Latent Class Analysis With Covariates Utilizing Fuzzy Clusterwise Generalized Structured Component Analysis

• In Event: Marriott Roundtable Session 29
  In Roundtable Session: Advances in Clustering and Classification

Abstract

Latent Class Analysis (LCA) utilizing maximum likelihood estimation (MLE) has been ubiquitously applied. However, MLE frequently faces identification issues due to a large number of parameters. When LCA includes covariates, the estimation would suffer more severe problems. To avoid this issue, previous research applied a clustering algorithm to implement LCA in generalized structured component analysis, called gscaLCA. As an extension of this, the current study proposes a method of implement gscaLCA with covariates by using logistic regression. The introduced method was applied to substance use data with two covariates (gender and education level) to demonstrate the feasibility. The results showed that either males or low education level samples more likely tends to be high substance user class.

Authors

• Seohee Park, American Board of Internal Medicine

• Seongeun Kim, University of North Carolina at Greensboro

• Ji Hoon Ryoo, Yonsei University