Partial least squares as a tool for scientific inquiry: comments on Cadogan and Lee
本文回应Cadogan和Lee对PLS科学适用性的质疑，指出不同PLS变体在正确使用下均适合科学研究，并批评“PLS-SEM”框架的某些支持者误导了反思性测量和模型评估。
Purpose In their paper titled “A Miracle of Measurement or Accidental Constructivism? How PLS Subverts the Realist Search for Truth,” Cadogan and Lee (2022) cast serious doubt on PLS’s suitability for scientific studies. The purpose of this commentary is to discuss the claims of Cadogan and Lee, correct some inaccuracies, and derive recommendations for researchers using structural equation models. Design/methodology/approach This paper uses scenario analysis to show which estimators are appropriate for reflective measurement models and composite models, and formulates the statistical model that underlies PLS Mode A. It also contrasts two different perspectives: PLS as an estimator for structural equation models vs. PLS-SEM as an overarching framework with a sui generis logic. Findings There are different variants of PLS, which include PLS, consistent PLS, PLSe1, PLSe2, proposed ordinal PLS and robust PLS, each of which serves a particular purpose. All of these are appropriate for scientific inquiry if applied properly. It is not PLS that subverts the realist search for truth, but some proponents of a framework called “PLS-SEM.” These proponents redefine the term “reflective measurement,” argue against the assessment of model fit and suggest that researchers could obtain “confirmation” for their model. Research limitations/implications Researchers should be more conscious, open and respectful regarding different research paradigms. Practical implications Researchers should select a statistical model that adequately represents their theory, not necessarily a common factor model, and formulate their model explicitly. Particularly for instrumentalists, pragmatists and constructivists, the composite model appears promising. Researchers should be concerned about their estimator’s properties, not about whether it is called “PLS.” Further, researchers should critically evaluate their model, not seek confirmation or blindly believe in its value. Originality/value This paper critically appraises Cadogan and Lee (2022) and reminds researchers who wish to use structural equation modeling, particularly PLS, for their statistical analysis, of some important scientific principles.