proc glimmix with continuous outcome variable; unbalanced dataset

[kayec]

Hi, I am a relatively new to SAS but am trying to run a GLIMMIX model to take account of the flexible family (link) options not available in R....

I have a longitudinal dataset of ~98,000 observations on ~30,000 individuals (user). The data is unbalanced as some individuals only have 1 observation and some have many over the period of observation.

The outcome variable is continuous and non-normal (A). There is up to 6 independent variables which are categorical or continuous (eg X).

The code I am running is

proc glimmix data=work.P3 ;

model A = X /

s dist=expo link=log ;

random intercept / subject=user ;

run ;

When I run this base model I get the following issues:

ERROR: Integer overflow on computing amount of memory required.

ERROR: The SAS System stopped processing this step because of insufficient memory

Can someone advise what I should do to overcome this in SAS.

Many thanks in advance.

[SteveDenham]

My first suggestion would be to Google Liang Xie and Larry Madden's paper "%HPGLIMMIX: A High Performance SAS Macro for GLMM Estimation,"  Journal of Statistical Software, 58(8).  A copy of the macro can be obtained at http://www.jstatsoft.org/v58/i08.

My second suggestion is to check the available memory for your instance of SAS.  You may have to step up the memsize= global option.

If neither of these work, you may need to sample and bootstrap your estimates in a model-averaging context.

Steve Denham