2014 Senate Election Prediction Results
Nate Silver and Sam Wang are two of the most popular prognosticators of elections in the United States. They are also known to be openly critical of each other’s methods. But now that the elections are over, let’s take a look and see how each of them fared.
Wang suggests using the Briere Score to evaluate predictions, so I’m going to use that as one measure. This measure compares the predicted win probabilities to the actual outcomes and takes the average the squared differences. I also looked at the mean squared error of the predicted margins of victory as another measure of accuracy.
Silver amassed a Briere score of .14656 based on the 12 closest senate elections (Kentucky, Arkansas, Louisiana, Georgia, Colorado, Alaska, Iowa, Kansas, North Carolina, New Hampshire, Minnesota, and Virginia). For those same races, the Silver put up a mean squared error of 38.727 for margins of victory. Wang fared worse by both these measures with a Briere score of .17018 and a mean squared error of 44.981.
I’ve summarized the results in the graphic below. Each square represent the squared error of margin of victory within a state, and the color of the square corresponds to which direction Wang or Silver was wrong in. For instance, Silver predicted a 5 point win for the Republicans in Arkansas, but they actually won by 17.13. So the square is very red because they Republicans exceeded Silver’s expectations. Alternatively, Louisiana was just slightly bluer than both Wang and Silver projected, so the square gets a slight hint of blue.
So I’m declaring Silver the winner of this round in the batter of Silver vs Wang.
However, if you look at all of the other popular senate predictions, Silver was not the winner. According to Sam Wang, that title goes to Drew Linzer of the DailyKos for his Senate predictions based on Briere score and followed by the Washington Post. FiveThirtyEight came next in a group that included the Huiffington Post and Betfair.com with Briere scores of 0.14. (These Briere scores are slightly different than the ones I calculated, as I used a few more states.) This was then followed by the Upshot and the Princeton Election Consortium. So while Silver, the highest profile of all the prognosticators, won the Silver vs Wang battle, he sort of finished in the middle of pack in terms of predictions. At least he can take solace in the fact that he beat his old employer!