That online sampling can only mean one thing
Many Democrats and Democratic-leaning voters have some rough memories of the last election. It was, according to polls, going to be a cakewalk for Clinton. One election forecaster was so confident in Clinton’s victory and the 99% chance of victory his model foretold that he offered to eat a bug if she lost. He subsequently enjoyed a cricket dipped in honey, live on CNN [Pacific Standard]. So with Joe Biden currently up in the polls by 8 points as of this writing [FiveThirtyEight], there’s a certain unwillingness to believe.
A common theme you hear is the idea of “shy” Trump voters, who are enthusiastic about voting for the President but won’t admit it to a pollster. The polls were wrong last time, or so the thinking goes, due to voters who lied to pollsters or simply decided to hang up when the pollster called. This could be happening again, and lead to Joseph Robinette Biden snatching defeat from the jaws of victory.
Meet the Shy Trumper (or Don’t)
The idea of the ‘shy Trumper’ isn’t new. It’s a new version of an old idea, the ‘Shy Tory’ concept from the UK [Wikipedia]. Polls tended to consistently underestimate the Tory share of the vote (as recently as 2015, below), and so the idea took root that perhaps voters were simply unwilling to admit to a pollster that they were going to vote Tory. It’s not even new in the US context, either - there’s a political meme called the ‘Bradley Effect’ minted right here in California [Wikipedia], based on LA Mayor Tom Bradley’s shocking loss in the 1982 gubernatorial election. The thinking had been similar - voters who were unwilling to vote for a nonwhite candidate but wouldn’t admit it on the phone.
This all falls under a general phenomenon called ‘social desirability bias’. The theory is simple, and psychological in nature - people like to please other people, and tend to avoid conflict. When a pollster calls them, they may not be willing to express opinions they sense might upset the human on the other end of the line. Good luck getting a stranger to tell you their most racist opinion, or that they don’t like the new Taylor Swift.
There are ways to detect social desirability bias that rely directly on the psychological theory. If poll respondents don’t want to tell a stranger something embarrassing, perhaps they’d be willing to enter it into a form on a website? The website will never judge you. If there is social desirability bias around a certain question, that often leads to what are called ‘mode effects’. Mode effects different ‘modes’ for taking a survey like a live phone interview vs. an online survey will yield different results. Social desirability bias isn’t the only source of mode effects, but it should result in mode effects.
Thanks to FiveThirtyEight, I can look at the real data and find…that Trump voters don’t seem to be shy. FiveThirtyEight helpfully provides all their 2020 general election polls for download [FiveThirtyEight], and includes structured data on methodology, such as whether a poll was online, used a live interviewer, SMS, etc. When you break apart recent polls (I used all polls after June 1 2020), it’s strikingly obvious that there is zero difference in reported Trump support between polls involving live interviewers and automated polls.
Shy Trumpers Probably Aren’t Real
This doesn’t decisively settle the debate over ‘Shy Trumpers’ - nothing can settle an argument in this day and age. There are more sophisticated ways to check for this type of bias using surveys designed for this purpose [APSR], which would strengthen that argument. But in 2020 the Shy Trumper theory doesn’t pass the sniff test, either in the data or from a theoretical point of view. The President is, after all, the President - expressing support for the incumbent in a Presidential election is a pretty common point of view. The theory might have been easier to believe in 2016, but for a reelection campaign it’s a stretch.
The biggest source of uncertainty in 2020 remains the possibility not that the polls are wrong, but that they will change. A fundamental turnaround in the virus situation, or in the economic response, or even in the perceptions of either, could lead to a swing back in the incumbent’s favor. Historically there have been big swings in opinion after the party conventions in the summer when normal voters ‘tune in’ [Wikipedia] - god knows whether that will still happen when the convention is just a well-produced Zoom call. And while it may seem unlikely, events could also swing opinion even further towards Biden.
But as far as we know, there’s no reason to think the polls are wildly wrong.