Towards machine-learning enhanced QCA: optimizing coverage and empirical significance

Abstract Qualitative Comparative Analysis (QCA) relies on case knowledge and existing theories to select conditions and calibrate sets. However, as the number of cases grows, researchers’ case knowledge tends to decrease, posing significant challenges to large-N QCA practitioners. In response, this article proposes machine-learning enhanced QCA (mlQCA) for condition selection, set calibration, and iterative model development in large-N QCA. This approach specifically incorporates Extreme Gradient Boosting (XGBoost) because its tree-based models have algorithms similar to QCA sufficiency tests. By using mlQCA, practitioners can identify explanatory conditions and calibration cut-offs that optimize the coverage of QCA solutions. Drawing on a classic study on voting participation, the article illustrates how mlQCA can leverage empirical information to produce substantively important results. Two versions of mlQCA are presented. The radical version makes extensive use of automation tools, while the conservative version prioritizes theoretical engagement and allows QCA practitioners to decide the extent to which automation tools are used.