| Summary: | Estimating a demand system under the assumption that preferences are homogeneous may lead to biased estimates of parameters for any specific individual and significantly different expected consumer surplus estimates. This paper investigates several different parametric methods to incorporate heterogeneity in the context of a repeated discrete-choice model. The first is the classic method of assuming utility to be a function of individual characteristics. Second, a random parameters method is proposed, where preference parameters have some known distribution. Random parameters logit causes the random components to be correlated across choice occasions and, in a sense, eliminates IIA. Simulation noise is discussed. Finally, methods are proposed to relax the assumption that the unobserved stochastic component of utility is identically distributed across individuals. For example, randomization of the logit scale, which is a new method, allows noise levels to vary across individuals without the added burden of explaining the source using covariates. The application is to Atlantic salmon fishing, and expected compensating variations and changes in trip patterns are compared across the models for three policy-relevant changes in fishing conditions at the Penobscot River, the best salmon fishing site in Maine.
|
|---|