Luke Ford

AI helped me with this post, which has gone through several versions.

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 and where the terms of women's employment in it were a live subject at the dinner table. 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 decades later, 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, and there she did 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 and gets corrected by feedback 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 on his blog 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. Someone did.

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 rather than commanded 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 on a weekly deadline. 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.

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 even if the world stands still.

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 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.

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.

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, and the book delivers. The same bonds that produce intimacy and obligation produce dependency, unequal labor, and stigma. His position is that sociology's job is to explain how the benefits and burdens get distributed.

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. The decline came from younger women. 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. He concluded that marriage is becoming rarer and more durable at once, and stable marriage is concentrating among people who already have the resources that make it stable. He then did the thing that distinguishes him: 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 make the case against his framing, that divorce has become a poorer measure of instability as cohabitation, serial cohabitation, and third marriages have risen, and that a falling divorce rate can be used to paint a rosy picture.

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 does not dispute the correlations. His 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 sharply 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.

His antagonist in this argument has been W. Bradford Wilcox (b. 1970) of the University of Virginia and the Institute for Family Studies, and the argument has not stayed methodological. 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.

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. Cohen went at the one component of the work that could be checked by anybody: a household survey Goffman reported conducting with a man she calls Chuck. Her article in American Sociological Review described 217 homes in the neighborhood, 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, and that the reporting fell short of ordinary standards. 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. 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.

Ten years later he is still counting. In December 2025, having been invited to talk with a National Academies panel on corrections and retractions, he sat down and 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 counted the qualitative and mixed papers too. 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.

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, 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 in November 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.”

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 and granted declaratory relief. 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. 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.

Cohen told reporters: “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. … When I have a popular reply it can be viewed by 100,000 people or more, which, while small in the grand scheme, is very satisfying as an individual act of resistance.”

His other public campaign was quieter. In May 2021 he circulated an open letter to the Pew Research Center, signed by roughly a hundred and fifty demographers and social scientists, urging it to stop using labels like Silent Generation, Baby Boom, X, Millennial, and Z. The argument: the boundaries are arbitrary, they invite stereotyping, and they blur the difference between the effects of age, the effects of a period, and the effects of a birth cohort. Pew’s director of social trends, Kim Parker, replied in July that “Generations are one of many analytical lenses researchers use,” acknowledged the risk of age stereotyping, and offered to discuss it. In May 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, readers advised to treat them as reference points.

Cohen's own illustration of the distinction is the essay he wrote 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. Its rationale is a sociological claim about sociology: journals do not only transmit findings, they allocate prestige, impose delay, restrict access, and convert unpaid academic labor into corporate revenue. A preprint archive cannot flatten the hierarchy. It can let work circulate before the hierarchy has processed it, and let a researcher in a country whose libraries cannot buy the packages read the same paper as a researcher at Harvard.

By the time the Kohli Foundation gave it the 2025 Infrastructure Prize for Sociology, ten thousand euros and an invitation to Fiesole, the archive held around seventeen thousand papers. It accepted 3,162 new ones in 2025, twenty percent more than the year before, and in early 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. In February 2026 he wrote up the case that had unsettled him. Someone submitted 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. Attached was 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, interpreted the results, 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, against himself as much as anyone. He is not persuaded that using such tools is unethical, because all knowledge is learned from someone else and no norm requires citing everything you read. He thinks training may be fair use and that displacing the original producers' income would not be. He grants that an archive that assigns digital object identifiers without peer review is an attractive place for machines to deposit and to harvest. He also 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 in March 2026, drafted by a subcommittee and revised by the full moderation team. One key passage in his announcement reads: “We feel the need to express our humanity in this process.” The archive will make human judgments, will not defer to automated detection, and promises fairness without any claim to algorithmic purity.

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.

His third book, Citizen Scholar: Public Engagement for Social Scientists, appeared from Columbia University Press in January 2025, and is the closest thing he has written to a statement of method. Its argument is that values shape the questions anyone asks. The obligations that follow are disclosure of where you stand, honesty about uncertainty, deference to evidence, and clarity about the edges of your expertise. A man may speak as a researcher, a teacher, an activist, an officer of an institution, a litigant, or a private citizen, and the trouble starts when academic authority is used to smuggle him from one role into another.

He practices the disclosure part relentlessly. In June 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 had 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 that he had grown. He would protest that he was fifty-eight. She insisted. Two weeks later 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. In July 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 that he would dislike a high-functioning Sikh country, 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 rather than race, 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 once told him that his own position was not shouting fire in a crowded theater but “saying quietly and unhysterically, 'there's smoke.'” Cohen ends by saying that 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 has kept a running account of how he is seen. Updating the media page on his website, 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 the observation 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, meanwhile, has filed his career under a crusade for equality.

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. 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 that 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, 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 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

Opening scene (1986 newsroom). All detail is from his own account — the Selectric, the carts and paired pages, razor-blade editing, pointing at the engineer, two versions per story plus quarter-hour headline versions, ~30 hours/week at ~$5/hour, the job obtained through his parents’ friendship with the WHCU news director, WVBR training via a school-district internship, high school graduation 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 (a review due every weekend, around 1988) and his Mastodon note to the same effect.

Parents. Marshall Cohen’s Cornell page (PhD Michigan 1965; geometric topology, combinatorial group theory). Avis Cohen’s obituary post, 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, NACS, the AAAS fellowship (2008), Distinguished Scholar-Teacher (2011–12), the robotics afterlife of central pattern generators, the Bianchi comparison, the “all A’s” story, “Thank god, I’m fine!” and “You’ve grown!” — verbatim quotes used: “Being a full time mom was wearing very thin.” (originally from her 2013 GGSTEM essay, quoted in his post) and “Thank god, I’m fine!” There is also a Transmitter obituary linked from his post. One softening on my part: his post says Lenneberg died by suicide. I wrote only that Lenneberg died before she finished.

Lehman Alternative Community School. Confirmed by Wikipedia and by his old school’s alumni blog, which titled its profile of him “How ACS Inspired Philip Cohen’s Crusade for Equality” (November 22, 2017) (search that title if the URL has moved).

Degrees, positions, awards. His CV and Wikipedia. Date discrepancy to resolve: 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. Blog version with the 18% unadjusted and 8% adjusted figures and the reader objection about serial cohabitation and redivorce. The three thousand preprint downloads before journal publication.

Success sequence and Wilcox. The failure of the success sequence; on marriage markets and partner availability, Who are you gonna marry?; on the Supporting Healthy Marriage evaluation and the 27 hours of services, This Supporting Healthy Marriage, I do not think it means; the 2025 post whose title accuses Wilcox of lying.

Goffman. His comment as filed; the original blog post with the arithmetic; the ASR rejection; the Social Forces review (95:4, June 2017, p. e5) and the full text. Goffman’s “ethnographic survey” reply is as paraphrased by Jesse Singal and quoted in Cohen’s post. The 217/308/144 figures are from her 2009 ASR article. Paul Campos’s Chronicle piece is the other place the arithmetic is laid out. Note: I deliberately 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 letter, because the 2012 letter to Social Science Research was organized by Gary Gates and others. His account of the review process.

ASR transparency audits. The December 2025 audit, with the 2025, 2023 and 2020 counts, the National Academies panel invitation, the single retraction from an ASA journal this century, the 2018 Publications Manual lacking the word retraction, the fifty-dollar-bill joke, the advice about blinded OSF links, and the author conceding in the comments. Earlier audits: Sociologist, scientist: toward transparency, accountability, and a sharing culture and Quick update on American Sociological Review’s failure to meet current social science standards. Sociological Science’s reproducibility policy, which he holds up as the comparison.

ASA. The two-stage exit: end of his elected term on the Committee on Publications and the resignation, source for the doubled subscription fees, the 2018 Sage contract, the library wealth-transfer argument, the invented rule barring him from discussing his own proposal, his list of what ASA does well, and the verbatim quote “Sage throws the leadership a nice party at the conference.” The petition against the 2019 ASA letter to the White House. The junior scholar’s objection and the sibling’s comment are both in the comment thread of the resignation post. I did not name the commenter who says they grew up with him, and did not assign a gender or relationship beyond their own words; the display name is “Ness Blackbird.” Also note: there is a widely circulated 2025 “Letter of Resignation from the American Sociological Association” over the Sociologists for Palestine resolution. It is by a different sociologist.

Trump. The tweet, the block fifteen minutes later, and the June 6, 2017 air traffic control post: complaint coverage at Forward and Lawfare. The courthouse scene, including the numbered chip, the steps on Pearl Street, and the verbatim quote “Who knows what he did while we were in here?” His explanation that he replies fast without reading, and the verbatim quote “And honestly, they may be right.” Case history: SDNY May 2018; Second Circuit, 928 F.3d 226 (2019); vacated as moot by the Supreme Court in 2021. Full plaintiff list: Buckwalter-Poza, Cohen, Figueroa O’Reilly, Gu, Neely, Papp, Pappas.

Pew. Open letter, May 26, 2021. Signatory count is reported variously as ~150 and ~170; I used “roughly a hundred and fifty.” Kim Parker’s July 12, 2021 response, source of the verbatim quote “Generations are one of many analytical lenses researchers use.” Pew’s new policy, May 22, 2023. His Washington Post op-ed, “Generation labels mean nothing. It’s time to retire them.” The turning-fifty essay.

SocArXiv. The archive’s own public figure is about 17,000 papers. See SocOpen news and essays. It accepted 3,162 papers in 2025, up 20%, and was running at roughly 4,300/year in early 2026: SocArXiv new submissions at record pace. Kohli Prize (€10,000, ceremony November 2025 at the European University Institute, Fiesole): SocArXiv wins Kohli Foundation for Sociology 2025 Infrastructure Prize and Kohli Foundation. The AI material is the freshest and most useful thing in the file: the case of the paper he would once have accepted, at Where should SocArXiv draw the AI line?, and the policy announcement at SocArXiv releases AI policy, source of the verbatim quote “We feel the need to express our humanity in this process.” The November 2025 moratorium on AI-topic papers and the November 2025 decision to stop accepting arts and humanities work are linked from those posts.

Citizen Scholar. January 2025.

GLP-1 post — source for the insurance detail (state plan covers GLP-1s only for diabetes; about $6,000 out of pocket for the year) and the verbatim quote “never comment on anyone’s body unless invited.” I omitted his weight, BMI, and body-composition numbers, which are in the post. They are his to disclose and I saw no reason to relay them.

Lyman Stone post — source for both Stone quotations (Cohen links the original X posts), Stone’s clan-organization defense, the Murray comparison, the hiring recommendation, and the footnote about Stone calling him a liar. The Murray line, verbatim “saying quietly and unhysterically, ‘there’s smoke,'” comes from What they say about race, where Cohen reports Murray confirming it to him. I gave Stone’s own framing in his own terms and left the adjudication to the reader.

Self-assessment. The anonymous exchange about giving him a voice, and his remark that outreach eats research time (verbatim quote used: “Why give Philip N. Cohen a voice in this discussion?”).

Extrapolations I 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. The one interpretive move I want to flag explicitly is the paragraph pairing his mother’s distributed-control science with his own institution-building. I say in the text that he has not drawn that line in print, and I did not assert a causal relationship. Her claim about hierarchy and nervous systems is quoted from his own post.

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 rather than shrinking him: 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 rather than as evidence about him: 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 rather than ancestry.

Two moves then take the configuration apart. Cohen delivers the verdict rather than submitting the case, 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 rather than analyzing Cohen with it.

Two claims I want to flag as mine rather than Boltanski’s. 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 rather than smuggled in as exegesis. 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 rather than complaining.

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 rather than the substance of the work, 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 rather than because they are hiding anything, 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 rather than eliminating the need for 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 rather than the institutionalist terms it draws on.

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 rather than known to be absent, 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

I tried to enter a conversation and I succeeded.

My interlocutor, the great auditor, can’t point out a mistake. He condemns my first pass on genre.

Bliss was in that dawn to be alive.

Cohen criticizes me for producing a slop biography. He is right. My first pass fails to add anything to his story. It is the jumbled mess he describes.

It is also a breadcrumb dropped in an online wilderness.

Breadcrumbs along the way is my motto.

When I am curious, I read what’s at hand, and then I might post something. The mediocrity of my initial effort embarrasses me and I feel driven to improve it.

If I have time and interest, I return to the post and to add value. The tenth edition is always better than the first one.

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 is real, and it comes from membership rather than from anything outside.

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 rather than being asked to conclude that a particular one is bad. 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 rather than leaving it to the market. 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 rather than party: 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 rather than a solution. Academic freedom protects the analysis. It does not underwrite the verdict.

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. I have paraphrased rather than quoted.

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 rather than whether this one is) 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.

Biographical: born April 19, 1938, Providence; Berkeley, Johns Hopkins, Duke, dean at the University of Illinois at Chicago, then FIU; now Floersheimer Distinguished Visiting Professor of Law at Cardozo.

Frame discipline. Fish only. No Boltanski, no Power, no Becker. The temptation I refused is the obvious one, running interpretive communities over the Goffman dispute and the AI queue as a problem about evidence. The Power essay already lands on that conclusion by another route, and repeating it would give you a second vocabulary for a finished argument. This essay takes the one thing Fish has that nothing else in the file has, which is a published position on the exact question Cohen wrote a book about.

Cohen material. The April 3, 2020 lecture is at his YouTube channel: the disclaimer at 1:14, the four conditions from 3:02, the mask forecast at 6:51, the CDC budget at 7:40, the NSC office at 8:11, the 2019 exercise at 9:02, “total lie” at 9:40, the montage from 10:43, the verdict at 12:46, the market-versus-federal-direction argument at 16:09, the self-correction about outbreak units at 18:11, the lag argument at 22:36, and the instruction at 26:50. The chapter headings on the video, including the one naming what Trump did wrong, are his own and are quotable as his framing. Citizen Scholar (Columbia, January 2025) for the disclosure position. The resignation letter, the community college instructor, and the junior scholar are all in the November 6, 2021 post and its comments. The March 2026 archive policy and the February 2026 post about the submission are at socopen.org. The Bluesky thread is dated July 27, 2026.

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 that 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, and it would import a fight neither Cohen nor you needs.

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 rather than say it. 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 rather than open a second frame.

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).

Frame discipline. Becker only. Otto Rank stands behind the argument about the artist who submits to a larger design, and I left him out rather than open a second frame.

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). 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 rather than the legal position.

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 rather than an accident that befell it.

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 rather than a claim about what happened in his head.

‘Philip N. Cohen: The End of Men Is Not True’ (Oct. 13, 2012)

Working off the YT transcript and running it through AI.

0:00 Thanks to the organizers; he notes that Hanna Rosin has already left and will have to catch it online.

0:29 Sets up the talk: partly about the book, partly about the same claims circulating outside it, with two general points and a lot of data.

1:03 First point: women are not dominant. He declines to run through the specific errors and distortions, says they are on his blog, and argues instead that the story is wrong at its foundation.

1:41 Why the baseline is worth establishing: the theory of how we supposedly arrived here treats women’s rise as continuous and inevitable, which he calls disempowering.

2:20 The claim that young women out-earn young men, which he says has reached meme status and is false.

2:33 Earnings by age across all full-time workers: close in the late twenties, and no reversal.

2:47 The underlying study compared full-time year-round, never-married, childless, metro-area workers aged 22 to 30.

3:15 His objection: that conditioning set is engineered to find the population with the least measurable labor market inequality.

3:35 Break the same group out by education and men earn more within every level; the pooled group is oddly composed, with many Latino men and many White women.

3:56 The rhetoric of fifty percent, which he distinguishes from a tipping point. A milestone is a round number. A tipping point is where a trend accelerates.

4:20 Managerial and professional occupations are majority female, but the category holds the 89 percent of nurses and 78 percent of elementary and middle school teachers who are women alongside engineering at 12 percent.

4:53 Managers alone are far more male; top executives are 26 percent female, inflated by people running small businesses.

5:07 Degrees. On Rosin’s line that women are beginning to crowd men out of science and engineering, he says beginning to means it has not happened.

5:31 Field composition among 25-to-39-year-olds runs from about 75 percent female in psychology to 20 percent in engineering.

5:56 Wives out-earning husbands sits around 28 percent of married couples, and can be pushed near 40 percent by restricting the base to employed or full-time women.

6:43 The full distribution rather than the fifty percent cutoff, 1970 to 2010. Couples where the wife earns nothing fall from 46 percent to 18.7. Couples where she earns everything rise from 0.6 percent to 2.3.

7:32 Newlyweds in the American Community Survey: 36 percent of college-graduate wives out-earn their husbands in the first year, 44 percent among African American women.

8:19 Rosin’s claim that sexual assault rates have fallen so far that criminologists cannot plot them.

8:38 He grants the decline is large and appears in both police reports and the National Crime Victimization Survey.

9:02 New Jersey has the lowest state rate of reported forcible rape at 11.2 per 100,000, which he says is more than twice the median for European countries.

9:36 Rosin’s reversal, that female power will bring more female violence.

9:54 The only rising series is arrests of female juveniles. Victimization data show female offenders down by about half in twelve years, and intimate partner homicide down sharply, with the steeper fall among male victims.

10:46 Alabama. He pauses on the claim that women dominate Alexander City and Auburn, and says he can say they do not without ever having been there.

10:58 The line the talk turns on: if the facts do not support the generalization, the anecdotes are not illustrations of a trend, they are little stories.

11:11 The first female mayor was elected eight years earlier, is a traditional southern Republican from a political family, is stepping down, and five men are running to replace her.

11:45 One woman on a six-member city council; fourteen of fifteen department heads male.

11:55 Against the trailer-park anecdote: 77 percent of men in their twenties in Alexander City are employed, against 53 percent of the women.

12:12 He interrupts himself to say he is not claiming the men are better, only describing a normal pattern.

12:24 Auburn. Council, mayor, and department heads; the within-couple income distribution close to the national one; women at about 70 percent of men’s earnings; men holding about 70 percent of management.

13:18 The velocity claim, that change is too fast to stop and will carry women past parity into dominance.

13:45 Rosin’s list of firsts she calls unthinkable twenty years earlier.

14:06 All of them had already happened, except an Ivy League president, and that was 1993. Marlene Sanders and Jessica Savitch anchoring; Shirley Chisholm and Pat Schroeder running; Roseanne Barr; Penny Marshall on her fourth feature by 1992; Maggie Lena Walker founding a bank in 1902.

14:29 His point: movement toward equality takes politics, social movements, and policy, and does not arrive on its own.

14:47 Women’s labor force participation flat since the late 1990s.

14:57 The pay gap narrowed sharply in the 1980s and early 1990s and has crawled since.

15:04 College-educated women are doing worse relative to college-educated men than they were twenty years earlier.

15:11 Occupational segregation: no progress in twenty years for women without degrees, modest progress for women with them.

15:23 Women entering management: large gains in the 1970s and 1980s, close to nothing since.

15:34 Conclusion. We should think harder about why we are stalled than about how far we have come.

15:56 What he finds wrong in the narrative is the implication that we can sit and watch women collect their gains.

16:03 Google Books mentions of feminism and feminist, both peaking in the 1990s.

16:12 Close: culture, politics, and policy, rather than waiting for technology to work its magic.

The method is complete at age forty-five. This is October 2012, three years before the Goffman comment, four before the archive, nine before he quits the association, and everything the later career runs on is already here: the receipts parked on the blog so the talk can make an argument, the attack on a conditioning set, the correction of a factual detail nobody bothered to check, the willingness to take apart a book whose politics his audience shares, and the flat prosecutorial delivery.

The best thing in the talk is the point about the conditioning set at 2:47. The statistic requires you to hold constant full-time work, year-round work, never having married, having no children, living in a metro area, and being between 22 and 30. His objection is that the controls are the finding. If the gap disappears only after you remove marriage, children, part-time work, and everywhere that is not a city, then marriage, children, part-time work, and geography are the gap. That is the same move he later runs against the success sequence, and it generalizes past gender to any argument that produces equality by subtraction.

The distinction at 3:56. Fifty percent is a round number in base ten and nothing happens at it. A tipping point is a claim about acceleration and has to be shown. Most trend journalism converts the first into the second without noticing, and the conversion is invisible because the arithmetic looks like evidence.

The line at 10:58 is the one to quote. If the facts do not support the generalization, the anecdotes are not illustrations of a trend. It is the clearest statement I have seen of the quarrel between reported texture and aggregate measurement. Rosin was not claiming that Alexander City has a female-majority council. She was reporting a felt reversal of standing inside particular homes among particular men, and Cohen answers with employment rates and department heads. He is right that the aggregates do not show dominance and she is right that aggregates are not what she went to see. The boast at 10:46, that he can say it never having been there, is the whole epistemology in one clause, and it is the same claim that made the ethnographers furious three years later.

The mayor correction at 11:11 is his strongest hit and the model of what he does better than anyone. The claim that a town got its first female mayor that year was checkable in five minutes, was wrong by eight years, and had survived whatever fact-checking the book received because it sounded like the thing being argued. The 14:06 list is the same weapon aimed at chronology: every item presented as unthinkable twenty years earlier had already happened, some of them long before. Trend writing runs on an assertion of novelty and the assertion is the least examined sentence in the genre.

The stall thesis is the part that has aged best. Convergence in pay, segregation, management, and participation happened in the 1970s and 1980s and mostly stopped, and he said so in 2012 and has been broadly right since. It also carries the politics, and he states the link at 14:29: automatic progress requires no movement, and stalled progress requires one.

Two weak spots he would not tolerate in anyone else.

The New Jersey comparison at 9:02 does not work. Cross-national counts of reported rape are shaped by legal definitions and reporting propensity more than by incidence, which is why Sweden looks like a catastrophe and why the countries with the lowest reported rates are often the worst places to be a woman. He built his career on the principle that you check whether the instrument changed before you compare, and he skipped it here for a debating point against a rhetorical flourish. He had already conceded the substantive claim at 8:38.

The Google Books slide at 16:03 is worse. Frequency of the words feminism and feminist in a corpus of published books is a fact about publishing and about corpus composition, and he uses it as a proxy for the vitality of a political movement, with no adjustment and no discussion of what the corpus contains. It is the move he spent the previous fifteen minutes taking apart. It is also the last slide, which is where a speaker reaches for the thing that feels true.

One detail Tom Wolfe would have led with. This is a law review symposium at Boston University, the format is adversarial by design, the subject of the prosecution left before the prosecution began, and Cohen opens by pointing at the empty chair and joking about it. He is the expert witness in a proceeding with no defendant present, which is a fair description of most of his public career, and the joke suggests he knows it.

‘Philip Cohen addresses National Academies on “Against the Generations”‘ (July 30, 2019)

0:06 Opens by thanking the organizers, notes he is the last speaker before lunch and will try to be brief and lively.

0:30 Says the committee’s remit gave him the chance to make a pitch that has been gnawing at him for four years.

0:57 The complaint: generation talk encourages broad generalization, stereotyping, clickbait, character judgment, and echo-chamber thinking. He will illustrate with family demography and the recession.

1:30 The Pew Research Center’s generation scheme, which he calls the leading purveyor of these memes.

1:43 First objection: these are cohorts rather than generations, the categories are irregular, and the boundaries have no justification. Making the last two spans fifteen years each avoids people sharing a generation with their own children, and families skip generations anyway.

2:13 His own family has no baby boomers; his brother was born in late 1964.

2:32 Once you name them and assign characters, you impose qualities on populations with no basis, and it snowballs into confirmation-bias clickbait that nobody assesses.

3:02 He acknowledges the incentive problem. Abandoning terms that generate millions of clicks would be brave and admirable.

3:24 The 2015 Pew survey that gave people a lineup of generation names and asked which they belonged to. Only 40 percent of Millennials picked Millennial, and respondents could select more than one.

3:53 Uptake was low except among boomers, and the response was not self-reflection. He points out that if racial categories produced this, social science would treat it as a problem for the theory.

4:29 A disclaimer he repeats: he admires Pew, uses their material constantly, including in this talk, and thinks they should get more money. The trouble is an addiction they have to sell.

4:45 Pew’s announcement moving the millennial endpoint back to 1996 and naming Generation Z. On their line that cut points are not an exact science: they are not a science at all, and there is no question they are trying to answer.

5:22 On their claim that the boundaries are not arbitrary: they are completely arbitrary, and there is no agreed formula for the span.

5:42 The baby boom is the only cohort officially designated by the Census Bureau, and it is a demographic event that produced a cohort rather than a generation.

6:06 When Pew shortened the millennial span it generated stories treating the arrival of more Millennials than Gen Xers as news, when the span had simply had more years in it.

6:30 The census work: a birth rate spike in 1946 holding above that level through 1964, the only prolonged rise to a stable high level of fertility in the record, breaking the long-run trend.

7:08 The accident that started everything. The event happened to last roughly the length of an actual generation, which put the culture on a treadmill of naming them. Had the boom run eleven years instead of nineteen, he does not think the conversation would exist.

7:33 He realizes he never checked the clock.

7:45 The celebrity slide. The Silent Generation runs from Noam Chomsky to Debbie Harry. Donald Trump and Keanu Reeves are in the same generation.

8:09 The line between X and Millennial separates Venus and Serena Williams, sisters one year apart. He says the Gen X is all over Venus, you cannot miss it, this is like astrology, and Serena is a pure millennial.

8:43 Age-specific fertility rates from the American Community Survey, 2001 to 2017, by birth cohort, with dots marking where each cohort stood in 2010.

9:13 Women born in the early 1970s hit 2010 in their late thirties with their path already written. Late 1970s cohort, same.

9:37 The early 1980s cohort, the first Millennials, were at peak fertility when 2010 arrived, and fertility declined after.

9:52 The late 1980s cohort saw fertility fall from ages 23 to 24 to 25, which he calls extremely unheard of, then rebound into their early thirties, possibly close to catching up on completed fertility.

10:33 The youngest group hit 2010 as teenagers, and their curve rotated clockwise, falling off faster than earlier cohorts, with signs of a post-recession rebound.

11:09 The point: all three of those lines are Millennials, and collapsing them loses the entire story.

11:16 He notes his five-year groupings are arbitrary too. Use one-year or ten-year groups, or groups chosen for a reason.

11:28 Male employment rates by single year of birth from the Current Population Survey.

11:42 Men born in 1982: about 47 percent employed at eighteen, rising to their mid-twenties, then falling as the recession hit them in their late twenties, which he calls ordinary recession behavior.

12:01 Men born in 1987: a scarring event, with employment rates flat from 21 to 23.

12:18 Men born in 1992: 2010 hit at eighteen, and a twenty-point difference in employment at that age against the earlier cohort. All three are Millennials.

12:44 Marriage. Propensity to marry before thirty, controlling for age and sex.

13:00 The decline between people born in 1945 and 1965 is enormous, and both are baby boomers, so the group does not hold together on this measure.

13:20 The question science should be asking: when do we want to group people, and how.

13:29 First positive suggestion: ask whether the question is a cohort question at all.

13:41 On “kids these days,” a thing you could say at any point in history. General Social Survey data showing the under-35 share identifying as liberal is consistently higher.

14:09 If you are using a cross-section, you are describing right now, and labeling those people Millennials adds only the baggage.

14:26 Second suggestion: if you want cohorts, break them into five or ten year groups sized to your sample and eyeball it first.

14:34 Attitudes toward easier divorce, adjusted for age, sex, and education, by the four standard generations, then by decade of birth.

15:01 The decade version shows what the standard grouping hides. Late Millennials are markedly more pro-divorce than early Millennials, and the silent-to-boomer shift happens earlier than expected.

15:19 Naming advice: describing people as born in the eighties or nineties is fine. Calling them the tennis generation or the butterscotch generation assigns a character with no basis.

15:41 Third suggestion: cohort analysis is important where events at a point in the life course have lasting consequences.

15:56 Examples. Children of the Great Depression as the classic. Digital natives, if you study education.

16:14 The story about his daughter texting to ask what her phone number was, and the other daughter’s ruling that if it can send email it is a phone.

16:38 Post-9/11 schooling for security research; leaving school in a recession for labor economics; and, for politics and culture, the high school students shaped by the Parkland moment.

17:07 The close: none of those experiences supports assembling one thing, calling it a generation, and having it make logical sense. He urges the committee not to use the labels at all.

17:39 First question, from John: is there systematic research showing that generation language promotes stereotyping and bad social outcomes?

18:02 His answer: he knows of none. Most sociologists do not use these as real categories in the research, only in the discussion section when making findings seem relevant. He suggests experimental psychology could test what the labels evoke.

18:39 Second question: if Pew hired you, what would you tell them to do other than stop?

19:00 His answer: do not give up cohort analysis. Use decade cohorts and routinize them; the labels exist, the intervals are regular, and they already carry cultural salience. He is not telling Pew to delete their account after a thousand reports.

19:37 His own practice: a script that breaks age into decades and years into decades as a first pass at trends before looking finer.

20:07 A committee member’s comment: cohort and age questions apply universally across the social sciences in a way most topics do not.

20:58 Third question, from Dana, on the great-and-terrible framing: pushing him toward what is redeeming in generational analysis, and noting that many people think astrology has scientific underpinnings.

21:47 His reply that astrology has no scientific value, followed by the crack about a selected expert audience.

22:01 The substantive answer: cohort analysis and the life course perspective are important. Job changes, health insurance at a given age, school crowding, the racial composition of a childhood neighborhood. He wants to see the shadow of events across lives at the cultural level.

22:40 A concession. The millennial idea was that these people came of age in a new era around 2000, which he calls a powerful way of thinking and an opportunity to apply social science.

23:02 He is not discouraging cohort thinking, he wants to improve it, and he acknowledges an institutional juggernaut built around the concepts.

23:34 A return to the Pew survey: if you live in a world where everyone knows what Millennials are, that may indicate you are not interacting with most people.

24:01 The closing self-assessment: this is not his most important work. Understanding demographic change, how it relates to inequality, and communicating it so people can make smarter decisions is his most important work, and he counts this as part of that.

This is the campaign two years before the open letter, delivered to the body most likely to act on it, and it works. He got Pew’s policy change in 2023 and this is where you can watch him earn it: no accusation of bad faith, praise for the target three separate times, a concrete alternative offered under questioning, and a closing line that ranks his own complaint below his real work.

Compare it with the Rosin talk seven years earlier and the difference is the target’s status. Both talks run on the same two moves. First, take a category and show that the things inside it behave differently, which is the argument at 9:13 through 12:44 and the argument about the 22-to-30 conditioning set in 2012. Second, check a claim that nobody checked because it sounded right, which is the Pew self-identification survey here and the Alexander City mayor there. What changed is that in 2012 he was correcting a journalist and in 2019 he is advising a National Academies committee, and the register adjusts without the argument changing at all.

The strongest passage is 6:30 to 7:33 and it is the part nobody quotes. The baby boom is a real demographic event with a documented spike, a documented plateau, and a break in the long-run trend, and the Census Bureau designated it for that reason. Everything downstream is imitation of a thing that happened once. And then the historical accident: the event lasted about as long as a human generation, which licensed the treadmill. Had it run eleven years, he says, there would be no Generation X, no Millennials, no Z. That is a claim about how a measurement artifact became a cultural ontology, and it is the most original thing in either talk.

The Venus and Serena slide is doing more work than the joke suggests. Sisters born a year apart, raised in the same house by the same parents on the same courts, sorted into different generations by a boundary somebody chose. It is a within-family control. He throws it away as comedy at 8:09, which is a habit of his: the sharpest evidence gets delivered as a laugh line.

He concedes the key empirical question at 18:02 and nobody notices. Asked whether there is research showing generation labels cause stereotyping or harm, he says he knows of none. The talk’s opening claim, at 0:57, is that the labels drive stereotyping and character judgment. Under questioning he admits that is an intuition. It is a reasonable intuition and it may be right, and it is unverified, and this is the man who spends his December afternoons counting whether other people’s claims can be checked. He does the honest thing and answers straight, and the honest answer undercuts his own frame.

The astrology answer at 21:47 is the misfire. The questioner was making a subtle point, that a scheme with no scientific basis can still organize how people understand themselves and each other, which is an argument that generation labels might be socially real even if demographically empty. He hears it as a claim that astrology is scientific, corrects her, gets a laugh from the room, and only reaches the substance after the moment has passed. The correction was directed at something she did not say.

And the concession at 22:40 is the largest one he has ever made in public on this subject. The millennial idea, that a cohort coming of age around 2000 encountered something new, is called a powerful way of thinking and a good opportunity for social science. That is close to the thing he spent seventeen minutes urging them to abandon, and it means his position is narrower than the title. He objects to the naming, the character assignment, and the arbitrary cut points, and not to the underlying thought.

The distinction between a demographic event and a generation, with the eleven-years counterfactual, is portable and nobody has credited him with it. And the decade-cohort script he mentions at 19:37, the thing he runs before looking at any trend, is a small biographical detail that shows the habit underneath the arguments.

Transcript flags. At 4:29 “picking up Hewitt” is picking on Pew. At 4:50 “296” is 1996. At 5:12 “your medicines are talking their story ins” is a mangling, probably epidemiologists and historians. At 8:23 “Generation Next” is Generation X, and “Carrie” should be career or possibly aura. At 13:00 “married by age three” is thirty. At 14:43 “general any ‘el” is Generation X and Millennial. At 15:27 “quartzion” is cohorts. At 19:21 “cultural violence” is cultural salience. At 22:53 “melayu” is Millennials. At 24:27 “dr. otter” and “dr. Capelli” are the remote panelists, and Peter Cappelli is the likely spelling.

‘Philip N. Cohen COVID lecture April 3 2020’

0:00 Opens by greeting his social problems class, says he has been missing them, and notes it has been a month since he first opened the class to discuss the then-emerging epidemic.

0:23 Sets the plan: where we are, how we got here, where we are going, then discussion.

0:48 Records this on the afternoon of April 3, 2020.

1:07 Structure: global trajectory, US trajectory, and what comes next.

1:14 The disclaimer he returns to twice more. He is not an expert on pandemics or global public health. He is a social scientist who can understand what he reads, and a teacher.

1:36 Etymology of pandemic, from the Greek for all and people.

1:44 The World Health Organization declared a pandemic on March 11.

2:11 Coronaviruses affect humans and other species, ranging from common colds to serious disease. The novel virus was found in Wuhan late in 2019; the disease is COVID-19.

3:02 Why this one is bad. No human immunity.

3:18 Highly contagious, and he says twice that we do not know exactly how it spreads.

3:34 Asymptomatic transmission, which defeats the strategy of avoiding people who look sick.

3:58 Relatively high mortality. Most people survive and most cases are mild, and the rate is higher than seasonal flu. He says we do not know the rate, that it depends on how much medical care you get, and that it is maybe around one percent.

4:31 Confirmed cases worldwide, probably reaching a million as of the day before, with 80,000 new cases reported yesterday. Exponential increase.

5:11 The global map. Spread followed travel from China to Europe and the United States, with an Iranian outbreak from a cluster of travelers.

5:42 Countries without rigorous testing are unknowns, so low counts in Kenya or Zambia say nothing about incidence.

5:59 Droplet transmission and the origin of the six-foot rule, which comes from how far droplets travel when you cough.

6:16 Asymptomatic spread, plus strengthening evidence of aerosolized droplets that may linger in air, and cases where no liquid droplets appear to have been exchanged.

6:51 Current recommendations: six feet, hand washing. And a forecast, on April 3, that a cloth mask recommendation is coming to protect others from what you may be expelling before you have symptoms.

7:23 The section on what the Trump administration got wrong before the outbreak. First, a general orientation against effective government, expressed in CDC cuts from just under eight billion in 2018 to just over seven in 2019, with a proposed budget near six and a half.

8:11 The pandemic preparedness office in the National Security Council eliminated in a reorganization, and a Center for Strategic and International Studies report from the previous fall noting it remained unclear who would be in charge in a cross-border biological crisis.

9:02 The 2019 federal pandemic exercise, run on a novel influenza scenario, which found the government unprepared and produced recommendations that were ignored.

9:40 Trump’s claim that nobody could have seen this coming. Cohen calls it a total lie, then adds that Trump did not know because he does not know a lot, while people who paid attention did.

10:05 The happy talk section, and Trump’s position that one or two hundred thousand deaths would count as total victory.

10:43 A montage of Trump clips: he always knew it was a pandemic, it is totally under control, one person from China, fifteen cases going down to close to zero, it will disappear like a miracle.

12:46 His verdict on the clips: none of it was true, and the messaging existed to make Trump feel he was doing a good job, which sent the public a signal that this was not serious.

13:09 Operational failures. Testing first. Without it you cannot count cases.

13:25 The standard first-line response: isolate, contact trace, quarantine contacts.

13:57 The CDC’s first test did not work, and the country did not adopt the World Health Organization test that might have permitted faster testing at scale.

14:13 Community spread arrived before anyone could trace it, with cases of unknown origin.

14:22 Containment failures: no restriction on movement between affected and unaffected areas, or within towns, schools, nursing homes, and malls.

14:46 No uniform system for stay-at-home orders. Most places now have rules or strong guidance, not everywhere, and all of it too slow.

15:19 Supply failures. Personal protective equipment, and the fact that it has to be changed constantly.

15:44 Ventilators, and the shortage of both machines and people to operate them. The federal stockpile was insufficient, some units did not work, and distribution was poor.

16:09 His central policy criticism: too much left to the market and to the producing companies rather than the federal government seizing control of production and distribution.

16:42 The log-scale doubling chart, days since a country’s tenth death against deaths, each gridline a doubling, a straight diagonal meaning exponential growth.

17:22 South Korea at about 37 days sits between 128 and 256 deaths. Spain at 24 days is near 10,000. Italy, France, and recent outbreaks in Brazil and Canada at 13 or 14 days follow the same course.

18:03 The United States slope is steeper than the others, which he calls the disturbing part.

18:11 A qualification that improves the analysis. The country is large enough that it is more like Europe, a set of separate outbreaks rather than one.

18:28 The same chart by state. New York on Spain’s path, Washington the best at 29 days and under 256 deaths, Michigan, New Jersey, and Louisiana rising dangerously.

19:11 What a straight line means, and what bending looks like. Spain and Italy show early curvature if the numbers are accurate, and he notes the fuzziness.

19:36 China is left off because it would break the graph.

19:53 Flattening the curve. The red distribution shows unmitigated spread piling up cases past the health system’s capacity.

20:23 The logic of closing schools, restaurants, and bars: slow the rate so the sick arrive at a hospital that can take them.

20:54 Even without a vaccine, slowing infection reduces total damage.

21:11 Back to the state chart, with an explanation of the variation. Washington’s first outbreak was largely confined to a nursing home, and the state closed and distanced early, producing a shallower slope. California similar. New York and Michigan did too little early and got explosive outbreaks.

22:19 The window argument: most of the country does not have a bad outbreak yet and can still change its slope.

22:36 What happens next. People arriving at hospitals now were infected ten days to two weeks ago, six to eight days to symptoms and another five to seven to hospitalization, so today’s admissions date to roughly when schools closed.

22:59 Spread continues because nothing stops you driving from New York to Missouri.

23:15 His prediction: many thousands of Americans will die and that is inevitable at this point, with no break in the slope visible.

23:35 How bad it gets over months remains open, and mortality is not a fixed biological rate. It depends on diagnosis, hospital resources, and staffing.

24:10 The Javits Center temporary hospital, and his forecast of more converted convention centers.

24:27 When it ends. Absent a vaccine, nobody is immune, so tamping the numbers down may let people out with waves of local outbreaks and repeated shutdowns.

25:00 The exit condition: few enough cases plus testing and tracing capacity, then normal life with distancing and rapid isolation when someone gets sick.

25:30 Global spread means travel restrictions and quarantine requirements will last much longer than domestic measures.

25:54 Vaccines. A best case around eighteen months, then testing for safety and efficacy, then production and distribution of billions of doses.

26:18 No treatment yet; patients fight it with their own immune systems, and treatment research is underway.

26:34 What this means for students, prefaced by another disclaimer: this is not his expertise, he cares about them, and he is not the person to give personal advice.

26:50 Follow real public health advice. Do not listen to the president or the politicians in his orbit. Listen to your governor, local public health officials, and the CDC.

27:14 It is not only about you but about whom you might affect.

27:21 Take care of yourselves mentally and physically, and do not let physical distance become total social distance.

27:37 Helping others, with time or money or online, can be empowering.

27:52 The close. The world will be different, think about who you want to be in it, and humanity has gotten through bad things before.

This is the only recording in my file so far where Cohen is not prosecuting anybody’s arithmetic.

Watch what he does with uncertainty here. Inside twenty-five minutes he says we do not know exactly how it spreads, we do not know the mortality rate, we do not know how many are infected in countries without testing, we do not know how many will die, we do not know how long it takes, the numbers have fuzziness in them, and China would break the graph so he left it out. That is a man narrating the limits of his evidence in real time to nineteen-year-olds. The audits and the Rosin talk and the Pew talk all run on the opposite posture, certainty about a number somebody else got wrong. The pedagogical setting produces the epistemics his critics say he lacks, which suggests the register is a choice about audience rather than a feature of the man.

The disclaimer at 1:14 is the thesis of Citizen Scholar five years before the book. He tells them he is not an epidemiologist, that he is a social scientist who reads well, and that he is a teacher. He is claiming authority to synthesize while disclaiming authority to know, and then he spends twenty minutes doing exactly what he claimed. That is the role boundary the book argues about, drawn in public, at the moment it was hardest to draw, when a professor with a microphone and frightened students could have taken the larger authority for free.

Then he takes a different one. At 9:40 the lecture turns and stays turned. Ignored warnings, a total lie, the montage, the verdict that the messaging existed to make the president feel good about himself, and at 26:50 a direct instruction to his enrolled students not to listen to the president. He would say the roles are separable and the sourcing supports every claim, and the sourcing does. His own book says the trouble starts when academic authority carries a man from one role into another without the audience noticing the crossing. Here he crosses in front of a captive audience holding his grade book, in the same voice he used for the doubling chart. He announces the crossing at 26:43 by saying this is not his expertise, which is the disclosure his method requires and which does nothing about the power differential. Someone should ask him about it.

He was right about the things his training equipped him to judge and less reliable on the rest. The market-versus-federal-direction argument at 16:09 is his own field and is a live policy dispute he states as settled. The eighteen-month vaccine best case at 25:54 was the consensus and turned out pessimistic, and the reason it was wrong sits outside a demographer’s competence. The forecasts he got right are the ones that follow from arithmetic he could do: exponential slopes do not break on their own, hospital admissions lag infection by two weeks, tamping down produces waves rather than an ending, travel restrictions outlast domestic ones, and thousands would die. The one he got right by a day or two is the mask recommendation at 6:51; the CDC issued it that same day.

Two moves.

At 18:11 he corrects his own chart while presenting it. The American slope looks worse than Europe’s until you notice the country is the size of Europe and holds several separate outbreaks, so the aggregate curve is an artifact of the unit. That is the same objection he makes to generation labels and to the success sequence: the category is doing work the data are not. He applies it to his own slide, unprompted, in front of students who would never have caught it.

And 22:36, the lag argument, is the finest teaching in the video. Today’s hospital is a photograph of two weeks ago. Everything you can see is already history and everything you do today shows up in two weeks. That is a demographer’s habit of mind, the same one behind checking his divorce prediction against data that arrive years later, and it is the single most useful thing an ordinary person could have learned in April 2020.

This belongs in the biography as the missing 2020 entry. Every other document has him fighting somebody. This one has him teaching, and the teaching is patient, careful about what it does not know, and warm at the end in a way the blog never is. The essay currently derives his character from the news desk clock and the audit tables. This is the third setting, and it is the one where the students are, which is where most of a professor’s life happens. Two lines at 27:21 and 27:52, about not letting physical distance become social distance and about deciding who you want to be afterward, are the only places in anything I have read by him where he speaks to a person’s wellbeing rather than to the record.

Transcript flags. “philip gon” at 0:00 is Cohen. “COBIT 19” and “Kovach 19” are COVID-19. “pandemics rent” at 8:37 is pandemic event or similar. “first actor” at 21:37 is first outbreak. “organ bed” at 14:56 is garbled, likely congregating. “an S case scenario” at 26:02 is a best case scenario. “how many people really happen in Kenya” at 5:42 is have it. The chapter titles (“WHAT TRUMP DID WRONG: CUT CDC FUNDING”) are his own.

‘Family violence and abuse in COVID context lecture’ (Apr. 6, 2020)

0:00 Announces a short video lecture on family violence and abuse and the implications of the pandemic.

0:18 Context for why now: he has been revising the violence chapter of his textbook, due out in about half a year.

0:45 The headlines he is responding to, that home is not always safe and that lockdowns are associated with rising domestic violence around the world.

0:59 His first move, and the lecture’s governing posture: we do not have the information to draw those conclusions yet, but we know enough to raise questions and think the issues through.

1:15 Historical backdrop. Violent victimization of people aged 12 to 17 by family members, from the National Crime Victimization Survey, falling from about 14 per thousand to just over two.

2:00 Why it declined, prefaced by a warning that attributing violence trends to particular social changes is speculative.

2:29 Three candidate causes. Shelters and hotlines that provide distance from abusive partners.

2:47 Women’s rising education, employment, and the cultural acceptance of independence and single parenthood, which make leaving easier both structurally and in terms of values.

3:10 The decline in marriage and cohabitation, which reduces exposure to the main risk factor, presence of a spouse or partner.

3:28 Lifetime prevalence. Something over a quarter of both men and women report violence in a relationship, and the gender disparity widens sharply for severe physical violence, contact sexual violence, and stalking.

4:18 Defining the sphere. Family violence means caring or intimate relationships, with partners, children, and elders.

4:48 The structural claim. Secrecy and isolation are made possible by how families are built, literally by the walls and figuratively by the sphere of privacy people value highly, which is the same privacy that makes the problem hard to reach.

5:22 The framing he says he always reinforces: this is not about bad apples, it is a structural feature of family life, which does not mean everyone experiences it and does mean we should ask what our institutions do to enable and reproduce it.

6:04 Child abuse and neglect defined as an act or failure to act by a parent or caretaker resulting in or risking physical or emotional harm, sexual abuse, or exploitation.

6:32 The perpetrator chart, overwhelmingly parents and family members.

6:52 Two glosses. Neglect is definitionally a failure of whoever holds responsibility. The high figure for mothers reflects that many children live with their mothers only.

7:12 Risk factors: parental mental health and impulse control, any history of violence including as a victim, violence between adults in the home.

7:49 Poverty, which raises risk partly through material deprivation and partly through whatever produced the poverty.

8:14 Weak support networks, and the importance of having people around who notice and can intervene.

8:38 A clip from a Cleveland television station on advocates’ concerns, including the governor’s worry about children not seen by anyone outside the family, a center CEO reporting steady calls and heightened fear, and a closing line about checking in on each other.

10:52 The pandemic’s implications, opening with a swipe at the genre of first-world-problems content, the boredom and the same shows and the boring food, which does not tell you the story of families where things are going badly.

11:28 Economic shock. Roughly ten million new unemployment claims in recent weeks as of April 6.

11:45 Isolation from support networks, since family and friends are also sheltering. Who checks in, who notices, who helps, where does somebody escape to.

12:16 Services may be unavailable, government and nonprofit alike, for lack of staff and resources.

12:42 Mandatory reporters. Teachers, clergy, and in some states health care providers are the points where the outside world can reach into a family, and those interactions have been cut.

13:08 His caution again. All of this is theoretical; there is no evidence yet, and reason for concern.

13:23 Intimate partner violence against adults, down from about 16 per thousand women in the mid-1990s to about four, roughly a three-quarters reduction, with a decline for men from a much lower base.

14:04 Homicide, which the survey cannot capture because you have to be alive to answer it. In 2018, a little over a thousand women killed by intimate partners against about 360 men, with male homicide concentrated among associates and strangers.

14:47 The counterintuitive prediction. Lockdown may reduce male homicide more than female homicide, because it removes exposure to acquaintances and rivals while increasing time spent with partners.

15:15 The typology. Situational couple violence, arising from specific disputes without a pattern of escalation, triggered by stress and economic shocks, which is why he expects more of it now.

15:50 Coercive controlling violence, formerly called intimate terrorism, a campaign for control and domination with an escalating pattern.

16:12 Why the pandemic worries him more here. Escaping this pattern usually requires outside support, and isolation removes it.

16:36 The specific route: decisions about who goes out and when, and the negotiation of rules for mobility, feed into controlling behavior.

17:05 Slightly less than half of intimate partner violence is ever reported to police.

17:18 Reasons given for not reporting. Fear of reprisal or of getting the offender in trouble is the most common, covering both punishment for reporting and loss of the family’s economic support. Handling it personally is about 20 percent. Third is lack of faith in police.

17:58 Whether this changes under lockdown he calls speculation.

18:14 Rape and sexual assault. About two thirds by acquaintances and intimate partners, about 24 percent by strangers.

18:59 The two directions the pandemic pushes. More isolation with partners raises one risk while less circulation lowers the other, and the net is unknown.

19:24 Interventions, introduced as things there is reason to believe may help or that many people think are good and are trying.

19:41 Rapid assessment checklists for police at the scene, formalizing judgments about severity and escalation and routing victims toward services and escape planning.

20:13 Civil protection orders, with some success.

20:25 Court-mandated psychological treatment, with a mixed record.

20:35 Specialized family violence courts with judges who develop expertise, again mixed.

20:55 Victim services broadly, including escape planning, assessment, and counseling.

21:16 The forward question. Given a quarter century of decline, what does the present crisis imply.

21:36 The modernity argument. First, the construction of childhood as a protected period of innocence recognized by families, institutions, and the state.

22:19 Second, the rise of state power. Outsourcing dispute resolution to formal authority may reduce interpersonal violence, while raising its own questions about surveillance and oppression.

22:53 Third, the Pinker argument that the rise of reason as an ideal reduces violence because you have to justify what you do.

23:19 The momentum claim: childhood exposure predicts adult violence, so a generation raised with less of it may pass less of it on.

23:42 The unsettling. Two months ago he would have named climate change or Trump; now the pandemic raises the question of whether history has a direction at all, and big crises unsettle any narrative they do not fit.

24:16 The close. He has no answer for most of what he raised, and hopes the context and the background statistics will let students work out what is happening.

On April 3 he played a montage of presidential clips, called a public statement a total lie, and told enrolled students not to listen to the president. On April 6, on a subject where the political temperature is comparable and the policy stakes are immediate, there is nothing. The word Trump appears once, at 23:48, inside a subordinate clause listing the kinds of events that unsettle a story about progress. The rest is data, definitions, a typology, and seven separate admissions that he does not know.

The lecture where he editorialized was the one outside his competence. The lecture where he taught was the one inside it. Whatever governs his register, it is not confidence in the material. My guess, and it is a guess, is that the pandemic lecture had a villain available and this one does not, or has one he is unwilling to name in front of undergraduates whose homes he knows nothing about.

The epistemic discipline here is the best in anything of his I have read. He opens with the headline, tells them the headline cannot yet be confirmed, and then spends twenty minutes explaining what would have to be true for it to be right. He flags speculation at 2:19, at 13:08, at 17:58, at 18:59, at 21:30, and at 24:16. That is a professor teaching students how to hold a live question open, in April 2020, when every incentive ran toward certainty.

Two things in it have aged well and one has not.

The measurement point at 12:42 is the one that mattered most and the one nobody outside the field made in public that spring. Teachers, clergy, and clinicians are the layer through which the state sees inside a family, and closing schools removed it. Which means a fall in reported child maltreatment in 2020 would measure the removal of the observers rather than a change in the thing observed. Reports did fall. Anyone reading those numbers without his warning would have drawn the opposite conclusion from the correct one, and the subsequent literature on intimate partner violence has been a mess for the same reason, with calls to police and hospital presentations and hotline volume moving in different directions because each measures a different mixture of incidence and access.

The structural claim at 4:48. 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. That is the argument of his textbook and the subtitle of his essay collection compressed into thirty seconds, and it is the reason he can teach this subject without moralizing: nobody has to be a monster for the structure to produce the outcome.

The prediction at 14:47 is the falsifiable one and it looks wrong. He reasoned from exposure structure: male homicide victims are killed by acquaintances and rivals, so lockdown should cut male homicide more than female homicide. American homicide rose sharply in 2020, by roughly a third, the largest single-year increase on record, and the increase was concentrated in exactly the non-intimate gun homicide his exposure logic predicted would fall. He modeled a pandemic as a reduction in interaction, and 2020 turned out to be a reduction in some interactions and a collapse in others, including in policing and in the institutions that suppress street violence. The exposure model was too clean for the year.

Three weaknesses, stated the way he would state them about somebody else.

The mother figure at 6:32 gets a chart and a single sentence of explanation, that many children live only with their mothers. That is a real explanation and an incomplete one. Exposure time, the poverty of single-parent homes, and neglect definitions that by construction fall on the primary caregiver all belong in the answer. He knows all of this. He is presenting a number that gets used badly and giving it a nine-word gloss, which is close to the thing he spends his blog objecting to.

The interventions section is his weakest four minutes anywhere. Some promise, some success, rather mixed, mixed results, and an introduction that says many people think these are good so they are trying them. No citations, no effect sizes, no discussion of what the evaluations actually found. Set that beside the ferocity he brings to counting whether a journal article shipped its code and the asymmetry is hard to miss. The standard he applies to other people’s claims about marriage promotion is not the standard he applies to his own summary of what works against family violence.

And he leaves the sharpest tension in the field lying on the floor. At 17:18 he reports that the most common reason women give for not calling police is fear of what calling will cost them, either reprisal or the loss of the household’s income, with distrust of police third. At 19:41 he recommends a set of interventions that mostly route through police and courts. Those two slides are four minutes apart and he never puts them together. A student who noticed would have asked the best question of the semester.

This is the textbook being written in public. He says at 0:18 that he was revising the violence chapter, which makes the lecture a draft of the third edition delivered live, and that is a detail for the biography that otherwise treats the textbook as a finished object.

Transcript flags. Every variant of koban, kovat, Cova, köppen, and Coppa is COVID. At 0:28 “should be halved in half year” is should be out in half a year. At 8:16 “as important networks” is and support networks. At 15:26 “vance” is violence. At 18:11 “not mr. speculation” is that is just speculation. At 21:16 “if we had to speak decline” is we have had this steep decline. At 21:57 “a life of that pain” is a life without pain. At 23:33 “West violence” is less violence. At 23:54 “just history really have a direction” is does history really have a direction. At 6:39 “pare” is parents.

‘Health Disparities and the COVID-19 Pandemic (old version)’ (Apr. 20, 2020)

0:00 Introduces himself, the subject, and the plan: define health disparities, connect them to the pandemic, and let students take it further. Gives his contact information and invites students to get in touch.

0:24 The CDC definition he says he likes: preventable differences in the burden of disease, injury, violence, or opportunities to achieve optimal health, experienced by socially disadvantaged populations.

1:06 Examples from the news, including a nursing home death, junk food, the effect of federal policy on opioid deaths, maternal mortality, and preventable cancers in Alabama.

1:57 The framework, the theory of fundamental causes, which he says sounds larger than it is. It concerns the relationship between socioeconomic status and health outcomes, and asks why general inequality stays so persistently attached to inequality in illness and death.

2:24 The improving baseline: water and sewer systems, nutrition, refrigeration and canning, medical advances, and rising education that lets people absorb health information.

3:00 Life expectancy in the United States rising from a little under 50 to almost 79 across about 120 years, with the caveat that the figure assumes nothing changes.

3:36 The theory’s answer. Income and education bring money, knowledge, prestige, power, and beneficial social connections, and the key insight is that as the health system evolves, higher status keeps converting into better outcomes. He calls it a moving target.

4:31 A study of people born in 1960, asking the odds of surviving from 50 to 85 by income quintile. Roughly 77 percent for women and 66 for men in the top fifth, against about 30 percent in the poorest fifth.

5:33 Smoking by education. A steep gradient from graduate degrees at the low end to GED holders at 36 percent, which he attributes to what education teaches and to the circles and resources it brings.

6:28 Infant mortality by race. In 1850, 340 per thousand Black infants and 217 White. By 2017, 11 and 5. Enormous progress for both, with the gap persisting.

7:22 What the theory predicts, and the routes: living conditions, nutrition, health care, occupations, and discrimination running from housing to schooling to the health system.

7:53 A qualification. Persistence so far does not make persistence inevitable, and it does mean nobody should be surprised to keep seeing it.

8:02 The opioid epidemic, and a map showing overdose deaths concentrated in Appalachia, the Upper Midwest, and other places where the industrial economy collapsed.

8:38 The route he describes: lost jobs, fewer opportunities, and opioids introduced through workplace injury, snowballing into mass addiction across prescription drugs, heroin, and fentanyl.

9:11 Background on the pandemic, the coronavirus family, and the Wuhan discovery in late 2019.

9:58 Transmission. Droplets, surfaces, hands to mouth, with two hedges: good evidence of asymptomatic spread and evidence of aerosolized particles that float rather than fall.

11:14 Why this one is bad. Novel, so nobody is immune. Highly contagious. Asymptomatic transmission, which means the infected keep moving instead of lying still, which he notes is much better for containment. Relatively high mortality, rate unknown.

12:03 The counts as he speaks: over 160,000 deaths worldwide, about 40,000 in the United States, which he says has the worst epidemic in the world.

12:26 The inversion. This epidemic began in richer parts of the country, in coastal metros, and spread from there, which is not what the previous discussion would lead you to expect.

12:54 About 80 percent of counties have cases, with a visible pattern of a dark county surrounded by lighter ones as the disease moves from a city to its suburbs.

13:30 Concentrated outbreaks. Nursing homes, where people cannot move away, infection goes unnoticed until late, and residents are the most vulnerable. Several thousand deaths already in New York State facilities.

14:04 Prisons and jails, with the largest outbreak in the country at Marion in Ohio at 1,800 infected, and another at Cook County Jail.

14:37 Factories, with meatpacking singled out. About 800 cases at a Smithfield plant in South Dakota, another at a plant in Greeley, Colorado.

15:20 The sources of disparity. Physical concentration first.

15:31 Exposed workers. Health care staff, and service workers in restaurants, groceries, and delivery, and then the general category of anyone who cannot work from home, which he contrasts with his own position recording a lecture in his studio.

16:21 Severity once infected: asthma and breathing problems, high blood pressure, compromised immunity, and age, which accumulates all of them.

17:03 A clip from Elise Gould of the Economic Policy Institute. The CDC says stay home when sick; 27 percent of private sector workers have no paid sick leave; over 30 million workers; the United States is one of two OECD countries without a federal mandate; taking time off can mean missed bills and calling in sick can mean dismissal; the workers without it are concentrated in restaurants, hotels, transportation, and tourism, with under half of leisure and hospitality workers covered and 30 percent of the bottom tenth of earners.

18:38 His gloss: the law does not require paid sick leave, which puts workers and everyone around them at risk.

18:55 Bureau of Labor Statistics data on who can work at home, by occupation, education, and race. Managers, professionals, sales, and office workers far more than transportation, production, service, and construction. Over half of college graduates against about 5 percent of those without a high school degree. Asians and Whites most able.

19:51 The consequence: the instruction to stay home opens an immediate inequality between who complies and who goes out and takes the risk.

20:16 The Black-White difference in COVID deaths, with the caution that these are very early days for the data.

20:42 Five jurisdictions reporting deaths by race, Washington, Chicago, Louisiana, New York State, and New York City, all showing Black residents dying at higher rates.

21:14 The explanation through the theory. Prior living conditions had already produced worse baseline health, and he notes these are deaths rather than infections, so most of the people counted had an existing condition.

21:38 Education and discrimination in access to jobs with benefits, flexibility, and the option to work from home, all unequally distributed.

21:52 Employer policies, including sick leave.

22:05 The health system, both the care people did or did not get before, since many have no insurance, and the hospital and doctor they reach once infected.

22:42 The close. This is a taste of one disparity, and the pattern shows the kind of thing to look for as the epidemic unfolds.

Thoughts

Three lectures in eighteen days, and the register changes each time.

April 3, on epidemiology, where he twice said he lacked expertise: a montage of presidential clips, the phrase total lie, and an instruction to enrolled students about which officials to believe. April 6, on family violence, squarely his field: no politics at all. April 20, on health disparities, also squarely his field: the most structurally radical content of the three, and not one word of partisan heat.

The lecture argues for federal paid sick leave mandate. Universal health coverage. That racial disparity in employment and benefits is discrimination rather than sorting. Those are contested political positions with serious opposition, and he advances all three inside a framework any medical sociologist would recognize as standard. A student leaves this lecture with a more consequential set of policy commitments than he leaves the April 3 lecture with, and without ever being told what to think about anybody.

Except once. At 17:03 he stops teaching and plays three minutes of advocacy video from the Economic Policy Institute, a labor-aligned think tank, and thanks the presenter afterward. He does not tell students what EPI is or who funds it. That is the same move as the April 3 montage with a fig leaf: a professor who selects an advocacy clip has made the argument, and the fig leaf is that somebody else’s mouth delivered it.

Cohen might say the paid sick leave numbers are correct, checkable, and drawn from BLS, and that this is what distinguishes EPI here from the Institute for Family Studies work he attacks. That defense holds for the statistics. It does not cover the framing, the policy conclusion, or the decision to let an outside organization narrate three minutes of a course. Anybody who thinks he is right about IFS should still want him to say out loud that he applies a different procedure to a think tank whose conclusions he shares.

Now the part that would matter to him more, which is that he undersold his own evidence twice by skipping the measurement work he demands from everyone else.

The infant mortality slide at 6:28 is presented as a gap that persisted. Run the numbers he put on the screen. In 1850, 340 against 217, a ratio of about 1.6. In 2017, 11 against 5, a ratio of about 2.2. The absolute gap collapsed by a factor of thirty and the relative gap widened by about forty percent. That is a stronger result than the one he stated and it is the result his theory predicts, since fundamental causes is a claim about relative advantage surviving absolute improvement. He had the better fact on the slide and told the students the weaker one.

The race and mortality slide at 20:42 has the same problem in the other direction. Those spring 2020 jurisdictional reports were crude counts and shares. The Black population in the United States is younger than the White population, and COVID mortality rose steeply with age, so age-adjusting those figures makes the disparity larger, not smaller. He shows unadjusted numbers to introductory students without saying they are unadjusted. This is the man who spent a decade arguing that a claim arrives with its instrument attached, who three weeks earlier corrected his own slide on live video when he noticed the American outbreak curve was an artifact of treating a continent-sized country as one unit. Here the caveat would have strengthened his case and he skipped it.

There is a defensible explanation, which is that this is twenty-three minutes for an introductory course and compression is the job. It is also the pattern from the family violence lecture, where the interventions section ran on some promise and mixed results with no effect sizes. His rigor is a function of whether he is prosecuting or teaching, and it runs in the direction you would not predict.

Two things here have aged well.

The prediction at 12:26 is his best public call of the pandemic and nobody has credited him with it. He notes that the epidemic inverted the usual gradient by starting in rich coastal metros, and then argues from the theory that the gradient will reassert through exposure, comorbidity, and treatment access. That is what happened. By the later waves the mortality burden had moved down the education and income distribution and stayed there. Set it beside the homicide prediction from April 6, which was wrong, and the difference is instructive: this one applies a well-tested theory to structural position, and that one was a bespoke exposure model built for the occasion. He is reliable when he reasons from the frame and less so when he improvises.

And the outbreak trio at 13:30 is the analytic core of the lecture, delivered so fast that most students would have missed it. Nursing homes, prisons, and meatpacking plants are three institutions with nothing in common except that they hold people in place who cannot leave. He names confinement as a variable, and the mechanism operates identically whether the confinement is medical, penal, or economic. That is a better sentence about the American pandemic than most of what was published that spring.

Two small things for my file.

At 16:08 he interrupts a list of workers who cannot work from home to note that he is delivering the lecture from his own studio. That is one clause, unforced, in front of students who mostly could not have caught it. It is the disclosure ethic operating without an audience to perform for, and it is the best evidence in this file that the ethic is a habit.

And the opioid map at 8:02 is the seed of published research. His 2024 work on marital status and deaths from alcohol, overdose, and suicide grows from exactly this material, and here he is teaching it to undergraduates four years earlier. The initial biography treats his research program and his teaching as separate strands. This is the place they are visibly the same strand.

YT transcript flags. Every instance of köppen, koban, covin, cope at, and Cova is COVID. At 0:00 “Philip Conn” is Cohen. At 5:52 “37% of them smoke” for graduate degrees is 3.7 percent; the gradient is roughly tenfold and the transcript flattens it to nothing, so this one will wreck the sentence if you quote it. At 4:48 “income Phipps” is income fifths. At 6:39 “1853 forty” is 1850, 340. At 6:49 “217 of the white babies” needs the per-thousand restored. At 8:08 “rotavirus epidemic” is coronavirus. At 10:39 “spread among people our age” is probably aerosols, given what follows. At 11:31 “de jure disease” is the disease. At 13:22 “the poor Daria’s first” is the poor areas first. At 19:41 “they say they couldn’t work from home” is could. At 21:22 “were already half conditioned” is have already conditioned. At 21:29 “that Charlie just showed” is that chart. The named facilities are correct: Marion Correctional Institution in Ohio, Cook County Jail, Smithfield in South Dakota, and the Greeley plant in Colorado. Elise Gould of the Economic Policy Institute is correctly identified. The 160,000 global and 40,000 American death figures place the recording on or about April 19 or 20, 2020, which matches the upload date.

‘COVID-19, preprints, and the information ecosystem’ (Jun. 8, 2020)

0:00 Introduces the talk on COVID-19, preprints, and the information ecosystem, gives his contact information, and invites people to get in touch.

0:13 The cartoon he opens with: you judge how bad things are by which specialist appears on the news. A volcanologist means trouble, an archaeologist means relax, and for this crisis a biologist sits close to the volcanologist end.

0:42 The ecosystem model, borrowed from ecology. The foundational layer holds government agencies, news organizations, and scientific publishers.

1:08 The access layer. Devices, subscriptions, paywalls, and language.

1:21 The marketplace layer, where demand is generated: people’s information needs, the production and movement of information, and its use.

2:01 The influencer network, running on individual decisions about whom to trust, both for one’s own needs and for the people relying on you.

2:23 The impact layer, and his summary that the whole thing is large, complicated, and hard to grasp.

2:45 Peer review has not always worked the way we assume.

2:53 Einstein and the Physical Review in 1936. He was offended to learn the editor had sent his paper out for review, the paper turned out to need correction, and he withdrew it rather than authorize anyone else to see it.

3:28 The general case before the mid-twentieth century: editors decided what to publish and consulted experts at their own discretion. Gatekeeping and authority existed; editors held them.

3:51 The origin story he offers. After the Second World War, facing constricted research funding, academics saw that they could lose control of the process to government agencies, and the mystique of anonymous peer review arose from the insistence that only academics could decide what was trustworthy.

4:31 Watson and Crick on the double helix was not peer reviewed. The editor decided to publish it.

4:38 The cycle of scientific work: search and discovery, developing ideas, designing projects, gathering data, analyzing, publishing, and the results returning to the pool.

5:22 A qualification. Nothing published is right forever, things get superseded and added to, and the process expands rather than circling.

5:42 Where journal peer review sits, between analysis and publication.

5:58 Publishers coordinate it using their own status criteria, who the researchers are and where they work, alongside the reviewers’ judgments, and the process is a black box conducted in secret.

6:17 Other evaluation points in the system: awards, grant funding decisions, and hiring.

6:51 Where open science advocates have intervened. Preregistration, which lets designs be reviewed before results exist.

7:23 Sharing materials and methods in repositories.

7:34 Publication beyond journals, including preprints, news media, and social media.

7:50 Two definitions of a preprint. The finished draft sent to a journal before review, and the accepted manuscript before typesetting.

8:25 The main servers. arXiv from math and physics in the 1990s, now over a million papers; bioRxiv and medRxiv; and SocArXiv in the social sciences, which he directs.

9:11 The pandemic explosion. A paper by Fraser and colleagues counts 16,000 COVID papers from January through April, of which 6,000 were preprints.

9:36 The follow-up finding that many were published in journals quickly with few substantive changes, which he reads as evidence they were in decent shape when posted.

9:46 Some were shared tens of thousands of times and shaped the discourse. Some turned out to be wrong, and the controversy raised the question of whether people should be reading and sharing them.

10:08 The medical sensitivity problem, and the caution notice bioRxiv and medRxiv now carry: preliminary reports, not certified, not to be relied on for clinical practice, not to be reported as established.

10:38 How journalists handle them anyway, using their own reliability indicators: who the researchers are, where they work, what outside experts say, their own reading, and newsworthiness against the cost of waiting weeks or months.

11:21 Preprints that became public knowledge. Excess deaths exceeding the official COVID count, indicating undercounting. The estimate that an earlier lockdown would have saved tens of thousands of lives, from medRxiv, on the front pages of the Washington Post and the New York Times.

12:01 A third case, the men-doing-more-housework finding, published by the Council on Contemporary Families, which does its own vetting outside both journals and preprint servers.

12:19 Retractions from the New England Journal of Medicine, The Lancet, and Annals of Internal Medicine, alongside a withdrawn bioRxiv preprint that had claimed a human-made origin for the virus on faulty genetic analysis.

12:49 His summary. Errors appear in preprints and in published papers, different systems of trust and authority are at work, and everything is moving fast.

13:00 The system runs on trust, normally formalized into predetermined authorities, prestigious journals and big universities, so that status, reputation, and legitimacy do the work.

13:27 A constant churn of deciding whom to believe, of which peer review is one part.

13:33 His verdict on the crisis. Preprints succeeded at accelerating research and getting knowledge out, and they taxed the system of trust, making explicit what is usually implicit.

13:54 Transparency as the emerging criterion. Whether researchers share data and code is now one of the ways we decide whether to believe them.

14:12 What a researcher can do. Publicly endorse work and share only what you think is reliable.

14:29 The Plaudit tool, a browser extension that lets anyone with an ORCID endorse a piece of scholarship. He has endorsed the Watson and Crick article, and he endorses a paper on pandemic dog adoptions when sharing it on Twitter. If people already trust him, his endorsement carries something.

15:14 Understand the ecosystem you work in and allocate your labor toward organizations that support open science: post preprints when ready, and decide carefully which journals to edit, review, and write for.

15:53 Closing reading list. The National Academies’ Open Science by Design, a resource for journalists on covering preprints, and his own report on scholarly communication in sociology.

This is the most theoretically interesting thing of his I have read, and he does not seem to notice.

The historical section from 2:45 to 4:38 is an argument that peer review is roughly seventy years old, that it arose from a professional turf defense rather than an epistemic discovery, and that the two most consequential papers he names, Einstein’s 1936 submission on the gravitational wave and the Watson and Crick’s double-helix paper, went through editors, he says, rather than referees. The claim is broadly correct and it is more radical than anything in his blog. Anonymous peer review is a mid-century institution built when academics realized that if they did not control quality certification, somebody with appropriations authority would.

Einstein’s 1936 submission to Physical Review was sent to an anonymous referee, whose criticism led Einstein to withdraw it before a revised version appeared as “On Gravitational Waves” in 1937.

Notice what that does to his own program. If the mystique of peer review was a professional guild move dressed as epistemology, then the transparency criterion he names at 13:54 has to answer the same question about itself. Is checkability a discovery about how knowledge works, or is it the certification standard preferred by the faction that finds it easy to meet? Quantitative researchers with code and data can satisfy it at the cost of a day’s file management. Ethnographers, historians, and theorists cannot satisfy it at any cost. A standard that redistributes authority toward the people who already hold the tools is exactly what his own history predicts would emerge from a crisis in an existing certification regime. He tells that story about the generation before him and does not turn it on himself. That is the essay somebody should write, and he handed over the materials.

The Plaudit passage at 14:29 is the strangest four minutes in this file. Follow what he proposes. Peer review is a black box run by publishers on status criteria. The replacement he demonstrates is a browser extension where he clicks endorse and then says that if you already trust him for other reasons, his tweet might mean something to you.

That is status criteria with the intermediary removed. The anonymous referee at least did not benefit personally from the paper’s reception. A named endorser trades on his reputation, accrues from being seen to endorse, and is embedded in a network of people he will need favors from. He replaced a black box with a visible reputation market and described the visibility as the improvement, which it partly is, and never addressed the thing that makes reputation markets go wrong, which is that they reward endorsing the people who can endorse you back. A man who spent his career pointing out that prestige networks distort what gets believed just proposed prestige networks as the correction.

He also endorsed Watson and Crick with a browser extension in 2020 and offered it as a demonstration. The joke is good. It also quietly concedes that the tool has no way to distinguish an endorsement that costs the endorser something from one that costs nothing.

Two things here are sharper than his usual work.

The Fraser finding at 9:36 is the best empirical defense of preprints anybody made that year. If work posted before review appears in journals shortly after with few substantive changes, then review added little to those papers, and the months of delay bought the reader nothing. That is a testable claim about the value added by an institution, of the kind he demands from everyone else, and it comes with a citation rather than an assertion.

And the 12:19 slide is the honest move. He puts preprint failures next to retractions from the New England Journal of Medicine, The Lancet, and Annals of Internal Medicine, which is the right comparison and the one an advocate would have skipped. The Surgisphere retractions from The Lancet and the NEJM had happened days before this recording, and both journals had cleared the same peer review he was being asked to defend as the safeguard preprints lack.

Two weaknesses.

The ecosystem model from 0:42 to 2:23 does no work. Five layers, no argument, no prediction, and nothing later in the talk uses it. Strike it and the talk loses nothing, which is the test he would apply to somebody else’s framework.

And the accuracy question goes unanswered. He says at 9:46 that some preprints turned out to be wrong and at 12:49 that some published papers did too, and he never tells you the rates. Whether preprints are wrong more often than reviewed papers is answerable, at least roughly, and it is the question anyone watching would want settled. He has the data literacy to know that the presence of errors on both sides is compatible with any ratio between them. This is the counting he does not do when the answer might be inconvenient, and it is the same pattern as the interventions section in the family violence lecture.

For my file, this is where the archive stops being an administrative project and becomes an argument about what certification is. The biography covers SocArXiv through its founding, the prize, and the AI crisis, and the crisis reads as an accident that befell a working institution. This talk shows the position was always there: certification is a historically contingent arrangement that some professionals built to keep control, and preprints tax the trust system by making the implicit explicit. Six years later he wrote a policy conceding that his moderators cannot state the rule and will use human judgment. The taxing of the trust system turned out to have no floor, and he described the direction of travel in June 2020 while assuming it stopped somewhere convenient.

One line to carry forward: at 13:33 he says the crisis made explicit what is usually implicit. That is the whole of his career in eight words.

Transcript flags. Kobe, kovat, COBIT, koban, Cobin, and club in nineteen are all COVID-19. At 8:48 “social kind” and “sai archive” are SocArXiv. At 8:26 “the major print print” is preprint. At 12:09 “bedding” is vetting. At 12:53 “systems of crust” is trust. At 11:52 “a pre current published” is a preprint. At 14:32 “the plaudit tool” is Plaudit, correctly named. “Orchid ID” is ORCID. At 15:44 “review right for” is review and write for. The Fraser paper is Fraser et al. on the preprint literature during the pandemic; the version circulating in mid-2020 was a preprint.

A smoothed transcript: 2:45, on peer review as a historical artifact. “Now, it’s important to realize, when we talk about peer-reviewed science and the way we assume science works, that it has not always been this way. There’s a famous interaction between Albert Einstein and the journal Physical Review, where he was extremely distressed to discover that the editor of the journal had sent his paper to somebody else to review it before deciding to publish it. It turned out that this particular paper was wrong and had to be corrected. But Einstein was offended, incensed: on the basis of this incident, I prefer to take my paper elsewhere. He hadn’t authorized it to be shown to anybody else.

“So that’s 1936. Peer review, and especially anonymous peer review, was completely offensive to Einstein. Of course, he’s Einstein. But it’s the general case that 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. Which doesn’t mean there was no gatekeeping, and it doesn’t mean there was no authority and no expert review. It was editors who were in charge of it. They decided who would review what, and they were the gatekeepers.

“Then, 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. That’s really when we see the mystique of peer review come up, where academics insisted that only academics could decide what was valuable. That’s when we got this idea of the anonymous peer review system, and especially the idea that this was essential for science to be considered trustworthy.

“There’s a good example. The Watson and Crick paper on the double helix structure of DNA. Not peer reviewed. The editor just decided to publish it.”

5:42, on where review sits and what else does the same work

“The peer review process as we know it in journal publishing happens at that point between the analysis of the data and the publication of the results. That’s when we bring in these expert peers, sometimes anonymous. The effort is usually coordinated by publishers, especially journal publishers, and they’re going by their own status criteria: who are these researchers, are they reputable people, where do they work. And then they’re also consulting their peer reviewers. The process is very black box. We don’t know exactly how it works. It all happens in secret.

“Now, there are other places where evaluation goes on in this system that we don’t think of as peer review in that specific sense. After results are published they can be elevated; they can be given awards. We have grant funding agencies that evaluate research ideas and whether or not to fund them. We have people who decide who’s going to get jobs, and which are going to be the important jobs. So there’s a lot of peer evaluation that goes on in the system besides that moment of publication we focus on.”

7:50, what a preprint is

“There are two main definitions of preprints that are relevant. One is the finished draft of scholarly work, not yet peer reviewed. Think of the version of the work that scientists are sending off to a journal to be reviewed. Another definition is the version after it’s been reviewed and accepted, but before it’s been typeset and published by the journal. People sometimes share that, and that can also be called a preprint. So preprints are essentially before the typeset version that comes from a journal, at some stage of development.”

“The major servers: arXiv, the first big one, out of math and physics, started in the 1990s, now well over a million papers. bioRxiv and then medRxiv in the life sciences and medicine. In social science, SocArXiv, which I direct. And a number of other discipline-specific archives working on similar principles, archiving and disseminating preprints more rapidly and openly than traditional journals.”

9:11, the pandemic volume, and the Fraser finding

“In the current pandemic environment there’s been an explosion of research, and much of it in preprints. This paper by Fraser et al. finds that in the first four months, January through April, of sixteen thousand papers published having to do with COVID-19, six thousand were preprints. So a huge volume of work, pre-peer-review, coming out available to the public.

“They also did some follow-up and found that a lot of these studies are published quite quickly in journals with few substantive changes, indicating the preprints were in pretty good shape when they were posted. Some of these preprints have been shared thousands of times, or tens of thousands of times, and really helped shape the discourse on COVID-19. Some of them have turned out to be wrong, and there’s been controversy with some of them, and that’s part of what’s raised the issue of whether we should be reading preprints and talking about them and sharing them.”

10:08, the warning label, and what journalists do instead

“Because we’re dealing with medicine, and people are very sensitive about it, you’re going to give clinical advice, you can talk about specific medications, doctors might be looking at these. medRxiv and bioRxiv, which are run by the same organization, are putting this caution right on their homepage: preprints are preliminary reports that are not certified, they should not be relied on to guide clinical practice, and they should not be reported as established information.

“But preprints are all over the news. So how is that happening? Journalists and news organizations are using their own indicators of reliability. Who are these researchers, where do they work. They might talk to other experts who are not involved in the research and say, what do you think, is this reasonable, should we write a story about this. They have their own assessment. The science journalists and reporters and editors have their own expertise, and they can read and decide what to trust. And then there’s the newsworthiness. How important is it? Do we need to get this out there right now, instead of waiting weeks or months for it to be published in a journal?”

11:21, preprints that became public knowledge

“Some of the things you’ve probably heard that are now established knowledge in the public sphere around COVID-19 have come from preprints. For example, the estimates that the number of deaths going on in the country far exceeds the number counted by the official statistics, indicating we were undercounting infection-related deaths. Page one of the Washington Post. Page one of the New York Times: if we had done the lockdown sooner, tens of thousands of lives would have been saved. That’s also from a preprint, published on medRxiv. So these things have been tremendously influential, and lots of people know about them.

“Here’s another one. Men are doing more housework, say researchers. This one is not on a preprint server. It’s published by the Council on Contemporary Families, which does its own vetting of unpublished research. They publish it themselves, but not in a journal, and not with a formal peer review process.”

12:19, errors on both sides

“Now, some of these preprints have been wrong. But then again, some of the published articles have been wrong. Here are some retractions from major journals, the New England Journal of Medicine, The Lancet, Annals of Internal Medicine, which turned out to be wrong and had to be retracted after getting a lot of attention. I include here one bioRxiv preprint that was withdrawn after suggesting that the coronavirus was made by humans, because of a faulty analysis of the genetics of the virus.

“So there have been errors in preprints and among published papers, and we have different systems of trust and authority, and the whole thing is moving very rapidly.”

13:00, the passage the essay turns on

“So what do we trust? What do we believe we know? This whole system runs on trust. Normally it’s formalized, and we have authorities that are predetermined to be things we should trust: Nature and other important journals, and big universities, and so on. We use the status and the reputation and the legitimacy of the people who did the study, and the people who published the study, and the people who report on the study. There’s a constant churning process of deciding who we’re going to believe. Peer review is just one part of that.

“This crisis has really shown that preprints have been tremendously successful at accelerating the pace of research and the efficiency of getting that knowledge out there. And it has really taxed the system of trust, and made more explicit part of what’s usually implicit. One of the things that we now look at is transparency. One of the ways that we evaluate whether or not we should believe a piece of research is: are they sharing their data and code and so on. Have they made their process transparent. If they have, that indicates that they’re more trustworthy.”

14:12, the replacement he proposes

“I want to talk a little bit at the end here about what you can do if you are a researcher, who works not just on this but on anything. One thing you can do is take steps to publicly endorse work, and make sure you only share work in your networks that you think is reliable. For example, the Plaudit tool. I really like it. It’s a browser extension you can install, I use Chrome, where if you have an ORCID ID you can endorse any piece of scholarship. You can see here I have endorsed the Watson and Crick article. And also, if I’m going to share a paper on Twitter, here’s a paper I shared about dog adoptions during the pandemic, I’m giving it my endorsement. If you trust me already for other reasons, then my tweet might mean something to you.

“So if you’re a researcher, I just urge you to spend some of your time doing the work of publicly endorsing things that you think are good.

“I also think it’s important for us to try to understand this ecosystem that we work in, and then to make decisions about how we allocate our own labor to support those organizations and institutions that support open science and support our values. To practice open science ourselves. To publish preprints when they’re ready, to help accelerate and transmit knowledge out to other researchers and the public. And to decide who we’re going to edit journals for, review for, write for. To think of ourselves as part of this system, and try to push the system in a positive direction while we do our work.”

Notes on the editing

Removed “sort of,” “you know,” “um,” and repeated false starts. Repaired the speech-recognition damage: Kobe, kovat, COBIT, koban, Cobin, and club in nineteen all became COVID-19; “social kind” and “sai archive” became SocArXiv; “the major print print” became preprints; “bedding” became vetting; “systems of crust” became trust; “a pre current published” became a preprint; “review right for” became review for, write for; “orchid ID” became ORCID. Joined sentences he left hanging. Nothing was added, no order was changed, and where a phrase was ambiguous I kept his looser version rather than improve it.

At 3:11 he delivers Einstein’s line as though quoting: on the basis of this incident, I prefer to take my paper elsewhere. The documented letter of July 27, 1936, reads that on the basis of this incident he prefers to publish the paper elsewhere. His rendering is close and not exact. If you are quoting Cohen, use his words as smoothed above. If you are quoting Einstein, use the letter, which is reproduced in the gravitational-wave histories I sent with the peer review essay. Do not let the two blur, since the whole point of that passage is what counts as a faithful record.

One flag on the Council on Contemporary Families passage at 12:01. He presents CCF as a third model, an organization that vets and publishes outside both journals and preprint servers, without saying whether he has any relationship to it. CCF has published his work and he has been associated with the organization.

What I cut and why. The ecosystem model (0:42 to 2:38) is five labeled layers with no argument attached, and nothing later in the talk uses it. The virus background (9:11 area, in the other lecture’s language) duplicates material he covered better on April 3. The closing reading list is three citations, all recoverable from the slides at https://osf.io/5mue7/. The retraction slide carries numbers that never made it into the audio.

If you want a verbatim passage rather than a smoothed one, the two places where the exact wording is likely to matter are 3:51, the after-World-War-II turf-defense sentence, and 13:33, the taxing of the system of trust.

In the early summer of 1936, Albert Einstein (1879–1955) and his assistant Nathan Rosen (1909–1995) sent a paper to Physical Review arguing that gravitational waves do not exist. The editor, John Torrence Tate (1889–1950), had published Einstein before without sending the work to anybody. This one struck him as suspicious. He mailed it to Howard Percy Robertson (1903–1961), a relativist at Princeton, who read it and sent back ten typed pages explaining that the authors had made a mathematical error.

Tate forwarded the report and asked for a response. Einstein wrote back that he and Rosen had submitted the manuscript for publication and had not authorized anyone to show it to specialists beforehand, that he saw no reason to address the comments of Tate’s anonymous expert, and that he preferred to publish elsewhere. Tate answered that he could not print a paper the author would not let his editorial board see. Einstein took it to the Journal of the Franklin Institute and never submitted to Physical Review again.

What happened next is the part that gets left off.

Rosen went to the Soviet Union. A young Polish physicist named Leopold Infeld (1898–1968) arrived at Princeton to replace him. Robertson, back from a sabbatical in Pasadena and by all accounts sociable, befriended Infeld, got him talking about what Einstein was working on, and expressed doubt. The two of them went through the manuscript together. Infeld took the objection to Einstein. Einstein revised the paper before returning his galleys, reversed its conclusion, changed the title, and thanked Robertson in a footnote for friendly assistance in clarifying the original error. He never learned that Robertson had been the referee.

The correction arrived. It traveled through a man who happened to be in Princeton that year and happened to like talking to graduate students. The institution built to carry it had been refused at the door.

Philip Cohen tells the first half of that story in a lecture he recorded in June 2020 on preprints and the information ecosystem, and he uses it to make a claim that is larger than his listeners probably noticed. He follows it with Watson and Crick on the double helix, which no referee read; the editor decided to publish. Then the general point. Before the middle of the twentieth century, journal editors decided what to publish and consulted experts when they felt like it. Gatekeeping existed. Authority existed. Editors held both. And then, he says, after the Second World War, facing the prospect that research funding decisions might be made by government agencies without academic control, academics insisted that only academics could judge what was true, and the mystique of anonymous peer review was born.

That is an argument that the central credentialing procedure of modern science is a professional turf defense that later acquired an epistemology. He states it in ninety seconds, in a lecture for undergraduates, and moves on.

The history is right in structure and wrong in date, and the correct date makes his case stronger.

Melinda Baldwin, a historian of science who teaches at his own university, published the definitive account in Isis in 2018, “Scientific Autonomy, Public Accountability, and the Rise of ‘Peer Review’ in the Cold War United States.” External referee reports were an optional feature of journal editing and grant making through the nineteenth century and most of the twentieth. The idea that refereeing is a requirement of scientific legitimacy arose in the Cold War United States, and the decisive moment came in the 1970s rather than the 1940s. American science was under sustained congressional attack over how it spent public money. Scientists needed to appear accountable to the people funding them while retaining control of funding decisions. The term peer review, which had been migrating out of grant administration, was pressed into service as the thing that made science credible as such. Nature did not require formal review of everything it published until 1973. The Lancet came later still.

So the sequence is: an existing certification regime comes under external pressure, the profession selects a standard that answers the pressure while keeping the decision rights inside the profession, and the standard is then described as constitutive of knowledge rather than as the settlement it was.

Cohen tells that story about the generation before him. At the end of the same lecture he proposes the successor standard and does not turn the story on it.

The passage runs about twenty seconds. The crisis has taxed the system of trust, he says, and made explicit what is usually implicit, and one of the things we now look at is transparency. Are they sharing their data and code. Have they made their process inspectable.

Ask of that standard the question Baldwin asks of the last one. Who can meet it, at what cost, and who decides whether it has been met.

A quantitative researcher working with public survey data can meet it for the price of a day’s work. That figure comes from Cohen’s own comment section, where a sociologist who had published in a journal with a mandatory policy reported that assembling his first replication package took a full day despite tidy file management, and that most people default to available upon request because collation for a stranger is labor rather than concealment. One day, once, and then the habit is cheap. Every subsequent paper is a matter of directory hygiene.

An ethnographer cannot meet it at any price. The field notes are the instrument, the consent forms usually forbid release, and even redacted material carries the risk of deductive disclosure in a neighborhood where everyone knows who the researcher followed for two years. An archival historian cannot meet it either. The archive is the data, and it sits in boxes in a building, often behind permissions granted to individuals, and citation to box and folder is the most a reader will ever get. An interview researcher who promised confidentiality is barred by the promise. A demographer working inside a Census research data center is barred by federal law. A theorist has nothing to deposit.

That is a standard whose compliance costs run from one day to infinity depending on what kind of evidence a field is built out of, and it arrived in a discipline whose quantitative wing had spent thirty years gaining ground on its qualitative wing.

Political science ran the experiment first and the record is public.

The American Political Science Association wrote data access and research transparency into its ethics guide in 2012. In October 2014, twenty-seven journal editors signed a joint statement committing to require, by January 15, 2016, that cited data be available in a trusted digital repository at the time of publication, with exceptions for restricted material at editorial discretion. The following autumn a petition circulated asking for a delay. It gathered more than seven hundred fifty signatures within days and close to a thousand within a week, including senior figures and sitting editorial board members.

The signatories did not say they wished to be believed without evidence. Their complaint was about discretion. The rules were clear enough for people with datasets and unspecifiable for everyone else, which meant that ten editors could produce ten interpretations, and a scholar planning a project could not know whether a given kind of fieldwork would close off a set of journals. The standard did not tell them what compliance was. It told them who would decide.

The association’s past, current, and future presidents answered with a joint statement supporting implementation, and the ground they chose is the one Baldwin would predict. Their legitimacy as scholars speaking to the public, they wrote, rests partly on maintaining common standards for evaluating evidence-based knowledge claims. External legitimacy, purchased with an internal certification rule, in the middle of a decade when the social sciences were being asked what their findings were worth. That is 1975 with different equipment.

The deadline held at some journals and slipped at others, and the association’s qualitative and multi-method section opened a multi-year deliberation to work out what transparency could mean for research that cannot deposit anything. The deliberation produced sensible documents. The distributional question it was convened to settle has not been settled anywhere.

Now the counter-case.

Showing that a standard has winners does not show that it is false. Peer review’s origins in a guild’s anxiety about congressional appropriations tell you nothing about whether referees catch errors, and Robertson caught a real one. In the same way, the fact that checkability favors people with spreadsheets does not make checkability arbitrary. There is a difference between a claim a stranger can test and a claim he must take on the author’s word, the difference is not a matter of taste, and every discipline that has looked hard at its own literature in the past fifteen years has found things that a stranger’s inspection would have caught earlier.

The second point. If your evidence cannot be inspected, that is a fact about your evidence. It may be an unavoidable fact, forced by ethics and by law, and it still bears on how much weight a reader should place on your conclusion. A discipline that answers the transparency demand by saying its methods are different has said something true and has also said trust me, which is the position everyone occupied before 1970 and which produced a literature nobody can now audit.

And the third point runs against Cohen’s own side. Compliance costs fall hardest on people with the least slack. A day of file management is a day for a tenured professor with course relief. It is a week that does not exist for an adjunct teaching five sections, and the standard has no way to see the difference.

Peer review answered an accountability crisis with a rule that kept the deciding inside the profession. The test of any successor is whether it moves the deciding outward, and this is where Cohen’s program deserves better than the easy debunking, because on its face it does exactly that. A replication package is not an opinion delivered by a credentialed stranger to an editor in confidence. It is a public artifact. Anyone can download it. The authority to check has been handed to whoever wants it. That is a real answer to the guild problem, and nobody else in his discipline proposed one.

Except that the thing that got adopted is not checking.

What his audits count is whether a package exists. He said so in 2020, that he had assessed them at a glance and had not tested anybody’s code, and he has never claimed otherwise. The proportion of quantitative articles in the discipline’s flagship journal arriving with packages went from four of fifteen to twelve of eighteen across five years, and what that number measures is documentation. Nobody has run the code. Nobody has recomputed the tables. The public who received the authority to check has not exercised it and shows no sign of intending to.

Which leaves a standard that certifies compliance with a documentation regime, adjudicated by journal editors and by a professor clicking links on a December afternoon. The outward transfer of authority is real in principle and dormant in practice, and while it sleeps, the rule functions the way Baldwin’s model says these rules function: as a gate, operated by people whose evidence happens to fit through it, carrying the language of universal accessibility.

Then documentation became free to produce.

In the winter of 2025 Cohen’s archive began receiving papers with everything in place. Abstracts, real citations to real work, competent models, code in public repositories, and disclosure statements listing every tool used to generate the analysis and draft the text. One of them he described in detail because it would have been accepted a few years earlier. Dull, unoriginal, thin in its literature review, empty where it claimed a contribution, and immaculate on every checkable dimension.

A standard that certifies the existence of an audit trail becomes worthless at the moment audit trails cost nothing. That is what happens to any proxy once the thing it stood for stops being expensive.

The policy his committee published in March 2026 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. A small group of trusted individuals, embedded in a community, deciding what to publish on criteria they cannot fully specify, sometimes wrong, answerable to nobody but their own standing.

That is John Tate at his desk in 1936.

The narrative Cohen told his students in June 2020 ran from editorial discretion, through a guild’s invention of peer review under political pressure, to transparency as the successor standard. He was describing a sequence he thought he was at the end of. Six years later the third term collapsed and the first term returned, and the whole circuit closed inside a single career.

The counting moved a journal that three years of committee work had failed to move, and a literature with code in it is better than a literature without. What the history says is narrower and harder to shrug off. Every certification standard is a settlement about whom to believe, arrived at by people with something at stake, under pressure from outside, and every one of them gets described afterward as a discovery about the nature of knowledge. He can see this clearly when the subject is the generation that trained his teachers. The materials for seeing it about his own generation are on his own slides.

Robertson’s ten pages never persuaded Einstein of anything. What persuaded him was a man he liked, at his own institution, talking to his assistant over the course of an autumn. Every certification regime since has been an attempt to make that transmission work without depending on who happens to be in Princeton that year. None has managed it. Each has had to decide, in the attempt, whose evidence counts as evidence.

Notes

The Einstein episode. Daniel Kennefick’s research, first published in Physics Today in September 2005 and developed in Traveling at the Speed of Thought (Princeton, 2007), is the source for Robertson’s identification as the referee. The submission was “Do Gravitational Waves Exist?”, sent to Physical Review in June 1936; Tate wrote to Einstein on July 23 with the referee report; Einstein replied on July 27. Verbatim quotation used: “comments of your anonymous expert,” from that letter, which is reproduced in full at https://arxiv.org/pdf/1609.09400 and https://arxiv.org/pdf/1610.08803. Tate’s reply of July 30, that he could not publish a paper the author was unwilling to have shown to the editorial board, is in the same sources. The Infeld and Robertson sequence, the reversal of the conclusion, the retitling to “On Gravitational Waves,” and the acknowledgment thanking Robertson are documented in both arXiv histories and in https://www.science.org/doi/10.1126/science.347.6226.1085. The detail that Tate had published Einstein, Podolsky and Rosen in 1935 without refereeing is from https://www.discovermagazine.com/the-sciences/einstein-vs-physical-review.

The history of peer review. Melinda Baldwin, “Scientific Autonomy, Public Accountability, and the Rise of ‘Peer Review’ in the Cold War United States,” Isis 109, no. 3 (September 2018): 538–558, https://www.journals.uchicago.edu/doi/abs/10.1086/700070, with a copy at https://gwern.net/doc/statistics/peer-review/2018-baldwin.pdf. Her earlier “Credibility, Peer Review, and Nature, 1945–1990,” Notes and Records 69 (2015), and John Burnham’s 1990 JAMA piece on the evolution of editorial peer review, are the standard companions. Nature dated at 1973 and The Lancet later; the term appearing in JAMA in 1970; Colin Norman’s “Peer-Review under Review” in Nature in 1975. She is on the faculty of the University of Maryland’s history department.

DA-RT. The joint statement of twenty-seven political science journal editors, October 2014, with the January 15, 2016 deadline and the restricted-data exception. The APSA ethics guide change dates to 2012 and the initiative to an ad hoc committee formed around 2010. Petition coverage at https://www.insidehighered.com/news/2015/11/16/political-scientists-seek-delay-transparency-standards-publications, with signature counts at https://dialogueondartdotorg.wordpress.com/faqs/ (more than 750 as of November 8, 2015) and https://politicalbehavior.wordpress.com/2015/11/11/statement-on-da-rt-and-the-jets/ (almost 1,000 by November 11; the organizers were Hirschmann, Htun, Mansbridge, Thelen, Wedeen, and Wood). Verbatim quotation used: “common standards for evaluating evidence-based knowledge claims,” from the joint statement of the APSA past, current, and future presidents, November 24, 2015, linked from https://www.dartstatement.org/2014-journal-editors-statement-jets. The Qualitative Transparency Deliberations are at https://qualtd.net/page/about/, and the discretion objection I paraphrase is from a contributor post at https://www.qualtd.net/viewtopic.php?t=67.

Cohen material. The June 2020 lecture, “COVID-19, preprints, and the information ecosystem,” slides at https://osf.io/5mue7/, for the Einstein and Watson-Crick examples, the postwar turf-defense account at 3:51, and the transparency criterion at 13:54. The 2020 audit with the glance caveat, the December 2025 audit with the counts, and the commenter’s day of file management are all linked in my earlier notes. The AI submission and the March 2026 policy are at socopen.org.

The concession structure, which is the essay’s exposed flank. Three counter-arguments run before the payoff: that genealogy is not refutation, that a field’s inability to share is a real fact about its evidence rather than an exemption, and that compliance costs fall hardest on the least resourced. A hostile reader will say the concessions are placed to be survived. The honest answer is that the first one is decisive and the essay says so, and that the argument turns on the narrower claim about who adjudicates rather than on the claim that transparency is arbitrary. If you want the essay less exposed, strengthen the second concession, since it is the one most likely to be true.

The claim that is mine. That what got adopted is documentation rather than checking, that the public who received the authority to inspect has not used it, and that the standard therefore operates as a gate while carrying the rhetoric of universal access. Cohen concedes the first part of that (he has never claimed to run the code) and would dispute the conclusion. The strongest reply available to him is that a dormant right to check disciplines behavior even when unexercised. I have not found anyone who has tested it.

The audit measures a regime whose effects do not depend on the regime being used. Tax compliance runs far ahead of the audit rate. Agencies write differently under freedom of information statutes whose requests mostly never come. A firm’s filings are shaped by a short-seller who never appears. If that holds for replication packages, then counting deposits is counting something real, and my complaint about dormancy is a complaint about the wrong variable.

The trouble with the reply is that deterrence is doing four different jobs in it, and they have different requirements and different failure conditions.

The first is a forcing function that needs no inspection at all. To assemble a package you have to run the pipeline from raw data to published table in one pass, and that pass finds errors whether or not anybody else ever looks. Cohen’s commenter described this without naming it: a full day of work for a first package, despite tidy file management. Much of what takes a day is discovering that your own directory does not do what you thought it did. This version of the reply is the strongest, because it operates at a zero inspection rate, and it is the one Cohen’s audits are aimed at, since what they count is whether the package was built.

The second is deterrence proper, which needs a nonzero perceived chance of being caught, and which targets choices made before any file exists: the outlier you would have dropped, the specification you settled on after seeing which one worked.

The third is conditional inspection, and it is the strongest structural point in Cohen’s favor. Nobody checks at random. People check when a finding is surprising, contested, or load-bearing for something. So the unconditional rate at which packages get downloaded badly understates the discipline, because the probability of inspection is high exactly where a wrong result would do damage and near zero where it would not. Michael LaCour’s fabricated study was exposed because David Broockman obtained the data and found that it matched an existing survey too closely. The right was dormant across the whole literature and got exercised at the one point that mattered.

The fourth is norm signaling, which does not require inspection either, and which is probably how most compliance with most rules works.

Now the evidence, which exists, and which separates these cleanly.

In March 2015 the American Journal of Political Science adopted third-party verification. Every accepted quantitative manuscript goes to the Odum Institute at Chapel Hill, where staff run the code and compare the output to the tables in the paper before the article is published. Authors know this. They know it with certainty, they know it will happen within weeks, and they know publication waits on it.

By February 1, 2018, Odum had processed 149 manuscripts and completed 127. Of those 127, eight reproduced all the analyses on the first attempt. Six percent. The mean number of resubmissions was about 1.8 and the maximum was five. A later account gives a pass rate around ten percent, and the process adds fifty to sixty-five days to publication.

That number kills the forcing-function version of the reply. These are authors operating under near-certain inspection by a named institute on a known timetable, which is the most favorable condition the argument could ask for, and ninety-four percent of them handed over packages that did not run. Whatever anticipatory self-auditing the deposit requirement induces, it does not survive contact with a stranger’s machine. If it fails under certainty, it is not doing much under dormancy.

The economics numbers point the same way from a different angle. When Kingi, Vilhuber, Herbert, and Stanchi went through 303 papers in American Economic Journal: Applied Economics, ninety-eight of them, about a third, relied on confidential or proprietary data and could not be reproduced by an outsider at all. Of the eligible attempts, roughly forty-three percent fully reproduced the analysis and another forty-two percent partly did. That corpus was published under a data availability policy the American Economic Association had held since 2005. Fifteen years of a deposit requirement, and under half of the checkable papers checked out.

So the deposit is a weak proxy for the thing Cohen cares about, and the gap is measurable, and it has been measured, and the measurement was not done by him.

What the AJPS record also shows, and this is the part that cuts against me, is that the requirement does enormous work anyway. The errors got found. They got fixed before publication rather than after, without a retraction, without a correction notice, without anybody’s reputation entering the transaction. That is close to the ideal case for research correction and it is invisible in every count of retractions, including Cohen’s count of one this century.

But look at what produced it. Not the right to check. Somebody checking. Funded staff, fifty to sixty-five days per article, a five-hundred-thousand-dollar Sloan grant in 2018 to automate part of the workload. The improvement came from labor, and the labor is the thing that does not happen when a package sits in a repository with a right attached to it.

Two things survive for the deterrence reply, and I want to state them because they are not weak.

The AJPS numbers refute the forcing function and say nothing about deterrence proper. A package that fails to run is evidence about file hygiene. It is not evidence about whether the author dropped an inconvenient case, and it could not be, because those decisions leave no residue in the deposited files. The archived script documents the path taken, never the search. Whatever restraint the possibility of inspection exerts on specification choices is not something a verification failure rate can measure, in either direction.

And the conditional-inspection point stands untouched. A regime under which most packages are never opened and the consequential ones reliably are is a good regime, possibly better than random auditing, and its performance cannot be assessed from the average.

What would settle it is three studies, none of them expensive.

Compare AJPS packages deposited before March 2015 with packages deposited after, re-run today by a third party. If the post-policy packages verify at higher rates years later, the regime changed author behavior durably. If the improvement disappears once you account for the Odum round-trips, then verification worked and the requirement by itself did nothing. Somebody proposed this in 2019 and I cannot find that it was completed.

Compare journals that require deposit without verification against journals that verify. Sociology, political science, and economics now sit on opposite sides of that line, and the comparison is available.

And look at the download logs. Dataverse and openICPSR know how often each package has been retrieved and by whom. The dormancy rate is not a matter for speculation. It is sitting in server logs at institutions that would hand it over to a researcher who asked.

Which brings the whole thing back around on Cohen. He counts a proxy. The relationship between his proxy and the outcome he cares about is measurable. It has been measured in two neighboring disciplines and never in his own. And the answer his own standard would demand is one sociology journal paying somebody to run the code, which costs money and two months per article, and which nobody in sociology has proposed.

As of February 2018, three years into the policy, the journal reported that it had accepted no manuscripts with qualitative analyses since the policy took effect. There are other possible explanations, including submission mix and editorial preference, and it is a single journal. It is also the cleanest available datum on the distributional question, and it points where the petitioners said it would.

Sources

AJPS verification figures from the journal’s own policy document, “The American Journal of Political Science Replication and Verification Policy,” dated February 27, 2018: https://ajps.org/wp-content/uploads/2019/03/ajps-replic-and-verif-policy-2-27-18.pdf. Source for the March 26, 2015 announcement, the 149 processed and 127 completed manuscripts, the eight first-attempt successes, the mean of 1.8 resubmissions and maximum of five, and the sentence about no qualitative manuscripts having been accepted. The roughly ten percent pass rate and the division of labor between Odum and the Qualitative Data Repository at Syracuse are at https://wholetale.readthedocs.io/en/stable/project/cases/political-science.html, which also notes the 2018 Sloan grant of five hundred thousand dollars. The fifty to sixty-five day publication delay is from a conference report at https://www.csescienceeditor.org/article/science-editor-symposium-reproducibility-reporting-guidelines/. The workflow case study is Thu-Mai Christian et al., “Operationalizing the Replication Standard,” International Journal of Digital Curation 13, no. 1 (2018), https://ijdc.net/article/view/555.

Economics figures from Kingi, Vilhuber, Herbert, and Stanchi, “The Reproducibility of Economics Research: A Case Study,” working paper at https://sylverieherbert.github.io/files/Replication_aejae.pdf, published as Herbert, Kingi, Stanchi, and Vilhuber, “Reproduce to Validate,” Canadian Journal of Economics 57, no. 3 (2024): 961–988. The eligible-attempt percentages in the working paper text are ambiguously worded and should be checked against the published version before you print them. The AEA’s data availability policy dates to 2005, and prepublication verification under the Data Editor began in 2019: https://aeadataeditor.github.io/.

The LaCour case is the standard reference for consequential inspection of an available dataset. Broockman, Kalla, and Aronow’s May 2015 report is the primary document.

The proposed pre- and post-policy comparison at AJPS is described at https://wholetale.org/internship_projects/2019-analyzing-ajps.html. I could not find a published result and would not assume there is none without a proper search.

I have not checked whether anybody has published dormancy rates from Dataverse or openICPSR logs. If nobody has, that is a small original paper sitting unclaimed, and it is the direct empirical answer to the question you asked. It is also the sort of thing Cohen could do in an afternoon and has not.

‘Families, Inequality, and Sociology in Pandemic Times’ (Aug. 1, 2020)

0:04 Opens with the discipline’s premise: sociology looks for where macro and micro meet, where biography meets history, and much of that happens in families. Families are where social structure gets internalized into personality early in life.

0:24 The example that carries the talk. People marry when they have confidence in the future, and many have children when they feel able to shape their own destiny. If people are being laid off and housing, health care, and education are out of reach, they will not make long-term commitments.

1:07 The Gini index and half a century of rising American inequality.

1:19 Runaway wealth at the top. He grants that Bezos doubling his fortune does not affect him personally, and argues the aggregate effect corrupts other parts of the economy, since a few firms set wages and conditions for millions.

1:46 The formulation: the richest people make the rules for themselves and everyone else works around it. We want Gates to solve malaria because he has more money than many governments, which would be good, and it is unfortunate that this is where we ended up.

2:04 The weak social floor. How far people are allowed to fall, and the toll even on those whose necessities are currently met, since losing them easily wears on health, anxiety, and fear for children.

2:25 The parenting consequence. Do you let your children be whoever they want, or do you fear they will land on the wrong side of the gap.

2:43 Self-efficacy again, the sense of controlling your own destiny, which he says modern society prizes and which shapes every decision and relationship when absent.

2:55 The decline in marriage. When women’s earnings and education accelerated from the 1970s, some feared feminism would end the family, which he says now seems quaint.

3:16 The arithmetic under the fear. In the 1950s more than 80 percent of Americans married before 25, so marriage was near universal and there was nowhere for the rate to go but down once opting out became possible.

3:38 What nobody anticipated: richer women marrying more. He attributes it partly to better economic prospects among their available partners, and to men increasingly judging partners by earnings.

3:54 Why poorer people and African Americans marry less. He calls it an unbalanced gamble, a long-term commitment in a universe offering no reason to think long-term decisions will be reciprocated by reality.

4:09 Middle-class anxiety, driven by college costs and fear that children will end up broke, which he says contaminates opportunity with fear and produces cynicism, materialism, and a duty to keep children on track that limits their range.

4:31 The other side. People look for a realm they can control, and families are part of that. The haven in a heartless world, which he attributes to an old book. Emotional connections over material principles, pulling together, trusting each other. That is the ideal and some get near it.

4:57 How it is achieved lower down: extended family, especially among immigrants, fictive kin, informal care networks, people watching each other’s children.

5:24 The correction he stops to make. He does not want race collapsed into class. Take two college graduates with mid-level incomes, one White and one Black. Because of history and context the Black graduate is likelier to have poor relatives and friends, so people come knocking for help, and likelier to be less secure and to face racism from work relationships to policing to exclusion in public. Track only class and you miss it.

6:06 Impression management, part of the human condition, and heightened on social media.

6:12 The connection back. When 80 percent married before 25 you did not have to justify the arrangement. More choices mean justifying them to others and to yourself, which feeds the presentation of lifestyle.

6:35 His refusal to call it all bad. The humble brag about the messy office, the real brag about the cooking or the clothes, and the point that presenting a controllable image gives people some real control.

6:53 The Zoom green screen as the pandemic case: creating the visual context for relationships that exist only online.

7:05 A shallow cynical dimension and a positive one, since making your own image can be empowering.

7:13 Teenage anxiety. Adolescents may know rationally and cannot fully grasp that everyone is presenting a positive image, so they conclude they are the only sad one when everyone is posting one happy minute from yesterday.

7:31 The unanswerable question: how much is people having fun and how much is producing the image of fun.

7:38 Large shocks push things further in the direction they were already going, or that is an illusion and a confirmation bias, and everywhere he looks he sees inequality growing.

7:51 Education. Children with computers, devices, wi-fi, and parents pulling away from the rest.

7:58 Home. Some are trapped and getting more family time, which is basically good even when tiresome, and some are trapped in an abusive hellscape.

8:13 The measurement problem, stated as sharply as he has ever put it. Many tools for measuring family violence are activated only when people leave home, so tracking is limited.

8:29 School as the eight-hour daily respite from home, and how much that matters for children.

8:42 The magnification argument. Those with enough rooms for everyone to work and their own computers are riding it out, and the existing advantage is grossly magnified.

8:56 The saving that comes from not traveling, not buying coffee, and not eating out, being readied for a tutor or a private pod teacher next year.

9:13 Permanent cuts. Workplaces doing without certain workers may never restore them, and an office without a receptionist, or without an office, takes down building staff, parking attendants, and coffee shop workers.

9:38 A one-line aside on the advice to try something new, and that it is not voluntary.

9:42 Long-term medical effects, including possible lifetime consequences of infection in cases that look mild now, which given disparities in infection rates becomes future health inequality.

10:02 Fertility. No data yet, and certainly many are postponing children in a country where birth rates were already falling. Some delay becomes births forgone through breakups, restarting, and conception difficulty at older ages, with ripples like fewer kindergarteners in six years.

10:35 Postponed and cancelled weddings, and the open question of how many are temporary.

10:41 Children not going to camp, not traveling, not making friends or seeing new things, who will be different adults.

10:50 Single people, and whether they are meeting anyone.

10:56 Some of what is happening is the things that are not happening, and that absence will have effects nobody can predict.

11:03 The millions infected or bereaved, and how prolonged risk might change personalities or culture permanently.

11:16 The close. If the political system can function at all, he expects the crisis to produce universal health care, child care, and paid family leave, and says those are for a different talk.

Five recordings across four months and this is the one where the theory shows. The other four were a professor explaining things he had checked. This one is a man saying what he thinks families are, and the position turns out to be more conservative than his reputation allows and more radical than his critics would grant.

Start with what he says at 0:24 and repeats at 2:43 and again at 3:54. People marry when they are confident about the future. People have children when they believe they can shape their own destiny. Poor people do not marry because the gamble is unbalanced, a long-term commitment made in a universe that gives no reason to expect long-term decisions to be honored.

That is a claim that marriage requires conditions, and that the conditions have been withdrawn from most of the people being lectured about it. Hand that argument to a reader who thinks family breakdown is the central American problem and he will find that Cohen agrees with him about what has been lost and disagrees about the address to send the complaint to. The disagreement with Wilcox is about whether you can produce confidence in the future by exhorting people who have no reason for it. This talk is the clearest statement of his side of that argument I have seen from him, and it contains no accusation against anybody.

The haven in a heartless world, at 4:31, is quoted from Christopher Lasch’s 1977 book, and Cohen names it only as an old book. It is a strange citation to leave anonymous, since Lasch’s argument was that the haven had been hollowed out by the therapeutic professions and the market, and Cohen is using the phrase as an ideal that some families reach. Whether he is endorsing Lasch, correcting him, or reaching for a familiar phrase is impossible to tell from the recording, and it is the one place where a reader would want a footnote.

Two passages carry the analysis.

At 5:24 he stops the flow to say he will not collapse race into class, and gives the argument about two college graduates with the same income, one of whom fields requests for help from poor relatives and is exposed to racism from the workplace to the police. That is the wealth-versus-income point and the kin-network point, delivered in ninety seconds without a chart, and it is the single most portable thing in the talk. It also answers, in advance, the objection that his structural story is a class story wearing a racial costume.

At 8:13 he names the thing that broke the data. Many measures of family violence activate only when people leave home. Teachers, clinicians, and clerks are the observers, and the lockdown removed them, so the numbers that came out of 2020 measure the observers as much as the violence. He said the same thing in April and it is sharper here. Anyone reading the pandemic literature on family violence without that sentence in hand will draw the wrong conclusion from a falling number.

And 10:56 is the line of the talk. Some of what is happening is the things that are not happening. That is a demographer’s habit turned into a sentence a general reader can carry: absences have consequences, births forgone, friendships never made, meetings that did not occur, and none of it registers in any count of events.

Now the weaknesses.

The confirmation-bias moment at 7:38 is admirable and unfinished. He says large shocks push things further in the direction they were already going, then says that might be an illusion, then says everywhere he looks he sees inequality growing, then proceeds for three minutes on the assumption that it is growing. Naming the bias and then not acting on it is worse than not naming it, because the acknowledgment buys credibility that the subsequent argument does not earn. On the substance he was partly right and partly wrong, which is what you would expect: pandemic-era learning loss and childcare disruption fell hardest on poorer children, and household income at the bottom rose sharply through the transfer programs, and wage compression at the low end over the following two years ran against the story he was telling. A man who spends his Decembers counting other people’s evidence made a directional forecast here on the strength of what he could see from his window, which he says at 8:35, and never went back to it in public that I can find.

The Bezos passage at 1:19 concedes and then abandons. He grants that one man’s wealth doubling does not affect him, then says the aggregate corrupts the economy, and the connective tissue is one sentence about a few firms dictating wages. That is the weakest link in a chain the whole talk hangs from, and he would demolish it in anyone else’s argument. The concentration story about employers is empirically serious, monopsony research is real, and it has nothing to do with the personal fortunes he opens with.

And the ending is a wish presented as a forecast. There is every reason to believe, he says, that the crisis will produce universal health care, child care, and paid family leave. In August 2020 that was a hope with a political condition attached, and he attaches the condition and then discards it in the same breath. None of the three arrived. The one that came closest, the expanded child tax credit, ran for six months and lapsed. It is the only sentence in five recordings where he says something will happen because he wants it to, and he closes on it.

For my file, this is the piece the biography most needed and did not have. Everything else in the record shows him checking somebody. This shows him theorizing, and the theory is coherent: structure enters the person through the family by way of confidence about the future, inequality operates by removing that confidence, and the removal shows up as decisions not made rather than as visible damage. That is a thesis, it is the same one underneath the divorce paper and the marriage-market work and the pronatalism arguments, and he has never stated it this clearly in print as far as I can see.

Two smaller things. The view count is the highest of the five, at nearly five thousand, and it is the least data-dense of them, which is either a fact about audiences or a fact about him. And the description, a short talk about what is on my mind as a family and inequality sociologist today, is the only one of the five that promises opinion rather than instruction, which suggests he knew the difference.

Transcript flags. At 1:09 “genie index” is Gini. At 4:42 “an old book” is Christopher Lasch, Haven in a Heartless World: The Family Besieged (1977), and you should name it if you quote the passage. At 8:02 “some people are trapped at home are but” is garbled. At 9:38 the Ivanka Trump reference is to the “Find Something New” campaign launched in July 2020, which dates the recording. At 11:24 “true universal health care” may be real universal health care. The talk appears to be scripted or read from notes rather than delivered live, given the absence of the verbal stumbles that fill his classroom recordings.

‘Joanna Pepin interviews Philip Cohen’ (Aug. 23, 2020)

0:00 Joanna Pepin introduces herself as being at the University at Buffalo and introduces Cohen and the textbook.

0:25 First question: what led him to become a family sociologist.

0:33 His answer, that inequality questions kept routing him back to families.

“I was interested in inequality, and a lot of the inequality questions I was thinking about kept leading me back to families, especially things having to do with the intersections of race, class, and gender. If you ask why women do more housework and child care, part of that is because they’re single parents, and that’s a family question. Why are more parents single now than they used to be? Who decides who’s married, and why are people married, and what are the consequences of that for health and well-being and income? I just kept coming back to family topics.”

1:20 The policy version of the same point.

“What is the government going to do? Is it going to give people money if they’re poor? Is that going to be based on their family structure? Is it going to penalize people for their family choices? Is it going to let people get married if they’re 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.”

1:52 Pepin notes the textbook’s life course approach and asks for the biggest transformation of American families.

2:06 His answer: two changes that arrived together over a couple of centuries, fewer children and women working outside the home. Fewer children means less care work at home, partly because women had other opportunities and partly the reverse, since women doing other things had fewer children.

2:41 Why he calls it foundational.

“Having fewer children changes the relationship between parents and children. It means we’re investing more in each child than we used to, so kids spend more years in school with more resources. It also means our relationships are longer. People are living longer, so you have fewer children and now you have a seventy- or eighty-year relationship with your children, and there’s only one or two of them, instead of having a forty-year relationship with eight or ten kids. It’s a whole different relationship.”

3:27 Pepin asks about a common misperception he finds himself correcting.

3:40 The decline narrative first, with a hedge.

“There’s a common sense that families are going down the drain in general, that in the good old days things used to be better. Some things used to be better for some people, some things used to be worse for some people. It’s hard to generalize that way.”

3:58 Then divorce, which is the substantive answer.

“Divorce is more common than it used to be a long time ago, like a hundred years ago or fifty years ago. But in the last ten or twenty years, divorce is actually becoming less common. That’s surprising to people, and it clues us into some important changes. People are getting married later, people are getting married after they have more education, there’s more selectivity in marriage. Some people are choosing not to get married, or they don’t have anybody they want to marry.”

4:39 What the decline is made of, stated without comfort.

“You have fewer people getting married just because that’s what you do, fewer people getting married because they’re stuck with somebody, fewer people stuck in marriages because they can’t support themselves on their own and so they can’t leave an abusive marriage. When you put them together, the fewer marriages and the marrying later have turned into a decline in divorce, even though divorce is more available than it used to be, which is great, and it’s happening less. But partly because marriage is available to fewer people. So it’s a mixed bag.”

5:31 Pepin asks what he has had to update for the third edition.

5:48 Same-sex marriage, and the measurement problem underneath it.

“In the second edition, same-sex marriage was just coming out, and we didn’t really have a lot of good data about what was going on with it. That’s improved. I worked really hard to try to get better information. It’s still pretty hard. A lot of states don’t record the gender of the people getting married, which seems odd, but they just have spouse one and spouse two on their marriage certificates. So the data is still blotchy. But we’re starting to learn more, and in a lot of ways we’re learning that same-sex married couples look pretty much the same as different-sex married couples.”

6:42 The faster-moving revision.

“At the same time, the rapid change that’s happening is on sex and gender and language and culture and attitudes. Transgender identities, non-binary identities. The language had to get revamped even though the last version of the book was only a couple of years ago. 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.”

7:24 Pepin asks him to speculate on the pandemic’s impact, noting that data collection is hard right now.

7:43 His framing principle, delivered as an aside.

“Things that are happening now will tend to continue happening. That’s just a way that society works. So working at home is going to be more common now forever.”

8:11 The disrupted side. Cancelled weddings, postponed childbearing, and the observation that postponement becomes permanent for some people, so the drop in marriage and childbearing will probably be large this year and next.

8:42 Inequality, prefaced by self-awareness.

“In terms of the subjects I pay most attention to, I think inequality is going to be exacerbated. I sound a little like a broken record on inequality, because I think everything’s related to inequality.”

8:56 The question underneath his whole program, stated more plainly here than anywhere else in the recordings.

“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. When gay marriage was illegal, it was, okay, there’s a bunch of people who can’t have the kind of family they want. But now we have people who are stuck with their families that they don’t want to be stuck with. We have trans young people who are living with families that don’t accept them. We have people living in abusive situations.”

9:39 School at home, and the magnification argument: children whose parents have education, resources, spare rooms, and computers against children whose parents have none of that.

9:57 The summary.

“All the gaps and inequalities we have, whether class and race or between happy and abusive, I just feel like the gaps are opening up right now, and it’s going to get more unequal before we get out of this.”

10:32 The close, on what follows for the discipline: a lot of work to do, not only to study these things but to fix them, and a hope that a semester of teaching and learning will turn up ways to address some of them.

Thoughts

Eleven minutes, a friendly interviewer who is his own former doctoral student, and no adversary anywhere in the frame. It is the softest recording in the file and it contains the two clearest statements of his position that exist on tape.

The first is at 4:39. He is asked about a misperception and gives the divorce decline, which is his best-known finding, and then does something almost nobody does with their own famous result: he itemizes what the good news is made of. Fewer marriages entered out of default. Fewer people trapped with someone. Fewer people unable to leave an abusive marriage because they cannot support themselves. And then marriage available to fewer people, which is the part that does not fit the story, and he says it in the same breath and calls the whole thing a mixed bag. A researcher protecting a headline finding leaves that last clause out. He puts it in, to a sympathetic interviewer, with no one pressing him.

The second is at 8:56, and it is the sentence I would build any essay on him around. Who gets to have a family. He has spent a career being read as the sociologist who says the traditional family is not in decline, and what he says here is that his real subject is the people outside the frame entirely: not counted in the marriage rate, not counted in the divorce rate, absent from the data because the thing they would have been counted for never happened. That is the same insight as some of what is happening is the things that are not happening, from three weeks earlier, and it is the same insight as the marriage-market work with the interviewer sitting across from him. The pattern of his mind is to ask what the denominator excludes.

And notice how he uses same-sex marriage in that passage. He reaches for it as the historical example of people barred from the family they wanted, and then pivots to people trapped in families they cannot leave, including trans teenagers living with parents who do not accept them and people in abusive homes. The two cases are structurally identical in his account: the constraint runs on access to resources and to other people, and it operates whether it locks you out or locks you in.

Three small things.

The androgyny detail at 7:02 is the best single piece of evidence in the file for how a textbook works as an artifact. He had a whole section on androgyny in an edition published two years earlier, and by 2020 the concept had been superseded in his own book by non-binary identity. That is a discipline’s vocabulary turning over inside one author’s revision cycle, reported without defensiveness and without celebration. Whatever a reader thinks about the substance, the mechanic is the same one that produced his complaint about generation labels: the category changes, and the trend line moves with it, and somebody has to decide whether the world moved or the instrument did.

The marriage certificate detail at 6:07 is a measurement anecdote he throws away. States record spouse one and spouse two rather than sex, so the administrative record of who married whom is unusable for the question everyone wants answered. He calls the data blotchy and moves on. That is the same category as the mandatory reporters vanishing from the family violence data and the observers vanishing from the child maltreatment counts. His most durable contribution may be that he keeps noticing when the instrument, rather than the world, produced the number.

And at 7:43 he states a forecasting rule so casually that it sounds like nothing: things happening now tend to keep happening. He then applies it to remote work, where it held, and in the same minute applies the opposite reasoning to weddings and births, where he expects the disruption to produce permanent losses through postponement. Both turned out roughly right, and the two claims run on different logics that he never reconciles. It is the closest thing he has to a stated method for talking about the future, and it is four seconds long.

Two more things.

He is a broken record on inequality and says so at 8:47, and the self-awareness does not change the fact. Asked what the pandemic will do to families, he answers that inequality will grow, which was the answer he gave to a different question three weeks earlier and the answer he gives at every level of the analysis from Bezos to the spare bedroom. A frame that explains everything explains less than it appears to, and his own joke is the acknowledgment that no correction follows.

And the interview format flatters him. Pepin was his doctoral student, published with him on the marriage-market work, and asks five questions with no follow-up. Every recording in this file has him talking to people who either agree with him or have no standing to push. There is no tape anywhere of Cohen being interrogated by someone who disagrees, which is a gap in the record on a man whose reputation rests on interrogating others.

For my file, this is the interview to mine for direct quotation, because it is the only one where he speaks in his own voice about what he is trying to do rather than explaining a chart. The biography currently derives his position from his writing. Three of these passages, the origin story at 0:33, the mixed bag at 4:39, and who gets to have a family at 8:56, would let him state it himself, and quoted speech does something in a profile that paraphrase of published work cannot.

Transcript flags. At 0:03 “university of buffalo” is the University at Buffalo, which is how the institution styles itself. At 0:11 the textbook title is garbled as “diversity in equality and social change” and is The Family: Diversity, Inequality, and Social Change. At 7:28 kovid is COVID. At 8:18 and 8:23 “that we’re gonna” is “that were going to,” which reverses the meaning of both sentences and matters: people who were going to get married and were going to have children. At 10:32 “upgrade news” is garbled beyond recovery and appears to be Pepin’s transitional remark; leave it out. The third edition he describes working on published in 2021, and the fourth in 2024, so this catches him mid-revision on the third.

‘COVID-19 Pandemic Overview’ (Aug. 25, 2020)

0:00 Opens the lecture, and lays out the plan: history and terms, the global situation, where the United States fits, the American failure, the overlapping crises, and the future.

0:37 Pandemic from the Greek for all people, meaning the worldwide spread of a new disease. The World Health Organization declared this one on March 11, 2020. The American epidemic is part of the global pandemic.

1:14 Precedents. The 1918 influenza pandemic, which he notes probably did not come from Spain, infected about a third of the world and killed around 50 million people including more than half a million Americans, against a much smaller world population. The 1957 and 1968 influenza pandemics killed about a million each. SARS in 2003 killed about 800 and was extremely frightening before it was controlled. H1N1 in 2009 killed about 200,000.

2:53 The virus. Discovered in 2019, probably from a bat to a person, possibly through another species, in or near Wuhan. SARS-CoV-2 causes COVID-19.

3:38 Why it is bad, in four parts.

“It’s a novel disease, which means we are not immune to it. As far as we know, nobody was immune when this started. And it’s very highly contagious, so it spreads very easily between people. All they have to do is interact or be close to each other, and if they’re not taking any precautions the virus transmits and nobody is immune. To make matters worse, it transmits asymptomatically. There are people who show no signs that they are sick who can transmit it, so it’s hard to avoid people who are transmitting the disease. It transmits through aerosols, which are tiny airborne particles you can get just from breathing, so you don’t have to spit or cough or sneeze on somebody to transmit the virus.”

4:27 Mortality. Seasonal influenza kills somewhere between 0.3 and 0.7 of every thousand infected. He puts COVID at roughly ten times that, perhaps six to ten per thousand, with the rate uncertain, and notes the seriously ill who survive.

5:12 The counts in late August. The world above 20 million confirmed cases and approaching 25 million, with more infections than confirmed cases. The United States above five million. Global deaths in the hundreds of thousands and American deaths approaching 200,000, which he calls about a quarter of the global total.

6:09 The world distribution. Started in China, spread to Europe, then the United States, which now has the most confirmed cases. India approaching three million, Brazil above three million, Russia approaching a million, South Africa at half a million.

6:58 Comparison with peer countries, defined as those above thirty thousand dollars of annual income and five million population. The United States has more cases per thousand than any of them by a wide margin, around 17, with Israel next and then Singapore. On deaths per million, five large rich countries were then higher: Sweden, Italy, Spain, the United Kingdom, and Belgium. He immediately says the cumulative comparison is misleading without the trajectory.

7:59 The trajectory chart, and the central finding of the lecture.

“In late April, when we had our peak, there were those five countries with higher peaks than us, including much higher, like Italy and Spain, many more deaths at that time per person in their populations. We all started to decline after that. But the key thing is that those countries had a faster decline and a more continuous decline. By June our daily deaths had stopped declining, and actually started rising again in July and August, which is virtually unique among the rich countries. Now we’re at about ten times or more deaths per day, relative to our population, than the other rich countries of the world.”

9:39 Racial and ethnic disparities. Black Americans with the highest death rate at 85 per hundred thousand, White, Asian and Pacific Islander Americans the lowest, Hispanic and American Indian populations between. The causes he lists: discrimination, inequitable access to good health care, work in front-line and essential occupations involving contact with the public, inequalities in income, education, and wealth, and housing quality and crowding.

11:02 CDC funding, cut from above seven and a half billion two years earlier by about a billion for the 2020 budget.

11:35 The National Security Council pandemic office, closed, and the warning from the Center for Strategic and International Studies that it was unclear who would be in charge at the White House in the event of a pandemic. He adds that this turned out to be true.

12:04 The ignored exercise. His administration ran a functional scenario gaming out a global pandemic originating in China, which produced recommendations that went unused, against the president’s claim in March that nobody could have foreseen it.

12:44 Published warnings. Ed Yong’s 2018 piece asking whether America was ready, on underfunded public health programs and stretched hospitals. Biden in the fall of 2019 saying the country was not prepared.

13:19 His summary of the foreknowledge question.

“This was not that much of a surprise. The details are always a surprise, but the fact that this was inevitable at some point, and that we were not prepared, was known.”

13:25 The messaging failure, framed as motive: the president more concerned with how it made him look than with dealing with the epidemic.

13:53 A montage of presidential clips running roughly ninety seconds, longer than the version he used in April. Totally under control. One person from China. It goes away in April with the heat. Eleven or twelve cases and many of them in the shade. Fifteen going down to close to zero. It will disappear like a miracle.

15:08 The compounding failures. Not enough testing from the beginning, so nobody knew who was infected. Without that, no contact tracing. Without containment, spread, with nothing stopping travel from New York to New Orleans for Mardi Gras. Numbers too large for tracing to work. Slow and weak policy once the federal government caught up, with no strong lockdown and no internal travel restrictions. Mismanagement of equipment and staff when New York was overwhelmed, so more people died than needed to.

16:53 Institutional outbreaks in nursing homes, prisons, and factories, largely unchecked at first, which then seeded the surrounding communities, which he notes are largely low income and minority.

17:17 The cultural failures: happy talk, mixed messages, a cynical attitude toward science and expertise from the top, conspiracy theories, and what he calls useless hero talk.

“This sort of useless hero talk. Oh, we’re warriors, we’re tough, we’re strong. 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.”

17:56 The Italy case study, which is the analytical heart of the lecture. A steeper and higher spike than the American one, arriving about a month earlier, followed by a much steeper decline, so that by late August the United States had more than ten times Italy’s ongoing death rate while Italy prepared to reopen schools.

19:01 The Italian sequence, step by step. Two deaths at the start of February. First, a few northern towns locked down with checkpoints and barricades, which failed because the virus had already left. By early March, the whole northern region. Within about a week, all schools and sporting events nationwide. By the second week of March, a national lockdown.

19:50 What that lockdown contained, and the comparison he draws.

“Their lockdown was much stronger than anything we had in this country. You could not leave the house unless you were going to the doctor, to get necessary supplies, or to do an essential job, and the police would stop you and you had to explain what you were doing. We had nothing like that in this country. You could not travel between regions of the country. By and large you couldn’t go from one state to another. We never had that restriction.”

20:17 All non-essential workplaces closed, including factories, for five or six weeks. Then phased reopening as cases fell ahead of the death curve: travel within regions, some factories, most businesses with distancing and masks, then pools and gyms in late May, then theaters by June 15.

21:26 The endpoint he wants students to see.

“Now they have a low enough number of cases that they can do testing and tracing, isolate people who get infected, and contain the epidemic. We never reached that level of containment in this country. We never had that level of closure, and we never had a small enough number of cases that we could jump on them, isolate them, and stop outbreaks as they occur. So our epidemic is relatively uncontrolled still at this point. It’s only controlled to the extent that people are wearing masks and staying home as much as possible.”

22:05 Why the failure. Beyond the presidential decisions, a partisan attitude toward the science.

22:23 Four survey questions he says have correct answers, with party gaps. That social distancing slows the spread, agreed by 76 percent of liberal Democrats and 47 percent of conservative Republicans. That the virus spreads more easily than most other infectious diseases, 78 against 58. That there is not enough testing, 82 against 31.

23:20 His diagnosis.

“Science can answer these questions, but because of the way our politics corrupts our science, or our science fails to penetrate into our political bubbles, we are impaired from being able to do smart things.”

23:36 Misinformation. The claim that the pandemic was engineered for profit, the 5G theory, the claim that masks harm health, and fake cures, including a hydroxychloroquine post the president shared. He notes how hard this makes it for anyone who is not a full-time debunker.

24:42 Racism and xenophobia. White Americans treating it as a disease of non-White people, the same way people in Texas and Arizona treated it as a New York problem. Presidential language about the China virus, vandalism against Asian businesses, attacks on Asian people, and a clip of a rally crowd cheering the phrase kung flu.

25:47 The southern border, where he notes an inversion.

“At this point it’s probably better for Mexicans than for Americans to have the border closed.”

26:05 A clip of the president claiming that public health authority at the southern border prevented a catastrophe, followed by his correction that the American side, and the Rio Grande Valley in particular, has been extremely hard hit.

26:44 The economic crisis. Non-farm employment rising from about 130 million in the early 2000s, dropping in 2008, recovering past 150 million, then falling by 20 million in a single month in April.

27:56 Food lines in Queens set beside New York City lines from 1931.

28:10 Education. Online math participation dropped about 20 percent in higher-income zip codes in March and more than 60 percent in low-income ones, and the richer areas recovered by May while the poorer ones never did. Estimates of about seven months of lost learning on average without in-class instruction, more for Black and Hispanic students and much more for low-income students.

29:28 The other side of the choice, which he refuses to resolve. Hospitalization rates five to eight times higher for Black and Hispanic children than White children. And the exposure of the household: 4 percent of White seniors live with a school-aged child against 11 percent of Black seniors and 13 to 19 percent in other groups.

30:35 His conclusion on schools: there are no good choices.

30:42 Family violence, and the measurement problem he has now stated three times across the recordings.

“We don’t have great data on this, because 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. One study found that there were fewer visits to the emergency room with injuries related to domestic violence, but the injuries that were showing up were more serious. So it looks like a lot of people are having abuse happen in their homes but are not making it to the emergency room unless their injuries are very serious.”

31:21 The gender impact. Women with school-aged children were the group most likely to leave the labor force between February and April, with some evidence of men taking on more of the domestic work when home.

32:12 The mental health section: fear, isolation, separation from friends and from services, and despair at the accumulation of bad news.

32:44 A pause to address the students directly, which is the only moment of its kind in any of these recordings.

“I just want to pause to say here, if you are anxious and depressed, if you are thinking about suicide, you are obviously not alone. A lot of people are, and a lot of people are not talking about it. There are the numbers there for you to call and seek help at any time, and you should not feel like you’re the only one, or that you are burdening anybody by asking for that help. So please do.”

33:08 Survey figures on young adults reporting serious thoughts of suicide, rising through the 2010s and then a much higher figure in a June 2020 survey of adults under 25. He says he was astounded by it.

33:52 The worst-case arithmetic, flagged as speculation. About 177,000 deaths and 5.7 million confirmed cases, a case fatality rate near three percent, an infection fatality rate perhaps ten times lower if actual infections run to 50 million. Herd immunity possibly around 70 percent, which for 330 million people would mean roughly 230 million infected and perhaps 700,000 deaths, unless treatment improves or a vaccine slows transmission.

35:58 When it ends. Waves for months and hopefully not years. Tracing only works once numbers are low, since nobody can call a hundred thousand contacts a day. Limited world travel for an unknown period. Vaccines under test with unknown efficacy and safety, unlikely to affect students personally inside eighteen months, which he calls speculation. And a caveat on his own arithmetic: herd immunity is theory, and it may not arrive at 70 percent.

37:39 The advice. Follow real public health advice from state and local government rather than every fad, and then a clause: your sociology professor is probably a good source too.

38:03 Stay home if you can, granting that it is boring, frustrating, and lonely, and may be impossible to pull off.

38:27 Take care of yourself, since nobody alive has experienced this. Watch your friends, neighbors, and relatives, work out what they need, and provide it.

38:45 If distressed, find ways to help others, online, with money, through volunteering or activism, which gives a sense of purpose.

39:16 The close. The world, politics, society, and science will be different, career plans may have to change radically, and it is a time to think big about what should be done afterward.

This is the longest and most political of the recordings so far, and it is the opening lecture of a required introductory course. Not a guest talk, not an elective seminar, not a video for people who sought him out. The first thing a nineteen-year-old sees in Social Problems, fall semester, includes ninety seconds of presidential clips, a rally crowd chanting a racial epithet, a partisan knowledge slide, and at 37:57 the instruction that their sociology professor is a good source of pandemic advice.

Every factual claim in it is sourced and checkable. The clips are real, the CDC appropriations are real, the exercise happened, the Yong article exists. The question this recording raises is is that a man who wrote a book about the boundary between the scholar’s role and the citizen’s role crossed it in the first week of a course, in front of students who cannot leave, and the crossing came packaged with a recommendation that they treat him as an authority.

His defense is that there was no neutral version of that lecture in August 2020, and a professor who withheld the trajectory chart to appear balanced would have failed at the only job Fish thinks he has. That defense covers the charts. It does not cover the montage, the rally clip, or the clause at 37:57.

Now the strongest thing in the lecture, which is not the politics.

The Italy sequence from 17:56 to 21:26 is the best twenty minutes of teaching in this entire file. It is a counterfactual done properly. Same disease, worse start, higher peak, earlier arrival, and a better position five months later. Then the specific timeline of what Italy did, escalating from a few towns to a region to a national shutdown, with the detail that police stopped people and required an explanation. Then the reopening sequence, with the observation that cases were falling faster than deaths because deaths lag. A student comes out of that able to answer the question that dominated American argument all year, which is whether the outcome was fixed by circumstance. Cohen shows it was not, and he shows it with a case rather than an assertion.

The line at 17:43 is the one to keep. You fight a virus by acting weak and staying home. That is a sociologist noticing that a culture’s heroic register was the wrong tool, and it says in one sentence what a great deal of pandemic commentary took paragraphs to miss.

Three failures of his own standard.

The partisan knowledge slide at 22:23 does the thing he attacks in other people. He introduces four items as questions that have correct answers, and one of them is whether there is enough testing. He calls it uncontrovertibly true, just a true fact. Sufficiency is a judgment about a threshold, resource allocation, and what testing is for. There were serious people arguing at the time about testing strategy, targeting, and turnaround. He may be right about the policy and he has converted a policy position into a factual item and scored a partisan gap on it, which is the precise operation he objects to when the Institute for Family Studies presents a family ideal as an empirical conclusion. The other three items are closer to factual, and the causal story he attaches at 22:29, that Republicans believe their political leaders instead of the science, is an assertion about why the correlation exists rather than a finding.

The suicide comparison at 33:08 is a measurement error of the kind he built his reputation on catching. The earlier figures he cites are twelve-month prevalence. The June 2020 figure is thirty-day prevalence, from an online panel with a low response rate. A thirty-day number that exceeds a prior twelve-month number is partly an artifact of the windows, and any comparison of the two overstates the change by an unknown amount. He says he was astounded, and part of what astounded him was the instrument. The underlying concern was real, the direction was probably right, and the number as presented does not support the size of the jump.

To his credit, the passage at 32:44 is the most careful thing he does in any of these recordings. He stops the lecture, speaks to the students directly, tells them they are not alone, that many people are not talking about it, that seeking help is not a burden, and points them to the numbers on the slide. That is right, and it is the only moment across six hours of tape where he steps out of the analytic role to address a person rather than a cohort.

The worst-case arithmetic at 33:52 is his most honest passage and his least accurate forecast. He shows every step, labels it pure speculation, states the herd immunity assumption, and then at 37:13 comes back and says the assumption itself is unknown and may fail. American deaths passed his worst case of 700,000 in the autumn of 2021 and have since gone well beyond it. His worst case was optimistic, and the reason is the assumption he flagged: immunity at 70 percent did not stop transmission once variants and waning protection arrived. A forecast that fails on the premise its author identified as the weak point is a better performance than most, and it is still a failure.

For my file, two things.

This lecture and the April 3 one are the same lecture five months apart, which makes them the only place in the record where you can check him against himself on a short horizon. In April he said thousands would die, that nothing would break the slope, that waves and repeated shutdowns were coming, and that travel restrictions would outlast domestic ones. By August, 177,000 dead, the slope unbroken, waves arriving, and he repeats the same forecast with more evidence behind it. The stable predictions are the ones that come from arithmetic. The unstable ones come from assumptions about biology he had no standing to make and said so.

And the family violence measurement point now appears in three separate recordings across five months, in April, in June, and here. He said it before the data came in, he kept saying it, and it is the thing about the pandemic he was most reliably right about and least credited for.

Transcript flags. At 5:49 the world death figure reads as over 80,000, which is wrong by a factor of ten and should be over 800,000; check the slide before quoting. Kobit, covet, corvette, cobin, and chronovirus are all COVID or coronavirus. At 3:10 “somebody ate a bad or a bad bit somebody” is bat. At 3:28 “the viruses stars” is SARS-CoV-2. At 24:17 “a bad high druxy chloroquine” is hydroxychloroquine. At 13:03 “even joe biden running for president last fall” places that quote in fall 2019. At 14:35 “we’re a bridge” is garbled. At 33:36 “26 of adults under age 25” should be checked against the slide, since the underlying survey reported 25.5 percent for ages 18 to 24. Slides are at https://osf.io/9vw2a/ and will settle the death figure, the disparity rates, and the suicide chart.

‘Introduction to Social Problems’ (Aug. 29, 2020)

0:00 Lays out the lecture: what sociology and social problems are, the main theoretical perspectives, and the research methods.

0:24 Sociology as the science of society, the study of individuals, groups, and the social structures that emerge from their interaction, with attention to the role of those structures in creating problems and solutions.

1:00 A social problem defined as a condition or pattern of behavior with negative consequences for individuals, for the social world of families, education, and the economy, or for the physical world of housing and environment.

1:27 Gallup’s most-important-problem question in August 2020, where coronavirus dominates, followed by government and the economy, with everything that would normally rank high pushed down.

2:14 The objective and subjective duality.

“Objectively, does the particular condition exist? Do we have a pandemic? Yes, we have a pandemic, we have a virus which is spreading, which has killed a lot of people around the world. So that’s objective reality. But do we think it’s a problem, and how do we shape or define that problem in our collective minds? Do we think the problem is the pandemic? Do we think the problem is the way the government is dealing with it? Do we think the problem is the government telling us what to do? What is the problem?”

3:10 Social constructionism, with the illustration he uses to mark its limit.

“Social constructionism is the view that social problems only become problems when they’re perceived to be problems. If a meteor hit us and we were all obliterated, that wouldn’t really be a social problem. That would be the end of all problems. It’s called constructionism because it’s acknowledging that social reality is the meaning we attach to it. There is a reality out there beyond our consciousness, but in terms of society, what matters is how we construct it.”

4:06 The back and forth is iterative. Some problems emerge fast, like a pandemic, and some persist a long time, like racial inequality, and the defining, blaming, and fighting over solutions goes on continuously.

4:45 Stages of the process, with the caveat that they are categories for thinking rather than a sequence, and that some problems do not move through them in order.

5:09 Stage one, claims making. Somebody has to come forward and say this is a problem. With a pandemic there may be no single actor, but the disparities and consequences require someone to stand up.

5:53 Stage two, getting weight behind the claim, especially from government agencies willing to accept the definition and take on the role.

6:23 Stage three, negotiation and power struggle over whether the response is adequate, effective, and moving in the right direction.

7:01 Stage four, when invested groups conclude the established system will not produce the results they want and have to work outside it or change it radically. He notes that not every social problem reaches this.

7:43 Why courses on social problems focus on protest: it is the most explicit version of the process, where people gather and literally state what they think the problem is.

8:03 A photograph he took at the Women’s March the day after the 2017 inauguration, and the observation that people bring signs, friends, and relatives.

9:04 What protest does beyond making claims.

“In this process of making claims and having protests, people are also building identities around their definition of problems, and who else thinks it’s a problem, and who defines it the same way we do. The formation of a social group comes out of that process of making the demand that something be considered a problem.”

9:23 A photograph he took at the March for Life, where teenagers from religious schools came from around the country, and the same analysis applied.

“It’s not just people saying, hey, we think you should stop women from having abortions. It’s this process of we are all in this together, we all believe this, we reinforce it. I’m looking at my friend, my friend is really into it, I’m really into it, we’re really into it together. And it’s building solidarity and identity.”

10:12 C. Wright Mills and The Sociological Imagination, and the link between personal experience and the social world, described as the intersection of biography and history.

11:00 Personal troubles against public issues.

“If I’m hungry right now, that’s a personal trouble. I’m hungry, maybe I just need to go in the other room and have some food, or maybe I don’t have enough food. But it’s still one person who doesn’t have enough food. The point at which it becomes a public issue is when it affects a lot of people and we start to acknowledge that.”

11:29 The illustrations: cars lined up for free food in San Antonio, and bags of food at a drive in Des Moines.

12:07 Theory defined as a set of assumptions or propositions used to explain, predict, or understand, stated explicitly and systematically.

13:37 Why the theories were already in place.

“We may even be able to make predictions. We’ve seen pandemics before, we understand that they’re likely to take a certain course. If we could already in January, February say some intelligent things about what was likely to happen, it’s because we had experience and we already had theories about how pandemics work.”

14:04 Macro and micro as a continuum. Macro asks why this happens in one society rather than another, or now rather than in the past. Micro asks why an individual gets in line at a food drive.

15:05 Functionalism, traced to Durkheim, and the body metaphor: organs with distinct functions maintaining a balance, and institutions like family, politics, and religion working in some harmony to meet social needs.

16:18 Change as the problem, from that perspective, producing anomie.

“When society changes rapidly, when the social order changes rapidly, it disrupts that balance, and we end up not knowing what the rules are, what the norms are, how we should behave, how our institutions should work together, where individuals fit into the changing social order. This produces a state that Durkheim described as anomie, or normlessness.”

18:04 The functionalist question: where did the problem come from, and is it serving some function, the way a fever serves the body against an infection.

18:41 Merton, and the argument that crime may help mark the boundary between good and bad behavior, and poverty may motivate people to avoid destitution. He calls the poverty version cold-hearted and grants there is a logic to it.

19:29 “There’s a potentially faulty logic to this, which is that if you see something happen in society you assume there must be a good reason for it. Every society seems to have poverty and inequality, so it must be good, right? Like human beings have lungs, everybody’s got lungs, let’s see why we have lungs. You can see the potential logical problem in this metaphor is you tend to define things as good and then try to rationalize them after the fact, working from the assumption that they are good. So this tends to be politically a conservative theory, because it tends to be describing the status quo as somehow good, and it’s suspicious of rapid change.”

20:12 Conflict theory as the opposite, with a teaching disclosure: when we teach sociology we build up this opposition because it is easier to learn that way.

20:31 Society defined by conflict rather than order, held together through power and coercion, with social problems emerging from conflict between groups.

21:53 The attitude toward change reverses. Conflict theories are positive about social change, especially where the tension can be harnessed, and especially if the less powerful side wins.

22:33 Marx introduced with a qualification.

“Karl Marx was not an academic sociologist. He was a writer and philosopher, and really a political activist in the nineteenth century. He wanted to change society in what he saw as a progressive direction. But his framing, and the way he identified how society works, is really the basis for what we’re now simplifying into conflict theory.”

23:01 Workers who own nothing but their labor and owners who own capital, the proletariat and the bourgeoisie, brought together non-consensually and unequally.

23:56 Alienation, set beside anomie as a parallel condition arriving by a different route: workers do not control the process because the owner makes the rules, and do not own the product.

24:46 The contemporary illustration: iPhone factory workers in China who cannot afford iPhones, do not set how the factory runs, and produce for people in other societies.

25:24 Where the two theories part on the same fact.

“When the subordinate group, the weaker group, gets an awareness of their position, that is when they are ready to make change. This is what concerns or frightens functionalists, and what excites Marx and the conflict theorists.”

26:10 Feminist perspectives, introduced with a caveat that they are not all conflict theories, though they lean that way because of the focus on inequality between the genders.

27:14 Patriarchy defined as a male-dominated society or subsystem, family, workplace, or school, with the term originating in the family and extended outward.

27:55 Intersectional feminism, its origin in the position of Black women, and its extension to class and sexuality. Patricia Hill Collins, who he notes retired from his own department, on Black feminist thought. Angela Davis, from activism into academia. Audre Lorde, writer, activist, and poet.

30:00 An aside on the migration of the term.

“One of the interesting things about this is that for something that kind of came from academia, intersectional feminism has really emerged into politics in the last few years in a really interesting way, where you see people actually holding up signs at rallies that say intersectional feminism. Which if you’re old like me strikes you as a big change in the world.”

30:33 Interactionism as the third of the big three, a micro-level theory about how people create and maintain social reality through language, words, and symbols.

31:34 George Herbert Mead, and society as organized and patterned interaction among individuals, so that society is real at the macro level and built from interaction at the micro level.

32:12 The self as both mental and social.

“You can’t have a self if you never interact with anybody else. Our ability to see others in relation to ourselves is reflective. Once we have social interactions and they shape our identity and our personality, we know that when we act a certain way, when we say or do something, we can use our imaginations to predict how people will react, or to understand how they react when we’re seeing their reaction. So this reflective ability is what really creates the self.”

34:07 The summary table in speech: functionalists on order held together by institutions, conflict and feminist theories on tension and coercion and the engines of change, interactionists on definition, anticipation, and labeling.

35:20 The line that governs the whole lecture.

“The key thing is that most people are not one or another of these theories. We use different kinds of ways of thinking to ask different kinds of questions.”

36:33 Research methods. Survey research first, as probably the most common, with the tradeoff stated in one breath: you can ask many people efficiently and get answers to questions like whether older people support Trump more than younger people, and it is not a rich question and will not tell you about their hopes and dreams.

37:23 Secondary data analysis, and his own practice.

“Most sociologists are not actually doing those surveys. We’re using the surveys that other people have done. This is what we call existing data analysis or secondary data analysis. Someone has done a survey, or we’ll pool all of our resources, through the government, and say let’s do a big survey, and then we’ll all analyze it. 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.”

38:02 Experiments, less common in sociology because of ethics and construction difficulty, and not necessarily reflective of the world, and useful for specific questions about response and learning under different conditions.

38:55 Qualitative research: in-depth interviews, observation, participation. And its cost.

“Qualitative research is much more time intensive, in the sense that you’re spending a lot of time studying a relatively small number of people, and there’s a lot of interpretation that has to go into making these results meaningful.”

39:47 Historical and comparative methods, comparing countries or one country across time, on questions like why some societies became rich, what causes civil wars, and what happens when democracy breaks down.

Twenty-four thousand views, twenty-five times the audience of anything else in this file, and it is the recording with the least of him in it. Somebody’s syllabus is assigning it. The largest audience Cohen has ever reached with a lecture that I have seen so far came from doing the ordinary work of his job well, and it dwarfs the reach of the fights.

The lecture is also the answer to a question the rest of the file raises. Two days earlier he opened Social Problems with presidential clips and a rally chanting a slur. This is the next lecture in the same course, and there is no politics in it at all. When he teaches conservative theory he teaches it, and when he criticizes it he criticizes the logic rather than the politics.

That passage at 19:29 is the best thing in the lecture. He gives functionalism its strongest form, the fever fighting the infection, then Merton on crime, then poverty, which he calls cold-hearted while granting the logic. Then he takes it apart on the ground it stands on: the body metaphor licenses you to assume that whatever exists has a purpose, and then you reason backwards to the purpose from the assumption. Everybody has lungs, let us find out why we have lungs, therefore every society has poverty, let us find out why poverty is good. He identifies the conservatism of the theory as a consequence of the metaphor rather than as a partisan property, which is the honest version of the criticism.

And then he does something that his blog almost never does. He gives conflict theory the identical treatment structurally, notes at 20:22 that the opposition between the two is a teaching device rather than a fact about the field, and closes at 35:20 by telling students that nobody holds one of these positions and that they are question-generating instruments. A man who is often accused of prosecuting from a fixed position spends forty minutes teaching that theoretical positions are tools you pick up and put down.

Three things carry beyond the classroom.

The construction passage at 2:14 is the most useful thing in the lecture for a general reader. The pandemic exists. What the problem is remains open: the virus, the government’s handling, or the government telling people what to do. Anyone who lived through 2020 watched that fight and mostly did not have the vocabulary for it, and he hands it over in ninety seconds without indicating which answer he prefers. Set that beside the previous lecture, where he had made his own answer unmistakable, and you have the two halves of the citizen scholar problem in a single week of one course.

The March for Life passage at 9:23 is the one that would surprise readers who know him only as a combatant. He photographed the march, and he analyzes the teenagers from religious schools with the identical apparatus he applies to the Women’s March, and the analysis is respectful. The solidarity is real, the identity work is real, and the mechanism is the same on both sides. He does not editorialize. For your purposes that passage answers the charge that he cannot handle a movement he disagrees with as a sociological object rather than a target.

And at 37:23 he tells his students something about himself that the rest of the record only implies. He does not collect data. He analyzes what other people collected. That sentence explains an enormous amount: why measurement is his obsession, why he audits instruments rather than gathering fresh evidence, why his reflex when a number moves is to ask whether the survey changed, and why his disputes with ethnographers run the way they do. A researcher who lives entirely inside other people’s instruments will be exquisitely sensitive 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 the same source, and he states it in one aside while explaining research designs to freshmen.

Two weaknesses, both small.

The qualitative passage at 38:55 is thin next to the survey passage. Interviews, observation, participation, time intensive, lots of interpretation required. It is accurate and it does not convey why anyone would do it or what it can reach that a survey cannot. The survey section gets its tradeoff stated with real specificity. The qualitative section gets a cost with no corresponding account of the benefit, and a student would come away unsure why that method exists.

And intersectionality gets nine minutes of names and a promise to return, which is fine for an intro lecture, and it is the one section where he tells students to go look people up rather than explaining the concept. He names Collins, Davis, and Lorde and never states what the concept claims, beyond that different inequalities interact. Given how contested the term is in public argument, a definition would have served students better than a reading list.

Transcript flags. At 15:24 “email durkheim” is Émile Durkheim (1858–1917). At 17:14 and 25:14 “enemy” is anomie. At 18:46 Robert Merton (1910–2003) is credited with “dysfunctionalism,” and the term he actually coined is dysfunction, alongside manifest and latent functions; check the slide, since the concept of latent function is what the crime and poverty examples illustrate, and the poverty argument is usually credited to Herbert Gans (1927–2023) in his 1971 essay on the uses of poverty. At 19:50 “let’s see why we have loves” is lungs. At 22:37 Karl Marx (1818–1883) is dated correctly in the chapter title. At 23:03 “the economic director of capitalism” is probably the economic structure. At 31:35 George Herbert Mead (1863–1931). C. Wright Mills (1916–1962) at 10:19, The Sociological Imagination published 1959. Patricia Hill Collins (b. 1948), Angela Davis (b. 1944), Audre Lorde (1934–1992). At 5:12 “emergency claims making” is emergence and claims making, two stages run together.