Bayesian Latent Class Analysis: Sample Size, Model Size, and Classification Precision

The current literature includes limited information on the classification precision of Bayes estimation for latent class analysis (BLCA). (1) Objectives: The present study compared BLCA with the robust maximum likelihood (MLR) procedure, which is the default procedure with the Mplus 8.0 software. (2...

Full description

Saved in:
Bibliographic Details
Published inMathematics (Basel) Vol. 11; no. 12; p. 2753
Main Author Mindrila, Diana
Format Journal Article
LanguageEnglish
Published Basel MDPI AG 01.06.2023
Subjects
Artificial intelligence
Bayes estimation
Bayesian analysis
BLCA
Book publishing
Classification
Comparative analysis
Entropy
Latent class analysis
latent variable modeling
Machine learning
Markov chains
Markov processes
Mathematics
Missing data
Monte Carlo method
Optimization
Parameter estimation
person-oriented analyses
Probability
Software
Statistical analysis
Statistical methods
Structural equation modeling
Variables
New York
United States
United Kingdom
Online AccessGet full text

Cover

More Information
Summary:The current literature includes limited information on the classification precision of Bayes estimation for latent class analysis (BLCA). (1) Objectives: The present study compared BLCA with the robust maximum likelihood (MLR) procedure, which is the default procedure with the Mplus 8.0 software. (2) Method: Markov chain Monte Carlo simulations were used to estimate two-, three-, and four-class models measured by four binary observed indicators with samples of 1000, 750, 500, 250, 100, and 75 observations, respectively. With each sample, the number of replications was 500, and entropy and average latent class probabilities for most likely latent class membership were recorded. (3) Results: Bayes entropy values were more stable and ranged between 0.644 and 1. Bayes’ average latent class probabilities ranged between 0.528 and 1. MLR entropy values ranged between 0.552 and 0.958. and MLR average latent class probabilities ranged between 0.539 and 0.993. With the two-class model, BLCA outperformed MLR with all sample sizes. With the three-class model, BLCA had higher classification precision with the 75-sample size, whereas MLR performed slightly better with the 750- and 1000-sample sizes. With the 4-class model, BLCA underperformed MLR and had an increased number of unsuccessful computations, particularly with smaller samples.
ISSN:2227-7390
2227-7390
DOI:10.3390/math11122753