Ah, yes. Guilty of not following the link, I’ll retract the comment.
The study is interesting. Gelman and Silver are highly respected people, but I’m not sure of real-life applicability of far-in-the-tails estimates. The basic issue is that a situation in which an event in the tails is viable is so different from “normality” that the assumptions of the initial analysis no longer apply.
For example, in the study there is a non-zero probability that California would be tied. The state of the world in which this happens doesn’t resemble the state of the world in which the polls (on which the analysis is based) were conducted.
If you follow the link where I say “one study” you can see a graph by state with values ranging like I said between 10^-7 and 10^-11
Ah, yes. Guilty of not following the link, I’ll retract the comment.
The study is interesting. Gelman and Silver are highly respected people, but I’m not sure of real-life applicability of far-in-the-tails estimates. The basic issue is that a situation in which an event in the tails is viable is so different from “normality” that the assumptions of the initial analysis no longer apply.
For example, in the study there is a non-zero probability that California would be tied. The state of the world in which this happens doesn’t resemble the state of the world in which the polls (on which the analysis is based) were conducted.