Making Progress on Causal Inference in Economics
Enormous progress has been made on causal inference and modeling in areas outside of economics. We now have a full semantics for causality in a number of empirically relevant situations. This semantics is provided by causal graphs and allows provable precise formulation of causal relations and testable deductions from them. The semantics also allows provable rules for sufficient and biasing covariate adjustment and algorithms for deducing causal structure from data. I outline these developments, show how they describe three basic kinds of causal inference situations that standard multiple regression practice in econometrics frequently gets wrong, and show how these errors can be remedied. I also show that instrumental variables, despite claims to the contrary, do not solve these potential errors and are subject to the same morals. I argue both from the logic of elemental causal situations and from simulated data with nice statistical properties and known causal models. I apply these general points to a reanalysis of the Sachs and Warner model and data on resource abundance and growth. I finish with open potentially fruitful questions.