Files
Abstract
This dissertation consists of three essays on theoretical and empirical issues related to modeling choice-based samples. The research objectives include investigating the joint effect of endogenous and exogenous stratification in modeling recreational preferences, estimating recreational demand functions by relaxing the basic and naive assumption of randomnes for a choice-based clustered sample, and modeling heterogeneous recreational preferences in a latent-class model. In this dissertation, it is shown that welfare estimates of willingness to pay are extremely sensitive to model specification when controlling for sampling related problems.In the first essay, it is shown that estimating regional demand models by pooling differentsamples without correcting for such differences causes model misspecification wheneach sample belongs to a different population. Estimating a weighted regression using pseudolikelihood improves the efficiency of estimates after correcting for heteroskedasticity. The estimates still remain biased as the weights interact with covariates to explain part of model misspecification. The comparisons between weighted and unweighted models go unnoticed because results from both models are rarely reported. By reporting results from both models it is shown that it is best to use unweighted regression when the coefficient on interactions with the weight variable are jointly insignificant. However, the model needs to be respecified if these interactions are jointly significant but the estimation still proceeds using an unweighted regression.In the second essay, the dependence between individuals surveyed at a particular sitein a choice-based sample is modeled. This dependence is due to some observed or unobserved site-specific effects. Individuals surveyed at a given site are most likely correlated rather than independent. The above argument is used to develop a mixture model where site specificrandom effects follow a standard normal distribution. When evaluating policy changes such as opening a new site, developing an existing site, or closing an old site, significant site effects show that the expected mean calculations which are used in deriving welfare estimates are sensitive to assumptions about the sampling procedure.In the third essay, heterogeneous recreational preferences in a latent-class model areconsidered. This class is based on frequency of visits to a National Forest and is treatedas latent due to arbitrariness in defining how many visits constitute high or low frequency.The results show different marginal effects for the two populations; high frequency visitors,who take frequent short duration visits mostly to general forest areas; and low frequencyvisitors who take less frequent, long duration trips mostly to developed sites. This information on market segregation between high and low frequency visitors can be of importance to the USDA Forest Service because differences in consumer surpluses across classes provide potential scope for differential pricing policies.