Longitudinal data consist of time-sequences of measurements, counts or categorical responses taken from one or more experimental units or subjects. For example, in a medical study the subjects would be patients, and the measurements on each subject might correspond to some measure of their health status at hourly intervals, such as systolic blood pressure (a continuous measurement), or whether or not the patient is in pain (a categorical response). Longitudinal data are therefore closely related to time series data. However, longitudinal studies differ from most classical time series studies in two respects. Firstly, as in the above hypothetical medical example, measurements are typically made on each of a number of subjects who can be regarded as a sample from some underlying population. Secondly, covariate information is usually available, either as explicit measured covariates on each measurement or on each subject, or through the allocation of subjects to two or more distinct treatments in a designed experiment. Furthermore, the inferential focus in longitudinal studies is usually on the mean response, and how this varies with time, treatment or other covariate information. The correlation structure of longitudinal data is usually of secondary interest, but must be accommodated in the analysis of the data to ensure valid inferences about the structure of the mean response.