Testing Departures from Time Homogeneity in Multistate Markov Processes

SUMMARY Many investigations are concerned with phenomena which evolve in continuous time but are observed only at some discrete points in time. Time homogeneous Markov models are often chosen to represent the processes underlying such observations. We propose two test statistics for examining the adequacy of such models. The first test assesses the overall goodness of fit while the second examines local departures in the direction of time inhomogeneity. Data on transitions between two states of disease before death are analysed. They concern breast cancer patients affected by spinal metastases.