ABSTRACT
This chapter considers a class of state space models for handling multivariate observations at unequally spaced times. Also included is the situation where some of the multivariate observations are missing. For example, if three variables are measured at each observation time, sometimes one or two of the three variables may be unobserved. The use of a state space representation allows the modelling of various within subject error structures. Assuming Gaussian errors, the Kalman filter can be used to calculate the exact likelihood for given values of the nonlinear parameters. A nonlinear optimization program can be used to obtain maximum likelihood estimates of the parameters. The state space approach makes predictions for individuals straight forward. Once the parameters of the model have been estimated using the data from all the subjects, the state space model can be used to make subject specific prediction.
