Saturday, February 9, 2008

Predicting Who the Unpledged Democratic Superdelegates Will Support

We have now put together a mostly complete dataset on the Democratic superdelegates so that we can potentially get some idea of which candidate (Clinton or Obama) the unpledged superdelegates might support. I noted in an earlier post the types of information we are using to generate these predictions and which delegates are not included in our model (specifically, those from DC and the territories). I also noted in that post all of the reasons to be cautious about these predictions. Our model is able to correctly predict 72% of those who have already endorsed one candidate or the other, but that means we would be wrong for at least 1 of every 4 delegates.

We are using the Democratic Convention Watch site to provide us with information about who the superdelegates are and which candidates they are supporting.

Several superdelegates have not yet been designated by the DNC. In our dataset, we have 305 superdelegates that have not yet endorsed either candidate, though we only have enough information to generate predictions for 288 of them. However, this still means that we can provide estimates of how likely most of the unpledged superdelegates are to support Clinton or Obama. We have presented this information in the figure below:

The red line divides those more likely to support Clinton (on the right) and those more likely to support Obama (on the left). The pattern in the figure is pretty clear: our model estimates that there are a lot more unpledged superdelegates who are likely to support Obama over Clinton than vice versa. In fact, the Obama advantage is about 2 to 1. This is significant since Obama presently trails Clinton by nearly 100 superdelegates.

The list of unpledged superdelegates and their likelihood of supporting either candidate is available here. Clinton's most likely supporters come from states like MI, CA, and OH. Obama's most likely supporters come from VT, WY, KS, and MT.

Two things to note. First, there are reports that Patrick Lynch will be endorsing Obama next week. Our model indicates that he was slightly more likely (52%) to support Clinton over Obama. So, we've already gotten our first one wrong. However, in our defense, our model showed Christine Gregorie was more likely to endorse Obama than Clinton, and she did. Second, we have generated a prediction for Howard Dean and according to our model, he is more likely than anyone to vote for Obama. In reality, as DNC chair, his vote is likely to go to whoever has the most delegates.

We will try to update these estimates as more superdelegates pledge their support to one candidate or the other.

Once again, many thanks go to Alicia Prevost and Caitlin Zook who helped put all the data together.

UPDATE: I've posted more detailed information on the methodology used to create these estimates here.


Marc said...

This is great! Could you please provide some more details about the model that you are using here?

Brian Schaffner said...

I'll put together a more involved post tomorrow about the methodology behind this. Thanks.

Anonymous said...

This, and the rest of your blog, are all great stuff. Keep it up!

I echo the request for more methodology background, just because this result will be counter-intuitive for so many folks that I want to point to the post.

marry said...

Blogs are so informative where we get lots of information on any topic. Nice job keep it up!!

Technology Dissertation