Multivariate statistics for dummies: A guide for community ecologists

Since its development, the progress in community ecology, by virtue of its multispecies nature and the multitude of structuring forces, has been tightly linked with the progress in multivariate methods. The ecological practitioner, however, is often confronted with a smörgåsbord of different statistical methods, each with there own assumptions, strengths, and weaknesses. We are often left wondering: Is one method better than the other, does it matter what method I choose, how much does it matter, how will information loss affect my results? I applied the majority of currently used multivariate methods to metacommunity theory, i.e. the relative importance of local environmental and regional dispersal processes in determining community structure. I analyzed 158 data sets with information on multispecies communities, together with environmental and spatial information. I compared all the methods (and their subtle variants) to each other, both based on ordination results, significance structure, and sizes of explained variation components. I also investigated the effect of information loss by reducing the information present in the species data from abundances, to presence/absences, to species diversity (in several of its forms). In terms of metacommunity structure, the majority of the methods gave similar results, although the power in a Mantel test framework was much lower compared to the other methods. The loss in explanatory power when moving from abundances to presence/absences, and then to diversity puts a question mark on the usefulness of the diversity concept in community ecology. Finally, although the general patterns are similar, each method extracted slightly different patterns at an individual data set level, each of them of potentially equal importance. To encourage this open-minded analysis strategy, I provided a unified analyses strategy using the statistical R package.