35.6 Choosing a Matching Strategy

35.6.1 Based on Estimand

  • ATE: Use IPTW or full matching
  • ATT:
    • If many controls (Nc>3Nt): k:1 nearest neighbor without replacement
    • If few controls: subclassification, full matching, or odds weighting

35.6.2 Based on Diagnostics

  • If balanced: proceed with regression on matched samples
  • If imbalance on few covariates: Mahalanobis matching on those
  • If imbalance on many covariates: Try k:1 matching with replacement

35.6.3 Selection Criteria

  • Minimize standardized differences across many covariates
  • Especially prioritize prognostic covariates
  • Minimize number of covariates with large (>0.25) imbalance (Diamond and Sekhon 2013)

Matching is not one-size-fits-all. Choose methods based on the target estimand, data structure, and diagnostic results.

References

Diamond, Alexis, and Jasjeet S Sekhon. 2013. “Genetic Matching for Estimating Causal Effects: A General Multivariate Matching Method for Achieving Balance in Observational Studies.” Review of Economics and Statistics 95 (3): 932–45.