Stock Identification with the Maximum-Likelihood Mixture Model: Sensitivity Analysis and Application to Complex Problems

Simulations were performed to evaluate the bias and precision of stock composition estimates from the maximum-likelihood mixture model using hypothetical multilocus characters. Bias and precision were examined in relation to the number of stocks being resolved, the number of loci available, and the difference in allelic frequency among stocks at each locus, using Monte Carlo simulations with different levels of sampling error in the mixture and learning samples. Model performance improved with increasing stock separation and number of loci available. Bias was not affected by the number of stocks resolved in simulations where mixture contributions from individual stocks remained constant. These results provide guidelines for reducing the complexity of genetic stock-identification problems by summing estimated mixing proportions for individual stocks within groupings based on stock similarity. The trade-off between improved accuracy and level of grouping can be examined graphically to determine the most use...