Nonlinear Approach to PLS Path Modelling: Methodology, Software and Application
Francesco Costigliola · Università degli Studi di Napoli Federico II · 2017
This thesis proposes a flexible nonlinear alternative to the PLSPM algorithm which tackles two main issues identified and motivated throughout this study: (i) the presence of linearity assumptions; and (ii) the path direction's incoherence within the inner model estimation phase. The proposed approach can be seen, when it comes to the inner model, as a data-driven estimation approach. In fact, the algorithm adapts to the form assumed by the inner relationships among composites by means of a piecewise estimation method. As detailed and motivated along this work, another added value is represented by the possibility of defining a non-symmetrical weighting system designed to accommodate a coherent path direction modelling among composites. The customer satisfaction application to the energy supply market shows how using the proposed nonlinear approach to PLSPM allows the definition of a more precise business strategy. The results obtained are very promising and the proposed Nonlinear PLSPM approach achieved two main goals: (i) the relation defined in the theoretical model are free from the linearity assumption; (ii) the results provided set the basis for a more suitable interpretation of the relation between composites, based on the natural patterns present in the data.