Agent-Based Social Simulation (ABSS), i.e., the study of social systems through Agent-Based Modeling, has great potential in serving as policy support. However, challenges in the modeling for policymaking process prevent this potential to be fully realized. For instance, the models might not be deemed sufficiently credible or appealing to decisionmakers.This licentiate thesis aims to contribute to the understand of these challenges, and to identify challenges and opportunities in how to increase the credibility of ABSS models in policymaking.

First, we investigate the modeling of human behavior and what challenges there are that may affect models’ suitability for policymaking. Properly representing individuals’ decision-making can serve to increase model accuracy, model descriptiveness and recognizability of the modeled system, which in turn can increase model credibility. Taking models of the COVID-19 pandemic as an example, we performed reviews that analyzed what aspects of human behavior were modeled, and how these aspects relate to what conclusions can be drawn from the model. The studies found that many of the aspects that seem relevant were rarely included in the studied models. Three challenges were identified with regards to being able to build more descriptive behavior models within the time constraints posed by decision-makers: improvements to modeling tools and software, model reuse, and data availability.

Second, we look at the verification and validation (V&V) of ABSS models. The proper evaluation of models is crucial for model credibility to be ensured. In the study of COVID-19 models, V&V activities were found to be rarely documented (if performed at all). To this end, we suggest the continuous use of a ”V&V plan” during the modeling process,and proposes an accreditation framework for institutions to be able to perform external validation.

Third, we consider the combination of models and model results. Model combination can lead to more credible models as it allows modelers to use well-established, well validated models to represent parts of a system, while combined simulation results often are more robust than those of a single model. A literature survey was performed, identifying six different types of combinations: Ensemble Modeling, Meta Analysis, Model Merging, Models as Modules, Model Integration and Model Chaining. For each type, we identified purpose and examples, as well as existing challenges for the combination of both simulation models in general and ABSS models in particular.