Maybe the first question you want to ask is, what's wrong with you? I.e., what is a Fish from Illinois doing, writing about a House race in Arizona? OK, I'll get that out of the way first.
My daily trip to FiveThirtyEight took me to the House model Nate has been working with. He's been writing about his methodology lately, and I wanted to take a look at some details to see where he's going with it. AZ-5 is just one of many, many races he has data for, and it's one I chose more or less at random, from the depressing "Lean Takeover" list - seats we can expect to lose, but shouldn't give up on. This one has a 76% chance of going over to the Republicans. I wanted to see what that 76% is all about, so I took a look at the details.
This is an interesting race. It appears that the Democratic incumbent, Mitchell, had been holding onto a steady lead until the end of August, at which point, there is a dramatic reversal. What's wrong with this picture?
I'm not sure. Having chosen this race at random, I don't know anything about the district or the candidates. I don't know of any news stories - scandals - that could account for this. Perhaps the answer lies in the polling. You have to look a little farther down for that.
Now here's where it gets really interesting. A poll conducted Aug. 25-29 by a Republican-leaning pollster shows the challenger Schweikert surging ahead by a margin of 6 points. A subsequent poll, conducted Aug. 31-Sept. 2 by another Republican-leaning pollster shows the gap opening up to 8 points. Can this be the cause of the dramatic reversal in the potential outcome?
But wait - there's more! The most recent polls on this race, conducted by two Democratic-leaning pollsters, show Mitchell with a 1-point, then a 3-point lead. Fish is confuzzled. Silver still has the race pretty solidly in the R column. Now, I won't pretend I know enough about what Nate Silver does to second-guess his methods. I know the polling data shown on this race isn't the only thing he uses in his prognostications. But from where I sit, it looks as though Mitchell has regained the lead, and it is widening.
Now, I am not a stranger to statistical anaylsis or to numerical modeling. As an electrical engineer, I once did quite a bit of it, developing a predictive model for traffic overloads in a network switch. I was pretty proud of it, since it did what I wanted it to do, and proved to be a pretty good predictor when checked against the actual data. Nonetheless, I had to build certain assumptions into the model that occasionally fell flat on their faces when there were things I didn't account for.
I hope for Mitchell's sake, Nate's model has the same quirks. We need to save every seat we can.
Update [2010-10-6 14:45:18 by Fish in Illinois]: Thanks to Dave in AZ for pointing out the timing of the primary election - and thanks to the Rescue Rangers for pulling this one out of the slush pile.