The purpose of this dissertation is to study the specification of the choice set and its influence on the parameters of the utility function in a random utility model of recreation demand. In addition, the dissertation investigates the implications of the choice set specification for measuring welfare losses following beach closures. The random utility model is widely used in recreation demand modeling. It estimates the probability of visiting a site, conditional on the travel cost and the attractiveness of the site, as measured by the quantity and quality of its attributes. An index of the attractiveness of all sites in the choice set, called inclusive value, divided by the travel cost coefficient is a measure of the consumer surplus obtained from a choice situation. The random utility model is, therefore, widely used for measuring the welfare losses from closing a site or the welfare gains from quality improvements.
I employ three sets of data. The University of Delaware designed an Internet-based survey of a random sample of households from the Mid-Atlantic region. Knowledge Network of Menlo Park, CA administered the survey. The survey collected data on historical beach visitation and, separately, on visitations in the 2005 beach season. In addition, the survey asked respondents about their familiarity with seven beach regions in the Mid-Atlantic. I also collected data on travel costs and beach characteristics.
I estimate three random utility models with different specifications of the choice set based on the information about beach familiarity. A conventional random utility model estimated on the full choice set of 66 beaches serves as a baseline. Then, I fit a second conditional logit model estimated over only the set of familiar beaches. Finally, I estimate three mixed logit models with error components on the dummy variables for familiar and unfamiliar beaches. The mixed logit models differ in the level of completeness of the specification of the utility for unfamiliar beaches.
The results from the baseline model are consistent with prior expectations. High travel costs, private beaches, beaches that allow vehicle access and beaches in New Jersey reduce the probability of visitation. This probability grows with beach length and width, the availability of commercial boardwalk, an amusement park, a park within the boundaries of the beach community, as well as the status of the beach as a state park. The coefficients on developed beaches and Atlantic City are also positive. Only one coefficient, that on the variable indicating good surfing conditions, is statistically insignificant.
In the conditional logit model estimated over only the familiar beaches the implicit prices of several beach attributes fall significantly. This implies that the variables Atlantic City, length, private beach, park within, and beach width have a larger influence on choice set formation than on the utility of beachgoers. Once Atlantic City and larger beaches enter people’s choice set, their effect on the probability of choice is significantly reduced.
The mixed logit models find that the utilities of familiar beaches are correlated with each other. In other words, familiar beaches are close substitutes for each other, while unfamiliar beaches are not. If a familiar beach is closed, the welfare loss will be mitigated by the presence of close substitutes.
The welfare losses from beach closure at less visited beaches tend to be driven by changes in the travel cost coefficient. A conditional logit model estimated over the full choice set will provide unbiased estimates of those losses. The tendency of losses to rise in choice set formation models relative to the baseline model, because a smaller number of substitutes are available, is not a simple function of beach familiarity, as is implied by the received literature. For example, a beach has a larger substitution effect, if a neighboring beach has a strong substitution effect.
The substitution effect increases with the scale of beach closure. When closing individual beaches, in only about a third of beaches the losses are larger in the conditional logit model estimated over only the familiar beaches relative to the baseline model. When closing whole beach regions, the welfare losses from the choice set formation models always exceed the losses from the baseline model. Therefore, when evaluating large-scale events that result in closure of a group of beaches, a simple conditional logit model estimated over the full choice set will significantly underestimate the true losses.