Embedding a Field Experiment in Contingent Valuation to Measure Context- Dependent Risk Preferences: An Application to Wildfire Risk

This paper contributes towards the development of an approach that would generate welfare measures that accommodate non-expected utility risk preferences. Combining the merits of elicitation approaches used in field experiments with contingent valuation, we embed an experimental design that systematically varies probabilities and losses across a survey sample in a willingness to pay elicitation format, where a hypothetical situation is described that closely resembles the actual policy context. We apply the proposed elicitation and estimation approaches to estimate the risk preferences of a representative homeowner who faces probabilistic wildfire risks and an investment option that reduces losses due to wildfire. Based on prospect theory, we estimate parameters of probability weighting, risk preferences and use individual characteristics as covariates for these parameters and as utility shifters. We find that risk preferences are consistent with non-expected utility theory. We also find that the use of individual characteristics as utility shifters, as is standard non-market valuation, attenuates some of the effects of the probability weights. Prior experience with wildfire is associated with risk preferences that are closer to expected utility theory. Introduction When stated preference approaches are used to elicit willingness to pay in probabilistic contexts, modeling underlying individual risk preferences based on expected utility (EU) would lead to biased welfare estimates if in fact risks preferences are consistent with non-expected utility (nonEU) (e.g. Jindapon and Shaw 2008, Shaw and Woodward 2008, Riddel and Shaw 2006). EUbased welfare measures would either overstate or understate predicted net social net benefits of a proposed policy depending on how much true risk preferences deviate from linear risk preferences and the direction and the magnitude of shifts in probabilities brought about by the policy. This has serious implications for the application of contingent valuation (CV) to a variety of contexts in which policy outcomes are probabilistic, such as those policies that address