This is strange. In a post-2016 study of why the polls were wrong (and nowhere were they wrong as consistently as WI), it was claimed that the pollsters underestimated the share of white working class voters. However, this does not bear out in the last Marquette Law School poll, which had Clinton +6. In that poll, 46% of voters were college graduates. Similarly, in the CNN exit poll, 45% were college graduates. In the Marquette Law School poll, 36% were high school graduates or below, and in the CNN exit poll, only 20% were high school graduates or below. So if anything, the 2016 Marquette Law School poll overestimated the education level of the electorate, yet still got it wrong.
This Marquette Law School poll finds 55% are college educated, which is reasonable given that it's a midterm. But if education does not explain why they were wrong in 2016, what was the reason? Could it be that Republican voters were afraid of being called bigots or voting for someone the media had so strongly shamed?
The exit polls consistently overestimate the education of the electorate itself though.
Can you provide a source for that? The exit polls are adjusted to match the actual result, which must mean that the demographic factors are adjusted until their bias is removed.
Here’s a good article on this: https://www.nytimes.com/2018/02/27/upshot/trump-losing-college-educated-whites-he-never-won-them-in-the-first-place.html
There are two distinct issues at play here, I think. One is that some people lie about their education level to pollsters in a way that they don't lie about other demographic characteristics. I don't have an example in front of me right now, but we've talked on this forum before about how some polls seem pretty unrealistic in terms of the distribution of education levels of respondents, when you compare to the actual percentage of the population with college degrees or higher. There's no real way to square some of the numbers with overall turnout if everyone's telling the truth, unless you assume that there was ~100% turnout of those with college degrees, which obviously didn't happen. This is a problem for both the exit poll and regular telephone polls, though I'm not sure if we really know if the scale is different in the two cases.
The other problem is, even if everyone were telling the truth about their demographics, the exit polls have to weight all the demographics in the right way to account for differential response rates, in a way that gets the topline number right. But it's not like there's just one unique way to do that, and so there have been attempts to investigate ways in which the exit polls might have gotten the demographic mix of the electorate wrong, including in the link you gave.
