Predictions and Results

November 7, 2012
Posted by Jay Livingston

“Well, Howard, my predictin’ days is over,” said Muhammed Ali, even though the champ had been fairly accurate at predicting how many rounds an opponent would last. 

My post yesterday quoted Peggy Noonan predicting a Romney win.  On Fox last night, when one of the anchors asked Noonan if she were “surprised” by the results, she dodged the question and said only that she was disappointed.  Are her predicting days over?  Unlikely.

I contrasted Noonan’s “all the vibrations are right” methodology with Nate Silver’s thorough and complicated Bayesian model.  Conservatives accused Silver of liberal bias (and effeminacy), and they offered their own unbiased, clear-eyed  predictions. 

How did the prognosticators do?  Well, if somebody had bet on the Bayes, they would have cleaned up.  Here’s a scorecard.  I filled in some of the blanks in the grid that Neil Sihnababu posted at VoteSeeing .  Right now, the electoral vote stands at 303-206, with Florida not yet called, though most experts have put it in the Obama column.  Assuming that Florida is blue, Silver’s electoral prediction is perfect.  As for the battleground states, they all went for Obama (again, assuming Florida).  So did Nevada, adding one extra wrong prediction to Steve Forbes’s clear-eyed predictions.

(Click on the table for a larger view.)

You’d think that Karl Rove, George Will, and the others who did about as well as P’lod would follow Ali’s example.  Or at least, they would adjust their models, if they had any.  Or they would stop scoffing at science when it doesn’t tell them what they want to hear.  But don’t bet on it.

UPDATE:  Gabriel Rossman dresses up in his St. Paul costume (left over from Halloween, presumably) to deliver the same message. [Note for non-stats people who click on the link to Gabriel’s blog:  CLT = Central Limit Theorem.]

1 comment:

maxliving said...

The three most recent xkcd's have all been on a statistics/polling theme:,,