The use of parametric models fitted on micro data in predicting the rates for a cohort of individuals

Francesco C. Billari, Università Bocconi
Laurent Toulemon, Institut National d'Études Démographiques (INED)

Parametric models have traditionally been used in the graduation of age (or duration) patterns of demographic behaviours. In this paper we focus on family formation or dissolution. We assume that micro-level data are available from retrospective surveys providing event history data and we are interested in the prediction of the rates for a cohort of individuals. This is particularly relevant when the quantum of events is not 100%. For instance, for final childlessness levels a prediction is necessary when a cohort is still through reproductive ages. We here assess the use of parametric models fitted on micro-level data. Our assessment is conducted by artificially (and progressively) censoring data and comparing actual estimated rates for a cohort with the prediction arising from the artificially censored data. The main advantages of this approach are: the use of relatively standard statistical tools; 2) to forecast parameters only; 3) to have interpretable parameters.