Saturday, November 5, 2011

Selection Bias Alert

The New York Times reports on the issue of why we do not have more STEM graduates despite so much effort to increase in science. The article includes this paragraph:
“You’d like to think that since these institutions are getting the best students, the students who go there would have the best chances to succeed,” he says. “But if you take two students who have the same high school grade-point average and SAT scores, and you put one in a highly selective school like Berkeley and the other in a school with lower average scores like Cal State, that Berkeley student is at least 13 percent less likely than the one at Cal State to finish a STEM degree.”
My guess is that this conclusion comes from a regression where they controlled for SAT and GPA, but there are a lot of unobserved variables that could drive this correlation. I'm not a "selection bias hawk" who thinks that any non-experimental evidence is worthless, but in some cases there is good reason to think that a regression coefficient is biased in a direction that would invalidate the conclusion and I think this is one of those cases.

No comments:

Post a Comment