Stupid predictions No. 1: Democrats look to lose about 60 House seats this fall
Updated: Aug 25, 2022
Updated Aug. 25, 2022 to correct previous chart and text that omitted the fact that President Harry Truman lost party control of the House in his first midterm.
I think congressional elections matter less than they may have in the past. We are a deeply divided nation where a large congressional majority (and thus real change), especially in the Senate is very unlikely. So the frenzy over who will control the house by a few dozen votes is likely misplaced.
But congressional elections are easy for the media to cover and they are fun for math geeks like me, whether or not they matter, I love to analyze and make ridiculous guesses about how they’ll turn out.
In this post I'll look just at the U.S. House of Representatives. I’ll summarize House election history and some common indicators used to forecast the next election. I’ll also introduce my own measure that may more accurately predict the unique case of the first midterm election in a president’s term. That model projects a loss of 60 seats for the Democrats, which is a bit more than other indicators would suggest. That would still be the best midterm performance under a Democratic president since Carter in 1978, but would still put Republicans back in control of the House, with a 50-vote majority.
The era of consistent party control of the House is over
We’ve had a new dynamic in House elections since 1990. The chart below (which shows Democratic advantages as positives) shows mostly stable party control over the long run, with occasional short turnarounds. From 1898 (and a few years before) through 1930, Republicans dominated the House with a brief exception. After 1930, Democrats dominated for all but 2 of the next 60 years. Since then, we’ve seen less stability. Republicans have controlled the House just over half the time since 1990, but margins for both parties have been narrower than in the past. The four changes in control of the House in these 30 years are the most since the eight changes in the three decades leading up to the Civil War. Given the Democrats’ current margin of ten seats, another change seems likely this fall.
House elections are predictable, to an extent
Pundits and data nerds generally rely on three indicators to predict the results of House elections. This great article from The Economist explains all three and shows how both national and local factors affect individual races, and thus how Congress looks overall. Nationally, the generic ballot poll question (will you vote for a Democrat or Republican?), the president’s approval rating, and the results of both special elections and Virginia’s off-year election are decent predictors of the actual national party share of the House vote.
All of these indicators have their limits as predictors of House seat changes and party control, because the share of national vote doesn’t determine the change in the number of seats or which party controls the House. Members of both parties get elected with 70 or 80 percent of the votes in their districts. That runs up the national share of votes but the one additional seat counts the same as a seat won by one vote. The relationship between vote share and seats is even less predictable in years like this that follow congressional redistricting.
The chart below shows the relationship between generic ballot results in June and the actual seat change in November back to 2002. The blue line shows that a more Democratic generic ballot is generally related to an increase in Democratic seats. The formula in the lower right shows that a one percentage point increase in the generic ballot translates to six additional seats on average. If we plug this June’s average of generic vote polls (a 2.5 percentage point Republican lead) into the formula, we would expect a 43-seat Republican gain. The R-squared result, however, shows that the generic ballot explains only 36 percent of the change in seats. The boxed years show that two of the three first midterms this century have varied greatly from the average and suggest that a first midterm might be even more responsive to the generic ballot. In fact, if you just extended this line to the -2.5 percentage point Democratic disadvantage in the generic ballot, you'd expect the Dems to lose 153 seats!
With consistent results back to the 1940s, one might expect presidential approval to be a better predictor than 20 years of generic ballots, but it isn’t. June presidential approval has virtually no relationship to November’s house election results. The results from Virginia’s House of Delegates are somewhat better but still explain just 11 percent of variation (about a third as much as the generic ballot). For what it’s worth, June presidential approval suggests just a 4-seat Democratic loss and last year’s Democratic loss of 16 Virginia seats would predict a Democratic loss of 23 House seats.
First midterms are different
Midterm elections, particularly the first one for a newly elected president, are a special type of House election. Presidential elections usually provide a big increase in House seats for the winning party, but presidents actually in office always look worse than they did as candidates, so It’s typical for president’s parties to lose seats in the House. Since 1900, the presidential party has gained seats just twice, after a depression and a terrorist attack. That leaves 16 first midterms with party losses, ranging from 9 to 153 seats, with a median loss of 56. It’s less common for the president’s party to lose control of the House. Of the 19 first midterms since 1900, just five presidents (shown in blue in the chart below) have done so. But three of the last four presidents did lose House control after two years in office.
Precise prediction is more difficult for midterms, particularly first ones, because there’s less data. However, for the four first midterms for which I could find data, each of the last four Presidents has lost (or gained in Bush’s case) House seats when all three traditional indicators (generic ballot, presidential approval, and Virginia elections) are unfavorable (or favorable for Bush). The fact that all three are negative for Biden suggests Democrats will lose seats in November, but there isn’t enough data for them to predict how many.
Given the lack of data, prior first midterms may be the best predictor of the next one. The chart below shows the relationship between actual first midterm results and the average of all midterms up to that point. Again, the last four midterms are highlighted and they show that Democrats have under-performed against this model in three of the last four elections (as indicated by the actual results being below the trend line). This model, which still explains only 19 percent of the variation in the number of seats changing hands, predicts a 52-seat Democratic loss. Adjusting for the recent trend of large Democratic losses about a 60-vote loss results in my thoughtful and fearless (yet still stupid) prediction of this November’s House election.
What are the pros saying?
A lot of people get paid to forecast elections, but few attempt what I do here, which is project to the seat. Most prognosticators are more comfortable with ranges and probabilities. FiveThirtyEight.com, for example, runs data through thousands of simulations to come up with the chances either party will win a majority of the House. Today they give Republicans an 83 percent chance of flipping the House. Most of their simulations fall between a 40-seat Republican win and a 10-seat Democratic win (which translates to a range from no change to a 50-seat Republican gain).
The Cook Political Report looks at individual seats and ranks them as likely for or leaning to one party the other and tossups. Cook sees a range of Republican gains from 19 to 41 seats (which would leave Republicans in control of the House by 9 to 31 seats). They give Republicans a 29-seat advantage in likely seats and the same lead in likely plus leaning seats. Forecasts from the University of Virginia’s Larry Sabato are nearly identical to Cook’s.
So those with more data, analysts, and skills indicate that my forecast is about as favorable results as Republicans could get.
But this year’s election is different!
Recent Democratic action in Congress, the Ukraine War, inflation, and the Supreme Court’s abortion decision could all affect voters’ thinking in the coming months. However, today party loyalty rules everything and these events are more likely to confirm decisions we’ve already made than to change them. Every election has big events (wars, recessions, and scandals, for example) and none of this year's events feel out of the range of normal to me.
You can rest until November; I’ll tell you now how it came out
The result from the historical series of first presidential midterms is my favorite estimate for Biden’s first midterm. Unlike the other common indicators, this one is based just on first midterms, which are a different animal than other Congressional elections. It also has a few more data points than I’ve been able to put together for the other indicators. So I’ll be looking for a swing of 60 seats toward Republicans, for a 50-seat majority starting next January. I’ll also be watching for a frenetic lame duck session and crossing my figures that no Supreme Court vacancies appear before the election.
Stay tuned for Senate predictions (those are hard; I’m still sharpening my darts) and an absurdly early prediction for the 2024 presidential election!