Predicting 2012 the Nate Silver way
posted at 2:34 pm on November 5, 2011 by Karl
That title might be unfair, depending on how you view it, as I’ll explain later.
At the NYT, 538’s Nate Silver has set up a model for calculating the odds of various GOP candidates winning the 2012 presidential election. He wrote a long explanation for of the model the NYT magazine, and a short one for his blog. Silver’s model relies on three basic factors: (1) Pres. Obama’s approval ratings a year in advance of the election; (2) GDP growth in 2012; and (3) the ideology of the eventual GOP nominee. Let’s look at each factor in turn.
Approval rating: Silver asserts that a “president’s approval rating toward the end of his third year *** has been a decent (although imperfect) predictor of his chances of victory,” presumably based on his prior research. On the other hand, Gallup maintains that approval ratings at this juncture are not strongly predictive of an incumbent president’s re-election chances, and don’t become predictive until we move well into the election year. Thus, while it would be nice to believe Silver’s findings on approval ratings because they suggest Obama’s odds would be less than one in three, I don’t really buy it. Rather, I think Silver demonstrated the unremarkable theory that a sitting president near 50% has a good chance of reelection because not all the undecideds vote against the incumbent.
The economy: Silver chooses GDP growth, so I’ll have to get a little wonky to explain his thinking. Here’s Silver in the long explanation:
Growth rates during an election year are a good but imperfect indicator of electoral performance. The two times that economic activity actually shrank during an election year, 1980 and 2008, the incumbent party lost badly. The two times that it grew by more than 6 percent, 1944 and 1972, it won overwhelmingly. But Eisenhower won a landslide in 1956 despite tepid 1.8 percent growth, and George W. Bush won in 2004 with only 2.9 percent. The economy grew about 5 percent in 1968, but that wasn’t enough to save Humphrey.
Some political scientists have tried to explain these exceptions by resorting to an alphabet soup of economic indicators, conjuring obscure variables like R.D.P.I.P.C. (real disposable-personal-income per capita), which they claim can predict elections with remarkable accuracy. From the standpoint of responsible forecasting, this is a mistake. The government tracks literally 39,000 economic indicators each year. Although many (say, privately owned housing starts in Alabama) are obscure or redundant, perhaps two or three dozen of them are looked at regularly by economists.
When you have this much data to sort through but only 17 elections since 1944 to test them upon, some indicators will perform superficially better based on chance alone, the statistical equivalent of the lucky monkey from a group of millions who banged out a few Shakespearean phrases on his typewriter. Conversely, indicators like the unemployment rate have historically had almost no correlation with election results despite their self-evident importance. The advantage of looking at G.D.P. is that it represents the broadest overall evaluation of economic activity in the United States.
What’s going on in that passage is Silver’s criticism of the “Bread and Peace” forecasting model created by Douglas Hibbs — criticism that’s a bit overblown. He’s rhetorically over-the-top in that passage because even his own criticism shows that disposable income growth is slightly more predictive than GDP growth. Disposable income growth is not all that esoteric a concept; it’s essentially whether you’re finding more money in your pocket every payday as the election approaches. Conversely, I could abbreviate Silver’s presumed variable as R.P.C.G.D.P.G.L.I.A. — real per capita GDP growth, less inflation, annualized — to make it sound more esoteric than it really is. (Also, if you look at the examples Silver cites as problematic, one might hypothesize that wars had something to do with them — a factor Hibbs accounts for, but Silver does not).
GOP nominee ideology: Silver thinks this factor may help Obama and it may the most, er interesting. From Silver’s blog:
I will have more detail on how the ideology scores are calculated in a subsequent article, but they are based on a combination of three statistical systems: (i) DW-Nominate scores for candidates like Mrs. Bachmann who have been in Congress; (ii) CFscores, developed by the political scientist Adam Bonica, which estimate a candidate’s ideology based on his fund-raising; and (iii) surveys, which have asked voters to assess the ideology of the candidates on a five-point spectrum from very liberal to very conservative.
In the long explantion, Silver notes the “difference between Romney and Perry amounts to about 4 percentage points at the ballot booth.” However, the general consensus among political scientists is that the difference between a moderate and conservative candidate is about 1% or 2%, not 4%. It will also be interesting to see the guts of Silver’s relative rankings of the GOP nominees. Romney, Cain and Perry have not been in Congress and thus do not have easily comparable DW-Nominate scores. Looking at ranking by fundraising, I can show you a June 2011 measurement, based on data from prior campaigns, that shows — as Silver posits — that Perry is to the right of Cain. But that chart also suggests Santorum is to the left of Romney and Mitch Daniels is to the right of Perry. Or I could show you the October 2011 measurement, based on data for this cycle, that places Cain well to the right of Perry and just barely to the left of Bachmann — contra Silver’s assumption. And Silver does not reveal his polling source, so it cannot be evaluated at this time. Silver’s placement of Cain to the left of Perry seems to conveniently match the current poll positions of the NotRomneys, so I would await further explanation.
Indeed, a larger criticism (for now) is that Silver, for all of the explanation in the NYT magazine in the blog, still leaves out details of how he ranked the GOPers and, more importantly, how each of the three factors are weighted. I suppose a hacker could bust open the interactive app at the NYT site, but Silver is someone fond of demanding transparency of others while occasionally opaque himself. In the past, he has eventually come around to transparency, so I would hope Silver is merely dragging out the reveal to provide more content for his blog.
Beyond that, it is disappointing that he chose to have his model predict odds of the various GOPers winning, rather than, say, a share of the two-party vote. That would have allowed easier comparison with other competing models. Given Silver’s approach, it will be easy for him to dismiss a GOP win, even by a more “extreme” candidate, not as a failure of his approach, but a simple case of a candidate “beating the odds.”