Evaluating the role of data quality when sharing information in hierarchical multi-stock assessment models, with an application to Dover Sole

An emerging approach to data-limited fisheries stock assessment uses hierarchical multi-stock assessment models to group stocks together, sharing information from data-rich to data-poor stocks. In this paper, we simulate data-rich and data-poor fishery and survey data scenarios for a complex of dover sole stocks. Simulated data for individual stocks were used to compare estimation performance for single-stock and hierarchical multi-stock versions of a Schaefer production model. The single-stock and best performing multi-stock models were then used in stock assessments for the real dover sole data. Multi-stock models often had lower estimation errors than single-stock models when assessment data had low statistical power. Relative errors for productivity and relative biomass parameters were lower for multi-stock assessment model configurations. In addition, multi-stock models that estimated hierarchical priors for survey catchability performed the best under data-poor scenarios. We conclude that hierarchical multi-stock assessment models are useful for data-limited stocks and could provide a more flexible alternative to data-pooling and catch only methods; however, these models are subject to non-linear side-effects of parameter shrinkage. Therefore, we recommend testing hierarchical multi-stock models in closed-loop simulations before application to real fishery management systems.