Examining Class Separation Contribution by Observed Indicators in Latent Class Models: A Multiple Testing Procedure. Issue 1 (2nd January 2020)

• Record Type: Journal Article
• Authors: Raykov, Tenko; Harrison, Michael; Marcoulides, George A.
• Is Part Of: Structural equation modeling. Volume 27:Issue 1(2020)
• Page Start: 88
• Page End: 96
• Publication Date: 2020-01-02
• Subjects: Class separation -- entropy -- item response probability -- latent class indicator -- latent class analysis -- multiple testing adjustment -- variable-specific entropy; Multivariate analysis -- Periodicals; Social sciences -- Statistical methods -- Periodicals; 519.535
• DOI: 10.1080/10705511.2018.1554446
• Languages: English
• ISSNs: 1070-5511

Abstract

Abstract : A multiple testing procedure for examining the contribution of latent class indicators to class separation in finite mixture models is discussed. The method permits testing whether an observed indicator contributes to class differentiation in a latent class model, and can also be used for interval estimation of item response probability differences across classes or functions of these probabilities. The approach capitalizes on what may be treated as local information about item response category contribution to class formation, and can similarly be utilized when interested in examining the possibility of dropping an item from a multi-component measuring instrument under consideration. The procedure is readily used with popular software and is illustrated with a numerical example from a latent class modeling setting.