Identifying Functional Regions in Australia Using Hierarchical Aggregation Techniques

Identifying Functional Regions in Australia UsingHierarchical Aggregation Techniquesgeor_631 24..41

WILLIAM MITCHELL* and MARTIN WATTSCentre of Full Employment and Equity, University of Newcastle, University Drive, Callaghan, NSW2308, Australia.*Corresponding author. Email: [email protected]

Received 31 March 2009; Revised 19 October 2009; Accepted 19 October 2009

AbstractThis paper continues our work focused on developing a new socio-economicgeography for Australia such that the chosen spatial aggregation of data is basedon an analysis of economic behaviour. The underlying hypothesis is that thedevelopment of a geographical classification based on underlying economicbehaviour will provide new insights into critical issues of regional performance,including unemployment differentials, the impact of industry, infrastructure andchanges in local public expenditure on local labour markets. As a precursor todetailed work on the 2006 Census of Population and Housing data, we establishthe proof of concept in this paper of the Intramax methodology using 2001Journey-to-Work data from the Australian Bureau of Statistics (ABS) for the stateof New South Wales. The functional regionalisation generated by the Intramaxmethod is then tested using ABS labour force data. We compare 2001 ABS Censusof Population and Housing data aggregated by the ABS labour force regions to thesame data aggregated using our functional regions. The results demonstrate thepotential value of this technique for the development of a new geography.

KEY WORDS spatial aggregation; functional economic areas; intramaxalgorithm; spatial clustering

IntroductionThis paper continues our work focused on devel-oping a new socio-economic geography for Aus-tralia such that the chosen spatial aggregation ofdata is based on an analysis of economic behav-iour (see Watts et al. 2006). The underlyinghypothesis that has motivated this work is thatthe development of a geographical classificationbased on an analysis of labour market behaviourwill provide new insights into critical issues ofregional performance, including unemploymentdifferentials, the impact of industry, infrastruc-ture and changes in local public expenditure onlocal labour markets. A systematic understand-ing of levels of interaction between neigh-bouring areas, will facilitate an assessment ofthe adequacy of the administrative geographical

demarcations currently used by the AustralianBureau of Statistics (ABS) to collect and dis-seminate their labour force data.

Recent Australian studies have analysedspatial patterns of unemployment, housingand related socio-economic phenomena usingadministratively-defined Australian StandardGeographical Classification (ASGC) spatialaggregations typically at the Statistical LocalArea (SLA) and/or Statistical Region (SR) level(for example, O’Connor and Healy, 2002;Lawson and Dwyer, 2002; Baum et al., 2005;Mitchell and Carlson, 2005; Yates, 2005; Yateset al., 2006a; 2006b; Mitchell and Bill, 2006;Gregory and Hunter, 1995). Most Australianresearchers are reluctant to acknowledge thatthe interpretation of these spatial data can be

24 Geographical Research • February 2010 • 48(1):24–41doi: 10.1111/j.1745-5871.2009.00631.x

mailto:[email protected]

compromised by the Modifiable Areal UnitProblem (MAUP), although this problem haslong been recognised by geographers. Openshaw(1984, 3) says that ‘the areal units (zonal objects)used in many geographical studies are arbitrary,modifiable, and subject to the whims and fanciesof whoever is doing, or did, the aggregating’ andresulting analyses are fraught. In short, thespatial groupings must be justified so theseaggregations should be based on informedchoice. However the principles for groupingareas are not clearly defined. Indeed the‘optimal’ grouping does not maximise an objec-tive function, otherwise formal mathematicalmethods might be appropriate. Rather theproblem is somewhat idiosyncratic and, ignoringany distance constraints, there are a very largenumber of possible solutions, given that none ofthe following are specified a priori: (a) thenumber of regions; (b) the number of SLAs ineach region; and (c) the allocation of areasbetween the regions. This means that some formof algorithm is required, which is inevitablysomewhat arbitrary in its specification.

Just as geographical regions may be definedby physical features, we hypothesise that ameaningful socio-economic geography shouldbe defined by socio-economic features of space.It is most unlikely that that these ‘regions’ willcorrespond exactly to a demarcation basedon administrative/political criteria. Significantissues arise when erroneous geography is used.First, a poorly delineated geography invokesmeasurement error. Thus, a local measure suchas SLA unemployment, may be unrelated tosocio-demographic and policy variables at asimilar scale, and lead to spurious causalitybeing detected and misguided policy conclu-sions being drawn. Second, analysing errone-ously aggregated spatial data with standardstatistical tools will yield results that may notonly lack economic meaning but also suffer biasdue to spatial correlation. Based on an earlierpilot study (Watts et al., 2006) we hypothesisethat studies relying on ABS administratively-based Census of Population and Housing areasor the Labour Force Survey regions producemisleading inferences when applied to socio-economic analysis or policy. Watts et al. (2006)found significant spatial correlation in keylabour market variables and sensitivity to spatialaggregation, when they compared 2001 AGSCgeography with experimental commuting areasgenerated using the 2001 Census of Populationand Housing data.

Unfortunately, only limited attempts havebeen made in Australia to create such a ‘space’(see Watts et al., 2006). Journey-to-work (JTW)data provides information about the interactionbetween a large number of spatial units and is auseful basis for defining a functional regionalisa-tion. The theoretical basis for demarcatingregions based on commuting behaviour is out-lined in Watts et al. (2006). It is applicable to anyof the possible aggregation methodologies thatare available. A region is conceived as a geo-graphical area within which there is a highdegree of interactivity (commuting by residents)and is thus the appropriate spatial scale to capturethe interplay between labour supply and demandin a particular localised setting. These spatialmarkets result from both costs of mobilitybetween jobs and the limitations of informationnetworks (Hasluck, 1983). Employers andworkers who interact within a functional regionare assumed to be well informed and able torespond quickly to changes in market conditionsrelative to those outside any particular region.While Hasluck (1983) is critical of such anattempt to create regionalisations; on balance, wesupport Green (1997) who sees commuting clus-ters as revealing the boundaries of local labourdemand and supply and hence as a sound basisfor an ‘an alternative geography’ for labourmarket analysis.

Different terminology has been employed toidentify these areas including Commuting Areas,Local Labour Market Areas, Functional LabourMarket Areas and Commuting Zones. Early on,Berry (1968) referred to these aggregations asfunctional regions. In this paper and consistentwith the extant literature (on intramax tech-niques) we use the term functional region todescribe our regional aggregations based on JTWdata. In this paper, and consistent with our con-jecture above, we seek alternative aggregationsof the JTW data which reflects economic behav-iour (commuting interaction) rather than admin-istrative structures.

Previous work by Watts (2004) and Wattset al. (2006) used a non-hierarchical, rules-baseddemarcation method first developed by Coombeset al. (1986) to determine a new ‘behavioural-based’ geography (for a detailed description ofthe method see Coombes et al., 1986, 948–52and Papps and Newell, 2002, 9–14). In summary,the Coombs et al. algorithm is based on: (a) the apriori specification of the magnitudes of anumber of parameter values; and (b) a complexsequence of stages in which areas are identified

W. Mitchell and M. Watts: Identifying Functional Regions in Australia Using Hierarchical Aggregation Techniques 25

© 2009 The AuthorsJournal compilation © 2009 Institute of Australian Geographers

as foci according to particular criteria, which arethen combined according to a weighted interac-tion function and then further combined intotemporary or proto groups of areas, again byreference to the interaction function and othercriteria. These proto-groups are then entirely dis-membered if the associated value of an objective(spline) function does not exceed a critical value.

There are two obvious shortcomings of thealgorithm. First, the choice of foci and proto-groups as the sequence of steps is implemented isdependent on the set of arbitrarily specifiedparameter values. The sensitivity of the solutionto these values is hard to gauge without extensiveexperimentation. Second, the dismembermentprocess would appear to generate a final set ofgroupings with numerous singleton groups, butalso some very large groups, at least using Aus-tralian SLAs. A simplified version of the 1986algorithm was first used on the 2001 UK Censusdata (Bond and Coombes, 2007), but these defi-ciencies appear to remain (Watts, 2009).

Figure 1 presents the mapping for New SouthWales derived from Watts (2004). Spatial auto-correlation measures were calculated using

unemployment rates at the postal area level in2001, to examine the regions generated by Watts(2004), termed ‘commuting areas’ or local labourmarkets (LLMs). Global spatial autocorrelationindices are formal measures of the extent towhich near and distant observation items arerelated. Global statistics can be decomposed toprovide local measures of spatial association(LISAs), which reveal statistically significantclusters of above average values (hotspots) andstatistically significant below average concentra-tions (coldspots) for the phenomena under inves-tigation. LISA maps at the postal area levelrevealed considerable spatial heterogeneity inlabour force outcomes, and highlighted a keymotivation for the development of the new geog-raphy. They revealed that Statistical Region (SR)boundaries which roughly approximate ABSlabour force geography conflate areas with sta-tistically significant heterogeneity (that is bothhotspots and coldspots), while finer commutingareas generally did not. However, as the Sydney‘commuting area’ comprised the entire SydneyMetropolitan Area (see Figure 2) the ‘commut-ing area’ geography seems to miss much of the

Figure 1 Functional regions for NSW derived using the rules-based Coombes approach.

26 Geographical Research • February 2010 • 48(1):24–41


significant (non-random) heterogeneity in labourmarket outcomes within Sydney. Notably itcreated a geography which conflated unusuallylow and statistically significant clustering ofunemployment rates in the eastern and northernsuburbs and clusters of unusually high unem-ployment rates in Sydney’s inner and outer west.1

In this paper we take an alternative approach tofunctional regionalisation and deploy the Intra-max method. JTW flows between areas can bedepicted as a square matrix with each row denot-ing an origin (residential location) and eachcolumn representing a destination (workplacelocation) (see Table 1). The Intramax method is ahierarchical clustering algorithm (Masser andBrown, 1975) that maximises ‘the proportion ofthe total interaction which takes place within theaggregation of basic data units that form thediagonal elements of the matrix, and thereby tominimise the proportion of cross-boundarymovements in the system as a whole’ (Masserand Brown, 1975, 510). Masser and Scheurwater(1980, 1361) say that the ‘intramax procedure isconcerned with the relative strength of interac-tions once the effect of variations in the size ofthe row and column totals is removed . . . relativestrength is expressed in terms of the differencebetween the observed values and the values thatwould be expected on the basis of the multipli-cation of the row and column totals alone.’

The Intramax method is concerned with howthe aggregation impacts on interaction flows(journeys) across the regional boundaries. Themain diagonal elements of the JTW matrix rep-resented in Table 1 (at any stage of aggregation),capture the journeys that begin and end in thesame region, whereas the off-diagonal elementsshow journeys that cross regional borders.

Masser and Brown (1975, 509) say that ‘the mostimportant distinction that must be made in thegrouping procedure is between the proportion ofinteraction in the diagonal as against the non-diagonal elements of the basic flows matrix’(emphasis in original).

Barros et al. (1971, 140) refer to the ‘strengthof interaction’ as the proportion of total journeysthat cross regional boundaries. Clearly, as weaggregate smaller regions into larger (functional)regions, the proportion of interactions that crossboundaries should decline and a rising propor-tion of interactions thus would be consideredintra-regional.

As a way forward, we seek to define our func-tional economic regions (based on labour marketcommuting behaviour), by maximising ‘the pro-portion of the total interaction which takes place’between the individual regions and thus minimis-ing ‘the proportion of cross-boundary move-ments in the system as a whole’ (Masser andBrown, 1975, 510).

The results reveal the Intramax technique to bea very useful for understanding the operation ofthe Sydney labour market and demarcating func-tional regions. At each stage of the clusteringprocess fusion occurs between regions in such away as to maximise commuting flows or interac-tion – providing valuable insight into those SLAswhose labour markets are most ‘linked’ inmetropolitan and non-metropolitan New SouthWales. Overall the technique collapses many ofthe standard labour force regions (used by theABS in the dissemination of its statistics) in met-ropolitan Sydney (but preserves established dis-tinctions between east and west) and splits manynon-metropolitan labour force regions, particu-larly around major regional centres. An applica-

Table 1 JTW flow matrix with j regions.

DestinationOrigin

Region 1 Region 2 . . . Region j Total

Region 1 1 to 1 1 to 2 . . . 1 to j a jj

1∑Sum of flows out of Region 1

Region 2 2 to 1 2 to 2 . . . 2 to j a jj

2∑. . . . . . . . . . . . . . . . . .Region j j to 1 j to 2 . . . j to j a jj

j∑

Total aii

1∑Sum of flows into Region 1

aii

2∑ . . . aiji

∑ n aijji

= ∑∑Total Interaction



tion to 2001 Census of Population and Housingdata reveals that, on average, regional unemploy-ment rates are higher and labour force participa-tion rates are lower when applying this newgeography. The enhanced behavioural homoge-neity of the functional regions also reduces theintra-region dispersion (measured by the stan-dard deviation in unemployment rates). Thissuggests that the strength of commuting flowsbetween contiguous areas is capturing the eco-nomic interaction between them.

The paper is organised as follows. The nextsection describes the data used in this study. Wethen outline the Intramax method of functionalregion demarcation. The following section pre-sents the regionalisation based on the applicationof the Intramax method and analyses some of theimplications of the resulting geography. Con-cluding remarks follow.

DataAustralian Bureau of Statistics Journey to Workdata from the 2001 Census of Population andHousing was used at the Statistical Local Arealevel for New South Wales and its cross borderinteractions (the area chosen to allow a directcomparison with Watts, 2004, which used theCoombes approach).2 Using the 2001 JTW datacube from the ABS Census of Population andHousing, we deleted the following destinationlocations (columns): Sydney (undefined); nofixed address; migratory and off-shore; unde-fined New South Wales; not stated and not appli-cable. Lord Howe Island was also deleted. Inaddition, the corresponding residential (origin)locations were deleted (rows). The remainingJTW matrix thus had 197 SLAs. To protect con-fidentiality small flows are randomised by theABS. Consequently all entries of less than 6 wereset to zero.

To help in the explication of the Intramax tech-nique, Table 1 provides a schematic representa-tion of the square JTW flow matrix where therows are designated as origins and the columnsare destinations.

The summary details of the 197 ¥ 197 matrixfor NSW journey-to-work flows between SLAsare detailed in Table 2. Zero elements constitute88 per cent of the JTW matrix (no commutingbetween pairs of places) which means that the‘network depicting the commuting patterns isrelatively unconnected’ (Brown and Holmes,1971, 61), with many SLAs not being linked bycommuting flows.3

The Intramax methodThe Intramax method considers the size of theinteraction (JTW flows) to be ‘of fundamentalimportance’ (Masser and Brown, 1975, 510). Toexpress this concern the method considers the‘interaction matrix’, that is, the JTW matrix, tobe a ‘form of contingency table’ and then formu-lates the ‘objective function in terms of the dif-ferences between the observed and the expectedprobabilities that are associated with these mar-ginal totals’ (Masser and Brown, 1975, 510).

If we view Table 1 as a contingency table thenthe expected values of each element are derivedas the product of the relevant column sum (Equa-tion 3) multiplied by the ratio of the row sum(Equation 2) to total interaction (Equation 4).For example, the expected flow out of Region 2into Region 1, a21 in Table 1, where aij is theelement in row i and column j of the contingencytable (JTW matrix), is given as

a a

a

aa a ni

i

jj

ijji

ii

jj

21 1

2

1 2* = =⎛⎝⎜

⎞⎠⎟∑

∑∑∑ ∑ ∑ (1)

This is the ‘flow that would have been expectedsimply on the basis of the size of the row andcolumn marginal totals’ (Masser and Brown,1975, 512).

The row sum of the JTW matrix is

a ai ijj

* = ∑ (2)

The column sum of the JTW matrix is

a aj iji

* = ∑ (3)

The total interaction n is the sum of the row sums

n aijji

= ∑∑ (4)

Table 2 Summary JTW matrix characteristics andinteractions

Number of regions 197Percentage of zero elements

in JTW matrix (%)88

Total interaction (trips) 2,449,385Intra-regional interaction

(trips) – Trace of matrix1,099,613

Percentage of intra-regionaltrips (%)

44.9



The null hypothesis for independence betweenthe row and column marginal totals of a contin-gency table is defined as:

H a

a a

n

a a

no ij

ijj

iji i j: * * *= =

∑ ∑(5)

If the grand total of the flows is normalised suchthat n = 1 and a a aij i j

* = * * then Masser and Brown(1975, 512) note that ‘the difference betweenobserved and expected values (a aij ij− *) for theflow between zone i and zone j may be taken asa measure of the extent to which the observedflow exceeds (or falls below) the flow that wouldhave been expected simply on the basis of thesize of the row and column marginal totals.’

The objective function of the hierarchical clus-tering algorithm using a non-symmetrical JTWmatrix, is defined as

max ( *) ( *),I a a a a i jij ij ji ji= − + − ≠ (6)

In the Flowmap software4, which was used toperform the Intramax, Equation (6) is modifiedas follows:

max ,I T O D T D O i jij i j ji j i= ( ) + ( ) ≠ (7)

Where Tij is the interaction between the originSLA i and destination SLA j; Oi is the sum of allflows from origin i to j; and Dj is the sum of allflows from destination j to origin i.

In relation to Equation (7), Goetgeluk and deJong (2005, 9) say that ‘the proportional amountof within group interaction is maximised in eachstep of the procedure . . . two areas are fused thathave the strongest relative relations’ in terms ofcommuting flows.

At each stage of the clustering process, fusionoccurs between the regions that have the stron-gest commuting ties (interaction), as representedby Equation (7). The stepwise procedure thencombines the clustered interaction and the matrixis reduced by a column and a row. The remainingactual and expected commuting flows arere-calculated and the i,j combination of regionsmaximising (7) is again calculated, and so on.With N regions (197 in our study) after N-1 (196)steps, all regions would be clustered into a singlearea (the state of New South Wales) and by con-struction, all interaction would be intra-zonalwith one matrix element remaining. In contrast to

the Coombes algorithm, there is no dismember-ment of groups of regions during the operation ofthe algorithm.

Masser and Brown (1975) place a contiguityconstraint on the maximisation process to elimi-nate the possibility that clusters between non-contiguous regions would form. There is verylittle chance that contiguity would not be satis-fied in the Intramax clustering process. So asintra-zonal flow increases (at higher levels of thedissolution process) the newly forming clustersare almost certain to be contiguous. However,with respect to commuting, there is no logicalreason why two non-contiguous regions couldnot belong to the same local labour market. TheIntramax algorithm as well as other algorithms,would not identify that, in these circumstances,commuting entailed crossing a boundary out ofthe region and then re-entering the region, sinceonly the identity of the origins and destinationswould be recorded. Peculiarities of the housing,occupational and transport patterns overlayingemployability could generate such a result. In ourdata, the contiguity constraint is not enforced butthe results deliver functional economic regionswith the constituent SLAs being contiguous.

ResultsRegions with large relative JTW flows (as com-pared to the expected flows) are fused in the firststages of the clustering process and those withsmaller relative intra-region JTW flows are fusedin the later stages. Figure 2 produces a dendro-gram for the Sydney MSR (excluding theHunter) and Figure 3 produces a dendrogram forthe Gosford-Hunter region.

Recall from the previous work of Watts et al.(2006) that the Coombes methodology, whileyielding significant aggregations of SLAs, didnot generate fine spatial demarcations of func-tional economic regions within the Sydney MSR.In this context, Figures 2 and 3 reveal some veryinteresting patterns. In the first stages of theaggregation process, the SLAs of Burwood andStrathfield fuse to form the basis of an inner-westSydney cluster with the SLAs of Ashfield,Concord and Drummoyne (all of which are in theInner Western Sydney Statistical Subdivision inthe 2001 ASGC) and then collect the LeichardtSLA soon after. Around 58 per cent of all theJTW flows in and out of these clusters are intra-zonal for this new aggregation. The SLAs ofCanterbury and Marrickville also fuse early andincorporate the Bankstown SLA soon after toform an inner south-west Sydney cluster. Around



Figure 2 Dendrogram for Sydney-Illawarra.



Figure 2 (Continued ).

Figure 3 Dendrogram for Central Coast and Hunter functional regions.



55 per cent of the JTW flows in and out of theseclusters are intra-zonal. Figures 4 to 7 map thesubsequent clustering whereby the inner westand inner south-west fuse to form a new clusterand around 65 per cent of its total flows areintra-zonal.

To the south and east of the Sydney CBD,other clusters form. The Botany Bay SLA ini-tially fuses with the SLAs of Randwick andWaverley with the Woollahara SLA. At this pointintra-zonal flows in each constitute around 48per cent of total flows. The two clusters fuse(52 per cent intra-zonal) then fuse again with theSouth Sydney SLA. In turn, this grouping fuseswith Sydney CBD (the SLAs of Sydney-Innerand Sydney Remainder) and the new aggrega-tion has around 65 per cent intra-zonal interac-tion. At a later stage of the grouping this entireblock forms a new cluster and the proportion ofintra-zonal JTW flows are around 78 per cent forthis new group. There are still cross-boundaryflows (22 per cent) but there is a strong sense of‘intra-action’ in this grouping.

To the south-west, the SLAs of Kogarah andHurstville fuse first as do the SLAs of Suther-

land Shire-East and Sutherland Shire-West. TheRockdale SLA joins the first of these clustersand together they fuse with the Sutherland Shirecluster to form a 5 SLA cluster, quite distinctfrom the near inner south-west and inner-westand south and east clusters discussed above.This larger southern cluster has around 57 percent of its total flows as intra-zonal. At a muchlater stage of the aggregation the southerncluster fuses with the previous cluster incorpo-rating Sydney CBD, inner west and inner south-west and inner south and east to form a clusterwhere 80 per cent of the total flows are intra-zonal.

In the north, the SLA of Hornsby looks east tothe Ku-ring-gai SLA and to the north-east theSLAs of Hunter’s Hill and Ryde fuse (both clus-ters containing 50 per cent intra-zonal flows).These north and north-east clusters in turn fusewith about 60 per cent of their flows being intra-zonal. Closer to Sydney, on the north side, theSLAs of Lane Cove and Willoughby fuse as dothe SLAs of Mosman and North Sydney (againwith about 50 per cent of their total flows beingintra-zonal). These clusters fuse and then fuse

Figure 4 Comparison of ABS Labour Force Region geography and Functional Region Geography (Cluster 2), NSW (Source:ABS, JTW Custom matrix, 2001 and ABS, SLA to LF concordance 2004).



again with the north-east and northern clusterwith about 60 per cent of their flows beingintra-zonal.

Interestingly, nearby north-side SLAs, Manlyand Warringah (east of the Pacific Highway) fuseearly and then attract the Pittwater SLA which isfurther north along the coast. Together this newcluster of 3 northern coastal SLAs represents adefinable cluster containing about 50 per cent ofintra-zonal flows. They subsequently fuse withthe other north/north-east cluster where about 60per cent of total flows are intra-zonal.

At a later stage of the clustering process thenorth/north-west regions fuse with cluster south/south-west/east including the Sydney CBD toform a larger cluster where the proportion ofintra-zonal flows exceeds 80 per cent. This is astrongly defined cluster.

A further interesting point is that the SLAs inthe outer western parts of the Sydney Metropoli-tan Statistical Region – the SLAs of Holroydand Parramatta fuse and then look south to fusewith the Auburn SLA. Similarly, the SLAs ofBlacktown-North and Blacktown-South fuse first

then look north-east and fuse with the BaulkhamHills SLA. These two larger clusters in turnaggregate and the percentage of intra-zonal flowsto total flows for this new cluster exceeds 55 percent.

Nearby, the SLA of Blacktown-South Westlooks further to the west and fuses with the PenrithSLA and this cluster fuses further to the west withthe Blue Mountains SLA and to the north with theHawkesbury SLA. About 55 per cent of the totalflows of this new cluster are intra-zonal. Thisnorth-western/western cluster aggregates into theHolroyd/Parramatta/Blacktown-North and Southand Baulkham Hills cluster with 65 per cent ofintra-zonal flows.

Further to the south-west, the SLAs ofCamden and Wollondilly fuse, then fuse with theCampbelltown SLA. The SLAs of Liverpool andFairfield also fuse. These two south-westernclusters aggregate to form a well-defined south-west cluster with about 55 per cent intra-zonalflows.

In turn, at the next significant grouping stagethis south-west grouping aggregates with the

Figure 5 Comparison of ABS, Labour Force Region and Functional Region Geographies (Cluster 2), Sydney MetropolitanRegion (Source: ABS, JTW Custom Matrix, 2001 and ABS, SLA to LF concordance 2004).



north-west/outer-west cluster to form a stronglydefined grouping in outer Sydney and about 70percent of total flows are intra-zonal. When thissouth-west cluster eventually aggregates with theother large Sydney cluster some 90 per cent oftotal flows for this large region are intra-zonal.

The Illawarra SLAs are quite distinct. TheSLAs of Kiama and Shellharbour fuse as do theSLAs of Shoalhaven-Part A and Shoalhaven-PartB. The Kiama/Shellharbour cluster fuses with theWollongong SLA and then with the Shoalhavenpair.At this level of aggregation, intra-zonal flowsconstitute about 70 per cent of total flows for thisIllawarra region. This region clusters with thelarge Sydney region at around 92 per cent intra-zonal, just after the big outer west/south-westcluster fuses with the larger Sydney cluster.

Of interest is the way in which the SLA ofWingecarribee (part of the Illawarra StatisticalDivision) stands separate with around 78 per centof its total flows being intra-zonal and looksfurther south into the rural SLAs of Boorowa,Gunning, Yass, Queanbeyan, and the ACT toform a larger cluster.

Another interesting result is associated withthe SLAs belonging to the Gosford and Hunterregion (see Figure 3). The SLA of Gosford istypically considered to be part of the Sydneymetropolitan area (Watts et al., 2006) but theseresults challenge that assertion. The SLAs ofGosford and its northern neighbour, Wyong, ini-tially fuse to make a cluster with 52 per cent oftotal flows intra-zonal. The SLAs of Cessnockand Singleton fuse as do the SLAs of Dungogand Maitland and these two clusters fuse to forma new grouping (with 50 per cent of its total flowsintra-zonal). The SLA of Port Stephens lookswest to this cluster first. The SLAs of NewcastleInner and Newcastle Remainder fuse first thenlook south to fuse with the Lake Macquarie SLA.This aggregation then fuses with the lowerHunter cluster (including the SLA of PortStephens). This new functional region has about58 per cent of its total flows intra-zonal). TheUpper Hunter SLAs – Muswellbrook and Scone– fuse (48 per cent intra-zonal) and fuse with thelower Hunter/Lake Macquarie cluster (74 percent intra-zonal).

Figure 6 Comparison of ABS Labour Force Region geography and Functional Region Geography (Cluster 2). NSW (Source:ABS, JTW Custom Matrix, 2001 and ABS, SLA to LF concordance 2004).



The Upper Hunter SLA of Gloucester looksnorth to the fusion between the SLAs of Tareeand Great Lakes and these in turn form a mid-north coast/hinterland cluster with the SLAs ofHastings, Kempsey and Nambucca (88 per centintra-zonal). They further aggregate north ratherthan south. These results clearly indicate that theSLA of Gosford looks northwards rather thansouthwards in terms of its dominant commutinginteraction.

To render the concept of functional regionsoperational, some level of clustering (number ofsteps) has to be chosen and the resulting region-alisation defined. The exact point at which westop the algorithm is a matter of judgement andcannot be determined in any rigid way.5 A con-vention adopted in the literature is to define a‘stop criterion’ as some level of clustering(number of steps) where homogeneity within acluster is lost. Goetgeluk (2006, 11) states that alarge increase in the intra-zonal flows during thefusion process does not generally indicate ‘amerger of two rather homogenous zones.’ The‘stop criterion’ would thus use the regionalisation

that was defined ‘just before the high increase inintra-zonal flows’.

We can determine this point from an exami-nation of the dendrogram and the fusion report(summarised in Table 3) provided by the Intra-max process. Prior to the start of the fusionprocess, Table 2 tells us that the total number ofintra-zonal flows is 1,099,613, representingmarginally under 45 per cent of all the commut-ing flows (trips). Table 3 presents a truncatedversion of the fusion report for the Intramaxprocess starting at Step 69 (which involved thefusion of the two Newcastle SLAs and was thefirst significant jump in the cumulative intra-zonal interaction). The large jumps or ‘breakpoints’ in the fusion output occur when thereare significant increases in the cumulative intra-zonal interaction as a result of a fusion. Thesestart occurring from Step 168, then Steps 170,174, 176 and 182. In terms of Steps 168 and170 these involve the inner west clustering withthe inner south (168) and then the SutherlandShire clustering with the inner south (170).Further analysis is required to determine

Figure 7 Comparison of ABS Labour Force Region geography and Functional Region Geography (Cluster 2), SydneyMetropolitan Region (Source: ABS, JTW Custom Matrix, 2001 and ABS, SLA to LF concordance 2004).



whether these pre-clustered regional aggrega-tions are interactive enough to be classified asfunctional regions in their own right. Wesuspect not at this stage and, as a consequence,we considered Step 173 to be a reasonable stoppoint such that 78 per cent of flows were nowconsidered intra-zonal by this regionalisationand 24 clusters emerge as distinct functionalregions. In making this decision, we decidedthat the fusion determined by Step 174 (outersouth-west with west) was creating an aggrega-

tion that probably blurred important labourmarket differences. This is pure conjecture atthis stage and further analysis of occupation andindustry data is required to test its veracity.

We identified 3 useful regionalisations thatwill provide the basis for further analysis andtesting. Cluster 2 (24 regions and 78 per centcumulative intra-zonal flows), which is our pre-ferred stopping point in this paper, is the inter-mediate level. Cluster 1 defined 40 functionalregions with cumulative intra-zonal flows mar-

Table 3 Fusion report for the Intramax aggregation process, NSW SLAs 2001.

Step Dissolved Area Enlarged Area PercentageIntrazonalInteraction

CumulativeIntrazonalInteraction

No ofRegions

0 44.89 44.89 19769 Newcastle-Inner -> Newcastle-Remainder 0.3% 47.4% 12887 Shellharbour -> Wollongong 0.7% 48.5% 11092 Lake Macquarie -> Newcastle-Remainder 1.3% 50.3% 10593 Sutherland Shire-West -> Sutherland Shire-East 0.4% 50.8% 104101 Wyong -> Gosford 0.5% 51.9% 96105 Pittwater -> Warringah 0.4% 52.4% 92106 Camden -> Campbelltown 0.3% 52.7% 91122 Blacktown South-West -> Penrith 0.3% 53.9% 75127 Liverpool -> Fairfield 0.4% 54.6% 70133 Holroyd -> Parramatta 0.4% 55.2% 64134 Blue Mountains -> Penrith 0.3% 55.5% 63136 Willoughby -> North Sydney 0.4% 56.2% 61138 Waverley -> Randwick 0.4% 56.7% 59140 Cessnock -> Newcastle-Remainder 0.9% 57.8% 57141 Hurstville -> Sutherland Shire-East 0.6% 58.4% 56142 Baulkham Hills -> Blacktown South-East 0.4% 58.8% 55144 South Sydney -> Randwick 0.9% 59.7% 53148 Auburn -> Parramatta 0.3% 60.1% 49151 Bankstown -> Canterbury 0.3% 60.7% 46152 Campbelltown -> Fairfield 0.7% 61.4% 45154 Blacktown South-East -> Parramatta 1.5% 62.9% 43157 Ryde -> Hornsby 0.4% 63.3% 40158 Sydney-Inner -> Randwick 1.8% 65.2% 39161 North Sydney -> Hornsby 1.2% 66.5% 36163 Penrith -> Parramatta 2.0% 69.2% 34165 Warringah -> Hornsby 1.1% 70.4% 32168 Canterbury -> Randwick 3.7% 74.0% 29170 Sutherland Shire-East -> Randwick 3.8% 77.9% 27173 Shoalhaven-Pt A -> Wollongong 0.1% 78.0% 24174 Fairfield -> Parramatta 2.1% 80.2% 23176 Hornsby -> Randwick 6.3% 86.4% 21182 Parramatta -> Randwick 10.4% 96.9% 15183 Gosford -> Newcastle-Remainder 0.4% 97.2% 14190 Wollongong -> Randwick 0.9% 98.2% 7191 Newcastle-Remainder -> Randwick 1.3% 99.6% 6195 Hastings-Pt A -> Randwick 0.2% 99.8% 2196 Orange -> Randwick 0.7% 100.00% 1

Source: ABS, JTW custom matrix, 2001. Percentages are rounded up so may not add up to 100%.



ginally over 63 per cent, whereas Cluster 3defined 15 regions with cumulative intra-zonalflows marginally under 97 per cent. The judg-ment as to whether one of these Clusters is usefulwould be based on whether the resulting betweenregional variance of labour market variables,such as unemployment and labour force partici-pation, represented a significantly greater shareof the total variance than those defined accordingto administrative boundaries. However thesemeasures are aspatial.

The maps in Figures 4 to 7 compare the Func-tional Regions (Cluster 2) generated above withthe standard ABS Labour Force areas (whichroughly conform to Australian Standard Geo-graphical Classification Statistical Regions),used to disseminate regional statistics from theLabour Force Survey (see ABS, 2004).6 In aggre-gate for New South Wales approximately thesame number of regions exists under bothsystems: 21 Labour Force regions and 24 Func-tional Regions. However the spatial distributionof these regions is significantly different betweenthe two geographies.

Three general points can be made: (a) Manymore regions are generated in non-metropolitanNSW when using the Functional Regions, par-ticularly in the north and north-west, comparedto the standard Labour Force areas; (b) Largeregional towns (such as Wagga Wagga, Albanyand Dubbo) emerge as epicentres of functionalregions; and (c) Fewer functional regions emergein Sydney than currently exist as the standardABS Labour Force areas.

Other more specific comments include:

1. The Functional Region geography dividesInner Sydney in two and combines EasternSuburbs and Inner Sydney to form oneeastern Sydney suburban Functional Region(see Figures 3 and 5);

2. Canterbury Bankstown, Inner westernSydney and eastern Inner Sydney combineto form an inner-west functional region;

3. Fairfield Liverpool and outer South WesternSydney Statistical Regions (SRs) combine toform an outer South Western Sydney func-tional region;

4. Central Western Sydney SR, North WesternSydney SR and Central Northern Sydney SR(east side) combine to form one WesternSydney functional region;

5. Wollongong Statistical Region Sector andNowra SR form one functional region alongthe south coast;

6. Part of Illawarra SR (excluding Wollongong)combines with Goulburn; Wingecarribeecombines with the South Eastern SR to formone functional region;

7. Newcastle statistical region sector andHunter combine to form one functionalregion;

8. The Hunter SR divides at Gloucester/GreatLakes to combine with Port Macquarie toform a mid-north coast functional region(extending to Nambucca);

9. A functional region emerges in Dubbo andsurrounds (formerly part of the Northern,Far-West North Western SR), also incorpo-rating Merriwa, formerly in the Hunter SR;

10. Orange and Bathurst and surrounds (part ofthe Central West SR) form one functionalregion;

11. Murray-Murrumbidgee SR splits into fourseparate functional regions: one situated inthe far north west in Wentworth and sur-rounds; one encompassing Hay and theMurray Murumbidgee region; one situatedin and around Albury and one in WaggaWagga and surrounds;

12. The Far West North Western SR is brokeninto eight Functional Regions. Tenterfieldcombines with Richmond Tweed (Lismore)SR to form one functional region; a func-tional region emerges around Armidaleand one around the Tamworth, as well asseparate functional regions encompassingDubbo and Broken Hill, and their sur-rounds. In the south east of the CentralWest SR a functional region emergesaround Bathurst. Cobar (Far West NorthWest) combines with Lachlan, Parkes,Bland, Forbes, and Weddin all regions for-merly in the Central West SR to form onefunctional region. Finally part of the North-ern SR forms one functional region encom-passing the towns of Walgett, Narrabri,Moree Plains, Bingara and Yallaroi.

13. Murrurundi (formerly part of the HunterSR), merges with part of the NorthernFar-North Western (encompassing Manilla,Barraba and Gunnedah) to form a functionalregion;

14. The SR of Richmond Tweed and Mid NorthCoast separate to form three functionalregions, one situated around Lismore andsurrounds (combining with Tenterfield asmentioned), another encompassing CoffsHarbour and surrounds and a third centredon Port Macquarie (and combining with



Gloucester/Great Lakes in the south, for-merly part of the Hunter SR).

Comparison of labour force estimatesTable 4 uses 2001 Census of Population andHousing data to compare measures of unemploy-ment, labour force participation and the ratio ofpart-time to full-time employment generatedusing the ABS labour force region geography andthe Functional Region geography.7 The resultsindicate that, on average, the design of the ABSlabour force regions is such that these geogra-phies tend to produce lower estimates of theunemployment rate and higher estimates oflabour force participation when compared tothose of the newly created functional regions.Measures of standard deviation in unemploy-ment rates indicate that there is less dispersionbetween SLAs comprising a functional region acompared to that in those SLAs in a labour forceregion. The intra-regional deviation in rates is 1.8for the functional regions compared to 2.1 for thelabour force regions, which suggests that unem-ployment rates within this new geography aremuch more alike, or homogeneous. While it isdifficult to assert the primacy of one geographyover another, the particular groupings of SLAsdevised by the ABS for the dissemination of itslabour force statistics are not as internally homo-geneous as are those of the Functional Regionsdevised using the Intramax technique. Thepattern of the ABS labour force regions is prin-cipally based on the need to maintain statisticalreliability and this requires equal distribution ofpopulation and labour force8. The outcome ofthis is the proliferation of LF regions in metro-politan Sydney versus non-metropolitan NSW.The important point to note is that the MAUPproblem may be resulting in tangible differencesin measures of regional labour force outcomes.Keeping in mind the need to ensure statisticalreliability in regional estimates, the particulardelineation of spatial boundaries deservesserious consideration in the dissemination ofAustralian labour force statistics.

ConclusionIn exploring the best way to delineate regionallabour markets such that the resulting geographyhas inherent ‘economic meaning’ we have devel-oped spatial demarcations (termed functionalregions) based on a hierarchical aggregationtechnique known as Intramax. This technique isapplied to Journey to Work data which explicitlycaptures the economic interactions of firms and

workers across space. The technique, in all threestages of clustering, delivers very interestingresults. It establishes a new geography represent-ing the space over which supply (workers) anddemand (firms) are seeking to interact as shownby the maximisation of commuting flows. It alsohelps us to better understand the ways in whichthe regions are linked, as illustrated by the hier-archy visible in the dendrograms. The latterapplication is particularly useful within metro-politan areas where labour market flows havepreviously been represented by sizeable matricesof data, which are difficult to order and compre-hend. The hierarchical aggregation techniqueclearly indicates which regions are connected bytheir labour market flows and the order in whichthey connect. This has obvious applications forplanners and policy makers broadly interested inregional interactions (in terms of economic activ-ity and housing), as well as those expressly inter-ested in labour market problems.

The Intramax technique emphasises labourforce flows and optimises SLA groupings basedon higher than expected interactions betweenneighbouring areas (regions), and appears toprovide a much closer approximation of a locallabour market. Mapping the 24 functionalregions (cluster 2) provides an informative cri-tique of the current labour force area designa-tions. These functional regions generallycollapse metropolitan and split non-metropolitanlabour force regions (with the splits often centredaround major regional towns). This wouldsuggest the need, perhaps, for the ABS to reas-sess its current labour force area designations(keeping in mind its requirements for statisticalreliability in delivering regional estimates from anational survey). The functional region patternoutlined in this paper reduces intra-region disper-sion in unemployment rates, which is to say thatthis technique tends to group areas that are morehomogeneous. A simple application using 2001Census of Population and Housing data revealsthat, on average, the emerging unemploymentrates are higher and labour force participationrates are lower than those of the standard labourforce regions, an indication that the MAUPproblem may be resulting in observable differ-ences. Given that ABS regional labour force esti-mates are widely used by policy makers andpractitioners it is important that these findingsare given due consideration.

Further work remains to be done in applyingthis technique to the 2006 Census of Populationand Housing data to develop an updated geogra-



Tabl

e4

Com

pari

son

ofA

BS,

Func

tiona

lR

egio

nan

dL

abou

rFo

rce

Reg

ion

Geo

grap

hies

,200

1.

Func

tiona

lR

egio

nG

eogr

aphy

UR

LFP

PT/F

TE

mp.

Stan

dard

Dev

iatio

n(U

Rra

te)

Lab

our

Forc

eR

egio

nG

eogr

aphy

UR

LFP

PT/F

TE

mp.

Stan

dard

Dev

iatio

n(U

Rra

te)

%%

%%

%%

Can

terb

ury-

Ban

ksto

wn

6.9

62.2

39.7

1.8

Inne

rSy

dney

6.2

70.6

36.5

1.2

Sout

hSy

dney

,Inn

erSy

dney

5.5

68.5

39.8

1.3

Eas

tern

Subu

rbs

4.8

67.7

42.6

0.9

Hur

tsvi

lle,K

ogar

ahan

dSu

ther

land

Shir

e4.

565

.844

.91.

3St

Geo

rge

Suth

erla

nd4.

565

.844

.91.

3L

ower

Nor

ther

nSy

dney

,Eas

tern

Cen

tral

Nor

ther

nSy

dney

and

Nor

ther

nB

each

es3.

768

.046

.70.

5C

ante

rbur

y-B

anks

tow

n8.

456

.840

.70.

7

Haw

kesb

ury,

Bla

ckto

wn,

Blu

eM

ount

ains

and

Bau

lkha

mH

ills

6.4

65.6

42.4

2.6

Fair

field

Liv

erpo

ol10

.659

.137

.13.

1

Fair

field

-Liv

erpo

olan

dO

uter

Sout

hW

este

rnSy

dney

9.0

62.4

39.6

3.4

Out

erSo

uth

Wes

tern

Sydn

ey7.

067

.443

.02.

3W

ollo

ngon

gan

dSh

oalh

aven

9.4

56.6

57.4

2.1

Inne

rW

este

rnSy

dney

5.4

63.4

40.8

1.6

Yas

san

dY

arro

wlu

mla

5.3

64.0

50.8

1.9

Cen

tral

Wes

tern

Sydn

ey8.

160

.039

.52.

6G

oulb

urn

8.4

58.0

64.7

3.8

Nor

thW

este

rnSy

dney

6.5

66.7

43.0

2.1

Gos

ford

-Wyo

ng8.

556

.858

.41.

9L

ower

Nor

ther

nSy

dney

3.9

68.8

39.1

0.5

New

cast

lean

dPo

rtM

acqu

arie

9.8

58.3

58.4

1.9

Cen

tral

Nor

ther

nSy

dney

3.6

68.2

50.6

0.4

Gre

atL

akes

and

Has

tings

12.7

48.6

70.0

3.7

Nor

ther

nB

each

es3.

469

.649

.80.

4C

offs

Har

bour

and

Gra

fton

13.5

52.9

71.7

2.4

Gos

ford

-Wyo

ng8.

556

.858

.41.

9Ta

mw

orth

&Su

rrou

nds

8.9

60.2

50.1

2.4

New

cast

leSe

ctor

10.2

57.6

60.3

0.9

Arm

idal

e&

Surr

ound

s8.

958

.554

.32.

0H

unte

rex

clud

ing

New

cast

le8.

256

.352

.51.

8N

arra

bri,

Mor

eePl

ains

and

Surr

ound

s9.

364

.044

.82.

3W

ollo

ngon

gSt

atis

tical

Reg

ion

9.0

58.7

54.8

1.9

Ric

hmon

d-Tw

eed

and

Tent

erfie

ld12

.453

.369

.81.

7Il

law

arra

excl

udin

gW

ollo

ngon

g8.

853

.065

.73.

5D

ubbo

and

Surr

ound

s7.

661

.047

.31.

5So

uth

Eas

tern

6.8

61.7

54.1

2.9

Lac

hlan

,For

bes,

Park

esan

dC

olba

r7.

060

.844

.41.

2R

ichm

ond-

Twee

dan

dM

id-N

orth

Coa

st12

.952

.070

.52.

3O

rang

e,L

ithgo

wan

dSu

rrou

nds

7.7

61.3

49.7

2.1

Nor

ther

n,Fa

rW

est-

Nor

thW

este

rnan

dC

entr

alW

est

8.3

60.5

49.0

2.3

Wen

twor

than

dC

entr

alD

owlin

g9.

956

.451

.23.

7M

urra

y-M

urru

mbi

dgee

6.3

63.7

46.6

1.7

Tum

ut,G

unda

gai

and

Surr

ound

s6.

460

.046

.61.

1A

lbur

y,H

ume

and

Surr

ound

s7.

263

.549

.51.

6M

urra

y-M

urru

mbi

dgee

and

Surr

ound

s4.

964

.942

.41.

6St

anda

rdD

evia

tion

2.5

4.8

9.6

Stan

dard

Dev

iati

on2.

45.

59.

2A

vera

ge(o

fre

gion

s)8.

160

.551

.41.

8A

vera

ge(o

fre

gion

s)7.

262

.148

.52.

1A

vera

ge(i

nag

greg

ate)

7.2

62.2

47.6

3.4

Ave

rage

(in

aggr

egat

e)7.

262

.247

.63.

4

Sour

ce:

AB

S,C

ensu

sof

Popu

latio

nan

dH

ousi

ng,2

001.



phy. Exploring the occupation dimensions ofthese local labour markets (see Bill, Mitchell andWatts, 2008) may also be a useful future appli-cation of the method demonstrated here.

NOTES1. Subsequent analysis reveals that a different parameteri-

sation of the rules-based Coombes algorithm can yieldmore spatially disaggregated commuting areas in theSydney Metropolitan Area. The authors will publish acomparative analysis of these two techniques in duecourse.

2. Watts (2004) developed a regional demarcation for NSWbased on the Coombes et al. algorithm (Coombes et al.,1986), but cross-border commuting was ignored and theadjustment of small flows to zero was not undertaken.It is considered that these nuances in the dataset useddo not explain the differences in the regionalisationsproduced.

3. Of course, in many of these instances, the SLAs are manykilometres apart, so that daily commuting would beunfeasible.

4. The Flowmap software is available from http://flowmap.geog.uu.nl. We also thank Tom de Jong whogave excellent software support.

5. All grouping algorithms entail the adoption of arbitraryjudgments. Intramax is less vulnerable to this criticism.

6. The labour force geography illustrated in the above mapsis based on an SLA to labour force region concordanceprovided by the ABS, for 2004 see: http://www.ausstats.abs.gov.au/ausstats/subscriber.nsf/0/AE3E261E3BA2506ECA257139007746CE/$File/61050_sla%20to%20lfs%20region%20concordance_jul2004.xls.

7. Labour force measures are drawn from the ABS 2001Census of Population and Housing as it correspondsto the Journey-to-Work matrix used to construct theFunctional Regions, and it is sufficiently spatiallydisaggregated.

8. The ABS (2004) notes that ‘LFS regions were originallyestablished after extensive consultation with major usersof labour force data. Estimates for these LFS regionswere first released in 1985. Factors in the design of theLFS regions included: the sample sizes required to yieldreliable estimates; the need for consistency with theASGC; and the need for comparability with other statis-tical collections. LFS regions are determined, in part, bythe expected sample size for each region. If the regionsare too small, then the accuracy of estimates will not beacceptable: relative standard errors on estimates willbe very large, and the estimates will not be reliableor useful. The regions represent a compromise betweenuser interest in small area data and the design limitsof the sample’. See: http://www.abs.gov.au/AUSSTATS/[email protected]/94713ad445ff1425ca25682000192af2/44a9700b59346b98ca256f5600768aac!OpenDocument

REFERENCESAustralian Bureau of Statistics, 2004: Feature Article –

Labour Force Survey of Regions, Australian LabourMarket Statistics (Cat No. 6105.0), July 2004, Canberra.

Barros, R., Broadbent, R.A., Cordey-Hayes, M., Massey,D.B., Robinson, K., and Willis, J., 1971: An operationalurban development model for Cheshire. Environment andPlanning, 3, 115-234.

Baum, S., O’Connor, K., and Stimson, R., 2005: FaultLines Exposed: Advantage and Disadvantage across Aus-tralia’s Settlement System, Monash University ePress,Clayton.

Berry, B.J.L., (1968) A synthesis of formal and functionalregions using a general field theory of spatial behaviour. InBerry, B.J.L., and Marble, D.F. (eds.) Spatial Analysis.Prentice-Hall, Englewood Cliffs, 419-428.

Bill, A., Mitchell, W.F., and Watts, M., 2008: The occupa-tional dimensions of local labour markets in Australiancities. Built Environment, 34, 291-306.

Bond, S., and Coombes, M.G., 2007: 2001-based Travel toWork Areas Methodology. mimeo. http://www.statistics.gov.uk/geography/downloads/2001_TTWA_Methodology.pdf

Brown, L.A., and Holmes J.H., 1971: The delimitation offunctional regions, nodal regions and hierarchies by func-tional distance approaches. Journal of Regional Science,11, 57-72.

Coombes, M.G., Green, A.E., and Openshaw, S., 1986: anefficient algorithm to generate official statistical reportingareas. Journal of the Operational Research Society, 37,943-53.

Goetgeluk, R., 2006: Dynamic clusters in migration patterns:Intramax-analyses of inter-municipal migration flowsbetween 1990 and 2004. Paper presented at the ENHRconference Housing in an expanding Europe: theory,policy, participation and implementation, Slovenia,July.

Goetgeluk, R., and de Jong, T., 2005: Migration Analysis asa Political Instrument, the case of the Leiden and Bulbregions in Randstad Holland. In Goetgeluk, R., andMusterd, S. (eds.) Beyond Residential Mobility: LinkingResidential Choice with Urban Change. Open House Inter-national, 30, 75-81.

Green, A., 1997: Alternative approaches to defining locallabour markets for urban and regional policy. In Turok, I.,(ed.) Travel to Work Areas and the Measurement of Unem-ployment, Conference Proceedings, Edinburgh.

Gregory, R.G., and Hunter, B., 1995: The macro economyand the growth of ghettos in urban poverty Australia. Dis-cussion Paper No. 325, Economics Program, ResearchSchool of Social Sciences, Australian National University,Canberra.

Hasluck, C., 1983: The economic analysis of urban labourmarkets. Journal of Industrial Affairs, 10, 25-31.

Lawson, J., and Dwyer, J., 2002: Labour market adjustmentin regional Australia. Research Discussion Paper, 2002–04, Economic Group, Reserve Bank of Australia.

Masser, I., and Brown, P.J.B., 1975: Hierarchical aggregationprocedures for interaction data. Environment and PlanningA, 7, 509-523.

Masser, I., and Scheurwater, J., 1980: Functional regionali-sation of spatial interaction data: an evaluation of somesuggested strategies. Environment and Planning A, 12,1357-1382.

Mitchell, W.F., and Carlson, E., 2005: Why do disparities inemployment growth across metropolitan and regionalspace occur? Australasian Journal of Regional Studies, 11,25-40.

Mitchell, W.F., and Bill, A., 2006: Who benefits fromgrowth? Disadvantaged workers in growing regions. Aus-tralian Journal of Labour Economics, 9, 239-256.

O’Connor, K., and Healy, E., 2002: The links between labourmarkets and housing markets in Melbourne. AustralianHousing and Urban Research Institute (AHURI),Swinburne-Monash Research Centre.



http://flowmap.geog.uu.nl

http://flowmap.geog.uu.nl

http://www

http://www.abs.gov.au/AUSSTATS

http://www.statistics

Openshaw, S., 1984: The Modifiable Areal Unit Problem.Concepts and Techniques in Modern Geography(CATMOG), 38. GeoAbstracts Ltd., Norwich.

Papps, K.L., and Newell, J.O., 2002: Identifying FunctionalLabour Market Areas in New Zealand: A ReconnaissanceStudy Using Travel-to-Work Data’, IZA Discussion Paper443. Institute for the Study of Labour, Bonn.

Watts, M.J., 2004: Local labour markets in New South Wales:Fact or Fiction? In Carlson, E. (ed.) A Future that Works –Economics, Employment and the Environment, Proceed-ings of the 6th Path to Full Employment Conference,Newcastle, NSW.

Watts, M.J., 2009: Rules versus hierarchy: an application ofFuzzy Set Theory to the assessment of spatial groupingtechniques. The International Conference on Adaptive andNatural Computing Algorithms (ICANNGA), Kuopio,Finland, April.

Watts, M.J., Baum, S., Mitchell, W.F., and Bill, A., 2006:Identifying local labour markets and their spatial properties.The ARCRNSISS Annual Conference, Melbourne, May.

Yates, J., 2005: Are occupational choices affecting housingchoices? The Australian Social Policy Conference, 2005,‘Looking Back, Looking Forward – A Quarter of a Centuryof Social Change’, University of New South Wales, 20–22July 2005.

Yates, J., Randolph, B., and Holloway, D., 2006a:, HousingAffordability, Occupation and Location in AustralianCities and Regions. Australian Housing and UrbanResearch Institute (AHURI), Melbourne.

Yates, J., Randolph, B., Holloway, D., and Murray, D.,2006b: Housing Affordability, Occupation and Location inAustralian Cities and Regions, Final Report, AustralianHousing and Urban Research Institute (AHURI),Melbourne.



Identifying Functional Regions in Australia Using Hierarchical Aggregation Techniques

Documents

Transcript of Identifying Functional Regions in Australia Using Hierarchical Aggregation Techniques