STUDIES:   

900/

“Because there are inevitably unobserved factors that jointly influence the decisions to stop and use force, analyzing only stops introduces selection bias. Therefore, for the conclusions in Fryer (2019) to hold, there must be no racial bias in the decision to detain civilians — in other words, to deem these estimates credible, analysts must assume away racial bias in a study of racial bias."

Toward a General Causal Framework for the Study of Racial Bias in Policing Toward a General Causal Framework for the Study of Racial Bias in Policing Figure 3: Selection on the outcome. Encounters are grouped into cells according to the race of the encountered civilian, whether a stop occurred, and whether the stop resulted in a police behavior such as use of force. The outcome-selection approach focuses on cells labeled (1) and (2), ignoring the remaining faded regions. Analysts argue that anti-white bias exists when cell (1) is larger than cell (2), or anti-minority bias when the reverse is true. However, this approach fails to account for the rate of minority and white encounters — the size of the Di = 0 and Di = 1 boxes — and is uninformative about any causal quantity of interest. Figure 3: Selection on the outcome. Encounters are grouped into cells according to the race of the encountered civilian, whether a stop occurred, and whether the stop resulted in a police behavior such as use of force. The outcome-selection approach focuses on cells labeled (1) and (2), ignoring the remaining faded regions. Analysts argue that anti-white bias exists when cell (1) is larger than cell (2), or anti-minority bias when the reverse is true. However, this approach fails to account for the rate of minority and white encounters — the size of the Di = 0 and Di = 1 boxes — and is uninformative about any causal quantity of interest. Stop selection studies and outcome tests. Encounters are grouped into cells according to the race of the encountered civilian, whether a stop occurred, and whether the stop resulted in a police behavior such as use of force. The stop-selection approach and outcome tests both focus on cells labeled (1)–(4), ignoring the remaining faded regions. In this stop-selection approach, which implicitly assumes away racial bias in the police stops, a positive outcome (Yi = 1) represents some subsequent police behavior, such as the use of force. Stop selection studies and outcome tests. Encounters are grouped into cells according to the race of the encountered civilian, whether a stop occurred, and whether the stop resulted in a police behavior such as use of force. The stop-selection approach and outcome tests both focus on cells labeled (1)–(4), ignoring the remaining faded regions. In this stop-selection approach, which implicitly assumes away racial bias in the police stops, a positive outcome (Yi = 1) represents some subsequent police behavior, such as the use of force.
Toward a General Causal Framework for the Study of Racial Bias in Policing
Journal of Political Institutions and Political Economy
Knox, Dean / Mummolo, Jonathan
Vol. 1 Issue 3, pp 341–378, 2020
10.1561/113.00000018
Thoughts?
Comment on this study on its Twitter thread.