Simulation study of generalized linear mixed effects models on fishery data

Wheeler, Melissa Doloras (2012) Simulation study of generalized linear mixed effects models on fishery data. Masters thesis, Memorial University of Newfoundland.

[English] PDF (Migrated (PDF/A Conversion) from original format: (application/pdf)) - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (48MB) [English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. (Original Version)

Abstract

The generalized linear model (GLM) is a class of versatile models suitable for several types of dependent variables. GLMs are commonly used to model maturity data. Generalized linear mixed models (GLMM) are a useful extension of the GLM with the addition of random effects. GLMMs have previously been used to improve the estimates of the maturities and provide better predictions of maturities in the near future. Dowden (2007) used GLMMs to model a Atlantic cod maturity data set. His research found that GLMMs improved maturity estimates and forecast accuracy over the GLM commonly used. The results also revealed potential year effects in the cod data. This may be due to actual year effects or some other source such as sampling error. In general it is unknown whether year effects are present in a data set. In this practicum we first provide an overview of Dowden's results. Then we conduct a simulation study to investigate which GL 11M provides the most accurate estimates of the simulated maturities and parameters under a range of simulation factors including the presence of year effects. The two GLMMs used to model the simulated data are an autoregressive (AR) mixed model and a AR mixed effects model with random year effects (AR YE). In this research we find the AR YE model appears to be more appropriate than the AR model when the presence of year effects are unknown. The AR YE model's estimates are similar or better than the AR model's and it also tends to be either as efficient or more efficient depending on the presence or size of the year effects.

Actions (login required)

View Item