Modeling recreation site choice: Do hypothetical choices reflect actual behavior?

We examine the ability of revealed preference (RP), site-specific stated preference (SP), transferred SP, and joint RP-SP models to predict aggregate and individual recreation site choice in a holdout sample. For two statistical comparisons, the RP model provided the most accurate predictions of individual choices. However, the transferred SP model, applied directly or estimated jointly with the RP data, performed best in three aggregate and one individual prediction test. These findings suggest that data from well-designed and conducted SP surveys from one site can be combined with site-specific RP data from another site to generate improved models of recreation site choice.