Convergent and external validity of risk preferences elicitation methods: Evidence from Vietnam

• This study investigates the validity of risk preference elicitation methods • We conduct a field survey and experiment among farmers in rural Vietnam • Unlike other studies on supporting the use of self-assessment in survey, we find that self-assessment has limited validity • Multiple price lists and loss-gain tasks appeared to perform relatively well. • Our findings support the use of incentivized methods as the first option and hypothetical methods as the second option • Further research is needed to confirm their robustness across different contexts This study investigates the validity of various risk preference elicitation methods among farmers in rural Vietnam by conducting a field survey and experiment. Elicitation methods include four hypothetical tasks and three incentivized tasks. The hypothetical tasks contain a set of self-assessment questions, multiple price list and a set of hypothetical questions taken from a Vietnamese household survey. Most participants show no difficulty understanding the elicitation tasks. Most elicitation tasks provide evidence that participants are, on average, risk averse. Participants appear less risk-averse in the self-assessment of risk attitude than other methods. Unlike other studies on supporting the use of self-assessment in survey, we find that self-assessment has limited validity as it has the least or no relation with other measures as well as low predictive power of risky behaviors. These differences could reflect contextual factors between developed and developing countries. The findings further indicate that the MPL and loss-gain tasks, particularly in incentivized settings, capture behavioral patterns more consistent with real-life decisions. Accordingly, incentivized MPL and loss-gain tasks may be most suitable for researchers seeking an in-depth understanding of risk preferences, while their hypothetical versions can serve as practical tools for quick assessments or for use as control variables. While these results are encouraging, further research is needed to confirm their robustness across different contexts.