How to Implement Empirical Results of Complex Longitudinal Analysis Models into Microsimulation and Test the Sensitivity of Such Implementations

• Activity Number: 538 - SPEED: Predictive Analytics with Social/Behavioral Science Applications: Spatial Modeling, Education Assessment, Population Behavior, and the Use of Multiple Data Sources
• Type: Contributed
• Date/Time: Wednesday, August 1, 2018 : 10:30 AM to 11:15 AM
• Sponsor: Social Statistics Section
• Author(s): Dawid Bekalarczyk* and Petra Stein
• Companies: and University of Duisburg-Essen
• Keywords: Microsimulation; Panel analysis; Prediction; Sensitivity Analysis; Survey Data; Longitudinal Analysis

Abstract

Microsimulation models are commonly used to predict future developments in relevant societal areas (health, traffic, finances, demographic transition etc.). Since it is hardly possible to predict unforeseeable influences, different scenarios are defined. The outcomes resulting from simulating these scenarios are then compared (quasi-experimental design). Typically, the starting point for developing a scenario are theoretical considerations. In the next crucial step, statistical models are formulated and estimated empirically to specify those scenarios. It is shown how to implement elaborated statistical models, estimated on complex survey-based panel data and how by doing this the quality of scenario-specific microsimulation outcomes can be improved essentially. Furthermore, strategies are presented which allow evaluating the stability of those scenario-specific results (sensitivity analysis). These aspects are worked out in an ongoing sociological research project funded by the German Research Foundation. The aim of this project is to predict occupational status developments of 3rd generation migrants in the context of the demographic transition in Germany.