Detecting Heterogeneous Regression Patterns: Linear and Quadratic Effects in Mixture Models

• In Event: Methods and Advancements in Mixture Modeling

Abstract

This study explores the consequences of utilizing a linear regression model within a mixture framework when a quadratic pattern exists. Through a Monte Carlo simulation study, we evaluated the performance of regression mixture models under various sample sizes and latent class specifications. Results indicate that the quadratic regression mixture model effectively detects distinct regression shapes. Smaller sample sizes favored the quadratic model with a single class, while larger samples selected the 2-class quadratic mixture model, improving parameter estimation accuracy and reducing bias. Although limited to linear and quadratic relationships, the findings underscore the importance of proper model specification for capturing heterogeneity. Future research could explore non-linear patterns and diverse sample sizes for a comprehensive understanding of model performance.

Authors

• Junyeong Yang, The Ohio State University

• Minjung Kim, The Ohio State University

• Eunsook Kim, University of South Florida

• Radhika Sundar, University of South Florida