Uncovering Correlation Sensitivity in Decision Making Under Risk

Allowing risk preferences to depend on the correlation between lottery outcomes can explain behavioral anomalies, while empirical evidence is limited and mixed. Using the framework of correlation sensitivity, we classify preferences into three types and adapt a choice task to categorize subjects. Experiments show that aggregate choices exhibit correlation sensitivity opposite to regret and salience theory predictions. Clustering analysis reveals that a correlation-sensitive minority drives these patterns, while most subjects display no sensitivity. We further disentangle deliberate within-state comparisons from incidental payoff comparisons, finding that both contribute to correlation sensitivity, with deliberate comparisons exerting slightly stronger effects.