Pollsters are perplexed. Many believed that the polling errors we saw in 2016 had been adequately addressed in time for the 2020 presidential election. But once again, the polls underestimated support for Donald Trump (and support for Republicans across the board). Now, more than three months out from the election, we still don’t have a great sense as to why.
A number of theories may offer some clues, though. For instance, one popular explanation is that pollsters’ likely voter models were off. Survey screening for likely voters may have failed to adequately gauge voter enthusiasm. Or attempts to contact inconsistent or infrequent voters — who tend to be harder to reach in surveys — may have failed in reaching those more favorable to Trump. Also, due to the pandemic, Democrats chose to limit typical methods to increase voter turnout, like door-to-door canvassing, which may have affected actual turnout. Then again, maybe the polling error was due to sampling problems. If Democrats were more likely than Republicans to stay at home during the pandemic, they would more likely be available to take surveys. Of course, it’s not necessarily an either-or situation. Both of these theories could be true (not to mention a whole host of other explanations), but it’s also possible that something bigger is at play here since the polls misfired in similar ways in both the 2016 and 2020 presidential elections: Pollsters may be systematically missing certain types of Republican support.
This is a relatively new phenomenon, but pollsters have increasingly found evidence of partisan nonresponse — that is, particular types of Republicans are just less likely to take surveys, so these voters’ opinions are not reflected in survey data. This was especially relevant in understanding Trump’s support, too, as many of these voters broke for him and other Republicans in 2020.
But why are some Republican voters more reluctant to take surveys? As the director of polling at the Cato Institute, I, as well as other pollsters, am studying this and currently have two working theories for why this is happening. First, Republicans are becoming more distrustful of institutions and society, and that may be extending to how they feel about pollsters. Second, suburban Republican college graduates are more likely to fear professional sanction for their views and are therefore self-censoring more, including in surveys. Now, of course, understanding who isn’t responding to a survey is inherently difficult because … well, they aren’t taking the survey, and at this point, we don’t know whether these two things are happening independently or are part of the same phenomenon. However, it indicates to me that some of the polling error we saw in 2020 is part of a long-standing issue that isn’t unique just to Trump.
1) Republicans are losing confidence in institutions, and Trump accelerated this distrust.
Long before Trump took office, Republicans were already losing trust in our society and its institutions. But there are now signs that lack of trust could be driving the nonresponse and distrust we see among Republicans in polls. In his examination of what drove survey nonresponse in the 2016 election, Alexander Agadjanian, now a Ph.D. student at the University of California, Berkeley, found that in the past 40 or so years, Republicans’ trust has declined considerably. Initially, they were much more likely than Democrats to say that people could be trusted, but now the gap between the two parties has narrowed. If we look at more recent data, from 2018, we see that trust among Republicans is now at the same level as trust among Democrats: In 1972, 56 percent of Republicans said other people could be trusted, as did 41 percent of Democrats. In 2018, those figures declined to 35 percent among both groups — a 21-percentage-point decline among Republicans and a 6-point decline among Democrats.
Declining trust in institutions also breaks down along very partisan lines with more Republicans than Democrats saying they lack faith in institutions. Take the share of Republicans who believe the national news media has a positive effect on the country. That figure plummeted from an already-low 24 percent in 2010 to just 10 percent in 2017, according to the Pew Research Center. For academia, the decline in confidence among Republicans has been even more dramatic. In 2010, Pew found 58 percent of Republicans thought colleges and universities had a positive effect on the country, but that share dropped to 36 percent by 2017. In contrast, Democrats’ confidence in both the media and higher education has gone up slightly since 2010, by 5 points and 7 points, respectively.
So, why are Republicans losing confidence in institutions?
According to Gallup, the most common reason Republicans gave for their low confidence in universities was that they believed they were too political and biased. Similarly, a 2017 Cato Institute/YouGov survey I worked on found that Republicans tended to believe many major news outlets had a liberal bias while an outlet like Fox News had a conservative one.
To be clear, these trends predate Trump, but he likely also accelerated them. Think of all the times he called the news media an “enemy of the American people” and claimed that the polls were “fake” — that is, unless the results were favorable to him.
And this perception that knowledge gatekeepers like the media and academia are politicized may have given some Republicans the impression that other institutions — like polling — are politicized too. Or at least this is a working theory I have. Take the fact that media organizations and colleges or universities are often frequent sponsors of polls (e.g., CNN, The New York Times, Monmouth University). In fact, in my analysis of the polls included in the Real Clear Politics polling average one month before the 2020 election, I found that 79 percent of the polls were sponsored by either a media outlet or a college or university. Consider that the sponsors of these polls often explicitly identify themselves when they contact respondents and ask them to participate in a survey. If most Republicans believe journalists and academics are politicized, it stands to reason they might assume the polls they sponsor are politicized, too.
Taken together, it’s plausible at the very least that as Republican confidence in societal institutions plummets, so does their trust in polls and pollsters more generally. And, as a result, some Republicans — in particular, Trump voters who have lower levels of social trust — are less likely to take surveys.
2) Republicans, particularly college graduates, worry they will be ostracized for their political views.
But Trump’s anti-establishment rhetoric and Republicans’ declining trust in societal institutions is probably not the whole story either. Polls of the 2018 midterm elections — held in the middle of Trump’s presidency — performed reasonably well. This indicates that something else may also have been at work.
One possible explanation? Republicans may be more likely to opt out of election polls because they increasingly fear retribution for their views. A Cato Institute/YouGov survey I helped conduct in July found, for instance, that 62 percent of Americans have political views they are afraid to share given the current political climate. Republicans were overwhelmingly likely to say they self-censored their political opinions (77 percent) compared with Democrats (52 percent).
Not only were many Republicans afraid to express their political opinions, but those with more education were also more likely than Democrats to say they feared getting fired or missing out on job opportunities if their opinions became known. Interestingly, Republicans with a high school education or less (27 percent) were about as likely as their Democratic counterparts (23 percent) to fear their political views could harm them at work. But Republicans with college degrees (40 percent) and post-graduate degrees (60 percent) were far more concerned than Democrats with college degrees (24 percent) and post-graduate degrees (25 percent) in this regard.
Several other studies have also found that more educated, affluent, white suburban Republicans were hesitant to share their political views. Public Opinion Strategies, for instance, found that Trump voters were more likely to keep their vote a secret from their friends (19 percent of Trump voters versus 8 percent of Biden voters) and that this demographic was more likely to be college-educated white women. Wes Anderson of OnMessage, a Republican consulting firm, also told me that in their research they found that higher-income, college-educated white voters were more likely to say they knew someone who was uncomfortable telling people they were voting for Trump, and what’s more, they were more likely to say that description could apply to themselves.
There is also some evidence that these voters might be less likely to reveal their voting preferences in a live telephone interview, although we want to be careful about putting too much stock into “shy” Trump voters. That said, the pollster Morning Consult did find in 2015 and 2016 that more affluent and educated voters were slightly less likely to indicate that they would vote for Trump in a telephone interview, which carries social desirability pressure, than in an online survey. In 2020 they found limited evidence for an education effect but slightly higher Trump support among higher-income households online. A team of academics also found in an online survey of registered voters that Republicans were about twice as likely (12 percent) as Democrats (5 percent) to say they would probably or definitely not share their true voting intentions for president with a pollster in a telephone poll.
But why are these Republicans resistant to telling pollsters their true preferences? They’re likely afraid their opinions might get out, even though polls are conducted confidentially. In the aforementioned study, the voters who admitted being unlikely to tell pollsters who they were voting for were asked why that was the case. As one respondent put it, “I don’t believe the information would be confidential and I think it’s dangerous to express an opinion outside of the current liberal viewpoint.”
Another said, “I would not give my real opinion for fear of reprisal if someone found out.” These quotes are not just cherry-picked either. The researchers categorized all the open-ended responses and identified six primary reasons why respondents wanted to keep their opinions private, and four of those six reasons involved a fear that one’s political opinion could be traced back to them and prove harmful. Republicans considerably outnumbered Democrats in these fears.
It’s possible, of course, that this second theory of mine isn’t telling the whole story either. It may be that two separate phenomena are occurring simultaneously. Perhaps a sliver of college-educated affluent suburban Republican voters are reluctant to express their views with pollsters, as was detected by Public Opinion Strategies, OnMessage and Morning Consult. And it’s possible that this group is entirely different from another set of low-trust Republican voters who refuse to take surveys. Or maybe these two theories overlap more than we realize.
Nevertheless, this second theory does help explain, in part, why Republicans did better than expected in more affluent suburban House districts. Most Republican incumbents in suburban districts won, and instead of losing seats nationally, as was forecasted based on pre-election polls, Republicans actually picked up a net gain of 12 House seats (and flipped 15) — primarily in suburban areas.
It’s easy to think something may be unique to Trump’s being on the ballot considering we saw such a large miss in the 2016 and 2020 polls (the 2018 midterm polls, on the other hand, performed well), but although Trump did play a role in creating a contentious environment, we should avoid jumping to the conclusion that polling will return to normal now that he’s out of office.
That’s because our country is in the middle of an uncivil war, full of partisan rancor and loathing. This predates Trump, of course, but many believe he brought a person’s political identity to the forefront in both 2016 (immigration) and 2020 (immigration, but he also painted Democrats as politically extreme) to the extent that people are now changing their views on race and gender to match their political party. This, in turn, has meant political discourse in the U.S. is harder. For instance, a Yahoo News/YouGov survey conducted in late May 2020 found that a majority of Americans (52 percent) thought Trump was a racist. With those stakes, agreeing to disagree is simply hard for many to do. And those who think that Trump administration policies did more good than harm, on balance, may opt to not participate in polls in order to keep those views private given the stakes.
But this isn’t unique to Trump. Other elections where political identity was pivotal to the outcome have also produced polling misses. Take the Brexit vote, the 2016 referendum in the United Kingdom to either leave or stay in the European Union. This vote, too, became strongly associated with issues of identity and immigration and had a surprising electoral outcome because polls systematically underestimated the conservative vote for Leave, which in the end won by nearly 4 points (52 percent Leave versus 48 percent Remain).
Notably, Leave support had actually dropped in the polls after a far-right extremist assassinated Labour lawmaker Jo Cox, a staunch Remainer, because of her views on immigration and globalization. But SurveyMonkey’s Chief Research Officer Jon Cohen told NBC News that he thought the drop occurred because many Leave voters didn’t want to be associated with Cox’s murder; they may have just opted out of surveys even though they were still voting in favor of Leave because they were concerned about immigration. And this is a trend I think is likely to continue as long as sensitive, politically divisive issues like immigration, race, identity and citizenship frame the way voters think about what their vote means.
Provided this interpretation of the data is accurate, pollsters will have to contend with societal forces much larger than the industry itself. The perception that societal institutions are politicized, the belief that a growing illiberal zeitgeist will punish dissenting viewpoints, the inherently sensitive and salient issues of immigration and identity — these all combine to undercut the social trust needed for accurate polling. Without Trump on the ballot, the issues at stake could very well change. But it seems unlikely that the highly contentious issues over the past four years will fade into the background simply because Trump is no longer in the White House. Indeed, these issues will likely remain in the forefront in future election cycles, especially if he runs for office again. And unless pollsters can regain respondents’ trust, today’s reluctant Republican survey-takers may continue to conceal their political positions, only to reveal them at the ballot box.