Causal assumptions and causal inference in ecological experiments. (December 2021)

• Record Type: Journal Article
• Authors: Kimmel, Kaitlin; Dee, Laura E.; Avolio, Meghan L.; Ferraro, Paul J.
• Is Part Of: Trends in ecology & evolution. Volume 36:Number 12(2021)
• Page Start: 1141
• Page End: 1152
• Publication Date: 2021-12
• Subjects: counterfactual causality -- potential outcomes -- excludability -- exclusion restriction -- interference -- noncompliance; Ecology -- Periodicals; Evolution (Biology) -- Periodicals; 576.8
• DOI: 10.1016/j.tree.2021.08.008
• Languages: English
• ISSNs: 0169-5347

Abstract

Abstract : Causal inferences from experimental data are often justified based on treatment randomization. However, inferring causality from data also requires complementary causal assumptions, which have been formalized by scholars of causality but not widely discussed in ecology. While ecologists have recognized challenges to inferring causal relationships in experiments and developed solutions, they lack a general framework to identify and address them. We review four assumptions required to infer causality from experiments and provide design-based and statistically based solutions for when these assumptions are violated. We conclude that there is no clear demarcation between experimental and non-experimental designs. This insight can help ecologists design better experiments and remove barriers between experimental and observational scholarship in ecology. Highlights: Causal inferences require causal assumptions. To formalize the assumptions required to draw causal inferences from experimental data, scholars have leveraged insights about causal inference in observational settings. Even carefully designed experiments may face challenges in satisfying four important causal assumptions. Ecologists sometimes acknowledge and address these challenges but do not have a cohesive framework for understanding them. When the validity of a causal assumption is questionable, ecologists can apply design-based and statistically based solutions. Despite popular wisdom, no clear demarcation exists between experimental and non-experimental designs. When inferring causal relationships from data, experimentalists need to be just as careful as non-experimentalists in assessing the validity of their assumptions.