What to do when there is no signal in the noise

The takeaway: General election polls that come out this early are scarcely predictive. This is even truer at the state level. What is predictive, though, is the president’s approval rating. Lucky for poll-watchers, it’s showing the same thing as the head-to-heads. And given the current hyper-polarized political environment, early polls this time around could be more predictive than usual.
Perhaps the only thing worse than taking presidential general election polls seriously when the election is over 500 days away is saying that “a global anti-elite surge improves the president’s chance of victory” and thus the polls are unhelpful. Of course, this is just a dig at a bad NYT op-ed, and has little to do with my point. But you don’t have to be as silly as the journalist who wrote that article to get this whole 2020 data thing wrong. Even tweeting about head-to-head matchups without noting that they have zero predictive power at this point in a race is probably going to get you in some hot water as far as accurately predicting things goes.
So here are a few ways to think about all these polls that are showing Trump, for example, down seventeen percentage points versus Joe Biden in Pennsylvania (give me a break).
Three guiding thoughts:
In the past, national polls have had zero relationship to eventual vote share at this stage in the primary. That’s partially because (a) nominees haven’t even been selected and candidates matter, (b) not a whole lot of voters are tuned in to politics and/or (c) the state of non-political “fundamentals” indicators like the economy and…. the economy (it’s the only one) are still very different than they’ll be next November. And there is real political science evidence to this point. In a book from political scientists Christopher Wlezien and Robert Erikson, polls taken even 300 days out had a correlation with actual results of 0 (that’s the lowest it goes, as far as absolute values are concerned). So read them at your own risk.
It’s probably a good idea to totally ignore the state-level polls. What we know about state-level polling is that they have much higher (a) error and (b) variance than the national polls. This is because it is slightly harder to get the target population right (especially when the polling universe is likely voters) and because things like weighting frequencies by education make a much bigger difference but fewer outlets do it. That’s why, for example, the polls in 2016 were so “wrong”. I jokingly remarked on Twitter that early state-level polling might even lower your IQ. Ok… half-joked.
This could be changing. Because (a) vote choice is more stable when parties are polarized, as they are today, and (b) a larger share of Americans than ever before at this point in a cycle are telling pollsters that they’re interested in the next election, preferences could be closer to what they’d be in a general election. That’s made clear in the following figure from political scientist and genius polling analyst Charles Franklin. But then again, maybe all is as it should be—that is not at all predictive. We can only wait and see…

If you’re going to read—or even share (*shudder*) early polls, at least keep these three things in mind.
Editor’s note:
Thanks for reading my thoughts on this subject, and thanks for subscribing! Your membership adds up and makes all this newslettering possible (reminder: I do all this work independently). Please consider sharing online or with a friend; the more readers, the merrier. Reminder that, apart from getting premium articles, subscribers can comment below each post and participate in exclusive threads. I welcome all discussion!
As always, send me your tips about what you’d like to read about next. Or what you don’t want to read. Or your feedback otherwise. Also, dog pictures are nice, so please send them to me? I’m elliott@thecrosstab.com, or @gelliottmorris on Twitter.
Thanks all,
—Elliott