Barring any changes due to polls released in the next few hours, our Election Lab forecast gives Republicans a 98 percent chance of gaining at least the six seats they need to win a majority. This is to say, in 98 percent of the simulations from our model, Republicans win six seats or more.
According to the model, the most likely scenario, which occurs 49 percent of the time, is that Republicans win 53 seats. They have a 36 percent chance of winning more than that, however. If they did, 54 seats would be the most likely outcome. There is also about a 15 percent chance that they win 51 or 52 seats.
Here’s what we expect in the key states (see also Chris Cillizza’s rundown):
Alaska. We give Republican Dan Sullivan an 81 percent chance of beating Democratic Sen. Mark Begich. That Begich has a 19 percent chance reflects the somewhat noisy polls in that state. A victory for him is certainly not ruled out.
Arkansas. By early spring, our model gave Republicans the edge in Arkansas. That has been borne out since. Republican Tom Cotton has led in nearly every poll that wasn’t released by the Democratic Party. We give him a 99 percent chance of defeating Democratic Sen. Mark Pryor.
Colorado. Early on, this state was one we saw as leaning Democratic, and a model based purely on election fundamentals would still give Democrat Mark Udall the edge. But he has trailed in the large majority of polls over the past month. We give Republican Cory Gardner a 98 percent chance of winning. That Udall seems to be underperforming the fundamentals could reflect aspects of the race besides the factors we take into account. Or perhaps it reflects that, as in 2010, the polls are underestimating his vote share. Could Colorado be a pure miss for the polling averages this year? We return to this subject below.
Georgia. We give Republican David Perdue a 79 percent chance of having more votes than Michelle Nunn as of Election Day. The polls have moved somewhat in his direction in the last week or two. However, we are forecasting only Perdue’s share of the major-party vote, so we cannot say whether he will win the 50 percent needed to avoid a runoff. Pollster estimates nearly even odds that this race goes to a runoff.
Iowa. This, too, was a race that we believed tilted toward the GOP as of the spring. Republican Joni Ernst has since built a consistent enough lead in the polls that we give her an 88 percent chance of beating Democrat Bruce Braley. This is more optimistic for Ernst than other forecasting models, to be sure.
Kansas. The number on Election Lab factors in the chance that Greg Orman or Pat Roberts wins and that Orman caucuses with the majority party, as he has said he’ll do. Right now, the chance that Roberts wins or Orman wins and caucuses with the GOP is 98 percent. By itself, the chance that Roberts wins is right at 50 percent. So the race is a tossup, but the GOP is likely to have enough seats for the majority regardless.
Louisiana. We give Republican Bill Cassidy a 98 percent chance of winning more votes than Mary Landrieu as of Election Day. That is all our forecast can speak to. We are not forecasting whether either candidate will get to the 50 percent needed to avoid a runoff.
New Hampshire. Despite some apparent tightening in the polls, we expect Democratic Senator Jeanne Shaheen to win. The model gives her a 98 percent chance.
North Carolina. An early version of our model was bullish on Democratic Sen. Kay Hagan. That continued to be true after Republican Thom Tillis’s nomination, since it is rare for a state legislator like Tillis to beat an incumbent senator. (His status as speaker of the House may not be helping him here either.) But the polls have narrowed somewhat of late, although Hagan leads in most of them. We currently give Hagan a 77 percent chance, so a Tillis victory here would be unlikely, but certainly possible. If N.C. goes for Tillis, then Democrats are likely in for a long night.
It is worth noting that many of these races conform to our early expectations, largely because the fundamentals of this election favor Republicans and the polls tend to shift toward the underlying fundamentals as the campaign goes on. It’s also worth noting that the other forecasting models tend to see the same candidates winning these races as we do, although with greater uncertainty (something we discussed here). Averaging the models summarized at The Upshot, the predicted chance that Republicans control the Senate is 81 percent.
Of course, any model based mainly on poll averages — as are most models at this point — risks getting a few races wrong. Nate Silver noted this week that the 538 model should miss two or so races. The same is true for us. We estimate that there is only a 19 percent chance that we call every race correctly. There is about a 1-in-3 chance that we call one race incorrectly and a 1-in-3 chance that we call two races incorrectly.
If that is true, why can we be so confident that the GOP will control the majority? This is because it is unlikely for these misses to occur in enough states, and exactly the right states, to give Democrats a majority. Any bias in the polls is hard to predict year-to-year and, as Silver and Harry Enten found, may actually end up favoring the Republicans in key races anyway. (It is also worth remembering that one potential miss is Kansas, which would give Orman a win but would say nothing about which party he’ll caucus with.)
After the election, we’ll be evaluating the forecast in several respects — not only in terms of its performance of a model, as we did here for earlier elections — but also in terms of what it can tell us about the dynamics of the 2014 election. It is instructive, for example, to ascertain how candidates did relative to the fundamentals. Who did better or worse than they arguably “should” have done? In fact, forecasting a race incorrectly can actually be more instructive than getting it right — not only for what it might reveal about polling practices but for what it might tell us about the impact of get-out-the-vote efforts or other factors.
It’s also important to emphasize that the outcome of Tuesday’s election likely won’t support the many grand conclusions that analysts will draw from it. If the GOP does take a Senate majority, it will largely square with the underlying fundamentals. As Sean Trende has noted, it tells us little about 2016. There won’t be any policy mandate for the GOP, for the same reasons that the 2008 and 2012 elections didn’t provide a mandate for Obama and the Democrats. In fact, if past is precedent, the biggest mistakes on Election Night won’t come from the forecasting models, but from pundits too eager to impute significance to the results.
