multiple regression/Johnson Neyman output with data that has had multiple imputation

[Charlotte Davis]

Hi,

I am responding to a reviewer who wanted me to rerun an analysis using multiple imputation on MCAR data. There was 16% data lost and although there is little change to the outcome using imputed data, but I now feel I have to impute the data. My original analysis was in SPSS and reported Johnson-Neyman probe following an interaction. Multiple imputed data does not won't work using Hayes' Process macro so...

I had read that I could run a moderation analysis with Johnson-Neyman output using Stata using the mi estimate command. However, having had a try, I find margins is not supported with mi estimate. Can anyone help with the commands?

It's a behavioural experiment. The analysis involves variable task heart rate response to film(0/1) at levels of self-control with baseline heart rate as a covariate.

Just to have it confirmed whether I can get this done more quickly in Stata than I can write up a justification for not imputing data at all would be really helpful. However, guidance on commands and syntax would, of course, be brilliant!

Thanks,

Charlotte

[Andrew Musau]

You may want to check out mimrgns from SSC, by Daniel Klein.

ssc install mimrgns, replace
help mimrgns

[Charlotte Davis]

Thank you Andrew, that's really helpful. Though I can see from a previous discussion, it's going to be problematic with an interaction. But still, really useful to know what can't be done...

Thank you.