Pooled Wald statistic for multiply imputed sets

ccoman — Sat, 01 Aug 2020 14:38:18 GMT

Hi everyone,

I have generated m = 30 multiply imputed data sets due to missing data using proc MI. Then, I used proc LOGISTIC to perform uni- and multivariabele logistic regression by each imputation set. Then I use proc MIANALYZE to pool parameter estimates and odds ratios from all 30 MI-sets. This all works fine, however, I just can't seem to find a way to pool the Wald test's p-values for the covariate(s). Currently, I have 30 separate Wald chi-squared P-values for every imputation set. Is there an easy way to pool/combine these P-values into one overall p-value for each covariate? I read about Rubin's rule and Fisher's method to combine P-values, but this goes way over my head. Anyone who might have any advice?

Thanks for your help!

Re: Pooled Wald statistic for multiply imputed sets

SteveDenham — Mon, 03 Aug 2020 12:31:38 GMT

Take a look at the example here , which shows how to combine the imputations into a pooled estimate. Note that the results contain a probt value for each covariate. This is the equivalent of your pooled Wald tests in each of the BY _imputation runs of PROC LOGISTIC.

SteveDenham

topic Pooled Wald statistic for multiply imputed sets in Statistical Procedures

Pooled Wald statistic for multiply imputed sets

Re: Pooled Wald statistic for multiply imputed sets