Sam Wang (who’s a biologist in his day job and analyzes surveys for fun) pointed me to his poll aggregates and predictions for the presidential race. Wang writes:
My current view is that modelers who put in econometric assumptions face the challenge of ascertaining how much new information their additional variables (GDP, unemployment) add which is not already in the data implicitly. For certain they add noise. To my taste, I would use the polls to establish a prior for November, then add the other stuff, for instance to calculate a future storm path.
Wang gives Obama a 90% chance of winning the election. I don’t see Wang’s popular vote forecast anywhere, but I’m guessing he’s giving Obama an expected share of about 53% of the two-party vote. In contrast, Doug Hibbs’s model gives Obama 47.5% and, as Hibbs has noted, the consensus is somewhere around 50-50.