Characterizing modeling contexts to support adequate and valid model choice

There is widespread consensus across disciplines that model validity can only be judged with respect to a specific purpose (Oreskes et al., 1994; Barlas, 1996; Jakeman et al., 2006; Augusiak et al., 2014). While e.g. Edmonds et al. (2019) classify modeling purposes into discrete classes (description, explanation, prediction, theoretical exposition, illustration, analogy, and social learning), we discuss that the choice of valid models and methods in addition requires considering structural knowledge about the system, (potential) data availability, computational resources, degree of generalization and identifiability, as well as precision, accuracy and robustness required to answer the research question. Together, this can be termed as the ‘modeling context’. Discussions about valid modeling approaches and adequate methods of validation and calibration of agent-based models are often complicated by blurred descriptions of modeling context. In our contribution, we delineate the main dimensions that characterize the modeling context and illustrate how they affect the adequate choice of valid models and methods in agent-based simulation throughout the whole process of simulation starting from conceptualization and parameterization over potential estimation and empirical validation as well as simulation and uncertainty analysis to interpretation.