People wait in line to vote on Election Day in Las Vegas.

Polls have faced strident criticism after President Trump outperformed pre-election surveys nationally and in key states. Now that nearly all votes have been counted, a clearer picture of poll accuracy has emerged, showing that they struggled much more in certain states, while they were much more accurate in others.

A review of polling in 10 key states finds that public surveys underestimated Trump’s vote margin against Joe Biden by about 3.6 percentage points from an average of final-week polls tracked by the website RealClearPolitics, similar to the size of errors in those states four years ago. That pattern was similar to national polls, which underestimated Trump’s vote margin against Biden by an average of 3.3 points, which is larger than the error in national polls four years ago.

The overall accuracy of what polls showed varied quite a bit across key states, with surveys sharply overestimating Trump’s standing in some while giving an accurate image of the race in others. Notably, some of the most difficult states to poll in 2016 proved to be challenging again in 2020.

Here’s how polls did in key states:

States where polls showed a Biden lead, but he won by a narrower margin

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Wisconsin

7.4 point miss

2020

2016

6.9

Michigan

3.5

4

Pennsylvania

2.7

3.1

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Wisconsin

7.4 point miss

2020

2016

6.9

Michigan

3.5

4

Pennsylvania

2.7

3.1

Actual election result

Polling average in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Wisconsin

7.4 point miss

2020

2016

6.9

Michigan

3.5

4

Pennsylvania

2.7

3.1

Polling problems were again concentrated in the Upper Midwest. An average of surveys tracked by RealClearPolitics conducted at least partly in the final week before Election Day in Michigan and Wisconsin showed Biden with leads of six and eight percentage points, respectively, before he eked out a three-point victory in Michigan and a 0.6-point win in Wisconsin. In 2016, polls showed Clinton with smaller leads in each state before Trump won each by less than one percentage point.

In Pennsylvania, polls showed Biden with about a four-point lead on average, and he ended up carrying the state by a little more than one point. In 2016, polls showed Clinton with a two-point edge in Pennsylvania, but Trump won by less than one point — a similar level of error as this year.

States where polls showed a tight race, but Trump won

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Florida

5.5

2020

2016

1.4

Ohio

8.2

6

Iowa

6.2

6.3

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Florida

5.5

2020

2016

1.4

Ohio

8.2

6

Iowa

6.2

6.3

Actual election result

Polling average in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Florida

5.5

2020

2016

1.4

Ohio

8.2

6

Iowa

6.2

6.3

In Iowa and Ohio, polls showed Trump and Biden very close, but Trump carried both states by eight points. Polls looked similar to 2016 in those two states; then Trump had small polling leads and carried each by high single-digit margins. Error was larger in Ohio in 2020 than it was four years ago, although fewer polls were conducted this year.

In Florida, polls on average showed Biden with a two-point lead this year, yet Trump carried the state by more than three points. That resulted in a larger error in 2020 than in 2016, when polls indicated an even tighter race with an eventual one-point win for Trump.

States where polls accurately showed a tight race

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Arizona

1

0.4

North Carolina

2.5

2020

2016

3.9

Georgia

-1.2

1

Polls accurately showed a tight race

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

Arizona

1

0.4

North Carolina

2.5

2020

2016

3.9

Georgia

-1.2

1

Actual election result

Polling average in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

1

Arizona

0.4

North Carolina

2.5

2020

2016

3.9

-1.2

Georgia

1

Polls told an accurate story in other key states. In North Carolina, surveys in the final week before the election showed the race roughly split (Biden by one point), and Trump won the state by a one-point margin. Error here was smaller than in 2016, when polls showed a narrow race, but Trump won by almost four points.

In Arizona and Georgia, polls showed a nearly split race this year, and Biden won by narrow margins. Polls were accurate in those states four years ago, too.

States where polls accurately showed a clear Biden lead

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

New Hampshire

0.6

2020

2016

1.8

Actual election

result

Polling average

in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

New Hampshire

0.6

2020

2016

1.8

Actual election result

Polling average in final week

Gap between

polls and results

Polls favored Clinton/Biden

Polls favored Trump

8

4

0

4

8

0.6

2020

2016

New Hampshire

1.8

In New Hampshire, polls showed a clear Biden lead of eight points, and he won the race by more than seven points. In 2016, error was slightly larger when surveys showed Clinton with a much smaller two-point lead in the state and she won by less than a half a percentage point.

What this means

Some of the states with the largest polling errors this year also had the largest errors in 2016 — particularly in the Upper Midwest. And as in 2016, when polls missed, they tended to underestimate Trump’s support rather than the Democrat’s. Even so, in only a handful of states did polls give an inaccurate overall picture of the competitiveness of the contest after all the votes were counted. State-level polling has not been very precise in the past, and this year’s state polls were no exception.

There are various potential reasons for this overall pattern of errors: 2020 was a complicated election year with century-high turnout and a huge shift in how people voted amid the coronavirus pandemic. A majority of all voters cast ballots before Election Day — by mail or in person — and early and mail-in voters backed Biden by wide margins, while Election Day voters largely backed Trump. That may have complicated efforts to identify the always-uncertain population of “likely voters.”

It’s also possible that Biden supporters were easier for pollsters to reach than Trump supporters in polls this year for various reasons. It’s worth noting that past elections show polls have not consistently underestimated Republican candidates more than Democrats — in 2012, surveys understated President Barack Obama’s reelection support.


Election officials in Milwaukee help voters on Nov. 3. (Alex Wroblewski/Bloomberg News)

After the 2016 election, pollsters made some changes to address polling problems evaluated in a report by members of the American Association for Public Opinion Research. By 2020, more pollsters were weighing their samples to match population benchmarks for educational attainment, and there was much more polling in battleground states ahead of Election Day, especially in key Midwestern states. Yet polls still underestimated Trump support.

Final vote totals provide a clearer picture of where surveys were more and less accurate, as well as a road map for efforts to understand potential sources of errors and ways those errors can be addressed. Survey researchers — including those at The Washington Post — have begun investigating those issues, and research over the coming months may shed more light on the factors behind 2020 poll accuracy. Before the election, the American Association for Public Opinion Research assembled a task force to examine the accuracy of polls as it did after the 2016 election, and the Pew Research Center is conducting its own review. In a blog post summarizing possible causes of 2020 poll errors, senior Pew researchers noted that “this effort will take time.” But pollsters are committed to understanding what happened in 2020 and improving what they do in the future.

Scott Clement contributed to this report.

About the report: The 2020 poll averages in this article were calculated as a simple average of the Democratic-Republican vote margin for polls tracked by RealClearPolitics in which the final date of interviews was in the last week before Election Day (Oct. 26 or later). If a polling firm conducted more than one survey in that period, only its final poll was included. The 2016 poll averages included surveys tracked by both RealClearPolitics and HuffPost. Election results are from Edison Research as of Dec. 9.