Some insights into data weighting in integrated stock assessments

Abstract

The results of fishery stock assessments based on the integrated analysis paradigm can be sensitive to the values for the factors used to weight each of the data types included in the objective function minimized to obtain the estimates of the parameters of the model. These assessments generally include relative abundance index data, length-composition information and conditional age-at-length data, and algorithms have been developed to select weighting factors for each of these data types. This paper introduces methods for weighting conditional age-at-length data that extend an approach developed by Francis (2011) to weight age- and length-composition data. Simulation based on single-zone and two-zone operating models are used to compare five tuning methods that are constructed as combinations of methods to weight each data type. The single-zone operating models allow evaluation of the tuning methods in terms of their ability to provide unbiased estimates of management-related quantities and the correct data weights in the absence of model mis-specification, while the two-zone operating models allow the impacts of model mis-specification on the performance of tuning methods to be explored. The results of assessments are sensitive to data weighting, but the choice of method for data weighting is most consequential when there is model mis-specification. Overall, the results indicate that arithmetic averaging of effective sampling sample sizes from the McAllister and Ianelli (1997) approach is inferior to other methods, and the new method for computing effective sample sizes for conditional age-at-length data seems most appropriate.