According to the “shy Trump supporter” hypothesis, polls overstate the size of Hillary Clinton’s lead because some members of Trump’s “silent majority” decline to state their choice for president to pollsters. Although they may support Trump for partisan or policy reasons, they are embarrassed to admit their support for fear of being associated with Trump’s socially disapproved views on race and gender.
Figuring out whether shy Trump supporters exist is hard. Some indirect evidence comes from comparisons of live-interviewer phone and online surveys. Trump support was lower in phone surveys in the primaries — possibly because shy supporters are embarrassed in front of live interviewers — but that pattern is more muted now. Then again, this difference could arise because different kinds of people take each type of survey.
So how can we tell if people support Trump if they are ashamed to admit it? The “list experiment” is a way of asking about sensitive topics that may avoid the biases that can threaten direct questions.
The list experiment I ran involved a nationally representative survey of 5,290 adult Americans conducted by Reuters-IPSOS. (Note: this was not restricted to likely voters.) Survey respondents were randomly split into two groups.
The first group was asked, “Here is a list of three things that some people would do and some people would not. Please tell me HOW MANY of them you would do. We do not want to know which ones of these you would do, just how many. Here are the three things:”
- If it were up for a vote, I would vote to raise the minimum wage to 15 dollars an hour.
- If it were up for a vote, I would vote to repeal the Affordable Care Act, also known as Obamacare.
- If it were up for a vote, I would vote to ban assault weapons.
The treatment group was given the same list, plus one additional item:
- If the 2016 presidential election were being held today and the candidates were Hillary Clinton (Democrat) and Donald Trump (Republican), I would vote for Donald Trump.
The key logic of the list experiment is that shy Trump supporters never have to directly admit their support.
I then compared the average number of items reported in the two groups: 1.523 in the first group and 1.847 in the second. The difference is 0.315, which indicates 31.5 percent support for Trump.
What happens when we just ask people directly who they would vote for? Respondents split 32.5 percent for Donald Trump and 38.6 percent for Hillary Clinton, while 28.8 percent reported that they would vote for other candidates, wouldn’t vote, or were undecided.
If anything, the list experiment estimate is lower than the direct estimate, an indication that the Shy Trump Supporter hypothesis is probably wrong. Trump’s fans do not appear to be embarrassed to admit their support.
What if we split up the sample? Presumably, strong Democrats and strong Republicans aren’t ashamed to admit for whom they’re voting — it’s the people in the middle who are likely to be shy about it. The graph below shows the list experiment and direct estimates of support for Trump, broken out by partisan identification. The vertical bars are 95 percent confidence intervals, which capture the uncertainty in the estimates.
For all seven groups, the difference between the direct and list experiment estimates is small. In many cases, the list experiment estimate is below the direct estimate, providing more fine-grained evidence against the idea of Shy Trump Supporters.
At the rally in Wisconsin, Trump assured his supporters that there is a “big, big undercurrent out there.” When I used a survey tool specifically designed to measure such an undercurrent, I was unable to find evidence of it.
The study provides no hint of a silent majority that withholds its opinions from pollsters but will nevertheless turn out to vote for Donald Trump on Election Day.
Alexander Coppock is Assistant Professor of Political Science at Yale University.