Ordination Methods and the Evaluation of Ediacaran Communities

Abstract

Analysis of paleocommunity data poses a challenge because of its multivariate nature, containing counts of many species in many samples. Comparison of the abundance of a single species among all samples provides only incomplete information, whereas attempting to consider every species is impractical. Ordination methods are analytical techniques that reduce the original multivariate dataset to a few important components by creating new synthetic variables designed to explain the maximum amount of original data variability. The ultimate goal is to order the samples along ecologically or environmentally meaningful gradients in order to interpret differences in community structure. This chapter describes three of the most widely-used ordination methods, principal components analysis (PCA), detrended correspondence analysis (DCA), and non-metric multidimensional scaling (NMDS), explaining the methodology of each and outlining their strengths and weaknesses for analysis of paleoecological data. The techniques are illustrated using Ediacaran paleocommunity data from Mistaken Point, Newfoundland. PCA relies on assumptions that are inappropriate for ecological data, such as the requirement that species abundances change in a linear fashion along the environmental gradient, and is not well suited for community ordination. In contrast, DCA and NMDS both perform well with ecological data; DCA incorporates a more ecologically-realistic measure of distance between samples but some of the detrending methods have been criticized, whereas NMDS only assumes a monotonic relationship between compositional similarity and gradient distance. The two methods also have complementary strengths, with DCA typically better at extracting the primary gradient and NMDS better at resolving the overall pattern.