But the gains have been already cancelled by Romney’s better performance in the first debate. You could spin this in two ways. One one hand, you could argue that the “47%” comment did move the polls, and that ceteris paribus it would have reduced significantly Romney’s chances of winning. On the other hand, you could say that ceteris should not be expected to be paribus; polls are expected to shift back and forth, and regress to the mean (where “the mean” is dictated by the fundamentals—incumbency, state of the economy, etc), and that if 47% and the debate hadn’t happened, other similar things would have.
Silver’s model already at least attempts to account for fundamentals and reversion to the mean, though. You could argue that the model still puts too much weight on polls over fundamentals, but I don’t see a strong reason to prefer that over the first interpretation of just taking it at face value.
He basically jumped to fame when he predicted the result of many of the Obama-Clinton primaries far more accurately than the pundits. He then got right 49 out of 50 states in Obama-McCain (missing only Indiana, where Obama won by 1%). He also predicted correctly all the Senate races in 2008 and all but one in 2010. In the House in 2010 the GOP picked just 8 seats more than his average forecast, which was well within his 95% confidence interval. (All info taken from Wikipedia.) I do not know of any systematic comparison between his accuracy and that of other analysts, but I would be surprised if there was someone better.
So? That just means that some of the people who trade on intrade also made the mistake Will alludes to.
Nate Silver’s model also moved toward Obama, so it’s probably reflecting something real to some extent.
But the gains have been already cancelled by Romney’s better performance in the first debate. You could spin this in two ways. One one hand, you could argue that the “47%” comment did move the polls, and that ceteris paribus it would have reduced significantly Romney’s chances of winning. On the other hand, you could say that ceteris should not be expected to be paribus; polls are expected to shift back and forth, and regress to the mean (where “the mean” is dictated by the fundamentals—incumbency, state of the economy, etc), and that if 47% and the debate hadn’t happened, other similar things would have.
Silver’s model already at least attempts to account for fundamentals and reversion to the mean, though. You could argue that the model still puts too much weight on polls over fundamentals, but I don’t see a strong reason to prefer that over the first interpretation of just taking it at face value.
Has there been any analysis of how accurate Silver’s predictions have been in the past?
He basically jumped to fame when he predicted the result of many of the Obama-Clinton primaries far more accurately than the pundits. He then got right 49 out of 50 states in Obama-McCain (missing only Indiana, where Obama won by 1%). He also predicted correctly all the Senate races in 2008 and all but one in 2010. In the House in 2010 the GOP picked just 8 seats more than his average forecast, which was well within his 95% confidence interval. (All info taken from Wikipedia.) I do not know of any systematic comparison between his accuracy and that of other analysts, but I would be surprised if there was someone better.
Silver makes and changes his predictions throughout the campaign season. Which predictions is this referring to?
The ones on the eve of election day.
Why? That might mean that, I don’t see how it would necessarily mean that.