Comparison of Criteria for Choosing the Number of Classes in Bayesian Finite Mixture Models

Kazem Nasserinejad, Joost van Rosmalen, Wim de Kort and Emmanuel Lesaffre

PLOS ONE, 2017, vol. 12, issue 1, 1-23

Abstract: Identifying the number of classes in Bayesian finite mixture models is a challenging problem. Several criteria have been proposed, such as adaptations of the deviance information criterion, marginal likelihoods, Bayes factors, and reversible jump MCMC techniques. It was recently shown that in overfitted mixture models, the overfitted latent classes will asymptotically become empty under specific conditions for the prior of the class proportions. This result may be used to construct a criterion for finding the true number of latent classes, based on the removal of latent classes that have negligible proportions. Unlike some alternative criteria, this criterion can easily be implemented in complex statistical models such as latent class mixed-effects models and multivariate mixture models using standard Bayesian software. We performed an extensive simulation study to develop practical guidelines to determine the appropriate number of latent classes based on the posterior distribution of the class proportions, and to compare this criterion with alternative criteria. The performance of the proposed criterion is illustrated using a data set of repeatedly measured hemoglobin values of blood donors.

Date: 2017
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (8)

Downloads: (external link)
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0168838 (text/html)
https://journals.plos.org/plosone/article/file?id= ... 68838&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pone00:0168838

DOI: 10.1371/journal.pone.0168838

Access Statistics for this article

More articles in PLOS ONE from Public Library of Science
Bibliographic data for series maintained by plosone ().