Abstract Context The recent decline in biodiversity, which is of particular concern in agriculturally managed landscapes, pushes researchers and practitioners to evaluate potential solutions, such as agricultural diversification. This requires the ability to provide robust conclusions about causal relationships in agroecosystems, which is hindered by limitations of conventional research methods. Findings from controlled plot experiments do not always translate to real-world conditions, while ana