The Effects of Chance Correlations on Partial Least Squares Path Modeling

Abstract

Partial least squares path modeling (PLS) has been increasing in popularity as a form of or an alternative to structural equation modeling (SEM) and has currently considerable momentum in some management disciplines. Despite recent criticism toward the method, most existing studies analyzing the performance of PLS have reached positive conclusions. This article shows that most of the evidence for the usefulness of the method has been a misinterpretation. The analysis presented shows that PLS amplifies the effects of chance correlations in a unique way and this effect explains prior simulations results better than the previous interpretations. It is unlikely that a researcher would willingly amplify error, and therefore the results show that the usefulness of the PLS method is a fallacy. There are much better ways to compensate for the attenuation effect caused by using latent variable scores to estimate SEM models than creating a bias into the opposite direction.

Strategic Improvement for Quality and Satisfaction of Hospital Information Systems

“…IPA was then employed to compare the perceptions of HIS attributes by physician users between high-satisfaction and low-satisfaction groups. The above statistical analysis were conducted by using R software [37] with the matrixpls package [38]. …”

A structured review of partial least squares in supply chain management research

“…PLS thus has been the subject of much debate. One half of the research community argues that PLS has its advantages when it is correctly used, and might even be a "silver bullet" (Hair et al, 2011); the other half is strictly against its use, arguing that it is inferior to traditional CBSEM techniques (Antonakis et al, 2010;Rönkkö, 2014;Rönkkö and Evermann, 2013). Researchers opposing the use of PLS criticize the bias and inconsistency of parameter estimates, its inability to model measurement errors, and the lack of an over-identification test, which would allow for testing a model causally (Hwang et al, 2010;Peng and Lai, 2012;Rönkkö and Evermann, 2013).…”

“…The growing number of articles published using PLS in SCM (e.g., Caniëls et al, 2013;Hartmann and de Grahl, 2012; and the controversy regarding the application of PLS in various disciplines (e.g., Hair et al, 2011;Henseler et al, 2014;Rigdon, 2012;Rönkkö, 2014;Rönkkö and Evermann, 2013) suggest the need to compare and contrast how PLS is being used in the SCM literature. Thus, a structured review that is targeted directly at SCM research -one that includes the pros and cons of applying PLS -seems warranted.…”

Empirically measuring salience of objects for use in pedestrian navigation

“…They propose that it is not a structural equation modeling method at all, cannot be used to develop measurement models, does not provide results which are more reliable than simple sum scores, is not suitable to do exploratory research, has no sample size requirements and does not allow for statistical tests of the significance of path coefficients estimated. Rönkkö (2014) goes on with a simulation study investigating the so-called problem of capitalization on chance correlations of error terms 30 . Thereby, they abandon mixed or purely formative models at all.…”

“…On the other hand, the need to establish crowd-sourcing based data sources for geographical data in general and data on LMs in particular has been claimed for some years now (cf. Raubal, Mark, & Frank, 2013;Richter & Winter, 2011a, 2014Winter et al, 2010). Hirtle and Raubal (2013, p. 145) define this type of information:…”