Tuesday, 30 October 2012

Questions Answered

Shorter Dylan Byers: Nate Silver offers multiple predictions with appropriate caveats and explanation of uncertainty.  The only comment on the only prediction I was able to concentrate on doesn't agree with my gut.  Now a word from people with no statistical training who agree with me.

Shorter Byers next week: why do liberals always think they're smarter than me?


Tomsk said...

That's so bad it almost reads like a parody of all the other stupid attacks on 538 and other poll prediction sites.

Nevertheless, there is an interesting psychological point lurking somewhere beneath the nonsense. I think a lot of people are looking at the 75% (or whatever) chance of Obama winning and thinking that means Obama is miles ahead, because they're subconsciously confusing the predicted chance of winning with the predicted share of the vote, which is clearly almost a dead heat. Among the many depressing consequences of a Romney win would be the cries of "538 was wrong" even if the result is well within the margins of error given on the site.

Having said that, Silver did completely screw up his 2010 UK election prediction (managing to do worse than a UNS model), so his record is not quite as perfect as it's made out to be, although he did say his method in that case was an experiment and at least he conclusively proved that there isn't enough polling data in the UK to ever do significantly better than UNS.

SpaceSquid said...

I hadn't considered the idea of people assuming a direct translation from probability to vote ratio, but that's an excellent point.

The fundamental problem here seems to be the old issue of people not understanding the difference between a model being wrong, and a model giving the greatest probability to an event that doesn't ultimately happen.

Byers is light years away from being able to judge whether the former is the case, of course, so he has no choice but to pretend the two are identical.

Well, he had a choice to shut the hell up. He could have chosen that.