Power analysis for 3-level models

[emdags]

Wondering if anyone can share helpful references for conducting power analyses for three-level models (preferably including longitudinal models). Thank you!

[Rick_SAS]

The main procedure for analysis of power for GLM-type models is the GLMPOWER procedure. It enables you to compute power and sample sizes for ANOVA models, repeated ANOVA models, GLM models with continuous covariates, and more. It uses the same syntax as the GLM procedure in SAS.

I am not an expert at using this procedure, but there is an example of a repeated measures ANOVA analysis in the example section of the doc.

[emdags]

Thank you! I should have specified that I am looking for 3-level power analyses for longitudinal multilevel models that generate individual level estimates (not population averaged means) and using the MIXED (for linear outcomes) or GLIMMIX (for binary/poisson distribution outcomes) procedures.

[Rick_SAS]

I see. In that case, I refer you to a previous thread about power analysis for mixed models that discusses two books and a website that might be useful for you. 

[Rick_SAS · Accepted solution]

That thread is from 2013, so I also want to provide some newer info:

- support for repeated measures and also summarizes how much of the mixed model realm that PROC GLMPOWER supports 

- a discussion of power for generalized linear models with a discrete response 

In general, GLMPOWER is suitable if you omit random effects and assume no missing data. I don't have further recommendations for procedures that apply to your case study. If you are motivated and a good programmer, you can obtain empirical estimates through simulation or use SAS/IML to implement the special formulas mentioned in the R documentation/code that you link to.

[emdags]

Thank you so much! The first source on repeated measures was particularly helpful.