My advice to everyone about reading the polls was to ignore them until after Labor Day.
Guess that’s expired. Time for updated advice.
Look: I don’t exactly disagree with those who say that there’s too much emphasis on horse-race coverage in most media outlets and that people should become political actors themselves instead of treating politics as entertainment.
But I also know that most people following political news really want to know who is going to win. And there’s nothing wrong with that! If you’re attentive enough that you’re reading here, you probably already know who you’re voting for — indeed, most of us knew which party we would vote for long before the candidates were selected — and you have a pretty good idea of what the stakes are and the policy preferences of the candidates.1
So while, yes, I’d like more policy coverage, no one has to feel guilty for wanting to know the main unknown — that is, who will win. And now that we’re in the final weeks, the polls are the best hints at that unknown.
But there are still good and bad ways to do consume the polls. Obsessing about each new published survey isn’t just a waste of time; it will actively make you less informed about what’s going on. There’s a lot of ways to misinform yourself, especially this cycle. And to be fair, I’m as guilty of violating all of this as anyone, just as I failed to ignore the polls up until now. But I do try to follow my own advice.
The best guideline to follow is the simple one: Polling averages, not single polls. Seriously: If you are capable of ignoring the individual polls, you’ll be better off. Our brains aren’t built to be able to handle all of the data points involved individually, and without the polling averages we’ll all emphasize what we want to see (including that pessimists will “see” the soundings that give them bad news!). Even with the averages plus individual polls, it’s just very easy to start believing that one or another individual survey is what’s “really” happening.2
Even worse, looking at individual polls makes us susceptible to arguments about why this or that poll should be discounted. Avoid Truthers! We got them from both sides in the weeks leading up to the 2022 midterms. Republicans for several cycles have tried to build momentum by emphasizing good-for-them polls, and in 2022 many Democrats were convinced that low-quality Republican-sponsored (or Republican-sympathetic) surveys were drowning out good data. Here’s the thing: You’re not an expert. You can’t evaluate whether Fox News polls are solid (generally, they have an excellent reputation) or whether some outfit you’ve never heard of is reliable or not (some yes, some no). Leave it to the “neutral” experts. For example, the 538 team is generally pretty good at it. They could be wrong, yes. But for virtually all of us, second-guessing the aggregators is only going to tell us what we’re looking for, not get us closer to the truth.
Those are the main two guidelines, and I think most experts would agree. But I’ll add a few more:
I’m basically ignoring the prediction models this time around. I just can’t shake my core belief that This Time May Be Different in ways we just can’t quantify. It’s not just that treating this as a normal open seat presidential election may not capture the situation properly, or that both the economic facts and perceptions of the economy are unusual. It’s also that the late substitution of Kamala Harris for Joe Biden (plus a few other oddities) make it hard to guess whether voters will make up their minds at the same pace they usually do. In other words, none of things that the models give us that basic polling averages cannot see are apt to be reliable this time around, including their estimates of uncertainty.
Instead, take the head-to-head polling average. Understand that there’s a fair amount of normal polling error that makes estimates more of a range than a single prediction, and also realize that the more time remains, the more the chance that something changes. And that’s about it. I’m glad the models are out there, and up through 2016 I thought they were helpful, but this time (and in the pandemic election) I think we’re better off emphazing a general feeling of uncertainty than trying to quantify it.3
I also advise sticking to national polls, not state polls and then just adjusting for the likely GOP favoring bias in the electoral college. Yes, it does appear that as long as the election remains close, it will come down to seven competitive states. But we can’t rank them (most likely Trump to most likely Harris) with much confidence, and for a variety of reasons state polls just aren’t as accurate as national ones.
To be sure: We don’t know what Harris needs to do nationally to win the electoral college. The bias (that is, how much Harris has to win by nationally to win the electoral college if all states move together) could be as much as five percentage points or it could even slightly favor the Democrats, as it often did before 2016. But in my view, dealing with that by focusing on state polls winds up more likely to mislead than just following the national polling and admitting we can’t really know how that translates into the electoral college.
And if you can’t ignore them, be at least very cautious about the crosstabs. Remember that (all things equal) the smaller the sample size, the less certainty we can have about the results, and so estimates for how subgroups can easily be pretty unreliable – especially if it’s a combined group, such as (1) young (2) Black voters or (1) older (2) Anglo (3) married (4) women (5) without college degrees. It’s not that it’s impossible to study such groups, but if you take a general-purpose survey and then look for unusual subgroup results, you’re likely to find one created by chance, not reality. And since “group has surprising views!” is news, a lot of outlets have an incentive to highlight such results.
So that’s it. Once again: I certainly understand the impulse to dig deep into all of this, and I’m not going to advise anyone to ignore the horse race.4 But a daily glance at a national polling average along with a daily reminder of the uncertainty involved in projecting from that to the final results is going to get you most of what you need.5 And if you want to be better informed the marginal returns from learning more about policy (or perhaps downballot candidates, or governing processes) are going to be far higher than those from squeezing more out of the polling information that’s out there.
Truth be told, you may know more about the actual policy direction of a second Trump administration than Donald Trump does; outside of a handful of issues, he doesn’t pay much attention to such things and may not realize that the people who he will place in key positions come with a deep policy agenda.
A corollary of this: Don’t read about individual polls. There are plenty of outlets that do a good job of writing up their polls, and plenty that do a less-than-good job. But just as it’s a mistake to spend too much time obsessing about individual polls, it’s overkill to read the write-ups.
I think Nate Silver does good work; I think G. Elliott Morris does good work. But if you read (for example) 538’s explanation of their methodology, I think you can see more than one place where they’re forced to make arbitrary decisions on how to treat the oddball events of this cycle. That’s not their fault, and as I said, I’d rather have the models than not…but I haven’t been following them the way I did in past cycles.
And I should note here that I’m speaking only to ordinary news consumers. How campaigns consume polls is an entirely different topic, as are the ways that the news media can use polls to inform their overall coverage.
Daily? Yeah. No, really, you can stop there. I mean, I can stop there. Probably. Okay, that’s the advise here I’m least likely to follow myself.
This seems very important "It’s also that the late substitution of Kamala Harris for Joe Biden (plus a few other oddities) make it hard to guess whether voters will make up their minds at the same pace they usually do." I'm curious why you prefer not to look at state polls -- like PA or WI?