Machine Learning in Sociology

Abstract

Abstract Sociologists are increasingly turning to machine learning (ML) for data-driven discovery and predictive modeling. ML methods help classify data, compute new measures, predict outcomes and events, make causal inferences, and collaborate within a common-task framework. Although predictive analytics has become a mainstay of public policy analysis and evaluation, the contributions of ML to theory building are less widely appreciated. ML-derived data classifications can reveal patterns that require a new theory, while predictive performance metrics can point to shortcomings of existing theory and motivate inductive theorizing. ML equips research analysts to venture outside the “general linear reality” of classical statistics and the deductive framing of much social science research. Both quantitative and qualitative sociologists also scrutinize ML applications by industry and government to reveal implications for distributive justice, social inequality, and algorithmic bias. In short, ML is now an integral part of sociological inquiry as a discovery-enabling analytical tool as well as a controversial object of study.