Abstract: 

While there is much concern about the effect of race on policing, identifying causal effects is difficult due to endogenous police-civilian interactions.  Much of the existing literature has either modeled how officers initiate interactions or imposed assumptions of selection-on-observables.  This paper identifies effects by exploiting as-good-as-random variation in the race of police officers dispatched to over 1.2 million 911 calls.  We do so using administrative data from a large city in which protocol dictates and operators confirm that neither they nor police officers can exercise discretion in the assignment of officers to calls.  This protocol dictates operators must first dispatch the beat officer if that unit's computer signals it is available, or otherwise dispatch the next-closest available officer as observed on the live-location computer map.  We show empirical evidence that conditional on police beat by time fixed effects, the race of the dispatched officer is uncorrelated with call characteristics and predicted use of force.  Results indicate black officers use force 40 percent less often than white officers, and use gun force 65 percent less often.  In addition, black officers use force less often even in all-white neighborhoods.  Moreover, while white officers use force at higher rates as they respond to calls in more black neighborhoods, the opposite is true for black officers.  As a result, we estimate dispatching a same-race officer results in a 44 to 62 percent reduction in use of force.  We find similar effects for gun force, driven by the white officers' much higher use of their guns when dispatched to predominantly black neighborhoods.  These results indicate that at least in our setting, officer race is an important determinant of use of force, including force in which an officer fires his gun.  This has important implications for citizen perceptions of policing and for the outcomes of high-stakes interactions between police and civilians.