Who Shows Up

I. The Chair

On September 15, 2025, the United States House of Representatives voted on whether to table a motion to discharge H.J. Res. 117 from the Foreign Affairs Committee. I realize that sentence contains approximately four layers of procedural abstraction, so let me unpack it: a member of Congress had introduced a resolution to terminate a national emergency that the President had declared regarding Brazil. The resolution was sitting in committee. A different member moved to force it out of committee and onto the floor for a vote. A third member moved to table (which is Congress-speak for "kill") that motion. The House then voted on whether to kill the motion to force the vote on the resolution to terminate the emergency.¹

The vote was 200 to 198. Thirty-four members of Congress were not in the room.

A margin of two, in a chamber with thirty-four empty chairs, means the outcome was shaped not only by who voted but by who didn't. If three of those absent members had shown up and voted nay, the motion fails, the discharge proceeds, and the resolution reaches the floor for debate. Three people. Out of thirty-four who, for reasons the Congressional Record does not require anyone to disclose, were somewhere else.²

H.J. Res. 117 went back to the committee it came from, where (as of this writing) it remains, which is where resolutions go when they are no longer intended to become law but are not quite dead enough to admit it.

The attendance question found me accidentally. I was building a model that predicts congressional votes and decomposes each prediction into its explanatory parts (money, ideology, party loyalty, bill content; the whole apparatus is described here and here), and one of the 239 features is attendance rate, which I included because it seemed like it might correlate with something and because I am, at this point in the project, constitutionally incapable of seeing a variable and not throwing it into the feature set. The model did not find attendance particularly interesting. But when I broke it down by vote type (contested versus routine, where "contested" means the final margin was less than 10 percentage points), a pattern emerged that has nothing to do with money or ideology or any of the things the model was designed to measure, and everything to do with a question that is, I think, more fundamental: when the outcome is uncertain and your vote might actually matter, do you show up?

The answer depends on the size of the room.

II. The Bit

Before the data, I want to establish what, exactly, a member of Congress is being asked to do when they cast a vote, because the difficulty of the act is underappreciated in a way that matters for understanding why some people choose not to do it.

A roll-call vote is a binary. Yea or nay. One or zero. A single bit of information, in the Shannon sense³, applied to a document that might be 800 pages long, written over months by committee staff, amended fourteen times (twice at 11 PM by members who attached unrelated provisions because the legislative vehicle was moving and their provision needed a ride), and presented to you for a final decision that you will be associated with, publicly, permanently, in a way that will be quoted in mailers and attack ads and interest-group scorecards for as long as you hold office and possibly longer.

Your options are: yea, nay, or not voting.

There is no button for "this bill is 89% of what I want, and the 11% I don't want includes a provision that will hurt people I represent, and I have been awake since 2 AM trying to figure out which percentage wins."⁴

The contested votes (701 out of 1,087 in the 119th Congress) are the ones where this compression does the most damage. Nobody agonizes over a blowout. You agreed with the bill, you voted for it, you went to lunch. But the close votes, the ones that pass or fail by a handful, are close precisely because the underlying question was genuinely hard, and a genuinely hard question is exactly the kind of thing that a yes/no vote is worst at capturing. The harder the question, the more the binary lies about the answer.

Which means that not voting is, in a strange way, the only response that doesn't force you to flatten your position into one bit. The member who skips a contested vote might be lazy, or traveling, or sick, or fundraising in Scottsdale. They also might be the only person in the chamber taking the bill's complexity seriously enough to refuse to reduce it, which is a form of intellectual honesty that the system interprets as failure and that voters interpret as dereliction and that the data I built interprets as a gap in attendance. The two look identical in a spreadsheet, and the inability to tell them apart is itself worth noticing.

III. The Average

Here is the finding. Two numbers, followed by the reason neither of them matters.

For every sitting member of Congress, I computed two attendance rates: how often they voted on all roll calls, and how often they voted on contested roll calls (margin under 10%). The difference is the gap. Positive gap means you show up more for contested votes than routine ones. Negative gap means you show up less.

The average House member has a gap of +0.59%. The average senator has a gap of approximately zero⁵. These averages suggest that Congress is working roughly as designed. Boring. Reassuring. Move along.

Except the averages are concealing the variance. And the variance is the entire story.

The Senate's standard deviation on the attendance gap is 0.08%. The House's is 1.26. That is a ratio of 14.8 to 1.

The entire range of Senate behavior, from the senator with the most negative gap to the senator with the most positive, spans 0.64%. The interquartile range of the House (the middle 50% of members, not even the extremes) spans 1.00 percentage points. The Senate's full range fits inside the House's middle. Every senator, regardless of party, ideology, state, tenure, or committee load, attends contested votes and routine votes at essentially the same rate. The compression is total.

The House is a different institution entirely, despite sharing a building and a job description, and the behavioral spread is ~fifteen times wider. Some members attend contested votes at rates five points higher than their baseline. Others attend at rates one to two points lower. And the party breakdown is identical within each chamber: House Democrats average +0.57%, House Republicans +0.61%, Senate Democrats approximately zero, Senate Republicans approximately zero. Both parties in the House show the same average and the same wide spread. Both parties in the Senate show the same tight compression.⁶

The compression also doesn't scale down the way you'd expect it to if visibility were the mechanism. You might expect that members from small delegations (Wyoming's single representative, New Hampshire's two, who attend contested votes at identical rates) would behave more like senators, since they're visible within their state in a way that one of California's 52 members is not. There is a weak gradient in the expected direction: two-member delegations cluster tighter than thirty-member ones. But it's noisy, and California's 52-member delegation shows less internal variance than New Jersey's 12. The compression isn't about how many colleagues know your name. It's about the institution you're inside. A small delegation within a large chamber is still a large chamber.

Whatever is producing this pattern is not partisanship. Not ideology. Not delegation size. Not individual character.

One hundred people in a chamber. One absence shifts the margin by a full percentage point. Leadership runs a whip operation that can track, by name, who is present and who is missing. The architecture makes you visible.

Four hundred and thirty-five people. One absence shifts the margin by 0.23%. The whip operation is, by structural necessity, coarser. The architecture makes you anonymous. And when you are anonymous, you get to choose.

IV. The Scatter

So who chooses what?

The members at the positive extreme (attending contested votes at rates well above their baseline): John James, Republican from Michigan's 10th district, gap of +5.2%. Juan Ciscomani, Republican from Arizona's 6th, +3.6%. Hillary Scholten, Democrat from Michigan's 3rd, +2.8%. Josh Gottheimer, Democrat from New Jersey's 5th, +3.4%.

James represents a district with a Partisan Voter Index (PVI) of R+3. Ciscomani's is EVEN. Scholten's is D+4. Gottheimer's is D+2. Competitive districts, all of them. Members whose opponents are, right now, somewhere in an office building with a binder, cataloguing every vote and every absence and every possible angle of attack for the next cycle.⁷

For these members, contested-vote attendance is a survival strategy. You need the record. You need to be able to say, in a debate, "I was there, I voted, here's why." Ambiguity is a luxury you cannot afford when your district is split and every voter who switches is a voter you lose. Their positive gap looks less like civic virtue and more like electoral math.

Now the other end:

John Rutherford, Republican from Florida's 5th, gap of -1.3%. PVI: R+10. Max Miller, Republican from Ohio's 7th, -1.3%. PVI: R+5. Joe Courtney, Democrat from Connecticut's 2nd, -1.2%. PVI: D+8. Steve Womack, Republican from Arkansas's 3rd, -1.1%. PVI: R+13.⁸

Safe seats. Nobody is building a binder on Steve Womack's attendance record, because nobody is running against Steve Womack with any realistic expectation of winning, because Arkansas's 3rd district is thirteen points to the right of the national average and Steve Womack is a Republican, which means the general election is a formality and the primary is the only contest, and primaries do not typically turn on attendance rates for contested procedural votes about Brazilian sanctions.

The pattern is almost too clean: members who can afford to be absent are absent slightly more often; members who cannot afford to be absent are present slightly more often. One chamber's architecture permits both behaviors. The other chamber's architecture permits neither.

And then there is Lucy McBath.

McBath is a Democrat from Georgia's 6th district. PVI of D+25, which makes it one of the safest Democratic seats in the country after the 2023 court-ordered redistricting that redrew Georgia's congressional map. She is a gun violence prevention champion whose son Jordan Davis was murdered in 2012, and she flipped a competitive version of the district in 2018, then moved to a neighboring seat and beat a Democratic incumbent in a primary when redistricting redrew the lines under her. Her gap is -1.3%. Fourth most negative in the entire House. By the safe-seat logic, this is unremarkable — she's in a D+25 district, she can afford to miss votes. But that's not the story.

In March 2025, McBath suspended an exploratory campaign for governor to care for her husband, who was experiencing complications from cancer surgery. She is herself a two-time breast cancer survivor. The attendance gap in her record is not strategy. It is not complacency. It is a woman whose husband was sick, in a job that requires her to be in Washington while her life was happening somewhere else.

I bring this up because the spreadsheet does not know any of it. The spreadsheet sees -1.3% and files her next to Rutherford and Womack and Miller, in the same column, with the same sign, as if the number means the same thing for all of them. It does not. The gap is identical. The lives inside the gap are not. And the distance between the number and the person is, I think, the thing that everyone who works with data about human behavior should be required to stare at for a while, at regular intervals, as a kind of spiritual hygiene.

One more name, briefly: Brian Fitzpatrick, Republican from Pennsylvania's 1st. Bucks County. PVI of R+1. Chairman of the Problem Solvers Caucus, ranked the most bipartisan member of the House for five consecutive years, former FBI agent. His interest group ratings are a portrait of a man being drawn and quartered by his own coalition: LCV 51, Chamber of Commerce 89, AFL-CIO 56, NRA B-rated at 80.⁹ For every contested vote, at least one of those organizations will be disappointed in him, and several of their donors fund his campaign. His gap is -0.34%. He shows up for 98.4% of contested votes. The institution offers him an exit on every one of them, and his coalition gives him every reason to take it.

I find myself admiring this former FBI Special Agent, moderate Republican's representation of his constituents.

V. The Exit

But the question that matters here is not "why do some members skip contested votes?" Political scientists have studied that for decades; Eleanor Neff Powell at Wisconsin calls it "strategic abstention" and her book Where Money Matters in Congress maps the financial incentives in detail. A 2020 ProPublica analysis found 128 votes on bill passage where absent members later submitted statements claiming they would have voted against their own party's position. Fifty-six of those contained no clear reason for the absence.¹⁰ The individual behavior has a name and a literature.

The question the data raises that the literature hasn't answered is: why does one chamber produce a behavioral spread fifteen times wider than the other?

What strikes me is how absent this dimension is from the conversation about congressional accountability, which focuses almost entirely on how members vote and almost never on whether they show up to vote at all. We track every yea and nay. We score them. We build scorecards and ratings and indices (I built one). We do not, as a matter of routine, ask whether the member was in the room when it mattered.

And here is where the pattern becomes uncomfortable in a way I've been circling for several sections:

In almost every other profession, not showing up is a fireable offense for far lower stakes. In Congress, not showing up can be the most sophisticated move available. You can't be attacked for a vote you didn't cast. You can't be primaried over a position you never took. The vote that doesn't exist in your record is the vote that can't be used against you. And the chamber designed to be the closest expression of popular will, the one with two-year terms and proportional representation and all the structural features that were supposed to make it maximally answerable to the public, is the one where skipping the hard votes is, architecturally, the easiest to get away with. The system permits, and possibly rewards, the precise behavior that representation is supposed to prevent.

The magnitude matters, though. The most negative gap in the House is -1.6%. The fifth most negative is -1.2%. In absolute terms, that's three or four more missed contested votes than the baseline rate would predict over the course of a Congress. Three or four votes. Quiet. Marginal. The kind of thing that wouldn't survive a news cycle. But it concentrates in the chamber where individual absence is invisible, and it is nearly absent from the chamber where individual absence is not.¹¹

VI. Two Rooms

Given a free exit from a painful binary choice, a measurable number of people take it. This is not a scandal. It is a species. Put any population of human beings in a room large enough that individual absence becomes invisible and some percentage of them will find the door, not because they are worse than the ones who stay but because the door is there and the vote is hard and the record is permanent and the hallway is quiet. The Senate closes the door. The House leaves it open. And the data, which is very good at telling you what happened and very bad at telling you why, shows the consequences of open doors with a clarity that surprised me, given that almost nothing else in this project has been clear about anything.

The number 435 is not a law of physics. Neither is 100. Someone chose them, in different centuries, for reasons that had nothing to do with whether your representative would show up when the vote was close. But the choices have consequences that propagate forward through time in ways the choosers could not have anticipated, and one of those consequences is that the chamber designed to sit closest to the people (two-year terms, proportional representation, district lines drawn and redrawn to keep the representative tethered to the represented) is the chamber where, on the votes that matter most, the system permits you to not be there. The architecture that was supposed to produce accountability produces, in this one specific dimension, its opposite.

I keep thinking about McBath. Her gap is -1.3%, same as Rutherford's, same as Miller's. The spreadsheet files them in the same column, with the same sign, as if the number means the same thing for all of them. It does not. Rutherford is in a safe seat and chose the exit. Miller is in a safe seat and chose the exit. McBath is in a safe seat AND was in a hospital waiting room. The distance between their -1.3% and her -1.3% is the entire width of a human life. This is either a profound limitation of quantitative social science or the most obvious thing in the world, depending on whether you've ever been the person the number is about.¹²

Whether your representative was in the room for the last contested vote is a question you have almost certainly never thought to ask. I hadn't. The accountability infrastructure (the scorecards, the ratings, the interest group grades, the model I built) tracks every yea and nay with forensic precision. It does not track whether the member was present when the yea or nay was called.

We have built an elaborate, expensive, multi-institutional system for watching how Congress votes, and we forgot to check whether Congress showed up, especially when it mattered.¹³

Footnotes

1. The vote was technically Roll Call No. 265 in the 119th Congress, House, on the motion to table a motion to discharge H.J. Res. 117 from the Committee on Foreign Affairs. The national emergency in question was Executive Order 14323, signed July 30, 2025, in which the President declared that the Government of Brazil posed a threat to the United States and imposed sanctions on a Brazilian Supreme Court justice who was overseeing the criminal trial of former President Jair Bolsonaro. If you are wondering how sanctions against a Brazilian judge became a national emergency of the United States, the answer involves 50% tariffs, a diplomatic crisis, allegations of censorship and judicial overreach, and a former president's ankle bracelet, and is largely outside the scope of this piece, which is about attendance. I am going to exercise the same strategic restraint that several members of Congress exercised on September 15 and not wade into it. ↩

2. The Congressional Record notes absences but does not require explanations. In the House (unlike the Senate), members can submit "Personal Explanations" after the fact, stating how they would have voted, which has the practical effect of allowing a legislator to be simultaneously absent from a vote and on the record about it, a feat of bureaucratic quantum mechanics that I find both impressive and faintly enraging. ↩

3. Claude Shannon, 1948, "A Mathematical Theory of Communication." A bit is the minimum unit of information required to resolve a binary uncertainty. Shannon explicitly excluded meaning from his theory: "The semantic aspects of communication are irrelevant to the engineering problem." A yea/nay vote is a bit in the Shannon sense. Whether it captures the meaning of the legislator's position is, as Shannon would say, a different department. For more on Shannon and the Congressional Yield Index model, see The Plumbing. ↩

4. Ranked-choice voting, which has become the reform of choice (no pun intended, though I'm not giving it back) for people who find the binary reductive in elections, doesn't actually help here. RCV solves a candidate selection problem: it lets you express a preference ordering over multiple people competing for the same seat. The legislative binary is a different animal. You're not choosing between competing versions of a bill; you're choosing between a bill and the absence of a bill. There are mechanisms that could, in theory, address this. Line-item voting would let members vote on individual provisions. Weighted voting would let them register intensity ("yes, but weakly" versus "yes, and I'll fight for it"). The European Parliament uses a system where members can vote for, against, or abstain, and the abstention is formally counted as a distinct third category rather than lumped in with absence, which at minimum forces you to show up and announce that you're conflicted rather than letting you pretend you were at the dentist. But every alternative to the binary introduces its own compression. Line-item voting fragments legislative bargains that depend on logrolling (you vote for my provision, I vote for yours; separate the provisions and the deal collapses). Weighted voting creates a problem of strategic exaggeration that game theorists have been writing papers about since roughly the invention of game theorists. And none of these reforms are on the table in any serious way in the U.S. Congress, because the people who would have to vote to change the voting system are the people who won their seats under the current one, which is a structural irony so foundational it barely qualifies as irony anymore. ↩

5. For the statistically inclined: Senate gap mean is -0.005%, median is -0.022%, standard deviation is 0.085%. House gap mean is +0.589%, median is +0.288%, standard deviation is 1.257%. The Senate min-to-max range is 0.64 percentage points. The House interquartile range is 1.00 percentage points. One additional detail worth noting: final-passage votes (the substantive ones, the ones that become your permanent record) show the largest positive gap of any vote type, at +0.82%. Members make more effort for contested passage votes than for contested amendments or procedural motions. Nobody runs an attack ad about a missed amendment vote on a rule. Christopher Coons (D-DE), who chairs the Ethics Committee, ranks on Defense Appropriations, and sits on 19 committees total, has a gap of -0.07%, which tells you roughly how much individual circumstances matter in the Senate: they don't. ↩

6. This is the strongest evidence that the pattern is structural rather than behavioral. If the gap were driven by partisanship, you'd expect party means to diverge within the same chamber. They don't. If it were driven by ideology (moderates dodging contested votes because they're cross-pressured), you'd expect a correlation between DW-NOMINATE scores and the gap. There is one, but it's weak and inconsistent. Chamber membership predicts the variance of the gap more powerfully than any individual-level variable I tested. That's a result you get when the behavior is about what the institution permits, not about what the individual wants. ↩

7. Opposition research firms maintain databases of every roll call vote, every missed vote, and every procedural motion for every member of Congress. A House member in a swing district can assume, with near certainty, that any absence from a contested vote will appear in a future opposition research book, if not in a direct-mail piece with a bold red font and an unflattering photograph. The members at the positive end of the attendance distribution are not unaware of this. ↩

8. Joe Courtney's profile complicates the safe-seat narrative in a way worth a digression. He is the ranking member of the Armed Services Subcommittee on Seapower and Projection Forces, which is a title that sounds like it was generated by a military acronym algorithm but which translates, in practice, to "the person in Congress most responsible for submarines." His district includes the Electric Boat shipyard in Groton, Connecticut, where General Dynamics builds the Virginia-class submarine. Forty-five percent of his campaign funding comes from defense-related industries. He authored the two-submarines-per-year production policy. He is, by all accounts, deeply engaged in his work. His attendance gap of -1.2% probably reflects the reality that a ranking member of an active subcommittee faces scheduling conflicts with floor votes that a backbencher does not. Contested votes, incidentally, show no strong day-of-week clustering (they happen roughly equally Tuesday through Thursday), which eliminates the "everyone leaves for their districts on Friday" alternative explanation but does nothing to eliminate the committee-scheduling one, which is the explanation that keeps me from calling this pattern anything more definitive than "a gap in attendance that is consistent with multiple hypotheses, the most interesting of which I cannot distinguish from the most mundane." ↩

9. Fitzpatrick's DW-NOMINATE score of 0.141 places him at the leftmost edge of the Republican caucus. His district voted for Kamala Harris in 2024 despite being nominally R+1, making it one of three Republican-held seats in the country where the Democratic presidential candidate won. Democrats have been trying to unseat him since 2016 without success. His career missed-vote rate is 0.5%, better than the House median of 2.1%. The scorecard reads like a custody dispute: every organization that rates him has a plausible claim on his vote and a plausible grievance about the last time he voted against them. ↩

10. The combination (absent for the vote, on the record afterward with the popular position, no stated reason for the absence) is consistent with strategic abstention. It is also consistent with genuinely having had bad sushi and genuinely holding that position. Distinguishing these from roll-call data alone is, so far, not something anyone has cracked, which is the identification problem that haunts this piece. ↩

11. The companion post found that financial features account for 21.3% of the model's vote predictions, but the model can only measure votes that were cast. Every absence is a hole in the training data. A legislator who votes on 95% of contested roll calls and shows low financial influence has been tested. A legislator who votes on 91% and shows low financial influence might be clean, or might have removed themselves from the sample on the votes where the pressure was highest. The lobbying-to-bill linkage for the 119th Congress isn't built yet. When it is, the question becomes testable: do members disproportionately miss contested votes on bills where their donors have lobbied? Whether the answer, when it arrives, will be distinguishable from several equally plausible alternative explanations is something I have learned not to bet on. ↩

12. I think the distance is the most important thing in this piece. The model has 239 features and 2.3 million training votes and a defection AUC of 0.765, and it cannot distinguish a woman in a hospital waiting room from a man in a safe district who went home early. The spreadsheet, which is the tool I built to understand these people, cannot see it. I mention this because I think every person who has ever been represented by a number (which, if you have a credit score or a GPA or a blood pressure reading or a margin of victory, is you) already knows this, and does not need me to tell them, and I am telling them anyway, because the alternative is to end this essay on a data point, and I find that I can't. ↩

13. The model I built to study all of this is XGBoost, which is a gradient-boosted decision tree ensemble. Decision trees are, literally, trained to answer sequences of binary questions. The tool I used to understand Congress is, at its architectural core, a machine that processes the world in yea/nay splits. Whether this makes it particularly suited to the domain or particularly limited by the same constraint it's trying to measure is a question I find philosophically interesting and have decided not to think about too hard, because the alternative is an infinite regress, and I have a cat that needs feeding. ↩