The state of play in the 7 states that will decide the election
Plus, re-examining the possibility of an Electoral College split favoring Harris.
If nobody released another national poll between now and the election, I wouldn’t care one bit.
Seriously. National polls have shown Donald Trump gaining ground lately — Kamala Harris’s lead in our national polling average is down to just 1.2 points. But the popular vote is merely a beauty contest. If you know the winners in the seven major battleground states — Pennsylvania, Michigan, Wisconsin, Georgia, North Carolina, Arizona and Nevada — you can correctly identify the Electoral College winner about 99 percent of the time. So those national polls are only interesting to the extent they help us anticipate trends in state polls — and at this late point in the race, it’s better just to look at state polls directly.
So that’s what today’s Model Talk is about. Our last tour of swing state polling was an incredibly popular feature, so we’re warming up the bus to head on the road again. Except this time, we’re skipping Florida, which has drifted away from Harris to the point it will only matter in the event of some rather unlikely polling errors.
First, though, I wanted to take another look at our model’s conclusion that Harris is very unlikely to win the Electoral College while losing the popular vote. Obviously, a split the other way around is much more likely: Donald Trump won the Electoral College despite his loss in the popular vote in 2016, and in 2020 the Electoral College / popular vote gap was even wider — nearly 4 points in Trump’s favor, although not quite enough for Trump to win given Joe Biden’s robust margin in the popular vote.
At the same time, there’s been a bit of weirdness in state versus national polls lately. Over the past week, our national polling average has drifted toward Trump by a further 1.1 percentage points. But in the aforementioned seven swing states, the shift has been smaller: an average shift of 0.3 points toward Trump:
Furthermore, as we described this week, likely voter screens have tended to have opposite effects in state and national polls, hurting Harris in the former while helping her in the later. Many pollsters seem to lack confidence in their results this cycle: there may be some herding where it's safer to show a tie or Harris +1 or Trump +1 instead of taking the risk of going out on a limb. And at least one high-quality set of surveys, from the New York Times, have consistently suggested that Harris could win through the Blue Wall states (Michigan, Wisconsin, Pennsylvania) despite having mediocre numbers for her in their national polls.
Even given all this uncertainty, our model has generally shown only about an 0.3 percent chance of Harris winning the Electoral College while losing the popular vote. That’s not zero, but it’s obviously unlikely — about 300:1 odds against. I’ve written before that I’d get a bet down on those odds if I could.
So maybe there’s something structural about the model that’s been understating this possibility?
Well, perhaps. In the 40,000 simulations that we run every day, the model considers the possibility of a wide range of polling errors correlated with various demographic, geographic and political variables For instance, in some simulations, Harris overperforms among Hispanic voters, which could lead her to winning swing states like Arizona and Nevada. In other cases, Trump overperforms in the Midwest, devastating Harris’s Blue Wall pathway (this is basically what happened to Hillary Clinton in 2016).
In other simulations, one candidate overperforms in blue states or red states — which won’t affect the Electoral College but would impact the popular vote. The way we had things set up, though, there wasn’t really a way for one candidate to systematically overperform in purple states even though there are clearly realistic mechanisms where this could happen: for instance, if their ground game is more effective. So I made a very subtle change (as described in the footnotes1) to how the model accounts for state correlations to provide for this possibility. An Electoral College split having Harris remains very unlikely, but it increased to an 0.7 percent chance in today’s forecast of 0.2 or 0.3 percent as in recent simulations.
I don’t love making any changes to the model in midstream, but I think this change is pretty clearly correct: it describes a realistic scenario for polling error, and the whole point of running all those simulations is articulating the various ways in which the polls could go wrong. And over the long run, the effect should be neutral: it will slightly increase the chance of an Electoral College / popular divide favoring either Harris or Trump, or other Democrats and Republicans in future races.
All right then, let’s embark on our swing state tour. Last time, I started with what was then Harris’s best swing state, Michigan. As much as I’d like to mix things up by going Trump-to-Harris instead of Harris-to-Trump, I’m going to stick with this itinerary because I think it’s the most intuitive way to run through the circuit. If Harris wins the three Blue Wall states, she probably (although not certainly2) wins. If Trump does, it doesn’t technically clinch the White House for him, but it probably does for all practical purposes — Harris would have to sweep all the other battlegrounds. If they’re split, there are more good scenarios for Trump than bad ones, but the map would get very interesting.
Michigan: Harris win probability 56%. Tipping-point probability 15%.
Here’s why Democrats are understandably feeling insecure. Michigan remains Harris’s best swing state, and yet it's only barely better than a coin flip for her.
With that said, Michigan is traditionally the bluest of the Blue Wall states, and it’s one where some of the higher-quality pollsters have (with the exception of highly-rated AtlasIntel) shown Harris with some semblance of a real lead instead of a tie. It’s also had some of the more consistent polling throughout the cycle: Harris emerged with a lead in Michigan almost immediately after replacing Biden and has never relinquished it in our polling average.
Polls show Arab American voters, typically a Democratic group, instead splitting their votes about evenly between Harris and Trump. (Undoubtedly this reflects their critiques of Biden/Harris on the Israel-Palestine war.) That’s more of an issue in Michigan which has the country’s largest share of Arab American residents.