New Evidence on the Finite Sample Properties of Propensity Score Matching and Reweighting Estimators

By Matias Busso, John E. Dinardo, J. McCrary. Social Science Research Network, 2009

164 citations recorded by the source corpus.

Currently available asymptotic results in the literature suggest that matching estimators have higher variance than reweighting estimators. The extant literature comparing the finite sample properties of matching to specific reweighting estimators, however, has concluded that reweighting performs far worse than even the simplest matching estimator. We resolve this puzzle. We show that the findings from the finite sample analyses are not inconsistent with asymptotic analysis, but are very specific to particular choices regarding the implementation of reweighting, and fail to generalize to settings likely to be encountered in actual empirical practice. In the DGPs studied here, reweighting typically outperforms propensity score matching.

Subfields: Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/2c28e061342962bba586b3f0e0501664979c7129

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Statistics and Probability

Citations in the corpus

Estimation of Multivalued Treatment Effects under Conditional Independence (2013, methodology): “Having selected models for both the predicted probabilities and the conditional mean, we now use poparms to estimate the means of the potential-outcome distributions for each value of w in the sample.”
Oaxaca-Blinder as a Reweighting Estimator (2011, background): “See for example Busso, Dinardo, and McCrary (2009).”
Bias-Corrected Matching Estimators for Average Treatment Effects (2011, background): “For other Monte Carlo studies of matching type estimators see Zhao (2002), Frölich (2004), and Busso, DiNardo, and McCrary (2009).”
What Do Emissions Markets Deliver and to Whom? Evidence from Southern California's Nox Trading Program (2012, background): “Recent work has demonstrated that, when there is good overlap in the distribution of propensity scores for treated and control units, reweighting estimators outperform nearest neighbor or kernel matching in 26 finite samples (Busso et al., 2009).”
The Finite Sample Performance of Inference Methods for Propensity Score Matching and Weighting Estimators (2018, methodology): “Radius matching (see, for instance, Rosenbaum and Rubin 1985; Dehejia and Wahba 1999) uses all nontreated observations with propensity scores within a predefined radius around that of the treated reference unit, which trades off some bias to increase efficiency.”
Radius matching on the propensity score with bias adjustment: tuning parameters and finite sample behaviour (2014, methodology): “Previous simulation studies on propensity score methods (Frölich 2004; Busso et al. 2009a, b) do not even include radius matching.”
The Effectiveness of Public-Sponsored Training Revisited: The Importance of Data and Methodological Choices (2014, unknown): “Third, optimized versions of both kernel or radius matching improve upon simple implementations, unless sample sizes are really small, and they typically do not perform badly even when inverse probability weighting ðIPW, the preferred method in Busso et al. ½2009, 2011 Þ performs better.”
IDENTIFYING CAUSAL MECHANISMS (PRIMARILY) BASED ON INVERSE PROBABILITY WEIGHTING (2014, methodology): “…results in Propositions 1–4, such that the weights of the observations in either treatment state add up to unity, as advocated in Imbens (2004) and Busso et al. (2009b).10 For example, the normalized estimators of the direct effects under treatment and non-treatment are given by ŷ 1ð Þ ¼ P Yi…”