Fighting COVID-19 Misinformation on Social Media: Experimental Evidence for a Scalable Accuracy-Nudge Intervention

Fighting COVID-19 Misinformation on Social Media: Experimental Evidence for a Scalable Accuracy-Nudge Intervention Gordon Pennycook Jonathon McPhetres Yunhao Zhang Jackson G. Lu David G. Rand

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Fighting COVID-19 Misinformation on Social Media: Experimental Evidence for a Scalable Accuracy-Nudge Intervention 2020 Psychological Science 10.1177/0956797620939054 pennycook2020-fighting-covid-19-misinformation-o pennycook2020-fighting-covid-19-misinformation-o.txt, the project's staged copy 554be8c08bacd7da264c7c0d3d48f417d49206b9 cc-by social-psychology

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them why people share false claims they could have identified as false, and whether the gap can be closed by making accuracy salient Quotation withheld because this record is not cleared for redistribution. whether the inattention account of political fake-news sharing extends to COVID-19 misinformation, where the content is largely apolitical and the stakes are life and death Quotation withheld because this record is not cleared for redistribution. a measured dissociation between accuracy judgment and sharing intention on the same headlines, and a randomised test showing that one accuracy judgement about an unrelated headline raises sharing discernment 2.8-fold Quotation withheld because this record is not cleared for redistribution. mechanism Quotation withheld because this record is not cleared for redistribution. truth discernment that is present when people are asked about accuracy and nearly absent when the same people are asked what they would share Quotation withheld because this record is not cleared for redistribution. behavioural-experiment behavioural-experiment Quotation withheld because this record is not cleared for redistribution. the account is not that people want to spread falsehood, nor that they cannot tell the difference. It is that the sharing context puts other things in front of accuracy - social approval, partisan alignment, the fact that news sits between baby photos - so the preference exists and is simply not applied at the moment of decision a preference that is never consulted Quotation withheld because this record is not cleared for redistribution. the treatment is one accuracy judgement about a politically neutral headline unrelated to the topic, framed as a pretest. It supplies no information about COVID-19 and no instruction about sharing; what it does is raise the accessibility of the concept of accuracy, and the effect is a change in what the participant attends to rather than in what they know a prime that carries no content Quotation withheld because this record is not cleared for redistribution. if the nudge works by directing attention to accuracy, then its effect on a headline should track how accurate that headline is perceived to be - pushing sharing up for the clearly true and down for the clearly false. Measured across the thirty items using Study 1's ratings as the accuracy scale, r(28) = .76 the item-level signature of an attention effect Quotation withheld because this record is not cleared for redistribution. the evidence is restricted to the United States, and the sample is quota matched rather than probability sampled so it is not nationally representative Quotation withheld because this record is not cleared for redistribution. one particular set of thirty headlines was used, and misinformation in other forms - e-mails, text posts, memes - is untested Quotation withheld because this record is not cleared for redistribution. sharing intentions were hypothetical and the induction was performed in a lab context, so extension to naturalistic social media is an inference Quotation withheld because this record is not cleared for redistribution. accuracy nudges are described by the authors as far from a complete solution Quotation withheld because this record is not cleared for redistribution. one false headline turned out on inspection to be misleading or unverified rather than untrue, and was retained after checking that removing it changed nothing Quotation withheld because this record is not cleared for redistribution. the preregistration specified clustering standard errors on participant only, and the published analysis clusters on participant and headline Quotation withheld because this record is not cleared for redistribution. priming - raising the accessibility of a concept with an unrelated task - carried from cognitive psychology into a platform-deployable intervention against misinformation Quotation withheld because this record is not cleared for redistribution. the nudge frame from behavioural economics: change what is salient at the moment of choice rather than what the chooser knows or is allowed to do Quotation withheld because this record is not cleared for redistribution. the accuracy induction disguised as a pretest, using a single politically neutral headline unrelated to the topic of the sharing task eliminates: demand characteristics and social desirability as the explanation of the treatment effect - a participant who cannot see what the manipulation is for cannot comply with it · outcome: held Quotation withheld because this record is not cleared for redistribution. apolitical COVID-19 content used in place of the political headlines the prior literature used, with partisanship measured and tested as a moderator eliminates: partisan motivation as the thing distracting people from accuracy - the effect survives when the content gives partisanship nothing to work with · outcome: held Quotation withheld because this record is not cleared for redistribution. a post hoc item-level analysis correlating each headline's treatment effect against its perceived accuracy from the other study's participants eliminates: a general suppression of sharing as the explanation - an attention effect must push true headlines UP and false ones down, and the correlation is r(28) = .76 · outcome: held Quotation withheld because this record is not cleared for redistribution. all participants retained in the main analyses regardless of failing the three attention screeners, per the preregistered plan, with the key result reported across levels of attentiveness eliminates: the effect being an artefact of who was excluded - dropping inattentive respondents would trade representativeness for a cleaner sample and the authors show the result does not depend on the trade · outcome: held Quotation withheld because this record is not cleared for redistribution. one false headline reclassified after data collection as misleading rather than untrue, and the analysis rerun without it eliminates: a defective stimulus carrying the result - the item is disclosed, the removal is tested, and it is retained on the strength of the test rather than quietly · outcome: held Quotation withheld because this record is not cleared for redistribution. linear probability models used for interpretability, with logistic regression run alongside eliminates: the functional form as the source of the result - the readable model and the conventional one are reported to agree · outcome: held Quotation withheld because this record is not cleared for redistribution. the self-reported sharing measure checked against a headline-level relationship between reported willingness and actual shares on Twitter eliminates: a hypothetical intention being read as behaviour with no evidence at all - the link is external and at the level of the item · outcome: partial Quotation withheld because this record is not cleared for redistribution. results also reported excluding the few headlines that were political or that did not contain a clear claim of fact eliminates: a handful of atypical items driving the interaction · outcome: held Quotation withheld because this record is not cleared for redistribution. a preregistration deviation declared in the paper - the plan specified clustering on participant only - with the preregistered version run eliminates: an undisclosed analytic choice; the deviation is named and the preregistered specification is reported not to change the conclusion · outcome: held Quotation withheld because this record is not cleared for redistribution. three screener questions embedding a subtle instruction inside a block of text, placed before each of the main tasks eliminates: participants who click through without reading - on a task about whether people can tell true headlines from false ones · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. a leave-one-out check on the stimulus set: one false headline judged after the fact to be misleading rather than untrue was removed and the analysis re-run eliminates: that the result rests on one badly-classified item - the paper re-ran without it, found no change, and retained it while saying so · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. truth discernment truth discernment Quotation withheld because this record is not cleared for redistribution. inattention inattention Quotation withheld because this record is not cleared for redistribution. accuracy nudge accuracy nudge Quotation withheld because this record is not cleared for redistribution. CRT CRT Quotation withheld because this record is not cleared for redistribution. demand characteristics demand characteristics Quotation withheld because this record is not cleared for redistribution. sharing intentions sharing intentions Quotation withheld because this record is not cleared for redistribution. partisanship partisanship Quotation withheld because this record is not cleared for redistribution. quota sampling quota sampling Quotation withheld because this record is not cleared for redistribution. truth discernment - WHEN THE QUESTION IS ACCURACY: veracity moves the answer at Cohen's d = 0.657 Quotation withheld because this record is not cleared for redistribution. truth discernment - WHEN THE QUESTION IS SHARING: d = 0.121 from the same people on the same headlines, and 32.4% more would share a false headline than call it accurate Quotation withheld because this record is not cleared for redistribution. sharing discernment - IN THE CONTROL CONDITION OF STUDY 2: not significantly different from zero, d = 0.050, p = .24 Quotation withheld because this record is not cleared for redistribution. sharing discernment - AFTER ONE ACCURACY JUDGEMENT ABOUT AN UNRELATED HEADLINE: d = 0.142, p = .003, and 2.8 times the control level Quotation withheld because this record is not cleared for redistribution. the sample - AS DESIGNED: quota matched to the US public on age, gender, ethnicity and region Quotation withheld because this record is not cleared for redistribution. the sample - AS THE AUTHORS QUALIFY IT: not obtained by probability sampling and therefore not to be treated as nationally representative Quotation withheld because this record is not cleared for redistribution. extends → the inattention-based account of political fake-news sharing, carried here to COVID-19 misinformation where the content is largely apolitical and the stakes are health Quotation withheld because this record is not cleared for redistribution. critiques → the view that partisan motivation is what distracts people from accuracy on social media - the effect appears on apolitical content and is not moderated by partisanship Quotation withheld because this record is not cleared for redistribution. qualifies → its own intervention - accuracy nudges are far from a complete solution, and the sharing measure is hypothetical Quotation withheld because this record is not cleared for redistribution. confirms → the analytic-cognitive-style literature on truth discernment - CRT scores relate to discernment here in the same pattern previously seen with political headlines Quotation withheld because this record is not cleared for redistribution. partisanship —distinct-from→ inattention partisanship inattention differs in: whether the thing pulling attention from accuracy is political identity or the sharing context itself · the relation itself is separately evidenced · strength of the relation: 0.95 · differs in (stated by the tagger, no quotation): whether the thing pulling attention from accuracy is political identity or the sharing context itself · the relation stated: the effect appears on apolitical COVID-19 content, so partisanship is not the key factor distracting people from accuracy Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span says partisanship is not the key factor; distinct-from is loose. Quotation withheld because this record is not cleared for redistribution. veracity moves accuracy judgments at d = 0.657 and sharing intentions at d = 0.121, from the same people on the same headlines veracity moves accuracy judgments at d = 0.657 and sharing intentions at d = 0.121, from the same people on the same headlines Quotation withheld because this record is not cleared for redistribution. 32.4% more people were willing to share a false headline than rated that headline as accurate 32.4% more people were willing to share a false headline than rated that headline as accurate Quotation withheld because this record is not cleared for redistribution. Rating the accuracy of a single non-COVID headline raised sharing discernment 2.8-fold. Rating the accuracy of a single non-COVID headline raised sharing discernment 2.8-fold. Quotation withheld because this record is not cleared for redistribution. in the control condition sharing discernment was not significantly different from zero, p = .24 in the control condition sharing discernment was not significantly different from zero, p = .24 Quotation withheld because this record is not cleared for redistribution. the treatment effect was not moderated by cognitive reflection, science knowledge, partisanship, distance to an epicentre or medical maximising the treatment effect was not moderated by cognitive reflection, science knowledge, partisanship, distance to an epicentre or medical maximising Quotation withheld because this record is not cleared for redistribution. the treatment's effect per headline correlates at r(28) = .76 with that headline's perceived accuracy, which is the signature an attention account predicts the treatment's effect per headline correlates at r(28) = .76 with that headline's perceived accuracy, which is the signature an attention account predicts Quotation withheld because this record is not cleared for redistribution. the intervention requires no decision by a platform about what content to censor the intervention requires no decision by a platform about what content to censor Quotation withheld because this record is not cleared for redistribution. the difference in accuracy judgments or sharing intentions between true and false headlines - a within-person contrast, so a higher score means greater sensitivity to truth relative to falsity truth discernment Quotation withheld because this record is not cleared for redistribution. more people were willing to share a false headline than rated it accurate - the size of the gap between the two questions 32.4% Quotation withheld because this record is not cleared for redistribution. Cohen's d for the effect of veracity on accuracy judgments, against 0.121 for sharing intentions 0.657 Quotation withheld because this record is not cleared for redistribution. times higher sharing discernment in the treatment condition than the control, from a single accuracy judgement about an unrelated headline 2.8 Quotation withheld because this record is not cleared for redistribution. the item-level correlation between a headline's perceived accuracy and the size of the treatment's effect on sharing it .76 Quotation withheld because this record is not cleared for redistribution. Cohen Cohen Quotation withheld because this record is not cleared for redistribution. truth discernment truth discernment Quotation withheld because this record is not cleared for redistribution. sharing intentions sharing intentions Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1177/0956797620939054.