What the Midterms Told Us

What do we actually know?

26

Comments

There is a mass of various posts and conversations on this site with the basic goal of trying to provide some key understanding of American political outcomes. This was the point, for example, in these two posts:

One of the things I was trying to get at was that once the dust settled after the midterms, there would be a lot of punditry trying to tell us how the outcomes provide a blueprint for the future of the Democratic Party (you know, the notion that a candidate “like X” is the way forward!).

But the evidence suggests, more or less in line with what I was saying months ago, that the issue for voters was Trump, not some specific kind of Democrat. This was the point I was trying to make in a series of posts that included the following:

Partly, this is because of the way midterm cycles go.

To wit, social science does not have the same kinds of precision that, say, chemistry does, but that does not mean we cannot systematically study and understand social phenomena. We know, for example, the observable patterns of midterm elections and the relationship of presidential popularity to those outcomes.

G. Elliott Morris provides the basics (and, to my point above about Trump, note that the title of his piece is, “Trump wasn’t on the ballot, but Trumpism was”).

By my calculations, we can expect the party in power to lose roughly 7.6 points of support from the last election, in terms of swing in vote margin over the intervening two years.

[…]

The president’s party lost House margin in 19 of the 20 midterms from 1946 to 2022, and the average loss was 7.6 points. The lone exception was 2002, when Republicans gained 4.6 points. This is about as ironclad a rule as they come.

So, as a minimum, it is pretty much as much of a guarantee that social science can provide that presidents lose seats at midterms. 2002 is the only outlier, but that is because 2001 was an outlier year for reasons that should be obvious.

Here are the data.

Now, of course, the GOP lost more than the 7.6 average. Morris breaks it down as follows:

Morris is planning ten more posts to go through all of this, but I think it is important to understand that 12.1 points of the swing towards Democrats is explainable by the average midterm shift (7.6) and Trump’s (un)popularity (4.5). That means, in simple terms, the things that a lot of pundits and consultant types are focused on, like candidate quality and the question of moderation/messaging, as well as turnout and other factors, are located in those 3.4 points.

This all conforms to the basic set of variables from the posts I linked above:

  1. Partisanship
  2. Prevailing national conditions
  3. Candidate quality
  4. Other factors, such as campaign spending

We start with a baseline of partisanship (more on that below) and then seek to understand deviations from preexisting partisan leans via the remaining factors. #2 is captured by the roughly 12.1 points above, with #3s and 4 accounting for 3.4 points.

The fairly uniform nature of the partisan shift across the country underscores that the main drivers were not about specific races in the main (certainly not in the way that provides an obvious blueprint).

As political scientist Seth Masket noted:

There is really no obvious formula for how Democrats should win going forward. One thing Democrats did well this year was to recruit candidates who could run strongly in their particular states. But even within regions, there was a great deal of variation. James Talarico ran in Texas as a white man preaching Christian love for strangers and those we disagree with; John Ossoff ran in Georgia as a white Jewish man who had no time for racist and antisemitic bullshit. They both won by double digits. Keisha Lance Bottoms got elected governor as a Black woman in Georgia; Scott Colum got elected senator as a Black man in Mississippi. All these things were considered unthinkable just a few years ago. Which of these is the obvious type of candidate for Democrats in 2028? Who knows? And I haven’t even gotten into the progressive Muslim in Michigan or the moderate minister in Kansas or the pro-choice pediatrician in South Carolina.

Indeed.

Also, indeed:

I doubt most Americans could tell you what Democrats stood for this year, no less who their leaders were. Yet that party just won massive record gains. It was enough to simply not be the party associated with Trump.

(Which is also why I don’t think that Congressional leadership matters as much as a lot of people think, especially electorally.)

Masket also provides an excellent illustration of why I place partisanship at the top of my variable list.

Below is a scatterplot showing the vote share for Biden in each state in 2020 paired up with the vote share for Harris in 2024. The correlation coefficient is .99. The red line represents the 2020 vote share, showing that Harris underperformed Biden almost uniformly across every state.

That is partisanship in clear evidence, but again, with one candidate underperforming another.

The same is true, but with candidates overperforming, in 2024 (the correlation coefficient is .94):

Masket’s piece is about some of his pet peeves concerning election coverage (they are peeves I share). One specific one that I agree with is that analysts need to stop talking about a “realignment.”

As Masket notes,

The term “realignment” gets thrown around a lot, but it’s generally understood to mean a durable shift in party coalitions. The shift of Black voters to Democrats and southern whites to Republicans, peaking in the 1960s, can be seen as a realignment. We’ve seen some minor realigning over the past decade-plus as Americans without a college degree have shifted from mostly supporting Democrats to mostly supporting Republicans, and vice versa for those with a college degree, but it’s been pretty gradual and hardly an upset of the whole party system.

1994 was a realignment election, insofar as we saw a substantial resorting of party affiliation around geography after that election that was durable. While there may have been some shifting in the two coalitions (like Hispanics who voted R in 2024 returning to the Ds), there is not a lot of evidence that some significant sorting of the coalitions has taken place.

This is why I am less convinced than some pundits that Texas, for example, is blue (if by that it means some permanent change to the state’s partisan lean) or that some magic formula has been unlocked by Democrats in the Midwest.

See also:

Early Mod v Prog take: DSA candidates and Blue Dogs both beat the Democratic Party average last night. Edge is small esp. once you add controls for demographics and other candidate characteristics (e.g., anybody would have crushed Paxton; Collins would have done well against any Maine Dem).

— G Elliott Morris (@gelliottmorris.com) 2026-11-05T04:15:13.634Z

We shall see, of course, but vibes ain’t data.

More to come, no doubt, but I think that the best way to understand the outcomes from earlier this month is that it was a profoundly anti-Trump election more than anything else by far.

26 responses to “What the Midterms Told Us”

  1. vibes ain’t data.

    Preach!

    I’m curious to hear your thoughts on how the 15.5 point swing is partitioned. If I understand the figure and commentary, it appears that these four components are considered to be orthogonal. I wonder about that, and whether/how interaction effects are at play.

  2. One point that I think gets overlooked in post-election commentary is that it often takes years, even decades, to fully understand any particular election’s long-term significance.

    Take, for example, 1964–LBJ’s landslide where he won every state in the country except five in the Deep South plus Goldwater’s home state of Arizona. Despite LBJ’s oft-quoted remark that his signing the Civil Rights Act had just lost the South to Republicans for a generation, he in fact won the majority of the South–just not the Deep South. He also overwhelmingly won the Northeast, though most of those states would almost immediately snap back to voting Republican after the election, and they didn’t transition to becoming fully Democratic until 30 years later. On the other hand, all those states he won in the Mountain West and Great Plains–the Dakotas, Nebraska, Kansas, Wyoming, Idaho, Utah, etc.–were basically one-offs. They were Republican strongholds going into the election, and they would remain so afterwards. His winning them was just a byproduct of one strong election cycle and had no future significance.

    That’s what’s so hard to determine about any election in the moment: how much of it is a reflection of the particular circumstances and how much reflects longer-term realignments.

    When people talk about 2010, they most often mention things like the passage of the ACA, the floundering economy still suffering the aftereffects of the Great Recession, and Obama’s declining popularity (though he wasn’t actually all that unpopular–his Gallup approval rating was dead-even on Labor Day). What rarely gets discussed is that a lot of the Republican wins were in states and districts that had been voting Republican at the presidential level for decades. Along with 1994 and 2014, it somewhat represented state and local races catching up to what had been going on nationally for a long time, mainly the shift of Southern, rural, and working-class whites away from the Democratic Party. For that matter, many of the gains Dems made in 2006 and 2008 could be seen in retrospect as a final gasp of a fraying coalition.

    Strong election victories mask these long-term shifts in voting, on the one hand temporarily keeping voters who are shifting away from the party within the fold, and on the other attracting voters from outside the party which could be a one-off but also could end up being permanent.

    When it comes to 2026, nobody knows the future significance of the seemingly unlikeliest of the Dem wins such as South Carolina, Kansas, and Mississippi. Of course it’s worth mentioning that the new Senators will next be up for reelection during presidential years (Sherrod Brown, due to its being a special election, will be up in two years). One thing I will say is that I don’t think it’s comparable to 2006 in holding on to voters who were rapidly shifting away from the party long-term. I think it reflects something new to the Trump era in particular.

    1
    1. That’s what’s so hard to determine about any election in the moment: how much of it is a reflection of the particular circumstances and how much reflects longer-term realignments.

      So many people think this is just a pendulum swing that will swing back. Not me, I think the GOP is Trumpified and will stay that way, so the GOP was also rebuked, not just Trump.

      My model analogy for this is the rebuke to Harding/Coolidge/Hoover that occurred in 1930, to be extended in 1932.

  3. Is there any polling data available yet on the voting groups breakdown and their motivations? Particularly in the seats flipped to Dem?

  4. Now do the huge realignment in the West since 1995.

  5. Michael Reynolds Avatar
    Michael Reynolds

    Vibes ain’t data, in politics vibes are more important than data. It’s all about the story.

    1. You know that, when Stephen Colbert was talking about “truthiness,” he was making a joke, right?

      1
      1. Like many of the best jokes, Colbert’s works because it’s true.

        1
      2. Michael Reynolds Avatar
        Michael Reynolds

        In the last 10 years not a single true datapoint has escaped from Donald Trump’s lips. He’s serving his second term as president because he can tell a story. It’s a stupid story, but when you’re trying to attract stupid voters that’s what you go with.

        There is great political power in the story that we are a blue tsunami, chosen by the voters to change, repair and revenge. And the game is ‘power,’ not ‘accuracy.’

        1. @Michael Reynolds: While I think I understand what you are saying, it mostly comes across to me as a non sequitur vis-a-vis the post (and the conversation I thought we were having).

    2. If you interpret “vibes” to confirm personal preferences, it’s entirely possible to end up falling flat on your face.
      A narrative you can sell to others is nice; a narrative you sell to yourself is risky.

      2
    3. @Michael Reynolds: It seems worth noting that I was able to use my understanding of the data to shape a pretty good picture of how things like Trump’s attempt to insert himself into the narrative would turn out (as well as sketching out a reasonable scenario for the elections as a whole).

      And if narrative were king, we would see far more volatile elections over time than we do. Again, I refer you to Masket’s two graphs above: that isn’t narrative-driven.

  6. Whether/how people vote –> vibes

    Understanding/characterizing voting behavior –> data

    1. @Mimai (and @Michael Reynolds): If the data can predict the vast majority of the vote without knowing either the vibes or the narrative, then I would suggest that the data are more powerful than vibes/narrative.

      Give me enough data about a given voter, and it is possible to very reasonably predict their voting habits with a reasonable margin of error.

      1. I think prediction and explanation are getting blended here. Demographics, party ID, and voting history can predict most votes without telling us why people hold those partisan loyalties, and that “why” often involves identity, perception, and narrative. Those things aren’t the opposite of data — they’re measurable.

        Presidential approval, which accounts for 4.5 points of the swing in your own breakdown, is essentially a quantified vibe, is it not?

        So I’d argue that vibes aren’t a competitor to data. They’re part of what we can measure and use data to understand.

        1
        1. @Mimai: I suppose the issue then becomes what we mean by “vibes.” I was referring more to the vibes deployed by pundits and consultant types.

  7. Michael Reynolds Avatar
    Michael Reynolds

    One of the disconnects we often have is that most of you are talking to each other; I assume a larger audience. But just for once, since we’re all just talking to each other, I’ll admit that there is almost always an element of outward-facing propaganda when I write. I’m not always trying to convince you people, I’m talking to ‘them,’ knowing of course that ‘them’ may not exist.

    1. @Michael Reynolds When I write my posts, I assume a broader audience.

      When I engage in one-to-one interactions in the comments, I guess I am a weirdo who assumes that we are actually talking to one another.

  8. This current blue tsunami may be strong enough to elect a Dem prez in ’28. If so, he’ll likely take a hit in the ’30 midterms and be replaced by a GOP after at most two terms. WIKI. dates this “Divided Government Era” back to 1969. Prior to that the norm at the national level was a “sun and moon” model of a dominant party and a weaker opposition party. Our current situation is the oddity.

    Republicans have literally screwed up everything they’ve touched except maybe border security. We should not be in a Divided Government situation. I would hope that Dem leadership would take advantage of the current situation to create a new era of Dem domination. But I wouldn’t bet on it.

    1. This current blue tsunami may be strong enough to elect a Dem prez in ’28.

      The issue is less whether this wave can be ridden (for example, 2010 did not lead to Obama losing reelection in 2012), but whether Trump remains unpopular, which I think is quite likely.

      And yes: the main space for any serious changes will be 2029 if the Ds have a trifecta. You can’t count on maintaining that trifecta into 2031.

      1. And yes: the main space for any serious changes will be 2029 if the Ds have a trifecta. You can’t count on maintaining that trifecta into 2031.

        When people keep seeing something they expect to keep seeing it. So for 40 years or so, or perhaps 80 I dunno, we have seen the pendulum swing back and forth, the parties alternating control – thus people expect that to continue.

        But I think not, I think after 2 more years of Trump, Stephen Miller, Musk/Thiel the GOP will be too discredited to come back. I think if you are waiting for the next GOP President or HOR you will have a long long wait. Senate also if DC becomes a state.

        Also, demographic trends are bleak for the GOP.

        ETA: Thus I support getting rid of the filibuster as I am not worried about what a GOP trifector might pass.

        Also relevent:

  9. @Steven L. Taylor
    Layne’s Law of Debate:
    Every debate is over the definition of a word. Or
    Every debate eventually degenerates into debating the definition of a word. Or
    Once a debate degenerates into debating the definition of a word, the debate is debatably over.

    Not just words though. Frames of reference too. And concept creep and shifting foci and and and. In other words, a typical online discussion. fwiw, I think there is consilience to be had… but who wants that!

    1
    1. @Mimai: There is consilience to be had, and this is my preference.

      These little boxes do make things difficult.

      From my perspective, it goes a bit like this: I write 100s (or 1000s) of words about subjects I tend to know better than the readers, and then I try to further explain myself and engage the best I can in the comments. I will confess to sometimes getting prickly.

      I actually thought I avoided too much descent into bickering over words when I explained what I meant by “vibes” in the OP rather than arguing with you about what you meant.

      1. Funny enough, I wasn’t arguing with you at all. Nor did I find any of your words prickly* in nature.

        I am a data guy myself, which is why my very first post was to signal boost your quip: “vibes ain’t data.”

        Your post was well articulated, and I agree with pretty much all of it — letter and spirit.

        Michael’s point about vibes seemed to address a related but different (and important) issue. And in typical style, it was too dismissive of data, which was unnecessary for the point to be made.

        And off we went…

        *fwiw, I sometimes get prickly on your behalf when I see your 1000s of words, backed up by extensive domain expertise, get hand-waved away or memory-holed entirely. Everyone is an armchair social scientist.

        1. @Mimai: I got some of the comment wires crossed, so my apologies.

          Let me say, however, that I appreciate your comment very much!

  10. I can imagine one realignment that may well bear out. That is a shift in Mexican-American voters in Texas to Democrats. I think that is a real possibility.

    This will have serious consequences. It will both shift the Electoral College balance of power to Democrats, and probably shift the Democratic party in a more religious, social conservative direction. I’m not sure I understand what it means for MfA, etc.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.