Hi Larouche,

If it is possible to do imputation separately then I would suggest to use R for Multivariate Imputation. The link to R's package "mice: Multivariate Imputation by Chained Equations" is below:

https://cran.r-project.org/web/packages/mice/index.html

I hope it will work.

Best of luck

Dear Firdos and Levente,

Thanks so much for your suggestions! Learning R is definitely on my to-do list, but for this project we would like to have results to share soon because the final report to our sponsor is due in July. So if I can implement Rubin's formula in excel I think this could do the trick. Would you be able to orient me to a good source for this formula?

Best regards,

Richard

Thanks Levente!

It is indeed a very tedious process to perform the analysis individually for each individual datasets and pool the estimates after. I wonder if an alternative solution might be to compute the mean of the multiply imputed values (e.g., n = 20 for 20 imputed datasets) into a single imputed dataset, and then running the models. Then we would not have to run each model 20 times. Any thoughts?

Thanks Levente!

In the meantime, I discovered another problem with the SPSS multiple imputation procedure: it does not handle multinomial and ordered categorical variables well enough. For example, I had created dummy variables for my 3-level parental education variable and, for some of the participants, SPSS imputed both "less than high school" and "college" as highest level of education which is a problem as the categories should be mutually exclusive. Because of all these issues, we will be using R instead.