Why the Polls were Off…and What They Did Tell Us
Down the home stretch, most polls had Biden in the lead with a 10-point advantage. After all the votes were in, the margin in the national popular vote was about 3 percentage points. This is disconcerting to heavy users of polling data — political campaigns and the media, in particular — leaving them to ponder why the polls were off…again! Could it be that rural voters are harder to reach? That calls to landline phones reach a different audience than cell phones? That voters changed their minds during the final days of the campaign?
The apparent answer is, none of the above. There is a simpler, more obvious explanation, and it is revealing about how people think and vote.
Polling services have become increasing sophisticated — sampling with precision, updating outreach methods, accounting for demographic patterns, and fine-tuning the margin of error. The one source of error they don’t account for is that some people lie. What happened in 2016 and again in 2020 is that a certain percentage of responders said they favored Biden, but voted for Trump in the privacy of the voting booth.
There’s no mystery as to why some people would hesitate to declare who they will be voting for a candidate who lies, engages in corruption, endangers the health of millions, supports domestic terrorists, and undermines democratic institutions. One can only stretch so far in excusing Trumpian behavior because of his policies on immigration, abortion, or the economy. In some social circles and some households, Trump voters felt pressure to dissociate from the Trump brand of white supremacy, misogyny, conspiracy theories, and rejection of science.
Surveys tell you what people say. Election results tell you what people do. Pollsters count on there being little or no different between the two, and that’s typically that case. But as with everything Trump, the situation is hardly typical. What pollster attributed to the margin of error is interpretable information. This so-called error was anything but random. In both national and state polls, the direction was consistent in underestimating support for Trump, and was fairly consistent in magnitude. The pattern was seen in 2016, and was even more pronounced in November of 2020.
If this explanation about a consistent gap between declared votes and actual votes is correct, we would see the discrepancy replicated in exit polls. And that is indeed the case.
The exit poll of over 15,000 voters conducted by Edison Research provides strong evidence of this. To account for the record number of mail-in ballots, the survey reached voters at polling places, at early voting sites, and by phone. With women voting in equal or greater numbers than men, the exit poll indicated that women went for Biden 56% to 43%, while men opted for Trump 49% to 48%. These numbers don’t add up. They point to a 52% to 46% margin of victory, when the actual difference was about 3%. With almost identical results, the CBS News exit poll suggested that Biden’s 3-point popular vote advantage should have been at least twice that. These exit polls correspond more closely to pre-election polls than to the actual results.
The Center for American Women and Politics kept a running scorecard of the gender gap in every reputable national poll during the run-up to the election. Averaging the results of the seven national polling services that surveyed voters in October 2020, women were favoring Biden by a 57–36 count. The 21-point margin did not hold up; according to the exit poll, women voted 56–43 for Biden. The diminished margin is consistent, however, with the explanation that women who professed to be undecided proved to be Trump voters. Men were fairly consistent in what they told pollsters, favoring Trump 47–1/2% to 45% in pre-election polls and 49% to 48% in exit polls.
One conclusion to draw from these data is that women are more truthful than men, or at least, less inclined to mislead. With a highly consistent pro-Biden percentage, only a tiny percentage of women mis-reported their preference when polled, or had a change of heart. What evidently eroded the gender gap is that about 1 in every 6 women who were voting for Trump reported that they were undecided. This leads one to surmise that men were the poll-deceivers. The data suggest that men cast their ballots for Trump by a greater margin than reported — as much as 5 percentage points.
It only takes a few dissemblers per hundred to account for the discrepancy: a 3-point margin masquerading as a 10-point margin. Pollsters call this margin of error. But when the margin of error is always off in the same direction, it’s more of a trend.
The difference between polling numbers and voting numbers tells the same story as the national picture: an incremental shift toward rejecting Trump as a leader, but rock-solid maintenance of a conservative coalition — bolstered by the populist agenda of gun rights, anti-abortion, white nativism, anti-elitism, and American exceptionalism — that fended off a Blue Wave.
A question being asked in the aftermath of this election cycle is, Should we stop putting so much faith in the polls? To maintain the integrity of their trade, pollsters need to make one more adjustment when one of the candidates stands out as morally reprehensible and politically irresponsible. They need to ask the follow-up question, “Who are you really voting for?”
