Here’s how we arrive at that conclusion. As described before, we estimated two models to predict the outcomes for 2014. One was based on previous Senate elections from 1952-2012 and the other on more recent elections (1980-2012). There is substantial agreement between the models on the order of elections with respect to the chances of the Republicans winning. Both models identify the same road to Republican control.
The graph above — crafted by the excellent Chris Ingraham — shows our estimates of Republican victory in each of the 36 Senate elections scheduled for November. Following the advice of Monkey Cage contributor Andrew Gelman, we averaged the estimated chance of a Republican win produced by the model based on 1952-2012 and the model based on 1980-2012.
The basic math goes like this. The Republicans hold 30 seats that are not up for election in 2014. To gain control of the Senate, they need 51 because Vice President Biden would cast tie-breaking votes in a 50-50 Senate. Thus the Republicans need to win 21 (of the 36) elections this fall. Both of our models identify the same 17 elections where the Republican chances of winning are quite good. If the Republicans win those 17, then they need four more.
After the 17 seats that look likely to go Republican, the model settles on these four elections as most likely to go Republican: Alaska, Iowa, Louisiana, and Montana. In later posts, we’ll analyze these races individually. For now we note that all four seats are currently held by Democrats, with two (Begich in Alaska and Landrieu in Louisiana) having incumbents seeking reelection. Of course the Republicans could lose all four of these elections and still gain control of the Senate by winning a different set of four. Our point is that any other four pose more difficult challenges than these four. (The next three in line are Michigan, Arkansas, and North Carolina).
It may be surprising that Arkansas, where Democrat Mark Pryor is seeking reelection, is not in this top four. Our initial estimates give the Republicans only about a 1 in 3 chance of defeating Pryor, and identify Landrieu and Begich as more vulnerable Democratic incumbents. By contrast, the Cook Political Report and Larry Sabato’s Crystal Ball identify Pryor as the most vulnerable or one of the must vulnerable Democratic incumbents this year. Cook views the race as a toss-up and Sabato sees it leaning Republican.
In this case, Cook and Sabato may have a better take on Pryor’s reelection chances than our model does at this moment. One factor in our model is the previous vote for the Senate seat up for election. The rationale is straightforward: the better a party does in the previous election, the better a party is likely to do in the current election, especially when it is running the same candidate. But Pryor’s case this year may be misleading. In the previous election in 2008, Pryor won reelection with 100 percent of the major party vote because the Republicans did not contest the election. Our model makes an adjustment for an uncontested previous election, but this may not be enough.
Moreover, the Cook and Sabato predictions also take into account Pryor’s competition, Representative Tom Cotton, who qualifies as a high-quality candidate by any usual definition in political science. Our model does not yet include challenger characteristics. If Cotton can defeat Pryor, then taking control of the Senate would only require winning three of the four elections held in Alaska, Iowa, Louisiana, and Montana.
As we begin to incorporate other information into the model, it is likely that the model will become less optimistic about Pryor’s chances. Hotly contested Senate elections –as these are forecast to be – can obviously turn on the dynamics of individual contests, which we will be monitoring moving forward. What we’ve done here is map the electoral landscape and identify those states where those dynamics may prove most important.