Rising complexity and falling explanatory power in ecology
Quantitative analysis of historical research output has provided several insights into the trajectory of scientific disciplines including a better understanding of shifts in topic selection, shifts in patterns of scientific collaboration and the consequences of selective publication processes. While reviews on the status of ecology have appeared in different journals, none have involved a quantitative analysis of the published record. Here we analyze over 18,000 articles published by the preeminent ecological societies for historical trends in ecology’s statistical methods and outcomes. We used pattern matching regular expressions to search the full text of each article to quantify the number of R2 and p-values. The generated data set was subsequently used to assess trends in these values across time and in sub-domains determined through an unsupervised machine learning approach.
We found that research in ecology is becoming increasingly complex, reporting a growing number of p-values per article, and that the explanatory power of ecology measured by the coefficient of determination (R2) is falling rapidly. We provide three potential mechanisms for the increasing complexity and decreasing explanatory power: 1) the low hanging fruit of ecology have been picked bare 2) we are progressing towards the true mean explanatory power of ecology or 3) there has been a steady shift in the publication bias within the discipline, or at least in the journals reviewed. Irrespective of the causal mechanism of the observed decline in R2-values, the observed trend provides an impetus to evaluate procedures and motivations of the discipline.