For Eighty Years One Number Has Predicted the Midterm. It Never Once Asked Which Party.

The rule that does not care who you voted for

I have built puzzles for thirty years, and the first thing you learn is that a rule is only a rule if it holds in both directions. A theme that works when the answer helps one solver and breaks when it helps another is not a theme. It is a mistake wearing a costume. The grid does not know who is solving it. The rule applies or it does not.

American midterms have a rule like that, and it is one of the most durable numbers in our politics. Since 1946, the party that holds the White House has lost seats in the House in 18 of the last 20 midterm elections (American Presidency Project, 2022). The average loss across those contests is about 25 House seats. Only two presidents in that whole stretch, Bill Clinton in 1998 and George W. Bush in 2002, watched their party gain ground in a midterm. Everyone else paid.

I want to be plain about something before I go one sentence further. This is not a partisan observation. It is the opposite. The rule has fallen on both parties with the same weight, and any honest read has to show that.

The receipts, both colors

Look at the recent record and you see the ruler working on everyone. In 1994, Bill Clinton's Democrats lost 54 House seats. In 2006, George W. Bush's Republicans lost 30. In 2010, Barack Obama's Democrats lost 63, the largest midterm House loss since 1938. In 2018, Donald Trump's Republicans lost about 40 (American Presidency Project, 2022; Ballotpedia, 2026). Four midterms, two parties, one pattern. The number did not check anyone's registration.

Then read it all the way down, because the honest read is also the more interesting one. In 2022, Joe Biden's Democrats lost only about 9 House seats, far under the historical average, and held the Senate outright (American Presidency Project, 2022). The rule is a strong tendency, not a law of physics. It bends. Sometimes it bends hard. A president's party can beat the average, and twice in eighty years a president's party has beaten it so thoroughly that it added seats. So when someone tells you the midterm outcome is already written, the record says otherwise. The record says the tendency is real and the exceptions are real, and a careful reader holds both.

Where the number sharpens

Here is where the ruler gets its precision, and this is the part I would circle if I were grading it. The size of the midterm loss tracks the president's job approval, and the relationship is steep. Gallup ran the numbers across midterms from 1946 to 2014 and found that when a president's approval sits below 50 percent, the president's party loses an average of 37 House seats. When approval sits above 50 percent, the average loss is only 14 (Gallup, 2018).

Thirty-seven against fourteen. That gap is the whole game. It is why the single most watched number in an election year is not a horse-race margin at all. It is the approval line, because approval is the input that has moved the output for eighty years, no matter who was sitting behind the Resolute Desk.

What the number reads today

So where does the input sit right now, in the honest, non-partisan language this house insists on. In July 2026 the president's approval is running in the high 30s across the major public polls. Pew Research Center, in a survey conducted July 6 to 12, put approval at 34 percent against 64 percent disapproval (Pew Research Center, July 2026). A Washington Post and Ipsos poll conducted July 8 to 13 put it at 37 percent (Washington Post and Ipsos, July 2026). CNBC's July 8 to 12 survey had overall approval at 40 percent, with approval on the economy at 38 percent (CNBC, July 2026). A running average of qualifying national polls in late July had approval near 39 percent against 58 percent disapproval, with independents at 34 percent (US Polling Data, July 2026).

None of those numbers is a verdict. Every one of them is a snapshot of a mood in the middle of the summer before an election, and moods move. Approval in July is not approval in November, and I have watched enough late shifts to respect the distance between the two. But by the ruler that has held for eighty years, a president's party heading into a midterm with approval in the high 30s is in the range that historically produces losses closer to 37 seats than to 14. That is not a prediction. It is what the number has meant before, stated without heat.

The citizen's built-in check

I find something genuinely American in this pattern, and I do not mean that as a partisan can. The midterm is the country's mid-course correction, a scheduled moment where the people who were not on any ballot two years earlier get to weigh in on how it is going. It has disciplined presidents of both parties. It disciplined a Democrat in 1994 and 2010 and a Republican in 2006 and 2018. The mechanism does not love or hate an administration. It just asks the same question every two years and counts the answers.

US Political Rank keeps the long view on this in its report on what history says about the 2026 House, and on the record of the elections that reshaped the country in the most consequential midterms. The through-line in both is the same ruler I trust in a puzzle. It holds in every direction or it is not a rule. This one holds.

Read it down, then decide

My whole trade is refusing to publish an answer I have not read to the bottom. The gloomy version of this column would quote the 37-seat figure and stop. The rosy version would quote the 2022 exception and stop. Neither would be true, because the truth is that both are in the same dataset and a serious person has to carry them together.

The pattern is real, and it has taught both parties humility. The exceptions are real, and they keep the pattern from becoming a prophecy. The approval number is low right now, and the approval number can climb. What I would tell any solver applies here too. Do not grade the puzzle on one square. Read the whole grid, keep the same ruler for every entry, and let the count in November do what a count is for.