Estimating Power and MDES for Propensity Score Model

[Bob Fitzgerald]

I have been asked by a proposal reviewer to calculate power and MDES for a propensity score analysis for key outcome variables. I have searched via Google and in the archives of this list but cannot find any references on the "correct" approach. A number of authors suggest that matching on the propensity score is similar to blocking in a randomized trial, and hence PSM has greater power than a simple test might indicate but there does not appear to be general agreement on an approach.

Would appreciate any references or hearing about personal experiences from other researchers facing a similar requirement from a funder. I use Stata 13.1, so am particulary interested in learning about solutions based on the Power and Sampling Size module.

Thanks for your attention.

Bob

[Paul Jasper]

Hi Bob,

I am in a similar situation, and looking for a way of how to do power calculations and estimate MDES for propensity score matching approaches. Doing some research via Google, I have come across this blog post: http://blogs.worldbank.org/impacteva...score-matching, and this article: http://www.tandfonline.com/doi/abs/1...044790#preview, but that's pretty much it.

Have you had any responses to your earlier post? Also, would you mind sharing some of the references you mention above?

Thanks and kind regards,

Paul

[Bob Fitzgerald]

Paul,

I did not find an authoritative reply. Though several folks (Elizabeth Stuart among others) agreed it's an important issue, no one could point to established precedent. Among the citations you may find useful are Stuart, E. 2010. “Matching Methods for Causal Inference: A Review and a Look Forward.” Statistical Science. Vol. 25:1–21, and Ho, D., Imai, K., King, G. and Stuart, E. 2007. “Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference. “ Political Analysis 15: 199-236.
Please do let me know if you make any progress or come upon promising leads.

Bob