Distinguishing niche and neutral processes: issues in variation partitioning statistical methods and further perspectives

Abstract

Variance partitioning methods, which are built upon multivariate statistics, have been widely applied in different taxa and habitats in community ecology. Here, I performed a literature review on the development and application of the methods, and then discussed the limitation of available methods and the difficulties involved in sampling schemes. The central goal of the work is then to propose some potential practical methods that might help to overcome different issues of traditional least-square-based regression modeling. A variety of regression models has been considered for comparison. In initial simulations, I identified that generalized additive model (GAM) has the highest accuracy to predict variation components. Therefore, I argued that other advanced regression techniques, including the GAM and related models, could be utilized in variation partitioning for better quantifying the aggregation scenarios of species distribution.