David RandContributions
- Successful replication of the negative effect of repetition on moral condemnation using expanded headline set with larger sample, confirming robustness across headlines and participants
- Test of moderating mechanisms: Two conditions (Full: includes accuracy prompts; Ecological: omits accuracy to mirror real social media) reveal the moral repetition effect is robust regardless of whether accuracy is primed, though baseline moral condemnation is higher when accuracy is discussed
- Generalizability evidence: Effect replicates with 100 balanced-partisan false headlines (vs. original 12), addressing "generalizability crisis" in psychology and showing effect is not idiosyncratic to particular stimulus set
- Pre-registered design with transparent reporting of inconclusive hypothesis (H2) where expected moderation did not emerge, modeling open science standards
Method
Two-stage study with familiarization and judgment phases. Participants (N=1,600: 813 Democrats, 787 Republicans, Mage=42.8) recruited on Prolific. Stage 1: participants rated 6 randomly selected false headlines on engagement, humor, writing quality (familiarization phase). Stage 2 (after distractor demographics/politics survey): rated 12 headlines—the 6 from Stage 1 (repeated) and 6 novel—on moral condemnation ("How unethical would it be to share?"), perceived accuracy, liking, popularity, and sharing intentions.
Two conditions:
- Full Condition (N=750): All outcome measures including accuracy judgment (replicates Effron & Raj closely)
- Ecological Condition (N=750): All measures except accuracy, modeling real social media where accuracy is not explicitly prompted
Headline set: 100 false headlines pre-tested for partisan balance, with 6 originally used by Effron & Raj (updated where outdated) plus 94 new. Headlines randomly sampled for each participant to test generalizability beyond the original 12.
Results
Hypothesis 1 (Replication): Confirmed. Repetition decreased moral condemnation by 4.66pp in Full Condition (β₁ = −4.66, CI: [−5.70, −3.63], t(99) = −8.57, p < 0.001, multi-level model; nearly identical to original β = −4.64). Fixed-effects regression model confirms identical effect (β₁ = −4.66).
Hypothesis 2 (Moderation by accuracy prompts): Inconclusive. No significant difference in repetition effect between conditions (β₂ = −0.91, CI: [−2.43, 0.61], p = 0.24). Although difference was not significant, point estimate in Ecological condition was actually larger (β₁ = −5.62) than Full Condition, opposite to hypothesis that accuracy prompts would strengthen the effect.
Hypothesis 3 (Effect persists without accuracy): Confirmed. Repetition decreased moral condemnation in Ecological Condition (β₁ = −5.62, CI: [−6.73, −4.52], p < 0.001), comparable in magnitude to Full Condition and original study.
Exploratory findings:
- Baseline moral condemnation lower without accuracy: Moral condemnation of novel headlines was 3.71pp lower in Ecological vs. Full Condition (p = 0.008), suggesting accuracy prompts increase baseline ethical concern
- Illusory truth effect confirmed: Repeated items believed 3.40pp more true than novel items (p < 0.001), approximately 1pp smaller than moral condemnation effect
- Liking increased by repetition: Repeated items liked 2.90pp more than novel items (p < 0.001)
- Perceived popularity increased: Repeated items estimated to have been seen by more people (β = 3.14pp, p < 0.001)
- No effect on sharing intentions: Repetition did not significantly predict sharing intentions (β = 0.02, p = 0.35)
- No moderation by political concordance: Effect of repetition did not differ for headlines politically aligned vs. misaligned with participant's partisanship (exploratory, underpowered analysis)
Notes
Strengths: Pre-registered design published as registered report (Stage 1 accepted before data collection), large sample (N=1,600, 100 participants larger than pre-registered n=1,500), realistic headline set expanded and updated with approval of original authors, two analytical models (multi-level and fixed-effects with clustered SE) addressing methodological concerns about random-intercepts-only models, clear presentation of inconclusive results (H2) rather than p-hacking, high statistical power (99% to detect original effect size, 95% to detect 75% reduction in effect).
Limitations and questions: The lack of moderation by accuracy prompts is surprising theoretically. Authors hypothesized that highlighting accuracy would increase the transgression of sharing falsehoods and thus buffer against moral desensitization; instead, effect is robust regardless. Exploratory evidence suggests baseline moral condemnation (of novel headlines) is higher when accuracy is discussed, possibly because awareness of falsity primes ethical concern—but the reduction with repetition is unaffected. This suggests the repetition mechanism (possibly normalization/fluency effects on perceived commonality, see Reber & Unkelbach on processing fluency and normativity) operates downstream of accuracy considerations. However, on real social media, users often lack explicit falsity awareness, so the Ecological Condition may be more ecologically valid than the Full Condition. The null effect on sharing intentions despite strong effect on moral judgment deserves follow-up: do people share things they judge unethical? The authors note this disconnect but do not resolve it. Political concordance analysis underpowered (52% power for half-moderation); future work should test whether partisan-aligned false news elicits less moral condemnation when repeated.