IdeaLog

In retrospect, 2020 polls were fairly accurate

Posted on December 2, 2020

Many commentators have concluded that the 2020 General Election was a disaster for the polling business. And yes, both public and private polls skewed toward Democrats by a few percent, which was not very different from polling in 2016. Nationally, the final polling forecast had Biden winning by 8 percent: 53.4 to 45.4, while Biden actually won by 4.4 percent: 51.3 to 46.9. In other words, the consensus of national polls was off by only 3.6 percent.

Further, when we compare the final polling averages to Presidential and Senate results in the 13 states that were surveyed the most, the polls accurately called the winner 18 times and were wrong only 4 times. (Sources: polling averages from FiveThirtyEight; election results from Politico.)

Pres vote         Pres polls         Senate vote     Senate polls
Biden/Trump Biden/Trump   Dem/Repub     Dem/Repub

AZ        49.4 to 49.1     50.7 to 48.1     51.2 to 48.8     52.6 to 47.4

CO       55.4 to 41.9     54.6 to 42.8     53.5 to 44.2     51.7 to 44.0

FL         47.9 to 51.2     50.9 to 48.4

Ossoff/Perdue Ossoff/Perdue

GA       49.5 to 49.3     50.1 to 49.1     47.9 to 49.7     49.0 to 49.3

IA         45.0 to 53.2     48.5 to 50.0     45.2 to 51.8     48.2 to 49.6

ME       53.1 to 44.0     55.2 to 43.3     42.4 to 51.0     51.0 to 49.0

MI        50.6 to 47.8     53.5 to 45.5     49.9 to 48.2     52.3 to 45.3

MN      52.6 to 45.4     53.7 to 44.6     48.8 to 43.5     54.8 to 42.5

NC       48.7 to 50.1     50.5 to 48.8     46.9 to 48.7     50.5 to 47.3

NV       50.1 to 47.7     52.3 to 46.2

OH       45.3 to 53.3     49.2 to 49.8

SC        43.4 to 55.1     45.9 to 53.4     44.2 to 54.5     46.6 to 51.7

WI        49.6 to 48.9     53.7 to 45.4

State polling on the presidential race was more than 3 percent off in Iowa, Ohio and Wisconsin, but just barely. The Maine Senate election is the only race one where state polls were substantially wrong.

Of course, the issue doesn’t end there. Why were Trump votes underrepresented in both national and state polls? It was not insufficient weighting by educational attainment which was suspected as the cause of polling errors in 2016.

It seems reasonable to say that there is a segment of Trump/Republican voters who were underrepresented in the polls because they are people who are less likely to cooperate. (Or, put another way, they are jerks.) The non-college-educated whites (and to a lesser extent, other subgroups) who willingly answered polls skewed a few percent toward Biden/Democrats because they are more cooperative people.

Participating in a poll is, at least in part, a question of “social trust,” in other words, a faith in people and institutions. Trump supporters include a disproportionate share of people who have low social trust, and the Trump presidency has worsened the problem by encouraging his people to be suspicious of and uncooperative toward conventional political institutions like polling.

Maybe these most uncooperative Trump people will stop voting when Trump is not, himself, on the ballot. After all, the polls had little trouble with accuracy in 2017, 2018 and 2019. Perhaps the Georgia election will tell us or perhaps we’ll have to wait for 2022.