The frustrating thing about regression analysis is that it is an incredibly powerful tool that yields limited results. All it tells us is the strength of correlations. You have to draw conclusions about the causal mechanisms at work yourself.

Regardless, we see that a large portion of the GOP fight can be explained very well using only demographic variables. This is what I believe Cost picked up on when he found that northern conservatives voted for Romney, while southern conservatives voted against him. In the north, the conservatives tend to be non-evangelical. In the south, they tend to be evangelical (in Florida, they’re split).

Why this is the case is open to interpretation. The simplest answer is anti-Mormon bias, but that seems a bit too easy. After all, the alternatives are a pair of Catholics. The other possibility — and this is a problem with regression — is that religion could be a stand-in for ideology, and that, regardless of self-identification, a self-described conservative evangelical Republican is significantly to the right of a self-described conservative who is non-evangelical. Or it could be some third possibility: Perhaps evangelicals and non-evangelicals alike in heavily evangelical counties vote against Romney for an additional reason.

The other interesting observation is Romney’s decreased vote share in African-American counties. Again, this is susceptible to many interpretations.