The 2020 polling autopsies are still being written, and there are many unanswered questions: How did Republicans win 27 of 27 House districts identified as tossups by the Cook Political Report? How did Sen. Susan Collins win Maine by nine points when she didn’t lead in a single major poll for months? How are President Trump and Joe Biden separated by a little under four points in the national popular vote when Nate Silver’s FiveThirtyEight average showed an 8.4-point Biden lead on Election Day?
The NYU economist Arpit Gupta compared the FiveThirtyEight presidential forecast to that of online betting markets, where people trade contracts reflecting the likelihood a candidate will win. He found that the “markets are pretty well calibrated—state markets that have an estimate of 50% are, in fact, tossups in the election. 538 is at least 20 points off—if 538 says that a state has a ~74% chance of going for Democrats, it really is a tossup.”
Prediction markets aren’t perfect, and some are so thinly traded that they can move with a few bets. But at least everyone knows they’re bets. The polling industry and statistical models make claims to objective expertise, yet they again clearly missed Republican strength.
Combined with media conformity, this can lead pundits and the public astray. The prognosticators make projections—shaped by assumptions about demographics, turnout and response rates—which influence the press. These views are reinforced on social media, especially Twitter . Everyone runs in the same direction, and the press sets the conversation in a way that can further influence pollsters and campaign decisions.
Consider the impact. Radical Democratic ideas like court-packing may not have received as much backing from party leadership if forecasts had been more realistic about Democratic odds of winning Senate seats in Iowa, and North Carolina as well as House seats in Florida, California and Texas. That warped discourse may have increased polarization.