Using estimation of demand for the George Washington/Jefferson National Forest as a case study, it is shown that in a stratified/clustered on-site sample, latent heterogeneity needs to be accounted for twice: first to account for dispersion in the data caused by unobservability of the process that results in low and high frequency visitors in the population, and second to capture unobservable heterogeneity among individuals surveyed at different sites according to a stratified random sample (site specific effects). It is shown that both of the parameters capturing latent heterogeneity are statistically significant. It is therefore claimed in this paper, that the model accounting for site-specific effects is superior to the model without such effects. Goodness of fit statistics show that our empirical model is superior to models that do not account for latent heterogeneity for the second time. The price coefficient for the travel cost variable changes across model resulting in differences in consumer surplus measures. The expected mean also changes across different models. This information is of importance to the USDA Forest Service for the purpose of consumer surplus calculations and projections for budget allocation and resource utilization.