Modelli ad Equazioni Strutturali e Reti Probabilistiche Bayesiane: due approcci a confronto nello studio di relazioni causali

Causality is a concept extremely important in science, but its definition is quite controversial and its detection is not exempt from problems. There are different approaches to deal with causality. Two of them are the Structural Equation Models (SEMs) and the Probabilistic Bayesian Networks (PBNs). SEMs are confirmative models. Given a causal structure, they test if it is coherent with data. In this context they are estimated using the Partial Least Squares Path Modeling technique in order to obtain the scores of latent variables. PBNs are inductive methods. Their attempt is to extract the causal scheme deriving from data, without presupposing any knowledge. Both models presents advantages and disadvantages regardless of the causality approach they refer to. SEMs are best suited for quantitative data and when there is a solid theoretical knowledge on the subject of analysis. PBNs are preferable for nonlinear analysis or uncertain causal scheme. In the thesis a possible integration of the two methods is proposed in the analysis of data deriving from a satisfaction and customer loyalty survey for “Customer American Satisfaction Index” (ACSI). Results suggest that the SEMs are more suitable than the PBNs and that the integration of the two statistical models is advantageous only in part. This is related to the kind of data, since the ACSI survey is structured for a PLS-PM analysis. Thus, it could be very interesting repeat the comparison for different types of data.