Addressing dependency in the sportfishing valuation literature: Implications for meta-regression analysis and benefit transfer

- Dependency avoidance changes the estimated partial effects in the metadata.
- Different models and dependency treatments result in different model coefficients.
- MRFBTs tend to predict better toward center of the distribution of observed values.

Meta-regression analysis is a statistical summary or synthesis of a body of evidence. However, when primary studies provide more than one estimate, the presence of dependence in the metadata has implications for the statistical efficiency of estimated moderator variables. Previous meta-analyses have adjusted for within study dependence through ad hoc procedures (e.g., selecting one estimate per study and study average) or regression-based methods (e.g., weighted and panel data models). This paper defines dependency based on the underlying primary data (i.e., from the same sample) and examines the effect of different models and treatments on meta-regression estimation and implications for benefit transfer performance. The models are applied to the sportfishing literature that contains 140 papers providing 833 estimates of access values for fishing in the United States and Canada. The different methods of adjusting for dependency within the sportfishing metadata result in differences in the estimated model coefficients; hence, different transferred values and transfer errors.