There are roughly three schools of thought on how to project an election. By far the most popular way is vibes. That’s not Gen Z slang, by the way. Here’s Wall Street Journal columnist Peggy Noonan, former Reagan aide and lukewarm conservative Boomer darling, on the eve of the 2012 Presidential election: “Romney’s crowds are building. He looks happy and grateful. His closing speech has been positive, future-looking, sweetly patriotic. His closing ads are sharp… There is no denying the Republicans have the passion now, the enthusiasm… In Florida a few weeks ago I saw Romney signs, not Obama ones. From Ohio I hear the same… all the vibrations are right.” Thirteen years later, pundits have new ways of assessing these vibrations besides signs (though people definitely still use signs too). They look at their (algorithmically personalized and bubble-reinforcing) twitter feeds and TikTok algorithms to see which candidate people are posting about. They report out which candidate’s canvassers have called them or knocked on their door more. They ask people they know who they are voting for, and then if that feels too unrepresentative, they also ask their cab drivers. They look for indicators everywhere, choose some arbitrarily, use them to craft a narrative, and then derive a prediction from that narrative’s logical conclusion. This method has a very bad track record, but pundits love it anyway, and if they’re honest, their readers do too.
If you want something a little more rigorous than pure vibes, you might consider constructing a fancy polling model. This has many advantages; you can make fancy charts and graphics, you can process and balance tens of thousands of data points seamlessly instead of the three or four that your brain has space for, and best of all, you can spit out a precise win probability for each candidate. People love precise probabilities, it feels so much better to say “seventy point nine percent” than “probably but like, not for sure for sure.” Who cares if your model’s weighting function and house effects are completely arbitrary, the underlying polls use sketchy methodologies, and your error margin is so wide that the model will spit out 50-50 no matter what. Your readers get a number, which they love, and which is completely unfalsifiable after the fact! Events with a 29.1% chance of occurring do occur, you know, 29.1% of the time. In the case of Tuesday’s NYC Mayoral Primary Election, though, this method runs into a few problems. Polling error grows with each candidate, so if a poll of a two-person head-to-head has an error range of 3%, a race with nine candidates will have a much larger one. Ranked choice voting also makes polling incredibly difficult; in order to project where votes will go when a candidate is eliminated, you need a representative sample for each candidate, nine times as many as you’d normally need. And there’s just a sample size problem generally; there have been exactly three high-quality independent polls of this race since April.
There’s a third thing you can try, which is just… know everything? Just have really granular knowledge of every single segment of the electorate, every neighborhood, every assembly district, every ethnic group, every congregation. Have a spreadsheet of how every precinct in every district voted last cycle, but also a mental model of demographic and ideological shift since then built from deep immersion and reporting in all of these places, so that if someone asks you “hey, how much would increased early vote turnout in Bed-Stuy affect Mamdani’s margin there,” or “what if no one votes for Adrienne Adams even in Rochdale Village,” you have an immediate and relatively precise answer, just off the top of your head.
Knowing everything is obviously impossible for national elections, and it’s pretty difficult for local ones too, but luckily, New York City has at least one such knower willing to implement this method to give us some dazzling predictions: Michael Lange, of the substack The Narrative Wars. Lange was exit polling voters outside a polling place on the Upper West Side last weekend when he ran into my parents. “Also ran into @ghostrunnerblog’s parents. Who were lovely!” he generously tweeted, somehow receiving 3 retweets and 27 likes, because he’s absolutely blowing up right now. I took the opportunity to reach out, and he agreed to come on the podcast and chat with me about how he sees this election. You can listen to our conversation here.
Sometime very soon, he’s scheduled to release his official win projections. When we spoke last Monday, he told me he saw Mamdani having about a 33% chance of winning. Since then, a whole bunch of things have happened, and I’m excited to see whether they have revised his chances up or down. Either way, getting to hear his thought process was incredibly enjoyable and educational for me, and I imagine it will be for you too!
The election is tomorrow, at which point this will all of course be completely obsolete, so go listen now!