The Loneliest Point

On designing (academic) institutions for imaginary people

This is a story about designing institutions for people who don't exist, and what happens to the people who do. It starts with fighter jets, but it ends with graduate students and postdocs and tenure committees, all sitting in cockpits built for someone else.

On a single day in the late 1940s, seventeen United States Air Force pilots crashed their planes. The planes were mechanically sound. The pilots were experienced. The weather was fine. For months, the Air Force had been watching its pilots lose control of aircraft that, by every diagnostic measure, worked perfectly. They checked the electronics, the hydraulics, the engines. Everything checked out. The official explanation, applied with the institutional creativity of a military bureaucracy, was "pilot error." Seventeen times in one day.

The cockpits of those planes had been designed in 1926, when army engineers took measurements from a few hundred male pilots, averaged each body dimension, and used the result as a blueprint. Seat height, pedal distance, windshield placement, helmet shape: all calibrated to a single set of numbers. For three decades, every American military cockpit was built to these specifications. By the late 1940s, with jets replacing propeller planes and the margin for error shrinking accordingly, someone suggested that maybe the pilots had simply gotten bigger. The solution, naturally, was to re-measure everyone.

The resulting study, conducted at Wright Air Force Base in Ohio in 1950, was the most comprehensive anthropometric survey of pilots ever attempted. Researchers recorded 140 bodily dimensions from 4,063 men: the length of their thumbs, the circumference of their chests, the distance between their eyes and their ears, and dozens of other measurements that collectively described the shape of a person in extraordinary detail. When they were done, they had a magnificent new average, computed with far more data and far more precision than the 1926 version. The thinking was obvious: build a better average, build a better cockpit.

A twenty-three-year-old researcher named Gilbert Daniels was not so sure. Most people encounter Daniels through Todd Rose's 2016 book The End of Average, which tells the cockpit story in full, but the original source is a 1952 Air Force technical note with the beautifully understated title "The 'Average Man'?" As an undergraduate at Harvard, Daniels had measured the hands of 250 students and then compared each person's hand to the average hand. None of them matched. Not one. These were similar people at the same university, and the average hand described exactly zero actual hands. "When I left Harvard," he later said, "it was clear to me that if you wanted to design something for an individual human being, the average was completely useless."

I encountered Daniels's study for the first time in grad school, in the context of a completely different problem. We were trying to define the "typical" galaxy in a photometric survey by averaging across a dozen parameters: luminosity, color index, effective radius, Sérsic index, and so on. A professor asked how many galaxies in the sample actually matched the average profile across all parameters simultaneously. The answer was very close to zero, which was confusing to some people in the room until someone pointed out that it was basically the same arithmetic as the Air Force cockpit study. I remember thinking this was a cute analogy. I did not expect to spend the better part of the next decade watching the academic institutions I was so familiar with make the same mistake.

Daniels did the obvious thing, which in an institutional context means the thing nobody else thought to do. He took the ten physical dimensions most relevant for cockpit design (height, chest circumference, sleeve length, and so on) and defined the "average pilot" as anyone whose measurements fell within the middle 30 percent of the range on each dimension. This is generous. The middle 30 percent is a wide band. If you're anywhere near the middle on any one measurement, you count. The number of pilots who qualified as average on all ten dimensions was zero. Out of 4,063. When he loosened the requirement to just three dimensions, any three, fewer than 3.5 percent qualified. Pick any three measurements you like. Neck circumference, thigh circumference, and wrist circumference. Height, arm length, and torso length. It doesn't matter which three. Virtually nobody is average on all of them simultaneously.

The arithmetic is not subtle. If you have a 30 percent chance of being "average" on any single dimension, and the dimensions are roughly independent of one another, then the probability of being average on all ten is 0.3 multiplied by itself ten times. That's 0.3 to the tenth power. Which is 0.0000059. Six in a million. The more dimensions you add, the worse it gets. Twenty dimensions gives you 0.3 to the twentieth, which is about three in a hundred billion. On the Discworld, million-to-one chances famously crop up nine times out of ten. This is not a million-to-one chance. This is six in a million, which apparently doesn't qualify.

Mathematicians call this concentration of measure, though you don't need the formal machinery to see it. In one dimension, a Gaussian distribution (the bell curve, the thing you've seen on every statistics textbook ever printed) piles most of its probability near the mean. That's the whole point of the bell curve. The average is where the action is. But add dimensions and something counterintuitive happens. In high-dimensional space, the probability mass migrates away from the center and concentrates in a thin shell at a fixed distance from the mean. In a hundred dimensions, a typical sample from a Gaussian distribution is very far from the mean. Not in any one direction, necessarily. But in the aggregate, across all dimensions simultaneously, it's out near the surface of a sphere. The center, the place where the average lives, is practically empty.

This is one of those results that feels wrong the first time you encounter it. You've spent years building intuition in one and two and three dimensions, where the peak of the distribution is the most populated place, where "average" and "typical" mean roughly the same thing, where the center of the bell curve is the center of the action. Then you step into higher dimensions and all of that breaks. The peak is still there. It's still the single most probable point. But the volume near the peak becomes vanishingly small compared to the volume slightly further out, and volume wins. It's like finding out that the summit of a mountain is technically the highest point but the mountain has no top, just a hollow cone where the peak used to be, and everyone is standing on the rim.

Think of it this way. In one dimension, the bell curve is like a hill. Most people are near the top. If you step away from the peak, you can only go left or right. Two directions. In two dimensions, the hill becomes a mound, and you can step away in any compass direction: left, right, forward, backward, and every diagonal in between. A whole circle of directions. In three dimensions, you can also go up and down, and every diagonal through three-dimensional space. A whole sphere of directions. Each new dimension adds more ways to be away from the center without adding any new ways to be at it, because the center is always just one point. By the time you reach a hundred dimensions, the number of directions pointing away from the peak is so astronomically larger than the volume near the peak itself that the probability mass drains out to a thin shell, like the skin of an orange, at a specific distance from the center. The center itself, the point where every coordinate is at its mean value, is a desert. If you dropped a pin on a random point drawn from a hundred-dimensional Gaussian, the odds of it landing near the middle are essentially zero. Not small. Not unlikely. Essentially zero.

The average is the loneliest point in the space.

The Air Force, to its credit, solved the problem within a few years. The solution required no new mathematics, no new materials science, no breakthrough in engineering. They made the seats adjustable. They made the pedals adjustable. Helmet straps, harnesses, everything that had been fixed got a range of motion. They designed cockpits that could accommodate pilots from the 5th to the 95th percentile on each dimension independently. Pilot performance improved immediately. Crash rates dropped. The adjustable seat, which is now in every car you've ever sat in, exists because a twenty-three-year-old lieutenant did a simple calculation that the entire military establishment had failed to do for three decades. This is usually where the story ends, with the adjustable seat. A design parable. Don Norman probably has a chapter on it. But the cockpit problem didn't stay in aviation. It migrated into every system that compresses people into a single profile, and the place where it continues to do a lot of damage without anyone noticing is the one I know best.

Take a complex, high-dimensional population, compress it to a single point (the average), design your system around that point, then act confused when the system fits nobody. The Air Force had the excuse of doing this in 1926, before high-dimensional statistics was a standard part of anyone's training. We do not have that excuse. What we have instead is a collection of institutions that find the average too useful to give up, not because it describes their populations accurately but because it makes those populations administrable. One cockpit is cheaper than an adjustable one. One number per institution is cheaper than a multi-dimensional evaluation. James C. Scott called this legibility: the process by which institutions simplify messy, high-dimensional populations into categories they can see and manage, not because the simplification is accurate but because it makes governance possible. The average is the purest form of legibility, a single point standing in for a cloud of thousands, and it outlived its accuracy by about seven decades. And nowhere has it outlived it longer, or more destructively, than in the academic pipeline.

The institutional scaffolding of that pipeline runs on compressed numbers. The U.S. News & World Report college rankings, for instance, have spent four decades compressing American higher education into a single ordered list. The methodology uses up to 17 factors: graduation rates, faculty salaries, class sizes, peer assessment scores, student debt levels, and various other metrics that get weighted, summed, and squeezed into a single number per institution. Princeton is first. MIT is second. Harvard is third. This has been roughly true for years, through multiple methodology overhauls that U.S. News periodically trumpets as "the most significant changes in our history," and which reliably leave the top ten almost untouched.

In 2022, a Columbia math professor named Michael Thaddeus published an 11,000-word analysis showing that several data points Columbia had submitted to U.S. News were, to use his precise language, "inaccurate, dubious, or highly misleading." Columbia had claimed that 83 percent of its undergraduate classes enrolled fewer than 20 students. Thaddeus pulled the university's own class directory and found the real number was somewhere between 63 and 67 percent. Every discrepancy was in Columbia's favor. Columbia dropped from 2nd to 18th in the next cycle, then withdrew from the rankings, citing the "outsized influence" of the list. It had not been concerned about this outsized influence when it was ranked 2nd. Temple University's business school dean was convicted of fraud for submitting false data. Yale's law school left. Multiple medical schools at Harvard, Stanford, Columbia, and Penn pulled out. The institutions best positioned to evaluate the methodology concluded it was useless, and U.S. News responded by saying it would rank them anyway using publicly available data.

Thaddeus himself noted it: if a university can drop sixteen places in a single cycle based on how it counts class sizes, the problem is the ranking methodology, not the university. But broken or not, the rankings set the stage on which academic careers are built, because departments at highly ranked universities get more applicants, more funding, and more prestige, and the people who work inside those departments inherit the same compression. The university sits in its cockpit. The scholar sits in another one inside it. Cockpits all the way down.

The scholar's cockpit starts at the entrance. I know, because I sat in it.

You leave home at eighteen to go to college. Maybe you move one state over, maybe one country over. You finish, and if you're good enough and stubborn enough, you get into a PhD program. It's in a different city. You move again. Your stipend is $27,000 (mine was £15,000 a year, in a country where the word "stipend" is apparently a polite way of saying "less than minimum wage but you should be grateful"). Your rent is $1,800 a month. You do the arithmetic and it doesn't work, but you do it anyway because this is what you've wanted since you first understood what research was. Your friends from college are making twice your salary in jobs that don't require them to live in a specific city, but you're doing important work, and the stipend is temporary. Five years, maybe six. You'll manage. Your parents help, if they can. If they can't, you take on debt that a doctoral stipend was supposedly designed to prevent. Meanwhile, you teach undergraduates, run experiments, write papers, attend conferences you can barely afford, and try not to think about the fact that you are twenty-six years old and your net worth is negative.

The conferences deserve their own sentence, because they are a small-scale model of the whole system. You fly to a city you've never been to, sleep on a friend's couch or four to a hotel room because the conference hotel costs more than your weekly take-home, print a poster at whatever copy shop is still open at midnight because your university's print shop closes at five, and then stand next to it for two hours hoping that someone on a hiring committee wanders by. This is called "networking." You spend money you don't have to be visible to people who might one day decide your future. The people with family money do this more easily. The people without it either do it at a cost the system prefers not to calculate or they don't do it at all, and then someone on a hiring committee notes that their CV has fewer conference presentations. The system produces the disparity and then measures it.

And then there is the person standing between you and the degree. Your supervisor controls your funding, your authorship, your access to equipment, your letters of recommendation, and in many cases the timeline on which you are permitted to graduate. If your supervisor expects you in the lab on weekends, you are in the lab on weekends. If they email you at eleven on a Thursday night and expect a reply by midnight, you reply by midnight. Your holiday can wait because a paper deadline can't. Your work shows up at a conference without your name on it and you learn about it afterward. You don't complain, because the person you'd complain to is the person you'd be complaining about, and that person writes the letter that determines whether you have a career. Structurally, this is a tremendous concentration of power in a single individual with minimal institutional oversight, and the results are what you'd predict. A 2019 global survey by Nature of over 6,300 PhD students found that 21 percent reported experiencing bullying, and the most frequently reported perpetrators were supervisors. A separate study surveying over 2,000 people in academic science found that the primary targets of abusive supervision were graduate students and postdocs, and the primary perpetrators were PIs and department heads.

For women, the power asymmetry produces an additional, predictable outcome. You learn which professors not to be alone with. You hear about it from the older students when you arrive, informally, in the way that things which should be institutional knowledge get transmitted as gossip instead. You adjust your behavior around certain colleagues, avoid certain office hours, learn to read certain tones of voice. A 2018 report by the National Academies of Sciences, Engineering, and Medicine found that between 20 and 50 percent of female students in science, engineering, and medicine reported sexual harassment by faculty or staff. The most common form was not what makes the news. It was what the report called "gender harassment": the comments about what women can and can't do, jokes you're expected to laugh at, the persistent small acts of communicating that you are a guest in someone else's field. Women who reported it were retaliated against roughly three-quarters of the time. The system's proposed solution, naturally, was a formal reporting mechanism that the data showed almost nobody used because the data also showed it made things worse.

I should be specific about where I'm standing while I say all this. My PhD supervisors were good, and so is my postdoc PI. They gave me room to work, treated me like a person, and never made me feel like I owed them my weekends. I didn't fully understand how unusual this was until I watched friends and colleagues go through the version where it isn't. The midnight emails, the missing authorship, the holiday that couldn't wait, the letter of recommendation wielded like a leash. I heard these stories more often than any other kind, always from people too dependent on the person in question to say any of it out loud. The system worked for me in the ways that depend on individual people. Where it didn't was structural: the moves, the distance from family, the two-body arithmetic, the cost of living in cities the stipend wasn't designed for. No amount of supervisor decency fixes those, because they're built into the cockpit.

But assume you get through it. You keep your head down, or you're lucky, or your supervisor happens to be a decent person, which many of them are, though the decent ones are not the ones who shape the incentive structure. You finish the PhD. You are now qualified for a postdoc, which is a two-to-four-year position in another city, possibly another country, paying more than your graduate stipend, though in some countries the word "more" is doing a lot of heavy lifting. You move again. Your partner, if you have one, either follows you or doesn't. The relationship survives or it doesn't. You are told this is normal.

The field calls this the "two-body problem," borrowing the term from orbital mechanics, which is either charmingly self-aware or deeply oblivious depending on your mood. In physics, the two-body problem has an exact solution. In academia, it doesn't. Your partner has a career that doesn't transfer, or they're also an academic and you're both applying to different job markets in different countries, each of you hoping the other lands somewhere within driving distance. Or they follow you, give up their job, start over in a new city where they know nobody, and then you do it again two years later. Or you try long-distance, which works for a while, the way anything works when you're young enough to mistake endurance for sustainability. The system has a word for the person who follows an academic across borders and starts over repeatedly. The word is "trailing spouse." It is not a term of endearment.

You start thinking about having children, and you do the math on that too: the salary, the lack of parental leave, the fact that the position ends in two years and the next one might be in a different time zone. You defer. Then you do another postdoc, because the job market demands it. You move again. You are thirty-two and you have moved four times in eight years. This is described in hiring committee conversations as "flexibility."

If you survive the postdoc years, and you're still standing, you get a tenure-track position. You move again. You set up a lab, write grants, teach courses, publish papers, and prove your worth in six years. The grant writing alone is a full-time job grafted onto the two full-time jobs you already have. Federal funding rates in most physics and engineering programs hover around 15 to 25 percent, which means that for every successful proposal you are statistically expected to write four or five that go nowhere. Each one takes weeks, sometimes months: you write the narrative, assemble the budget, coordinate with collaborators whose schedules don't align with yours, and format everything to the agency's specifications (which changed since last year), and you submit it knowing that the most likely outcome is a polite rejection and a set of reviewer comments that may or may not be coherent. Then you do it again. The hours you spend writing unsuccessful proposals don't appear on your CV. They don't count toward tenure. They are the invisible scaffolding on which the visible research rests, and the system pretends they don't exist.

Each stage of this process has assumed the same person: someone young, white, male, unattached, geographically rootless, financially subsidized by sources the system prefers not to name. The people who can't afford to play this game don't show up in the attrition statistics, because they never entered the pipeline. They're not a data point. They're a ghost.

And then comes the tenure decision itself, which is its own cockpit. The "ideal" tenure-track faculty member is supposed to be simultaneously excellent at research, teaching, and service, with research weighted most heavily but teaching and service also supposedly mattering. That's at least three dimensions, and, just as with the pilots, very few actual humans fit the average profile across all of them. A 2012 study tracked nearly 3,000 of the people who made it this far, assistant professors in science and engineering at 14 U.S. universities, and found that about 64 percent were promoted to associate professor at the same institution. A third of the survivors didn't survive. Women were disproportionately likely to end up in adjunct positions. Men were more likely to either get tenure or leave academia entirely.

Scholars of color face what's often called the "service tax": they're expected to be the diversity representative on committees, mentor students of color, engage with the local community, and serve as the visible proof that the department takes inclusion seriously. This labor is real. It is also not what the tenure cockpit is calibrated to measure. When tenure review comes around, these scholars are evaluated on the same research-heavy rubric as their white colleagues who spent those same years with their doors closed, writing papers. The rubric doesn't have a line item for "kept the department from being embarrassingly homogeneous." The advice given to junior faculty of color, frequently and without apparent irony, is to stop doing the very service work their departments rely on them to do. "Your colleagues don't want you here. You are not a good fit. But you will do a great job elsewhere." That's allegedly an actual quote from a dean to a scholar denied tenure at Penn State. The dean presumably said this with a straight face.

The system assumes, at every stage, that you are a single point in a low-dimensional space. You have research output. You have a publication list. You have an h-index. These are the coordinates the cockpit was built for. The dimensions it doesn't measure, the ones that determine whether you can actually survive inside it, are treated as external to the problem. Whether you have a partner whose career exists. Whether you have a family that exists outside your department. Or a disability that makes the constant relocation an ongoing medical crisis, because moving countries means finding new doctors and new prescriptions and navigating new insurance systems every two years, assuming the postdoc comes with insurance at all. Whether you are a person and not a CV. The pipeline was designed for someone who is, in every dimension except research output, alone.

In 2006, two theoretical physicists named Oliver Rosten and Francis Dolan started two-year research fellowships at the Dublin Institute for Advanced Studies. They became friends. They had a shared sense of humor and, as Rosten later put it, a similar outlook on the absurdity of existence. Dolan suffered from depression. Two-year fellowships are short; in theoretical physics, that gives you a little over a year to produce something before the next application deadline. Dolan was separated from his partner by the moves. He struggled with isolation at subsequent positions. The postdoc system is not designed with mental health conditions in mind. It destroys the two things those conditions most require: continuity of care and a stable support network. Every two years, your doctors change, your friends scatter, your daily geography resets. If you're healthy, this is disorienting. If you're not, it can be something else. Dolan secured a new position in Crete. Around the time he was due to start, in 2011, he returned to the UK and died by suicide.

Rosten had by then left academia. In 2014, he finished a paper on conformal algebra, a topic that substantially overlapped with Dolan's research interests, and dedicated it to his friend's memory. In the acknowledgements, he wrote that he was "firmly of the conviction that the psychological brutality of the post-doctoral system played a strong underlying role in Francis' death." He submitted the paper to the Journal of High Energy Physics. The science was accepted. The acknowledgement was flagged. An editor told him to remove it. Rosten refused. The editor responded: "The required corrections concern the last paragraph of the acknowledgements. We would remove it completely. I guess there were more basic problems in Dolan's life than the pressure put by physics work. Certainly people, say in business, behave more brutally than in academia." A second journal accepted the science and rejected the acknowledgement. A third told him: "In a scientific paper we discuss about science, not about life." Rosten withdrew the paper twice rather than remove the paragraph. The European Physical Journal C finally published it in 2017, acknowledgements intact, almost three years after it first appeared on arXiv. The system couldn't accommodate a single paragraph about what it had done. The cockpit killed someone and then three journals tried to redact the obituary.

Like most people, I first read about Rosten and Dolan in late 2017, when the paper was finally published. It was devastating. I was in my first year of graduate school, still learning how the system worked, and here was someone stating plainly what I was only beginning to suspect. A few months later, I sat in that seminar about galaxy surveys and learned about concentration of measure for the first time. I didn't connect the two. The math was abstract, the death was real, and I failed to see they were the same arithmetic wearing different clothes. It took years.

Sometimes the cockpit even ejects people who won. In May 2023, a theoretical physicist who writes the popular blog 4gravitons landed the thing that the entire pipeline is supposed to lead to: a permanent research position, at CEA Paris-Saclay. Permanent. In theoretical physics, that word has the gravitational pull of a small star. He'd spent a decade moving through the system, PhD through multiple postdocs across multiple countries, and this was supposed to be the end of it. He and his wife relocated from Denmark to France. She's a teacher, and they timed it so she could start with the school year.

Under EU law, the spouse of any EU citizen can work in any member state. In practice, France took five months to even open his wife's file, and when it finally responded, it granted her the right to remain in the country but not to work or travel. She had given up her job, her friends, and her daily routine to follow a permanent academic position to a country that then informed her she was welcome to exist in it but not participate. They left. He resigned the permanent position and they returned to Denmark, where he moved into data science and freelance science journalism. "Academics don't get to choose where to live," he wrote. "People do, though, especially in places like the EU. I can choose for us to live in Denmark, to build a life in a country that has treated us well. I just have to leave academia to do it."

The cockpit was built for a person without a spouse, or at least without a spouse who has a career, or preferences, or legal personhood that the host country's bureaucracy is willing to recognize. It was built for someone who exists in one dimension: research output. Every other dimension, the partner, the language barrier, the five months of administrative silence, the right to participate in the economy of the country you live in, is outside the spec. He left because the cockpit didn't have a seat for the person next to him.

These are the stories that get told, because the people in them had names and blogs and papers and enough visibility for their departures to register. But for every physicist who enters the pipeline and leaves, there is someone who looked at the pipeline from the outside and never entered at all. The undergraduate who loved research but watched their advisor work seventy-hour weeks for a salary their roommate's starting offer in tech would double. The first-generation student who couldn't picture six more years of negative net worth. The person who ran the numbers on the two-body problem before the two-body problem had a chance to run them. These people don't write blog posts about leaving academia. They don't appear in attrition studies. They chose something else before the system had the chance to choose for them, and the system has no way to count what it never saw. The pipeline's most efficient filter is the one at the entrance, and it operates by reputation.

The field has a name for all of this: the leaky pipeline. The metaphor is revealingly passive. A pipeline leaks as though by accident, as though the fluid just happens to seep out through minor imperfections, as though someone would fix it if only they noticed. A global study tracking nearly 400,000 scientists across 38 countries found that a significant number leave within a decade of starting their research careers. Women with comparable scientific output face a 35 percent higher risk of leaving academia than men. Forty-two percent of mothers in STEM leave full-time STEM employment within three years of having children, versus 15 percent of fathers. In physics specifically, women make up a shrinking share at every successive rung: a smaller percentage of postdocs than PhD students, a smaller percentage of associate professors than assistant professors, a smaller percentage of full professors than associate professors. The pipeline was never sealed. It was designed, like the cockpit, for a person who doesn't exist, and the people furthest from that fictional average are the first ones out. The 0.3 to the tenth applies here too, just with different dimensions: research productivity, geographic mobility, financial cushion, childcare access, mental health, safety from the person who controls your career, willingness to work for less than your training is worth. The more of these dimensions you measure, the emptier the center gets.

The political response to all of this has been instructive. Oklahoma recently announced it would phase out tenure at public regional universities and community colleges entirely. Not adjust the cockpit. Remove the seat. The reasoning was accountability, which is a word that, in an educational policy context, means "we've decided teachers are the problem." Several other states (Florida, Georgia, Texas, Kansas) have implemented mandatory post-tenure reviews. The permanent appointment is now permanently provisional. None of these states has proposed making the pipeline more livable: longer postdoc contracts, portable benefits, spousal employment support, housing assistance, any of the adjustable-seat equivalents that the Air Force implemented seventy-five years ago. The diagnosis is that scholars have too much security, and the prescription is less. In the language of the cockpit study, the pilots keep crashing and the response is to make the seat smaller. The tenure system was designed to protect scholars from political pressure. The political response to its failures has been to apply more political pressure. This has a certain symmetry.

There's a detail about the original cockpit study that usually gets treated as a footnote. The 4,063 pilots were all men. They were almost all white. The population had been pre-filtered along dimensions of race and gender before anyone started computing averages. The "average pilot" was the average of a subset that had already been selected for demographic similarity. And the average of that narrow, pre-filtered, relatively homogeneous group still described nobody. Daniels proved that the average was empty even under the most favorable possible conditions for the average to work.

The math tells you what happens when you relax those conditions. Every dimension of human variation you add is another factor in the exponent. Gender, race, socioeconomic background, disability, the life circumstances that shape how people show up in institutional systems: each one multiplies the improbability of anyone landing at the center. The 0.3 to the tenth was computed over a population of physically fit men of similar age and background. Extend the same calculation to the actual population that has to sit in the cockpit, the one that includes women, people of different body types, people the 1926 engineers didn't bother to measure, and the exponent grows. The average gets lonelier. The systems built around it fail more people, and they fail hardest the people who were never in the sample to begin with. The exponent doesn't care whether the designers acknowledge this. It just keeps multiplying.

Seven years before Daniels published his cockpit findings, there was a contest. In September 1945, the Cleveland Plain Dealer announced a search for "Norma," offering a $100 war bond to the Ohio woman whose body most closely matched a statue of the same name displayed at the Cleveland Health Museum. The statue had been sculpted by Abram Belskie based on measurements from 15,000 young women, almost exclusively white and able-bodied. It represented, in the language of the time, "the norm or average American woman of 18 to 20 years of age." Nearly 4,000 women sent in their measurements. The number who matched Norma's dimensions was, by now predictably, zero. The winner, a twenty-three-year-old theater cashier named Martha Skidmore, was the closest approximation, and even she wasn't close. The medical establishment reviewed the results and concluded that the problem was not with the statue. The problem was with American women, who were out of shape. Experts recommended more strenuous physical education programs to make women's bodies conform more closely to Norma's. The average didn't describe anyone, so the solution was to fix the people. Scott would have recognized the move. In Seeing Like a State, he documents the same logic applied at continental scale: Soviet planners who redesigned agriculture around a theoretical average farm, city planners who bulldozed neighborhoods to match a geometric ideal, foresters who replanted diverse ecosystems as monoculture rows. In every case, when the model and the population disagreed, the population was wrong. Norma's doctors were working from the same playbook. They just had a smaller budget.

If that sounds like a relic of 1945, listen to the way academic career advice is dispensed. Be more productive. Be more mobile. Publish in higher-impact journals. Write more grants. Say yes to the postdoc in a country where you don't speak the language; it'll look good on your CV. A senior physicist once told me, at a conference dinner, that the trick to surviving the postdoc years was to "just not have a life for a while." He meant it as encouragement. Stop doing service work; it won't count. Find a more flexible partner, or don't have one. The advice is always to reshape the person. Never to reshape the seat. When a scholar can't fit the cockpit, the system doesn't ask whether the cockpit was designed wrong. It asks whether the scholar was committed enough. Norma's doctors would have recognized the logic immediately.

I keep coming back to the arithmetic. 0.3 to the tenth power. Six in a million. That's how many people are average on ten dimensions when "average" is defined as the middle thirty percent on each. The statistical point is about as subtle as a brick. It requires no measure theory, no functional analysis, no blackboard and three hours. It's multiplication. The kind you learn in fourth grade. And the conclusion it supports is the same one Daniels reached in 1950: the more dimensions along which you measure people, the more certain you can be that nobody is average. The average is an artifact of the compression. A place in the space that the math says should be full and the data says is empty.

There's a version of this that's comforting, the version where you say "everyone is unique" and put it on a poster. That's not what the math says. The math says something colder, which is that the center is uniquely empty. Everyone is unique in the trivial sense that they occupy a different point in the space. But the specific point called "the average," the one we build our systems around, the one we design our cockpits for, is the point that nobody occupies. The thing we keep designing for is fictional.

The galaxies come back to me here. The average galaxy in our survey, the one with the mean luminosity and the mean color index and the mean effective radius, didn't correspond to any object in the actual sky. It was a ghost. A mathematical construct that we almost mistook for a representative example. The professor who asked the question already knew the answer. He liked to run show-of-hands polls during lectures. That one caught most of us.

The Air Force figured this out seventy-five years ago. The solution was adjustable seats, adjustable pedals, adjustable helmets. They did this because pilots were dying. Seventeen in a day. The failure mode was visible, dramatic, and countable. Someone wrote a report. The seats moved. But nobody in the academic pipeline is crashing in a way that produces a report. Francis Dolan is a paragraph in an acknowledgements section that three journals tried to delete. The 4gravitons author is a line on a spreadsheet somewhere in France marked "resigned." The scholars who never entered the pipeline, the ones who looked at the stipend and the geography and the odds and did something else, don't appear in any dataset at all. The Air Force had the advantage of a failure mode you couldn't ignore. Seventeen planes on the ground in one day. The academic pipeline's failure mode is people quietly disappearing, and the system interpreting their absence as evidence that the system works. The ones who remain must have been the ones who belonged. Anyone who left must not have wanted it enough. The ones who died, well, there were probably more basic problems in their lives than the pressure put by physics work.

A tenure system that expects a scholar to move across the country, away from family, into a city they can't afford, to work for a salary below what their training would earn them in industry, and then to perform optimally for six years under conditions of maximum personal disruption, is not selecting for the best scholars. It is selecting for the people who can tolerate the cockpit. And when someone can't, or won't, or leaves, or never shows up, the system doesn't count them. It counts the people who are left and calls the result a meritocracy.

We are still building fixed cockpits. For scholars, for the people who follow them across borders, for everyone who has to pass through a system that was designed around a point in high-dimensional space where nobody lives. The cockpits have gotten nicer. They have digital dashboards and updated methodologies and equity statements and "the most significant changes in our history." But the seats don't move.

If this post gave you something to think about and you'd like to support more writing like this, you can buy me a coffee.

If you have a Bluesky account and want to share your thoughts, you can leave a comment on this post's thread.

References

1. Todd Rose, "When U.S. Air Force Discovered the Flaw of Averages," Toronto Star, January 16, 2016. Excerpted from The End of Average (HarperOne, 2016). thestar.com
2. Gilbert S. Daniels, "The 'Average Man'?" Technical Note WCRD 53-7, Wright Air Development Center, Wright-Patterson Air Force Base, Ohio, 1952. apps.dtic.mil
3. Todd Rose, The End of Average: How We Succeed in a World That Values Sameness (HarperOne, 2016). See also: "Beyond Average," Harvard Graduate School of Education, August 2015. gse.harvard.edu
4. Michael Thaddeus, "An Investigation of the Facts Behind Columbia's U.S. News Ranking," February 2022. math.columbia.edu
5. "Former Dean at Temple Convicted in Rankings Scandal," Inside Higher Ed, December 6, 2021. insidehighered.com
6. "Michael Thaddeus," Wikipedia. wikipedia.org
7. "U.S. News & World Report Best Colleges Ranking," Wikipedia. wikipedia.org
8. "A More Detailed Look at the Ranking Factors," U.S. News & World Report, September 22, 2025. usnews.com
9. "U.S. News Shakes Up Rankings Methodology — but Top Colleges Held Their Spots," Higher Ed Dive, September 18, 2023. highereddive.com
10. "Academic Salaries and Stipends." academicsalaries.github.io
11. Oliver Rosten, "On Functional Representations of the Conformal Algebra," The European Physical Journal C 77, 477 (2017). See also: "The Story Behind a Moving Academic Acknowledgement," Academia Obscura, August 29, 2017. academiaobscura.com
12. "Calling Attention to a Postdoc's Struggles and Suicide," Inside Higher Ed, August 8, 2017. insidehighered.com
13. Andrew Grant, "A Paper on Field Theory Delivers a Wake-Up Call to Academics," Physics Today, August 3, 2017. physicstoday.aip.org
14. 4gravitons, "Well That Didn't Work," January 19, 2024. 4gravitons.com
15. 4gravitons, "Why We Are Leaving France: The Misadventures of a Trailing Spouse," January 26, 2024. 4gravitons.com
16. Deborah Kaminski and Cheryl Geisler, "Survival Analysis of Faculty Retention in Science and Engineering by Gender," Science 335, 864-866 (2012). science.org
17. Toni Feder, "When Tenure Fails," Physics Today 76(10), 44 (2023). physicstoday.aip.org
18. "The Future of Tenure," The Chronicle of Higher Education, May 2024. chronicle.com
19. "Scientists Are Leaving Academia at Unprecedented Rates," Nature, October 4, 2024. nature.com
20. "Why Women Leave Academia: A Longitudinal Study of the Leaky Pipeline in German Sociology," Higher Education, 2026. springer.com
21. "The Leaky Pipeline," Institute of Physics, University of Amsterdam. iop.uva.nl
22. "The Leaky Pipeline in Physics Publishing," ResearchGate, 2020. researchgate.net
23. "Tenure Can't Stay the Same," Inside Higher Ed, March 16, 2026. insidehighered.com
24. Dahlia S. Cambers, "The Law of Averages: Normman and Norma," Cabinet Magazine, Issue 15. cabinetmagazine.org
25. "Norma and Normman," Weird Universe, March 3, 2024. weirduniverse.net
26. Mayo Oshin, "One Size Fits None: Why Striving to Be Above Average Limits Your Potential." mayooshin.com
27. "The PhD Experience," Nature PhD Survey 2019. nature.com
28. Sherry Moss and Morteza Mahmoudi, "STEM the Bullying: An Empirical Investigation of Abusive Supervision in Academic Science," EClinicalMedicine 40, 101127 (2021). pmc.ncbi.nlm.nih.gov
29. National Academies of Sciences, Engineering, and Medicine, Sexual Harassment of Women: Climate, Culture, and Consequences in Academic Sciences, Engineering, and Medicine (Washington, DC: The National Academies Press, 2018). nationalacademies.org
30. "National Academy Report on Sexual Harassment," 500 Women Scientists, 2018. 500womenscientists.org
31. James C. Scott, Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed (New Haven: Yale University Press, 1998).