Joseph Cesario's Lab
Are Black Americans shot by police more than we would expect?
(The data described on this webpage are a portion of the data in a paper now published: Cesario, Johnson, & Terrill, SPPS. Click HERE for a copy of this paper. This webpage provides a summary of a portion of that research designed for a non-academic audience.)
​
In recent years there has been unprecedented public concern over police shootings, with a particular emphasis on the possibility that Black citizens are more likely to be shot compared to White citizens. The idea that police are biased against Black citizens in their deadly force decisions has been raised almost universally in media reports of police shootings and, it seems, is widely accepted by the public. Indeed, perhaps the central point of the Black Lives Matter movement has been to highlight officers' greater likelihood of shooting Blacks compared to Whites. As to why this happens, the most common view is that this is due to racism on the part of police; hence "Black Lives Matter," because some people (police officers) think they matter less than White lives.
​
In all this, people generally take it as a fact that Blacks are shot "more than we would expect" compared to Whites. But is this really the case? The title of this page is intentionally ambiguous to illustrate the important point that prompted these analyses: "More than we would expect" given what?
​
To highlight why this is a key question, we can start by looking at raw numbers: From 2015 - 2016, 1,051 White citizens were killed by police gunfire and 510 Black citizens were killed by police gunfire. From these numbers, it is clear that Whites are shot more than Blacks, by a factor of just over 2.0 (1051 / 510 = 2.06). Stated differently, of those citizens identified as either White or Black, Whites made up 67% of police killings by gunfire and Blacks made up 33%.
​
Everyone immediately recognizes, of course, that these raw numbers aren't very meaningful. Because Whites make up a greater proportion of the population, we would expect more Whites to be killed simply due to their greater numbers overall.
​
This brings us to the key point: We have to compare these numbers to some other value. In other words, we have to ask: "given what?" Are Blacks or Whites shot more than we would expect given what? The most common answer to this question is to adjust raw shooting numbers by population numbers. That is to say, are Blacks killed by police gunfire more than we would expect given the population proportions of Blacks and Whites in the U.S.?
​
Certainly adjusting for population proportions is better than nothing. And when we do this, we get the familiar statistic: Blacks are about 2.5 times as likely to be killed by police as Whites. This is seen in the rightmost bar in the figure below:
The question, however, is whether adjusting shooting values by population proportions makes good sense. This is certainly one way to adjust the raw data, but doing so necessarily requires one of two assumptions to be true: (1) That police are equally likely to shoot any citizen at any moment in time and in any situation, or (2) If police are more likely to shoot citizens in some situations rather than others, that Black and White citizens are equally likely to occupy those situations where deadly force is likely to occur.
As for the first assumption (police use deadly force equally across all situations) this seems wrong on the face of it; surely police are more likely to use deadly force in crime-related situations than not, or in dangerous neighborhoods than safe neighborhoods, or when serving arrest warrants than when talking to a 5th-grade classroom. To adjust the raw shooting numbers on population proportions assumes that none of this is true and that in fact, an officer buying a cup of coffee is as likely to shoot the cashier selling him the coffee as he is to shoot a citizen with an outstanding warrant who has just been pulled over for speeding. Not only does common sense suggest this is wrong, the data do not support this assumption.
​
Granting that officers are more likely to use deadly force in some situations more than others, what about the second assumption then (Blacks and White are equally likely to occupy those situations where deadly force is likely)? The data on criminal activity, particularly violent crime, are clear that this assumption is simply not true. As I show below, this is supported both by official crime statistics, self-report victimization data, and data from sources such as the CDC.
​
In other words, the decision to compare fatal shooting statistics against population proportions is a decision that carries with it some unreasonable assumptions.
​
When we consider the nature of police shootings (that they overwhelmingly occur in crime-related contexts, for example), a very different picture emerges. If police are in fact more likely to use deadly force in some situations rather than others, then the correct benchmark is compare fatal shooting statistics against involvement in those situations, not population proportions. In other words, instead of asking, "Are Blacks killed by police gunfire more than we would expect given the population proportions of Blacks and Whites?", one might instead ask "Are Blacks killed by police gunfire more than we would expect given involvement in violent crime of Blacks and Whites?" To the extent that police are in fact more likely to shoot a citizen during a criminal rather than non-criminal interaction, this becomes a reasonable way of studying police shootings.
​
​
Are Blacks or Whites more likely to be killed by the police given each group's criminal involvement with police?
​
There are many types of crime data against which we could benchmark police shootings (and our paper goes through many of these). For now I describe just one way, which is to look at the crime statistics as taken from the FBI's Summary Report System data.
​
In terms of criminal activity, the most relevant (in order of decreasing severity) include murder/non-negligent manslaughter, violent crime (which includes murder, rape, robbery, aggravated assault), and weapons violations. These three types of crime, on face, seem to be good candidates for this analysis as officers likely have a heightened concern with their own safety when dealing with citizens suspected of these activities.
​
I start with the most violent activity, murder/non-negligent manslaughter. Here the question is, "Are Blacks or Whites more likely to be killed by police given the rates at which each group is arrested for killing people?" When we benchmark police shootings against these values, a very different picture emerges. Here, Whites are about 2.6 times as likely to be killed by police:
​
Turning next to arrests for any violent crime, the question becomes "Are Blacks or Whites more likely to be killed by the police given the rate at which each group is arrested for violent crime?" Here we see again a bias in the direction against Whites, with Whites about 1.3 times more likely to be killed by police:
​
Finally, we can benchmark police shootings against weapons violation arrests. This is an interesting variable because, unlike arrests for murder, weapons violation arrests may be more subject to biased policing on the part of officers. When a murder occurs, police will generally attempt to arrest someone for the crime. By comparison, officers have more leeway to initiate the chain of events that might lead to a weapons arrest (e.g., stop and frisk).
​
In this case, the question is, "Are Blacks or Whites more likely to be killed by the police given the rate at which each group is arrested for weapons violations?" Once again we see that Whites are relatively more likely to be shot by police, as Whites are 1.5 times as likely to be killed than Blacks:
​
In sum, once we take into account the frequency with which Blacks and Whites interact with police in a criminal context, we do not find evidence of bias against Black citizens in being killed by police. As we demonstrate in our in press paper, this holds regardless of the database from which crime rates are drawn (e.g., FBI's SRS, FBI's NIBRS, NCVS, etc.) and regardless of how violent crime is defined. These general conclusions also hold when we look at shooting of unarmed citizens as well.
​
Is this just biased policing?
​
One reasonable question is whether biased policing accounts for these results. If it's the case that police are simply more likely to arrest Black citizens than White citizens, this would conceal actual anti-Black bias. Therefore the question could be put, "Isn't using arrest data simply wrong to begin with, given that we "know" that police are more likely to arrest Black citizens?"
​
In the supplement to our paper, we outline the many reasons why this is not the case. One very direct way of testing this is to approximate criminal activity not from policing records but from the CDC's death by assault records. The CDC's data on deaths by various kinds of assault provide a bias-free estimate of violent criminal activity, because (1) these data are uncontaminated in any real way by biased policing, and (2) the far majority of murder cases are within-race, not between-race.
​
Given this, we can again ask, "Are Blacks or Whites more likely to be killed by police given the rates at which each group commits murder?", with our estimate of murder rates with the CDC data. And here we observe the same pattern:
Indeed, there is even less anti-Black bias when using the more objective CDC data compared to, say, weapons violations, the latter of which is much more subject to distortion by biased policing.
​
Summary
​
Taken in total, the above analyses -- while in no way definitive -- are consistent with the idea that police officers do not broadly show anti-Black bias in their use of deadly force. The finding of greater proportional deaths at the population level appear to be consistent with Blacks' greater involvement with police in criminal contexts.
​
​
​
"Big Picture" Point
​
In general, these analyses highlight the fact that comparing a group's outcome to their representation in the population is questionable at best, despite the ubiquitous practice of doing so. There is a near-constant refrain of disparities between the outcomes a group obtains and their proportion in the population: females are 6% of CEOs despite being ~51% of the population, Black Americans are about 2% of mathematics PhDs despite being ~13% of the population, and so on. These comparisons are typically made as a means of demonstrating that bias or discrimination or disparate treatment has occurred somewhere along the way, requiring interventions (such as "implicit bias" training) to reduce these disparities.
Underlying this argument is the assumption that absent disparate treatment we would expect all groups to receive the same outcomes. There is little justification for this belief, however, as different groups have different average interests, abilities, histories, characteristics, and so on, all of which impact the average outcomes of different groups.
​
For example, in the case of female CEOs, in order for the 51% number to be at all meaningful as a measure of disparate treatment, it requires that females and males have exactly the same interest in being CEOs to begin with.
​
To the extent that groups differ in any characteristics beyond disparate treatment (implicit or otherwise), then comparing a group's outcome with their population proportion is in error. Such a comparison only makes sense if one holds the belief that all groups are equal in every relevant characteristic for success in some domain. Although this assumption of human distributions is held by many intellectuals, evidence suggests otherwise:
​
Dr. Thomas Sowell
Data & Code
I've left all analyses out of the main text, to keep this webpage easy to read. All data and code can be downloaded from the Data & Materials webpage, under the publication:
​
-
Cesario, J., Johnson, D.J., & Terrill, W. (2019). Is there evidence of racial disparity in police use of deadly force? Analyses of officer-involved shootings in 2015-2016. Social Psychological and Personality Science.
​
You can also find the data analysis script at the OSF PAGE for this publication.
​
​
​
Caveats/Footnotes
The issue of police use of deadly force is a difficult topic. In our manuscript, we highlight many of the limitations and complications of our research. For now, I briefly list these:
​
*The results presented here are strictly about fatal shootings by law enforcement. These results have no implications for any other type of policing behavior, including non-lethal use of force.
​
*These results have no implications for any specific instance, officer, or police department. The analyses are strictly at the nationwide level.
​
*The results do not speak in any way to the causes of racial differences in criminal activity, nor whether race (as opposed to socio-economic status or some other characteristic) is the relevant variable at all.
​
*These results do not imply that all officers are perfect, that all officers make perfect decisions, or that all officers are perfectly trained.
*This is not the first attempt to take crime rates into account in understanding police shootings and race bias. Our manuscript goes into detail on the previous literature on this topic. For example, Dr. Peter Moskos provides a similar analysis and comes to similar conclusions; however I believe the analyses presented on this webpage are more comprehensive and use a more reliable database of police shootings. Other researchers, such as Dr. Cody Ross and Dr. Francis Smart use different approaches and statistically control for crime by including crime rates in multilevel or other types of regression models. These researchers come to different conclusions than those presented here, although (1) the goals and underlying assumptions of my approach and their approach are different and (2) there are disagreements (which we describe in our manuscript) about some classifications, e.g., that if "unarmed" shootings show racial disparities then this is evidence of police bias. Moreover, other research on crime rates, such as that of Dr. David Klinger, shows conclusions that are similar to those presented here.
​
​