Over the last few weeks, pollsters across the country have released polls in key Senate races—and my forecast model, SwingSeat, has eaten them and come up with new probability estimates in each state. These polls were, on the whole, really good for Republicans and the model thinks they currently have roughly 3-to-1odds of holding the Senate in November. That’s not a guarantee (Trump had a 25 percent win probability heading into Election Day 2016), but it is a real advantage.
I’ll walk through the most important races and explain why the model thinks what it does in each race. Feel free to jump around to whatever state you want, but I’ll start with the states where the GOP got good polls, move to the states with good news for Democrats and then head to the states with weird, middling news.
RACES WHERE THE GOP SAW GOOD POLLS
Florida: Rick Scott Beats Fundamentals
Florida Atlantic University recently published a poll showing Florida Gov. Rick Scott leading Democratic Sen. Bill Nelson by 6 points. Our model ate this poll (along with the relatively dense record of existing polls) and decided that Scott is a slight favorite (55 percent win probability) to win the race.
The FAU poll improved Scott’s overall standing. But the real story here is that the race was close in the first place.
I’ve made this point before, but the fundamentals of this race (the stuff that doesn’t include head-to-head polling) look reasonably Democratic. Florida is a swing state, but a Democratic incumbent is running in a midterm when an unpopular Republican is in the White House. In other traditional swing-y states such as Pennsylvania, Wisconsin, and Ohio, this formula has allowed Democratic candidates to amass huge leads.
But Scott has managed to turn this into one of the closest races in the country. His approval rating, according to the latest Morning Consult poll, is 54 percent. Scott got high marks for his response to recent natural disasters, he’s been getting his ads out early and is trying a sort of balancing act between being Trump’s favorite governor and someone who can appeal to a pre-Trump vision of the GOP. All of that seems to be working in his favor right now.
It’s unclear who’s going to win this race—a 55 percent win is only slightly better than a coin toss. But the fact that Scott is close is a big win for Republicans either way. Even if Scott loses, he’ll successfully divert Democratic attention and resources to the state. And if he wins, it’ll make it very difficult for Democrats to take the Senate—they’d most likely have to hold all their seats, take Arizona and Nevada and take either Texas or Tennessee.
Tennessee: Blackburn Increases Her Lead
A recent Gravis poll out of Tennessee put Republican Rep. Marsha Blackburn ahead of former Democratic Gov. Phil Bredesen by 4 points. That poll pushed the model’s estimate towards Blackburn, giving her roughly 3-to-2 odds of winning that race. That’s not a big advantage, but it’s a change from earlier this year when nearly every poll showed a narrow Bredesen advantage.
This Blackburn advantage intuitively makes sense. The fundamentals suggest that she has a real advantage. Morning Consult puts Trump’s approval at 56 percent in Tennessee—meaning that it’s the sort of state that still likes Republicans even when national conditions are unfavorable. And a quick fundamentals-only forecast (which uses factors like presidential approval, incumbency, and statewide partisan lean, but excludes head-to-head polls) suggests that Blackburn should have roughly 2-to-1 odds of winning this sort of matchup.
Put simply, the model is suggesting that Blackburn is only underperforming (or, alternatively, Bredesen is overperforming) this view of the fundamentals. And if she continues to improve her position, the GOP will benefit greatly—Tennessee is one of the few Democratic pickup opportunities on the map.
Missouri: The Closest Race in the Country
Now that Florida and Tennessee have moved towards the GOP, Missouri takes the crown as closest Senate race in the country. As my colleague Andrew Egger has already noted, a recent Missouri Scout poll showed Republican Josh Hawley and Democratic Sen. Clare McCaskill tied at 47 percent. The model ate that poll, looked at past data and concluded that McCaskill has an 54 percent win probability. That’s an extremely close race.
A close race in Missouri makes sense. In Morning Consult’s latest state-by-state estimates of Trump’s approval rating, 52 percent of Missouri registered voters approved of Trump. If we assume that Senate races are typically referendums on the president (they are) and that McCaskill will get a modest bonus for being an incumbent (and, based on her past performances, you wouldn’t expect an abnormally large incumbency bonus), you get . . . an extremely close race.
Wisconsin: Vukmir gets a great poll, but one poll isn’t enough
The biggest (or maybe second biggest—we’ll talk through Indiana later in the piece) was a Marquette Law poll showing Leah Vukmir, the newly minted establishment-friendly Republican Senate candidate, trailing Democratic Sen. Tammy Baldwin by only two points in Wisconsin.
But SwingSeat still thinks Baldwin has a big advantage—a 93 percent win probability, to be exact. There are two reasons for this.
First, the model has enough information in this race to avoid over-reacting to new data. If you take a look at FiveThirtyEight or RealClearPolitics, you’ll see that we already have numerous polls that put Baldwin ahead of Vukmir by high single- or even double-digit margins. The model takes those polls into account, and one good poll for Vukmir wasn’t enough to change its mind.
Second, there are diminishing returns to high margins. Think about it this way—the model thinks Baldwin is ahead by high single digits right now and her win probability is 93 percent. If we got four new polls tomorrow that put Baldwin ahead by 30, how much would you increase her win probability? You wouldn’t increase it much, because there’s not much room above 93 percent.
Similarly, Decreasing Baldwin’s lead from about 15 points to something closer to 10 points doesn’t change her win probability much because it’s high in both cases. But if Vukmir keeps getting good polls and the race starts to look better for her, the model will move and her win probability will really start to increase.
New Jersey: Menendez Corruption Leaves Many on the Fence
A few months ago, Fairleigh Dickinson published a poll showing Democratic Sen. Bob Menendez with an 4-point lead. Frankly, that left me scratching my head a bit. Menendez’s recent trouble with the law (more on that here) caused his overall approval to take a hit, but a 4-point lead is very close for a Democratic senator in New Jersey.
But then Gravis and, more recently, Quinnipiac, published estimates pegging his lead at 10 and 6 points. And the model translates this (plus the rest of the data) into a roughly 76 percent win probability for Menendez.
It’s important to note that the margin isn’t the only thing matters in these polls. Gravis, Quinnipiac, and Fairleigh-Dickinson all found that a large slice of voters were still undecided in this race. If Democrats were running a more generic candidate, some of those undecided voters would probably be in the Democratic camp (New Jersey is a very blue state).
But they aren’t, and there’s an outside chance that the Republican, Bob Hugin, actually wins this race. But even if he doesn’t, the GOP should be happy about these polls. New Jersey is an opportunity for them to expand the playing field and try to get Democrats to spend time, money, and energy on a race that they shouldn’t be worried about.
STATES WHERE DEMOCRATS SAW GOOD POLLS
Indiana: An Outlier in a Data-Sparse World
Trafalgar, a Republican pollster, recently released a poll showing Democratic Sen. Joe Donnelly leading Republican Mike Braun by 12 points. The poll is recent, there weren’t many undecided voters, and there haven’t been many other polls in the state lately. So SwingSeat is relying heavily on that poll, and currently gives Donnelly roughly an 85 percent win probability.
This probability doesn’t match what you’ve seen from handicappers or what you’ll likely see from other Senate models as time goes on. But the model is doing what it can with the polling data that’s available, and it’ll self-correct as it gets more data. If I had to guess, I would say that the next few polls in Indiana would show a tighter race and that the model will catch onto that relatively quickly.
But I could be wrong. I was initially uncomfortable with how bullish my model was on Sen. Jon Tester’s chances in Montana or Sen. Joe Manchin’s chances in West Virginia. But I’ve gotten more data in both states, and it all seems to indicate that both senators have solid (though not unbeatable) leads. Maybe Trafalgar is a weird outlier, or maybe other polls are going to move in that direction as time goes on.
Either way, this is what a model is for. If a probabilistic model always agreed with your intuitions, it would be boring and useless. So we’ll see how this race evolves as time goes on.
Texas: A Decent Poll that Hasn’t Changed Much
A new Marist poll shows Texas Sen. Ted Cruz ahead of Democratic Rep. Beto O’Rourke by only 4 points. Marist is a high-quality pollster, and this is a good result for Democrats. But SwingSeat hasn’t changed much—it still thinks Cruz is a roughly 4-to-1 favorite. There are a couple reasons for that.
First, we already have a solid amount of data in Texas. That makes it hard for any one poll (unless it’s a clear outlier) to shift the model too far in either direction. The model already knows that Cruz is up by mid-to-high single digits, so a poll showing a 4 point advantage isn’t going to move the overall estimate as much as, say, a 12 point Democratic advantage would in a sparsely polled state like Indiana.
Second, the model has already seen some polls that are in this general neighborhood. Over the course of the election, we’ve seen polls that have put Cruz ahead by 11, 10, 12, 9 and 7. But we also saw polls that had him up by only 3, 5, 2 and 6 points. SwingSeat will get more myopic as the cycle moves on, but some of those better polls haven’t fully left the data.
Third, the model has gotten a little bit more sour on Cruz as time has gone on. At various points, his win probability was closer to 85 percent, and now it’s 80 percent. It’s dangerous to get too precise with these probabilities (they’re all approximations), but if we get more polls along these lines, Cruz’s probability will continue to slide.
The bottom line is that O’Rourke has a decent chance of winning this race. At his worst moments in 2016, Donald Trump had a lower win probability than O’Rourke currently does. But Cruz is an incumbent in a red state who has led every poll taken so far—he’s still the clear favorite with roughly 4-to-1 odds of holding the seat.
Michigan: Abandon All Hope, Republicans Who Enter Here
The story in Michigan is short—Debbie Stabenow has led Republican John James by double digits (16 and 18 point margins) in recent polls. Most of the time, candidates don’t come back from that sort of a deficit. Obviously, something bizarre could happen and James could pull it out. But past data strongly suggests that Democrats have this one in the bag.
WEIRD IN-BETWEEN RACES
Minnesota: Something for Everyone to Like?
The Minnesota special Senate election has mostly been flying under the radar, but two recent polls suggest that this race could be close. A recent Emerson poll showed Republican Karin Housley trailing Democratic Sen. Tina Smith (Smith was appointed to replace Al Franken, who resigned amidst a sexual misconduct scandal) by 4 points, and a Suffolk poll showed the Democrat ahead by only 7 points. Those aren’t landslides margins, and the model translates that sort of a lead into roughly 4-to-1 odds for Smith.
But those same two polls showed Amy Klobuchar, the state’s Democratic senior senator, ahead by 20 and 24 points.
So what’s going on here?
My working theory is that Minnesotans are less familiar with Smith than Klobuchar. In the Emerson poll, the Republicans in both races got 28 and 26 percent of the vote, while Smith and Klobuchar got 32 and 50 respectively. Similarly, in the Suffolk poll, the Republicans got 37 and 34 percent, while Klobuchar and Smith got 54 and 44 percent. That suggests that some Klobuchar voters are currently undecided on Smith.
I tend to think that they’ll end up voting for Smith. “Midterms are referendums on the president and his party” is one of my mantras, and I think the Minnesota special Senate election will end up moving towards left (Trump is currently unpopular in Minnesota). But the model is a data-processing tool and it’s reacting rationally to a set of polls that say that Smith’s lead is small and that many voters are undecided. And even if Housley loses, she may (like Hugin) force Democrats to spend resources on a state that they would ordinarily wouldn’t need to pay attention to.
Programming Update: New Mexico, Third Parties, and Labor Day
At this point in the piece, Libertarians and third party enthusiasts are probably upset that I haven’t mentioned former New Mexico Gov. Gary Johnson. Johnson officially jumped into the New Mexico Senate race as a third party candidate, and an Emerson poll has shown that a significant portion of the electorate currently favors him.
As of now, SwingSeat is only built to handle two candidates. When I built the model there were no major third party candidates, so it seemed like no big deal. But for now, Johnson is counted as the Republican (because he’s leading the GOP candidate in the Emerson poll) and the “undecided” category is adjusted to reflect the fact that some voters plan on casting their ballot for the GOP candidate.
I’ll be building out a feature that deals with Johnson (and other possible last-minute, third-party entrants) in the next week or two. My tentative plan is to add that feature (and possibly others depending on how my tinkering goes) sometime around Labor Day. After that, I don’t plan on making further changes to the model. I think that it’s okay to make improvements to your model midstream, but at some point the risk of putting your thumb on the scale becomes too great. And for me that point is somewhere close to Labor Day.
