Jay Cost: Under/Over Is Dems +19

But there is a decent chance of Republicans holding the House.

Last night I read a great Cost piece. I thought I got the link from RCP, but I can't seem to find it now. So I'll paraphrase the important bits from memory.

First, you can't just average the polls showing a modest Dem lead (Gallup, ABC/WaPo, Pew, Democracy Corps) with those showing a big Dem lead (CNN, FoxNews, Newsweek, etc.). Statistically these polls are too divergent -- one set of data is right and the other is wrong. It's not a high side of things/low side of things deal; either the low lead is right, or the high lead is right.

Forced to choose, Cost chooses the low end, based entirely on Gallup's proven reliability in methodology.

Based on Gallup's polling, this should translate (I think) to a 54-46 Democratic victory in the two-party vote.

It's difficult, he says, to predict how many seats that translates to. There was a model that worked well up until 1994, a model that showed a great deal of volatility and many changes in seats. But since 1994, there has been much less volatility, and a differential in raw vote totals equals far fewer seats won.

He can use a model based on the races since 1994 to guess the new math of translating vote share to seats won, but with so few elections since then (just five), there's just not nearly enough data to have confidence in the model.

Nevertheless, he plows forward anyway with best guesses. Assuming a 54/46 split, Dems gain 19 on average. About two thirds of the time they will gain something between 15 and 24. One third of the time they will gain fewer (11 to 14) or higher (say, 25-30), with probabilities split equally between these, more or less. There's a five percent chance the number of seats won will be beyond these parameters, too.

So, altogether, based on Gallup's polling (and not, I think, averaging it with the other polls in that range, which show a lesser Dem lead; Gallup has the biggest Dem lead of any of the low-end polls), the Republicans have about a 1 in 6 chance of holding the House, a two thirds chance of losing it from anywhere between 1 and 15 seats, and a one in six chance (about) chance of losing it by 16 to 20 seats (or perhaps a few more).

I don't think his models admit for any correction in Democratic skew in polling, because I would imagine that factor is already built in. Not sure about that, though.

Robert Novak also predicts 19 as his most likely Dem pickup. Michael Barone has it as 16, and he made that prediction two weeks ago. He didn't seem to want to adjust it when he mentioned that last night on Hannity & Colmes; I imagine this means he already assumed the polls were out of whack two weeks ago, so that the recent polling showing a smaller Democratic lead just bring the numbers closer to what he thought they were all along.

We should know pretty early. If the Republicans have any realistic chance of holding the house, they'll have to pick off one of two vulnerable Democrats in Georgia (GA 8 and GA 12). We've got ten seats more or less already lost due to scandal, and any realistic scenario means we're going to have to pick up at least one Dem-held seat to overcome the inevitable losses from seats now thought to be toss-ups. Georgia's polls close early, like at 7 or so, so we should have some idea if we're in this thing or not.

Unless, of course, those races are too close to call until the wee hours.

posted by Ace at 02:45 PM

| Access Comments