Joint Chum Technical Committee. Accuracy and Precision of Genetic Stock Identification for Estimating the Stock Composition of Mixed-Stock Chum Salmon Fisheries in Northern Puget Sound and Southern Georgia Strait. TCCHUM (92)-2. February 18, 1992.

Summary

1. Simulations were conducted to quantify the accuracy and precision of the WDF 21-locus, 36-stock baseline (1989 version) used to estimate the stock composition of mixed-stock catches of chum salmon from northern Puget Sound. The baseline provided accurate estimates with acceptable precision of the proportional contribution of U.S. and Canadian chum salmon stock groups to mixed-stock samples with stock compositions representing commercial fisheries in northern Puget Sound and southern Georgia Strait.

2. For the models and mixture compositions examined in this report, mixture sample sizes of 100 did not provide estimates with acceptable accuracy or precision. There was usually a large increase in precision (decrease in variance of the estimates) when the mixture sample size increased from 100 to 200. There were more modest gains in precision for increases in sample size over 200. The minimum sample size for a mixture sample should be 200 fish.

3. For the simulations in this report, both 7-loci and 21-loci analyses provided accurate estimates of the total U.S. and total Canadian contribution to mixed-stock samples representing commercial fisheries in northern Puget Sound and southern Georgia Strait. In most cases, 21 loci provided more precise estimates of stock composition than 7 loci.

4. The largest bias of the stock contribution estimates occurred at extremes of stock composition, such as when one or more stock groups contributed 20% or less to the mixture sample. We recommend that methods of bias correction be examined for mixture samples where one or more stock groups are estimated to contribute 20% or less to the sample.

5. Three different GSI models were used in the simulations for this report. There were differences in the models' treatment of the baseline electrophoretic data, number of loci analyzed, analytical method used to estimate stock composition, and bootstrapping procedures used to estimate standard errors. The three GSI models produced similar stock composition estimates when analyzing the same specified stock mixture, especially for the total U.S. and total Canadian contribution