RDP 2002-04: Labour Market Adjustment in Regional Australia 4. Explaining Employment Growth in Regional Australia

In an effort to explain why employment has expanded in some regions and contracted in others, we consider some economic characteristics of regions. We then nominate summary measures of these characteristics and attempt to assign a probability to them being present in regions where employment is growing and where it is contracting.

4.1 Factors Influencing Regional Employment

A region's industry composition relative to that of the national economy is one of the principal factors cited in explanations of regional disparities in employment growth (Malizia and Ke 1993; Bradley and Gans 1998; Garcia-Mila and McGuire 1998). If a region's industry composition is skewed toward industries that have recorded a change in employment nationally, regional employment is likely to be disproportionately affected.^[19] A region's industry composition may also be subject to structural change that is more pronounced than in the national economy and may accentuate regional variations in employment growth. Both ABARE (2001) and the Productivity Commission (1999) present evidence that industry composition and structural change are important contributors to regional variations in Australian employment growth.^[20]

Industry composition also has a bearing on the way in which a region responds to shocks. Regions with a diversified industry structure may be less exposed to industry-specific shocks and so, on average, may experience more rapid employment growth than regions with narrow industrial structures (Duranton and Puga 1999). Diversified regions may also benefit from agglomeration economies that permit them to take advantage of intra and inter-industry linkages that stem from many industries being present in a locality.^[21] Bradley and Gans (1998) identify a role for diversity in the relative economic performance of Australian cities.

Furthermore, aspects of geography can be important for employment growth. Prominent among these is the proximity of a region to product and factor markets, since an important factor in a firm's location decision is its access to customers and potential workers (Duffy 1994; Ellison and Glaeser 1999). The size of a region's population may also affect employment growth if there is a critical size below which particular services cannot be maintained (Glaeser and Shapiro 2001). When the population falls below this critical level, some firms may be forced to exit and subtract from employment in the affected region (Productivity Commission 1999).

Related to a region's size and proximity to key markets is the issue of amenity. Amenity – such as attractive physical and cultural characteristics – may influence the location decision of households and businesses. To the extent that amenity attracts people to a region and generates demand, it may also be associated with higher employment growth (Salt 2001). In fact, Salt argues that this has been a significant factor in explaining ‘coastal drift’ in the growth of Australia's population and employment.

Another standard proposition in the regional science literature is that regions with higher levels of human capital experience more rapid growth over the long run (Glaeser and Shapiro 2001). Regions with skilled labour forces may have more success attracting firms to their regions, may be more likely to take advantage of economic opportunities, and may be favoured by skill biased technical change (Bradley and Gans 1998; NIEIR 2001).

Finally, government policies can directly influence regional employment growth, particularly if there is an explicit policy of regional development, as has occurred at various times in Australia.^[22] Government decisions about where to locate public services and utilities also have the potential to influence the growth path of particular regions. Reflecting this, the Productivity Commission (1999) has explored the significance of such government decisions, and broader government policies on the regulation of markets, on regional employment outcomes in Australia.

4.2 The Data

To determine which initial characteristics of a region influenced whether it experienced expanding or contracting employment, we use small area data from the ABS Integrated Regional Database (IRDB). The IRDB combines small area data from the Census of Population and Housing and a range of government agencies. (See Appendix A for details.) Its distinctive feature is that it permits a region to be defined at a given level of aggregation (for our purposes an SLA), and the retrieval of all available data at that level of aggregation. The data available for SLAs are primarily from the census, although we also draw on some data from government agencies to compile a demographic and economic profile of each region at the beginning of our sample period in 1986. From this information, we derive a list of variables capturing the characteristics of regions that may have influenced employment growth.

Industry employment share: The share of regional employment in 1986 in each of the following industries taken separately: agriculture, mining, manufacturing, utilities, retail trade, accommodation, and property and business services.^[23]

Industrial diversity: A modified Herfindahl index that increases as a region's industrial diversity increases to match the diversity of the Australian economy.

Structural change: An index showing the extent to which the industrial composition of employment changed between 1986 and 1996.

Remoteness: The Commonwealth Department of Health and Aging's Accessibility and Remoteness Index of Australia. The Department has allocated a score between 0 and 12, where 0 is the least remote SLA (SLAs in capital cities) and 12 is the most remote SLA, with the degree of remoteness based on the SLA's proximity to service centres of different size.

Human capital: The proportion of a region's population aged 15 and over with a skilled vocational qualification, TAFE qualification, or an undergraduate degree.

Coastal dummy: A region is allocated a value of 1 if it borders the Australian coastline, and is not remote. A region is allocated a value of 0 otherwise.^[24]

Size dummy: A region is allocated a value of 1 if it has a population below 5,000 in 1986 and 0 otherwise.

State dummies: We also include a dummy variable for each state, besides New South Wales, to control for any state-specific effects over our sample period.

While each of these variables (besides the state dummies) follows directly from our review of the factors that influence growth, we do not, however, include a variable for the influence of government policy because we do not have data to identify it, or the set of regions directly affected by the actions of government.^[25]

4.3 A Modelling Strategy

In our attempt to explain the role played by the characteristics of regions in their employment growth, we first distinguish between regions in which employment has expanded and those in which it has contracted. Because employment growth and contraction are mutually exclusive events, we choose an estimation technique that imposes this as a restriction.^[26] Labelling employment growth as 1 and employment contraction as 0, the binomial logit specification defines the probability of a region experiencing employment growth as:^[27]

where X is a vector of regional characteristics.

Because a binomial choice model is a non-linear function of its coefficients, the estimated coefficients provide information about the direction of the effect, but not readily interpretable information about its size. In order to better gauge the relative importance of variables, we present results in terms of an odds ratio associated with a particular variable. This shows the factor by which the odds favouring y=1 change with each 1 unit increase in variable i, holding all other variables constant:

For regional characteristics that are continuous variables, the odds ratio for each X variable is interpreted as the amount by which the odds favouring y=1 change with each 1 percentage point increase in that variable at its mean. For characteristics captured by dummy variables, the odds ratio is interpreted as the amount by which the odds favouring y=1 change when the dummy variable changes from zero to unity.

4.4 Results

The results from the estimation of the binomial logit model are presented in Table 2. The odds ratios for each variable are listed. An odds ratio greater than unity indicates that increases in that variable increased the probability that a region's employment grew between 1986 and 1996. Simple z tests are used to determine whether the odds ratio is statistically different from unity, and a likelihood ratio test of the null hypothesis that all the odds ratios are equal to unity is rejected at all conventional levels of significance. Further, almost 80 per cent of observations were predicted correctly using this model.

Estimation suggests that a number of initial regional characteristics increased the probability that a region experienced an employment expansion rather than an employment contraction. Larger initial shares of employment in the growing service industries of accommodation, cafes and restaurants, and property and business services, increased the probability that a region experienced employment growth. By contrast, larger employment shares in utilities reduced the probability that a region experienced an employment expansion.^[28]

Regions that experienced relatively large structural changes in their industrial composition of employment also experienced, on average, more rapid employment growth than others. This is, in fact, contrary to popular claims about the effects of structural change.^[29] It suggests that between 1986 and 1996 many regions (at least at the level of aggregation we are examining) were successful in changing their industrial structure to include a greater role for industries that are either labour intensive or have recorded above-average rates of employment growth.

Consistent with the observation that structural change has been associated with employment growth, we also find that diversified regions tend to grow more quickly than highly specialised regions. Our diversity index is, however, highly correlated with a region's share of employment in agriculture, raising the possibility that it is specialisation in agriculture that has inhibited employment growth rather than specialisation per se, and that the high correlation between the two variables may also account for the insignificance of the agriculture variable.^[30] In an attempt to account for this, we recalculated the diversity index to exclude agricultural employment. Not only did we find that the diversity index remained significant when agriculture was excluded, but that a region's share of employment in agriculture became significant. Furthermore, the odds ratio indicated that as the share of agricultural employment in total employment increased, so did the probability that a region experienced an employment contraction between 1986 and 1996.

Coastal regions and less remote regions were also more likely to experience employment growth than others, suggesting that both firms and households have been attracted to locate in regions with greater amenity, thereby generating higher growth in local employment.

Point estimates suggest that regions in Queensland, Western Australia, and the Northern Territory were more likely to grow than regions in New South Wales, while regions in Tasmania, Victoria and South Australia were less likely to grow. However, only the odds ratios for Queensland and Western Australia were statistically significant. Because a myriad of factors could be behind these state-specific effects, we merely suggest Queensland and Western Australian regions were more likely to grow than regions in other states for reasons that could not otherwise be accounted for in our model.

Each of these findings accords with the arguments for growth posited in the regional science literature. There are, however, some exceptions. We could not find an independent influence on employment growth for either regional population size or regional human capital.^[31] It may be that many small regions recorded contractions in employment not because of their size but because of other characteristics that they shared. The lack of significance of the human capital variable may be because our measure is a poor proxy for it, or because the influence of human capital may not be detectable over the time horizon of our sample.

Footnotes

Reflecting this, shift-share analysis has been a standard form of assessing regional employment outcomes. (See Hunter (1994) for an Australian example.) [19]

In particular, they show that regions that specialised in agricultural production tended to record slow employment growth, while regions with a large services sector tended to record rapid employment growth. [20]

A tenet of economic geography is that economic benefits (such as knowledge spill overs) occur when firms within a region specialise in the production of a narrow range of goods or services. However, there is increasing evidence that regions benefit most from agglomeration economies when multiple specialisations are present (Duranton and Puga 1999). [21]

For example, in the early 1970s, the Department of Urban and Regional Development and the Cities Commission investigated potential growth centres and introduced a range of public works programs designed to stimulate regional growth. Prominent among their initiatives were the financing of the regional growth centres of Albury-Wodonga and Bathurst-Orange, the national estate program and finance for urban renewal. For an historical perspective on regional policy in Australia see Harris and Dixon (1978). For a recent discussion of the potential and pitfalls of regional development policy in industrialised economies see Braunerhjelm et al (2000). [22]

See Appendix A for detailed information about the definition and construction of all the explanatory variables used in this paper. [23]

In most overseas studies, amenity is proxied by variables relating to weather. Given that data on weather are not available for small areas in Australia, we have chosen a region's coastal location because it encompasses cultural and physical advantages. [24]

Data on state government expenditure on regional development are available. However, these data do not provide information about the amount of expenditure per SLA. [25]

We are primarily interested in the question of what initial characteristics influenced the probability of a region growing or contracting between 1986 and 1996. However, explanations of the rate of employment growth in a region may also be of interest. Consequently, in Appendix B, we report the results of an OLS regression with a region's rate of employment growth as the dependent variable, and the same set of regressors. The results of the OLS regression are broadly consistent with the results of the binomial logit model. [26]

We could also have used a binomial probit model, which has a normal, continuous probability distribution. However, for practical purposes it usually does not matter because the two distributions tend to generate similar probabilities. (See Greene (1993) for a detailed discussion of the circumstances under which the functional form does matter.) [27]

The insignificance of a region's share of employment in agriculture is unexpected, and contradicts earlier research suggesting that falling employment levels in the agricultural sector have detracted from regional employment growth in recent decades. It turns out that this anomaly can be resolved only if we identify specialisation in agriculture and specialisation in other industries separately. We discuss this adjustment to the model a little later on. [28]

This is not to say that for some regions structural change did not have a pronounced negative impact. In recent work, the Productivity Commission (1999) argued that structural change (as defined by a broad structural change index) had an ambiguous impact on regional employment growth between 1986 and 1996 because it depended on the nature of the change, and how it interacted with other regional characteristics such as human capital and natural endowments. Nevertheless, our results suggest that for many regions structural change presented an opportunity for growth. [29]

This is part of a broader problem with the variables we have constructed. A number of the variables in the model are highly correlated, raising the possibility that multicollinearity contributes to their statistical insignificance. For example, population size is negatively correlated with both the diversity index and the share of employment in manufacturing. To check the robustness of the results, variables were excluded from the model one at a time to determine whether the significance of the variables with which they were correlated was altered. We found that of the insignificant variables, only the manufacturing variable was sensitive to the exclusion of correlated variables. From this we infer that regions with higher shares of employment in manufacturing may have been more likely to experience employment growth between 1986 and 1996 than the results in Table 2 imply, which is consistent with evidence presented by ABARE (2001) that manufacturing was a regional growth industry during this period. [30]

As it turns out, regional human capital is one of the few variables for which significance is altered when an OLS regression for employment growth rates was estimated. The OLS results suggest that higher proportions of skilled people in the local population were associated with higher employment growth rates. [31]