A 2022 Midterm Election Forecast

Staff writer Asher Maxwell created his own election forecast for the upcoming 2022 U.S. midterm elections. On November 8, 2022, millions of Americans will be voting, all 435 seats in the House of Representatives and 35 of the 100 seats in the Senate are up for grabs. Check out the model below alongside Maxwell’s analysis as of October 27th, 2022 to understand who is likely to win these seats across the nation The model will continue to update itself until election day.

Note: The model updates to reflect new polling and decreased uncertainty daily. This publication’s commentary reflects the model’s forecast as of 2:00 pm on October 27, 2022.

The Race for the Senate

ASHER MAXWELL: The model favors Republicans to win control of the Senate, giving the GOP a 64% chance of winning a majority of 51 seats or greater. Currently, the model expects Republicans to win exactly 51 seats, but many races could reasonably go either way.

The model projects five races to have a vote margin within two percentage points. These five form the core of the Senate battleground. Upsets in favor of either side are possible in the other races, but we’re very likely to only see seat flips in Pennsylvania, Arizona, Nevada, Georgia, or New Hampshire. The only one of those races with a current Republican Senator is Pennsylvania, which also happens to be the closest race of the cycle. The model currently believes that the Democrat, Lieutenant Governor John Fetterman, will win the race by 0.15 percentage points, but we are only 51% confident that he will be the winner. The incumbent in the race, Pat Toomey, is not running for re-election, and the Republicans have nominated TV personality Dr. Mehmet Oz, who has significantly narrowed the race recently. 

The other four races have Democratic incumbent senators running for re-election. In two, Nevada and Georgia, we favor Republicans by about one percentage point, giving them about a 60% chance to win each seat. In Georgia, current Democratic incumbent Raphael Warnock is slightly trailing his political newcomer opponent, Herschel Walker. But there’s an important caveat for the Georgia race. If neither candidate gets over 50% on election day, then a runoff election will be held where only those names are on the ballot. We currently expect the race to go to a run-off, which adds further uncertainty to the outcome of the race and the Senate majority overall. In Nevada, Republican Adam Laxalt is currently favored to beat the incumbent Catherine Cortez-Masto. The race has been close all summer, but lately, polls and other indicators have begun to slightly favor Laxalt.

The final two of the five closest races are Arizona and New Hampshire, where the Democratic incumbents are both ahead of their Republican opponents. In Arizona, Democratic Senator Mark Kelly, who first won his seat two years ago, is facing Republican Blake Masters. In recent elections, Arizona has been a very competitive state, and in a Republican-leaning year, it should be a race that Republicans are favored in. But Senator Kelly is a strong incumbent running against a controversial opponent. A similar situation is playing out in New Hampshire, where Democratic incumbent Maggie Hassan is favored against retired General Don Bolduc. The race has tightened lately, but most experts believe Senator Hassan will be re-elected.

The last group of truly competitive races are Ohio, Wisconsin, North Carolina, and Colorado. In the first three states, the Republicans are favored to hold these seats with a probability of between 80-90%. In the last, the Democratic incumbent, Michael Bennet, has around a 86% chance of winning. 

It’s important to note just how uncertain the race for the Senate is. If Democrats win just four of the closest five seats, then they win the majority. If Republicans win more than one, they win control. But, most elections will turn out to favor one party across the board. So if Democrats are having a slightly better night than expected, they could easily win four of those seats or sweep all five. None of this accounts for the possibility that an upset will occur in one of the four races with 80-90% certainties. Given these uncertainties, we can say that Republicans are only generally favored. 

The Race for the House

The future of the House is more certain. Republicans have a 75.3% chance of winning a majority of seats in the House. Right now, our model predicts that the GOP will gain eighteen seats on a path to a sixteen-seat majority. 

However, just as with the Senate, there are many races for the House that are extremely close, enough so that a Democratic majority is certainly plausible. In order to win, Democrats would need to win all or almost all of the seats they are currently favored in and nearly sweep all of the close races.

The Republicans have been gaining in the projected seat margin in the past few weeks on the backs of their improving poll numbers. But this year's election is particularly difficult to predict due to redistricting. Many incumbents will be running in new districts with unfamiliar communities drawn into their constituency.  

The tipping point district, or the 218th most Republican-favored district, tells us which is the closest seat Republicans must win to get to the magic number of 218 seats. In other words, if Republicans were to win the House by one seat, the tipping point seat would be the closest seat that Republicans win. In our model, Colorado’s eighth congressional district is the tipping point district and the Republican, Barbara Kirkmeyer, is expected to beat by three percentage points the Democrat, Yadira Caraveo, in this district created in the latest round of redistricting. 

The closest race in our model is Minnesota’s second congressional district, where the Democratic incumbent is expected to beat her Republican opponent in a rematch from the 2020 cycle by merely fifteen hundredths of a percentage point. 

Methodology

My model uses several variables to predict the margin of each Senate and House race in the country. The model combines a calculated partisan score for each state and the predicted national environment. From there, district-specific factors like incumbency, district realignment, candidate quality and elasticity are added to our projected margin to arrive at the final prediction. I determined how the district has changed over time (realignment) and how responsive it is to changes in the national environment (elasticity) by looking at results in past election cycles for each district. I calculate my own value for realignment by looking at how the partisanship of a state or district has changed over time. The elasticity scores for a district that I use are taken from FiveThirtyEight’s data. 

I calculate incumbency advantage based on how much better incumbent candidates have tended to perform versus non-incumbent candidates in the past three election cycles (2016, 2018, and 2020). This results in about a three point advantage for incumbent Senate candidates and a one point advantage for incumbent House candidates. However, often there are candidates who tend to substantially outperform the average incumbency advantage. For those candidates, I adjust for candidate quality based on their fundraising and history of outperforming expectations. 

All of these factors are added together to calculate a projected final margin between the Democratic candidate’s and the Republican candidate’s vote share. I refer to this number as the fundamentals margin as it is based on the fundamental characteristics of the race and is not paying attention to some of the nuances of the race. 

For my House model, I only use the fundamentals margin, but the Senate model incorporates race-specific polling to try to account for some of the complexities of a race. I generate my own polling averages by adjusting and weighting polls for their historical accuracy and methodology. If a pollster has a history of being accurate, I will give it more weight in the average. And if a pollster has a history of being biased, I will adjust their polling margin for their past bias. These polling averages are then combined with the fundamentals calculation above for a final margin. The more and better polls a Senate race has had, the higher the model weights the polling average when combining it with the fundamentals margin. 

The decision to not incorporate district-level polling to the House model is simply due to how little polling there is and how inaccurate it tends to be. This does leave the model with a major blindspot as often it struggles to account for the strength and weaknesses of individual House candidates and the nuances of individual races.

From there, my model takes the projected margin and generates a probability for how likely each candidate is to win. These probabilities are based on how the model would have performed in past elections and how difficult to predict each district has historically been.

What Makes This Model Unique

My model differs from other models such as FiveThirtyEight or The Economist in several key ways. First, I attempt to adjust polls for past inaccuracies. If a pollster has a history of favoring one party’s candidates over another, my model corrects their polls for this cycle. My model also adjusts for the methodology of polls. For example, if a firm polls registered voters as opposed to likely voters, my model attempts to correct for registered voters’ tendency to favor Democratic candidates. 

Unlike other models, my forecast does not incorporate any expert ratings in order to avoid groupthink. FiveThirtyEight and The Economist will incorporate qualitative expert ratings into their predictions.

Finally, my model is generally less complex than the other models. For obvious reasons, I do not have the time to calculate and incorporate more complex variables. However, in my opinion, this simplicity can be beneficial. In the 2020 election, my model performed better than both FiveThirtyEight’s and The Economist’s model in the presidential elections, and performed about the same in the House and Senate elections. For the presidential election, my model was three points more accurate on average than the FiveThirtyEight model. For the congressional races, all three models showed Democrats ending up with 52 Senate seats (they won 50), and all three models showed Democrats winning somewhere between 239-246 seats on average (they won 222). 

Asher Maxwell is a staff writer for On the Record. He is currently a freshman in the College studying government from Nashville, Tennessee. Asher is particularly interested in elections, economics, and environmental policy.