regression analysis and presidential elections


Via the Gelman social science statistics blog, there’s a nice discussion of the Hibbs peace-bread model. In a nutshell, the incumbent party’s two party vote share is mainly a function of (a) the nation’s recent economic performance and (b) casualties in unilaterally started wars (e.g., Iraq, Korea, Vietnam). Here’s the regression plot.


It’s interesting to speculate on the points way off the regression line: 1996, 2000, 1968, 1952. I think there are two processes at work. First is that there were years where third parties suppressed the challenger vote (1996 – Dole/Perot, 1968 – Humphrey/Wallace,  2000 – Gore/Nader). The second hypothesis is the effect of individual political skill and charisma. The two most extreme outliers came when one of the major candidates (Clinton, Eisenhower) not only benefited from good conditions, but had unusually appealing personnas.

Hibbs also has a web site with a paper with projections for the 2008 election and estimates that the final vote will be about 45% to 48% for the Republican. Not a bad estimate considering how Obama (the current likely nominee) pulls independents and the economy is tanking. It’d be interesting if McCain can beat this spread if Hillary (who has high negatives) is nominated.

Written by fabiorojas

March 18, 2008 at 1:27 am

%d bloggers like this: