ABSTRACT
Multiple time series on independent units occur routinely in biostatistics and econometrics, among other disciplines. In the classical longitudinal study, multivariate quantitative outcomes may be collected over time on multiple subjects (living organisms) and their contemporaneously observed explanatory variables or covariates. This chapter describes the cross-sectional extension that is required in the classical state-space (SS) formulation. It discusses appropriate Kalman smoothing, filtering, and prediction recursions, as well as maximum likelihood estimation. The linear mixed model is perhaps the most frequently employed type of random effects model in longitudinal data. This is sometimes also called the Laird-Ware model and it has a SS representation. The chapter discusses methods for handling missingness in responses only, followed by two methods for handling missingness in both responses and covariates.
