Commuting Areas in Australia

1
(No Transcript)
2
Commuting Areas in Australia

• Martin Watts, Scott Baum, William Mitchell
  Anthea Bill
• Centre of Full Employment and Equity
• The University of Newcastle
• http//e1.newcastle.edu.au/coffee
• Griffith University

3
Introduction

• ABS Census data available at Census District (CD)
  level.
• BUT highest spatial disaggregation for time
  series LF Survey data Statistical Region Sector
  (SRS SR), a group of SLAs.
• Do ABS SR(S)s represent economic areas so that
  economic social data meaningfully interpreted
  in spatial analysis?
• If not, interpretation of these data is
  compromised, due to Modifiable Areal Unit Problem
  (Coombes, 2002).
• LM policy by particular administration, say
  local Council, on edge of SR could have impact
  across and within SRs.

4
Introduction

• Travel to Work Areas (TTWAs) form of spatial
  disaggregation.
• TTWA geographical area within which a high
  percentage of commuting by residents occurs.
• Site for interplay of labour supply demand
  appropriate area over which labour market
  statistics defined (Coombes, 2002).
• Spatial markets delineated by both costs of
  mobility between jobs and limitations of
  information networks (Hasluck, 1983).
• Each TTWA largely self-contained (closed) from
  rest of economy, but some cross boundary
  commuting.

5
Introduction

• Using measures of closure interaction based on
  commuting patterns, Coombes et al (1986)
  algorithm identifies TTWAs based on UK Census
  data.
• Algorithm adopted with some amendments in
  international studies Spain (Casado-Diaz, 2000),
  New Zealand (Papps Newell, 2002), Denmark
  (Andersen, 2002, USA (Tolbert Sizer, 1996),
  Britain (Coombes et al, 1997), NSW (Watts,
  2004).
• Different terminology employed across studies, so
  Commuting Areas (CAs) used to identify basis of
  disaggregation agnostic about economic
  significance.

6
Introduction

• Generalisation of Coombes algorithm, (European
  Regionalisation Algorithm) recommended by
  Eurostat (1992) for grouping areas in European
  countries (Coombes, 2000).
• Eurostat (1992) principles for defining LLMs in
  declining order
• (1) autonomy maximised self-containment
• (2) homogeneity minimised geographical area
• (3) coherence recognisible boundaries
• (4) conformity alignment with admin boundaries.
• Tradeoff between containment, interaction size
  of geographic area.
• Key? question of extent of intra-CA interaction
  or connectedness.
• CA LM policy should impact mainly within CA,
  (high closure rate) (Coombes Openshaw, 1982, in
  Papps Newell, 2002).

7
Introduction

• Three tasks of this project
• First brief exploration of conceptual basis for
  CAs by reference to literature which analyses
  economic behaviour across spatial LMs.
• Second refine Coombes algorithm to identify CAs
  across Australia by clustering 1311 SLAs from
  2001 Census based on aggregate JTW data, via use
  of Matlab (3 hour program).
• Third (to do), explore suitability of CAs as
  basis for new geography of Australian regions by
  reference to
• calculation of CA descriptive spatial
  statistics for socio-ec LM variables
• spatial modelling of CAs cf SRs
• explore how results inform spatial economic
  theory

8
Theoretical Framework

• Morrison (20052261) 2 views of operation of
  urban LMs.
• Local unemployment in spatial sub-markets of
  urban area due to deficient local labour demand.
  Solution to attract private public jobs within
  close proximity to jobless. (Challenged by LM
  accounts literature significant in-commuting)
• City viewed as single market with employment
  relationships, irrespective of locations of
  residents employment opportunities.
• Spatial distribution of U reflects residential
  clustering effects of a housing market reacting
  to an unequal income distribution.
• Low wage, vulnerable section of labour force can
  only commute a limited distance, due to domestic
  circumstances /or income.

9
Theoretical Framework

• BUT if social networks with weak ties, better
  job-search outcomes because more likely to access
  high quality information from distant parts of
  social system (Granovetter, 1973).
• Increased jobs in specific locations, via
  increased AD, not guaranteed to reduce local
  unemployment rates, due to pervasive job
  competition.
• Once assumption of homogeneous labour is dropped
  than presence of spatially seamless and
  segmented markets becomes a source of insight
  rather than conflict (Morrison).

10
Theoretical Framework

• Derivation of exhaustive mutually exclusive
  spatial groupings challenged (Hasluck, 1983
  Webster, 1997 Newell Perry, 2003 Morrison,
  2005).
• With specific reference to Coombes et al (1986),
  Morrison (2005) conceptual basis for an
  empirical delimitation of the local labour market
  is still rather weak.
• Areas typically defined via commuting patterns
  of those employed in spatial cluster of
  worksites. BUT separate spatial groupings may not
  be well defined, because employees cross local
  geographical boundaries.
• Thus interdependence overlap between adjacent
  CAs. How much cross- boundary commuting is OK?

11
Theoretical Framework

• Contiguous groupings often consolidated to
  overcome boundary crossing, but problem of
  defined areas too large for commuting, so lack of
  internal integration.
• Thus, CAs not internally homogeneous wrt labour
  force stats, so local pockets of high low
  unemployment (Webster, 1997).
• Seamless model job loss job growth in
  different places in CA leads to reconfiguration
  of commuting patterns. Thus spatial LM measures
  across CA more uniform after impact of initial
  shocks diffused over area.
• BUT implicit assumption of intra-CA homogeneity
  challenged by recognition that space matters,
  given, as noted above, (i) spatial concentration
  of low cost housing (ii) limitations on
  commuting of some workers (iii) operation of
  networks.

12
Theoretical Framework

• Corvers et al (2006) compare spatial groupings
  by strength of internal adjustment processes.
• Third, boundaries of spatial groupings may
  exhibit a lack of continuity over time due to
  investments in housing and transport, as well as
  the prevailing local economic conditions.
• This creates problem for conduct of time series
  analysis, whereas reform of admin areas less
  frequent (Newell and Perry, 20034).
• BUT if admin boundaries have no relevance to
  economic processes, then no research benefits to
  using these boundaries.

13
Spatial Grouping Methodologies

• Cluster analysis
• Hierarchical methods.
• Rules-based (Coombes, 20001504).
• Core areas from which final regions are built up.
• Remaining (residual) non-core areas are
  unallocated, unless linked (in)directly to core
  area.
• Final number of regions not known at outset.
• Coombes et al (1986) use non-hierarchical,
  rules-based procedure.
• Upper limit on number of CAs (foci) determined by
  specific criteria in 1st stage of algorithm.
• Foci subsequently dismembered if do not satisfy
  particular criteria, when they are tightened.

14
Coombes Algorithm

• Determine algorithm values. JTW gt 200 km
  omitted.
• Define base spatial units (1311 SLAs in Australia
  2001 Census)
• Identify spatial units to act as LM foci based on
  associated job ratio supply-side
  self-containment criteria. (548 foci)
• Amalgamate foci with high degree of interaction,
  if some foci have inadequate degree of self
  containment (503 foci)
• Expand these foci to form proto CAs by allocating
  other (non) foci to them with which they have a
  high degree of interaction.
• Allocate remaining (residual) non-foci SLAs to
  proto CAs
• Iteratively dismember proto CAs not satisfying
  minimum value of objective function reallocate
  associated SLAs. (394 CAs)

15
Commuting Areas

• Low values for job ratio containment
  parameters (similar parameter magnitude to
  Coombes, but lower min. population)
• 394 CAs (Case A)
• Major capital city and ACT groupings
• 172 SLAs in Brisbane
• 99 ACT
• 62 Melbourne
• 57 Adelaide
• 50 Queensland
• 49 Sydney (excluding Hunter!)
• 274 singleton SLAs

16
Commuting Areas

• 394 CAs (Case A)
• Higher parameter values 313 CAs (Case B)
• 313 foci for CAs (Case B) coincide with 313 of
  394 foci for Case A. Numbers of SLAs in CAs based
  on these 313 foci in Case A less than numbers in
  corresponding CAs under Case B.
• With one exception, orderly breakup of 313 CAs
  occurs when Case A considered.
• Mitchell North belonging to 62 SLA grouping with
  Melbourne Inner in Case A, belongs to a grouping
  around Greater Shepparton Pt A in Case B.

17
End of Talk
18
End of Talk
19
End of Talk
20
End of Talk
21
End of Talk
22
Summary Statistics in preparation

• Simple ave. self containment (grouping) ratios
  90.
• BTRE classification (415) high sc ratios too
  87.
• Both need to be rechecked.
• Weighted average sc ratio of diagonal elements
  of grouped JTW matrix to total JTWs overcomes
  impact of singleton SLAs.
• Interaction measure, based on algorithm
  criterion, is being constructed for all pairwise
  combinations within each grouping.
• Spatial Statistics

23
Concluding Comments

• This paper has provided a preliminary exploration
  of conceptual measurement issues from using
  rules-based algorithm to identify CAs in
  Australia.
• Are these CAs the basis of a new ABS geography?
• Yes if they add to our understanding of spatial
  economic processes, but at this stage this
  question remains unanswered.

24
End of Talk

• End of Talk

25
End of Talk

• Cluster analysis
• Initial areas to final set of regions in one
  step, using similarity of their statistical
  properties, measured by affinity matrix.
• Can specify required number of groupings at
  outset, but cannot ensure all regions meet
  minimum statistical properties.
• Constraint, plus need for contiguity reduces
  available options, so results are likely to be
  sub-optimal.
• Hierarchical methods
• Can impose number of required regions, or minimum
  statistical requirements.
• Contiguity can be imposed which, make the
  grouping sub-optimal, but reduced computational
  demands.
• Major deficiency is initial area groupings
  severely constrain options available later
  (Coombes, 20001505), because preserved
  throughout procedure.

26
End of Talk

• Rules-based (Coombes, 20001504).
• Core areas from which final regions are built up.
• Remaining (residual) non-core areas are
  unallocated, unless linked (in)directly to core
  area.
• Final number of regions not known at outset.
• Other procedures commence with each basic area as
  potential region, so if not linked to others
  during procedure, remains as single area region,
  rather than being unallocated under rules-based
  approach.