Earlier this month I wrote a piece entitled Friends Don’t Let Friends Watch the Early Vote. I wanted to re-up that recommendation, notwithstanding the fact that I have kept an eye on these findings, as I always do. The simple fact is that all of these efforts amount to figuring out how big a share of the pie you have before you know how big the pie is. That doesn’t mean you can’t glean anything by analyzing these things or that the knowledgable people doing it are silly or mendacious. They’re not. It’s just that you can’t draw any firm conclusions from it and our desire to see what we want in these entrails greatly outweighs the hints we can get from them. As I noted before, this is especially the case because the partisan dynamics of who votes early, in person, by mail, on Election Day, etc., is in so much flux.

There are lots of arguments about “prediction markets.” Some of the most important are about the opportunities they create for corruption and the general social ills tied to what is essentially gambling. But the one that perpetually puts a bee in my bonnet is the frequently-advanced idea that prediction markets are superior to polls as predictors of election outcomes. I was reminded of this by the upset Lula win in yesterday’s Brazilian presidential election. Kalshi had candidate Flávio Bolsonaro with 76% odds of winning until the first returns came in and he lost. Obviously, polls and every other kind of prognostication have misses too. Those don’t discredit the whole enterprise. The problem with the argument about prediction markets and polls is both empirical and logical. Empirically they do not appear to be better. But the comparison is difficult if not impossible since polls don’t generate odds or probabilities. They approximate current levels of support. It’s easy to look fast and think those two things are the same. But they’re not. Logically, prediction markets cannot be “better” than polls since they are mostly derivative of polls; they are overwhelmingly reliant on polls.