iv-estimation

Use IV and 2SLS estimation to identify causal effects with endogenous treatments. This approach requires valid instruments and exogenous controls, providing first-stage diagnostics like F-statistics and overidentification tests for robust reporting.

Apply propensity score matching when selection on observables is credible. PSM estimates ATT through nearest-neighbor matching and balance checks, whereas 2SLS is necessary when unobserved confounding causes endogeneity.

Valid 2SLS requires first-stage F-statistics to detect weak instruments, Sargan or Hansen tests for overidentification restrictions, and Wu-Hausman tests to confirm endogeneity, ensuring robust causal identification and reporting.

IV estimation identifies the Local Average Treatment Effect for compliers affected by the instrument, not the Average Treatment Effect. Proper interpretation requires first-stage diagnostics and overidentification tests to validate the LATE estimate.

IV 2SLS and propensity score matching workflows support Python, R, and Stata implementations. Provide an instrumented dataset with outcome, treatment, control, and instrument variables to obtain causal estimates and diagnostics.

Check for weak instruments using the first-stage F-statistic during 2SLS estimation. A low F-statistic indicates weak instruments that compromise causal identification, requiring exogenous controls and valid instruments for robust results.