SAS Code: Joint Models for Continuous and Discrete Longitudinal Data

Since different distributions and link functions have to be used for the different outcomes, we use a special device available in the SAS procedure PROC GLIMMIX, i.e., the ‘byobs=(.)’ specification that can be used to specify both the distribution in the ‘dist=’ option and the link function in the ‘link=’ option. Thus, before we start with the main analysis, two variables need to be created to specify the outcome distribution and link function for each observation in the dataset. For example, assume ‘distvar’ is a variable denoting a Gaussian distribution for the continuous measurement and a Bernouilli one for the binary measurement. Then, the option ‘dist=byobs(distvar)’ specifies a normal distribution for the first measurement of every subject, and a Bernouilli distribution for the second one. The following code creates a distributional indicator for the continuous outcome ‘Pupil Size’ and the binary outcome ‘Toe Pinch’, as well as a link function indicator.