If 2022 Is a ‘Normal’ Midterm by One Metric, Democrats Will Actually Gain Seats in the House
For those of you who are new to my page, my name is Noah Rudnick and over the last few years, I’ve been writing about elections, from the 1860s, to modern primaries to the future. If you remember anything, then it’s been the fundamental-based House models I made in 2018 and 2020. 2018 saw my prediction within a seat of the final total and the model caught the biggest upset of the cycle: Oklahoma’s 5th. Then in 2020, the model was the only quantitative one I could find publicly the correctly predicted that Republicans would gain seats in the House, rather than lose even more.
You know what happened next. The success of the 2018 and 2020 models catapulted me to star status and soon I was driving expensive and fast cars, high rolling in Vegas, and a constant featured panelist on cable news. This, I thought to myself, is why I got into modeling House elections in the first place.
There was a lot that went into the successful 2020 model that thought Republicans would capture seats and it was findings over the last three years, and I really wanted to summarize it all here for future modelers, IEs, campaigns etc. to look into and aggregate them all in one place. You can skip this section if you’re not interested but I’ll summarize it one point and then link the post about it:
§ This article looks at open seat and incumbency adjustments as well as the impact of late fundraising and spending on post-election adjustments: https://medium.com/@rudnicknoah/house-model-behind-the-curtain-potential-error-sources-3e15f37b0e33
§ Adjustments to the topline cash spent number so you make sure that you are capturing the campaign cash spent accurately — and an innovative look at how to include a scandal penalty: https://medium.com/@rudnicknoah/q3-house-model-update-part-2-navigating-the-wild-west-of-campaign-finance-1b6551fb190e
§ I do not have an official writeup on this yet but there was a significant correlation between overperformance and the age of the candidate. The younger a candidate is from their opponent, the more of a boost they receive. This holds especially well when looking at shifts from 2018 to 2020 and the age of the challenger in each. There’s a few theories for this in my opinion. People like change but younger candidates also will usually have more energy to hustle. They raise more cash and they could also attend more events. Being more tech savvy could lead to more effective digital campaigning as well. It’s tough to find a why, but if you aren’t including age of candidates in the model, you’re missing out on the biggest effect I’ve found besides money.
§ The spending difference accounts for about $400k in difference equals 1% but you need to reset the money every cycle starting at zero and set the candidate’s usual performance to how much they’ve spent before: https://medium.com/@rudnicknoah/the-hot-seat-arizonas-2nd-bf1fd823f15e
§ Performance is tied to ideology, where a more moderate candidate will get a bonus to overperformance but only in relation to how much more moderate they are from their opponent. There is also a possible gender effect: https://medium.com/@rudnicknoah/a-dimes-worth-of-moderation-84a1fa45d53b
From college, through a few jobs, and my marriage, it’s been a lot of fun to keep working on the model and I even have some future analyses using the 2018 and 2020 data even more. However, it’s probably time to hang up my hat and retire, and not build a model for the 2022 cycle. I wrote this months and months ago but I decided to publish after I read this recent piece by Nate Silver (my poker and pundit buddy from earlier) asking if this will be an “asterisk election,” or one outside the bounds of fundamentals. But I think if anything, the fundamentals point to the Democrats holding the House and if they don’t then it will be proof that fundamentals aren’t everything. It also brings up the biggest question: What should the baseline be?
2022 is a whole new beast for modeling House races and the past few cycles were some of the easiest to model. This is because we have all new maps and can not rely on nearly a decade of past races for that districts. Also because the maps are coming later, demographic information for the maps might be less exact and will not be compiled in an easy format to get that or past electoral information for the districts. You also have different incumbency effects. Before, I had a standard districtwide bump for incumbents but challengers that represented an area got a bonus for the share of the population they had repped before. But now with new maps, it would take a lot of data that I’m not sure if it’s out there to see how much of a bonus incumbents get in a new area based on media market overlap or other factors. I think it’s really hard to figure this out and encourage more to try but it could definitely skew results too much.
The last reason is that it is just not as much fun to do the kind of modeling I was doing before. My model worked slightly differently than others in that it created a partisan baseline in the district and all of the modeling was to predict split ticket voting for the House. It worked great in 2018 and well in 2020 but virtually all of my 2020 error was that split ticketing decreased. All of the factors I mentioned before still moved the needle but only got you so far and even in safe seats for each side we saw split ticketing decrease. As races align more and more to the national results, it is less interesting to me as an exercise to keep diving into the split ticket aspect of House races. It doesn’t mean that has disappeared completely and age, money, etc still matter on the margins but it feels like after finding those effects, I’m just not as motivated to spending too much time figuring out what matters for .4% of the vote.
Something I see a lot talks about historically or normally for midterms. There’s a small collection of articles saying the same thing in the hyperlinks of each word of this ending. I’ve posted about this before but we don’t have a lot of actual historical precedent. Sure, in all midterms there have been losses but modeling it has shown to me that a lot of what matters is on a seat by seat basis and not some grand macro-prediction. But even with that in mind, rising polarization and exposure matter a lot and when filtering down to past examples, we have only had 4 midterms in past memory that have come right after a redistricting cycle shown below, with a deviation too high to make any conclusions about. When talking about a normal or historical average, there really is no comparison for each election to the next and when you filter to even one of the broadest definitions, we only have 4 similar ones in history.
Now I didn’t want to go empty-handed on modeling this cycle and I really wanted to see what a House breakdown would look like in “normal” circumstances. So, I created a very simple polarization model. This is not intended to be a specific prediction of the outcome, or a hard set of rules but rather a look at why it’s tough to make comparisons even to recent history. In past large wave cycles in a midterm, a lot of the raw seat gain came down to exposure and the ability to ticket split of 20%, 30% or more.
The table below breaks out the seats into grouped buckets based on how much of the raw share the president got in that seat the last presidential election and the year of the midterm. The number in green is what percentage of the seats the opposition party won in that bracket. For example, in 2018 there was a Republican president, and Democrats during that midterm won 82.4% of seats that Trump got between 45 and 46%. Credit for the past data comes from DailyKos for past cycles here and a spreadsheet maintained by Drew Savicki and Aaron Moriak found here. The raw data for every seat and its category can be found here:
A few things stand out here, starting with the overall polarization. The opposition party just doesn’t win seats that would have a lot of split ticket voting like it used to be back in 2006 or 2010. On the other hand, they win virtually every seat that the losing presidential candidate took in the previous election and democrats in 2018 didn’t lose any seats that Clinton got less than 45% in. This is my main sticking point when people talk about how many seats flip in a ‘normal’ midterm, because it ignores this polarization and goes back to a time when a wider battlefield meant that exposure was much different.
The other point I want to make is that the 2018 Democratic performance was pretty weak, and it makes sense with the result that they also lost red senate seats. Part of this could have been due to the high third party in 2016, meaning that a Trump 48% seat could have still been a win by a few points. Democrats also generally have a weaker midterm coalition. I know there’s some theories that they’ll do better now with a new voting base among college educated voters and it could help but the suburbs really are still pretty split and reliance on nonwhite and younger voters could lead to a weaker performance overall than this model could capture. Even before I started working on this however, I published the 2020 House model with a blog post titled, “The House Model Doesn’t Think 2018 Was a Blue Wave. What Does That Mean For 2020?” It focused on the fact that Democrats really only won where the model thought they would based on where they spent and not on the back of a true wave, making them more vulnerable to Republican gains in 2020, which is exactly what happened. Now Republicans face the same challenge where they should not rely on a midterm wave, but really fight for areas with good candidates and cash with polarization increased.
So let’s take the simple polarization model to its natural conclusion by tallying up the number of seats in each category and assigning the share won in each year to that number and then tallying up the number of seats the out party wins.
The biggest point here is that Democrats are forecasted to hold even or pick up a few seats, and if Republicans want to win, they need to do better than the 2018 benchmark. This isn’t set in stone and even a polarized year like 2014 would deliver the House by a comfortable margin. Nor is polarization the only thing that matters as I’ll talk about later. But I think that if you keep taking this polarization to its conclusion, having more Biden won seats should deliver the House, this isn’t 2006 or 2010 anymore and if you need an “asterisk”, it would just be negative partisanship.
These new maps have seen a shift since the old ones, mostly with the number of safer but not totally safe Democratic seats up and more very safe Republican seats increasing. I think that while the earlier discourse talked about how these marginally pro-Biden seats made a big wave possible, it hasn’t really contended that if there isn’t a wave, it could make them a bulwark against Rs taking the House in a neutral environment or even a slightly positive Republican one.
So when I say that Democrats are favored to win the House in a “normal” year, does that mean it’s a foregone conclusion? Of course not! Jeff Hauser has a really good piece about how believing too much in fundamentals leads to people not trying or learning how to avoid things here: https://thebaffler.com/outbursts/the-do-nothing-discipline-hauser. All of the factors I outlined above about candidate age, past experience repping an area, and especially money can clearly make differences. 2018 and 2020 showed more than anything how important it was to have a real spending advantage. One good look at this is the Virginia governor’s race in 2021. Republicans ran Youngkin who acted as a more moderate candidate and had enough cash to spend in every market. Plus, there was strong evidence that there was differential turnout which means that Republicans showed up more than Democrats did, all things that could be repeated in House races in 2022.
The below chart shows the difference in Virginia gubernatorial elections for the margin and the difference from the margin in the presidential election the year before. Red dots indicate when a Republican president and those dots are above the 0% line which means Democrats overperformed and vice versa. 2013 and 2017 saw extreme polarization where the results looked close but 2021 saw Youngkin outperforming by levels similar to the Clinton and Bush era. In conclusion, simple polarization should be the default when we talk about a ‘normal’ election, but it should not be taken as the only possible variable, and anything can happen.
While this will be the extent of my addition to the 2022 House discourse, at least in a longer analytical form, I wish good luck to everyone attempting to model this election. Just from a lack of data standpoint, it’s going to be harder to get than the past few cycles and what I’ve learned is to always be improving and testing new aspects because it can always be better, and you never know which variable is suddenly a great explainer. Don’t be afraid to spend a lot of time on data collection and things that are still outstanding questions and get in the weeds on local news coverage, FEC reports, and past electoral data. The past few years creating and writing about the House model has been really rewarding and if you do it right by building something that provides good advice to stakeholders and strives to be better, you’ll feel better than aiming to get every one right. Stay open-minded and thanks to everyone who has followed what I’ve been writing on House races for the last 5 years. Here are some other House models you should check out instead:
Split/Ticket’s 2022 House Model created by Harrison Lavelle, Armin Thomas and foremost Lakshya Jain: https://split-ticket.org/category/house-2022/
Decision Desk HQ 2022 House Model: https://forecast.decisiondeskhq.com/house
CNalysis 2022 House model by Chaz Nuttycombe and Jack Kersting: https://projects.cnalysis.com/21-22/house
538 model by Nate Silver: https://projects.fivethirtyeight.com/2022-election-forecast/house/
