The Convergent and External Validity of Risk Preference Elicitation Methods

We evaluate the convergent and external validity of several commonly used risk preference elicitation methods with and without controlling for measurement error using the obviously related instrumental variable (ORIV) approach (Gillen et al., 2019). Risk preferences are elicited in a large sample of the Dutch population (N = 4, 282) and linked to field behavior in financial, occupational, and health domains based on register data and survey questions. We find that controlling for measurement error improves the correlation between methods, suggesting that not accounting for measurement error can partly explain the lack of convergent validity among risk preference elicitation methods found in previous studies. At the same time, we find clear differences between revealed and stated preference methods in terms of their external validity. Stated methods correlate well with most types of field behavior and correlations are of economic significance. In addition, controlling for measurement error increases the strength of the relationships found. Revealed methods are at best weakly related to field behavior, even when controlling for measurement error. The difference between revealed and stated methods appears not to be driven by the higher complexity of the incentivized tasks used to elicit revealed risk preferences.