Things the President Doesn't Know About Racial Disparities
On July 7, 2016, in a speech on police shootings of black men by police in Louisiana and Minnesota, President Obama discussed the contribution of racial disparities in the criminal justice system to African American mistrust of law enforcement. Statistics the President cited included that African Americans are 30 percent more likely to be pulled over than whites and African Americans and Hispanics who are pulled over are three times as likely to be searched as whites; last year African Americans were twice as likely to be shot by police or arrested as whites; African Americans who are arrested are 75 percent more likely than whites to be charged with sentences carrying mandatory minimums; while African Americans and Hispanics make up 30 percent of the general population, they make up over half of the incarcerated population.
The President observed, however, that modifications of practices can address such disparities and discussed an expert task force he had convened that had carefully reviewed the data and provided a range of recommendations to improve policing. He also approvingly cited sentencing reforms being considered in Congress.
The President did not say exactly how modifications to practices might reduce the racial differences he cited. Presumably he did not know the reasons why such modifications, as well as Department of Justice (DOJ) initiatives such as those regarding Ferguson, Missouri, may well increase those differences.
In May 6, 2016 Federalist Society Blog post titled “Is the Disparate Impact Doctrine Unconstitutionally Vague?” (PDF available here), I explained that certain civil rights law enforcement policiesare based on government beliefs about effects of particular practices on statistical disparities that are the exact opposite of reality. For example, the government encourages lenders and public schools to relax standards in order to reduce the severalfold racial differences in adverse borrower outcomes like rejection of a mortgage application and adverse school discipline outcomes like suspension and expulsion. I explained that relaxing standards or otherwise reducing the frequency of an adverse outcome, while reducing percentage differences in rates of experiencing the corresponding favorable outcome, tends to increase percentage differences in the adverse outcome itself.
To illustrate the point, I posited a situation where pass rates on a test are 80 percent for whites and 63 percent for a minority group and thus the pass rate is 27 percent higher for whites than minorities. If the cutoff is lowered to the point where 95 percent of whites pass, assuming normal test score distributions, the minority pass rate would be about 87 percent. With the lower cutoff, the pass rate would be only 9 percent higher for whites than minorities.
That lowering test cutoffs tends to reduce percentage differences in pass rates is well known and underlies the common belief that lowering cutoffs tends to reduce the disparate impact of tests on which some groups outperform others. But there is also a part of the picture virtually no one understands.
While lowering the cutoff reduced the percentage difference in pass rates, it increased the percentage difference in failure rates. The minority failure rate was initially 1.85 times (85 percent greater than) the white rate (37 percent/20 percent). With the lower cutoff, the minority failure rate would be 2.6 times (160 percent greater than) the white rate (13 percent/5 percent).
Anyone having difficulty grasping the example should look at Table 1 (at 9) of my amicus curiae brief in Texas Department of Housing and Community Development, et al. v. The Inclusive Communities Project, Inc.
The pattern whereby reductions in the frequency of an outcome tend to increase percentage differences in rates of experiencing the outcome, while reducing percentage differences in rates of experiencing the opposite outcome, is an inherent feature of normal risk distributions. It can be found in almost data showing rates at which groups fall above or below various points on a continuum of a quantifiable factors as well as in the myriad situations where, in point of fact, reductions in the frequency of an outcome are accompanied by increasing percentage differences in rates of experiencing the outcome and decreasing percentage differences in rates of avoiding the outcome. Extensive examples and explanations of the pertinent patterns, as well as the implications of the failure to understand them in the law and social and medical sciences, may be found in my “The Mismeasure of Health Disparities,” Journal of Public Health Management and Practice (July/Aug. 2016) and “Race and Mortality Revisited,” Society (July/Aug. 2016). See also my October 8, 2015 letter to American Statistical Association urging it to explain the matter to arms of the federal government.
But government agencies enforcing laws against discrimination in lending or school discipline remain utterly unaware that relaxing standards tends to increase percentage differences in adverse outcomes. Indeed, they repeatedly show that they believe that relaxing standards will have the opposite effect. And they consistently monitor the fairness of lending and discipline practices on the basis of percentage differences in adverse outcomes. Thus, for some time there have existed perverse situations where complying with government encouragements to relax standards increases the chances that the government will sue one for discrimination.
Sometimes, racial disparities are appraised in terms of comparisons between the proportion a group makes up of the population and the proportion it makes up of persons experiencing some outcome, as in the President’s statements regarding the minority proportion of persons incarcerated. This was the approach of the report on police and court practices of the City of Ferguson, Missouri that DOJ issued in March 2015 and the suit it brought against Ferguson in February 2016. DOJ attributed the fact that African Americans made up a much higher proportion of persons experiencing adverse interactions with the police and courts than that 67 percent they made up of Ferguson’s population to what the agency regarded as over policing and unduly harsh court procedures.
But, as I explained in the May 6 post, a corollary to the pattern whereby reducing the frequency of an outcome tends to increase percentage differences in rates of experiencing it is a pattern whereby reducing the frequency of an outcome tends to increase the proportions groups more susceptible to the outcome make up of persons experiencing it. Thus, the type of relief sought by DOJ in the suit against Ferguson – and which it subsequently secured in a consent decree entered in April 2016 – will tend to increase, not decrease, the proportion African Americans make up of persons experiencing adverse interactions with Ferguson’s police and courts.
Reforms of police practices recommended by the President’s task force are of the type that, by adding circumspection and discouraging unnecessary use of force, tend generally to reduce arrests and the use of force. Thus, they will tend to increase percentage racial/ethnic differences in arrests and being subject to the use of force and increase the proportions racial minorities make up of persons experiencing those outcomes.
This does not have to happen. If racial bias plays an important role in racial difference in criminal justice outcomes and reforms reduce racial bias, all measures of racial disparity should decrease. But whether any such effects will be sufficient to counter the effect of generally reducing adverse outcomes on standards measures of disparity remains to be seen. And it is by no means improbable that percentage differences in adverse criminal justice outcomes (and the proportion African Americans make up of persons experiencing those outcomes) will increase even if in any racial bias has been reduced. Observers who attribute racial differences to bias will nevertheless assume that bias has increased.
The same points hold for sentencing reforms being considered in Congress that are aimed at generally reducing the prison population. The ways reforms pertain to different types of crimes may play into the matter in various ways. But there remain statistical reasons to believe that generally reducing the size of the prison population will increase the proportion African Americans and Hispanics make up of that population.
Notably, quite consistently across the country the jurisdictions that have substantially reduced public school discipline rates while believing that doing so will reduce percentage racial/ethnic differences in discipline rates have found those differences to increase. And this has occurred notwithstanding that such jurisdictions are presumably doing everything else they can think of to reduce racial/ethnic differences in discipline rates.
More directly pertinent to criminal justice issues, a June 2016 report of the Sentencing Project found that substantial reductions in youth incarceration rates between 2003 and 2013 were accompanied by increased racial percentage differences in youth incarceration rates. That the report found the increase in disparities to occur “despite” the reductions reflects how poorly the matter is understood by an organization of putative expertise in the subject.
Meanwhile, New Jersey is considering a bill requiring that all criminal justice legislation include a racial impact statement. It is a fair assumption that in the New Jersey legislature – as in the U.S. Congress – not a single member understands that legislation that decreases adverse criminal justice outcomes tends to increase percentage racial/ethnic differences in experiencing the outcomes (and the proportion racial minorities make up of persons experiencing those outcomes) or that legislation that increases adverse criminal justice outcomes tends to have the opposite effect.
In the July 7 speech, the President emphasized that the statistics he cited were facts. They may well be facts. But they are facts that neither the President, nor virtually anyone else analyzing or opining about racial/ethnic differences in criminal justice outcomes, understands how to interpret.
A fuller treatment of this matter may be found in my July 25, 2016 letter to the American Statistical Association urging it to explain to the President that his beliefs about the likely effects of modifications of practices on the racial disparities he cited are mistaken.
* * * * *
James Scanlan is an attorney in Washington, D.C.