ABSTRACT
In time series analysis, the word realization is used to describe independent time series with the same correlation structure. In longitudinal data analysis, different subjects are usually assumed to generate data with errors that are independent realizations of the same correlation structure. The correlation function for an AR(1) error process, and three independent simulated sets of serially correlated errors taken at unequally spaced time points. The horizontal line in each simulation is the true mean of the observations. When observations can be taken on subjects at arbitrary time points, there must be an underlying continuous time process. Suppose observations are taken every four hours on a group of subjects, and it is of interest to know whether there is a fluctuation in the mean level with a period of 24 hours. There are several possible ways to formulate this problem.
