If the 2016 presidential polls had been prescient, President Hillary Clinton would be fighting for her FY18 budget priorities right about now. But instead, today the federal government is foundering under the weight of scandal after scandal emanating from the White House. Something was very clearly wrong with the polls predicting her victory, specifically the state-level polls, and organizations like the American Association for Public Opinion Research (AAPOR) have been digging into the data to figure out what happened.
In May, AAPOR — one of the leading professional organizations for public opinion and survey research practitioners — released its eagerly awaited audit of 2016 presidential election polling. It identified a series of reasons why the polls were off, several of which have been long suspected. Some people were unwilling to admit that they planned to vote for Trump. Others made up their minds at the very last minute in a way that broke against Clinton. But there was one finding in particular that was unanticipated and crucial for those thinking about rebuilding after this loss — by not taking educational attainment into account when weighting their samples, pollsters missed a major shift among voters.
What Weighting for Educational Attainment Means
Pollsters can’t call every voter in America, nor guarantee that the folks who answer their phones are a perfect representation of the electorate. So they weight their polls, meaning that if they only talk to three men and seven women, they take that into account and adjust their data to reflect the actual proportion of men and women in the electorate. Pollsters weight their polls against all kinds of characteristics to more accurately reflect the country — but one of the reasons last fall’s state polls were so wrong is many pollsters didn’t weight their samples by educational attainment.
In the past, they haven’t needed to — as the AAPOR report explains, while historically people with lower levels of education have been underrepresented in polls, especially relative to highly educated people, it hasn’t been considered essential to properly weight for them. That’s because when you graph support for Democratic presidential candidates by education level, it traditionally looked like the letter U: people on either end of the education spectrum — with either the most or the least — have been the most supportive of Democrats, with folks in the middle less likely to vote for Democrats. So even if pollsters couldn’t talk to a bunch of people with lower levels of formal education, they could still produce meaningful results because people at the educational attainment poles generally voted in a similar way.
Why It Matters
But in the 2016 presidential election, the relationship between Democratic support and education level started to shift, and the U-shaped pattern pollsters have relied upon for years started to morph into a diagonal line. Instead of voters at the educational attainment poles mirroring each other, last year certain voters with lower levels of educational attainment were significantly less likely to support Hillary Clinton, while highly educated voters were more likely to do so. Specifically, compared to how they voted in 2012, there was a huge drop-off in support for the Democratic nominee among folks who graduated high school but didn’t go on to any sort of postsecondary education. In pollster parlance, people are grouped by their highest level of education, so these folks are called “high school graduates.”
Using the results of the last two presidential elections by Congressional district and educational data from the Census Bureau, the chart below compares the shift in Democratic presidential support from 2012 to 2016 and the share of high school graduates in each district. In districts with the highest percentages of high school graduates, Clinton did worse than Obama did four years ago. The correlation between education level and Democratic vote choice can be measured by a correlation coefficient — a number between negative one and one that measures the strength and direction of a relationship between two variables. The closer that number is to 1 or -1 (a perfect correlation), the stronger the relationship is. Last year, the correlation coefficient for the percentage of high school graduates and change in Democratic support was -0.79 — meaning that there was a really strong negative relationship between how much of the population was made up of high school graduates and support for the Democratic presidential nominee. As the former increased, the latter decreased. In districts where nearly half of the population graduated high school but didn’t go on to college, Clinton did about eight to ten points worse than Obama. But in places with a smaller percentage of high school graduates, like the highlighted district where they only made up 16.3% of the population, she did almost ten points better than Obama.
Though the correlation was strongest with high school graduates, it’s clear that the relationship between educational attainment and support for the Democratic presidential nominee shifted in 2016 across the board. It wasn’t just that folks at the lower end of the education spectrum were less likely to vote for Clinton than they were for Obama — higher educated folks were also more likely to vote for her than him. That’s why the formerly symmetrical relationship between educational attainment and vote choice is starting to look more linear than U-shaped — because both poles changed. That said, as the table below highlights, the change wasn’t perfectly consistent across every level of education, especially among statistically small subgroups like people who dropped out by ninth grade. Where the relationship was strongest was among high school graduates (who were less likely to vote for Clinton than Obama) and college graduates who didn’t go on to graduate school (who were more likely to vote for Clinton than Obama).
It wasn’t just that folks at the lower end of the education spectrum were less likely to vote for Clinton than they were for Obama — higher educated folks were also more likely to vote for her than him.
Interestingly, while voters in the much-discussed white working class (high school graduates who identify as non-Hispanic white) were less likely to vote for Clinton, that correlation coefficient was slightly lower than that of all high school grads. That means the relationship between education level and vote choice was slightly stronger overall than it was just among white voters at the lower end of the education spectrum. Less attention has gone into potential differences in people of colors’ vote choice by educational attainment, an avenue of future research that could produce meaningful results.
Where We Go From Here
This data tracks with AAPOR’s finding that a meaningful realignment took place in the relationship between educational attainment and Democratic presidential support in 2016. For polling professionals, that means they should be thinking about whether and how to weight educational attainment in their samples, rather than assuming that support among less educated voters will mirror that of their more educated peers. But this data means something much more for the progressive community. It’s clear that there was an outsized shift in support away from Clinton in districts with high percentages of high school graduates. These voters’ concerns reached critical mass in 2016, and their departure from the Democratic coalition significantly degraded Clinton’s viability in a way that the polls didn’t take into account. But now the pollsters know — and so too does the left. Determining what caused this shift, how it happened, and how it can be fixed will be critical for rebuilding a big tent coalition that can win going forward.
Ryan Pougiales is a senior political analyst at Third Way.
