A random number generator determined the “favorite" in our forecast
So fine: let's go to the tiebreaker round.
When I say the odds in this year’s presidential race are about as close as you can possibly get to 50/50, I’m not exaggerating.
At exactly midnight on Tuesday, I pressed the “go” button for the final time on our election model this year. I knew it was going to be close. I felt like I was spinning a roulette wheel. (Appropriate, I guess, in a year when I published a book about gambling.) We’d decided ahead of time to run 80,000 simulations instead of our usual 40K.
And after 80,000 simulations, Kamala Harris won the Electoral College in … 40,012 of them, or 50.015 percent. The remaining 39,988 were split between Trump (39,718) and no majority — a 269-269 tie — which practically speaking would probably be resolved for Trump in the U.S. House.
Here’s how it went down:
Harris jumped out to a huge early lead, ahead 50.7%-49.3% after the 18,000th simulation — but then Trump + no majority mounted a thrilling comeback. But on simulation #79,281, Harris went on a winning streak, claiming 15 of the next 17 simulations to turn a 5-sim deficit into a 8-sim lead and never looking back. Trump closed to within single digits again as late as simulation #79,603, but couldn’t seal the deal.
Obviously, this is quite ridiculous. If I’d closed out my 37 browser tabs and left my computer running all night, I have no idea who would have “won”. We don’t talk about it much, but there’s a small amount of error introduced into the model because it’s probabilistic rather than deterministic. The margin of error on 40,000 simulations is about ± 0.5 points of win probability for either candidate; after 80,000, it declines, but only to ± 0.35.
But I guarantee you: there are literally going to be people who say, “NATE SILVER PREDICTS A HARRIS WIN” as a result of this. Literally.
It’s not because my default is to hedge or just throw some extra uncertainty parameters in the model for no reason. This is my fifth presidential election — and my ninth general election overall, counting midterms — and there has never been anything like this. In 2008, 2012, and 2020, Democratic candidates were clear Electoral College favorites in our model, and I spent a lot of time arguing with people who claimed they weren’t. Hillary Clinton was a slightly less clear favorite in 2016, and our model was bullish on Donald Trump relative to betting odds and other forecasts — but a favorite she was. I suppose I’ll never get over that one.
In 2010, we correctly had Democrats favored to keep the Senate (where they started with a huge majority) but lose the House. 2014 was a boring and predictable Republican hold. 2018 was the inverse of 2010, with Democrats as heavy favorites in the House but big underdogs in the Senate. In 2022, our final Senate forecast was close1 — though not quite so close as this one — but the House wasn’t particularly so.
Even this year, the model definitively showed Joe Biden as an underdog, and the post I wrote introducing in June it emphasized that point: Democrats were trying to convince themselves the race was a toss-up when it wasn’t. But Democrats did the smart thing and shoved Biden aside. And now they’re rewarded by being able to determine the future of the country by means of … a coin flip?
Actually, it won’t be a coin flip. You will determine it with your votes. I’ll sometimes literally flip a coin to decide where to go out to dinner — heads for Italian, tails for Thai. But who becomes the next president is a question of, uh, slightly more importance. I’m not going to give you a speech about voting, but if you care about the outcome and haven’t, you probably should.
From the model’s standpoint, though, the race is literally closer than a coin flip: empirically, heads wins 50.5 percent of the time, more than Harris’s 50.015 percent.
If you’ve been regularly reading this newsletter — and if not, there’s still time to subscribe, and we have some ambitious post-election plans! — I hope you’ll feel like we’ve had the bases covered. Eli and I were somehow able to publish pretty much every story we wanted to, except the one that got away about how Allan Lichtman’s 13 Keys to the White House, if followed strictly based on how he’s applied them in the past, actually clearly predict a Trump victory. (Now that I’m on Team Harris by 12 simulations, I suppose I shouldn’t. There’s a one-paragraph verison in the footnotes2; for the record, Lichtman claims the keys support Harris.)
We’ve written about how the close polls don’t necessarily predict a close result. How there could be another systematic polling error favoring either Trump or Harris that turns Ann Selzer into a hero or a goat — and about how other gutless pollsters have basically just given up and decided to copy off the students who did their homework
We’ve given you two tours of the 7 swing states and guided you through the 128 paths to victory. We even took a detour to Alaska at one point.
Those stories hold up well, because this election has ultimately been stable — if uncertain — after a tumultuous beginning. People go nuts over small swings in the forecast, but the model has spent 87 of 98 days since we relaunched in the 60/40 range in one direction or another.
There are no Daily Doubles on the electoral map — or if there are, you don’t know where they’re placed in advance. In 6 of the 7 key swing states, our final polling average is within 1.2 points; the only one with something resembling a clear lead is Arizona for Trump. But even there, some polls show Harris leading amidst evidence of a late rebound for her with Latino voters. And Harris has closed comparatively strongly in Wisconsin and Michigan. Here’s how it all translates on the map:
So if you offered me a bet on the election, who would I pick?
Well, I wouldn’t. And not because I don’t like gambling. But at the odds sportsbooks typically offer, you have to win 52.5 percent of the time to have a +EV bet. And Harris’s “lead” is smaller than that. I emphatically don’t think you should trust your “gut” when it comes to elections. I also don’t really believe in indicators like early voting or rumors of internal polls that people tend to overweight.
But what if you offered me a free bet at 50/50? If I’m right, you buy me a nice sushi dinner. If I’m wrong, you razz me a bit?