> Did you make the usual adjustments
> for education, experience, etc.,
> or did you just look at average
> wage levels?
I don't remember "work experience" being an explicit variable in the PUMS, but the general answer to your question is yes. (I mean, I'd expect experience to lower the gap further and correlated with age and holding jobs for longer spells. And that would make the claim I'm making stronger.)
I used GLS regression. And since I had such a large data set, I included all sorts of interactions and dummies, used robust standard errors, experimented with instrumental variables to get around some obvious chicken-and-egg problems, etc.
As statisticians say, I had a lot of "power" in my tests. So I'm pretty confident about these results. It wasn't up to me to publish them though.
> There are a lot of educated, suburban
> white women, so in absolute terms,
> this is not a small problem.
Yes, absolutely! Still, the gender interaction with race/ethnicity and all other markers of class is striking. And that's something that doesn't receive much attention.
In fact, and this will surprise some, when you include all these interactions between gender and a bunch of other demographic variables included in the PUMS, the estimate of the gender ("sex" in the PUMS) slope is positive, meaning that being a woman is associated with higher earnings/income than men for the baseline (male, 16, no education, urban, etc.).
However, if you predicted earnings/income and had women age, get married, get educated, have children, live in suburbia, etc., the algebraic sign would become very negative. Professional women (doctors, lawyers, etc.) had very large income/earnings gaps with respect to men otherwise demographically identical. Again, all in relative terms (%).
When I presented this to some colleagues (feminists), they were shocked at first that being a women predicted a higher income for the baseline. I knew they would be. But then I asked them to list a few stereotypes for me. And we ran the simulations. They came around.
We ran simulations (predicted income/earnings) for a black or Hispanic single mother of 2 with elementary or middle school education living in urban NJ, a white female lawyer, married with 2 children, from suburban NY, a similar white male lawyer, etc. All sorts of hypothetical characters. They were all spot on, according to my audience.