Inequality in Australia: Does region matter?

Inequality in Australia: Does region matter? Riyana Miranti, Rebecca Cassells, Yogi Vidyattama and Justine McNamara Corresponding author: Riyana Mira...
Author: Adrian Sharp
18 downloads 0 Views 1MB Size
Inequality in Australia: Does region matter? Riyana Miranti, Rebecca Cassells, Yogi Vidyattama and Justine McNamara

Corresponding author: Riyana Miranti NATSEM University of Canberra ACT 2601 Australia [email protected] Phone: +61 (0)2 6201 5319 Fax: +61 (0)2 6201 2751

PAPER PRESENTED AT THE 2ND GENERAL CONFERENCE OF THE INTERNATIONAL MICROSIMULATION ASSOCIATION, OTTAWA, CANADA, JUNE 8 – 10, 2009

JUNE 2009

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

ABOUT NATSEM The National Centre for Social and Economic Modelling was established on 1 January 1993, and supports its activities through research grants, commissioned research and longer term contracts for model maintenance and development. NATSEM aims to be a key contributor to social and economic policy debate and analysis by developing models of the highest quality, undertaking independent and impartial research, and supplying valued consultancy services. Policy changes often have to be made without sufficient information about either the current environment or the consequences of change. NATSEM specialises in analysing data and producing models so that decision makers have the best possible quantitative information on which to base their decisions. NATSEM has an international reputation as a centre of excellence for analysing microdata and constructing microsimulation models. Such data and models commence with the records of real (but unidentifiable) Australians. Analysis typically begins by looking at either the characteristics or the impact of a policy change on an individual household, building up to the bigger picture by looking at many individual cases through the use of large datasets. It must be emphasised that NATSEM does not have views on policy. All opinions are the authors’ own and are not necessarily shared by NATSEM.

Director: Ann Harding

© NATSEM, University of Canberra 2009 All rights reserved. Apart from fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright Act 1968, no part of this publication may be reproduced, stored or transmitted in any form or by any means without the prior permission in writing of the publisher. National Centre for Social and Economic Modelling University of Canberra ACT 2601 Australia 170 Haydon Drive Bruce ACT 2617 Phone Fax Email Website

2

+ 61 2 6201 2780 + 61 2 6201 2751 [email protected] www.natsem.canberra.edu.au

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

CONTENTS About NATSEM

2

Author note

5

Author note

5

Acknowledgements

5

An appropriate citation for this report is:

5

General caveat

5

Summary

6

1

Background

7

2

Data and methodology

9

2.1 Data

9

3

4

2.2 Spatial microsimulation methodology

10

2.3 Calculating inequality

13

Validation

16

3.1 Small area validation

17

3.2 Aggregated data validation

19

Results

20

4.1 Small Area Results

20

4.1.1 New South Wales

20

4.1.2 Victoria

24

4.1.3 Inequality and Characteristics of SLAs

27

5

Conclusion and policy implications

35

6

References

37

Boxes, figures and tables Figure 1

Lorenz Curve

15

Figure 2

Comparison between Census Gini coefficients and spatial microsimulation estimates for persons, New South Wales, 2006

18

Figure 3

Comparison between Census Gini coefficients and spatial microsimulation estimates for persons, Victoria, 2006

19

Figure 4 Gini Coefficients by Statistical Local Area, New South Wales, 2006

22

Figure 5

Gini Coefficients by Statistical Local Area, Sydney, 2006

23

Figure 6

Gini Coefficients by Statistical Local Area, Victoria, 2006

25

Figure 7

Gini Coefficients by Statistical Local Area, Melbourne, 2006

26

Table 1

Benchmark tables used for SpatialMSM/09C

11

Table 2

Number and characteristics of failed SLAs

13

Table 3

Comparison of Gini coefficient estimates from the 2005-06 Survey of Income and Housing and SpatialMSM/09C

20

3

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Table 4 Table 5 Table 6 Table 7

4

Average proportion of persons in each Gini coefficient group by selected characteristics, all New South Wales, 2006

33

Average proportion of persons in each Gini coefficient group by selected characteristics, all Victoria, 2006

33

Average proportion of persons in each Gini coefficient group by selected characteristics, balance of New South Wales (NSW) and Sydney, 2006

34

Average proportion of persons in each Gini coefficient group by selected characteristics, balance of Victoria and Melbourne, 2006

34

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

AUTHOR NOTE All authors are affiliated with the National Centre for Social and Economic Modelling (NATSEM), University of Canberra. Riyana Miranti is a Research Fellow, Rebecca Cassells is A/g Senior Research Fellow, Yogi Vidyattama is a Senior Research Officer and Justine McNamara is A/g Principal Research Fellow.

ACKNOWLEDGEMENTS This study has been funded by NATSEM. The authors would like to thank the Australian Bureau of Statistics for providing the data that are used in this study. We would also like to acknowledge the involvement of other members of NATSEM’s Social Inclusion and Small Area Modelling Team (Cathy Gong and Robert Tanton) who contributed to the development and the validation of the spatial weights used in this paper.

AN APPROPRIATE CITATION FOR THIS REPORT IS: Miranti, R., Cassells, R., Vidyattama, Y. & McNamara, J. 2009, ‘Inequality in Australia: Does region matter?’, paper presented at the 2nd International Microsimulation Association conference, Ottawa, Canada, June 8 – 10.

GENERAL CAVEAT NATSEM research findings are generally based on estimated characteristics of the population. Such estimates are usually derived from the application of microsimulation modelling techniques to microdata based on sample surveys. These estimates may be different from the actual characteristics of the population because of sampling and nonsampling errors in the microdata and because of the assumptions underlying the modelling techniques. The microdata do not contain any information that enables identification of the individuals or families to which they refer.

5

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

SUMMARY Previous research has argued that the economic boom is not spread evenly across all regions in Australia, and consequently, it is possible that relatively stable national income inequality has been accompanied by increasing gaps between the affluent and the poor at a regional level. This research explores the extent of inequality throughout small areas of Australia using equivalised household disposable income data. Because disposable income data are not available at a small area level, a spatial microsimulation model has been developed in order to calculate inequality. We create household synthetic data at the Statistical Local Area (SLA) level using SpatialMSM/09C, NATSEM’s most current spatial microsimulation model. SpatialMSM/09C draws together data from the 2006 Australian Bureau of Statistics Census of Population and Housing, and the 2003-04 and 2005-06 Surveys of Income and Housing. We estimate inequality at a small area level for two states in Australia – New South Wales and Victoria using conventional Gini coefficient methodology. We also examine the differences in inequality between the densely populated capital cities of each state and the balance of these states. In order to gain greater insight into the driving forces of inequality in these small areas, we examine other characteristics of areas with high income inequality, including immigrant populations, public housing, and poverty and unemployment rates. We find that there are distinct pockets of small areas with high income inequality in both states and their capital cities. However, there are marked differences in what appears to be driving this inequality between the capital cities and the balance of each state and also between each state, highlighting the complexity of income inequality at a small area level.

Key words Income inequality, Spatial microsimulation, Gini coefficient, Microsimulation, Small area estimation

6

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

1

BACKGROUND

Measuring income inequality has long been of interest in applied social and economic research in Australia. In comparison to other OECD countries, the latest data available in the year 2000, shows that Australia sat at around the middle rank (ranked 12 out of 26 reported OECD countries from the highest to the lowest inequality) with a Gini coefficient of 0.305. This was substantially higher than Denmark, which had the lowest inequality of 0.225, but much lower than Mexico, which had the highest inequality of 0.48 (OECD 2006).1 While this provides us with a picture of where Australia falls internationally in terms of income inequality, much more can be said about the nature of inequality in Australia, and in this paper we focus on a sub-national analysis of Australia’s income inequality. Previous Australian studies mostly consider Australian income inequality at a national level, with only a few authors studying this phenomenon at a regional level. However, as there is increasing interest in studying regional diversity in inequality, as discussed by Athanasopoulous and Vahid (2003), policy makers are now interested in examining income inequality within and between regions (Gregory and Hunter 1995; Lyold et al. 2000; Chotikapanich et al. 2005). Just as the fruits of the recent economic boom are not spread evenly across all regions in Australia (Miranti et al. 2008; Saunders et al. 2008; Vu et al. 2008; Meagher and Wilson 2008), it is likely that the average Gini we see nationally is in fact much higher (or lower) in some areas. There is some support for this notion in previous literature. Studies on inequality decomposition in Australia (those which separate “within” and “between” regional inequality) have found that inequality in Australia is due more to inequality within region rather than between regions (Athanasopoulous and Vahid 2003 ; Chotikapanich et al. 2005). Using Theil inequality decomposition to study inequality in Australia for the years 1986, 1991 and 1996 Athanasopoulous and Vahid (2003) find that inequality in Australia is due more to inequality within regions (New South Wales, Victoria, Queensland, Western Australia, South Australia, and combined Australian Capital Territory, Northern Territory and Tasmania) rather than between regions. Further, using a Gini inequality decomposition, it has been found that within state/territory inequality contributes more to national inequality than between state/territory inequality (Athanasopoulous and Vahid 2003; Chotikapanich et al. 2005). Some Australian research has attempted to measure income inequality at a small area. Most of this research uses Statistical Division (SD) level data – a more aggregated geographical level than we use in this study. For example, Maxwell and Peter (1988) measure income inequality in 1976 and 1981 using Australian Census data and further examine the determinants of these inequality differences; McGillivray and Peter (1991) measure income inequality and examined its determinants in 1976, 1981 and 1986. O’Hagan (1999) examines income inequality and population movements in Victoria from

1

Gini coefficients are calculated for persons using household disposable income.

7

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

1981 to 1996 at a statistical division level. However, some current research focuses on more detailed regional disaggregation, although this has often been limited to selected states only. For example, Athanasopoulos and Vahid (2003) measure income inequality in selected and combined small areas (using statistical subdivisions) in New South Wales and Victoria (and some larger areas of other states) whilst, Trendle (2005) calculates income inequality and examines sources of regional income inequality for Local Government Areas (LGAs) in Queensland. This research expands Athanasopoulos and Vahid (2003) with two improvements. First, our research explores “within” inequality for small areas in Australia using equivalised household disposable income data calculated using spatial microsimulation techniques, whilst Athanasopoulos and Vahid (2003) use gross income. As disposable income data is not available at a small area level, a spatial microsimulation model is used to calculate inequality. We create household synthetic data at the small area level using SpatialMSM/09C, NATSEM’s most current spatial microsimulation model. SpatialMSM/09C draws together data from the 2006 Australian Bureau of Statistics Census of Population and Housing, and the 2003-04 and 2005-06 Surveys of Income and Housing. Second, our unit of analysis used in this paper is the Statistical Local Area (SLA), which is a smaller geographical unit than any that has been used in previous Australian studies. We estimate inequality at a small area using conventional Gini coefficient methodology. We limit our analysis to two states - New South Wales (NSW) and Victoria, as the SLAs in these two states are relatively comparable in terms of population size, and therefore issues associated with different levels of heterogeneity in geographical units of different population sizes (known as the Modifiable Areal Unit Problem) are minimized. These states are also the two most populous states in Australia, containing around 58 percent of Australia’s total population in 2006. We calculate Gini coefficients for each SLA in New South Wales and Victoria. There are three objectives in calculating these Gini coefficients. Firstly, to provide valuable information about regional inequality at a small area level at a more disaggregated geographical level than what has been done previously. Secondly, to explore another use of spatial microsimulation and demonstrate its benefits. Previous NATSEM spatial microsimulation techniques are applied to calculate poverty at a small area (see for example Tanton et al. 2008; Miranti et al. 2008) and housing stress at a small area (see for example Cassells et al. 2008). And finally, although, a comprehensive multivariate analysis of determinants of inequality is beyond the scope of this paper, we also examine some small area characteristics which previous research has found to be associated with inequality at a higher level of spatial disaggregation (poverty rates, unemployment rates, proportion of people living in public housing, proportion of Indigenous people, and proportion of people with at least a bachelor degree). The remainder of this paper is organised as follows. Section 2 describes the data and spatial microsimulation methodology (SpatialMSM/09C) used in this paper. Section 3 describes the validation of our results, which are then discussed in Section 4 and section 5 provides a conclusion.

8

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

2

DATA AND METHODOLOGY

2.1

DATA

All data used in this study are originally sourced from the Australian Bureau of Statistics. All income data are sourced from the Survey of Income and Housing (SIH). For the spatial microsimulation analysis, the 2006 Census and the ABS 2003-04 and 2005-06 Surveys of Income and Housing are blended together. The spatial microsimulation methodology is discussed further in section 2.2. Validation is conducted using national, state, and small area level data published by the ABS, using the 2006 Census and the latest available Confidentialised Unit Record Files (CURF) survey data (2005-06 SIH). The 2003-04 SIH has a sample size of 11,361 households whilst the 2005-06 SIH has a sample size of 9,961 households. The sample used by the ABS for the SIH covers occupied private dwellings only. In contrast to the Census, the SIH has rich, detailed information about a range of socioeconomic variables, including disposable income. This detailed information on the SIH allows Gini coefficients to be generated using equivalised disposable income, in comparison to Census data which only provides income in ranges and available as gross income only. However, the SIH does not provide a detailed geographical disaggregation. Therefore, the SIH is suitable for analysing inequality at a larger geographical area such as national or state level, but not for small areas; while the Census is suitable for analysing many household characteristics at a small area level, but does not provide enough income data to create acceptable measures of inequality. Our spatial microsimulation techniques bring these two data sources together, using the Census to provide reliable small area benchmarks, which are used to reweight the SIH data. The Gini coefficients are calculated at person level using household income, as we assume income sharing within households and, prior to conducting the analysis, negative household incomes are recoded to zero to follow the standard approach of the ABS (Li 2005).2 Disposable household income is chosen since this is a better measure for income distribution analysis as it measures resources available to households, (Lyold et al. 2000) and as argued in Harding (1997) there is some evidence that the income tax system has become more progressive and provides an offsetting force to growing inequality of gross income. Nevertheless, in the small area validation section, we will also use gross household income since the income in Census data is only available as gross household income. In common with other research, we equivalise disposable household incomes, so that rankings of income will then take into account the differences that household size and composition make to standards of living. Equivalence scales give ‘points’ to each adult and

2 Li (2005) and Saunders et al. (2008) argue, some analysis has shown that the expenditure patterns of those households with zero and negative incomes are inconsistent with their reported low income.

9

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

child in the household, and then the household’s disposable income is divided by the sum of these points so that incomes can be compared across different types of households. Here we use the modified OECD equivalence scale, which assigns the following values: 1.0 point for the first adult; 0.5 for each of the remaining adults and 0.3 for each dependent child in the household. It should be noted that for the purposes of calculating equivalised income, dependent children are defined as only those children aged less than 15 years, in common with current Australian practice. The spatial unit used in this paper is the Statistical Local Area (SLA). The SLA is one type of standard spatial unit described in the Australian Standard Geographic Classification (ASGC) 2006 and is based on the boundaries of incorporated local government bodies where these exist (ABS 2007a). The 2006 Census data covered 1426 SLAs in Australia. There are 200 SLAs in New South Wales, 210 in Victoria, 479 in Queensland, 128 in South Australia, 156 in Western Australia, 44 in Tasmania, 96 in the Northern Territory, and 109 in the Australian Capital Territory. There are two main reasons why the SLA is used as the unit of analysis in this study. First, the SLA is the smallest unit in the ASGC where there are not substantial issues with confidentiality, as occur with Census Collection Districts. Second, SLAs cover the whole of Australia (as opposed to Local Government Areas which do not cover areas with no local government) and cover contiguous areas (unlike some postcodes) (McNamara et al. 2008). As mentioned earlier, we limit our study to SLAs in New South Wales and Victoria, as those SLAs are more comparable in terms of population size in comparison to the other states (both states have population coefficients of variation of 0.9) .3 To examine the characteristics of these SLAs, most data we use (proportion of immigrants, indigenous, managers and professionals, female labour force participation, bachelor degree and above) are from the Basic Community Profile (BCP) and Expanded Community Profile (XCP) tables of Census data published by the ABS. These data provide Census characteristics of persons, families and dwellings. The proportion of persons who live in public housing is provided through a special data request from the ABS, while the poverty estimates are calculated using the same spatial microsimulation technique that is used to calculate the Gini coefficient. Although the validation of poverty rates is beyond the scope of this paper, we have validated the poverty estimates, and the validation suggests that these estimates are reliable.

2.2

SPATIAL MICROSIMULATION METHODOLOGY

The most current NATSEM spatial microsimulation model is used to calculate inequality at a small area level – SpatialMSM/09C. New versions of NATSEM’s spatial microsimulation

3 The population size of the SLAs used in this study, range from 730 to 133,838 persons for New South Wales and 589 to 94,889 persons for Victoria.

10

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

model are continually being produced as we find new and improved methodologies and specifications that improve the small area weights. Spatial microsimulation is essentially the calculation of a set of small area weights. By combining detail available on data-rich surveys with detail available on the geographically rich Census, we are able to create synthetic data that accurately estimate certain socioeconomic phenomena that are closely related to the benchmarks used to calculate these weights. We combine two available data-rich surveys – the 2003-04 and 2005-06 ABS Surveys of Income and Housing (SIH) in order to maximise the sample size available for modelling and also in order to be compatible with NATSEM’s static microsimulation model STINMOD4, which provides the added potential of policy analysis if the research question calls for this. We then select and derive a set of data that is directly comparable to data available in the 2006 Australian Census (either through publically available data or special request data gained from the ABS). These tables are known as “benchmark” tables, and currently comprise the variables shown in Table 1. As shown in Table 1, most of the benchmark tables are at household level, and only three of them are “person” level benchmarks. Table 1

Benchmark tables used for SpatialMSM/09C

No

Benchmark Table

Level

1

All household type

Household

2

Age by sex by labour force status

Person

3

Tenure by weekly household rent

Household

4

Tenure by household type

Household

5

Tenure by weekly household income

Household

6

Persons in non-private dwellings

Person

7

Monthly household mortgage by weekly household income

Household

8

Dwelling structure by household family composition

Household

9

Number of chidren aged under 15 usually resident in household

Household

10

Number of adults usually resident in household

Household

11

Weekly household rent by weekly household income

Household

12

Gross equivalised weekly household income by age

Person

Source: ABS Census Population and Housing 2006

Both the survey data and the Census are adjusted and manipulated in order to gain alignment for use in the reweighting process. Income values are uprated on both surveys using average weekly earnings in order to coincide with 2006 dollar values and mortgages and rents are also uprated using a factor derived from the Consumer Price Index. Extensive work has been undertaken for all benchmark components to ensure that they have the same definition and coverage on both the Census and the SIH. For example, the SIH 4

Information about STINMOD can be found in Lloyd (2007).

11

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

includes no non-classifiable households (that is, households that contain no persons aged over 15 years or where the collector could not make contact, or where there was insufficient information on the Census form), while some of the person-level Census tables include non-classifiable households. Thus for the age by sex by labour force status benchmark we request a special table from the ABS that exclude non-classifiable households in order to gain the best possible comparison. The reweighting process as it is applied at NATSEM is described in Chin and Harding (2006). The procedure used is a SAS macro called GREGWT which uses an iterative constrained optimisation technique to calculate weights that best represent all the Census benchmarks. The procedure is a generalised regression procedure outlined in Bell (2000). It is used within the Australian Bureau of Statistics to benchmark survey datasets to known population targets, generally at the national or state level. In contrast, this process is used within NATSEM to create a synthetic household microdata file for each SLA in Australia, containing a set of synthetic household weights which replicate, as closely as possible, the characteristics of the real households living within each small area in Australia (Harding et al. 2008). Because the reweighting process is an iterative process, there are areas where the procedure does not find a solution. If there is no solution found after 30 iterations, then the process has not converged. Those SLAs where the process does not converge are usually SLAs where the population is quite different to the sample population – for example, industrial estates or inner city areas. For 29 areas, however, we find that the GREGWT criterion for non-convergence is too strict: even after iterating 30 times and not converging, the estimate we obtain from the weights is still reasonable when compared with the benchmarks. In order to maximise the number of SLAs for which we can produce valid data, we develop a new criteria for reweighting accuracy, which uses the total absolute error (TAE) from all the benchmarks. If the absolute total error from all the benchmarks is greater than the population in that SLA, then the accuracy criteria has failed, and the SLA is dropped from any further analysis. Generally, the convergence criteria and the accuracy criteria provide the same results when an area has obviously not converged; but for marginal areas, the area may reach the maximum number of iterations but still provide a reasonable total absolute error. In the final results, we have applied the TAE criteria rather than the GREGWT convergence criteria. From Table 2, it can be seen that the proportion of persons living in ‘failed’ SLAs in 2006 is very small. Only 0.79 per cent of the total Australian population in 2006 are lost in the reweighting process. While the acceptance rate of SLAs is overall very high (especially when considered in population terms), we loose almost a third of the Northern Territory population in this reweighting process, however as our research concentrates on New South Wales and Victoria, this does not affect our results. It should also be noted that validation of our Gini coefficient estimates (discussed in Section 3) results in the exclusion of some additional SLAs.

12

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Table 2

Number and characteristics of failed SLAs

State/Territory

Total SLAs

Failed SLAs

Proportion of failed SLAs

Proportion of persons living in failed SLAs out of all persons within state/territory

New South Wales

200

2

1.0%

0.34%

Victoria

210

7

3.3%

0.52%

Queensland

479

45

9.4%

0.75%

South Australia

128

7

5.5%

0.32%

Western Australia

0.87%

156

17

10.9%

Tasmania

44

2

4.5%

0.15%

Northern Territory

96

53

55.2%

28.37%

109

16

14.7%

0.61%

1422

149

10.5%

0.79%

Australian Capital Territory AUSTRALIA

Source: SpatialMSM/09C applied to SIH2003-04 and SIH2005-06, ABS Census Population and Housing 2006

2.3

CALCULATING INEQUALITY

There are various ways to measure inequality (see Vidyattama (2008) and ABS (2006b) for a summary of measures of inequality including the Theil and Atkinson Index). This paper uses a Gini coefficient which measures disparity between each person in the population and every other person in the population. We use Gini coefficients to measure inequality for two reasons as follows (i) the Gini coefficient is the most commonly used summary measure (Athanasopoulous and Vahid (2003) and ABS (2006)); and (ii) Gini coefficients are published (at the national and state level) and thus allow us to validate our spatially microsimulated small area Gini coefficient estimates. 5 As discussed in Athanasopoulos and Vahid (2003), the Gini coefficient satisfies the three basic criteria of acceptable inequality measures proposed by Sen (1993), 1) Scale independence: which states that if every individual’s income in a society changes at the same proportion, the inequality measure in that society will not change; 2) Invariance to replication of population: which means if the population size is doubled by keeping the exact characteristics of the original population, the inequality measure will not change; 5 The Gini coefficient is the only statistical measure of income distribution included in the published output from the ABS Survey of Income and Housing as (i) it is not overly sensitive to extremely low incomes and (ii) it is relatively simple to interpret (ABS 2006). However, although the calculation of Gini coefficient is considered simple, we should also acknowledge its weakness. For example, ABS (2006) mentions that the Gini coefficient is too sensitive to relative changes around the middle of the income distribution as the calculation of Gini coefficient reflects the ranking of the population and the ranking is most likely to change around the middle of the distribution. In addition, it is possible to have two coefficients that are not comparable if the Lorenz curves attached to these two coefficients intersect (Vidyattama 2008).

13

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

3) Compliance with the Pigou-Dalton principle of transfers: which means that inequality is expected to increase if there is an income transfer from a poorer person to a richer person and to decrease if the transfer is from a richer person to a poorer person (see Athanasopoulos and Vahid (2003) for further detail). There are several ways to calculate the Gini coefficient. Assuming that we will use equivalised disposable household income, basically, the Gini coefficient is calculated using the following formula (ABS 2006b, pp. 6).

⎛ 1 ⎞n G = ⎜⎜ 2 ⎟⎟∑ y i − y j ⎝ 2n μ ⎠ i , j

(1)

Where; n = the number of people in the population μ = the mean equivalised disposable household income of all people in the population And yi and yj are the equivalised disposable household income of the ith and jth persons in the population. Alternatively, the Gini coefficient can also be calculated by examining the Lorenz curve. The Lorenz curve is a curve with the horizontal axis showing the cumulative proportion of the persons in the population ranked according to their income and with the vertical axis showing the corresponding cumulative proportion of equivalised disposable household income.

14

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 1

Lorenz Curve

Proportion of total income

G =0 when all people have the same level of income

A

B

Lorenz curve

Proportion of population, ranked from poorest to richest Source: modified from Athanasopoulos and Vahid (2003)

The Gini coefficient is a measure of the area between the Lorenz curve and the 45 degree line, calculated with this formula:

⎡ A ⎤ G=⎢ ⎣ A + B ⎥⎦

(2)

The Gini coefficient has a value between zero and one. A value of zero means perfect equality, a situation in which everyone in the population lives in a household with the same level of equivalised income. A value of one indicates perfect inequality, a situation where one person holds all the income. Smaller Gini coefficients indicate a more equal distribution of income. In this paper, as explained earlier, we apply the weights generated from the spatial microsimulation model to calculate Gini coefficients at a small area. Therefore, the estimates of Gini coefficients are calculated using a weighted Gini formula as adapted in Harding and Greenwell (2001). Applying household weights in each SLA to calculate Gini coefficients is challenging as we do not have actual unit record data per person to calculate the Lorenz curve. Therefore, as highlighted in Harding and Greenwell (2001), the income distribution is determined by a ranking of a people by their equivalised household income. Consequently, if a household has five people, their equivalised income will be counted five times, not once.

15

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Consequently, the weighted Gini coefficient for each SLA is calculated with the following

⎡ N ⎛⎡ n ⎞⎤ ⎡ y j w j ⎤⎤ ⎢ ∑ ⎜ ⎢∑ y j w j − ⎢ * wj ⎟⎥ ⎥ ⎥ ⎟⎥ ⎢ j =1 ⎜⎝ ⎢⎣ j =1 ⎣ 2 ⎦ ⎥⎦ ⎠ formula: G = 1 − 2 ⎢ ⎥ N N ⎢ ⎥ y w * w ∑ ∑ j j j ⎢ ⎥ j =1 j =1 ⎣ ⎦ (3) Where; yj is the equivalised disposable household income of the nth household in the population. wj is the person weight of the nth household in the population (the person weight is defined as household weights created for each SLA multiplied by the number of people that reside in the household). The person weights are proportional to the population from which the household sample observations are drawn. N is the total households in the population

3

VALIDATION

As the estimates of the Gini coefficient are calculated using spatial microsimulation techniques, for which the methodology is still being developed and refined, we undertake a set of validation procedures in order to check the accuracy of our synthetic estimates. As a first step, and after already excluding those SLAs (two SLAs in New South Wales, and seven SLAs in Victoria) which had failed (either using the GREGWT convergence criteria or our additional accuracy criteria as described above), we further exclude one SLA in New South Wales and five SLAs in Victoria where the estimated population size is less than 30 persons, as these population sizes are likely to be too small to produce reliable estimates of small area Gini coefficients. This additional criterion gave us 197 SLAs in New South Wales and 198 SLAs in Victoria. We use two procedures to validate the Gini coefficient estimates: 1) Small area validation against estimates of Gini coefficients calculated using the 2006 ABS Census of Population and Housing equivalised gross household income data. We calculate the R2 of the relationship between the estimates of Gini coefficients from the spatial microsimulation and Census data. Further we also calculate the Spearman rank correlation to measure the rank correlation between the two estimates. 2) Aggregated data validation against direct estimates of Gini coefficients at a capital city and balance of state level, calculated using the 2005-06 ABS Survey of Income and Housing and equivalised disposable household income data.

16

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

3.1

SMALL AREA VALIDATION

We use data directly from the 2006 ABS Census of Population and Housing (the source of the regional data for this project) as a point of comparison for our small area Gini coefficient estimates. The Census contains equivalised gross household income, and these data are only available in categories (in this case there are eight groups of income), not as a continuous variable. Some earlier literature has used Census data to calculate Gini coefficients (Needleman 1978; Athanasopoulous and Vahid 2003). Census data allow us to calculate an approximate level of Gini coefficients using equivalised gross household income, and then, using the same calculation with our synthetic data, compare outcomes on the actual versus the synthetic measures using gross household income (so we compare ‘apples’ with ’apples’ although for the discussion of the results later on, we will only analyse Gini coefficients based on equivalised disposable household income). To calculate Gini coefficients using the Census data, we need to apply the median value from each income range as the variable is categorical. For our purpose, we use median income for each income range, calculated from the 2005-06 Survey of Income and Housing. Consequently, we have different median income values for each income range for New South Wales and Victoria and we allocate all persons within each income range the same income value as we assume that there is an even distribution of income within each income range, and that the data are not skewed. For example, there are 1617 persons in the SLA of Melbourne – remainder that fall in the Census income category of $1-$149 per week. Thus, for the purpose of calculating the Gini coefficient, we assume that each of these 1617 persons have the same equivalised gross household income of $73.10 which is the median equivalised income of this income range in the 2005-06 SIH. Figures 2 (for New South Wales) and 3 (for Victoria) show the regression/trend lines of the relationship between Gini coefficients from the Census data (Y-axis) and Gini coefficients from the spatial microsimulation data (X-axis) . These two figures show a consistent and positive bias of the spatial microsimulation estimates. The spatial microsimulation estimates are consistently higher than the Census estimates which may be related to the difference in whether the data is categorical or continuous, as the Gini coefficients calculated from the Census are based on grouped income data in contrast to the Gini coefficients calculated from spatial microsimulation which are calculated based on continuous income data.

17

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 2

Comparison between Census Gini coefficients and spatial microsimulation estimates for persons, New South Wales, 2006

0.500

Gini coefficient gross household income (Census data)

0.450

0.400 y = 0.918x R2 = 0.897

0.350

0.300

0.250 0.250

0.300

0.350 0.400 Gini coefficient gross household income (SpatialMSM/09c)

0.450

0.500

Source: SpatialMSM/09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

Despite these differences, we find that our estimated Gini coefficients and Census Gini coefficients are closely matched for most SLAs, both for New South Wales and Victoria. For New South Wales the relationship between these two estimates has an R2 of 0.897, while for Victoria, the R2 is 0.908.6 Thus we are confident that the weights we have used are giving reasonable results for the small areas in these two states.

6 We also conducted some sensitivity analysis by deleting SLAs which had a 0.05 point difference between the two Gini coefficient estimates. We found that this did not substantially improve the R2 (it increased to 0.910 for New South Wales and 0.914 for Victoria).

18

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 3

Comparison between Census Gini coefficients and spatial microsimulation estimates for persons, Victoria, 2006

0.500

Gini coefficient gross household income (Census data)

0.450

y = 0.901x 2 R = 0.908

0.400

0.350

0.300

0.250 0.250

0.300

0.350 0.400 Gini coefficient gross household income (SpatialMSM/09C)

0.450

0.500

Source: SpatialMSM/09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

In addition, we also calculate the Spearman rank correlation which is used to measure the linear relationship between the two sets of ranked data (in this case the ranks between the spatial microsimulation estimates and Census estimates) to measure how tightly the ranked data clusters around a straight line (Altman, 1991). The Spearman rank correlation has a range of value between -1 and +1. A perfect correlation of +1 or -1 reflects a linear positive or negative correlation between the ranks of the two data sets, whilst in contrast, a correlation close to zero means there is no linear relationship between the ranks. The Spearman rank correlation is 0.958 for New South Wales and 0.953 for Victoria, reflecting a high and positive linear relationship between these two ranked data.

3.2

AGGREGATED DATA VALIDATION

We also have conducted additional validation using equivalised disposable household income from the 2005-06 Survey of Income and Housing data. Income collected from the survey is argued to provide more accurate estimates of the distribution of income than income collected from the Census as interviewers are involved in collecting data directly from the survey respondents, whereas for the Census, the respondents complete the Census questionnaires without an interviewers’ guidance (Maxwell and Peter 1988). As Gini coefficients cannot be calculated directly from the SIH at a small area level using our regional estimates, we aggregated the Gini coefficients by SLA to the state and capital city and balance of state levels in order to compare these results directly with results available at this geographic level from the SIH.

19

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Table 3

Comparison of Gini coefficient estimates from the 2005-06 Survey of Income and Housing and SpatialMSM/09C

State New South Wales

Victoria

Capital city/ Balance of state All Sydney Balance of state All Melbourne Balance of state

Australia

SpatialMSM/09C 0.322 (+) 0.324(+) 0.300(+) 0.306 0.308(-) 0.290(+) 0.308(+)

2005-06 SIH 0.317 0.321 0.287 0.306 0.309 0.274 0.307

* * * *

Note: + (-) indicates where the estimates from the spatial microsimulation are higher or lower than the estimates directly from SIH 2005-06; * indicates that the coefficients have been calculated by NATSEM. The Gini data at the capital city and balance of state level are not available from the ABS publication. Source: ABS (2007b; 2008) and SpatialMSM/09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

Table 3 indicates that while the Gini coefficient estimates from spatial microsimulation tend to be higher than estimates from the 2005-06 SIH, the results are generally aligned. The slightly higher estimates from SpatialMSM/09C are possibly due to benchmarking with Census data, which may have a greater amount of persons reporting lower incomes than the sample in the SIH, however overall the results are very promising. We also find that the Gini coefficient estimates for the balance of each state have greater differences than those estimated for each capital city. This may be due to the higher contribution of capital city SLAs when the weights are created or the fact that the Census coverage is higher than the SIH, therefore it is able to capture more heterogeneity in the data including extreme values especially in the balance of state areas.

4

RESULTS

The aggregated data validation above (Table 3) shows that in general the capital cities have a higher Gini coefficient when compared to the balance of each state. This finding confirms previous research (Lyold et al. 2000; Bray 2001). This may reflect that the capital cities are more heterogeneous in terms of income than the balance of state areas, as cities have a predominance of upper-middle income households together with very low income households. Table 3 also shows that incomes in New South Wales are distributed relatively more unequally than incomes in Victoria. The following section will discuss these differences further at a small area level.

4.1

SMALL AREA RESULTS

4.1.1

New South Wales

Figures 4 and 5 show Gini coefficients by Statistical Local Area for the whole of New South Wales and Sydney, the state capital. In this paper, we apply natural breaks to the data

20

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

when creating the maps as we are more interested in examining the income distribution of SLAs within each state, than comparing one SLA relative to the other SLAs within a state.7 Figure 4 shows the natural breaking of Gini coefficients for New South Wales SLAs. The 197 areas are ranked and then divided into five categories according to where the greatest differences are in the data. Similarly Figure 5 applies the same natural break categories for the whole of New South Wales to Sydney. The palest colour on the map represents areas that have the lowest income inequality category (the lowest Gini coefficient) while, in contrast, the darkest colour on the map represents areas with the highest income inequality (the highest Gini coefficient). The missing data on maps represents the excluded SLAs due to inadequate estimates and population sizes, as discussed earlier. The Gini coefficients in New South Wales vary between 0.262 and 0.369. Comparing figures 4 and 5, over 50 per cent of SLAs in the highest income inequality category (29 SLAs) lie within the capital city - Sydney (16 SLAs). SLAs with high income inequality are mostly clustered in Sydney, with some additional high inequality SLAs scattered throughout the state balance. In Sydney, these SLAs run in a horizontal corridor, from east to west, starting at the inner city suburbs of Waverley, Woollahra, Randwick, Ashfield and Strathfield, and flowing out along the western motorway (M4) and the major train line, towards the western suburbs of Auburn and Parramatta. For the balance of New South Wales, most SLAs with high inequality are concentrated in the remote far west and north western areas of New South Wales. However, there are several small areas with high income inequality scattered throughout New South Wales, including Conargo, Urana, Palerang and Mid-Western Regional. Further, Newcastle, a major urban area, just north of Sydney (unable to be identified readily on the map) is also an SLA that falls within the highest income inequality grouping.

7

There are various ways to represent data on maps (see Miranti et al. 2008 for a summary).

21

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 4 Gini Coefficients by Statistical Local Area, New South Wales, 2006

Source: SpatialMSM/09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

22

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 5 Gini Coefficients by Statistical Local Area, Sydney, 2006

Source: SpatialMSM/09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

23

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

4.1.2

Victoria

Figures 6 and 7 show the geographic distribution of income inequality for Victoria and Melbourne, by applying the natural break classification for the whole of Victoria. Figure 6 shows the 198 Victorian SLAs divided into five categories and similarly for Melbourne (Figure 7), using the same categories. As with New South Wales, the palest colour on the map represents the areas with the lowest income inequality and the darkest colour, areas with the highest income inequality (the highest Gini coefficient). Figure 7 shows that the Gini coefficients in Victoria vary between 0.247 and 0.407. Figure 7 also shows that there are only five SLAs in Victoria that fall within the highest category of inequality, and all of these SLAs lie within the Melbourne city statistical division. These SLAs are clustered in the inner city area of Melbourne and include the SLAs of Melbourne – Remainder, Port Phillips – West, Stonnington, Yarra – North and Yarra – Richmond. No SLAs in the balance of Victoria fall into the highest inequality category, however, from Figure 7, we can see that there is a large cluster of SLAs in the west of the state that fall into the second highest category of income inequality – for example, West Wimmera, Hindmarsh, Moyne, Yarriambiack, Grampians and Corangamite. From these data it can be seen that in Victoria, all high inequality SLAs fall within the capital city - Melbourne, rather than the balance of Victoria. However for New South Wales, the high inequality SLAs are spread evenly between the balance of state and capital city, with a little over 50 per cent of high inequality SLAs found in Sydney (although it should be noted again that the definition of ‘high inequality’ in these maps differs between the two states due to the differences in natural breaks). These results lead us to the next question of what the characteristics of SLAs with high inequality are, and whether these characteristics vary between states and the balance of state and capital city.

24

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 6

Gini Coefficients by Statistical Local Area, Victoria, 2006

Source: SpatialMSM09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

25

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Figure 7 Gini Coefficients by Statistical Local Area, Melbourne, 2006

Source: SpatialMSM09C applied to 2003-04 and 2005-06 SIH, ABS Census Population and Housing 2006

26

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

4.1.3

Inequality and Characteristics of SLAs

A conclusive empirical study using econometric techniques to analyse the determinants of inequality is beyond the scope of this paper, but in this section we present some characteristics that are likely to be highly related to the variation in income inequality that exists across SLAs. We firstly examine the overall characteristics of each Gini coefficient grouping in each state and then use a framework of capital city and balance of state in order to gain greater insight into these areas. We seek to answer the following research questions: (i) Are there similar patterns between New South Wales and Victoria? (ii) Do high inequality SLAs in capital cities have similar characteristics to high inequality areas in the balance of the state? It is important to note that when we compare the two states that the states do not have the same classifications of income inequality, as each states Gini coefficient distribution classification is calculated separately using natural breaks. Previous research in Australia discusses several indicators that are considered as determinants of inequality (Maxwell and Peter 1988; McGillivray and Peter 1991; Trendle 2005). In this section, we analyse some of these indicators namely the proportion of immigrants in the population, the proportion of Indigenous persons in the population, the proportion of the population whose occupation is managers or professional, female labour force participation rates, unemployment rates, the proportion of the population with post school qualifications, poverty rates (not as a determinant, but at least correlated with inequality) and the proportion of the population living in public housing. Interestingly, the relationships of many of these variables to inequality, particularly regional inequality are ambiguous. The proportion of immigrants in the population may have an ambiguous correlation with regional inequality as it is argued that there are two different groups of overseas migrants in Australia. The first group are an older generation of immigrants such as Italian immigrants who arrived in Australia after WWII in the 1950s and Vietnamese who arrived in the mid-1970s after the Vietnam War, as discussed in Greig et al. (2003). These immigrants are characterised by limitations such as language barriers, low education levels, and racial discrimination, which in turn has resulted in a concentration of low paid occupations. This group of immigrants are more geographically dispersed and consequently their presence is expected to have a positive correlation with inequality. On the other hand, the relationship between the second group of immigrants who are a younger generation that tend to be more concentrated in specific regions and earning similar levels of income. However, the relationship between this group of immigrants and inequality are ambiguous. It is expected that this group of people has a negative correlation with inequality (McGillivray and Peter 1991). Nevertheless, although it is noted that the majority of these younger generations of migrants are highly skilled, yet Greig et al. (2003) find that their qualification may not be readily recognised. In addition, these immigrants are likely to be concentrated in a few urban areas, and strongly attached to their community, and some experience limited labour market opportunities or what Greig

27

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

et al (2003) argue as being limited by the “ethnic mobility trap” (Greig et al, 2003 pp.128), and consequently having a positive relationship with inequality. The proportion of Indigenous persons in SLAs is argued to have a positive correlation with regional inequality as these populations tend to have lower levels of educational attainment, do not have as many opportunities to work in highly skilled occupations (Trendle 2005), and are more likely to be unemployed and dependent on government benefits. Previous research has found a positive relationship between female labour force participation and inequality with two different reasons. However, these differences are likely to be due to the differing types of income used to measure inequality. Trendle (2005) finds that female labour force participation has a positive correlation with inequality (measured using personal income) as this group of workers may have a higher probability of experiencing a career break/disruption or to work part time in order to care for children or families and earn lower incomes. However, McGillivray and Peter (1991) argue if those females who enter a labour market in a region are married, the number of families with double incomes in that particular region will also increase, which is expected to lead to greater inequality (measured using family income). Unemployment rates are also found to have an ambiguous relationship with inequality. Trendle (2005) proposes that unemployment rates have a positive correlation with inequality as higher unemployment rates are associated with low income. However, Maxwell and Peter (1988) also argue that an opposite relationship can exist, if the region lacks high income earners, as these people prefer to live in more prosperous regions. Therefore if the income effect is higher than the unemployment effect, we may expect to find a negative relationship between unemployment and inequality. Much of the previous literature discusses the relationship between a country’s income distribution and the stage of economic development (Kuznet 1955; Williamson 1965 for international literature and Amos 1986; Maxwell and Peter 1988 and McGillivray and Peter (1991) for Australia). The stage of development is usually represented by ‘real’ average income which is adjusted to reflect different costs of living across areas in Australia (Maxwell and Peter 1988). Since it is difficult to find a comprehensive cost of living index, Maxwell and Peter (1988) suggest using the proportion of the population with a post school qualification to represent the level of development. A high proportion of this variable may represent a higher level of regional development (Maxwell and Peter 1988). The proportion of the population with post school qualifications is expected to have a negative association with inequality (Maxwell and Peter 1988; Trendle 2005). Nevertheless, the relationship between inequality and the proportion of people with post school qualifications may also be ambiguous as Amos (1986) as cited in McGillivray and Peter (1991, pp. 137) argues, there is an “augmented inverted U” relationship between income inequality and the level of development, whilst McGillivray and Peter (1991, pp. 140) argue “short-run oscillations” between income inequality and the level of development which proposes a possibility of positive association between income inequality and the level of development. Glaeser, Resseger and Tobio (2008) find that an increase in the share of college graduates increases

28

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

inequality (using the Gini coefficient as a proxy) in urban areas. In this paper, we use the proportion of people with at least a bachelor degree to represent post school qualifications. The proportion of the population who are employed as either a manager or professional represents a highly skilled workforce, and these persons generally have higher incomes than employees in other occupations (Lyold, 2000). People who are employed as managers and professionals are also most probably those people with a post school qualification. Glaeser, Resseger and Tobio (2008) argue that the inequality of human capital within a region is related to returns to skill, and that those with a post school qualification will be concentrated in places where the return to human capital is higher (mostly in urban areas). Thus, it is also likely these persons with highly skilled occupations are concentrated together, and follow a similar pattern for those people who have a post school qualification, thus areas with a high proportion of people employed in professional and managerial occupations may also have high inequality. We are unable to find any literature detailing the relationship between inequality and public housing, however in the Australian context, it is likely that persons living in public housing also have very low incomes. The presence of public housing within small areas is likely to increase income inequality, as it increases the number of persons in the lower end of the household income distribution, however this will also depend upon where the public housing is primarily located. The relationship between inequality and poverty is usually ambiguous. The literature tends to discuss how income is distributed and the impact of income distribution on poverty, rather than vice versa (Miranti 2007). Nevertheless, as poverty identifies those people at the bottom end of the income distribution, it is interesting to examine the relationship between poverty and inequality, to examine whether those areas with high inequality are also (or sometimes) those areas with high poverty. Table 4 shows the average proportion of persons in each Gini coefficient group by selected characteristics for all of New South Wales. From Table 4 it can be seen that SLAs in the highest inequality group are characterised by,on average, high proportions of immigrants, Indigenous persons, people working as managers and professionals, female labour force participation, people having a bachelor degree or higher, and living in public housing (in comparison to other Gini coefficient groups). It is interesting that the lowest inequality group in New South Wales also has a high proportion of immigrants, and high female labour force participation. Table 5 shows similar results for Victoria. In general, SLAs which fall into the highest inequality group in Victoria, except for the proportion of population who are Indigenous (no clear pattern), show the same pattern as in New South Wales and have high average proportions of immigrants, people working as managers and professionals, female labour force participation, persons with a bachelor degree or higher, and persons living in public housing. To answer whether high inequality SLAs in each capital city have similar characteristics to high inequality SLAs in the balance of state, Tables 6 (for New South Wales) and 7

29

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

(Victoria) show the average proportion of persons in each Gini coefficient group by selected characteristics for the balance of state and capital city. Table 6 shows some clear differences in characteristics between Sydney and the balance of New South Wales, and enables the patterns in Table 4 to be further fleshed out. The average proportion of immigrants has an opposite relationship for Sydney and the balance of NSW, with the highest average population of immigrants in the balance of New South Wales in the lowest inequality group, yet for Sydney, the highest average proportion of immigrants is in the areas of the city with the highest levels of income inequality. These differences possibly reflect the two different immigrant populations discussed above – with the older immigrant population more likely to be scattered throughout the regional areas of New South Wales, and the newer immigrants more concentrated in the inner city suburbs of Sydney. The effect of average Indigenous populations for each area also differs, with an increasingly larger average proportion of Indigenous persons as we move up the inequality groupings (a relationship that is supported in the literature). In the highest inequality grouping in the balance of New South Wales, around 15.2 per cent of the population are Indigenous, whereas, for the lowest grouping, only around 3.7 per cent of the population are Indigenous. For Sydney, overall, there are much lower average populations of Indigenous persons throughout the SLAs, ranging from only 0.7 to 1.7 per cent, and the pattern of Indigenous persons for each inequality grouping appears to be the reverse of that in the balance of New South Wales, with the highest proportion of Indigenous persons in the lowest inequality grouping. The greater presence of persons working at higher occupational levels (managers and professionals), has a positive relationship with inequality levels, increasing as inequality does for both the balance of New South Wales and Sydney. The large presence of managers and professionals (nearing 50 per cent), in areas of high inequality would also suggest that these persons are more highly paid, however on further investigation of areas in the balance of New South Wales that have these qualities, there appears to be two areas that are contributing heavily to the weight of this average – Conargo and Unincorporated Far West. These areas have very low populations and are also predominately farming areas, suggesting that most persons reporting themselves as managers and professionals are likely to be holding these positions within the agricultural industry, and likely to be paid less than managers and professionals in city areas, within other industries. The average female labour force participation has a U shaped pattern, where the highest average participation rates are in both the lowest and highest income inequality groupings. The high average female labour force participation in both the lowest and highest inequality groupings is likely to represent two different female populations which will be valuable for future research. For the inner city suburbs, it’s likely that the group of women living here are mostly single or partnered professionals working full-time without children. For the lowest income inequality areas, yet high female labour force participation, the group of women here are likely to be women with children in the suburbs.

30

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

The presence of highly qualified persons (those with a bachelor degree or higher), as found in Glaeser, Resseger and Tobio (2008) for urban areas in the US, appears to be associated with high levels of inequality in Sydney, however has no real effect on the balance of New South Wales, with this characteristic remaining relatively flat across all groupings of inequality. The lower rates of highly qualified persons in the balance of New South Wales, together with the high proportion of self-reported managers and professionals supports the above theory, that these persons represent a different classification of managers and professionals than those in the capital city, and are more likely to be farm managers and the like. The average proportion of persons living in public housing does not appear to be very strongly correlated with inequality groupings. For Sydney, public housing fluctuates between groups and there is no clear pattern. This could be due to public housing being located primarily in areas that have not too dissimilar private housing income distributions. For the balance of New South Wales, the highest inequality grouping also has the highest average proportion of persons in public housing – at 5.2 per cent. In results not shown, poverty rates and unemployment rates are also examined for each area. For unemployment rates no identifiable pattern is found, however for poverty rates, there is a positive relationship between poverty and inequality in the balance of New South Wales only. As the proportion of persons in poverty increases, so do the levels of inequality. For Sydney however, there is no clear pattern between poverty and inequality. Table 7 also enables the patterns in Table 5 to be further investigated, with a break-down of the balance of Victoria and Melbourne. Inequality increases as the average proportion of immigrants generally increases, however this does fluctuate a little. Overall, there is a higher proportion of immigrants within Melbourne than in the balance of Victoria. The proportion of Indigenous persons appears to have little effect on inequality, and in contrast with New South Wales there are very small populations of Indigenous persons within Victoria. Similar to New South Wales, the greater presence of persons working in the higher occupational levels (managers and professionals) has a positive relationship with inequality levels, increasing as inequality does for both the balance of Victoria and Melbourne. This may reflect the previous literature which finds that those highly skilled people tend to concentrate in certain areas. The average female labour force participation has a U shaped pattern for Melbourne, where the highest average participation is in both the lowest and highest income inequality groupings. In the balance of Victoria, the average female labour force participation decreases as inequality increases, suggesting that increased female labour force participation is associated with lower inequality levels in these areas. The presence of highly qualified persons (those with a bachelor degree or higher), has a very similar effect on inequality in Melbourne and the balance of Victoria as it does for the like areas in New South Wales. For the balance of Victoria, the population with a bachelor degree or higher remains relatively flat across all groupings of inequality. For Melbourne,

31

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

the proportion of persons with post-school qualifications is highest on average in areas with the greatest inequality. The proportion of persons in public housing appears to have a negative relationship with inequality in the balance of Victoria, with the highest proportion of persons in public housing also in the second lowest income inequality group. As with Sydney, this could reflect the location of public housing in the balance of Victoria, with areas that have similar income distributions. For Melbourne, public housing has an opposite relationship, with the highest proportion of persons in public housing (around 10 per cent) also in the group with the highest inequality. This relationship is the opposite to that experienced in Sydney, and is likely to be due to the majority of Melbourne public housing being located in the inner city areas, alongside households with very high incomes. In results not shown, poverty rates and unemployment rates are also examined for each area. As is the case for New South Wales, no pattern is found between unemployment and inequality. For poverty however, similar patterns emerge in Victoria, with inequality increasing as poverty increases in the balance of Victoria, yet in Melbourne there is no clear pattern between poverty and inequality. To conclude this section, we find that New South Wales and Victoria mostly do have similar patterns in terms of the characteristics of areas with high inequality with the exception of the indigenous characteristic.

32

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Table 4

Average proportion of persons in each Gini coefficient group by selected characteristics, all New South Wales, 2006

Gini coefficient – natural breaks

Immigrants

Indigenous

Managers and professionals

Female LFPR

Bachelor +

Public housing

%

%

%

%

%

%

1 (lowest inequality)

16.37

2.61

28.56

56.25

21.43

3.42

2

12.21

3.27

30.59

51.28

20.27

3.94

3

12.73

3.06

33.97

51.12

22.13

3.54

4

15.53

4.62

39.70

52.04

25.55

3.33

5 (highest inequality)

23.79

7.19

46.17

54.59

33.46

4.51

Source: ABS Census Population and Housing 2006

Table 5

Average proportion of persons in each Gini coefficient group by selected characteristics, all Victoria, 2006

Gini coefficient – natural breaks

Immigrants

Indigenous

Managers and professionals

Female LFPR

Bachelor +

Public housing

%

%

%

%

%

%

1 (lowest inequality)

17.98

0.56

25.48

59.97

22.09

1.57

2

14.29

0.71

30.11

55.43

23.71

2.47

3

14.13

1.07

33.41

51.41

21.96

2.55

4

16.99

0.53

44.22

53.17

32.62

2.08

5 (highest inequality)

30.18

0.31

56.23

60.79

54.69

9.69

Source: ABS Census Population and Housing 2006

33

Inequality in Australia: Does region matter ? NATSEM JUNE 2009

Table 6

Average proportion of persons in each Gini coefficient group by selected characteristics, balance of New South Wales (NSW) and Sydney, 2006

Gini coefficient – natural breaks

Immigrants

Indigenous

Managers and professionals

Female LFPR

Bachelor +

Public housing

NSW balance

Sydney

NSW balance

Sydney

NSW balance

Sydney

NSW balance

Sydney

NSW balance

Sydney

NSW balance

Sydney

1 (lowest inequality)

9.05

22.46

3.66

1.73

28.49

28.61

53.00

58.95

19.00

23.45

3.46

3.39

2

8.68

25.22

3.74

1.54

30.86

29.60

50.65

53.58

18.90

25.25

3.34

6.14

3

8.23

33.00

3.61

0.59

33.64

35.48

50.56

53.67

19.94

31.98

3.54

3.58

4

5.99

36.40

6.50

0.52

38.86

41.52

51.65

52.89

18.85

40.21

2.96

4.13

5 (highest inequality)

5.90

38.34

15.17

0.71

44.66

47.39

54.44

54.70

19.37

44.90

5.17

3.98

Source: ABS Census Population and Housing 2006

Table 7

Average proportion of persons in each Gini coefficient group by selected characteristics, balance of Victoria and Melbourne, 2006

Gini coefficient – natural breaks

Immigrants

Indigenous

Managers and professionals

Female LFPR

Bachelor +

Public housing

Vic – balance

Melbourne

Vic balance

Melbourne

Vic balance

Melbourne

Vic balance

Melbourne

Vic balance

Melbourne

Vic balance

Melbourne

1 (lowest inequality)

7.30

21.03

0.84

0.48

26.91

25.07

57.82

60.59

22.60

21.95

3.00

1.16

2

8.64

21.54

0.93

0.43

30.09

30.15

53.07

58.48

21.03

27.15

3.44

1.22

3

9.64

32.67

1.22

0.49

34.76

28.17

51.98

49.21

20.33

28.27

2.56

2.50

4

8.23

28.67

0.70

0.30

42.64

46.32

51.89

54.89

21.48

47.48

1.68

2.61

Nil

30.18

nil

0.31

nil

56.23

Nil

60.79

nil

54.69

nil

9.69

5 (highest inequality)

Source: ABS Census Population and Housing 2006

34

Does region matter? NATSEM JUNE 2009

5

CONCLUSION AND POLICY IMPLICATIONS

This research has applied a spatial microsimulation technique to show how this methodology can be used to derive small area estimates of Gini coefficients for New South Wales and Victoria, the two most populous states in Australia. We apply regional weights created by SpatialMSM/09C, which utilises the 2006 Census data and the 2003-04 and 200506 Confidentialised Unit Record Files (CURFs) of the Survey of Income and and Housing (SIH). We validate our results using small area validation against Gini coefficients calculated from the 2006 Census equivalised gross household income data and conduct aggregated data validation against direct estimates calculated from the 2005-06 Survey of Income and Housing using equivalised disposable household income data. The results show that although there are slight differences in terms of magnitude, the weights give reasonable results for the vast majority of small areas in these two states, with the broad regional rankings being very similar across both the Census and the synthetic estimates. The aggregation to state average shows that whilst the Australian Gini coefficient is 0.308, it is higher for New South Wales (0.322) and slightly lower for Victoria (0.306). Our research has shown that there are clear groupings of SLAs with high income inequality in Sydney, Melbourne and the balance of New South Wales. Further, while a comprehensive multivariate analysis of the determinants of inequality is beyond the scope of this paper, we conduct some further analysis of our results to examine various characteristics of small areas which previous research have found to be correlated with inequality. We focus on questions relating to possible differences in characteristics associated with inequality between New South Wales and Victoria, and between capital city and balance of state SLAs. Our findings show that, (with the exception of average proportion of Indigenous population which has a positive correlation with inequality only for New South Wales), small areas in New South Wales and Victoria which fell into the highest inequality group are characterised by, on average, high proportions of immigrants and people working as managers and professionals, high female labour force participation and post-school qualification rates, and high proportions of persons living in public housing. People working as managers and professionals and persons with a bachelor degree or higher may reflect the characteristics of the population at the top of income distribution, while the proportion of the population living in public housing and the Indigenous population represent the characteristics of people at the bottom of the income distribution. The proportion of immigrants and female labour force participation may reflect characteristics of people both at the bottom and the top of the household income distribution (assuming there are very different types of immigrants and female labour force participants). Thus as expected, the highest inequality small areas are characterised by a mixture of characteristics of people at the top and the bottom of the income distribution.

35

Does region matter? NATSEM JUNE 2009

A comparison between Sydney and Melbourne and the balance of New South Wales and balance of Victoria shows interesting findings. We find that the high inequality group in each state capital is similarly characterised by high average proportions of immigrants, people working as managers and professionals, persons having an educational attainment of bachelor degree and above, and high female labour force participation. In Melbourne, the high inequality small areas are also characterised by a relatively higher proportion of persons living in public housing although this is not the case for Sydney. The characteristics of high inequality small areas in the balance of each state have generally rather different patterns between these two states, although both have a relatively high average proportion of people working as managers and professionals and high poverty rates. It is interesting to see that high inequality areas in the balance of Victoria are also characterised by a higher average proportion of immigrants, however this is not the case for the balance of New South Wales. High inequality areas in the balance of New South Wales are also characterised by a higher average proportion of Indigenous people whilst this is not so for the balance of Victoria. This is likely to be due to the overall lower proportion of Indigenous persons in Victoria than New South Wales. High inequality small areas in Sydney and the balance of New South Wales share similar patterns of a high average proportion of people working as managers and professionals and high rates of female labour force participation. In contrast, high inequality small areas in Melbourne and the balance of Victoria share similar patterns of high average proportions of people working as managers and professionals. In answering the question posed in the title of this paper – “Does region matter?” - the findings of this research show that regional inequality does differ considerably. It is important to examine the characteristics of small areas to understand both ’between‘ and ’within‘ regional diversity, as both are important in making regional policy more effective. By knowing which small areas are more unequal in regards to income, together with the characteristics of these areas, policy makers and service providers are able to better understand the differences in the characteristics of high inequality areas in order to better target programs or policies. Future work in this area will concentrate on an econometric examination of the determinants of regional inequality in order to examine the variation of inequality across small areas in more depth, including taking into account the spatial autocorrelation that exists between small areas. Other planned work includes applying an inequality decomposition technique to separate ’between‘ and ’within‘ regional inequality and to apply the spatial microsimulation technique further in order to model policy changes in terms of examining the impact of income distribution policies on inequality at a small area level.

36

Does region matter? NATSEM JUNE 2009

6

REFERENCES

ABS (Australian Bureau of Statistics) 2006a, Income and Housing Survey (2005-06), 2nd Edition, Basic Confidentialised Unit Record Files. ABS (Australian Bureau of Statistics) 2006b, Household Expenditure Survey and Survey of Income and Housing: User Guide, 2003-04, cat no. 6503.0. ABS (Australian Bureau of Statistics) 2007a, Australian Standard Geographical Classification (ASGC), cat. no. 1216.0. ABS (Australian Bureau of Statistics) 2007b, Household Income and Income Distribution, cat no. 6523.0. ABS (Australian Bureau of Statistics) 2008, Australian Social Trends, cat no. 4102.0. Altman, D.G. 1991, Practical Statistics for Medical Research, London, Chapman&Hall. Amos, O.M. 1986, Substate and SMSA Personal Income Inequality and Regional Development, Review of Regional Studies, vol.16, pp. 23-30. Athanasopoulos, G. and Vahid, F. 2003, Statistical Inference and Changes in Income Inequality in Australia, The Economic Record, vol.79, no. 247, pp. 412-424. Bell, P. 2000, GREGWT and TABLE Macros - Users Guide, Canberra: Australian Bureau of Statistics. Bray, J.R. 2001, Social Indicators for Regional Australia, Department of Families, Housing, Communituy Services and Indigenous Affairs Policy Research Paper No 8. . Cassells, R., McNamara, J., Wicks, P. & Vidyattama, Y.,2008, ‘Children in housing disadvantage in Australia: a small area analysis’, paper presented at the European Network of Housing Researchers Conference, Dublin, July. Chin, S.F. and Harding, A. 2006, Regional Dimensions: Creating Synthethic Small-area Micro Data and Spatial Microsimulation Models, NATSEM Technical Paper 33. Chotikapanich, D., Flatau, P., Adam, M. and Lloyd, R. 2005, 'Trends in Income Inequality and Poverty in Australia in the 1990s: Regional Perspectives', paper presented at the Australian Social Policy Conference Sydney, 20-22 July. Glaeser, E.L., Resseger, M. and Tobio, K. 2008, Urban Inequality, NBER Working Paper No. 14419. Greig, A., Lewins, F. and White, K. 2003, Inequality in Australia, 1st ed, Cambridge University Press. Gregory, H., and Hunter, B. 1995, The Macro Economy and the Growth of Ghettos and Urban Poverty in Australia, Centre for Policy Research Discussion Paper 325. Harding, A. 1997, The Suffering Middle: Trends in Income Inequality in Australia, NATSEM Discussion Paper 21. Harding, A. and Greenwell, H. 2001, 'Trends in Income and Expenditure Inequality in the 1980s and 1990s', paper presented at the 30th Annual Conference of Economists, 24 September 2001. Harding, A. and Greenwell, H. 2002, 'Trends in Income and Consumption Inequality in Australia', paper presented at the The 27th General Conference of the International Association for Research in Income and Wealth, Stockholm, Sweden, August 18-24. Harding, A., Vu, Q.N., Tanton, R. and Vidyattama, Y. 2008, 'Improving Work Incentives for Mothers: the National and Geographic Impact of Liberalising the Family Tax Benefit Income Test', paper presented at the The 37th Australian Conference of Economists, Gold Coast, 30 Sept- 3 Oct 2008. Hunter, B.H. 2003, Trends in Neighbourhood Inequality of Australian, Canadian and United States of America Cities since the 1970s, Australian Economic History Review, vol.43, no. 1, pp. 22-44. Kuznet, S. 1955, Economic Growth and Income Inequality, American Economic Review, vol.45, pp. 128. Li, Y. 2005, 'Impact of Demographic and Economic Changes on Measured Income Inequality', paper presented at the Australian Social Policy Conference 2005, Sydney, Australia.

37

Does region matter? NATSEM JUNE 2009

Lloyd, R., Harding, A. and Hellwig, O. 2000, Regional Divide ? A Study of Incomes in Regional Australia, Australasian Journal of Regional Studies, vol.6, no. 3, pp. 271-292. Lloyd, R. 2007, ‘STINMOD: use of a static microsimulation model in the policy process in Australia’ in A. Harding and A. Gupta (Eds) Modelling our Future: Population Ageing, Social Security and Taxation. Elsevier: Amsterdam, 315-334. Maxwell, P. and Peter, M. 1988, Income Inequality in Small Regions: A Study of Australian Statistical Division, The Review of Regional Studies, vol.18, pp. 19-27. McGillivray, M. and Peter, M. 1991, Regional Income Inequality in A Developed Nation: A CrossSectional Study of Australian Sub-State Regions, The Review of Regional Studies, vol.21, pp. 137151. Meagher, G. and Wilson, S. 2008, Richer, but More Unequal: Perceptions of Inequality in Australia 1987-2005, Journal of Australian Political Economy, vol.61, pp. 220-243. Miranti, R. 2007, The Determinants of Regional Poverty in Indonesia 1984-2002, The Australian National University, unpublished PhD thesis in Economics. Miranti, R., Harding, A., Ngu, Q.V., McNamara, J. and Tanton, R. 2008, 'Children with Jobless Parents: National and Small Area Trends for Australia in the Past Decade', paper presented at the 37th Australian Conference of Economists Queensland. Miranti, R., McNamara, J., Tanton, R. and Harding, A. 2008, 'Poverty at the Local Level: National and Small Area Poverty Estimates by Family Type for Australia in 2006', paper presented at the 'Creating Socio-economic Data for Small Areas: Methods and Outcomes' Workshop, Canberra, 19 September 2008. Needleman, L. 1978, On the Approximation of the Gini Coefficient of Concentration, The Manchester School, vol.46, pp. 106-122. OECD 2006, OECD Factbook 2006: Economic, Environmental and Social Statistics. O'Hagan, R. 1999, Income Disparities and Population Movements in Victoria, The Australasian Journal of Regional Studies, vol.5, no. 1, pp. 87-100. Saunders, P., Hill, T. and Bradbury, B. 2008, Poverty in Australia : Sensitivity Analysis and Recent Trends, Social Policy Research Centre Report no 4. Sen, A. 1993, On Economic Inequality, Oxford : Clarendon Press. Tanton, R., Vidyattama, Y., McNamara, J., Vu, Q.N. and Harding, A. 2008, 'Old, Single and Poor: Using Microsimulation and Microdata to analyse Poverty and the Impact of Policy Change among Older Australians', paper presented at the UNU-WIDER Conference on Frontiers of Poverty Analysis, Helsinki, 26-27 September. Trendle, B. 2005, Sources of Regional Income Inequality: An Examination of Small Regions in Queensland, Review of Urban and Regional Development Studies, vol.17, no. 1, pp. 35-50. Vidyattama, Y. 2008, Patterns of Provincial Economic Growth in Indonesia, The Australian National University, unpublished PhD thesis in Economics Vu, Q.N., Harding, A., Tanton, R., Nepal, B. and Vidyattama, Y. 2008, Advance Australia Fair ?, AMP NATSEM Income and Wealth Report No 20. Williamson, J. 1965, Regional Inequality and the Process of National Development, Economic Development and Cultural Change, vol.4, pp. 3-47.

38