Calibrating an Opinion Dynamics Model to Empirical Opinion Distributions and Transitions

Agent-based models of opinion dynamics enable the investigation of societal phenomena that psychological theories of individual opinion change trigger in artificial societies. Many of such models are validated on the phenomenological level only (e.g., can they produce polarization) and not with respect to opinion data about real-world attitude distributions known from social surveys or individual attitude change observed in panel surveys. In this work, we use an existing agent-based model which builds on repeated random pairwise interaction as introduced in many early opinion dynamics models with a function of individual opinion change function that includes the contagious and assimilating opinion adjustment, idiosyncratic opinion change, and motivated cognition, which is a generalization of the concept of bounded confidence. Depending on its parameters, this model creates different opinion distributions. By arguing that institutional trust is formed by an exchange of experiences and opinions, we then match the opinion distributions from the seven institutional trust questions in the European Social Survey to the simulated ones. Goodness-of-fit measures allow to calibrate the model parameters to reproduce 1,235 empirical distributions (for different countries and years) and further on, the calibration procedure is extended to data about individual opinion change from the Swiss Household Panel where participants reported their trust in the government on a yearly basis. Overall the opinion dynamics model reproduces most of the empirical distributions to a high degree, additionally, matching agents to individuals shows also a high level of concordance. In the analysis, we show that the concept of motivated cognition is a crucial part to achieve this level of accuracy. However, in both calibrations, we expose the shortcoming of the model to reproduce individuals with extreme and neutral attitudes.