CANCELLED - Accounting for time, space and phylogenies in ecological research

Background/Question/Methods

A core assumption of many ecological analyses is statistical independence i.e. that observations are not inherently linked to one another. In ecological systems, this assumption is often violated, as observations can be correlated across time, space, and phylogenies. Here, we develop a novel framework to account for such correlation structures (non-independence) within linear models and apply this framework to a series (N = 10) of high-profile datasets describing population abundances through time e.g. the Living Planet Index. We then compare our newly developed model against widely applied biodiversity change models used in the literature and assess how inference changes when accounting for temporal, spatial and phylogenetic non-independence.

Results/Conclusions

Our results show that addressing sources of non-independence not only improves inferential validity, but also has a substantial impact on the overall trend in biodiversity change data, regularly inflating noise around trends and altering the trend's magnitude. In some cases, trends even reversed (e.g. a negative trend became positive). These findings raise questions about the validity of inference in existing biodiversity change research.