Explaining the Spatial Patterning of Racial Disparities in Traffic Stops Requires a Structural Perspective: Further Reflections on Stelter et al. (2022) and Ekstrom et al. (2022)

Recent studies have made clear that Black drivers in the United States are more likely to be stopped by police than White drivers and that the size of the disparity varies widely from one place to another. Now, two research teams have independently analyzed the same large data sets to investigate the role of racial bias in explaining the geographical patterning of racial disparities. The two teams’ findings are illuminating, both in their agreements and in their differences. First, the agreements. Both Stelter and colleagues (2022) and Ekstrom and colleagues (2022) found that counties with higher average levels of racial bias tend to have larger racial disparities in traffic stops. Both teams found that the association is significant for both explicitly measured and implicitly measured racial bias. And both teams found that disparities are higher in counties with larger White populations. As both teams note in different ways, these county-wide associations are unlikely to be explained by “a few bad apples” responsible for stopping Black drivers. Instead, these findings call for explanations at the level of communities. But what, exactly, explains these effects?

Here the teams diverge in the questions they ask. Stelter and colleagues contrasted (a) measures of stereotypes linking Black Americans to weapons with (b) measures of evaluations linking Black and White people with “good” and “bad” in general. If disparate traffic stops resulted from stereotypes regarding disproportionate involvement of Black Americans in crime, they reason, then the stereotypes measure should be a stronger predictor than the evaluation measure. But it was negative evaluations of Black Americans that proved the more robust predictor. Ekstrom and colleagues drew a different contrast. They focused on evaluation measures and compared (a) the extremity of racial biases within racial groups with (b) the prevalence of racial bias between groups. They found that more extreme biases among White participants (or any other racial group) were not associated with more disparate traffic stops. Instead, they suggested that the associations are explained by the proportion of a county’s population that is White, combined with the fact that White participants display higher anti-Black biases on average than other groups.

Here, we come to a substantive disagreement, because Stelter and colleagues found that the extremity of racial bias among White respondents was significantly associated with disparate stops, independently of county racial demographics. A second disagreement concerns the shared and unique contributions of explicit and implicit measures. Stelter and colleagues found that when both were in the model, explicit measures remained significant predictors but implicit measures did not. Ekstrom and colleagues, in contrast, found that implicit measures were the more consistent unique predictors.

These differences raise the question of which analytical choices may explain the discrepancies. In fact, the two teams made a lot of different choices. For example, Stelter and colleagues analyzed only the racial biases of White respondents, whereas Ekstrom and colleagues’ main analysis measured racial biases among all groups and used poststratification to weight the estimates to reflect the demographics of each county. The two teams also calculated their dependent variables differently. Stelter and colleagues computed the difference between (a) the percentage of Black drivers stopped out of all drivers stopped and (b) the percentage of Black residents in a county out of all residents in a county. Ekstrom and colleagues, in contrast, computed the difference between (a) the number of Black drivers stopped divided by the total Black population in the county and (b) the number of White drivers stopped divided by the total White population in the county. These different ways of measuring racial disparities might lead to different results.

In reflecting on these two articles, we can see agreement on the general finding of an association between racial-bias measures and racial disparities in police stops, but we can see disagreements when the authors partition variance in ways that might speak to causal pathways. Some of this is easily resolved. A diligent analyst can (and should!) systematically localize the differences between analytic models. But the hard—and crucial—part relies not just on statistical models but also on theory. Neither team lays out an explicit causal theory that guides their analytic choices. Both teams indicate that the results suggest something about communities, but what is it? Answering that question may require going beyond these two data sets, and it provides the opportunity to collaborate with sociologists, policy and legal scholars, or historians. Many of today’s racial disparities can be traced to structural inequalities. We believe it is important to consider the structural factors associated with both the predictors (implicit and explicit bias, population demographics) and outcomes (policing).

The finding that racial disparities in police stops are higher in counties with larger White populations is provocative, but it raises the question of why these areas may have higher White populations in the first place. In the southern United States, modern Black populations can be traced in part to the legacy of slavery, and in this region, both modern Black populations and historical enslaved populations are associated with greater racial bias among Whites (Acharya et al., 2016; Payne et al., 2019). In the northern United States, racial demographics are tied to the Great Migration, as Black Americans left southern farms for northern industrial cities throughout the 20th century (Logan et al., 2015). And in the western United States, small Black populations sometimes reflect the absence of both slavery and widespread migration by Black Americans (Tolnay, 2003). The demographic patterns revealed in these articles might have different historical meanings across different regions.

In some cases, large White populations reflect residential segregation. The continued presence of highly segregated White enclaves can be explained by the legacy of structurally discriminatory housing policies (e.g., redlining) as well as institutionally discriminatory real-estate practices (e.g., racial steering, “block busting,” restrictive covenants) and even outright violence (e.g., Massey & Denton, 1993). Understanding the structural factors that led to the racial demographics of the counties surveyed is important to understanding what these demographic associations mean.

Structural racism shapes not only population demographics but also the nature of policing. Indeed, some of the earliest predecessors of organized community policing in the Americas were the slave patrols found throughout the European colonies (Alexander, 2010). After the end of the American Civil War and the failure of Reconstruction in the south, “Black Codes” were designed to criminalize, and thereby control, formerly enslaved Black Americans. Vagrancy laws, for example, forced Black Americans to sign labor contracts or be imprisoned at hard labor. The criminalization of recently emancipated Black Americans in the south can be tied to regional efforts to secure sources of manual labor, such as convict-leasing programs (Blackmon, 2008). In northern industrial cities, the criminalization of Black Americans at the turn of the 20th century was part of the ruling elite’s efforts to maintain control of their workforce, which largely consisted of poor European immigrants and Black migrants (Muhammad, 2010). Across various time periods, including the current day, the hampering of marginalized racial groups with various fines and fees has been used to support municipal budgets (e.g., Shoub et al., 2021).

Many of these historical and contemporary practices can be quantified, and some have been already. Stelter and colleagues and Ekstrom and colleagues have identified a compelling association that cries out to be understood better. Taking up that challenge will require not only additional statistical work but also a broader perspective on psychological theory and history to make plain how our shared past is shaping choices made now.