Computational Social Science Methodology, Anyone?

Abstract. This article reviews computational social science methods and their relation to conventional methodology and statistics. Computational social science has three important features. Firstly, it often involves big data; data sets so large that conventional database and analysis techniques cannot handle them with ease. Secondly, dealing with these big data sets has given rise to analysis techniques that are specially developed for big data. Given the size of the data, resampling and cross-validation approaches become feasible that allow both data-driven exploration and checks on overfitting the data. A third important feature is simulation, especially agent-based simulation. Here size also matters. Agent-based simulation is well known in social science, but modern computer equipment and software allows simulations of unprecedented scale. Many of these techniques, especially the resampling and cross-validation approaches, are potentially very useful for social scientists. Given the relatively small size of social science “big data” is useful to explore how well these techniques perform with smaller data sets. Social science methodology can contribute to this field by exploring if well-known methodological distinctions between external validity, internal validity, and construct validity can help clear up discussions on data quality (veracity) in computational social science.