Forecasters: New Hampshire predicts, Perry could win, and the electoral college may or may not be biased
By Ezra Klein,
I spent the afternoon at a political forecasting panel. The big takeaway, I think, is that you shouldn’t believe any forecasts yet. What matters is what happens in the election year, and so anything that’s using the president’s current poll numbers, or the current state of the economy, is basically useless, and any model that’s trying to predict the president’s poll numbers next year, or the state of the economy right before the election, is opening the door to so much error that it, too, is basically useless.
But the panel at the American Political Science Association conference did have some interesting papers making more limited claims. Among them:
New Hampshire is the most predictive primary: Helmut Norpoth and Michael Bednarczul presented a study looking at which state’s primary did the best job predicting the eventual results of the presidential election. The answer? New Hampshire, which was second only to Massachusetts in predicting the vote total for the incumbent party, and which easily beat every other state in predicting the vote total for the opposition party.
Surprisingly, you couldn’t improve on New Hampshire’s predictive ability by adding any other primary or caucus on top of it. The authors had a few hypotheses for why new Hampshire so easily bested the competition, but the only one that really struck me as convincing was that the state’s primary includes independents. This would also help explain why New Hampshire performed better for the out-of-power party, as the independents often choose to vote in the more contested of the two party’s primaries.
Rick Perry might be the only Republican who can beat Barack Obama: That was the conclusion of a paper by Andreas Grafe and Scott Armstrong that tried to identify 58 biographical factors that can predict presidential success. These range from height and weight to whether a candidate has published a book, lost a sibling, and been blessed with a popular first name. This model, the authors say, returns the right result for 27 of the past 29 elections, and it suggests that the only Republican candidate who is likely to beat Barack Obama is Rick Perry. The panel, however, didn’t seem particularly impressed with this paper, and noted that it was built to fit past elections and has not had any actual success predicting real elections. If you look at the paper, you’ll see that Mitt Romney is barely ranked as a stronger candidate than Michele Bachmann, and that’s enough for me to pretty much discount this model totally.
Is the electoral college biased towards the Republicans or the Democrats? Brian Gaines and Neil Thomas Bear tried to take a rigorous look at this and basically found that there have been so few elections in which the electoral college was even close to mattering that it was almost impossible to tell. Democrats, they said, clearly think the electoral college hurts them -- the National Popular Vote initiative, which would undermine the electoral college, has had very little luck in Republican states and quite a bit of luck in Democratic states -- but that might just be because the 2000 election looms so large in their minds.