Un Drôle De Type: The Schelling Model, Calibration, Specification, Validation and Using Relevant Data

This paper identifies a potential blind spot in ABM, linking aspects of methodology and data use. The relative neglect of “specification” (empirical justification of model components like particular agent decision processes) combined with a relative paucity of qualitative data in ABM draws attention away from the possibility that agents may make decisions in heterogeneous ways with uncertain implications for macroscopic system properties. Using the Schelling model as a simple and well known example, this paper considers the role of specification as complementary to calibration and validation, the way that different kinds of data (qualitative and quantitative) map on to different aspects of ABM methodology to justify a model empirically and the possible implications of systematically heterogeneous decision making. Some preliminary simulation results are presented and discussed for mixtures of heterogeneous decision “types” grounded in existing secondary data. The paper also considers exactly what the Schelling model can (and cannot) be taken to show and how the legitimacy of various claims made for it relate to its empirical justification.