The performance of estimators based on the propensity score

Abstract

We investigate the finite sample properties of a large number of estimators for the average treatment effect on the treated that are suitable when adjustment for observed covariates is required, like inverse probability weighting, kernel and other variants of matching, as well as different parametric models. The simulation design used is based on real data usually employed for the evaluation of labour market programmes in Germany. We vary several dimensions of the design that are of practical importance, like sample size, the type of the outcome variable, and aspects of the selection process. We find that trimming individual observations with too much weight as well as the choice of tuning parameters are important for all estimators. A conclusion from our simulations is that a particular radius matching estimator combined with regression performs best overall, in particular when robustness to misspecifications of the propensity score and different types of outcome variables is considered an important property.

A simple and successful shrinkage method for weighting estimators of treatment effects

“…The third one is shown to perform decently in Huber et al (2013). Note that the third rule restricts the weights, and not the propensity scores directly.…”

“…The choice of making the number of replications inversely proportional to the sample size is motivated by the fact that simulation noise depends negatively on the number of replications and positively on the variance of the estimators, which depends negatively on the chosen sample size. Hence, the simulation noise is constant if the Monte Carlo samples are chosen inversely proportional to the sample size (Huber et al, 2013). Our Monte Carlo study consists of three parts.…”

Leave the Drama on the Stage: The Effect of Cultural Participation on Health

“…Second, the set of control variables is collapsed into one single number, the propensity score, which reduces the problems associated with a highdimensional covariate vector (e.g. Huber et al, 2013). Third, matching allows assessing the comparability of treated and control individuals.…”

Are the Children of Uneducated Farmers Doubly Doomed? Farm, Nonfarm and Intergenerational Educational Mobility in Rural China

“…20 There is substantial monte-carlo evidence in favor of these estimators for estimating causal effects with non-experimental data. Busso et al (forthcoming) provide evidence that NIPW performs best among a large set of matching and propensity score estimators in estimating a binary treatment effect, and Huber et al (2013) provide extensive evidence from empirical Monte-carlo that the BC-RM estimator due to Lechner et al (2011) performs very well among a wide set of estimators. While NIPW reduce biases in the estimates compared to the OLS estimates by using appropriate weighting, the MB-NIPW estimator is especially useful, because it minimizes the biases arising from selection on unobservables.…”

“…9 This framework allows us to use appropriate subsamples as "treatment" and "comparison" groups to explore a set of important questions related to the role of family background in children's schooling using recently developed matching and propensity score weighted estimators (Millimet and Tchernis (2013), Huber et al (2013)). 10 First, we provide evidence of substantial heterogeneity.…”