As the generic ballot closed over the course of the summer, the battle for the House of Representatives has moved into the forefront of political analysis. House races tend to develop late, and it is too soon to predict with specificity what the outcome is going to be. But we can probably set some reasonable bounds for expectations at this point.

Let’s start with the possibility that the Democrats might hold the House. To do so, they would have to lose fewer than five seats. The president’s party has managed to avoid losses in excess of four seats, five times since 1882 (before 1882 we did not have a single Election Day, or even Election Year, for Congress). One of those times, 1902, was an election that involved an increase in the size of the House; the president’s party lost ground in terms of its share of the House. So, out of the 34 midterm elections dating back to 1882, four have involved a president’s party losing fewer than five seats. That’s about 12% of the elections, which seems a reasonable starting point for expectations.

Does this year look like any of those elections? In 1934, the country was enjoying sharp economic growth in Franklin Delano Roosevelt’s first term. In 1962, the country was in an economic boom, and the incumbent president had a job approval rating of around 70%. In 1998, the country was again enjoying strong growth, and the president had a job approval of around 65% on Election Day. In 2002 growth was mediocre, but George W. Bush’s job approval was still stratospheric in the wake of the 9/11 attacks; Republicans also enjoyed a relatively favorable redistricting year.

Probably the closest example we might have is 1986, when Democrats only gained five seats in the midterm: Reagan’s job approval was 63% in late October, but 47% by early December (the next time Gallup polled). The Iran-Contra story broke shortly before the election; we don’t know where opinions of the president landed on Election Day. We do know that Republicans got blown out in the Senate; their House losses were probably minimized because Democrats already enjoyed solid majorities in the House.

That’s not this year. Unemployment is low, but economic growth and inflation remain problems in most voters’ minds. The president’s job approval is not in the 60s, it is in the low 40s. To be sure, this time could be different. But we should acknowledge that this has to be the argument: Things are fundamentally different than in elections we’ve had in the past 140 years. It could happen! In fact, I can think of reasons it might: Donald Trump and the January 6 committee may be transforming the election from a traditional referendum on the party in power to a choice between the parties, the Dobbs decision may supersize Democratic turnout, or polarization could severely constrict the range of outcomes. Things sometimes are different – things with even a 99.9% chance of not happening happen one time out of 1,000 – but overall, “this time it’s different” has a poor track record.

What about large Republican gains? Former House Speaker Newt Gingrich recently stated that he expected Republicans to gain 40-70 House seats. This generally wouldn’t be unusual for an election featuring a president with a 43% job approval rating, but it runs into some real challenges as well. The following chart shows the distribution of seats for the upcoming Congress, sorted by President Biden’s vote share.

There, the horizontal axis represents Biden’s vote share, while the bars represent the number of seats where Biden’s job approval falls within a given range. Since Biden won 52% of the vote nationally (with third parties excluded), that’s marked with a red line; it’s where seats probably shift from being Republican-favored to Democrat-favored in a neutral environment.

The outstanding feature of this chart should be fairly obvious: There is a significant valley right at the national average. Joe Biden only won five districts with 52% of the vote. This is in part due to gerrymandering. Both Republicans and Democrats hoped to avoid creating districts that were uncertain for their party and/or winnable for the other party.

One upshot of this is that in a neutral or close-to-neutral environment, there aren’t many winnable seats for either party. This will change over the course of the decade as party coalitions shift, but for now, absent a wave election for either party, the seats just aren’t there. Put differently, in the swingiest of swing seats where Biden won between 51% and 53% of the vote (rounded), there are just 19 seats. Of those seats, 10 are held by Democrats, seven are held by Republicans, and one is a newly created district.

Of course, there are some “gimmes” for Republicans and Democrats as well. There are currently 15 seats either held by Democrats (AK-AL, GA-6, TN-5, AZ-2, ME-2, FL-4, FL-13, FL-7, WI-3, OH-9, FL-15, PA-8 and TX-15) or newly drawn (TX-38, MT-1, FL-15) where Biden received less than 50% of the vote (rounded). While Democrats will likely win some of these this cycle, for the most part these are Republican targets regardless of the environment. At the same time, there are six seats that are either occupied by Republicans (IL-13, CA-22, CA-27) or newly created (OR-06, NC-14, TX-37) where Joe Biden won in excess of 54% of the vote (rounded).

We can see this better if we color-code our chart to take account of which party won the seats.

Republicans may hold some of those districts, and there are some eliminated districts to consider as well (e.g., Republicans had two districts eliminated in Illinois, one in West Virginia, etc). But the bottom line is that the basic distribution of districts right now isn’t particularly conducive to either major Republican or Democratic years.

To make this clearer, in a neutral year, where Republicans won all the districts where Joe Biden received 52% of the vote or less and lost all of the districts where he did better, they would win 224 seats. As an aside, if they won the seats where he received 51% of the vote or less, they’d narrowly claim control of the House, with 219 seats. And if Republicans did a bit better than a purely neutral environment, they’d receive 232 seats.

But one consequence in the near-elimination of swing districts is that there are substantial “levies” on either side of the valley. The levy on the Democrats’ side is particularly steep. So in a universe where Republicans win the popular vote by four points, sweeping all of the districts that Biden won with 54% of the vote or less, the levy would break and the Republican majority would jump from 232 seats to 245 seats. Winning by six points leads to smaller gains as we work down the other side of the levy, leading to a majority of 252 seats. Of course, these are approximations; Republicans would probably win some seats beyond Biden +10 and lose some where Biden performed worse in this scenario. This is just to illustrate how the playing field works.

There’s a similar “levy” on the Republican side as well, although the slope is much gentler. Recall that sweeping the districts that were 52% Biden or less yields 224 seats, while sweeping those that were 51% Biden or less yields 219 seats. Winning the 50%-Biden-or-less seats would give Republicans 206 seats. If the cutpoint were 49% Biden or less, Republicans would win 198 seats. Note the asymmetry. Even with a cutpoint of 47% Biden or less, Republicans would get 191 seats to the Democrats’ 244. In other words, because of the different placements of their levies, Republicans could lose the popular vote by around ten points and have more seats than Democrats would have if Republicans won the popular vote by six points.

We can view this graphically by taking the previous chart and limiting it to districts where Biden won between 48% and 56% (52% +/- 4%). Note that there are substantially more districts on the right side of the line than on the left.

So where does this leave us? It’s plausible for Democrats to hold the House, but it really would require some reordering of our understanding of what makes elections work. Even elections that look extreme or unusual, like 2002 or 1998, weren’t unusual; the presidents were just popular, much as Joe Biden is not.