November 14, 2012

How good were US election predictions?

Neil Sinhababu has compiled a list of US election predictions by professional pundits, ranked on how well they predicted the overall Electoral College results, the ten marginal states, and the popular vote (as a tie-breaker).

Remember that since people such as Sam Wang, Simon Jackman, and Nate Silver obtain their predictions deterministically from poll data, and publish them in fine detail, people who have any additional sources of information not represented in opinion polls should be able to do better on average, since they can also use the poll summaries as inputs to their own thinking. On the whole, though, the people with additional sources of information mostly did worse.  If we say that Florida was too close to call, we find ten predictions that were otherwise accurate. One was the Intrade betting market, seven were deterministic models, and only two were individuals.

In the future, as happened with baseball after Moneyball, the journalists should start to use the statistical predictions more effectively.  They probably won’t beat the models by much, but they should be able to avoiding doing much worse and sometimes do slightly better.

 

avatar

Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »