Here is a story you may already know. In 1870, a German chemist analysing the iron content of vegetables misplaced a decimal point, recording spinach as having ten times the iron it actually contains. The error was copied from textbook to textbook for decades, nobody rechecked the figure, Popeye was built on it, and a generation of children were force-fed a vegetable on the strength of a typo. It is a perfect parable about the dangers of citing without verifying - which is exactly why it has appeared in academic textbooks, peer-reviewed journals, university lectures and countless student essays as the cautionary example of unchecked citation.
It is also not true.
When the criminologist Mike Sutton went looking for the original evidence, he found that the decimal-point story could be traced no further back than a 1972 inaugural lecture by the nutritionist Arnold Bender, popularised by a 1981 British Medical Journal Christmas article whose author, when Sutton asked him for his source, said he couldn't remember - possibly the Reader's Digest. There was no decimal error. Spinach's modest iron content was correctly known by the 1890s. And Popeye, in the only strip where his creator has him explain himself, eats spinach for its vitamin A. Sutton coined a term for what had happened: a "braced supermyth" - a false claim made almost unkillable because the very people using it are sceptics invoking it as a warning about the need to check your sources, without having checked this one.
That recursion is the subject of this article. Academic knowledge is supposed to be transmitted through citation: each claim traceable back, link by link, to the evidence that supports it. In practice, the chain very often behaves less like a ledger and more like a game of telephone. A careful, hedged finding is summarised by a second author, who trims the caveats. A third repeats the summary with more confidence, citing the second. A textbook states it as established. A journalist quotes the textbook. A student, reading the journalist, cites "research shows." At no point did anyone lie. And yet the thing being repeated at the end may bear only a passing resemblance to the thing that was found at the beginning. We'll call it the citation echo: trace how it works, look at the real numbers on how often it happens, walk through some famous cases, and then turn to the one part of academic training explicitly designed to stop it - the literature review.
The anatomy of an echo
The sociologist of science Bruno Latour described what happens to a scientific statement as it travels. When first made, it carries what he called modalities - qualifying phrases that tie it to its evidence: "our data suggest," "in this sample," "under these conditions," "although further work is needed." As the statement is repeated, the modalities are shed, one at a time, until it stands alone as a bare fact that no longer seems to need a citation at all. Robert Merton called that final stage "obliteration by incorporation": the idea becomes so absorbed into common knowledge that its origin disappears. This is sometimes a healthy process - nobody cites Newton when they mention gravity. It becomes a pathology when the modalities being shed were doing real work, and the bare "fact" left standing is one the original researchers would never have signed.
A typical echo runs through five stages:
- The cautious finding. A study reports a result in a specific population, under specific conditions, with stated limitations. "In our sample of 30 undergraduates, we observed a modest association between X and Y, which may reflect…"
- The simplification. A second paper, reviewing the area, summarises it in a clause: "Smith (2010) found that X is associated with Y." Population, effect size and caveats gone.
- The confident repetition. A third paper, citing the second (or citing Smith without reading Smith), writes: "X has been shown to cause Y (Smith, 2010; Jones, 2014)." Association has become causation; one study has become a consensus.
- The textbook. "It is well established that X causes Y." Sometimes with a citation, often without; always without the original conditions.
- The public version. A news article, a TED talk, a management-training slide, a student essay: "Research proves that X causes Y." By now the claim may be cited to the textbook, the talk, or nobody.
Economists have a name for the underlying dynamic. In a 1992 paper, Sushil Bikhchandani, David Hirshleifer and Ivo Welch described information cascades: situations where each person, seeing that many others before them have accepted a claim, rationally concludes the claim is probably right and accepts it too - adding one more endorsement for the next person to see. Once a cascade is under way, the number of people repeating a claim tells you almost nothing about how much evidence underlies it, because most of them are repeating each other. A hundred citations can rest on a single study. Sometimes, as we'll see, on none.
How often does it actually happen? The numbers
This would be an amusing curiosity if it were rare. The research on citation practice suggests it is not.
Consider first the most basic question: do people read the papers they cite? Two UCLA researchers, Mikhail Simkin and Vwani Roychowdhury, devised an ingenious way to estimate this without asking anyone. They tracked misprints in citations - a wrong page number, a mangled volume - to a set of famous papers, on the logic that if you copy a reference from someone else's bibliography, you copy their typo too. In one case, a paper with around 4,300 citations had accumulated 196 misprinted references, but only 45 distinct misprints: the other 151 were copies of someone else's mistake. Modelling the propagation, they concluded in their memorably titled paper "Read Before You Cite!" that only around 20% of citers appeared to have read the original; across a dozen high-profile papers they later estimated that 70–90% of citations were copied from other reference lists. The estimate has been debated - a copied reference doesn't strictly prove an unread paper - but no one has produced evidence that the practice is rare.
Next: when people do cite, do they represent the source accurately? Here the evidence is unusually good, because medicine has been auditing itself for decades. In 2015, Hannah Jergas and Christopher Baethge pooled 28 studies of "quotation accuracy" - studies where someone actually went back to the cited source and checked whether it said what the citing paper claimed. They found total quotation error rates of around 25%, with 11.9% classified as major: the source did not support the claim, or contradicted it. Their 2025 update, pooling 46 studies, put the overall rate at 16.9% with 8% major errors, found no improvement over time, and found that error rates were higher in lower-impact journals. Roughly one in twelve claims in the medical literature, in other words, is attributed to a source that does not actually support it - and this is the field with the strongest checking culture. Nobody has done comparable audits of most humanities and social-science fields, but there is no obvious reason to think they would fare better.
Finally: what does an echo look like when you map it completely? That question was answered with extraordinary thoroughness by the Harvard neurologist Steven Greenberg in a 2009 BMJ paper. He chose a single claim in his own field - that β-amyloid, the protein associated with Alzheimer's disease, is produced by and damages muscle in a condition called inclusion body myositis, a belief that had shaped research and treatment trials for two decades - and built the complete citation network: every paper that made the claim, every citation supporting it. The network contained 242 papers, 675 citations, and over 220,000 distinct citation paths. Then he traced them back to the primary data. The results were striking. The four primary studies supporting the claim - all from the same laboratory, two of them probably reporting the same data - received 94% of all citations to primary evidence. The six studies containing data that weakened or refuted it received 6%. Greenberg named three distinct mechanisms at work:
- Citation bias: supporting papers cited; contradicting papers ignored.
- Amplification: the claim's apparent support swollen by hundreds of papers - reviews, model studies, introductions - that repeated it while presenting no data of their own.
- Invention: most troubling of all, papers that had stated the idea as a hypothesis were later cited as having demonstrated it. The claim was upgraded from speculation to fact by citation alone.
He then obtained the text of grant applications funded by the US National Institutes of Health through freedom-of-information requests, and found the same distorted claim being used to justify research funding. The echo had reached the money.
Three famous echoes
The mechanisms are clearest in cases that have been forensically unpicked. Three follow - each a claim you may well have encountered in a lecture, a textbook or your own essay.
The thirty-eight witnesses
In March 1964, Kitty Genovese was murdered outside her apartment building in Queens, New York. Two weeks later The New York Times reported that thirty-eight respectable neighbours had watched from their windows for over half an hour and done nothing. The story prompted John Darley and Bibb Latané's celebrated experiments on the bystander effect - the robust finding that people are less likely to help when others are present - and it has opened the chapter on helping behaviour in introductory psychology textbooks ever since.
In 2007, Rachel Manning, Mark Levine and Alan Collins went back to the trial transcripts and archive material. They found no evidence for thirty-eight witnesses, no evidence that any witness saw the murder, and no evidence that witnesses did nothing. The attack occurred in two locations, largely out of sight; several people did call the police or shout; one neighbour went down to Genovese and held her as she died. The number 38 appears to have come from a police list of people interviewed, not people who watched. Yet the story continued to appear in textbooks four decades on, because each new textbook was citing the previous one, or Darley and Latané's summary of the newspaper account - never the primary record. The researchers' careful distinction is worth preserving: the bystander effect is real and well replicated. The story that launched it is a parable, and its uncorrected repetition, they argued, had narrowed the questions psychologists thought to ask about groups and helping for forty years.
"93% of communication is non-verbal"
You have almost certainly heard that only 7% of communication is the words, 38% is tone of voice and 55% is body language. It appears in management training, public-speaking courses, business textbooks and thousands of student essays, usually cited to "Mehrabian (1967)" or, more often, to nobody.
Albert Mehrabian's 1967 studies were real, careful, and about something very narrow. Participants listened to single words - such as "dear," "maybe," or "terrible" - spoken in tones that matched or contradicted their meaning, sometimes alongside photographs of faces, and judged the speaker's feelings. The finding concerned how people resolve inconsistent signals about attitude and liking, in a lab, with one-word stimuli. Mehrabian himself has spent decades pointing out that the 7-38-55 formula says nothing whatever about communication in general, and that applying it to a lecture, a negotiation or a job interview is nonsensical - if it were true, you could follow a talk in a language you didn't speak with 93% comprehension. It made no difference. The numbers were memorable, the source was rarely consulted, and each repetition cited the last. This is the echo in its purest form: a real study, a real author, a completely transformed claim.
The 10,000-hour rule
In 1993, Anders Ericsson and colleagues published a study of violinists at a Berlin music academy, finding that the most accomplished students had accumulated, on average, around 10,000 hours of deliberate practice by the age of twenty - more than their less accomplished peers. It was a paper about the importance of a particular kind of structured, effortful practice. Fifteen years later, Malcolm Gladwell's Outliers turned it into "the 10,000-hour rule": the idea that 10,000 hours of practice at anything produces mastery.
Ericsson objected in print, repeatedly and eventually at book length. The 10,000 figure was an average, not a threshold; some of the best violinists had done far less, others far more. It described practice at age twenty in students who were still improving - it was not the amount required for mastery, which for elite performers often ran to double that. And it applied to deliberate practice - focused, feedback-driven, uncomfortable - not to hours spent doing anything. Meanwhile, the evidence itself moved on: a 2014 meta-analysis led by Brooke Macnamara found that deliberate practice explained about 12% of the variance in performance overall - 21% in music, 26% in games, but under 1% in professional work - a long way from the popular version. None of this has dented the rule's popularity. Students cite it in essays on expertise, education and management to this day, usually to Gladwell, occasionally to Ericsson, almost never having read either.
A note on circular echoes
The internet has added a new form. In 2008, a seventeen-year-old edited the Wikipedia entry for the coati - a South American mammal related to the raccoon - to say it was "also known as the Brazilian aardvark." He made it up. Over the following years the phrase appeared in national newspapers and a book from a university press; when Wikipedia editors later tried to remove the claim as unsourced, those publications were cited as evidence that it was true. The cartoonist Randall Munroe coined the term citogenesis for this loop: an unsourced claim is published, repeated by a "reliable" source, and then cited back to that source as verification. Students who use Wikipedia as a starting point (which is fine) rather than a citation (which is not) should understand that its footnotes can be part of the echo rather than a way out of it.
Why it happens: the honest reasons
It would be comforting to blame laziness, and laziness plays a part. But the citation echo is mostly produced by ordinary incentives acting on ordinary people.
- Secondary sources are easier and better written. A textbook explains the finding clearly in a paragraph; the original is thirty pages of methods and hedges behind a paywall. Rational people read the paragraph.
- Citations are a currency of confidence. A claim with a reference attached looks supported, whatever the reference says. Writers under pressure to appear authoritative reach for one; readers rarely follow it.
- Simplification is a service - until it isn't. Summaries exist to strip detail. The trouble is that a summariser can't always tell which details were load-bearing. "In a sample of 30 undergraduates" feels like clutter; it is actually the boundary of what the study can tell you.
- Positive, clean findings travel further. Greenberg's 94/6 split is the general pattern: results that confirm an appealing story get cited; messy, null and contradictory results do not. The literature that reaches you has been filtered for narrative.
- Memorable numbers are sticky. 38 witnesses. 7-38-55. 10,000 hours. A tenfold decimal error. Precision is persuasive even when - especially when - it is spurious.
- Nobody's job is to check. Peer reviewers rarely verify that cited sources say what they're claimed to say; Jergas and Baethge's central recommendation was simply that editors start spot-checking. Until then, the only person likely to trace a chain is a reader who has decided to.
What this means for students
Notice where the student sits in the five-stage echo: at the end. The student essay is where every upstream simplification lands, and it is written by the person with the least time and the least access to the originals. That produces two recurring failures markers know well.
The first is the textbook fact stated as research finding. "Research shows that 93% of communication is non-verbal (Mehrabian, 1967)." A marker who knows the field sees instantly that the writer has not read Mehrabian, because nobody who had could write that sentence. The citation, meant to signal rigour, signals its absence.
The second is the laundered secondary citation. The student read about Smith (2010) in Jones (2018) and cites Smith directly, as though they had read it. This is why referencing conventions insist on "Smith (2010, cited in Jones, 2018)": not as pedantry but as honesty about where in the chain you actually stand. It also protects you - if Jones misrepresented Smith, the error is visibly Jones's, not yours.
The deeper problem is that a piece of writing assembled from echoes cannot do the thing that earns marks at university. Descriptive writing repeats what others have said; critical writing evaluates it - and you cannot evaluate a finding you have only met second-hand. You don't know its sample, its method, its effect size, or its own authors' caveats. You cannot weigh it against a contradictory study you have never seen because it was cited 6% as often. You can only pass the echo along, slightly louder.
The literature review: the discipline that exists to break the chain
All of which explains why every dissertation, and most serious essays, require a literature review - and why it is so much harder than it looks. A literature review is not a list of what people have said. It is the systematic attempt to go back up the chain: to find the original studies rather than the summaries of them, to compare what different authors have made of the same evidence, to notice where a "consensus" is actually one lab cited ninety times, to weigh strong studies against weak ones, and to synthesise all of this into an honest map of what is actually known, what is disputed, and what is merely repeated. In Greenberg's terms, a good literature review is a small claim-specific citation network built by hand - and the moment you find your own version of the 94/6 split is the moment you have something genuinely worth writing.
Concretely, that work involves:
- Finding originals. Following every important claim to the primary study - using the reference list of the secondary source as a map, not a destination. Your library's database access exists for exactly this.
- Reading for modalities. Recovering what Latour would call the shed qualifiers: sample, setting, measure, effect size, and the authors' own stated limitations. These are the boundaries of the claim.
- Comparing interpretations. Setting two or three secondary accounts of the same study side by side. Where they diverge, someone has simplified - find out who, and how.
- Looking for the 6%. Actively searching for studies that failed to replicate, contradicted or qualified the claim, because citation bias guarantees they will not find you.
- Evaluating evidence, not counting it. Ten papers repeating a claim are not ten pieces of evidence for it. Ask how many independent primary studies exist and how good they are.
- Synthesising, not summarising. Organising the field by question, debate and quality of evidence rather than paper by paper - so the review argues a position about the state of knowledge instead of reciting it.
This is difficult, genre-specific work, and most students have never seen it done well before being asked to do it. A literature review that argues looks quite different from one that lists, and the difference is hard to grasp from a description. As we've argued throughout this series, the fastest route into an unfamiliar genre is to study a well-constructed example first - and a literature review that traces claims to their sources, weighs contradictory evidence and synthesises a field is precisely the kind of example most students have never encountered. UKEssays' literature review service provides model reviews written by academics in the relevant field, showing how primary and secondary sources are distinguished, how competing interpretations are compared, and how a body of evidence is organised into an argument rather than a catalogue; it also offers feedback on your own draft. Used as intended - study how the review handles its sources, then go back up your own chain and build your own - it's a way of seeing what a broken echo looks like on the page before you attempt to break one yourself. The usual rule applies: the model is a scaffold; the review you submit must be yours, because the field you're mapping is one you'll need to actually know when your examiner asks about it. And there's a satisfying irony in the fact that the one thing a model literature review cannot do for you is the one thing that makes a review good - going back and reading the originals yourself.
A source-checking routine that fits real life
You cannot trace every claim to its origin; nobody can. But you can adopt habits that catch the echoes most likely to hurt you:
- Trace anything load-bearing. If a claim carries weight in your argument, read the original. If it's a passing mention, cite honestly as secondary.
- Be suspicious of memorable numbers. Any suspiciously round or precise statistic - 93%, 10,000, 38, "tenfold" - deserves a trip to the source. Spurious precision is the echo's fingerprint.
- Check the verb. Where a source says "suggests," "is associated with" or "in this sample," and your draft says "proves," "causes" or "people," you have shed a modality. Put it back.
- Read the limitations section. It's where the original authors tell you what the echo will later omit. Two minutes, enormous return.
- Ask "how many labs?" When a claim seems universally accepted, count the independent primary studies. If the answer is one, say so.
- Use Wikipedia and textbooks as maps. Start there, follow the references out, cite what you reach. Never cite the map.
- Write "cited in" without embarrassment. It's a mark of honesty, not weakness - and it's the convention that keeps the chain visible for the next reader.
The bottom line
Knowledge doesn't move through the academic world as a set of verified facts; it moves as a set of claims, each carrying a reference that most readers will never follow. That is not a scandal - it's how any large system of knowledge has to work - but it means that the number of times a claim has been repeated is no measure of the evidence beneath it. Thirty-eight witnesses who never existed launched a research tradition. A study of single spoken words became a law of all human communication. An average became a rule. A decimal error that never happened became the standard warning against decimal errors. In each case the chain was long, every link was made in good faith, and nobody went back to the start. The literature review is the part of your education designed to make you the person who does - who finds the original, restores the caveats, hunts for the contradicting six per cent, and writes down what is actually known rather than what is usually said. It is slow, unglamorous, and the single most valuable intellectual habit a degree can give you. The echo stops with whoever decides to check. Let it be you.
