Performance comparison of generalized linear model, generalized additive model and spatial interpolation for standardization of catch rate through a simulation study

Catch rate standardization is a necessary step before catch rate can be used in stock assessment models. Generalized linear model (GLM) and generalized additive model (GAM) are commonly used to standardize catch rate. However, there are many practical limitations when using the GLM and GAM. Because in most situations, catch rate data are spatially correlated in the study field, spatial interpolation (SI) is applied as an alternative way to analyze catch rate as an abundance index. In this study, we compared the performance of GLM, GAM and SI through a simulation study which is based on the real fishery independent survey of yellow perch in Lake Erie. For each combination of sample size and error magnitude, 100 simulations were conducted to estimate correlation coefficients between the “real” abundance and standardized catch rate from GLM, GAM and SI. We found that their performance was improved when the sample sizes increased, but it degraded when the magnitude of simulated errors increased. In general the GLM performed consistently better than GAM. SI performed better than GLM and GAM only when the simulated errors were low, and SI was more sensitive than GLM and GAM to the magnitude of the simulated random errors.