Hostility has a trivial effect on persuasiveness of rebutting science denialism on social media

The paper examines whether hostility in social media conversations affects the persuasive impact of science denialism and rebuttals by advocates for science. Science deniers often use strategies like fake experts, logical fallacies, impossible expectations, cherry-picking, and conspiracies, which can reduce public support for behaviors such as vaccination and adoption of GM foods. Given that online debates are often hostile, the authors hypothesized that hostility would undermine persuasiveness by lowering perceived competence of speakers (per persuasion theories like the elaboration likelihood and heuristic–systematic models). They also considered expectancy violations theory: hostile rebuttals might be less persuasive if they negatively violate audience expectations, whereas neutral responses to hostile misinformation might positively violate expectations. The authors further predicted that rebuttals, even if hostile, would still outperform no response. Potential audience moderators were considered: verbal aggressiveness, need for cognition, issue involvement, and social media use frequency, as these could influence reliance on tone versus argument strength.

Prior work shows incivility is perceived negatively in workplaces, politics, and science communication. Hostility may conflict with perceived scientific competence, which can serve as a cue in persuasion models. Expectancy violations research suggests effects depend on whether messages align with audience expectations and situational cues (e.g., prior hostility). Despite potential downsides, rebuttals of misinformation are generally effective and format often matters less than correcting misinformation. Audience similarity and perceptions of authenticity can buffer negative reactions to insensitive or aggressive language; highly involved or high need-for-cognition audiences may focus more on arguments than peripheral cues like tone. Social media environments are characterized by frequent hostility, potentially shaping norms and reactions. Previous studies on civility, corrections, and aggressive styles in science debates provide mixed evidence regarding tone effects, motivating systematic experimental tests across topics (vaccination, GM foods).

Four preregistered online experiments with U.S. adults recruited via Prolific tested the impact of hostility in social media debates about vaccination (Experiments 1–3) and GM food (Experiment 4). Participants read a fictitious back-and-forth social media discussion (#YourVoice) between a science denier spreading misinformation and an advocate rebutting with refutational messaging and scientific facts. Hostility was manipulated via inclusion of insults, swear words, and capitalization; neutral versions omitted these features while keeping arguments constant. Designs: Experiment 1 used a 2 (denier: hostile vs neutral; between) × 2 (advocate: hostile vs neutral; between) × 2 (time: pre vs post; within) mixed design. Experiment 2 used 2 (denier: hostile vs neutral; between) × 2 (advocate: hostile vs absent; between) × 2 (time; within). Experiment 3 combined all (2 × 3 × 2) with advocate levels hostile vs neutral vs absent. Experiment 4 replicated Experiment 3 with topic changed to GM food. Total N = 3,226 (Exp1 n=521; Exp2 n=310; Exp3 n=1,200; Exp4 n=1,195). Primary outcomes were attitudes toward the behavior (vaccination or GM foods) and intentions to perform it (vaccinate or buy GM foods), measured pre and post; outcomes were transformed to POMP (0–100). Manipulation checks included perceived politeness; mediators included perceived competence (all but Exp1, where expectancy measures were also collected); moderators measured included verbal aggressiveness (all), need for cognition and issue involvement (Exp3–4), frequency of social media use (Exp1–2), and social desirability (Exp3–4). Analyses: preregistered ANCOVAs compared post values across conditions controlling for baseline; nonparametric robustness checks were also conducted. Internal random-effects meta-analyses computed standardized mean differences adjusted for baseline; equivalence testing (TOSTER) used SESOI d = ±0.2. Mediation used PROCESS (Exp1 expectancy mediated moderation; Exp3–4 competence mediation). Robustness checks excluded speeders and high social desirability responders; attention checks were used (96–99% passed). Preregistrations: Exp1 vp626; Exp2 7zy4r; Exp3 7ui2w; Exp4 tv3sw (AsPredicted). Data collection ran 09/2020–07/2021. Ethics: negligible-risk, nonidentifiable data; informed consent; debriefed; IRB exemption at University of Erfurt.

- Hostile vs neutral misinformation (denier): Internal meta-analysis showed slightly higher intentions after hostile than neutral misinformation (d = 0.12 [0.06, 0.19]); attitudes showed a non-significant small effect (d = 0.06 [-0.07, 0.20]). Equivalence tests indicated both effects were within trivial bounds (|d| < 0.2). - Neutral vs hostile rebuttal (advocate): Across experiments, neutral rebuttal outperformed hostile rebuttal on attitudes with a small effect (d = -0.13 [-0.21, -0.04]; negative indicates hostile worse), while intentions showed a trivially small effect (d = -0.08 [-0.20, 0.04], equivalence supported for triviality). - Rebuttal vs no rebuttal: Both neutral and hostile rebuttals significantly mitigated misinformation compared to advocate absent. Neutral vs absent: attitude d = 0.53 [0.30, 0.76], intention d = 0.42 [0.32, 0.53]. Hostile vs absent: attitude d = 0.39 [0.24, 0.54], intention d = 0.34 [0.17, 0.51]. - Perceived competence: Hostility substantially reduced perceived competence. Meta-analytic effects: denier hostile vs neutral d = -0.95 [-1.14, -0.77]; advocate hostile vs neutral d = -0.80 [-0.98, -0.62]. - Mediation: Across experiments, perceived competence mediated the small adverse effects of hostility on attitudes and intentions for both denier and advocate. Expectancy violations did not mediate effects (Exp1 found no expectancy differences by tone alignment). - Moderators: No consistent moderation by verbal aggressiveness, need for cognition, issue involvement, or social media use; a single social media use interaction in Exp1 did not generalize. - Manipulation checks confirmed hostility manipulations were perceived as less polite (very large ds > 1.7). Misinformation alone (no rebuttal) reliably reduced attitudes and intentions across applicable experiments (ds ~0.35–0.56).

The findings show that hostility in social media discussions plays a minor role in shaping persuasive outcomes (attitudes and intentions) among U.S. audiences: neutral rebuttals are only slightly more persuasive than hostile ones, and hostile tone in misinformation does not meaningfully blunt its impact. Crucially, rebuttals—regardless of tone—consistently mitigate the damage of misinformation relative to silence, supporting the strategic value of engaging and correcting. However, hostility exacts a cost in source impressions, markedly lowering perceived competence, which is central to trust and may indirectly reduce persuasive impact. The absence of expectancy-violation effects suggests that hostility may be normatively expected online, especially from non-authority personas. Weak direct effects may reflect offsetting mechanisms: hostile language could increase perceptions of authenticity, convey anger that signals norm violation, suggest dominance, or increase attention and engagement, balancing competence losses. Despite small direct effects, the consistent competence penalty counsels against hostility when reputation and trust are priorities. Practically, science advocates need not fear catastrophic persuasive losses if provoked into hostile tone, but should still prefer civility to sustain perceived competence and long-term trust. The results also reinforce that leaving misinformation unchallenged is detrimental, as the strongest negative outcomes occurred when advocates were absent.

Across four preregistered experiments on vaccination and GM foods, hostility had trivial-to-small effects on persuasion: neutral rebuttals were only modestly more effective than hostile ones, while both forms of rebuttal clearly outperformed no response. Misinformation remained impactful regardless of tone. Hostility substantially reduced perceived competence of both deniers and advocates, and competence perceptions mediated small persuasive disadvantages. Key contributions include cross-topic tests, internal meta-analyses with equivalence testing against a prespecified SESOI, and identification of competence as a consistent perceptual mechanism. For practice, advocates should prioritize rebutting misinformation and, when possible, avoid hostility to maintain perceived competence and trust. Future work should test different cultural contexts and platforms, measure real behaviors beyond intentions, examine nonverbal and multimodal hostility, explore interactions with speaker identity (e.g., officials vs lay users), and assess long-term reputational and polarization outcomes.

- Samples were all U.S. adults; cultural context and norms around hostility may limit generalizability. - Measured intentions and attitudes, not actual behaviors (e.g., vaccination uptake or purchases). - Focused on verbal hostility (insults, profanity, capitalization) in text-only scenarios; findings may not generalize to nonverbal, audiovisual, or physically aggressive expressions. - Fictitious social media scenarios may differ from real platform norms, relationships among discussants, and interactive dynamics (e.g., audience becoming targets of hostility). - Generalization to other public discussion formats (e.g., televised debates with visual cues) is limited. - Manipulated tone while holding arguments constant; in real interactions, hostility may alter argument content or coherence. - Moderator analyses found little evidence of audience characteristics shaping tone effects, but other unmeasured factors may matter.