Accommodating Complex Substitution Patterns in a Random Utility Model of Recreational Fishing

We employed a cross-nested logit (CNL) model that permits a rich pattern of substitution among alternatives within a closed form choice model. The specification we employed is ideal for applications with many choice alternatives, such as the 431 fishing sites in this application. The CNL model provided a significant improvement over multinomial and nested logit model specifications at explaining observed recreational fishing site choices by residents of northern Ontario, Canada. Results from two scenarios illustrated the implications of using the CNL model on spatial substitution patterns and welfare measures associated with attribute change scenarios. The CNL model forecasts demonstrated that the relative change in fishing site use was lower at the most affected sites and higher at sites near the affected sites than was predicted by the multinomial logit model. No consistent pattern was found in mean or variance of welfare estimates associated with hypothetical attribute changes.