Summary of Work Packages 2-3: A Theoretical Assessment and A Practical Assessment Marja Tammilehto-Luode, Statistics Finland Philippe Guiblin, Office of National Statistics, United Kingdom GISCO Working Party 25 October 2001

Plan for the presentation Point of departure The two parallel approaches Methods Tests with empirical data Results GISCO Working Party 25 October 2001

Point of departure: A need for more comparable territorial divisions for statistics Why? to visualise data more effectively to combine and compare data on different spatial units to combine or compare data on different spatial scales to make better statistical/spatial analysis of data (to test spatial patterns and trends GISCO Working Party 25 October 2001

Finland 20 NUTS3-areas Belgium 45 NUTS3-areas GISCO Working Party 25 October 2001

Objectives to find alternative comparable building blocks/territorial division for European system of small area statistics to summarise studies of theoretical assessments of systems of statistics by regular and irregular tessellation to present a practical pilot study -to product grids and blobs -to test their capabilities -to test both systems with the same user case GISCO Working Party 25 October 2001

The study of two parallel approaches Regular tessellation approach Statistics Finland’s responsibility study of methods to construct grid-based statistics tests of candidate methods with empirical data construction of a prototype tests on a prototype with a user case Irregular tessellation approach Office of National Statistics in UK’s responsibility study of methods of zoning design tests with ZDES (Zone Design System) by Leeds University with empirical data tests on a prototype with a user case GISCO Working Party 25 October 2001

Irregular tessellation From diversified input areas to harmonised building blocks and comparative output areas Regular tessellation Irregular tessellation Input areas Building blocks Output areas GISCO Working Party 25 October 2001

Towards a system of grid-based statistics Methods -Points to grid cells - need for best practices -Polygons to grid cells - different methods give different results - need for standardisation Size of a grid cell dependent on -quality of data -confidential grid cells -scale of the study -compatibility of different kind of source data -disk space, processing speed Map projection -rectangular -common geo-reference system - with a same origin GISCO Working Party 25 October 2001

Towards a system of polygon-based statistics Optimum grouping of the source areas -equal values of selected variable (population) -similar degree of heterogeneity (homogeneity) -weighted accessibility on certain input variable (shape) The automated zone design program (Openshaw 1977) -optimising an objective function -optimising under constraints Design functions -equality population zoning -shape design function -homogeneity design function -correlation -, distance -, spatial autocorrelation functions GISCO Working Party 25 October 2001

Test areas and data sets The Finnish data sets cover the Helsinki region -56 municipalities, 168 postal code areas and 168 grid-squares (10 km x 10 km) and 13 003 grid-squares (1 km x 1 km) -1998 census population counts of each area level The British data set covers the region of Wales and sub set of the county of the South Glamorgan -6376 EDs for the Wales and 817 EDs for the South Glamorgan -1991 census population counts by EDs GISCO Working Party 25 October 2001

Tests: Data from two countries, different types of geo-references, different types of variables grids tests of candidate algorithms to convert polygon-based data to grids construction of prototype visualisation of results delineation of urban areas blobs (polygons) tests of zone function with constraints of equal population and a single weighted shape construction of prototypes visualisation of results delineation of urban areas GISCO Working Party 25 October 2001

Building up a system of grids - Results From points to grids -Description of best practices From polygons to grids -Regiongrid-algorithm maintains the original data structure - similar statistics to ‘real grids’ -Larger grids keep the structure better than smaller -Polygrid and Pointgrids algorithms overestimate data to small grids and underestimate data in larger grids Delineation of Urban areas -The urban area by 1km x 1km grids is 45% smaller than that defined using the data on NUTS5 areas in Finland - The urban area of the UK test area, South Glamorgan is a little bit larger by 1km x 1km building blocks than by NUTS5 areas GISCO Working Party 25 October 2001

GISCO Working Party 25 October 2001

GISCO Working Party 25 October 2001

GISCO Working Party 25 October 2001

GISCO Working Party 25 October 2001

GISCO Working Party 25 October 2001

Building up a system of blobs Software -Visual Basic program, AZM, developed by David Martin (Southampton University) Constraints -only equal population zoning and a shape constraints - perimeter squared/area were used -no other source of heterogeneity was considered such as geography and/or social class Delineation of Urban areas - Production of an ‘optimised’ boundary system GISCO Working Party 25 October 2001

AZP Case Study: Cardiff Area Input Areas : 818 ED’s min. pop: 64 max. pop: 1030 Design Constraints Thresholds: Target pop.: 500 Min. pop. : 1500, Homogeneity : off Shape: on Automated Output Area Design (AZP) Building Blocks min.pop: 500 max. pop: 1909 Delineation of Urban / Rural Areas GISCO Working Party 25 October 2001

AZP on South Glamorgan Enumeration District Input Areas GISCO Working Party 25 October 2001

AZP on South Glamorgan Enumeration District Building Blocks GISCO Working Party 25 October 2001

Enumeration Districts Urban / Rural Delimitation AZP on South Glamorgan Enumeration Districts Urban / Rural Delimitation GISCO Working Party 25 October 2001

Urban / Rural Delimitation AZP on Helsinki post-codes Urban / Rural Delimitation GISCO Working Party 25 October 2001

GISCO Working Party 25 October 2001

Results and deliverables Reports of theoretical and technical assessment of two alternative building blocks for a system of geo-statistics of Europe Prototypes of grid (regular)-based and polygon(irregular)-based statistical system Results of delineation of urban areas with the building blocks Recommendations GISCO Working Party 25 October 2001

Conclusions - grid-based statistics A lot of advantages - a relevant alternative The critical points - characteristics of input data - what is a optimum/minimum size of the grid - what is a projection system Harmonised methods for converting grids needed The resultant grid size should be the same or coarser than the one for the input data The common georeferenced system needed - UTM applicable projection system Confidentiality problems - special disclosure control methods needed GISCO Working Party 25 October 2001

Conclusions - polygon-based statistics Advantages of the zoning design approach Software available and technical feasible for the whole of Europe Theoretically can be used for any level of geography During the tests some problems with the great numbers of input areas - the more areas the greater the number of possible combinations and thus the more time needed for processing Face the definition of optimality (optimisation of an objective function) Using multiple objective functions makes it more difficult to find a good optimum GISCO Working Party 25 October 2001

Conclusions - both approaches Both methods propose their own way to harmonise data production To perform well both approaches: - Need the provision of a ‘good’ initial set of areal units - Need to incorporate confidentiality restrictions (for the dissemination of statistics) - Face the definition of optimality (e.g. optimal grid-size, optimisation of an objective function) Need to deal with limitations due to processing speed and disk space Need a good GIS software environment GISCO Working Party 25 October 2001

Conclusions - future studies Tests with more /different data sets -different variables -different kind of georeferences -different scale of analysis Development of methods -polygons to grids -AZM Case studies -to compare results with those by administrative areas -to make analysis which are not possible by administrative areas GISCO Working Party 25 October 2001

marja.tammilehto-luode@stat.fi philippe.guiblin@ons.gov.uk GISCO Working Party 25 October 2001