LogFAQs > #934396407

And this is really relevant because this is analogous to 538 in 2012 - should their model be considered worse because it called more states correctly than the model itself would have expected itself to call? And I think the answer is no.

There is something being changed between the statements "Hillary has a 70% chance of winning" and "hold 100 elections, and Hillary wins 70 of them on average." I think that is where the error slips in that lead to the absurd results we're seeing in the thought experiments. Namely, the 2nd statement here subtly implies that seeing the outcome of election 1 doesn't change the probability (in the eyes of the observer) of the other 99 elections. It does change, because the 100 elections are highly correlated events. 538 gets this correct, btw.

---

September 1, 2003; November 4, 2007; September 2, 2013
Congratulations to DP Oblivion in the Guru Contest!