In the comments to Lott’s post where he failed to notice that the study on guns and road rage reported a multi-variate regression in table 1, someone asked him about my post pointing out his mistake. Lott’s response:
I haven’t bothered looking at Lambert’s page, but he typically
doesn’t have a clue what he is talking about. Assuming that he
actually made the claim that Hemenway et al reported regressions in
the tables, that is definitely not the case. The Tables just show
the means, or conditional means. They are not regression estimates.
Clue: you can’t have a regression where you have a dummy variable
for both male and female (you can only pick one or the other)
because otherwise you will have perfect collinearity with the
intercept term. The package will force you to drop one of the three
Clue: the table contains bivariate and multivariate (in the columns labeled “multivariate”) results. Lott is just looking at the columns that present the bivariate results.
Anyway, it looks like the penny finally dropped, because he has now deleted his claims about multivariate regressions, the comment that asked about my post, and his response above. As usual, Lott tries to rewrite history to cover up his mistakes.