A Field Guide to Polling: Election 2020 Edition

Even outside the U.S., there is substantial evidence that polling hasn’t witnessed a substantial decline in accuracy. A comprehensive review of polling accuracy published in 2018 found that “relying on vote intention polls from more than 200 elections in 32 countries over a period of more than 70 years, there is no evidence that poll errors have increased over time….”

In 2016, problems with polls in a few key Midwestern states led many people to underestimate the chances of a Donald Trump victory. As a consequence, the immediate post-election assessment was that there had been a complete polling meltdown.

But that “insta-narrative” turned out to be oversimplified. The 2016 election was not, in fact, an industry-wide failure for the polls.   Rigorous national surveys – designed to measure the popular vote rather than capture the effects of the Electoral College – were quite accurate by historical standards. An average of the final, publicly released national polls suggested that Hillary Clinton would win the overall popular vote by 3 percentage points, and she ultimately won by 2 points.

A second issue was some states that turned out to be key to Trump’s victory had few, if any, public polls in the field in the last few days before the election. All evidence suggests that more undecided voters swung to Trump than to Clinton in those last days, a trend most state polls were unable to detect as they had already stopped interviewing.

Pew Research Center’s American Trends Panel (ATP) is also a probability-based online panel. We select a random sample of the public and reach them via snail mail at their home addresses. Those selected are invited to complete surveys online, and those who do not have internet access are provided with tablets and data plans to facilitate their participation. (You can read about the details of this approach in our report on building the ATP here.) Increasingly, because of the challenges facing phone polling, we are moving our U.S.-based surveys over to this online probability panel.