Model-based Clustering of Sequential Data with an Application to Contraceptive Use Dynamics

ABSTRACT

Multi-state models describe the transitions people experience as life unfolds. The transition probabilities depend on sex, age, and attributes of the person and the context. Empirical evidence suggests that attributes that cannot be measured directly may at most be inferred from a long list of observable characteristics. A cluster-based, discrete-time multi-state model is presented, where transition probabilities are estimated simultaneously for several subpopulations of a heterogeneous population. The subpopulations are not defined a priori but are determined on the basis of similarities in behavior in order to determine which women exhibit similar characteristics with respect to method choice, method switch, discontinuation and subsequent resumption of contraceptive use. The data are from the life history calendar based on the Brazilian Demographic and Health Survey 1996. The parameters of the model are estimated using the EM algorithm. Seven subpopulations with heterogeneous transition probabilities are identified.