Semiparametric and semi-nonparametric welfare impact analyses with applications to natural resource valuatio

This dissertation contains two theses. Demand based valuation and contingent valuation are the two methods discussed and used to value the demand for natural resources.;The first thesis begins with a discussion of the problems of demand based valuation in the conventional approach. A new modeling method, the extended Average Derivative Estimator (ADE) modeling method, is introduced to estimate the average Marshallian surplus and the average Hicksian compensation due to a price change. The new modeling method has the property of nonparametric estimation in that no parametric functional forms are assumed for preferences. The estimated welfare impacts are consistent with the maintained hypothesis of individual utility maximization. The method can also be used to estimate the average welfare impacts when observations are censored, which is often the case in the recreational travel cost model. The asymptotic distributions of the ADE estimates are developed. A Monte Carlo simulation is used to test the ADE modeling method. A recreational travel cost data is used to carry out an empirical estimation of the average Marshallian surplus due to a price change.;The second thesis addresses the econometric model specification and estimation of the random utility models for the contingent valuation of environmental quality improvements. It is suggested that the random utility interpretation and Cameron's dual interpretation of the discrete choice model require a different econometric model specification in order to estimate the true nonparametric model. Theoretical implications of a random utility model are derived to guide the selection of the functional form. The semi-nonparametric estimation is used to empirically estimate the six random utility models under various transformations and constraints to demonstrate the importance of relaxing the parametric assumptions. Formula to recover the average mean willingness to pay from the random utility model needs to be consistent with the estimated econometric model. When there are missing variables in the model specification, asymptotic distribution of the average willingness to pay is affected by the missing variables. Statistic inference from the model with the missing variables can be biased.