SILENT MAJORITY!

 Recently there has been a lot of discussion of the possibility of "hidden Trump voters"--people who intend to vote for Trump but aren't willing to admit that in a survey.  For example, a story in the NY Times said "while the effects of a hidden Trump vote are certainly overstated by the president’s allies, that does not mean that no evidence exists that polls are missing some of his voters. A small percentage of his support is probably being undercounted, and has been in the past, public opinion experts said. And in states like North Carolina, where the margin of victory could be narrow, the undercount could make a difference between a poll being right or wrong."  The idea is that they think support for Trump will meet with disapproval or hostility, so they are unwilling to admit it in public.  

You can make objections against this argument--for example, Nate Silver points out that there are "many places in America where you would face no social stigma whatsoever for showing outward support for Trump (often the opposite, in fact)."  But you could defend it by saying that it isn't just local climate that matters, but the general hostility to Trump in the mainstream media--if people don't know a person they assume that he/she is or might react negatively to an expression of support for Trump.

Looking back over previous American elections since surveys began, there's one case in which it seems that these kind of pressures would have been particularly strong--George Wallace supporters in 1968.  Mainstream opinion was overwhelmingly against Wallace, and this was when the media was dominated by a few "gatekeepers"--TV networks, newspapers, and news magazines.  He didn't have a major party behind him, and his supporters were a minority in most of the country.  Also, at that time the Gallup poll still conducted their surveys in person, and you might expect social pressure to be stronger in a face-to-face interview than over the phone. Nevertheless, the polls in 1968 did not underestimate support for Wallace--the last two polls before the election showed 15% support for Wallace among registered voters, and he actually got 13.5% of the vote.  

One of those polls had an experiment which suggests that they were concerned about the possibility of underestimating the Wallace vote.  A randomly selected half of the sample was not asked to say who they would vote for, but given a "secret ballot" which they marked and put in an envelope.  It wasn't truly secret, since it was connected to their survey, but that technique had been found to produce more accurate predictions of the vote in France, where standard methods underestimated the Communist vote.  Comparing the "secret ballot" to the standard question:

 

All respondents:

                    Nixon    Humphrey     Wallace        Undecided

Standard        41.1%        34.3%            15.9%            8.8%

Secret            42.9%        35.4%            18.0%            3.7%

 The biggest difference is that more people made a choice when they were asked to mark the ballot.  Adjusting for that, support for Wallace was a little higher with the ballot, although the difference is not statistically significant.  

 

Registered voters:

  

                    Nixon    Humphrey     Wallace        Undecided

Standard        43.1%        34.3%            15.6%            6.9%

Secret            44.2%        36.5%            15.7%            3.6%

 

The difference in estimated support for Wallace disappears.  That means that most of the difference in the first table involved people who weren't registered--30% of them chose Wallace with the ballot, and only 18% in the standard question (that difference is statistically significant).  

 I broke it down by region--Southern states (defined as states that were part of the Confederacy) versus others--but no differences in the effect of the ballot were visible for either registered or non-registered respondents.  

It seems reasonable that people who weren't registered voters would be less informed about politics and less confident in their choices, and therefore more influenced by what they think other people might think.  This gets at what I think is the major problem with the "hidden Trump voters" argument.  Survey interviewers are instructed to be neutral and just record your answers without expressing surprise or disapproval.  So for someone who's in an environment where Trump is unpopular, it's a rare opportunity to speak your mind without catching flak for it.   Similarly, for someone who thinks the media is biased against Trump and sees the polls as part of the media, it would be a chance to tell them what he really thinks.

[Data from the Roper Center for Public Opinion Research]