Sides: The forecast is the result of three things: the factors in the model, the importance of those factors in previous elections (1980-2012), and where those factors stand today. National conditions are already not that favorable to the Democrats: it's a midterm year, the economy is not growing very rapidly, and the president is not very popular. Based on Senate elections in 1980-2012, national conditions alone give the GOP good odds of taking the Senate in 2014. When we added measures of candidate quality to the forecast, those odds shot up to 82 percent. This is because many of the GOP recruits this year — like Tom Cotton in Arkansas and Bill Cassidy in Louisiana — are experienced politicians who typically do better than candidates with no previous record of winning elective office.
FIX: Iowa and Michigan look stronger for Republicans in the model than most people think. North Carolina looks stronger for Democrats. Why the discrepancy in each of those states?
Sides: Good question. The way to think through states like Iowa, Michigan and North Carolina is to compare them in terms of the factors currently in the model.
In Iowa, as in several other states, we had to make an assumption about what kind of GOP candidate will emerge to take on Democratic Rep. Bruce Braley. We assumed that it will be a candidate with previous political experience similar to GOP candidates in previous open-seat races. That would be someone who has been elected to a statewide office or to the U.S. House. It seems more likely, however, that the eventual nominee will have, at most, elective experience at the state legislative level. This would increase the odds for Democrats.
In Michigan, our model is pretty much in line with others. It suggests a close race, with Republicans the slight favorites. The two candidates — Democratic Rep. Gary Peters and former Secretary of State Terri Lynn Land — are both strong candidates, and polls currently show a close race.
North Carolina is clearly the forecast that deviates most from others. We currently give Kay Hagan a good chance of winning. Three things are working in her favor. North Carolina isn't as strongly Republican as some other key states. She is an incumbent, and incumbents do have advantages in congressional elections. Third, the most experienced GOP candidate (and current front-runner), Thom Tillis, serves in the state legislature, and historically state legislators have not done as well in Senate elections as U.S. House representatives or candidates who have won a statewide office.
FIX: How did you decide on the factors that go into the model? Can they/do they change?
Sides: The model currently relies on factors that political science research and even casual observation would highlight as important predictors of congressional elections — things like presidential approval, state or district partisanship, incumbency status and candidate experience. There are two components that we'll add soon.
The first is fundraising and the second is polls. (Indeed, some preliminary analysis suggests that factoring in fundraising will shift outcomes in key districts and states. The Iowa Senate race will tilt toward the Democrats, for example.)
The second component is polls. Combining a forecasting model with a polling average is nothing new, of course. But as we get closer to the campaign, the polls will become more predictive and thus the model will weight those more and more heavily.
FIX: What surprised you most in this latest model run?
Sides: What surprised us is that the current model is so bullish about the Republicans' chances in the Senate. However, as we've noted all along, the current model is based only on conditions today. If conditions change or the polls shift because of unexpected events in a race, our forecasts will change too.