That's the interesting question raised by polling expert David Moore in Mark Blumenthal's pollster.com blog. Moore's finding: The polls on average were pretty steady throughout October. But there was wide variation between polls during any given week. Then, in November, the average results stayed the same, but there was much less variation between polls. Blumenthal advances one theory as to why not: He argues that pollsters take special care to weight the results correctly on their last poll before the election, because they know they will be judged on how close it comes to the actual election results. That's certainly plausible. But I do not dismiss, as much as Blumenthal does, an alternative theory, which is that opinion congealed during the last week. As the AP-Yahoo series of 10 polls in the 12 months before the election show, there was a lot of movement back and forth between candidates that wasn't registered in the overall standings very much because movements to and fro tended to cancel each other out. Maybe that movement stopped in the last week.
This theory is not mutually exclusive with Blumenthal's; both could be right. If so, they buttress the argument that the shrewd Noemie Emery makes, that Barack Obama realizes that his victory was contingent, that he could easily have lost the nomination if Hillary Clinton's campaign had thought to seriously contest the caucuses or could easily have lost the general election if the financial crisis had not come along in mid-September and obliterated (with help from the candidate) the tenuous lead that John McCain had developed in the first half of September. So, Obama, in her view, is proceeding with appropriate caution in staffing his administration and choosing which election promises to keep and which to throw under the bus. Which is to say that Obama is guessing that my theory, instead of or in addition to Mark Blumenthal's, explains why the poll results tended to converge in the last week before the election.