Ballots are still being counted across the country, but the results so far have provoked a lot of public scorn for how far off polling their predictions were. This is not in any way a defense of bad polls: however, it’s important to remember to (1) follow the trend instead of the outlier and (2) mind the margin of error.
A poll is a snapshot of the race in a moment in time. So, relying on an average of polls or focusing on trends – is there a pattern of movement in multiple surveys over time? – often gives a better sense of the state of a race. RealClearPolitics.com aggregated all polling in the battlegrounds (AZ, FL, GA, IA, MI, MN, NC, NV, OH, PA, TX, WI) over time. Looking at the final polling averages in each state compared to the vote share as of Nov. 4, 2020, the averages underestimated President Trump’s strength by as much as 6.06 points (Ohio) and little as 0.64 points (Arizona). They overestimated Joe Biden’s strength by as much as 4.49 points (Minnesota) and little as 0.03 points (Michigan). The only state that has been called so far that the RCP averages misread is Florida, which barely went to Biden in the RCP trend.
Also remember each poll has a margin of error. The polls tracked by RealClearPolitics.com have margins of error of roughly 2- to 6-points. That means Biden’s lead in Florida (+0.9) could have swung 4.9 points either way, giving Trump a win there of as much as 4-points.
Even remembering the importance of the margin of error, major improvements are needed in polling. There were blaring misses with down ballot races – just ask newly re-elected Senators Lindsey Graham and Susan Collins whom polls suggested were on the road to political retirement. Let’s hope there’s lots of soul-searching about methods, practices and changes among polling professionals before the next election.