GRADE-OF-MEMBERSHIP TECHNIQUES FOR STUDYING COMPLEX EVENT HISTORY PROCESSES WITH UNOBSERVED COVARIATES.

The analysis of multivariate event history data with both observed and unobserved covariates usually requires strong parametric assumptions about the distribution of unobserved covariates affecting the process and the nature of the functions describing the time dependency of the transition rates. In social science applications, there is frequently inadequate theory to specify the parametric form of either the mixing distribution or the hazard fund ion. In this paper, we present an analytic strategy to describe such processes in very general terms, without having to make either type of parametric specification. The use of "fuzzy set" logic to describe the complex multivariate state for individuals enables us to make estimates without such specification. The illustrate the model, we analyze data on the use of long-term care and residential services from a large California study population followed for 4.5 years. The analysis illustrates the model`s basic structure, its ability to describe very general types of multivariate discrete-state/discrete time stochastic processes, and its ability to deal with analytic problems like dependent competing risks, missing data, and the confounding of experimental conditions in a nonrandomized selection of cases and controls. [ABSTRACT FROM AUTHOR]