Misinformation Distorts Perceived Social Consensus, and Interventions Help Correct It

Contributions

1. Second-order belief distortion as a misinformation mechanism: Demonstrates that exposure to false headlines increases participants' beliefs about what others believe (second-order beliefs) significantly more than it increases their own direct belief (first-order beliefs). For true headlines, no such asymmetry emerges.

2. Illusory truth effect extends to meta-perceptions: Shows that the illusory truth effect—increased perceived truthfulness from repeated exposure—generalizes beyond personal judgment to judgments about collective belief, with comparable effect magnitudes for first and second-order beliefs.

3. Interventions work on both belief types: Documents that media literacy tips and descriptive norms, both designed to target first-order beliefs, equally or nearly equally reduce second-order belief distortion, revealing a downstream social benefit of these interventions.

4. Mechanism for misinformation's social impact: Identifies second-order belief distortion as a critical mechanism through which misinformation influences collective action, social coordination, and democratic functioning by warping perceptions of what society endorses.

Method

Four experiments across 6,273 U.S. participants recruited via Lucid, CloudResearch Connect, and Prolific with quota sampling for demographic representation:

Experiment 1 (N=646): Participants rated accuracy of 40 headlines (20 true, 20 false) using a 6-point Likert scale, randomly assigned to either report personal belief ("How accurate do you think this headline is?") or projected social belief ("How accurate do you think the average American would find this headline?").

Experiments 2A, 2B, 2C (N=2,023 combined): Tested whether repeated exposure increases second-order beliefs by showing participants statements/headlines in an initial phase, then re-rating them along with novel items. Participants randomly assigned to first-order or second-order belief questions. Experiments 2A used non-political trivia (N=989), 2B used news headlines with CloudResearch (N=985), 2C replicated with Lucid (N=1,049).

Experiment 3 (N=2,604): Tested media literacy tips and descriptive norms interventions. Participants shown 39 political headlines (20 false, 10 misleading, 19 true) with images and source attribution. Randomly assigned to three intervention conditions: control (no intervention), media literacy tips (7-item checklist: be skeptical of headlines, consider source, watch for formatting, check photos, assess motivation, check for humor, examine website domain), or descriptive norms (information emphasizing that Americans value accuracy and evidence-based decision-making). Measured both first and second-order beliefs via 6-point Likert scales.

Results

Experiment 1: For false headlines, second-order beliefs were significantly higher than first-order beliefs (M_second = 0.338 vs. M_first = 0.302, p = 0.009, Cohen's dz = 0.65). For true headlines, no significant difference (M_first = 0.577 vs. M_second = 0.582, p = 0.610, Cohen's dz = 0.12). The asymmetry indicates individuals systematically overestimate others' susceptibility to misinformation.

Experiments 2A–2C: Repeated exposure significantly increased both first and second-order beliefs across all three experiments with comparable effect sizes (Exp 2A: b_1st = 0.048, b_2nd = 0.045; Exp 2B: b_1st = 0.013, b_2nd = 0.013; Exp 2C: b_1st = 0.028, b_2nd = 0.028). No three-way interaction between exposure, belief type, and headline veracity, confirming the illusory truth effect generalizes to second-order beliefs with equal magnitude.

Experiment 3: Media literacy tips significantly increased discernment (improved discrimination between false and true headlines: b = −0.047, p < 0.001 for false vs. true). Descriptive norms also improved discernment but with smaller effect (b = −0.023, p = 0.028). Critically, both interventions were similarly effective for first-order versus second-order belief reduction—no significant three-way interactions, indicating interventions work through shared mechanisms.

Notes

Strengths: The paper elegantly documents a previously underappreciated pathway through which misinformation operates—not solely by changing individual beliefs, but by distorting perceived social consensus. The experimental design is rigorous (four experiments, large N, preregistration evident from methodological clarity), and the finding that common interventions work on second-order beliefs is both surprising and practically important. The asymmetry between false and true headlines (higher second-order than first-order for false, equal for true) is a particularly clean empirical signature of the mechanism.

Limitations and open questions: The paper works entirely in laboratory settings with hypothetical headlines; ecological validity to real-world social-media contexts remains untested. The mechanisms driving the asymmetry are discussed (Bayesian Truth Serum, probability weighting, better-than-average effect) but not definitively resolved empirically. It remains unclear whether correcting second-order beliefs translates to actual changes in social coordination or collective action. The interventions, while effective, do not fully eliminate the distortion, suggesting additional mechanisms or complementary strategies may be needed.

Significance: This work opens a distinct research agenda at the intersection of misinformation, meta-perception, and social dynamics. By showing that misinformation harms collective coordination through perceived-norm distortion, it extends the scope of misinformation research beyond individual epistemics and connects to long-standing literatures on pluralistic ignorance, spiral of silence, and collective action problems in democracy.