Life histories: real and synthetic

Frans Willekens, Max Planck Institute for Demographic Research

Life history data are generally incomplete. Respondents enter observation late (left truncation) or leave early (right censoring). In survival analysis, these limitations are considered in the estimation of hazard rates. Rates are estimated from data on different respondents with different observation periods (observation windows). In multistate modeling, transition rates also integrate information on different individuals. By combining data from different but similar individuals, life histories can be modeled. The life history that results is a synthetic life history. It is not observed and it does not tell anything about a particular individual. It tells something about the population the individual is part of. A synthetic biography summarizes information on several individuals. The collective experience is summarized in transition rates. The individual is a fictitious individual, referred to as virtual individual or statistical individual (Courgeau, 2012). A population of virtual individuals is a virtual population. The life history of such an individual is not directly observed but is an outcome of a probability model, the parameters of which are estimated from empirical data. Life histories are generated from models using microsimulation in continuous time. Several life course indicators may be derived from transition rates. They include probabilities of significant transitions, probabilities of having reached particular stages in life, expected durations of stages of life, and expected ages at significant transitions. The methods are illustrated using data from the German Life History Survey (GLHS). It is a subsample also used by Blossfeld and Rohwer (2002) in their book Techniques of Event History Modeling. In the paper, references are made to R packages for multistate modelling and analysis, in particular mvna, etm, msm, mstate, ELECT and Biograph.