Were the 2020 Presidential Polls Really Wrong?

Maybe the polls were right but the vote was wrong.

Polls leading up to the November 3, 2020 elections accurately predicted that President Joe Biden would get more votes than President Donald Trump but overestimated the size of Biden’s advantage. The polls had what appears to be the highest polling error in 40 years for the national popular vote with an average error of 4.5% for national polls and 5.1% for state-level presidential polls. The discrepancies were even larger for senatorial and gubernatorial races. In states Trump won by more than 5% in 2016, polls averaged 5.3% too favorable for Biden. In states Clinton won by more than 5% in 2016, polls averaged 3.5% too favorable for Biden. These apparent errors were consistent throughout the campaign.

An analysis of the polling failures was conducted by the American Association for Public Opinion Research. The report indicated that the errors were probably NOT caused by:

• Methods of interviewing (telephone, mail, online, in person) or sampling (random-digit dialing, voter registration lists, online recruiting).
• Late-deciding voters voting for Republican candidates.
• Failure to weight by education (a factor suspected in 2016 polling errors.
• Incorrect assumptions about the composition of the electorate in terms of age, race, ethnicity, gender, or education level.
• Incorrect assumptions about the proportions of Election-Day versus early voters.
• Respondents’ reluctance to tell interviewers they supported Trump.

The report did suggest that at least some of the polling errors in 2020 were caused by:

• Too many Democrats and too few Republicans responding to the polls.
• Corrections for nonresponse that were used successfully in previous elections were not effective for the 2020 election.
• Incorrectly accounting for the larger than expected turnout, especially involving first-time voters.

I’m wondering if the polls were actually correct in gauging voter preferences and it was the actual vote that was skewed. Rather than Trump’s support being underestimated by the polls, maybe what really happened was that Biden’s support in the actual vote was suppressed. Perhaps efforts by Republicans to make voting harder for minorities, mainly Democrats, were successful and led to the five-point differential. Political pollsters can control how they select their sample of respondents and how they elicite their responses, but they cannot control what actually happens in the contest they are analyzing.

If turnout hadn’t been so exceptionally high because of mail-in voting and other initiatives, Biden might even have lost. If this hypothesis were correct, it would have significant implications for the next election given all the new efforts being taken in some States to restrict voting under the guise of election security. Ongoing election audits that are trying to find evidence of election fraud are focused on people who voted but shouldn’t have. Perhaps we should explore evidence of election-system fraud by focusing on people who wanted to vote but were prevented. People have different reasons for not voting, but in 2020, there were more non-votes than votes for either candidate in 12 states.

So, here’s the conundrum:

Were the hundreds of polls conducted by scores of independent, professional, political pollsters using different methods and groups of respondents all incorrect for one or more reasons such that the results were skewed in exactly the same way?

OR 

Was the actual vote itself skewed by impediments to voter participation that largely affected potential Democratic voters such that Biden’s support was lower than measured by pre-election polls?

It would certainly be a good idea to consider these questions before future elections. Unfortunately, the data for determining what actually happened isn’t available and is not likely to be available from future elections.