Adjusted methods (Backdoor criterion)

  • A set of approaches used to control for confounding and isolate the causal effect of an exposure on an outcome.
  • Based on the backdoor criterion, these methods require a valid adjustment set to block non-causal paths.
  • The key premise is that the causal graph is known or assumed.
  • These methods rely on adjusting for confounding variables using conditioning sets.

Overlap: Widely overlaps with Causal ML (which builds upon these), Causal Discovery (which may infer the graph), and Quasi-Experiments (when combined with temporal or spatial structure)

Key feature: Control for confounding through observed covariates

Usage: Widely used when experimental data is unavailable but good covariate data exists


Table of contents