Integrating Information about Complex Systems

Abstract

Opus is a project funded by the European Union Statistical Agency (Eurostat). The project is developing a methodology for the coherent and consistent integration of information from multiple sources about complex systems (based on Bayesian statistical models) and a series demonstration of this approach. Most, but not all, of these are transport applications. Alongside the modeling frameworks being developed, the project has developed a framework for storing a complete audit trail, covering the specification and fitting of statistical models, and making use of a formal metadata specification to represent statistical models. This framework builds on ideas about metadata for processes that have been developed in other statistical projects. Practitioners are often skeptical about results derived from models. In practice, all analysis of survey or sample data involves some form of model, even if it is hidden in implicit assumptions. The approach to this skepticism is to open up the model to scrutiny and to present information about the reliability of results, under the title of provenance and reliability. All this comes from the metadata about the statistical model. The approach is generic, but the presentation needs to be domain specific and tailored to the level of expertise and understanding of the user of the results. This paper presents the motivation behind this approach and outlines the metadata structures and functionality being built to deliver this support to users of results from the Opus methodology.