A critique for meta‐analyses and the productivity– diversity relationship

Abstract

It is an exciting time to be an ecologist. Over the past several decades, our discipline has matured from one focused on the assembly of case studies based on natural history, to one that has seen improved conceptual frameworks and mathematical models that help explain ecological phenomena from species coexistence to elemental cycling. The maturation of our discipline has been fostered by many things, including improved technology, increased availability of data, and emergent methods for analyzing large data sets. One factor that has played a central role in the modern synthesis is metaanalyses. Gurevitch et al. (1992) introduced metaanalyses to ecologists and catalyzed their entrance into the ecological literature as a powerful statistical means to assess the generality of pattern and process. Soon after, the U.S. National Science Foundation established the National Center for Ecological Analysis and Synthesis (NCEAS) whose mission it is to bring together ecological data sets so that we could synthesize pattern and process using meta-analysis and many other analytical tools. NCEAS was so successful that it was soon after mimicked by other scientific disciplines (e.g., NESCent, the National Evolutionary Synthesis Center). However, when our initial honeymoon with ‘‘synthesis’’ was over, criticisms began to surface, exposing the inherent warts and flaws of a growing discipline. Some argued that data sets were now being analyzed, and syntheses performed, by researchers who knew little about (or perhaps had never even seen) the systems they were trying to understand. Such ‘‘remote ecology’’ reduces an appreciation for natural history, and may lead to incorrect conclusions because one doesn’t understand the intricacies and contingencies of each system that reveal how pattern is linked to process. Some argued that meta-analyses were proliferating more rapidly than the methods needed for quality control, and that the concatenation of data sets was leading to a propagation of errors.