Georgia: Rasmussen has Chambliss up, 50-46.
Minnesota: the tallies on the Secretary of State’s page haven’t changed from the pre-recount Coleman lead of 215. Whether that means they won’t change the tallies there until the recount is completed, I have no idea.
I also have no idea why Nate Silver says Coleman’s lead is down to 172. The Star-Tribune still says 174.
I don’t think we can read anything into that: The polls are reporting more “likely voters” than the number who voted in the general election. I think it’s pretty obvious that the run-off will see lower participation than the general, so a lot of those poll respondants should be scratched off the list. The question is, which ones?
I dunno, but they came up with essentially the same results as a poll a week or so ago (either Research 2000 or SurveyUSA, can’t remember which) which had Chambliss up 49-46.
Unless the pollsters have gone back and figured out something about older runoff and special elections, they should probably leave their LV screens unchanged, because as of two years ago, nobody had figured out how to poll one of these things. Turnout is way down from an Election Day election, and everything depends on which side has less of a drop-off.
I think how I’d think of it is like this: there’s a whole bunch of possible elections that could happen in 12 days, and the mean outcome of those possible futures is around 49-46 or 50-46, Chambliss. But there’s a really wide distribution, a really high variance, amongst all these possible elections, which means it’s much closer to a tossup than if you’d had two reputable pollsters give the same guy a 3-4 point lead in a regular election.
That isn’t how it is, but it’s a useful thought experiment, IMHO.
Coleman’s lead is down to 136, but since 46% of the ballots have been recounted, Norm’s chances of surviving seem to be picking up.
Running down the list of counties that have been recounted, it doesn’t look like there’s a tremendous bias either way in terms of whose territory’s been recounted.
If you adjust for the percentage of the vote in a county that’s been recounted, the recounted ballots probably went for Coleman by ~15,000 in the original voting. Out of nearly 2 million ballots recounted, that isn’t much of a tilt. So the geography of the recount isn’t fixing to put Franken over the top.
His best chance is to close the gap to the low double digits in the recount, and then maybe the disputed ballots that will be considered in December will put him over.
Or maybe not. Right now, if I had to bet, I’d bet that Coleman hangs on by the skin of his teeth.
Huh.
Can you say “mandate”?
Right. Nate Silver was arguing earlier that since Coleman had been out-challenging Franken by some amount, the ultimate result was likely to skew against him slightly (on the grounds that most of the challenges are pretty specious). But now Franken is slightly ahead (by like 9 out of 800-some-odd total challenges) in challenges, so any result from them is likely to be a wash.
With 64% of the ballots recounted, Coleman’s lead is at 120.
Updated county tallies here. Maybe there’s something going on at the precinct level in the half-dozen sizable partially-recounted counties, but at the county level, there’s no evidence to back up the Franken team’s claim that the recount areas have been mostly Coleman territory.
I wonder if they keep going with the recount over the weekend. I guess we’re about to find out.
WARNING: Unconfirmed blog post.
These people claim the Coleman camp is contesting ballots merely because they voted for McCain and Franken.
With that being said, I’m at a loss to figure out where “voter intent” was ambiguous on any of those. 
ETA: Another story on the same site says the Minnesota AG is considering posting the ballots online for public viewing.
New GOP standard for voter intent - all ticket-splitters really meant to vote a straight GOP ticket!
The Franken campaign has said that if challenges with obvious resolutions are included, that Coleman’s lead is down into double digits. Ballots like these seem to be what they have in mind.
I noticed that the ballots in that video were all from one county, Fillmore County, where only about 10,000 votes were cast. It’s quite possible that one Coleman recount-watcher got overzealous in challenging these ballots, in which case this really means nothing. Or it could be that this was part of a more widespread effort on the part of the Coleman camp to challenge ballots with Franken votes on specious grounds, in which case the Coleman team’s challenges are hiding a sizable batch of Franken votes.
IMHO, the single overzealous challenger theory has to be the null hypothesis here, unless and until we see evidence that such bogus challenges were widespread.
Guess we’ll just have to stay tuned…
I’ll accept that. I can’t understand the motivation though. Did he seriously think these were really questionable? Maybe just a frustration that the votes weren’t going like s/he had hoped?
Or more along the conspiracy theory vain, maybe an attempt to marginalize a Franken victory should it occur? “Coleman won the election! Coleman won the recount! It was the FRANKEN that stole the election by counting the invalid ballots!” (I can has tinfoil hat smiley?)
Meanwhile, Coleman’s lead is back up to 180, but given the number of challenges, it’s kinda hard to tell what’s really going on with that.
Of course, if ballots which voted for both McCain and Franken are invalid, then so too must ballots for both Obama and Coleman be invalid. OK, we can stop the recount now: Obama won Minnesota, so clearly Franken did, too.
Update: Nate’s broken out his statistical mojo again, and projects that Franken will win by 27 votes, plus or minus some large unspecified error. Without knowing more about the regression techniques he’s using, I’d guess that the error is Gaussian, with a standard deviation of the square root of franken_net, or 15.55. That means that Coleman winning would be a 1.74-sigma result, which has a probability of somewhere between 1 and 2%.
Nate says that the error bars are quite high, so even if his analysis is correct, Franken’s a very slight favorite.
franken_net is roughly equal to 0.837. (Third-to-last paragraph at the link.)
Ah, right, I meant to say the square root of the total number of votes gained.
And he doesn’t say what he means by “quite high”, so I was making an educated guess at what they would be. Really, I’d prefer to see his breakdown of the error bars, and the probabilities they lead to (since my assumption of Gaussian errors might also be off).
Ahhh, now I understand.
He does say that even if he’s right, Franken should only be considered a slight favorite. Which implies that Nate’s MOE is a lot bigger than 27 votes.
Given that he doesn’t give any numbers for that, though, that might just be a hunch. His statistics-fu is strong, but his unstatistical hunches aren’t much better than any of the rest of us. Again, though, if he gives any actual numbers, I’ll happily defer to that.
Back in Georgia, things aren’t looking so great for Jim Martin. RCP’s summary shows four polls taken since Election Day:
Date Chambliss Martin Pollster
11/13 Chambliss 49 Martin 46 Research 2000
11/19 Chambliss 50 Martin 46 Rasmussen
11/20 Chambliss 51 Martin 45 Research 2000
11/24 Chambliss 52 Martin 46 PPP
Not a great trend, huh? Chambliss’ numbers are creeping up; Martin’s at 46 and going nowhere.
A DSCC poll released yesterday shows the race at Chambliss 48, Martin 46. But (a) it’s a Dem poll, and (b) it still shows Martin stuck at 46.
The wildcard is still that since this is a special election, there’s no telling who will show up, and everything will really depend on that. But the odds definitely favor Chambliss.