The usefulness of algorithmic models in policy making

Abstract

Governments increasingly use algorithmic models to inform their policy making process. Many suggest that employing such quantifications will lead to more efficient, more effective or otherwise better quality policy making. Yet, it remains unclear to what extent these benefits materialize and if so, how they are brought about. This paper draws on the sociology and policy science literature to study how algorithmic models, a particular type of quantification, are used in policy analysis. It presents the outcomes of 38 unstructured interviews with data scientists, policy analysts, and policy makers that work with algorithmic models in government. Based on an in-depth analysis of these interviews, I conclude that the usefulness of algorithmic models in policy analysis is best understood in terms of the commensurability of these quantifications. However, these broad communicative and organizational benefits can only be brought about if algorithmic models are handled with care. Otherwise, they may propagate bias, exclude particular social groups, and will entrench existing worldviews.