538's track record is statistically very good. The problem is with applying statistics at all. Odds help us make sense of similar, repeatable events. They do not bear much intelligible meaning for events that are infrequent, and none for ones we view as singular and existential.

A vandal is coming tomorrow to burn down a house in your neighborhood. You are informed that the odds of it being your house have lowered from 30 percent to 20. What should you do with this information? Tweak your insurance and adrenaline levels proportionally?

Election modeling on this view bears the same sort of weight and actionability as someone chiding you for being afraid of airplane crashes or terrorism.

We make decisions about singular events whose stakes are incommensurable with the normal run of daily life all the time, and do so a lot better than behavioral economists think — but not using anything like an expected-value framework. (fixed typo)

If this blows up, here's my Soundcloud (via @daily_barbarian): https://www.thenewatlantis.com/publications/put-not-thy-trust-in-nate-silver

Muting the word "Bayes" and logging off for a while