An approach to the estimation of the distribution of marginal valuations from discrete choice data

Models such as the mixed logit are often used to measure the distribution of the marginal value of a good based on discrete choice panel data. There are however serious specification and identification issues that are rarely addressed. The consequences for results may be dramatic. This paper points out the issues and presents an approach to dealing with them that may be applied under some circumstances. The issues and the approach are illustrated using a dataset designed to measure the value of travel time.