Small Samples and Propensity Score Methods

• Activity Number: 670
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Ana Kolar*+ and Vasja Vehovar
• Companies: University of Ljubljana and University of Ljubljana
• Keywords: propensity score methods ; small samples ; causal inference

Abstract

Propensity score (PS) methods for estimating causal effects are becoming increasingly important for observational study designs. The development of PS methods over the past three decades also resulted in guidance on the process of estimating causal effects. However, this guidance is well established only for the case of large data sets, whereas the question of "how large" data sets should be remains for the most part unanswered. Particularly in social science and medical research, where we often face relatively small samples, minimum sample size requirements are not clear. The paper presents results of an intense simulation study investigating small sample properties of PS methods. The results show that the required ratio of control to treatment units - group ratio, strongly depends on the treatment group's sample size. For treatment samples consisting of only 8 units, a seven times bigger group ratio is required, in comparison to moderately large samples. At the same time, when the number of covariates is increasing, the needed group ratio for small samples increases much faster than the needed group ratio for moderately large samples.