Causal analysis in control–impact ecological studies with observational data

Abstract Randomized experiments have long been the gold standard in determining causal effects in ecological control–impact studies. However, it may be difficult to address many ecologically and policy‐relevant control–impact questions‐such as the effect of forest fragmentation or protected areas on biodiversity through experimental manipulation due to scale, costs and ethical considerations. Yet, ecologists may still draw causal insights in observational control–impact settings by exploiting research designs that approximate the experimental ideal. Here, we review the challenges of making causal inference in non‐experimental control–impact scenarios as well as a suite of statistical tools specifically designed to overcome such challenges. These tools are widely used in fields where experimental research is more limited (i.e., medicine, economics), and could be applied by ecologists across numerous sub‐disciplines. Using hypothetical examples, we discuss why bias is likely to plague observational control–impact studies in ways that do not surface with experimental manipulations, why bias is generally the barrier to causal inference, and different methods to overcome this bias. Satellite‐, survey‐ and citizen–science data hold great potential for advancing key questions in ecology that would otherwise be prohibitive to pursue experimentally. However, to harness such data to understand causal impacts of land, environmental and policy changes, we must expand our toolset such that we can improve inference and more confidently advance ecological understanding and science‐informed policy.