topic Re: Propensity score match - multiple imputation in Statistical Procedures

Propensity score match - multiple imputation

bollibompa — Tue, 16 Jun 2015 10:05:49 GMT

Hi,

I have a dataset with some missing information and have performed av PROC MI generating 5 datasets and further estimated p-score for these. Are there any way of pooling/avaraging the p-score from these 5 datasets and match to a control using the avaraged p-score? Any procedure in SAS that could handle this?

Thanks

/Thomas

Re: Propensity score match - multiple imputation

Rick_SAS — Tue, 16 Jun 2015 12:39:56 GMT

An average p-value doesn't make sense. The p-values give information about whether the data support the null hypothesis. For each set of data, you run an analysis, generate parameters, and you get p-values, but those p-values are conditional on the data.

You could report the distribution of the p-values or you could make a statement like "for each of the imputed data sets, the p-value for XYZ was significant."

Re: Propensity score match - multiple imputation

SteveDenham — Tue, 16 Jun 2015 16:34:20 GMT

Multiple imputation almost always involves two steps--first imputing the missing values (PROC MI) and then combining the results in a meaningful way (PROC MIANALYZE). I suspect that if you look through the documentation for PROC MIANALYZE, especially the examples, you will find precisely what you are looking for.

Steve Denham