President-elect Trump’s victory on Election Day shocked the nation, not only because he had never run for public office before, or because he had faced a series of scandals regarding his treatment of women – but because the most widely respected polls predicted that he would lose to Hillary Clinton.
Stats guru Nate Silver issued his final forecast on Election Day. On his blog, FiveThirtyEight, Silver gave Clinton a 71 percent chance of winning the White House. He forecasted that she would win 302 electoral votes. She only won 228.
Polls underestimated Trump in part because Trump voters were slightly reluctant to tell pollsters that they planned to vote for the brash businessman candidate. Susquehanna University Polling & Research Inc. found that Trump performed better in polls where respondents talked to a recording rather than a live person; female Trump voters often opted out of being polled at all.
A common refrain is that pollsters have a hard time predicting who is a likely voter and who will stay home on Election Day. But it’s not just the pollsters who have that problem – registered voters themselves are largely unable to predict if they will cast a ballot. The RAND survey asks respondents to state how likely they are to vote, and the potential voters were frequently wrong about their own future behavior.
The traditional political polling model does not work as well in the digital age. When 66 percent of millennials do not have landline phones, any landline poll is going to under-sample young adult voters. Internet polling is famously unreliable because people can vote in the poll multiple times to skew the results.
Poll aggregator RealClearPolitics claims that “the national polls were actually a touch better in 2016 than in 2012.” The polling averages had President Obama winning a second term by a larger margin than he actually won. The 2012 polls were off, but they were off in the correct direction. 2016 seems like a worse year for polling because the statistics had the outcome wrong. After all, the point of polling is to predict who wins – and forecasting the margin of victory is far secondary.
The American Association for Public Opinion Research stated, “As final results continue to be tabulated it would be inappropriate for us to participate in conjecture.” Finding precise polling errors is an intense process, and the association says it will take six months to report on what went wrong.
President-elect Trump cited his great poll numbers throughout the primary race as reasons to be optimistic about his candidacy. When general election polls favored Clinton, he started claiming they were rigged. Trump surrogate Sarah Palin told Bill O’Reilly: “I just don’t put a whole lot of faith in polls…Like I’ve said before, polls are only good for strippers and cross-country skiers.”
Her silly statement was widely mocked on Election Day, until the results came in and we saw that the joke was on us. Trump outdid all the major polling forecasts. Pollsters have a long way to go to prove their relevance and trustworthiness in 2020.
