship prior to 1950. The initial decline is modest and statistically detectable only in the last decade of the early period, at which point multiple comparison tests indicate that the 1930 estimate is lower than in all previous decades. The decline in magnitude continues in the contemporary period, starting with a dramatic drop but more gradual decline thereafter. We find significant differences in the magnitude of the relationship for each “decade” pair except 1980 and 1990, and 2000 and 2014–2018. The specific decade distinctions are indicated by numeric superscripts in the figure. All told, the change in the slavery–inequality relationship for the region is steady after an initial drop, and well underway by 1930.

Spatial Variation in the Extent and Timing of Change

To address our second set of hypotheses, we examine state variation in changes to the slavery–inequality relationship over time. We anticipate that declines will begin first in the upper South, followed by the lower South, and finally reach the Mississippi Valley region either in line with state variation in the dominant crop during slavery and/or the developmental maturity of systems of slavery. We find general support for the maturity hypothesis with a few notable caveats. We primarily use visual representations in our discussion of results, but a table with coefficients is available in Table A2.

We identify three classes of temporal trends across the 15 states: states that “never” exhibited a significant slavery–inequality relationship, “declining” states that show a clear decline in the significance of the relationship (in addition to the magnitude), and “delayed decline” states that exhibit a decline at a later start date than observed for the region and longer persistence of significance. The categories we employ to describe the various observed spatiotemporal patterns are inevitably imperfect, as they can only capture broad trends. Yet, we rely on multiple points of evidence to classify the states and identify exceptions when appropriate.

The never states were unexpected; however, the gradual change in the timing of decline across the region is consistent with our understanding of how the legacy of slavery diffused at different paces depending on the broader state context. We map the three classes of temporal trends in Figure 7 to emphasize the spatial connections across the subregions associated with our second set of hypotheses.

The never states are shown in light gray and include Delaware, Maryland, South Carolina, Virginia, and West Virginia. They are concentrated in the northeastern portion of the region, except for South Carolina. We also show the temporal trend for Maryland as an example of how this relationship unfolds over time among the never states (panel a of Figure 8; additional figures are available at our project website: https://apl.wisc.edu/shared/changes-in-legacy). The most defining feature of this category is that the relationship is largely nonsignificant. Additionally, the multiple comparison tests indicate little to no statistical difference in the relationship between any of the decade pairs. These five states present a temporal pattern that is unanticipated by the literature on the legacy of slavery but could be understood in connection with the diffusion of the legacy of slavery to new places where sophisticated ways of institutionalizing racism had more time to develop. It is possible that the local legacy of slavery manifested in these states and subsequently diffused, but this trend is censored from our analysis because the diffusion occurred prior to our study period (i.e., pre-1900).

The next group of states demonstrate declining significance in the slavery–inequality relationship and include the medium-gray states of Florida, Georgia, Kentucky, Louisiana, North Carolina, and Tennessee. These states exhibit the longest, most gradual pattern of decline. North Carolina exemplifies this category and shows a statistically significant relationship well into the contemporary period, suggesting persistent local attachments to the institution of slavery (panel b of Figure 8; additional figures are available at our project website: https://apl.wisc.edu/shared/changes-in-legacy). Yet, there are also significant declines in the magnitude of this association. These states are primarily southeastern, except for Louisiana, and farther south than the never states (see Figure 7). This geographic pattern is consistent with the diffusion of racist structures on the basis of the maturity of slavery within a state. The inclusion of Tennessee in this category—as opposed to being among the states where the relationship was most strongly attached to local historical enslavement concentration—is especially suggestive of the maturity hypothesis.

Moving westward and farther inland, we reach the delayed decline states, namely, Alabama, Arkansas, Mississippi, and—most notably—Texas. These states are depicted in dark gray (see Figure 7). States characterized by a delayed decline maintain a relatively constant and significant slavery–inequality relationship even into the contemporary period. Multiple comparison tests indicate a significant difference in the magnitude of the coefficients for only Mississippi and only in 2014–2018 compared with 1970 (panel c of Figure 8; additional figures are available at our project website: https://apl.wisc.edu/shared/changes-in-legacy). We emphasize that while the trend appears to be similar visually between the declining and delayed states (e.g., North Carolina and Mississippi), the multiple comparison tests suggest distinct differences in the timing and extent of the decline. The location of these delayed states suggests that the legacy of slavery has remained contained to places with higher historical concentrations of enslaved people in the most western portion of the South, which is also where slavery and its associated systems of racialized oppression were least developed. This provides further evidence for explaining the transferability of the legacy of slavery on the basis of the maturity of local systems.

Finally, these findings also clarify understanding of our first set of hypotheses regarding the timing of change in the slavery–legacy relationship. Our models that allow for temporal and spatial variation in the relationship suggest considerable deviation in the extent and timing of decline. For some states (i.e., never states), there is no decline; for others, the decline is substantial, yet the timing of that decline occurs as early as 1930—consistent with the regional results—and as late as 2014–2018. No single event spurs the diffusion of the legacy of slavery, and regional results can be deceptive given that they only reflect an average of the disparate trends across states.

Supplemental Analyses: Black–White Inequality in Illiteracy and Educational Attainment

We provide evidence from supplemental models focusing on educational outcomes—namely, illiteracy and below-average educational attainment rates—to confirm that results are robust to the type of outcome examined. We find substantively similar results (available upon request). The regional analysis, again, suggests an initial decline in the legacy of the slavery relationship in 1930, followed by a gradual decline in the contemporary period (with the exception of the change between 2000 and 2014–2018, which is twice as large as any other difference between neighboring decades in the period; see Figure A1).

There are some differences regarding specific state patterns, but the overall trends remain the same: declines happen first in the northern and eastern sections of the South, and then move through the Deep South. Critically, while Tennessee shifts to delayed decline, which is more consistent with the spatial pattern of cotton crop dominance, the pattern for Texas continues to support an interpretation of the spatial variation on the basis of the maturity of the slavery system (figures are available at our project website: https://apl.wisc.edu/shared/changes-in-legacy). These results suggest that the observed changes manifest in multiple institutions rather than within the economic system alone. Further, these results buttress our general conclusions regarding the timing of changes and the importance of spatial variation in understanding the mechanisms that explain those changes.

Discussion

The underbelly of historical racist violence in the United States has been on full display in recent debates. Slavery is at the center of this violence when considering the Black American experience. Research demonstrates that this historical institution remains imprinted on contemporary society in many ways, including on spatial patterns of Black–White inequality (e.g., Baker 2022; Gabriel et al. 2021; Kramer et al. 2017; O'Connell 2012; Reece 2022). We know there are factors that temper the contemporary strength of local legacies of racist violence (Acharya et al. 2016; Gabriel and Tolnay 2017; O'Connell et al. 2020; Petersen and Ward 2015; Reece and O'Connell 2016); however, there is little to no empirical evidence on the timing of changes in the strength of these relationships or how changes might vary across the Southern region, which impedes our ability to understand the mechanisms of change proposed by previous research (i.e., migration; see Gabriel and Tolnay 2017; O'Connell et al. 2020). We advance knowledge regarding the legacy of slavery through an analysis of the temporal and spatial dynamics of the slavery–inequality relationship using INLA models.

In terms of our first set of hypotheses concerning temporal trends, we find a pronounced decline in the relationship between 1970 and 1980, and an earlier region-wide shift by 1930. This gradual decline is consistent with previous research on population change suggesting that early demographic changes did not result in immediate social change (O'Connell et al. 2020). It took time for Whites to alter the local institutions that ultimately shape the extent of White advantage and Black disadvantage; however, the initiation of the decline was earlier than expected. O'Connell (2022) only identified a significant decline in the magnitude of the enslaved concentration association starting in 1950 when explaining the construction of Confederate monuments. Our results for the region, instead, are more consistent with Sacerdote's (2005) analysis of economic distinctions between people who were enslaved and people who were “free” prior to the abolition of slavery. The timing of changes in the legacy of slavery—namely, its spread across places—appears to differ depending on the specific institution involved, which is also reflected in slight differences between our focal (i.e., economic security) and supplemental analyses (i.e., educational attainment).

Moreover, the initiation of this decline as early as 1930 suggests that the impetus for change was not squarely rooted in racial equity gains made during the Civil Rights Era. The Civil Rights Act of 1964 and Voting Rights Act of 1965 provided legal teeth for the anti-racist movement, yet efforts for racial equity were afoot far earlier and, in the South, decades before these monumental legislative achievements (Brown 2003; Jones 2010; Park 2020). The decades surrounding the turn of the twentieth century are marked with community organizing and direct-action protests. The naming and work of the NAACP with W. E. B. Du Bois also emerged in the early decades of the century, the period proclaimed “the Progressive Era.” We can understand the spread of the legacy of slavery among some Southern places in terms of these early civil rights efforts and their interplay with racist countermovements (e.g., Owens et al. 2015), legislation (e.g., Blackmon 2008), and migration patterns (e.g., Tolnay and Beck 1992; Tolnay at al. 2018). Moreover, we can understand the gradual, region-wide decline in terms of the ongoing civil rights movement in play throughout the nation's history. We anticipate that a variety of factors inform the exact timing of change in the legacy of slavery, yet we suspect anti-racist movements are a key force that initiates a longer chain of events (i.e., racist countermovements, legislation, migration) (see also Andrews 1997; Santoro 2008).

Future studies could assess the robustness of our findings regarding the timing of decline by creating outcome variables that are measured consistently over time, perhaps utilizing restricted-access microdata. Our outcomes are substantively comparable but are not the same measure. Therefore, we are unable to fully determine whether the early period of results would be the same had we been able to analyze poverty rather than sharecropping, for example. However, we expect that the broad trends would remain given the consistency of our conclusions across economic and educational inequality, and because we observe declines within each period using the same measure and not just between periods using different though comparable measures.

Critically, and in terms of our second set of hypotheses, we argue that there is meaningful spatial variation in when the slavery–inequality relationship changed. For some Southern states, there was never a consistent significant county-level relationship between historical enslaved populations and later Black–White inequality, at least starting in 1900. Other states exhibit a clear decline in significance, and one that roughly aligns with the regional average. Still, a third group of states shows a delayed decline compared with the regional average, and one most consistent with the late decline hypothesis.

The observed spatial patterning is most consistent with the argument that the maturity of the system of slavery explains spatial variation in the timing of the diffusion of the legacy of slavery. Specifically, places within the region with relatively younger connections to racial slavery experienced late-period declines in the slavery–inequality relationship. We argue that states where the infrastructure supporting slavery was older had more sophisticated forms of institutionalized racism that could be infused into new social domains by Whites more easily than “rawer” forms of racism that characterize newer systems of slavery (for research examining the intensity and continuity of “historical racial regimes,” see Baker 2022). As a result, we argue that the legacy of slavery declines through institutional and spatial diffusion in the upper South first and moves to the Deep South, where the diffusion process started much later. Given these results, any theory intending to explain changes associated with the legacy of slavery must contend with spatial variation.

Our interpretation of these results is consistent with previous research suggesting that White population change immediately following Reconstruction facilitated the spread of the legacy of slavery (O'Connell et al. 2020). Growing settlements of White populations necessitated the creation of additional institutions (e.g., schools, governments). The early stage of White development at which this was occurring would have opened the door for the institutionalization of the legacy of slavery in diffuse places. Our argument adds critical new insight to this theoretical development by suggesting that the eventual incorporation of the legacy of slavery into new places occurred sooner in states where institutionalized chattel slavery existed earlier. This insight concerning the South has parallels with racist conditions in the Northeast, where racism was necessarily more insidious after the abolition of slavery (see, especially, Clark-Pujara 2016; Melish 1998). As we direct our attention northward, conceptualizing the legacy of slavery as just one part of a broader “historical racial regime” (Baker 2022) would contribute to a more cohesive theory of institutionalized racism that persists throughout the United States and not just in the Southern states.

We suggest several areas of future research that address limitations of our analyses. We use the state to define our spatial regimes, yet states are not the only social space shaping the county-level slavery–inequality relationship. We encourage future work to explore other geographic scales. Specifically, our findings suggest that future work might investigate spatial regimes based on the historical development of slavery (see Berlin 1998). Although we suspect analyses at different scales will reveal more about the slavery–inequality relationship, we assert that states are a valuable starting point, especially given our broad, theory-generating objectives. Counties in the same state are united by a shared legal, political, and—to an extent—cultural context (Baker 2022). We likely capture much of that variation by focusing on states.

We also see a benefit to extending research on how White migration out of the South contributed to the spread of the institutionalized legacy of slavery throughout the United States (for a historical analysis of the social impacts of the Great Migration, see Gregory 2005). Analyses directly linking migration flows between counties would advance understanding of the diffusion process while also expanding the geographic scope of this literature. Similarly, research on legacies would benefit greatly from direct measures of the institutionalized racism that is associated with the legacy of slavery (Cunningham et al. 2021). Finally, international migration has received scant attention in discussions of migration and the legacy of slavery but could play an important role in shaping local understandings of race (e.g., Joseph 2015), as well as the composition of the local population and its implications for inequality (e.g., Hamilton 2019; Lieberson 1980).

Our study illustrates the potential for advanced spatiotemporal modeling techniques, which are still relatively new to the social sciences, to enhance demographic research investigating temporally and spatially motivated questions. We hope our application of the INLA approach prompts more demographic research at the spatiotemporal frontier. The INLA approach—rooted in a Bayesian perspective and born out of the mathematical sciences (Rue et al. 2009)—has become more widespread, but primarily in the biological and geospatial sciences. The recent introduction of an R-INLA package has facilitated uptake of this modeling approach. R-INLA is fast, free, well supported, and easily adaptable relative to other approaches, such as maximum likelihood estimation or Markov chain Monte Carlo models. Further, practitioners can find well documented instructions for each model and examples to follow on the website of R-INLA (http://www.r-inla.org/).

Our findings support the general argument that Black–White inequality in the U.S. South persists and is more deeply entrenched in some places than in others. The legacy of slavery has been a key force in places with some of the highest levels of contemporary Black–White inequality (net of other factors) (e.g., Kramer et al. 2017; O'Connell 2012; Reece 2022). Importantly, our study finds a regional decline in this geographic connection, in terms of magnitude and significance, and a dissipation in some states much earlier than in others. Certainly, declines in the local connection to slavery do not mean the legacy of slavery is irrelevant (see also O'Connell et al. 2020). We hope our findings prompt future empirical and theoretical attention to the pathways through which the legacy of slavery permeates across places, including those with indirect links to slavery, and the social, economic, and political institutions reinforcing racial hierarchies. The local manifestation of the legacy of slavery has declined, yet institutionalized racism in its various forms has by no means disappeared from the American landscape.

Acknowledgments

We thank the Applied Population Lab at the University of Wisconsin–Madison, and especially Rozalynn Klaas, for assistance with data preparation and website construction. We also thank the Population Association of America for the opportunity to present this research and receive helpful feedback from our peers. This research was completed with support from the Center (P2C HD047873) and Training Grants (T32 HD007014) awarded to the Center for Demography and the Center for Health and Aging at the University of Wisconsin–Madison.

Notes

References