Estimating the relative contributions of distinct populations in a mixture of organisms is a common task of fisheries and wildlife managers. These "Mixed Stock Analyses" commonly use genetic data. Generally, component learning samples allow for conditional maximum likelihood estimation of contribution rates in a finite mixture model. In application, the presence or absence of rare components in a specific mixture is determined using nonparametric bootstrap confidence intervals. This is subject to increased Type I and II errors compared to a likelihood ratio test of zero contribution (note that the potentially contributing components are considered known).
The likelihood test is demonstrated in a study of simulated harvest mixtures of sockeye salmon (Oncorhynchus nerka) from the Kenai River, Alaska. Genetic data are used to estimate mixture contributions from a known baseline of 44 potentially contributing populations. The more fundamental problem of using bootstrap confidence intervals as a test of detection for discrete data is briefly discussed.
|