Earlier this week, I edited my Senate forecast model so that it takes in “fundamentals”—that is, data from sources other than polling as well as polling data. A few days later, pollsters (most notably Fox News) released loads of new polls. Here’s how they changed the topline and the state-by-state level estimates.
The Topline: Republicans Maintain an Edge, but their Senate Majority is At Risk
THE WEEKLY STANDARD’s new model gives the GOP a 60 percent probability of holding the Senate in November . The outlook is mostly unchanged from earlier this week: The model still thinks that the most likely scenario is that the GOP wins 50 seats by holding their safe seats, defending Texas and Tennessee and taking North Dakota. But there are a wide range of scenarios—Democrats could sweep all the competitive seats, or Republicans could make solid gains.
Interestingly, the older version of the TWS model (which only used polls when they were available) put the GOP’s odds of winning at roughly 3-to-1. That’s far from a guarantee—as I’ve emphasized many, many times, Donald Trump had a lower than 25 percent chance of winning the White House at various points in the 2016 cycle.
So why is there this gap between what the polls say and what the new iteration of the model says? The answer: in key races, fundamentals stack up to help the Democrats.
The model is guessing that polls will move towards the Democrats
I’ll take the races state by state:
In Missouri, the Fox News poll (one of the best in the business) had the race at a tie. And the older version of the forecast (which only used polls) basically agreed and put the race as a tossup. But the new version of the model weighed incumbency, president Trump’s approval rating, and the effect of a midterm year (all of which benefit McCaskill) against the redness of the state (which helps Republican Josh Hawley) and decided that the odds were 2-to-1 in favor of McCaskill instead of 50-50.
In Arizona, the fundamentals pushes the estimate toward Kyrsten Sinema by a similar margin. Three polls—from FOX, Data Orbital and OHPI —show a low single digit race for the Senate in Arizona. The older, poll-based model responded by giving Sinema a 70 percent win probability. Adding fundamentals to the mix pushes that estimate to 85 percent.
In Indiana, recent polling has moved the race towards Republican Mike Braun. Fox News in particular showed a two point Braun lead, other previous polls put Democratic senator Joe Donnelly ahead by single or even double digit margins. That poll put Braun’s win probability at roughly 1-in-4. In this case, the fundamentals adjustments still favor Donnelly, but the gap isn’t wild—Braun has a 1-in-3 chance in the older, polls-focused version of the model.
The best number for the Democrats is in Florida, where adding in fundamentals decreases Republican Rick Scott’s win probability from over 50 percent in the older, poll-reliant model to something close to 15 percent in the newer version. So far, polling has continuously suggested a close race (two recent polls put Scott and Nelson ahead by two and one points respectively). The model thinks this race will break towards Nelson—though it’ll adjust his probability downward if the polls fail to show evidence of that.
The model helps the GOP in Texas. The older, polls-heavy version of the forecast would have looked at the newest poll (Dixie Strategies with a four point advantage for Cruz) and concluded that Democratic Rep. Beto O’Rourke had something closer to a 1-in-4 chance of unseating Cruz in Texas. But the fundamentals cut that probability down to about 15 percent. In Tennessee, adding fundamentals didn’t make much of a difference.
Both the polls-focused and the newer version of the model give Marsha Blackburn a roughly 60 percent win probability in Tennessee. The fundamentals knock Republican Kevin Cramer’s chances in North Dakota a bit (Fox put him ahead by four) but it still thinks he’s a roughly 2-to-1 favorite to win that race.
The Big Ifs of a Wide Playing Field
Right now, Democrats in numerous key contests have a few advantages stacked on each other (i.e. incumbency, President Trump’s low approval rating and the fact that it’s a midterm election). Those advantages help us better predict where races are going in the aggregate.
But the model isn’t magic, and this election is closer than the conventional wisdom might suggest. Just one candidate who defies the fundamentals could change the math for chamber control.
For instance, the model thinks that Claire McCaskill will gain ground between now and Election Day. But maybe she won’t. As time goes on, the strength of these fundamentals factors will diminish and the model will become more reliant on polling. If Hawley opens a real lead in the polls instead of McCaskill, his win probability will move up and he’ll likely bring the projected number of GOP seats up with him. Similarly, the fundamentals think that Texas Sen. Ted Cruz will gain a bit of support as the campaign rolls on. But maybe O’Rourke will continue to gain momentum, and the model will have to adjust the GOP win probability downwards.
Moreover, the fundamentals will become less powerful over time. So if McSally, O’Rourke, Hawley and others maintain their current polling (and Trump’s approval rating doesn’t take a dive) then they’ll likely gain steam as the election goes on.
Put simply, a fundamentals-defying candidate on either side (and in some cases fundamentals-defying boils down to just keeping polls where they are) in one race could really change what the odds end up looking like. The average simulation result gives the GOP 50.6 seats in January, but the range of outcomes is very wide.
