A tool for managing information to improve services: probabilistic expert systems

Within social surveys, the questionnaire is a well-known tool to gather data. Statistical techniques characterised by different levels of complexity, can be used to get out of the collected data differently detailed information to support decisions. For instance, an average is the first and simplest calculation that can be done. However, it carries partial information, since it stands for the central tendency only. Variability is the main feature of any variable; therefore it is necessary to measure (i.e. estimate) it. If the survey refers to quality scores, estimating variability allows to have a first idea of possible improving margins. It is necessary to know, explain and control variability in order to appropriately suggest improvement hypotheses and to measure corrective actions effectiveness. In this paper we propose, by an application to a public administration survey, the use of probabilistic expert systems, PES (Cowell et al., 1999), as a tool for managing complex multivariate and highly structured information. PES are constituted by a graphical representation (the graph, where variables are represented by nodes and relations between variables by directed edges) and by an inferential engine allowing to efficiently propagate information through the network once some kind of evidence occurs to a node. PES can integrate subject-matter knowledge and statistical information obtained from the questionnaire producing a knowledge instrument. Having this kind of knowledge helps making managerial decisions (possible scenarios can be simply produced with a PES by simulating different interventions in one or more variables) and planning improvement actions.