Polling can’t be fixed because only the very nice still answer the phone

I only did two sessions of phone-banking in this election cycle, but even in that very small sample, I came away impressed by how few people still answer their own phones and how elaborate the technology has become for avoiding calls from strangers. I was phone-banking with a group that had pretty sophisticated software, which did the dialing for us, in the background, but even so, mostly I was just leaving voicemail messages. Now and then I got through to an app called Google Assistant, which is basically just a way to hang up on people in realtime without the awkwardness of having let them hear your voice. Years of scam callers also seem to have disinhibited people from just hanging up the old-fashioned way.

I don’t blame people for not wanting to talk to strangers. I hate it myself, and only undertook to call strangers because I believed democracy was in peril. But if no one has to answer their phone any more, the people who do still answer their own phones must be different from you and me. They must not mind somehow. Maybe they enjoy human connection just for its own sake? (I’m reminded of the New York Times profile last year of a woman with a genetic mutation that has made her immune to any feelings of pain or anxiety.) Maybe the people who still answer their phones are . . . nicer? More open to experience? More available to human interaction? Is “sociotropic” the right word? Something like that.

Statisticians probably aren’t ever going to be able to correct for this difference by adjusting for race, gender, income, age, or any other trait. After the widespread pollster failure of 2016, many pollsters decided that their error in 2016 was failing to realize that the non-college-educated were different from the college-educated, and in 2020, almost every pollster did adjust for that. It didn’t help. One rough estimate I saw on Twitter is that in 2016, state polls were off by about 5 percent, and in 2020, they seem to have been off by about 7 percent.

If I’m right, polls from now on are always going to suggest a more humane outcome than will really happen. There will be no way to compensate, because you can’t add extra weight to the opinions of people who don’t answer their phone. They don’t answer their phones! Multiply zero by as large a factor as you want, it’s still zero. In the future, my advice is that whenever you read a poll result, figure out which candidate a mean person would be more likely to vote for, and silently add to that candidate’s numbers an extra 5 to 10 percent. Or more, if you think people are getting meaner.

UPDATE, 1pm: On Twitter, @15c3PO points out to me that David Shor, the political consultant famous for having been unjustly fired, has made a similar suggestion: because distrustful people didn’t answer their phones, the Hillary Clinton campaign failed to realize in 2016 that socially liberal messaging wouldn’t play well with non-college-educated voters. Says Shor:

The actual mechanical reason was that the Clinton campaign hired pollsters to test a bunch of different messages, and for boring mechanical reasons, working-class people with low levels of social trust were much less likely to answer those phone polls than college-educated professionals. And as a result, all of this cosmopolitan, socially liberal messaging did really well in their phone polls, even though it ultimately cost her a lot of votes.

And, @15c3PO adds, reporter Matthew Zeitlin also had the same idea as me, but sooner. The day before the election, Zeitlin wrote on Twitter: “Just putting this out there but isn’t it possible that the low social trust non-response problem has gotten worse and thus the swing toward Biden in the Trumpy demos and the states that swung big toward Trump is something of a mirage…(we’ll find out!).”