The 2018 Senate races come down to a battle between a near-immovable object and a strong, possibly unstoppable, force.

The immovable object is the map. In 2018, Democrats are defending 26 Senate seats (I’m counting Bernie Sanders and Angus King as Democrats) and Republicans are defending only 9. And in five of the states Democrats are defending (Montana, North Dakota, West Virginia, Missouri, and Indiana), Donald Trump won by a wide margin. It’s very difficult for Republicans to lose with this sort of map.

But Democrats have a strong force on their side: anti-Trump sentiment. With two weeks to go, Trump’s approval rating is 43 percent, and the Democrats lead by about 9 points in congressional generic ballot polls. In a presidential year, numbers like that would likely translate into a big disadvantage for Trump and down-ballot Republicans.

The analytical challenge is figuring out if anti-Trump sentiment is strong enough to overcome the GOP-friendly map—or if the Republicans will be able to exploit the map, adding to their 51-seat majority by defending their vulnerable seats and taking some red-state ones.

To that end, early this year I built a Senate forecast called SwingSeat. Every day, I feed this statistical system new head-to-head polling and other data (e.g., historical polling data, which states have incumbents running, past results in each state, presidential approval ratings, etc.), and the model spits out forecasts for every race and for which party will control the chamber. We can use this model (which has been cranking away since June) to chart the outlook for Senate control and show how the race has evolved over time.

Forecast: likely Republican control with a chance of an upset

At the time of publication, SwingSeat projects Republicans are roughly 4-to-1 favorites to hold the Senate. The most likely result is 52 GOP seats, probably attained by holding Texas and Tennessee, unseating Heidi Heitkamp in North Dakota, and winning two more seats. The model isn’t exactly sure where those wins come from—sometimes the GOP takes tossup states (like Nevada and Missouri), and sometimes it wins upsets in leans-Democratic states (like Indiana, Arizona, Florida, or Montana) or some combination of the two. Either way, per the model, 52 GOP seats is for now the best prediction.
But the spread of possible outcomes is quite wide, and a Democratic win is still plausible. If a few Republican candidates stumble at the last minute (remember Todd Akin and Richard Mourdock in 2012?), if the Democrats see a late surge, or if the polls have systematically underestimated the Democrats (as they underestimated Donald Trump in 2016 and Barack Obama in 2012), Democrats could take enough seats to win a majority.

The pivotal race seems to be Tennessee. If Democrats manage to hold all their seats except North Dakota and take Tennessee, Arizona, and Nevada, they’ll win a narrow majority. And the model shows an outside chance that Republican losses could be worse (e.g., Democrats also hold North Dakota and/or pick up Texas or the Mississippi special Senate seat). If the Democrats were to have such a truly exceptional night, they might end up with 52 or 53 seats. Those are far from the most likely outcomes, but we can’t rule them out based on the data we’ve seen so far.

The model also thinks a GOP landslide is possible. In some simulations, the model reverses all those factors and imagines some Democratic faceplants (Kyrsten Sinema in Arizona and Heidi Heitkamp may be doing that right now), a last-minute improvement in Trump’s numbers, and a polling error that works in the GOP’s favor. In those cases, it’s plausible that the GOP holds on to all its seats, adds North Dakota, Missouri, Indiana, Florida, and possibly another seat where the Democratic incumbent has been running strongly (e.g., Montana or West Virginia). The exact numbers vary, but a truly great Republican night could end with a GOP count in the mid-to-high 50s.

How the GOP got into the driver’s seat

Earlier this year, the race for Senate control looked more competitive. The model put the GOP win probability a little below 60 percent (it’s hovering around 80 percent right now) and projected that a 50-50 split (with Vice President Mike Pence breaking ties in favor of the GOP) was the likeliest result. But there was a real shift in mid-to-late September.
What happened? Why did the race move towards the Republicans? There are a few possible answers.

The most obvious is Brett Kavanaugh’s confirmation fight. Right around the time that Christine Blasey Ford (who accused Kavanaugh of sexually assaulting her when they were both teenagers) and Kavanaugh testified before Congress about the allegations, the Republican win probability began to climb. Much of this was due to movement in the red states. Marsha Blackburn’s win probability in Tennessee began to climb then, as did Ted Cruz’s in Texas. Polling has been sparse in North Dakota so it’s hard to know how much Kavanaugh may have changed the numbers there. Still, it’s worth noting that around the time of the hearings Republican Rep. Kevin Cramer suddenly started posting double-digit leads against Heidi Heitkamp.

But the Kavanaugh fight didn’t change the dynamics of every race. West Virginia’s Joe Manchin, the lone Democrat who voted for Kavanaugh, hasn’t seen any slippage in his numbers (some of his most recent polls have been quite strong). The race between Democratic incumbent Joe Donnelly and Republican Mike Braun may be tightening in Indiana, but the model didn’t see any major changes in that race until well after the Kavanaugh hearings. In Missouri, the race has remained tight before and after the hearings, and there hasn’t been enough data in Montana to come to a firm conclusion either way. In the swing states, it’s hard to discern any trend. Republicans have gained in Arizona and Nevada but lost ground in Florida, and those changes don’t line up neatly with the Kavanaugh and Ford testimonies.

This pattern makes sense intuitively. The Kavanaugh fight likely fired up both sides. In some very red states (North Dakota, Tennessee, Texas) the GOP base is bigger, so its enthusiasm overwhelmed any gains on the left. In other red states, Democrats with a stronger brand (Joe Donnelly, Jon Tester) seem to have weathered it, and Manchin may have benefited in West Virginia from supporting Kavanaugh. In more marginal states, any increase in Republican enthusiasm for Kavanaugh was likely met by increased Democratic enthusiasm against him. This doesn’t explain Missouri, where the polls didn’t move against Claire McCaskill, but it can account for much of the Kavanaugh-related movement (or lack thereof) elsewhere.

Kavanaugh explains a lot but doesn’t explain everything. Martha McSally had started to gain ground in Arizona’s Senate race before the Kavanaugh hearing. Cruz, Kevin Cramer, Blackburn, and others have managed to stay strong after the Kavanaugh hearings left the front page. And Democratic senator Bill Nelson has been making last-minute gains in Florida.

In many cases, we’re seeing races move toward what we might expect given the basic political conditions. Texas, Tennessee, and North Dakota are red states—it’s possible that they were always going to head right and that the Kavanaugh spectacle gave them an extra push in that direction. This drift toward the fundamentals would explain why Arizona is tightening (the blueness of the year and the light redness of Arizona don’t quite cancel out, but they come close) and why Nelson (a Democratic incumbent running in a purple state in a Democratic year) has been showing strength in Florida.

These fundamentals are a very rough guide. There’s a wide range of possible outcomes, and the facts on the ground—what the candidates actually do and say—will matter a lot. But it’s possible to see some of the current GOP advantage as something the Republicans were always going to have with these candidates, on this map.