9.15-9.45

Introduction: the challenge of OPUS

Prof John Polak, Imperial College

A pervasive problem in many areas of social policy analysis is the need to integrate information from multiple, incomplete data sources (e.g. conventional sample surveys, census records, operational data streams and data generated by technology systems to name but a few) in the estimation of underlying parameters of interest. Researchers working in different areas have developed a wide range of approaches of varying levels of sophistication and applicability but no standard and generally applicable methodology exists. The OPUS project proposes a Bayesian framework for such information integration problems which is designed to enable the use of existing structural knowledge (in the form of existing models) and existing measurement knowledge (in the form of characterisations of sampling and non-sampling errors) to inform the data integration task.

9.45-10.45

The OPUS Methodology (theory: into practice)

Miles Logie, Minnerva/ CharlesLindveld, Imperial College – including demonstration

The OPUS methodology is based on Bayesian Belief Network (BBN) concepts, which it supports through a series of abstractions of real world processes and exploiting survey and modelling information, together with expert domain knowledge about structure. OPUS is generic in nature, but has paid particular attention to problems that are typical of transport system applications, notably those of scalability. London Case Study – The OPUS methodology has been applied to integrate a variety of sets of transport data for London . These have been applied at differing levels of detail but with a focus on a sub-region of North London .

11.15-12.00

Explaining Statistical Models through Metadata

Andrew Westlake, SASC

It can be difficult for end users of results from statistical modelling to understand the form or implications of the modelling process, particularly when modelling complex systems and sometimes without access to the original modellers. OPUS has developed a metadata structure to store information about the forms of model used and the fitting processes applied, together with explanations of the purpose and assumptions of the models. It focuses on, but is not restricted to, Bayesian methodologies. This is supported with functionality for exploring the model structures, accessed through a web interface.

12.00-12.30

Comparing methods for data combination

Dean Judson, Staff Chief, Immigration Statistics US Bureau of the Census

The demand for information at smaller geographies, with more disaggregation, and on a more timely basis is growing. In response to the demand, a wide variety of ‘data suppliers’ (ranging from local governments to academic departments to consulting firms) have attempted to provide such information. Sometimes the attempts have been extremely ad hoc, at other times the attempts have generated new thinking on old topics. A common way to provide answers that are not easily obtainable from standard statistical products is the way of information integration. This presentation will compare and contrast information integration techniques that have been used in the US , noting potential strengths and important weaknesses of each. The presentation will apply the lessons learned to a discussion of the OPUS project.

12.30-12.45

Discussion of OPUS methodology

Lunch - In the break, before and after lunch, the OPUS tools will be on show