Drawing Credible Directed Acyclic Graphs for Causal Inference

Causal directed acyclic graphs (DAGs) are intelligible representations of real-world data-generating processes that facilitate causal inference by providing (automatized) guidance for assessing whether a causal effect is identified with the observed data and for selecting covariates that remove most, if not all confounding bias. However, less attention has been paid to the process of constructing causal DAGs. Methodological work often relies on toy examples that have limited practical utility for applied researchers working in complex contexts. This article introduces and demonstrates a stepwise, iterative procedure for drawing credible causal DAGs, which is designed to guide researchers in identifying important sources of confounding while also incorporating research design features of quasi-experiments or random experiments, as well as threats to validity (e.g., measurement error, treatment non-compliance). Although constructing a complete DAG that fully captures the data-generating process is difficult and rarely achievable in practice, we argue that developing a credible DAG—one that includes all plausible sources of confounding—is adequate for applied research. The proposed iterative drawing procedure is directly aligned with the goal of constructing credible causal DAGs.