topic Re: ANOVA with a random effect from clusters in Statistical Procedures

ANOVA with a random effect from clusters

fengyuwuzu — Fri, 28 Oct 2016 17:31:04 GMT

For clustered data (multiple test sites from the same subject), if I want to do one way ANOVA to compare the baseline values in each group, is the followint the right way to add the random effect from clustering in the ANOVA model ?

proc glm 
  data=mydata;
  class treatment subjid;
  model change = treatment;
  random subjid;
run;

Re: ANOVA with a random effect from clusters

lvm — Sun, 30 Oct 2016 15:58:29 GMT

You really (REALLY) don't want to use GLM for any situation with random effects. This was OK before procedures like MIXED and GLIMMIX were written, but GLM should be avoided now. For your example, use:

proc mixed data=mydata;
  class treatment subjid;
  model change = treatment;  
  random subjid;
lsmeans treatment / diff;
run;