Nate Silver is the statistician and poll-watching guru who rose to prominence in 2008 by developing a poll aggregation model that correctly predicted 49 of 50 states in that year's presidential contest. He has since been hired by the New York Times and gained more prominence with his up-to-the-minute analysis on new polls and the state of the presidential race.
Conservatives have attacked Silver relentlessly for his model that predicts that President Obama with a 70%+ chance to win the election even as national polls narrowed throughout October. This has caused Silver's New York Times colleague Paul Krugman to explode in hysterical hyperbole about a conservative "attack" on "objective methodology."
This is clearly wrong. There is undeniably a subjective aspect to Silver's model - he weights the polls, for example, according to what he believes to be the most reliable pollsters. There are other subjective parts to the methodology as well. As Manhattan Institute scholar Ted Frank writes,
Silver lets the fact that Gore outperformed his polls by so much influence his model of how to predict undecided predilections for the incumbent and how to calculate house effects, rather than tossing it out as a case where polls didn't capture Election Day sentiments. That's a subjective decision to choose a particular objective rule, not an inherently objective decision.
If Silver's model is wrong, election day will bear that out. The accusation that Silver's model consciously favors Democrats is a strange one. Silver built his reputation, after all, on his model's predictive power in the 2008 election. It would be strange that he would willingly sell out the thing that made him famous for the vague and indeterminate benefit of propping up Democrats in advance of what might be an embarrassing electoral defeat.
Silver's model's confidence in an Obama victory also does not rule out a Romney victory entirely - Romney would win once out of every four times if the election were run over and over as an experiment. To borrow from the advanced statistics of the sportscasting world - which Silver hails from originally - surprising things can happen.We're one week away from the election and national polls show a tight race. That may be analogous to, say, a tied football game with two minutes to go with one team possessing the ball on their own side of the field, but near midfield. Instinctively we might say that it's a close game, but advanced statistical models actually would give one of the teams an overwhelming chance to win the game. Luckily for us, this happened yesterday!
With two minutes to go, the Tennessee Titans had the ball in a tie game with the Indianapolis Colts. AdvancedNFLStats gave the Titans a 78% chance of winning what appeared to be a close game. This isn't because their statistical model is weighted towards the Titans - it's because they've built a model that says that a team who has the ball and is driving down the field towards the end of a tie game has an overwhelming chance to win based on the historical evidence.
What happened, however, is that the Colts stopped the Titans' drive, forced overtime and won. That doesn't mean that AdvancedNFLStats' model is wrong. It means that this was one of those times that the team with less of a chance to win in the final two minutes actually did pull it out.
What this all means is that it's certainly a possibility that Silver is wrong. It's unlikely, however, that he's conciously weighting the polls in favor of Barack Obama. If his model is wrong, then conservative statisticians should develop their own modelers - and this would be a very good thing for election forecasting. The answer to Nate Silver isn't to go back to traditional political punditry - which is notoriously unreliable - but it's to develop ever more statistical models in the hopes of finding a better way than Silver to forecast.