Inference from censored Markov chains with applications to multiwave panel data (Q1115073)

We consider some problems of inference from censored discrete-time Markov chains. The censoring mechanism is general enough to be useful in applications and maintains the conditional multinomial sampling model for successive transitions. Procedures are given for nonparametric estimation of transition probabilities and tests of the intrinsic-time and time- homogeneity hypotheses. These procedures apply to multiwave panel data when continuous-time Markov chains are only periodically monitored. To illustrate this we build on a result of \textit{B. Singer} and \textit{J. E. Cohen} [Math. Bisoci. 49, 273-305 (1980; Zbl 0425.92011)] by showing that a nonhomogeneous Markov chain without an intrinsic-time is a reasonable model for transition data from a WHO malaria survey.