How agent heterogeneity, model structure and input data determine the performance of an empirical ABM - A real-world case study on residential mobility

Agent-based models (ABMs) have tended to become more complicated in recent years, going along with challenges of fully understanding model behaviour. While partly answered using simple and abstract models, the question of the role of different aspects of model detail for controlling model outcomes still has not been explored with empirical ABMs. We therefore use a detailed model built upon an empirical survey of residential mobility to explore the effects of three domains of model detail - agent heterogeneity, structural model detail and detail in the data used - on model performance. Evaluation is done by measuring the ability to predict empirical patterns for different topics, population, relocations and vacancies at different scales. Our diverse results indicate that a good data basis is crucial, heterogeneity strongly controls small-scale patterns and different design aspects must be tested thoroughly. This outcome calls for caution when interpreting the results of such models.