COMPUTATIONAL SOCIAL SCIENCE

Department of Computational Social Science Seminar Abstract

Friday, 4th December: 3.00pm

On Qualitative Empirical Techniques in Agent-based Computational Social Science

Armando Geller, Department of Computational Social Science

Strengthening the empirical foundations of agent-based social simulation is critical if we want the field to be taken more seriously by established social science traditions. There are many ways to found a model empirically. For example, input parameters can be matched against empirical values, model distributions can be fit to real data, and topologies can be taken from GIS data files. Fitting simulation output to real time-series data is one of the most commonly used cross-validation techniques. In contrast, the use of qualitative data is underused in informing the design of agent-based computational social science models. The advantages of such techniques have not yet been fully understood nor have they been broadly endorsed. However, given recent interest in the approach by a number of scholars, I deem it appropriate to introduce and critically discuss some relevant concepts and techniques. Amongst these are qualitative data and the cognitive-behavioral sphere; from actor narratives to agent rules: formalization procedures for qualitative data; defining cognitive spheres: what agents reason about; social mechanisms and internal validation; cross-validation via narratives. In summary, by not making use of qualitative data in agent-based computational social science we neglect a rich body of data that provides information to answer what is an essential question in the social sciences: How and why do humans what they do?