On Fri, 27 Oct 2000, Bryan Caplan wrote:
> Unfortunately Bill is too busy to weigh in here. I still haven't
> untangled your simulation, but I'm skeptical that measurement error
> could radically effect the results I showed (or M/H's much more detailed
> results).
Notice your results are stronger than anything in the entire Bell Curve:
you at least contolled for education and some important characteristics
such as marital and fertility status. M/H make no attempt to control
for income, health, marital status, fertility, experience, region of
residence, gender, nationality, and so on, and so on. Even if they had,
there would still be some pretty serious endogeneity problems, but as
it stands their analysis is completely worthless as far as I can tell.
Recall too that folks like Heckman have shown the results in M/H change
dramatically when the analysis is done right.
> If you're right on this, then I'd better start greeting
> almost all results with greater skepticism - in the real world, what is
> better measured than IQ and education?
Well, it's not so much measurement error as the other endogeneity problem
(and recall that was modest) that caused the inconsistency in my little
experiment. That said, I'm not sure education, at least, is particularly
well measured in most datasets, as we generally ignore quality measures.
And recall M/H "solve" the colinearity problem they have between education
and IQ by *dropping* education. Some solution! Granted, they also repeat
most of their regressions with a high school only sample, but notice the
results generally change markedly when that sample is contrasted with the
whole sample. Also recall they admit that education causally increases
performance on IQ tests (although I looked up their cite, a 1989 AER
article, and that piece actually assumes such a relationship exists, but
differences it out so they don't have to measure it, so I don't know what
M/H were talking about). That implies that a simultaneous approach is
necessary, particularly recalling that the endogeneity problem here is
compounded by the colinearity problem.
Perhaps you could explain why you find the evidence they array compelling.
Maybe I'm missing something.
Chris Auld (403)220-4098
Economics, University of Calgary <mailto:[EMAIL PROTECTED]>
Calgary, Alberta, Canada <URL:http://jerry.ss.ucalgary.ca/>
