When re-checking a source changes an entry, the change is recorded here.
Changed "hundreds of versions" to more than two hundred experiments, and attributed liking without memory to the later studies that found it.
The original overstated the count and credited Zajonc's first studies with a later finding.
"Very few finished on time" became fewer than a third finishing by the date they had predicted.
That is the study's figure.
Added that the two best-known accounts of the 1996 disaster contradict each other on several points.
Krakauer's and Boukreev's books disagree, so the detailed sequence is contested.
Described the 1977 follow-up as Nisbett and Wilson ran it: students who saw a lecturer behave warmly rated his appearance and accent more favourably, and denied the warmth had influenced them.
The entry described a different study from the one that was run.
Corrected Smith's lone worker: he could perhaps not make one pin a day, and certainly not twenty.
The entry misquoted Smith.
Attributed the sabotage and the lost decade at Microsoft to Kurt Eichenwald's reporting.
Both were stated as fact rather than as one journalist's account.
Removed the claim that a combined Squares and Diamonds box outsold the original.
No source supports it.
Changed "handed player selection to measurement" to leaning on measurement, and replaced "against teams spending three times as much" with the American League record of twenty straight wins on about a third of the Yankees' payroll.
The original overstated both points.
Changed "a fifty-year-old word" to a medical word dating from 1874.
The word was 47 years old when the campaign began.
Limited the law to stable systems, where work leaves at roughly the rate it arrives.
It does not hold for every queue.
Corrected the Berkeley case: department by department, the gap vanished, and the small difference left slightly favoured women.
The entry said the gap reversed, which overstates it.
Replaced "the original audio tapes were published" with Thibault Le Texier's examination of the study's archives, recordings included, in 2018 and 2019.
The tapes were not published; they were examined in the archives.
Gave the exact figure, 31 of the 87 chains that formed went the wrong way, and added that in longer sequences many cascades eventually break.
The figure comes from the 1997 paper, and the later research changes how far the finding reaches.
Attributed the figure of over 90 per cent of US employers using hiring algorithms to the study's authors as their estimate.
It was stated as fact rather than as the authors' estimate.
Replaced "four teams building a compiler ship a compiler with four parts" with Conway's own 1968 example, five people on a COBOL compiler and three on an ALGOL one, producing five and three phases.
The four-teams line is a later paraphrase, not Conway's own example.
Removed the claims that a crowd that discusses becomes worse than its smartest member and that crowds beat experts.
Both were stronger than the evidence, and a 2017 study found discussion among equals improved accuracy.
Described the twenty-point gap as a difference in median answers rather than averages.
The study reported medians.
Rebuilt the entry around the relay assembly test room: the change in the women's group incentive, the two workers replaced in the second year, the Depression, the 1978 statistical re-analysis, and the participants' own accounts in 1983. Added that the phrase was coined in 1953.
The earlier entry told only the lighting story, while the records show several things changing at once.
Corrected Galton's sample to 928 adult children and their parents.
The figure of 1,078 belongs to Pearson and Lee's 1903 study of fathers and sons.
Corrected the 2015 project's sample to 100 studies from three journals published in 2008, about a third of which replicated, and moved power posing and similar failures to the separate efforts that tested them.
The entry misdescribed what the 2015 project covered.
Changed "only the founder" bears the cost to the company and the team carrying it.
The original overstated who pays.
Softened the claim: anxiety lowers interview scores somewhat, and has little to do with how people later perform in the job.
The studies found a modest effect, not the strong one the entry described.
Replaced "largely explained by family" with what the 2018 replication found: a link about half the size of the original, which fell by about two thirds once family background, early ability and home environment were taken into account.
The entry overstated the replication's finding.
"Tripled" became "more than doubled".
In the mug study, owners asked about 7 dollars and buyers offered about 3.
Added that full-price ticket buyers went to more plays only in the first half of the season.
That is what the theatre study reported.
Restated the two conditions for trustworthy intuition as Kahneman and Klein gave them: an environment regular enough to be predictable, and long practice in it with quick, clear feedback.
The entry gave a different pair of conditions from the 2009 paper.
Changed the date of the campaign from around 1920 to early in the twentieth century.
Prohibition began in 1920, so the original date was impossible.
"Matched" became "closely predicted".
The study measured how closely brief ratings predicted later ones, not an exact match.
"Every student" became the 39 students in the study, and "most personality tests" became "many".
Both were stated more broadly than the evidence allows.
Removed "anyone could take it" and "one in a thousand", and dated anonymous marking from the Song dynasty onwards.
The exams excluded women and some groups, the pass-rate figure had no source, and anonymous marking was not used throughout.
Removed the claim that Langley had Harvard credentials.
Langley had no degree; his standing came from his scientific reputation.
Removed the "three to five times" ratio, and attributed the five-to-one figure to John Gottman's research on couples in conflict.
The ratio had no source in the research cited.
Removed the claim that evaluators rate charismatic people as more competent even where charisma is irrelevant, and described the training effect as medium in size.
The removed claim had no source, and the study reported an effect of medium size.
Described Heidi versus Howard as a classroom exercise rather than a published study.
It was never published as a study.
Added Howard Wainer's Pennsylvania schools data and the later lottery-based research on New York City's small high schools.
The earlier version stated more than Wainer's data showed, and the lottery studies found that New York's small schools did raise graduation rates.
Added that a blunder raised liking only for someone already seen as highly competent.
The original study found the opposite effect for an average performer.
Rewrote the entry. It said founders who had failed before did as well as founders who had succeeded. It now says previously successful founders had the highest success rate, about 30 per cent, while previously failed founders, at about 22 per cent, did little better than first-time founders, at about 18 per cent.
Re-reading the study showed the entry had its finding the wrong way round.
Added that the effects are typically small and often fade.
A 2005 review by Jussim and Harber found exactly that.
Updated the 2022 revision: it lowered most of the estimates, and the structured interview came out as the single best predictor.
The entry described the revision inaccurately.
The size of the effect changed from roughly three times to 63 per cent more.
The 63 per cent comes from the storage box study by Norton, Mochon and Ariely; the larger figure belonged to a different experiment.
The entry no longer says a complete warship was built from scratch during a single banquet. It now says a galley was assembled from ready-made parts, and that the accounts disagree on the details.
The historical accounts do not support the stronger version.
Corrected how the study was done: interviews with 120 people who hire at elite firms, and nine months observing one firm's recruiting. Removed the examples of sailing, skiing and wine.
The entry misdescribed the method, and the examples were not in the study.
Limited the popular version, that we blame other people's character and excuse our own behaviour by the situation, to negative behaviour.
A 2006 meta-analysis of 173 studies found the split holds mainly when the behaviour is negative.
Narrowed the finding to people who found their job through a personal contact: most of them heard about it from someone they saw only occasionally.
The original claimed it for everyone in the sample.