rjlimit warning not disappearing/smc fcs

[Farah Seedat]

Dear Prof Bartlett, 

I am a first time user of the smcfcs command in STATA and I was hoping you could help me with a problem.

I continuously receive the warning "valid imputed values may not be generated", even if I set the rjlimit to 100,000. The warning is only for one subject.

Overall I have 384 cases, with only 60 that are full observed (due to the nature of the predictors, which vary from 142-347 observed). 

I am imputing 5 continuous predictor variables as the outcome is fully observed and I am using smcfcs due to the non-linear relationships between predictors and response (although this is only based on 60 observations) as well as the inclusion of an interaction.

Is there a point at which it is safe to limit the rjlimit and no longer increase it?

I am happy to provide further information if necessary. 

Thanks very much for your time. 

Best wishes, 

Farah 

[Jonathan Bartlett]

Hi Farah

Thanks for the question and your interest in the package. If it is failing for just one subject out of 384, I would personally not worry about it. Even when it fails it still imputes the covariate(s) from their conditional distribution given the other covariates. Typically one subject's data (imputed or observed) has limited influence on final estimates, and so imputing a missing value from not quite the right conditional distribution probably won't cause a big problem. Of course this is not a rigorous argument, but is just a personal judgment.

Best wishes
Jonathan

[Farah Seedat]

Dear Jonathan, 

Thanks very much for your response and confirming it is okay! 

Best wishes, 

Farah