Voters’ level of education closely tracked with the candidate they backed in key states in the 2016 election, but state-level polling failed to place enough emphasis on education to accurately reflect that, according to a new report.
While voters with a college degree were more likely to support Hillary Clinton, those who didn’t attend college tended to back Donald Trump. National-level polls predicted Clinton would win the popular vote by 3 percentage points, when she actually won it by 2 ― so those were fairly accurate. But polls were way off the mark in key states such as Michigan (6 points off), Wisconsin (7 points) and Pennsylvania (5 points) ― and Trump’s victories there helped him gain the Electoral College votes necessary to become president.
State-level polling didn’t survey enough voters without a college education ― which may be a big reason they overestimated Clinton’s support, a topic discussed in the report issued Thursday by the American Association for Public Opinion Research on what went wrong in 2016.
The report found many state-level pollsters weren’t accounting for the fact that those with college degrees are significantly more likely to take surveys. While pollsters weight their survey responses to reflect a representative balance of people based on demographics such as age, gender, race and political party, not all weight them based on education.
“National polls are twice as likely to weight for education as state polls,” noted Courtney Kennedy, director of survey research at the Pew Research Center and one of the polling experts on the committee that wrote the report.
In the past, education has not been as strong a predictor of the presidential vote. In 2012, 50 percent of college graduates and 51 percent of non-college graduates supported Barack Obama, according to exit polls. Obama in 2008 and George W. Bush in 2004 also each won while garnering similar levels of support from both college graduates and non-college graduates. However, in 2016, Trump won 51 percent of non-college graduates, but just 42 percent of those with college degrees.
As Nate Silver pointed out in November, Clinton won by a larger margin than Obama did in 48 of the 50 most-educated counties in the United States. So while Clinton won the educated vote, many lower-education voters flipped from Obama to Trump.
Education levels of poll respondents is complicated to measure because there are almost as many ways of weighting the data as there are pollsters. The report states that after the election, committee members attempted to weight their surveys based on education levels and found a “very small change in the pre-election vote estimates with no systematic improvement.”
The report concludes with the sentiment that election polling is a unique beast — Americans shouldn’t distrust all polls because of this example — and pollsters are working to adjust their methodology for more accurate political polls in the future.
For now, committee member Mark Blumenthal, now at SurveyMonkey and previously at HuffPost, advises pollsters to be open and honest about their data and methodology and to work collaboratively with committees like AAPOR to ensure everyone is held to the same standards.