Signals of Uncertainty and Misspecification in Latent Class Analysis

• In Event: Methods and Advancements in Mixture Modeling

Abstract

Our approach aims to address the limitations of modification indices in latent class analysis. By introducing gradient descent-based sensitivity analysis, we seek to improve the reliability of study conclusions. Through iterative adjustments guided by the objective function, gradient descent aligns model predictions with observed data, highlighting potential inadequacies in the original model. This prompts further investigation and potential refinement to mitigate misspecification. Adopting gradient descent enhances the robustness and validity of findings in LCA scenarios, providing researchers with a powerful tool for model improvement.

Authors

• Zachary K. Collier, University of Connecticut

• Yiqin Cao, University of Delaware

• Joshua Sukumar, University of Connecticut