Nate Silver’s newest critique of presidential election forecasting models has been making the rounds. He was kind enough to publish my response to his critique late last week while I was traveling, so I want to highlight it now. The essence of my response is this:
* Undoubtedly these forecasting models could be improved in various ways. I agree with several of Nate’s specific criticisms. (Thus, contra Jon Bernstein, I don’t think I’m “giving the models a pass.”)
* The models — despite their limitations — rarely predict the wrong winner, so the lay consumer of these models will not be grossly misled most of the time. By lay consumer, I mean someone interested enough in politics to care who wins the presidential election, but not much interested in “mean squared error” and other commonplaces of the “nerdfight.”
* Most political science knowledge about the economy’s role in elections, the effect of campaigns, and voter behavior does not come from forecasting models.
* There is no easy way to separate the effects of the economy and the campaign — a point I made previously.
Although I am not a forecaster, I’ve been dabbling with a simple forecasting model as part of the book on the 2012 election. I will have a bit more to say about this subject in the weeks ahead.