Abstract
Taxonomists and evolutionary biologists frequently use a combination of morphological measurements to distinguish between species and investigate local adaptation. However, the entire set of characters often displays various degrees of collinearity. This paper discusses the effect of using collinear data in morphological taxonomy and ways to handle multicollinearity in a classification context, with special consideration for small sample size. In addition, I propose a robust and easy-to-use combination of dimension reduction using partial least squares (PLS) with traditional discriminant methods for morphological data. To do this, I investigated morphological variation patterns among four monogenean populations from the Pacific Ocean using the correlated morphological features of the sclerotized attachment organ. The new approach yielded better prediction results (lower classification error rates) than the traditional dimension reduction method based on principle component analysis (PCA) and is also much more robust for small sample size. This emphasizes that PLS may be more efficient than PCA in dealing with correlated data and extracting the most relevant morphological differences among groups.