Controlling biases in targeted plant removal experiments.

Targeted removal experiments are a powerful tool to assess the effects of plant species or (functional) groups on ecosystem functions. However, removing plant biomass in itself can bias the observed responses. This bias is commonly addressed by waiting until ecosystem recovery, but this is inherently based on unverified proxies or anecdotal evidence. Statistical control methods are efficient, but restricted in scope by underlying assumptions. We propose accounting for such biases within the experimental design, using a gradient of biomass removal controls. We demonstrate the relevance of this design by presenting (1) conceptual examples of suspected biases and (2) how to observe and control for these biases. Using data from a mycorrhizal association-based removal experiment, we show that ignoring biomass removal biases (including by assuming ecosystem recovery) can lead to incorrect, or even contrary conclusions (e.g. false positive and false negative). Our gradient design can prevent such incorrect interpretations, regardless of whether aboveground biomass has fully recovered. Our approach provides more objective and quantitative insights, independently assessed for each variable, than using a proxy to assume ecosystem recovery. Our approach circumvents the strict statistical assumptions of, for example, ANCOVA and thus offers greater flexibility in data analysis.