Improvements to PLSc: Remaining problems and simple solutions

Loading...
Thumbnail Image
Journal Title
Journal ISSN
Volume Title
School of Science | J Muu elektroninen julkaisu
Date
2016
Major/Subject
Mcode
Degree programme
Language
en
Pages
102
Series
Abstract
The recent article by Dijkstra and Henseler (2015b) presents a consistent partial least squares (PLSc) estimator that corrects for measurement error attenuation and provides evidence showing that, generally, PLSc performs comparably to a wide variety of more conventional estimators for structural equation models (SEM) with latent variables. However, PLSc does not adjust for other limitations of conventional PLS, namely: (1) bias in estimates of regression coefficients due to capitalization on chance; and (2) overestimation of composite reliability due to the proportionality relation between factor loadings and indicator weights. In this article, we illustrate these problems and then propose a simple solution: the use of unit-weighted composites, rather than those constructed from PLS results, combined with errors-in-variables regression (EIV) by using reliabilities obtained from factor analysis. Our simulations show that these two improvements perform as well as or better than PLSc. We also provide examples of how our proposed estimator can be easily implemented in various proprietary and open source software packages.
Description
Keywords
partial least squares, structural equation modeling, composite variables, latent variables, measurement error, reliability, correction for attenuation, capitalization on chance, errors-in-variables regression
Other note
Citation
Rönkkö, Mikko & McIntosh, Cameron N. & Aguirre-Urreta, Miguel I. 2016. Improvements to PLSc: Remaining problems and simple solutions. Unpublished working paper. 102 pages.