Beyond Zar: An Evaluation of Frequentist, Bayesian, and Alternate Methods for Classifying Fish through Otolith Chemistry

As computers have become more powerful, statistical methods to classify populations have proliferated. Methods of classifying fish to habitat and natal area now include Bayesian classification, Neural Networks, Random Forests, often in preference to linear (LDA) and quadratic (QDA) discriminant analysis. LDA and QDA have been treated as old-fashioned and inadequate and been cast aside to novel methods. Believing that it is best to use the appropriate method regardless of current fashion or convenience, we compared the ability of these methods to classify fish through simulations of populations with different otolith chemistries. Depending on the circumstances, each of these methods has value, but LDA and QDA perform well.