Detailed Notes

1. psmatch2 (Leuven and Sianesi, SSC)

• Use: psmatch2 treatment varlist, out(matched) method(knn) caliper(0.1)

• Pros: Highly flexible, supports many matching options (nearest neighbor, kernel, caliper).

• Cons: No direct support for estimating ATE; standard errors must be bootstrapped.

2. psmatch (Built-in)

• Use: psmatch treatment, neighbor(1)

• Pros: Simpler syntax; works well for quick ATT estimation.

• Cons: Limited to nearest-neighbor matching; less flexible than psmatch2.

3. teffects (Built-in)

• Use:

  • teffects psmatch — Matching

  • teffects ipw — Inverse Probability Weighting

  • teffects ra — Regression Adjustment

  • teffects aipw — Augmented IPW

• Pros: Direct estimation of ATE, ATT, ATU; robust to some model misspecification; good diagnostics.

• Cons: More parametric assumptions (especially RA and AIPW).

4. ipw (User-written)

• Use: ipw treatment x1 x2 x3, pscore(ps) logit

• Pros: Focused on reweighting; can be followed by outcome models using svy commands.

• Cons: Requires user management of weights and modeling.

5. Entropy Balancing (ebalance from SSC or tebalance from reghdfe)

• Use: ebalance treat x1 x2 x3 or via tebalance ebalance

• Pros: Exactly balances covariate moments (mean, variance, skewness); ideal for pre-processing.

• Cons: Doesn’t perform matching; must be followed by outcome modeling (e.g., OLS on weighted sample).