It has become fashionable to speak of a Democratic comeback, but we’re not really seeing one in our forecasting models. Certainly there are some individual races — particularly on the East and West Coasts, as well as some gubernatorial contests outside these regions — that look better for Democrats than they did a few weeks ago. But we’re showing Republicans gaining ground where they need to gain it to maintain decent chances of taking over the Senate. We also show improvement for them in the House forecast this week.
Our model now estimates that the Republicans have a 72 percent chance of taking over the House, up from 67 percent last week. Moreover, they have nearly even odds of a achieving a net gain of 50 seats; their average gain in a typical simulation run was between 47 and 48 seats. However, the playing field remains very broad and considerably larger are possible, as are considerably smaller ones.
Republican gains this week are mostly the result of factors at the local level; the national environment is roughly stable. The expert forecasters whose judgment we incorporate into the model continue to revise their characterizations of races, and in almost all cases, the changes work toward the benefit of Republicans. Cook Political, for instance, in a break from its usual convention of not classifying seats held by incumbents as worse than toss-ups, this week decided to classify 10 seats currently held by Democratic incumbents, including the seat held by Alan Grayson in Florida’s 8th Congressional District, as leaning toward the Republicans.
Polls of individual House districts, while more varied, also generally contained good news for Republicans this week. A series of polls conducted by Penn Schoen Berland for The Hill, for instance, found 11 of 12 first-term Democrats that tested trailing their Republican opponents. The polls ought not have been terribly shocking to Democrats — our model already had the Democrats favored to lose all but one of these seats — but they nevertheless confirm that the principle of “last hired, first fired” often holds in “wave” elections, and than many of the Democrats who won their seats for the first time in 2006 or 2008 will likely lose them this year.
Republicans also released a number of internal polls this week. The model is careful about how it handles these polls — in some cases, the release of an internal poll may actually harm the candidate’s forecast, or may change the way that the model evaluates other surveys released by the same polling firm. But, for instance, the model is persuaded that Republicans are favored in districts like the Florida’s 2nd, Alabama’s 2nd, and New York’s 23rd, where credible Republican internal polls show their candidates ahead.
Other polls need to be viewed more skeptically. One survey released this week, for instance, found long-time Democratic incumbent John Dingell trailing in Michigan’s 15th Congressional District, which includes the college town of Ann Arbor and is ordinarily reliably Democratic. But the poll used a small sample, was conducted by automated script, was commissioned by a group that has little track record of releasing polls into the public domain in the past, and contradicted a mid-September poll in the district that showed Mr. Dingell ahead by 19 points.
In general, an individual poll of a House district that appears dubious usually will turn out to be so. Right now, the model regards Mr. Dingell as having only a 1-2 percent chance of losing his seat, although those numbers will increase if other polling shows a tight race or if some of the forecasters change their classification of the race from “Safe Democrat.”
Likewise, I would be skeptical of some hopeful-seeming polling for Republicans in districts like Massachusetts’ 4th and Arizona’s 7th, where two members of the Democratic leadership — Barney Frank and Raul Grijalva — are running for re-election.
Still, the fact remains that if Democrats were to lose 60 or 70 seats — outcomes that are entirely possible — there are going to be some “shocking” outcomes in individual districts. They just may not necessarily be in the districts where some enterprising candidate has gotten cute by releasing an dubious poll.
So why all the talk about a Democratic comeback? Some of it may be because the news media had been a little bit too sure of itself in forecasting Democratic doom before, and now is (over)compensating. In consideration of the large number of seats in play, the differences of opinion among pollsters in modeling turnout, and the fact that the various indicators that analysts use to forecast House races are not always in strong alignment with one another, this U.S. House election is frankly one that calls for some hedging and humility. The 95 percent confidence interval on our model runs between a Republican gain of 17 seats and 78.
We do show Democrats with a slightly stronger number on the generic ballot than they had a few weeks ago — down about 6.5 points among likely voters rather than 8. (This is in spite of a Gallup poll which shows a considerably larger advantage for Republicans and a change to the procedure our model uses to calculate its likely voter adjustment). But when one looks at a more diverse set of indicators, as our model does, there are few signs of Democratic momentum.