topic Re: Standard Deviation from Proc Glimmix in Statistical Procedures

Standard Deviation from Proc Glimmix

mekono — Fri, 22 Jan 2021 14:34:20 GMT

I'm using Proc Glimmix to get the adjusted means cost for my research question. The output gives the standard error estimate but is there a direct was to get the standard deviation from the model? I know I can calculate it based on the standard error but the deviations are looking larger than expected.

proc glimmix data= final_population;
class patient_group survival gender (ref="0") daibetes (ref="0") smoking (ref="0") obesity (ref="0") ;
model costs=age gender daibetes smoking obesity propensity_score / dist=binary link=logit solution;
random intercept / subject=patient_group;

lsmeans patient_group / pdiff ilink oddsratio cl ;
run;

Re: Standard Deviation from Proc Glimmix

jiltao — Fri, 22 Jan 2021 15:09:00 GMT

Standard error is the statistic to measure the uncertainty of the parameter estimate. Standard deviation is not relevant here and might not be available from PROC GLIMMIX..

Re: Standard Deviation from Proc Glimmix

mekono — Fri, 22 Jan 2021 15:49:13 GMT

Thanks for your reply. That makes sense. What would be the best way to get
the covariate adjusted mean and standard deviation for a model?

Re: Standard Deviation from Proc Glimmix

SteveDenham — Mon, 25 Jan 2021 14:02:45 GMT

As @jiltao said (and you really, really need to listen to what Jill says), the standard error is the proper measure of dispersion for both the parameters of the model and for the covariate adjusted mean (least squares mean). I suspect you are looking for the standard deviation so that you can calculate some sort of scaled effect size. The usual methods (Cohen's d, eta squared, omega squared) aren't appropriate in the mixed model setting. Search this community for other methods, one of which was posted in the last month. It involves looking at the change in the response to a unit change (unit here could be a standard deviation of the raw input) while holding all other factors in the model fixed at their respective means.

SteveDenham