Brendan NyhanVerified account

Among NBA players, the correlation between height and salary (a proxy for being good at basketball) is negligible. Doesn't mean NBA is wrong for favoring tall people http://rpubs.com/msluggett/189114 …

Grad students are hardly a sample so select as to be equivalent to NBA players. GRE is a lot of time and money for students- do we have any evidence that it *does* do something over and above GPA?

don't know. you would need a dataset using it to predict grad school success on a sample that doesn't condition on admission and matriculation

And until then... assume the null, right?

No. We don’t need that. There are well-understood corrections for selection effects. In I-O this is called range restriction. It is easy to correct for in cases like the GRE.

Just chiming in bc I had that discussion before--range restriction is afaik *not* the same as collider bias caused by conditioning on joint outcome. These are two distinct problems!

I suspect there may be some of both going on in GRE prediction problems

This is a fundamental flaw in all studies of this topic I've seen. Has anyone seen a good study? Tough to get data....

Well sure, but there's still plenty of variability within that sample. Some students are let in despite marginal scores. Those scores should have predicted *something* we care about.

not necessarily. post-treatment bias doesn't work that way! depends how selection takes place, etc.

surprised that folks who made it into grad school didn't learn about selection on the DV.

We had a problem set about this exact question in 787 stats at Michigan on selection models. I can never remember which way the right interpretation went!

Makes me wish Walter was on Twitter.

On second thought: no. I say too many dumb things on here.

This was with the great John Jackson because I'm OLD!