Power polls

Nate Silver’s predictions for the US elections, on his fivethirtyeight blog, have received a lot of attention since the actual election results agreed extremely well with his predictions. Silver’s results have been held up as an example of the power of numbers as opposed to, for example pundit opinions, and rightly so. Silver’s results shouldn’t be a surprise. Much of the hard work was already done by the pollsters, and the aggregated predictions published by Silver and others mainly serve to underline the power of polling. Even much simpler analysis, such as a simple five-poll average  gave very accurate results.

While they are not really the same thing, his analysis has something in common with what we try to do in Complex Systems: good data combined with reasonable assumptions and a careful application of statistical principles can give simple, but powerful, results.
On the topic of pundits, the vindification of numerical analysis, and the wildly innacurate predictions of some pundits might suggest that, come the next election, numerical analysis will supplant those pundits. But I doubt this will happen. This conclusion misunderstands the role of TV pundits. Their main role is not to give accurate predictions. Their task is to make TV interesting, by providing lively discussion, and to give arguments in support of a particular point of view. In fact, it might even be better for them to not give accurate predictions: if every network lined up pundits all repeating more or less the same careful estimates, there wouldn’t be as much entertainment value.