In a pre-election essay, I contrasted the prognostications of poll-influenced prediction models with the wisdom of the betting crowd. I suspected that bettors—many of whom believed their candidate would win—were overly influenced by the false consensus effect (the presumption that most others see the world as we do).
But score this round for the wisdom of the betting crowd, which better anticipated the election closeness, including Trump’s Florida victory. That’s a contrast with Britain’s Brexit vote, where polls indicated a toss-up (a half percent edge for “Remain”) while the betting markets wrongly estimated a 90+ percent Remain chance.
The U.S. polls had a rougher-than-usual year, with the final Biden margin likely near 4.5 percent rather than the predicted 8 percent. The pollsters also struggled in key states. On election eve, the average poll gave Trump a 0.8 percent edge in Ohio; he won by 8 percent. In Florida, the polls gave Biden a 2.5 percent edge; he lost by about 3 percent. In Wisconsin, the polls favored Biden by 8.4 percent; he won by less than a percent. With so few people now responding to pollster calls and texts, precision is increasing a challenge (even with pollster adjusting results to match the voting demographics).
But lest we dismiss the polls and sophisticated forecasters, let’s grant them three points.
First, they got many of the specifics right. FiveThirtyEight, for example, correctly anticipated 48 of the 50 presumed state outcomes. Some of its correct predictions even surprised its creator:
Second, polls, as Silver has said, could be worse and they would still greatly exceed conventional seat-of-the-pants wisdom. In a University of Michigan national survey in September, 4 in 5 Republicans incorrectly anticipated a Trump victory.
Third, although the models missed on some details, credit them with the big picture. “Biden’s Favored in Our Final Presidential Forecast, but It’s a Fine Line Between a Landslide and a Nail-Biter,” headlined FiveThirtyEight in its final election forecast.
And Biden did win. As I write, though, the betting markets—mindful of fraud allegations and legal challenges—still give Donald Trump a 12 percent chance of victory. But this, says Silver, “is basically a market saying there's a 12% chance that the sky isn't blue.”
Consider, say the media analysts who have called the election: How would widespread voter fraud account for Donald Trump’s doing much better than predicted by the historically reliable polls (and for down ballot Republicans doing better yet)? And how could that be so across America in countless local municipalities, including those with Republican-elected officials?
Take my community—Holland, Michigan, a historically Republican town where Betsy DeVos grew up and has a home just a bike ride from my own. With a changing demography that now closely mirrors the nation, our last three presidential elections have been razor close, with Donald Trump narrowly edging Hillary Clinton in 2016 . . . but with Biden defeating Trump by 11 percent.
Likewise, our surrounding county—one of the nation’s most reliably Republican counties—voted Trump by 30.2 percent in 2016 but only 21.5 percent in 2020. Neighboring Kent County, also leans Republican and is the home of Republican-turned-Libertarian Congressman Justin Amash, gave Biden 50,000 more votes than Clinton received four years ago. Is it conceivable that the Republican-friendly voting officials in countless such places across the U.S. consistently committed Biden-supporting fraud?
All in all, it was not the best of years for pollsters and modelers, though it was a worse year for John and Joan Q. Public’s expectations of their candidate’s triumphant success. Winston Churchill once called democracy “the worst form of Government except for all those other forms.” To borrow his sentiment, polls and models are the worst forms of prediction, except for all the other forms.
(For David Myers’ other essays on psychological science and everyday life, visit TalkPsych.com.)
