Abstract

Latent class analysis (LCA), which uses binary variables, and latent profile analysis (LPA), which uses continuous variables, are methods for extracting and interpreting latent groups based on individuals' response patterns. These methods are widely applied in social sciences, including psychology. Both methods require the essential assumptions to ensure valid results, which is the local independence assumption. Violating this assumption can lead to biased parameter estimates and incorrect identification of latent classes. However, many domestic studies tend to overlook the importance of this assumption and omit procedures to verify whether it holds. This study examines the local independence assumption in LCA and LPA and discusses methods for detecting and modeling violations. It first explains the meaning and role of the local independence assumption in both methods and analyzes how violations affect parameter estimation. Then, the study classifies and explores various techniques for detecting violations and modifying models. Specifically, this study organizes LCA-based detection and modeling methods according to different estimation approaches and investigates how these principles can be extended to LPA. Finally, the study integrates the discussions to highlight its implications and limitations.