Perhaps the most prominent proponent of the shy Trump voter theory is Republican pollster Robert Cahaly, founder of the Trafalgar Group. Cahaly recently remarked to Politico, “We live in a country where people will lie to their accountant, they’ll lie to their doctor, they’ll lie to their priest, and we’re supposed to believe they shed all of that when they get on the telephone with a stranger?”
It turns out that “shy voter” is the most generous of euphemisms: As Cahaly’s words lay bare, by “shy voter” he means prevaricator. Liar. We’re not talking about misremembering — like when people tell a pollster they eat more vegetables than they do or attend religious services more frequently than they do. No, we’re talking about baldfaced lying. Proponents of this theory claim that some Trump voters are embarrassed to admit their support for Trump to a pollster. The average Trump supporter, sporting his red MAGA hat, might be surprised by this characterization.
To believe that lying to pollsters explains pollsters’ failure to capture Trump’s true support means believing that only Trump voters are liars — or at least that there are a lot more Trump-voting liars than Biden-voting liars. If supporters of both candidates lied to pollsters about their vote intention, the lying would cancel itself out, and the polls wouldn’t show a bias. So the “shy Trump voter” theory is really this: the “Trump voters are liars” theory.
This theory finds a welcome audience among some of my Democratic friends, who are understandably nervous after being shocked by the 2016 results and perhaps too readily accept the idea that Trump voters lie, but it’s apocryphal.
Let’s examine the 2016 election, where the “Trump voters are liars” theory began. In 2016, the national polls were pretty accurate. The RealClearPolitics final average showed Hillary Clinton winning by 3.2 percentage points, and she won the popular vote by 2.1 points, only slightly less. Yet the polling averages were substantially off in certain states, such as Iowa, Ohio, Wisconsin and Michigan — all of which Trump won — and Minnesota, which Clinton only barely carried.
If the “Trump voters are liars” theory was true, then the Trump-voting liars only existed in those Midwestern states where the polls underestimated his support. That simply doesn’t make sense. You could logically argue that Trump voters didn’t want to admit they were Trump voters in very Democratic states where they were surrounded by Clinton voters — but the polls didn’t underestimate Trump’s support in those states. In Virginia, for example, the RealClearPolitics final average showed Clinton winning by 5 points, and she won by 5.4 points. In California, the polling underestimated Clinton’s support, showing her with a 22.3-point lead when she won by 28.8.
The more likely explanation is that pollsters underrepresented non-college educated voters in 2016 — which meant most of the political class failed to notice the magnitude of Trump’s surge. The Midwest states where polling was off have particularly large populations of White voters without a college education. The Pew Research Center estimates that 60 percent of voters in 2016 lacked a four-year college degree, but that number was even higher in Pennsylvania (64 percent), Wisconsin (63 percent), Michigan (62 percent), Iowa (65 percent) and Ohio (66 percent). White voters without a college education vote Republican at higher rates than White voters with a college degree. This education gap was much larger in the 2016 election than it had been before, so the underrepresentation of non-college voters caused a larger error. In Michigan, for example, the final polling average showed Clinton leading by 3.4 percentage points, but Trump won by 0.3 points. In Wisconsin, Trump won by 0.7 points, but the polling suggested Clinton would win by 6.5. And in Pennsylvania, the polling average showed Clinton leading by 1.9, while Trump won by 0.7.
Why did pollsters underrepresent non-college educated voters? Because we do not have reliable data to tell us what share of the electorate is college-educated. The best data on educational attainment comes from the Census Bureau’s Current Population Survey (CPS), but this is a survey of the whole population, not just of those who vote. Non-college educated Americans vote at lower rates than college-educated Americans, so pollsters thought they were justified in including fewer of them than the CPS reports. But non-college educated voters are also harder to reach when we do surveys, and we simply were not including enough of them, even accounting for their lower voting rates.
In 2020, though, most pollsters are weighting their data on educational attainment to correct for this problem. There are also fewer voters undecided or choosing a third-party candidate in 2020 than in 2016, leaving less room for movement. In 2016, 9.4 percent of voters were undecided or favored a third-party candidate, compared to just 5 percent now. As a result, Clinton only stood at 46.8 percent of the vote in the final 2016 polling average, while Biden is at 51.1 now. And if the polls are wrong this year, there’s more likely a different explanation than lying Trump voters: false assumptions about turnout, higher absentee voting leading to more votes not being counted — or something else we have not yet thought about.
So with the lying Trump voters theory debunked, Trump voters need not ask for absolution from the polling priests. As for the prevaricator in chief (who is anything but shy) — well, that’s another story.
Watch Opinions videos: