Why The Polls Were Wrong – Again

A pollster talks about why predicting who will win an election, or by how much, has become increasingly difficult.

By Kristen Cabrera & Shelly BrisbinNovember 5, 2020 2:48 pm

A question on many minds right now is, why didn’t pre-election polls accurately predict the outcome? For many, it seems that for the second presidential election in a row, pollsters got it wrong.

Joshua Blank is manager of Polling and Research at the Texas Politics Project. He told Texas Standard that pollsters are still trying to figure out what went wrong with their predictions of a massive win for Joe Biden and other Democrats, and how wrong they actually were. Biden appears to have a path to victory in the presidential race, but his contest with President Donald Trump was far closer than most polls showed.

“The polls were off,” he said. “There’s no doubt about that. If you look across the competitive states – it doesn’t matter whether it was a Democratic-controlled state or a Republican-controlled state… ultimately, it seemed that they were all biased against Trump by about 4 points on average.”

By “biased,” Blank means polls whose results contain systematic errors that over, or under-count a candidate’s support – not a poll that intentionally under-estimates a candidate’s strength.

“Ultimately, we’re OK with random error, because random error cancels out,’ he said. “Systemic error is a problem. And that’s what I think pollsters are going to be spending the next few days, weeks and months trying to uncover.”

Blank says there are three theories about why this year’s polls were wrong; shy voters, nonresponsive voters and flaws in the “likely voters” model.

Blank said there is little evidence of the existence of “shy” Trump voters – people who didn’t want to say they were supporting the president. 

A “nonresponse bias” could have played a role in inaccurate polling, Blank said. If people choose not to respond when a pollster contacts them, their perspective would be missing or underrepresented in the final poll. 

A third possible source of error is the “likely voter” model. Pollsters know a lot about the demographics of a population as a whole, but they don’t know who will actually decide to vote in an election.

“In a historic election with historic turnout and multiple historical crises facing the country at the same time, it’s very difficult for pollsters to go out and say, this is what the electorate’s going to look like,” Blank said. 

If you found the reporting above valuable, please consider making a donation to support it here. Your gift helps pay for everything you find on texasstandard.org and KUT.org. Thanks for donating today.