Ouch. Post-convention polls by Time and Newsweek give Bush, respectively, a 10-point and an 11-point lead.
First, this discounts the claim, peddled by Democrats after the Democratic convention, that Kerry's small convention bounce is due to the ultra-polarized electorate.
Secondly, there has been much discussion around the blogosphere of various mathematical models used to predict election turnout. Fitting models to things is easy - as an applied math guy I should know - enter the data, run some best fit algorithm, and immediately you have a prediction for the future. But is there any reason to believe elections can be modelled in this manner? Trying to model, say, interarrival times of customers in a business as a Gaussian variable will give you bad results no matter how much you work it -because interarrival times are exponential. Not every approximation you come up with will match reality.
In fact, this swing - a 7 point lead for Kerry three weeks ago into an 11 point lead for Bush now - without any measurable change in the variables that go into these models - suggest that election outcome is a phenomenon which cannot be modelled using macroeconomic variables and other such data. I know little about this line of research, but has anyone ever made an argument for these models that did not go along the lines of "picking these parameters, the error in the previous 10 elections is small, so it will probably be small in the next election?"
Anyway: I, actually, did not think Bush's speech would have a large impact. I saw party spin that falls flat of reality - things in Iraq are going terribly and the more Bush pretends everything is hunky dory, the more disconnected he seems from the real world. The american people seem to disagree.