Using Multiple Imputations in Propensity Score Matching

[Ishmam Al-Quddus]

Hello everyone,

I want to see the impact of a treatment, hence, using the propensity score matching method. However, I have missing data in my data set. I have imputed the data set. Now I am struggling with the integration of Propensity Score Matching using the imputed data as well. Please, I will be grateful if you could help me out. I am using STATA 13.0.

[Rich Goldstein]

the error message sends you to a list of "approved" estimation commands

but, in addition, there is an option to the -mi estimate- command called "cmdok"; for more, look at

help mi estimate

you don't say what you did to get to this stage but a "warning" - propensity scores should not use outcome variables but -mi- should; for advice on what to do in this situation, see Leyrat, C, et al. (2019), "Propensity score analysis with partially observed covariates: how should multiple imputation be used?", Statistical Methods in Medical Research, 28(1): 3-19

[David Radwin]

Another approach, which doesn't require imputation at all, is to match on missing values. See, for example,
Hansen, B. B. (2004). Full matching in an observational study of coaching for the SAT. Journal of the American Statistical Association, 99(467), 609-618.