using a different residual variance structure in proc mixed

[mamzolo]

Dear all,

I'm currently fitting a mixed effects model in SAS using proc mixed. The only issue I have is that I have a special data set in a sense that the observed data come with a variation measure.

What I want to do, is to incorporate this additional variation in the estimation of the residual variance. Thus, instead of having a residual variance of the form sigma^2*I, I have (sigma^2 + S2)*I where S2 is the variation measure.

Any ideas of how I can account for this structure in SAS?

A sample of the model in SAS is of this form:

proc mixed data=data;

class run;

model y=x;

random intercept/sub=run;

repeated / ;                                              /*I believe the variation S2 should come in here*/

run;

[SteveDenham]

This is going to be a lot easier if you have all of the data that generated this additional variation measure.  It would then be a straightforward MIXED analysis, I think.  But I am guessing not...

So, some more detail on how S2 is derived and then on how it is supplied to the dataset are going to be very helpful.  I am hoping that S2 is some sort of variance calculated on a per run basis, and if so, we can address it through the use of a WEIGHT statement.

Steve Denham

[mamzolo]

Thanks Steve.

Indeed S2 is a variance calculated for each run.

I have just been able to solve this problem (partly) by using the PARMS statement. I had to create a data set with S2 values and use this data in PARMSDATA option.

I haven't checked whether it works but I'm keeping my fingers crossed.

Thembile Mzolo

[SteveDenham · Accepted solution]

I think that is a good plan--I assume these are read in and then the HOLD option is used to keep them fixed at the calculated values?

Steve Denham