Continuous-time Microsimulation in Longitudinal Analysis

Abstract

In longitudinal analysis, individuals are followed in time and are observed either continuously or at points in time. The time to event, the sequence of events and the factors that influence timing and sequence constitute the object of study. Time is a continuous variable (exact time) or discrete variable (time interval). Longitudinal data are used to estimate parameters of event history or life history models. Individual life histories can be represented by sequences of states and sequences of events, that are transitions between states, and described by multistate transition models. The parameters of these models are transition intensities when time is represented by a continuous variable and transition probabilities when time is discrete (for details, see Willekens, 2001). Microsimulation contributes to longitudinal data analysis in a number of ways (Wolf, 1986, 2001). First, it generates individual event histories that are fully consistent with a set of transition intensities (probabilities). Second, it produces estimates of the full distribution of an outcome, in addition to the expected value that is produced analytically by most models. Third, it is helpful in examining the potential seriousness of defective data. Fourth, it may play a role in the imputation of missing