Modeling Demand for Outdoor Recreation Settings with Choice Based Data Accounting for Exogenous and Endogenous Stratification

Estimating regional demand models by pooling different samples without correcting for such differences causes model misspecification as each sample belongs to a different population. Weighted regression using Pseudolikelihood to account for differences in sample population with adjustment for heteroskedasticity improves efficiency but the estimates are biased. We estimate regional demand for National Forest settings types in the southeastern states of U.S using weighted and unweighted regression. Using estimation of demand for National Forests as a case study, we resolve problems relating to inference about the data generating process when different samples are pooled together. We show that though efficiency of weighted estimates improves after correcting for heteroskedasticity, they still remain biased as the weights interact with covariates to explain part of model misspecification. In this paper, we show that it is best to use unweighted regression including interactions with weights as covariates.