This is the post that I made to Thomson Reuters’ internal social media site, called the Hub, that precipitated a barrage of hateful and racially charged attacks from BLM supporters within the company. When I brought those attacks to the attention of Thomson Reuters’ Human Resources department, my post was censored, and I was told I was not allowed to use any company communications channels (email, teams, the Hub, etc.) to discuss the attacks I had experienced. Receiving no support from HR, I raised the issue with my colleagues and senior leadership over email, for which I was fired. Below is the Hub post that precipitated this chain of events, in full.
BLM Spreads Falsehoods That Have Led to the Murders of Thousands of Black People in the Most Disadvantaged Communities
I believe the Black Lives Matter (“BLM”) movement arose out of a passionate desire to protect black people from racism and to move our whole society towards healing from a legacy of centuries of brutal oppression. Unfortunately, over the past few years I have grown more and more concerned about the damage that the movement is doing to many low income black communities. I have avidly followed the research on the movement and its impacts, which has led me, inexorably, to the conclusion that the claim at the heart of the movement, that police more readily shoot black people, is false and likely responsible for thousands of black people being murdered in the most disadvantaged communities in the country.
Over the last few years I’ve also seen support for the BLM movement grow within Thomson Reuters. A search of the hub shows dozens of messages and posts supportive of BLM, including an entire series of official TR events supporting BLM and organized in cooperation with BLM organizers. A similar search of Reuters News shows extensive positive and uncritical news coverage of the BLM movement. Unfortunately, in both our internal discussion and external coverage there seems to be a nearly absolute failure to examine the scholarly research, coming out of the most respected institutions in our country, which demonstrates the acute damage that the BLM movement is doing to many black communities, and the possibility that structural and systemic patterns in our society which have historically disenfranchised blacks are propelling the movement’s whirlwind rise.
Thomson Reuters must do better to resist simplistic narratives that are not based in facts and evidence, especially when those narratives are having such a profoundly negative impact on minority or marginalized groups. And, as one of the most important and respected media institutions in the world, Reuters News has a special responsibility to correct widely-repeated falsehoods that are spread as a result of structural and systemic patterns in our society which have historically disenfranchised blacks.
The Falsehood at the Heart of the BLM Movement
The BLM Movement became nationally recognized with street demonstrations following the deaths of a number of black suspects at the hands of police in 2014. Since then, BLM’s highest profile demonstrations have been protesting police related deaths of blacks, and the movement again made international headlines after George Floyd died in police custody in 2020. Wikipedia has a “Timeline of notable events and demonstrations in the United States” regarding the BLM movement, and the vast majority of notable events and demonstrations concern someone being fatally shot by police or otherwise dying while in police custody.
Many have noticed that the rise of BLM has coincided with the rise of ubiquitous smartphone usage, meaning that nearly everyone is carrying around a video recorder at almost all times. This meant that police shooting statistics that may have once seemed abstract suddenly became visceral and real, and police shooting incidents that may once have been mired in conflicting accounts suddenly had documentary footage showing exactly what happened. These videos are disturbing. Almost everyone feels horror when seeing a video of someone being killed.
At BLM protests, and from BLM proponents, we have since heard that “it’s open season” for police to kill black people, and that police are “hunting” black people. According to the BLM website itself, “Black lives are systematically and intentionally targeted for demise” by “state-sanctioned violence against Black people''.
Putting aside, for the moment, whether language suggesting an intentional genocide is hyperbolic, it’s clear that BLM activists and proponents are emphatically asserting that police are targeting blacks with lethal force: shooting and killing them in circumstances that they wouldn’t shoot and kill whites. And, this claim, that police are targeting blacks with lethal force, seems to be, if not the most important claim of the BLM movement, at least one of the most important claims.
The only problem: it’s completely untrue.
According to the Washington Post’s database of police shootings, over the last five years there have typically been between 30% and 100% more unarmed whites killed by police than unarmed blacks, with an average across the last five years of 39% more. For instance, in 2020 there were 457 whites shot and killed by police, compared to 243 blacks. Of those, 24 of the whites killed were unarmed compared to 18 blacks. (It’s worth noting that in the vast majority of police shootings of both blacks and whites, police gunfire was justified in response to an armed and threatening suspect. [EDIT: Additionally, a study examining every police shooting of unarmed suspects over a two year period suggests that even most of those shootings were possibly justified due to suspects physically attacking officers, attempting to grab their firearms, etc.])
If there are more unarmed whites than blacks shot by police each year, what is the basis for the claim that blacks are being targeted by police with lethal force? The idea is that because blacks are 13% of the population, while whites are 76% of the population, if police were not targeting blacks with lethal force, whites would be shot by police at a rate 5-6 times the rate that blacks are shot. In other words, we have to “benchmark” the higher number of shootings of whites to their larger population in order to have a fair comparison. While these disproportionate numbers certainly point to some kind of problem, is the problem police bias?
To start approaching an answer to that question, we must consider the issue of benchmarks in more detail. Police are not supposed to distribute lethal force randomly throughout the population in order to ensure equal application to each racial group. Instead, police are supposed to use lethal force only in response to threats of serious violence during encounters with criminal suspects. Thus, if lethal force were applied by police without any bias whatsoever, we would expect the number of applications of lethal force for each racial group to be proportional to the number of high risk encounters members of each racial group have with police officers, and not with the population overall. The correct benchmark for measuring bias in police use of lethal force is the number of high risk encounters for each group, and not the population of each group.
This is a critical distinction because there are definitive reasons to believe that police have very different rates of high risk encounters per member of different racial groups for reasons related to entirely legitimate policing objectives. For instance, as the evidence in the following section demonstrates, on average, violent crime rates are dramatically higher in predominantly black communities than they are in predominantly white communities. This violence takes a severe toll on those communities, can traumatize residents there, makes it virtually impossible for children to focus on school and academic success, and worse. Because, on average, there is so much greater violence in predominantly black neighborhoods, in order to protect and defend the (mostly) black residents in those communities, police are disproportionately required to confront criminal suspects in those communities. Therefore we should expect there to be more encounters in those communities for the purpose of achieving entirely legitimate and laudable policing objectives.
As another example of why it’s important to use a proper benchmark, there is a substantial body of evidence establishing that members of different racial groups resist arrest at very different rates. Because the vast majority of potentially dangerous encounters happen when a suspect resists arrest, a greater rate of resisting arrest will be expected to increase the number of police shootings, even if police have absolutely no bias when deciding when to shoot.
Therefore, if we want to investigate whether there is bias in the application of lethal force, we need to look at the rate of police shooting per potentially violent encounter with criminal suspects—and not per member of a group’s overall population (most of whom are law abiding, peaceful citizens). When you do so, the supposed anti-black bias disappears completely, and possibly, even reverses.
This investigation can be carried out in two main ways: (1) consideration of high-level descriptive statistics and (2) econometric analysis that controls for circumstances of encounters. There is a considerable and growing body of research worth discussing, much of which I have studied. While below I have space only to review the key findings from the research, I welcome further discussion and analysis with anyone who would like to investigate this topic in more detail with me.
A preliminary step when doing a statistical investigation is to consider the high-level descriptive statistics, if for nothing else, than as a sanity-check. Although the descriptive statistics often won’t have the granularity to give definitive answers, you can at least discover broad patterns worth investigating. In this case, we are interested in considering whether the number of police shootings of blacks is disproportionately large in relation to the number of potentially violent encounters between police and black suspects, but unfortunately we do not have reliable statistics about the number of potentially violent encounters nationwide. When faced with a lack of data it is common to look for proxy data that we hypothesize will be highly correlated with our missing data. As alluded to above, in this case, the obvious proxy for potentially violent encounters with suspects would be actually occurring violent crime, for which we do have data.
Here the evidence is very clear. For instance, the Wall Street Journal reports that “African-Americans made up 53% of known homicide offenders in the U.S. and commit about 60% of robberies, though they are 13% of the population.” As for non-homicide violent crime, the Justice Department’s National Crime Victimization Survey shows that whites commit about 48% of nonfatal violent crimes and blacks commit 35%. When you look at just serious nonfatal violent crimes, whites commit about 41% and blacks commit about 43%.
In other words, depending on the type of violent crime, whites either commit a slightly greater (non-fatal crimes) or slightly smaller (fatal, and serious non-fatal crimes) percentage of the total violent crime than blacks, but in all cases roughly in the same ballpark. But, as referred to above, over the past 5 years, police have killed 39% more unarmed whites than unarmed blacks. There are many more whites killed by police, even though whites account for a similar absolute number of violent offenders. Thus, if the number of potentially violent encounters with police reflects the violent crime rates, then the raw statistics suggest that there is actually a slight anti-white bias in police applications of lethal force.
But, what if the violent crime rate does not actually reflect the frequency at which police officers face risk of grievous injury from suspects? Can we find a proxy variable that more directly reflects the frequency at which police officers face risk of grievous injury and thus must use lethal force? Perhaps the most direct measure of the danger of grievous injury that police face is the rate at which they are actually murdered by criminals. Thus, if we benchmark police shootings against the number of police murdered by criminals, we should obtain a very good indication of whether police use lethal force more readily in response to lower levels of threat for one group than another. This yields similar results: “Adjusted for the racial disparity at which police are feloniously killed, whites are 1.3 times more likely than blacks to die at the hands of police.” [EDIT: Here is an even better source, that I’ve only come across recently, looking at police killings by racial group in proportion to the rate that police are killed by members of those groups.]
In other words, if you measure police shootings against a legitimate benchmark, one that is actually related to how often police need to use lethal force for entirely lawful, ethical and moral reasons—such as defending themselves or others from grievous injury—there appears to be a clear anti-white bias.
However, looking at the descriptive statistics like this leaves a lot of questions unanswered. For example, perhaps police shoot whites at a higher rate per violent offender, because, hypothetically, whites are more violent in confrontations with police on average, and police are simply responding to legitimate threats. Thus, to really investigate if there is bias, it’s necessary to look through thousands of examples of police confrontation, code them according to the circumstances, weapons involved, behavior of the suspect, and whether there was a shooting, and see if police on average use lethal force more readily in the same circumstances for one group than the other.
Roland G. Fryer Jr. is a star economist at Harvard University. He was one of the youngest professors to achieve tenure at Harvard, received a MacArthur “genius” grant, and won the most prestigious award for a young American economist, the John Bates Clark medal. Without a doubt, he is one of the top researchers in the field of economics. He also is black, grew up poor, personally witnessed episodes of his peers being roughed-up by police, and, initially at least, supported the BLM movement. He set out to lay the empirical and intellectual foundations of the BLM movement by conducting a study exactly like that described above.
In what he describes as “the most surprising result of my career”, his study didn’t find evidence for anti-Black or anti-Hispanic disparity in police use of force across all shootings, and, if anything, found a slight anti-White bias. Fryer was so shocked that he disbanded his original research team, hired an entirely new team, and repeated the entire data annotation and analysis process from scratch. He found the same results.
This, perhaps, should not have been as surprising to Fryer as it was because it confirms exactly what the raw descriptive statistics reviewed above implied. (Fryer also found that the result did not change when you ignored the police’s reporting of the circumstances, further adding to the robustness of his findings, and rebutting a possible concern that police dishonestly exaggerate the threat of white suspects at a lower rate than with black suspects.)
It’s worth mentioning here that while Fryer’s results raise the possibility that police shoot and kill whites more readily, his results also show that police more readily use non-lethal force against blacks. What could explain why police would more readily shove or hit blacks, but might more readily shoot and kill whites? That remains an unanswered, and largely uninvestigated question.
Unsurprisingly, Fryer’s study precipitated considerable criticism from researchers sympathetic to the BLM movement. In the Appendix at the bottom of this post, I examine some of that criticism as well as other research in the field. While much of that criticism seems to be motivated, at least in part, by the political and social agendas of the critics, it’s safe to say that Fryer’s study is not the final word on the subject. More research is needed.
Nevertheless, thus far, Fryer’s research finding that there was no bias in shootings stands as the gold standard for investigating the question of police bias in use of force. Although there are limitations to Fryer’s study, no properly designed study controlling for the circumstances of shootings has, before or since, produced any findings to the contrary.
Unfortunately, it’s not uncommon for news-media and even researchers themselves to report research findings sloppily or falsely. As an example, consider a recent study that is widely, but falsely, cited to support the contention of police bias in shootings. This study is based on the National Violent Death Reporting System (NVDRS), which contains only data about encounters that led to someone’s death. Because the data contain no information about the total number of police encounters (including both those that result in death and those that do not) it’s impossible to calculate, for any racial group, the rate that any particular type of police encounter will result in a police shooting—and the researchers make no attempt to do so.
Nevertheless, ABC news misleadingly reports that “In the new study, black Americans were three times more likely to be shot and killed by police officers during interactions where the victim appeared to pose little or no threat to officers, the researchers found.” In fact, the study did not investigate the likelihood of black suspects being shot during interactions where they posed little or no threat to officers, because the dataset contained no information about how many such interactions there were and thus calculating such a likelihood would be impossible. Instead, the study found, yet again, that the number of such shootings was disproportionate to the black population, which, as discussed above, gives us no information about bias. (In fairness to ABC News, the paper appears to be almost deliberately written to make this sort of misleading reporting more likely. See the Appendix below for a more in-depth exploration of the issues with this study and research in this field in general.)
A simple example should suffice to illustrate how crucial this distinction is: according to the Washington Post’s database of police shootings, police shoot and kill 10 times more unarmed, fleeing men than unarmed, fleeing women, a disparity that dwarfs any racial disparity in the data. Since we are dealing with unarmed, fleeing men and women, we can assume that both the men and women posed no threat to the officers. Can we thus infer that any difference must be due to lethal gender discrimination? Is it “open season” for police to “hunt” and kill men?
Or, alternatively, is it possible that police shoot so many more men who pose no threat than women, simply because police have many more encounters with male suspects who pose no threat than they do with female suspects who pose no threat? The answer is, of course, that if the rate of fatal police error is exactly the same for both men and women, we’d expect vastly more men to be shot simply because police have so many more encounters with male suspects.
In order to investigate whether one group of suspects is more likely to be shot than another in similar circumstances, you must know the number of such circumstances where nobody is shot, and this is what sets Roland Fryer’s study apart from all others. He had access not just to death reports, but to incident reports in general, including those where lethal force was not used. And he was able to code the specifics of the circumstances according to 290 variables. This allowed him to calculate the rates that a given set of circumstances would lead to use of lethal force for different groups. And the result clearly showed there was no detectable bias towards shooting black suspects.
I have been unable to find any study that supports the narrative of anti-black bias by police in the application of lethal force while properly accounting for the circumstances of shootings. The raw statistics, and the studies that account for those critical factors, both seem to agree that police do not more readily shoot blacks.
When these facts are pointed out to BLM proponents, one common response is to say that police are more likely to confront black people because of bias or racism. This artificially creates more police confrontations with blacks, so even if police are not more likely to shoot a black person in any given confrontation, because there are more confrontations, it results in an excessive number of shootings. In other words, police aren’t legitimately responding to vastly higher crime rates in many black communities in order to protect the residents. Rather, racial bias causes black communities to be “over-policed,” which causes more confrontations with police officers.
But, even a cursory examination of crime rates shows the flaws in this argument. As an example, consider the city where I live, Boston. Every year Boston has dozens of murder victims. Here are pictures of the victims:
Please take a minute to look at them. These were all human beings whose lives were cut short by the brutal violence of neighborhood criminals. Their families and friends will never stop grieving for them. Each murder victim leaves an indelible mark on the entire community. Each leaves hundreds of neighborhood children traumatized, unable to focus on school and building the skills they need to be successful in life, always on guard, wondering if they will be next.
If you look through their faces, you’ll quickly notice that there are hardly any pictures of whites among Boston’s murder victims, despite the fact that there are roughly twice as many whites as blacks in the city. That’s because nearly all the murders happen in predominantly black neighborhoods, like Dorchester and Roxbury. In my neighborhood, Jamaica Plain, right next to Dorchester and Roxbury, but skewing somewhat wealthier and whiter, there are few if any murders each year.
The reason that police have more confrontations in predominantly black neighborhoods in Boston is because that is where the great bulk of violent crime is occuring. (Murders are a valuable proxy for violent crime in general because murders, unlike other crimes, rarely go unreported, and those reports can’t be inflated. As a result, they are not as susceptible to statistical manipulation, biased police reporting, differences in rates of calling the police, etc.) These neighborhoods are plagued and traumatized by the most violent criminals. I wonder how people can claim that the reason there are more encounters and arrests in these neighborhoods is not because there is vastly more violent crime, but rather because these neighborhoods are “over policed”. If there is not vastly more violent crime in these neighborhoods, why do almost all the murders happen there?
Looking at an anecdotal example like Boston is instructive, but there are also systematic investigations of the question of whether black communities are “over policed”. The Justice Department’s Bureau of Justice Statistics released a report looking into exactly this question on a national scale. “It found that for nonfatal violent crimes that victims said were reported to police, whites accounted for 48% of offenders and 46% of arrestees. Blacks accounted for 35% of offenders and 33% of arrestees. Asians accounted for 2% of offenders and 1% of arrestees. None of these differences between the percentage of offenders and the percentage of arrestees of a given race were statistically significant.” [Emphasis added.]
In plain english, the number of arrests for violent crime is proportional to the number of violent crimes actually committed by each group. Black people are not arrested at a rate disproportionate to the number of crimes committed, suggesting that black neighborhoods are not “over policed”. Instead, the reason more blacks are arrested for violent crimes is because black neighborhoods suffer more from violent crime.
In turn, the primary reason there are more arrests, confrontations with police and, consequently, police shootings in predominantly black neighborhoods is because police disproportionately encounter perpetrators of violent crime there.
Evidence and Falsehoods
In summary, the only evidence that I’ve been able to find that controls for the circumstances of police shootings suggests that police do not more readily shoot blacks than whites (though possibly, shoot whites slightly more readily than blacks). And, the counter argument that so-called over-policing leads to more encounters, and thus more opportunities for confrontations that result in a shooting appears to contradict the data as well.
The core grievance of the BLM movement, that police are much more prone to use lethal force against black suspects, appears to be unambiguously false. The truth is that the best available evidence suggests that they are not.
Ferguson Effect: Devastation Inflicted by BLM Falsehoods
The effect of BLM’s falsehood that police more readily shoot black suspects has been the devastation of many low-income black communities. In 2014, after the shooting of Michael Brown in Ferguson, Missouri, the BLM movement’s anti-police rhetoric and propaganda found a receptive audience. As police were demonized with falsehoods, their morale declined and their willingness to engage in proactive policing, such as street stops for suspicious behavior and other forms of policing designed to prevent firearms crimes, plummeted. Police officers reported that they were scared or unwilling to confront suspects because any confrontation could escalate into a situation where they would need to use force. Any such situation could turn into a media circus where they would be scapegoated, their careers would be ended, their friends and community would cut all ties with them, and possibly, where they would even be wrongfully convicted and imprisoned. Without community support, many police officers reduced or even eliminated entirely their proactive policing. Thousands simply quit. Fewer police stops led to more guns and more criminals on the street. Murder rates, especially murder rates in low income black neighborhoods—where the police were most reluctant to confront criminal suspects—spiked.
This pattern of false anti-police rhetoric followed by reductions in proactive policing and spiking rates of violent crime, especially in predominantly black neighborhoods, was termed “the Ferguson Effect”. Initially, researchers sympathetic to the BLM movement were skeptical of whether the effect existed, but there is now a growing consensus that the Ferguson Effect is both real and devastating.
Evidence and Magnitude
After completing his landmark study on police shootings, and absorbing the shock of his results, Roland Fryer, the star black Harvard economist who, initially, at least, supported BLM, undertook a second effort: to verify or debunk the Ferguson Effect, and quantify its magnitude. After an exhaustive statistical analysis, he concluded that not only was something like the Ferguson Effect real, but in just the five cities he examined, it caused a staggering 900 excess murders, and 34,000 excess felonies that would not have otherwise occurred—and it was expected to cause hundreds more murders in those cities in the following years. Extrapolated to other cities and time periods this result suggested thousands of additional murder victims nationwide.
Other researchers also studied the question. One of the field’s most prominent researchers, Richard Rosenfield, was initially skeptical, but after re-examining the data, ended up changing his mind. ‘“The only explanation that gets the timing right is a version of the Ferguson effect,” Rosenfeld said. Now, he said, that’s his “leading hypothesis”.’
Incredibly, the study that I’ve seen most commonly cited to refute the Ferguson Effect states the following:
No evidence was found to support a systematic post-Ferguson change in overall, violent, and property crime trends; however, the disaggregated analyses revealed that robbery rates, declining before Ferguson, increased in the months after Ferguson. Also, there was much greater variation in crime trends in the post-Ferguson era, and select cities did experience increases in homicide. Overall, any Ferguson Effect is constrained largely to cities with historically high levels of violence, a large composition of black residents, and socioeconomic disadvantages. [Emphasis added.]
In other words, the Ferguson Effect has not been experienced broadly throughout our entire society. Instead, it’s been focused in exactly the cities you’d expect: those with large numbers of residents living in low income, predominantly black neighborhoods plagued by violent crime. Far from refuting the Ferguson Effect, this study actually bolsters the theory even further.
As an example of how this study is cited, a CNN article says the Ferguson Effect “has been challenged in academic research as anecdotal rather than data-driven and evidence-based”. In contrast, according to CNN, a data-driven approach found that “any Ferguson Effect is constrained largely to cities with historically high levels of violence, a large composition of black residents, and socioeconomic disadvantages.”
It’s hard to see how challenging the validity or importance of the Ferguson Effect because the devastation is only felt in low income black neighborhoods is not overtly racist. The direct implication seems to be that those neighborhoods don’t really matter. But, there are hardly any studies that challenge the Ferguson Effect, so CNN used the one that was available.
“The Minneapolis Effect”
In 2020, the theory was tested again when protests and riots swept across the country following George Floyd’s death while in police custody. The covid pandemic lockdowns had been underway for months by then, and many kinds of crime were predictably down as a result of fewer people being out and about. However, as anti-police rhetoric and propaganda increased after Floyd’s death, once again, police reduced proactive policing and murders spiked. This time, even more than in 2016. One top expert in the field estimates that the result of de-policing during June and July of 2020 alone resulted in an additional 1,520 murders. He explains:
“Crime rates are increasing only for a few specific categories—namely homicides and shootings. These crime categories are particularly responsive to reductions in proactive policing. The data also pinpoint the timing of the spikes to late May 2020, which corresponds with the death of George Floyd while in police custody in Minneapolis and subsequent anti-police protests—protests that likely led to declines in law enforcement....police officers have scaled back on proactive or officer-initiated law enforcement, such as street stops and other forms of policing designed to prevent firearms crimes.”
Other top researchers in the field concur. De-policing as a result of false anti-police rhetoric is causing a massive spike in homicides, mostly in predominantly black neighborhoods. The graphic below shows that there was no spike in murders for three months after pandemic lockdowns started (shaded pink), but that murders suddenly spiked after protests following George Floyd’s death (the red line).
Putting Harm In Perspective
The growing consensus in the field of criminology that a decline in proactive policing is resulting in drastic increases in murders, is so substantial that even strongly left leaning media outlets, like CNN, which have typically been supportive of the BLM movement, are starting to acknowledge the role of de-policing.
Regardless of whether you call it the Ferguson Effect or the Minneapolis effect, if you add up the estimates of murders from the different studies in various cities and time periods, you get something in the neighborhood of 2,500 additional murders on the lowest end, but, possibly, well over 10,000 on the high end.
While it may be difficult to pin down an exact number, what’s clear is that thousands of black people have been murdered as a result of BLM’s falsehoods villainizing the police, and the resultant anti-police sentiment that makes police even more wary of confronting criminal suspects.
It’s worth taking a moment to put these numbers in perspective:
18 unarmed blacks shot by police annually
26 unarmed whites shot by police annually
2500 (at least, but possibly well over 10,000) additional murders—mostly black—as a result of the de-policing prompted by BLM falsehoods
8000 blacks murdered by criminals annually
It would take roughly 140 years for police to shoot as many unarmed black people as have been murdered as a result of BLM falsehoods in just the past few years. But, the thousands of additional black murder victims are just the tip of the iceberg of devastation that BLM falsehoods have inflicted on black communities. For each victim murdered by criminals there are dozens of lives derailed; hundreds of children traumatized.
Perhaps even greater than the deaths and trauma that result directly from BLM’s falsehoods, is the damage done by drawing attention away from the real solutions to the approximately 8,000 black people murdered annually. The tragedy of the BLM movement is not just the additional murders and devastation to low income black communities that its falsehoods have caused directly, but also how those falsehoods retard progress on tackling the violence which was already plaguing those communities before BLM even came along.
How could we possibly have gone so wrong?
How Systemic and Structural Patterns that Disadvantage Black People Propel the BLM Movement
As the above research has emerged proving the enormity of the devastation wrought by the BLM movement’s falsehoods, I’ve spent considerable time wrestling with the question of how our society could get something so important, so utterly wrong.
My Hometown: Newton, Massachusetts
As I’ve considered that question, my mind keeps coming back to where I grew up, a wealthy suburb of Boston called Newton. It’s rare for there to be even a single murder in Newton each year. According to city-data.com, the last murder was well over a decade ago. By and large, Newton residents simply don’t need to worry about the safety of themselves or their children.
Nevertheless, everyone I know in Newton is supportive of the BLM movement, the city has held BLM rallies, the mayor and other officials have made statements of public support, and as you drive through the city’s neighborhoods you will often see BLM yard signs. I have the strong impression that Newton as a whole is very supportive of the BLM movement.
I try to imagine how residents in a city like Newton would react if instead of zero murders annually in their city, residents were being murdered by the dozens, every single year—as actually happens in the nearby Boston neighborhoods of Roxbury and Dorchester. If Newton were plagued by such criminal violence, would we be hearing calls from concerned Newton parents to reduce or even eliminate proactive law enforcement, such as street stops and other forms of policing that are designed to prevent firearms crimes?
It’s almost impossible even to imagine Newton parents calling for less officer-initiated policing if their—mostly white—children were the ones in danger and being murdered. So, why do so many Newton residents think that is the right answer in neighboring Dorchester and Roxbury, where other people’s—mostly black—children are being murdered?
Why Do People Support BLM Despite It’s Devastating Impact on Black Communities?
I know many BLM supporters, and I fully believe their sincerity when they profess to be against racism. Nevertheless, to someone familiar with the facts, it seems like it would take an almost willful blindness not to see the dangers that BLM’s fraudulent villainization of police poses to violence plagued communities. Fatal police shootings of unarmed blacks account for approximately 0.18% of black homicides each year—less than one fifth of one percent of the black people murdered each year by neighborhood criminals. The communities where those murders are occuring are being crushed by violence, but not from the police. How could anyone who actually cares about the black people living in those communities not at least suspect that the radical decreases in proactive policing, and policing in general, called for by the BLM movement would have deadly consequences for those communities?
The best explanation I can come up with for why a person (white, black or any other race) would support the BLM movement, is ignorance of even the most rudimentary facts. For example, support for BLM correlates very highly with being more liberal, and a recent survey found that among those who describe themselves as “Very Liberal”, more than 50% believe law enforcement killed 1,000 or more unarmed black men in 2019. Nearly 8% believed they killed more than 10,000! According to the Washington Post, the real number of unarmed black men shot and killed by police in 2019 was 11. That’s a difference of 3 orders of magnitude. It’s impossible to reason intelligently when your beliefs about the relevant facts are so completely divorced from reality.
For those reasons, I don’t believe that anti-black racism is a primary factor in explaining why so many people support BLM. Rather than racism, rank ignorance appears the likely culprit.
The Role of Press, Politicians and Academia
But, if ignorance of the most rudimentary facts is the true culprit, that raises the question, how have the normal channels for educating people about social policy questions failed so completely? After all, most people believe what they read and hear in the media. Most people are not experts. If they read in the media over and over again that police are targeting blacks with lethal force, and they never read anyone contesting that claim, then they’ll tend to assume it’s an accepted fact.
But, it’s not quite so easy to excuse the media for failing to inform them of the key facts. Verifying, scrutinizing and reporting on the veracity of people’s claims is literally the job description of the news media. That’s what they’re there for. That’s what people depend on them to do. Whereas consumers of news media may (problematically) assume that job is being done for them, reporters are never supposed to assume that claims are true without research, and it’s literally their job to provide the key background facts to their readers.
It may be tempting to excuse the news media with the defense that they have just been reporting on newsworthy events. BLM activists have been organizing protests, posting on social media, writing books, and so forth. Even if journalists disagree with activists it is still their job to report on what they are saying. And, particularly, it’s the news media’s responsibility to report on what black voices are saying right now because of the legacy of oppression that black people have suffered. Therefore, the news media bear no responsibility for their audience’s failure to grasp even the most basic facts. They’ve just been reporting the news.
The problem with this argument is two fold. First, as mentioned above, it is never the news media’s job to uncritically report falsehoods. Where people are asserting misinformation, it’s the news media’s job to accompany reports of those claims with key facts contextualizing that information. But the second reason is perhaps more disturbing.
Disregarding Black Scholars
There has been a sizable contingent of highly respected black scholars, intellectuals and public figures who have been studying the challenges faced by low income black communities for decades and who have been vocally pointing out the falsehoods of the BLM movement. Examples include Thomas Sowell, Glenn Loury and John McWhorter. Black scholars like these have been struggling to get their message out, and their message is scathing. Sowell, for instance, describes BLM as “self serving” and is despairing about the damage the movement is doing to black communities:
“Even though I’m regarded as pessimistic, I was never pessimistic enough to think that things would degenerate to the point where they are now, where adult human beings are talking about getting rid of the police, where they’re talking about reducing the number of police, reducing the resources put into police work, at a time when murder rates have been skyrocketing over what they were just a year ago in 2019. And, what is frightening, is how many people in responsible positions are caving in to every demand that is made, repeating any kind of nonsense that you’re supposed to repeat.”
Unfortunately, these critics have found it virtually impossible to get a similar amount of air-time, op-eds, positive references to their research, and glowing biopics in corporate media and academia as BLM activists and proponents receive daily from the news media, professors and politicians.
There is nothing accidental about this. Famed journalist and former NYT editor Bari Weiss published this account from a high school teacher on her blog:
Since the BLM protests often came up in our discussions, I thought of assigning Glenn Loury, a Brown University professor and public intellectual whose writings express a nuanced, center-right position on racial issues in America. Unfortunately, my administration put the kibosh on my proposal.
The head of school responded to me that “people like Loury’s lived experience—and therefore his derived social philosophy” made him an exception to the rule that black thinkers acknowledge structural racism as the paramount impediment in society. He added that “the moment we are in institutionally and culturally, does not lend itself to dispassionate discussion and debate,” and discussing Loury’s ideas would “only confuse and/or enflame students, both those in the class and others that hear about it outside of the class.” He preferred I assign “mainstream white conservatives,” effectively denying black students the opportunity to hear from a black professor who holds views that diverge from the orthodoxy pushed on them.
Why have the voices of black activists spreading devastating falsehoods been relentlessly promoted by the media, schools, universities and politicians while highly esteemed black scholars whose research would have exposed those falsehoods, and the damage those falsehoods are inflicting, been largely ignored? The news media has not simply been reporting the news. Academics have not simply been researching in pursuit of truth. Rather, through careful choices of which voices they elevate and promote, they’ve been shaping a narrative. And that narrative neglected the voices, and crucial insight, of some of the most distinguished black professors and leaders in the country.
Structural and Systemic Patterns that Disadvantage Black People
The history of this neglect may not be written for a long time, seeing as how BLM’s ascent appears to continue unabated. But when that history is someday written, it will not be kind to those who played key roles in enabling and promoting falsehoods of such tremendous destructive force.
Ultimately whether journalists’, editors’, and media owners’ decisions to promote destructive falsehoods instead of investigating and reporting the truth was motivated by racial animosity may be unknowable. We can’t look inside their heads to see what was going on.
But, even if it turns out that overt racial animosity did not play a role in promoting BLM, it seems there could hardly be a better example of how structural and systemic patterns that come from a history of racism continue to disadvantage black people today. Nor can I think of any contemporary example of such patterns that can match the sheer destructive impact that BLM has had on many black communities.
How has the press so totally failed to report the key facts about BLM’s falsehoods? Maybe it’s just because so few of the key media owners, editors, and journalists, were the ones at risk of being murdered as a result of those falsehoods. They were mostly wealthy and mostly white, living far from the impacted communities. The incentives weren’t aligned for those key people in power to produce hard hitting journalism because it wasn’t their lives on the line. It wasn’t necessarily any person’s antipathy for blacks that has propelled the BLM movement to such heights, but rather the structure of the system itself, where the most powerful decision makers were insulated from the devastation wrought by their delinquent click motivated journalism. And how was that system, where the key decision makers were insulated from the harmful effects of their decisions, established? It’s hard not to see, as the primary culprit, the legacy of years of oppression that have ensured that so many of the key decision makers would be wealthy and white.
Absolution for BLM?
Some BLM proponents may object that, even if they were to concede that the core BLM premise that police more readily shoot blacks is false, and even if they were to concede that BLM rhetoric asserting that falsehood has resulted in the murders of thousands of black people, there is still a lot of positive that has come out of the movement. After all, if you read Fryer’s research, for example, you will see that while he found that police were possibly more likely to shoot white suspects, he also found that police were substantially more likely to use some sort of non-lethal physical force against black suspects. Aren’t such disparities worthy of discussion and examination?
And, I would tend to agree that poorly understood phenomena like that are worthy of further research. Why would police possibly more readily shoot and kill white suspects, but less readily shove or hit them? Personally, I find the model that Fyer proposes to explain this strange discrepancy to be unconvincing, but I certainly agree with him that it’s an important area for further research. Should we grant the BLM movement absolution for the devastation wrought by its falsehoods, because it is simultaneously drawing attention to these unanswered and important questions? That might be a more compelling argument if I saw any evidence that the BLM movement was spurring real research into these questions as opposed to obscuring them, or possibly, in some cases, making it impossible to research them at all.
The Power of a Name
One of the most pernicious aspects of the BLM movement is its name. A more accurate and descriptive name for the movement might be something like the “Anti Proactive Policing Movement”. The movement’s rhetoric, its focus on organizing around police shootings, and its actual impact are all, to a significant degree, centered on the reduction of proactive policing. If the movement were named honestly, by its policy objectives, there would be much more room for people to disagree with those policy objectives and for a rational dialogue to ensue. Instead, by adopting the name Black Lives Matter, the activists who invented the phrase put everyone who disagrees with their misguided policy objectives in the position of defensively distinguishing their disagreement with the policy from their agreement with the slogan. It was a good rhetorical strategy, but it wasn’t good for the black communities they were trying to help.
Also troubling is how the name elides the considerable difference of opinion within the black community itself, both in regards to the methods and rhetoric of the movement and the policy objectives the movement promotes. Many black people do not share BLM’s hostility to police and don’t want less police or less policing.
The truth is that, with the exception of a small handful of marginalized misfits, today nearly everyone in America believes that black lives matter. In my entire life, I have never met a human being who said or implied that black lives don’t matter. But, nobody should support the Black Lives Matter movement: it’s a poisonous falsehood, uncritically promoted by corporate media, that is devastating many black communities.
Thomson Reuters Must Do Better
Ultimately, I don’t believe it’s Thomson Reuters’ role to affirm or renounce social movements. Nevertheless, over the past few years, witnessing open and pervasive support within Thomson Reuters for a movement that is having such a devastating impact on the most disadvantaged black communities has made the work environment feel untenably hostile for me. I have frequently felt the imperative to speak out against the anti-black bias and devastation but have instead held my tongue because I was scared of the consequences. Indeed, a few months ago, my feelings of alienation reached a point that I couldn’t tolerate it anymore, and I asked to take unpaid leave. Even then, I kept my concerns to myself and didn’t share them with my manager and leadership, because, I feared, I could be fired for even letting them know why I want to take a leave, and possibly even informally blacklisted in the job market if rumors spread to other employers. But, when I made the decision to return to Thomson Reuters after my leave, I knew I could only justify returning to myself if I had the courage to stand up for the truth. I cannot live with myself in an environment where people freely express uninformed support for a movement inflicting such destruction in the most disadvantaged black communities, without, at the very least, offering an alternative perspective based on research and evidence. Perhaps more importantly, I cannot ethically work at a company that is the home for Reuters News, one of the most important and widely respected news agencies in the world, without working to bring attention to potentially severe problems in our reporting.
I’ll also note that just like I understand that it is not Thomson Reuters’ role to affirm or renounce social movements, I also understand that it is doubly not the role of Reuters News to do so. Reuters News has a commitment to remaining as objective and unbiased as possible so as to retain the great trust that our readership places in us. However, like just about every other corporate media outlet, I believe Reuters News needs to do better in informing its readership about the basic facts so the readers can make up their own minds about the truth or falsity of BLM’s claims, and the magnitude of the devastation that the movement is inflicting on so many black communities. Readers can only make up their own minds intelligently if they have the key information, and, as discussed above, polls of people’s beliefs about even the most rudimentary facts show that the key information is not being widely disseminated and absorbed by readers.
Thus, I believe that we at Thomson Reuters need to do better. I also believe that we have a strength of community and sense of purpose that means we can do better. I hope we do.
Last time I posted on the hub speaking out against systemic and structural societal patterns and institutions that harm marginalized groups and against the ways that people at Thomson Reuters may be unintentionally helping to perpetuate such patterns and institutions, someone flagged my post by ticking the “Report abuse” checkbox and my post was temporarily taken down. One of the sad facts about fighting against such patterns and institutions is that often people perpetuating and promoting them, frequently unintentionally, are more interested in silencing examination of how their actions disadvantage minorities than they are in dialogue and self-examination. If you believe this post should not be taken down, please let HR know, and please forward this post. If it is taken down, I would be more than happy to provide a PDF version to anyone who asks, for dissemination throughout the company. The fight against bias that harms marginalized groups is too important to let it be thwarted by those who wish to silence open discussion. Discussion of race, and the ways that institutions in our society continue to perpetuate racial disadvantage are uncomfortable and difficult, but it is essential that we have them nonetheless.
For people who have not studied statistics or econometrics recently, or at all, it may be difficult at first glance to understand the problems in many of the studies that purportedly show bias in police shootings. However, one of the most common problems should be straightforward to understand even for someone with no background in statistics with just a few minutes studying some illustrative hypothetical examples. In this Appendix, I present such examples, as well as a more in-depth discussion of some of the key research in the field.
Meaning and Identification of Bias
If there is bias in police shootings, that means that if all but race were equal in two situations, the black person would be more likely to be killed than the white person. In other words, in a particular type of situation, for example, where a large unarmed male suspect is physically resisting arrest, police would be more likely to shoot the suspect if he is black, than if he is white (or vice versa). Thus, if we look at all the police encounters with large, unarmed, male suspects, and we find that there is a higher rate of shootings for black suspects than white suspects, that suggests that there is bias.
Why did I specify specific adjectives like “large”, “unarmed”, and “male” in my example? Because police use force in response to a perceived threat, and because different sizes, strengths, attitudes, and weapons of a suspect can radically change the threat they pose, it’s critical to compare apples to apples. There is a much lower rate of police shootings when police are arresting small, unarmed, women than there is when arresting large, armed, men. Ideally we’d also like to know other details such as the time of day, whether police came to the scene as a consequence of a report of illegal activity, whether the suspect was resisting arrest, whether there were innocent third parties endangered by the behavior of the suspect, and so forth. The more variables we can control for that might have an impact on use of lethal force, the more we can isolate the impact, if any, of race specifically.
Unfortunately, oftentimes the available data does not have all the granularity you would like to have in the ideal world, and instead just uses gross categories like “unarmed” and “male”. (The Fryer study was an exception, and through an extensive manual labeling process obtained data for 290 relevant details for each example.) For simplicity's sake, in the following examples, we’ll assume we are working with a dataset that only codes incidents according to race and whether the suspect is armed.
Important Note: In all the hypothetical examples below, I’ve drastically inflated the rate of shootings from anything remotely realistic, so that shootings are visible in the charts. Please do not walk away thinking that police shoot 8% of unarmed suspects, when the true number is likely closer to 0.001%. I’m just making up extreme numbers in order to make possible a visual illustration of a key mathematical fact.
Consider the following hypothetical data suggesting that there is not a racially biased application of lethal force. Percentages represent the rate of killing per encounter.
For both groups there were about 500 encounters, of which about 40 resulted in a fatal shooting, so the rate of fatal shootings is the same for each group. This means that when faced with an unarmed suspect resisting arrest, there is no evidence that police responded differently to the suspect based on their race.
Compare the graph above with the graph below.
In this new hypothetical, there are again the same number of shootings for each group, but this time we can see that there were many more police encounters for whites than for blacks. This suggests that the rate of shooting was higher for blacks (8% vs just 5%). Thus this graph supports the hypothesis that there is bias in police shootings.
Now consider another hypothetical:
In this chart, there are twice as many shootings of blacks than whites. But, in this chart, there are also twice as many encounters with blacks than whites, so the rate at which an encounter leads to a shooting is the same. This suggests that in any given encounter police are no more likely to shoot the suspect based on their being black, and thus this data does not support the hypothesis that police more readily shoot black people (though it says nothing about whether there is biased “over policing” as discussed at length in the main body of this post).
With that understanding of bias, let's look at a more complicated hypothetical example. Here we have data for both armed and unarmed shootings. Again, the rate of shooting for each racial group in each category is the same. Armed blacks and armed whites are both shot at a rate of 50%. And unarmed blacks and unarmed whites are both shot at a rate of 8%. In this hypothetical, there are approximately twice as many armed whites (250) killed as armed blacks (125), however, that does not imply any bias on the part of police, because whites are having twice as many encounters with police—in any given encounter, police are no more likely to shoot a white suspect than a black suspect.
Now, imagine a scenario identical to the above scenario, except now we are dealing with a dataset that only has information about encounters that lead to a fatal shooting. We have no data about encounters that do not lead to a shooting.
Because we have no data about non-shooting encounters, in this graph there are no blue bars, however the red bars are exactly the same length as they were in the previous graph. Notice also that the percentages listed are different. The lack of encounter data makes it impossible to calculate a rate of shooting for each scenario type. Because it is impossible to calculate a rate of shooting, I’ve inserted a different percentage instead: the percentage of cases of each scenario type for each racial group. For instance, of the black suspects who were killed, 76% of them were armed, and 24% of them were unarmed. And for the white suspects who were killed, 86% of them were armed and only 14% of them were unarmed.
One may look at those numbers and think: the percentage of blacks shot while unarmed is much higher than the percentage of whites shot while unarmed, therefore police are more likely to shoot unarmed blacks. However, as we know from the previous graph, police were not more likely to shoot unarmed blacks than whites in any given encounter, and the number of shootings is perfectly consistent with completely unbiased application of lethal force.
Indeed, in this hypothetical, the only reason a greater percentage of the black people shot are unarmed, is because there are many more armed whites shot than armed blacks. The number of unarmed encounters and the number of unarmed shootings is the same for black and white, indicating that there is no bias.
Incorrect Use of Study Results to Argue for Bias
The example above of a mistaken inference of bias illustrates exactly the problem with many of the studies that people use to (falsely) argue that there is racial bias in police shootings. Two in particular which I’ve recently seen cited to support the claim of bias are both based on the National Violent Death Reporting System (NVDRS). One, which I described above, was published in 2020 and the other was published a few years earlier in 2016. When used to argue that police more readily shoot blacks, they both suffer from the same problem: NVDRS has no information about the number of police encounters, and thus it is impossible to calculate the rate of shooting per encounter.
For example, the 2016 study states that:
Black victims were significantly more likely to be unarmed than white or Hispanic victims. Black victims were also significantly less likely than whites to have posed an immediate threat to LE. White victims were significantly more likely than black victims to be killed in incidents related to mental health or substance-induced disruptive behaviors and more likely than black or Hispanic victims to be involved in potential “suicide by cop” incidents. Hispanic victims were also more likely than black victims to be involved in a potential “suicide by cop” incident. Incidents involving black and Hispanic victims were more likely than those involving white victims to have at least one black LE officer involved in the fatal injury. [Emphasis added.]
It’s a common but catastrophic type of statistical mistake to infer that police more readily shoot unarmed black people who pose no immediate threat based on the first two sentences in bold above. To understand why, consider that during the study period there were approximately the same number of unarmed whites killed, 40, as blacks, 39. Therefore, if there were the same number of police encounters of the type that leads to an unarmed shooting, then the rate per encounter would be nearly identical for whites and blacks, indicating an absence of bias.
This is precisely the same inferential error that I illustrated in the hypothetical example above. There, as here, it was impossible to calculate rates of shootings because the dataset had no data about the number of encounters. Instead, in the hypothetical, and in the study above, a different set of percentages is calculated: the percent of suspects killed who were armed, and the percent killed who were unarmed. Because a greater percentage of blacks who are fatally shot are unarmed, to a reader without a background in statistics, those percentages have a tendency to give the impression that there was some bias, when in fact it says literally nothing about the presence or absence of bias.
But, since there were the same number of unarmed blacks and unarmed whites fatally shot, for these numbers not to suggest bias, there must also have been the same number of encounters that led to such shootings for both groups. Is it plausible that there would be the same number of encounters? If policing resources for reducing violent crime are focused on neighborhoods that experience the most violent crime, as they should be, then we would expect the number of police encounters to be proportional to the amount of serious violent crime committed. In that case, the answer is yes, since according to the evidence reviewed in the main body of my post, blacks and whites commit roughly the same number of serious violent crimes and are arrested for those crimes in similar numbers (despite the fact that whites are a greater percentage of the population).
If the rates of police shooting of unarmed suspects are similar, why would unarmed suspects be a greater proportion of blacks shot than of whites shot? One possible explanation is for exactly the same reason this was true in the hypothetical examples above. If there were more armed whites shot than armed blacks, this reduces the proportion of whites shot who are unarmed. (But, just like in the hypothetical, this doesn’t actually tell us anything at all about whether there was bias in police shootings of unarmed suspects.)
Interestingly, in the italicized portion of the quote above, the study itself hints at a perfectly plausible explanation for why there would be proportionately more armed whites encountered by police than armed blacks: police have many more encounters with armed whites suffering from substance abuse or mental health issues, and more encounters with armed whites intending “suicide by cop”. Because police have so many more encounters with armed whites posing a threat, it makes perfect sense that unarmed whites would be a smaller proportion of whites killed compared to unarmed blacks.
Thus, this study gives us no evidence whatsoever to support the hypothesis of bias in police shootings. This is not merely splitting hairs, or nitpicking. Without data on the number of police encounters, it's impossible to calculate the rates of shootings, and thus impossible to identify bias. The data examined by this study is perfectly consistent with a complete absence of bias.
The second study based on the NVDRS suffers from exactly the same problems. The number of whites, 16, and blacks, 17, in the category of “likely unarmed” was, again, almost identical. Thus if there were the same number of encounters with police of the type that leads to unarmed shooting, then the rates of shooting per encounter would be identical. This would indicate a lack of bias.
(It’s also worth noticing that according to the Washington Post police shooting database, over the last five years, police have killed 39% more unarmed whites than blacks. However, in 2012, the subject year of the first study, and 2015, the subject year of the second study, there were almost the same number of unarmed blacks and whites shot by police, suggesting the possibility that these were unusual years. It appears that both papers reach back 4-5 years to find periods that don’t exhibit the pattern of the last 5 years where substantially more unarmed whites have been killed than blacks. In neither paper do the authors explain why they choose to examine the years they choose, or why they didn’t simply use all the data, or all the most recent data.)
Comparison to Fryer Study
It’s worth contrasting the methods in the papers above, which are incapable of examining whether there was police bias, with the method in Fryer’s paper, which although drastically more labor intensive, is designed specifically to determine if there is police bias. Fryer looked at thousands of police arrest reports, and coded each report according to over 290 variables. Crucially, he took a random sample which included cases where police did not shoot suspects. This allowed him to calculate whether there was a different rate of shooting for different racial groups when controlling for relevant variables like the behavior of the suspect. As he says, “A simple count of the number of police shootings that occur does little to explore whether racial differences in the frequency of officer-involved shootings are due to police malfeasance or differences in suspect behavior.” That is the crux of the issue with using the two NVDRS studies above to argue that there is police bias. The NVDRS data simply does not contain the data necessary to support that contention.
One of the most interesting facts about Fryer’s study is that in addition to running a proper statistical test for bias, as an experiment he also calculated the same flawed statistics that are frequently falsely used as evidence of bias (like those discussed above). When he used those improper methods to estimate bias with his data, his data showed the same supposed bias as other studies using those improper methods. However, when using proper methods that bias disappeared completely:
Perhaps the most striking finding is when one replicates the analysis in Ross (2015) across all five datasets: calculating the probability of being black, unarmed, and shot by police divided by the probability of being white, unarmed, and shot by police. A quantity greater than one is consistent with racial bias. Using the data from Ross (2015), this ratio is 3.28. Using the data from the Post database I get 6.20 and 5.99 if using the data in Fryer (forthcoming). In other words, if I ignore the detail available in the Fryer data and simply report the descriptive statistics reported in Ross (2015), I could conclude that the data provided evidence of even more racial bias than that reported in Ross (2015). Yet, when using the simple statistical framework that economists have used for more than a half century to analyze racial differences on myriad dimensions – from wages to incarceration to teen pregnancy – the evidence for bias disappears. The differences in results on police shootings in America seem to be driven by differences in what qualifies for a valid research design and not differences in datasets.
The fact that the improper methods showed the same supposed bias when applied to Fryer’s dataset, but the bias completely disappeared and even reversed when using proper methods, underscores the point that the studies using flawed methodologies that purportedly show bias actually carry essentially no evidentiary weight for establishing bias.
Limitations of Fryer Study
Given that Fryer’s study upended unwarranted assumptions held by many people about the biased application of lethal force, it’s not surprising that the study precipitated a torrent of criticism. While much of that criticism seems to be motivated, at least in part, by the political and social agendas of the critics, there are some important limitations and caveats worthy of discussion.
Data Limited to Houston
The most obvious limitation of Fryer’s study is that the most complete data that he examines comes from just one city, Houston. Critics have raised the concern that there may be selection bias in how the city was chosen, and/or that the city may not be representative of patterns in other cities.
However, there are sound reasons to believe that police use of force is no different in Houston than in other major cities, and also that there was no selection bias in the choice of Houston. In regards to selection bias, according to Fryer, he selected Houston because it had the most comprehensive set of officer-involved shooting data:
“The most comprehensive set of officer-involved shooting data is from the Houston Police Department. For this reason, we contacted HPD to help construct a data set of police-civilian interactions in which lethal force may have been justified. If we had the data from other cities, we would definitely use it.”
This means that Houston didn’t somehow self-select into the study by being the only police department willing to share their data. Rather, Fryer identified and approached Houston because Houston had the most complete data.
In regards to Houston being idiosyncratic, it’s important to remember that Houston shows the same surface level disparities that other cities show. Further, as described above, when Fryer calculated the improper statistical tests typically used to falsely demonstrate bias, Houston showed the same supposed bias as other cities. It was only when he did a proper econometric test that the disparities disappeared.
Therefore, there is no reason to suspect that Houston is not representative of a general pattern in police shootings. And, at the very least, Fyer’s findings in Houston are a definitive demonstration that the surface level disparities and improper statistics used to falsely argue for bias in other cities, actually carry no evidentiary weight for establishing bias.
Nevertheless, in order to prove conclusively that there is no bias in police shootings nationally, ideally studies like Fryer’s would be conducted in other cities.
Police Bias in Initiating Interactions
If police are biased towards perceiving black subjects as more threatening than they are, they may stop black subjects who pose less of a threat more often than white subjects. Since the black subjects turn out to be less dangerous and more compliant, there would consequently be a lower rate of escalation, and thus a lower rate of police shootings per stop, compared to whites, thus biasing the data. One paper critical of Fyer’s study explains the problem like so:
“This could occur if officers have a higher threshold for stopping white civilians during the unseen first stage of police-civilian contact, meaning that white civilians observed in the data are incomparable because they tend to pose a greater threat to police than observed minorities.”
Another critic says:
“if even a small subset of police have propensities to more frequently encounter black relative to white individuals, then analyses of pooled encounter-conditional data will fail to detect systemic anti-black racial disparities in the encounter-conditional use of lethal force by the larger subset of police.”
Fryer himself has discussed this limitation in his original paper and subsequently:
“I totally agree that deciding who to stop in a police stop is highly problematic and there certainly may be racial bias in that decision. So let’s think about the officer-involved shootings in which there’s a robbery in progress or a violent crime. Those are less likely to be plagued by selection bias in the decision of who to harass or stop. Analyzing only those cases yields similar results.”
In other words, if you are concerned that police bias in initiating encounters with suspects were driving Fryer’s finding that there wasn’t bias in police shootings, then you can look just at encounters where police exercised no judgement or discretion in their choice of whether to initiate an encounter. For example, if police are called to a location to respond to a violent crime or robbery that is in progress, then they do not have an opportunity to exercise discretion (and therefore bias) as to whether they engage a suspect at all. But, when you look at only those cases where police do not have an opportunity to introduce bias into the encounter rates, the findings do not change: the data still do not show bias in police shootings.
It’s also worth mentioning that even if police were introducing bias in the encounter rates by stopping more non-threatening black subjects, and thereby decreasing the proportion of encounters that are at high risk of leading to a shooting, you would expect Fryer’s econometric analysis to control for that, at least to the extent that the 290 variables that his team coded were capable of distinguishing between non-threatening subjects and threatening ones, and thus isolating the impact of race.
Bias in charging decisions
In addition to police bias in initiating encounters there is another kind of police bias that could impact Fryer’s study results: police characterization of encounters.
“A second concern is the reliability of police department reports. There are two types of potential bias. First, police officers may bend the truth about the context of a particular interaction so as to justify their own actions; for instance, indicating a suspect was threatening when they were calmly following an officer’s commands. This type of bias is less of a concern in Fryer (forthcoming) because the qualitative results are identical whether or not one includes contextual factors about the encounter recounted by police.”
Because Fryer’s results were the same whether he ran his analysis including or excluding police characterizations, the bias in those characterizations cannot account for his results. However, there is one type of related bias Fryer was not able to devise a method to test or correct for:
“A second type of bias is that officers may be more likely to charge black suspects with crimes such as resisting arrest or attempted assault on a public safety officer rather than misdemeanors, relative to whites, for identical behavior. This type of bias is an important limitation of Fryer (forthcoming) because it implies that the counterfactuals coded from arrest data may themselves contain bias. It is unclear how to estimate the extent of such bias or how to address it statistically.”
Therefore, to conclusively prove a complete absence of police bias in shootings, future researchers would need to find a way to conduct a study that, like Fryer’s, could properly account for circumstances and encounter rates, while simultaneously not being biased by any differences in propensity to charge (if they exist).
“Not Peer Reviewed”
Perhaps the flimsiest criticism of Fryer’s research is that it wasn’t published in a peer reviewed journal. The reality, however, is that, anticipating the incendiary nature of his research findings, Fryer endeavored to subject this study to a much more rigorous and exhaustive process of peer review than most, or even any, journals, typically offer:
Fryer first presented his data, and collected feedback from his peers, in a seminar last summer at the National Bureau of Economic Research. He presented the work again to colleagues at Harvard, Brown, University College London, and the London School of Economics. Aware of the paper’s incendiary nature and importance, he sent the paper out to 50 colleagues asking for feedback. He also hired a second research team to recode all the data that he used for his analysis, just to make sure the results would replicate. Fryer’s paper was not “peer-reviewed.” It was reviewed, though, very thoroughly, and by a large number of his peers.
Not the Final Word… But, the best evidence we have today
In addition to the studies discussed above, I’ve reviewed others, and would be happy to discuss the research field with anyone at Thomson Reuters who is interested. Regardless, as is clear from the limitations discussed above, Fryer’s study is far from the final word on police shooting bias. There are plenty of unanswered questions left to be explored, and it’s certainly conceivable that at some point in the future, a new study that also uses proper methods to control for circumstances of encounters will contradict Fryer’s study by showing some evidence of police shooting bias.
Nevertheless, in my investigation so far, I have not been able to identify a single study using valid research design that has found bias in police shootings. To date, Fryer’s study is the only study using valid research design to test for bias, and it provides strong evidence for an absence of bias in police shootings (or, if anything, a slight bias towards shooting whites). As described in the main body of this post, these findings are also consistent with an analysis using high-level descriptive statistics. Thus, the best evidence available today strongly suggests that police are not biased towards shooting black suspects.
Thank you to the anonymous contributors who worked with me to write and research this blog post.