We went late into the evening waiting for absentee ballots in Milwaukee County last night. But Milwaukee County executive David Crowley will become the Democratic nominee for governor in Wisconsin, despite it seeming like a certainty that Francesca Hong would be the winner. As of yesterday, every recent poll had shown Hong at least 20 points ahead. The two most recent public polls had her 22 and 25 points ahead. (The most recent poll from Crowley’s own camp showed Hong 19 points ahead!) We’ll come back to the question of why these polls have been so wildly off. But there was a different casualty from the night’s results that I personally enjoyed witnessing because it’s a hobbyhorse or pet peeve of mine — and it seems to be as well of the numbers crunchers/data nerds whose Twitter feeds I watch on elections night. I’m talking here about the “prediction markets.”
An earlier iteration of political prediction markets were a thing back in the aughts. But as they’ve returned in a slicker, more highly capitalized form in the last few years, there’s been a growing argument that they’re not only predictive about political races and possibly as good as polls but actually better than polls. This has always struck me as not only wrong but a category logical error.
Needless to say, after these two races where polls were consistently wrong by double digits, it might seem hard to make an argument that they’re better. But that’s not the argument. The issue is that these supposed wisdom of crowds (or perhaps inside information) data markets are really just emanations of the polls. So what me and my data pals were watching last night was that Kalshi and Polymarket were close to certain that Hong was the winner last night until the first results started coming in, at which point the Hong contract collapsed. But the real reveal was that Hong and Crowley wildly oscillated back and forth over the course of the night. And that only really made sense if you were moving your money on the latest headlines or the topline number on the results sights. In other words, it all pretty clearly showed that the betters (sorry, investors … lol) were buying on the basis of the polls. Once it was clear that the polls were wildly off, they were completely adrift. The markets “knew” they were very wrong and were starved of data for how to get right.
The oscillating back and forth was, again, the tell. The reason I mostly ignore the topline results and certainly the media accounts on election nights and focus on these number crunchers is that if you are good with numbers and have access to historical and regional data you can get a good sense of what’s happening very early. So for instance, Candidate A may be up by 20 points in initial results early in the evening. But you might be able to see that in key districts or counties that have most of their votes in, Candidate A simply isn’t getting the numbers they need. Watching the oscillations of the betting markets last night, it was clear the big traders weren’t even making much use of this somewhat technical but hardly secret information. Once they lost the polling signal, they didn’t seem able to get access to the available results signal that was there for the taking.
Needless to say, this doesn’t mean these markets don’t have some predictive value. To the extent they are emanations of the polls, it’s not the polls speak with one voice — though they pretty clearly were here and last week in Michigan. They can be a good summary of what the polls are saying, if the wisdom of crowds thing is working. We should also always been adding contextual data to the polls. You’d figure in that, whatever the polls say in the Texas Senate race, Texas is a very hard nut for Democrats to crack. So that gets figured into the predictions even if it’s not explicitly in the polls. There are also lightly polled or non-polled races where you want, at least in theory, the wisdom of crowds to intelligently add in non-polling factors. It’s not a matter of one being better or worse. It’s about reliance. It’s a bit like asking whether it’s your car or the gas it runs on that is better at getting you to work in the morning.
So, valuable? Sure, though I guess in the very marginal sense of telling you what you’re going to learn anyway by the end of election night. But the idea that they are better than polls — have some ability to replace them as our way of establishing probabilities for the electoral future — is not so much wrong as a kind of category error since they are overwhelmingly just a reflection of those polls. And often not even a very expert one at that.
