Most of what any of us believes about science we have not checked. Nobody has a particle accelerator in the garden. We accept the mass of the Higgs boson, the safety of a vaccine and the effect of a drug on trust — trust in the people who measured, and in the people who told us what they found. That is not a weakness of science. It is how a division of labour works, and the Czech philosopher Jan Frei sees more in it: in the communicating of truth, he argues, “a moment of guaranteeing and of trust” is necessarily present (as the editors of Spor o pravdu, 2016, p. 11, summarise his chapter; our translation). It means the chain of retelling between a measurement and a reader carries real weight, and it is worth asking what happens to a finding as it travels along it.
The short answer, from three cases looked at closely in the last few days, is that it leans. And it leans in one direction.
A lecture about trust
In 2020 the physicist Catherine Beauchemin gave a public lecture on why trust in health research is eroding. She named three impressions that do the damage: experts disagree, experts change their minds, experts have an agenda. She drew a line between scepticism, which science needs, and cynicism, which is what scepticism becomes when trust has collapsed and nothing is checked any more. And in the questions at the end she said something every scientist who has spoken to a journalist will recognise: that shortening a statement does not just make it shorter — it can change what it says, and the person doing the shortening may not notice.
It is a good lecture, and she is right. Which is what makes the next part interesting.
Checking the lecture against its own sources
Three of her examples can be traced to the papers they come from.
- She cites a study of ten years of a leading medical journal: of 363 articles testing an established practice, 146 overturned it. The figure is exact. In the lecture the overturned practices were found to do more harm than good. The study defines a reversal more modestly: a practice found to be no better than a simpler alternative.
- She describes a surgical sponge whose large trial showed more infections, more emergency visits and more hospitalisations. The first two differences were statistically significant. The third was not.
- She cites a check of eighteen published genetic analyses of which sixteen could not be reproduced. The paper counts two reproduced in principle, six partly and ten not at all. Sixteen is defensible if partial means failed — but it is the strictest available reading.
None of this is dishonest, and none of it changes her argument. But notice the direction. Three small drifts, all toward the stronger claim. Not one toward the weaker.
The same thing, elsewhere
A long piece in the New York Times Magazine in May 2026, on the interstitium — the fluid-filled spaces running through our tissues — reported a study in which dye injected at an acupuncture point spread along a meridian in almost all volunteers. The study says fifteen of nineteen injections, not people, and it used two dyes, one of which did not follow the line. The article mentions the one that did.
And the same thing in our own notes. Beauchemin said that 60% of the fifteen scientists with the most retracted papers work in medicine, and none in physics. The first reading here dismissed that as an anecdote: a list of fifteen names says nothing about fields unless you account for how much each field publishes. On checking, the anecdote was right. A survey of 4,449 retracted articles across 42 databases found medicine, the life sciences and chemistry over-represented among retractions relative to their share of the literature, and physics, mathematics and engineering under-represented. The dismissal leaned too — toward the more confident, more sceptical-sounding position. The correction is left visible here on purpose.
And even in a book that is entirely about truth. The Czech volume Spor o pravdu [The Dispute over Truth] ends with an English summary written by the same editors as its Czech introduction. The introduction says of scientific findings that what is mistaken “as a rule falls into oblivion” (p. 12, our translation); in the English summary, theories “disappear”. The shift is tiny, and the same people wrote both. It still goes the same way — from something people do to something that happens to the world.
What a count measures
The list itself, though, deserves a second look. It counts retractions. It does not count harm.
Today one physicist does appear near the top: a plasma physicist in Tehran with over ninety retractions. Many followed the discovery that he and his co-authors had suggested reviewers whose email addresses led back to themselves; others were for overlap with earlier published work. When the first batch came to light in 2017, Retraction Watch noted that the majority of his papers had been cited between zero and ten times. It inflated a publication list. Very little was built on it.
At the top of the same list is an anaesthesiologist whose fabricated trials of a resuscitation fluid, hydroxyethyl starch, entered the meta-analyses that guided intensive care. When a 2013 analysis removed his seven trials, the fluid was associated with higher mortality and more kidney failure. His numbers had helped keep a harmful treatment in use.
On the leaderboard these are the same kind of entry, differing only in the count. A number can be exact and still measure the wrong thing — and the more exact it is, the easier it is to forget to ask what it is a number of.
Why the lean has a direction
It would be easy to call this carelessness, but carelessness is random, and this is not. A summary has to choose what to drop, and the qualifications are what go first: the second dye, the non-significant third outcome, the definition that makes a word modest. What survives the cut is the version with the cleanest edge. It is also, reliably, the stronger version. A retelling rarely makes a finding weaker than it was; it has no reason to add the doubt back in.
Nor is this a fault of journalists or popularisers. The child psychotherapist Ingrid Strobachová describes the same pressure in a consulting room, where nothing is published: “A strong pressure towards simplification arises here, and it is hard to resist. So we convert things into symbols, principles, explanations” (Strobachová 2016, p. 225, our translation). It acts wherever something complex has to be passed on.
That is where cynicism finds its opening. A reader who once discovers that a confident summary overstated its source is tempted to conclude that everything is overstated, and to stop reading sources at all. That is the lazy form of scepticism: it has the posture of checking without the work. Beauchemin put the right version plainly at the end of her lecture: be sceptical of health research, and more sceptical still of the random website, and keep trusting the results that have earned it many times over — the old vaccines, antibiotics, a cast on a broken bone.
The practical rule is small. When a finding reaches you in one clean sentence, the interesting question is not whether it is true. It is what was taken out to make it that clean. Look for the second dye.
—
Sources
- Beauchemin, C. A. A. “A physicist’s adventures in virology.” Perimeter Institute Public Lecture, 4 November 2020. Video. Paraphrased from the recording.
- Prasad, V., et al. “A decade of reversal: an analysis of 146 contradicted medical practices.” Mayo Clinic Proceedings 88(8), 2013, 790–798.
- Bennett-Guerrero, E., et al. “Gentamicin-collagen sponge for infection prophylaxis in colorectal surgery.” New England Journal of Medicine 363(11), 2010, 1038–1049.
- Ioannidis, J. P. A., et al. “Repeatability of published microarray gene expression analyses.” Nature Genetics 41(2), 2009, 149–155.
- Li, T., et al., incl. A. Ahn. “In Vivo Visualization of the Pericardium Meridian with Fluorescent Dyes.” Evidence-Based Complementary and Alternative Medicine, 2021, 5581227. On the magazine article: Cooper, A. Z. “Inside the Interstitium.” The New York Times Magazine, 11 May 2026.
- Grieneisen, M. L., Zhang, M. “A comprehensive survey of retracted articles from the scholarly literature.” PLoS One 7(10), 2012, e44118.
- The Retraction Watch Leaderboard (state of 28 August 2026), and Retraction Watch’s reporting on the Tehran case (2017, 2024).
- Zarychanski, R., et al. “Association of hydroxyethyl starch administration with mortality and acute kidney injury in critically ill patients requiring volume resuscitation.” JAMA 309(7), 2013, 678–688.
- Trlifajová, K., Frei, J., et al. Spor o pravdu [The Dispute over Truth]. Filosofia, Prague 2016 — Introduction (pp. 9–13), English Summary (pp. 243–244), and Strobachová, I., “Co je pravdivost v psychoterapii?” [What is truthfulness in psychotherapy?] (pp. 223–233). Czech quotations in our translation.
