Creating a new framework for census workplace data David Martin 1 , - - PowerPoint PPT Presentation

Creating a new framework for census workplace data

David Martin1, Samantha Cockings1, Andrew Harfoot1, Bruce Mitchell2, Ian Coady2

1University of Southampton, 2Office for National Statistics

NTTS Conference, 11 March 2015

Presentation overview

• 2001 and 2011 census output areas/automated zone design
• Residential vs workplace populations
• Creating 2011 census workplace zones and workplace data
• National classification of workplace zones

Census output areas

• Census output areas (England, Wales, Northern Ireland)

designed to reflect residential locations

• Automated zone design procedure first used for output area

creation in 2001, then again in 2011

• Building blocks representing smallest postcodes built from

Thiessen polygons and boundary data around addresses

• Iterative recombination of building blocks and evaluation

against objective functions, retaining best combinations – Population thresholds and targets, compactness statistics, intra-area correlations

3

4

Photos: David Martin

Why worry about non-residential population?

• Understanding workplace population patterns is of

particular relevance where population “exposure” is concerned: e.g. delivery of daytime services, transport planning or emergency planning

• Distribution of workplace population/ workplaces is very

different

• Extraction of different population bases from place of

residence and place of work statistics – usual residents, workplace population and “daytime” population

• Only four workplace tables produced for 2011 output areas,

NOT e.g. mode of travel to work, industry classification

5

0.00% 20.00% 40.00% 60.00% 80.00% 100.00% 120.00% 20000 40000 60000 80000 100000 120000 140000 160000 Frequency Workplace population

2001 OA workplace population totals (mean residential population 300)

Frequency Cumulative %

Basic logic for workplace zones

• Some output areas have very large workplace populations

and can be subdivided into two or more workplace zones

• Some are suitably sized and can also be workplace zones
• Some have very small workplace populations and need to

be combined with others to create a workplace zone

• This can be set up as an automated zone design problem

and implemented using procedures very similar to those used for output areas

• Control over size allows much more workplace data to be

published: 25 tables at workplace zone level

Workplace zone design

• Workplaces need to be protected from disclosure risk:

– 3 or more postcodes and 200 or more workers – homogeneity based on industry type – spatial compactness

• Boundaries for subdivision built from postcode Thiessen

polygons and map features (not individual workplaces)

• 53,578 workplace zones of which 30.9% from subdivision

(over 625 workers); 4.5% matched output areas; 62% from mergers (below 200 workers); 2.9% complex

Classification of workplace zones: the concept

• Classify workplace zones according to key characteristics

as measured by 2011 Census

• Same concept as geodemographic classification of output

areas, but for workplaces and workplace populations

• Adopt similar method:

– definition of dimensions; selection, transformation and standardization of variables; hierarchical k-means clustering; interpretation and naming of clusters

Workplace zone classification

11

Data and methods

• 504 initial candidate variables; 63 chosen for further

examination; 48 in final selection; particular attention paid to patterns of between-variable correlation

• Box-Cox transformation and range standardisation applied

before K-means clustering

• Development of a 2-tier hierarchy; supergroups divided

into groups (currently being processed)

• Multiple diagnostic approaches to evaluation of clusters

Correlations

Transformations

Evolution of clusters…

7-cluster solution: highly qualified city workers 5,183 workplace zones

+ Black / Asian

• White British

Pub trans Travel > 20km Highly qual, EU origin High density / split OAs Young females Employees IT / finance / professional Higher managerial

Photo: David Martin

7-cluster solution: rural occupations 10,858 workplace zones

White British Agriculture, Forestry, Fishing Work from home / no fixed place V low density Low female Elderly workers Self employed Mining & Quarrying / Manuf / Utilities / Construction Low pub trans

Photo: David Martin

Creating a National Statistics product

• Workplace zones and data published for England and

Wales

• Workplace zones now being constructed for Scotland and

Northern Ireland

• Classification design based on England and Wales data:

detailed subdivision of 7 groups being evaluated

• Finally, roll-out of UK-wide classification of workplace

zones