2022 House Forecasting: Pre-Election Write-Up

The model can be found here.

The model methodology is outlined here.

The GitHub repository can be found here.

Election Eve. It's like Christmas Eve for nerds. While there is still one more model update to go before the model is finalized, Republicans have an 82% probability of winning the House of Representatives from the Democrats. The GOP is also currently projected to win 233 seats in total (with a standard deviation of 15 seats).

A word of caution here that there have been far fewer House polls this year than previous years. It's roughly half of that from 2018. These polls are very useful for the model in close elections, so we could see some surprising results tomorrow night. However, the GOP is still the strong favorite going into the night.

Is it impossible for the Democrats to win a majority? No. But based on where things stand, the probability of Democrats maintaining control of the House is roughly the same as rolling a 6 on a die. Meanwhile, the GOP wins if a 1 through 5 are rolled. You’d rather be the GOP at this point.

Below is the breakdown for the individual congressional races. If Republicans were to only win Solid, Likely, and Leaning GOP seats, they could retake the majority in the House of Representatives without winning a single Tossup district.

I'm extremely pleased with how the model performs. It's extremely more robust than the 2018 model. With that being said, I think there may be a few blind spots that will need work for the next iteration. Models aren't infalliable. They're merely simplifications of the real world used in an attempt to derive meaning about the real world. I'd like to walk through some specific cases where these simplifications may not be representative of the real world, and as I think you'll see, these blind spots are generally a wash because they work both for and against the parties in different situations.

ME-02: We'll start with a congressional district that vexed me in 2018 and I believe continues to do so. This is a unique case. Maine, like Alaska, has rank choice voting, meaning that if the candidate you vote for as your first choice does not make it as one of the top two candidates, they go down your list of preferences until they get to a candidate in that top two. However, unlike Alaska where they provide polling results based on these preferences, Maine does not. As a result, there’s an Independent candidate garnering about 9% of the vote, but there isn’t polling information to say who those 9% of voters will support next. My best guess is that the model is currently underestimating the incumbent Democratic candidate’s chances, with the model placing this as a Likely GOP district when it should probably be a Tossup. (ME-01 is not really effected because it’s not a competitive district.)

VA-02: This could also be applicable to districts like NV-04 and WA-08, but I'm going to use VA-02 as the poster child. The incumbent Democrat is sitting on a boatload of money - so much so that she could fund her entire race many times over. The GOP candidate is not underfunded and actually has a considerable amount of cash for a House race, but comparatively, it's nowhere close to the Democrat. I've put in some (vey high) caps on the financial data and used a logarithmic transformation to reflect the diminishing marginal return of cash in the race. However, I would like to further study and address this in future iterations of the model. This district is currently rated as Lean Democrat; it's probably close to a 50-50 Tossup.

TX-15, TX-28, TX-34: These are Rio Grande Valley districts where the GOP are trying to flip three historically Democratic districts. In fact, Mayra Flores (TX-34) has already done this in the special election earlier this year, but that was under the previous congressional district lines. This new district will be more favorable to the Democrat. My model is more bullish on these Republican Latina candidates than other models because I use the previous Presidential results when calculating my Leans while most use the previous two Presidential results. I found one year to be more accurate in the aggregate. Because this area of the country moved so far to the right in 2020, it will be a test to see if that pattern sticks. It's a bold bet by my model, and while I think all three have a good chance of winning on Tuesday, these districts should probably be closer to Tossup and Lean Republican rather than Lean and Likely Republican.

There are other districts that people might quibble about. However, I believe the highlighted examples above illustrate where my model might have some blind spots when the results start to roll in on Tuesday.

You need 218 seats for a majority. Let's walk through the easiest path to the majority for each party.

Republicans: The Republican path is a lot simpler than the Democrat path. The GOP simply needs to win the Solid, Likely, and Lean Republican districts, and that will get them to 225 seats. This is a seven seat cushion without even having to win a single Tossup district. Even if we subtract the model blind spots previously discussed of ME-02 and the Rio Grande Valley districts, the GOP still has a three seat cushion. To put it simply, the GOP has a lot of combinations on the table for them to get to 218 seats.

At this point for the GOP, it becomes a game as to how high they can run the table. Can they pick off seats in NV, OR, and IL that were drawn as gerrymanders but can backfire on the Democrats if the generic ballot gets into the GOP +6-7% range? (The generic ballot is currently just under GOP +5% right now, so it's within the realm of possibilities currently.) Can they pick up Democrat leaning districts that don't have much polling, such as CT-05, MN-02, NY-18, OR-04, WA-08, and CO-07? There have also been rumblings about Democrats spending heavily in districts that voted almost 20 points for Biden in 2020, like NY-25, PA-12, and CA-26. Though unlikely, if the GOP has an outstanding night, maybe they can pick off one. Finally, the crown jewel for the GOP would be to knock off chair of the DCCC, Sean Patrick Maloney, in NY-17, where he finds himself in a surprisingly competitive race.

Democrats: The path to 218 is possible for Democrats, but the possibilities look really narrow when you compare to the possibilities the Republicans have to get to 218. Let's assume Democrats win all Solid, Likely, and Lean Democrat districts as well as all of the Tossups. That gets Democrats to 210 seats, meaning they will have to pick off at least eight more seats from the seats currently favoring Republicans. Let's give Democrats ME-02 and the three Rio Grande Valley districts for reasons previously discussed, bringing their total up to 214. The easiest path I can see for Democrats would be:

1: RI-02: The GOP has a very good recruit with Allan Fung, who has been leading in the polling. Let's assume that his polling edge has been overstated and that because this is a very Democrat friendly district, the Democrat takes this one.

2: OH-09: This is a GOP leaning district, but the GOP has some scandals. Let's assume the Democratic incumbent can hang on here.

3: MI-08: There's been virtually no polling here. There's a decent Democratic incumbent who has considerable funding. Let's assume he can pull this one out.

4: PA-17: This is actually a D+1 district. If Democrats are to the point where they're getting 217 seats, there's probably been a polling error understating Democrat support. If that's the case, they should probably win this district.

This will get you to the required 218 seats. Some other possibilities are: OR-05, CA-27, NM-02, IA-03, NJ-07, and NE-02 as a real long shot. There really isn't much room for error for the Democrats, and they're fighting historic headwinds.

The 90th percentile for Democrats in the model currently is 223 seats. All the districts I just listed would put the Democrats at 224. Is it do-able? Sure. Is it also as likely as the GOP getting over 253 seats, which would give them their largest majority in the House of Representatives since the 1920s? Yes. Yes, it is.

Everybody wants to know how the night is going to play out. While I would caution that changes are not felt uniformally across the nation, let's look at some races that can guage where the night might be heading.

We'll probably get our first look after 7PM. By this time, VT, IN, KY, VA, SC, GA, and most of FL will start reporting results. The GOP doesn't need VA-02 to get to a majority, but if it looks like they're going to win VA-02, then they're probably on their way to the majority with a bit of cushion. If it looks like the GOP could win IN-01 or VA-07, then we're getting into territory where the GOP is probably going to have a very good night. If the GOP can win VA-10 or GA-02, we're probably looking at a major wave for the GOP where they cruise to a historic number of House seats for the post-War Era of American politics.

There really aren't many good districts by 7PM that would be a good sign Democrats may be able to hold the majority. The best case for Democrats at this point would be to lead in VA-02 and wait for most of the East Coast to start reporting at 8PM. If it looks like Democrats can hold onto all of their New England seats that look to be in trouble (RI-02, ME-02, CT-05) and can start showing positive signs in Tossups (PA-07, PA-08, NC-13), then there may be some hope for Democrats. If the Democrats get a majority, it is going to be a slim majority, so it's not going to be determined early in the night. It probably won't even look possible until the next day since it can take western states like CA a long time to count ballots.

One data point I'm interested in is some FL districts. The FL early vote has looked absolutely awful for the Democrats. While we need to proceed with caution about looking at the early vote, Democrats losing a district that seems quite safe in FL could spell doom for their night. FL counts votes fast. If the GOP were to pick off an unexpected district, I would suspect it to be FL-09 or FL-23. If either of these are lost by the Democrats or even if the races look really close, that could be a very bad for the FL Democratic Party. I wouldn't want to extrapolate to the national level with this because FL looks to unique when it comes to the early vote, but it wouldn't be positive for Democrats overall.

Finally, I've created a no tossup prediction using 270towin. While I did use my model to inform some of my picks, these are really gut decisions. I'm sure my own biases taint these results as my pick is 241 GOP - 194 DEM, which is considerably higher than my model's average expected seat count of 233 GOP - 202 DEM. However, this was a fun exercise to look back after the election and see how wrong I was.

So, get the popcorn ready, settle in for a long night, and may the odds be ever in your favor.