Multiple regression analysis is a technique that measures correlations between variables, but adjusts things so that each variable’s influence is shown “all other things being equal.” Let’s imagine that Trump voters tend to be against both abortion and immigration, and most voters who are against abortion are also against immigration. But the few people who are fine with abortion and against immigration still tended to vote for Trump, and those who are against abortion and fine with immigration tended to vote for Clinton. In that case, multiple regression analysis will show a strong association between views on immigration and Trump voting, but only a weak relationship between views on abortion and Trump voting; one might reasonably guess in this case that immigration mattered more to voters than abortion did.
Regression analysis can be as much art as science, if the goal is to present a clear and informative explanation of causes and effects. For example, there are questions on the ANES survey about how much the respondent likes or dislikes Hillary Clinton. This question is an extremely good predictor of whether the respondent voted for Trump, but it’s also not very interesting; it’s obvious that someone’s general feelings about the candidates will inform how they vote; we would rather know why they feel the way they do.
The goal, then, is to find a small-ish number of narrowly-worded questions that predict voting behavior as well as possible. After hours of tinkering, this is the model I’m most comfortable with:
- Disliking Obamacare made white Obama voters 42% more likely to vote for Trump.
- Believing that illegal immigrants raise the crime rate made white Obama voters 35% more likely to vote for Trump.
- Believing that the police treat blacks just as well as they treat whites made white Obama voters 34% more likely to vote for Trump.
- Liking North Carolina’s transgender bathroom bill made white Obama voters 30% more likely to vote for Trump.
- Thinking people get offended too easily made white Obama voters 22% more likely to vote for Trump.
- Thinking it is sometimes justified to torture suspected terrorists made white Obama voters 21% more likely to vote for Trump.
- All of these findings were highly significant. After adjusting for these questions, the following things were not significant:
- Gay marriage.
- Gay wedding cakes.
- Worrying about the economy.
- Distrusting politicians.
- Free trade.
- Sending troops to fight ISIS.
- Believing the world is changing too fast.
- General authoritarianism.
- General tolerance.
- The following things hovered on the edge of significance depending on the specifics of the model:
- Believing that blacks would be as well-off as whites if they tried harder might have made white Obama voters slightly more likely to vote for Trump. This question has often been considered a way to uncover “hidden racism” that respondents aren’t otherwise willing to admit to. I should note that using the question in this way is controversial; the fact that very few political scientists are conservatives make research in this area problematic.
- Believing that women who complain about discrimination are asking for special favors might have made white Obama voters slightly more likely to vote for Trump. I’m not sure if this question is used as a way to uncover “hidden sexism”, but it seems to me that it’s somewhat parallel to the “blacks try harder” question.
- Believing that illegal immigrants take jobs away from Americans might have made white Obama voters slightly more likely to vote for Trump.
- I ran the same model for Hispanic Obama voters, and got very similar results except that these voters are less concerned about illegal immigrants raising the crime rate (they are still significantly concerned about that, though.)
- I ran the model for both Obama and Romney voters together; as I mentioned before, this tells us more about Republicans and less about “Trumpism.” Loyal Republicans, it turns out, care much more about Obamacare, abortion, and gay wedding cakes than Obama/Trump voters do.
- For the most part, specific political issues seem to matter more than personality traits or worldviews. Specific racially-loaded political issues seem to matter more than broad racial views.
- It seems to me that the concerns of Obama/Trump voters cluster around a fear of criminal violence and an association of criminal violence with otherness.
- There is little evidence that economic issues (other than Obamacare) played much role influencing Obama/Trump voters in the 2016 election. As I will discuss in the third part, that’s not necessarily the same as saying that economic issues don’t matter to these voters.