Saturday, February 9, 2008

Obama Election Regression Results

Update: has a discussion of Poblano's model as well as other regression results.

Poblano at Daily Kos has a nice empirical analysis of the two way vote share between Hillary and Obama using multivariate regression.

Not surprisingly, some key variables include: caucus vs primary, African American share, youth share, fraction with college degrees, fundraising, southern baptists (proxy for Southerness).

Interestingly, the Latino share does not matter once one conditions on the other variables (e.g. college). What I like about this analysis is that it 1. goes deeper than set of cross tabs on exit polls can by conditioning on multiple variables 2. aggregates info across states (Latinos in CA vs NM) and 3 can be used as a forecasting tool.

It suggests that had Obama campaigned in Florida he would have narrowed Hillary's lead by about 7 percentage points. It also forecasted a much closer race in WA than it appears to have been. Interestingly it suggests that Obama has a good chance at winning Ohio (by 10 pts) but losing Texas by 10.

Interesting stuff!

(btw, its nice to be on board!)