GLM CPUE STANDARDIZATIONS: INTERPRETATION RISKS FOR MODEL FITS FROM ALTERNATIVE SOFTWARE PACKAGES

SUMMARY Marginal means of CPUE by year from General Linear Model (GLM) fits to CPUE data from fisheries are commonly used as “standardized” CPUE estimates of population abundances where gear, area, and/or other effects have been removed by the statistical analysis. Often there is a lack of balance in the number of observations by factor and year sometimes including instances where gear, area, or other factors or combinations are missing for a year or years. This imbalance diminishes the robustness of the GLM estimates of population trends. Software packages differ in the way they handle these missing strata in the calculation of marginal means. This can lead to “fitted” marginal means that may radically misrepresent the actual population abundance trend. Analysts must be vigilant of the conventions applied by different software packages when interpreting fitted results. RESUME