Choice Attributes and Behavioral Errors in Decision Making Under Risk : An Empirical Evaluation

We empirically evaluate the performance of three heteroskedastic error models, namely the Contextual Utility, the Stronger Utility, and the Entropy models, relative to the classical Fechner specifications. The three heteroskedastic specifications are different in how complete they incorporate the choice information into their specifications, with the former including fewer attributes than the latter two. We account for unobserved heterogeneity by modeling all structural parameters as individual-level random coefficients that follow a population distribution. Our analyses reject homoskedasticity in favor of Entropy and the Stronger Utility models. The inferior fitness of Contextual Utility implies that merely allowing for heteroskedasticity does not necessarily improve the goodness-of-fit of a model. What matters is which form of heteroskedasticity is introduced. Finally, our results show that the estimation of risk parameter under Expected Utility Theory is robust across alternative stochastic models, whereas the risk parameters estimation under Rank Dependent Utility Theory exhibits some sensitivities