8.7 Propensity Score Matching

  • Propensity score: Probability of receiving the treatment given the observed covariates
  • Propensity score matching developed as part of Rubin causal model (Wikipedia contributors 2016)
  • Criticized by LaLonde (1986), defended by Dehejia and Wahba (1999) and critisized again by King and Nielsen (2015):
    • Presentation in the ‘international’ methods colloquium: “We show that propensity score matching (PSM), an enormously popular method of preprocessing data for causal inference, often accomplishes the opposite of its intended goal – increasing imbalance, inefficiency, model dependence, and bias.”
    • Small summary here: Same propensity score ≠ Same covariate values; Prop. score matching mimics completeley randomised experiment vs. other matching methods that mimic blocked/stratified randomised experiments
  • Nonparametric matching methods
    • Genetic Matching (Diamond and Sekhon 2013)41
  • New interesting developments… Generalized Full Matching

References

Dehejia, Rajeev H, and Sadek Wahba. 1999. “Causal Effects in Nonexperimental Studies: Reevaluating the Evaluation of Training Programs.” J. Am. Stat. Assoc. 94 (448): 1053–62.

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

King, G, and R Nielsen. 2015. “Why Propensity Scores Should Not Be Used for Matching.” Copy at Http://J. Mp/1sexgVw Export BibTex.

LaLonde, R J. 1986. “Evaluating the Econometric Evaluations of Training Programs with Experimental Data.” Am. Econ. Rev.

Wikipedia contributors. 2016. “Matching (Statistics).” https://en.wikipedia.org/w/index.php?title=Matching_(statistics)&oldid=747816352.


  1. “Our method, Genetic Matching (GenMatch), eliminates the need to manually and iteratively check the propensity score. GenMatch uses a search algorithm to iteratively check and improve covariate balance, and it is a generalization of propensity score and Mahalanobis Distance (MD) matching (Rosenbaum and Rubin 1985). It is a multivariate matching method that uses an evolutionary search algorithm developed by Mebane and Sekhon (1998; Sekhon and Mebane 1998) to maximize the balance of observed covariates across matched treated and control units.” (Diamond and Sekhon 2013, 2)