Principal investigator: Bruno Benevit
Author Roland G. Fryer Jr
Original title: An Empirical Analysis of Racial Differences in Police Use of Force
Location of the Intervention: United States
Sample Size: 5.000.000 police stops
Sector: Public security
Primary Variable of Interest: Police violence
Type of Intervention: Racial discrimination
Methodology: Logit
Summary
Until recently, information on shootings involving police officers was scarce and lacked detail about the incidents. Simply counting the number of police shootings did not sufficiently explore whether racial discrepancies in the frequency of these events were the result of police misconduct or differences in the behavior of suspects. In this sense, this study investigated racial differences in the use of police force in the United States. Employing an optimization model, the results revealed that in situations involving the use of non-lethal force, Black and Hispanic individuals are up to 50% more likely to experience some form of force in interactions with the police. Regarding the use of police force involving lethal force, no significant racial differences were found that imply an increase in the incidence of shootings, either in the raw data or when contextual factors are considered.
- Policy Problem
The problem of police brutality against Black individuals in the United States is a historically sensitive issue that has generated widespread public debate. The country's history is marked by regrettable incidents in which African American citizens have been victims of police violence, raising concerns about unequal treatment and the selective use of force by authorities. Furthermore, the issue is also related to other forms of racial inequality in the country, such as disparities within the criminal justice system.
These cases highlight the relevance of persistent racial inequalities in the United States and raise scrutiny of how authorities address this issue. Identifying how this problem manifests itself is fundamental to improving police practices, training, and policies governing interactions between the police and the community. However, data on this type of incident is scarce and, when available, may be underreported during the compilation process due to potential conflicts of interest with the police officers themselves and their departments (Fryer, 2019).
- Policy Implementation Context
This study used four separate datasets to examine racial differences in police use of force, including two constructed specifically for the research. All results were conditional on an interaction. The study sought to address the existence of biases in data collection arising from reporting by police officers themselves, involving potential conflicts of interest that could lead to underreporting.
With this goal in mind, the four data sources employed sought to jointly portray racial differences in the use of police force conditional upon an interaction. The first two data sources, originating from the program Stop, Question, and Frisk The New York City Police Department and the Police-Public Contact Survey (PPCS) provided information on the use of nonlethal force, observed from both the police and civilian perspectives, respectively. The other two databases included summaries of events related to police shootings in 16 locations across the United States and information on civilian-police interactions in Houston where the use of lethal force might have been legally justified. Using this dataset allowed for the investigation of racial differences in police shootings, both in terms of extent and intensity, by analyzing databases from distinct sources.
- Evaluation Details
The program Stop, Question, and Frisk The first dataset, from New York City, comes from a New York Police Department practice in which officers approached and questioned pedestrians, and could also search them for weapons or illegal items. This dataset contained approximately 5 million observations. This database provided detailed information on a wide range of uses of force, from the placement of civilians' hands to the use of batons. The second dataset, PPCS, is a triennial survey of a nationally representative sample of civilians, which included, from a civilian perspective, a description of interactions with the police, also covering the use of force. Both datasets are publicly available and easily accessible.
The third and fourth datasets were compiled by a team of researchers for the purposes of the study. The third dataset considered event summaries from all incidents in which a police officer fired their weapon at civilians – including both hits and misses – in three major Texas cities (Austin, Dallas, and Houston), Denver, Seattle, nine large Florida counties, Jacksonville, and Los Angeles County, in order to construct a dataset in which it is possible to investigate racial differences in police shootings. Given that all individuals in these datasets were involved in a police shooting, the analysis of this data alone allows for the estimation of racial differences only at the intensive level (e.g., whether officers fired their weapons before or after the suspect attacked).
Finally, the fourth dataset contains a random sample of interactions between police and civilians from the Houston Police Department based on arrest codes in which the use of lethal force is most likely justified: attempted murder of a law enforcement officer, assault on a law enforcement officer, resisting arrest, escape from arrest, and interference with an arrest. This data comes from arrest reports ranging in length from two to 100 pages. A team of researchers was responsible for reading the arrest reports and collecting nearly 300 variables from each incident. Combining this with data from shootings involving Houston police officers, we can estimate both the extent (e.g., whether or not a police officer decides to shoot) and the intensive scope.
- Method
The study presented estimates of the extent of racial differences in police use of force using four datasets presented separately. To this end, the logistic regression (Logit) method, commonly used for the analysis of binary outcome variables, was employed. Regarding the outcome variables examined, the probability of police interaction conditioned on the race of civilians was initially observed, considering different "risk sets," that is, the sets of civilians who could potentially be stopped. Furthermore, these interactions were disaggregated considering the level of force used by the police officers, separating them by violent use of non-lethal force: (i) placing hands on a civilian, (ii) forcing against a wall, (iii) handcuffing, (iv) drawing a weapon, (v) pushing to the ground, (vi) pointing a weapon, (vii) using pepper spray or striking with a baton. Additionally, racial differences in involvement in shootings were estimated. These were estimated with the addition of several specifications regarding the covariate vector. When controls were incorporated into the models, demographic characteristics, behavior of the civilians approached, characteristics of the approach, characteristics of the police officers, and fixed year effects were considered.
The analysis was completed through the development of a discrimination test, based on the models from the studies by Knowles, Persico, and Todd (2001) and Anwar and Fang (2006). The Knowles et al. test verifies racist preferences by analyzing the success rate of searches conducted by police officers on different races. Their model assumed that the police sought to maximize the number of successful searches, taking into account the cost associated with these approaches. If racial bias existed, the cost of searching drivers would differ between races, implying a different rate of successful searches. All the study's results were conditioned on an interaction.
- Main results
The estimates were found using data on police interactions from the program. Stop and Frisk Data from New York City indicated significant racial disparities. When raw data was considered, Black and Hispanic civilians were more than 50 percent more likely to have an interaction with police involving the use of force. Controlling for variables representing baseline characteristics, approach characteristics, civilian behavior, and fixed effects of precinct and year, the odds ratio for Black and Hispanic civilians compared to white civilians was 1,178 and 1,122, respectively. Specifically regarding Black civilians, this group is 21% more likely than white civilians to experience an encounter where a weapon is drawn by the officer. Additionally, the odds ratio of nonlethal use of force for Black civilians compared to white civilians ranged from 1,175 to 1,275, decreasing with the intensity of force used in the encounter.
The results considering the data from the Police-Public Contact Survey (PPCS) are qualitatively similar to the results from the New York data, but quantitatively different in terms of magnitude. Black and Hispanic individuals were approximately 1,3 percentage points more likely than white individuals to report any use of force in a police interaction compared to the average of 0,7 percent for white individuals. That is, the odds ratio was 2,769 for black individuals and 1,818 for Hispanic individuals. The author justifies the differences in estimates between the data. Stop and Frisk The discrepancies in the basic probabilities of use of force and national representativeness of PPCS versus the specific nature of Stop and Frisk in dense urban areas can be explained by the differences in the basic probabilities of use of force and the national representativeness of PPCS versus the specific nature of Stop and Frisk in dense urban areas, as well as limitations of the location variables present in the PPCS database.
Evidence from the models of Knowles, Persico, and Todd (2001) and Anwar and Fang (2006) revealed that, controlling for civilian behavior, police officers exhibit approaches with greater use of force against Black civilians compared to white civilians. When assessing the probability of suspects involved in shootings possessing weapons considering the race of the police officers involved, no statistically significant difference was found between the behavior of white and Black police officers towards Black suspects. However, surprisingly, it was found that Black police officers were 16 percentage points more likely than white police officers when the suspect was white.
Observing these results, it was found that this behavior is consistent with the proposed model for optimizing police officers, insofar as the utility costs for police officers associated with involvement in shootouts are considerably higher compared to the use of non-lethal force.
- Lessons in Public Policy
Police violence and its racial correlation has become one of the most controversial topics in American debates, eliciting a range of emotions from outrage to indifference. The lack of substantial data complicates the understanding of racial disparities in the use of police force, compounded by the complexity of interactions between police and civilians. Beyond the scarcity of data, the analysis of police behavior faces challenges such as the reliability of existing data and the impossibility of randomly assigning race.
As police departments in the US explore community policing models that induce greater compliance and confidentiality Regarding best practices and the reduction of implicit discriminatory bias by officers during police encounters, the results found in this study highlighted how the process of racial discrimination in police encounters occurs. Currently, few police departments collect data on less severe uses of force or apply explicit punishments for misconduct in these tactics, implying incentives that are reflected in low accountability for discriminatory conduct. This scenario highlights the need to increase the expected cost of the excessive use of non-lethal force for police officers.
References
ANWAR, S.; FANG, H. An Alternative Test of Racial Prejudice in Motor Vehicle Searches: Theory and Evidence. American Economic Review, v. 96, no. 1, p. 127–151, 2006.
DELAVANDE, A.; ZAFAR, B. University Choice: The Role of Expected Earnings, Nonpecuniary Outcomes, and Financial Constraints. Journal of Political Economy, v. 127, no. 5, p. 2343–2393, 2019.
FRYER, RG An Empirical Analysis of Racial Differences in Police Use of Force. Journal of Political Economy, v. 127, no. 3, p. 1210–1261, 2019.
KNOWLES, J.; PERSICO, N.; TODD, P. Racial Bias in Motor Vehicle Searches: Theory and Evidence. Journal of Political Economy, v. 109, no. 1, p. 203–229, 2001.
