In any election as close as the one we had last November, countless variables could have decided the outcome. People will tend to seize on the ones that serve their ideological goals. It’s a somewhat different question, however, to figure out why the polls (particularly the state polls) were off by so much. At The Upshot, Nate Cohn takes a whack at trying to answer that question. There is evidence that a lot of late-deciders went for Trump which is a thing the pollsters can’t be faulted for failing to predict. This could be explained by the so-called Comey Effect, named after former FBI Director James Comey’s decision to link Anthony Weiner’s sexting with minors to Clinton’s private email server in the last week of the election. It could also be explained by the Shy Voter Theory that postulates that people were a little ashamed to admit they were going to vote for Trump and weren’t truly undecided. Maybe it was a little of both.
But there’s a more potent explanation available about why the polls were wrong, which is that they may have been incorrect all along due to a failure to anticipate the importance of educational attainment in candidate preference. Poorly educated people are less likely to respond to surveys which results in them being underrepresented in most polls. But, until the 2016 election, this didn’t tend to skew the results because the correlation between education and how people vote wasn’t all that strong.
The tendency for better-educated voters to respond to surveys in greater numbers has been true for a long time. What’s new is the importance of education to presidential vote choice. Mrs. Clinton led Mr. Trump by 25 points among college-educated voters in pre-election national polls, up from Mr. Obama’s four-point edge in 2012.
This made it a lot more important to weight by education. In the past, it barely mattered whether a political poll was weighted by education — which is probably part of why so many didn’t do so.
It’s a pretty simple theory to understand. If surveys exclude a population that is fairly evenly divided in its voting preference that is not likely to have a big impact on the results, but if they miss a segment of the electorate that is heavily skewed in one direction, then that could cause a large error. Just as a pollster might have to give more weight to Latino respondents in a poll if they haven’t succeeded in contacting enough of them, they may have needed to weight poorly educated respondents more heavily in their surveys.
If this theory is true, it could provide guidance for better polling in the future, but it also tells us something about how the Democrats should respond to their traumatic losses. I’ve had people tell me that rural and working class voters want higher education for their children just as much as anyone else, but it should be intuitive that educational attainment is a lower priority for parents who haven’t gotten a higher education themselves.
The Democrats, going all the way back to Bill Clinton on the campaign trail in 1992, have responded to the impact of globalization on manufacturing and job loss by talking about retraining and access to education. Of course, the cost of college has soared in the intervening years so now the Democrats are competing to come up with the most generous affordable college plans. Setting aside the merits, these appeals are least likely to have political success among people who don’t have a higher education and resent the hell out of the fact that their kids will need one.