With the 2022 election cycle coming to a close, we’ve made our final picks for Senate, House, and Governor. In line with historical midterm precedent, the predictions suggest that tonight’s results will benefit Republicans — though not overwhelmingly, as we’ll discuss later on. Our forecast shows the GOP winning 50 Senate seats to the Democrats’ 49, just short of an outright majority with Georgia going to a December 6th runoff. In the House, Republicans can expect to flip the chamber with a clearer 234 to 201 majority. When it comes to governorships, we foresee Republicans with a 28 to 22 post-election edge. But we’d be lying if we told you we were supremely confident in any of those numbers.
With the acceleration of partisanship, the national environment has begun to play a bigger role than ever before in determining the results. The types of titanic overperformances that once let members like Collin Peterson represent Trump +30 seats as Democrats are all but gone, and while members like Mary Peltola, Joe Manchin, Susan Collins, Brian Fitzpatrick, and Jared Golden were quite common not so long ago, they are now the electoral exceptions. This means that the vast majority of the individual races in tonight’s elections will mostly be determined by the national environment. And to be frank, it is more difficult than ever to know what, precisely, we’re going to be looking at.
The consensus estimates generally have the Republicans winning the popular vote by anywhere between 1 and 4 points, if you aggregate partisan and nonpartisan public polling alike. FiveThirtyEight’s generic ballot aggregator has the popular vote at R+1.2, while their model estimates something closer to R+2.5. Meanwhile, our nonpartisan tracker, which excludes partisan polls from the aggregate, pegs the national environment at a statistical tie. The difference between R+4 and R+2 or between R+2 and R+0 may seem negligible, but it could actually change the outcome of 20 to 30 House elections and 2-3 Senate races, to say the least. In short, the consequences and implications of even a minor miss are huge.
The arguments for a Republican overperformance are clear and obvious. Joe Biden’s approval rating is at an abysmal 44%. Inflation is at 8%, and the stock market has taken a massive beating over the last year, denting the 401(k) retirement accounts of many middle-aged voters. The president’s party rarely gains seats in a midterm and usually loses a healthy amount, as the in-party lets their enthusiasm wane and dissatisfied independent voters recoil against the president and his agenda, choosing to elect a check on the executive branch two years after putting him into the White House. The Virginia and New Jersey elections from 2021 indicate that the Republican Party can make serious gains in blue areas, and that the bottom really could fall out for Democrats, especially considering that Biden’s approval rating is similar to what it was back then, all while gas is more expensive.
This, combined with the well-documented nature of polling underestimating Republicans of late, is why respected outlets like the New Yorker have run pieces suggesting that the GOP is poised for a blowout — virtually everybody who’s ever seen any previous midterm election thinks that the Republican party is more likely than not to have a good night, and party insiders themselves are supremely confident of scoring a rout. But that contradicts the polling data we do have, which indicates a tight race for the Senate and a close national environment overall in which a massive GOP majority isn’t a foregone conclusion. Moreover, as Nate Silver points out, insider sentiment isn’t really a convincing argument for anything, especially because they’ve been very wrong more often than not over the last decade (2012, 2016, and 2020 are good examples of this).
There’s good reason to think about the other argument, then, if only for intellectual rigor. Is it possible that we may be underestimating Democrats? The August special elections and the post-Dobbs primary turnout numbers indicate that we are on track for a much, much closer election than polling would indicate, and that Democrats may actually be underrated. Democratic turnout has been robust in primaries, special elections, and early voting. Since June’s Dobbs v. Jackson decision, which overturned Roe v. Wade, we have had five special elections in Nebraska, Minnesota, New York, and Alaska, and Democrats have overperformed both Joe Biden’s 2020 margins and polling in every single one of these races. If special elections were taken as the primary indicator, they would indicate a Democratic-leaning environment for November. If primary turnout was used to predict November, it would point to a very narrow Republican victory in which the House would be a tossup.
Moreover, our tracker of nonpartisan generic ballot polls has the national environment at a tie. If there is a red wave, some high quality outlets like ABC, NBC, and The Economist simply are not picking up on it. There’s a case to be made that this Democratic overperformance might then actually be what ends up happening — after all, when was the last time a party had robust primary turnout and overperformed in a ton of August specials, only to get blown out in November? We can’t think of any, and we daresay that few such cases exist.
With that said, a Democratic overperformance is not our forecasted outcome. Simply put, we believe that polling, while imperfect, is still likely to be the best general signal, as it provides a direct, consistent, and constantly-updated measurement of how voters intend to vote. Polling also suggests that Republicans are on track to retake the majority in one or both chambers. Our nonpartisan tracker shows that there has been a material shift of about four points towards Republicans since the late summer Democratic peak, and this probably has to do with a shifting issue environment and an acceleration of Republican campaigning and advertising. In August, Democrats had the airwaves to themselves amid a string of legislative and electoral victories, and so independents were mainly seeing Democratic ads. However, in the fall, Republicans began to ramp up the pressure, and voters responded accordingly, leading to the more GOP-leaning picture we have today. So if we had to describe the national environment that best matches our ratings, we’d probably pick somewhere in the ballpark of R+2.
But polls can be wrong, and it is a mistake to use them as the only metric, or to use them without the appropriate error bounds. As Bleeding Heartland’s Daniel Guild notes, the average generic ballot poll has missed by about three points per cycle since 1994. If polls underestimated Republicans massively, we could be looking at a 54-seat GOP Senate and a 250 seat GOP House majority. Due to the structure of the Senate map, this would make a filibuster-proof Republican majority a real future possibility, and it could become a centerpoint of the 2024 elections. Meanwhile, if polls underestimated Democrats to that degree, then the Democrats could end up at 52 seats and still control the House if things broke correctly for them. That is a very wide set of outcomes on a tight electoral map, and while the aforementioned scenarios are all unlikely, they shouldn’t be ignored in analysis either, especially when things currently rest on a knife’s edge.
So at the end of it all, we still believe Republicans are on track for a good night. But considering the environment was D+4.5 in 2020, an R+2 result would lead to an average shift of R+6.5 — those numbers would bode well for the GOP, as our forecast implies, but they would not be indicative of a red tsunami on par with a watershed midterm like 2010, or even a wave similar to 2014’s. This is not an environment that resembles either of those two elections, and the majority of the gains the GOP makes are likely to be done on persuasion. They may do quite well there, especially given Biden’s low approvals, but this is harder than making gains that come about as a result of one party not turning out to vote, which was a large component of Democratic losses in 2010 and 2014.
Ultimately, forecasting elections is impossibly difficult, and anyone who tells you with confidence that they already know what is going to happen tonight is probably overconfident. One thing we hope to have conveyed through our articles, our ratings, and our modeling is simply that elections are fantastically noisy, sparse, and difficult to model, and that doing so without data is near impossible, especially because we can fall into the trap of predicting our priors (or really just predicting the last election). To that end, we hope to have brought you a few new data-driven insights throughout the course of this cycle.
I’m a software engineer and a computer scientist (UC Berkeley class of 2019 BA, class of 2020 MS) who has an interest in machine learning, politics, and electoral data. I’m a partner at Split Ticket, handle our Senate races, and make many kinds of electoral models.
My name is Harrison Lavelle and I am a political analyst studying political science and international studies at the College of New Jersey. As a co-founder and partner at Split Ticket, I coordinate our House coverage. I write about a variety of electoral topics and produce political maps. Besides elections, my hobbies include music, history, language, aviation, and fitness.
Contact me at @HWLavelleMaps or firstname.lastname@example.org