A comparison of hierarchical models for relative catch efficiency based on paired-gear data for US Northwest Atlantic fish stocks

Abstract

Selectivity and catch comparison studies are important for surveys that use two or more gears to collect relative abundance information. Prevailing model-based analytical methods for studies using a paired-gear design assume a binomial model for the data from each pair of gear sets. Important generalizations include nonparametric smooth size effects and normal random pair and size effects, but current methods for fitting models that account for random smooth size effects are restrictive, and observations within pairs may exhibit extra-binomial variation. I propose a hierarchical model that accounts for random smooth size effects among pairs and extra-binomial variation within pairs with a conditional beta-binomial distribution. I compared relative performance of models with different conditional distribution and random effects assumptions fit to data on 16 species from an experiment carried out in the US Northwest Atlantic Ocean comparing a new and a retiring vessel. For more than half of the species, conditional beta-binomial models performed better than binomial models, and accounting for random variation among pairs in the relative efficiency was important for all species.