How the class action lawsuits top expert lies with statistics
If U.S. District Judge Shira Scheindlin rules against the New York Police Department in the racial profiling trial that concluded last week, she will rely heavily on the arcane statistical models of Columbia law professor Jeffrey Fagan, the anti-cop advocates favorite expert witness.
Let us take a careful look at Fagans role in the trial, the actual numbers behind his scholarship and the methods he used to reach his conclusions.
The plaintiffs in Floyd vs. New York charge that the NYPDs practice of stopping, questioning and sometimes frisking suspicious individuals is driven by race, not crime. Scheindlin cannot reach that verdict based on the cases 12 individual complainants alone.
Even if she were to conclude that the named plaintiffs were all stopped because of their race — a finding that would require ignoring the considerable evidence supporting their stops — that judgment still would leave her far from the requisite inference that their police encounters were emblematic of the 4.4 million stops that the NYPD has conducted since 2004.
Thats where Fagan comes in. The academic was tasked with showing en masse that those 4.4 million stops were made not because the officers suspected that the individuals stopped were engaged in criminal activity, but because most of those individuals were black and Hispanic.
The models he constructed to prove such bias were an apt symbol of the lawsuit itself: wholly detached from the realities of crime and policing in New York.
The Center for Constitutional Rights and the elite law firm of Covington & Burling, the attorneys in Floyd, faced an inconvenient truth: The stop rate for blacks is actually lower than their violent crime rate would predict.
Blacks, who constitute 23% of the citys population, committed 66% of all violent crimes and 73% of all shootings in 2011, according to victims and witnesses, but they were only 53% of all stop subjects.
Whites, who constitute 35% of the citys population, made up 9% of all stops in 2011, though they committed only 5.5% of all violent crimes and 2.5% of all shootings.
Fagan, however, needed to show that race, not crime, predicts police activity. Given the facts arrayed against such a proposition, it was no surprise that for days Scheindlins courtroom was filled with debates over “exponentiated coefficients,” “P values,” “Z scores,” “zero-inflated models” and “Vuong tests,” as Fagan tried to explain and defend his computer formulas.
The judge candidly admitted to being lost at times, as anyone would be who was not steeped in advanced statistics.
But you dont have to have a Ph.D. in econometrics to spot the flaws in Fagans techniques. For starters, he fed into his statistical black box all of the 4.4 million stops from the last eight years, even though his own extremely superficial analysis of the forms that officers fill out after a stop had concluded that nearly nine-tenths of those stops appeared to be lawful, and only 6% unlawful.
Those lawful stops should have been excluded from his regression analysis, since they cannot form the basis for concluding that the officers making the stop substituted race for reasonable suspicion.
The errors in Fagans models get worse from there. He refused to consider the race of New Yorks criminal suspects in evaluating whether the police were making stops based on skin color rather than behavior.
Such information, however, is an essential component of any racial profiling analysis. Males are 91% of all stop subjects, but no one accuses the police of sex discrimination, because it is acceptable to acknowledge that males commit the lions share of crime.
Fagan justified his refusal to take criminal suspect data into account on the ground that the data are incomplete, but the race of suspects is known for 98% of the citys shootings, 98% of drug crimes, and 85% of all violent crimes — precisely the offenses that overwhelmingly drive the NYPDs deployment decisions. Only in property crimes is the race of a majority of suspects unknown, but there is no reason to think that the racial makeup of unknown property offenders differs from the makeup of known property offenders.
Fagans treatment of the relationship between crime and police response was even farther from reality. His model used crime data from the previous calendar month to predict stops in the current month: if a stop was made on May 31, for example, he assumed that April crime data — not Mays — should explain it.
Such a huge time lag ignores the essence of the NYPDs data-driven policing revolution, in which the most up-to-the-minute crime information determines tactics. Officers on one tour will be working off information gathered on the previous tour, not just off of crime data from six weeks ago. A gang shooting will immediately trigger a local influx of officers, who will make an elevated number of stops to try to apprehend the shooter and avert a retaliatory shooting.
In Fagans model, however, if there were no shootings in the previous month, the spike in stops from this months shooting will appear unmotivated by crime and thus likely to be a function of race.
Fagan made no distinction between domestic and gang homicides, even though the former, unlike the latter, have no impact on street patrols. His model does not properly represent Commissioner Ray Kellys biggest policing innovation: so-called Impact Zones, where a high concentration of rookie officers walk the beat in the citys most dangerous neighborhoods and make stops if they witness suspicious behavior. Impact Zones are located virtually exclusively in minority neighborhoods; the higher number of officers available to observe criminal activity will result in more stops. Fagans model, however, will see only the stops without accounting for the strategy behind them.
With these and other important details of the NYPDs operations stripped away, resulting in a set of mathematical equations that are blind to the ways that gang culture influences communities and law enforcement, Fagans “negative binomial regression analysis” purported to show that the number of stops in a neighborhood is a function of the race of the residents, not of local crime conditions.
No wonder when, at the end of the trial, he finally estimated the exact number of stops that would result from an increase in an areas black population, he reached a result wildly out of sync with the real world. His model predicted that a census tract with an 85% or higher black population would experience 120 stops a month; the actual average in such tracts is 19 stops. The plaintiffs offered no examples of tracts with 120 stops.
The citys attorneys and expert witnesses did a superb job of discrediting Fagans methods, but Scheindlin has previously expressed support for his findings and it is unlikely that she will reject them. Like the rest of the trial, the time and energy spent on Fagans analyses were a sad diversion from the real problem facing the citys black and Latino residents: the persistence of senseless victimization, despite New Yorks record-breaking crime drop.
During closing arguments last Monday, no one referred to the fatal shooting just two days earlier of an innocent 14-year-old girl returning from a friends birthday party in Queens. A gunman had sprayed the bus on which the girl was riding with bullets; one of the shots entered her right temple. The jud-ge did her best to keep such incidents far from the courtroom, but they occasionally seeped in anyway, when officers described the mob violence or the reports of a gunman at large that had preceded a stop.
The NYPDs response to last weekends bus murder in South Jamaica will generate more stop data that can be used against the department in the next racial profiling lawsuit, but it is precisely such high-intensity policing that has brought the citys murder rate down nearly 80% since 1993. The vast majority of lives saved thanks to that falling homicide rate have been black and Hispanic.
The last two decades of racial profiling discourse have been a massive exercise in changing the subject: It is a lot more comfortable to talk about the allegedly racist police than about black-on-black crime. Until society is willing to address the family breakdown that generates such mindless violence as last weekends bus shooting, however, it will continue to fall to the NYPD to provide the social control in neighborhoods where fathers no longer do so. If the Floyd plaintiffs are successful in their suit, they will strip the citys most vulnerable residents of their essential lifeline from fear and predation.
Thats a wrenching reality ignored by the abstruse statistical models of the plaintiffs star number-cruncher.
Original Source: http://www.nydailynews.com/opinion/great-stop-and-frisk-fraud-article-1.1354173