Why the President's Party Almost Always Loses the Midterm, and the Exceptions

If you want a law of American politics, here is one that has held for most of a century: the president's party loses the midterm. It does so after landslides and after squeakers, after good economies and bad ones, under popular presidents and unpopular ones. It does so with the regularity of the tide, and with about as much interest in what the cable news panel thinks about it.

There are exceptions, and they are the interesting part. Let me set out the pattern with numbers, test the explanations against it and see what the exceptions teach.

Define the terms

Four terms, defined briefly, so that nobody argues past anyone.

• A midterm election is a federal election held two years after a presidential election, in the middle of the president's four-year term. The House of Representatives is elected in full, and one-third of the Senate.
• The president's party is the party that holds the White House, whatever the voters think of the person.
• A seat change is the net number of House seats a party gains or loses, comparing the new House to the one before the election.
• A majority in the House is 218 of its 435 seats.

I use the House for the main count, because its 435 members all stand for election every two years and the numbers are cleaner. The Senate, where only a third of the seats are up, depends too heavily on which states happen to be on the ballot.

The pattern

Since 1934, there have been 23 midterm elections through 2022. In 20 of them, the president's party lost House seats. In three, it gained. Twenty out of 23 is 20 divided by 23, which is about 0.87, or roughly 87 percent. In other words, a president's party has lost House seats in about seven midterms out of every eight.

The losses range from modest to enormous. Here are some of the largest and most familiar, in the party's own terms.

Add those four together, using 55 for 1946. 55 plus 47 is 102. Add 54 and the total is 156. Add 63 and you reach 219. (Sources differ by a seat or so in some years, depending on how vacancies are counted, so read the total as roughly 219.) Four midterm elections, spread across 64 years, took about 219 seats from a president's party. That is about as many as the 218 that make a majority of the House. To put it differently, those four elections moved, in aggregate, a whole House of Representatives' worth of majority.

And here are the three exceptions in the House.

Look at the size of the gains. Nine, four or five, and eight: roughly 22 seats across all three elections. The exceptions are exceptions in direction and not in size, which is itself informative. The president's party has never had a midterm in which it gained as many seats as the four losses I listed took away. A party can sometimes hold its ground. It has never conquered new ground at the midterm.

The pattern also has a feature that the single-year count hides. The president's party often suffers worst when it has just won big. Seats won in a landslide are seats won on the margin, and margins do not last. That was the story of 1966, which I tell in Two Years After Goldwater, the 1966 Midterm That Put the Right Back in It, and the story of 1946, which is in Had Enough? The 1946 Midterm That Ended Sixteen Years in the Wilderness. The 1938 midterm, two years after the greatest landslide of the century, cost the president's party more than 70 House seats.

Three explanations

Political scientists have offered three main explanations, and I will state each at its strongest.

Surge and decline

The first explanation, associated with the political scientist Angus Campbell, is called surge and decline. In a presidential year, the contest is high-profile and turnout is high. The surge brings to the polls a group of marginal voters who are not regular participants and who lean toward the winning candidate. In the midterm, with no presidential race, those voters stay home. The electorate shrinks to the committed, the coattail seats revert, and the president's party declines.

The argument fits the seat-by-seat pattern, and it explains why the president's party loses especially after a big win: a bigger surge means a bigger decline. It does not explain why the other party would not equally lose in the next presidential year.

The referendum

The second explanation is that the midterm is a referendum on the president. Voters who are unhappy with the way things are going take it out on the president's party, and the party takes the hit regardless of the particular representative. Because voters are rarely perfectly happy, the president's party is usually in some trouble.

It fits the cases in which the president was unpopular. It has trouble with cases in which the president was popular and the party lost anyway.

Balancing

The third explanation is balancing: voters want a check on the president and prefer divided government. Even voters who approve of the president hedge by supporting the other party in Congress. On this view, the midterm is a deliberate correction, and the voter is trying to keep power from collecting in one place.

I find that one the most congenial, for a reason that will become clear in a moment. It requires that voters be sophisticated enough to think in terms of the whole government, not only the person. But it is hard to prove, because voters do not tell us why they voted, and a pollster's question about "balance" invites the answer.

What the exceptions suggest

Here is where the exceptions earn their keep. Whichever explanation you favor has to survive the three counterexamples, and that is a test I like because it does not require anyone to read minds.

1. Surge and decline predicts losses even for a popular president, so it struggles with the exceptions, unless something kept turnout up.
2. The referendum fits the exceptions well, if the president was popular in each case.
3. Balancing has the opposite problem: it predicts losses for popular presidents too.

Now the question is whether the president was popular in each of the three years, and my answer is that in each case the president held unusually strong standing by the standards of his time, a point that I will not put in percentages because I would rather not assert a figure I cannot cite to a source. That makes the referendum explanation the best fit for the exceptions, and the other two the best fit for the rule. My conclusion is a compound one: the pattern is mostly structural, and the exceptions are mostly about the president.

What the three exceptions had in common

Let me walk through each, and then set out what they share.

1934. Franklin Roosevelt's first midterm came in the depths of the Depression, two years after he and the Democrats had swept in. The New Deal's early measures were in place and the economy had begun to recover from its low point. The Republicans, who had lost both chambers in the preceding elections, had few seats to defend and no persuasive alternative. The Democrats gained nine House seats and nine Senate seats.

1998. The Clinton administration was in the middle of the scandal that led to impeachment, and the Republican congressional majority was pursuing it. The economy was strong. Voters appeared to react against what they saw as an overreaching effort, and the Democrats gained a few House seats, four or five depending on the count, an unusual result in the sixth year of a presidency. The House voted to impeach after the election, in December, and the effort ended in an acquittal in the Senate in early 1999.

2002. President George W. Bush's party gained eight House seats and also picked up Senate seats, a little more than a year after the attacks of September 11, 2001. National security dominated the campaign, and the Republicans held an advantage on it. The fight over the creation of the Department of Homeland Security was a defining issue, and the president campaigned hard.

Here is what they share.

1. The president had an unusually strong standing with the public. In each year, the president was, by most accounts, unusually well regarded by the public.
2. A national issue overshadowed local ones. The Depression, the impeachment fight and the aftermath of 9/11 were issues the whole country was thinking about, and they made the midterm a national election more than a collection of local ones.
3. The opposition was in a weak position on the issue that mattered. In 1934 the Republicans had no alternative to the New Deal that voters trusted. In 1998 the opposition party was seen as overreaching. In 2002 the opposition could not easily challenge the president on security.
4. The size of the effect was small. None of the three produced a gain larger than nine seats.

If you need a rule of thumb from this, it is that a president's party can resist the midterm penalty when the president is popular and the opposition's argument is weak. And the corollary is the part the commentators prefer to skip: the opposition's argument matters as much as the president's popularity. An opposition that has a strong, simple case against the governing party, as the minority did in 1946, 1966, 1994 and 2010, can overcome even a popular president's advantage. I make that argument at greater length in Five Conservative Comebacks Side by Side, and the Ingredient They All Shared.

The objection

A point-by-point post owes its readers the strongest rival view. Here it is.

The pattern is much weaker than you suggest. Twenty-three midterms is a small sample, and the explanations are conjectures that fit the numbers after the fact. The three exceptions may be nothing more than noise. A president's party might lose seats in most midterms simply because it holds more seats at the start, after winning the presidential year, and has more to lose. That is not a profound fact about voters; it is arithmetic.

That objection has force, and I accept the last part in particular. A party that has just won big has more marginal seats, and in that sense regression to the mean is built into the pattern. It is why I did not describe the pattern as a law of human nature. I called it a regularity.

But the objection does not erase the regularity; it names one of the mechanisms behind it. Whatever combination of surge and decline, referendum and balancing is at work, the observed result is that the president's party loses seats about 87 percent of the time. If the cause is arithmetic, then the arithmetic is the finding. A farmer does not need to know why the river floods in spring to build his barn on the hill.

What the pattern says about the Founders' design

Now the constitutional point, which is the reason I find the topic so interesting. Article I, Section 2 says that the House of Representatives shall be composed of members "chosen every second Year by the People of the several States." Two years, not four, not six.

The Founders did that on purpose. In Federalist 52, James Madison wrote: "As it is essential to liberty that the government in general should have a common interest with the people, so it is particularly essential that the branch of it under consideration should have an immediate dependence on, and an intimate sympathy with, the people. Frequent elections are unquestionably the only policy by which this dependence and sympathy can be effectually secured."

Put that beside the data. The Framers designed a chamber that would be returned to the voters at a rhythm that does not match the presidency. They did not plan for the midterm penalty, because they did not plan for parties as we know them. But the result is what Madison described: the House is the branch that most readily answers to the electorate's changing mood. When the voters feel that the government has gone too far, the two-year clock gives them a chance to say so before the habit sets in.

That is why I think the pattern vindicates the design. A government in which a landslide is permanent is a government in which the voters cannot correct a mistake. A government in which every victory is subject to a midterm review is one that must keep earning its majority. The president's party loses because the people are not required to wait four years to pass judgment, and that is a feature.

For the mechanics that follow a midterm, including how the old Congress behaves in its last weeks, see What Happens in a Lame-Duck Congress, and Why the 20th Amendment Shortened It, and for what a president can do when the voters send him a Congress of the other party, Divided Government Again, What the Constitution Gives a President Facing It. The numbers that decide who controls each chamber are set out in The Numbers That Run Congress, From 218 to 60 to 290 to 38 and Back, and the case of a party that turned the midterm penalty into a Senate majority is in From Lost Senate Races to a Majority, What 2014 Learned From 2010 and 2012.

How to use the pattern, and how not to

A word about handling a statistic like this. A base rate is not a forecast. If someone tells you that the president's party will lose seats in the coming midterm because it usually does, he is telling you where to start, not where to end. The starting point is a rate of about 87 percent. From there, the question is whether this particular president and this particular opposition look more like the rule or like one of the three exceptions, and you answer that by looking at the issues and the candidates, not at the calendar.

Here are the habits I would recommend to any reader who follows elections.

1. Start with the base rate. Know that the default is a loss for the president's party.
2. Ask which explanation applies. Is this a referendum, a correction or a decline in turnout?
3. Look for the three conditions of the exceptions. Is the president unusually strong, is a national issue dominant, and is the opposition's argument weak?
4. Weigh the size of any swing against the size of the majority. A party with 235 seats that loses 18 is down to 217, one seat short of a majority, and the arithmetic of 218 is unforgiving.
5. Distrust anyone who is certain. The sample is small and the world is large.

The case in brief

1. In 20 of the 23 midterms from 1934 through 2022, the president's party lost House seats, about 87 percent of the time, and four large losses (about 55, 47, 54 and 63 seats) add up to roughly 219.
2. The three modern exceptions in the House, 1934, 1998 and 2002, produced small gains of 9, 4 or 5, and 8 seats.
3. Surge and decline, the referendum and balancing each explain part of the pattern, and the exceptions fit the referendum explanation best.
4. The exceptions shared a popular president, a dominant national issue and a weak opposition argument.
5. The pattern vindicates the Founders' two-year House terms, since a chamber that answers to the voters at a different rhythm from the presidency gives them a regular chance to correct a majority.

What you can do: look up the House seat count for your party and its opponent at Congress.gov, then calculate how many seats would have to change hands for control to flip, and write the number down.