Comparing the Robustness of Stepwise Mixture Modeling With Continuous Nonnormal Distal Outcomes

Area of Science:

• Statistics
• Psychometrics
• Data Analysis

Background:

• Latent class analysis (LCA) mixture modeling is used for complex data structures.
• Evaluating model performance with auxiliary distal outcomes requires robust methods.
• Previous studies have not fully addressed non-normality in distal outcome distributions.

Purpose of the Study:

• To compare the robustness of four latent class analysis mixture modeling approaches under various conditions.
• To assess the performance of maximum likelihood (ML) with homoskedasticity (ML_E) and heteroskedasticity (ML_U), BCH, and LTB.
• To investigate the impact of non-normality on distal outcome estimation.

Main Methods:

• Monte Carlo simulations were used to evaluate model performance.
• Four distinct latent class analysis mixture modeling approaches were tested.
• The study considered non-normality in the distributional form of distal outcomes.

Main Results:

• The Bayesian Content-based (BCH) approach consistently yielded the most unbiased estimates of class-specific distal outcome means across all simulation conditions.
• The performance of ML_E, ML_U, and LTB varied depending on the simulation conditions.
• Non-normality significantly impacted the accuracy of estimates for some approaches.

Conclusions:

• The BCH approach is recommended for latent class analysis mixture modeling when dealing with auxiliary distal outcomes, particularly when non-normality is present.
• Researchers should carefully consider the distributional assumptions of distal outcomes when selecting a modeling approach.
• This study provides valuable insights for applying robust latent class analysis techniques.