ePrivacy and GPDR Cookie Consent by Cookie Consent
The leading association
of public opinion and
survey research professionals
American Association for Public Opinion Research

2020 Pre-Primary Polls on Democratic Presidential Race Were Accurate, AAPOR Study Finds

CHICAGO, IL, October 12, 2020 - Pre-election polls on the 2020 Democratic presidential primaries accurately tracked the tumultuous nomination contest in most states, a new study by the American Association for Public Opinion Research (AAPOR) said today.

The study found that the pre-primary polls correctly identified the candidate who won each primary contest four times out of five (81%), roughly equal to the accuracy of pre-primary polls in the presidential primary races since 2008.

This report is the first from the AAPOR Task Force on 2020 Pre-Election Polling. Building on the 2016 AAPOR report on poll performance that year, the 2020 group is also analyzing pre-election polls during the general election and will issue a report after final official votes totals are certified from the Nov. 3, 2020 voting.

“The 2020 Democratic presidential primaries were a case study in how difficult pre-primary polls are to conduct well,” said Josh Clinton, the task force chair. “The pollsters who did track the 2020 primaries did a remarkably good job under difficult and changing circumstances.”

The main challenge in the 2020 Democratic primary was the major shift in preferences in the race from Feb. 26 through March 3.On Feb. 26, 2020, Rep. James Clyburn endorsed former Vice President Joe Biden, giving Biden a boost going into the South Carolina primary on Feb. 29. Then two of Biden’s opponents—Amy Klobuchar and Pete Buttigieg—dropped out of the race and endorsed Biden shortly before the precincts opened on Super Tuesday, March 3.

As a result of these sudden changes in the dynamics of the race, the pre-primary polls on the Super Tuesday states were the least accurate, with only 61% correctly identifying the winner, the study found. The study said that for primaries other than South Carolina and Super Tuesday, the pre-primary polls were the most accurate since 2000 with 93% correctly identifying the winner.

“Survey research contributes much more to society than just pre-election poll numbers,” said AAPOR President Dan Merkle. “The value of survey research in general should not be judged solely on the accuracy of pre-election polls. But pre-election polls are an extremely visible case of the use of polls. AAPOR has decided to conduct these studies to provide a solid base of data on which to judge the polls’ election performance.”

Another measure that AAPOR and others have used to judge the accuracy of pre-election surveys is average absolute error. The absolute error examines how well the winning margin in the certified vote compares to the margin between the first- and second-place candidates for each poll.

The study found that the average absolute polling error on the margin of victory in polls conducted up to two weeks before each election day was slightly larger in 2020 (10.0) than in recent primary contests because of the campaign events occurring immediately before South Carolina and Super Tuesday. Excluding the pre-election polls for South Carolina and Super Tuesday, the polls had the lowest average absolute error in the last 20 years. The overall absolute polling error of 10.0 decreases to 7.0—lower than every pre-election primary polling average except 2004 (which was also 7.0).

The metrics of correctly identifying the winner and of average absolute error do not always follow the same pattern by state, by type of poll or by how the poll was conducted. Both have been used for more than a decade in analyzing the performance of pre-election polls.

Other findings of the report:

  • How the poll’s interviews were conducted did not seem to have a significant impact on how well the poll results matched the primary results, although the primary season is a difficult time to make such a comparison. In terms of average absolute error, polls conducted by live human interviews, polls conducted completely online and polls that used IVR (interactive Voice Response) technology were equally accurate.
  • Looking at the polls by election day, those on the Iowa caucuses, the New Hampshire primary and the Nevada caucuses had the lowest absolute error, followed by the polls in the post-Super Tuesday states. Polls in South Carolina and the Super Tuesday states had larger absolute errors due to the quick and dramatic changes in candidate preference immediately prior to these elections. 
AAPOR offers extensive resources to journalists on election polling on AAPOR’s For the Media page. Detailed election polling guidance is available here.