Election polls say they are within three percentage points of their prediction. For an educated guess surveying a large population, that’s pretty precise and one of the reasons polls have been a mainstay in elections for nearly a century.
But after the 2016 presidential election, where nearly every poll predicted the victory of the then Democratic nominee Hillary Clinton over Republican nominee Donald Trump, the nation seemed stunned by the result. So this time around, when candidates and media organizations regularly cite polls, people understandably wonder: Can these numbers be trusted?
UM-Dearborn Political Science Associate Professor Julio Borquez — whose polling experience began with 1982’s Michigan gubernatorial election (that was James Blanchard vs. Richard Headlee for those who are curious) as a research assistant in UM-Ann Arbor’s Institute for Social Research — has an answer to that question.
Julio Borquez: Election predictions go back a long way, but the modern polls really started with George Gallup, who was experimenting with probability-based surveys in the 1930s. He got the 1936 presidential election right (FDR over Alf Landon) when the Literary Digest infamously got it wrong. Gallup’s sample size was much smaller than the Digest’s, but he took into account selection bias and complete coverage of the population. Literary Digest worked from telephone directories and vehicle registration lists. If someone had a telephone or owned a car in 1936, they probably had money, and probably leaned Republican. Gallup had an interest in probability-based polling because he had this ideal notion that survey research could serve representative democracy by giving voice to public preferences. If preferences could be gauged quickly, candidates could then be attentive and responsive.
JB: Yes. Even though we can all be a bit cynical toward political candidates — especially with this year’s polarizing presidential election — candidates do want to learn more about voters and their base of support to determine what voter priorities and concerns are. Conducting polls is also a good way to gather information to properly inform campaign strategies. Polls give insight to how your campaign is performing and since they can survey a large group of people in a short time frame, they give some insight on if you need to quickly adjust campaign messaging. You don’t want to be out of touch with voters.
Polls also inform news coverage — which also doesn’t have the luxury of time — and allow media organizations to claim objectivity because they are citing poll data. Of course, what is done with those numbers may be used out of context. But the numbers are the numbers.
JB: It has to do with survey methodology. This is usually pretty consistent across polls, but there are some outliers. For example, most polling organizations have real callers who ask the survey questions. But Rasmussen doesn’t do this and that makes it an outlier. Rasmussen relies heavily on robocalls and recorded messages where you hit buttons to respond. What this means is that Rasmussen has more success calling landlines, which leads to more responses from older voters who typically lean more conservative and this is reflected in their numbers. But outliers are few and far between.
If you are concerned about bias in numbers or just want a variety of polling options, the website FiveThirtyEight is a good one because it shares the latest polls and polling trends from several organizations. I also recommend the polls done by the New York Times, the Washington Post and CBS.
JB: With the amount of by-mail voting we are experiencing this year because of the pandemic, we are in a bit of uncharted territory. People are voting by mail in much larger numbers than ever before. But polls do ask how people are voting and if people have already submitted their vote if they are doing it by mail, so it is taken into consideration. I don’t believe the polls will be affected too much by this; but election night might be a whole different story. We’ll have to see what happens after Nov. 3.
JB: I see those news stories that keep reporting how Hillary Clinton was up by double digits 16 days out, just like Joe Biden. That’s true, but 2020 is different.
Before I get into this, I do want to point out that even though the 2016 presidential election was a polling surprise, the national numbers weren’t far off the mark: Hillary Clinton did win the popular vote. The issue was with the state-level polls, which obviously matters with the Electoral College system, and incorrectly predicting which state would go red or blue.
After that election, the American Association for Public Opinion Research commissioned a study to do a post mortem on what went wrong. There were two main issues with state-level polls that year. The last poll before the election was done a week out in some instances, which was a problem because 2016 had many late-deciding voters in key states. The other problem was not taking upper education bias into account. People with more formal education are more likely to take part in polls. They were also more likely to vote for Hillary Clinton in 2016. Some survey organizations got the weighting wrong because they didn’t correct for the overeducated sample. And it mattered. Believe me, every pollster out there is very aware of this.
JB: Anything can happen, especially in 2020. But a polling surprise is less likely. There’s a steadiness in public opinion numbers and fewer undecided voters this time around, which makes life a little easier for polling. Also, looking at campaign fundamentals — those are the cards the candidates are dealt that they can’t do much to change things like incumbency or the state of the economy — one candidate has an advantage this time around; things were more of a mixed bag in 2016.
President Trump’s approval rating has been in the low 40s for months; in 2016, he ran as an outsider who was going to shake things up in Washington for the benefit of the American people. Even though some of his messaging continues to say this, it’s a harder sell in 2020. The stock market is still performing well, but there are economic concerns because we are in a public health crisis which the majority of Americans don’t think President Trump is handling well. These are a few factors going into the low approval rating, which has been consistent.
When you combine this with the lower number of undecided voters, many of whom took a chance on Donald Trump in 2016, it makes for a more probable outcome that former Vice President Joe Biden will win.
JB: It’s not outside the realm. But President Trump’s campaign will need to get everything right from here on out. He needs to stop putting the spotlight on his liabilities and instead focus on his strengths he may have.
It also depends on voters. Polls predict outcomes based on people’s indicated behavior or preferences. If behavior suddenly changes, even the most methodically sound polls won’t have accurate data. If people decide to stay home or not submit their ballot after indicating they were going to vote for a particular candidate, that makes a difference too. That’s why it’s important — no matter what the polls say — to vote.
