Among New York City precincts with similar levels of crime, precincts with a high proportion of minority residents experience significantly higher rates of stop and frisk than whiter precincts.
Last Friday, New York City filed an appeal of the US District Court ruling that the City’s Stop & Frisk policy, as practiced, utilized “indirect racial profiling” and was, therefore, unconstitutional. Mayor Michael Bloomberg and Police Commissioner Ray Kelly initiated their media defense that same weekend. Kelly made the rounds on the Sunday talk shows, and Bloomberg took to the OpEd pages of the Washington Post to defend Stop & Frisk.
This PR blitz gives insight into the line of defense the City will take in its appeal, which gives us an opportunity to refine a rebuttal. What follows won’t be a legal rebuttal, rather it will be a data rebuttal. First I’ll summarize Bloomberg’s and Kelly’s positions. Then I’ll present the data which may be helpful in supporting the legal rebuttal.
While the Mayor’s and the Commissioner’s defense of Stop & Frisk revolved around a number of issues, I'd like to highlight a couple where hard data can be clarifying.
1. Benchmarking - Bloomberg and Kelly insist that witness and victim descriptions of a violent crime suspect's race or ethnicity should be the benchmark against which race-neutrality should be measured. This isn’t exactly new, as it has been an ongoing defense used by the Mayor. The New York Times reported on a Bloomberg statement back in June indicative of both his defense of Stop & Frisk and his scorn for those who oppose him.
“They just keep saying, ‘Oh, it’s a disproportionate percentage of a particular ethnic group,’ ” [Bloomberg] said dismissively of the practice’s critics. “That may be, but it’s not a disproportionate percentage of those who witnesses and victims describe as committing the murder. In that case, incidentally, I think we disproportionately stop whites too much and minorities too little.”
He added: “It’s exactly the reverse of what they say. I don’t know where they went to school, but they didn’t take a math course. Or a logic course.”
2. Diverse Institutions Can't Be Biased – The Mayor and the Commissioner both insist the NYPD can’t be engaged in “indirect racial profiling” because they have black and Hispanic cops. As Kelly described how the judge's decision was an unfair indictment of hard working officers protecting the City, he told Bob Schieffer on Face the Nation, "It’s also ironic, here, that the New York City Police Department is the most diverse police department in the United States. We have police officers born in eighty-eight countries."
Bloomberg wrote:
As a black Brooklyn detective with nearly 20 years on the job recently told the Daily News, "Stop-and-frisk is never about race. It's about behavior." If an officer sees someone acting in a manner that suggests a crime is afoot, he or she has the obligation to stop and question that person. That's Policing 101, and it's practiced all over the country.
The Mayor need not worry. Math will be involved. First, we will examine the problem with the Mayor's benchmarking standard. Then, we will examine the social science regarding bias to establish that diverse institutions can indeed be biased. Finally, we will review the data that shows it. I compiled the data that follows from the NYPD’s 2012 report on
Crime and Enforcement Activity in New York City, their 2012
Reasonable Suspicion Stops report, plus the invaluable New York Civil Liberties Union 2012
dataset of UF-250 Stop, Question and Frisk Reports.
BENCHMARKING
Stop & Frisk defenders point out that the racial and ethnic distribution of stops mirrors the racial and ethnic distribution of witness descriptions of violent crime suspects. But they leave out a number of crucial details.
Violent crimes account for 37% of the City’s total crime. The focus on violent crimes as a benchmark means that the remaining 63% of property crimes are ignored. Further complicating the picture is that the proportion of suspects identified as black is much higher than the proportion of arrestees who are African American. The race or ethnicity of a suspect is reported in 70% of crimes. The remaining 30% have no description of race or ethnicity. This further skews the benchmark Bloomberg and Kelly insist on using.
To understand how this is problematic, we can impute the racial distribution of those 30% of suspects based on the racial distribution of arrestees. The numbers are eye-opening.
NYPD Commissioner Ray Kelly claims that the distribution of race or ethnicity based on descriptions by violent crime victims and witnesses is the best benchmark for evaluating the race-neutrality of Stop & Frisk. However, arrest rates imply a dramatically different distribution of suspects whose race or ethnicity is not known, suggesting a racial bias in suspect descriptions.
While black suspects comprise nearly 65% of known suspects, they make up less than half of that (32%) among unknown suspects. Meanwhile, the representation of white suspects is more than double, moving from 7% of known suspects to 15% of unknown suspects. This disparity, at a minimum, suggests a reporting bias among witnesses. Ignoring this disparity when implementing policing strategies will cause the bias to be amplified. This is what makes Bloomberg and Kelly’s benchmark of known suspects so troublesome.
What would serve as a better benchmark for measuring race neutrality of Stop & Frisk is a comparison to the distribution of crime by police precinct. In a perfect policing strategy, the distribution of Stop & Frisk occurrences by precinct would match the City’s distribution of crime by precinct. This benchmark naturally accounts for more policing in higher crime areas. By evaluating police precinct demographics along with absolute levels of crime, we can develop a better picture of potential bias by evaluating police activity relative to criminal activity.
This is the data framework I will use to highlight the bias that occurs in Stop & Frisk.
DIVERSE INSTITUTIONS CAN'T BE BIASED
Bloomberg and Kelly can’t comprehend that racial bias is even possible with a diverse police force. However, communities can experience disparate impacts of police activities without explicit racism. It would serve New York City well if Bloomberg and Kelly made a closer review of social science research on implicit bias.
Captain Tracey Gove of the West Hartford (CT) PD, writing in the magazine The Police Chief, had this to say regarding implicit bias:
The research and general findings suggest that implicit biases are held by all and, interestingly, race does not affect results. For example, of the 50,000 African Americans who have taken the Race IAT, about half of them had stronger associations with whites than with blacks. To some, this is not surprising; it has been argued that “we live in North America, where we are surrounded every day by cultural messages linking white with good.” According to Mahzarin Banaji, a psychology professor from Harvard and a leader of IAT research, “You don’t choose to make positive associations with the dominant group, but you are required to. All around you, that group is being paired with good things. You open the newspaper and you turn on the television, and you can’t escape it.”
In other words, the belief is that media bombardment in which a certain race is consistently linked with crime, deviance, and so on may form the basis for implicit biases. This conclusion is contrary to existing assumptions that discrimination and bias are intrinsic characteristics held only by ignorant, pernicious individuals. The research on implicit bias indicates that discrimination and bias are based more on those social issues and influences.
Implicit bias favoring whites by both white and minority officers is a real and legitimate phenomenon. They’ve been assimilated into an organization that continuously associates being black or Hispanic with criminality. When assessing perceived risk on the beat, their framework for assessment is guided first by their role as a cop. Department policy and strategy dictate their actions, and biases in strategy will manifest as biases in police outcomes.
In 'Stopping the Usual Suspects: Race and the Fourth Amendment', NYU Law Professor Anthony Thompson argues, among other things, that explicit bias isn't the only basis for racism when reviewing patterns of policing.
Although the Court appeared to assume in Terry and Whren that police officers can make assessments of criminality independent of whatever attitudes the officers may have about race, the social scientific research shows that the stereotypic judgments and biases that an individual brings to an event fundamentally shape perception. Research suggests that negative attitudes toward African Americans create a perceptual norm of viewing African Americans as more prone to criminal conduct. As a result of a phenomenon that social scientists call the "principle of least effort," individuals confuse category members with each other and remember more positive features about members of their own groups than those of other groups because such mental processes involve less mental energy than differentiating among group members.
Moreover, in our effort to predict and understand behavior, we often reduce our perceptions to culturally embedded stories about groups. One such story frequently applied to people of color is that they are more prone to engage in criminal and violent activity than whites. If one believes this as fact, then it is reasonable to assume that conduct engaged in by people of color will more likely be criminal or suspicious than the same actions by whites. The threshold for labeling conduct as "criminal" lowers when viewing conduct by people of color. A practical consequence of this behavioral principle is that when race is "a" factor in the description of individuals suspected of crimes, it may become "the" determining factor in the course of the investigation.
These distortions do not creep in only at the stage of perception. Cognitive psychologists suggest that the perceiver's biases also may distort the way that she samples, encodes, stores, and retrieves information. The processes of retrieving data and recalling information tend to bolster one's existing beliefs.
The effects of these phenomena are not limited to police officers whom one can easily characterize as "biased." Of course, some law enforcement officers consciously act on the basis of racial bias in denominating behavior as "suspicious." Such officers embrace stereotypes and allow personal biases to dictate their behavior. But "dominative racists" are not the only class of discriminators. Especially as it has become less socially acceptable to acknowledge racial prejudices and because people increasingly tend to view themselves as egalitarian, discriminatory treatment is often the product of unconscious racism.
Policing tactics don't just enforce law, but establish or reinforce community norms. Take frisking, for example. Frisks require a higher level of suspicion by an officer than a stop. Most of these stops are taking place on the street, so the officer's actions are public. To a rational witness, seeing someone frisked imparts a higher impression of a suspect's guilt than a mere questioning would. Over-policing people of color in this manner leaves higher impressions of minority guilt within the community. This is turn fuels increased suspicion of any person of color in the community. Community impressions of suspicion and guilt then reinforce those policing actions. Police action teaches the community how to perceive risk. Community perception of risk fuels policing actions. It is a feedback loop that will naturally amplify bias.
DATA HIGHLIGHTS BIAS
One way we can evaluate bias in policing activity is by analyzing the outcome of frisks. Here the question isn't whether this is a legitimate tactic in policing. Rather the question is in how it is applied. Who is considered suspicious? It's not enough to say that most perpetrators of crime are black or Hispanic. That assumes criminality is inherent in race or ethnicity. It ignores the context of the situation. Frisking is an act which requires a higher level of suspicion than a mere stop. This is an on-the-spot determination by the officer. If suspicion is race-neutral, then there should be correlation between the decision to frisk and a subsequent decision to arrest. As this chart shows, there is no correlation between frisk rates and arrest rates.
Frisks require a higher standard of suspicion and should, theoretically, predict higher arrest and summons rates. However, as practiced by the NYPD, frisks during Stop & Frisk aren't predictive of arrests or summons at all.
What we find when the data is disaggregated is rather stark. In 40 out of 75 police precincts (53%), a
white suspect is more likely to be arrested after a frisk than a non-white suspect. Yet non-white suspects are frisked at a higher rate in 71 of the 75 precincts (95%). If frisks are true measures of heightened suspicion, then there is an obvious bias against non-white suspects.
Another way to evaluate this is to look at people in similarly situated circumstances. One such group involves people living in precincts with a comparable number of crimes. Rather than focusing on the race of a suspect, I looked at the racial and ethnic composition of police precincts. If the frisking data does indicate a bias against minorities in assessing suspicion, then evaluating neighborhood demographics can highlight biases when the pool of potential suspects looks different even when the absolute level of crime is the same.
To begin, I rank ordered the 75 precincts by their total number of crimes and then grouped them into quintiles of 15. The 15 precincts with the highest number of crimes were grouped together, then the next 15, and so on. Within each quintile, I subdivided the precincts into groups with white populations above the median (30.3%) and below the median. This allows browner precincts to be compared to whiter precincts with comparable numbers of crimes.
In a perfect policing strategy, the distribution of stops throughout the city would match the distribution of crimes throughout the city. By comparing each quintile's distribution of stops and distribution of crimes, we can determine where over-policing (higher distribution of stops than crimes) and under-policing (higher distribution of crimes than stops) occurs.
The chart below measures the demographic composition of the police precincts on the x-axis and the level of policing relative to the level of crime on the y-axis. Precincts with white residents comprising more than the median level of 30.3% of the population are to the right of the y-axis. Precincts with a white residential population below 30.3% are to the left of the x-axis. Over-policed precincts are above the x-axis, while under-policed precincts are below it.
Intra-quartile comparisons show precints of color are consistently over-policed relative to the precinct level of crime by an average of 20%. Conversely, whiter precincts are under-policed by an average of 25%. In the 2 lowest crime level quintiles (5-Blue, 4-Green), precincts of color have the highest rate of over-policing at 42% while the comparable whiter precincts in quintiles 5 (Blue) and 4 (Green) are under-policed by 5% and 38% respectively.
We see a strong correlation between the racial and ethnic composition of a police precinct and the level of policing that occurs even when the absolute levels of crime are taken into account. Whiter precincts are consistently under-policed compared to their crime levels, while browner precincts are consistently over-policed. Frisking rates remain 25% higher in browner precincts regardless of crime level.
IMPLICATIONS AND NEXT STEPS
All of this suggests Stop & Frisk is practiced in a racially biased manner. It establishes a threshold of suspicion for minority residents that is significantly lower than the threshold for white ones. The frisking data shows minorities are consistently viewed more suspiciously by the NYPD. The over-policing data shows how that suspicion bias is amplified in the decision to stop someone in the first place. Of significant concern is how this in manifested in the lowest crime areas. In practice, Stop & Frisk equates the mere presence of African Americans or Hispanics in low crimes areas as suspicious.
The practical implication is that a community of color is over-policed regardless of the community's actual crime level. And the truism will hold; the more the NYPD looks for crime, the more crime the NYPD will find. Criminal possession of a weapon is the suspected crime most often cited on the UF-250 Stop & Frisk reports. While over 98% of frisks fail to find a weapon, these frisks make it significantly easier to bust someone on a secondary offense like possession of marijuana. So while studies have shown marijuana use at similar levels for both whites and people of color, over-policing of minorities creates a criminal justice system filled with a disproportionate number of minorities incarcerated for possession.
Any system or policy, not just policing, creates areas of bias. Some are by design, but many are not. Recognizing that bias will exist opens up an organization's ability to actively look for it, measure it, and address it.
An efficiency metric like stops per arrest could help the NYPD look for areas of bias. The chart below looks at the 13 most suspected crimes cited on the Stop & Frisk report, accounting for 96% of the stops.
Analyzing disparities in efficiency metrics like stops per arrest could help the NYPD reduce stops by 14%. Understanding the disparity can help the NYPD identify areas of potential bias.
Criminal possession of a weapon was the most cited crime, listed on 24% of reports. There were over 12 stops per arrest for white suspects as opposed to 22 stops per arrest for minority suspects. This large gap in policing efficiency ought to be further analyzed. Why are officers wrong so much more often with minority suspects than with white suspects? Field observation and officer interviews could help the NYPD close that gap. Just eliminating the efficiency gap in stops for criminal possession of a weapon means overall stops could be reduced by 54,000. Seven of the 13 suspected crimes listed above have efficiency differences greater than 2. If the NYPD were able to close those gaps, they could reduce stops by 14%. Interestingly, 3 of those gaps actually adversely impact white suspects.
Finally, the NYPD should evaluate the variability is stops per arrest for given suspected crimes. Why does the NYPD need 35 stops per arrest for robbery but only 8 for assault? There may be some legitimate reason for different rates for different crimes, but an absolute difference of 27 suggests ample room for improvement.
Crime in America is framed by race and ethnicity. Bans on racial profiling are positive signals to citizens that unequal treatment shouldn't be tolerated. But these bans aren't a magic wand that makes biased policing strategies disappear overnight. A modern day Bull Connor does not need to lead a department for it to practice biased policing strategies. Despite Mayor Bloomberg's protestations, Stop & Frisk is indirect racial profiling. Rather than doubling down and alienating the people they are professing to protect, Bloomberg and Kelly need to listen to their community and our courts. A policing strategy that annually stops more people than the total population of Minneapolis AND St. Paul hurts New York. It breeds resentment and drives a wedge between the NYPD and the people they are supposed to protect and serve.