Risk Ratio (or Prevalence Ratio) in SAS Proc Glimmix Is it possible to obtain  risk ratio in proc glimmix. And  if yes, how do I specify the base. E.g base 'male' in variable 'gender'.
Below is a template of my model:
proc glimmix data = mydata method=laplace;
class  wave gender race city education;                                     
weight wgt;
model response = gender age  race education city
/ dist=poisson link=log;
random intercept / subject = wave type=vc;
run;

 A: Hmm... I'm a bit confused about the random effect by person in these data. Observing more than one event in a time frame makes estimation of risk confusing whereas it's much safer to call it a rate ratio. Intuitively, think about the differences in risk ratios versus rate ratios for flare ups of herpes among herpes simplex adolescents comparing groups differing an average consumption of one alcoholic beverage per week.
If you are interested in rate ratios, you would use for each obseration (one per person level) an offset of the log of time they were observed at risk. This has the impact of creating a denominator of units of time in the modeled outcome, hence the fitted values are considered events per unit time. This should remove the need for use of mixed effects.
If you are indeed interested in risk ratios, you may want to only consider a maximum of 1 event per individual, usually the first to be observed, as the outcome (still using, of course, an offset to standardize the modeled outcomes to units of time).
If you are interested in prevalence ratios, you must define a particular cross-sectional time, such as "1 year after infection", or "in 2005", or "at age 30" and evaluate yes/no indicators of whether or not an individual had an event. Because it is a prevalence ratio you would want to change the specifications of the GLM slightly so that the modeled outcome is a log (hence coefficients are log-relative-prevalences), however, you need a binomial variance since a person either does or does not have the event of interest. This is something like dist=binomial; link=log. 
Prevalence ratios are usually estimated because they are the "best" you can do with cross sectional data. For longitudinal data, you omit repeated measures. Estimating prevalence from incidence is sometimes very interesting. Ron Brookmeyer at UCLA does great research on the stuff.
