Election Prediction Recap

In trying to model any phenomenon, it always possible to bring in too many correlated independent variables. With enough variables, it is always possible to fit past data. However, the question is whether the true dynamics have been captured. The simplest model that can reasonably explain the data is usually the safest.

The professionall analysts in this past election did not fair too well. As it looks now that there will be a net 65-seat pickup for Republicans in the House of Representatives. Larry Sabato, Director of the University Center for Politics, predicted a gain of 55 seats. Nate Silver, political statistician supported by the New York Times, used elaborate simulations to predict a 54-seat gain. The RothenbergÂ’s Political Report faired better with a 55-65 seat gain prediction.

Using a simple linear regression model, we were able to predict a 69-seat gain with a +/- 10 seat standard deviation. The plot below shows the advantage in the actual Democratic vote versus the pre-election Gallup generic preference poll. The indicated point shows the result of this past election. Note that the linear fit very closely predicted the outcome. We submit here the humble thesis that some rather simple models have been adequate to explain events like the House election, where the 435 seats available allow the statistical means to prevail.

Leave a Reply

You must be logged in to post a comment.