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.