I have been working on a post-2020 “Lessons for Investors” column, when I stumbled upon an interesting Nate Silver post mortem discussion (above + ABC). His pre-election forecast was a 90/10% likelihood of a Biden victory. While that was correct, lots of the rest of the polling industry forecasts were wildly off.
Silver notes “Polls did call every state but Florida and North Carolina correctly in the presidential race, and everywhere but North Carolina and Maine correctly in the senate.” But he notes the polls were off by a surprising amount — the national polls were off about 3-4 % in the presidential race, and about 5% in Congressional races. Note that Biden won Wisconsin by less than 1% when the polls had him +8%.
All told, the polling industry did not distinguish itself this year.
But what really caught my attention in Silver’s discussion was how various peoples’ living patterns and differing responses to the pandemic were impacted by their political beliefs. About 37% of jobs can be performed at home, many of which, Silver observed, are “white collar, knowledge sector jobs held by college-educated professionals — a group that mostly votes for Democrats these days.”
That raises the issue of oversampling. Poll response rates for democratic voters shot up about 5% once the pandemic hit in March (from 12% to 17%). Hence, it was not “shy Trump” voters, but more likely, a function of how people in different industry sectors tend to vote, and how people of different party affiliation tended to respond to the pandemic. With the benefit of hindsight, those two factors emerge as a likely factor in the undercount of Republicans for the second straight presidential election.
Whenever we have a broad miss by a large group of analysts, it is worth diving into it to see how and why the analysts went astray. We’re all sick to death of the election by this point, but it is always useful to see what lessons can be learned.