You already suspected this. Now there’s a number for it, and the number is worse than you thought.
A 2022 study in Nature, “Quantifying hierarchy and dynamics in US faculty hiring and retention” by K. Hunter Wapman, Sam Zhang, Aaron Clauset and Daniel Larremore, mapped the doctoral training and employment of 238,676 tenure-track professors across 387 PhD-granting US universities over the decade 2011 to 2020. It’s the most complete picture of academic hiring anyone has assembled, and it says the thing nobody on a hiring committee wants said out loud: where you got your PhD predicts your career more cleanly than almost anything about the work you did there.
Start with the concentration. Just 20.4 percent of PhD-granting universities produced 80 percent of all tenure-track faculty in the country. Narrow it further and it gets starker: five universities, Michigan, Wisconsin-Madison, Berkeley, Harvard and Stanford, trained about one in eight American professors between them, 13.8 percent out of hundreds of doors. If you didn’t train at a top-fifth institution, four out of five of your future colleagues did.
The authors reconstructed a prestige ranking purely from who hires whom, then asked how often people move up it. The answer, depending on the field, is that only 5 to 23 percent of faculty end up at a university more prestigious than the one that granted their doctorate. In the humanities, upward mobility was about 12 percent. For most people the hierarchy is a one-way street: you’re hired at your level or below, almost never above. Prestige flows downhill, and it rarely climbs back.
The authors are direct about what this implies. Prestige is overvalued in hiring, and a researcher has little chance of landing somewhere more elite than where they trained. Your doctoral institution is less a starting line than a ceiling.
The same study followed who leaves, and attrition is not evenly spread. Faculty trained at more prestigious institutions stay longer, and the group with the lowest attrition of all is white male faculty from top programs. The paper also finds that recent gains in women’s representation come mostly from demographic turnover and earlier hiring changes, and that on current trends most fields will not reach gender parity without further change. The hierarchy is not only who gets in. It is also who stays.
I want to be careful, because a bleak number invites overreading.
This is US, tenure-track, PhD-granting universities only. It says nothing direct about liberal-arts colleges, industry, national labs, or other countries, and careers outside that slice may look different.
“Prestige” here is defined by the hiring network itself, who places graduates where, not by an outside measure of quality. That is deliberate, and it is also circular by construction: the network draws its own hierarchy. It tells you how the system sorts people, not that the sorting tracks merit. The authors’ own reading is the opposite, that the sorting overweights pedigree.
And it is a snapshot of one decade. It measures placement, not the quality of anyone’s scholarship. Plenty of the best work in any field is done by people the hierarchy placed far from the top. The study does not dispute that. It shows that doing the work and being placed by the work are two different games.
So what do you do with a fact you can’t change? You can’t re-run your PhD at Berkeley. That door closes the day you finish.
The answer is narrow and a little cold. If the one variable that most predicts your placement is fixed, then the variables you can still move are worth more, not less. Your output is one. Whether the right people can find that output is the other, and it is the one most people leave entirely to chance. I run a company premised on that second lever, so discount me accordingly. But the logic holds without me: in a system that hands a large advantage to a pedigree you can’t acquire, the small controllable edges are the whole game for everyone who didn’t get the pedigree.
None of this promises that being findable will vault you up the hierarchy. This paper is fairly clear that almost nothing vaults you up the hierarchy. The claim is more modest: the prestige you were assigned at 27 is doing a lot of quiet work in rooms you’ll never sit in, so the parts of your visibility you actually control are worth taking seriously. The alternative is to leave the one lever you’ve unpulled, and hope the committee reads your papers instead of your CV’s letterhead. How often do you think they do?
]]>Everyone gives the same advice to a new PhD student. Find a great mentor, and do what they do. The first half is right. The second half, according to a large study of how scientific careers actually unfold, is close to backwards.
The paper is Yifang Ma, Satyam Mukherjee and Brian Uzzi, “Mentorship and protégé success in STEM fields”, in PNAS (2020). They assembled genealogical records on nearly 40,000 scientists who published 1,167,518 papers in biomedicine, chemistry, math and physics between 1960 and 2017, and asked a question most of us only answer with anecdotes: does having a great mentor actually change where you end up, and if so, what is it that gets transmitted?
Their headline: mentorship is associated with a 2x to 4x higher chance that a protégé wins a major prize, gets into the National Academy of Sciences, or becomes what they call a superstar, compared with matched protégés who had similar talent and similar starting odds. Not a small nudge.
Here is the clever part of the design. They found clusters of mentors with similar records and reputations, who attracted protégés of similar talent. Within each cluster, one mentor had something extra: they went on to win a prize. Crucially, the skill that made them prizeworthy existed before the prize was awarded, which means their protégés were exposed to it early, before anyone knew to call it special.
And the authors are specific about what that hidden skill is. In their words, it is “skill in creating and communicating prizewinning research.” Read that twice. It is not only about having good ideas. It is about creating and communicating them. The thing that separates the future laureate from an equally credentialed peer is partly the ability to make important work legible to other people, and that ability rubs off on the students standing nearby. I run a company built on the second half of that sentence, so take my enthusiasm with salt. But it is their sentence, not mine.
Now the finding that should change behavior. Protégés did not succeed most by working on their mentor’s topics. They succeeded by studying original topics, and by coauthoring only a small fraction of their papers with the mentor. The students who broke away and staked out their own ground did better than the ones who stayed in the mentor’s lane. Mentorship also predicted something related: protégés who pioneered their own research topics, and who were more likely to be midcareer late bloomers than instant stars.
So the transmission is real, but it is not imitation. You absorb the mentor’s tacit skill for making and communicating important work, and then you point it at a question that’s yours. Copying the topic is the mistake. Copying the craft is the point.
I like this paper, and I don’t want to oversell it.
It is observational. Nobody randomly assigned students to great mentors. The authors work hard with matched comparisons to rule out the boring explanation, that great mentors simply attract great students, and that matching is the backbone of the whole argument. But matching narrows a gap, it does not close it. Some of the 2x to 4x could still be selection they could not see.
The models explain 34 to 44 percent of the variance in protégé success. That is a lot for social data, and it also means most of what decides a career still sits outside the model. Mentorship matters. It is not destiny.
And it is STEM, mostly the elite tip of STEM, measured through prizes and academy membership and “superstardom.” Those are the outcomes at the top of a very steep pyramid. Whether the same pattern holds for a solid, non-prizewinning career, or in the humanities, this paper cannot say.
If you’re early, the practical read is unusually clean. Choose a mentor for the tacit skill you can only learn by proximity: how they frame a problem, how they decide what’s worth doing, how they make it land with other people. Then don’t spend your twenties as their coauthor. Take the craft and leave the topic.
If you’re the mentor, the uncomfortable implication is that the most valuable thing you pass on isn’t the project you hand a student, and not the papers you write together. It’s a way of working that you might not even know you’ve, including how you make your work visible. You can’t teach that by keeping students on your own topics. You teach it by letting them leave.
I keep coming back to that one word in the abstract: communicating. The study says the hidden, mentorable skill behind a celebrated career includes the ability to make the work known. If that is right, then “just do good science and it will speak for itself” is not merely incomplete advice. It is the one thing the laureates apparently did not believe. Did yours?
]]>You submit a paper and then you wait. Weeks pass, then months. When the reviews finally land, one of them is three sentences long and the other clearly read a different paper than the one you wrote. Now turn the desk around. Your own inbox is holding four review requests, three from editors you’ve never met, and you’re the same tired person who owes everyone else a review by Friday. If either of those scenes made you wince, you already understand the problem this paper is chewing on.
The paper is a 2025 preprint, “AI and the Future of Academic Peer Review” by Sebastian Porsdam Mann and colleagues (arXiv:2509.14189). Its argument, said plainly and up front: a language model can responsibly take over narrow, supervised parts of peer review, and it must not take over the human judgment that makes a review worth reading.
The authors are blunt about the state of things. They describe “long publication delays, escalating reviewer burden concentrated on a small minority of scholars, inconsistent quality and low inter-reviewer agreement,” on top of the usual systematic biases. And the discouraging part: “decades of human-centered reforms have yielded only marginal improvements.” So the load really is piled on a few people, which is the second scene above, and the fixes we already tried barely moved it.
Here is the useful half. The paper argues that “targeted, supervised LLM assistance can plausibly improve error detection, timeliness, and reviewer workload without displacing human judgment.” Think of the mechanical jobs: checking whether a proof holds, flagging a citation that is missing, catching a statistic that cannot be right, pointing a reviewer at the three claims out of fifty that actually need a human to think hard. The authors go further and sketch fine-tuned, retrieval-augmented, and multi-agent systems that could make review “more reliable, auditable, and interdisciplinary.” Keep the word supervised in view. It is carrying most of the weight in that sentence.
Now the half that should worry you. The authors list the failure modes as “hallucination, confidentiality, gaming, novelty recognition, and loss of trust,” and they refuse to file these under details to patch later. In their word, these problems are “constitutive” of what makes review legitimate, which makes them “governance choices as much as technical capacity.”
Two of them are worth stopping on. Confidentiality: your unpublished manuscript is a secret you handed to one editor, and pasting it into a commercial chatbot quietly breaks that promise for everyone who trusted you with it. And novelty recognition, which is my shorthand for the model faking competence: a language model is very good at producing a confident, fluent review of a paper it doesn’t actually understand, and a genuinely new idea is close to the one thing it never saw in training. A tired human at least half-knows when they are bluffing. The model does not know at all.
Where do I land? The paper’s own recommendation is the sober one, and I agree with it: reject both “uncritical adoption [and] reflexive rejection,” and run instead “carefully scoped pilots with explicit evaluation metrics, transparency, and accountability.”
Two caveats before you quote it at your next lab meeting. First, this is a preprint. It has not itself been through peer review, which is a small irony worth saying out loud, so treat its framing as a well-argued position and not a settled result. Second, “supervised” is cheap to write and expensive to enforce. The thing that goes wrong is not a rogue AI seizing the journal. It is a swamped reviewer who lets the model draft the whole review and then skims the output, and no editor can see that from the outside.
If you submit papers, the confidentiality problem is already yours, today, policy or no policy: someone in your review pile may be feeding your manuscript to a model right now. If you review papers, the honest question the paper hands you is which parts of your own reviewing are the mechanical parts a machine could take, and which parts are the judgment you’re willing to sign your name under. That line is the entire debate, compressed.
The paper doesn’t draw the line for you, and I’m not going to pretend I can either. But once you’ve waited most of a year for two careless reviews, “let the machine help, carefully” stops sounding reckless and starts sounding like the least bad option on the table. I might be wrong about how careful anyone will actually bother to be.
]]>A talented postdoc from a smaller country takes a two-year contract in Boston, then a better one in Zurich, and somewhere along the way stops planning the trip home. You have probably watched this happen to someone. Countries that train researchers keep losing them to countries that pay better and cite better, and everyone from the European Commission down has spent decades trying to reverse the flow. The obvious lever is money: hand the returning scientist a grant, and maybe they stay.
Does that lever actually work? A new paper in Quantitative Science Studies is, as far as I can tell, the first study to check the question quantitatively, and the short answer is: a little, and mostly for the people you would least expect.
The paper is Marco Seeber and colleagues, “The effect of public funding on long-term relocation of experienced researchers”, published in 2025. They studied the Marie Curie Career Integration Grant, or CIG, a scheme under the European Union’s Seventh Framework Programme that was built to counter the European brain drain to third countries and to help experienced researchers settle for the long term inside an EU member or associated country. It is exactly the kind of “come back and stay” incentive that governments love to announce and rarely measure.
Here is the finding, stated the way the authors state it. Getting the grant was associated with a 9.4% higher chance of long-term relocation in the host institution, and an 8.2% higher chance in the host country. Not nothing. Not enormous.
How did they measure it? They took the applicants to three CIG calls, from 2011, 2012, and 2013, reconstructed each person’s career from Scopus records and their CVs, and then compared where the funded applicants ended up against the applicants who applied and did not get the money. So the comparison is not grantees versus the general population; it is grantees versus their own rejected peers, which is a much fairer test.
The part I find most interesting is who the grant helped. The effect was concentrated in the applicants who had the fewest other options: people without a tenured position, scientists in the soft sciences, people who were not returning to their home country, and people moving to institutions that are not highly ranked. For a well-funded scientist heading to a top lab, the grant changed little, because that person already had other money and other offers. For someone with none of that, it moved the needle. A subsidy matters most to the person who cannot easily replace it. That is almost boring to say out loud, but it is good to see it hold up in the data.
And one honest negative that the authors report: the grant showed no relationship with landing a tenured position, and none with scientific productivity. It helped people stay in place. On this evidence it did not make them more successful once they stayed.
Now the caveat, because this is where I get twitchy. This is a comparison, not an experiment. Nobody flipped a coin to decide who got funded. The people who won CIG grants won because a committee judged their proposals stronger, and stronger proposals tend to come from people who were already more likely to build a stable career wherever they landed. The authors are careful with their verbs, “related to,” “associated with,” and so should we be. A 9.4% gap between grantees and rejected applicants could be the grant, or it could be the same qualities that got them the grant in the first place. Matching and careful career reconstruction narrow that gap. They do not close it.
It is also one programme, run by one funder, over a handful of years, on one continent. Whether an Israeli or Canadian or Brazilian version of the same idea would buy the same 8% is an open question, and this paper cannot answer it.
So why should you care? If you’re an early or mid-career researcher weighing whether a reintegration grant is worth the application effort, this gives you a realistic number instead of a brochure promise. The grant nudges your odds. It doesn’t rewrite your career, and it’ll do the most for you precisely if you are the applicant with no fallback. If you care more broadly about where science concentrates, the sobering reading is that even a targeted, reasonably designed grant buys single-digit percentage points against the pull of richer, better-cited places. The gravity is strong.
I would like to see the randomized version of this, if any funder is ever brave enough to run one. Until then, single digits is what we’ve. Would you move countries for an 8% nudge, or does that just confirm you were leaning that way already?
]]>Years ago, at Automattic, I looked at what made a blog post collect likes. The strongest predictor I found wasn’t the topic, the length, or the time of day. It was whether the same author’s earlier posts had collected likes. Likes bred likes. If you already had them, you got more. If you didn’t, you mostly didn’t.
Sociologists have a name for this: the Matthew effect. The rich get richer. In citation studies it goes by “preferential attachment” or “cumulative advantage,” and it is the standard answer to a question every researcher asks: why do a handful of papers get cited thousands of times while most sink without a trace?
A recent preprint says that answer is incomplete. In “Community-centric modeling of citation dynamics explains collective citation patterns in science, law, and patents” (arXiv, January 2025), Sadamori Kojaku and colleagues argue that whether a paper keeps getting cited is not mainly about the paper. It is about the community the paper sits in, and where that community’s attention happens to drift. Success, in their framing, is less about you and more about the crowd you run with.
Pure cumulative advantage has an awkward blind spot. If citations only flow to papers that already have citations, then a paper ignored at birth should stay ignored forever. But that is not what happens. Some papers sit unread for years, sometimes decades, and then suddenly surge. The literature calls them sleeping beauties, and the delay is called delayed recognition. The authors note that models of individual citation trajectories fail to reproduce this. You cannot get a sleeping beauty out of a machine whose only rule is “the popular get more popular.”
Their move is to stop modelling a paper as a lone object with intrinsic pull, and start modelling the citing side. A citation, in this view, is a decision made by the authors who are writing now, and those authors cluster into communities in a high-dimensional “knowledge space.” Communities drift through that space over time. As a community moves, the work near its current location gets cited, and work it has moved away from goes quiet. That drift produces recency even though the model is never told about time. There is no explicit aging term. Papers age because attention wanders off.
It also explains sleeping beauties in a way rich-get-richer cannot. A paper published in a sparse, out-of-the-way corner of knowledge space waits. Years later a community wanders into that corner, and the old paper wakes up. The model reproduces this across all three corpora it was tested on: science, law, and patents, using a newly available U.S. case-law dataset for the legal side. The three domains share the same heavy-tailed patterns, which is part of why the authors think they are onto a general mechanism rather than a quirk of academia.
The model keeps three ingredients working together, as the authors describe them: relevance, which is how close a paper sits to where a community’s attention is; cumulative advantage, the old rich-get-richer pull, still present, just no longer alone; and fitness, something like intrinsic quality. Preferential attachment is not wrong here. It is one of three, not the whole story.
Two cautions before anyone gets excited. First, this is a preprint. It hasn’t been through peer review as I write this, and preprints get revised, sometimes heavily. Read it as a strong hypothesis, not a settled result. Second, and more important, it is a model. It reproduces collective, aggregate patterns: the shape of the distribution, the abundance of sleeping beauties across a whole field. It says almost nothing about your specific paper. Knowing that attention drifts through communities doesn’t tell you when, or whether, a community will drift toward your particular corner. A model that explains the crowd is not a plan for the individual.
Carefully, maybe one thing. If citations track where communities are looking, then being legible to the community you actually belong to isn’t vanity. It’s how you get seen at all. That is not the same as saying “promote harder and you’ll get cited.” The paper makes no such claim, and I won’t put it in the authors’ mouths. It’s closer to a reframing: a paper nobody in your community can find, or place, is a paper sitting in an empty corner of the space, waiting for a drift that may never come.
I find the collective framing convincing, and I might be wrong about that. If the mechanism is real, the uncomfortable implication is that a lot of what we call a paper’s quality is really a statement about its neighbourhood. Would your best-cited work have done as well one community over? I genuinely don’t know. Neither, quite, does the model.
]]>“Be more interdisciplinary.” Every funding call, every keynote, every senior colleague who has run out of specific advice tells you the same thing. Reach across fields. Borrow from the neighbors. It’s the safest thing you can say to a young researcher, because nobody will ever call it wrong. Safe advice and useful advice aren’t the same thing, though, and I’ve always wanted to know whether crossing fields actually makes your work land harder, or just makes it harder to do.
A 2025 paper by Alex J. Yang in the Journal of Information Science (doi.org/10.1177/01655515251330614) gives a straight answer, at least on average. Looking at 38 million papers across all fields from 1960 to 2020, Yang finds a clear positive link between how interdisciplinary a paper is and its “disruptive potential.” Papers that pull ideas from further apart tend to reshape their area more, not less. And the link has been getting stronger over time, more so in STEM than in the social sciences and humanities.
Two words are doing a lot of work here, so it is worth being precise. Interdisciplinarity is an ex-ante input: what you drew on, read off from how varied your references are. Disruption is a post hoc score: whether later work cites your paper instead of the things it built on (you eclipsed your predecessors) or alongside them (you extended them). One is fixed the day you publish. The other only makes sense years later, once the citations arrive.
Now the part that undercuts the slogan. Over the same 60 years, interdisciplinarity went up and disruption went down. And bigger teams, which produce more interdisciplinary work, are less likely to be disruptive. So “be interdisciplinary” and “be disruptive” have been drifting in opposite directions across science as a whole, even while, paper by paper, the two move together. Yang also reports that the interdisciplinarity-to-disruption coupling is stronger inside larger teams, which does not obviously square with larger teams being less disruptive overall. I’m reading this from the abstract, so I would check the full text before leaning on that one.
To explain how interdisciplinarity feeds disruption, Yang points at a few candidates: how diverse the team is, how varied the references are, and delayed citation recognition. Those are the plausible machinery behind the correlation. They’re also, notice, three more things measured from the same paper trail, which is worth keeping in mind before you treat any of them as a recipe.
Both quantities are built from citation data, and citation data flatters some things and hides others. “Interdisciplinary,” here, mostly means you cited far-apart fields. Stapling a neuroscience reference onto an economics paper counts, whether or not you did anything real with it. The disruption index is a clever proxy for “this changed the conversation,” but it’s a proxy: it’s sensitive to how many references you’ve, to a field’s citing habits, and to which database you trust. And “delayed citation recognition,” one of the mechanisms Yang points to, is a polite way of saying interdisciplinary work often sits ignored for years before it disrupts anything, if it ever does. The average is positive. The variance around that average is the part that lands on you personally.
So what does a researcher deciding how far to reach do with this? Not much, if you read it as “reach further, score higher.” A 38-million-paper average is a fact about the population, not a promise about your next paper, and the team-size paradox says the lever isn’t simply adding collaborators and citations from more fields. The honest reading is narrower. When a problem genuinely needs a second field, the evidence says crossing over tends to pay off in impact, and pays off more now than it used to. When the problem does not need it, importing a distant reference to look interdisciplinary is just noise that a citation counter happens to reward. And the payoff is not evenly spread: the effect is stronger in STEM than in the social sciences and humanities, so the same reach across fields is not worth the same everywhere. If you work outside the hard sciences, the average that this paper reports is quieter than the headline sounds.
I like this paper because it refuses the easy version of its own headline. Interdisciplinarity is linked to disruption, yes. Disruption is also falling while interdisciplinarity rises. And bigger teams do more of one and less of the other. That is messier than “cross fields, win bigger,” and messier is usually closer to true. Whether your particular bridge between two fields disrupts anything, the index will only tell you in a decade. Would you make the same bet if the payoff were that slow, and that uncertain?
]]>I read a lot of papers. Lately I catch myself doing something I never used to do. Before I trust a result, I check whether the paper is real: is the journal a real journal, are the authors real people, does the data look like data or like something a machine dreamed up on request. That check used to feel paranoid. Now it feels like basic hygiene, and I don’t like that it does.
That unease has a name and a number now. In November 2025, Bernhard Sabel and Dan Larhammar published a Perspective in Royal Society Open Science called “Reformation of science publishing: the Stockholm Declaration”. Larhammar is a past president of the Royal Swedish Academy of Sciences, so this is not a blog rant, it is a formal shot across the bow. Their message, stated flatly: fake and low-quality science is now being produced at industrial scale, and they call it “arguably the largest science crisis of all time, threatening to erode people’s trust in research.”
The authors name three threats to the literature. First, for-profit publishers, who charge steep fees for open access and take, by the authors’ own figure, between 25% and 35% profit from academia. Second, predatory journals, which publish a flood of low-quality papers with no real peer review. Third, and newest, paper mills.
The headline number is the number of fake papers: “hundreds of thousands per year,” which the paper attributes to two earlier studies. If that estimate is anywhere near right, the counterfeit is no longer a rounding error in the scientific record. It is a meaningful fraction of it.
A paper mill is a business. It manufactures articles with fabricated data, increasingly using AI-generated text, tables, and images, and then, in the paper’s own words, it “sells authorships to scientists under pressure to publish and bribes editors to publish these papers.”
The buyer is not a cartoon villain. It is usually a researcher who needs one more line on a CV to keep a job, land a grant, or finish a degree, working inside a “publish or perish” culture that counts papers instead of reading them. The paper’s point: the incentive comes first, the fraud follows. The metrics we chose, publication count, impact factor, h-index, get gamed, because any scoreboard eventually does.
The declaration was drafted at a conference at the Royal Swedish Academy of Sciences on 9-10 June 2025, and signed by more than two dozen researchers, including the image-integrity sleuth Elisabeth Bik and a board member of PubPeer. It asks for four things:
I believe the diagnosis. The cure I’m less sure about.
The scary number does a lot of work, and it arrives as an estimate carried across a citation, not something I can audit from the page. I would have liked a table. The four recommendations, meanwhile, are the ones the reform movement has been making for years. The paper knows this: it prints its own table of predecessors, Leiden 2015, DORA 2012, Plan S 2018, and more. Good ideas usually don’t fail for lack of being restated one more time. There is even a reply already in the same journal, titled “Why the Stockholm Declaration will never work”, which argues the plan rests on assumptions that are too optimistic. Read it next to the declaration, not instead of it.
Two smaller notes, for honesty. The lead author founds and holds a stake in a science-integrity foundation, disclosed in the paper. That doesn’t make him wrong, it is context. And the paper carries a line I keep re-reading: “We have not used AI-assisted technologies in creating this article.” You have to declare that now. That’s the world we write in.
If you’re a researcher trying to be found and cited, and your work is honest, you might file this under someone else’s problem. It isn’t.
Fakes don’t only add noise, they tax trust, and the tax is flat. When a reader, a reviewer, or an AI answer engine can’t cheaply tell your real result from a manufactured one, the cheap move is to discount all of it. Skepticism gets applied to the whole field, and the honest pay it right alongside the frauds. The entire reason to put real work where people can find it is that being found is supposed to mean being believed. Erode the trust base and visibility stops paying: you can be the most discoverable paper in your subfield and still lose, if the currency of the field turns out to be counterfeit.
So maybe the check I started doing, is this paper real, isn’t paranoia anymore. Maybe it is just what reading looks like now. I would rather it went back to being paranoid. Do you think it’ll?
]]>You have seen the badge. Once or twice a year your LinkedIn feed fills with the same post: someone you know, usually a professor, announces they made the “World’s Top 2% Scientists” list. There is often a certificate. Sometimes a screenshot of a spreadsheet with one row highlighted in yellow. The comments say congratulations, and they mean it.
The list is real, and it is not junk. It comes from John Ioannidis and colleagues at Stanford, who built a set of science-wide author databases of standardized citation indicators out of Elsevier’s Scopus. The logic is reasonable. Raw citation counts are unfair to compare across fields, so they normalize them, fold six measures (total citations, three author-position citation counts, the h-index, and a co-authorship-adjusted h-index) into one composite, and publish the top slice inside each field. That slice is what everyone now calls the top 2%. It comes in two flavors, a single-year snapshot and a career-long one.
Before you print the certificate: in at least one field, the list tracks a US postal code about as closely as it tracks merit.
A 2025 paper in the Journal of Informetrics took the 2024 edition of the list, picked one field, and counted. Péter Sasvári and Gergely Ferenc Lendvai, both at Ludovika University of Public Service in Hungary, chose legal studies, partly because “Law” is one of the fastest-growing fields on the list: since the first edition it has grown 811% in the career-long ranking and 491% in the single-year one. They pulled every law scholar from both rankings, 766 people in all (393 in the single-year list, 373 in the career-long), then scraped roughly 50,000 of their publications from Scopus to see who was citing whom.
The United States is heavily overrepresented. In the career-long ranking, 213 of 373 legal scholars were based in the US, about 57%. In the single-year ranking it was just under half, around 46%. The same person tops both lists: Cass R. Sunstein, whom the authors call “leagues above his peers.” The best-ranked non-US scholar, David Weisburd, sits far behind. In the single-year data Sunstein’s rank is 242 and Weisburd’s is 4,787. Across the board the US median rank beats the non-US one (131,187 against 163,484 in the single-year set, where a lower number is better).
The gap is not really about who writes better law. It is about where the writing lands. The top 10 journals for US scholars are, all ten of them, American law reviews, and not one is published by a “Big 5” commercial house (Elsevier, Sage, Wiley, Taylor and Francis, Springer). They are the in-house reviews of elite law schools: Harvard, Chicago, Yale, Stanford, Penn. The Harvard Law Review is the single most popular venue, and articles there pull a mean of about 101 citations each.
Non-US scholars publish somewhere else entirely. Their top journals are European and international ones, with far thinner citation means. Their most common venue by article count is actually Nature, but those articles average just 5.39 citations. Not a single Global South journal appears in any of the top-journal tables. And when Sasvári and Lendvai mapped the citation network, US scholars mostly cited other US scholars, a dense, inward-looking web. The authors’ description of the ideal candidate for the list in law is worth quoting in full: “one has to fit three criteria: publish frequently in Ivy League universities’ journals, cite U.S. scholars, and it definitely does not hurt to be American either.”
Put the pieces together and, in this field, the ranking is not really measuring the scientist. It is measuring the neighborhood the scientist publishes in. Citations pool where the papers already cluster, the composite standardizes that pool, and the ranking hands the trophy back to where it started. Call it the Matthew effect with a passport: citations accrue to those who already have them, and here “those” tend to carry a US affiliation. The authors describe an American “citation ecosystem that reinforces its central position,” frequently at the expense of Global South scholarship, which stays on the periphery of both the publishing and the citing.
I want to be careful, because it’s easy to over-read one paper. This is a single field, and law may be the strangest field you could pick for citation counting. A large share of US legal scholarship lives in student-edited law reviews that behave nothing like a physics or a medical journal, many of them without clean metadata for the databases to even read. Law is also split by legal system, common law against continental, so a scholar’s work is often local by nature. Sasvári and Lendvai point out that researchers working in Islamic, customary, or Jewish law are “visibly non-existent” on the list, and that Scopus itself under-indexes non-English work. None of this is claimed to generalize. I wouldn’t assume the same US share turns up in molecular biology or applied mathematics.
The authors are also not telling you to burn the list. Their words: “Completely disregarding ranking such as this one is, of course, not necessary.” What they are saying is narrower, and more useful. In legal studies the list carries a loud geographic accent, so it should not be read as the measure of who the best legal scholars are.
Use it for less than the certificate invites. If you’re on it, enjoy it. It’s a fair sign your work gets read and cited, which isn’t nothing. If you’re ranking other people with it, for hiring, promotion, or a tenure file, remember that in at least one field it rewards a postal code about as much as a mind. “This person made the top 2%” honestly means “this person publishes in well-cited venues.” That is worth something. It is not the same sentence as “this person is among the best scientists alive,” even though the yellow highlight invites you to read it that way.
And if the world’s top 2% is this sensitive to where you happen to publish, what is it we’re actually ranking?
]]>Eugene Garfield is the reason we count citations at all. He built the Science Citation Index, he gave us the impact factor, and in 1980 he wrote a short piece in Current Contents with a plain title: “Premature discovery or delayed recognition-why?” In it he named a thing every researcher half-suspects is real. Some papers sit almost uncited for years, then wake up and get cited a lot, long after most of the people who could have used them have moved on. Anthony van Raan later borrowed a nicer name for these papers from Perrault and the Brothers Grimm: sleeping beauties.
Here is the joke. Garfield’s own paper on delayed recognition was itself delayed-recognized.
That’s the finding of a 2025 preprint by Tariq Ahmad Mir and Marcel Ausloos, Forsaking your own (arXiv:2512.16943). It is a preprint, so treat it as a strong claim, not a settled one. But the claim is clean, and I’ve been chewing on it for a week.
Garfield published the paper in 1980. By 2004 it had collected about 10 citations. That is 0.4 citations a year for a quarter century, from the single most cited name in the field that invented citation counting. For 23 of those years it never once cracked more than one citation in a year. The authors call it a deep sleep of 28 years.
Then it woke. Counting up to 15 September 2025, they find 45 citations in Scopus, 93 in Web of Science, and 205 in Google Scholar, which they reconcile to 214 unique citing papers (234 if you count Garfield citing himself). A paper that averaged one citation every two and a half years for decades is now a normal, respectably cited paper. On the beauty coefficient, a standard sleeping-beauty score, it lands at 159.55 and 144.62 across two citation peaks, in 2018 and 2023. Whatever threshold you pick, it qualifies.
What woke it up? Van Raan’s 2004 paper, the one that rechristened these papers “sleeping beauties.” After 2004 Garfield’s paper gets cited far more, and Mir and Ausloos note that in 141 of the 187 citations it collected after van Raan, roughly three quarters, it is co-cited with van Raan. The old paper rode in on the coattails of the new one.
There is an uncomfortable lesson here, and it is not about Garfield. Van Raan does not even cite Garfield’s 1980 paper. The revival happened anyway, because a well-placed, well-cited paper made the topic legible again and people went looking for the roots. Garfield’s paper needed a prince. It got one by luck, not because the work finally spoke for itself.
That’s the part I keep turning over. We tell ourselves that good work eventually gets found on merit. This is a case study in the opposite. The idea was correct. The author was famous. The paper still sat there for 28 years, and what saved it was somebody else’s visibility, not its own quality.
I run a company whose whole premise is that good research gets ignored and that being right isn’t enough. So I’m the last person you should trust to be skeptical here, and I want to be anyway.
“Delayed recognition” is not a fact of nature. It is a definition, and the definitions are somewhat arbitrary. Garfield’s own rule of thumb was 10 or fewer citations at age 10 and a tenfold jump by age 20. Van Raan added his own thresholds for the depth and length of the sleep. The beauty coefficient is yet another formula. Move the cutoffs and some sleeping beauties stop being beauties.
The authors are honest about a messier problem too. Garfield published in Current Contents, which was not peer reviewed, so Web of Science and Scopus record its metadata badly. Web of Science at one point piled the citations of 52 different articles from one 1980 issue onto a single paper. Every count above came from manual cleanup across three databases. And, again, it is a preprint. None of that kills the story. It just means the exact numbers matter less than the shape, and the shape is not in doubt.
Read this as comfort or as a warning. I mean both.
The comfort: silence is not a verdict. Garfield’s paper was not wrong for 28 years and then suddenly right. It was the same paper the whole time. If your work is being ignored, that’s information about attention, not about quality.
The warning: attention does not show up on its own, and it did not show up for Garfield either. His paper waited a quarter century for an accident. Most papers never get a famous prince to co-cite them into daylight, and most of us don’t have Garfield’s name on the byline. If nobody knows the work exists, its merit never gets a turn.
So the question the paper leaves me with isn’t whether your good work will be recognized. It’s who, exactly, you’re counting on to be your prince, and whether it’s wise to wait for one to wander by.
]]>There is a flattering idea floating around science-of-science, and it goes like this: your best work doesn’t dribble out evenly across a career. It arrives in a burst. A few golden years when almost everything you touch lands, and around them, before and after, the ordinary stuff.
That burst has a name. Liu et al. called it a “hot streak” in 2018: a bounded stretch in a creative career during which a person produces their highest-impact work. The unsettling part of their result was that the streak seems to strike at a random time. You cannot schedule it. You get one, if you get one at all. I find this idea seductive, and I distrust things I find seductive, so a new paper caught my attention.
Hongkan Chen, Lutz Bornmann, and Yi Bu asked a narrow, testable question. Not “when does your best work happen,” but “when your best work happens, is it also your boldest?” (Humanities and Social Sciences Communications, 2025) Their answer, from the careers of 21,271 economists: yes. “Disruptiveness of scientific publications is higher during researchers’ hot streaks than in their other periods of careers.” During the streak you’re not only cited more. You are also more likely to publish work that shoves the field in a new direction instead of reinforcing the old one.
That word needs a definition, because it is doing a lot of work. Some papers consolidate what we already know; later work leans on them and on their predecessors together. Other papers disrupt: later work cites the new paper and quietly stops citing what came before, as if the new paper made the old scaffolding unnecessary. The disruption index puts a number on that difference. A high score means your paper eclipsed its own references. The claim here is that hot-streak papers score higher on it.
First, the boldness is random too. Just as hot streaks land at unpredictable times, so does disruptive work: “we also observe the randomness rule for the occurrence of disruptive work.” You do not graduate into disruptiveness at a fixed career age.
Second, and this is the one I keep chewing on: “researchers’ most disruptive years are oftentimes earlier than their most highly cited years.” Your boldest paper and your most-cited paper are usually not the same paper, and the bold one tends to come first. The recognition lags the risk, sometimes by years.
I like this paper, and I don’t fully believe it yet.
It is economists. 21,271 of them is a large sample, but it is one field with its own citation culture. I wouldn’t bet the pattern transfers unchanged to molecular biology or to poetry.
The disruption index is contested. People argue about what it actually measures and how much it wobbles with database coverage and reference counts. Read “more disruptive” as “scored higher on one specific metric,” not as a verdict handed down by nature.
It is a correlation. Hot streaks and disruption travel together in this data. The paper doesn’t, and can’t, tell you that being disruptive triggers the streak, or that the streak makes you disruptive, or that some third thing drives both.
And one lever in the paper is not random at all: volume. “A higher publication volume of researchers increases the likelihood of entering hot streaks and producing disruptive publications.” Publish more, and you raise your odds of both. That is the least glamorous sentence in the paper and possibly the most useful.
If you’re staring at your own CV and wondering whether your good years are behind you or still ahead, this paper offers a small, cold comfort. The good years don’t seem to be something you earn on a schedule and then lose. They arrive when they arrive. What you can do is keep the tap open, because output feeds the odds. And if you’re sitting on a paper that feels riskier than your usual, the timing result hints that you might be earlier in a streak than the citation counts will admit for another few years.
The authors put the practical conclusion plainly, aimed at funders as much as scientists: “both scientists and funding agencies can assume the randomness of phases with important papers in scientific careers. But when these phases occur, impactful and disruptive papers can be expected.”
I’m glad they stopped there and didn’t try to sell me a recipe for manufacturing a hot streak. Because here’s what still nags at me. If the streak is random, and the disruption is random, and the two merely happen to move together, then the comforting story might collapse into something much plainer: sometimes a scientist has a good run, and during a good run good things cluster. Is that a law of creative careers, or is it just what any run of luck looks like from the inside? I’m not sure the data can tell those two apart. Are you?
]]>