(This article was reprinted in the online magazine of the Institute for Ethics & Emerging Technologies, December 13, 2016.)
There is plenty of analysis on why Trump narrowly won the crucial states that gave him an electoral college victory—Wisconsin, Pennsylvania, Michigan—even though Clinton won the popular vote by more than 2.5 million votes. But what was particularly striking was how, even if we control for race and income, educational levels best predict how people voted in the election. Of course this was expected, but I was shocked by how much a difference education made in terms of voter preference.
Nate Silver provides the overwhelming statistical evidence for the effect of education in “Education, Not Income, Predicted Who Would Vote For Trump.” In 48 of the 50 most educated counties in the country—almost all of which lean Democratic—Clinton did better than Obama did in 2012. And this holds true even in those educated counties than lean Republican. But in 47 of the 50 least educated counties in the country—almost all of which lean Republican— Clinton did worse than Obama had done in 2012. And this held true even in those less educated counties than lean Democratic.
Now you might think that income levels rather than education was more important. In reply Silver notes:
How do we know that education levels drove changes in support — as opposed to income levels, for example? It’s tricky because there’s a fairly strong correlation between income and education. Nonetheless, with the whole country to pick from, we can find some places where education levels are high but incomes are average or below average. If education is the key driver of changes in the electorate, we’d expect Clinton to hold steady or gain in these counties. If income matters more, we might see her numbers decline.
And what did Silver find? In high-education, medium-income white counties Clinton did better than Obama, while high-income, medium-educated white counties Clinton did worse than Obama. In addition, highly educated majority-minority counties shifted toward Clinton, while medium-educated majority-minority counties shifted toward Trump. Thus Silver concludes:
In short, it appears as though educational levels are the critical factor in predicting shifts in the vote between 2012 and 2016. You can come to that conclusion with a relatively simple analysis, like the one I’ve conducted above, or by using fancier methods. In a regression analysis at the county level, for instance, lower-income counties were no more likely to shift to Trump once you control for education levels. And although there’s more work to be done, these conclusions also appear to hold if you examine the data at a more granular level, like by precinct or among individual voters in panel surveys.
This conclusion was confirmed and expanded on by exit polls as described in Harry Enten’s piece, “Even Among The Wealthy, Education Predicts Trump Support.” Exit polls of white voters show clearly that every bit of education means less support for Trump as can be seen here:
|EDUCATION LEVEL||CLINTON||TRUMP||TRUMP MARGIN|
|High school or less||27%||69%||+42|
|Some college or associate degree||29||65||+36|
|FAMILY INCOME||NO COLLEGE DEGREE||COLLEGE DEGREE OR MORE||DIFFERENCE|
The conclusion here is straightforward. The more educated you are, even controlling for income and other factors, the more likely you were to vote for Clinton, and the less educated you were the more likely you were to vote for Trump.