Estimating the parameters of system dynamics models using indirect inference

Abstract

There is limited methodological guidance for estimating system dynamics (SD) models using datasets common to social sciences that include few data points over time for many units under analysis. Here, we introduce indirect inference, a simulation-based estimation method that can be applied to common datasets and is applicable to SD models that often include intractable likelihood functions. In this method, the model parameters are found by ensuring that simulated data from the model and available empirical data produce similar auxiliary statistics. The method requires few assumptions about the structure of the model and error-generating processes and thus can be used in a variety of applications. We demonstrate the method in estimating an SD model of depression and rumination using a panel dataset. The overall results suggest that indirect inference can extend the application of SD models to new topics and leverage common panel datasets to provide unique insights.