Machine Learning to Analyze Single-Case Data: A Replication and Extension

Abstract

Recently, Lanovaz et al. (2020) have found that machine learning algorithms may adequately control for Type I error rate and power when analyzing single-case graphs. However, the study limited most of its analyses to simulated datasets. To replicate and extend this study, we applied the four machine learning models developed by Lanovaz et al. (2020) to a previously published nonsimulated dataset. On average, the four models produced lower proportions of false positives than well-established methods to analyze AB graphs (i.e., the dual-criteria and conservative dual-criteria methods). These results support the use of machine learning to analyze single-case graphs, but further replication by an independent research team using educational and clinical data remains necessary.