I'm confused by the existence of this article. Comparing the accuracy of two probabilistic predictors is pretty much a solved problem. (At least in situations w

I'm confused by the existence of this article. Comparing the accuracy of two probabilistic predictors is pretty much a solved problem. (At least in situations where the two predictors have both made predictions about a fairly large number of events in the same set, S.) Just add up the log scores for both predictors and see which does better. [1] [1] https://en.wikipedia.org/wiki/Scoring_rule#Logarithmic_scoring_rule https://en.wikipedia.org/wiki/Scoring_rule#Logarithmic_scori... Another thought: Although the author didn't lie about its meaning, I thought the graph "Stated Probabilities Compared with Average Portions" was visually misleading. At a casual glance, it appeared that 538 was all over the place with its senate predictions. But if you look closer, you can see that many of the red data points are at 0 or 0.5 or 1. This suggests that these data points represent the outcome of just 1 or 2 elections, which isn't really enough to say that 538 is doing particularly badly.