Bayesian Multilevel Latent Class Profile Analysis: Inference and Estimation for Exploring the Diverse Pathways to...

Area of Science:

• Statistics
• Educational Psychology
• Data Analysis

Background:

• Multilevel latent class profile analysis (MLCPA) is crucial for longitudinal studies.
• Conventional maximum likelihood (ML) estimation faces challenges with small samples and boundary issues.
• The underflow problem can occur in MLCPA due to multilevel structures.

Purpose of the Study:

• To propose and evaluate Bayesian estimation for MLCPA as an alternative to ML estimation.
• To investigate the underflow problem in MLCPA.
• To compare the performance of Bayesian and ML estimates in various simulation conditions.

Main Methods:

• Developed a Bayesian estimation approach for MLCPA using non-informative priors.
• Conducted extensive numerical simulations to compare Bayesian and ML estimates.
• Analyzed longitudinal academic performance data from the Progress Monitoring and Reporting Network.

Main Results:

• Bayesian estimates are preferred when latent classes are well-separated.
• ML estimates are preferred when latent classes overlap.
• Identified distinct student academic proficiency trajectories and school-level latent groups.

Conclusions:

• Bayesian estimation provides a viable alternative for MLCPA, especially in challenging scenarios.
• Findings highlight variations in academic proficiency and inform educational policy.
• The study offers new perspectives on academic patterns and interventions.