Proximal OPE

Proximal methods use proxy variables to identify causal effects when standard ignorability fails. They require strong assumptions about proxy quality and bridge functions.

Assumptions

• Proxy variables are correlated with the unobserved confounder.
• Bridge functions are identifiable and well-posed.
• Sufficient variability to solve the moment conditions.

Practical guidance

• Use ProximalPolicyValueEstimand to make proximal assumptions explicit.
• Inspect bridge fit diagnostics (MSE, conditioning) before trusting estimates.

Limitations

• Sensitive to proxy misspecification.
• Often requires careful feature engineering.