BLUP (best linear unbiased prediction) method for predicting random effect(s) has been widely used for many years in a number of areas, e.g. tree or animal breeding and hospital or school performance measurement. However there is no a simple way of specifying fixed effects to obtain the prediction and the 95% confidence interval for the cluster variable (i.e. random effect). My application requires prediction for binary and continuous outcome adjusted for the known difference among the clusters with a number of fixed effects.
LSMEAN statement provides OM and BYLEVEL options to obtain marginal means with procedure MIXED / GLIMMIX; however it doesn’t accommodate for random variable(s).
The ESTIMATE statement supports both fixed and random effects. If a fixed effect is not specified it employs the default values (0 applied to a numerical variable and the equal proportion to each level of a categorical variable); the predicted values may make sense for applications with 0 orientation and balanced experiments. In general users have to specify all effects to get marginal means for the random variable(s), and the job can be very tedious.
Therefore it is suggested that LSMEAN statement is expanded to accommodate for random variables. The alternative solution is to add OM and BYLEVEL options to the ESTIMATE statement.