The BGLIMM Procedure

A GLMM model has various types of parameters: fixed-effects parameters (in the MODEL statement), random-effects parameters (in the RANDOM statement), parameters in the covariance matrix, and missing data variables in the response.

For the fixed-effects parameters, PROC BGLIMM generates initial values that are based on the optimization of the posterior density functions. To assign initial values to the fixed-effects parameters, you can use the INIT= option in the MODEL statement. If there are multiple fixed effects, you can provide a list of numbers, where the length of the list is the same as the dimension of the fixed effects. Each number is then given to all corresponding fixed-effects parameters in order.

For all the individual random-effects parameters, PROC BGLIMM sets the initial values to zero.

The initial values for either the or matrix are set to the identity matrix.

For missing data in the response, PROC BGLIMM uses the sample average of the nonmissing values as the initial value. The procedure requires some response value to be nonmissing—if all values of a particular variable are missing, PROC BGLIMM issues an error and stops.