Non-Metric Partial Least Squares

Partial Least Squares Structural Equation Modeling Approach for Analyzing a Model with a Binary Indicator as an Endogenous Variable

“…Hair et al (2012b), Jakobowicz & Derquenne (2007), and others espouse that this situation is a problem because OLS is used to regress the indicator-binary, in this case-on the (continuous) estimate of the latent variable and OLS is not appropriate for a binary dependent variable. Statisticians such as Tenenhaus, Vinzi, Russolillo, and others (see Russolillo, 2012;Tenenhaus & Hanafi, 2010;Vinzi, Russolillo, & Trinchera, 2010a;Vinzie et al, 2010b) hold an opposing view. In their view, adopting mode A is a statistical decision to use a covariance-based criterion that sets the outer weights to the covariance between the indicator and the estimated latent variable (i.e., the regression coefficient when all scores are standardized) and is not in any way meant as a model of the data (and similarly for mode B, which is a statistical decision to use a correlation-based criterion).…”

An optimization method for evaluating variations in paint coating thickness by infrared thermography

“…For our application, the PLS regression (PLSR) looks for modeling a linear relationship between the predictor (the 2D thermal response matrix) and a predicted (a column-vector composed of a time series). The goal of this technique is to create orthogonal scores vectors (also called latent vectors or component) while maximizes the covariance between and [18]. In this study, the application of PLSR to the pulsed thermography data arises from the necessity to decompose the thermal data obtained during the inspection into a new sequence of images less contaminated by noise and excitation inhomogeneity while maintaining physical coherency [10].…”

GNM-NIPALS: Estimación general no métrica y no lineal por mínimos cuadrados parciales iterativos

“…Se sabe que remplazar cada variable cualitativa por su correspondiente matriz indicadora y luego desarrollar un ACP conlleva problemas de comparación de pesos entre las variables numéricas y las indicadoras, afectación (disminución proporcional) de la inercia en los primeros factores debido a la ortogonalidad de las indicadoras e incremento innecesario de la dimensionalidad (matrices esparcidas) dificultando la capacidad de síntesis en el análisis. Russolillo (2012) [4] presenta el método NM-NIPALS (Non Metric -Nonlinear estimation by Iterative PArtial Least Squares) y desarrolla algorítmica-mente el ACP en una matriz de datos mixtos que contiene n individuos y p * = p + q variables con diferentes escalas de medida; q de ellas cualitativas.…”