Does Affirmative Action Litigation Reduce Police Killings of Civilians?


Can a lawsuit over who a police department hires change how often its officers kill civilians? Cox et al. (forthcoming) use federal death records for nearly 3,000 U.S. counties from 1960 to 2012, comparing 75 counties where a city police department faced an affirmative action hiring lawsuit between 1969 and 2000 with counties that did not. Before a lawsuit, police killed about 1.5 non-White civilians a year in the average litigated county. Over the long run, that number fell by about 40 percent, and by 2000 roughly 50 deaths a year were being averted. The decline took years to appear, and departments also became more diverse.
Why This Article Was Selected for The Policy Scientist
Police use of deadly force remains one of the most contested issues in American public policy. Black men are roughly 2.5 times as likely as White men to be killed by police, and reform debates keep returning to whether changing who serves in police departments changes how they police. The question is timely: the Supreme Court's 2023 decision ending race-conscious college admissions and the Justice Department's 2025 retreat from consent decrees with the Minneapolis and Louisville police departments have narrowed the legal and federal tools for changing police institutions. This article extends McCrary's influential research on court-ordered hiring quotas, moving from workforce composition to the most severe outcome of policing. The authors have published extensively on police violence, including work on protest, collective bargaining, and segregation. The staggered difference-in-differences event study, supported by newer estimators, placebo tests, and measurement-error simulations, is a credible causal design, although historical death records undercount police killings. The article is forthcoming in American Economic Journal: Applied Economics, a leading applied economics journal.
Full Citation and Link to Article
Cox, R., Cunningham, J. P., & Ortega, A. (Forthcoming). The impact of affirmative action litigation on police killings of civilians. American Economic Journal: Applied Economics.
Central Research Question
The study asks whether the threat of affirmative action litigation against municipal police departments, which began in 1969 and produced some of the most aggressive court-ordered racial hiring quotas in American history, reduced police killings of civilians, particularly non-White civilians. It also asks how any reduction came about: through more diverse police departments, through changes in how often police arrested non-White residents, or through federal scrutiny itself.
Previous Literature
McCrary (2007) shows that court-ordered hiring quotas substantially increased the share of Black officers in litigated departments, and Miller and Segal (2012) find that departments facing litigation increased Black and female employment, though by less than departments under court-ordered affirmative action. Other work links officer characteristics to police conduct. Hoekstra and Sloan (2020) find that White officers dispatched to largely Black neighborhoods use force 60 percent more often than Black officers and use gun force twice as often, and Ba et al. (2021) find that Black and female officers in Chicago use force less often. Miller and Segal (2019) find that affirmative action plans increased reporting of violence against women and reduced intimate partner homicide, and Harvey and Mattia (2022) link employment discrimination litigation to lower crime victimization, especially among Black residents. Devi and Fryer (2020) show that federal pattern-or-practice investigations can change police behavior. Few studies, however, had rigorously examined how police diversity affects police killings.
Data
Police killings come from the National Vital Statistics System Multiple Cause of Death files for 1960–2012, which record deaths due to "legal intervention" by race and county, excluding legal executions. Dates of litigation come from McCrary's compilation of employment discrimination suits filed from 1969 to 2000. The sample contains 2,985 counties, 75 of which contain at least one litigated city. These counties are only about 2.5 percent of the sample but held roughly a third of the U.S. population and 46 percent of its non-White population in 1960. They differ sharply from other counties: their average 1960 population was 815,406, compared with 39,774, and about 85 percent experienced a racial uprising in the 1960s, compared with 7 percent of control counties. Mechanism analyses draw on federal police employment surveys and Uniform Crime Reporting arrest data. Because government records capture only about half of police-involved deaths, the study also uses state-by-race corrections for underreporting published in The Lancet.
Methods
The study uses a difference-in-differences event study that exploits variation in when, and whether, a city in each county was threatened with litigation, comparing changes in non-White police killings in litigated counties with changes in counties never threatened. Models include county fixed effects and region-by-year fixed effects. The main estimates use a Poisson model suited to rare counts, checked against least-squares models, population-weighted mortality-rate models, and the Callaway and Sant'Anna estimator, which avoids biases from staggered treatment timing. Because only the threat of litigation is observed, not whether a quota followed, the estimates are intent-to-treat effects. The key assumption is that killings in litigated and non-litigated counties would have followed parallel paths absent litigation; pre-litigation estimates near zero support this for non-White deaths but not for White deaths. Weaknesses are substantial. Litigation occurs at the department level, but deaths are measured at the county level, and treated counties are far larger and more urban than controls, although results hold for large, urban, and uprising-affected subsamples and with reweighted controls. Litigation could also have changed how deaths were recorded, so the study tests suicide reporting, runs placebo tests in never-litigated counties, and estimates how much reporting would have to change to erase the results.
Findings/Size Effects
In the year before litigation, police killed an average of 1.457 non-White civilians per treated county. Pre-litigation differences are negligible (a Poisson estimate of −0.005), and effects over the first seven years are small and statistically insignificant. Effects turn negative by the second year, become significant by the fifth, and level off after about 11 years. In year 11, non-White police killings were 44 percent lower, about 45 fewer deaths across 70 counties. The long-run (years 16–25) Poisson coefficient of −0.508, significant at the 1 percent level, corresponds to a 40 percent decline; the mortality-rate model implies 37 percent and the OLS model 46 percent (significant at the 5 percent level), with nearly identical Callaway and Sant'Anna estimates. By 2000, litigated counties were averting nearly 50 non-White deaths a year, and more than 800 from 1970 to 2000.
Not all effects point the same way. Non-White killings rose, though not significantly, in the year of litigation, and the racial gap in killings rose significantly that year, mainly in counties litigated before 1980, when lawsuits coincided with racial unrest. The long-run decline is concentrated outside the South. White deaths appear to fall by about 43 percent (36 deaths) right after litigation, but that estimate fails the parallel-trends test.
The measurement checks are reassuring. Litigation had no effect on recorded suicides, and corrected counts make the long-run decline more than 70 percent larger. To erase the results, litigation would have had to increase underreporting by about 46 percent, or change the reporting gap between treated and control counties by more than 400 percent.
On mechanisms, litigation increased Black and female representation in police departments over the long run. Non-White arrests fell significantly beginning in year 11, driven by property crime arrests, which were 16 percent lower that year, while quality-of-life arrests did not change significantly. If the decline in killings ran entirely through property arrests, each 1 percent drop in those arrests would correspond to a 2.75 percent drop in killings. Weaker evidence suggests the largest declines occurred where departments remained least representative, with diminishing returns once Black officers exceeded about 30 percent of a department.
Conclusion
The study finds that the threat of affirmative action litigation produced a large, gradual, and durable decline in police killings of non-White civilians: about 40 percent in the long run, or roughly 50 deaths averted per year by 2000. The decline coincided with more diverse departments and fewer non-White arrests, but the study cannot separate the effect of new personnel from the effect of federal scrutiny itself. It does not establish effects on White civilians, and it cannot rule out reporting changes, although implausibly large ones would be needed to explain the results. The findings apply most directly to large, urban, non-Southern counties in the decades after the civil rights era and may not transfer to today's more diverse departments or legal environment. Future research using modern use-of-force data, department-level outcomes, and records of whether lawsuits produced quotas could separate the roles of diversity, oversight, and policing practices.



Comments