The Investor's Business Daily/TIPP poll (which was closest to the mark in predicting the 2004 outcome -- 0.4% off the actual result) now says this is a three-point race.
Before he wrote this, I thought Rove understood data. But polls are (more-or-less) random samples, and a group of pools will produce a distribution of outcomes. It is of course the case that one poll will come closest to the population outcome; the fact that a particular poll does says nothing about the skill of the pollster. Now, if one pollster comes closest ten times in a row, we can be sure that something real is going on--that she has insights about sampling that the others don't. But one correct call is nothing but luck.
Every quarter, the Wall Street Journal picks a "best economic forecaster," which is based on close an economist's forecast to predicting economic conditions in a quarter. Check out how well that forecaster does in the following quarter. If you made any decisions based on who wins the award for one quarter, you may well be disappointed in the following quarter.