Correlation or Causation? Identification! - Directed acyclic graphs as an identification framework in econometrics

I explore the merits and limitations of Pearl’s directed acyclic graphs (DAG) approach for identification in applied economics. Researchers rely on causal inference frameworks to derive causal effects from observational data. Econometricians use either the potential outcomes (PO) or the structural equation modelling (SEM) framework. I argue that Pearl’s structural causal modelling (SCM) framework based on DAGs offers certain advantages unattainable in the other two. DAGs’ main strengths are their accessibility, clarity, transparency, and testability, and the current scepticism by economists against their use is largely unfounded. The SCM framework is unsuited to entirely replace the current causal inference methodologies in econometrics, but has tremendous potential to support identification and counter endogeneity problems in complex causal settings and when working with big data. I revisit Fryer (2019) and Knox et al. (2020) as an example for a successful application of DAGs and add to Knox et al.’s critique of Fryer’s identification strategy