proc mixed model yielding equal standard errors across groups

[Jest]

Hi,

I am running a mixed model that has 4 groups with various blood levels and comparing each of the three groups to one control. For some reason when I include all the groups in one proc mixed model, with the control as my reference group, the solution for fixed effects yields equal standard errors for each group. However, when I run separate mixed models including only the control and one group, I get different standard errors. Is there any reason for this?

This is the code I am using that includes all the groups, 0, 1, 2, 3,4 and 0 (control as reference).

PROC MIXED data=data.data COVTEST NOITPRINT;

CLASS ID;

CLASS Group (ref="0");

MODEL level = Day Group/SOLUTION ddfm=bw;

repeated /subject = ID;

lsmeans Group / pdiff; run;quit;

results look something like the below, is there any reason for the equal standard errors. This is the first time I encounter this....Is it better to run separate mixed models instead of one model with all the groups.

Solution for Fixed Effects
Effect	Group	Estimate	Standard Error	DF	t Value	Pr > |t|
Intercept		3.4658	0.7105	8	4.88	0.0012
Day		0.2978	0.05967	47	4.99	<.0001
Group	2	-5.7163	0.8129	8	-4.45	0.0031
Group	3	-4.8571	0.8129	8	-4.73	0.0065
Group	4	-1.4241	0.8129	8	-0.53	0.7088
Group	1	0	.	.	.

[StatsMan]

Your model as written is fitting a single common residual variance for all observations in your input data.  You could try adding GROUP=GROUP as an option to the REPEATED statement to fit separate residuals for each level of your GROUP variable.  

That's essentially what is happening when you run the model separately for each group.