I'm "lucky" enough to be two hours delayed here at the airport, so I'll take advantage of it by getting today's update out.
Today's polls are relatively inconsequential, but look like decent news on balance for Barack Obama and less so for Hillary Clinton.
|GA||Rasmussen||3/20/08||McCain +13||McCain +20|
|MN||Rasmussen||3/19/08||Obama +4||McCain +1|
|NY||Quinnipiac||3/17/08||Obama +11||Clinton +10|
The Georgia poll is actually the one I find most interesting, even though neither Democrat will carry the state in a competitive election. The results are identical, almost to the percentage point, to the Survey USA poll in February -- the only other poll conducted in the state. But our regression model was convinced that Clinton's -21 in the SUSA poll was an outlier, and that the underlying demographics aren't bad for her in the state, which Bill Clinton carried in 1992. This may be some evidence that the Clintons' Southern coattails will only carry them so far outside of the Arkansas.
The Minnesota poll is either good or bad news for Obama, depending on what you want to compare it to. It is certainly better for him than the Survey USA poll over the weekend, which had him losing the state, but down from Rasmussen's previous poll in Minnesota, which had him at a healthy +15 in mid-February. The upper Midwest is one of those regions in which Obama needs to maintain his advantage over Clinton to win an argument about electability; so far, he has been, although he hasn't been putting the crooked numbers on the board that he did in February.
By the way, I'm also working on some tweaks to the model, which center around the idea of calibrating the polls based on extremely short-term polling trends. For example, we have only one poll of North Dakota, which showed Obama winning. But that poll came at a time where he was polling exceedingly well pretty much everywhere. Should we adjust those numbers downward? Actually, we probably should. But on the other hand, in a state like Massachusetts, which was polled twice in the last week but relatively little before then, we might want to adjust upward for the short-term trend. That's the sort of thing I'm looking into. I'm also looking at ways of using national polling data as a way to help calibrate the state-by-state results.
UPDATE: Make that three-and-a-half hours late. This is why airports have bars, right?