On the night of Nov. 3, as early election returns came in, many observers argued that once again the election pollsters had been wrong. That had changed by Friday. In the presidential race at least, nearly every state had gone in the direction pollsters had predicted, once mail-in and early votes were counted.

For instance, Nate Silver at FiveThirtyEight issued 56 forecasts: 50 states, the District of Columbia, three congressional districts in Nebraska and two in Maine. Biden won in every place he was favored — with three exceptions: Florida, North Carolina and Maine’s second congressional district.

However, some state polls did underestimate support for President Trump. For example, polls in Iowa showed a 1 percent Trump advantage, but he won by about 8 percent. Polls in Pennsylvania showed a 5 percent Biden advantage, but he won by only about 1 percent. Some observers blame “shy” Trump voters — those who tell pollsters they’re undecided, or won’t vote, or will vote for Biden, but then cast their ballots for Trump. Are they right?

Here’s how we did our research

After studying 2016 election polling, the American Association for Public Opinion Research reported that it found no evidence of “shy” Trump voters. More recent studies challenged that, suggesting such voters might exist. Survey firms that correctly predicted the 2016 election have argued that they did so because of how they dealt with pollster-shy voters.

We fielded an experiment designed to measure whether prospective voters feel social pressures when telling pollsters which candidate they plan to support; social psychology research shows that can bias polls. We fielded our experiment to a national sample of about 1,698 respondents from Oct. 23-29 using Qualtrics Panels. Though not a probability sample, quotas and weights ensure age, region, gender, race and education match U.S. census estimates of the same.

Respondents were randomly assigned to answer either which candidate choice would make the “worst” or which would make the “best” impression on others:

Please select the answer you believe will make the WORST impression/BEST impression on others — even if this answer does not actually describe your true feelings.
If the 2020 presidential election were being held today, would you vote for Donald Trump, the Republican, or Joe Biden, the Democrat?

By comparing the answers from the two groups, we can estimate what social pressure respondents are feeling. If fewer respondents say Trump would make the best impression on others than would say the worst impression, that would suggest they feel social pressure to reject Trump — which might make Trump voters less willing to answer pollsters honestly. For comparison, elsewhere in the survey we also directly ask which candidate respondents planned to vote for.

About 6 percent more say that voting for Trump would make the worst impression than say either they would vote for him or that voting for him would make the best impression (all percentages reported are weighted). In other words, potentially about 6 percent could misreport their Trump support because of social pressure.

However, social pressure can influence not just who someone says they will vote for; it can also influence who they do vote for. If someone intends to vote for Candidate A, and also feels social pressure to vote for Candidate A, they will probably answer the polling question accurately.

Which individuals feel social pressure?

To sort out who was likely to vote for the person from those likely to tell pollsters one candidate but vote for another, we went back to the original respondents. If someone had been asked which would make the best impression, we now asked them which would make the worst and vice versa. Respondents who answered differently depending on how the question was framed feel social pressure. Individuals who say “Biden” makes the best impression and “Trump” makes the worst feel pressure to say “Biden.” Individuals who say “Trump” makes the best impression and “Biden” makes the worst feel pressure to say “Trump.”

We found that 22 percent answered differently, depending on the way the question is framed — suggesting they feel pressure to say one candidate or the other. Within this 22 percent, however, 8 percent said “Trump” would make the best impression and “Biden” would make the worst impression. In other words, 8 percent might have been shy Biden voters, not shy Trump voters. They felt social pressure to say “Trump” in surveys. Many probably did plan to vote for Trump. Another 14 percent of the 22 percent said “Biden” would make the best impression and “Trump” the worst. These are potential shy Trump voters, but many likely did vote for Biden.

Which pressured individuals are shy voters?

To figure out which are likely to say one thing in a survey and do another when casting ballots, we next divided respondents by their party affiliations: Democrats, Republicans or independents.

Of the 14 percent who felt pressure to say Biden, 10 percent are Democrats. As nearly all Democrats voted for Biden, these probably did in fact vote for Biden. Similarly, of the 8 percent who felt pressure to say Trump, 5 percent are Republicans and probably did vote for him.

Only 1 percent of people who feel pressure to say Biden are Republicans. These might be shy Trump voters, because they might be planning to vote for Trump as nearly all Republicans did. But 1 percent of people who felt pressure to say Trump are Democrats. This is important because a) there aren’t very many likely shy voters and b) the number of shy Biden voters cancel any polling bias caused by shy Trump voters.

The remaining 5 percent of the 22 percent of people that felt social pressure to say one candidate or the other are independents. Among this 5 percent, 3 percent felt pressure to say Biden and 2 percent felt pressure to say Trump. But we think them unlikely to be shy Trump voters because, while a few more felt pressure say Biden, exit polls indicated that only 40 percent of independents voted for Trump. In other words, most independents who felt pressure to say Biden probably voted for Biden. These cancel out any polling bias created by independent shy Trump voters.

Many voters felt social pressure to say they support one candidate or the other. But looking closely, we find about the same number of likely shy Trump voters and likely shy Biden voters. And of course, some voters who said one thing to polls and did another in the voting booth might have been genuinely undecided when polled.

We expect researchers to find that record-breaking turnout and record-breaking levels of early and by-mail voting made it unusually difficult for pollsters to identify likely voters and that these difficulties account for many polling misses.

Ryan L. Claassen is professor of political science at Kent State University.

John Barry Ryan is associate professor of political science at Stony Brook University.