model lane
DAGs and Causal Inference
A project integrating directed acyclic graphs (DAGs) with conventional econometric techniques, covering collider bias, the front-door criterion, M-bias, and falsification testing.
pipeline
- data (not covered)
- model (covered)
- eval (covered)
- serve (not covered)
- monitor (not covered)
EconometricsCausal inferenceDAGsOLSInstrumental variablesMonte Carlo
linksThis project uses Monte Carlo simulations to show where a standard regression goes wrong, like when you adjust for a collider or run into M-bias, and how reasoning with a DAG helps you choose which variables to adjust for.
It also covers falsification testing, where you check a DAG against the data by evaluating the conditional independencies that the DAG implies.
The Spot the Confounder game in the lab is built on the same ideas.