Does sample size heterogeneity affect growth model selection?

Abstract

Abstract

To solve the problem of limited age-size data in aquatic organisms, bolstered sample, using simulated data with normal distributions has been proposed. Bolstered is believed to correct the effect of selecting a biased model, by size sample heterogeneity. To analyze these claims, simulated age-length data were generated under nine different scenarios with heterogeneous sample sizes, and then these same data were bolstered with more simulated data to homogenize sample sizes by age. Several candidate models were then evaluated for each data set (original and reinforced data) using a multi model selection (MMS) approach, with additive and multiplicative errors. The Akaike index and Akaike difference were then estimated. The bolstered samples were repeated 100 times. The MMS approach always selected the model that generated the data as the best model with heterogeneous sample sizes. While with the bolstered samples, the generating model was not always selected as the best, and in some cases, it was classified as the worst. It was observed that variance heterogeneity and the assumed error structure do affect the selection of the best model between the original and boosted samples. It is proposed that both the candidate model and the error structure be subjected to the MMS approach when modeling the growth of aquatic species, and that an effort be made to complement the limited data and not reinforce them with simulated data that may have a source bias.