Joseph Antonelli

Research Summary Following high‐profile police killings, many U.S. cities banned officers from using chokeholds and other neck restraints. The evidence for such bans, however, is limited. To test whether use‐of‐force policies prohibiting neck restraints are related to fewer police killings, we use three modeling approaches to analyze 2183 U.S. cities between 2009 and 2021. Police killings were lower in places that adopted neck‐restraint bans and the bans were associated with less crime and fewer assaults on officers, net of controls. Because officer labor unions can affect use‐of‐force policies and the frequency of police killings, we also analyzed them, finding unionization increased the likelihood a city had a neck‐restraint ban and had a null or negative association with police killings. Policy Implications Adopting a neck‐restraint ban is likely an effective way to reduce deaths due to police use of force with minimal collateral consequences. The bans operate through a diffuse discouragement of many types of lethal force or as a part of an array of use‐of‐force policies. Their direct relationship to asphyxiation deaths remains unclear. Officer unionization is unlikely to change the frequency of police killings, except through its association with stricter use‐of‐force policies.

We propose a doubly robust approach to characterizing treatment effect heterogeneity in observational studies. We develop a frequentist inferential procedure that utilizes posterior distributions for both the propensity score and outcome regression models to provide valid inference on the conditional average treatment effect even when high-dimensional or nonparametric models are used. We show that our approach leads to conservative inference in finite samples or under model misspecification and provides a consistent variance estimator when both models are correctly specified. In simulations, we illustrate the utility of these results in difficult settings such as high-dimensional covariate spaces or highly flexible models for the propensity score and outcome regression. Lastly, we analyze environmental exposure data from NHANES to identify how the effects of these exposures vary by subject-level characteristics.