Sure, it's easy to understand why we see it in nationally representative samples, but why do we also see it also in highly selected samples, such as folks who showed up to take the New York firefighter's hiring test? (1999 gap: 1.04 s.d. 2002 gap: 0.96 s.d.)
In contrast, consider average height. Chinese people on average are less tall than black and white Americans. Yet, the average height of Chinese NBA players (such as Yao Ming) has usually been well above the NBA average. Short Chinese guys just don't play in the NBA.
So, why don't we see this kind of non-representativeness of the sample among FDNY test-takers? Just as 6'3" Chinese forwards generally decide to continue playing in China rather than try to make the Lakers, would-be minority test-takers could have estimated their scores from practice exams and then decided not to bother to show up and waste their time taking the test if they were likely to only end up way down the hiring list. This kind of self-selecting behavior would reduce the racial gap.
And yet, we saw the usual one-standard deviation gap. Why?
Well, one reason is that affirmative action promotes the more competent sort of minorities into higher realms, the equivalent of Yao Ming skipping playing center in the NBA to play keeper in the Intergalactic Quidditch League, leaving only short Chinese guys to try to make the NBA.
Another reason, however, is that, for a black or Hispanic, taking an FDNY test is like buying a very, very long-lived lottery ticket.
If the damages in Vulcan Society are set at, say, $20 million, the contingency fee lawyers will presumably grab about $7 million, and several hundred or more black and Hispanic test-takers who came close enough on the test so that they would have been hired if there had been no disparate impact will get checks in the mail adding up to $13 million.
Wouldn't it be totally awesome to get a five-figure check in the mail for something you wasted time on and failed at a decade ago? So, you can see why so many minorities who didn't have a chance of getting a good score took the firefighters test — because there was always a sizable chance under Disparate Impact theory that a judge would change the rules years after the game was played and send them money.
In contrast, dumb white guys wouldn't be as likely to bother showing up because they know nobody is going to change the rules in their favor.
Thus, the Fundamental Constant of Sociology endures.