Bootstrapping Cases Vs. Residuals: Which Is Better for Estimating Growth of Skipjack from Tagging Data?

Our study explores the advantages and shortcomings of two different bootstrapping techniques in the context of growth curve estimation: resampling of residuals and resampling of cases. In tagging experiments, samples are often unbalanced with higher numbers of small (young) fish compared to larger (older) fish. With case bootstrapping, bootstrap samples that, by chance, lack either large fish or long times at liberty will result in unrealistic estimates of the asymptotic length, L_∞. As such, resampling residuals rather than cases allows us to maintain sample sizes of larger fish and longer times at liberty, resulting in more accurate estimates of L_∞. We illustrate this with a simulation modeled after Skipjack tuna. We explore the factors that affect the performance of each method and highlight instances where it is preferable to resample residuals over cases. These results hold if the model has been correctly identified. However, we caution against the use of bootstrapping when that is not the case and explore as an example the use of Fabens’ method in the presence of measurement error. In this situation, both techniques can lead to highly unreliable confidence intervals so caution is advised.