20868: At the Risk of Loss: Improving Discriminant Validity of Self-Reported Measures of Risk Preference (session 25431)

We argue that self-reported measures of risk preference potentially lack discriminant validity because they may reflect not only individual risk aversion but also other constructs that play a role in decision-making, particularly individual loss aversion. The lack of discriminant validity is a serious problem, especially in studies that use self-reported measures of risk preference in contexts where the expected effects from the constructs that compose risk preference diverge. In such a scenario, replacing construct-specific, incentivized measures of risk preference with self-reported measures can lead to significant endogeneity problems when used either as predictors or control variables. In our study, we analyzed three samples from two independent studies. Our results demonstrate that even after accounting for incentivized risk preference measures that remove latent construct associations and focus on measurement-related confounds, self-reported risk preferences maintain a substantial association with incentivized loss aversion measures. This is an indication that the self-reported measurement captures a mixture of both risk and loss aversion. We also observe that the strength of associations depends on how individuals interpret risk-taking: whether they interpret it as either being related to higher variances or higher losses. Based on this finding, we propose a simple procedure that utilizes this individual heterogeneity of the strength of the confounding effect to enhance the robustness of conclusions drawn from analyses using self-reported risk preference measures.

view paper (if available)