Contribution
This is a blog essay (“A rant”, by its own title) by the statistician Andrew Gelman. It extends his June 11, 2025 post on scientists writing things they know are false without regarding them as lies. It adds two domains, paid expert testimony and law-enforcement public statements, adopts a commenter’s name for the shared pattern, and gives an incentive account of why it persists. The novelty is the comparison across domains and the framing of the question around incentives. No new evidence is offered.
Key insight
According to Gelman, falsehood that is visibly false (“long term” for a 3-day study; “assault” contradicted by video) persists because it functions as a term of art inside a community that accepts it and imposes no cost. Because honest actors compete against it, a Gresham’s-law dynamic favours those willing to “make shit up”. The remedy he proposes is not persuading the liars but withdrawing the tolerance of bystanders.
Assumptions
Premises and authorities:
- The examples are false as characterized. The video-game paper’s “long-term” framing for a 3-day study, and the power-pose abstract’s claim with no measure of power, are taken from his earlier post. The Ludmerer testimony comes from a screenshot, with a reported fee of over half a million dollars (per Wikipedia, linked). The Lander and Padilla characterizations are taken from Ezra Klein (NYT), Philip Bump (Washington Post) and Sam Adams (Slate), which cite video.
- Kinsley’s rule (“The scandal isn’t what’s illegal, the scandal is what’s legal”) frames the concern as practices that are permitted, not transgressions.
- The term “reckless disregard for the truth” comes from commenter Raphael. It was preferred to “bullshit” because the misstatements seemed “pretty deliberate”. “Reckless” is used because readers “might actually read the title or abstract … and take it seriously”. Gelman does not invoke the legal (actual-malice) sense of the phrase.
- The incentives as stated in §4d: (a) it helps your case, (b) captive media pass it along, (c) it is common enough not to feel bad, (d) no professional consequences.
Key results
What the post argues:
- §1 Scientists. The two examples show authors whose paper “wrapping” misdescribes the content, and Gelman’s concern is that “the authors involved never even saw the problem”.
- §2 Doctors. Ludmerer (2002), a Washington University professor of medicine who consulted for the tobacco industry, testified to having “no opinion” on whether smoking contributes to lung cancer. Gelman says Ludmerer “felt he was playing by the rules”.
- §3 Cops. DHS called Lander’s conduct “assaulting law enforcement” and Padilla’s “lung[ing]”, and these accounts are disputed by video according to the quoted journalists. Bump could not get ICE to source the “413 percent increase in assaults” claim. He notes the figure was 300 percent in mid-April. He found CBP year-to-date assaults 20 percent lower than in 2024, and news releases documenting assaults on 12 individuals against 66,463 ICE arrests (all figures Bump’s, quoted).
- §4a Motive. One lies when one’s arguments are not good enough, because adding “fake evidence … makes your case stronger”.
- §4b Mechanism. A propagation channel is needed: journals (“include some statistically-significant p-values, then you can slip just about anything into the title and abstract”), courts, and social, partisan and nonpartisan media repeating official statements.
- §4c Acceptance. The phrases “function as terms of art”, and critics are “outsiders” who “don’t really count”. The public visibility of the falsehood shows it is accepted behaviour.
- §4d Consequences and conclusion. There are none beyond “the occasional angry columnist or blogger”. Lying is negative-sum. Bystanders who say “Not my problem”, “the general point is correct” or “the other side does it too” tip the incentives. His closing line: “I can ask the rest of you to stop tolerating it.”
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | A paper described as “long-term” studied effects over 3 days; a power-pose abstract claimed people “instantly become more powerful” without measuring power | moderate | author’s reading of the papers, via his earlier post; the papers are not cited by name in this post |
| C2 | Ludmerer testified in 2002 to “no opinion” on smoking and lung cancer and was paid over $500k | moderate / weak | screenshot of testimony (moderate); fee “reportedly”, via Wikipedia (weak) |
| C3 | DHS falsely described Lander as assaulting officers and Padilla as lunging | weak (in this post) | quoted press accounts that cite video; the video is not examined by the author |
| C4 | ICE’s “413 percent” assault figure is unsourced and inconsistent with other data | moderate (as Bump’s finding) | Bump’s documented inquiry and CBP data, quoted |
| C5 | These falsehoods persist because they are accepted as terms of art and carry no consequences | weak | informal argument from three anecdotes; no measurement of prevalence or sanctions |
| C6 | People making such statements do not recognize them as lies | assertion | author’s belief (§4c), in tension with §4a and the Interlude |
| C7 | Journals will pass nearly anything in a title or abstract given standard style and significant p-values | assertion | author’s experience; no data in the post |
| C8 | Lying is negative-sum, and bystanders’ tolerance sustains it | informal argument | Gresham’s-law analogy |
Concepts
- Reckless disregard for the truth — as Gelman uses it: stating what one has no basis to believe, or knows to be false, in a setting where others may rely on it, and doing so as accepted practice. This is not the defamation-law standard, which the post does not discuss.
- Terms of art — Gelman’s name for false phrases that insiders understand not to mean what they say (“long term”, “assaulting”, “lunging”).
- Kinsley’s rule — “the scandal is what’s legal”, applied to science.
- Gresham’s Law (of discourse) — liars drive out honest argument in the short term.
Connections
The post continues Gelman’s June 11, 2025 statmodeling post (“Scientific poetic license”) and his long-running critique of social-psychology practice (power pose). The terminology debate draws on Frankfurt’s “bullshit”, which the post names only as a commenter’s suggestion and rejects. Within this record, its §3 subject matter, DHS and ICE public statements during enforcement operations, overlaps the context of the Minnesota show-cause transcript (LIT-012), where the court questions the government’s unexplained “safety concern”. The two documents do not reference each other.
Bearing on the record
No THEORY document in this record is affected. For the Anthology of the SOTA the relevance is procedural, not a practice. §1 and §4b describe titles and abstracts that overstate what a paper shows, and review processes that let them through. That is the failure the anthology’s reading scheme is designed to guard against: a Skimmed reading may not source a practice, and a Read note must rate claims by their support (ANTH-ADR-025). The post gives anecdotal support for that design and no new instruction for ML practice. It does not warrant an ANTH-SOTA document.
Limitations
- The evidence is three to five anecdotes, chosen to illustrate rather than sampled. Nothing establishes how common the pattern is in science, testimony or policing.
- The police examples depend on quoted journalism, and the underlying video and DHS statements were not examined by the author or in this reading.
- The treatment of intent is internally inconsistent (deliberate versus not recognized as lies).
- The mechanism is conceded not to have worked in the testimony case.
- The source papers in §1 are not cited in this post.
- Gelman says it is a “rant”. The Substack text is assumed identical to the blog original (unverified).
Open questions
- How often do titles and abstracts overstate a paper’s design or measures, and does review catch it? An audit comparing claim language in abstracts against the methods sections, across a field sample, would settle it.
- Do any sanctions (journal corrections, bar or medical-board discipline, agency retractions) follow such statements at measurable rates?
Corrections to the seeded skim
- The skim describes §4 as “4a–c” (motive, mechanism, standard operating procedure). The post has a fourth part, §4d “Lack of consequences”, and a closing argument the skim omits. Gelman raises an “internal contradiction”: if everyone knows the claims are false, why does it matter? He answers that “not everybody lies”. By “a form of Gresham’s Law, liars win” in the short term, lying is “a negative-sum action” that adds noise to the channel, and tolerant bystanders “are part of the problem”. The post ends with a call to readers to “stop tolerating it”. This is the post’s conclusion, and the skim misses it.
- The skim says the post explains the term was chosen because the misstatements “seem deliberate”. That is right for the Interlude, but the post is not consistent on intent. §4c says “I don’t even think the people saying these false statements recognize them as lies”, while §4a says of Ludmerer “I can only assume he didn’t believe it—i.e., he was lying”. The skim presents one intent claim where the post makes two in tension.
- In §4b Gelman concedes that the medical-testimony case probably did not work (“I can’t believe his testimony was very effective to anyone”). So the propagation mechanism is not claimed for all three cases. The skim presents the motive–channel–acceptance account as uniform.
- Venue: the Substack publication is titled “The Future of Statistical Modeling”, not simply “Gelman’s Substack mirror”. The header date “2025-06-01” is a month placeholder; the post is dated 2025-06-24.