Friday, February 15, 2008

Accuracy of Democratic Superdelegate Predictions

A week ago, I used statistical models to generate predictions for how likely each unpledged superdelegate was to vote for Hillary Clinton or Barack Obama. You can find the estimates here. Since I produced those estimates, 29 superdelegates (for whom we generated predictions) have pledged their support for one candidate or the other. Thus, we have our first chance to see how well our model does at predicting how unpledged superdelegates might vote. You can find that information in this table:


If you do the math, you'll see that the model correctly predicted 65.5% of the delegate decisions correctly. Thus, the model got nearly two out of every three superdelegates correct.

When the model did miss, it tended to miss on the the side of Obama. Only twice did I predict a superdelegate would support Clinton when he or she actually support Obama. The other 8 mistakes were in predicting support for Obama that actually went to Clinton.

I will use these new endorsements, along with information about superdelegates who have changed their votes and information about additional superdelegates that I initially had to leave out of the model to generate new predictions over the weekend. Hopefully with this additional model, I will be able to provide an even more accurate picture of who the unpledged delegates are likely to support.

No comments: