Social Science Statistics Blog

« Mother Nature Estimates Using...? | Main | Stats of the Union »

24 January 2007

The Goal of Causal Inference

Jim Greiner

I’ll be giving the talk at the Gov 3009 seminar in early February, and I’ll be presenting a paper I’m writing with Don Rubin on applying the potential outcomes framework of causation to what lawyers call “immutable characteristics” (race, gender, and national origin, for example). I’ll be previewing some of the idea from this paper on the blog.

One key point from this paper is the recognition that in law (specifically, in an anti-discrimination setting), the goal of causal inference may be different from that in a more traditional social science setting. A sociologist, for example, might study the effect of tax breaks for married couples on marriage rates; the obvious goal of the study is to see whether a contemplated intervention (tax breaks) has a desired effect. An economist might evaluate a job training program for a similar reason. In anti-discrimination law, however, we study the effect of units’ perceived races (or genders or whatever) on some outcome (e.g., hiring or promotion), but we have no interest in intervening to change these perceptions. Rather, we’re contemplating action that would mitigate the effects we find. The “intervention” we’re considering might be compensating the victim of discrimination, as is true in an employment discrimination suit. Or it might be ceasing a certain type of government action, such as the death penalty. But we’re not interesting in implementing a policy promoting or effectuating the treatment that we’re studying.

Posted by James Greiner at January 24, 2007 1:14 PM

Comments

Notification

Post a comment