Wednesday, November 07, 2012

Don't mess with the nerds

By Frank Moraes

Monday evening, Nate Silver released his final predictions for the presidential race. The most remarkable aspect of the prediction was that Florida, which two weeks earlier had been solidly in Romney's favor, slipped into a slight lean towards Obama. How slight? Silver predicted that Obama would win Florida by 0.022 percentage points. Earlier that day, I predicted that Florida would go to Obama, based upon the clear trend that had been evident the previous two weeks. Regardless, I predicted even odds that Obama would get 332, 303, or 294 electoral votes. Right now, we aren't sure which way Florida will go -- just as Silver predicted. But we do know that Obama won re-election with either 332 or 303 electoral votes.

I don't bring this up to show how smart I am. In general, all I did was follow Nate Silver's work and understand it more than most people. For example, I heard a lot of people claim that they thought Obama would win 313 electoral votes because that was the average prediction that Silver released. But that number was extremely unlikely. What was much more helpful was the electoral vote distribution that indicated that Obama had a 20% chance of getting 332 votes, 16% chance of getting 303 votes, and 13% chance of getting 345 votes. Here it is:



As you may recall, Dean Chambers of Unskewed Polls, predicted a 359-179 vote Obama loss. I went over to his website this morning and it is like a ghost town. But I noted that he had updated his prediction. As of Monday, he had Romney winning by the much less ridiculous margin of 275-263. What does that mean? It means in the end he abandoned his claim that the polls were skewed and went along with them, but gave the benefit of the doubt to Romney in Ohio and Florida.

Needless to say, this is not science; this is partisanship. We saw the same thing throughout the right wing echo chamber. George Will predicted Romney would win 321 votes. Dick Morris predicted 325 votes. What was most interesting about these conservative predictions was how extreme they were. While liberals were mostly predicting modest Obama wins (290 and 294 were the most common), conservatives were all predicting blow outs. The reason for this seems to be that conservatives had to assume that the polls were systematically wrong. In order to predict an Ohio win for Romney, they had to predict that a number of other unlikely states would go for him too. Thus the 321, 325, and 359 predictions.

It isn't just Nate Silver who did well last night. Drew Linzer of Votamatic had been predicting a 332-206 Obama win for weeks. What all this means, of course, is that as always the polls were right. The models were better than the straight polling averages because they took parameters like trends into account. Real Clear Politics, for example, had Romney up in Florida by 1.5% on Monday. Even still, a simple poll average was enough to predict 49 out of 50 states. There were no surprises last night in the presidential race.

The bottom line here is: don't mess with the nerds. They may be boring. They may be unfathomable. But they have special Nerd Powers. You'd be foolish to play poker with them. And masochistic to publicly challenge their splendid statistical models of the real world.
 

(Cross-posted at Frankly Curious.)

Labels: , , , , , , , , ,

Bookmark and Share

0 Comments:

Post a Comment

<< Home