Rating our certainty: how confidence judgments amplify belief polarization

Although metacognitive reflection (e.g., thinking about our confidence or certainty in a decision) is often assumed to improve decision-making, recent research suggests that eliciting confidence ratings can sometimes produce counterintuitive results, including the amplification of subjective beliefs. We report three experiments investigating how trial-by-trial confidence judgments affect the extremity of beliefs. In Experiments 1 and 2, participants provided subjective ratings of how much they liked a series of paintings and the extent to which they agreed with various social axioms, either accompanied by confidence ratings or not. Across both studies, participants who were asked to rate their confidence, gave significantly more extreme responses relative to those who did not provide confidence ratings. By contrast, in Experiment 3, when participants provided objective estimates of the age of the paintings, confidence ratings led to less extreme estimates compared to a control group. We found no evidence that the effect of eliciting confidence ratings was moderated by the average confidence associated with a judgment. We situate these findings within the broader literature on metacognition and belief polarization, discussing possible mechanisms, such as demand characteristics, selective rationalization, and cognitive biases, that may explain why confidence judgments exacerbate rather than reduce extreme subjective beliefs. We conclude with implications for interventions seeking to address polarized thinking, highlighting that confidence measurement alone may backfire unless paired with strategies to encourage genuine self-reflection and open-mindedness. Public surveys, classrooms, and social-media polls now routinely ask people not only what they think, but how sure they are. The hope is that eliciting these confidence judgments has no effect on people’s underlying beliefs. Yet that assumption has not been empirically tested. Here, we asked whether confidence prompts affect people’s beliefs and judgments. Across three experiments, participants repeatedly expressed a judgment either about personal preferences (e.g., “Do I like this painting?”) or about a verifiable fact (e.g., “When was this painting created?”). Half of the participants also rated their confidence on every trial. Confidence prompts polarized subjective opinions - people’s subjective ratings became more extreme - while simultaneously softening objective estimates. These findings show that measuring confidence in people’s beliefs may unintentionally modify those beliefs.