Arguments have been made that measurement error associated with survey biomass estimates leads to positive bias in GLM estimates of the fishing effect parameter λ for the island closure experiment. The reason for this bias is examined, and found to arise from the assumption of normally distributed survey errors with constant variance. If instead these distributions are log-
normal, with constant CVs, the bias does not occur. Tests plotting λ estimates based on either recruit or spawner biomass survey estimates against the corresponding random year effects estimates from the GLMs suggest that the measurement error bias for these analyses is insubstantial.
History
Department/Unit
Department of Mathematics and Applied Mathematics, University of Cape Town