Estimating conditional certainty equivalents using choice-experiment data

Expected-utility theory is embraced by some researchers because of its theoretical and empirical tractability, although empirical testing has exposed systematic behavioral inconsistencies that violate the axiom of independence in the theory. In particular, empirical evidence suggests that people making choices under uncertainty do not have a unique certainty equivalent when single-stage games are turned into multistage games, even if the likelihood of outcomes remains unchanged. We show analytically why a violation of the independence axiom is not inconsistent with random-utility models and data obtained from discrete-choice experiments. Results from a health care application are used to demonstrate the feasibility of estimating certainty-equivalent functions that can be used to test the validity of the independence axiom by modeling data from choice-experiment surveys. Further, we show how estimating certainty-equivalent functions with a simple survival model can also help us understand choice-experiment respondents' implicit perceptions of the likelihood of future outcomes. This information is particularly valuable when only part of the risk details in a multistage game can be provided to respondents who complete a discrete-choice experiment either because the actual risk levels are unknown or because risk levels vary significantly from person to person, neither one of which is uncommon in health applications.

Duke Scholars

Author Juan Marcos Gonzalez Population Health Sciences

Author F. Reed Johnson Population Health Sciences