Read on Twitter
1.Another tweetstorm! Economics is engaged in needed self-examination on race. I want to address one thread of this, germane to the policing discussion: the obsession with distinguishing “statistical discrimination” from “taste-based” discrimination or “prejudice.”
2. As an outsider (a law prof) who publishes in econ journals and reads a lot of econ articles, I was really taken aback when I first encountered this (and still am). In my field, criminal justice, it’s particularly jarring; it is a straight-up defense of racial profiling.
3. In its traditional form, at least, the idea is that racial discrimination is A-OK, or at least less troubling, so long as cops (or employers, etc) are using race “rationally” as a proxy for some legitimate concern, like employee productivity or crime risk.
http://4.So , let’s say a cop says “Sure, I use race to decide who to stop and search. I stop black people more often because they are black. But it’s not because I don’t LIKE black people. It’s just that I know are more likely to be criminals.”
5.The layperson hears this and thinks: “What a racist POS.” The lawyer hears this and thinks “It’s surprising that this racist cop just openly admitted to unconstitutional racial discrimination. This will be an unusually easy lawsuit!”
6.But some economists apparently hear this and think “Oh, well, as long as it’s not because he doesn’t LIKE black people. If there are stats showing racial differences in crime rates, it’s perfectly rational for him to use race to decide who to stop!”
7.Seriously: some of the most influential and cited economics papers on criminal justice, from THIS CENTURY, use this approach. And countless mediocre ones too. I review papers and attend paper presentations with these approaches ALL THE TIME.
8.Papers on policing, on sentencing, on parole decisions and bail decisions…MANY start from the premise that statistical discrimination is OK, or at least somehow less objectionable. The entire literature on “hit rate” or “outcome tests” turns on this assumption.
9.Hit-rate or outcome-test studies dominate econ studies of crim justice disparities. Studies finding equal hit rates for searches by race have been cited by police depts as evidence they don’t racially discriminate—unsurprising given the studies’ framing.
10. Yet their findings of “no bias” rest on a model that assumes that "unbiased" police DO racially profile (exactly as much as needed to maximize efficiency of searches).
11. In the recent lit on algorithmic prediction in criminal justice, a popular approach (from computer scientists too, not just economists) is to use equal predictive accuracy as the objective in racial disparity analyses. This approach has essentially the same premise.
12.Admission: Although criticizing statistical discrimination has been a theme of my work (esp legal work re: algorithmic prediction), some of my own econ papers nonetheless discuss why what we found seems to be stereotyping or taste-based discrim.
13.This is such a core concern of the field that it has seemed unignorable. It's one of the first questions at every seminar. We do emphasize that what economists label statistical discrimination would be equally bad and illegal, & say we’re just exploring mechanisms.
14.And there IS some merit to trying to break down how discrimination works (like, e.g., psychologists do w implicit bias studies). Sometimes it can help in assessing strategies to reduce it, or to see how well-intentioned laws can backfire.
15.For example, statistical discrim theory implies that decision-makers will rely on race more when they have less other (non-race) info available. (Stereotyping theory actually has the same implication, btw.) That insight has empirical support and is often policy-relevant.
16. Recognizing that doesn’t require DEFENDING statistical discrimination or seeing it as non-racist. It is racism, and it is racial discrimination, and it's gross, and it's illegal (for police, employers, and many other decision-makers).
17.But I worry that “understanding mechanisms” isn’t the only reason so many papers contain such a discussion. It's also because some (not all) economists won’t think discrimination findings reveal a *problem* unless there’s evidence it’s not “just” statistical discrimination.
18.These mechanisms don't differ in any morally meaningful way. Focusing on disentangling them risks implying to readers that they do. At least, we must counter that view *emphatically*--not just little caveats. I know I can do better on this (& am revising a paper to do so rn).
19.And in contexts where it *is* relevant/useful to parse mechanisms, it would help to come up with a better name for “statistical discrimination.” Maybe “racial profiling”? (Familiar for police, but works for other contexts too.)
http://20.As a law prof, I must note: the law is *also* far too restrictive in how it defines discrimination. The law is completely not up to the task of grappling with systemic racism. In constitutional cases, it fails even to incorporate disparate impact tests.
21.And the legal profession and legal academia also have a LOT of self-examination to do.
22.But given that fact: Economists, if your definition of racial discrimination is so narrow that it is even shocking to lawyers, it’s time to rethink it.
23.While composing this thread, I looked for other recent Twitter discussion of this issue. I was pleased to find some, e.g., from @Rodprime, @mioana, @AntoniaDiazRod, @mattyglesias. Didn't see this from criminal justice economists, but I've probably missed some (quick search).
24.But the best thing I found was a powerful passage in this open letter from Prof. William Spriggs ( @WSpriggs). Hopefully everyone in econ has already read it, but I hadn’t. https://www.minneapolisfed.org/~/media/assets/people/william-spriggs/spriggs-letter_0609_b.pdf?la=en.
25.Maybe I should have just amplified Prof. Spriggs’ words. In any case, I’ll close with them:
