Predictive Awesomeness

I was reading Larry Sabato's prediction for the GOP House gains in the 2010 midterms.  +47 seats.  He lays out a great case for the likely accuracy of the prediction, based on the validity of Gallup's likely voter model. That is predictive awesome - both in statistical validity (97%) and in outcome (+47 GOP seats).  It's a great read if you like that sort of numbers thing like I do.

But it reminded to take a look at something else.  Back on September 9th, I commented about not believing Gallup's numbers for the generic Congressional ballot for the period Aug 30th to Sept 5th.  My reasoning was that the numbers blew up to GOP +10% in the previous period and quickly reverted to GOP +0%.  I figured that that both numbers were statistical outliers and I expected the following set of numbers to look like this:

GOP 48%

Democrats 43%

For a net difference of GOP +5%.

Lo and behold - the Gallup numbers for Sept 6th to Sept 12th - GOP 48%, Democrats 43% for a net difference of GOP +5%.  More predictive awesomeness.  This time by me.  Normally I don't like to toot my own horn but how often does one get it exactly right?