Comparison of Uncertainty Estimates in the Short Term Using Real Data

Abstract

In response to increased interest in the Precautionary Approach, various approaches have been applied to characterize the uncertainty of fisheries assessment projection results. Using three case studies, a comparison of some commonly applied techniques was undertaken to determine if different methods give similar perceptions of uncertainty in the short term, with the same, or very closely similar structural models. The techniques for estimating statistical uncertainty included the delta method, the parametric bootstrap of data, the nonparametric bootstrap of residuals and Bayes. Each method was used to derive cumulative frequency distributions of SSB for 1998 and of change in SSB for 1998 relative to 1992. These comparisons were contrasted against the sensitivity of uncertainty estimates to fundamental structural assumptions such as separability. Results displayed measurable and often repeatable patterns in differences between methods of estimating uncertainty, suggesting that these differences were peculiar to the methodology and assumptions. The delta method displayed distributions with longer left tails. Results from Bayes and bootstrap percentile methods were similar. Bias adjusted results were more conservative. Often however, differences could be greater when fundamental structural assumptions were altered, indicating that structural relationships must be either clearly established or proper account taken of this model uncertainty.