Generalized linear mixed effects models with application to fishery data

Abstract

The generalized linear model (GLIM) represents a versatile class of models suitable for several types of dependent variables. GLIMs are popular models and are often an appropriate choice for modelling fisheries data. However, fishery data and corresponding models tend to be complex, because of the complexity of the populations the data are sampled from. In this practicum we use generalized linear mixed effects models (GLMMs), which are GLIMs in which some parameters are random effects to model two different fisheries data sets. The first involves a time series of biological samples used to determine fish maturity, and the second involves paired-trawl catch data to determine if there is a difference in catch rates between two fishing vessels. In this research we find that GLMMs improve estimates of maturities in a selected fish stock and can be used to model differences in catch rates between fishing vessels effectively. This research also suggests that prediction and forecast accuracies are improved by using GLMMs. We also provide some simulation results and found that, overall, GLMMs appear to perform better than GLIMs in terms of bias, coverage errors, and power tests.