Do Statistics Really Lie About Discrimination?
Ezra Klein, the liberal journalist, is suspicious of statistical controls -- too suspicious. He thinks they obscure the magnitude of certain social problems.
But statistical controls remain one of the best methods we have to illuminate the causes of social phenomena and possible responses to them. We should learn from the studies that use them, not discount them -- even when they disturb our ideological preconceptions.
Researchers use statistical controls to isolate the effects of variables they're examining. Consider a study on the best ways to quit smoking. If the researchers just collected data on how much people who chew nicotine gum smoke, and then compared it to how much non-chewers do, they might well find that nicotine gum is associated with higher levels of smoking. If they control for how much people smoked when they started chewing the gum, on the other hand, which has the effect of comparing smokers who start using the gum against those who don't, they might find a different and more instructive result.
Statistical controls are one reason that policy research is often confusing and difficult to read. But they’re necessary if we want to disentangle causes and effects. Klein understands that point in general. He cites two areas, however, in which he thinks studies that control for too many variables are misleading.
First, if we're trying to determine whether the criminal-justice system has a racial bias, he writes, then controlling for the income of people who get arrested might be a mistake. Blacks have lower incomes on average than whites. But if black people face disproportionate police harassment because many of them are poor, he argues, that doesn't mean the system isn't biased against blacks. It may mean that racism manifests itself in our society first by causing blacks to be disproportionately poor, and then by causing police to treat poor people badly.
Referring to a post by his Vox.com colleague Matthew Yglesias, Klein makes the same point about studies of the wage gap between genders. The gap looks a lot smaller when researchers control for the number of hours an employee works (and even smaller when they control for other variables). But that's not evidence, say the Voxers, that the wage gap isn't real. It's just an explanation of how the wage gap works. "Women work shorter hours because as a society we hold women to a higher standard of housekeeping, and because they tend to be assigned the bulk of childcare responsibilities," writes Yglesias.
It seems simplistic to say that women are "assigned" childcare responsibilities: Maybe women disproportionately prefer spending time with their children rather than with cubicle-mates. Maybe they also have different preferences from men on average when it comes to choice of occupation -- preferences that aren't merely the result of programming by a sexist society. But even if the Vox explanation was the full story, it would still be worth knowing that a lot of the discrimination that produces the gender wage gap is the fault of husbands and society at large, and not employers.
What both Yglesias and Klein ignore is that typically the raw wage gap between genders is used to make a case for cracking down on employment discrimination. That's the way President Barack Obama uses it, for example, when he's pushing intrusive legislation and executive orders on private companies intended to narrow the gap. That's the way feminist groups use it, too. And that's the context in which conservatives bring up the wage-gap studies that so aggravate Vox, the ones showing other variables may be at play. If the real answer is that men need to do more housework and child care, that strengthens the case that employers shouldn't be held accountable for the problem.
It's also helpful to know what might contribute to racial disparities in the criminal-justice system. For decades, the law has treated offenses involving crack cocaine much more harshly than those involving powder cocaine. Maybe racism explains that: The drug black people tended to use drew more punishment than the one whites preferred. But it's useful to know the mechanism -- because it suggests that sensitivity training for police officers won't solve the problem if we don't also change the sentencing laws.
That's how social science, and social progress, sometimes works.
This column does not necessarily reflect the opinion of Bloomberg View's editorial board or Bloomberg LP, its owners and investors.
To contact the author on this story:
Ramesh Ponnuru at firstname.lastname@example.org
To contact the editor on this story:
Timothy Lavin at email@example.com