The strange merger of the objective and the subjective

Have you noticed that all conservatives and Republicans, when they predict on the basis of an objective analysis the outcome of the election, predict a Romney victory, and that all liberals and Democrats, when they predict on the basis of an objective analysis the outcome of the election, predict an Obama victory? I’m the only conservative I know of who has written that I expect an Obama victory but hope for the opposite.

- end of initial entry -

Alexis Zarkov writes:

It seems we have become obsessed with polls, and predicting the outcome of the 2012 presidential election. Economist turned baseball-prognosticator turned poll-aggregator Nate Silver has received much attention with his regular New York Times column Five Thirty Eight. Silver, who is not a statistician (although the press frequently calls him that), predicts Romney will lose, giving ten-to-one odds against him. The liberals are almost giddy over Nate’s forecast.

Nate gained his reputation by calling all but one of the 51 races in the 2008 presidential election. Sounds impressive, but actually it’s not, as we shall see. Nate combines polls with his own home-grown bit of alchemy, which is far from transparent, to make his predictions. He puts in a lot of work, and uses a lot of technical machinery. Perhaps too much. Let’s see why. About 40 of the 51 races are easy to predict. Washington D.C. will go for Obama with the same certainty as the sun will rise tomorrow. I’d sooner expect an asteroid to appear out of nowhere and crush the earth. Obama will almost for sure win New York, California, Massachusetts, New Jersey, Illinois, Oregon, and so fourth. Romney will certainly take Texas, and the deep south along with Utah, Oklahoma, etc. Anyone who looks at prior RCP electoral maps and reads a little political science can easily rack up 41 reliable predictions. Then if you flip a coin, on average, you will get five more states right. So Nate has to do better than 46 to justify his elaborate machinery and reputation.

How much better? I can answer that. Using some mathematical statistics, I find that Nate has to get 50 out of the 51 races correct to claim significantly better performance than a coin flipper. He will be able to claim he has a 94 percent chance of getting any state out of the ten in play right. If he misses two, then he drops to 84 percent and that’s not good enough. One needs to reach over 90 percent to claim statistical significance. This race is much closer than 2008, so Silver is on thin ice. So is everyone else whether they use objective methods (Silver), or subjective methods (Stogie).

Posted by Lawrence Auster at November 06, 2012 06:05 PM | Send