The polls are getting downright weird

Could it be caused by racial depolarization?

This is a painting by M.C. Escher called Relativity. It depicts a world which makes no sense, in which the usual rules don’t apply and stairways you thought you could trust lead to nowhere. It’s also how I feel when I look at the polling on the presidential race right now.

If you closely watched the 2016 and 2020 elections, then by now you have a healthy skepticism of polls. On both occasions, surveys predicted that Donald Trump would receive a smaller share of the vote than he actually did. In 2016, that meant he won, whereas in 2020 it just meant Joe Biden won by less than expected. Like a lot of people who witnessed those elections, one of the simple heuristics I apply to interpreting the polls is to just shift them about 4% in Trump’s favor and see what happens. Right now, that would be the difference between Trump losing and Trump winning.

But that doesn’t mean that I’m saying that this is what I expect to happen. It’s just one possible outcome among many. One thing you can count on in any election is polling error, but assuming that it will be the same sort of error as last time is usually a mistake. Circumstances and pollster methodologies change. For instance, in 2020 one pollster I know of encountered a lot of voters who said something like “F*ck you, I’m voting Trump!” and hung up on them. (If you didn’t know, polling itself has become politicized, with many MAGA enthusiasts seeing pollsters as part of “the deep state”). The pollster used to count these people as undecided because they didn’t formally complete the survey questions, whereas now they count them as Trump voters. Changes like this can make polls more accurate.

You also shouldn’t assume that polling error is going to be unidirectional and uniform. Maybe it’ll mess up one way in one state or among one demographic and then in a different way somewhere else. Ultimately, you’re dealing with a lot of uncertainty - and you don’t even know what to be uncertain about.

This is why when I look at the polls right now, I don’t know what to think. Because there is some really weird stuff going on, and it’s not clear if it’s due to polling error or something a more fundamental shift in the electorate.

The first weird thing is that the polling from Pennsylvania has been strangely good for Harris compared to her results elsewhere. Pennsylvania is typically considered to the right of the national electorate, and also to its midwestern counterparts Wisconsin and Michigan. One of the reasons PA is considered the crucial state in this election - the “tipping point state” - is because if Harris wins it, she will almost certainly have also won MI and WI, and hence the election. Or so the conventional wisdom goes.

But in a few polls recently, Harris has been going gangbusters in PA in a way that seems disproportionate to how she’s doing elsewhere. She posted one 6% lead in PA in a poll which had her essentially tied to Trump in Wisconsin. Another found her tied with Trump nationally but leading him by 4% in Pennsylvania. The fact that these results are happening in not just one but across multiple surveys tells us that something is afoot.

Another aspect to this is divergence between the Midwest and the Sunbelt. A Times/Siena poll (the same one that found great results for Harris in PA) has Harris trailing badly in Georgia, North Carolina and Arizona. In the latter, she’s down by 5%. This is, again, weird. You would not expect to Harris to do so much better than expected in Pennsylvania while also doing so much worse than expected in Arizona.

(One of the greatest bugbears of mine in amateur commentary on U.S. elections is to assume that each swing state is its own little discrete universe, and that national polls or national comparisons are irrelevant. American politics is highly nationalized, meaning that trends in one state tend to affect all of them. The different demographic compositions of the states influences how these trends affect them differently, but the national trend matters).

A final piece of the puzzle is some surprising results outside of swing states. A recent poll has Harris surging in Iowa and only 4% behind Trump, which means she’s halved his victory margin since 2020. But in a separate survey in New York, Harris led by only 13%, a near-halving of Biden’s margin in that state in 2020.

So what to make of all of this?

One approach you can take is just to assume we’re dealing with a lot of statistical noise, and focus on the averages to try to get some sense of the real picture. But the fact that multiple surveys have now shown strange results makes me more inclined to think that we’re seeing something unexpected going on. The Midwest really does seem to be leaning more strongly towards Harris than we would expect. Strong results there along with weaker results in place like New York - a place with a lot of voters but no Electoral College impact - is consistent with a pattern in which Democrats’ usual disadvantage in the Electoral College is eroding. In extremis, it makes wacky things seem possible, like Harris losing the popular vote and winning the Electoral College.

These developments also seem broadly consistent with a story of racial depolarization, a situation in which white voters are getting more friendly to the Democrats and black/Hispanic voters are getting more friendly to Republicans. The places Harris is doing well in - like Iowa and Pennsylvania - are whiter than the average, whereas the Sunbelt battlegrounds in which she is struggling are more diverse. This might seem surprising, but it’s consistent with the results in 2020, in which modest racial depolarization also occurred, mostly among Hispanics.

Although it doesn’t quite gel with the common narrative of racial politics in the Trump era, this would actually be an electorally beneficial thing for Harris. There are just a lot more white people in America, and particularly in the swing states, than there are people of color. Small shifts in white support can make a huge difference to the overall result.

Or maybe it’s all just polling error. We’ll have to wait and see.