Beliefs Formation; Survey Experiment; Climate Change; Social Media Bundles.

D83, D82, D03, Q54

We study how exposure to competing climate-change content on social media affects beliefs, opinions, and actions. In a representative survey experiment with 1,633 adults in Italy, respondents were randomly assigned to view climate-advocacy content, climate-skeptical content, or an active control featuring gender-equality content. We elicit beliefs about the reality of climate change and its human causes before and after treatment, as well as post-treatment climate-related opinions, concerning behaviors and policy support, and actions, including sharing a post, signing a petition, and donating. To interpret belief dynamics, we develop a structural updating model that treats the interventions as real-world information bundles and decomposes posterior beliefs into a Bayesian-like underlying belief-updating component, common survey anchoring, and treatment-specific prior-independent responses, including experimenter demand effects and reactions to source identity, framing, and content. This allows us to disentangle information-related varia- tion, captured by the underlying belief-updating model, from non-informational variation. Although we document substantial underreaction to information, the advocacy treatment significantly increases underlying beliefs about the reality of climate change, but not about human causation, whereas the skeptical treatment shifts underlying beliefs toward skepticism in both domains. Standard average treatment effects understate the impact of skeptical content, especially for beliefs about human causation, because treatment effects depend on the distribution of prior beliefs and are compressed when respondents are clustered near the boundaries of the belief scale. Conditional simulations of the estimated updating rule imply asymmetric belief dynamics: under equally likely advocacy and skeptical signals, intermediate beliefs fluctuate closer to advocacy for climate-change reality but closer to skepticism for human causation. Finally, underlying posterior beliefs explain a substantial share of climate-related opinions and actions, while we also identify a backfire effect of skeptical content on support for taxing fossil fuels that operates independently of underlying beliefs.