-rctmiss- and relative risks

[Laura Myles]

Hi Listers,

I am analysing data from an RCT with a binary outcome.

In my primary analysis I used multiple imputation to estimate missing data; however, I would like to run some sensitivity analyses to explore possible departures from the MAR assumption. I appreciate that -rctmiss- supports logistic regression but I was wondering if there is any way to extrapolate relative risks, which I report in the primary analysis.

I thought I could use poisson (since robust SE are estimated) and then exponentiate the estimates, but I am not 100% sure this is OK within -rctmiss-.

Any advice welcome!

[Rich Goldstein]

here is a quote from the help file for -rctmiss-: "At present, estimation_command must use regress, logistic / logit or poisson." so, yes, it appears to be acceptable to use -poisson- within -rctmiss-

[Laura Myles]

Thanks Rich Goldstein - I wasn't sure if the confidence intervals would be accurate; but I suppose since robust SE are estimated it should be OK.