Population forecasting via microsimulation : the software-design of the MicMac-Project

Jutta Gampe, Max Planck Institute for Demographic Research
Sabine Zinn, Max Planck Institute for Demographic Research

In microsimulation life-courses of individuals are projected by randomly drawing their trajectories from a stochastic model which realistically portrays the propensity for individual transitions between relevant demographic states. Future population characteristics can then be obtained from the analysis of the aggregated simulated life-courses. This process has several key ingredients: A stochastic model that is able to realistically characterize individual behaviour over the life-course in settings that can be rather complex. Data sources, statistical models, and corresponding estimating procedures that allow to derive the empirical input for the microsimulation as estimated transition rates. And software that combines the input, allows to incorporate assumptions about future behavioural and institutional changes easily, performs the actual life-course simulations and provides simulation results in a format that will allow detailed further analysis. As part of the MicMac-project software is designed, which shall serve these purposes. It will contain a preprocessor to facilitate the estimation of relevant transition rates from data. This part is based on the freely available open-source statistical package R. The simulation according to the multistate model, which is the backbone of the MicMac microsimulation, is performed in a JAVA-based core system, which we jointly develop with the Research Group of Modelling and Simulation at the Computer Science Department, University of Rostock. Finally, a so called postprocessor, which again is based on the package of R, will provide tools for presentation of results. In this presentation we will illustrate the main features of the MicMac microsimulation software. We will outline the general multistate model on which the simulations are based, will demonstrate the link between input data and the simulation framework, and we will describe the main features of the microsimulation core. Finally some standard options for presenting results will be shown. The software will be demonstrated by using data from the Netherlands.