We can reject the idea that Trump’s victory last week was due to racism, as the New York Times‘s Nate Cohn’s data show. Obama voters flipped in favour of The Donald.
Is Trump’s win due to sexism?
Responding to this question is tough: existing studies are poorly designed, sampling from only a handful of white men in a single state.
A better, yet still flawed, method of proceeding is to determine whether states that have female Representatives, Senators, or Governors were less likely to vote Trump.
I was bored, so I ran this regression at the state level, with ‘Trump’ as a dummy outcome variable, and ‘female politician’ as a dummy explanatory variable. The coefficient is -0.163, but the p-value is 0.339, meaning an insignificant result.
Removing ‘governors,’ and focusing only on Congresswomen (i.e. federal politicians) yields a coefficient of -0.159, with corresponding p-value of 0.313. Again, this is insignificant.
In other words, there’s zero difference.
Of course, this is bad methodology because female political representation is correlated with other relevant factors, and the sample size is too small. Also, I do not cluster at the state level in the above regressions, though when I do so, this does not change either the effect size or significance by much.
What we really want is county-level data, and a conditional probability: the probability that Trump is elected, given that a county rejected a female politician who was at least as good as her male competitor.
This is difficult to do: what exactly constitutes a qualified politician?
And separating statistical, from authentic, prejudice is not easy.
A more direct approach would be to systematically use good polls, such as the CCES and Pew, correlating characteristics across individual voters, while controlling for demographic factors. This approach, of course, does not permit us to draw causal inferences.
A still better approach, but one that lacks external validity, would be to run psychological experiments with Trump vs. Hillary voters, to determine sexism. This would at least show causation, though it would require a lot of replication that probably is not worth it.
In summary, the answer is that nobody knows, because no one has satisfactorily attempted to falsify the hypothesis.