When studying complex systems the difficulty of analysis is mainly due to the complex network of hypothesized, but often hidden, and presumably causal (or at least predictive) relationships between tangible (i.e. manifest and directly observed) phenomena or intangible (i.e. theoretical and indirectly observed) concepts. It is somehow the problem of extracting knowledge from uncertain models rather than modeling uncertainty in a specific model defined on some a priori available information. The basic elements of causal networks (in a covariance-based framework) or predictive path models (in a component-based approach) are the manifest variables, the corresponding latent variables (or factors) and the network of dependence/causal relationships between the latter ones. Both the measurement model (manifest-latent links) and the structural model (latent-latent links) are usually specified according to theoretical hypotheses of the researcher and can be eventually (but only slightly) modified in case the statistical modeling of empirical data does not confirm the whole set of hypotheses thus providing a different or new evidence. Further knowledge may be extracted if induction by automatic learning is merged to the evaluation of probabilistic networks.
ESPOSITO VINZI, V. et ZARGOUSH, M. (2010). Knowledge Extraction by Investigating Model Uncertainty thorugh Predictive Path Modeling and Probabilistic Networks. Dans: Joint Meeting (GfKj-CLADAG’10).