The impact of employment specialisation on regional labour market outcomes in Australia.
This article examines the industry composition of employment across Statistical Divisions in Australia utilising census data from 2001 and 2006. We find some evidence to support the hypothesis that peripheral regions tend to have a higher level of employment specialisation than metropolitan centres, but there is little indication that employment specialisation, in general, grew over the period. From a policy perspective, we provide support for the findings of previous Australian researchers that higher levels of employment specialisation are associated with better labour market outcomes in a region, and that when policymakers assess different regional-development policies, they should give some consideration as to whether or not the implementation of their preferred policy will have an impact upon employment specialisation in the particular region.
Changing industry structure and an associated change in the industry composition of employment are ongoing features of all national economies. Commonly, these changes have been a decline in the importance of and employment levels in primary and secondary industries and a rapid growth in tertiary industries (see for example Robson (2006) and Garnet and Lewis (2007)). However, within each industry sector the experience of individual industries associated with that sector may be very different. For example in the Australian primary sector, drought may induce a decline in agriculture (Horridge et al., 2005), while external factors may stimulate mining (Connolly and Orsmond 2011). Further, across regions within national borders the nature of the changes taking place will display similar variability (Robson 2006).
Theoretical analysis (Krugman 1991; 1993; 2009; 2010) has attempted to explain both inter-temporal and geographic shifts in industrial-employment structures and the tendency towards either increasing or decreasing regional specialisation in employment. Government policy is capable of having an impact on the industrial structure, and consequently on the degree of employment specialisation across regions within an economy (Wren and Taylor 2009). Beer and Clower (2009) suggested that the desire to pursue policy intervention to change a region's industry structure is likely to evoke the question of whether or not the change generated is beneficial for that regional economy.
In the Australian context, this article explores the pattern of regional employment specialisation, how it has changed over a limited time span (between the 2001 and 2006 censuses), and whether--when formulating regional employment policy--the government should take into consideration the impact that their proposed policy would have on a region's employment specialisation. The study utilises a new dataset derived from the 2001 and 2006 Australian censuses. These data provide for a sub-state classification of regions and a detailed classification of employment by industry. Section 2 surveys previous research in Australia and internationally; it reviews the literature that deals with the theoretical underpinnings of the analysis. Section 3 considers issues of research method, including the definition of a region, selection of measures of employment specialisation, and description of the data. The research findings a re presented in Section 4 and are discussed in Section 5, after which some policy conclusions are drawn.
2. Literature Review
Empirical analysis of regional employment specialisation and its impact on regional economies is the focus of many international studies (for example Lilien 1982; Wren and Taylor 1999; Marelli 2004; Shearmur and Polese 2005; Suedekum 2006; Robson 2006; Belke and Heine 2006; Robson 2009). There is also a growing, but still relatively limited amount of literature that considers regional employment specialisation in an Australian context (Bradley and Gans 1998; Beer and Clower 2009; Stimson et al. 2009; Dixon and Freebairn 2009).
International research findings have been somewhat inconsistent. Lilien (1982) argued for regional policy intervention to offset identified disadvantages. Wren and Taylor (1999) found that all UK regions had increased in terms of their employment specialisation but that partly as the result of successful regional policy, the regions tended to converge towards the national employment pattern. Marelli (2004) found in his study of 145 European regions that specialisation had been decreasing, although he found evidence that increased specialisation had a positive impact on regional growth. Robson (2006) confirmed the increase in employment specialisation in UK regions, but found little evidence of convergence. Robson (2009) found a small but statistically significant relationship between an improved functioning of the regional labour market and increased regional employment specialisation. On the basis of their evidence, Shearmur and Polese (2005) concluded that diversification polices to enhance regional growth in Canada were difficult to justify However, Suedekum (2006) considered that his results based on German data supported diversification policies, despite not finding evidence of increasing employment specialisation in German regions.
Bradley and Gans (1998) attempted to explain the growth of Australian cities in terms of population and labour force growth. They included as one of their explanatory variables a measure of the industry specialisation of each of their selected cities, namely a modified Herfindahl index applied to city data segmented into 12 employment groups (Bradley and Gans 1998, p. 270). This broad classification of data was all that was available then. Their results indicated a negative relationship between growth and specialisation.
With a similar measure of employment specialisation, Beer and Clower (2009) utilised 2001 census data for 76 Australian cities (with populations greater than 10,000). While they found no relationship between labour force growth and their end-period employment specialisation measure (p. 383), they found a significant relationship between their measure of the change in employment specialisation and labour force growth (p.385).Their analysis suggests that the explanation for the difference between the findings of the two articles may be the substantially enhanced competitive nature of the prevailing economic environment at the time of the latter study. From the 1970s, with substantial reductions in tariff barriers, and moving through the floating of the Australian dollar in the 1980s and the microeconomic reforms of the 1990s, the Australian economy became more competitive globally and 'the economies of regional cities have reflected this transformation and those cities that have moved to greater levels of specialisation in their economies appear to have had the most certain economic growth' (Beer and Clower 2009, p. 386).
Stimson et al. (2009) used data from the 1991 and 2001 censuses to construct location quotients for mainland Local Government Areas as an index of employment specialisation over three employment categories. Based on these data they found limited support for a link between employment specialisation in a region, and in regional growth.
Finally, Dixon and Freebairn (2009) looked at regional differences in the composition of employment across 52 industry groups in regional Australia. They found that their measure of specialisation--the Coefficient of Regional Specialisation--indicated that the industrial composition of Australian states had tended to become less specialised over time. They attributed this to the tariff reductions which occurred throughout the period of their study.
In summary, past research based on different measures of employment specialisation produced evidence of constant, decreasing, and increasing levels of employment specialisation across regions. Within Australia, work undertaken at the city and Local Government Area levels with a relatively small number of industry groups returned findings of increasing specialisation; research at the state level, with a much larger number of industry categories, returned findings of falling specialisation. Evidence relating to the hypothesis that increasing regional employment specialisation will result in enhanced economic outcomes for a region has been found in some studies but not in others.
The Krugman hypothesis
The 'core-periphery' model, established in Krugman (1991; 1993) and revisited in Krugman (2009), is said to be driven by the choices open to both firms and individuals (RSAS 2008). Firms are attracted to the larger market by the potential economies of scale offered by the larger market and the likely saving in transport costs. On the other hand, individuals have an incentive to move to the region offering higher real wages and a greater variety of goods and services. The result of this is that the larger, more diverse core tends to grow and increase in diversity while regions at the periphery tend to become more specialised.
Increased specialisation at the periphery results from the interplay of the forces considered important in the Krugman model. Internal economies of scale accrue to the firm as it grows in size in one location and are associated with market size. External economies of scale, on the other hand, result from the organisation of the industry into agglomerations of firms that are mutually beneficial. When transport costs are high, firms remaining or establishing and growing in the periphery, will generally be those for which internal economies of scale are less important than the location of inputs (especially raw materials) and the presence of external economies. If transport costs fall, some firms in the periphery will relocate close to the large local market in order to take advantage of internal economies. However, as transport costs continue to decline, access to the market becomes less important and rising factor costs motivate firms to return to peripheral regions (Martin and Sunley 1999).
Associated with this, but often neglected in empirical work, is the prediction that specific industries tend to concentrate in particular regions (Krugman 2009). However, as Krugman notes: 'It seemed, from both an aesthetic and historical point of view, that a model in which agglomeration did not have to happen, in which results depend on parameters, would be more interesting than one in which emergence of an industrial core was preordained' (Krugman 2010, p.10). Thus if the 'centrifugal forces' of dispersed resources dominated the 'centripetal forces' of large markets, decentralisation could be the outcome.
While some research into the regional specialisation of employment has been set in the context of Krugman's 'core-periphery' model (for example Marelli (2004), Suedekum (2005), Coombes and Overman (2004), Dixon and Freebairn (2009)), this has not been the case with all such empirical work (for example Robson (2006; 2009), Shearmur and Polese (2005) and Lilien (1982)).
The Definition of Regional Types
Garnett and Lewis (2007), in their review of the major attempts at classifying regions in Australia, argue that many regional classifications failed to show the full impact of the changes taking place. They regarded an ideal classification of regions as one which groups areas so that they are internally as homogeneous as possible across all of the variables of interest. However, they also recognise that 'all methods of classification have flaws and anomalies' (p. 31). Following their brief review of various regional-classification systems that have been used in Australia they opted to utilise the classification system developed by ABARE and reported in Garnaut et al. (2001). This system is based on the ABS Statistical Local Areas (SLA's). According to ABS (2011), SLA's are the smallest ABS spatial units identified in the Australian Standard Geographical Classification (ASGC) that, in sum, cover the whole of Australia. In terms of hierarchy, SLAs are combined to form Statistical Subdivisions (SSDs), which in turn can be combined to form Statistical Divisions (SDs)--the largest sub-state classification. The ABARE (and hence Garnett and Lewis) approach defines each SLA in Australia as being either Capital City (of which there are eight), Other Metropolitan (where the SLA contains whole or part of an urban centre with a population of at least 100,000), Coastal (within 80 kilometres of the coastline--except for remote coastal areas in northern Western Australia and Queensland), Remote (as determined by the Accessibility/Remoteness Index of Australia) and Inland (the remainder).
As our purpose was to examine labour market activity, we considered that daily movements of workers between SLAs made it unlikely that SLAs could adequately represent individual labour markets (note, however, that Edwards et al. (2011) argue that the SLA level of disaggregation is the most commonly used in labour market studies). While the problem of interregional flows of labour persists at the level of SD, it was likely to be less of an issue. For example it is common for workers to commute between Orange and Bathurst (SLAs in the Central West SD involving 50 minutes travel by car), but it is far less common for workers to commute between Bathurst and Wagga Wagga, (the latter being an SLA in Murrumbidgee, an SD contiguous to the Central West), as this would require a commute of four hours each way. Nevertheless, there will be flows between SDs, especially between towns that are close to the SD boundaries. Therefore, we use data based on place of employment rather than place of residence. Our choice of SDs was further supported by a number of technical issues related to the data-collection processes of the ABS (see ABS 2011, pp. 11-12). First, SDs a re defined by the ABS on the basis of socioeconomic criteria. Outside the capital cities, they are defined as relatively homogeneous and characterised by identifiable social and economic links between the inhabitants, and between the economic units within the region under the unifying influence of one or more major town or city. Second, the boundaries of SDs are changed infrequently (15 to 20 years), making them the most stable regional unit within each state (ABS 2011, p. 11). This point is particularly important in the light of our econometric procedure (discussed below), as changes in the boundaries of both SSDs and SLAs over the period would have greatly complicated the process of estimation. Third, while we are restricted to the use of the classifications in the ASGC, the ABS has adopted a new classification system: the Australian Statistical Geography Standard (ASGS) replaces the ASCG. In this new system, Statistical Areas Level 4 (SA4) will be the broadest sub-state classification and are therefore comparable in size to existing SDs. Further, SA4s will be used for the release of labour force statistics (ABS 2011, p. 36).
Given the arguments put forward in Garnett and Lewis (2007) we felt that it was important to retain the essence of the ABARE classification; we classified the SDs by utilising the ABARE classification. This was done primarily by an inspection of the map of regional classifications provided in Garnaut et al. (2001), and the maps of SDs provided in ABS (2011). In the few uncertain cases, the ABARE classification of SLAs within specific SDs was consulted. As SDs often contained SLAs classified differently under the ABARE system, we had to make a value judgement about which of the two (and sometimes three) classifications most appropriately represented the SD. Table 1 reflects the outcome of our classification decisions for SDs by regional type.
Finally, we decided against the use of the alternative ASCG classification that covers all Australian Statistical Region Sectors (SRS), Statistical Regions (SR), and Major Statistical Regions (MSR) for three reasons. First, the classification was not based on SLAs and would preclude further use of the ABARE classification of regional types; second, the classification is based on the minimum regional population levels required to yield reliable survey results (ABS 2011, p. 25)--which is not necessary when census data are used--and not upon the economic and social homogeneity of the region; third, as previously discussed, the SD classification has the advantage of greater stability overtime.
Measures of employment specialisation
Our first measure of specialisation is designed to answer the question: regardless of any change with respect to the distribution of employment in Australia as a whole, has the employment in the region become more or less specialised relative to what the structure of employment in the region was previously? To answer this we need a measure of absolute specialisation for each region which we can compare overtime and across regions. In the literature, such a measure is provided by the Coefficient of Absolute Regional Specialisation (CARS):
CARS = [[[[summation].sup.n.sub.i=1] [([[chi].sub.ir] - [[??].sub.ir]).sup.2]/(n-1)].sup.1]/2/[[??].sub.ir] (1)
where [[chi].sub.ir] is employment in industry i in region r, and [[??].sub.ir] is the mean employment level overall industries in the region.
This measure has been used by Wren and Taylor (1999) and Robson (2006; 2009). It is regarded as a measure of absolute regional specialisation because the numerator in equation 1 is the standard deviation [sd] of regional employment across all industries from the mean employment level of industries in the region. When we divide the sd through by [[??].sub.ir] we have CARS, the coefficient of variation (CV) of employment in each industry. This provides a relative measure of absolute variation that a I lows us to compare variations in the absolute dispersion of employment among industries, across both time and space for regions, when mean levels of employment by industry may vary. An increase in CARS for a region, overtime, indicates that the region has become relatively more specialised in its employment structure, as the sd of employment across industries (the numerator in the index) relative to the mean level of employment for all industries (the denominator in the index) has increased. This can only occur when employment in some industries becomes relatively more important, that is as employment in the region becomes relatively more specialised overtime. Similarly, a higher value for CARS in region A than in region B, at a given time, implies that the employment structure in region A is relatively more specialised than in region B at that time.
A second measure of regional specialisation endeavours to quantify the degree of specialisation in regional employment relative to the degree of specialisation of employment in Australia as a whole. This measure is the coefficient of relative regional specialisation (CRRS) (Wren and Taylor (1999) and Robson (2006)) given by:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)
CRRS is regarded as a measure of relative regional specialisation because the numerator in the index is the standard deviation of the regional share of national employment in each industry from the mean of the regional share of national employment across all industries. Dividing the numerator through by the latter makes CRRS the CV of the regional share of national employment in each industry. A value of zero for CRRS would mean that the share of employment across all industries in the region concerned was the same as the share of employment across all industries in the Australian economy. The higher the value of CRRS, the more divergent the employment structure in the region is compared to the employment structure in Australia. CRRS is a relative measure of the variation in the regional share of national employment across all industries, which allows us to compare variations in the regional share of national employment among industries (across time and space) for regions when mean levels of employment by industry may vary. A decrease in CRRS for a region overtime indicates that the industry structure of the region has become more like the industry structure of Australia. In this sense, it provides us with the same information as the frequently used measure of the extent of convergence in regional employment structures known as the Coefficient of Regional Specialisation (CRS) (1). CRS is calculated as:
CRS = 1/2 [[summation].sup.n.sub.i=1] [absolute value of (([[chi].sub.ir]/[[chi].sub.r]) - ([[chi].sub.iA]/[[chi].sub.A]))] (B)
In our work we followed Wren and Taylor (1999) and chose to use CRRS rather than CRS as our measure of relative specialisation. This is because changes in industry specialisation within a region result from changes in specialisation that take place within a sector (primary, secondary, or tertiary), and changes in specialisation that take place between these sectors. Thus, the variance of employment within a region will be composed of the variance of employment that takes place in each of the sectors within that region and the variance of employment that takes place between these sectors. In future work we wish to examine sectoral shifts. As explained in Wren and Taylor (1999), the square of CRRS (and also CARS) is the population variance divided by the mean squared, making CRRS (and CARS) amenable to decomposition into intra-sectoral and cross-sectoral changes.
The 2001 and 2006 census data on employment by industry by place of employment for all SDs in Australia form the basic dataset on which our analysis is predicated (ABS 2006). The industry data for the 2001 Census were classified on the basis of the 1993 Australian and New Zealand Standard Industrial Classification (ANZSIC) into 52 distinct industry groups The 2006 Census industry data were classified in terms of the 2006 ANZSIC classification. The data for 2006 were harmonised to the 2001 equivalent classification of 52 industries using ABS (2008).The harmonisation process should be regarded as approximate, as the data are not presented at a sufficiently high level of disaggregation to achieve complete precision. Only data classified by industry (or sector) were used. Undefined (for industry), not classifiable, and not-stated data were omitted. In conducting this research we did not have access to similar data for earlier periods. Not withstanding the limitation that this places on our findings, we proceeded with the analysis on the basis of the importance this period has for regional Australia in particular. Between 2002 and 2006 Australia experienced one of its worst droughts on record (ABARES2012, p. 5; p.14). The drought covered most of Australia's important agricultural regions and significantly reduced farm production and, therefore, employment in the affected regions. Over approximately the same period mining--another primary industry--was experiencing one of its longest and strongest boom periods (Connolly and Orsmond 2011), substantially adding to employment in the affected regions.
The regional employment pattern in Australia
To establish a background against which our analysis of specialisation takes place we first examine the structural composition of changes in employment across Australia. Given that we employ data in our analysis of specialisation classified by 60 SDs and 52 industries, we need a means of summarising the data in order to keep the analysis manageable. To do this we chose to use our concept of regional types. For this purpose--for Australia and for each of the regional types--employment is the sum of employment in the relevant geographic SDs (reported by the ABS). Omitted from the data are employment figures for non-geographic SDs such as Undefined (Statistical Division), Offshore Areas (Statistical Division), and No Usual Address (Statistical Division). In this article we identify three employment sectors. The primary sector consists of the agriculture, forestry, fishing, and mining industries. The secondary sector comprises all of the manufacturing industries; the tertiary sector comprises the remainder and includes utilities and construction.
A formal technique for conducting this analysis is the classical shift-share approach (seefor example Baxendine et al. 2004). Shift-share analysis enables us to categorise changes in regional employment within an industry group ([SS.sub.i]) into three components:
[SS.sub.i] = [NS.sub.i] + [IM.sub.i] + [RS.sub.i] (4)
The first change component is the national share for a given industry in the region ([NS.sub.i]), which is equal to employment in the specified industry in the region in the base period ([e.sub.i,t-1]) multiplied by the national rate of growth in total employment overthe period of change ([E.su.t] - [E.sub.t-1]/[E.sub.t-1]) to give:
[NS.sub.i] = [e.sub.i,t-1] x ([E.sub.t] - [E.sub.t-1]/[E.sub.t-1]) (5)
The resulting estimate indicates how regional employment in the industry would have changed if it had changed at the same rate as total employment in the nation over the period. When summed overall industries in the region, the national share component indicates the extent to which regional growth as a whole results from growth in the national economy.
The second component is the industry mix ([IM.sub.i]), which is equal to [es.up.i,t-1] multiplied by the national rate of growth in the industry over the period ([E.sub.i,t] - [E.sub.i,t-1]/[E.sub.i,t-1]) less ([E.sub.t] - [E.sub.t-1]/[E.sub.t-1]) to give:
[IM.sub.i] = [e.sub.i,t-1] x ([E.sub.i,t] - [E.sub.i,t-1]/[E.sub.i,t-1] - [E.sub.t] - [E.sub.t-1]/[E.sub.t-1]) (6)
The resulting number reflects the change in employment in the industry in the region resulting from the fact that national employment in that industry was growing faster, or slower, than total national employment. If national employment in the industry is growing faster than total employment, we would expect this component to be positive for the region. On the other hand, if national employment in the industry is growing more slowly (or is declining) compared with total employment, we would expect this component to be negative for the region. Summing all [IM.sub.i] across the region provides an assessment of the impact of the region's industrial mix on total employment growth in the region. Where a region's industrial mix is dominated by slow-growing or declining industries the sum of [IM.sub.i] over all industries will be negative. A positive result therefore indicates that the region's industrial structure is dominated by fast-growing industries.
The third component is the regional shift ([RS.sub.i]), which is equal to [e.sub.i,t-1] multiplied by the rate of growth in the industry in the region over the period ([e.sub.i,t] - [e.sub.i,t-1]/[e.sub.i,t-1]) less ([E.sub.i,t] - [E.sub.i,t-1]/[E.sub.i,t-1] to give:
[RS.sub.i] = [e.sub.i,t-1] x ([e.sub.i,t] - [e.sub.i,t-1]/[e.sub.i,t-1] - [E.sub.i,t] - [E.sub.i,t-1]/[E.sub.i,t-1]) = (7)
The resulting number reflects the change in the region's employment in the industry resulting from the fact that in the region, the industry performed better (employment grew faster or declined more slowly--giving a positive figure) or performed worse (employment grew more slowly or declined faster--giving a negative figure) than the industry did nationally. This component is therefore an indicator of local competitive advantage in the industry compared with the industry nationally. An indication of the region's competitive advantage as a whole can be assessed by summing the [RS.sub.1] over all industries.
While the employment pattern of individual component SDs did not always follow those of the respective regional type, the data in Table 2 give us a general picture of the pattern of change in employment occurring between the two census dates.
Employment in the capital cities (Metro) grew strongly, fuelled by rapid national employment growth and a generally advantageous industrial mix. However, Metro's overall competitiveness declined over the period, largely because of the decline in the competitiveness of its secondary industry. National growth in the tertiary sector, supported by the concentration of tertiary activities in Metro's regional mix, and the group's relative competitive advantage in these activities, saw tertiary employment in Metro grow strongly; it more than offset the employment decline in both the primary and secondary sectors. However, it is apparent that employment growth in Metro was held back by the fact that secondary industry is declining nationally but still accounts for over 10 per cent of Metro employment.
Growth in the other metropolitan centres (Other Metro) benefited from national employment growth, an overall advantageous industrial mix, and increased competiveness in all industry sectors--including secondary. However, the shares of secondary industry (about 10 percent) and primary industry (about 4 per cent)--both industries being in decline--in the group's industrial mix disadvantaged Other Metro's employment-growth performance. As with Metro, it was growth in the services sector--where Other Metro benefits from national growth in the sector, the concentration of the tertiary sector in the regional group's industrial mix, and the regional group's competitive advantage in this sector--which accounted for the bulk of the group's total employment growth over the period.
Employment growth in coastal regions (Coastal) was driven by national employment growth and an increase in overall competitiveness--notwithstanding the decline in the competiveness of the primary sector. Coastal was disadvantaged by its industry mix, where the declining primary and secondary sectors each accounted for about 11 per cent of employment. On the other hand, tertiary employment grew strongly on the back of national growth in the sector, the importance of tertiary industry in the group's industrial mix, and the fact that in services this group was becoming increasingly competitive.
As a group, Inland had the second-worst employment outcome over the period--growing at just on 3 per cent. This low growth was due to the impact of national growth on the regional group, as it was disadvantaged by its industrial mix (primary accounted for over 13 per cent and secondary accounted for over 11 per cent), and its failure to increase its relative competitiveness. Competiveness in the primary and tertiary sectors both declined; however, competiveness rose in the secondary sector. The decline in primary-industry employment of over 13 percent in Inland can be largely linked to the consequences of the drought conditions experienced over the period of study. In particular, the decline in the competiveness of this industry in what are major primary-producing areas reflects the adverse growing conditions experienced throughout the period.
The Remote Group of regions experienced the lowest growth in employment. This group was entirely dependent on national growth for their overall increase in employment, as the aggregate industry-mix and regional-shift components were negative. Employment in primary industry declined. However, the decline was less than half a per cent compared with the 13.5 percent decline for Inland. This is probably explained by the fact that the Remote Group includes many of Australia's important mining areas and, although the farming and agricultural activities would have suffered because of drought, mining activity during this period was booming and the increased mining employment would therefore offset the decreased employment in farming and agriculture. This explanation is further supported by the fact that the primary sector in Remote became increasingly competitive over the period, reflecting that mining activity could only be undertaken where the particular resource was located.
Discussion of the last two groups highlights one of the major drawbacks with shift-share analysis. It is sensitive to the level of regional disaggregation and to industry detail. For example a shift-share analysis of the Pilbara (i)--one of the SDs making up the Remote classification--based on the 52-sector industry classification discussed above showed that total employment rose by 17 per cent over the period compared with a rise of just 2 per cent for Remote. Mining-related employment grew by nearly 65 per cent and nearly half of this was attributed to the regional shift component--reflecting the geographically fixed resource being mined. The growth in mining was accompanied by significant growth in construction and nearly three-quarters of this was from the regional-shift component, as the growth in mining provided a competitive advantage for construction activity in the region. Other industries in which employment grew and which experienced a positive regional-shift effect included transport and health services--both necessary inputs into mining and construction. This example illustrates the ability of increased industrial specialisation (in this case in mining) to call forth employment growth in other areas and thus sustain the employment growth of a region. We turn now to examine changes in employment specialisation more closely.
Employment specialisation in the regions of Australia
The general pattern of our findings is reflected in Appendix Table 1 (for CRRS) and Appendix Table 2 for (CARS), where the median value for each regional classification in each state for 2001 (in brackets) and 2006 is recorded. In order to establish whether our measures of employment specialisation varied across regional types, we estimated thefollowing model for each measure:
Specialisation = [alpha] + [[beta].sub.1] Time + [[beta].sub.2] Other Metro + [[beta].sub.3] Coastal + [[beta].sub.4] Inland + b[[beta].sub.5] Remote + [[epsilon].sub.rt] (8)
Specialisation is the value of the index of specialisation (CRRS or CARS), Time is a dummy variable taking the value of 0 for 2001 and 1 for 2006, and the remaining independent variables are dummy variables for four of the five regional classifications (defined above), with Metro being represented by the intercept, [alpha], and the [[beta].sub.i] are the relevant coefficients, and [[epsilon].sub.rt] is the error term. The results of this exercise are reported in tables 3 and 4.
The values for CARS are significantly higher in Inland and Remote areas, but are not significantly different from Metro in either the Other Metro or Coastal regions.
With respect to CRRS, the index is significantly higher for Coastal, Inland, and Remote. However, we note that Time is not significant for either version of the model, and so we can conclude only that for the Australian economy overall there was little change in the level of specialisation over the period.
When we examine the impact of regional type on the change in CARS, we observe that being Coastal or Inland tends to have been associated with decreases in the value of CARS.lhe explanation for this negative change will certainly differ between the two regional types. For coastal regions, the fall in absolute specialisation is derived more from the rise in tertiary activity and less from the fall in primary activity than would be the case for inland Australia where primary industry was devastated by drought throughout the period. In terms of relative changes, the change in CRRS is significantly (positively) related only to the remote areas of Australia.
The impact of changes in specialisation on labour market outcomes
The basic form of our model for analysing the impact of employment specialisation on labour market outcomes takes the form (after Robson (2009):
[Y.sub.rt] = [alpha] + [[beta].sub.1] [Specialisation.sub.rt] + [[epsilono].sub.rt] (9)
where [Y.sub.rt] is the dependent labour market performance variable that takes one of three forms in our analysis: the natural logs of (a) the unemployment rate, InUr; the non-employment rate, InNer; and the participation rate, InPr. The variable InNer is used because loss of a job in an area where there are few alternative employment opportunities (a predominantly agricultural region for example) might lead to a more rapid withdrawal from the labour force than in areas where the newly unemployed might reasonably hold out an expectation for employment in an alternative industry. The variable Specialisation refers to the use of either the natural log of CARS, InCARS; or the natural log of CRRS, InCRRS.
Our reported analysis was undertaken via a panel regression--accounting for fixed effects for all SDs and incorporating time-series dummies. The analysis also employs robust standard errors to account for any heteroskedasticity that may be present. Given the ABS definition of SDs as relatively homogenous economic units, we decided not to apply a spatial econometric modelling approach which would account for interrelationships between regions, as we expected these to be relatively weak. However, this assumption will be tested in future work, especially if we were to expand the study to smaller spatial units where such interrelationships are likely to become more important. With our model, we could find no statistical relationship between InPr and either explanatory variable; nor could we identify any statistical relationship between InCRRS and either InUr or InNer. However, as reported in tables 5 and 6, there exists a significant statistical negative relationship between InCARS and InUr, and between InCARS and InNer.
5. Discussion and Conclusions
We noted that despite considerable analysis at the international level, research on industry specialisation in employment in Australia is scarce. One analysis (Dixon and Freebairn 2009), based on state-level data, produced findings that appeared contrary to the usual predictions of the Krugman hypothesis. The evidence was of declining specialisation when increasing specialisation was expected. We use the same ABS industry classifications and census data from 2001 and 2006 (rather than employment survey data at the state level) and we re-examined the situation. While the restrictions of our data do preclude us from making any useful comment on the change of specialisation overtime--the issue considered by Dixon and Freebairn (2009)--we are able to examine specialisation across SDs classified by regional type, based on the classification introduced by Garnaut et al. (2001) and used by Garnettand Lewis (2007). We find that those regions that could be described as peripheral (Inland and Remote) do, on average, exhibit higher values for the measures of industry specialisation of employment and that this is in line with the expected results from testing the Krugman hypothesis.
In analysing employment data at the SD level we note that employment growth is associated with tertiary industry growth and, typically, a decline in employment in agriculture (and, to a lesser extent, in manufacturing). As the period examined (2001-2006) was one of extensive drought in Australia, it is likely that the decline in agricultural employment can be attributed to the adverse climatic conditions (for farming production) experienced over the period. It was further observed that the decline in agricultural employment--which typically would have led to a fall in our measures of specialisation--was frequently offset in SDs by substantial growth in tertiary employment and, particularly in remote regions, by the growth of the mining industry fuelled by the rise in Australia's exports to Asia--particularly China.
Finally, we examined the impact of industry specialisation in employment on labour market outcomes. We found no evidence of a significant impact of CRRS on any measure of employment outcome. This was not entirely unexpected, as CRRS is a relative measure of specialisation with the index indicating whether or not a regional employment structure was becoming more or less like the national structure. In itself, it does not say anything about whether or not the region was becoming absolutely more specialised. The impact of our measure of absolute specialisation, CARS, appeared to produce evidence in support of the prediction that both a lower unemployment rate and a lower non-employment rate were associated with greater absolute specialisation in a region. This finding supports earlier findings by Beer and Clower (2009) and Stimson et al. (2009).
From a policy perspective, our findings suggest that one consideration in deciding upon regional policy should be whether or not the chosen policy will encourage the development of industries that result in greater absolute specialisation of employment in a region, as this is likely to have a greater impact on reducing regional unemployment and non-employment than will policies that result in an increased diversity of employment. However, further research is required into the types of industries in which regions should specialise, and the appropriateness of policies that might bring this about. In particular, policies that encourage a clustering of industries in a region (as has occurred in the Pilbara) are preferable to policies that disrupt the industrial structure of the nation.
Appendix Table 1: Median Value of CRRS, 2006 and 2001 NSW Victoria Queensland South Australia Metro 44.91 45.07 35.37 44.91 (46.43) (46.07) (31.57) (36.79) Other 65.04 56.47 NC NA Metro (71.99) (57.19) NC NA Coastal 89.38 105.09 192.08 161.18 (84) (109.52) (163.87) (146.81) Inland 98.50 106.50 73.38 152.65 (96.14) (94.42) (71.38) (146.98) Remote 217.94 NA 186.99 303.22 (149.56) NA (150.08) (297.25) Western Tasmania NT ACT Australia Metro 74.46 84.97 138.85 170.55 (77.46) (79.86) (131.52) (154.05) Other NA NA NA NA Metro NA NA NA NA Coastal 115.52 169.42 NA NA (107.48) (129.62) NA NA Inland 161.52 NA NA NC (153.92) NA NA NC Remote 320.11 NA 130.08 NA (279.83) NA (142.35) NA Notes: NA indicates that no examples of the relevant regional type exist in the state; NC indicates that no calculation was made because of data difficulties associated with the identification of regional boundaries between the two censuses; brackets enclose the 2001 values. Appendix Table Table 2: Median Value of CARS, 2006 and 2001 NSW Victoria Queensland South Australia Metro 132.31 130.82 129.73 132.54 (125.92) (125.26) (121.78) (125.81) Other 137.69 131.40 141.86 NA Metro (130.84) (125.18) (143.92) NA Coastal 139.55 139.20 126.73 148.36 (133.88) (140.06) (126.71) (157.39) Inland 145.13 140.94 136.18 186.94 (148.70) (140.88) (142.66) (212.12) Remote 152.40 NA 163.50 148.57 (147.85) NA (134.10) (146.71) Western Tasmania NT ACT Australia Metro 132.54 146.68 155.49 224.93 (124.75) (133.24) (131.01) (191.97) Other NA NA NA NA Metro NA NA NA NA Coastal 143.37 121.68 NA NA (135.46) (118.67) NA NA Inland 218.39 NA NA 417.95 (199.27) NA NA (234.38) Remote 148.45 NA 179.85 NA (155.40) NA (155.01) NA Notes: NA indicates that no examples of the relevant regional type exist in the state; brackets enclose the 2001 values.
Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES) (2012), Drought in Australia: Context Policy and Management, Report to Client, prepared for the Australia China Environment Development Partnership, March, viewed 25 October, 2013, http://www.daff.gov.au/_media/documents/abares/ publications/client_reports/drought-in-australia-2012.pdf
Australian Bureau of Statistics (ABS) (2011), Australian Standard Geographical Classification (ASGC), viewed 10 November, 2011, www.abs.gov.au/AUSSTATS/ abs@.nsf/DetailsPage/1216.0July%202011?OpenDocument
Australian Bureau of Statistics (ABS) (2008), Australian and New Zealand Standard Industrial Classification (ANISIC), 2006 (Revision 1.0), cat. 1292.0, viewed 4 October, 2011, www.abs.gov.au/AUSSTATS/abs@.nsf/Latestproducts/9685 6082FF78FA68CA25711F001470AE?opendocument
Baxendine, S., Cochrane, W. Pool, I. and Poot, J. (2004), An Interpretation of New Zealand's Regional Employment Change by means of Classical Shift-share Analysis 1986-2001 Labour, Employment and Work in New Zealand, viewed 28 October, 2013 http://ojs.vidoria.ac.nz/LEW/article/view/1263
Beer, A. and Clower, T. (2009), 'Specialisation and Growth: Evidence from Australia's Regional Cities', Urban Studies, vol. 46, pp. 369-389.
Belke, A. and Heine, J. (2006), 'Specialisation Patterns and the Synchronicity of Regional Employment Cycles in Europe', International Economics and Economic Policy, vol.3, pp. 91-104.
Bradley, R. and Gans, J. (1998), 'Growth in Australian Cities', Economic Record, vol. 74, pp. 266-278.
Combes, P. and Overman, H. (2004), 'The Spatial Distribution of Economic Activities in the European Union', in Henderson, V. and Thisse, J. (eds), Handbook of Regional and Urban Economics, Volume 4, North Holland, Amsterdam, pp. 2845-2910.
Connolly, E. and Orsmond, D. (2011), The Mining Industry: From Bust to Boom, Research Discussion Paper, Reserve Bank of Australia, RDP 2011-08.
Dixon, R. and Freebairn, J. (2009), 'Trends in Regional Specialisation in Australia', Australasian Journal of Regional Studies, vol 15, pp. 281-296.
Edwards, B., Gray M., and Hunter, B. (2011), 'The Impact of Drought on Carers', Australian Journal of Labour Economics, vol. 14, pp. 199-214.
Garnaut, J., Connell, P., Lindsay, R. and Rodriguez, V. (2001), Country Australia, ABARE Research Report 2001.1
Garnett, A. M. and Lewis, P. E. T. (2007), 'Population and Employment Changes in Regional Australia', Economic Papers, vol. 26, pp. 29-43.
Horridge, M., Madden, J. and Wittwer, G. (2005), 'The Impact of the 2002-2003 Drought on Australia 'Journal of Policy Modelling, vol. 27, pp. 285-308.
Krugman, P. (1991) Geography and Trade, MIT Press, Cambridge, Mass.
Krugman, P. (1993), 'Lessons of Massachusetts for EMU', in Torres, F. and Giavazzi, F. (eds) Adjustment and Growth in the European Monetary Union, Cambridge University Press, Cambridge, pp. 241-269.
Krugman, P. (2009), 'The Increasing Returns Revolution in Trade and Geography', American Economic Review, vol. 99, pp. 561-571.
Krugman, P. (2010), 'The New Economic Geography, Now Middle-aged', paper prepared for presentation to the Association of Economic Geographers, April 16, viewed 26 September, 2011, www.princeton.edu/~pkrugman/aag.pdf.
Lilien, D. M. (1982), 'Sectoral Shifts and Cyclical Unemployment', Journal of Political Economy, vol. 90, pp. 777-793.
Marelli, E. (2004), 'Evolution of Employment Structures and Regional Specialisation', Economic Systems, vol. 28, pp. 35-59.
Martin, R. and Sunley, P. (1996), 'Paul Krugman's Geographical Economics and Its Implications for Regional Development Theory: A critical Assessment', Economic Geography, vol. 72, pp. 259-292.
Robson, M. (2006), 'Sectoral Shifts, Employment Specialization and the Efficiency of Matching: An Analysis Using UK Regional Data', Regional Studies, vol. 40, pp. 743-754.
Robson, M. (2009), 'Structural Change, Specialization and Regional Labour Market Performance: Evidence from the UK', Applied Economics, vol. 41, pp. 275-293.
Royal Swedish Academy of Sciences (Prize Committee of), (RSAS) (2008),Trade and Geography--Economies of Scale, Differentiated Products and Transport Costs, viewed 26 September, 2011, www.nobelprize.org/nobel_prizes/economics/ laureates/2008/ecoadv08.pdf
Shearmur, R. and Polese, M. (2005), 'Diversity and Employment Growth in Canada', The Canadian Geographer, vol. 49, pp. 272-290.
Stimson, R. J., Robson, A. and Shyy, T. K. (2009), 'Modeling Regional Endogenous Growth: An Application to the Non-metropolitan Regions of Australia', Annals of Regional Science, vol. 43, pp. 379-398.
Suedekum, J. (2006), 'Concentration and Specialisation Trends in Germany Since Re-unification', Regional Studies, vol. 40, pp. 861-873.
Wren, C. and Taylor, J. (1999),'Industrial Restructuring and Regional Policy', Oxford Economic Papers, vol. 51, pp. 487-516.
(1) Dixon and Freebairn (2009) provide an abridged list of authors utilising the concept from Hoover (1948). Dixon and Freebairn (2009) present the Krugman index, which is a simple transformation of CRS.
(i) The shift-share calculations for the Pilbara SD are available from the authors on request.
John Hicks, P.K. Basu and Christopher Sherley *
* Charles Sturt University
Table 1: Classification of Statistical Districts into Regional Types NSW Victoria Queensland Metro Sydney Melbourne Brisbane Other Hunter Barwon Gold Coast Metro Illawarra Sunshine Coast Richmond-Tweed West Moreton Coastal Mid-North Coast Western District Wide Bay-Burnett South Eastern East Gippsland Fitzroy Gippsland Inland Northern Central Highlands Darling Downs Central West Wimmera Murrumbidgee Mallee Murray Loddon Goulburn Ovens-Murray Remote North Western South West Far West Mackay Northern Far North Central West North West South Western Tasmania Australia Australia Metro Adelaide Perth Hobart Other Metro Coastal Outer Adelaide South West Southern South East Lower Northern Great Southern Mersey-Lyell Inland Yorke and Lower North Midlands Murray Lands Remote Eyre Upper Great Northern Southern South Eastern Central Pilbara Kimberley NT ACT Metro Darwin Canberra Other Metro Coastal Inland Rest ACT Remote Rest NT Source: Authors' assessments based on ABS (2011) and Garnaut et al. (2001). Table 2: Shift-share Analysis by Broad Industry Sector and Regional Type Regional Industry 2001 Percentage 2006 Group sectors of total Australia Primary 361792 4.9 336087 Secondary 945824 12.8 874856 Tertiary 6097366 82.3 6775634 Total 7404982 100.0 7986577 Metro Primary 56637 1.1 56287 Secondary 684781 13.7 615225 Tertiary 4270255 85.2 4750248 Total 5011673 100.0 5421760 Other Primary 30545 4.9 28825 Metro Secondary 71963 11.5 73652 Tertiary 522559 83.6 613843 Total 625067 100 716320 Coastal Primary 89365 13.6 80218 Secondary 77067 11.7 78576 Tertiary 489541 74.6 549524 Total 655973 100.0 708318 Inland Primary 103341 16.4 89294 Secondary 75694 12.0 73115 Tertiary 450484 71.6 485394 Total 629519 100.0 647803 Remote Primary 81904 17.0 81463 Secondary 36319 7.5 34288 Tertiary 364527 75.5 376625 Total 482750 100.0 492376 Regional Industry Percentage Percentage National Group sectors of total change share Australia Primary 4.2 -7.1 Secondary 11.0 -7.5 Tertiary 84.8 11.1 Total 100.0 7.9 Metro Primary 1.0 -0.6 4448 Secondary 11.3 -10.2 53783 Tertiary 87.6 11.2 335390 Total 100.0 8.2 393622 Other Primary 4.0 -5.6 2399 Metro Secondary 10.3 2.3 5652 Tertiary 85.7 17.5 41042 Total 100.0 14.6 49093 Coastal Primary 11.3 -10.2 7019 Secondary 11.1 2.0 6053 Tertiary 77.6 12.3 38449 Total 100.0 8.0 51521 Inland Primary 13.8 -13.6 8117 Secondary 11.3 -3.4 5945 Tertiary 74.9 7.7 35381 Total 100.0 2.9 49443 Remote Primary 16.5 -0.5 6433 Secondary 7.0 -5.6 2853 Tertiary 76.5 3.3 28630 Total 100.0 2.0 37916 Regional Industry Industry Regional Total Group sectors mix shift change Australia Primary Secondary Tertiary Total Metro Primary -8472 3674 -350 Secondary -105165 -18175 -69556 Tertiary 139631 4972 479993 Total 25994 -9529 410087 Other Primary -4569 450 -1720 Metro Secondary -11052 7089 1689 Tertiary 17087 33155 91284 Total 1466 40694 91253 Coastal Primary -13368 -2798 -9147 Secondary -11835 7292 1509 Tertiary 16007 5527 59983 Total -9196 10021 52345 Inland Primary -15459 -6705 -14047 Secondary -11625 3101 -2579 Tertiary 14730 -15202 34910 Total -12353 -18806 18284 Remote Primary -12252 5378 -441 Secondary -5578 694 -2031 Tertiary 11919 -28452 12098 Total -5910 -22379 9626 Table 3: Impact of Regional Class on CARS Variable Coefficient t statistic Intercept 139.387 17.583 *** Time 4.335 0.761 Other Metro -10.471 -0.817 Coastal -1.581 -0.168 Inland 18.995 2.048 ** Remote 23.352 2.549 ** R Square 0.156 Adjusted R Square 0.115 Observations 108 * = 10% level, ** = 5% level, *** = 1% level Table 4: Impact of Regional Class on CRRS Variable Coefficient t statistic Intercept 71.011 3.938 *** Time 13.836 1.119 Other Metro 0.266 0.011 Coastal 61.978 2.919 *** Inland 42.203 2.012 ** Remote 145.291 7.004 *** R Square 0.403 Adjusted R Square 0.377 Observations 120 * = 10% level, ** = 5% level, *** = 1% level Table 5: Impact of InCARS on InUr Variable Coefficient t statistic Intercept 0.049 0.038 InCARS -0.537 -2.069 *** Time -0.331 -15.589 *** R Square 0.939 Adjusted R Square 0.875 Observations 108 * = 10% significance level, ** = 5% significance level, *** = 1% significance level Table 6: Impact of InCARS on InNer Variable Coefficient t statistic Intercept 0.196 0.570 InCARS -0.204 -2.961 *** Time -0.082 -16.762 *** R Square 0.990 Adjusted R Square 0.980 Observations 108 * = 10% significance level, ** = 5% significance level, *** = 1% significance level
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Contributed Article|
|Author:||Hicks, John; Basu, P.K.; Sherley, Christopher|
|Publication:||Australian Bulletin of Labour|
|Date:||Mar 1, 2014|
|Previous Article:||Insights into the working experience of casual academics and their immediate supervisors.|
|Next Article:||Guest Editors' introduction to the special issue on FIFO Work.|