Q3 House Model Update: 2 Weeks to Go and Democrats Have the Initiative

We are more than halfway through October and this is a huge update with all the candidates reporting fundraising numbers and in the middle of outside group spending. I also have a huge update to the interactive model with a link here:

https://1drv.ms/x/s!AvfJM5SfJyk7hVFaln5R43JkWE79?e=RI4zd0

For more about it and how to use it, read up on the 2020 launch writeup at this link: https://medium.com/@rudnicknoah/the-house-model-doesnt-think-2018-was-a-blue-wave-what-does-that-mean-for-2020-f3fc6c69bf5e.

Right now, Republicans are expected to have 205.2 seats when the dust clears, though later on I’ll discuss why this snapshot may be a high point for them. The map and the chart below show the current ratings in both map form and laid out in a bar chart.

Democrats clearly have the lead here with a lot more seats solidly in their direction. The number of Likely rather than Safe Rs shows that they may still be playing whack-a-mole and the high number of tilts and tossup seats where neither party has more than a 30% chance to win says to me that this race is still pretty fluid. The below chart spells out all of the seats that are rated not safe by the model.

As usual, you are able to adjust the presidential toplines and the spending for the model to get your own version, which you can then map using the instructions on the downloadable model. If you keep reading, I’ll run through some different scenarios and give some insights behind a few of the predictions.

What If Everyone Spends Their Warchest?

The race isn’t over yet, and so things are still very fluid! Just recently and not yet in the model was an announcement from Michael Bloomberg’s group of huge buys in TX-22 and TX-24, both races the model currently has at Tilts R but would now suddenly flip. The other would be that candidates can still raise and spend more. I also have a feature that has the cash on hand for each representative so I wanted to look at what happens if theoretically those were all split. In this version, the House gets even bluer and republicans are down to a estimated 202.9 seats on average. The chart below shows every race change that would happen in the model if their accounts went to zero. They will also be fundraising in the meantime.

What Does a Biden Landslide Look Like?

Playing around, you might notice something about this model. The statewide totals seem to reflect on a national polling environment of a Biden win of 6 or 7 which at the moment is on the lower end of the spectrum. It is where the generic ballot sits (though quality polling has been sparse there) but it’s where I personally think it could wind up. There is another possibility in the presidential race which would be matching the current national polling for a Biden landslide. 538 currently rates those odds at a whopping 37% so it’d be negligent to not address this.

In order to see what a landslide would look like, I plugged in the state totals downloadable from my friends at LeanTossup and also included their current presidential toplines in the interactive version. Under this landslide scenario, a bunch of seats move. In this setup, the GOP will only hold 193 seats which would be a net 7 seat loss from the 2016 total. It shakes a lot of Republican seats loose and has more tentative Democratic wins with 200 solid ones and a majority rated at least likely.

If you compare the LeanTossup presidential numbers on my model to their House model (https://leantossup.ca/us-house/), then you will notice that they have more seats flipping for a reason I’ll talk about in the next section.

Where Could the Model Go Wrong?

A model is not always right, especially ones that eschews polling for fundamental measurements. And there will be errors, just like in 2018, and several of the seat predictions will look incredulous. If the model has error, it will likely come down to one of three options. The first is missing something about the candidate or the race. I added in new factors for this cycle and will be watching a lot more off candidate characteristics. This is the best sort of error because if it’s systematic it should reveal new and interesting factors in a race and can be incorporated in next time. The second is a money error, which I’ll get into more in a piece that should be out this week.

The third error would be something not with the split ticket or spending portions, but with the top of the ticket. This is by far the most underdeveloped part of the model, where it predicts how well Trump and Biden will do up-ballot and is the benchmark for all of the ticket splitting. This sort of error will come around if Biden does better in a weird niche place like Northeast Pennsylvania. It could also occur if the 2016 trends become reversed in some non-uniform way. For instance, the model is fairly bullish on Republicans keeping Indiana 5th. We have received several Democratic internal polls that show Biden up double digits there but the model thinks it’s a tight race or a Trump win. They’re internals so I’m not so worried but if Trump were to lose but plunge more in say, suburban seats, that the model doesn’t account for from current trends that it will be off there. This won’t come until 2022 btu along with candidate characteristics I can see myself having to adjust for that based on performance and even spending to correct what could lead to several misses.

What are Some of the Stand Out Predictions?

Well, there are some weird ones here. While you can ask me any seat on Twitter, for the sake of brevity I will compare a handful of seats here to the conventional wisdom.

· AR-2: This one isn’t really a break from conventional wisdom but was when it started. In fact, I’m sure there will be people saying that Tilts R here may even underestimate the Democrat. But I’m not going to miss a chance to point out that this seat was not on the radar beforehand and was first picked up by the model at the very beginning with potential: https://medium.com/@rudnicknoah/the-hot-seat-2020-arkansas-2nd-district-77083972bdb7.

· CA-25: Katie Hill spent a ton of cash so in my opinion it was just natural reversion and a bad special environment to bring us Mike Garcia. That shouldn’t discount him because the model only has his race at Tilts Dem but it thinks the conventional wisdom has swung too far in this seat that Biden is likely to win by over 15. Internal polling shows it closer but that doesn’t match what it should be and this district under-polled Dems in 2018 and led to a miss then too.

· CA-48: This one now matches the general thinking but the model originally pegged it at Leans or Likely R. I think anything short of a double digit win here should still be considered a win for the model and its factors due to a herculean rescue effort by outside groups for Rouda. Republicans may miss an offensive opportunity here because even with better candidate spending their outside groups are being outspent $2.6 million to $7.5 million. That number is the most spent in total by Dem outside groups of any seat this cycle and so the model is back to favoring Democrats but it had identified this one early while others marked him as safe.

· CO-3: The model rates this seat Likely R and it feels sort of like CO-3 with the hype. Last cycle it also got the seat dead-on because it predicted a low Republican share but also high third party. This time around, the open seat factor plus a more extreme nominee make it tempting, but the new Republican nominee is also a lot younger which backtesting showed was important. They’re also getting a lot more cash with help from outside conservative groups. But that’s not all. Kevin McCarthy’s recent alliance with the House Freedom Caucus and embracing controversial nominees means that this race is also seeing over $650k from the NRCC and their associated SuperPACs. This is not one Republicans want to lose and so in the end, will probably see a decent victory.

· FL-27: The model does want to take the night away from Donna Shalala. While the top of the ticket looks horrendous for Republicans, the model credited a lot of Shalala’s win last cycle to spending millions more than her opponent, mostly because of cash she loaned herself. This time she has not yet committed the same resources and so receives a penalty. It’s a test of the big money assumption with this iteration of the model.

· ME-2/MI-8 etc.: These Trump district/close races this time are very similar. Still Trumpy districts where top of the ticket could be within a wide range and also both where lots of spending last time led to underwhelming wins the model did not like. That’s why this time it thinks the GOP still has a shot even though neither is facing an opponent with cash or much outside backing. If these wind up farther left, I may look at the error here and other similar situations to craft a set of guidelines for any race where one party has ‘given up’ and not diverted appropriate resources and dock it in the margin.

· NY-22 & NM-2: The model doesn’t overthink these. These are Trump seats. They’re both rematches so we already know how their personal data fits in. This time Republicans are the ones outspending unlike 2018. It thinks they both go red.

· NY-19: OH NY-19. I get it, this seat is going to plague me forever. I don’t really know what to believe in for Upstate NY where Siena non-partisan polls have Biden doing double digits beyond what the model thinks he gets and it seems to be messing with all of them up there. This could be another one where the aforementioned ‘given up’ metric works but for now it will be either be the most wild random correct guess or more likely, another stain on the model.

· OR-4, WI-3, PA-17: An abundance of caution has made the raters move these but we didn’t see the alignment in these places in 2018 and the model doesn’t see much reason why we will here. Several of them have real money but are facing even bigger opposition and the GOP candidate in OR-4 is spending a lot on fundraising to reach the big numbers. I’ll go more into that in the future, but it doesn’t seem to help and it’s why I focus on the disbursements part of the equation.

· SC-2: This one is fascinating to me. Still sort of a red seat, it has an old Tea Partier pitted against a new, young candidate. The model gives a big bonus for that age gap and the Democrat has been outspending like crazy. It could still be a stretch but the model rates it at the same place it rated SC-1 last cycle so I don’t think to would be a surprise. A crushing loss means it’s back to the drawing board to on a one size fits all approach to age.

· TX-2: I know, I know, this is a recurring problem that I will attempt to explain in a piece coming out later this week and is part of a larger group that I believe I’ll need to take action on before the election

· TX-7, TX-32: The model got these in 2018 late, and against some polling, because of last minute Bloomberg spends for very tight wins. Now they face tougher opponents and especially in Texas 7th, where Republicans are really spending. Texas is pretty scattershot in the model so walking that tightrope in the model on funding and past results can be tricky. I think others would be obliged to call these safe and with Biden winning by so much they might be, but I wouldn’t count them out quite yet.

Conclusion and Updates

In conclusion, the Democrats seem to have a solid grip on control of the House. But under the surface I still believe this could be a large playing field and nothing from a GOP seat pickup to Democrats flipping a bunch can be off the table. The number of Tilts and Leans alone means that we could be in for some surprises and I don’t believe as many incumbents are safe as some people think. In the upcoming days I will be publishing a more in-depth look at how money is a factor in the model with a change in how I register it as well as other fun features that aren’t public. Then I will be updating IE spending more regularly as it ramps up for a final push.

Update 10/22:

It’s not in the public update yet but a huge $28.5 million DCCC spend in Democratic seats has moved the model prediction several seats to an expecetd GOP 200.2 seats. The rating changes are reflected here: