The 128 paths to the White House

From the Sun Belt to the Rust Belt, which maps are most likely?

Let’s try something a little different today: basically, a two-for-one. First, I’ll give you what’s essentially a full newsletter with some big-picture thoughts on where the presidential race stands and how to think about the possibility of a polling error. That part will run free for everyone.

Then we’ll get into the proper Model Talk portion of the newsletter, which I hope will give you insight into how the model thinks about the various plausible permutations of the map. The seven major swing states — for purposes of this column, I won’t get cute and count Florida as one — produce a total of 128 possible combinations (2 to the 7th power). Some of these are much more likely than others, however. The most likely combination — want to guess which one it was? — came up 15,273 times in the 70,000 simulations¹ we ran on Sunday morning. The least likely combinations only came up twice. Why are some so much more plausible? And how much can we tell from these seven states alone — do other states like Florida or New Hampshire or even exotic ones like Alaska ever really matter?

Anyway, I hope you’ll consider subscribing, whether to the free or the paid programs. We’ll be updating the model basically every day during these final 38 days of the campaign and publishing 4 to 5 newsletters per week, about half free and half paywalled. Most of them are quite meaty, like this one — we promise not to clog your inbox, although we may sneak in a non-politics newsletter here and there. (Don’t be surprised to see something about ESPN’s decision to fire my favorite NBA writer.) And we have ambitious plans for the post-election period, too.

The 5 ways the polls could go wrong

When I get cornered by strangers — or even by friends — I frequently get asked to provide my “real” prediction of the race. C’mon, Nate, Harris or Trump? I point out that our forecasts are probabilistic, which usually doesn’t help. But that’s really my best answer: as of this morning, for instance, Harris has a 56 percent chance of winning the Electoral College and Trump has a 44 percent chance.

Sometimes, there are material factors (like Joe Biden’s deteriorating public performances before he quit the race) that I don’t think the model is doing a good job of accounting for. When that’s the case, I’ll say as much. But there aren’t any such factors now, so those infernal probabilities are as well as I’m going to be able to do.

I’ll grant you that a 56/44 forecast is in kind of a weird, uncomfortable zone. To be clear, the forecast is that Harris will win the Electoral College 56 percent of the time — not that she’ll beat Trump by 12 points — in other words that the race is basically a toss-up. There is a strong likelihood that this will be a close election right up through Election Day.

Still, 56 percent is meaningfully different from 50 percent. Even a somewhat nitty (i.e. risk-averse) poker player would usually take an all-in spot where they win 56 percent of the time. A baseball team that wins 56 percent of its games goes 91-71, and that’s considered a strong season. So zooming out, you’d probably take this position if you were Kamala Harris. It certainly has Democrats in a much better situation than they ever had with Biden — and it should have Republicans wondering about a series of mistakes made by Donald Trump. It’s been almost three weeks since the debate, and the polls are holding steady — we haven’t seen the sort of slow deterioration for Harris in the numbers that we did following the DNC. 

But if you’re in the market for certainty? Sorry, but you’re not going to get it, at least not here at Silver Bulletin. Anybody who’s projecting this race with a great deal of confidence is just selling you hopium or snake oil. 

Some of the uncertainty is because there’s still some time left in the campaign for October Surprises. But especially with no more presidential debates scheduled, most of it is just because the polls are really close, will probably remain close, and are frequently in error to some degree even when they aren’t that close. So if it helps, let’s think about the various ways the polls could go wrong and what that means for the odds.

Our polling averages show all seven battlegrounds polling within 2 percentage points:

But if you squint, you can perceive a bifurcation between the Rust Belt or “Blue Wall” states of Pennsylvania, Michigan and Wisconsin, where Harris has been ahead in most of the recent surveys, and the Sun Belt trio of Arizona, Georgia and North Carolina, where Trump has usually had leads. The exception to the pattern is hot-and-sunny Nevada, which has polled strongly for Harris lately after having been a struggle for Biden earlier this year — and North Carolina has polled like a pure toss-up lately, too. These nuances, along with the fact the Blue Wall states are usually sufficient to get Harris to 270 electoral votes on their own, are why you’d slightly rather have her hand to play.

Nonetheless, the main storyline in the polling is that there’s some degree of racial depolarization. Harris is polling slightly better with white voters than Biden’s finish four years ago, but worse with Black, Hispanic and Asian American voters. That’s why polls narrowly project Trump to win Georgia, for instance, but Harris to expand on Biden’s margins in places with lots of white voters like Wisconsin or the 2nd Congressional District of Nebraska. Let’s go through five basic scenarios for polling accuracy:

1. The polls are mostly quite accurate. This does not mean they’ll be perfectly accurate, because they won’t be. But take this to imply that they’ll be in the upper half of the accuracy distribution by recent standards — not a big miss like in 2020. If the polls are basically accurate, then you’d rather be Harris. She’s ahead in the polls in states totaling 276 electoral votes, and that’s counting North Carolina in the Trump column when maybe we shouldn’t. But she’s still just barely ahead. A narrow Trump win — let’s say he holds all the states where he’s currently leading, but wins by a point in Pennsylvania — would be perfectly consistent with the recent polls and wouldn’t require any sort of systematic polling error.

2. The polls are accurate in the Sun Belt but underestimate Trump in the Rust Belt. This is Game Over for Harris. Maybe she holds onto Michigan, but Trump sweeps the rest of the swing states, and if the map really gets away from her, she could also be at risk of losing states like New Hampshire.

3. The polls are accurate in the Rust Belt but underestimate Harris in the Sun Belt. This is Game Over for Trump. Maybe he holds onto Arizona or Georgia, but Harris sweeps the rest of the swing states, and if the map really gets away from him, he could also be at risk of losing states like Florida.

4. The Rust Belt / Sun Belt gap is even wider than polls show. In other words, the polls are correct about racial depolarization but aren’t going far enough. This is basically the story that recent New York Times polls have been telling, for instance. They have Harris winning the Rust Belt states but losing Georgia, North Carolina and Arizona. This is the scenario where that exact 270 number looms really large, since the path through the Rust Belt remains intact for Harris on this map. It could even be the scenario where the Electoral College flips and favors Democrats — or more realistically, where Harris can get away with winning the popular vote by as little as half a percentage point.

5. Polls underestimate Trump in the Rust Belt but underestimate Harris in the Sun Belt. In other words, the racial depolarization narrative is a bust. It’s not crazy. There are years like 2022 when the polls don’t have much of an overall bias, but there are weird regional fluctuations. And some pollsters like AtlasIntel do show this map — their recent set of polls was mostly pretty bad for Harris, but surprisingly showed her best states as North Carolina and Nevada. In this scenario, Harris probably wins Nevada, but the other six swing states are really, really close, possibly recount close. If that sounds familiar, it’s because this looks a lot like the 2020 map, where the margin in five swing states was within a point and a half. But that map would make me a little nervous if I were Harris just because even though Biden won, 2020 featured a huge Electoral College / popular vote gap of nearly 4 points. If she replicated Biden’s 4.5-point popular vote win, she’d probably be fine, but any closer than that and it gets dicey.

This is basically how the model “thinks” about the race. It recognizes the possibility of polling errors, and it further expects those errors to be correlated to some degree — perhaps an across-the-board miss, but more likely in some regional clusters: Michigan and Wisconsin will tend to move together, as will Georgia and North Carolina. For paid subscribers, though, let’s go even deeper inside the model’s logic.

The most and least likely swing state scenarios

At the top of the column, I asked you to guess the model’s most frequently occurring combination involving the seven swing states. If you haven’t yet, this is your last chance. Ready?