Thursday, November 1, 2012

"President Obama is in electoral trouble"

In August, I saw a reference to an election forecasting model by Michael Berry and Kenneth Bickers (University of Colorado) that predicted a Romney win based on state unemployment data.  I couldn't find the full paper at the time, but after seeing another reference, I looked again and found it.  Their assessment is given in the quotation that provides my title.  The predict Obama will win only 218 electoral votes (with new data, they've revised that down to 208) and 47% of the popular vote.  It turns out that the state-level data doesn't have that much to do with the forecast--the important variable is the national unemployment rate.  According to their estimates, high unemployment costs a Democratic incumbent a lot of votes, which is why things look bad for Obama.  That's right, just Democratic incumbents--for Republican incumbents, the national unemployment rate doesn't matter.  Why do they get this result? The state-level data go back only to 1980, so the eight elections from 1980 to 2008 are the only ones they can use to estimate the model.  In that period, there was one Democrat who ran for re-election with a high unemployment rate (Jimmy Carter in 1980), and one Republican who ran for re-election with a high unemployment rate (Ronald Reagan in 1984).  Carter lost, and Reagan won big.  When you put it that way, the "evidence" that the national unemployment rate has different effects for Democratic and Republican presidents is really just one example.  

But since Berry and Bickers use state data (plus DC), they have fifty-one cases in which a Democrat ran for re-election with a high national unemployment rate, and fifty-one in which a Republican ran for re-election with a high national unemployment rate, which seems like a lot of evidence.  The problem is that the cases aren't independent--to a large extent, they're just duplicates of two cases (Carter and Reagan).  On some level, this is a fairly sophisticated statistical issue.  But on another level, it's pretty simple--really surprising results (Republican incumbents aren't hurt by high unemployment) usually mean you did something wrong. 

No comments:

Post a Comment