As anyone who lived through Election Night in 2016 knows, the political polls that year, especially in the presidential race, were off.
Yes, they nailed Democratic nominee Hillary Clinton’s win in the national popular vote, but the presidential race is a state-by-state contest decided by the rules of the Electoral College. And in state-by-state polls, many of Donald Trump’s voters were undercounted and it gave the impression that Clinton had the race in the bag.
Pollsters largely said they learned some important lessons from that election year. They made changes, particularly around how they would approach the education levels and economic classes of those they polled.
But polling in the 2020 election turned out to be even worse than 2016, according to a report commissioned by the American Association for Public Opinion Research (AAPOR) and released this week.
A task force looked at 2,858 polls, including 529 national presidential race polls and 1,572 state-level presidential polls in 2020. The conclusion: these polls overstated the margin between President Biden and Trump by 3.9 percent in the national popular vote and 4.3 percent in state polls. And collectively, it was the biggest polling error in 40 years for the national popular vote, and the biggest in 20 years for state-level votes.
The problems weren’t just contained to incorrectly gauging Trump voters. Actually, the polls were worse than the reality for Republicans running for governor and US Senate. The prime example might be here in New England where Maine Senator Susan Collins, a Republican, was easily reelected despite her Democratic challenger leading every credible poll for the previous 14 months.
But here is the real kicker: Those who studied the issue say they have no idea what exactly went wrong, or how to fix it for next time. These pollsters had already made the appropriate changes from the 2016 election by assuming a higher turnout from rural, less educated voters than they did before the Trump era. Further, while 2016 featured voters who made up their minds at the last minute, in 2020 minds were made up for a long time. In some cases, according to the report, voters had already cast their votes before they were even polled.
As the report frustratingly put it, “identifying conclusively why polls overstated the Democratic-Republican margin relative to the certified vote appears to be impossible with the available data.”
This suggests that the way that reporters and news consumers read these polls will have to change. Here are three ways, going forward, to do that.
1. Polls are still important, but we need to talk about them more broadly
There are a lot of polls now. And there are already a lot of polls ahead of the 2022 cycle. To say everyone should just ignore them is both never going to happen and not necessary.
While the polls were off by three percentage points, they did capture, in unison, trend lines. In 2020, they showed which states would likely decide the presidential election, and they were correct.
So going forward, polls do matter, but the way we talk about them should change: Reporters and others could say more often that races are “too close to call” based on the margin of error for each candidate polled, and avoid getting caught up in overhyping a race that appears more competitive than it might be because one candidate is up two percentage points versus four.
2. Build in a Democratic bias in the poll results
Another way to look at polls going forward is to read them with the assumption that Democrats might poll two or three points higher than is reality. This is not to say that the pollsters are bad actors. The report wondered, for example, whether a working-class white registered Republican from a rural county who agreed to participate in a survey instead of just hanging up is possibly more open-minded about voting for a Democrat. After all, Trump and many Republicans have made it a talking point that the polls are all wrong or all out to make Republicans look bad.
This might mean that some wouldn’t even bother to participate when asked.
In other words, if a poll shows a Democrat up by six points over a Republican, then maybe the race is closer to tied.
3. Be open-minded that bad polling is something that happened when Trump was on the ballot
It’s possible that Trump being on the ballot, and the unique voters he brought out, could be the reason these polls were wrong. (That is not an excuse for them to be wrong in the first place, just a theory.)
In 2018, after all, the polls were actually quite accurate in showing the major Democratic gains. The polls in the Georgia run-off Senate contests this January were also accurate. What neither of these had were Trump, himself, on the ballot.
Trump is not expected to be on the ballot again in 2022 and who knows about 2024. But it is possible that Trump’s presence offers a unique bad moment in polling.