Running numbers

From the Economist:  Running numbers

IN A short story called “Franchise”, Isaac Asimov dreamed up a computer that saved Americans from going to the polls. The machine was fed data, and interviewed one representative voter, before announcing a result that perfectly reflected what would have happened had the election been held. In real life Americans still form long lines to vote, but the fantasy is not so absurd. A group of data-hungry forecasters have recently become rather good at predicting what will happen on election day.

The most well-known of this bunch is Nate Silver, a blogger for the New York Times. After plugging this year’s data into his statistical model, Mr Silver predicted a large electoral-college victory for Barack Obama. The forecast was so out of step with the conventional wisdom that the race was tight that some accused him of liberal bias. In fact, he underestimated Mr Obama’s performance. But he got all 50 states right (compared with 49 in 2008).
Like Asimov’s computer, the soothsayers analyse reams of data, but they do not settle for one interview. Their models rely on the many state polls released each week. These are aggregated and weighted to form a picture of the electorate. Some in the media seem put out by the forecasters’ success, as it devalues their own analysis, full as it is of exciting but often trivial narrative, not dull regression analysis. With the boffins in the ascendant, the future of reporting may be more boring. But it should also be more accurate.