ReMoDe – Recursive modality detection in distributions of ordinal data

Abstract The detection of the number of modes in distributions of ordinal data is relevant for applied researchers across disciplines, from uncovering polarization to detecting incidence groups in clinical symptom scales. Yet, established modality detection methods are either purely descriptive or not developed for ordinal data. In the present paper, we attempt to fill this gap by proposing a recursive modality detection method (ReMoDe) which detects modes in univariate distributions through recursive significance testing. We provide a comprehensive review of existing modality detection methods and outline their potential pitfalls when applied to ordinal scales. Based on a benchmark of 172 simulated ordinal samples of different sample sizes, we demonstrate that ReMoDe outperforms other established modality detection methods. We furthermore present a stability test for our method as well as p ‐values and approximated Bayes factors for each detected mode. To make our method easily applicable for researchers, we introduce open‐source R and Python packages.