Key points are not available for this paper at this time.
Latent variable structural equation models have found a rather new field of application: the modeling of national customer satisfaction measurements. For example, the European Customer Satisfaction Index (ECSI) is based on a structural model that links latent variables such as quality factors, customer satisfaction, and performance factors together. To derive values for the ECSI from data, the partial least squares (PLS) technique is applied to the structural equation model. This technique has the advantage that the estimation results depend much less on distributional assumptions than, e.g. ML-techniques. An alternative approach that can be applied to fit structural equation models to data is to see them as an artificial neural network (ANN) and to use the learning methods developed for neural network analysis to obtain the estimated model. This technique is insofar quite natural as the concept of latent variables is a central element of such models. Moreover, ANNs allows for certain generalizations, e.g., in the functional form of the connections between variables of the model. The paper discusses how structural equation models can be imbedded into the ANN framework and which alternative specifications are feasible on this basis.
Hackl et al. (Sat,) studied this question.