Greetings from Tonopah, Nevada. I am meandering west at the moment en route to join a group hiking in the High Sierras. Soon I shall be lugging my camera and tripod at high altitude.
To get in the mood, my bedtime reading last night was the Supreme Court ruling on the New Haven firemen's case. I wrote a column about the case before the court ruled. A few things about the ruling--and about the case, now the ruling has taught me more about it--surprised me.
The court decided 5-4 (again) for the firemen and against the city, overturning Sonia Sotomayor. As Stuart Taylor says at National Journal, all nine disagreed with Sotomayor's summary judgment upholding the city's decision to void the promotions test. Even the minority wanted the lower court to take another look. But it seems I was mistaken when I said there was no dispute over efforts to ensure the test was fair. This is very much what the minority disputes. In their view, apparently, "disparate impact" remains strong prima facie evidence of "disparate treatment", or discrimination.
Ruth Bader Ginsburg, writing for the minority, seems to find it hard to believe that a fair test could have produced such a lopsided outcome. At first sight, that may seem reasonable enough--except for the evidence, related at length, of the efforts made to assure fairness. I think Ginsburg is grasping at straws somewhat when she talks, for instance, of possibly unequal access to test-prep materials. They are all firemen, after all.
If, without strong proof to the contrary, disparate impact reliably signals discrimination, race-based affirmative action will be with us for an awfully long time. In effect it would remain as a remedy not for racial discrimination in the workplace, but for unequal opportunity due to family circumstances and a failing education system. I cannot see that race-based affirmative action is the right response to those issues. Still less can I see it as the just response.
Happy holiday, and with luck I'll see you in the week begining July 13th.
It's worth repeating the statistics that so concerned Justice Ginsburg: On the lieutenant's exam, blacks and Hispanics accounted for 34 of the 77 applicants, yet the highest any one of them finished was 13th. 25 of 43 whites passed, 6 of 19 blacks passed and 3 of 15 Hispanics passed. On the captain's exam, there were 41 applicants, 25 white, 8 black and 8 Hispanic, with 16 whites passing, 3 blacks and 3 Hispanics. No blacks finished in the top 9.
As a matter of statistics, these results differ hugely from what you would expect if the scores were distributed randomly, particularly on the lieutenant's exam. In a random distribution, you'd expect 5 or 6 of the top 12 in the lieutenant's exam to be black or Hispanic, not zero. In fact, the top 12 scores on that exam would come from white applicants less than once out of every thousand trials.
The captain's exam isn't nearly as bad in statistical terms - there's about a 1 in 7 chance that in a random distribution you'd end up with no blacks in the top 9. But the captain's exam, like the lieutenant's exam, also has much higher pass rates for whites than the minority candidates (58 percent versus 26 percent for the lieutenant's exam and 64 percent versus 37.5 percent for the captain's exam).
This is a very strong statistical case for the proposition that the differences in the exam results were not random. If I told you in another context that the odds against something happening at random would be 1,000 to 1, it's likely that you'd conclude that the best explanation was that there was something non-random at work. It's no wonder that Justice Ginsburg thought something was wrong, and that's the whole point of the disparate impact test: When you see a result that you shouldn't see, you need to go back and see if there's a neutral explanation.
Keep in mind, too, that there were a lot of factors that tended to level the playing field here, notably the requirement for a certain amount of experience and the amount of notice available to potential applicants. If the test were neutral, these factors would be likely to minimize differences caused by previous education or family circumstances. (After all, a firefighter with a certain amount of experience already has succeeded in getting a job as a firefighter, a pretty significant hurdle to pass.) When you start eliminating other factors that might have given the white firefighters an edge, then the case that the disparate impact is related to something discriminatory in the test becomes stronger.
What the majority does, in essence, is ignore the statistical evidence because it is inconvenient. That is neither good law or good statistics.
Re: "This is a very strong statistical case for the proposition that the differences in the exam results were not random."
Not random, doesn't imply discrimination. Different results, or "disparate impact" is a lousy way of measuring discrimination. There would be a lot of different factors, some of which would be unknown even after the closest examination of the test and the candidates. Clearly "not random" results, and different pass rates, should at most be considered reason to suspect racial bias, not solid evidence for it, and certainly not enough evidence by itself to support a determination that there is racial bias.
randy_kahn misunderstands the issue. Ricci and the Supreme Court majority didn't take the position that the racial distribution of the test results was purely random. Rather, they assumed (as both sides in this case, and almost everyone, liberal and conservative, assumes) that there are real group differences. The question is how we deal with that fact.
A trivial example might be familiarity of Country & Western lyrics. If you tested a random sample of white, black and Hispanic Americans on their knowledge or C&W lyrics, the whites, as a group, would probably score higher than blacks and Hispanics as a group. If you decided which firefighters to promote based on their knowledge of C&W, your promotion decisions would be skewed toward whites. If you decided based on knowledge of rap, your promotion decisions would be skewed toward blacks. Music lyrics tests to promote firefighters would not be race neutral. The problem would not be that they would mis-measure familiarity with music lyrics. Rather, the problem would be that competence as a firefighter is not obviously, provably or even plausibly correlated with knowledge of music lyrics.
If there were no testable racial group differences, it would be impossible to construct a racially biased test. Indeed, if there were no objectively identifiable racial group differences, it would be meaningless to say that blacks as a group are disadvantaged. Obviously, there *are* racial group differences - randy_kahn identifies "previous education or family circumstances."
No one on the Supreme Court (or any other judge, or the litigants) "ignore[d] the statistical evidence." Rather, randy_kahn has convinced himself that math somehow solves a problem that is in fact a value choice. If only it were that easy.