Table 1 in Knox, Lowe, and Mummolo (2020) contains estimates of racial bias in police use of force when comparing: (i) stops involving Black civilians and (ii) stops involving Hispanic civilians. To estimate the bounds on these effects, we applied our statistical methodology (developed in the same study) which takes as an input a parameter, $ \rho $ , representing the severity of racial bias in the initial decision to stop a civilian. In the article, we state that (based on a separate analysis), $ \rho $ was set to 0.32 for comparisons involving Black civilians and 0.35 for those involving Hispanic civilians.

We have since discovered that, due to a coding error, $ \rho $ was in fact set to 0.35 for both sets of comparisons. As a result, the estimates of bias in the use of force for events involving Black civilians are still valid, but correspond to a different specification than the article indicates. Setting $ \rho $ to 0.32 results in slight numeric deviations from published results. This error does not affect any substantive conclusions.

We regret this error. Corrected estimates for the upper section of Table 1 appear below. The replication materials are updated to reflect this correction.

Table 1. Average Treatment Effect among Stops ( $ AT{E}_{M=1} $ ), by Severity of Force and Minority Group (Revised Results)

Note: Black vs. white comparisons. Bounds show possible values of $ AT{E}_{M=1} $ when the unknown proportion of discriminatory stops $ \rho =0.32 $ . Estimates scaled to indicate excess force per one thousand encounters. Estimates are bolded, and 95% confidence intervals are italicized.