A Bayesian Approach to Modeling Dynamical Systems in the Social Sciences

摘要

The paper presents a new modeling approach using longitudinal or panel data to study social phenomena and to make predictions of dynamic changes. While the most common way in social sciences to study the relations between variables is using regression, our modeling approach describes the changes in variables as a function of all included variables, using differential equations with polynomial terms that capture linear and/or nonlinear effects. The mathematical models represented by these differential equations are derived directly from data. The models can then be run forward to forecast future changes. A two-step model-fitting approach is applied to identify the best-fit models and included visualisation methods based on phase portraits help to illustrate modeling results. We show this approach on an example relating democracy to economic growth.