SOME PRELIMINARY EVIDENCE ON SAMPLING OF ALTERNATIVES WITH THE RANDOM PARAMETERS LOGIT

Random utility models rely on the properties of the logistic distribution for ease of estimation, but this distribution implies the independence of irrelevant alternatives (IIA). The random parameters logit model offers a means of avoiding the IIA assumption as well as greater heterogeneity among agents, recreational anglers or beachgoers in the current application. A problem often encountered in the estimation of random utility models with many alternatives is the necessity of sampling alternatives or otherwise reducing the number of choices. Research has shown that in the random utility model, such changes in choice set still lead to consistent parameter estimates. However, with the random parameters logit, there is greater need to sample but no theoretical evidence that sampling is justified. In this paper we show the impact of sampling in a random parameters logit model. We find that sampling does not appear to change the parameter estimates substantially. We investigate two data sets: a study of beach use in the Chesapeake Bay and a study of marine recreational angling behavior for the Northeast of the U.S.