AI helped me with this post, which has gone through several revisions.
Part Two
Ithaca, New York, sometime around 1986. An eighteen-year-old sits at a news desk in front of an IBM Selectric. To his right, a stack of tape cartridges, each one labeled on the front so the label can be read while the cart sits in the machine. To his left, single sheets of paper laid out on the table, each sheet paired with a cart. Each pair is a story.
Interviews get recorded on cassette in the field, on reel-to-reel in the office. Back at the station the useful seconds get copied onto the carts. If a clip needs trimming, you cut the tape with a razor blade and splice it. As he types each story he takes a cart off the stack on his right and sets it beside the page on his left. When it is time for the newscast he carries the pairs down the hall to the booth, hands the carts in order to the engineer, who is also the disc jockey, and keeps the pages to read. When he reaches the part with the quote in it, he points, and the engineer fires the cart. The cart rewinds to the start of the clip and waits, in case it is needed again.
He is a reporter at WHCU-AM/FM, the CBS affiliate, about thirty hours a week at something like five dollars an hour. He graduated from the local alternative high school in 1985 and did not feel ready for college. He had trained in news at WVBR, the Cornell student station, through a school district internship. He got the interview at WHCU because his parents were friends with the news director. He covers city council, school board, planning board, crime, and research news out of Cornell University.
The craft has rules. Each story needs two versions, each carrying a different quote, to run on alternating half-hour newscasts, plus one-sentence headline versions for the quarter hours. Same length every time, whether the story is a fire or a zoning variance. The quotes have to be gathered in person, which means walking up to the mayor after the meeting and managing an interaction with an adult who outranks you while listening closely enough to know whether you have two clean stretches of fifteen seconds. And the newscast is live. At the end of the local evening news he wraps the market report and the weather while the second hand sweeps toward twelve, because at six o’clock the CBS World News Roundup begins and his time is over.
In May 2026, when the CBS Radio Network shut down after ninety-nine years, Philip N. Cohen (b. August 12, 1967) posted the photograph of himself at that desk and explained the workflow to readers who have never seen a Fidelipac. He noted what Ithaca had then: fewer than thirty thousand year-round residents and three radio stations with full-time news reporters, plus reporters from the Ithaca Journal, the Ithaca Times, the Grapevine, and the student papers. Four or five reporters at a school board meeting. When something big happened, more came down from Syracuse and Binghamton.
He added a footnote his students have not read. He still runs an undergraduate class to fill the time allotted, and says thank you as the last second elapses.
Forty years later he is a professor of sociology at the University of Maryland, the author of the leading undergraduate textbook on American families, the founder of the largest nonprofit preprint archive in the social sciences, a former plaintiff against a sitting president, and the man most likely to tell you, with the arithmetic attached, that a celebrated study cannot be right. The habits from the news desk are visible in all of it: fixed deadlines, no retakes, an insistence on getting the quote in person, and a working assumption that a story exists to be handed to strangers.
He came from a house where science was the family trade. His father, Marshall Cohen, took a doctorate in mathematics at Michigan in 1965, did a postdoc at Princeton, where Philip was born, and then went to Cornell as a geometric topologist and combinatorial group theorist. His mother, Avis H. Cohen (1941–2026), took a bachelor’s degree at Michigan in 1964, moved with two small boys to Ithaca, and started graduate school at Cornell in the early 1970s. Writing about that decision in 2013, she gave two reasons. One was that she had already worked out a view of dynamical systems before enrolling and had found a new Cornell professor in neurobiology who shared it. The other was shorter: “Being a full time mom was wearing very thin.”
Her dissertation concerned locomotion in rats. She switched nerves across muscles and filmed the animals running with x-ray cinematography. Her adviser, the biolinguist Eric Lenneberg (1921–1975), died before she finished. She took the degree in 1977 and went to the Karolinska Institute in Stockholm for two years, the family with her, to do the work she is known for. Cut a short segment out of a lamprey’s spinal cord, isolate it in a dish, and the nerves keep firing in the pattern of swimming. No brain required. The capacity to generate the rhythm is distributed along the cord. Back at Cornell she worked with mathematicians on a model of these central pattern generators that neuroscientists still cite, and the finding turned out to matter most to engineers: robots walk better when the pattern lives in the limbs than when the brain has to compose each step. Her most cited paper is about a quadruped robot.
Cornell kept her for ten years as a research associate. She won her own federal grants and ran a lab. No tenure-track offer came. Maryland hired her in 1990 as a full professor, and she went on to found its interdisciplinary neuroscience program, direct a National Science Foundation project for women in science, and be elected a fellow of the American Association for the Advancement of Science. Her son’s judgment on the Cornell decade, published in June 2026 after her death, is split: passing on her in 1979 may have been reasonable, and passing on her by 1990, with the grants and the publications in hand, was not. He notes that Suzanne Bianchi, the Census Bureau demographer who became his mentor at Maryland and died in 2013, followed the same route. Faculty women who were never assistant professors.
Avis Cohen liked to say that her understanding of nervous systems bore on how people should think about their societies, and that the nervous system was less hierarchical than male scientists had assumed. She also told a story from her Illinois girlhood in the 1950s: that if she got all A’s, no one would want to marry her.
Her son has never, as far as I can find, drawn the obvious line in print between a mother who showed that coordination can be distributed and a career spent dismantling central authorities in his own discipline.
Before sociology there was more journalism and more politics. He reviewed local bands for the Ithaca Times, finding something to write about every weekend. He worked in the Ithaca left, the environmental and labor and alternative-media world that a college town of that size sustained in the 1980s. He did editorial work at the Michigan Daily as an undergraduate. He took his bachelor’s degree in American culture at Michigan in 1992, with high distinction, a master’s in sociology at Massachusetts in 1995, and a doctorate at Maryland in 1999 under Reeve Vanneman, with a dissertation on Black population size and the structure of American labor market inequality.
Asked in August 2020 by his former student Joanna Pepin how he became a family sociologist, he gave an answer that runs backward from the subject to the question. “I was interested in inequality, and a lot of the inequality questions I was thinking about kept leading me back to families.” Why do women do more housework and child care? Partly because they are single parents. Who decides who is married, and what follows for health and income? Will the government give money to poor people, and will it depend on their family structure? Will it penalize people for their choices? Will it let people marry if they are gay? “All these things that have to do with people’s well-being and social inequality go through family. So that’s how I got led into it.”
Two apprenticeships in the government’s statistical apparatus completed the training. He was a research assistant at the National Center for Health Statistics in 1997 and a statistician and demographer at the United States Census Bureau from 1998 to 1999. Anyone who has built a household roster for a federal survey learns the same lesson, and it turns up in his work for the next thirty years: the categories are constructed by people for purposes, they change, and when they change the numbers move.
He went to the University of California, Irvine, in 1999, and made his reputation there with Matt L. Huffman on the economics of gender at work. The two of them pushed past the familiar observation that women crowd into low-paying jobs. They examined whether an occupation pays less because it has come to be identified with women, and found evidence that it does. They joined individual earnings data to organizations, occupations, and metropolitan labor markets, which became his lasting methodological commitment: a worker’s education and experience are worth what the surrounding structure will pay for them. Their 2007 article in American Sociological Review on female managers and the gender wage gap won the section award for sex and gender; later work with Jessica Pearlman on workplace desegregation won an award from the organizations and work section. He extended the same logic into the home, studying housework, care work, paid domestic labor, and the way state policy and wage structures set the terms of bargaining between spouses.
At the University of North Carolina from 2005, and at Maryland from 2012, the subject shifted from labor markets to families, with inequality still the question. Families distribute money, time, care, housing, standing, and risk. They shelter people from markets and reproduce advantage between generations. An early paper with Lynne Casper and Tavia Simmons reconstructed historical levels of unmarried cohabitation from census data, and the exercise shows the pattern of his mind: before you explain a trend, establish that the instrument did not change.
Somewhere the research changed, he said at a Wisconsin workshop on February 26, 2025. “I’ve taken a pretty descriptive turn. I’m less about trying to discover new theories and mechanisms, and more about trying to figure out what’s actually happening, and find ways to communicate about that to other scholars and various publics.” That is a professional demotion accepted on purpose. “If you have a descriptive finding it’s considered to be less than, in a lot of academic circles, and so we end up doing more puffery just to make our work sound like it’s not just descriptive. Very off-putting to the public.”
Three words organize his textbook, The Family: Diversity, Inequality, and Social Change, now in its fourth edition and the standard assignment in the course. The mid-century arrangement of early marriage, a breadwinning husband, a homemaking wife, and several children was a historical episode resting on a particular labor market, legal order, and economic expansion. Its erosion widened women’s options and produced cohabitation, later marriage, same-sex marriage, one-person homes, childless adults, and stepfamilies. People with money, education, health, and support build flexible family lives; people without them experience the same freedom as instability.
The book gets revised on a cycle short enough to catch the language changing under him. Describing the third edition in that same August 2020 conversation, he said the data on same-sex marriage had improved and was still poor, partly because many states record only spouse one and spouse two on a marriage certificate, so the administrative record cannot answer the question everyone wants answered. And the vocabulary had turned over inside a single revision: “I used to have a whole section on androgyny, and now I don’t even talk about androgyny anymore. Now we have non-binary identities.”
His essay collection Enduring Bonds appeared in 2018 from the University of California Press. His wife, the historian Judy Ruttenberg, supplied the title, having found the phrase in Justice Anthony Kennedy’s opinion in Obergefell v. Hodges. The subtitle promises everything that makes families great and terrible. The same bonds that produce intimacy and obligation produce dependency, unequal labor, and stigma. His says sociology’s job is to explain how the benefits and burdens get distributed.
Talking with Pepin that August, he states the question. “I’m interested in the things that we don’t see about families, like who gets to have a family. We talk a lot as if everybody’s in a family, but not everybody’s in a family, and some people aren’t in families because they don’t have access to the resources or the people.” He reaches first for the years when same-sex marriage was illegal, and then turns the case over: people stuck with families they cannot leave, transgender teenagers living with parents who do not accept them, people in abusive homes. The constraint runs on access to resources and to other people, and it works the same way whether it locks a man out or locks him in.
In an interview recorded on December 4, 2020 he ran the chain from national mood to private decision in four sentences. “One of the things that is dislocating for people is the overall narrative of history: where is our society going? That has a lot to do with social mobility. Do we think our kids are going to do better than ourselves? If we’re safe with that assumption, then we feel ourselves the masters of our universe, and we’re comfortable making long-term plans.” Then the taxonomy follows. “Making banana bread is a short-term response. Redesigning your home, planning to have more children, making long-term career investments that you don’t expect to have short-term rewards, these are things that people do when they’re confident about the future. For family life, that means the lack of confidence in the future means fewer children, less marriage.”
The paper that made him famous among people who follow family statistics is “The Coming Divorce Decline.” He posted it to his own archive in September 2018, where three thousand people downloaded it, and Socius published it a year later. The proportion of married women divorcing each year fell about eighteen percent in the decade after 2008; adjusting for age, education, race and ethnicity, nativity, marriage order, and years married, it still fell about eight percent. Newer marriages carry lower risk because the people entering them are older, better educated, more often marrying for the first time, and less often bringing children into the marriage.
What distinguishes him is what he does with a result in his favor. Every year, when the new American Community Survey data came out, he checked his own prediction in public. And in his own comment section he let a reader argue that divorce has become a poorer measure of instability as cohabitation and serial cohabitation have risen. Asked in August 2020 to name a common misperception, he chose this one, and then itemized what the good news is made of: fewer people marrying because that is what you do, fewer stuck with somebody, fewer trapped in an abusive marriage because they cannot support themselves alone. Then the clause a headline-protecting researcher leaves out. The decline in divorce comes partly because “marriage is available to fewer people. So it’s a mixed bag.”
This is the ground on which he has fought his longest fight, against what its advocates call the success sequence: finish school, get a job, marry, then have children, and you will very probably avoid poverty. Cohen’s objection is that the sequence converts a description of who is not poor into an instruction for people who are, and that a public campaign urging the sequence supplies none of the things that make the sequence available, which are affordable schooling, steady employment, health care, housing, and a suitable partner. He points to the federal marriage promotion programs and their evaluations, where couples received their allotted hours of services and the measured effects on marital stability came out small or absent. His research with Joanna Pepin developed the argument into a question of supply: the availability of unmarried men differs between Black and White metropolitan populations, so the advocates who compare married people with unmarried people and attribute the difference to marriage are assuming that the men are there to be married.
In a talk posted on August 1, 2020, he said people marry when they have confidence in the future, and that poor people often do not marry because the gamble is unbalanced, a long-term commitment made “in a universe that offers you no reason to believe that your long-term decisions will be reciprocated by reality.” He put the same thing to undergraduates in a March 2025 lecture. “If you’re relatively poor, marrying somebody who’s relatively poor, what do you gain from it? Most poor people want to get married, but they’re either not persuaded by the prospects they have, or they can’t find someone who they can persuade to marry them. It’s a harsh way of putting it.” That is an argument about conditions, and it is the reason his quarrel with the family-restoration side is narrower than it looks. He agrees with them about what stable marriage requires. He disagrees about the address to send the complaint to.
He also grants that side its strongest evidence, and turns it. Maybe the decline in marriage is individualism and self-centeredness. “But on the other hand, maybe they care more about marriage. Maybe they idealize marriage more than they used to, so they are less eager to get into it because they have put it on a pedestal. They don’t think their life or their relationship is worthy of a marriage. Maybe they delay marriage not because they don’t want to, but because they want a better marriage.”
His antagonist has been W. Bradford Wilcox (b. 1970) of the University of Virginia and the Institute for Family Studies. Cohen accuses Wilcox of overstating the causal benefits of marriage, minimizing selection, and presenting a preferred family form as an empirical result, and his blog post titles now say that Wilcox lies. A reader who wants to evaluate the selection argument has to walk past the accusation to get to it.
The habit of taking apart an admired book was fully formed by 2012, and the first target was on his own side of the aisle. On October 13, 2012 he spoke at a Boston University law review symposium on Hanna Rosin’s The End of Men, and he opened by noting that the author had already left the room. The talk is the method in miniature. On the claim that young women out-earn young men, he showed that the underlying study compared full-time, year-round, never-married, childless, metro-area workers aged 22 to 30, and that within every education level in that group men earned more. The controls were the finding. On the rhetoric of fifty percent he drew a distinction: a milestone is a round number somebody likes, and a tipping point is a point “at which whatever trend you’re talking about accelerates,” which has to be shown. He checked a claim that a small Alabama city had just elected its first female mayor and found she had been elected eight years earlier, was a traditional southern Republican from a political family, was stepping down, and had five men running to replace her. And he stated the principle underneath all of it: “If the facts don’t support the generalization, then the anecdotes are not illustrations of a trend. They’re just little stories.”
He also said, of a town he had never visited, that he could say women do not dominate it “never having been there.” That clause is the whole epistemology, and it is the same claim that made the ethnographers furious three years later.
Two research controversies established his standing as a man who audits his own side. The first was Mark Regnerus‘s 2012 study of adults whose parents had same-sex relationships, which conservative litigators used against same-sex marriage. Cohen was among roughly two hundred researchers who signed a letter arguing that the comparison groups did not support the conclusions, and he spent years on the study’s funding, its review process, and its afterlife in court filings. Politically, that intervention cost him nothing.
The second cost him something. Alice Goffman’s On the Run, an ethnography of policing and warrants in a poor Black Philadelphia neighborhood, had been celebrated in the discipline and beyond it. In June 2015 Cohen went at the one component that could be checked by anybody: a household survey she reported conducting with a man she calls Chuck. Her article in American Sociological Review described 217 homes, 308 men aged eighteen to thirty found in residence, and 144 of them with warrants outstanding. Cohen did the arithmetic against census figures for comparable Philadelphia neighborhoods and found more than twice as many young men as those homes should hold. He noted that the residence rule, three days a week, could count the same man in two places. Goffman’s answer was that this was an ethnographic survey with different sampling and different definitions of who belongs to a home, designed to capture people whom conventional surveys never see.
American Sociological Review declined to publish his comment in August 2015. The editors agreed he had identified errors and chose not to correct the record. He put the comment on his university web page and eventually published a version as a review in Social Forces in 2017. What began as a dispute about one survey became his standing complaint about the discipline: that a request to inspect the evidence behind admired work gets treated as a breach of manners.
Seven years on, at the end of a long online workshop on February 18, 2022, a friendly questioner asked how open science applies to qualitative researchers, and he conceded the ethnographers’ central premise. “Just because you read my transcripts doesn’t mean you can analyze my data. The data is me. I was there, it’s the experience. There’s no such thing as reproducibility, because you couldn’t have the interaction and the experience that I had as me at that moment. So reproducibility is not really the issue. But maybe something like accountability is. You still might have lied in what you reported, or been wrong.” He then proposed the thing nobody has built: a restricted tier where a small number of vetted readers could see field notes and transcripts under agreement, on the model of Census research data centers.
Ten years after Goffman he is still counting. In December 2025, having been invited to talk with a National Academies panel on corrections and retractions, he went through Volume 90 of American Sociological Review article by article, clicking the links, and posted the results. Of eighteen quantitative papers, twelve provided code and data or information about obtaining the data, and six provided nothing. He gave his earlier tallies for comparison: four of fifteen in 2020, eight of twenty-five in 2023. He described his own methods as squishy and invited corrections, to be mailed to his office written on the back of a fifty-dollar bill. He noted that one article has been retracted from an American Sociological Association journal this century, in a case of documented fraud by a Dutch psychologist, and that the association’s 2018 guide for editors does not contain the word retraction. He also advised the journal’s staff to click their own links before publishing, since several papers carried blinded review links, and one pointed to a repository that required asking the author’s permission.
In the comments, one of the authors he had marked as providing nothing agreed with him, promised a replication package when teaching quieted down, and thanked him.
None of that is what most people who have encountered him have seen. His largest audience by a wide margin came from doing the ordinary work of his job. The first lecture of his introductory Social Problems course, recorded on August 29, 2020, has been watched more than twenty-four thousand times, roughly twenty-five times the audience of any of the fights. Somebody’s syllabus is assigning it.
The teaching is patient. His lecture of September 3, 2020 on the Gini index builds the measure from four named people holding stacks of cash, moves one person’s money, watches the line bend away from the diagonal, and only then counts squares by hand to arrive at a number. No integral appears. His justification for the exercise: “Being able to take all the different numbers, everybody’s income is a different number, and describe that whole distribution with one number that quantifies the amount of inequality, turns out to be super useful for analyzing society.” On the way through he notes that “the bottom 10 percent of the US is above half the world’s population in terms of income.”
The August 29 social problems lecture separates the objective condition from the fight over what the problem is: the pandemic exists, and whether the problem is the virus, the government’s handling of it, or the government telling people what to do is a separate question, settled socially. He teaches functionalism in its strongest form, the fever fighting the infection, gives Merton on the uses of crime, calls the parallel argument about poverty cold-hearted while granting its logic, and then takes the theory apart on the ground it stands on: “You tend to define things as good and then try to rationalize them after the fact, working from the assumption that they are good.” He photographed the Women’s March and the March for Life and analyzed both with the same apparatus, and the passage on the teenagers who came to Washington from religious schools is respectful and unmocking. He closes by telling students that “most people are not one or another of these theories,” that they are instruments for generating different questions.
He also tells them something about himself. “Most sociologists are not actually doing those surveys. We’re using the surveys that other people have done. I do a lot of research using surveys. I don’t conduct a lot of surveys.” A researcher who lives entirely inside other people’s instruments will be alert to what those instruments do and unmoved by the claim that presence in the field is its own warrant. His method and his quarrels have one source, and he states it in an aside while explaining research designs to freshmen.
The classroom is also where the family-structure argument gets its fairest hearing. In a lecture on race, ethnicity, and immigration recorded on February 11, 2025, he gives W. E. B. Du Bois (1868–1963) the founding position, that Black family patterns in 1900 Philadelphia followed from slavery and from urban discrimination, and then gives the reversal. “An analyst like Daniel Patrick Moynihan produced reports that essentially reinforced racist stereotypes by attributing racial inequality to family structure rather than the other way around. Rather than saying the family structure resulted from inequality, as Du Bois had said, he tended to look at it the other way.” And then he puts up a chart that cuts against the lazy version of his own side’s story. “If you go back to 1950, these rates were much more similar. So this is not a direct consequence of slavery, in the sense that we always had this big gap in marriage rates. The gap in marriage rates has in fact opened up, and the de-industrialization story that I told on the last slide is a big part of it.”
The same lecture contains the only place in the record where he uses his own family as course material. He puts his grandmother on a slide, photographed in front of the bar and liquor store she and her husband ran, and beside an earlier picture of her with her Polish siblings, and then places himself in the taxonomy. “The grandchildren of immigrants, like me. We may retain our identity as part of an immigrant family, but we come to see that ethnicity as family history. I don’t have an immigration experience. I had grandparents who had a Yiddish accent.”
And a month later, teaching love and romantic relationships on March 13, 2025, he arrives in the last thirty seconds at an argument he never publishes anywhere. Utilitarian love, the checklist model that dating apps are built for, was supposed to be the rational successor to the romantic myth. “It’s interesting that nowadays, when we have this concept of utilitarian love, it often reinforces endogamy. If you’re making that checklist of things that you want, it often ends up producing people who are kind of like you. Same religion, same area, maybe same social class background.” A spark can cross a line that a filter cannot, which is the individual-level route to the sorting his research measures at the aggregate: by 2022, about three quarters of American married couples had both spouses on the same side of the college divide.
The spring of 2020 pushed all of this into public view at once. Beginning on April 3, 2020 he recorded a series of pandemic lectures for his class and posted them. The first opens with a disclaimer he repeats twice: he is not an expert on pandemics or public health, he is a social scientist who reads well, and he is a teacher. What follows is careful about its own limits, saying repeatedly that nobody knows the mortality rate, that testing gaps make foreign counts unreadable, and that his own chart of American deaths looks worse than Europe’s until you notice the country is the size of Europe and holds several separate outbreaks. His best teaching in it is a demographer’s habit turned into something a general reader can use: the people arriving at hospitals today were infected around the time the schools closed, so everything visible is already two weeks old.
Then the lecture turns. He shows the CDC budget cuts, the eliminated pandemic office at the National Security Council, the ignored 2019 federal exercise, and a montage of presidential statements, and he tells a class that cannot leave the room not to listen to the president. His book on public scholarship argues that a man may speak as a researcher, a teacher, an activist, or a citizen, and that the trouble starts when academic authority carries him from one role into another without the audience noticing. In April 2020 he crossed in front of enrolled students, having twice told them the subject was outside his expertise. The disclosure his own method requires is there. What it does about the grade book is another question, and nobody has put it to him.
Three days later, on a subject squarely inside his field, there is no politics at all. The lecture of April 6, 2020 on family violence contains the observation that turned out to be his most durable pandemic call, and one that almost nobody else made in public that spring. Reports of child abuse and domestic violence were about to fall, and the fall would measure the removal of the observers: “The way we collect data on this is from people leaving their houses and doing things like going to work and going to school and going to the doctor, and a lot of that has not been happening.” Teachers, clergy, and clinicians are the layer through which the state sees inside a home. Close the schools and the layer disappears. He said it in April, said it again in June, and said it again in August, and anyone reading the 2020 numbers without that sentence in hand drew the wrong conclusion from a falling count.
The same lecture contains the structural claim his textbook rests on, stated in thirty seconds. The privacy that makes a home a refuge is the same privacy that makes it unreachable, and people value it most when things are going well. Nobody has to be a monster for the arrangement to produce the outcome.
He was wrong about things too, and the pattern of the errors is consistent. Reasoning from a well-tested framework worked: applying the theory of fundamental causes in April 2020, he predicted that an epidemic which had begun in rich coastal metros would reassert the usual gradient through exposure, comorbidity, and treatment access, which is what happened. Reasoning from a model built for the occasion did not: that same spring he predicted that lockdowns would cut male homicide more than female homicide because men are killed by acquaintances and rivals, and American homicide rose by roughly a third that year, concentrated in exactly the category he expected to fall. His August 2020 worst case for American deaths, which he labeled speculation and then flagged again as resting on an unknown herd-immunity assumption, was 700,000. The real figure passed it in 2021.
The best twenty minutes of the pandemic teaching is a counterfactual done properly. In his overview of August 25, 2020 he puts Italy beside the United States: a worse start, an earlier and higher peak, and a far better position five months later. Then the timeline of what Italy did, escalating from a few northern towns to a region to a national shutdown where police stopped people on the street and required an explanation, followed by a phased reopening as cases fell ahead of deaths. A student comes out of it able to answer the question that dominated American argument all year, which is whether the outcome was fixed by circumstance. And he names the cultural failure in one line: “That’s not how you fight a virus. Basically you fight a virus by acting weak and staying home, and we didn’t do that.”
The most self-implicating sentence he has ever spoken came two months later, on October 27, 2020, in a talk at the University of California, Irvine on what the pandemic was doing to American family demography. Births had not happened yet, so he had built proxies out of Google search indices, condom company earnings calls, and two states’ monthly vital statistics, and he warned the room about every one of them before showing it, including that he had not kept a record of the searches he tried that did not look good. Then he put up a chart of divorce searches and said this. “This is very speculative, but it does fit my presumption, which is probably why I’m showing it to you.” The file-drawer problem, confirmation bias, and motivated selection, admitted in one clause at the moment of committing them, in front of a room of demographers.
The association was the institution he tried hardest to fix from inside. He chaired its family section, served as secretary-treasurer of its population section, sat three years as an elected member of its Committee on Publications, and co-edited Contexts, its public-facing magazine, with Syed Ali. In 2019 the association’s leadership signed a letter to the White House opposing open distribution of federally funded research. He organized a petition against it that December, gathered more than two hundred signatures, got the Publications Committee to pass a resolution urging the governing Council to take up the question, and watched the Council ignore it. Committee members spent years trying to get the journals to adopt a policy that would merely tell readers whether a published article came with its data and code. Two subcommittees produced a mild version. The Council rejected it. At one point, he says, staff invented a rule that would have barred him from taking part in the discussion of his own proposal.
He resigned his membership on November 6, 2021 and published his reasons. Membership and the annual meeting cost too much, and the association’s own survey said so. The staff outlast the elected leadership and can stall anything. Publications are the largest source of revenue, which means library subscriptions, which means a transfer from teaching-heavy institutions to research-heavy ones, since publication in the association’s journals is dominated by elite universities. Subscription prices had more than doubled in eight years. The association signed a new seven-year contract with Sage in 2018 without notifying members in advance. And, in the sentence that gives the whole document its flavor: “Sage throws the leadership a nice party at the conference.”
Three months later, addressing a professional audience on February 18, 2022, he said: “Our association, the American Sociological Association, is pretty well captured by Sage, and organizationally not interested in changing this beyond what might be necessary for tinkering, daily functions, and PR.”
He also listed what the association does well. It publishes good research. It creates community. It has raised the standing of women and scholars of color through member elections. His conclusion was that these were no longer enough for him, and that the organization would have to hit bottom first.
The best objection came from his own comment section, from a junior scholar who agreed with the analysis and asked where new people would build professional networks if established people left. Cohen, a tenured full professor at a research university with a large blog audience, is the man in this dispute least exposed to the cost of exit. A commenter who said they had grown up with him added that they had always been the sibling who made trouble and started arguments about what was right, and that it was nice to see he had picked up the habit.
The other exit was involuntary. On June 6, 2017, the president of the United States posted about a plan to hand air traffic control to a private nonprofit. Cohen, who had learned that the fastest replies to a presidential tweet get the most attention, and who was verified, replied with a picture of Donald Trump (b. 1946) and a caption: “Corrupt Incompetent Authoritarian. And then there are the policies. Resist.” He had no view on air traffic control. Fifteen minutes later he found he was blocked.
He became one of seven individual plaintiffs in Knight First Amendment Institute v. Trump, filed in the Southern District of New York in July 2017, arguing that the interactive space beneath the president’s tweets was a public forum from which citizens could not be excluded for their views. The district court agreed in May 2018. The Second Circuit affirmed in July 2019. The Supreme Court vacated the judgment as moot in 2021, after Trump had left office and the account was gone.
The best image of him comes from a reporter at the courthouse in 2018. Leaving a hearing, Cohen hands the clerk his numbered chip, gets his phone back, and says: “Who knows what he did while we were in here?” Outside on Pearl Street he bounds down the steps to the photographers and holds the phone up to the cameras so they can shoot the blocked notice on the screen. He told a British education reporter that some of his friends thought the whole enterprise was a waste of time, playing into the president’s hands and feeding the outrage industry without improving anything. “And honestly, they may be right,” he said.
He told reporters in July 2019, after the Second Circuit ruling: “I often don’t read his tweets before replying. The point is not to have a dialogue with him, but to engage with the millions of people who read his tweets.” Looking back on it in February 2025, he described the campaign as strategy, and then delivered the line the room laughed at: “So I sued the president. Of course I did. Sociologist sues the president. I can’t tell you how many times people said something like, of course you did.” He also explained why the identities never separated. “I was doing it not separately from my scholarly identity, but as a sociologist. You don’t get to have different identities in different contexts unless you really work at it, with anonymity.”
The generations campaign ran the other way, from the inside out, and it worked. On July 30, 2019 he addressed a National Academies committee on generational issues in the workforce, and he arrived with a pitch he said had been gnawing at him for four years. The historical argument is the part nobody quotes and the best thing in the talk. The baby boom is a real demographic event, a birth-rate spike in 1946 that stayed above its old level until 1964, “the only time in history that we had a prolonged increase to a stable high level of fertility,” and the only cohort the Census Bureau has ever officially designated. Everything downstream is imitation. And then the accident: the event happened to last about as long as a human generation, which put the culture on a treadmill of naming them. “If the baby boom had been eleven years instead of nineteen years, I don’t think we’d be having this conversation now.”
The rest is the usual method. Pew’s own 2015 survey found that only about forty percent of millennials picked millennial from a list of generation labels, and the response was not self-reflection. The boundary between Generation X and the millennials separates Venus and Serena Williams, sisters one year apart, which he delivers as a joke about astrology and which is a within-family control. Cohort fertility during the recession runs three different ways inside the single millennial label, so collapsing them destroys the story. Male employment at age eighteen differs by twenty points across three birth years inside the same label.
Asked by a committee member whether any research shows that generation language actually promotes stereotyping, he answered straight: he knows of none, most sociologists do not use the labels as real categories in their research, and it would be an interesting experiment for somebody to run. That is the opening claim of his own talk conceded under questioning as an intuition, by the man who spends his Decembers counting whether other people’s claims can be checked. Pressed on what Pew should do instead of stopping, he offered decade cohorts and described the script he runs on any new dataset. And he ended by ranking the whole complaint below his real work: “this is not my most important work,” and understanding demographic change and communicating it so people can make smarter decisions is, and he counts this as part of that.
Two years later, on May 26, 2021, he circulated an open letter to the Pew Research Center, signed by roughly a hundred and fifty demographers and social scientists. Pew’s director of social trends, Kim Parker, replied on July 12, 2021 that “Generations are one of many analytical lenses researchers use,” acknowledged the risk of age stereotyping, and offered to discuss it. On May 22, 2023 Pew published a new policy: generational analysis only where historical data allow comparison of people at the same age in different periods, the standard labels used sparingly. He gave Pew its due afterward. “I like the Pew Research Center a lot. They do really good work. They just got addicted to generation clicks, which has happened to a lot of people. And then eventually they changed their process. Minor victory. Didn’t stop the train of generations, unfortunately.” The committee route first, the public campaign when the committee route stalled. It is the same sequence as the association, and this time the public phase worked.
His own illustration of the distinction is the essay he wrote in August 2017 on turning fifty. Being fifty was of no analytic interest. Having been fourteen in 1981 placed him at a particular intersection of technology, music, politics, and money.
He founded SocArXiv in 2016, after Elsevier bought the Social Science Research Network and the discipline discovered how a commercial owner can change the terms of a repository it has acquired. The archive is free, nonprofit, run through the Open Science Framework, and supported by the University of Maryland Libraries, which he describes without romance: “I serve at the pleasure of the dean of the libraries, basically. And that means they pay for it. Not that much, but not to diminish their contribution. It’s a low-budget operation.” He is equally modest about the ambition: “We’re not replacing journals, not eliminating peer review, we’re just trying to make the system better.”
The argument beneath it is historical, and he laid it out in a talk on June 8, 2020. Peer review as we know it is about seventy years old. “Before the middle of the twentieth century, journal editors were the ones who decided what to publish, and they consulted with experts at their own discretion.” Einstein withdrew a paper from Physical Review in 1936 because the editor had sent it out for review at all, and the Watson and Crick paper on the double helix was not refereed; the editor simply decided to publish it. Then, he says, “after World War II, as there was the threat of the constriction of research funding, academics realized that they could lose control of this process if it was handled instead by government agencies, without their direct influence.” The mystique of anonymous peer review arrived with that fear. Historians of science date the decisive moment later, to the funding fights of the 1970s, which makes his case stronger.
By February 2025 he had developed the point into an argument about scarcity. “It used to be that printing was expensive and difficult, and you could use peer review to decide what was going to be printed and shipped and bound and stuck on a shelf. That is not the case anymore. You cannot stop bad research by just saying it’s bad. That’s not how it works, and that’s not the goal of peer review anymore. What peer review is doing in real life today is filtering or curating and labeling research as good or bad. We’re not gatekeeping in the sense of preventing things from appearing. We’re helping people evaluate and decide what to read.”
He also has an argument for preprints that nobody else makes. Told that people fear their ideas will be stolen if they post early, he inverted it: “The problem of stealing ideas probably happens more in people’s inner circles, especially their mentors and advisers and committee members and journal peer reviewers. The people who are stealing ideas are usually not getting them from the public square. They have some private access to your ideas, and they’re abusing that position of trust.”
Underneath all of it is a theory of institutional trust he borrows from Anthony Giddens, and the image he uses for it is an elevator. Nobody knows how one works. Everybody gets in. “More than a fact check, which is a particular fact that something is true, we have veracity, which requires systemic trust. The fact may or may not individually be true, but it comes in a package which has some trustworthiness.” The inspector’s name may be on the certificate and the inspector is not there. “The person that we’re trusting is not there. They’re disembedded. They’re part of an institutional framework, not an individual person. This turns out to be absolutely essential to the functioning and survival of modern economies and democracy.” That is the fragility authoritarian politics exploits, and his proposal for what a scholar should do about it is the book’s whole argument in one sentence. “One thing we can do is be reembedding agents. We can show up at that moment, sort of in the elevator, and say, you don’t have to worry about this, because we’ve figured this out and here’s how we did it. And I’m a person who lives in your community, and I have some firsthand knowledge, and you can trust me for some reason.”
By the time the Kohli Foundation gave the archive its 2025 Infrastructure Prize for Sociology, ten thousand euros and an invitation to Fiesole, it held around seventeen thousand papers. It accepted 3,162 new ones in 2025, twenty percent more than the year before, and by February 2026 was running at a pace of about forty-three hundred a year.
Then the volume turned into a different problem than the one he built the thing to solve. Much of the growth was machine-generated. In November 2025 the archive stopped accepting papers about artificial intelligence to relieve its moderators. On February 22, 2026 he wrote up the case that had unsettled him: a paper the archive would have accepted a few years earlier, dull, unoriginal, thin on literature, competent in its statistics, coherent, relevant, with real citations, and a disclosure listing the tools used to generate the code, run the literature search, advise on the models, and draft the text. The author claimed to have formulated the question, chosen the variables, directed the analysis, and verified every data point and quotation, and offered an audit on request. To reject it, the archive would have to reject it for the tools.
He worked through the arguments in public. He is not persuaded that using such tools is unethical, since all knowledge is learned from someone else and no norm requires citing everything you read. He grants that an archive assigning digital object identifiers without peer review is an attractive place for machines to deposit and to harvest. He says he will not enter an arms race he expects to lose, and that a disclosure requirement he cannot verify is close to useless, since the chatbots read the rules too. And he says, more bluntly than most administrators would, that there is a great deal of bad work on his platform, that this does not damage the good work, and that where rejecting is more expensive than accepting, accepting may be the practical course.
The policy came out on March 9, 2026, drafted by a subcommittee and revised by the full moderation team. It refuses automated detection, declines to state a rule that could be written down completely, commits the moderators to judgment made by people using perception and experience, and invites appeals. One passage reads: “We feel the need to express our humanity in this process.”
That is a small group of trusted people, embedded in a community, deciding what to publish on criteria they cannot fully specify. It is also, structurally, the editorial discretion his own historical narrative began with. The man who spent a decade demanding that journals let readers check the code has ended up defending a queue of volunteers reading papers by hand and admitting that their standard cannot be written down completely.
On July 27, 2026 he read my long biographical essay produced with AI assistance, found it thorough and superficially accurate with correct facts linked to real sources, objected that it relied entirely on freely available material, and wrote five posts about it on Bluesky. The last one says: “We’re only a couple years into this hellscape, and already concepts like author, writing, document, research, and hallucination have been radically destabilized. If you think they haven’t, I’m jealous of your life.”
His third book, Citizen Scholar: Public Engagement for Social Scientists, appeared from Columbia University Press in January 2025, and it is a statement of method. He knows how the title sounds. “I hit on this concept of citizen scholar, which I’m aware is quite pretentious sounding. Who calls himself a scholar nowadays? The citizen scholar is somebody who accepts the duties of a scholar in society. It’s in the practice of citizenship that we generate knowledge.”
The tension the book is built on he states without pretending to dissolve it. “If you’re an activist, it’s not your job to list all the ways that you’ve been wrong before and all the ways you might be wrong right now. It might make you better in your activism, but it might also slow you down. Different from our role as scholars: being wrong is good and important in our scholarship.” And on positionality he takes the middle. “I don’t think I need to state my political opinions in every demography article that I write. On the other hand, I want people to be able to say, wait a minute, I think your work is not trustworthy because I happen to know something about your political beliefs. I think we can work with that tension instead of trying to avoid it.”
The book’s most useful passage for anyone who has ever written one is on the policy implications section at the end of a journal article. “You finally publish an article and you get two or three paragraphs to state your policy implications at the end. This feels like a great privilege, and it is. But if you look at actual policy implications that I have, they’re often so broad that anybody could have written them about almost anything. I find that kids in poverty do worse in school, therefore we should eliminate inequality. You could have said that already.” Then the case that proves it. “The terrible recent example is the expansion of the child tax credit in 2021, which lifted millions of children out of poverty. We had research to back that policy, the policy worked, and then they cancelled it. So just getting the research out is not enough. We have to actually do politics.”
And on what a university owes him, he takes a position with costs on both flanks. “I care more about having a university that protects my right to speak than one that agrees with me. I’m not trying to get my university to put out a statement about Palestine. I’m trying to get my university to put out a statement protecting people who make statements about Palestine.”
He practices the disclosure part relentlessly, and often when nobody is watching. Explaining on April 20, 2020 which workers cannot avoid exposure, he interrupted himself to note that he was delivering the lecture from his own studio. Reading a chart of households behind on rent on September 20, 2020, he stopped at a figure he could not explain and said so: “I don’t know what’s going on with the people over two hundred thousand dollars, why twelve percent of them are having a hard time paying their rent right now. That’s something to look into.”
The care extends to subjects where a lecturer has every incentive to pick a side. Teaching abortion to an introductory class on December 7, 2021, the week after Dobbs was argued, he played Justice Amy Coney Barrett’s question about safe haven laws, stopped the tape, restated her argument in its strongest form, and only then said where he thought the answer fell short. He also made the concession most people on his side will not make. “There is a sense that you cannot avoid the fact that in this policy we are in fact balancing the well-being of fetuses with the well-being of the people who carry them. You can’t pretend that there is no issue there.“ And he closed by addressing the students who disagreed with him. “Even if you are unsure about, or against, abortion rights, there are aspects of this politics which are disturbing to most people. I would never tell religious people what to believe. But there are absolutely trade-offs between some people’s rights and other people’s rights, and the way we manage those things is with politics. So there’s no easy answer. This is just my way of saying I understand this is difficult, so I’m sorry.“
On June 15, 2026 he published a tribute to his mother nine days after her death, including his own assessment of the promotion she did not get, and quoted the line she used near the end when asked how she was: “Thank god, I’m fine!”, which he notes tended to amuse the doctors treating whatever was wrong with her. In her last years she would tell him he had grown. He would protest that he was fifty-eight. She insisted. Twelve days later, on June 27, he published an account of a year on a weight-loss drug, including the detail that his state government health plan covers GLP-1 medications only for diabetes, so the year cost him about six thousand dollars out of pocket, and the advice that people should not comment on anyone’s body unless invited. A demographer who publishes on deaths from alcohol, overdose, and suicide, and on the mortality of the married and the unmarried, treating his own metabolism as a public data point with the insurance code attached.
The same instinct produces his least defensible work. On July 13, 2026 he published a post whose title asserts that Lyman Stone, a demographer associated with conservative family research, is a bigot, and set out the evidence: a 2023 thread in which Stone said the share of the population that is baptized and churchgoing is his primary measure of a good society, and a January 2026 post in which Stone wrote that “The Somali community is a scourge on the republic” and called for reviewing naturalization forms and deporting those with deportable offenses. Stone’s defense, quoted in the post, is that the objection concerns pastoralist clan social organization, and that plenty of other African immigrants do fine. Cohen answers that condemning a whole community for its supposed behavior is the oldest available cover for condemning it, and compares Stone to Charles Murray (b. 1943), who told him in 2014 that his own position was not shouting fire in a crowded theater but “saying quietly and unhysterically, ‘there’s smoke.'” Cohen ends by saying he would not hire Stone or publish his work and that his readers should not either.
Everything in that post is sourced, quoted, and linked, which is more than most such accusations offer. It is also a piece of writing whose conclusion is a hiring recommendation, from a man whose authority rests on his willingness to check arithmetic. Readers who think that judgments about evidence and judgments about character should travel separately will find the post an argument against its author’s method. Readers who think a scholar owes the public a verdict when the evidence supports one will find it the method’s fulfillment. He knows the reaction; he mentions, in a footnote, that Stone regularly calls him a liar.
He describes his own public register more harshly than his critics do. “A lot of my public work is sort of in the category of debunking, or being a crank and complaining about things. Trying to be more constructive than just that, but that’s a recurring theme.” What he wants that complaining to be taken for is stated in the next breath. “I’m trying to elevate what looks like us just complaining about stuff in public to a higher level of value, and say, no, actually what we’re doing is important. We’re discussing other people’s research.”
He has kept a running account of how he is seen. Updating the media page on his website in January 2024, he found himself mentioned in a piece about a journal editor, and in the anonymous comments beneath it, one reader asking “Why give Philip N. Cohen a voice in this discussion?” and another answering, with a laughing emoji, that he was the best sociologist of his generation. He posted both, said ouch, and added that time spent on public outreach is time not spent on research, so if he had fewer media mentions he might have more research. The alumni blog of his old high school has filed his career under a crusade for equality.
Sort the work by whether a stranger can check it and the pattern comes clear.
The checkable work has moved things. Anyone can recount the divorce rates in the American Community Survey, and the finding that marriage is becoming rarer and steadier has entered the standard account. Anyone can compare the number of young men Goffman reported in 217 homes with the number census data would predict, and the episode changed what sociologists expect to be able to inspect in celebrated ethnography. Anyone can count how many articles in a volume of the flagship journal come with their code, which is why the fraction has climbed from roughly a quarter to two-thirds in five years. Anyone can read the arbitrary birth years behind a generational label, which is why the largest survey organization in the country changed its policy. Anyone can watch a lecture and redo the arithmetic on the mayor of Alexander City. And anyone can download a paper from his archive without a subscription.
The unverifiable work has moved less. Whether a researcher at a think tank is lying or merely wrong, whether a man’s views disqualify him from employment, whether an association’s staff act in bad faith or merely in their own interest: these are judgments that ask the reader to trust the writer’s character, and they are exactly the sort of claim his own program says should be inspectable. He makes them anyway, in the prosecutorial register he learned reading copy against a clock, and they are the reason some colleagues who accept every one of his statistical arguments will not cite him.
His career amounts to an argument that sociology should be answerable to its data, its subjects, its students, the public that pays for it, and the readers who never get past the paywall. He has made that argument in journals, in a textbook, in a thousand blog posts, in federal court, in a letter to a survey organization, in front of a National Academies committee, in an undergraduate lecture hall, and finally in infrastructure, which is the form that outlasts an argument. What remains uncertain is whether the tool he built can survive a flood of writing that no one wrote, and whether the discipline that keeps improving its transparency numbers under his counting will keep counting once he stops.
He expects to be around for it. “It’s a disaster, it’s a catastrophe. If you’re studying anything in this space, you’re going to be studying this for a long time. I expect to be studying this for the rest of my career, however long that is.“
He is fifty-eight, turning fifty-nine in August. He still posts most weeks. He still ends class as the second hand reaches twelve, and says thank you.
Notes
Video sources. “The End of Men Is Not True,” the October 13, 2012 talk at a Boston University Law Review symposium, uploaded May 2016: conditioning set at 2:47, milestone versus tipping point at 3:56, “never having been there” at 10:46, the anecdotes line at 10:58, the Alexander City mayor at 11:11. “Against the Generations,” July 30, 2019, to the National Academies committee: Pew self-identification at 3:24, the baby boom as demographic event at 6:30, the eleven-years counterfactual at 7:08, the Williams sisters at 8:09, the concession that no research demonstrates harm at 18:02, decade cohorts at 19:00, “not my most important work” at 24:01. The April 3, 2020 pandemic update: the disclaimer at 1:14, the studio aside at 16:08, the instruction about whom to believe at 26:50. The April 6, 2020 family violence lecture: the measurement point at 8:13 and again at 31:02, the privacy claim at 4:48, the homicide prediction at 14:47. The Joanna Pepin interview, August 23, 2020: origin story at 0:33, policy version at 1:44, itemized divorce decline at 4:39 and “mixed bag” at 5:19, marriage certificates and androgyny at 5:48 and 7:02, “who gets to have a family” at 8:56. “Introduction to Social Problems,” August 29, 2020: objective and subjective at 2:14, the March for Life at 9:23, the functionalist logic at 19:29, “most people are not one or another of these theories” at 35:20, secondary data analysis at 37:23. “Measuring Inequality with the Gini Index,” September 3, 2020: what a summary number is for at 6:13, the bottom decile against the world at 13:00. The September 20, 2020 economic crisis lecture: the eviction moratorium analysis at 17:52, the unexplained rent figure at 12:55.
Quotations from the lectures are smoothed per normal journalistic and academic practice: filler removed, obvious transcription errors repaired, broken fragments joined. No words added, no order changed. Anything printed as verbatim should be checked against the audio. Known traps in this material: the Gini index appears throughout as “genie”; “Robert Wright” in the September 2020 economic lecture is Robert Reich; “email durkheim” is Émile Durkheim; “Daniel Patrick Mahan” is Moynihan; “web Deo” and “WB de boys” are W. E. B. Du Bois; “Anne swidler” is Ann Swidler; “goodmaker institute” is Guttmacher; “Franchesco Gino Michael Lor dri staple” is Francesca Gino, Michael LaCour and Diederik Stapel; “Rogers Brewbaker” is Rogers Brubaker; “Brian noek” is Brian Nosek; “Eric Schwarz” is Eric Schwartz; “the Bordan question” is Bourdieu; “Christian Styer and Sandra Perry” is Kristin Perry and Sandra Stier; “kathryn mckinnon” is Catharine MacKinnon; “Wong Kim Arc” is Wong Kim Ark; “doubles” is Dobbs; “hype amendment” is the Hyde Amendment; the smoking gradient by education reads as “37%” where the figure is 3.7 percent; and the global death count in the August 25, 2020 lecture reads as 80,000 where the slide said over 800,000.
Opening scene (1986 newsroom). All detail is from his own account of May 23, 2026: the Selectric, the carts and paired pages, razor-blade editing, pointing at the engineer, two versions per story plus quarter-hour headline versions, roughly 30 hours a week at about $5 an hour, the job obtained through his parents’ friendship with the WHCU news director, WVBR training via a school-district internship, high school graduation in 1985, the 6:00 p.m. CBS World News Roundup, the Ithaca media census, and the footnote about ending classes on the second. Earlier account of the Ithaca Times music reviewing, March 29, 2012, describing a review due every weekend around 1988.
Parents. Marshall Cohen’s Cornell page. Avis Cohen’s obituary post of June 15, 2026 is the source for her birth (November 29, 1941, Chicago) and death (June 6, 2026), the 1964 Michigan BS, the Cornell PhD in 1977, Lenneberg, Karolinska 1977–79, the lamprey work, the ten years as research associate without a tenure-track offer, the Maryland professorship, the AAAS fellowship (2008), Distinguished Scholar-Teacher (2011–12), the robotics afterlife of central pattern generators, the Bianchi comparison, the “all A’s” story, and “Thank god, I’m fine!” The line “Being a full time mom was wearing very thin” is originally from her GGSTEM essay of March 14, 2013, quoted in his post. There is also a Transmitter obituary. One softening on my part: his post says Lenneberg died by suicide; I wrote only that he died before she finished.
Degrees, positions, awards. His CV and Wikipedia. Date discrepancy still unresolved: Wikipedia gives UNC 2005–2011 and Maryland 2011–; his CV-derived material gives professor at UNC in 2011 and Maryland from 2012. I used 2012. Same for Contexts: Wikipedia says 2014–2017, his own site says 2015–2017.
Huffman collaboration. “Working for the Woman? Female Managers and the Gender Wage Gap,” ASR 2007, and Huffman, Cohen and Pearlman on workplace desegregation, 1975–2005.
Divorce decline. Paper, Socius 2019; preprint, September 2018; blog version of September 15, 2018; the three thousand preprint downloads.
Success sequence and Wilcox. May 17, 2018; May 8, 2018 on partner availability; April 7, 2014 on the Supporting Healthy Marriage evaluation; June 7, 2025.
Goffman. His comment as filed, June 22, 2015; the original post with the arithmetic, June 19, 2015; the ASR rejection, August 26, 2015; the Social Forces review (95:4, June 2017, p. e5) and its full text. The 217/308/144 figures are from her 2009 ASR article. Paul Campos in the Chronicle lays out the same arithmetic. I did not say the ethnography was fabricated, and neither did he.
Regnerus. I wrote “among roughly two hundred researchers who signed” rather than crediting him with organizing the 2012 letter to Social Science Research, which was organized by Gary Gates and others. His account of the review process, July 15, 2014.
ASR transparency audits. December 10, 2025. Earlier: November 30, 2020 and January 9, 2024. Sociological Science’s reproducibility policy is the comparison he holds up.
ASA. March 28, 2021 and the resignation of November 6, 2021. The junior scholar’s objection and the sibling’s comment are in that post’s thread; I did not name the commenter or assign a gender. A widely circulated 2025 ASA resignation letter over the Sociologists for Palestine resolution is by a different sociologist.
Trump. Complaint coverage at the Forward and Lawfare, July 2017. SDNY May 2018; Second Circuit, 928 F.3d 226 (July 2019); vacated as moot in 2021. Plaintiffs: Buckwalter-Poza, Cohen, Figueroa O’Reilly, Gu, Neely, Papp, Pappas.
Peer review history and the trust argument. On the dating: Melinda Baldwin, “Scientific Autonomy, Public Accountability, and the Rise of ‘Peer Review’ in the Cold War United States,” Isis 109:3 (September 2018) places the decisive moment in the 1970s funding fights, with Nature not requiring formal review of everything until 1973 and The Lancet later. Baldwin teaches in the history department at Maryland, which makes this one phone call. The Einstein correspondence of July 1936, and the fact that the referee’s objection reached him anyway through H. P. Robertson befriending his new assistant at Princeton, are documented in Daniel Kennefick’s work. The elevator and disembedded systems material is drawn from Anthony Giddens, whom Cohen credits.
The Bluesky paragraph. His five-post thread of July 27, 2026.
Extrapolations and interpretive moves. Made without citation: that a razor-blade splice, a live newscast, and a weekly music-review deadline train particular habits; that anyone building a federal household roster learns categories are constructed; that a tenured full professor bears less risk in leaving a professional association than a job candidate does. Two moves are flagged in the text itself. The paragraph pairing his mother’s distributed-control science with his own institution-building says explicitly that he has not drawn that line in print and asserts no causal relationship; her claim about hierarchy and nervous systems is quoted from his own post. And the reading that his most durable work is the checkable work is mine, offered as an argument.
The Set
The set has a center, and it is a library.
SocArXiv runs out of the University of Maryland Libraries, which means sociologist Philip N. Cohen serves at the pleasure of the dean of the libraries, as he puts it, and the dean pays for it, and the payment is small.
Around that center sit four groups that overlap.
The first is the open scholarship infrastructure world: the Center for Open Science, where Brian Nosek built the Open Science Framework that the archive runs on; MIT Libraries, where the Center for Research on Equitable and Open Scholarship hosted Cohen as a visiting scholar in 2018 and where Chris Bourg has been the most articulate dean in the movement; ASAPbio and Jessica Polka, who did for biology what SocArXiv attempted for sociology; Kathleen Fitzpatrick, who came out of the Modern Language Association arguing for open peer review before most sociologists had heard of a preprint. Cohen interviewed all four in the autumn of 2018, as pre-work for an invitational meeting hosted by the Association of Research Libraries and the Social Science Research Council, and wrote up the results in a report that carries the video. Librarians, a psychologist, a biologist, and a literary scholar. Not a sociologist among them, which tells you what holds the set together.
The steady collaborator is Micah Altman at MIT, who has coauthored most of Cohen’s scholarly communication work since 2021, including a 2023 letter in Science on peer review at NIH and a July 2025 essay in The Hill on the funding cuts. Elizabeth Popp Berman was on the original SocArXiv steering committee alongside Bourg.
The second group is family demography: Joanna Pepin at Buffalo, Kate Choi at Western, Brandon Wagner at Texas Tech, Ge Gao, Kelsey Drotning, Jeehye Kang, Jaein Lee, Hao-Chun Cheng, the doctoral line. Reeve Vanneman upstream as adviser, Suzanne Bianchi as the mentor who died in 2013, Lynne Casper as the early coauthor, Matt Huffman and Jessica Pearlman from the Irvine years. Stephanie Coontz is the senior figure whose book he reviewed warmly in 2026 and whose function in the set is to be the historian who established that the traditional family was a mid-century episode. Philip Levine, Melissa Kearney, Sarah Damaske, Jennifer Randles, and Jennifer Glass are the adjacent people whose work he engages, sometimes critically.
The third is the public sociology apparatus: Contexts, which he coedited with Syed Ali and Letta Page; The Society Pages and Sociological Images out of Minnesota, where Lisa Wade’s work sits and where Cohen’s blog posts circulate; Scatterplot and orgtheory, the discipline’s comment-section commons; the Council on Contemporary Families, which vets and releases research outside journals entirely and which he cites as a working model. Sociological Science, founded in 2014 by Jesper Sørensen and others precisely to publish fast with open access and a reproducibility policy, is the set’s demonstration project: the journal that does what he says journals should do.
The fourth is the reform wing that fights the American Sociological Association from inside. Committee on Publications members, the two hundred signatures on the 2019 petition against the association’s letter to the White House, the Publications Committee that passed a resolution the Council ignored.
What they value.
Access first, and it is a moral claim. The public paid for the research through grants and salaries, so the public owns it, and a paywall is a second sale of a thing already bought. Cohen says we do not aspire to have our work hidden from the people it most affects. The image that recurs is the researcher in a country whose libraries cannot buy the packages, reading the same paper as a researcher at Harvard.
Speed second. The set’s founding grievance is the twenty months between submission and publication at the discipline’s slowest journals, and the doldrums period in which a paper sits waiting for people whose arms are being twisted to read it while, as Cohen says, you hope there are people out there who want to.
Checkability third, and the rhetoric outruns the practice. Post the code, deposit the data, let a stranger recompute. Nobody recomputes. The value is held anyway, and Cohen has said in public that he cannot prove it.
Description fourth. The set prizes a fact a general reader can hold: divorce is falling, marriage is concentrating, the bottom tenth of Americans sits above half the world in income. The discipline treats descriptive findings as lesser, which Cohen says pushes authors into puffery.
The set’s exemplary man is the person who built the thing, who stood up a server, wrote the moderation policy, recruited the volunteers, and kept it running for ten years on a library line item. Cohen received a European prize in 2025 for that.
Second in the hero system is the man who pays a cost. The resignation letter published in public. The plaintiff’s name on the caption. The petition organized when the association’s leadership went the other way. Sacrifice is currency.
Third is the corrector who audits his own side. This is the set’s highest honor and its rarest performance, and Cohen’s standing rests on it more than on anything else: he went after Mark Regnerus, which was free, and then he went after Alice Goffman, which was not. He did it again in February 2024, in the Chronicle of Higher Education, where he took the side of a conservative critic against his own discipline while the discipline was under attack.
The villains. Elsevier, which bought SSRN in 2016 and gave the movement its founding trauma. Sage, which holds the association’s contract and, in the set’s most quoted line, throws the leadership a nice party at the conference. Springer, Taylor and Francis, Wiley: the five publishers who account for roughly eighty percent of recent sociology articles. And the association staff, who in this telling outlast the elected leadership, absorb reform proposals, and protect a revenue structure.
Since 2024 there is an outer villain too, and it changed the pitch of everything. In January of that year the Florida Board of Governors removed a general sociology course from the core-curriculum requirements of the state university system, and the state’s education commissioner said the discipline had been hijacked by left-wing activists. Cohen’s response in the Chronicle was not a defense of the discipline as it stands. It was an argument that sociology is vulnerable because it has not done the open science work that economics, psychology, and political science have already done, and that this vulnerability, while not the reason for the delisting, is one the field should urgently address.
The status games.
Prestige-journal placement still counts and cannot be admitted to count. The set’s members publish in American Sociological Review while arguing that its selection function is a status device. Cohen’s own resolution is to publish in Socius and Sociological Science where he can, both open access, and to post preprints of everything, which converts a compliance problem into a display.
Download counts are the substitute currency. Three thousand downloads of a preprint before the journal ran it. Seventeen thousand papers in the archive. The counts are visible, public, and comparable, which is what a status currency requires.
Being cited by a journalist ranks high and is not admitted to rank high. Cohen keeps a media page and updated it in January 2024 with an anonymous comment calling him the best sociologist of his generation, said ouch, and observed that outreach time is research time forgone.
And the highest-status move in the set is the audit. Counting other people’s compliance is a service to the community, a display of one’s own compliance, and an assertion of standing to judge. The December audits do all three at once, which is why they are the set’s signature genre.
The normative claims.
Research funded by the public belongs to the public. Delay in dissemination is a harm with victims. A finding that cannot be inspected is not yet knowledge. A professional association exists for its members, and one that transfers money from teaching-heavy libraries to research-heavy ones through its publishing contracts has inverted its purpose. Scholars are citizens and owe the public engagement. Values shape the questions anyone asks, so disclosure of standpoint is an obligation.
The Chronicle essay adds the one that binds the rest. You may not agree with our interpretation of the facts, Cohen writes, but you must know that our work, the data and methods, the funding sources, even our personal biases and opinions, is open to public scrutiny. Trust is not earned by being right. It is earned by being checkable, and the checkability is what the discipline owes in exchange for its standing.
The essentialist claims.
The first is about scientists: that a researcher, properly formed, wants his work checked, and that resistance to inspection indicates something about the resister. Under this assumption a person who will not share data is hiding.
The second is about institutions: that they follow revenue, and that a stated principle in conflict with an income stream will lose. It is usually right.
The third is about the public: that there is one, that it wants the research, and that access is the barrier. The evidence for this is thin. Removing a paywall does not create a reader who can evaluate an age-standardized rate, and the set has never addressed the gap between availability and competence. Cohen comes closest to noticing it in the Chronicle, where he grants that people judge sociology by how they feel about its conclusions rather than by its scientific merit, and calls that a problem of legitimacy. Then he proposes more openness as the remedy, which is the remedy his own diagnosis says will not reach.
The moral grammar.
The recurring word is owe. Researchers owe the public inspectability because the public funded the work. The association owes members an accounting. The journal owes readers the code. Obligation flows from having received something, which is why the argument is always about who paid.
Withholding is the characteristic sin and openness the characteristic virtue. In this world a boring correct paper posted publicly is better than a brilliant one behind a paywall, and Cohen’s willingness to say out loud that his platform contains a great deal of bad work follows. Quality is a lesser good than availability.
Confidentiality, human subjects protection, the ethnographer’s promise to a source, the reason a qualitative researcher cannot deposit a transcript: all of it enters as an exception to be managed. When Cohen was asked about it in February 2022 he improvised a restricted-access tier on the spot and said it was not his area. Nobody built it. Four years later it still does not exist, which tells you where the moral energy of the set is and is not.
Who is not in the set.
Ethnographers are not in it. Historians are not. Theorists are not. Anyone whose evidence cannot be deposited is structurally outside a movement whose central sacrament is depositing. The American Journal of Political Science required verification from 2015 and, three years in, had accepted no qualitative manuscripts at all under the policy. The set reads that as a coincidence. Everyone outside it reads it as the point.
And the activist wing is not in it either, which is the boundary most outsiders get wrong. Cohen writes in the Chronicle that the discipline includes a large activist wing that has successfully mobilized to win elections for leadership positions, and that this has magnified the target on our back. He names the association’s 2024 conference theme, Intersectional Solidarities, and its website promise to dismantle ongoing legacies of settler colonialism, as the sort of thing that makes an appealing target. He also notes that association membership had fallen to its lowest level since 1966.
His own position is stated in the negative twice in one paragraph, which is how you can tell it is the thing he cares most about. Sociology will not survive by becoming dispassionate, objective, or positionless. It will not survive as just another progressive-activist cause, or as what Christian Smith called the criminal investigative unit of the left wing of the Democratic Party. He quotes Smith’s line without rebutting it.
What he wants instead is a procedural settlement. Take a position, disclose it, and open the work to anyone who wants to prove you wrong. On the classroom, which is where the position is tested hardest, he is specific: his opinions are no secret to his students, he sued a president and won, and he always invites opposing views, does not censor those he disagrees with, and never grades for political perspective. His students need to know the divorce rate and the poverty rate. They do not have to share his moral perspective on any of it.
The line that carries the argument is about the other side’s demand rather than his own practice. In the same essay he notes that Florida’s replacement course promises to cover the horrors of slavery, and asks what kind of objective assessment that is. Objectivity, he writes, is what is self-evident to the speaker.
He also concedes, in one sentence, the charge that ought to worry him most. Some critics might say that public political pronouncements undermine not just his science but the reputation of the discipline.
The Voice
Cohen learned to write for a clock. Two versions of every story, each carrying a different quote, same length whether the story was a fire or a zoning variance, plus one-sentence headline versions for the quarter hours. At the end of the local news he wrapped the market report and the weather while the second hand came around, because at six the network took the air. Nothing in that job rewards a subordinate clause. Everything in it rewards knowing how long a thing takes to say and stopping when the time is gone.
Forty years later the sentences still land on the beat.
The default sentence. Subject, verb, number. “Marriage is becoming rarer and more durable at once.” “The rent eats first,” borrowed. “There’s less of everything demographic happening.” He builds paragraphs by stacking short declaratives and then, at the end, dropping in one shorter than the rest. The rhythm is a news reader’s rhythm: statement, elaboration, statement, stop.
Numbers as sentences. His signature move is to convert an abstraction into a quantity a person can hold. Twenty-nine of the years between eighteen and fifty-five spent married in 1960, eighteen now. Only eight of a hundred and twenty-seven papers reproduced on the first attempt. The bottom tenth of Americans sits above half the world’s population. Four of fifteen, then eight of twenty-five, then twelve of eighteen. Membership at its lowest level since 1966. He states them and moves, and the absence of adjectival help is what makes them land.
Concrete over abstract. He will say that making banana bread is a short-term response and redesigning your home is not. Printing was expensive and you could decide what got shipped and bound and stuck on a shelf. The Gini index arrives as four named people holding stacks of cash. Homophily arrives as liking people like yourself.
Register-jumping. This is the most distinctive thing in the prose and the hardest to imitate. He runs a technically exact sentence and then follows it with something conversational, sometimes crude, and the drop is deliberate. Software developers were much less affected than working-class jobs, thank you Zoom. He proposes a great methods assignment: condoms prevent births, so fewer condoms should mean more births, no, therefore people are having less sex. He tells a room that Google is not really your friend, it is okay to use Google tools because they are awesome, but they will not love you back in the end. The joke always does structural work, marking the end of a section or puncturing a claim.
The borrowed proverb is a subspecies of the same move. On arguing with Florida: it is a little like wrestling with a pig, we all get dirty, and the pig likes it. He uses it twice in one essay, once in the text and once pulled as a display quote, which suggests he knew it was the best thing in the piece.
Self-interruption as a habit. He corrects himself mid-sentence in public more than any writer I have read. Speculative, or I should say suggestive. It’s a harsh way of putting it. I realize later maybe that’s porn searches, I’m not sure. Don’t hold me to this methodology exactly. My methods are squishy. The corrections are the same reflex that makes him check whether a survey instrument changed before he explains a trend, turned inward and left in the draft.
The concession as a rhetorical engine. His strongest passages are built by giving the other side its best version first and then turning it. Maybe people don’t care about marriage anymore. Or maybe they care more and have put it on a pedestal so high their own relationship doesn’t qualify. Crime seems bad, and maybe crime helps a society mark the boundary of acceptable behavior, and poverty motivates people not to starve, which has a logic. Then the turn. This is a debater’s structure and he uses it in writing and in lecture.
The concession that does not turn is rarer and worth watching for, because it is where he is most honest. I can’t prove they’re wrong. I can’t expect readers to automatically embrace my scholarship if they hate my politics, they’re human, too. In each case he states the objection, declines to answer it, and keeps going.
Negation as a favorite figure. Constant. It is not so much that people are not getting married ever, it is that they are getting married later. Reproducibility is not really the issue, but maybe accountability is. It’s not the fact of having a position that paints a target on our discipline, it’s the particular positions we hold. The construction lets him kill a familiar reading and install his own in a single move, and he leans on it hard enough that it is a fingerprint.
Two verbal tics. He says “sort of” and “you know” constantly in speech and neither survives into his prose, which means he writes better than he talks and knows it. And he begins a striking number of written arguments with an admission of what he cannot do: we don’t measure this well, any number here is an estimate, nobody knows the rate.
Then the other voice.
The prosecutorial register. Post titles are the purest sample: Brad Wilcox lies a lot, part whatever. It’s not bigotry to say Lyman Stone is a bigot. The American Sociological Association is collapsing and its organization is a perpetual stagnation machine. The syntax is the same as everywhere else, declarative and short, and the content has moved from what a thing is to what a person is. “Part whatever” is doing specific work: it announces a series, signals contempt for the necessity of the series, and preempts the charge of obsession by acknowledging it first.
He still links everything. He still quotes the other side at length before answering. He still puts the arithmetic in. The prosecutorial posts are as heavily sourced as the audits, which is exactly why they unsettle people who would find an unsourced insult easy to dismiss.
There is a third register between the two, and the Chronicle essay is the best sample of it: the professional voice, aimed at colleagues, in which the prosecution is aimed at his own side. They are coming for sociology, and these are not good-faith actors, and also we have work to do, and also a conservative critic in the Wall Street Journal was right to call this a wake-up call. That combination is rare enough in academic writing to be worth naming. He gives the enemy no ground and his own tribe none either.
The teaching voice. A fourth mode, and the one with the largest audience. Patient, sequential, repetitive on purpose, and almost entirely free of the others. He builds from a case rather than a definition. He tells students what he is about to do, does it, and says what he did. He addresses them directly, warns them when a slide has a lot of numbers, tells them the take-home message is written in red. And in that mode he takes the opposing theory seriously in a way the blog does not: functionalism gets its fever-fighting-infection version, the March for Life teenagers get the same analytic respect as the Women’s March.
Email. No, don’t know, not to my knowledge, no, with care, see Popper, asked and answered, I disagree with the premise, yes. Twenty-three words for nine questions. This is the news-desk economy with the warmth stripped out, and it is a mode: he writes at length in his own comment section to people who ask him things. The venue determines the length.
Now the two things nobody says about the style.
The first is that his prose is a public-facing instrument built by a man who works entirely in private data. He spends his days inside survey microdata that a general reader cannot open, and he writes as though the reader is standing next to him at the terminal. The whole style is a bridge over a gap he never mentions. This is why description is his real subject: description is the only output of quantitative social science that does not require the reader to have a license.
The second is that the strongest and weakest sentences he writes are the same shape. Consider two. “If the facts don’t support the generalization, then the anecdotes are not illustrations of a trend. They’re just little stories.” And: “Brad Wilcox lies a lot.” Both are compressed, both are declarative, both refuse hedging, both were written by a man who believes that a plain sentence is more honest than a careful one. The first is the finest thing in his corpus. The second is the thing that keeps colleagues who accept every one of his statistical arguments from citing him.
The style has no setting for a claim that requires trust. It renders those claims at the same volume and in the same syntax as the checkable ones, and the reader cannot tell from the prose which is which. That is the direct cost of the virtue: a man who trained himself never to soften a true sentence has no equipment for softening an unproven one.
If you want to imitate him, three rules and one prohibition. Put the number in the sentence. Give the other side its best version before you answer it. Break the register once per section with something funny. And do not let the prosecutorial mode into a piece that also contains arithmetic, because he does, and it costs him every time.
Notes
The 2018 interviews. From cohen’s open science page: as an MIT CREOS visiting scholar he conducted short pre-work interviews with people on the invitation list for an ARL and SSRC invitational meeting held December 11 and 12, 2018, and videos from those interviews were included in a report he wrote for ARL. His page says “including” those four, and the report describes interviewing people from leading social science scholarly societies as well, so Bourg, Nosek, Polka, and Fitzpatrick are examples rather than the full list. The Scholarly Communication in Sociology report came out of the same residency.
The Chronicle essay. “How Sociology Can Save Itself,” February 7, 2024, in the print issue of February 16, 2024, and adapted in part from Citizen Scholar. Source for the Florida Board of Governors action, the education commissioner’s quotation, the association’s 2024 conference theme, the membership figure, the wrestling-with-a-pig line, the classroom paragraph, “objectivity is what’s self-evident to the speaker,” and “I can’t prove they’re wrong.”
The February 2022 talk. Scholars Strategy Network NYC, “Open Social Science and Public Engagement,” 82 minutes, slides here. This is the source for the improvised restricted-access tier at about 1:20:27 and for the ethnography concession at about 1:21:40.
One claim of mine that should be flagged as mine. That his remedy does not reach his own diagnosis: he says people judge sociology by how they feel about its conclusions, and then proposes openness, which addresses availability rather than feeling. He would answer that openness is a signal that earns trust.
‘Florida’s Shunning of Sociology Should Be a Wake-Up Call’
Jukka Savolainen writes Dec. 7, 2023:
I have taught undergraduate sociology courses since 1996. Through the decades, I have watched my discipline morph from a scientific study of social reality into academic advocacy for left-wing causes. Dozens of my fellow sociologists have also observed the transformation, including Christian Smith of Notre Dame, whose book “The Sacred Project of American Sociology” (2014) provides a brutally honest account of our troubled discipline.
Mr. Smith is disappointed that undergraduate sociology textbooks, rather than disseminate scientific findings, “function as recruiting tools and re-socialization manuals” to turn students into radical activists. He is equally disappointed with the discipline’s failure to come clean about its obvious political commitments. Publicly, the American Sociological Association describes sociology as a “scientific study of social life” interested in the “causes and consequences of human behavior.” Internally, ASA embraces and promotes social-change activism.
Each year, the association’s president chooses a theme for its annual meeting. Next year’s theme is brazenly political: “Intersectional Solidarities: Building Communities of Hope, Justice, and Joy.” The ASA sums it up as follows: “The 2024 theme emphasizes sociology as a form of liberatory praxis: an effort to not only understand structural inequities, but to intervene in socio-political struggles.”
In 2020 I started teaching a general-education course on the sociology of sports. Having never taught the subject before, I relied on a popular textbook. It didn’t take me long to discover that the book provided few interesting findings or persuasive arguments. Instead, it was peppered with tendentious grievances against capitalism, masculinity and “white privilege.”
‘How Sociology Can Save Itself: The recent political attacks in Florida represent a threat to the discipline — but also an opportunity.’ (Feb. 7, 2024)
Philip N. Cohen writes in the Chronicle of Higher Education:
Sociology has a tough row to hoe in this regard. For medical researchers or engineers, their social usefulness is often experienced tangibly. Social science may be useful, like when we do social surveys, make pandemic projections, or provide cultural analysis. But with sociology, people often judge us because of how they feel about our conclusions rather than their scientific merit. That feels great when people love us, and hard when they hate us, but either way it’s a problem of legitimacy. Contrary to our critics, we’re not going to get out of this situation by following those who want to make our research dispassionate, “objective” or positionless. But neither will we succeed as just another progressive-activist cause, or what the sociologist Christian Smith has called “the criminal investigative unit of the left wing of the Democratic Party.”
Like it or not, our work is part of a knowledge system that needs public support and trust. We earn this trust partly by our tangible usefulness, but also by demonstrating some key values of our scholarly approach: openness, transparency, and accountability. You may not agree with our interpretation of the facts, but you must know that our work — our data and methods, our funding sources, even our personal biases and opinions — is open to public scrutiny.
Our steadfast dedication to integrity and transparency should be a core asset in the public discourse. At our best we also openly embrace uncertainty, which is inherent to all scholarship. And our ethical standards require us to publish detailed instructions for anyone who wants to prove us wrong. These traits are certainly not demonstrated by those attacking the field. Unfortunately, as has been widely documented, American sociology sorely lags behind with regard to the practices of open science (and open scholarship more generally) — practices more widely embraced in, for example, economics, psychology, and political science — as well as open access in our publications. To be sure, this is not why our courses were delisted in Florida, but it’s a vulnerability we should urgently address. The openness imperative extends to the classroom…
I always invite opposing views, don’t censor those with whom I disagree, and never grade for political perspective. My “Family Sociology” students need to know the divorce rate, the poverty rate, the failure of the United States to provide work-family support like other rich democracies do, and the consequences of rolling back abortion rights, but they don’t have to share my moral perspective on these issues. My classroom is as safe as I can make it, but that does not include protection from hearing opposing or unpopular opinions and assumptions. We debate facts and opinions, and where to draw the line between them, but I won’t let that distinction be erased. I hope (and believe) this remains true in the majority of sociology classrooms.
The Great Delusion
“Sociology is interested in where the macro and micro meet, where biography and history meet, and a lot of that happens in families. It’s where social structure becomes internalized into our personalities early in life.”
That is Philip Cohen, in a recorded talk posted on August 1, 2020. It is also, almost word for word, the claim John J. Mearsheimer (b. December 14, 1947) makes in the first chapter of The Great Delusion and uses to demolish political liberalism. Humans arrive inside groups that form them before they can assess anything. The long childhood is the crucial fact. By the time a man can reason, the reasoning arrives to a mind already furnished.
So the frame starts with Cohen the sociologist as a Mearsheimer anthropologist, and Cohen the citizen scholar as the liberal individualist Mearsheimer is writing against.
The success sequence argument is an anti-atomistic argument. People who already possess schooling, employment, health, and a suitable partner cannot be converted into an instruction for people who possess none of them, because the instruction assumes a chooser standing outside his circumstances. The marriage market work is the same claim made arithmetically: you cannot select a spouse who is not there, and in most American metropolitan areas Black women are choosing from a smaller pool of unmarried employed men than White women are. And the confidence argument, which he stated in August 2020 and repeated to undergraduates in March 2025, is that people marry when they believe the future will honor a long commitment. Poor people often do not, because the universe has given them no reason to expect reciprocation.
That is a claim that the social order reaches into private decisions and produces them, which is what Mearsheimer says.
Cohen’s project rests on inspectability. Publish the code, count the packages, show the boundaries behind the generation labels, and the claim will be checkable by anyone. The rights language is unmistakable when you look for it. Everyone should be able to read the paper. A researcher in a country whose libraries cannot buy the packages should read the same article as a researcher at Harvard. He sued a president on the theory that a citizen cannot be excluded from a public forum for his views. And Citizen Scholar proposes that a man occupies separable roles, researcher and teacher and activist and litigant, and that the trouble starts when he moves between them without announcing it. Announce the movement and the problem is managed.
That is political liberalism: universal access, inalienable standing, and a self that composes itself from chosen roles and can be held to account for the composition.
Mearsheimer’s anthropology says the last part is not available to anyone. You do not assemble an identity out of roles. You were assembled, in a particular house, in a particular town, before you could evaluate the materials.
Cohen’s assembly is documented by Cohen. A mother who argued in public that the nervous system was less hierarchical than male scientists had assumed and that this bore on how people should think about their societies. That same mother running a funded laboratory at Cornell for ten years and never receiving a tenure-track offer, which made institutional injustice a fact of the household. An alternative high school built on participatory education. The Ithaca left of the 1980s, its labor and environmental and alternative-media world. A grandmother from Poland behind the counter of a liquor store, and a Yiddish accent one generation back.
Set that beside the man’s convictions at fifty-eight. Prestige hierarchies are suspect. Credentialing is a guild move dressed as epistemology. Institutions protect their revenue and call it standards. The people who own the machine should not define what the machine is for.
Mearsheimer’s claim is that no amount of subsequent argument produced any of that. The value infusion came first and the arguments arrived afterward, well made and sincerely held, to serve commitments already in place. Mearsheimer says this is true of everyone, himself included, which is why he thinks liberal universalism cannot work. It is an accusation only against a program whose founding premise is that first-order commitments should be inspectable, held by a man whose own first-order commitments arrived before he was old enough to inspect anything.
Cohen half sees it. In February 2025 he told a Wisconsin audience that you do not get to have different identities in different contexts unless you really work at it, with anonymity. He offered that as a fact about social media. Mearsheimer would say it is a fact about men.
Sort Cohen’s campaigns by outcome and the pattern is not the one he believes in.
The Pew campaign worked. He addressed a National Academies committee in July 2019 with an argument about arbitrary boundaries, and the committee thanked him. What moved the institution was a letter two years later carrying roughly a hundred and fifty signatures. A profession asserted a norm, and Pew, whose standing depends on that profession’s regard, adjusted.
The transparency counts worked. The share of quantitative papers in the flagship journal arriving with code went from four of fifteen to twelve of eighteen across five years. And the moment that shows why is in his own comment thread in December 2025, where an author he had marked as providing nothing agreed, promised a package, and thanked him. That is a man responding to shame inside a group whose regard he needs. It is not a man who read an argument and updated.
The Wilcox fight has moved nothing. Two demographers with access to the same data, both competent, arguing about selection into marriage, and neither has conceded an inch. Mearsheimer’s account of why is that they belong to different moral communities with different first principles, and reason cannot adjudicate between first principles. Cohen’s account of why is that Wilcox lies. His account requires bad faith. Mearsheimer’s does not require anyone to be dishonest, which is a point in its favor, since Wilcox’s readers do not believe he is dishonest and never will.
And the association. Three years inside the Committee on Publications, arguments made in the proper forum with the proper evidence, two subcommittees, a resolution passed, a Council that ignored it. Reason, applied at length, in the room built for it, moving nothing. Then exit and public argument, which cost him his membership and moved the association not at all, and moved his own coalition considerably.
So the record fits the anthropology better than it fits the program. Cohen’s counts change behavior when they activate group norms, and they change nothing when they collide with a rival group. He believes he is supplying reasons. What he is mostly supplying is shame, delivered inside a community that has already agreed what is shameful.
The last application is the machine.
The paper that arrived at his archive in the winter of 2025 satisfied every requirement inspectability can state. Abstract, real citations, competent models, public code, a disclosure listing the tools. What it lacked was a member. There was no community standing behind it, no socialization, no group whose regard the author needed, nothing that could be shamed.
The policy Cohen’s committee published on March 9, 2026 refuses automated detection, declines to state a rule that can be written down, commits the moderators to judgment made by people using perception and experience, and says: we feel the need to express our humanity in this process.
That is a group asserting membership at the point where argument ran out, which is what Mearsheimer says people do.
Now the objection. Mearsheimer’s anthropology under-predicts the divorce decline paper. Cohen produced a finding his own side did not want, that stable marriage is concentrating among people who already have resources, and then went further and itemized the losses inside the gain, and said out loud that divorce is falling partly because marriage is available to fewer people. He attacked a celebrated ethnographer whose politics matched his and paid for it. He conceded to a National Academies committee that his central claim about generation labels was an intuition with no research behind it. He told a room of demographers that a chart fit his presumption and that this was probably why he was showing it.
If socialization dominates and reason ranks last, this behavior needs explaining. The available answer is that his tribe is sociologists who check things, and that every one of those acts was loyalty to the smaller group. That answer works. It also works too well. Redescribe the tribe finely enough and any behavior becomes group loyalty, at which point the frame explains everything and predicts nothing.
Reason works inside a community that has already agreed on what counts, and it does something other than persuade even there. It supplies the terms in which shame can be applied. Cohen’s arithmetic is the instrument through which a professional group enforces a norm it already held. That is why the counts move journals and the verdicts move nobody.
Cohen has said the thing Mearsheimer would say, in his own words, in a classroom, teaching George Herbert Mead. You cannot have a self if you never interact with anybody else. He teaches it to freshmen every fall and then proposes a method that asks a man to stand outside himself and disclose where he stands.
Nobody stands there. That is the anthropology, and if it is right, the honest version of his program is smaller than the one he has written: not a universal right to inspect, which requires a public does not exist, but a professional community policing itself with numbers because numbers are the only shame it recognizes.
Notes
Frame. John J. Mearsheimer, The Great Delusion: Liberal Dreams and International Realities (Yale, 2018), chapters 1 through 3. The anthropology I used: the primacy of the social, the three sources of preference with reason ranked last, the long childhood and value infusion, the innate sentiments, and the resulting limits on anyone’s freedom to formulate a moral code. The Samuel Moyn (b. 1972) quotation on human rights is from The Last Utopia (2010).
Mearsheimer’s anthropology is separable from his foreign policy conclusions and from the controversies attached to his name.
Cohen material. Every quotation and every claim about him is sourced in the biography notes and in the lecture indexes already in this file.
The claims that are mine. First, that Cohen’s substantive demography is anti-atomistic in Mearsheimer’s sense while his public program is liberal individualist, and that the split runs through Citizen Scholar. Second, that sorting his campaigns by outcome shows reason failing at community boundaries and succeeding inside them, which fits the anthropology better than it fits his own account of why his arguments work. Third, that the March 2026 policy is a tribal move. Cohen would dispute the third. He might say: a community stating that it will exercise judgment is not the same as a community asserting that its judgment needs no reasons.
What would strengthen this. Mearsheimer is at Chicago and answers email. So does Cohen. The question that would produce a new sentence is the same one for both: whether a man can hold a first-order commitment he acquired before he could evaluate it and still demand that all commitments be inspectable. Neither has been asked it in these terms as far as I can find.
‘A Big Misunderstanding’
David Pinsof says intellectuals blame the world’s troubles on misunderstanding because that story makes intellectuals indispensable. If polarization, bigotry, war, and poverty all descend from bad beliefs, then the people whose trade is correcting beliefs hold the cure. Pinsof says beliefs follow interests. People understand what they have an incentive to understand. Stupidity is strategic. The hole you are studying is the hole you are stuck in.
Philip N. Cohen looks like a soft target for this. He has written a blog since 2008 whose recurring move is the correction. A conservative think tank releases a report on marriage and children, Cohen shows the model omits selection. A magazine writer announces a fertility collapse, Cohen posts the age-adjusted numbers. A study on same-sex parenting circulates through the courts, Cohen goes after the sampling. He founded SocArXiv to route sociology around paywalls and slow review. He has called for corrections and retractions in his own discipline and complained that sociology runs on trust because almost nobody checks anybody’s work. Read the surface and you get a man who believes the world goes wrong because people believe false things, and who has organized his career around fixing that.
The trouble is that Cohen has spent twenty years arguing the opposite about the subject he knows best.
Take marriage and poverty. The conservative account holds that poor Americans marry less, that their children suffer for it, and that the remedy runs through norms. Teach the sequence, honor the institution, and the outcomes improve. That account is a misunderstanding story with poor people cast as the ones who misunderstand. Cohen has rejected it for his entire career, and he rejects it on Pinsofian grounds. Marriage rates track the supply of marriageable men, which tracks employment, wages, incarceration, and mortality. His work with Joanna Pepin on divergent marriage markets makes the constraint visible. When federal marriage promotion money produced no measurable change in family formation across two decades, Cohen did not conclude that the message needed better framing. He concluded that people were responding sensibly to their circumstances and that if the goal were reducing child poverty, we already know the method, which is giving parents money.
Pinsof would recognize an ally here. Cohen grants poor Americans the presumption of rationality that Pinsof wants extended to everyone. He also grants it, in a colder register, to his opponents. He has written that the durability of marriage promotion after its failure suggests the program was never aimed at poverty. He treats the National Marriage Project and its allies as an ideological movement with religious and political objectives. He treats the peer review system as a set of incentives protecting careers and journal revenue.
Watch the direction Cohen’s explanations point. Downward, toward the people he studies, the account is structural. Poor women are not making errors, they face a market. Sideways, toward his rivals in the status order, the account turns epistemic. The vocabulary shifts to know-nothingism, propaganda, lies, a war on truth that the forces of truth must meet. His 2025 argument for the citizen scholar rests on the premise that scholarly speech, delivered clearly enough and often enough to lay audiences, changes what the public believes and therefore what the public does.
Pinsof predicts that asymmetry, and predicts where it lands. Competition runs hottest between near rivals. Cohen has no status contest with a single mother in Baltimore. He has a live one with Brad Wilcox, with the family scholars who testify for the other party, with the demographers who supply pronatalist columns, with the editors and associations who decide what counts as sociology. Against those rivals the epistemic charge is the sharper weapon, because it does not merely say the other man wants different things. It says he does not know things. Once you say that, his credentials no longer transfer, his testimony carries no weight, and his students inherit a discount. Methodological critique is the most efficient way to strip a colleague of standing while appearing to do nothing but arithmetic.
The critique can be a weapon and still be correct. Cohen’s technical objections have often held up, and a bad study remains bad regardless of who profits from saying so. Pinsof’s frame says nothing about whether the arithmetic is right. It says the arithmetic gets done selectively, on targets that pay.
That gives a test Cohen’s archive can answer. He has posted more than a thousand times. Sort the posts by whether the diagnosis is error or interest, then sort by the political valence of the target. My prediction is a gap: the marriage promoters get interest analysis when Cohen is explaining why their programs persist and error analysis when he is explaining why their claims are wrong, and the second is far more common than the first. Do the same with his correction and retraction requests. Count them by the direction the flawed finding cuts. If the calls fall evenly across the political map, the Pinsofian reading fails and Cohen has a strong defense. If they cluster, the open science project looks less like a general standard and more like an enforcement arm with a general standard’s name on it. This is countable. Somebody should count it.
The open science work contains the contradiction in miniature. Cohen’s diagnosis of why bad sociology survives is thoroughly incentive-based. Journals want prestige, associations want revenue, authors want jobs, reviewers want cover, nobody wants the trouble of checking. His remedy is transparency. Preprints, data, code, public criticism. Pinsof’s question follows on its own: if the cause is incentives, why does sunlight help? Cohen has a good answer available, which is that exposure raises the price of getting caught and therefore changes the incentives. That answer is Pinsofian and Cohen half-gives it. But the version he tends to give in public is the other one, that we are scientists, that truth is our business, that we owe the public accuracy. The first answer is a claim about costs. The second is a claim about virtue. He reaches for the second when the audience is watching.
Cohen’s standing question about conservative family research is whether it functions as science or as legitimation for a political program. Ask that question of Family Inequality, of SocArXiv, of the citizen scholar. The honest answer is that they function as both, which is also the answer about his opponents. What Cohen gets from the work, in addition to whatever public benefit accrues, is a position: the sociologist whose independence and technical care make him the one who says what the numbers show. That is a valuable seat, and holding it requires a steady supply of people who have gotten the numbers wrong.
Pinsof’s essay ends by announcing that nothing can be done, the world does not want saving, the study of human nature is the study of the hole. That conclusion buys its author the highest seat in the room at low cost. If everyone’s stated motives conceal a status play, the man announcing this is running one, and he would agree, and the agreement costs him nothing. A frame that fits every case discriminates between none of them. Its value has to come from the predictions it makes about particular people, in places where the prediction could fail.
Cohen has documented, in his own research, that a twenty-year federal campaign to change beliefs about marriage changed no behavior, because the constraints were never beliefs. Then he wrote a book urging scholars to write as though the truth mattered, on the premise that clearer explanation from credentialed people moves a public that has an interest in believing otherwise. The evidence he gathered about his opponents’ project applies to his own. If information does not move people whose circumstances point elsewhere, then a thousand well-sourced blog posts do not move an electorate whose coalition points elsewhere either.
Cohen may have an answer. The strongest one available to him is that the citizen scholar project was never about the public, that its working audience is other scholars, and that its effect is to shift what a discipline will let pass. That answer is closer to true, and it is also the Pinsofian answer.
Cohen’s blog, Family Inequality, contains more than a thousand posts. Again and again the action is epistemic. Someone makes a claim. Cohen checks the denominator, comparison group, sampling procedure, causal inference, historical trend, spreadsheet, funding disclosure, or peer-review process. A conservative think tank says marriage reduces poverty. Cohen asks whether selection into marriage explains the result. A writer discovers a generation with a new personality. Cohen asks where one cohort ends and another begins. Alice Goffman reports an extraordinary survey from a Philadelphia neighborhood. Cohen asks where the questionnaire and sampling information are. Mark Regnerus publishes findings about children with gay parents. Cohen reconstructs the study’s chronology, funding, peer review, and political use.
Cohen built an institutional program around this instinct. Since 2016 he has been the founding director of SocArXiv, the open social-science archive. He describes its purpose as opening social science to more readers while improving research and scholarly communication. His 2025 book Citizen Scholar: Public Engagement for Social Scientists extends the argument into a theory of scholarly vocation. Open scholarship, peer review, social media, activism, and public communication become parts of the same enterprise. Columbia University Press describes the book as an argument for a reciprocal relationship between professional scholarship and active citizenship.
The Pinsofian suspicion is obvious. Perhaps Cohen’s corrections are weapons in disputes over policy, prestige, and professional authority. A methodological objection can be completely correct and still be selectively deployed. The question is why his arithmetic gets done on this claim, by this person, at this moment.
Cohen’s archive lets us start counting.
One test is his own “marriage promotion” archive. It contains 42 posts across five pages, from his 2011 discussion of the federal Healthy Marriage and Responsible Fatherhood programs through his 2026 posts. The first four archive pages contain ten posts apiece and the fifth contains the final two.
I coded each by the principal diagnosis Cohen makes.
“Error” means that the main explanation is that somebody has made a factual, statistical, methodological, causal, or inferential mistake.
“Interest” means that Cohen explains the behavior or persistence of the target through ideology, religion, money, organizational survival, political objectives, professional prestige, constituency maintenance, or some related incentive.
“Mixed” means both explanations do substantial work.
“Neither” covers posts carrying the tag that do not really diagnose a marriage promoter in either fashion.
My first-pass coding produces 28 error-dominant posts, four interest-dominant posts, four mixed posts, and six that fit neither category. This classification has judgment in it. Another coder could move several border cases. The imbalance is much too large to disappear through reasonable recoding.
So the first Pinsofian prediction does surprisingly well.
When Cohen asks whether marriage promotion works, he reaches for error analysis.
The pattern is there from the beginning. In “Healthy Marriage and Responsible Fatherhood: Time for some results”, Cohen looked at the federal programs that had spent hundreds of millions of welfare dollars promoting marriage and fatherhood. The Building Strong Families experiment had randomly assigned more than 5,000 couples to services or a control group. Cohen’s summary of the principal result was one word: “Nothing.” His initial assessment was that the programs were a waste of money, pending further research. The question is framed as an empirical test. The program promised results. Where are the results?
The same structure recurs in “Marriage promotion: That’s some fine print”. Cohen attacks what he calls a selective literature review, the research design, migration problems, the absence of appropriate controls, and the distance between the evidence and the authors’ conclusion. He also notes that lead author Alan Hawkins had received marriage-promotion funding, but the core of the post is methodological. The finding is wrong because the analysis cannot establish what its authors say it establishes.
In “Who are you gonna marry? That one big assumption marriage promotion gets totally wrong”, Cohen and Joanna Pepin’s work on marriage markets supplies another kind of objection. White women aged 20 to 45 lived in metropolitan markets with about 118 unmarried White men for every 100 unmarried White women. Black women faced about 78 unmarried Black men for every 100 Black women. Cohen treats this as evidence for structural constraint over cultural deficiency. More important for marriage promotion, the people who have not already married are not interchangeable with the people who did. The hypothetical marginal marriages produced by a policy would not necessarily contain the economic advantages observed in marriages that already exist.
Five years later he formalized the point in “Partner prospects and the marriage promotion fallacy”. Women who had recently married had economically stronger partners than otherwise similar women who were cohabiting. Among the cohabiting women, those statistically closest to marriage were paired with better-positioned men than those least likely to marry. Cohen’s conclusion was that the hardest couples for marriage promotion to reach were also the couples likely to receive the smallest economic return from marriage. The average existing marriage therefore tells us little about the marriage a successful government program might produce.
The revealing sentence in that post comes before the models. Cohen calls it a “mind-boggling suggestion” that unmarried people might know something about the costs and benefits of the relationships available to them and “are already acting accordingly.”
That is Pinsof.
Poor people are not failing to understand marriage. They are responding to their circumstances.
But another Cohen appears when the question changes from “Is this claim true?” to “Why does this movement continue?”
That Cohen sounds even more like Pinsof.
His 2015 discussion of Melanie Heath’s One Marriage Under God is an example. In “Book review: One Marriage Under God”, Cohen accepts Heath’s account of marriage promotion as a coalition joining religious morality, market ideology, opposition to welfare, gender norms, and heterosexual family ideals. Poor single mothers may find the marriage classes useless. That does not mean the program is failing at everything it does. Cohen writes that if marriage promotion reproduces inequality and reinforces a particular normative model of family while failing to increase marriage or reduce poverty, the last fact does not necessarily mean the movement itself has failed.
His “The marriage movement has failed (long live the marriage movement)” makes the same move more aggressively. Cohen starts from the movement’s failure on its stated demographic objectives and then asks how its institutions continue. His explanation moves through ideology, organizations, political constituencies, fundraising, and the ability of movement entrepreneurs to redefine their projects. Success cannot be measured solely by whether national marriage rates rise if the institutions involved sustain themselves and their political position.
Brad Wilcox gets both Cohens in a single 2023 post.
In “The silly season of Brad Wilcox’s new book is upon us”, Cohen first asks how marriage promoters understand decades in which marriage continued to decline despite their campaigns. His explanation is sociological. Failure can sustain movements. Crisis creates demand for advocates. Followers confer prestige and resources. A movement need not reverse national marriage trends to reward the people and organizations within it.
Then Cohen turns to a statistic Wilcox supplied to David Brooks about happily married people being dramatically more likely to report being very happy with life. Now the incentive analysis disappears and the statistician arrives. Cohen retrieves the General Social Survey, reconstructs the comparison, explains the odds ratio, and notes that the predictor includes happiness in marriage while the outcome measures happiness in life. His punch line is perfect Cohen: “Happiness is very correlated with happiness.”
The sequence is almost a controlled experiment.
Why does the movement persist? Interests.
Why is this particular statement wrong? Error.
The second answer is much more common across the 42 tagged posts.
That supports one part of the Pinsofian case. Cohen knows how to explain ideological opponents without treating them as confused. He does it when the dependent variable is persistence. But when the dependent variable is a public claim, he usually returns to epistemology. The movement has interests. The proposition has an error. Correcting the proposition becomes the sociologist’s job.
The more difficult test is whether Cohen applies that job description across political lines.
In “Stop that feminist viral statistic meme”, Cohen attacked a politically convenient statistic in international feminism. Women, it was endlessly said, performed two-thirds of the world’s work, received one-tenth of its income, and owned less than one percent of its property. Cohen traced the claim backward through the United Nations, the International Labour Organization, Robin Morgan’s Sisterhood Is Global, and earlier documents. He concluded that the numbers had never been adequately substantiated. His opening made the political situation explicit: “And yes, I really am a feminist.” He told feminists to stop repeating the statistic and concluded that debunking it would not hurt feminism.
This strikes directly at the Pinsofian prediction that people strategically fail to notice errors useful to their own side. Cohen noticed one, investigated it, and publicized it. The post became one of the blog’s enduringly popular entries. In his 2012 retrospective he noted that it continued to receive traffic, including from anti-feminist websites, and repeated that he was a feminist and that women owned more than one percent of world property.
Hanna Rosin received similar treatment after The End of Men. In “Correct that error, Hanna Rosin edition”, Cohen compiled a list of factual errors he wanted corrected in the paperback. They concerned women in Congress, domestic violence, Auburn, sexual assault, marriage ages in Asia, cosmetic surgery, and Brazilian women’s earnings. His larger complaint was revealing. Rosin appeared to be using facts and numbers to confirm what she had already decided was true. Cohen added that he had not found errors running against her thesis.
In “Quick correction on that 90-percent-of-faculty-are-White thing”, Cohen turned on an anti-racist statistic circulating on social media. The claim was that nearly 90 percent of full-time professors were White. Cohen traced the figure to a confusion between “full-time professors” and the much narrower category of full professors. The politics of the people repeating the mistake did not rescue the statistic.
Alice Goffman is more damaging still to a simple partisan interpretation.
Cohen broadly shared the political concern underlying On the Run. He thought aggressive policing and mass incarceration imposed serious injustices on poor Black communities. Nevertheless, in “On the ropes” he became suspicious of the evidentiary structure of Goffman’s celebrated ethnography. He compared reported household numbers with Census figures, raised questions about extraordinary incidents in the narrative, objected to the unavailability of the dissertation, and questioned the survey attached to the project.
He pursued the survey in “On Goffman’s survey”. Cohen explicitly warned against trying to destroy Goffman personally or delegitimize her career. He nevertheless argued that there were errors and unresolved methodological problems in the survey and posted a draft formal comment intended for American Sociological Review. He even objected when a journalist described him as a “rival academic” who might benefit if Goffman’s work failed. Cohen wrote that he could not see how he benefited and noted that the criticism was annoying some of his own friends and colleagues.
Then he tried to move the correction into the formal scientific record. “Comment on Goffman’s survey, American Sociological Review rejection edition” reproduces the journal’s rejection letter and the reviewers’ responses to his submitted comment. Cohen’s complaint was institutional. If an article contains meaningful errors, future readers of the journal need a correction they can find.
The case matters because Goffman’s research did not cut in the conservative direction. It dramatized police surveillance and the criminal-justice system’s effects on poor Black men. Cohen had little obvious ideological reward for weakening it.
Cohen also corrects himself.
In “Correction: Black and White marriage pools”, he disclosed that he had reversed two rows in an Excel spreadsheet used for a figure he had presented for years, including in class. He repaired the old pages and ended: “Family Inequality regrets the error.”
In “Poverty is not just for single mothers”, he opened with another correction. A trend in the original post had been miscoded, showing percent married instead of percent single. Cohen called it a “major error,” fixed the figure, and apologized.
None of this proves political neutrality. It establishes something narrower and important. Cohen possesses a real accuracy norm that can override immediate ideological convenience.
The same is true institutionally. In 2021 Cohen organized an “Open letter to the Pew Research Center on generation labels”. Pew was not a conservative adversary. Cohen and the other signatories argued that labels such as Millennial and Generation Z imposed arbitrary cohort boundaries, encouraged stereotyping, and acquired false authority because Pew was a trusted research institution. The letter explicitly requested a “public course correction.”
SocArXiv’s withdrawal of an ivermectin paper is harder to place politically. In “Mexico City’s program to distribute ivermectin to 200,000 COVID-19 patients, and the bad science that propped it up”, Cohen explained that the SocArXiv Steering Committee had removed a widely downloaded study that it considered scientifically and ethically deficient. He took pains to say that he was “blissfully ignorant” of the Mexican political context and had trouble even determining how its factions mapped onto left and right.
Then there is Regnerus.
Here Cohen escalated much farther.
His first famous intervention, “Time travel: Regnerus study timeline suggests superhuman abilities”, noticed that Social Science Research received the paper on February 1, 2012, while the study’s own chronology had data collection continuing later in February. Cohen thought the paper was bad research that should not have been published and mocked the extraordinary speed of the review process.
He subsequently pursued the funding. In “‘More managerial than intellectual’: How right-wing Christian money brought us the Regnerus study”, Cohen used documents obtained through public-records requests to reconstruct relationships among Regnerus, the Witherspoon Institute, Brad Wilcox, consultants, and the research project. He emphasized Witherspoon’s political objectives and contrasted those records with Regnerus’s public description of the funder’s role.
His “Regnerus Affair timeline” assembled the chronology in one place, beginning with correspondence in 2010. One Witherspoon email said it would be useful to have the research completed before major Supreme Court decisions.
Cohen also pursued peer review. In “Paul Amato, Regnerus postscript”, he discussed emails showing that Regnerus had suggested Amato as a reviewer after Amato had been a paid consultant on the project. At that point Cohen said the documentary evidence that Amato reviewed the paper was suggestive.
Ten days later Amato supplied the answer. In “Paul Amato on reviewing Regnerus”, Cohen published Amato’s account acknowledging that he had been paid as a consultant, disclosed that role to the journal editor, and then reviewed the manuscript at the editor’s request. Amato wrote that in retrospect agreeing to review the paper “was not a good idea.”
Cohen then crossed the line from criticism to sanction. He wrote that the Regnerus paper should be retracted because its conclusions were demonstrably wrong, because Regnerus had misrepresented the funder’s involvement, and because the review process had been compromised. He announced that as long as James Wright remained editor he would neither submit papers to nor review for Social Science Research.
That looks very different from “Quick correction on that 90-percent-of-faculty-are-White thing.”
But it does not follow that politics explains the difference.
The paper entered litigation over same-sex marriage. A conservative foundation financed the project. People involved in developing the study also entered the publication process. Amato acknowledged being a paid consultant and reviewer. Regnerus eventually participated in a Supreme Court amicus brief invoking the research in the same political controversy he initially said the paper was not designed to resolve. Amato himself criticized both sides and called Regnerus’s later political use of the paper “disingenuous.”
Severity therefore cannot simply be counted as evidence of bias.
This exposes a problem with the original test.
If Cohen issues twenty corrections against the political right and five against the left, that does not establish selective enforcement unless he encountered comparable numbers of comparable errors from both directions. His beat is family demography. For decades one of the major organized political projects inside that territory has been conservative marriage promotion. We should expect his archive to contain more conservative marriage claims than conservative claims about agriculture, foreign policy, or monetary theory.
Equal raw counts are therefore the wrong null hypothesis.
The relevant denominator is opportunity.
Given a claim Cohen thinks is seriously wrong, what predicts what happens next?
Does it receive a passing correction?
A short blog post?
A technical replication?
A submitted journal comment?
A demand for correction?
A demand for retraction?
An investigation of funding and correspondence?
A boycott of the journal?
Years of follow-up?
That is a better Pinsofian test because it measures prosecutorial discretion.
Cohen’s friendly-fire cases establish that the law is real.
They do not establish that prosecution is evenly distributed.
Pinsof’s strongest insight concerns attention.
People choose what deserves inspection.
The intellectual’s scarce resource is scrutiny.
Cohen has only so many hours.
Who gets audited?
The pattern in Family Inequality suggests that political valence is part of the answer, but professional proximity may matter even more.
Brad Wilcox and Mark Regnerus operate in Cohen’s substantive territory. They make claims about marriage, family structure, poverty, children, and demography. Their work reaches journalists, policymakers, courts, foundations, students, and other scholars. Melissa Kearney’s The Two-Parent Privilege enters the same territory. Cohen’s 2024 “Review of Melissa Kearney’s The Two-Parent Privilege” grew out of several posts in which he challenged both her interpretation of family trends and the policy implications she drew from them. The political differences matter, but these are also competitors over who gets to describe the family to the public.
Alice Goffman complicates the political story because she was not on the right. She was, however, a suddenly prominent sociologist whose research had been praised by Cornel West, Malcolm Gladwell, Christopher Jencks, and other high-status intellectuals. Cohen subjected her to serious scrutiny and attempted to correct the formal record.
A more Pinsofian variable than left versus right may therefore be proximity in the market for intellectual authority.
Near rivals pay.
The archive gives additional reasons to take Cohen’s general standard seriously. His open-science campaign has increasingly moved away from individual adversaries toward rules applying to everyone.
In “Quick update on American Sociological Review’s failure to meet current social science standards”, Cohen audited an entire volume of American Sociological Review. His earlier 2020 check had found only four of fifteen quantitative papers supplying enough data and code for replication at a glance. In the 2023 volume, he found replication packages for eight of twenty-five quantitative papers, with several others providing only partial materials. This is quite different from selecting an ideological adversary and trying to break his paper. The unit of analysis is the journal.
He repeated the exercise for the 2025 volume in “Another update on American Sociological Review’s failure to meet current social science standards”. This time two-thirds of the quantitative papers supplied replication packages with code and at least some information about obtaining the data. Cohen welcomed the improvement while arguing that a third of quantitative papers without these materials remained unacceptable.
That post contains a revealing piece of evidence for the correction question. Cohen says that he has twice submitted comments to ASR intended to correct erroneous articles and that in both cases editors acknowledged errors but declined to correct the record. The Goffman case is public. Cohen says he has never publicly identified the other one. The blog therefore cannot supply a complete census of his attempts at formal correction.
More important is his proposed solution.
Cohen wants transparency and reproducibility standardized so that everyone confronts common rules. On his own open-science page, he summarizes the SocArXiv principle as making social science more open, accountable, and effective. In a draft chapter of Citizen Scholar on “Open Scholarship”, he reduces the principle to three commands: be accountable, be transparent, be open.
That answer is much closer to Pinsof than it first appears.
If bad science persists because authors want publications, journals want prestige, associations want revenue, reviewers want convenience, and almost nobody receives much reward for checking old work, exhorting everyone to love truth is unlikely to solve the problem.
Change the costs. Require code. Require data when possible. Make materials inspectable. Make errors easier to discover. Make concealment more expensive. Make correction normal.
By March 2026 Cohen was saying this. In “There is a tsunami of fake social science research coming for us, and the American Sociological Association doesn’t care”, he worried that generative AI would make fabricated social science increasingly cheap. His proposed answer included strong institutional rules and “enforcement mechanisms that create a credible threat of sanctions.”
That is incentive design.
Cohen also writes in another register.
His January 2025 “Write as if the Truth Really Matters” was adapted from Citizen Scholar. There he tells academics that the “secret weapon” of scholarship is transparency and says scholars should report findings that contradict their assumptions. He urges them to write publicly, publish openly, mentor, assist journalists and activists, sign letters, and defend scholarly institutions. The opponents of knowledge, he argues, weaponize ignorance, so scholars must be present as people at the intersection of scholarship and citizenship.
Pinsof would recognize the temptation.
When Cohen explains why institutions fail, he talks incentives.
When he recruits scholars to his cause, he talks duty.
Costs in the analysis and virtue in the speech.
Cohen has debunked a famous feminist statistic. He corrected an anti-racist claim about professors. He compiled errors in a feminist best seller. He challenged a celebrated ethnographer whose political conclusions he largely shared. He campaigned against Pew’s politically ambiguous generation categories. SocArXiv withdrew scientifically deficient work whose political alignment Cohen said he did not understand. He has audited whole volumes of his discipline’s flagship journal. He has publicly corrected his own work.
The strongest formulation of “open science as a partisan enforcement arm with a general standard’s name on it” does not survive those cases.
But a weaker and more interesting Pinsofian prediction survives. Cohen has a general standard. He also has to decide where to enforce it. Those are separate questions.
The marriage-promotion archive shows how his explanatory style changes with the problem. When the question is why poor Americans do not marry, Cohen emphasizes constraints, labor markets, wages, incarceration, partner availability, and the sensible responses people make to the options they have. When the question is why marriage-promotion claims are wrong, he overwhelmingly diagnoses errors. When the question becomes why the marriage-promotion movement persists despite failure, interests return. Religion, foundations, ideology, prestige, organizational survival, and political constituencies suddenly become explanatory again.
Cohen therefore already possesses the Pinsofian model. The question is what determines the expenditure of prosecutorial energy. Why does one bad statistic receive a quick correction while another mistake produces a replication, a public-records investigation, a journal submission, a demand for retraction, an institutional critique, and years of follow-up?
Methodological seriousness can be both a scientific virtue and a status weapon.
This changes how the Pinsof test should be run on the rest of Family Inequality. Political valence remains worth coding, but it should be crossed with at least three additional variables: seriousness of the alleged error, professional proximity of the target, and level of escalation.
Then ask which variable best predicts how much effort Cohen expends. If severity predicts escalation, Cohen wins. If professional proximity predicts it after severity is controlled, Pinsof gets something interesting. If political opposition predicts it even among comparably serious errors committed by comparably proximate scholars, the selective-enforcement argument becomes strong.
The 42 marriage-promotion posts give Pinsof his first piece of evidence. Cohen’s correction record denies him an easy victory. That leaves a better question than the one we started with.
Given that Philip Cohen has a general standard of truth, who gets the full audit?
The Family Inequality Blog
I traced Family Inequality back to its first surviving post, November 1, 2009, and then read through the major runs of posts, the annual retrospectives, the controversies that generated clusters of posts, and representative work from each period through 2026. There are roughly 1,300 posts, so I would not pretend that I have now read every sentence of every entry. But the longitudinal pattern is unusually clear because Cohen repeatedly takes stock of what he is doing, and because several subjects recur for fifteen years.
Philip Cohen starts as a sociologist explaining inequality to the public. He becomes a sociologist fact-checking the public use of social science. He then becomes a critic of the institutions that produce social science. From there he becomes an institution builder himself. Finally he arrives at something close to a theory and practice of the politically engaged public intellectual. His 2025 book Citizen Scholar is the explicit statement of what the blog became.
2009-2011: the empirical egalitarian
The blog begins while Cohen is at UNC Chapel Hill. His first post, “Whose right to sex education?”, already contains almost the entire moral architecture of the later blog. Children have unequal opportunities because ostensibly private family decisions operate within unequal institutions. Public policy should not simply ratify inequalities produced inside families. Claims about social policy ought to be judged empirically rather than by traditional morality.
The first two months are remarkable because there is almost no throat-clearing. He immediately writes about abortion, same-sex parenting, Black women’s employment, caregiving, child support, health insurance, infertility, poverty, sex education and gender inequality. The blog’s title is literal. The family interests him primarily as an institution through which resources, labor, risks and opportunities are distributed.
Early Cohen is more professor than polemicist. A typical post begins with a study or news claim, explains what was measured, adds a graph or qualification, then tells you why it matters. “How do they do it?” is a good example. He uses Bourdieu and stratification research to ask how families reproduce class position, but writes as though he is talking to an intelligent undergraduate rather than another sociologist.
There is already humor, but it is subordinate to explanation. He likes incongruous titles, parenthetical asides and slightly ridiculous examples. The recurring Disney, Smurf, bathroom-sign, height and baby-name material lets him show social construction operating on objects that readers ordinarily experience as natural or trivial. The method is nearly always the same: take something everyone recognizes, measure it, then make it strange.
By 2011 another characteristic becomes important. Cohen is willing to debunk claims made by people on his own side. “Stop that feminist viral statistic meme” tracks the famous claim that women perform two-thirds of the world’s work while receiving one-tenth of its income and owning one percent of its property. He follows the citations backward until the apparently authoritative statistic dissolves into estimates, extrapolations and assumption. Then he says explicitly that exposing the falsehood will not hurt feminism.
That post is essential for understanding him. His position is partisan commitment combined with an insistence that one’s own side does not receive an evidentiary exemption. In his 2011 retrospective, three of the most popular posts attacked a feminist statistic, conservative misuse of divorce data, and gender segregation within sociology itself.
2012-2015: the emergence of Cohen as prosecutor
The decisive event is the Mark Regnerus controversy.
Cohen moved from UNC to the University of Maryland in 2012. That same year the blog became much more than a running notebook on family sociology.
His post “Time travel: Regnerus study timeline suggests superhuman abilities” is the moment when his characteristic forensic mode becomes fully visible. He does not merely say that the Regnerus same-sex-parenting study is bad. He reconstructs the chronology from study documents and journal records and notices that the paper was submitted before data collection had officially ended and before the completed data file was delivered. The title is ridicule. The body is an evidentiary brief.
This becomes a major Cohen form: verdict, exhibits, reconstruction, methodological objection, institutional question.
His 2012 top posts include the Regnerus timeline, a letter signed by 200 researchers objecting to the study and publication process, a critique of claims about Asian American incomes, criticism of marriage promotion, and an investigation of duplicate or highly similar academic publications. By 2013, Regnerus remains central alongside attacks on another same-sex-parenting study, a dubious economics paper, conservative family-policy organizations and the financing behind Regnerus’s research.
This is a significant widening of the blog’s jurisdiction. The original question was “What does social science tell us about families and inequality?” The new question is “Who is producing claims about families and inequality, who paid for them, how were they reviewed, how are journalists representing them, and can the underlying evidence survive inspection?”
That change gives the prose more voltage.
Early Cohen often sounds curious. Middle-period Cohen sounds suspicious.
The best writing from this period combines the two. He starts with a presumption that something interesting might be true, then progressively narrows the possibilities. The reader gets to watch the claim fail.
That structure makes the polemics unusually effective because the insult generally comes after the demonstration rather than instead of it.
The same tendency crosses ideological boundaries. His 2015 engagement with Alice Goffman’s On the Run is important. “On the ropes” acknowledges the importance of the book’s subject, accepts much of Goffman’s political interpretation, rejects some attacks on her, then spends considerable time asking which observations can be verified and how her survey claims were produced. It is much more interesting than a conventional takedown because Cohen keeps saying, in effect, “I agree with you about the problem. Now show me how you know this.”
By the end of 2015 the blog had become institutionally consequential. Cohen reports that it had generated material for his textbook, was being used in teaching at more than 100 colleges and had led to a contract for a book of revised blog essays. Traffic peaked around this period.
2016-2019: from critic of research to critic of the research system
The next transformation is more important than the move into national politics.
Around 2015-2016 Cohen becomes increasingly preoccupied with peer review, academic publishing, replication, open data and access to scholarship. Individual bad papers increasingly interest him as symptoms of badly designed institutions.
“For (not against) a better publishing model” is characteristic. He criticizes the American Sociological Association’s defense of its publishing revenues, but he does not stop at denunciation. He proposes a different institutional arrangement for financing open scholarly publication.
That distinction matters. Cohen’s blogging has moved from criticism to governance.
In 2016 he founded SocArXiv. His own year-end assessment is unusually revealing. He says the blog had reached 909 posts and had contributed to his textbook, forthcoming essay collection, editorship of Contexts, professional leadership roles and ultimately SocArXiv. He identifies four dominant categories that year: Trump, academic publishing, demography and debunking.
So the blog is no longer merely commentary on his academic career. It is changing the career itself.
Trump accelerates another change. Cohen’s politics had always been plainly left-liberal and feminist, but before 2016 most political arguments were routed through family sociology. After Trump, politics becomes an independent subject.
By 2017 he is writing about racism, elections and authoritarianism. He is also one of the plaintiffs suing Donald Trump for blocking critics on Twitter. His account of the federal-court argument, “Trump Twitter suit argued in federal court”, is a completely different genre from the 2009 posts. It is first-person reportage. Cohen himself has become part of the event.
His 2017 retrospective makes the transition explicit. He was writing much less on the blog, only 54 posts compared with an average of about 130 annually from 2010 through 2016, while his Twitter posts received about 25 million impressions. Enduring Bonds, meanwhile, was assembled from the best of roughly 900 blog posts.
In 2018 the mixture is almost perfectly representative of mature Cohen: demography, Regnerus, Israel, sexual harassment inside sociology, Michael Kimmel, fertility, baby names, Trump litigation and open science.
Then comes an important technological shift. In “The blog’s decade”, he explains that Twitter has taken over the quick conversational function once performed by blogging, while OSF and SocArXiv have taken over data, code and formal working papers. The blog consequently becomes a place for longer arguments, announcements and work that fits between tweets and papers.
This explains a stylistic change visible after 2016. The surviving blog posts tend to be more substantial because the little observations have migrated elsewhere.
2020-2022: the epistemology phase
COVID brings Cohen’s various interests together.
He is a demographer working with mortality data, a teacher explaining risk, an open-science advocate publishing code, and a public sociologist trying to distinguish warranted claims from rapidly circulating nonsense.
His March 2020 post “COVID-19 graphs, with data and code” begins with an important sentence: he is “not an expert on pandemic analysis.” He then explains that he understands data better by making the figures himself and releases the figures, files and code through OSF.
That is almost a manifesto for his mature method. Expertise is real, but nonexperts can reason publicly. Claims should be inspectable. Data should be shared. Corrections should be visible. Readers should be able to reproduce the work.
During COVID he publishes codebooks, continuously updated datasets, figures, lectures and analyses of health disparities. He also uses pandemic publishing to ask what peer review does and does not certify.
At the same time his institutional skepticism hardens. In 2020 he asks whether sociology sufficiently signals scientific rigor and pushes transparency and accountability. In 2021 he finally leaves the American Sociological Association after years of trying to change its publishing policies and internal organization. “Why I’m leaving the American Sociological Association” is essentially an indictment with a bill of particulars.
There is a revealing symmetry here. Around 2012 he was investigating whether individual researchers and journals had lived up to scientific standards. By 2021 he is asking whether the professional organization of sociology deserves his support at all.
The 2021 retrospective says he was increasingly writing short papers on SocArXiv rather than blog posts, while working on Citizen Scholar. In 2022 he applies the same approach to ivermectin research, generational labels and other claims that sit at the border between social science, media incentives and politics.
2023-2026: from family inequality to demographic politics
The latest phase has a surprisingly coherent center: demographic crisis narratives.
Marriage promotion returns. Fertility becomes increasingly important. So do pronatalism, population decline, generational rhetoric and the politics of expertise.
Cohen had been arguing against moralized interpretations of marriage, family structure and fertility since 2009. What changes is the political context. Falling fertility is increasingly interpreted by conservative writers, think tanks and political figures as civilizational crisis. Cohen consequently starts treating demographic rhetoric as an ideological object in its own right.
His 2023 posts on the “retreat from marriage” metaphor, marriage promotion and Black-White marriage inequality revisit arguments he has been making for more than a decade.
By 2024-2026, pronatalism has become something like the organizing controversy Regnerus was in 2012-2014. Cohen follows the arguments, organizations, media representations and data. He increasingly connects population panic to nationalism, gender conservatism, immigration politics, Elon Musk and Trump. “Population implosion collapse chaos, pronatalist billionaire discourse edition” is an obvious marker.
The 2025 “Trusting Lyman Stone, and the gathering crisis of expertise” is even more revealing. A fertility dispute becomes a question about how journalists decide who counts as an expert, how think tanks manufacture authority and what happens to independent expertise as public research funding declines.
That is almost the whole fifteen-year journey contained in one post.
The prose has changed too.
The recent political writing is considerably harsher. In the 2026 “Pronatalism is not sending their best”, phrases such as “grifting,” “hate-mongering” and “White-maxxing” arrive before the demographic analysis.
That is different from 2011 Cohen.
The continuity is the data. The difference is the sequencing.
Early and middle Cohen often makes the reader watch him establish the facts before announcing the political judgment. Recent Cohen is more likely to announce the judgment immediately and then give you the evidence.
That makes the recent prose livelier for readers who already share his political premises. It sometimes makes it less persuasive for readers who do not.
The prose
His basic sentence is conversational and declarative. He likes “Here’s the problem,” “That’s it,” “Anyway,” “So,” “I don’t know,” “You see the point.” He frequently exposes the scaffolding of his reasoning rather than presenting a polished omniscient conclusion. This gives the prose an unusual sense that thought is occurring while you read it.
The second important device is quantitative deflation. Someone makes a grand claim. Cohen produces a denominator.
A social trend becomes less dramatic after population size is considered. A sex difference becomes less impressive once the overlapping distributions are shown. A marriage advantage looks different after selection is considered. A supposedly enormous occupational trend becomes small after the occupations are weighted by their number of workers. This is probably the single most characteristic intellectual move in the blog.
The third device is what I would call comic literalism. Cohen takes metaphors and categories too seriously on purpose. If something is called a “tipping point,” what tips at 50 percent? If demographers talk about generations, where exactly are their boundaries? If population decline produces “six great-grandchildren for every hundred Koreans,” what population process does that family-tree arithmetic leave out?
The fourth is source archaeology. Rather than arguing with the latest repetition of a claim, he often tries to locate the first version. The feminist one-percent statistic is the early example. The Regnerus material adds funding and publication records. Later versions examine journal contracts, institutional policies, think-tank credentials and media descriptions of experts.
This may be his strongest contribution as a blogger. He treats citations, affiliations, timelines and funding arrangements as sociological data.
The fifth is self-correction. He often leaves corrections, updates and failed predictions visible. That matters because his prose can be very confident. The transparency prevents confidence from becoming an imitation of infallibility.
His weakness is closely related to his strength. He enjoys prosecution.
Once Cohen becomes convinced that someone intellectually or politically disreputable, his prose becomes adhesive. Every subsequent error becomes evidence about the person. Regnerus and Bradford Wilcox are obvious examples. In recent years Lyman Stone occupies some of the same territory.
Sometimes that accumulation is justified. Sometimes it can make a substantive disagreement feel like a character dossier.
There is also a detectable shift from irony to contempt. Earlier Cohen’s jokes frequently puncture claims. Later political jokes sometimes punish people. “Time travel is definitely cheating” is funny because the joke crystallizes an evidentiary anomaly. “Evil-bananas snoozefest” tells you how Cohen feels before it tells you anything about the subject. The first form is stronger prose.
His best register, in my view, appears in pieces such as the feminist-statistic investigation, the Regnerus chronology, the Goffman review and the COVID data posts. He is politically situated but epistemically restless. He tells you what he believes, then makes himself do the work anyway.
The deeper trajectory
I do not see a major ideological conversion.
The Cohen of November 2009 is already feminist, egalitarian, supportive of gay rights, suspicious of religious conservatism, hostile to punitive social policy and convinced that inequality is socially structured. Those positions remain remarkably stable.
What changes is the radius of responsibility he assigns himself.
First he is responsible for explaining his research.
Then he feels responsible for correcting public misuses of social science.
Then he feels responsible for criticizing bad research inside sociology.
Then he feels responsible for reforming scholarly communication.
Then he builds SocArXiv.
Then he sues the president over access to a public communications forum.
Then he challenges his own professional association.
Then he writes Citizen Scholar, which turns this accumulated experience into an explicit argument about what scholars ought to do in public. His current University of Maryland biography describes public scholarship and scholarly communication as central parts of his work, alongside family and demographic research.
If you want a compact reading sequence that captures the evolution without reading 1,300 posts, I would use these as the spine:
- Whose right to sex education? (2009)
- Stop that feminist viral statistic meme (2011)
- Time travel: Regnerus study timeline suggests superhuman abilities (2012)
- On the ropes, Goffman review (2015)
- For (not against) a better publishing model (2016)
- 2016 year-end review
- Trump Twitter suit argued in federal court (2018)
- The blog’s decade (2019)
- COVID-19 graphs, with data and code (2020)
- Why I’m leaving the American Sociological Association (2021)
- Trusting Lyman Stone, and the gathering crisis of expertise (2025)
- Pronatalism is not sending their best (2026)
Philip Cohen began Family Inequality in November 2009 with a question about sex education. The subject was comfortably inside his professional territory. Cohen was a family sociologist and demographer. He wanted to know how institutions distribute advantages and disadvantages among parents and children. Yet the first post already contained something that would eventually become more important than its particular subject. Cohen was not merely reporting what research had established. He was intervening in an argument about what the evidence entitled people to do.
In “Whose right to sex education?”, Cohen argued that parental freedom can conflict with children’s equality when parents are permitted to remove their children from comprehensive sex education. He cited research on abstinence-only programs and moved quickly from empirical findings to a normative question. If the programs do not work, and if withholding information imposes risks on children, why should parental authority settle the matter? The movement from evidence to citizenship is already there, although Cohen did not yet have a name for it.
Sixteen years later he did. His 2025 book Citizen Scholar argues that the roles of researcher, teacher, activist and citizen need not be sealed off from one another. Scholars can bring professional methods into public life without pretending that professional expertise eliminates political values. The book discusses open scholarship, peer review, social media, activism and public intellectual life. It draws explicitly on Cohen’s own experiences.
Read backward through Family Inequality, Citizen Scholar looks like an attempt to theorize what happened to Cohen while he was blogging.
The Cohen of 2009 was already recognizably the Cohen of 2026. He was a feminist, an egalitarian, supportive of gay rights, hostile to racial inequality, skeptical of conservative family policy and comfortable drawing normative conclusions from social research.
What changed was his conception of the scholar’s jurisdiction.
He began by explaining society. He moved into checking claims about society. Then he began checking the scholars who made those claims, the journals that published them, the organizations that financed them and the media institutions that transmitted them. Eventually he began building alternative institutions for producing and circulating knowledge. Along the way he became a litigant defending access to a political forum and finally the author of a book arguing that public engagement belongs within the vocation of scholarship itself.
The blog changed the unit of analysis. It began with inequality. It ended up investigating authority.
One of the earliest signs of the transition came from Cohen’s own political side.
In 2011 he published “Stop that feminist viral statistic meme”. The target was the enormously successful assertion that women perform two-thirds of the world’s work while receiving one-tenth of its income and owning less than one percent of its property. Cohen identified himself as a feminist and then went after the statistic anyway.
The method is worth noticing. He did not begin with a rival ideological claim. He followed citations.
The statistic had acquired the appearance of established knowledge because respectable people kept repeating it. Feminist activists repeated it. Scholars repeated it. International organizations repeated it. Each repetition increased its apparent authority. Cohen started pulling on the thread. Where exactly did the numbers originate? What had been measured? Could the claim survive even simple arithmetic? He observed that American women alone earned enough income to make one part of the global claim implausible. The statistic disintegrated. Cohen concluded that exposing it would not hurt feminism.
This is a more important post than it first appears.
Cohen had discovered a genre.
A claim enters public circulation bearing the credentials of science. Someone cites someone who cites someone. Prestigious institutions launder uncertainty into fact. Journalists understandably assume that a statistic repeated by academics and international organizations has been checked. Political actors repeat it because it serves their purposes. Eventually nobody knows where it came from.
The sociologist can study that process too.
This requires a different kind of professional behavior. The scientist is no longer merely the person who produces a regression table and sends it to a journal. He becomes an auditor of the entire chain by which knowledge acquires authority.
That role became central to Family Inequality the following year.
In June 2012 Cohen encountered Mark Regnerus’s study of children whose parents had had same-sex relationships. The research appeared at a politically explosive moment. Opponents of same-sex marriage could cite it as evidence that children fared worse outside heterosexual married households. Cohen thought the study was badly designed. But what turned the episode into a prolonged investigation was not merely disagreement about methods.
He noticed the dates.
In “Time travel: Regnerus study timeline suggests superhuman abilities”, Cohen reconstructed the chronology printed in the study documents and journal record. The paper had been received by Social Science Research on February 1, 2012. Data collection was listed as continuing until February 21. The completed data file was delivered to the University of Texas on February 24. A revised paper reached the journal on February 29 and the article was accepted on March 12. Cohen’s joke was simple: speed is allowed, but time travel is cheating.
That sentence captures his best polemical technique. The joke works because the chronology has already made the joke unnecessary.
Cohen kept digging. Funding, consulting relationships, peer review, correspondence, publication decisions, political timing and Supreme Court briefs all became relevant. His later Regnerus timeline reconstructed the path from the Witherspoon Institute’s involvement in 2010 through the study’s publication and use in litigation. The research paper ceased to be an isolated text. It became one component in a social system connecting donors, academics, journals, advocates, journalists and courts.
This is where Cohen’s sociology folded back onto sociology itself.
A conventional methodological criticism asks whether a study measured what its authors claimed to measure. Cohen increasingly wanted to know who commissioned it, who reviewed it, how quickly the journal processed it, who promoted it and where it traveled after publication.
The sociology of the family was becoming a sociology of knowledge.
There is a temptation to interpret this simply as politicization. That misses what is distinctive about Cohen. He became more politically explicit while simultaneously becoming more demanding about scientific procedure.
The feminist-statistic post is crucial evidence. Cohen’s rule was not that conservative social science deserved scrutiny while progressive social science deserved solidarity. A politically useful falsehood was still a falsehood. Indeed, political usefulness created another reason to check it because useful claims spread farther.
This principle gradually widened the range of institutions that could become objects of criticism.
Academic journals were no longer neutral containers into which scholars deposited knowledge. Publication systems created incentives. Peer review conferred authority. Paywalls restricted who could inspect the evidence. Journal prestige depended partly on scarcity and rejection. Professional associations benefited financially from arrangements that Cohen increasingly regarded as contrary to the public purposes of scholarship.
By 2016 the complaint had become a constructive project. Cohen helped establish SocArXiv, a free open archive for social science research. He described the existing publishing system as slow, hierarchical, expensive and unnecessarily closed. The stated goal of SocArXiv was not merely cheaper distribution. Open access would make social science faster, more accountable and more transparent. Cohen has remained its founding director since 2016.
The sequence matters.
First Cohen criticized papers.
Then he criticized the machinery that certified papers.
Then he built machinery.
His own 2016 account of the blog makes the connection unusually explicit. After 909 posts, he credited Family Inequality with helping produce his textbook, an essay collection, his co-editorship of Contexts, leadership positions in the American Sociological Association and finally SocArXiv. The blog was no longer publicity appended to an academic career. It was producing the career.
This is one reason Family Inequality is an interesting document in the history of blogging. Blogs were often described as mechanisms by which academics could communicate finished research to outsiders. Cohen’s experience ran in the opposite direction too. Public communication altered what he researched, whom he encountered, which institutional problems he noticed and what he thought a scholar should do about them.
The medium rewarded a particular intellectual temperament.
Cohen likes anomalies.
Much of his best writing begins when a large public claim collides with a small inconvenient fact. Women own one percent of property. Really? A paper went through normal scientific review. Then why was it submitted before data collection ended? Millennials are fundamentally different from previous generations. Where exactly does one generation stop and another begin?
The prose often proceeds by quantitative deflation. Someone presents a dramatic transformation. Cohen asks for a denominator, comparison group, baseline or time series. Once those appear, the transformation frequently becomes less dramatic.
His long campaign against named generations works this way. Cohen argues that labels such as Millennials encourage stereotyping, character judgment and broad claims unsupported by the continuous patterns visible in demographic data. The objection is methodological, but it is also stylistic. He distrusts categories that make the world sound more discontinuous and exciting than the evidence warrants.
This produces one of the appealing tensions in Cohen’s prose. The writer enjoys drama. The demographer distrusts it.
His titles tease and exaggerate. The body of the post then often removes exaggeration from somebody else’s argument.
At his best, Cohen practices a kind of comic literalism. He takes a metaphor seriously enough to expose it. If something has reached a “tipping point,” what exactly tipped? If one sex now “dominates” employment, what measure produces the result? If a generation possesses a distinct personality, how can we locate its edge? If a research chronology requires the paper to precede the data, perhaps the researcher has mastered time travel.
The humor is strongest when it compresses the empirical objection.
It becomes weaker when it merely labels the opposing side ridiculous.
That change becomes more visible in the later blog. Earlier Cohen often allows readers to discover the absurdity with him. Some recent political writing announces contempt earlier. The 2012 Regnerus title is savage, but its savagery rests on a reproducible timeline. In later polemics, political classification can precede the demonstration. Readers who already agree with Cohen will experience this as welcome clarity. Readers who do not may feel that the verdict arrived before the evidence.
The change partly reflects what happened to the blog.
By 2019 Cohen was publishing far fewer posts. He reported 268,000 visits that year, down 37 percent from the blog’s 2015 peak, and only 39 new posts. Meanwhile his Twitter activity had generated about 25 million impressions. Twitter had taken over much of the quick conversation and provisional thinking once performed by the blog. The Open Science Framework housed data and code. SocArXiv housed papers and preprints. The blog remained for longer arguments, announcements and work that fit between social media and formal publication.
Cohen called this a communication strategy. It was really an emerging intellectual ecosystem.
Research paper, dataset, code repository, blog post and social-media argument no longer represented higher and lower versions of the same activity. They performed different functions. Public scholarship was not what happened after the research was complete. It included the entire process by which questions, evidence, criticism and revisions circulated.
COVID-19 made the model visible.
In March 2020 Cohen opened one analysis by saying that he was not an expert in pandemic analysis. Then he explained why he was working with the data anyway. He understood data better when he made the figures himself. He posted graphs, data files and code through the Open Science Framework under a license allowing others to use them.
There is a modest but consequential conception of expertise here.
Cohen does not argue that everybody is equally qualified. He announces the boundary of his expertise. But the boundary does not require silence. It requires transparency. Here is what I know. Here is what I do not know. Here are my data. Here is my code. Check me.
That formulation explains why open science became inseparable from his politics.
Cohen’s answer to distrust is not “trust experts.”
It is closer to “make expertise more trustworthy.”
That distinction also explains his increasingly antagonistic relationship with the American Sociological Association. Cohen had served sociology’s major professional organization in various capacities. By 2021 he announced that he would no longer renew his membership. His complaints concerned cost, organizational efficiency, publishing practices and what he regarded as failures to embrace open science. He emphasized that he had tried for years to change the organization from inside before leaving.
The critic of conservative research had become a critic of his own profession’s institutions.
Cohen repeatedly responds to disappointing institutions by trying to create better institutions. He wants peer review. He wants scholarly expertise. He wants professional communities. He simply does not treat their existing arrangements as sacred.
The same expansion of professional responsibility eventually carried him beyond academia.
When Donald Trump blocked Cohen and several other critics from his Twitter account, Cohen became one of seven individual plaintiffs in the Knight First Amendment Institute’s lawsuit against the president. A federal district court held in 2018 that Trump’s viewpoint-based blocking violated the First Amendment. A unanimous Second Circuit panel affirmed that conclusion in 2019. The Supreme Court later vacated the judgment as moot after Trump left office.
There is something almost too neat about this episode in Cohen’s trajectory.
The sociologist who began by blogging because the internet provided a public space for scholarship ended up in federal court arguing, as a citizen, about who had the right to participate in a public space on the internet.
The boundary between communication and citizenship had disappeared.
By the time Cohen wrote Citizen Scholar, he had accumulated a series of practical answers to the question the book asks. What is the role of professional scholars in civic life? The chapters move through description, open scholarship, peer review, social media and activism. The publisher describes the book as an argument for a reciprocal relationship between professional scholarship and active citizenship. Cohen’s examples include his own social-media work and the Trump lawsuit.
The important word is reciprocal.
The conventional image of public scholarship is one-directional. The professor possesses knowledge. The professor emerges from the university and gives some of it to the public. Engagement is dissemination.
Cohen’s career suggests a more disruptive model.
The public talks back.
Blog readers identify mistakes. Political actors misuse research. Journalists expose questions scholars have ignored. Critics force clarification. Bad studies reveal flaws in peer review. Paywalls become visible when someone outside the university wants to read an article. Twitter changes who can challenge whom. Political events generate new datasets. Litigation turns an academic’s theory of the public sphere into a concrete fight about access.
Public engagement therefore changes the scholar.
That is the real story of Family Inequality.
Cohen started the blog as a sociologist with things to say about families, gender and inequality. Blogging exposed him to the entire apparatus through which social facts become public facts. He began following claims backward toward their sources and forward toward their consequences. The trail led through datasets, citations, journals, donors, think tanks, newsrooms, professional associations, social-media platforms, political movements and courts.
Once he started following that trail, there was no obvious point at which professional responsibility ended.
The result is a conception of sociology that is considerably more demanding than “public sociology” understood as popularization. The citizen scholar does not merely explain sociology in accessible language. He inspects the conditions under which sociology deserves authority.
That includes inspecting his allies.
It includes showing the work.
It includes correcting mistakes.
It includes asking who paid.
It includes making data available.
It includes examining the institutions that certify knowledge.
Sometimes it includes building replacements for institutions that fail.
And sometimes it includes admitting that the scholar has left description behind and entered politics.
There remains a tension Cohen never completely solves.
The more one enters political combat, the harder it becomes to retain the peculiar authority produced by visible willingness to discover that one’s own side is wrong. Cohen’s 2011 demolition of the feminist one-percent statistic is powerful because the reader can see that his political allegiance does not determine his empirical conclusion. Some of his later writing risks reversing the order, identifying political enemies first and auditing their claims second.
That is the occupational hazard of the vocation he constructed.
A citizen scholar must possess convictions strong enough to justify entering public disputes and epistemic discipline strong enough to permit inconvenient conclusions once he gets there. Too little commitment produces irrelevance. Too much produces propaganda.
Cohen’s blog is valuable partly because it contains the experiment in public, including the strain.
What began in 2009 as a sociologist writing about family inequality became, post by post, an inquiry into something more fundamental: who gets to say what is true in public, what procedures make those claims credible, and what obligations follow when a scholar believes that the machinery producing public knowledge has failed.
The title never changed.
The subject did.
Family Inequality became a record of Philip Cohen discovering that the social organization of knowledge was itself an inequality problem, an institutional problem and finally a citizenship problem. Citizen Scholar supplied the theory at the end. The blog had spent fifteen years supplying the evidence.
Explaining the Normative
In the winter of 2025 a submission arrived at SocArXiv that satisfied every stated requirement. Abstract present. Title matching the contents. Real citations to real work. Complex statistical models, apparently competent, with tables and figures. Code in a public repository. And a disclosure listing every tool used to write the code, run the literature search, advise on the models, and draft the text.
Philip Cohen looked at it and knew it should be rejected. Then he sat down to write out why, and could not.
He published the attempt on February 22, 2026. He is not persuaded that using such tools is unethical, because all knowledge is learned from someone else. He thinks the training may be lawful. He grants that an archive issuing identifiers without review is an attractive deposit site. He will not enter a detection arms race. A disclosure requirement he cannot verify is close to worthless, since the machines read the rules too. Every route to a statable rule closes, one after another, in public, over about two thousand words.
The policy his committee published on March 9, 2026 stopped trying. It refuses automated detection, declines to state a rule that could be written down completely, commits the moderators to judgment made by people using perception and experience, invites appeals, and contains this sentence: we feel the need to express our humanity in this process.
Stephen P. Turner (b. 1951) has spent forty years arguing that this is what always happens.
His position, most fully stated in Explaining the Normative (Polity, 2010), runs against the entire family of accounts that treat norms as things with independent standing. The pattern he attacks is what he calls normativism: the move that begins with an ordinary fact about people doing something together, notices that the fact alone will not explain why the thing is binding, and then posits a normative element to supply the bindingness. Rules, conventions, shared understandings, collective intentions, presuppositions. Turner’s objection is that these are explanatory additions with no independent access to them. The normative element never shows up in experience. It is inferred from the very behavior it was introduced to explain, and it does no work that the ordinary causal facts about habit, training, and social experience were not already doing. His term for the operation is that normativists make explanatory tea from the twigs of the ordinary. Whatever you can say about how people came to act as they do, you can say in ordinary terms about how people learn things and become able to do things together.
He is saying that the binding is a fact about trained persons, acquired through experience, largely tacit.
Set that against Cohen’s program.
The transparency campaign is the attempt to convert the tacit competence of a discipline into stated rules that anyone can check. Publish the code. Post the data. Say what you did. The premise is that the norms of good social science can be written down and then enforced against the writing. His December 2025 audit is what that looks like in operation: he opened Volume 90 of American Sociological Review, went through it article by article, clicked the links, and reported that twelve of eighteen quantitative papers arrived with packages and six did not.
Turner’s question is what the count measures.
It measures compliance with a stated rule, and the stated rule was extracted from a prior competence about what makes a claim inspectable. The competence is the real thing. The rule is the twig.
You can watch the extraction fail at both ends of his career.
At the front end, the Goffman dispute, and the concession he made seven years after it. In February 2022, at the close of a long online workshop, a friendly questioner asked how open science applies to qualitative work, and Cohen said this: just because you read my transcripts doesn’t mean you can analyze my data, the data is me, I was there, it’s the experience, there’s no such thing as reproducibility because you couldn’t have the interaction I had as me at that moment. That is a statement that the relevant competence cannot be written down. Turner’s whole account of tacit knowledge is that what an expert knows is not a set of propositions the expert has failed to articulate, but a trained capacity that produces performances a stranger can evaluate only by having comparable training. Cohen grants it, and then rescues his program by substituting accountability for reproducibility. You still might have lied, or been wrong.
But accountability to whom, and judged how? Turner’s answer is: to trained members, on grounds they cannot fully state. Which is where the archive ended up four years later.
At the back end, the machine. The submission was the perfect test case because it isolated the variable. Every stateable rule satisfied, and the thing everyone in the field could see was missing had never been in the rules. Cohen looked for the norm behind the practice, found nothing there, and his committee wrote a policy that says in effect: competent people will recognize this, and we cannot tell you how.
That is a confirmation of Turner, delivered by the person with the strongest possible incentive to produce a rule instead.
Now the sharper application, which is about the word he keeps reaching for.
Cohen’s public case is normative. Journals should share data. The association should not sign contracts that transfer money from teaching-heavy libraries to research-heavy ones. Researchers owe the public inspectability because the public funds the work. Scholars should disclose where they stand. And when he leaves the association, in November 2021, the language shifts from should to a claim about what the institution has become: captured by its publisher, interested in change only for public relations.
Turner’s question about any such claim is what makes the obligation binding, and his answer is that nothing does, in the sense the claim requires. What exists is a group of people trained in similar ways, with similar habits, who find certain conduct intolerable and others’ conduct acceptable. When Cohen says the association is failing its members, the fact underneath is that a subset of sociologists formed in a particular way at a particular time find the arrangement intolerable, and another subset does not. Neither group is failing to perceive a norm that stands over both. There is no such thing to perceive.
This explains the pattern in the record better than his own account does.
The transparency counts moved the journals, because the people counted and the people counting share a formation. An author he marked as providing nothing appeared in his comments, agreed, promised a package, and thanked him. That is a trained member recognizing a demand as legitimate, not a man persuaded by an argument he had not previously encountered.
The Wilcox fight has moved nothing across a decade, because the formations differ. Cohen says Wilcox lies. Turner might say that neither man can supply the other with the tacit background that makes his own inferences feel compulsory, and that in the absence of that background, each reads the other’s persistence as bad faith. The accusation of lying is what normativism produces when the norm fails to bind: if the rule is real and public and he still will not follow it, he must be refusing.
The Pew campaign is the clean case in the other direction. In July 2019 he told a National Academies committee that generation labels drive stereotyping. Asked whether any research demonstrates it, he said he knows of none, that most sociologists do not use the labels as real categories in their research, and that it would be an interesting experiment for somebody to run. That is his own founding claim conceded as an intuition. Two years later a letter with roughly a hundred and fifty signatures moved Pew where the argument had not. What did the work was not evidence about harm. It was a professional group asserting that people trained as they were trained find the practice unacceptable, delivered to an organization whose standing depends on that group’s regard.
Turner’s model predicts all three outcomes. Cohen’s model predicts one and treats the other two as failures of good faith.
Two things must be said against the frame.
Turner’s account has a hard time with arithmetic, and Cohen’s best work is arithmetic. The 217 homes and 308 young men in Goffman’s survey against census figures for comparable Philadelphia neighborhoods is not a trained perception. It is a count, and a hostile reader with no sociological formation whatever can redo it and get the same number. Turner’s reply would be that agreeing on what constitutes a household, a comparable neighborhood, and a relevant discrepancy is precisely trained agreement, and that the wide sharing of the training makes it feel like no training at all. That reply is available and it is also the kind of reply that absorbs every counterexample. A frame that cannot lose is doing something other than explaining, and Turner is more vulnerable here than his admirers usually admit.
Second, Turner has been accused of proving too much: if norms do no explanatory work, it becomes unclear why anyone should do anything, and the position slides toward a description of what happens with no purchase on what ought to. Turner denies the slide, arguing that ordinary explanation of how people come to act is not a licence for indifference.
Cohen was formed, in a house where institutional injustice was a household fact and in a newsroom where a story existed to be handed to strangers, into a person who finds unchecked claims intolerable. He then built instruments that made his intolerance operable by others formed similarly: counts anyone in the group can redo, an archive anyone in the group can use. The instruments worked where the formation was shared and failed where it was not, and in 2026 they met a submission produced by something with no formation at all, at which point the only available move was to say that the people in the room would decide and could not say how.
His most durable contribution, on this reading, is not a rule about data sharing. It is that he took a tacit standard held by part of a profession and gave it a countable form, which is the one way a tacit standard can travel to people who do not yet hold it.
He would say the countability is the point and the formation is beside it. Turner would say the formation is the whole of it and the countability is how it spread.
The moderation queue in the spring of 2026, with five named volunteers reading papers by hand and stating that they cannot write the rule down, is the strongest evidence available for the second reading. And Cohen published it himself.
Notes
Frame. Stephen P. Turner, Explaining the Normative (Polity, 2010), which is the source for the critique of normativism, the argument that normative elements are explanatory additions with no independent epistemic access, and the tea-from-twigs formulation. Behind it sits The Social Theory of Practices: Tradition, Tacit Knowledge, and Presuppositions (Polity, 1994), which is where the argument against shared tacit frameworks is made in full, and Understanding the Tacit (Routledge, 2014).
Frame discipline. Turner only. I kept out Polanyi, whose Republic of Science is the obvious neighbor and whose tacit knowledge is a different and more sanguine doctrine than Turner’s.
The claims that are mine. Three. First, that the March 2026 policy is a confirmation of Turner’s thesis produced by the person least motivated to produce it, and that its evidentiary value lies precisely in that. Second, that Cohen’s accusation of lying against Wilcox is what normativism generates when a supposed norm fails to bind across formations, rather than a finding about Wilcox. Third, that Cohen’s achievement is better described as giving a tacit standard a countable form.
The Luc Boltanski Frame on Philip Cohen
In Paris, at the start of the 1980s, a sociologist sat with a box of letters that a newspaper had received and not printed. Luc Boltanski (b. January 4, 1940) had 275 of them, sent to Le Monde between 1979 and 1981 by people who wanted an injustice exposed. Working with Yann Darré and Marie-Ange Schiltz, he coded everything he could think of. The characteristics of the affair. How long it had been going on. Whether the letter came typed or on letterhead. Whether the writer had underlined heavily. Whether he had used insults.
Then he did the two things that made the study. He convened a jury of ordinary readers and asked each of them to rate every file for normality after one quick pass. And he defined the positions any denunciation has to fill: a victim, a denouncer, a persecutor, and a judge. Each position got coded for size, on an axis running from a single named individual to a collective.
What came back from the jury was a ranking by configuration. A letter in which a man speaks for himself about a persecutor who wronged him alone reads as private business, or as illness. The same complaint, with the victim enlarged into a category and the persecutor turned from a man into an office, reads as a cause. Boltanski called the work required to move from the first to the second desingularization, and the rise to generality. His paper's central sentence, in the standard English rendering, is that a normal denunciation is marked by the otherness of the denouncer and the victim, and that carrying one out requires the standing to speak up for someone else, to come symbolically to his aid.
Normality here is a judgment made by an audience. It runs loose from truth. A man can be right and be filed.
Boltanski built this instrument at the moment he was breaking with Pierre Bourdieu (1930–2002) over what a sociologist does. Bourdieu's sociologist unmasks. Boltanski's studies the unmasking, treating ordinary people as competent accusers whose grammar can be described. He got a theory of denunciation by declining to denounce. Anyone who picks up the instrument should keep that in view, because the model rates admissibility and says nothing about who deserves condemnation. Used as a verdict, it becomes the thing it describes.
Philip Cohen has spent twenty years filling the denouncer position in a discipline that has no established procedure for it. His work sorts along Boltanski's axis, and the sort explains something that complaints about his tone cannot.
Start with the case that works.
In December 2025, having been asked to meet a National Academies panel on corrections and retractions, Cohen opened Volume 90 of American Sociological Review and went through it article by article, clicking the links to see whether the code and data were there. Eighteen quantitative papers. Twelve arrived with replication packages or directions to the data. Six arrived with nothing. He counted the qualitative and mixed papers too, and he printed his earlier tallies for comparison: four of fifteen in 2020, eight of twenty-five in 2023. He added that the association's 2018 guide for editors does not contain the word retraction, and that a single article has been pulled from an association journal this century, in a documented fraud case involving a Dutch psychologist.
Now read the positions. The victim is anyone who wants to check a result and cannot, which is a category with no face. The persecutor is a procedure, a journal that permits the phrase available upon request. The judge is any reader holding the same volume, and the judge can overturn the denouncer, which Cohen invites when he calls his own methods squishy and asks that corrections be mailed to his office written on the back of a fifty-dollar bill. He appears in the file as a man who counted, and who can be recounted.
Boltanski's jury would have rated it high, and the discipline did. An author whose paper Cohen had marked as providing nothing arrived in the comment thread, agreed he was right, promised a package once teaching quieted down, and thanked him for what he does for the field. The share of quantitative papers shipping code has roughly tripled across the three audits.
The Pew campaign has the same architecture built by a different route. There Cohen enlarged the denouncer: an open letter in May 2021 carrying about a hundred and fifty signatures from demographers and social scientists. The victim is a category, everyone sorted by arbitrary birth years and described by traits assigned to the sort. The persecutor is a practice at a research organization, never a person. Kim Parker answered in July in the organization's voice, granting the risk of age stereotyping. In May 2023 the organization published a new policy: generational analysis only where the data support comparing people at the same age in different periods, the labels used sparingly.
Nobody was accused of anything. The configuration did the work.
The most exacting case in the file is the one where the victim is his mother.
Avis Cohen ran a lab at Cornell University for ten years on her own federal grants and got no tenure-track offer. Maryland hired her as a full professor in 1990. Her son published his account of this nine days after her death in June 2026, and every operation Boltanski describes is visible in the paragraph. He concedes the defensible version first: passing on her right after the postdoc may have been reasonable. He specifies the indefensible one: with the grants and the publications she held by 1990, it was not. He names no chairman and no colleague. The persecutor is a department that had no offer for her. And then he desingularizes the victim, setting Suzanne Bianchi beside her and naming the category they share, faculty women who were never assistant professors.
A grieving son writing about his own mother sits in the corner of the diagram Boltanski's jury filed as pathology, and Cohen gets out of that corner in four sentences. He knows how the filter works. He passed his mother through it.
The Goffman affair is where the configuration starts to move.
Cohen went after the one element of On the Run a stranger could test: a household survey reporting 217 homes on the block, 308 men aged eighteen to thirty in residence, and 144 of them carrying warrants. Census figures for comparable Philadelphia neighborhoods gave less than half that many young men. The residence rule, three days a week, could put the same man in two homes. He declined to say the ethnography was fabricated, which kept the persecutor from collapsing into a single young scholar: the file also indicted the referees who let the survey through, the committees that gave the book its prizes, and the professional habit of treating a request for the evidence as a discourtesy.
Then American Sociological Review rejected his comment. The editors agreed he had found errors and chose not to correct the record.
At that moment Cohen became the victim of his own affair, which is the position from which no denunciation travels. Watch the repair. He published the rejection. He put the comment on his university web page and later a version of it in Social Forces. And ten years on, in the audit, his rejected comments appear as evidence about journals: two comments submitted, errors conceded by editors both times, no correction either time, one case described and the other withheld. The injury got converted into a data point in a series. Boltanski would recognize the maneuver as the only legitimate route out of that corner. A denouncer who is also the victim can be heard if he makes his own injury an instance of a general condition.
The Wilcox posts run the other way.
The empirical dispute has a high-normality victim: people with no money being instructed that finishing school, working, marrying and then having children will keep them out of poverty, when the schooling, the work, the health care and the available partners are the things they lack. That victim is a category, and the argument about selection into marriage is checkable by anyone who understands what a comparison group has to do.
Then there is the title asserting that Brad Wilcox lies, and the tag on the end of it, part whatever.
Boltanski's team coded for insults. They also coded for duration, the accumulating file, the correspondent who writes again. Both markers are in that title, one announcing the insult and the other announcing the series. And the persecutor has contracted to a single named man at the moment the reader needed him to stay general. The consequence is measurable in the reading: a person who wants the selection argument has to walk past a character verdict to reach it. The selection argument invites him to check. The verdict asks him to trust.
The Lyman Stone post from July 2026 is the strongest file in the drawer on evidence and the weakest on form.
The victims are general and plural: Somalis, and everyone outside the denouncing party's religion. The evidence is sourced and linked, which is more than most such accusations offer, and it consists of Stone's own posts. Stone's defense is in the file too, that his objection concerns clan social organization.
Two moves then take the configuration apart. Cohen delivers the verdict, and he instructs the judge to act against the man: he would not hire Stone or publish his work, and neither should you. The judge position folds into the denouncer position, and the audience Boltanski's model reserves for weighing is handed a sentence to execute. Then Stone returns the same word, calling Cohen a liar, and Cohen answers in a footnote that he will not spell it out and that readers should ask Stone for receipts.
Two men trading the accusation of lying is the configuration ordinary readers file as a quarrel, whatever the record shows. None of this bears on whether Cohen has read Stone correctly. The jury in Paris was not rating truth.
The association posts are the mixed case.
Cohen had served three years on the Committee on Publications and failed to get a policy adopted that would have told readers whether a published article came with its data. He says the staff invented a rule that would have barred him from taking part in the discussion of his own proposal. That grievance is as singular as a grievance gets, one man wronged by named colleagues in a committee room.
His resignation post in November 2021 puts the money first. Dues and conference costs that the association's own membership survey identified as the largest problem. Subscription prices more than doubled in eight years. A seven-year contract signed with a commercial publisher in 2018 without notice to members. Library revenue that transfers money from teaching-heavy institutions to research-heavy ones, since publication in the association's journals is dominated by elite universities. The personal insult about the invented rule appears well down the page, after the victim has been established as somebody else.
And then the model's signature: the victim showed up. A community college instructor wrote in the comment thread that the subscription fees had always been prohibitive for faculty like her, that she had stayed out of commitment to the organization, and that she was now cancelling. A denunciation that has completed its rise to generality recruits the people it claimed to speak for.
The best reply in that thread also turns on configuration. A junior scholar agreed with the whole analysis and asked where new people would build networks if established people left. The question resingularizes Cohen. A tenured full professor at a research university with a large audience bears none of the cost of exit. He speaks for the community college instructor and he does not stand where she stands. Boltanski's requirement of standing to come to another man's aid cuts in both directions, and the junior scholar found the edge that cuts back.
Something in the 2026 material suggests Cohen understands all of this.
His archive spent late 2025 and early 2026 drowning in machine-written submissions, and had to start refusing papers it would once have taken. Read the March 2026 policy announcement next to the July post about Stone. The policy comes in the first person plural. It names the five people on the subcommittee. It states the competing interests it is trying to balance, the scholars who want a reliable literature, the volunteers facing the queue each day, the independent researchers who need guidance. It declines to promise a fair standard that could be written down completely, says the archive will make human judgments and will not defer to detection software, and invites appeals. The line everyone quotes from it is that the team felt a need to express its humanity in the process.
That is a man writing from the judge's chair, and writing carefully, four months before he wrote a hiring recommendation from the denouncer's.
A complaint about tone locates the trouble inside a man, where nothing can be done about it, and where the complaint doubles as an excuse for ignoring arithmetic that happens to be correct. Boltanski locates it in a system of places, and places can be occupied differently. Cohen's consequential work stands where the injured party is somebody else and the accused is a procedure. His inconsequential work stands where he is the injured party and the accused has a name and a face.
Two things follow.
The filter has a cost, and the cost falls on the people it looks most reasonable to protect. Requiring desingularization means that a man wronged by an institution, alone, with nobody to speak for him and no series to attach himself to, cannot make himself heard until he finds a category to stand in front of. Institutions benefit from that requirement. The editors who conceded Cohen's error and printed no correction were protected by his singularity for ten years, until he could file his injury as one of a pair inside a longer count. Most people never get the second instance.
And arithmetic is the cheapest desingularization available. A number anyone can recount arrives without an author. The reason Cohen's audits move journals while his verdicts move nobody is that a count of replication packages in Volume 90 has no interest in the outcome, no history with the accused, and no standing to establish. It does not need to come symbolically to anyone's aid. It just needs to be right, and to be checkable by a stranger with the same volume open on his desk.
Notes
Luc Boltanski, with Yann Darré and Marie-Ange Schiltz, “La dénonciation,” Actes de la recherche en sciences sociales 51 (March 1984), 3–40. Full text free at Persée: https://www.persee.fr/doc/arss_0335-5322_1984_num_51_1_2212. Corpus confirmed as 275 letters sent to Le Monde, 1979–1981 (summary; a pedantic correction of a common overstatement of the corpus size is at https://thesociologicalpages.org — see https://thesocietypages.org/theory/2021/01/06/shai-dromi-reads-luc-boltanski/, which notes the letters went to something like the paper’s investigative desk). The English-language development is Love and Justice as Competences (Polity, 2012), whose opening chapter is the source for the coding scheme (typewriter or letterhead, underlining, insults), the jury of ordinary readers rating each file for normality after a rapid first reading, the four actants, and the coding of each actant for size on a singular-general axis. The key sentence I paraphrase about the otherness of denouncer and victim is Boltanski et al. 1984, p. 32; the English rendering I worked from appears in a 2025 Open Book Publishers volume applying the model to denunciations in the arts. On normality as a function of how a political order relates private interest to the common good, see the discussion of Love and Justice as Competences pp. 193–94 in https://academic.oup.com/ips/article/16/1/olab024/6408727. On the Bourdieu break and the pragmatic sociology of critique: https://en.wikipedia.org/wiki/Luc_Boltanski and Boltanski, “Sociologie critique et sociologie de la critique,” Politix 10–11 (1990).
There was a fortieth-anniversary colloquium in Paris in March 2024 with Boltanski present, and a resulting volume, Théories et pratiques de la dénonciation publique (Le Bord de l’Eau, 2026): https://www.editionsbdl.com/produit/theories-et-pratiques-de-la-denonciation-publique/ and https://carism.assas-universite.fr/fr/colloque-scenes-denonciation-publique-medias-langages-societes.
I applied only the 1984 model and its direct extension in Love and Justice as Competences. I kept Distant Suffering and On Justification out, though both are tempting: the topic of denunciation in the former and the orders of worth in the latter would each pull the essay into a second frame. The single line I let in from outside is the historical note that Boltanski developed the model while breaking with Bourdieu, which sets up the instrument.
Two claims I want to flag as mine. First, that arithmetic is the cheapest desingularization available, because a recountable number arrives without an author. Boltanski does not say this; the singular-general axis in his paper runs over persons and collectives, not over kinds of evidence. It follows from his framework and it is the closest thing in this essay to an original contribution, so it should be defended as an extension. Second, that the requirement to desingularize systematically advantages institutions over the individuals they have injured. Boltanski implies something adjacent when he says the constraints of normality depend on how a political order relates private interest to the common good, and he is describing.
The essay reaches an unflattering conclusion about Cohen’s verdicts and a flattering one about his counts, and it reaches both by the same procedure, which is the only defense against the charge that the frame was chosen for the outcome. The Stone passage states his evidence, states Stone’s defense, and declines to say who is right, which keeps me out of a fight neither man needs me in. The guardrail paragraph is early on, that the model rates admissibility and not merit. Without it a reader can take the essay as a claim that Cohen’s accusations are false, which it does not make and which the frame cannot support.
The Audit Society: Rituals of Verification
Michael Power trained in the philosophy of the social sciences, then qualified as a chartered accountant and did audit work for Deloitte Haskins and Sells. He arrived at the London School of Economics as a lecturer in accounting and finance in 1987, which put him in Britain at the moment the word audit slipped its moorings. Medical audit. Environmental audit. Teaching audit. Technology audit. Quality audit. Value-for-money audit. A term that had meant a licensed professional checking a company’s books against evidence was suddenly available for any activity that somebody wanted held to account.
He wrote a pamphlet about it for Demos in 1994 called The Audit Explosion, and three years later a book of 183 pages, The Audit Society: Rituals of Verification, from Oxford. The argument has six moves, and the chapter titles carry them: the rise of audit, the audit explosion, audit and the dialectic of regulatory failure, audit knowledge and the construction of auditees, and then the last chapter, “Beyond Audit, Towards Trust.”
Audit expands where confidence in professional self-regulation has failed, which means audit demand is a symptom of withdrawn trust. Audit does not restore the trust whose absence created it; it relocates trust onto the audit process and the auditor, at a further remove and with less visibility. What audits check, once they scale, is the system of control, which Power calls control of control. Activities do not arrive auditable; they have to be made auditable, and the making reorganizes them around what can be documented. And the audited respond in one of two ways, either by producing compliance for the inspection while leaving the work untouched, or by internalizing the audit’s categories until the activity becomes the thing the audit can see.
Power has been careful about being read as a debunker. In a 2021 paper on the logic of the audit trail he wrote that accounting and auditing are “not inherently value-subverting and may be value-enhancing.” His complaint has always been narrower and harder to answer: the audit explosion has never been subjected to the scrutiny it applies to everything else.
Philip Cohen calls his own procedure an audit. In December 2025, reporting that two-thirds of the quantitative papers in the most recent volume of American Sociological Review came with replication packages, he described this as better than the results of his previous informal audits.
He counts replication packages by volume. He counts how many articles have been retracted from the association’s journals this century, and arrives at one. He counts the birth years behind generational labels and demonstrates that the boundaries float. He counts the association’s subscription price increases across eight years and reads its revenue structure out of its own audited accounts. He counted the households and young men in Alice Goffman’s neighborhood survey against census figures for comparable Philadelphia blocks. He counts his archive’s submissions in thirty-day windows and posts the Stata code that pulls the metadata. When his own divorce prediction was published under a title that promised a future trend, he went back every year as new survey data landed and checked it in public.
A sociologist who does this much counting of other people’s compliance is running a private regulatory regime with no budget, no statutory basis, and one employee.
Asked to meet a National Academies panel on corrections and retractions, Cohen’s message was that sociology almost never corrects or retracts, and that the reason is circular: the discipline’s practices make it impossible to know whether serious errors or fraud are occurring. Trust cannot be extended because nothing can be checked, and nothing can be checked because trust has been extended. Elsewhere he describes the professional reflex he ran into over the Goffman survey, where asking to inspect the evidence behind an admired book got treated as a breach of collegial manners. That reflex is the claim of professional self-regulation, and Power’s book is about what happens after the claim stops being believed.
Cohen does not count correct findings. He counts whether a package exists. In the 2020 version of the audit he said so, noting that he had assessed the packages at a glance and had not tested anybody’s code. The improvement he has documented, from four of fifteen quantitative papers in 2020 to eight of twenty-five in 2023 to twelve of eighteen in 2025, is an improvement in documentation. Whether the code runs, whether the data support the tables, whether the models were specified before or after the results were seen: none of that is in the count. Nobody has run those replications. The audit checks the control system.
Power’s term for the produced condition is auditability, and the tripled proportion is a record of auditability being manufactured across five years. The relation between auditability and correctness is an open empirical question that Cohen’s series does not address and could not address, and he does not claim otherwise.
The auditees behaved as Power’s fifth chapter predicts. Cohen’s tables catalogue the ceremonial forms: data available from the corresponding author on reasonable request, materials on the author’s personal website where they are not, a PDF of the code linked from a homepage, a repository link that requires asking permission, and blinded review links pasted into published articles by authors who had no interest in whether a reader could follow them. He tells the journal’s staff to click their own links before publishing. Each of those artifacts is what compliance looks like when the count is known and the substance is not.
Then the other response arrived in his comment thread. A sociologist who had published in a journal with a mandatory reproducibility policy wrote that assembling a replication package for the first time took a full day of work despite tidy file management, that people default to available upon request because collating for a stranger is labor, that most researchers keep a system only they understand, and that the experience had converted him: he plans to build packages from now on whether or not journals require them, and at fifty-nine he does not accept being set in one’s ways as an excuse.
Both of Power’s responses to audit, ceremonial compliance and internal conversion, appeared in the same thread on the same December morning, one in the tables and one in the comments. The compliance cost the convert described, a day of skilled labor per paper, appears nowhere in the count. Power’s observation is that audit generates work whose value is invisible to the audit that generated it.
If audit displaces trust onto the auditor, then the auditor has to be trusted, and audit has produced a new object of faith. Financial audit cannot escape this, because the evidence the auditor examined belongs to the client and the auditor’s report is an opinion delivered to a principal. The public gets the opinion and never the working papers.
Cohen’s evidence is a journal volume sitting on a library shelf. His caveat is that his methods are squishy. His invitation is that corrections be mailed to his office, written on the back of a fifty-dollar bill. His metadata code is posted. Anyone who thinks he has miscounted Volume 90 can open Volume 90. The audit’s inputs are as public as its output, which makes trust in the auditor optional, and optional trust is the only kind that survives contact with a discipline that has stopped extending it.
That is also the best available account of why his counts move institutions and his verdicts do not. A count anyone can redo carries no author with it.
The costs of his regime are visible in two places. The first is his own career, which has migrated from first-order work to second-order work. The man who established a reputation on the devaluation of women’s work, on metropolitan labor markets, on divorce rates and marriage markets, now spends December afternoons clicking other people’s links. He has said as much: time spent on public outreach eats research time, and with fewer media mentions he might have more research worth mentioning. Power’s audit explosion is second-order activity, checking, at the expense of the first-order activity being checked, and here it can be traced through one publication record.
The second is the archive, which publishes its own compliance costs. Rejection rates far above historical levels. Moderation burden up. Revised versions now passing through the queue, which is welcome and which doubles the reading. Volunteers thanked. Cohen is one of the few people running an inspection regime who audits the cost of his own inspection in public.
Then came the paper that took the argument to its end.
Somebody submitted work to the archive that would have been accepted a few years earlier. It had an abstract. It had a title that reflected the contents. It had an ORCID. It had cited references, and the citations and quotations appeared to be real. Its statistical models were complex and apparently competent, with graphical and tabular results. It was coherent and relevant to existing research. The code was in a public repository. And at the end there was an AI disclosure listing the tools used to write the code, run the literature search, advise on the statistics, and draft nearly all of the text, alongside the author’s claim to have formulated the question, chosen the framework and the variables, directed the analytical decisions, interpreted the results, and verified every data claim and every citation. He offered a methodology audit on request.
Every element of the audit trail was present. The paper was dull, unoriginal, superficial, thin in its literature review, and empty where it claimed a theoretical contribution.
Power’s 2026 essay on auditability in the digital age describes the general condition: notions of evidence and auditability have become fluid, the spread of algorithms through the systems being audited pushes audit toward becoming algorithmic in turn, and the evidentiary basis loses its independence, until at the limit the audit becomes self-referential and the auditor is left as a “de-skilled curator” of models and their outputs. Cohen reached the same place from the other direction and wrote it out in the first person. The audit trail was manufactured by the technology whose presence the trail exists to disclose. He noted that a disclosure requirement he cannot verify is worthless, since the chatbots read the rules and will help an author appear to satisfy them, and that any attempt at detection would be an arms race against the people who own the machines, lost at the cost of the archive’s work.
He conceded the further problem in the same breath. His platform accepts work without peer review and assigns digital object identifiers and the other trappings of scholarly legitimacy, which makes it an attractive place both to deposit synthetic papers and to harvest real ones. He built an institution that issues auditable form without quality judgment, and the form is now being produced at industrial scale by systems with no interest in the judgment.
The policy that came out in March 2026 reads like the last chapter of Power’s book. It is written in the collective voice with the five members of the drafting subcommittee named. It states the competing interests it tries to balance. It refuses automated detection, refuses the arms race, commits the moderators to human judgment using perception and experience, makes no promise of a rule that could be stated completely, and invites appeals. Cohen and his colleagues wrote that they felt a need to insist on their humanity in the process.
Beyond audit, towards trust, arrived at by the auditor.
Cohen spent three years as an elected member of the association’s publications committee trying to get a policy adopted that would merely tell readers whether a published article came with its data. Two subcommittees produced a mild version. The governing council rejected it. He organized a petition against the association’s letter opposing open distribution of federally funded research, got more than two hundred signatures, got the committee to pass a resolution, and watched the council ignore it. Persuasion inside the institution moved nothing across three years. Counting in public moved the proportion of quantitative articles shipping code from a quarter to two-thirds across five. A reader who takes Power’s book as an argument against auditing has to explain that record.
An audit trail records what a person can be made to show. It was always a proxy, and everyone involved knew it, and the proxy held for as long as producing the trail was more expensive than doing the work. That condition failed sometime in 2025. The archive’s queue is now full of papers whose trails are immaculate, and the only remaining question about them is the one no trail has ever been able to answer, which is whether anybody meant it.
Cohen built the trail, followed it to the end, and came out into a room where somebody has to decide.
Notes
Michael Power, The Audit Society: Rituals of Verification (Oxford University Press, 1997; paperback 1999), 183 pages, six chapters, the last titled “Beyond Audit, Towards Trust.” Table of contents and publisher description. The precursor pamphlet is The Audit Explosion (Demos, 1994). Biographical detail, including the philosophy training, the Deloitte Haskins and Sells years, and the 1987 arrival at LSE, is from https://www.lse.ac.uk/people/michael-power and the publisher’s author note.
The two later pieces I lean on: “Modelling the Microfoundations of the Audit Society: Organizations and the Logic of the Audit Trail,” Academy of Management Review 46, no. 1 (2021): 6–32, open access at http://eprints.lse.ac.uk/100243/, source of the verbatim phrase “not inherently value-subverting and may be value-enhancing” and of the argument that the audit trail is performative of the conditions of its own expansion. And “Auditability in the Digital Age,” International Journal of Auditing (2026), https://doi.org/10.1111/ijau.70038, source of the fluidity of evidence, the platformization risk, the non-independence of the evidentiary basis, and the verbatim phrase “de-skilled curator.” Also relevant: “Theorizing the Economy of Traces: From Audit Society to Surveillance Capitalism,” Organization Theory 3, no. 3 (2022), which won the 2024 James G. March Prize, and the new book Economy of Traces: Traceability, Tracking and the Accounts We Live By (Oxford, 2026). The self-critical retrospective is “Afterword: Audit Society 2.0?” (2022), http://eprints.lse.ac.uk/115189/.
Power’s own vocabulary carries the essay. I kept out the neighboring literature that would have turned this into a composite: Espeland and Sauder on reactivity, Meyer and Rowan on ceremonial conformity, Strathern on audit cultures, Muller on metric fixation. The two-response pattern I describe as ceremonial compliance and internal conversion is Power’s chapter 5 material on the construction of auditees, in his terms.
Two claims that are mine. First, that Cohen’s practice escapes Power’s regress of trust because his audit inputs are public in a way financial audit evidence never is. Power’s regress argument turns on the privacy of audit evidence and the delivery of an opinion to a principal; he does not consider the case where the audited material is a published artifact available to every reader. This is the essay’s contribution and it is arguable, since a reader still has to trust Cohen’s coding decisions about what counts as a package, which is why his squishy-methods caveat is doing more work than it appears to. Second, that the proxy held only while producing the trail cost more than doing the work, and that the condition failed in 2025. Power’s 2026 essay gestures at the same collapse without dating it.
The debunking pull is real and the essay resists it in three places: the paragraph early on where Power disclaims being read as a debunker, the sentence conceding that the relation between auditability and correctness is unknown in both directions, and the late paragraph laying out the three years of failed committee work against the five years of successful counting.
The obvious next move is to ask whether Cohen’s audits have improved the correctness of published sociology, and the honest answer is that nobody knows, because the replications have not been run. Saying more than that would require me to assert something the record does not support. It is also the strongest single sentence available to a critic of open-science reform, and the fact that Cohen has never claimed otherwise is to his credit.
Philip N. Cohen Responds
From Philip N. Cohen (@philipncohen.com), Bluesky, July 27, 2026, 2:41 PM. Five posts.
1/ Well now I have been slop biographied. It’s on the website of someone named Luke Ford, along with 475 other biographies that “AI helped” him to “write.” I’m probably the only person who has ever read it, including the “author.”
2/ At 13,000 words, it’s pretty thorough, and superficially accurate, with lots of correct facts linked to real sources. But next time you read a real biography, be grateful the didn’t just rely on free online content: blog posts, the subjects writings, CV, official websites, and Wikipedia page!
3/ Also, I don’t know if this is a glitch, or these are supposed to be separate chapters, but halfway through it has a bibliography, and then starts what looks like a new biography of Luc Boltansky, except it’s using incidents from my life to reflect on Boltansky’s key works.
4/ Then it does it again and starts what looks like a new biography of Michael Power, my life now reflected through The Audit Society. It wraps up with conclusions about me, so maybe these are chapters in my biography? Someone is definitely sampling hallucinogens (& by someone I mean me, right now)
5/ We’re only a couple years into this hellscape, and already concepts like “author,” “writing,” “document,” “research,” and “hallucination” have been radically destabilized. If you think they haven’t, I’m jealous of your life!
On July 26, 2026, I emailed Cohen:
Dear Professor:
I’m a blogger in LA with no credentials (I just like to read your papers and posts). If you have the patience and the inclination, I’d be grateful for any feedback, even if it is just “LOL.”
I’d love to learn where I was inaccurate or unfair or just missing the point. If you’d rather read this with the links inserted, let me know. I wrote this with assistance from AI.
I then included the text from early versions of these two posts: The Live Questions in Sociology and Turner v Turner. Other sociologists I emailed responded with substantial answers that improved my posts.
Cohen responded:
Hi –
My feedback is: stop using AI to write.
Cheers,
Philip
Cohen criticizes me for producing a slop biography. He is right. My first pass is the mess he describes.
It is also a breadcrumb.
Breadcrumbs along the way is my motto.
When I am curious, I read what’s at hand, and then I post something.
The mediocrity of my initial effort embarrasses me. I want to improve it.
If I have time and drive, I do improve it.
On Aug. 15, 2026, I email Professor Cohen (his replies are in brackets]:
[I have no idea what you’re doing here.
Answers below]
Your ASR audits count whether a replication package exists, not whether it runs. AJPS sends every accepted quantitative paper to the Odum Institute, and only about 6% reproduced fully on first
submission. Does that change what you think your counts measure?
[No]
In 2019 you told the National Academies you knew of no research showing generation labels cause stereotyping. Has any appeared? If not, what does that say about the campaign?
[Don’t know]
In 2022 you said of ethnography: “the data is me, I was there.” You also proposed a restricted tier where vetted readers could see transcripts under agreement. Did anyone build it?
[Not to my knowledge]
Would you now say the Goffman dispute was two knowledge systems failing to convert rather than a dispute about accuracy?
[No]
On April 3, 2020 you told your class twice you weren’t an expert on pandemics, then told them not to listen to the president. Your book says disclosure manages the crossing between roles. How is a nineteen-year-old supposed to discount the instruction by the right amount?
[With care.]
You’ve said the value of open data is hard to prove and you believe it anyway. What would count as evidence against it?
[See Popper]
You call the Wilcox dispute lying. What would you have to see to conclude instead that a competent demographer reading the same evidence honestly disagrees?
[Asked an answered]
Melinda Baldwin is in your history department and has written the definitive account of peer review’s rise in the Cold War, which dates it later than your talks do. Have you two ever discussed it?
[I disagree with the premise.]
And after March 2026, do you still think the criteria can be written down.
[Yes]
Is There a Contribution in this Blog?
A colleague of Stanley Fish’s at Johns Hopkins was approached on the first day of a semester in the late 1970s by a student with a question. She asked whether there was a text in this class. He told her yes, the Norton Anthology. She said no, she meant in this class do we believe in poems and things, or is it just us.
He worked out where she had been. She had taken Fish’s course on interpretive theory the previous term, and she had arrived in a new room wanting to know which game was being played in it before she committed to a position.
Fish (b. April 19, 1938) built a career on the argument her question contains. The words on the page do not carry their meaning independently of a community of readers who have agreed, without ever voting on it, what counts as a reading, what counts as evidence for one, and who is qualified to offer it. He calls this anti-foundationalism and denies that it is relativism. Members of a community are constrained, and the constraint comes from membership.
Seventeen years after that exchange he applied the position to his own profession in a book called Save the World on Your Own Time, published by Oxford in 2008. The academic job is a distinct enterprise with a distinct good, which is to introduce students to bodies of knowledge and traditions of inquiry and equip them to work inside them. Three rules follow. Do your job. Do not try to do somebody else’s job. Do not let anybody else do your job. Political questions belong in the classroom as objects of analysis, which he calls academicizing them, detached from their real-world urgency so that a student examines what would make a president good. Academic freedom is freedom to do the academic job, and not freedom to do whatever job a professor thinks the moment requires.
In January 2025, Columbia published a book by Philip Cohen making close to the opposite argument. Values shape the questions anybody asks, the value-free posture is a fiction, and a scholar owes the public disclosure of where he stands, honesty about uncertainty, deference to evidence, and clarity about the edges of his competence. He grants that a man occupies several roles, researcher and teacher and activist and litigant and citizen, and locates the trouble in moving between them without saying so.
Both men have written the book. Neither has been made to argue with the other. There is a recording that does it for them.
On the afternoon of April 3, 2020, Cohen sat down in front of a screen and recorded a lecture for his undergraduate social problems class at the University of Maryland. He opens by saying he has missed them. Ninety seconds in he says he is not an expert on pandemics or global public health, that he is a social scientist who reads a lot and can understand what he reads, and that he is a teacher trying to explain where things stand.
What follows for six minutes is the work of a competent explainer. The etymology of pandemic. The date of the World Health Organization declaration. Why this virus is bad, stated as four conditions with the uncertainty attached to each: no immunity, high transmissibility, asymptomatic spread, and a mortality rate he says is maybe around one percent while saying three times that nobody knows. He explains why the six-foot rule is six feet. He tells them a cloth mask recommendation is coming.
At the seven-minute mark the lecture turns. He puts up the CDC budget falling from just under eight billion to just over seven with a proposed cut below that. He puts up the elimination of the pandemic office in the National Security Council and a report from the previous fall noting that it remained unclear who would be in charge in a cross-border biological crisis. He describes the 2019 federal exercise that found the government unprepared and produced recommendations that went unused. Then he quotes the president saying nobody could have seen this coming, and says it is a total lie, and adds that Trump did not know because he does not know a lot.
Then he plays the montage. Two minutes of clips. Totally under control. One person from China. Fifteen cases going to close to zero. It will disappear like a miracle. When it ends he tells the class that none of it was true and that the messaging existed so that the president would feel he was doing a good job.
The rest is arithmetic again, and it is the best part. The log-scale doubling chart with days since the tenth death. The correction he makes to his own slide, that the American curve looks worse than Europe’s until you notice the country is the size of Europe and holds several outbreaks. The lag argument: the people arriving at hospitals today were infected around the time the schools closed, so everything visible is already history. Thousands will die and the slope will not break on its own.
At the twenty-seven minute mark he tells them what to do. Follow real public health advice. Do not listen to the president. Do not listen to the politicians in his orbit. Listen to your governor, your local public health officials, the Centers for Disease Control.
Then he tells them to look after each other, not to let physical distance become social distance, and to think about who they want to be in the world that comes after.
Fish’s rule permits almost all of that lecture and forbids a specific part of it, and the part it forbids is smaller than his critics assume.
The budget numbers are teachable. The eliminated office is teachable. The ignored exercise is teachable. The doubling chart, the lag, the flattening logic, the correction about outbreak units, all of it is the academic job done well, and a professor who withheld it because the subject had become partisan would be failing at the thing Fish says is the only thing.
Even the policy dispute survives. Cohen argues that the federal government should have seized control of ventilator production and distribution. That is his own field and it is a live argument with a serious case on the other side about procurement capacity and the costs of commandeering supply chains, and Fish’s method would have him teach it as an argument. He teaches it as a finding.
What the rule forbids is the last minute. The instruction to a captive class holding his grade in their transcripts, from a man who has twice told them he lacks the relevant expertise, about which public officials to believe.
Cohen has an answer. There is no neutral version of that lecture. A professor who decides to show the CDC budget has already chosen, and one who decides not to has also chosen, and the choice is invisible in the second case, which makes it worse. He disclosed his limits at the start and at the end. The instruction he gave was about competence: listen to your governor, whoever your governor is, and to the public health agency. And in April 2020 the cost of the professorial shrug was measurable in a currency other than pedagogy.
Fish’s reply is the one about structure. The disclosure changes nothing about the room. A nineteen-year-old who thought the administration was handling it adequately now knows what the man grading his papers thinks of people who hold that view, and the disclaimer does not restore his position. Fish would say Cohen is trading on an authority the university conferred for a competence Cohen has already denied having, and that saying so out loud is a courtesy. Academic freedom protects the analysis.
The men disagree about whether a role can be limited by an act of will and announcement.
Run Fish further and he takes something from Cohen.
The chapter of Fish’s work that applies here is the one on anti-professionalism, from Doing What Comes Naturally (1989). His claim is that attacking the profession from inside it is a professional move, the highest-status one available, and that the attacker’s standing to attack is a product of the institution he is condemning. There is no outside from which to speak. The critique of the guild is a genre of guild performance.
Cohen resigned from the American Sociological Association in November 2021 and published his reasons. Dues too high. Subscriptions doubled in eight years. A publishing contract signed without notice. Revenue transferred from teaching-heavy libraries to research-heavy ones because publication in the journals is dominated by elite universities. Every one of those claims is correct as far as I can check.
The letter carried because a tenured full professor at a flagship wrote it, and it reached an audience because he had spent a decade accumulating the standing the association helps distribute. He speaks for the community college instructor from a chair she will not occupy, and she arrived in his comment thread to say she was cancelling her membership, which is the effect he wanted and also the transaction Fish describes. A junior scholar in the same thread asked where new people would build networks if established people left. That question is Fish’s argument arriving by instinct from somebody with no theory and something to lose.
Fish would say the position is unavoidable and that Cohen should stop describing it as an escape. Then he would look at the archive, which issues digital object identifiers and the other marks of scholarly legitimacy without peer review, and observe that a man who left one credentialing body has spent ten years building another.
Fish’s model says interpretive strategies constitute their objects, which implies there is no evidence available to settle a dispute between communities that disagree about what evidence is. Cohen’s audits are an awkward case. He opens a volume of a journal, clicks the links, and reports how many quantitative papers arrived with code and data. Twelve of eighteen in 2025, against four of fifteen in 2020. An ethnographer who thinks the count is beside the point can redo the count and get the same number. The count travels across the boundary while its meaning does not.
Fish might say that the agreement about what a link is, what counts as a package, and what makes two counts the same count is not evidence of a world outside interpretation. It is the description of a community wide enough to include both parties, and the width was invisible until somebody tested it. What the audits show is that the community sharing a notion of a countable artifact is larger than the community sharing a notion of a valid ethnography.
That answer is also unfalsifiable. Every apparent counterexample gets absorbed by widening the community until it contains the agreement. A theory that cannot lose is doing something other than explaining.
Which brings the argument to July 2026.
Cohen read a long essay about his own life, produced with a machine’s help, and wrote five posts on Bluesky about it. He called it slop. He said it was thorough and superficially accurate with correct facts linked to real sources. He complained that it relied on freely available material, his blog and his writings and his CV and Wikipedia. He described losing his place halfway through, hitting a bibliography, and finding what looked like a second biography, of somebody else, using incidents from his life. His last post said that a couple of years into this the concepts of author, writing, document, research, and hallucination have been radically destabilized, and that he envies anyone who thinks otherwise.
Fish has been saying for fifty years that those concepts were never stable in the way the complaint assumes. Author is a set of practices for assigning credit and blame. Research is what a community recognizes as research. Document names a thing a community has agreed to treat as fixed. None of them ever rested on a property of the object, and all of them held firm anyway, because agreements that nobody has to defend feel like facts. What the machines did was withdraw the agreement. Fish would say nothing was destabilized in July 2026 that had ever been stable, and the panic is what it feels like when a settled practice becomes visible from outside.
Four months before the thread, Cohen’s own committee wrote the Fishian document.
The archive had received a paper that satisfied every requirement. Abstract, real citations, competent models, code in a public repository, and a disclosure listing every tool used to write it. Dull, thin, and empty where it claimed a contribution. Nothing checkable was missing.
The policy his subcommittee published in March 2026 refuses automated detection, declines to promise a standard that could be written down completely, commits the moderators to judgment made by people using perception and experience, and invites appeals. It says the group felt a need to insist on its humanity in the process.
That is a statement that competent members of a community can recognize the real thing without being able to specify the rule, which is Fish’s position on legal interpretation, literary interpretation, and everything else, arrived at by a man who spent ten years trying to build a machinery of verification that would make membership unnecessary.
The two halves of the argument turn out to be one argument. The reason disclosure cannot solve the classroom problem is the reason detection cannot solve the archive problem. Both proposals assume a place to stand outside the community from which the community’s practices can be certified. The professor announcing his standpoint is still the professor, holding the grade book, inside the room. The archive announcing a disclosure requirement is still the archive, and the rule gets read by the machines it was written to catch.
What Fish’s position does not supply is a way to act. He can say there is no outside and be right and have nothing to offer a man deciding on April 3, 2020, what to tell frightened students about whom to believe. Cohen decided. He said what he thought was true, he flagged the limits of his competence twice, he got the epidemiology broadly right and the vaccine timeline pessimistic, and he stepped over a line Fish thinks a professor should not cross. He has never claimed the line does not exist. His book is an argument about how to cross it in the open.
The student walked into the room at Hopkins and asked whether there was a text in this class. A student in a different room in April 2020 would have been asking a version of the same question with less time to spare. Is there a fact here, or is it just us, and either way, do I go home this weekend.
Cohen’s answer was that there is a chart, that the slope will not break on its own, and that he is telling them where he stands. Fish’s answer is that the chart persuades because they are already the kind of people charts persuade, and that the professor should have taught them what makes a chart persuasive and left the rest for after class.
One of those men had a class in front of him.
Notes
Frame. Stanley Fish, Is There a Text in This Class? The Authority of Interpretive Communities (Harvard, 1980), for the title anecdote and interpretive communities. The standard telling is that a Hopkins colleague was asked the question on the first day, answered with the Norton Anthology, received the clarification about poems and things, and then worked out that the student had taken Fish’s theory course the previous semester. Fish taught at Johns Hopkins from 1974 to 1985 as Kenan Professor of English and Humanities, which dates the story.
Save the World on Your Own Time (Oxford, 2008). The three rules, the mission statement about introducing students to bodies of knowledge and equipping them with analytical skills, the argument that academic freedom is freedom to do the academic job, and the concept of academicizing a topic by detaching it from its real-world urgency, are all from that book. The Bush example of academicizing (asking what makes a president good) is at around page 27; see the Chronicle discussion at https://www.chronicle.com/blognetwork/tenuredradical/how-to-do-your-job-without-really. Publisher summary at https://www.publishersweekly.com/9780195369021.
“Anti-Professionalism,” in Doing What Comes Naturally: Change, Rhetoric, and the Practice of Theory in Literary and Legal Studies (Duke, 1989), is the source for the argument I run against the resignation letter. The claim there is that anti-professionalism is generated by professionalism and depends on it, and that the critic’s standing is conferred by the structure he attacks.
The CDC appropriations figures, the elimination of the NSC directorate in the 2018 reorganization, and the 2019 federal pandemic exercise are documentary and were reported before the outbreak. The claim that ventilator production should have been federally directed is a policy argument with a real case against it, and Cohen presents it as settled. “Total lie” is a character verdict. I have kept those three categories separate in the text and have not told the reader what to conclude about the administration.
Two claims are mine. First, that the classroom problem and the archive problem have the same structure, because both proposed solutions assume a standpoint outside the community from which its practices could be certified. Fish would accept it; he does not say it, because he has never written about preprint moderation. Second, that Cohen’s audits are a limited counterexample to strong Fish, in that a countable artifact travels across a boundary its meaning does not. I give Fish the winning reply and then say the reply is unfalsifiable, which is the honest place to leave it and the sentence a hostile reader of Fish will quote.
What I left out. Fish’s position on faculty bodies voting political resolutions bears on the association material and would have doubled the length. His exchange with David Horowitz over the Academic Bill of Rights is the obvious next move if you want to place him politically.
Hero System
On December 6, 1973, an editor at Psychology Today telephoned a house in Vancouver. He had read a manuscript that had come across his desk a few weeks earlier and decided within the hour that he wanted to interview the man who wrote it. The man was in the hospital by the time Sam Keen’s got there. He had colon cancer and would be dead in three months, at forty-nine.
The first thing Ernest Becker (1924–1974) said when Keen walked into the room was that he was being caught in extremis, that this was a test of everything he had written about death, and that he had a chance to show how a man dies and what he surrounds it with.
His career had gone badly by the standards of the system he was in. He lost his first academic job at Syracuse after siding with a colleague in an academic freedom fight and never got tenure there. He moved through Berkeley, San Francisco State, and Simon Fraser, always somewhere else, always leaving. Two months after his funeral the book won the Pulitzer Prize.
What he argued in it is simple to state and difficult to hold. Man knows he will die, alone among the animals, and cannot bear it. Culture exists to give him a way of earning the feeling that his life counts for something beyond his body. Becker called the arrangement a hero system: a set of rules specifying what a man must do to be worth more than an animal that rots. Every society issues its own, every man is inside one, and the currency of one does not convert into the currency of another. The Aztec who took captives for the sun and the Quaker who would not lift his hand were both trying to be heroes. Neither could spend the other’s money.
The prize arriving after the funeral is the part of Becker’s biography that reads like an illustration of Becker. He had built a thing that would outlast him, and it did, and he was not there to receive the receipt.
In November 2025 a foundation in Europe invited Philip Cohen to open its awards with a lecture at the European University Institute, which occupies a former abbey in the hills above Florence. The prize was for infrastructure. The money was ten thousand euros, which for a full professor at a state flagship is not money. The archive he founded in 2016, SocArXiv, had by then taken in something like seventeen thousand papers and was accepting more than three thousand a year. The citation named him.
A man flies to a converted monastery in Tuscany to accept a medal for building a website that will keep working after he stops. The Beckerian reading is available in one sentence: Infrastructure outlasts its builder.
Nobody needs a theory of death anxiety to notice that a man in his late fifties has been putting his energy into a thing with no expiration date.
Cohen’s hero system promises that what he said will still be checkable after nobody remembers who said it. His sacred value is verification.
Watch what he does with his own standing. He posts his code. He tells readers his methods are squishy. He invites corrections, to be mailed to his office written on the back of a fifty-dollar bill. He publishes a paper predicting a trend and then returns every year, in public, to check the prediction against new data that could embarrass him. He publishes the rejection letters he receives. He tells the world that his own archive is full of bad work. Every one of those gestures spends down the personal authority that his profession issues as its wage.
He spends it to buy something else. A count of replication packages in a single volume of a journal has no author inside it. Anyone with the volume can redo the count and get the same number, and if the count is wrong the person who made it is beside the point. That is what he is building toward across everything: claims that do not depend on the continued existence, reputation, or good faith of any living man.
Set that against the system he fights. Under the trust regime, a finding is credible because a serious person vouched for it, and the serious person is serious because other serious people vouch for him, and the chain runs back through advisers and departments and prizes to men who are now dead but whose names still function. That system produces knowledge that lives in persons. It also means knowledge dies with persons, or worse, survives them as authority nobody can any longer inspect. Cohen’s objection to it is technical when he states it and theological underneath. He wants a literature that does not require anyone to be trusted, because everyone dies and trust dies with them.
His mother Avis Cohen cut a segment out of a lamprey’s spinal cord and found that it kept firing in the rhythm of swimming with no brain attached. The finding entered engineering. Robots walk better when the pattern lives in the limbs. Her most cited paper is about a quadruped machine, and most of the people building on her work do not know her name.
That is impersonal survival.
Cornell’s system rewarded something else, and by its rules she failed. She ran a lab there for ten years on her own federal grants and never received a tenure-track offer. Maryland made her a full professor in 1990 and she founded a program and won a fellowship and retired emerita, and none of that alters what happened in Ithaca. In the academic hero system, standing is a title conferred by an institution on a person, and a person without the title is a research associate for a decade.
She died on June 6, 2026. Nine days later her son published his account, which included the arithmetic of the grants and the papers she held by 1990, a concession that the earlier decision may have been reasonable, and a refusal of the later one. He named no chairman. He put Suzanne Bianchi beside her and named the category: faculty women who were never assistant professors.
He could not give her the title. He gave her the record.
Now take the word he lives by and hand it to other men.
Transparency, to a scholar in a study house on a Tuesday night, means something almost opposite to what it means in College Park. He is working a page that carries an argument between two rabbis dead nineteen hundred years, and the losing opinion is preserved with the loser’s name on it. His immortality runs through attribution. Ask him for the dataset and he will not understand the request; the tradition is the dataset, and it is a chain of persons.
To a reporter, transparency has a hole in the middle of it that is the point of the whole trade. He shows his readers everything except the one thing that would cost him his standing forever, which is the name of the man who told him. A colleague at the next desk says the editor wants the source. He says the editor can want. In his system the hero is the man who goes to jail. Cohen came out of that trade. He worked the police calls at five in the morning and reviewed local bands on a weekly deadline and learned in that building that the sacred act is refusing to reveal. Then he moved into a system where refusing to reveal is the cardinal sin. The same man, the same word, opposite content.
To a surgeon in a morbidity and mortality conference on a Thursday morning, transparency is a sealed room. Inside it he will say out loud that he took the wrong approach and the patient died, and nobody will use it against him, and that is what makes the room work. Outside it he will not say a word, and the reason is not cowardice. A surgeon who has been broken in public develops a hand that hesitates, and a hesitating hand kills people. The chief tells the resident afterward that what happens in that room does not leave it, and the resident, who has read about accountability, thinks the chief is protecting himself, and the chief is protecting the next patient. He may also be protecting himself.
To a woman entering names in a family history center, transparency is a sacrament. She is looking for a great-great-grandmother in a parish register scanned from a village that no longer exists, and the reason the church spent decades filming registers and cut a vault into a granite mountain to hold the film is that the record makes reunion possible, and reunion is the thing that defeats death. Records are salvation infrastructure. Ask her whether the record should be open and she will say yes with more force than any open-science advocate, and mean something no open-science advocate means.
To a case officer, transparency is what you inflict on the other side. His own opacity is his virtue and his obligation. He has names in his head belonging to men who will be killed if the names are written down, and the greatest crime available to him is disclosure. He is inside a hero system where the immortality is the service and the service is invisible, and the men who did the most are the men nobody will ever hear about.
And to a novelist, the demand to show your working is a category mistake. What she is trying to make is the thing only she could have made. Replication is the definition of failure.
Five systems, five sacred meanings, no exchange rate. There are dozens more. Cohen’s usage makes sense only inside a system that has already decided what a man’s life is worth and how he can earn it, and that outside that system the same word buys nothing.
The competing systems that bear on him are the ones in his own discipline.
The ethnographic system runs on presence. What makes the work is that one person went and stayed and saw, for years, at cost, and came back with something no procedure could have produced. Its saint is Erving Goffman (1922–1982). Its currency is the singular observer, which is why the tradition transmits through named masters and their students. In that system, a book is heroic because only its author could have written it.
Cohen went at Alice Goffman’s survey with census arithmetic and asked how 217 homes on one block held 308 men between eighteen and thirty. It is a fair question and he was right to ask it. Inside the system he was asking it of, the request reads as a demand that she prove she was not lying, which is an assault on the form of heroism she was practicing. Her answer, that this was an ethnographic survey with its own sampling and its own definition of who belongs to a home, designed to reach people conventional surveys never find, is what that system says when asked to convert its currency.
The family-restoration system is the other collision. Brad Wilcox and the people around him work inside a hero system where the immortality on offer is children, marriage, lineage, and the continuity of a people across generations, which is the oldest such system there is and the only one most men in history have had access to. Cohen’s demography reads, from inside it, as an attack on the last durable thing available to an ordinary man. And Wilcox’s success sequence reads, from inside Cohen’s, as a promise dressed as a finding, unfalsifiable where it counts. They have been at this for a decade and neither has moved an inch, and the reason is not that either lacks data. Each is defending the currency of the other’s afterlife.
Then in the winter of 2025 something arrived that neither system had a price for.
The archive started receiving papers that satisfied every requirement Cohen’s program had spent ten years installing. Abstract present. Title matching contents. Citations to work. Complex statistical models, apparently competent, with tables and figures. Code in a public repository. And at the end, a disclosure listing every tool used to write the code, run the literature search, advise on the models, and draft the text, with the author’s assurance that he had chosen the question and the variables and verified every number, and an offer to submit to an audit on request.
The paper was dull, unoriginal, thin, and empty where it claimed a contribution. Everything auditable was in order.
This is what an immortality project looks like when it succeeds. Cohen had set out to build a literature that did not depend on any person, and the machines produced exactly that: claims with a perfect trail and nobody behind them. The author had been removed from the claim, which was the goal, and what came out the other end was writing that no one needed.
He turned it down, and then spent months in public working out why. He is not persuaded that using the tools is wrong. He thinks the training may be lawful. He grants that his own archive, handing out identifiers without peer review, is an inviting place for machines to deposit and to harvest. He will not enter a detection arms race he expects to lose. He says a disclosure rule he cannot verify is close to useless, since the chatbots read the rules too. And the policy his committee published in March 2026 refuses to promise a standard that can be written down, commits its moderators to judgment made by people using perception and experience, and states that they felt a need to insist on their humanity in the process.
Becker’s term for the thing a man cannot look at directly is the vital lie, and his claim was that a man discovers what he was defending at the moment it is taken from him. Cohen spent thirty years engineering the person out of the claim, in the belief that persons are the weak point, which they are, because they die. What he found at the end of the trail was a room full of documents in perfect order that no person had meant, and his response was to reach for the one thing his system had never issued a receipt for.
He is fifty-nine in August. He counts the dead for a living.
Sources and notes
Frame. Ernest Becker, The Denial of Death (Free Press, 1973), Pulitzer Prize for General Nonfiction, announced May 1974, two months after his death. Escape from Evil (Free Press, 1975) is the posthumous companion and the place where the argument about heroism producing evil is fullest. Biographical detail from https://en.wikipedia.org/wiki/Ernest_Becker: born September 27, 1924 in Springfield, Massachusetts; died March 6, 1974 in Burnaby, British Columbia, aged 49; the lost job at Upstate Medical College in Syracuse over the Szasz academic freedom dispute; the subsequent posts at Berkeley, San Francisco State, and Simon Fraser. The deathbed scene and the December 6, 1973 phone call are from Sam Keen’s foreword to the 1997 Free Press edition, which is readable in the scanned copy at https://ia803107.us.archive.org/11/items/thedenialofdeath/The%20Denial%20of%20Death.pdf. The Keen interview appeared in Psychology Today in April 1974.
Frame discipline. Becker only. Otto Rank stands behind the argument about the artist who submits to a larger design, and I left him out.
Cohen material. All linked in the earlier notes.
The Erving Goffman sentence. I say the ethnographic tradition transmits through named masters and that its greatest name belongs to Alice Goffman’s father (Erving Goffman, 1922–1982, ASA president-elect at his death).
The Erving Goffman sentence. I say the ethnographic tradition transmits through named masters and that its greatest name belongs to Alice Goffman’s father (Erving Goffman, 1922–1982, ASA president-elect at his death). That is a claim about how the system authenticates, and I wrote it so it makes no claim about how she got her standing.
Invented dialogue. The five archetype vignettes carry the only dialogue in the essay, and every line of it is invented and spoken by an unnamed representative figure. No line of speech in this essay is attributed to a real named person unless it is sourced above.
Archetypes. Talmudic study partner, investigative reporter, surgeon at a morbidity and mortality conference, Latter-day Saints family history worker, intelligence case officer, plus a one-line novelist. None repeat the Rand essay’s set. The granite vault is the Granite Mountain Records Vault in Little Cottonwood Canyon, Utah. The M&M conference’s protected-discussion status varies by state statute and hospital policy, so I described the norm.
Two claims that are mine. First, that verification is a death-denial technology because an impersonal claim outlives the person who made it, and that Cohen’s program therefore buys durability by spending down authorship. Becker’s immortality projects are almost always ego projects in his telling; the version where a man seeks the survival of the claim at the cost of his own name is an extension, and it needs Rank’s artist to make it fully respectable. Second, that the machine-written submission is the logical completion of that project.
There is a version of this essay that reads his mother’s distributed-control science as the template for a decentralized archive and treats the son as executing the mother’s picture of the nervous system. It is seductive and it is unfalsifiable, and the biography already notes that he has not drawn the line in print. I used her work here for a different purpose, as an example of impersonal survival, which is a claim about what happened to her findings.
Expertise and Democracy
On April 3, 2020, Philip Cohen sat down in front of a screen to record a lecture for his introductory Social Problems class and said, twice, that he was not an expert on pandemics or public health. He was a social scientist who read a lot and understood what he read, and he was a teacher.
Twenty-three minutes later he told the same students to follow the real public health advice, to listen to their governor and their local health officials and the Centers for Disease Control, and not to listen to the president.
Both statements are honest. Both are the sort of thing his book on public scholarship recommends. And a nineteen-year-old sitting in his own bedroom with that recording had no way to act on the first while receiving the second. To discount a professor’s political instruction by the amount of his disclaimed expertise, you would have to know how far his competence extends, which is the thing you consulted him about.
Stephen Turner has spent twenty-five years on that problem. Liberal democracy rests on the political competence of ordinary citizens: on the idea that the people affected by a decision can evaluate the case for it. Expert knowledge is by definition knowledge the ordinary citizen does not have and cannot acquire without becoming an expert. So when a polity governs itself partly through claims it cannot assess, something in the legitimating story has broken, and no amount of goodwill on the experts’ part repairs it. Turner’s contribution is to refuse both easy exits. He will not say that expertise is a mask for interest, which is the debunking move, and he will not say that better science communication will close the gap, which is the professional move. He asks instead what publics do with expert claims, and builds his account from that.
What they do, in his telling, is make second-order judgments. Not judgments about the content, which is unavailable, but judgments about the source: who is saying it, who pays them, whether other experts contradict them, whether they have been right before. That is the real currency, and it is thinner than the discussion model assumes and more robust than the debunkers allow.
Turner also insists that experts are not one thing. He sorts them by the character of the audience that accepts them. At one end sit the fields whose authority is granted by nearly everyone, where a physicist’s claim is taken as settled by people with no capacity to check it. Further along are experts with a following they created, who persuade a self-selected group to accept a body of doctrine, and whose authority extends exactly as far as the conversion does. Further still are experts whose real audience is a patron, a funder, an agency, a bureaucracy, where public acceptance is incidental to the arrangement that sustains them.
Sort Cohen’s career by that scheme and the result is not what his own model predicts.
The generation labels campaign succeeded, and the success ran through an institution. In July 2019 he addressed a National Academies committee, which is a quasi-public expert body deliberating in a space no public attends. Two years later he circulated a letter carrying roughly a hundred and fifty signatures from demographers and social scientists to the Pew Research Center. In May 2023 Pew changed its policy. At no point did the American public express a view about birth-year boundaries. A professional group told a research organization that people trained as they were trained found a practice unacceptable, and the organization, whose standing depends on that group’s regard, adjusted.
The transparency audits work the same way. Cohen counts replication packages in a journal read by sociologists, and the count is legible to editors and authors. The share of quantitative papers arriving with code went from four of fifteen in 2020 to twelve of eighteen in 2025. In December 2025 an author he had marked as providing nothing appeared in his comment thread, agreed, and promised a package. That is one professional recognizing another professional’s demand as legitimate.
The blog, the Twitter presence, the archive: these reach a following he assembled, largely people already disposed to what he says. The 2020 pandemic lectures reach enrolled students, a captive and graded audience. The lawsuit reached a court.
Every one of those is an intermediary. Cohen’s book proposes a scholar who reembeds himself with a public and offers his own accountability as the guarantee. His actual influence has run through professional bodies, editorial staff, survey organizations, judges, and students who could not leave. The public in the citizen-scholar model is a public he has never had to persuade.
This is not a charge of failure. Turner’s point is that this is what expertise in a democracy generally looks like, and that describing it as public discussion is the delusion. Cohen’s most effective campaigns were guild actions on intermediaries, and they worked because they were guild actions on intermediaries.
Now the part he has seen.
In February 2025, at a Wisconsin workshop, he raised his own hardest question and attributed it correctly. We show our good acts in one area, he said, to generate legitimacy in another, and he called it the Bourdieu question of the exchange rate between different types of capital. Do I get to write an op-ed about Trump because I have written articles about divorce rates? He added that he could be wrong about it and was happy to argue either way.
Turner’s answer is that the exchange rate is unregulated because the only people who could regulate it are the people who cannot evaluate the currency. The public accepts a claim about pandemic policy from a family demographer on the strength of a credential earned in family demography, and the acceptance is reasonable given what the public can see, and there is no procedure available to the public for discounting it by the right amount. Cohen’s proposal, that a scholar be clear about the edges of his expertise, presumes a listener who can locate those edges independently. If the listener could, he would not need the expert.
Which is why the April 3 lecture is the case rather than an aberration. The disclosure was made. It was accurate. And it was inoperable by the people it was made to.
There is a second Turnerian argument running through Cohen’s work that he makes without knowing whose it is.
Turner’s worry in Liberal Democracy 3.0 is not experts as such. It is the fusion of expert knowledge with funding and administration, which produces a body of commissioned knowledge that shapes public life while sitting outside public discussion. Knowledge paid for by the public, generated inside institutions the public sustains, and then made unavailable to it.
That is Cohen’s open access case, exactly. Journals do not only transmit findings; they restrict access and convert unpaid academic labor into corporate revenue. The public supports the work through grants and salaries. We do not aspire to have our work hidden from the people it most affects. A researcher in a country whose libraries cannot buy the packages should read what a researcher at Harvard reads.
Every part of that is a democratic argument about commissioned knowledge, and it is stronger than the efficiency arguments usually made alongside it.
Then Turner’s distinction bites. Availability in principle is not availability in fact. A stranger can now download “The Coming Divorce Decline” for nothing. He cannot evaluate an age-standardized rate, cannot assess what adjusting for marriage order does to the estimate, cannot tell whether the American Community Survey supports the inference. The paywall was a barrier. Removing it does not create a reader. It relocates the problem from access to competence, and competence is the harder half.
Cohen has an answer to this and does not seem to recognize it as an answer.
In February 2025 he described what he called a descriptive turn in his own research: less concerned with discovering theories and mechanisms, more concerned with figuring out what is happening and communicating it. He said the profession punishes this. If a finding is merely descriptive it is considered lesser, so authors add puffery to make work sound like more than description, which he called off-putting to the public.
Description is the one form of social scientific output a lay reader can assess. When he tells a room that the bottom tenth of Americans sits above half the world’s population in income, or that the average American could expect twenty-nine of the years between eighteen and fifty-five in marriage in 1960 and eighteen now, he has handed over something a person can hold, compare against his own life, and argue with. The Gini lecture builds the measure out of four people holding stacks of cash and counts squares by hand, and a student with no statistics can follow every step and check it.
That is the democratic move, and it is a bigger one than open access. He presents it as a personal preference and a professional cost. On Turner’s terms it is the only route by which expert knowledge becomes evaluable by the people it describes, and the profession’s contempt for it is the profession choosing its own standing over public competence.
Now the hardest thing the frame produces, and it runs against him.
In February 2025 Cohen complained about a meme. Do your own research, he said, means do not listen to somebody else’s research, and it means not trusting experts and science and the accumulated wisdom of the institutional framework. He added that people who do their own research know less than people who do not.
Turner’s account of second-order judgment says where that disposition comes from. A citizen who cannot assess content watches the sources. What he chiefly watches for is disagreement, because visible disagreement among credentialed people is the strongest available signal that the credential does not settle the question. When two demographers with comparable training, comparable data, and comparable institutional standing spend a decade in public argument, and one of them titles his posts with the claim that the other lies, a lay observer draws the correct second-order inference. The field cannot resolve this. Therefore the credential is not decisive. Therefore I will have to decide.
Cohen believes the Wilcox posts demonstrate accountability. In the currency the public uses, they demonstrate that demography does not adjudicate its own disputes, which is precisely the premise of the disposition he deplores. The transparency program has the same structure at a larger scale: a professional telling the public, at intervals, in writing, that a third of the articles in his discipline’s flagship journal cannot be checked. That is true and it was worth saying and it is also, on Turner’s account of how publics reason, a withdrawal from the account that pays for expert authority.
He cannot have it both ways, and he is close to admitting it. In February 2025 he said he did not think his lawsuit undermined his research, and then granted that it would not be hard for somebody to find one of his articles on divorce and post about it on Twitter with an antisemitic comment. That is the exchange rate running in the other direction, and he sees it clearly when the cost is to him.
Two objections, and the first is his.
Turner’s model, pushed hard, says the public can only assess sources, so transparency is pointless because nobody reads the code. Cohen answered that in February 2022, and the answer is good. Even if people do not read your data and code and validate everything you do, he said, the fact that it is available says a lot. It says you are open to being held accountable, open to collaboration, not trying to slow other people’s work down. He added that this is hard to prove empirically and that he believes it anyway.
That is a second-order signal in exactly Turner’s sense. The availability is not a content channel. It is a costly display of willingness to be checked, and a public that reads sources rather than papers is precisely the public that can read a display like that. Cohen supplied the strongest reply to the strongest objection to his own program, in an aside, while explaining a repository to twenty people on a video call.
The second objection is about scope. Turner’s expertise writing is largely about experts inside the machinery of government: advisory bodies, regulatory science, commissioned reports, the apparatus through which states know things. Cohen holds no such position. He advised the Census Bureau once, on the enumeration of same-sex couples after the 2010 census, and testified to a Senate committee once, in 2007, and addressed a National Academies panel twice. He is otherwise a professor with a blog. A good deal of Liberal Democracy 3.0 simply does not reach him, and an essay that applied all of it would be inflating a private citizen into a functionary.
What survives is the part about publics, and it lands on the proposal at the center of his book.
Cohen’s answer to the trust problem, borrowed from Anthony Giddens (b. January 18, 1938), is that modern institutional trust is disembedded. The elevator inspector’s name is on the certificate and the inspector is not there. What a scholar can do, he says, is be a reembedding agent: show up at the moment, in the elevator, and say you do not have to worry about this because we figured it out and here is how, and I am a person who lives in your community, and you can trust me for some reason.
Read that sentence with Turner in hand and the trouble is visible. It works by converting an unevaluable claim into an evaluable person, which is the right kind of solution, because a person is something a non-expert can assess. And it works only where the personal presence is real. Almost nobody lives in your community. The scholar who shows up in the elevator shows up in one elevator.
Scale is what the proposal cannot supply. And scale is what the archive was built for: seventeen thousand papers, four thousand a year, available to anyone anywhere, an apparatus explicitly designed so that trust need not depend on knowing who deposited the work.
So the two halves of his career answer the same problem in opposite directions. The archive makes the claim impersonal so that no one needs to trust the claimant. The citizen scholar makes the claimant personal so that someone can trust him. Each is a reasonable response to a real difficulty. They cannot both be the answer, and nothing in the book explains which one is load-bearing.
The record suggests the first. The counts moved journals. The letter moved Pew. The archive is still there and still growing. The reembedding happened in classrooms, in comment threads, and in a courthouse hallway on Pearl Street, and it moved the people already in the room.
In July 2019 he sat in front of a National Academies committee and made his case about generation labels, and a member asked whether any research demonstrates that the labels do harm. He said he knew of none, that most sociologists do not use the categories as real ones in their research, and that it would be an interesting experiment for somebody to run. The committee thanked him. Nothing happened.
Two years later, a hundred and fifty signatures on a piece of paper, and the largest survey organization in the country changed its policy.
The argument in the room built for arguments did nothing. The guild, asserting itself to an institution that needed its regard, worked. Turner’s work on expertise is an attempt to describe a democracy in which that is the ordinary case.
Notes
Frame. Stephen Turner, “What Is the Problem with Experts?”, Social Studies of Science 31, no. 1 (2001): 123–149, which is where the typology of experts by audience is set out, and Liberal Democracy 3.0: Civil Society in an Age of Experts (Sage, 2003), which is the source for the argument about commissioned knowledge, quasi-public expert deliberation, and the transformation of liberal democracy by expert-staffed bodies. Also relevant: The Politics of Expertise (Routledge, 2014), which collects and updates the position.
The claims that are mine. Four, in ascending order of exposure. That Cohen’s influence has run entirely through intermediaries rather than through the public his book imagines. That his descriptive turn is a better answer to the democratic problem than his open access campaign, and that he does not present it as one. That the Wilcox fight, conducted publicly between two credentialed demographers, feeds the do-your-own-research disposition he deplores, because visible expert disagreement is the strongest second-order signal a lay observer has. And that the archive and the citizen scholar are opposite solutions to the same problem, one impersonalizing the claim and one personalizing the claimant, with nothing in the book adjudicating between them.
The third is the one that will draw fire, and it should. The reply is that the alternative, experts concealing their disagreements to preserve public confidence, is worse and produces a more brittle authority when the concealment fails. That reply is strong. It does not dispose of the point, which is about the cost rather than about whether the cost is worth paying.
What is missing, and it is the same thing as always. Turner is at the University of South Florida and has been publishing steadily; Cohen is at Maryland. Neither has been asked the question this essay turns on, which is whether an expert who tells the public his own discipline cannot check itself has strengthened public reason or withdrawn from the account that funds his standing. Both would answer it. Neither has been asked.
Convenient Beliefs
The most revealing sentence Philip Cohen has spoken in public was addressed to about twenty demographers on a video call on October 27, 2020. He had been showing them proxies for events that had not happened yet: Google search indices for contraception and pregnancy and wedding dresses, condom sales from earnings calls, two states’ monthly vital statistics. He warned them about every one before he showed it, including, at 5:10, that he had not kept a record of the searches he tried that did not look good. Then he put up a chart of divorce-related searches showing a decline followed by a rebound, and said this.
“This is very speculative, but it does fit my presumption, which is probably why I’m showing it to you.”
That is a man narrating a convenient belief at the moment of its formation, out loud, and then proceeding anyway. The interesting question is not whether he was fooling himself. It is why that clause has no successor: why he can see the operation with that clarity in one domain and be blind to a structurally identical operation at the center of his life’s work.
Turner’s answer, developed in the 2025 Boudon essay, is that convenient beliefs are absorbed through community practice and tacit formation, and by the time a man can inspect them they present themselves to him as things he reasoned his way to. The self-report is never the mechanism. Cohen could catch himself with the search-index chart precisely because that belief was a fresh one, formed that month, still visible as a preference. The load-bearing convenient beliefs of his career were installed before he had the capacity to inspect anything.
Take the deepest one first.
Cohen’s program asserts that transparency improves knowledge. Publish the code, deposit the data, and the literature gets better. He has counted the deposits for five years and reported the improvement: four of fifteen quantitative papers in American Sociological Review in 2020, eight of twenty-five in 2023, twelve of eighteen in 2025.
The relationship between deposit and correctness has never been tested in sociology. He knows this. In February 2022, addressing the Scholars Strategy Network, he put it as plainly as anyone could ask.
“It’s hard to empirically prove this, but I really believe it. Even if people do not read your data and code and validate everything you do, the fact that it’s available says a lot.”
That is a belief held without evidence, by a man whose entire public identity is the refusal to hold beliefs without evidence, stated to a professional audience, without apparent discomfort. Turner’s model explains the absence of discomfort. The belief is not a conclusion Cohen reached. It is the coordination device of the coalition he belongs to, the open science community, which cannot function without it, and which supplies him with collaborators, a platform, a library dean’s budget line, a European prize, and an identity.
And the evidence is available and points somewhere awkward. The American Journal of Political Science has required third-party verification since March 2015, sending every accepted quantitative manuscript to the Odum Institute to be run before publication. Of the first 127 manuscripts processed, eight reproduced all analyses on the first attempt. Six percent. Authors operating under certain inspection, on a known timetable, with publication contingent on it, submitted packages that failed ninety-four percent of the time.
That number is public, it is in a neighboring discipline, and Cohen has never engaged with it as far as I can find. Under his own standard, a man who counts things is obliged to ask what his count measures. Under Turner’s, the question does not get asked because asking it is inconvenient in the precise sense: it would require telling the coalition that its central instrument may be measuring documentation rather than reliability.
The second convenient belief is about his adversary.
Cohen and Brad Wilcox have argued for a decade about selection into marriage, using overlapping data, with comparable training and comparable institutional standing. Nothing has moved. Cohen’s explanation, stated in his post titles since at least June 2025, is that Wilcox lies.
Consider the alternative. A competent demographer with a different formation looks at the same evidence, weights selection differently, holds a different prior about what marriage does, and reaches a different conclusion in good faith. If that is what is happening, then two things follow that Cohen cannot afford. Demography does not adjudicate its own disputes, which undercuts the authority his transparency program depends on. And the boundary between correct method and correct politics, which his career rests on keeping distinct, is not as clean as he needs it to be.
The lying hypothesis costs him nothing and explains everything. Turner’s phrase for what is happening is the self-justification of actions to others: the belief that makes a man’s conduct explicable to the people he interacts with, rather than to the people he does not. Cohen’s peers are family sociologists and open science people. Wilcox’s are at the Institute for Family Studies and in conservative family policy. Neither man needs to persuade the other’s coalition, and each has a fully sufficient account of the other’s persistence that requires no revision of his own.
The third one is the biggest and Cohen states it himself without noticing what he has said.
In February 2025 at Wisconsin, he raised the question of the exchange rate between different types of capital and correctly attributed it to Bourdieu. Do I get to write an op-ed about Trump because I have written articles about divorce rates? He added that he could be wrong about it and was happy to argue either way.
That is the question, and having asked it he moved on. His answer, in Citizen Scholar, is that a man occupies separable roles and that the trouble starts when academic authority carries him from one to another without the audience noticing. Announce the crossing and the problem is managed.
The convenient part is the word manageable. If disclosure works, then a professor can hold his professional standing, his political commitments, and his public platform simultaneously without any of them contaminating the others, and the only obligation is a sentence at the front. That is a belief that permits everything he wants to do while costing him a sentence.
Test it against April 3, 2020. He tells his enrolled students twice that he is not an expert on pandemics or public health. Twenty-three minutes later he tells the same students not to listen to the president. The disclosure was made. It was accurate. And a nineteen-year-old holding a syllabus has no procedure for discounting the instruction by the amount of the disclaimer, because to do that he would need to know how far the professor’s competence extends, which is what he was consulting the professor about.
Cohen is not doing something cynical there. He is doing what his role permits, under a belief about roles that permits it, formed inside a professional community where that belief is the ordinary furniture.
Now Turner’s harder point. The Jewish intellectual in Boudon’s example who finds communism convenient grew up up in a milieu where leftist politics is the air, and the belief arrives long before any choice, and he reconstructs it afterward as principled reasoning.
Cohen’s formation is documented by Cohen. A mother running a funded laboratory at Cornell for ten years who never received a tenure-track offer, which made institutional injustice a household fact rather than a proposition. That same mother arguing in public that the nervous system was less hierarchical than male scientists had assumed and that this bore on how societies should be arranged. An alternative high school built on participatory education. The Ithaca left of the 1980s. A newsroom where the whole job was handing a story to strangers on a fixed clock.
Set that beside the convictions of the man at fifty-eight. Prestige hierarchies are suspect. Credentialing is a guild move dressed as epistemology. Institutions protect revenue and call it standards. Work should be handed to strangers.
Turner’s claim is that these beliefs were likely installed by formation, and that he has spent a career constructing the evidentiary architecture that makes them feel like conclusions. The counts are real. The arithmetic is correct. And the arithmetic was recruited to serve commitments that arrived first, which is the ordinary condition and not a scandal.
Cohen experiences his positions as reasoned because the buffered model of selfhood is what a modern intellectual has available. He can inspect a belief formed last month, which is why the divorce-search chart got its clause. He cannot inspect a belief that arrived through the walls before there was a self to inspect with.
Every essay written about Cohen by people who dislike him alleges a double standard: he audits others and not himself. Turner’s model says the unaudited beliefs are unauditable from the inside. What Cohen has done, which almost nobody does, is build machinery that lets other people audit him, and then handed it to them: the code posted, the methods called squishy, the corrections invited on the back of a fifty-dollar bill, the annual public check of his own divorce prediction.
That is the honest reading of his career under this frame. He cannot see his own convenient beliefs. Neither can anybody. What he has done instead is externalize the audit, which is the only available workaround for a creature that cannot inspect its own formation.
And then the 2026 problem.
The submission that arrived at his archive in the winter of 2025 satisfied every stated requirement. Real citations, competent models, public code, a full disclosure of the tools used. The policy his committee published on March 9, 2026 refuses automated detection, declines to state a rule that could be written down completely, commits the moderators to human judgment, and says: we feel the need to express our humanity in this process.
Read that as a coordination device rather than as a standard and it becomes legible. What the archive discovered is that its stated rules had never been doing the work. The work was being done by a coalition of trained people who could recognize their own kind, and the machine exposed the gap by satisfying the rules while belonging to nobody.
The convenient belief that collapsed in 2026 was the belief that the criteria could be stated. Cohen’s whole program had rested on it, and when it failed he did the thing Turner would predict: he asserted membership.
Is the transparency movement a discovery about how knowledge works, or is it the coordination device of a coalition that happens to find it easy to meet? Quantitative researchers with public survey data can comply for the price of a day’s file management. Ethnographers cannot comply at any price, and the American Journal of Political Science had accepted no qualitative manuscripts in the three years after its policy took effect.
Cohen conceded the premise underneath that question in February 2022, at the end of a long session, to a friendly questioner. Just because you read my transcripts doesn’t mean you can analyze my data. The data is me. I was there.
He said it about someone else’s method. Turner’s point is that it is true of his own, and that he will not be the one to notice.
Notes
Frame. Stephen Turner, “Boudon on Tocqueville” (2025), pp. 295–298, for the passage on what is convenient to believe as the product of multiple situational constraints, and for the Boudon example of the Jewish intellectual and communism as structural positioning. Turner’s related treatment of field theories, where location in a social field determines what people find convenient to believe and do, including the inconvenience of explaining why one rejects what everyone else appears to believe. Charles Taylor (b. November 5, 1931) on the buffered and porous self, from A Secular Age (2007), supplies the psychological half.
Outside evidence. The AJPS verification figures are from the journal’s own policy document of February 27, 2018: 149 manuscripts processed, 127 completed, eight reproducing all analyses on first attempt, and the note that no qualitative manuscripts had been accepted since the policy took effect. That last fact is a single journal over three years and has other possible explanations, including submission mix; it belongs in the essay as a datum rather than as a finding.
The claims that are mine. Three. That the divorce-search clause of October 2020 is a convenient belief caught in formation, and that its isolation in the record is the evidence for Turner rather than against him. That the lying hypothesis about Wilcox is convenient because the alternative, honest disagreement between competent demographers, threatens the authority his program requires. And that the March 2026 policy is the collapse of the belief that criteria can be stated, rather than a policy about machines.
The exposure, and it is the same one every time this frame is used. Convenient-belief explanations absorb their own counterexamples. Cohen produced a finding his coalition did not want, attacked an ethnographer whose politics matched his, conceded to a National Academies committee that his central claim about generation labels was an intuition with no research behind it, and narrated his own bias on a slide. The available reply is that his coalition is not progressives but sociologists who check things, and that all four acts were loyalty to the smaller group. That reply works, and it works too well: redescribe the coalition finely enough and any behavior becomes coalition maintenance. Say so in the essay rather than letting a reader find it. The frame’s value here is that it is being applied to a man who supplies the evidence against himself, which is the only condition under which it can be tested at all.
‘Mistakes at Work‘ (1951)
One article has been retracted from a journal of the American Sociological Association since the year 2000.
Philip Cohen turned that up in December 2025, going through Volume 90 of American Sociological Review article by article ahead of a National Academies panel on corrections and retractions. He reported it as a scandal, alongside the fact that the association’s 2018 guide for editors does not contain the word retraction. The one case involved documented fraud by a Dutch psychologist, which is to say it involved a man outside the discipline whose misconduct had already been established elsewhere.
Twenty-five years. Thousands of articles. One withdrawal, and that one imported.
Everett C. Hughes (1897–1983) would have said that the number is what a working profession looks like from the outside, and the thing to explain is why anybody expected it to be higher.
Hughes’s essay “Mistakes at Work,” begins from a fact he thought sociologists had been too polite to state. Every occupation has a rate of error and a rate of failure built into the work, and the people in it know the rate. The plumber floods a basement. The surgeon loses a patient who might have lived. The teacher writes off a child who could have been reached. What varies is not whether mistakes occur but who bears them and who judges them.
From that Hughes derives the arrangement he thought was the deep structure of professional life. Since error is inevitable and its costs fall on clients rather than on practitioners, the group needs some way to manage the risk of exposure. It cannot promise no mistakes. So it makes a bargain: the members will judge each other’s errors, seriously and by their own standards, on the condition that the judging happens inside the group. Guilt is collectivized. The occupation absorbs the failure rather than surrendering a member to lay judgment, and in exchange it accepts responsibility for policing what it will tolerate.
Hughes’s example was medicine, and the institution he pointed at survives today. The morbidity and mortality conference is a room in which surgeons state out loud that they chose the wrong approach and the patient died. What is said there does not leave. In most American jurisdictions it is protected by statute from discovery in litigation. The protection is the point: without it, nobody would say the true thing, and the group would lose its only route to learning from what it did wrong.
Sociology’s version of that room is the one Cohen has spent a decade trying to open.
The discipline does not retract. It also does not, in Hughes’s terms, do nothing. It reviews manuscripts, rejects most of them, staffs editorial boards, awards and withholds prizes, hires and declines to hire, and passes judgment continuously through channels no outsider can see. A paper that everyone in a subfield knows to be weak does not get retracted. It gets not cited. That is a sanction, applied by peers, invisible from outside, and Hughes would say it is the characteristic form professional discipline takes.
Hughes’s other great contribution is the pair of terms licence and mandate. A profession claims a licence: permission to do things others may not do, to open a body, to hear a confession, to ask a stranger about his marriage and his income and write it down. And it claims a mandate, which is the further and larger claim: the right to define what proper conduct in its area consists of, for everyone, including the clients and the public.
Cohen’s whole career depends on the licence and attacks the mandate.
Everything he does presupposes the licence. He works inside federal survey instruments the public cannot access in raw form, using training the public does not have, drawing inferences from age-standardized rates that a lay reader cannot evaluate. He told his own students in August 2020 that most sociologists do not conduct surveys, they use the ones other people made. He is authorized to speak about American families because a discipline certified him, and he has never proposed that the certification be opened.
And then he attacks the mandate at every opportunity. The discipline should not decide for itself what counts as adequate reporting. Journals should not decide what a reader is entitled to inspect. The association should not decide what its publishing arrangements owe its members. Editors who agreed that his comment identified errors and declined to correct the record had, in his account, no standing to make that call.
Hughes’s reply, and it is the strongest argument anyone has made for the other side of this fight, is that licence and mandate are one thing. A group that is permitted to do what others may not must be the group that determines what doing it well consists of, because nobody else knows. The discipline that trains a demographer to see that a household roster is an artifact is the only body that can say whether a given demographer has handled the artifact competently. Hand that judgment to a public that cannot make it and you have not democratized the standard. You have destroyed it and put nothing in its place.
The last five years supply an argument for Hughes.
Cohen’s transparency program made a promise: extract the discipline’s tacit standards into stated rules, publish the rules, and let anyone check compliance. Deposit the code, deposit the data. His counts show the compliance rising, from four of fifteen quantitative papers in 2020 to twelve of eighteen in 2025.
Then the machines arrived at his archive. A submission that satisfied every stated rule, real citations, competent models, code in a public repository, a full disclosure of tools. And the policy his own committee published on March 9, 2026 abandoned the project. It refuses automated detection, declines to state a rule that could be written down completely, and commits the moderators to judgment made by people using perception and experience. The sentence everyone quotes is the giveaway: we feel the need to express our humanity in this process.
That is a group of five trained volunteers asserting that they will recognize competent work and cannot explain how. It is the mandate, restated, by the man who spent ten years arguing that the mandate should be replaced with rules.
Now the case against Hughes.
Hughes’s bargain has a condition. The profession collectivizes guilt in exchange for policing itself. Absorb the errors, and judge them. Sociology’s record on the second half is where the argument turns.
Consider Cohen’s Goffman comment. In June 2015 he did arithmetic on a household survey reported in American Sociological Review: 217 homes on a Philadelphia block, 308 men aged eighteen to thirty found in residence, against census figures for comparable neighborhoods that give less than half that many. He submitted it as a comment. The journal declined it in August 2015. His account, uncontested, is that the editors agreed he had identified errors and chose not to correct the record.
An error acknowledged inside the group and left standing in the literature is not collectivized guilt; it is guilt evaporated. The morbidity and mortality conference protects the discussion because the discussion happens. Sociology has claimed the protection without holding the meeting.
And here Hughes’s own account supplies the test. His argument for internal judgment turns on the claim that only the group can assess the work. That justification lapses exactly where the assessment does not require professional judgment. Anyone can compare 308 against a census figure. The arithmetic is not a licenced act. When a profession invokes its mandate to protect a claim that a stranger could check with a calculator, the bargain has stopped protecting the work and started protecting the workers, and Hughes gives you no ground to stand on there.
That is the line. The bargain covers judgments that require the training. It does not cover arithmetic, and it does not cover a decision to leave an acknowledged error in the record. The first is a defence of competence. The second is a trade association.
Cohen may be doing Hughes’s work in a form Hughes did not anticipate, while describing it in terms that get it wrong.
Look again at what moved. In December 2025 an author he had marked as providing nothing appeared in his comment thread, agreed with him, promised a replication package when teaching quieted down, and thanked him. That is not a man complying with a rule enforced by an outside public. That is a member of a group responding to a peer’s judgment, delivered in a venue the group reads, on a standard the group has come to accept.
The counts function as a morbidity and mortality conference conducted in the open. The participants are all sociologists. The public does not read the December audits, and the six papers with nothing attached were not identified to a lay audience that could act on the information. What changed was that a professional standard acquired a countable form, and countable things can be applied to members who resist tacit pressure.
Cohen calls this accountability to the public. Hughes would call it the mandate operating through a new instrument, and note that the practitioner who conceded in the comment thread conceded to a colleague, not to a citizen.
If that is right, then the deepest convenience in his position is the belief that he stands outside the guild. He resigned his membership in November 2021. He kept the licence. And the campaigns that worked, the counts and the Pew letter with its hundred and fifty signatures, worked the way professional discipline has always worked, by a group telling an institution that people trained as its members are trained find something unacceptable.
The campaign that failed is the one aimed at a rival profession’s mandate. Ten years arguing with Brad Wilcox, whose authority is licenced by a different community with a different standard, and nothing has moved, because there is no shared bargain to appeal to.
In February 2022 Cohen conceded the ethnographers’ central premise to a friendly questioner at the end of a long session. Just because you read my transcripts doesn’t mean you can analyze my data. The data is me. I was there.
That is the licence, stated in one sentence, by the man most identified with the attack on the mandate. He said it about somebody else’s method. Hughes would say it is true of every method, including the ones that ship their code, and that the profession which forgets it will find that the rules it published can be satisfied by something that never trained at all.
Notes
Frame. Everett C. Hughes, “Mistakes at Work,” Canadian Journal of Economics and Political Science 17, no. 3 (August 1951): 320–327, reprinted in Men and Their Work (Free Press, 1958) and in The Sociological Eye (Aldine, 1971). The licence and mandate formulation is in “Licence and Mandate,” collected in The Sociological Eye, and developed in “Professions,” Daedalus 92, no. 4 (Fall 1963): 655–668. Two things to verify before printing. I have described the collectivization of guilt as Hughes’s own formulation and the term is used in the secondary literature on him as well; check the wording against the 1951 text. And I have not quoted him directly anywhere in this essay, which is deliberate, since I do not have the text in front of me and would rather paraphrase accurately than quote loosely. If you want him quoted, the plumber and the surgeon appear in the opening pages and the passage is short enough to transcribe.
The morbidity and mortality conference is Hughes’s example and also a live institution; peer review protection statutes vary by state and I have described the norm rather than the law.
The claims that are mine. Four. That the single retraction is evidence of a functioning bargain rather than of decay, which is Hughes applied and which Cohen would reject outright. That Cohen depends on the licence while attacking the mandate, and has never proposed opening the licence. That the March 2026 archive policy is the mandate reasserting itself under his own signature. And, the most interesting one, that his audits are the mandate operating through a new instrument rather than accountability to a public, on the evidence that the man who conceded conceded to a peer.
Hughes’s account justifies almost any professional self-protection unless you specify where the bargain stops. I have located the stopping point in his own justification: internal judgment is defensible because only trained members can assess trained work, so it lapses where the assessment requires no training.
What is missing. Nobody has asked Cohen whether he thinks the discipline should retract more or correct more, which are different remedies with different costs.
The Andrew Abbott Frame
In February 2025, at a workshop in the sociology department at Wisconsin, Philip Cohen told the room what had happened to his research.
“In my research I’ve taken a pretty descriptive turn. I’m less about trying to discover new theories and mechanisms, and more about trying to figure out what’s happening, and find ways to communicate about that to other scholars and various publics.”
He knew what it cost him.
“If you have a descriptive finding it’s considered to be less than, in a lot of academic circles, and so we end up doing more puffery just to make our work sound like it’s not just descriptive. Very off-putting to the public, in my opinion.”
Thirty-seven years earlier, Andrew Abbott (b. 1948) published an essay in Sociological Theory called “Transcending General Linear Reality.” Its argument was that sociology had confused a statistical technique with a picture of how social life works. The general linear model assumes that the world consists of fixed entities with attributes, that those attributes have independent and additive effects, that causal order does not matter, that a case has one meaning irrespective of its history, and that time enters only as a variable rather than as a sequence. None of that is true about people, and sociologists had begun writing as though it were, because it was what their machinery could represent. Abbott’s remedy was to recover narrative, sequence, and description as forms of knowledge in their own right rather than as preliminaries to the real work.
Cohen arrived at the same place from inside a career of running the models, and did not cite him. That is not a criticism. It is the interesting fact, and the reason to run both Abbotts at once, because Abbott’s other book explains why nobody claps when a demographer says he would rather describe.
The System of Professions (1988) argues that professions compete for jurisdiction: the right to define and treat a category of human trouble. The contests run in three arenas at once, and the arenas have different rules. In the workplace, jurisdiction gets settled by who does the work when nobody is watching. In public opinion, by who the public believes owns the problem. In the legal and administrative system, by who the state authorizes.
And the engine of a jurisdictional claim, in Abbott’s account, is abstraction. A profession survives by taking a problem stated in ordinary language and restating it in a system of its own, because the restatement is what only members can perform. Medicine takes a pain and makes it a diagnosis. Law takes a grievance and makes it a cause of action. The abstraction is not decoration on the expertise. It is the expertise, in the sense that matters for holding territory, because it makes the problem unintelligible to anybody outside.
Regression is sociology’s abstraction system. That is the thought that joins the two books.
When a family sociologist takes the question of why poor Americans marry less and restates it as an equation with covariates, controls, and an adjusted coefficient, he has done the jurisdictional work. The question came in as something a pastor, a novelist, a caseworker, or a mother could speak about. It goes out as something only a trained person can evaluate. Everything downstream, the standing to publish, to advise, to testify, to be quoted, rests on the conversion.
Now read Cohen’s career as a jurisdictional contest.
In the workplace arena, his claim is that the people doing sociology should be required to show their work. That is where the counts operate: Volume 90 of American Sociological Review gone through article by article, twelve of eighteen quantitative papers arriving with code and six with nothing, against four of fifteen in 2020. Nobody outside the profession reads those counts. They are a claim about who sets the terms of competent practice among practitioners.
In the public arena, his claim runs through the blog, the textbook, the media appearances, the lectures uploaded to a channel where the introductory Social Problems lecture has been watched twenty-four thousand times. The claim is that a family sociologist should be the person the public consults about American families, ahead of the think tanks and the trend writers.
And in the legal arena, he became a plaintiff. Knight First Amendment Institute v. Trump, filed in the Southern District of New York in July 2017, won at the district court in May 2018, affirmed by the Second Circuit in July 2019. He also testified to a Senate committee in 2007 and advised the Census Bureau on the enumeration of same-sex married couples after the 2010 count.
Abbott’s three arenas, all three occupied, by one man.
The competitors are visible too, and Abbott’s model tells you why the fights take the shape they do. The success sequence dispute is a jurisdictional contest with the Institute for Family Studies over who owns the problem of family instability. Cohen’s move is the classic defensive one: he attacks the rival’s abstraction rather than its conclusion. The success sequence, he argues, converts a description of who is not poor into an instruction for people who are, which is a claim that the opposing profession’s conversion is invalid. That is the argument a jurisdiction makes when a competitor reaches into its territory with an inferior method.
The generations campaign is a jurisdictional contest with marketing and journalism, and Cohen said so at the National Academies in July 2019 without using the word. The labels were never peer reviewed, he told the committee, because ridiculous marketing people made them up and everyone started using them, and they were never real categories in the way that scientific categories are. Then two years later, a letter with roughly a hundred and fifty signatures to the Pew Research Center, and in May 2023 Pew changed its policy. That is a profession asserting jurisdiction over a classification scheme, successfully, against an occupation with weaker abstraction.
The Goffman dispute is the internal case, and it is a boundary fight between two abstraction systems inside one profession. Her block is a place a person went and stayed. His census tract is a state simplification. Neither converts into the other. He conceded exactly this in February 2022, at the end of a long session, to a friendly questioner: just because you read my transcripts doesn’t mean you can analyze my data, the data is me, I was there.
Which brings the two Abbotts together at the point that makes the essay.
Cohen’s transparency program is a jurisdictional claim that runs against the ordinary direction. Every profession’s instinct, on Abbott’s account, is to raise the cost of evaluation, because unevaluable work is defensible work. Cohen proposes lowering it. Post the code. Deposit the data. Let a stranger recompute the number. That is a demand that his own profession make its abstraction system inspectable from outside, which is the one thing an abstraction system is built to prevent.
And the descriptive turn is the same move at the level of method. Description does not abstract. When he tells a lecture room that the bottom tenth of Americans sits above half the world’s population in income, or that the average American could expect twenty-nine of the years between eighteen and fifty-five in marriage in 1960 and eighteen now, he has handed over a fact a person can hold, compare against his own life, and dispute. No conversion has occurred. Nothing in the sentence requires a licenced reader.
So the two Abbotts turn out to describe one program, and the program is a partial jurisdictional disarmament. Open the certification, and simplify the output until a layman can check it. Cohen presents this as good practice. In Abbott’s terms it is a profession’s member proposing to give away the two things that hold its territory.
That explains the reception, which is otherwise puzzling. The transparency program has been fought at every turn by an association he served for years, whose Council rejected a policy that would have done nothing more than tell readers whether a published article came with its data. The descriptive turn, by his own account, is treated as lesser and pushes authors into puffery. Neither of those is about laziness or about tone. They are what a profession does when a member proposes to lower the walls.
Two objections, and both are serious.
The first is that Abbott’s model is silent on whether a jurisdiction deserves the territory it holds. Everything in it works the same way whether the profession’s abstraction tracks reality or merely mystifies it, which means the frame cannot tell you that Cohen is right and it cannot tell you that he is wrong. It tells you what kind of fight he is in. A reader who wants to know whether replication packages improve sociology will not get it here.
The second objection is Abbott against Abbott, and Cohen is the exhibit. If regression is the abstraction system that holds the jurisdiction, then abandoning it should cost a sociologist his standing. It has not cost Cohen his. He gave up mechanisms for description and remains a full professor at a flagship, with a textbook, a prize, and a large audience. Either the jurisdiction is held by something other than the abstraction, credentials, institutional position, and the sheer fact of employment, or Cohen has found a route Abbott’s model does not anticipate: reaching the public directly with unconverted facts, and converting that reach back into professional standing. The second reading is more interesting and it is available. The twenty-four thousand views on an introductory lecture are a jurisdictional asset of a kind that did not exist when The System of Professions was written.
What survives is a claim about him that nobody has made, and it is the reason to write this rather than a Hughes essay or a Turner essay.
Cohen’s two campaigns look unrelated. One is about repositories and code and journal contracts. The other is about how he writes his own papers. They are the same campaign, and it is a campaign against the conditions of his own authority. Open the box and simplify the contents, and the demographer becomes a man with a chart that anybody can argue with.
He has never said this, and I doubt he thinks of it this way. He describes the descriptive turn as a preference and the transparency work as ethics.
Abbott would say a profession that took both seriously would have to find some other reason to exist, and that the reason is not obvious.
Notes
Frames. Andrew Abbott, The System of Professions: An Essay on the Division of Expert Labor (University of Chicago Press, 1988), for jurisdiction, the three arenas of workplace, public, and legal claim, and the argument that abstraction is what enables a profession to seize and hold territory. And Abbott, “Transcending General Linear Reality,” Sociological Theory 6, no. 2 (Autumn 1988): 169–186, for the critique of regression as an implicit ontology and the case for recovering narrative and sequence. He develops the positive program in Time Matters: On Theory and Method (2001), which is where you would go if you extended this.
The joint argument is mine. Abbott does not, as far as I know, connect the two books by making regression the abstraction system that secures sociology’s jurisdiction. He treats general linear reality as a methodological error and jurisdiction as a structural account of professional competition. Welding them is the essay’s contribution and should be defended as an extension rather than presented as exegesis. If it is wrong, it is wrong because the profession’s jurisdiction may rest on data access and credentialing rather than on technique, which is the second objection in the text.
Where this is exposed. The second objection is the real one, and I have given it four sentences rather than burying it. If a sociologist can abandon the abstraction system and keep his standing, then the abstraction is not doing the work Abbott says it does, at least not in 2025. A hostile reader will go there first. The available reply is that Cohen abandoned mechanisms while keeping every other feature of the licence, including the training, the data access, and the credential, and that the descriptive turn is therefore a smaller disarmament than it looks. That reply is good enough to run and not good enough to settle it.
What would settle it. Abbott is emeritus at Chicago and has spent thirty-five years arguing that description is undervalued. Cohen announced a descriptive turn in 2025 without citing him. One email to Abbott asking whether he regards the open science movement as a jurisdictional claim, and whether he sees the descriptive turn in family demography as vindication or as something else, would produce the first new fact in anything I have written for you on this man.
The Legibility Frame
In a lecture on research methods recorded on August 29, 2020, Philip Cohen told his introductory students something about himself that he does not put in his books.
“Most sociologists are not actually doing those surveys. We’re using the surveys that other people have done. I do a lot of research using surveys. I don’t conduct a lot of surveys. I conduct a few, but mostly I’m getting the data that someone else has collected and analyzing it.”
The someone else is the United States government. The American Community Survey, the Current Population Survey, the vital statistics systems of the states, the National Crime Victimization Survey, the decennial census. Cohen worked inside two of those machines himself before he had a faculty job, as a research assistant at the National Center for Health Statistics in 1997 and as a statistician and demographer at the Census Bureau from 1998 to 1999.
James C. Scott (1936–2024) built Seeing Like a State (1998) around the observation that a modern state cannot govern what it cannot read, and that reading requires simplification. A forest containing hundreds of species used by villagers in dozens of ways becomes, on the state’s ledger, board feet of timber. A tangle of customary land rights becomes a cadastral map with owners and parcels. People with one name in the village and another in the family become permanent surnames on a tax roll. None of these simplifications describe the thing. Each makes the thing administrable, and the state then acts on the simplification rather than on the thing, sometimes with results the simplification could not anticipate.
Scott’s counter-term is metis, borrowed from the Greeks: practical knowledge acquired by doing, local to a place and a situation, held in a person, and resistant to being written down. The pilot who knows this harbor. The farmer who knows this field. Nobody can hand it over in a manual, which is why it survives outside the state’s instruments and why the state, which cannot see it, keeps designing around it and failing.
Cohen is not a critic of legibility. He is one of its skilled operators, and he is unusual among them in knowing exactly what the instruments do.
That is his signature move, repeated across thirty years. Before you explain a trend, establish that the instrument did not change. His early paper with Lynne Casper reconstructed historical rates of unmarried cohabitation from census data, which meant first asking whether the census had been counting the same thing across the period. His generation labels campaign is an argument that the categories are administrative fictions with arbitrary boundaries, and that the only cohort the Census Bureau ever officially designated was the baby boom, because a birth-rate spike in 1946 that held until 1964 was a real demographic event. His complaint about same-sex marriage statistics, which he has made in the same form in 2020 and again in 2025, is that most states record only spouse one and spouse two on a marriage certificate, so the administrative record cannot answer the question. And his most durable pandemic call, made on April 6, 2020 and repeated twice that year, was about an instrument going dark: reports of child abuse were about to fall because teachers and clergy and clinicians are how the state sees inside a home, and the schools had closed.
That last one is Scott’s thesis in a sentence, delivered by a demographer to undergraduates, six weeks before the data started coming in wrong.
So the frame does not make Cohen a villain. It makes him something more interesting: a man who works entirely inside the state’s simplifications and spends his professional life auditing what the simplifications do to the world they claim to represent. Legibility with a conscience.
Then in June 2015 he ran into metis and did not recognize it.
Alice Goffman’s On the Run reported an ethnography of a poor Black Philadelphia neighborhood, six years of presence, and inside it a household survey conducted with a man she calls Chuck. Her 2009 article in American Sociological Review described 217 homes on the block, 308 men aged eighteen to thirty found in residence, and 144 of them carrying warrants.
Cohen did the arithmetic against census figures for comparable Philadelphia neighborhoods and found more than twice as many young men as those homes should hold. He noted that her residence rule, three days a week, could place the same man in two households. The objection is correct, it is checkable, and any competent demographer would raise it.
Goffman’s answer was that this was an ethnographic survey, with its own sampling and its own definition of who belongs to a household, designed to reach people whom conventional surveys never find.
Read that as a defence and it sounds like evasion, which is how it has generally been read. Read it as a claim about metis and it becomes something else: an assertion that on that block, membership in a household was a thing you could only know by being there, that the state’s rule for who counts as resident does not describe how those men lived, and that a person who had spent six years watching them come and go knew something the roster could not hold.
She may still be wrong about the number. Being present does not make a count accurate, and a person with metis can misreport it. But the dispute was never settled, and Scott explains why it could not be. Cohen’s census tract and Goffman’s block are two knowledge systems with no exchange rate. His can be checked by a stranger and cannot see the men who move between houses. Hers can see them and cannot be checked by anybody who was not there. There is no third position from which to adjudicate, which is exactly Scott’s point about what happens when the map meets the territory.
The most striking thing in the record is that Cohen came to this himself, seven years later, and it has never been quoted.
In February 2022, at the end of a long online workshop for the Scholars Strategy Network, a friendly questioner asked how open science applies to qualitative research. He answered:
“Just because you read my transcripts doesn’t mean you can analyze my data. The data is me. I was there, it’s the experience. There’s no such thing as reproducibility, because you couldn’t have the interaction and the experience that I had as me at that moment. So reproducibility is not really the issue. But maybe something like accountability is. You still might have lied in what you reported, or been wrong.”
The data is me. That is a definition of metis, offered voluntarily, by the man whose census arithmetic is the reason a generation of ethnographers will not cite him.
And what he proposed next is the most constructive thing anyone has said about that fight. A restricted tier, where a small number of vetted readers could see field notes and transcripts under agreement, on the model of the Census research data centers where confidential quantitative data is examined under conditions. Not reproducibility, which is impossible. Not trust me, which is what the discipline had. Something between them.
Nobody built it. He said it was not his area and that it would be neat, and could not recall the name of the Qualitative Data Repository at Syracuse until a listener supplied it.
Now the objections, and there are two that matter.
Scott’s framework was built to explain the failures of high modernist state projects: forced villagization, collectivized agriculture, planned cities, schemes in which the simplification was imposed on people by force and the metis it destroyed was the metis of the people being planned. Cohen imposes nothing on anybody. He is a professor with a blog who counts things. Stretching Scott to cover an argument between two academics about a household survey risks making every methodological dispute look like Tanzanian villagization, which would be a misuse of a serious book.
The harder objection runs the other way, and it is the one Scott’s critics have always pressed. Metis is not automatically superior knowledge. It is local, and local knowledge is frequently wrong about anything beyond the locality. The person who was there has powerful reasons to believe his own account and no external check on it. Scott is not sentimental about this, but the book’s rhetorical energy runs toward the villagers, and an essay applying it has to say plainly that Cohen’s arithmetic was legitimate and that the legitimacy is the whole reason the episode matters.
Both parties in that dispute were doing something defensible. He was checking a number, which is the only thing a stranger can do. She was reporting what a place looked like from inside it, which is the only thing a stranger cannot do. The failure was institutional: American Sociological Review declined his comment in August 2015 while agreeing, on his uncontested account, that he had identified errors, so the discipline neither corrected the count nor articulated why the count might be beside the point. Ten years on it still has no procedure for a fight between two knowledge systems, which is why the same argument reruns every time an ethnography wins a prize.
Which leaves the ending, and it is the reason to run this frame rather than another.
In the winter of 2025 Cohen’s archive began receiving papers with everything the state’s instruments require. Abstract, real citations, competent models, public code, a full disclosure of the tools used to generate it. Perfectly legible submissions, produced by something that had never been anywhere.
The policy his committee published on March 9, 2026 refuses automated detection, declines to state a rule that could be written down completely, and commits the moderators to judgment made by people using perception and experience. Five volunteers, reading by hand, recognizing something they cannot specify.
That is metis, asserted as the last line of defence, by the man who spent a decade arguing that everything worth knowing about a paper could be written into a replication package.
He had already conceded the principle in 2022 and applied it only to other people’s methods. In 2026 it arrived at his own door.
Notes
Frame. James C. Scott, Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed (Yale, 1998), for legibility, state simplification, and the argument that administrative categories reshape what they measure. Metis is developed in chapter nine, drawing on Marcel Detienne and Jean-Pierre Vernant, Cunning Intelligence in Greek Culture and Society (1978).
Goffman’s side. Her “ethnographic survey” answer is as paraphrased by Jesse Singal. Paul Campos in the Chronicle lays out the same arithmetic Cohen did.
The claims that are mine. Three. That Cohen is a legibility agent who audits legibility, which is a more precise description than either his defenders or his critics use. That the Goffman dispute is a collision between two knowledge systems with no exchange rate rather than a dispute about honesty, and that this is why it never resolved. And that the March 2026 archive policy is metis asserted at the end of a career spent trying to make everything legible.
The failure mode. Scott’s book is about states destroying peasant livelihoods, and this essay applies it to a quarrel between professors. I have said so in the text rather than hoping nobody notices. The defence is that Cohen’s raw material genuinely is state simplification, that he says so himself, and that the argument is about what those simplifications can and cannot see. If a reader still thinks the scale mismatch is fatal, he is entitled to, and the essay should not pretend otherwise.
What is missing. The restricted-tier proposal Cohen made in 2022 is the one constructive idea to come out of a decade of that fight, and I cannot find that he or anyone else pursued it. The Qualitative Data Repository at Syracuse does something adjacent. Whether he knows that, and whether he thinks it answers the problem, is a question nobody has put to him and the answer would be worth a paragraph.
The Mary Douglas Frame
The paper had everything.
An abstract. A title that matched the contents. An ORCID for the author. Citations that were real, to work that existed, quoted accurately. Statistical models that were complex and, as far as the moderators could tell, competent, with tables and figures. Code in a public repository. And at the end, a disclosure listing the tools used to write the code, run the literature search, advise on the specification, and draft nearly all of the text, alongside the author’s statement that he had formulated the question, chosen the variables, directed the analysis, interpreted the results, verified every data claim and every citation, and would submit to an audit of his methodology on request.
Philip Cohen described it on February 22, 2026 as dull, unoriginal, thin in its literature review, and empty where it claimed a theoretical contribution. He also said, plainly, that SocArXiv would have accepted it a few years earlier.
Nothing on the checklist was missing. That is the fact to hold onto, because it is what makes the case worth thinking about at all.
Mary Douglas (1921–2007) opened Purity and Danger (1966) by refusing the explanation everyone expects. Dirt, she argued, is not a substance. Nothing is dirt in itself. Shoes are not dirty until they are on the dining table; food is not dirty until it is in the bedroom. Dirt is matter out of place, which means the idea of dirt presupposes a system of places, and where you find pollution you are looking at the boundary of a classification. The reaction to a polluting thing tells you what the classification exists to hold together.
Her second move is the one that bears here. Anomalies are not the same as bad specimens. A diseased animal is a poor member of its category. An anomaly is a thing the category system has no slot for, and Douglas catalogued the responses cultures make to them. The anomaly can be suppressed, or physically eliminated, or avoided, or declared dangerous, or labelled and confined. Or, in the rarer case, it can be made sacred, treated as the point at which the ordinary order shows its own limits and becomes the object of ritual. Her example was the pangolin among the Lele, an animal that had scales like a fish and climbed trees and bore single young like a human, and which the Lele, instead of treating as an abomination, made the center of a cult.
The response is the datum. What a group does with the thing that will not sort tells you what the sorting was for.
The submission that arrived at SocArXiv in the winter of 2025 was not a bad paper. Bad papers are the archive’s daily business and it accepts them, which Cohen has said in public more bluntly than any administrator needs to: there is a great deal of bad work on the platform, and it does not damage the good work. A bad paper is a poor specimen inside a category the system recognizes.
This was matter out of place. Every criterion satisfied and nothing in the category system corresponding to what was missing. Cohen spent two thousand words looking for the rule it violated and could not find one. He is not persuaded that using such tools is unethical, since all knowledge is learned from someone else and no norm requires citing everything one reads. He thinks the training may be lawful. He grants that an archive issuing identifiers without review is an attractive place to deposit. He will not enter a detection arms race. A disclosure rule he cannot verify is close to worthless, since the machines read the rules too.
Every route to a stateable objection closed, in public, in sequence. What was left was the certainty that the thing did not belong.
The archive’s response came on March 9, 2026, drafted by a subcommittee of five and revised by the full moderation team. It is a policing document in Douglas’s sense, and it is worth reading for what it declines to do as much as for what it does. It refuses automated detection. It declines to state a rule that could be written down completely. It commits the moderators to judgment made by people using perception and experience. It invites appeals. And it says: we feel the need to express our humanity in this process.
Douglas’s question is what the classification was protecting, and the sentence answers it.
Not correctness. The archive does not check correctness and never has, and Cohen says so. Not competence, since the submission was competent by every visible measure. Not the criteria, since the policy abandons the attempt to state them.
What is being protected is membership. The archive turns out to have been sorting for scholarship, and scholarship turns out to have meant work produced by a person who belongs to a community that produces work, and the identifiers, the abstracts, the citations, the code were never the thing. They were how the thing showed up. Something that could produce the showing-up without the belonging revealed that the whole apparatus had been reading indices of membership and calling them standards.
That is a discovery, and it was made by the person with the strongest interest in not making it.
The reason Douglas is the right instrument rather than a decoration on the audit-society reading is that she explains something the audit reading cannot. Michael Power’s account tells you that the archive was checking documentation rather than quality, and that documentation becomes worthless once documentation is free to produce. That is correct and it stops one step short. It cannot tell you why the response took the form it did. A pure audit failure would be met with better audit: tighter criteria, a detection tool, a verification step. Cohen considered all three and rejected all three.
What his committee wrote instead was a boundary rite. Five named people will look at submissions and know their own kind, on grounds they cannot articulate, and the policy says so in the register of a group asserting itself rather than in the register of a rule. Douglas would say that is what happens when a classification meets an anomaly it cannot suppress by refinement, and that the appeal to humanity is not sentiment but jurisdiction: a statement about who is inside.
Two things complicate the reading, and both belong in the essay.
The first is that Douglas has two responses available and the archive took only one. Nothing in the policy sacralizes anything. There is no move toward treating the machine-written paper as the thing that reveals the limits of the system, no ritual, no special category. It is policed and refused, which is the ordinary response and the less interesting one. If you want the sacralizing case, it is happening elsewhere in the culture, in the elaborate public fascination with what these systems can and cannot do, which has the shape of a cult around an anomaly. The archive is doing the mundane work of the boundary.
The second complication is the one that should keep the essay honest. Douglas’s framework has a well-known problem: it can render any prohibition as boundary maintenance, and once you have said that a rule protects a classification you have said something that fits every rule ever made. Applied loosely it explains dietary law, quarantine, immigration policy, and a preprint moderation queue with equal facility, which means it is not discriminating between them. The discipline it requires is to specify what the classification was protecting in this particular case, and to be able to say what would have counted as evidence against that reading.
Here is what would have counted. If the archive had responded by tightening its stated criteria, or by adopting a detection tool, or by requiring a verifiable disclosure, the anomaly would have been absorbed as a bad specimen and the classification would have held. Douglas would have been the wrong frame. The archive did none of those, considered and rejected each in writing, and reached instead for a claim about who its people are. That sequence is what makes the reading load-bearing rather than decorative.
There is one more thing, and it is the reason to write this rather than another essay about the same three documents.
Cohen spent a decade arguing that the discipline’s standards should be written down so that a stranger could apply them. The transparency counts, the replication packages, the audits every December, the whole program rests on the proposition that what a competent sociologist knows can be converted into criteria and the criteria checked by anyone.
The submission that broke it was not a violation of the criteria. It was the criteria satisfied by something outside the community, which is the only way to find out whether the criteria were ever the thing.
They were not. They were the visible residue of membership, and the residue can now be manufactured.
What the archive kept, at the end, is a room with five people in it who can tell.
Notes
Frame. Mary Douglas, Purity and Danger: An Analysis of Concepts of Pollution and Taboo (Routledge and Kegan Paul, 1966), for dirt as matter out of place, the argument that pollution beliefs mark the boundaries of a classification system, and the catalogue of responses to anomaly in chapter two, which she sets out as a list of options a culture has when confronted with something its categories cannot hold. The pangolin cult among the Lele is in chapter three and developed further in her Implicit Meanings (1975).
The claims that are mine. Three. That the submission was an anomaly, and that the distinction is what Power’s frame cannot capture. That the classification turns out to have been sorting for membership rather than for quality, and that the humanity sentence is the evidence. And that the sequence of rejections, detection considered and refused, criteria considered and abandoned, is what makes the reading testable.
Part Two
