Estimating a starting population for microsimulation in MICMAC using information-minimization and loglinear modeling techniques

Leo van Wissen, University of Groningen
Peter Ekamper, Netherlands Interdisciplinary Demographic Institute (NIDI)

Microsimulation is useful in cases where the number of dimensions of the state space is large. For instance, when simulating the future health status of the population the following variables may be relevant: age, sex, marital status, living arrangement, education, smoking behaviour, body mass index, and health status. It is often not possible to observe the starting population for the microsimulation from register or census data. Alternatively, we have a sample with all the relevant dimensions, but the sample is not representative. Or, we have pieces of information from different sources that we may want to combine. This paper describes a method to estimate a starting population for microsimulation based on information minimization techniques, and using elements from loglinear contingency table analysis. The method is applied to the estimation of a starting population for MICMAC using data from Italy.