Model selection for selectivity in fisheries stock assessments

• Poor choices for selectivity patterns can lead to bias.
• Selectivity patterns can be selected using residual analysis and formal model selection methods.
• Simulation studies that integrate model selection methods need to be the focus of future work to improve parameterization of selectivity.

The choice of how to model selectivity differs among approaches to fisheries stock assessment. VPA tends to make only weak assumptions regarding (age-specific) selectivity (asymptotic selectivity and temporal stability of selectivity for the most recent years). In contrast, selectivity is more parametric in “integrated” methods, and can be age-, length-, and age- and length-based. The use of parametric selectivity functions tends to reduce estimation variation because fewer parameters have to be estimated, but incorrect choices for the functional form for selectivity can lead to bias. This paper illustrates effects of poor choices for selectivity on the outcomes of stock assessments, outlines methods for evaluating whether a particular choice for selectivity is appropriate using residual diagnostics, and summarizes current ways to select among alternative functional forms for selectivity. This paper also provides a synthesis of the results of past simulation studies which have explored the ability to correctly parameterize selectivity.