# OK — So Is Nate Silver a Witch, or Not?

Eugene’s post does raise one of the truly “unforeseen consequences” of this election:  the world of polling, and statistical analysis, will never be the same.  Not to take anything at all away from Silver — who has proven himself to be a very, very smart guy — but the point is: it’s not that he’s some sort of unimaginably brilliant seer who can predict the future, it’s that he has worked out and implemented a new way to aggregate large amounts of polling data and to squeeze out the maximum amount of information possible out of them.  I’m told, by those who know a lot more about this stuff than I do, that he’s not the only one out there in the world of statistical analysis who uses this methodology (and there will surely be a whole lot more of them tomorrow and the day after).

And here’s what’s funny (in a “statistics-is-fun!” kind of way) about it:  If  Romney had won, say, or if  the electoral count  had not eerily and perfectly matched the predictions of Silver’s model, most people would have taken that as a sign that he didn’t really know what he was doing, and his stock, as it were, would have plummeted;  conversely, because Romney didn’t win, and because Silver’s electoral tally predictions were spot on, we’re all thinking, this morning: the guy’s a genius.

But the actual results of the actual election can’t, in and of themselves, offer an adequate test of what he’s doing.  Silver’s algorithms generate probability distributions; he can simulate 10 million elections, all of which produce different results but all of which are consistent with all of the available data.  In 494 of those, Romney got  392 electoral votes; in 173,218 of hem, he got  285; in 873,488 of them he got 2 30; and so on.  Of the 10 million, Obama won (as of the day before the election) 9 million of them – hence his final “prediction” that Obama had a 90.1% chance of winning.

Damned interesting – but it’s not shown to be “correct” when Obama actually wins, and it would not have been shown to be incorrect had he lost.  One-in-ten events happen all the time — once every ten times, to be precise.  Probability distributions are properties of populations, not of single occurrences; that the Giants won the World Series does not mean you were “wrong” when you gave 2-1 odds that they’d lose, for it may well be true that if the teams played 10 million series, the Giants would lose 6.66 million of them.  Obama’s loss would not – could not – have confirmed or falsified Silver’s methods.

So actually, in a nice bit of irony, hejust got lucky!  He had said: when you spin the wheel on Nov 6th, the most likely outcome is:  Obama 313, Romney 225. He put his money (not to mention his reputation) where his mouth was.   We spun, and . . . his number came up!  What a lucky guy!