# An Early Midterm Forecast

Political observers seem to be treating a Republican takeover of the US House next year as an inevitability. And maybe that's true, but I wanted to apply a bit of simple modeling to it to see if we can get some handle on how likely that is to happen. The quick version is that yeah, a Republican takeover is quite likely. But please bear with me, as there are some important details and caveats here.

To come up with a forecast, I'm using a model that predicts House seats loss for the President's party using a measure of economic growth -- real per capita disposable income. (Specifically, the growth in real per capita disposable income from the second quarter of the year prior to the election to the second quarter of the election year.) This works somewhat better than GDP for predicting midterm elections, although no measure of the economy is perfect. I also control for the president's Gallup approval rating as of Labor Day in the election year and the number of House seats the president's party currently controls. The data range from 1950 to 2018. You can view the data here.

The scatterplot below shows the relationship between real disposable income growth and House seats gained by the President's party. It's a noisy relationship, but a positive one nonetheless. And it reminds us that the President's party almost always loses seats in midterms, with the two exceptions being the unusually popular presidencies of George W. Bush in 2002 and Bill Clinton in 1998.

Adding in presidential approval and the number of seats held by the majority party adds to the explanatory power of the model. This model has worked reasonably well in the past, predicting Democrats gaining roughly 45 seats in 2018 (they gained 40) and 40 seats in 2006 (they gained 31). The R-squared for this regression is .67. The formula generated is:

Predicted seat gains for president's party = 5.65*RDI growth - .22*Number of seats held by president's party + 1.23*Gallup approval - 51.8

So now the question is, what are the right values to plug in for 2022? Obviously, we're just making educated guesses here. Many economic forecasts are fairly bullish for 2022. However, they're more likely to predict GDP than RDI. This Fed forecast suggests GDP growth of 4% for 2022. Let's just assume for a moment that RDI follows suit. (It's usually more like 2-3% growth in recent decades.)

What will Biden's approval rating look like by Labor Day? That's also really hard to know. But we know that Biden's approval has been around 43 percent for the past two months, and that's roughly where Trump and Obama were for a lot of their time in office. It definitely could go higher (or lower), but let's just assume for a moment that Biden will still be at 43% by Labor Day.

Plugging these values in, we get Democrats losing 25 seats. Given that they only hold 221 right now, that means Republicans taking control by a healthy margin. Of course, there are large error margins around all these estimates, and we really don't know what conditions will look like by next fall.

Here's a better way to look at all this. It shows a range of likely election outcomes given different levels of economic growth and presidential approval. Notably, Democrats are expected to lose House seats in almost all situations, except for presidential approval over 65% and economic growth over 4%, the combination of which is not particularly likely to show up.

In addition to the uncertainty about the variables in the model, there are a lot of other variables outside the model. Importantly, we don't know what the Covid pandemic will look like. The pandemic is likely responsible for some of Biden's recent unpopularity, and if there are significant outbreaks in the summer or fall of 2020, that could make the electoral situation worse for Democrats. On the other hand, if the better-case scenarios for the Omicron variant are true and this pandemic has mutated into a less deadly, more manageable (if more contagious) form, that could help Biden and Democrats a good deal by next year.

Another thing that's difficult to model in here is vote suppression and other anti-democratic activity that has become particularly acute in Republican-run states since the 2020 election. How many elections is that likely to affect in 2022? Probably not many, but it's very hard to know. Redistricting is also outside this model. Aggressive redistricting likely helped boost Republican congressional seats in the 2002 midterms, and might again next year.

So, TL/DR, midterm forecast models are inherently noisy, and the one for 2022 may be noisier than usual. It's still too early to make anything like a precise forecast. But the number of possible futures in which Democrats hold onto the House are very, very few.