Partial least squares path analysis

Abstract

We begin our review of modern path analysis tools with partial least squares path analysis software. PLS-PA has achieved near-cult-like stature within its circle of practitioners, but is not without its critics. Many issues arise from PLS-PA not being a proper statistical “methodology”—it has failed to accumulate a body of statistical research on assumptions, the role of data, objectives of inference, statistics, or performance metrics. Rather, PLS consists of a half dozen or so software packages that though only lightly documented seem to be able to conjure path estimates out of datasets that other methodologies reject as inadequate. This chapter explores whether PLS-PA software really possesses some “secret sauce” that makes it possible to generate estimates from weak data, or conversely, whether such imputed path structures may indeed be illusory.