Will election polls be more accurate now that Trump out of office?

A popular theory is that partisan non-response is going to disappear since Trump is not on the ballot. This misses the point.

Five of the biggest Democratic-aligned polling firms released a light autopsy of their 2020 pre-election polls yesterday. The firms — ALG Research, Garin-Hart-Yang Research Group, GBAO Strategies, Global Strategy Group, and Normington Petts — pooled their resources to investigate why their predictions of a Biden landslide were wrong. “Every one of us thought Democrats would have a better Election Day than they did,” the pollsters wrote. Though they did not discover “a single, definitive answer,” there are several hints in their data.

The Democratic pollsters identified four potentially large sources of bias.

First, they found that their polls understated the chance that low-propensity Republican voters would end up turning out. Among this group of people who typically vote very rarely, “the Republican share of the electorate exceeded expectations at four times the rate of the Democratic share,” they wrote. This meant that the polls ended up understating support for Trump and overstating support for Joe Biden.

A second culprit could have been late movement in the polls. Though the pollsters raise the point that trends throughout 2020 were incredibly stable, there was a slight shift toward Trump at the last minute in most battleground states, most notably in Florida. It’s possible that polls in the sunshine state were picking up on a broader trend across America — and one that polls didn’t detect in full.

Third, the authors posit that the polls were biased toward Democrats because left-leaning voters, who are more likely to be college-educated and trust scientists (in part because of political polarization), stayed home where they were more likely to answer pollsters’ calls and fill out surveys online. Conservatives, on the other hand, would have been comparatively harder to reach as they spent more time at work, at bars and restaurants, or doing other activities outside the house.

Finally, they raised the possibility that Republican-leaning voters are simply unlikely to answer polls because they are less trusting of other people — presumably a prerequisite to telling a random person a bunch of facts about your life.

---

Editor’s Note: If you found this post informative, please do me a huge favor and click the like button at the top of the page and the share button below. As a reminder, I’m fine with you forwarding these emails to a friend or family member, as long as you ask them to subscribe too!

---

These theories are easy enough theory to believe, but the remedies are less clear. Pollsters could try to train better likely voter models, but the bias introduced by poorly predicting turnout only skewed their estimates by one or two points — whereas they missed by four points in the average state and by much more in the northern battlegrounds. Neither do pollsters have a great estimate of how trusting the population “should” be, so they can’t adjust their polls to match the true distribution of trust like they can to make them representative by race or educational attainment.

The easier fix is probably to address the root cause of all of these issues: that more Republicans than Democrats refuse to complete an interview when a pollster contacts them. Why?

One theory is that this is primarily Donald Trump’s fault. By campaigning against experts and institutions and frequently claiming that public opinion polls were “rigged” against him, he may have incentivized his supporters to refuse interviews.

This may in fact be the case. In 2018, when Trump was not on the ballot, the average error in the polls was much smaller — and there was no uniform bias toward Democrats. There was still a familiar geographic pattern to the error (with Republicans beating their polls more in northern and more conservative states) but overall the industry did quite well. The polls also did alright in Georgia’s special elections to the Senate in January of 2021, though this could have been caused by a dearth of live-caller phone polling, which was particularly bad in November.

However, it’s worth noting that we only have a few data points to test this theory — and it’s a pretty high-stakes game to be making predictions based on two data points. It is entirely plausible that Trump only expressed a general distrust of institutions — including pollsters! — that already existed among the voters who were likely to vote for him. For example, a reader named Miles emailed me about why he doesn’t answer polls a few days ago:

I am a fiscal conservative who doesn’t respond to polls because I am tired of politicians telling people what they want to hear so they can get elected for another term. By not participating in the polls that I receive phone calls and emails for I am hoping that the polls will be viewed as unreliable and the candidates will talk about what they view as important instead of what they think people want to hear.

It is safe to assume that pollsters aren’t going to get Miles back any time soon. His objection seems pretty firmly established — a general opposition to the delegate model of representation that public opinion polls inherently encourage. It is also worth noting that he is kicking a sort of own-goal here by refusing to answer polls in hopes that politicians won’t rely on them. Given the way the Biden White House is emphasizing public opinion in pushing their agenda, I don’t see that happening any time soon.

Pollsters — and public-minded journalists, too — might find some success in emphasizing the normative importance of the data they collect. I reckon that many people, once they exposed to the critical role polls play in our democracy, will realize how valuable they are in giving a voice to the American people. Maybe then they will participate in them, feeling as if they are affecting the democratic process in a way they would not have if they hung up the phone. Journalists can also be clearer about the methods pollsters use when covering their data and emphasize the uncertainty in each poll, giving people a better understanding of the potential error in the data. These are no sure bet, of course, but it’s a better approach than just twiddling our thumbs until the next election without Donald Trump.