06 July 2009

Predicting the Waxman-Markey Vote and the Limits of Political Science

Nate Silver has put together an interesting exercise in which he has created a simple logistic model to retroactively predict the outcome of the House vote on Waxman-Markey and then applied it to the Senate. The model correctly predicts 401 out of 431 votes in the House, which is not a great result given that the vote turned on just four votes. In other words the error in the model -- 30 votes -- is 7.5 times larger than the number of votes that woul have had to change to alter the outcome -- 4 votes.

Silver uses six variables: voting record, lobbying money, district composition, district employment in carbon ntensive industries, district carbon emissions and district poverty rate. He does not include any variables for factors like spoils from the bill or receiving a call from President Obama or getting some facetime with Nancy Pelosi in the cloakroom, so of course it is incomplete. And this is the more general problem with trying to apply academic political science analysis to a specific political decision -- it can certainly be suggestive and interesting, but it is not necessarily directly relevant in particular circumstances where context matters so very much. The error in Silver's model versus the size of the vote differential tells you that.

Silver has applied his model to the Senate and finds only 52 Senators with a probability greater than 50% of supporting Waxman-Markey. Silver is correct to note that the Senate bill won't be the House bill and all sorts of dynamics could be in play. So I wouldn't write off Senate passage just yet, but I'd be very surprised if it happened in 2009. Silver comes to the right conclusion:
The question is how many ornaments the Democrats could place on the Christmas Tree before it starts to collapse under its own weight.