Some Zone Design Experiments with Socio-Economic Data

Seraphim Alvanides and Stan Openshaw
Centre for Computational Geography, School of Geography, University of Leeds, Leeds, LS2 9JT, United Kingdom

Empirical analysis so far has demonstrated that many results of census analysis are aggregation sensitive. Previous studies have appreciated the difficulties associated with the aggregation problem and suggested various geographical and statistical solutions. A general purpose zone design system known as ZDES has been developed; zone design takes regionalisation and redistricting one step further by fully automating the procedure. In that sense it is the process of aggregating N small areas into M larger zones so as to optimise a general function defined on the aggregate data, subject to specific restrictions on the nature of the output regions (viz. internal contiguity). However, unconstrained zone design provides a rather limited representation of reality and attention is focused on adding various additional constraints.

The paper describes extended experimentation work in zone design with socio-economic data. It is relevant to the current debate about the design of zoning systems for statistical purposes within the EC as well as for the 2001 census in the UK. The three main components of zone design: objective function, aggregation method and regionalisation restrictions are reviewed in accordance with the nature of the areal units being studied. Attempts are made to define suitable compaction, similarity and size constraints on the output zones. Furthermore, ways are suggested for defining suitable objective functions to develop standard zoning systems for particular purposes.

Various case studies are used to describe zone design experiments with Enumeration District (ED) data from the 1991 UK census. EDs are the finest areal units available at the moment providing a wide range of socio-economic census variables. Input coverages with 2000 to 5000 areal units, presenting problems of a moderate to medium degree of difficulty and computing time. Two serial algorithms are used to solve the zone design optimisation problem: a tabu search heuristic and a simulated annealing. These are embedded in a penalty function framework designed to impose various nonlinear constraints on the zoning systems. Alternative zonations of particular study areas are evaluated in order to demonstrate the utility of zone design in a wide set of geographical applications. It is also emphasised that zone design is more fundamental than the mere re-engineering of census EDs; it needs to be extended down to more micro spatial scales.

Finally, the paper seeks to bridge the gap between the available zone design system (ZDES3) and the user community. Potential pitfalls are illustrated and practical advice is provided to ZDES users. In this way it is hoped to raise awareness of what is now feasible and avoid zoning anarchy. Concluding, the paper demonstrates the importance of zone design as a tool in zone based spatial analysis and how it can be linked to HPC environments.