The proliferation of misinformation online necessitates the exploration of effective mitigation strategies. Social corrections—corrective comments by social media users—offer a promising approach. Previous research, primarily focusing on health-related misinformation in the US, suggests their efficacy in reducing misinformation spread. However, several critical questions remain unanswered. Firstly, does the effectiveness of social corrections generalize beyond health contexts and US samples? Secondly, how does the form and strength of corrections influence their impact? Thirdly, do 'social miscorrections' (incorrectly flagging true news as false) have similar, albeit negative, effects? Finally, what are the underlying psychological mechanisms that determine individual susceptibility to the influence of social corrections? This study addresses these gaps by conducting three pre-registered experiments across the UK, Italy, and Germany, investigating the effects of social corrections and miscorrections on diverse topics using a large sample size.

Existing literature demonstrates the potential of social corrections to combat misinformation, particularly in health contexts and US samples. Studies have shown that corrective cues can reduce perceived accuracy, engagement, and even alter attitudes towards misinformation. However, limitations exist. Research has primarily focused on health-related topics and US samples, neglecting the potential differences across diverse contexts and cultures. Operationalizations of social corrections have also varied, hindering the understanding of the role of correction strength and format. Importantly, the potential negative impact of miscorrections on true news has been highlighted but not extensively researched. Finally, the underlying psychological mechanisms remain unclear, with questions around the role of anti-expert sentiment, cognitive reflection, and susceptibility to social influence needing further investigation. This study builds upon this body of research by exploring these gaps across multiple countries and a wider range of topics.

Three online experiments were conducted in the UK, Italy, and Germany using Dynata, recruiting a total of 6621 participants (1944 UK, 2467 Italy, 2210 Germany). Participants completed a pre-treatment questionnaire assessing demographics, anti-expert sentiments, cognitive reflection capacities, and susceptibility to social influence. They then evaluated nine social media posts (six false, three true) presented in one of four conditions: control (no comments), low amplification correction/miscorrection, high amplification correction/miscorrection, and correction/miscorrection with link. Post-treatment, participants rated post accuracy, likelihood of liking, and sharing. The study employed a within-subjects design, with each participant evaluating posts under various conditions. The operationalization of amplification varied slightly across countries: UK (likes), Italy (number of comments), and Germany (number of comments; source cue manipulation for true news in Germany). Linear mixed-effects models were used to analyze the data, investigating the effects of social corrections and miscorrections on perceived accuracy, liking, and sharing, and exploring potential moderating effects of the individual difference measures. All experiments were pre-registered on OSF, and data and code are publicly available.

The results largely supported the hypotheses. Social corrections significantly reduced perceived accuracy of and engagement with false news posts across all three countries. Specifically, high-amplification corrections and corrections with links were consistently more effective than low-amplification corrections, although the low amplification condition still showed a significant effect in most instances. Moreover, social miscorrections significantly reduced perceived accuracy of and engagement with true news posts, across countries and conditions (though the low amplification miscorrection condition was not always statistically significant). Interestingly, there was no consistent significant difference between the low, high, and link condition effects. No significant moderating effects were observed for anti-expert sentiments, cognitive reflection capacities, or susceptibility to social influence. Across all three outcomes (accuracy, liking, and sharing) the main effects of the corrections on both false and true news were significant. The effect sizes of the interventions on these three variables were similar in magnitude across all three countries. The magnitude of the effect across various topics was also roughly consistent.

The findings demonstrate the double-edged sword nature of social corrections. While effective in reducing the spread of false information, they can also inadvertently undermine trust in accurate information when misapplied. The lack of moderation by individual differences suggests that social corrections exert their influence through a relatively straightforward mechanism, possibly due to recency effects in low-motivation settings, rather than through complex cognitive processes or social influence. The consistent effect size across countries and topics points towards the robustness of social corrections. However, the potential for miscorrections raises crucial concerns about their widespread implementation. The simplicity of creating effective (and ineffective) corrections implies a low barrier to participation, but also a significant risk of spreading further confusion if not implemented carefully.

This study provides robust cross-cultural evidence of the effectiveness of social corrections in reducing the spread of false news, while also highlighting the potential for negative consequences from miscorrections. Future research should focus on strategies to minimize miscorrections and further explore the underlying mechanisms of social corrections. Developing effective strategies to mobilize social media users to participate constructively in social correction remains a key challenge.

The study relies on self-reported data from an experimental setting, potentially limiting ecological validity. Future research could utilize observational data from real social media platforms to enhance external validity. The design's inclusion of both accuracy and sharing intention as outcomes raises questions about potential interaction effects as noted by Epstein et al. (2023), although the authors argue for the validity of using both measures. Further investigation into the interplay between these outcomes and the impact of social corrections is warranted.