A reader writes:
I'm with Nate Silver on this one. I plotted the 2008 presidential election results with each data point representing a poll-closing time (7PM EST, 7:30PM, etc.), to give a better idea than Nate's analysis of results over time. Doing this using the electoral vote count, I got an R-squared of 0.95, which is already very high. I suspected that doing it using the popular vote, given its more granular nature, would yield a higher value. And so it does. See the graph below (Obama votes on the bottom, McCain votes on the right), with a basically perfect correlation of 1. As Nate says, this doesn't disprove fraud (or the compelling nature of the election, come to that); it only shows that the graph does not prove fraud.
My reader's graph after the jump: