New Forecast Shows Democrats Losing 6 to 7 Senate Seats

The Democratic majority is in increasing jeopardy in the Senate, according to the latest FiveThirtyEight forecasting model. The Democrats now have an approximately 20 percent chance of losing 10 or more seats in the Senate, according to the model, which would cost them control of the chamber unless Gov. Charlie Crist of Florida, who is running for the Senate as an independent, both wins his race and decides to caucus with them.

In addition, there is an 11 percent chance that Democrats will lose a total of nine seats, which would leave them with 50 votes, making  them vulnerable to a defection to the Republican Party by a centrist like  Joseph I. Lieberman of Connecticut or Ben Nelson of Nebraska. On average, over the model’s 100,000 simulation runs, the Democrats are projected to lose a net of six and a half Senate seats, which would leave them with  52 or  53 senators. (Even though the G.O.P. primary in Alaska remains too close to call, that outcome is unlikely to alter the model.)

The forecasts are based on a program designed to evaluate current polling and demographic data, and to compare these present-day conditions to outcomes in United States Senate races over the past six election cycles. For instance, in recent  cycles, a Senate candidate with a 7-point lead in the polls 10 weeks before  the election won about 80 percent of the time, and a candidate with a 12-point lead won about 95 percent of the time. Although the model, which correctly predicted the outcome of all 35 Senate elections in 2008, is not quite this cut-and-dried, it is this recent  track record that forms the backbone of its projections.

Of late, the source of the Democrats’ problems has not necessarily been in high-profile Senate races where the Republicans have nominated inexperienced but headline-grabbing candidates, like  Sharron Angle in Nevada and Rand Paul in Kentucky (although the model regards both Ms. Angle and Mr. Paul as slight favorites). Instead, it has been in traditional swing states like  Missouri, New Hampshire, Ohio and Pennsylvania. The last time the Democratic nominee in Ohio, Lee Fisher, held the lead in any state poll, for example, was in June. Representative Joe Sestak, the Democratic nominee in Pennsylvania, has not led any poll there since May, and Robin Carnahan of Missouri has not held a lead since January. The Democratic nominee in New Hampshire, Representative Paul W. Hodes, has not led in any of 17 public polls in New Hampshire against his likely Republican opponent, Kelly Ayotte.

The Democratic candidate lags by single digits in each of these states, and victories there remain entirely possible (perhaps especially so in New Hampshire, where the Republicans have yet to hold their primary). But, at a time when they need to be drawing closer to their opponents as the clock ticks toward Nov. 2, these Democrats instead find themselves falling somewhat further behind. We are now close enough to Election Day that a deficit of as few as 5 percentage points may be difficult to overcome, especially in races where relatively few undecided voters remain.

Particularly vexing to the Democrats might be their standing in Missouri and Ohio, where the Republican incumbents —  Christopher S. Bond and George V. Voinovich — are retiring and identifiable members of the G.O.P.’s establishment have been nominated to replace them: Representative Roy Blunt of Missouri, the former Republican Minority Whip, and in Ohio, Rob Portman, the former congressman who served as trade representative and budget director in the Bush administration. And so far, the Democrats’ strategy of Bush-bashing does not seem to be resonating in these states. Mr. Fisher has only about a 20 percent chance of prevailing in Ohio, according to the model, and Mrs. Carnahan – once considered  a  strong nominee – has about a 1-in-10 chance of doing so in Missouri.

At a superficial level, states like  Missouri,  New Hampshire and Ohio might seem of  least concern to Democrats, since  the races there are for seats now held by retiring Republicans. But the Democrats need to keep some of these races in play to preserve a reasonably level playing field, especially since they appear to be almost certain to lose several seats they now hold. Foremost among these are North Dakota, where the popular Republican governor, John Hoeven, should romp to victory in a seat being vacated by Byron L. Dorgan, and Arkansas, where Senator Blanche Lincoln’s deficit in the polls exceeds 20 points (no recent Senate candidate, incumbent or challenger, has come back from such a significant margin so late in the race). Indiana, where Evan Bayh is retiring, is also more than 95 percent likely to flip to Republicans, according to the model. The outcome in Delaware, where there has been little polling, is more uncertain, but the model has established   Representative Michael N. Castle, the Republican nominee, as a 90 percent favorite.

Less clear are the Republicans’ prospects in California, Washington and Wisconsin,  traditional blue states where  incumbent Democrats are running for re-election. Although Barbara Boxer’s approval ratings in California have turned negative in many polls, the same ratings have remained decent for Patty Murray in Washington and Russ Feingold in Wisconsin. The forecast model in Wisconsin is somewhat skeptical of Republican chances there, particularly against Mr. Feingold, an idiosyncratic senator who has broken from his party’s position on many issues. It is these states – along with Illinois, where voters seem unwilling to commit to either the Democratic nominee, Alexi Giannoulias, or the Republican, Mark Kirk – that may determine whether Republicans indeed have a chance of taking over the Senate.

In one sense, the Republicans’ math remains quite daunting. There are 29 Senate contests in which the Republicans have at least a 5 percent chance of winning, according to the forecast: Republicans would need to win at least 28 of these in order to head into the 112th Congress with an outright majority. They must not only sweep essentially all of the Democratic-held seats, but also successfully defend all or almost all of their own. And in some of those, like Florida, Kentucky and perhaps North Carolina, Republicans remain quite vulnerable. Their chances would improve, of course, if  they are able to put in play a state like Connecticut, which falls just below that 5 percent threshold but where the Democratic nominee, Attorney General Richard Blumenthal, has sacrificed some of a once-formidable lead.

At the same time, the outcomes in individual Senate races are not uncorrelated: if Republicans tend to overachieve in some states, they will probably also overachieve in others. Certainly, if Democrats were to have another month as bad as the one they endured in August – one characterized by poor economic news and ethics scandals – the possibilities for a Senate takeover would rise further. But the reverse could also be true. It is not out of the question that the polling could shift back toward the Democrats: many voters do not begin paying attention in earnest to Congressional campaigns until after Labor Day, and the parties’ messaging strategies have yet to solidify. The Democrats retain long-shot chances – about 3 percent – of actually gaining one or more Senate seats and restoring a 60-seat majority.

It could also be that the polling somewhat overstates the degree of danger that Democrats face. Many of their poorer results, for example, come from polling companies like Rasmussen Reports that use automated scripts to conduct their surveys, rather than live operators, and which often poll in a blitzkrieg fashion, with all of their polling completed within a few hours. Although FiveThirtyEight has not found these “robo polls” to be less accurate than live-operator ones in recent elections, they are generally associated with lower response rates, and they may not be getting a representative sample of voters on the phone.

Nevertheless, the forecast model is carefully calibrated to account for these contingencies. No one pollster is allowed to dominate the ratings, for instance, no matter how widely or indiscriminately it polls, and pollsters whose surveys consistently lean toward one party have their results adjusted to bring them back toward the norm. The model is also careful about determining the extent to which the outcomes in different states are correlated with each other, and in estimating the degree of uncertainty associated with its forecasts.

A fuller description of the methodology behind the forecast model can be found on the methodology page; we also expect to convey certain facets of the model in more detail in coming posts. In the meantime, we invite readers to explore the interactive displays that  contain the model’s forecasts not just under current conditions but also at semimonthly intervals dating to Feb. 1. For the time being, we expect that our Republican readers will take more pleasure in doing so than our Democratic ones.