Saturday, February 23, 2008

Accuracy of Unpledged Superdelegate Predictions (Through March 3rd)

For the last several weeks, I have been using statistical models to generate estimates for how likely each unpledged superdelegate was to vote for Hillary Clinton or Barack Obama. You can find last week's estimates here. Last week I posted on how well the model had performed in predicting the decisions of the superdelegates who had decided since we generated the predictions. At that stage, the model had gotten 65.5% correct.

I generated the most recent predictions on February 16th. Since then, 36 superdelegates (for whom I generated predictions) announced their support for either Obama or Clinton. Using these endorsements, we can once again test the accuracy of the model. As always, I am using the Democratic Convention Watch site for data on which superdelegates have endorsed which candidate.

The model correctly predicted the support for 75% of these 36 superdelegates, slightly better than the first model performed. Overall, the models have correctly predicted 71% of the delegates since I began generating the estimates. Among the endorsements that the model correctly predicted were those of Senators Chris Dodd, Byron Dorgan, Jay Rockefeller, and Russ Feingold (all for Obama).

UPDATED on 2/4

NOTE: I will be using the updated endorsement information to generate new predictions after the March 4th primaries.

No comments: