Nate Silver, yesterday:

The closest recent parallel I can find to the Ted Stevens situation is that of former Ohio governor Bob Taft, who in August, 2005 was convicted on misdemeanor charges of failing to disclose gifts and golf trips paid for by lobbyists... He went from a 34-55 (-21) in a University of Cincinnati poll taken in April 2005 to a 26-65 (-39) in the same poll in August 2006, a net decline of 18 points.

Let's assume that Stevens will also suffer a decline in his net approval score of 18 points. Since he's at roughly 50/50 now, that would put him in the range of 40 percent approve, 60 percent disapprove. Our regression model uses approval ratings for incumbent senators as one of its inputs, and thinks a decline of this magnitude would cost a senator about 6 points in the polls ... actually, 5.8 points. So what we're going to do is apply a 5.8 point penalty to Stevens' numbers in Alaska.

Eric Kleefeld, today:
The first public poll of Alaska conducted entirely after GOP Sen. Ted Stevens was convicted on all counts in his corruption trial shows a probable Democratic pickup in this deep-red state -- but Stevens is doing surprisingly well for a newly-minted convicted felon.

The new numbers from Rasmussen: Dem candidate Mark Begich 52%, Stevens 44%, with a ±4.5% margin of error. Three weeks ago, Uncle Ted had taken a 49%-48% edge over Begich. So apparently getting convicted of a felony a week before the election can be quantified as taking five points off of a candidate's poll numbers.

Upshot: Nate Silver is a genius, with a margin of error of +/-1%.