Article
The scaling of income
distribution in Australia:
Possible relationships
between urban
allometry, city size, and
economic inequality
Environment and Planning B:
Planning and Design
0(0) 1–20
! The Author(s) 2016
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0265813516676488
epb.sagepub.com
Somwrita Sarkar, Peter Phibbs,
Roderick Simpson and Sachin Wasnik
The University of Sydney, Australia
Abstract
Developing a scientific understanding of cities in a fast urbanizing world is essential for planning
sustainable urban systems. Recently, it was shown that income and wealth creation follow
increasing returns, scaling superlinearly with city size. We study scaling of per capita incomes
for separate census defined income categories against population size for the whole of Australia.
Across several urban area definitions, we find that lowest incomes grow just linearly or sublinearly
( ¼ 0.94 to 1.00), whereas highest incomes grow superlinearly ( ¼ 1.00 to 1.21), with total
income just superlinear ( ¼ 1.03 to 1.05). These findings show that as long as total or aggregate
income scaling is considered, the earlier finding is supported: the bigger the city, the richer the
city, although the scaling exponents for Australia are lower than those previously reported for
other countries. But, we find an emergent scaling behavior with regard to variation in income
distribution that sheds light on socio-economic inequality: the larger the population size and
densities of a city, while lower incomes grow proportionately or less than proportionately,
higher incomes grow more quickly, suggesting a disproportionate agglomeration of incomes in
the highest income categories in big cities. Because there are many more people on lower
incomes that scale sublinearly as compared to the highest that scale superlinearly, these
findings suggest an empirical observation on inequality: the larger the population, the greater
the income agglomeration in the highest income categories. The implications of these findings are
qualitatively discussed for various income categories, with respect to living costs and access to
opportunities and services that big cities provide.
Keywords
Scaling, urbanisation, sustainability, urban planning, cities
Corresponding author:
Somwrita Sarkar, Faculty of Architecture, Design, and Planning, The University of Sydney, Sydney, New South Wales 2006,
Australia.
Email: [email protected]
2
Environment and Planning B: Planning and Design 0(0)
Introduction
As the world urbanizes itself faster than ever, it is becoming more and more important to
understand the principles behind urban spatial and socio-economic structure, in the hope
that we will be able to design and plan cities better. Cities are very special examples of
complex systems that are self-organized and designed at the same time. This separates
them from large classes of other systems that are primarily either completely selforganized (e.g., biological networks) or completely designed (e.g., complex engineering
systems such as airplanes or spaceships). In cities, centralized planning and policy-making
coexist with distributed responses to plans and policies from millions of inhabitants, and
together these produce the spatial and socio-economic structure of the city (Barthelemy
et al., 2013; Batty, 2007, 2013a, 2013b). Despite decades of research, a quantitatively
based theory of cities is still missing.
Recently, it was proposed that one possible way of quantitatively characterizing urban
spatial and socio-economic structure is through universal scaling laws of the form
(Bettencourt et al., 2007a, 2007b; Bettencourt and West, 2010):
YðtÞ ¼ Y0 NðtÞ
ð1Þ
where NðtÞ is the population at time t, and YðtÞ denotes material resources (such as energy,
infrastructure or dwelling stock) or measures of social and economic activity (including both
advantages such as incomes, patents, research and development activity and disadvantages
such as crime and pollution). With Yo as a normalization constant, the scaling exponent reveals the dynamics of whether an urban indicator scales superlinearly, linearly, or
sublinearly depending on whether the value of is estimated to be below, at, or above 1,
respectively.
Studies were performed across several Metropolitan Statistical Regions (MSAs) in the
USA, and other cities in China and Europe. It was found that all material resources
(such as infrastructure, road networks) showed economies of scale and scaled
sublinearly with population size, with the value of less than 1 or almost 1 (such as
dwelling stock). On the other hand, most social and economic indicators (such as
incomes, wealth, patents, crime) showed increasing returns and scaled superlinearly with
population size, with the value of consistently more than 1. This leads to the postulated
theory behind the existence of cities: urban agglomerations exist because it is inherently
advantageous for them to exist. As population grows, the per capita expenditure on
maintaining the urban system is less than the per capita socio-economic returns by way
of income generation and wealth (although negative aspects such as crime grow
superlinearly too).
This previous analysis was carried out over aggregated measures such as total
wages, total numbers of patents, or total bank deposits. However, while cities are wealth
and knowledge creators, they also demonstrate heterogeneities, inequalities of resource
distributions, and polarizations of social and economic indicators. In particular, the issue
of income and economic inequality has been in focus, historically (Sen, 1973, 1997), as well
as recently, witnessed through several timely publications and wide ranging public and
scholarly debates (Atkinson, 2015; Piketty, 2014; Stiglitz, 2012; Wilkinson and Pickett,
2009). Many of these studies highlight the significant costs of inequality. For example
Stiglitz (2012: xii) says: ‘‘We are paying a high price for this inequality—an economic
system that is less stable and less efficient, with less growth, and a democracy that has
been put into peril’’.
Sarkar et al.
3
Does urban allometry provide a new angle on the measurement
of income distributions and implications of inequality
In this paper we focus on the question: do income distributions also scale with city or
population size? More importantly, what is the nature of this scaling for different income
groups, especially incomes towards the bottom-most and top of the income range? In this
initial examination, we focus only on income as reported in the census. In future work,
the approach presented here can be extended to more in-depth and derived measures of
income, such as computed disposable incomes, incomes after deductions of housing and
transport costs, incomes in relationship with social need or functions, or other measures of
economic well being or equity (Atkinson, 2015; Sen, 1997).
Existing metrics of income inequalities consider given income distributions and
population groups, for pre-defined spatial units (city, country, or region). In this paper,
we explore a laterally different but related aspect: we measure how income in a particular
income category scales with respect to population size, i.e., the spatial scaling of income
distributions. We show preliminary possibilities for a new type of metric based on urban
allometry, taking into consideration conditions such as the different costs of living in
different sized cities, giving rise to differential qualities of life even on the same income.
Such considerations would bring in the spatial or geographic aspects of size and scaling into
the realm of measuring inequality that current indices such as the Gini do not consider. For
example, a large city A and a small city B could have similar Gini coefficients. But this would
hide the latent scaling we uncover in this paper: that along with population, the growth rate in
the number of people in different income categories could be different, and this could signify
that income or the number of people in a particular category is growing faster as compared to
another. While we do not present a new metric of inequality in this paper, we feel that the size
of an urban system is an important variable that should be considered in future metrics, and
urban allometry could be one of the ways in which such behavior is explored.
More generally, it is worth asking what insights into the measurement of inequality can
urban allometry provide, given different possible scaling behaviors. For example, if all
income categories show linear scaling, this would signify an ‘‘equal’’ situation—income
growth in each income category is proportional to city size, city size does not matter. On
the other hand, if the lower income categories show superlinear scaling, this would imply an
inequality, since it suggests that the numbers of people in the lower income categories are
increasing disproportionately with city size. As a third case, if the highest income categories
show superlinear scaling (as observed for Australia), this suggests an inequality of a different
kind, implying that the numbers of people in the highest income categories are increasing
disproportionately with city size.
Combinations of these behaviors, however, have to be treated with caution, and their use
in measuring economic inequality with city size variation will be the subject of future
research. In the Australian example, it is seen that the number of people in the highest
income category scale superlinearly, whereas the number of people in the lowest income
category scale linearly or sublinearly. Thus, a rich person is more likely to be found in a
bigger city, whereas a poorer person is equally likely or slightly less likely to be found in a
bigger city as compared to a smaller one. This could be interpreted as a ‘‘less unequal
situation’’, since it looks like people are better off in bigger cities. However, along with
this scaling behavior, one would also need to consider the relative proportions of
individuals in each income category, as well as the scaling of costs of living. Firstly, given
a city population, the absolute numbers of people in the lower or middle income categories
are much higher than the numbers of people in the highest ones. Secondly, on a given
4
Environment and Planning B: Planning and Design 0(0)
income, the basic costs of living are much higher in a bigger city than a smaller one.
Therefore, if proportions of individuals in each category are considered, and real incomes
are computed based on a combination of the numbers of people in a category and their
actual incomes and expenditures, then the agglomeration of a high proportion of income in
the highest categories becomes more evident. Finally, the internal structures of even similar
population cities in terms of housing and transportation, their social and economic
structures, taxation structures, etc. show large variation. Therefore, the scaling of income
distributions needs to be explored in different parts of the world.
The question of city size
A second reason exists for looking at not only income distributions but also all the other
socio-economic factors that, by way of superlinear scaling of negative socio-economic
effects, may work towards limiting the size of urban systems. In the previously proposed
model for urban growth (Bettencourt et al., 2007a), growth is shown to be governed by the
following differential equation:
dNðtÞ Y0
R
¼
NðtÞ NðtÞ
dt
E
E
ð2Þ
where resources Y from equation (1) are used for both maintenance and growth of the
population N. It is assumed that, on average, a quantity of resources R per unit time is
used to maintain the population, and a quantity of resources E per unit time is used to add a
new individual to the population. Then, with dN=dt providing the rate of change of
population, the model is of the form Y ¼ RN þ EdN=dt, that when rearranged gives
equation (2). The solution to this differential equation is analytically calculated as:
1
Y0
Y0 ½Rð1Þt 1
ð1Þ
NðtÞ ¼
þ ðN
ð3Þ
ð0Þ Þe E
R
R
The solution to equation (2) shows three different forms for 5 1, ¼ 1 and 4 1: When
5 1, population growth reaches a finite carrying capacity and then ceases. When ¼ 1,
the growth is exponential. When 4 1, the growth diverges at a critical time point, such that
there is unbounded growth in a finite amount of time. However, if resources are scarce,
unbounded growth cannot happen, thus the singularity condition is never reached in
practice. Thus, Bettencourt et al. propose that ever-faster and ever-higher rates of
innovation and wealth creation would be needed to sustain such growth, and would
produce cycles of innovation and growth.
Reasoning qualitatively about the model above, however, we see the implicit assumption
of bigger is better: that is, circularly, because higher and faster rates of wealth creation and
innovation would lead to ever increasing city sizes (population), and therefore, to sustain
such growth (which is assumed to be a desirable situation a priori), we need higher and faster
rates of wealth creation and innovation.
In this paper, we argue that the desirability of the unbounded growth assumption needs
deeper scrutiny. Of significant interest is the normative question: even if we had a system that,
in principle, had the capacity to grow through continuous cycles of innovation and wealth
creation, would such a system be desirable or stable, if in such a system along with the
positives of wealth creation and innovation, the negatives of increasing wealth inequality,
socio-economic polarization, increasing crime rates, or housing opportunity gaps also grow
by the very same natural implicit processes that drive the positives? And if such growth is
Sarkar et al.
5
achieved, in principle, what could be done to ensure social and economic wellbeing (as
measured through decreasing inequality of opportunities and outcomes for example)?
Contributions
In this paper, we provide empirical evidence for the differential scaling of income
distributions with increases in the sizes of urban systems. In simple words, we examine
whether the income category a person is in and the size of the city in which they live are
related. We examine the scaling of per capita income against population size for different
categories of income earners, across several urban area definitions based on social and
economic geography for the country of Australia. Instead of focusing on aggregate
measurements of income and wealth in cities, as previous studies have done, we focus on
specific income categories: how much income is earned in a specific income category, or
equivalently, what is the distribution of people in specific income categories, and how does
this measure scale with city or population size?
There is now considerable debate over the definition of what a city is, and that the way in
which city definitions are made could change the scaling behavior observed (Arcaute et al.,
2015; Louf and Barthelemy, 2014). Administrative boundaries often do not coincide with
economic or social activity, population density, or journey to work patterns, and this has
been shown to affect the scaling dynamics (Arcaute et al., 2015). Therefore, in order to test
the universality of our results, we perform the analysis over several urban area definitions,
based on statistical and economic labor force region definitions provided by the national
statistical agency, the Australian Bureau of Statistics (ABS), and using these to generate
both population count and population density based urban area definitions. The ABS is a
unique body in the sense that the geographic area definitions it provides are, to a large
extent, defined on the basis of labor markets and population counts and densities in large
regions. For more discussion of the finer points of area definitions used for the analysis, see
‘‘Methods’’ section.
We choose the country of Australia for several reasons. First, while studies of US,
Europe, and separately the UK have been reported (Arcaute et al., 2015; Bettencourt
et al., 2007a), Australia’s unique geographic position as an island-country-continent
merits study and has not been analyzed before (though New Zealand is not included in
our analysis, due to separate census bodies in the two countries). Second, even though
Australia is one of the most urbanized countries in the world, it is also one of the most
sparsely populated and one of the youngest: the urban structure is still nascent. The
population in Australia is not spread out as USA, Europe, or UK. The urban landscape
is dominated by the Sydney and Melbourne metropolitan regions, with 3–4 smaller regions
(Perth, Adelaide, Brisbane), followed by a few other smaller urban areas, and finally a vast,
mostly uninhabited continent in the middle. This results in a situation where almost the
entire population (more than 90%) is urban and lives agglomerated in a few large urban
areas, with the rest of the continent very sparsely inhabited. Third, because of this unique
position, a single national body, the ABS maintains data across several spatial scales, where
in a large number of ways the administrative boundaries concur reasonably with real
economic and social boundaries, population densities, and other bottom up indicators
that have recently been shown as important factors affecting the scaling behavior
(Arcaute et al., 2015). This allowed us to test the variation and sensitivity of the scaling
behavior across several stable urban area definitions. These special factors render the
Australian case study important: how does scaling behavior change with a significantly
different geographic prototype such as Australia?
6
Environment and Planning B: Planning and Design 0(0)
Methods
There is considerable debate in the literature on the geography that is adopted for the
analysis of scaling behavior. As recent research shows, changing the geographical
definitions can produce significant quantitative changes to the scaling exponent, and
implied qualitative changes of interpreted urban dynamics (Arcaute et al., 2015; Louf and
Barthelemy, 2014). For example, a range of indicators have been reported for which the
scaling exponent fluctuates between the superlinear and sublinear regimes, depending on the
way in which the geography of analysis is defined, and specifically for comparisons of CO2
emissions with city size, some studies have shown sublinear regime relationships, while
others have shown superlinear scaling relationships (Fragikas et al., 2013; Louf and
Barthelemy, 2014; Oliveira et al., 2014; Rybski et al., 2014). Thus, any claim of scaling
must rest on the behavior being tested over multiple possible realizations of reasonably
defined geographies. Both the ways in which data are measured and collected, as well as
underlying geography definitions and factors other than city size have been proposed as
reasons behind deviations, and therefore these need careful examination.
Australian Statistical Geography Standard
For the Australian case, the country is divided into eight States and Territories. The data for
the entire country are collected and organized by a single national statistical body, the ABS.
The ABS defines the Australian Standard Geographic Classification (ASGC) that in July
2011 changed over to the new system of Australian Statistical Geography Standard (ASGS),
that is a set of hierarchically organized levels of geographic units that correspond to spatial
scales into which the entire country is progressively divided into, without overlaps and gaps
(ABS, 2011a; Pink, 2011). The areas are defined with regard to several important factors
such as population cut-offs, social and economic activity, and labor force and housing
markets and sub markets, along with ensuring the best possible consistency with
administrative boundaries. However, the important fact to be noted is that the statistical
bases for defining these geographies is not purely administrative, but several related factors
as have been noted to be important in previous research (Arcaute et al., 2015). Australia is
highly urbanized, with its urban population concentrated in very few large urban centers,
making the derivation of the urban geography consistent in terms of many factors. Here we
describe in brief the organization of the ASGC and the main spatial scales, and the details of
how we have decided on a stable geography for our analysis.
The ABS classifies geographic structures into two classes: statistical ABS structures, and
non-ABS political and administrative structures into which fall state suburbs (SSC), postal
areas, and Local Government Areas (LGA). Here we consider the statistical ABS structures,
since these are derived on the basis of social and economic interactions primarily
concentrated within areas, rather than purely administrative divisions.
The smallest geographic ABS structures are the Statistical Areas Level 1 (SA1), with a
minimum population of 200, and a maximum population of 800, with an average of
approximately 400 people. From an administrative perspective, the SA1s closely reflect
(though are not identical with) the non-ABS state suburbs and postal areas. The criteria
for urban SA1s and rural SA1s are separately defined, with the urban SA1s characterized by
presence of different types of developments such as airports, ports, large sports and
educational campuses, roads, and large infrastructure. There are about 55,000 SA1s for
the whole of Australia. From continuous aggregates of SA1s, Urban Centers and
Localities (UCLs) are defined. Centers with core urban populations of more than 1,000
Sarkar et al.
7
are designated as Urban Centers, and centers with a core urban population of 200 to 1,000
designated as Localities. Populations contained within Urban Centers are used to describe
Australia’s urban population at the lowest geographic level. There are about 684 Urban
Centers in Australia.
We do not use Urban Centers to define the geography of the city unit for our analysis,
because several semi-urban and peripheral areas that are complete SA1s are predominantly
surrounded by rural territory, making these the smallest level complete urban areas. On the
other hand, many SA1s (for example, contiguous SA1s or UCLs within the Sydney,
Melbourne and other capital cities regions) cannot be considered to be separate entities,
since they are also parts of much larger urban agglomerations. Further, income data for
statistical areas are only available for the larger definitions of SA2, SA3, and SA4 (see below)
and are not available for SA1s.
The second level of geographic units is the Statistical Areas Level 2 (SA2), with a
minimum, maximum, and average population of 3,000, 25,000, and 10,000, respectively.
There are about 2,300 SA2s defined for the whole of Australia. Continuous SA1s are
aggregated to produce each SA2, and the SA2 is designed to reflect functional areas, with
the aim of representing a community that interacts together socially and economically. They
also coincide largely with the non-ABS structure of LGA.
From continuous aggregates of SA2s, Significant Urban Areas (SUAs) are defined, that
describe extended urban concentrations of more than 10,000 people. That is, these SUAs do
not represent a single Urban Center, but they can represent a cluster of related Urban
Centers, where the population interacts socially, lives, and travels for work. There are 101
SUAs describing urban concentrations in Australia. For reasons discussed below, we adopt
SUAs as the primary unit of defining geography in our work.
The third level of geographic units is the Statistical Areas Level 3 (SA3), with a
minimum, maximum, and average population of approximately 30,000 to 130,000. They
are designed as aggregates of whole SA2s, and reflect major regions within States and
capture regional level outputs. No SUAs are defined for SA3s, and SA3s may include
regional and rural areas.
The fourth and largest level of geographic units is the Statistical Areas Level 4 (SA4), with
a minimum, maximum, and average population of approximately 100,000 to 500,000. In
regional areas, SA4s contain populations of 100,000 to 300,000 whereas in metropolitan
areas, SA4s have larger populations ranging from 300,000 to 500,000. They are designed as
aggregates of whole SA3s. There are 106 SA4s in Australia. Whole SA4s aggregate to
Greater Capital City Statistical Areas (GCCSA) and State and Territory, with the
GCCSAs focused on major urban cities and extending to the urban periphery of these
large cities.
SA4s are not adopted as the unit of defining geography in this paper, because while
separate SA4s inside the GCCSA cannot be considered to be separate cities (for example,
it would be wrong, artificial and arbitrary to consider the different SA4s inside the Sydney
region as separate cities), while those that are outside the GCCSAs, can be considered to be
smaller but complete urban areas or cities.
For this work, we therefore work with the SUAs that include all the Urban Centers within
a region that interact closely socially and economically (including living and working within
the same SUA). The reasons for choosing SUAs are as follows: all major capital cities are
defined as single urban clusters (for example, Sydney, Melbourne, Brisbane, Adelaide and
Perth form the largest ones). In addition, smaller urban areas surrounded by predominantly
regional or rural areas, but that clearly house urban functions, are also defined as separate
cities.
8
Environment and Planning B: Planning and Design 0(0)
SUA definitions therefore do not suffer from the limitation as outlined in SA4 and SA1
definitions: (a) we are assured that no contiguous urban areas are arbitrarily defined as two
cities just on account of an administrative or census division, and (b) we are assured that
identifiable urban areas of all sizes are identified nonetheless, even if they are surrounded by
rural or regional hinterland, but are independently functioning urban regions. The statistical
reasons provided by the ABS for defining the SUAs are also close to the above arguments.
The ABS notes that ‘‘the regions of the SUA structure are constructed from whole SA2s.
They are clusters of one or more contiguous SA2s containing one or more related Urban
Centers joined using the following criteria:
. they are in the same labor market;
. they contain related Urban Centers where the edges of the Urban Centers are less than
5 km apart defined by road distance;
. they have an aggregate urban population exceeding 10,000 persons;
. at least one of the related Urban Centers has an urban population of 7,000 persons or
more.
Further, they also note in their naming conventions that when an SUA represents a single
dominant urban system according to above criteria, then it has a single name (e.g., Sydney).
On the other hand, if an SUA represents two close and related centers of considerable
importance, then they name it as a combination (e.g., Kalgoorlie-Boulder). Third, when
an SUA crosses a state or territory border, it is named after the largest centers on both
sides (e.g., Canberra-Queanbeyan).
Computation of incomes in separate income categories
For all the geographic area definitions discussed above, per capita weekly income statistics
are available. The ABS has 10 per capita income categories, defined on the basis of income
deciles, shown in Table 1 (ABS, 2011b). Categories (range identifiers) 01 and 02 represent
negative and nil incomes, respectively, and the other income categories represent positive
incomes. Thus, categories 01 and 02 will not be included in the analysis.
Using the 2011 Census Income Data retrieved using the TableBuilder facility (ABS,
2011c), counts of the number of people in a particular income category on census might
have been retrieved as per the SUA area definitions. That is, for each SUA, we have the
number of persons in each of the 10 (positive) income categories of Table 1. We have
computed the median earnings for each income category by multiplying the imputed
median income for each category for each SUA by the number of people in that income
category. We call this the computed median income per income category. Adding across all
income categories gives us the total annual computed median income per SUA for the census
year 2011.
As noted by the ABS (2011b), while it is risky to use the imputed medians for computing
small area incomes due to socio-economic variations and misreporting issues, they
nonetheless provide an appropriate and satisfactory representation for aggregated / large
areas, such as state/territory level, metropolitan / ex-metropolitan regions. We use SUAs as
they represent complete aggregated urban areas of Australia, and are large enough to fulfill
the definition of ‘‘aggregated / large areas’’ that will not suffer from the said problems. SUAs
represent entire aggregated metropolitan or urban entities at the larger geographic scale, and
hence the use of the imputed medians, unjustified at smaller SA1 or SA2 levels, may be
justified for the SUA level.
Sarkar et al.
9
Table 1. Personal income ranges, imputed medians (ABS), and computed means.
Weekly
ABS range personal
identifier income
Per annum
personal income
Imputed Computed Gap between median and mean
(absolute and % of the median
median mean
by which the mean is higher)
(weekly) (weekly)
01
02
03
04
05
06
07
08
09
10
11
12
Negative income
Nil income
$1–$10,399
$10,400–$15,599
$15,600–$20,799
$20,800–$31,199
$31,200–$41,599
$41,600–$51,999
$52,000–$64,999
$65,000–$77,999
$78,000–$103,999
$104,000 or more
$101
$0
$80
$263
$349
$487
$698
$896
$1,107
$1,363
$1,695
$2,579
Negative income
Nil income
$1–$199
$200–$299
$300–$399
$400–$599
$600–$799
$800–$999
$1,000–$1,249
$1,250–$1,499
$1,500–$1,999
$2,000 or more
$100
$250
$350
$500
$700
$900
$1,125
$1,375
$1,750
$20 (þ25%)
$13 (þ4.94%)
$1 (þ0.28%)
$13 (þ2.67%)
$2 (þ0.29%)
$4 (þ0.45%)
$18 (þ1.62%)
$12 (þ0.88%)
$55 (þ3.24%)
Notwithstanding the above observations, we also performed two extra validation steps to
check whether the computed median income could be considered satisfactory. First, the issue
of means versus medians is addressed. If the means of the income brackets are very far away
from the imputed medians, then the resulting analysis could be very different. Table 1 shows
the computed means for each bracket. Note that the mean for the highest bracket could not
be computed, since it is an open bracket. As seen, most of the brackets show little deviation
of the mean from the median, both in absolute and percentage differences of the mean from
the median. The largest deviation is in fact in the lowest reported income category.
Further, since the computed means are the mean of the range rather than the data, it
could be argued that the data mean could be different from the range mean. However,
quoting from the ABS fact sheet (ABS, 2011b): ‘‘The [income] ranges presented on the
2011 Census form were selected after analyzing data from the 2007–08 Survey of Income
and Housing (SIH), in which personal income was collected in actual dollar amounts.’’ Thus,
it is statistically unlikely that the 2011 income ranges resulting from the analysis of the 2007–
08 data would result in skewed data distributions. Re-doing the analysis with the mean
values of Table 1 instead of the imputed did not change the nature of our results, since
the mean and median for most categories are close.
Secondly, we also computed the actual total reported income in each SUA by aggregating
the small area incomes reported at the SA2 levels for the year 2011 (ABS Estimates of
Personal Income for Small Areas, ABS Catalogue 6524.0.55.002). That is, for each SUA,
we aggregated the SA2 level actual total personal incomes reported for the financial year
2010–2011. This information is collected by the ABS from the Australian Tax Office (ATO),
and is not reported in the income category distribution form, but in terms of actual total
incomes (including wages, superannuation, income transfers, and other tax categories). It
would be problematic if there were large deviations of the computed total income estimates
(computed using the imputed medians) from the actual reported total incomes. However, the
computed median total incomes were close to the reported total incomes, with the reported
incomes slightly higher in most cases (Figure 1). As is seen, the gaps between the two are not
so significant or large such that they would have the capacity to alter the basic results
reported in this paper. On performing the scaling analysis on both the computed total
Environment and Planning B: Planning and Design 0(0)
Millions
10
160000
Actual reported income
140000
Computed income
120000
100000
80000
60000
40000
Toowoomba
Darwin
Cairns
Townsville
Geelong
Hobart
Wollongong
Sunshine Coast
Central Coast
Canberra - Queanbeyan
Newcastle - Maitland
Gold Coast - Tweed Heads
Adelaide
Perth
Brisbane
Melbourne
0
Sydney
20000
Figure 1. Comparison of the computed total incomes (using ABS imputed medians) and the actual
reported total incomes.
income and the reported total income categories, we found the exponent as 1.03 for the
computed, and 1.02 for the actual, showing a very small difference.
Analysis of scaling behavior
For each SUA and for each income category, Matlab’s linear model fitting is used to
calculate the exponent, R-squared values, and all other related statistical parameters.
The log of total population is plotted against the log of the computed income in that
income category.
Results
The ABS carries out a detailed census once every five years. For this paper, we have worked
with the latest census data from 2011.
Geography: SUAs, population and density cutoffs,
and income categories
We have used SUAs to define the urban area geography of Australia (see ‘‘Methods’’
section). The five largest SUAs (continuous urban areas or ‘‘cities’’) in Australia are
Sydney, Melbourne, Perth, Adelaide, and Brisbane, with about 60% of the population of
the entire country residing in a capital city, and about 35% in Sydney and Melbourne alone.
Sarkar et al.
11
1.4
1.3
x 10
1.5
1
0.5
0
1.2
1.1
3
0.8
4 5
1.01
0.97
0
12
3 4 5 6 7 8 9 10 11 12
ABS Income Categories
6
7
1
0.9
0.96
2
0.98
8
1.01 1.05
9
10 11
Total
1.06 1.07 1.07 1.11
4
6
8
ABS Income Categories
1.03
10
(d)
β = 0.96, R 2 = 0.92
24
23
Melbourne
22
Adelaide
21
Sydney
Brisbane
Perth
20
19
18
17
16
8
10
12
14
log(Population)
16
β = 1.01, R 2= 0.98
22
Melbourne
Sydney
Brisbane
Adelaide Perth
21
20
19
18
17
16
15
14
12
(c)
log(Income bracket 5: $15,600 to $20,799)
log(Income bracket 3: $1 to $10,399)
Population Count
1.5
Scaling exponent β
(b)
6
2
8
log(Income bracket 12: more than $104,000)
(a) 1.6
10
12
14
log(Population)
16
β = 1.11, R 2= 0.84
25
Sydney
Perth
Melbourne
Brisbane
24
23
Adelaide
22
21
20
19
18
17
8
10
12
14
log(Population)
16
Figure 2. Scaling of income in all 101 SUAs. (a) Scaling exponents with 95% CI in 10 ABS income
categories. (b–d) Scaling behavior for income categories 3, 5, and 12 (roughly categorized as lowest, middle
lower, and highest, respectively), showing linear to sublinear behaviors in the lower income categories, with
superlinear behavior emerging for higher income categories.
These centers have populations greater than one million. We have performed the scaling
analysis firstly for all 101 SUAs, followed by considering alternate urban area definitions.
Several smaller subsets of the SUAs are considered by including high population density and
high population counts as cutoffs, and excluding very low population SUAs and very low
population density SUAs (especially those that are of an overwhelmingly regional character
surrounded by large rural hinterlands), and measured how the scaling exponent behavior
changes under urban area definitions that look at increasing population density and total
counts.
Scaling of income in different income categories: All SUAs
Figure 2(a) shows the scaling exponents for all the separate income categories and the total
income for all the 101 SUAs. Figure 2(b–d) shows the scaling behavior for ABS income
categories 3, 5, and 12 (roughly categorized as lowest, middle lower, and highest,
respectively). Total income scales just superlinearly, with ¼ 1.03. This is different from
the exponents previously reported, for the USA, European countries and China, where a
clear superlinear behavior was reported, with higher values (Bettencourt et al., 2007a).
12
Environment and Planning B: Planning and Design 0(0)
Table 2. Scaling exponents for all 101 SUAs.
ABS range identifier
Per annum personal income
95% CI
R2 values
03
04
05
06
07
08
09
10
11
12
Total
$1–$10,399
$10,400–$15,599
$15,600–$20,799
$20,800–$31,199
$31,200–$41,599
$41,600–$51,999
$52,000–$64,999
$65,000–$77,999
$78,000–$103,999
$104,000 or more
–
1.01
0.97
0.96
0.98
1.01
1.05
1.06
1.07
1.07
1.11
1.03
[0.98,
[0.91,
[0.91,
[0.94,
[0.98,
[1.02,
[1.03,
[1.04,
[1.02,
[1.02,
[1.00,
0.98
0.91
0.92
0.95
0.98
0.99
0.98
0.97
0.94
0.84
0.97
1.04]
1.03]
1.02]
1.03]
1.04]
1.07]
1.09]
1.11]
1.13]
1.21]
1.06]
However, in more recent work for the UK, a similar close to linear scaling of total income
has been reported (Arcaute et al., 2015). Table 2 summarizes the findings.
It is also observed that the lowest five income categories show sublinear or just linear
exponents, and the top five higher income categories show superlinear exponents. Here we
consider an exponent as sublinear, linear, or superlinear when the 95% Confidence Interval
(CI) error bar is safely below, at, around, or above the unity line, respectively. Thus, it is seen
that the first five income categories show linear to sublinear behavior, whereas the highest
five income categories show superlinear behavior, monotonically increasing especially for the
top five income categories.
Particularly interesting is the behavior of the highest income category, which is higher
than all the others by the largest successive difference between exponents (Figure 2(d)). A
slightly larger spread or variance of distribution of the data points can also be observed for
the highest income category, for some of the middle- and low-sized SUAs (R2 ¼ 0.84). In this
respect it is interesting to note that (disregarding for a moment the position of the largest
cities), Figure 2(d) shows some urban areas with lower populations can have significantly
high incomes, but only in the highest income category. A similar feature has been observed
also for total numbers of patents filed in the UK, where cities such as Oxford and Cambridge
with lower population counts still show high numbers of patents and innovation indicators,
suggesting that population size and wealth indicators may not necessarily be correlated
(Arcaute et al., 2015). It suggests that even smaller urban centers, when they specialize in
certain specific economic or knowledge-hub type activities, can show high incomes and
innovation indicators. In Australia, for example, several areas are specifically mining
areas or wine production areas that could show this behavior of higher incomes for lower
populations in the scaling plot. However, we observe this behavior only in the highest
income category; lower income categories do not show this behavior in our analysis for
Australia. Moreover, when we reconsider the largest cities in the plot, the scaling exponent is
superlinear for the highest income bracket.
Overall, this behavior implies that as SUAs grow bigger in population count, the incomes
in the highest income categories grow (disproportionately) faster than incomes in the lower
income categories. Incomes in the lower income brackets also grow, but linearly or
sublinearly. Equivalently, the numbers of people in the highest income brackets are
disproportionately higher in larger cities, but the numbers of people in lower income
brackets grow more or less proportionately with city size, or even sublinearly. In other
Sarkar et al.
13
words, a lower income person is equally likely to be found in a small or a big city, or slightly
less likely to be found in a bigger city than a smaller city. However, since the costs of living,
such as housing or transport, could be lower in smaller cities for the same income, this has
implications for wellbeing and quality of life indicators: it may suffice to earn lesser for a
good quality of life in a smaller city, while for a similar quality of life in a larger city, higher
incomes would be needed, or the same income would provide a lower quality of life in a
larger city. But, a higher income person is more likely to be found in a bigger city. For them
too, the cost of living would be higher, but so would the accrued relative advantages of being
able to afford a higher quality of life and access to opportunities of better education,
healthcare, or employment offered by a bigger city. As the inset of actual population
count distribution for the 101 SUAs in Figure 2(a) shows, the lower income categories
contain the largest sections of the population (this also holds for individual SUA
population distributions), showing that for a particular city, the largest sections of the
population are in the lower income brackets commanding a much lower proportion of the
total income, and the smallest sections of the population in the highest income brackets
command the largest portions of the total income. Given that total income scales just
superlinearly, this points to the emergence of inequality through the scaling behavior,
since it shows that the larger the city, disproportionately the more income it generates,
but a disproportionately larger amount of this income is in the highest income brackets.
To confirm this analysis, however, a more in-depth analysis of the urban area definitions
was needed. In recent work it has been shown that the manner in which urban areas are
defined can affect the values of the scaling exponents (Arcaute et al., 2015). In addition, in
Australia, not all the SUAs are equivalent, since the largest SUAs have populations
exceeding 1–4 million and densities close to 1,000 persons per square km, whereas the
smallest SUAs (some of them surrounded by regional or rural hinterlands), have
populations exceeding just 10,000, and densities close to 21 persons per square km.
Clearly, not all urban areas are equally ‘‘urban’’. Since the ABS SUA definitions are
based on the type of economic and social activities in an SUA, labor force and journey to
work region definitions (see ‘‘Methods’’ section), and because of the special manner in which
population is heterogeneously distributed across Australia, a deeper look into considering
alternate urban area definitions was warranted. In particular, we wanted to study how the
behavior of the scaling exponent changes if more ‘‘urban’’ areas are considered, and less
‘‘urban’’ areas are excluded from the analysis.
Scaling of income in different income categories: Population counts
and density cut-offs
In recent work, population density instead of population count has been proposed as one of
the factors that can be used to define urban areas (Arcaute et al., 2015). Figure 3(a) shows the
log–log plot of population count versus population densities (total population divided by total
area in sq km reported by the ABS per SUA). A general positive correlation is observed with
higher total population count areas showing higher population densities. However, as seen in
Figure 3(b), a plot of total population counts against population densities also shows some
outliers. For example, the Canberra-Queanbeyan region or Central Coast regions (clearly
urban by all measures of socio-economic activity) have lower total populations than many
larger total population count regions, but have higher density of population. Thus, we
empirically identified a cut-off point using a population density of 152 persons per sq km,
below which all SUAs had low total population counts as well as low population density. This
resulted in 66 SUAs that have either high total population count, or high population density,
14
Environment and Planning B: Planning and Design 0(0)
(a)
Sydney
(b)
8
Canberra-Queanbeyan
4
8
10
Perth
Brisbane
400
0
16
6
2
Population Count
1.5
Scaling exponent β
12
14
log(Population Count)
(d)
(c) 1.6
1.4
1.3
x 10
1.5
1
0.5
0
12
3 4 5 6 7 8 9 10 11 12
ABS Income Categories
1.2
1.1
3
1
4
5
1.00
0.9
0.8
600 Central coast
200
0.94
0
2
6 7
6
0.96
8
9
10 11
4
6
8
ABS Income Categories
Total
1.16
1.05
10
12
(f)
log(Income bracket 3)
3
Adelaide
22
Bathhurst
0
1
2
β = 1.00, R = 0.99
Density cutoff = 152 p/sqkm.
2
3
Population Count
(e)
20
18
16
14
8
10 12 14 16
log(Population)
β = 0.96, R2= 0.98
24
22
20
18
16
8
10 12 14 16
log(Population)
(g)
log(Income bracket 5)
5
Melbourne
800
log(Income bracket 12
Population Density
6
log(Income bracket 6)
log(Population Density)
1000
7
24
4
6
x 10
2
β= 0.94, R = 0.95
22
20
18
16
8
10 12 14 16
log(Population)
2
β = 1.16, R = 0.91
24
22
20
18
16
8
10 12 14 16
log(Population)
Figure 3. Scaling of income in high population density SUAs. (a) Log–log plot of population count versus
population density for all 101 SUAs shows the general positive correlation of higher population with higher
density. Sizes of circles scaled corresponding to total population counts. (b) Plot of population count against
population density shows some outliers. E.g. High population, lower densities, or low populations with
higher densities. The cut-off point for excluding all SUAs where both population counts and population
densities are low, which occurs at around 152 persons per sq km (Bathhurst). (c) Scaling exponents with 95%
CI in 10 ABS income categories for the top 66 highest density and population count SUAs. (d–g) Scaling
behavior for income categories 3, 5, 6, and 12 (roughly categorized as lowest, middle lower, and highest,
respectively), showing linear to sublinear behaviors in the lower income categories, with superlinear
behavior emerging for higher income categories for the 66 high population and high density SUAs.
or both. We note again that as compared to most other countries of the world, these numbers
on population density are particularly low. For example, the population density cut-off for the
UK for the purpose of defining urban areas was 1,400 persons per sq km, but even the highest
density SUA in Australia, Sydney, has a density of about 991 persons per sq km (Arcaute
et al., 2015). Therefore, our identification of urban areas based on population density has been
based on the relative comparative positioning of the SUAs with each other. We note here that
considering only the metropolitan areas will show higher population densities, for example
Sydney and Melbourne metropolitans sit at about 1,900 and 1,500 persons per sq km,
respectively, but here we consider the entire SUA (Sydney and its cluster of related urban
centers), which is more accurate considering journey to work patterns, since people from
urban areas outside the metropolitan region still travel daily to the city proper for work.
Figure 3(c–g) shows the behavior of the scaling exponents for these 66 SUAs. While the
earlier reported pattern in Figure 2 holds, the differences between the income categories
Sarkar et al.
15
(b) 1.6
Cut-off=30,000
1.5
1.5
1.4
1.4
1.3
12
1.2
1.1
3
4
1
5 6
7
8
9 10
11
Total
2
12
1.2
1.1
3
4
6
8
ABS Income Categories
10
0.8
12
4
1
5 6
7
8
9
10
11
Total
1.18
1.05
1.00
0.9
0.96
0
Cut-off=40,000
1.3
1.13 1.04
1.00
0.9
0.8
Scaling exponent β
Scaling exponent β
(a) 1.6
0.95
0
(c) 1.6
2
4
6
8
ABS Income Categories
10
12
Cut-off=80,000
Scaling exponent β
1.5
1.4
12
1.3
1.2
1.1
3
4
1
6
7
8
9
11
1.21
Total
1.04
1.01
0.9
0.8
5
10
0.95
0
2
4
6
8
ABS Income Categories
10
12
Figure 4. Scaling of income in SUAs by population count cut-offs. Scaling exponents with 95% CI in 10 ABS
income categories for SUAs over (a) 30,000 total population, (b) 40,000 population, and (c) 80,000
population, showing linear to sublinear behaviors in the lower income categories, with superlinear behavior
emerging for higher income categories.
becomes more pronounced, with the highest income category exponent going up to ¼ 1.16,
and the lowest one at ¼ 0.94, with total income ¼ 1.05. Thus, considering
higher population density and population areas show the scaling effect becoming more
pronounced.
Further, since the correlation between total population count and population density is
strong, we also apply population count cut-offs, to consider urban areas larger than a total
population count threshold. For example, there is only one SUA with a population of more
than 80,000 that has a population density of 130 persons per sq km (that is below the density
cut-off of 152 persons per sq km). Similarly, only four SUAs with a population of more than
40,000 and eight SUAs with a population of more than 30,000 have a population density of
lower than 152 persons per sq km. Figure 4(a–c) shows the behavior of the scaling exponent
for 45, 33, and 21 SUAs above the total population counts of 30,000, 40,000, and 80,000,
respectively. Once again, the same trend is observed: as SUAs with larger populations and
population densities are considered in urban area definitions, the scaling behavior shows the
exponent for the highest income bracket rising more sharply, with the inequality effect
becoming more pronounced. For example, when the SUAs beyond a total population
count of 80,000 are considered, the exponent for the highest income category goes up to
16
Environment and Planning B: Planning and Design 0(0)
¼ 1.21, whereas the lowest income category and total income holds at ¼ 0.95 and
¼1.04, respectively. Further, it is interesting to note that some of the error bars for the
other high income brackets actually graze or go below the unity line, showing the
substantially different behavior of the top income bracket. This could be a possible effect
from the low number of data points, a unique feature of the Australian dataset, because of
the very low number of urban areas housing a mostly completely urbanized populace.
However, the general trend of the monotonically rising exponent for higher income
brackets is maintained.
Thus, across all urban area definitions based on total population counts and population
density, we find that lowest incomes grow just linearly or sublinearly ( ¼ 0.94 to 1.01),
whereas highest incomes grow superlinearly ( ¼ 1.00 to 1.21), with total income just
superlinear ( ¼ 1.03 to 1.05). These findings support the earlier finding: the bigger the
city, the richer the city. But, we also see an emergent scaling behavior with regard to
variation in income distribution that sheds light on socio-economic inequality: as city size
grows larger, higher incomes grow more quickly than lower, suggesting a disproportionate
agglomeration of wealth in the highest income categories in big cities.
Separately, we compute the Gini coefficient for the all the SUAs with more than a
population of 30,000, using the standard definition of the Gini coefficient as the Lorenz
curve, computing the proportion of the total income of the population that is cumulatively
earned by the bottom x% of the population. A Gini coefficient of 0.47 results by considering
all the income categories (including the negative and nil income categories, for which we
have the population counts, but no earnings). We have considered a zero earning for both
the negative and nil earning groups, even though the imputed median for the negative
earning category by the ABS is $101. Since we could not include these two categories in
our scaling analysis, a Gini coefficient of 0.42 results by considering only the income
categories reporting positive income, the ones we have used for our scaling analysis. As
compared to the Australian average of 0.32 as reported by the ABS, and as compared to
developing country averages, this is a high Gini coefficient. One reason why the official Gini
coefficient of 0.32 is lower than 0.42 or 0.47 is that in our computations total income
reported in the census is considered, whereas the Australian Gini of 0.320 is based on real
income computations, and emerges lower than what our computations show. Nonetheless,
equivalently, the Gini coefficient does signify high levels of income inequality in urban areas
of the country.
However, the results reported in this paper present an empirical observation on inequality
from a completely different angle, separate from the Gini. The Gini coefficient looks at intrapopulation cumulative distributions of income, and these could be computed for a country,
or parts of a country (such as states, territories or metropolitan areas). But, it is computed to
measure income inequality for a specific geographic region. On the other hand, our findings
relate how incomes in specific income brackets scale with the sizes of populations, and
therefore, the sizes of cities or urban systems. This is different, because it allows the
observation of latent scaling behavior of income over different geographic regions, as a
measure of their population sizes.
In future research it will be very interesting to relate the theoretical basis of inequality
measurement, and existing measures of inequality (such as the Gini, Atkinson, or Theil
indices for example), to the observation on how income distributions behave as a
dependent variable with the city size as the independent variable. As of now, we are not
aware of any research that makes this connection, despite several empirical observations on
larger cities being more economically unequal when compared against the region or national
averages.
Sarkar et al.
17
Discussion
In this paper, we have examined the scaling of income in different income categories
measured against population size as a measure of city size for the country of Australia.
Australia represents a nascent urban system, still young and in early stages of growth, with a
highly urbanized populace, but concentrated in very few urban areas, presenting a unique
opportunity for this type of study. Using income and population data from the census of
Australia, under several urban area definitions based on total population counts and
population densities, it was found that while incomes in the lower income groups scale
just linearly or sublinearly, the incomes in progressively higher income groups scale
superlinearly. Moreover, the scaling of the income in the highest income category is
significantly superlinear and high, as compared to all the other income categories. In
general, there is a monotonically increasing trend for the scaling exponent for the high
income brackets. Australia is not considered to be one of the most unequal nations on the
planet. Recent studies have found that developing nations such as India and China, and
developed nations such as the USA, are much more unequal than Australia (Atkinson, 2015;
Piketty, 2014; Stiglitz, 2012; Wilkinson and Pickett, 2009). However, this paper shows
scaling of income distributions in a nation that is usually considered to be more equitable
than many others.
According to the ABS Household Income and Income Distribution of Australia (Cat.
6523.0, 2011-12), ‘‘the wealthiest 20% of households in Australia account for 61% of total
household net worth, with an average net worth of $2.2 million per household, whereas the
poorest 20% of households account for just 1% of total household net worth, with an
average net worth of $31,205 per household’’. Further, quoting again from the ABS
Household Income and Income Distribution of Australia (Cat. 6523.0, 2011-12), ‘‘while
the mean equivalised disposable household income of all households in Australia in 201112 was $918 per week, the median (i.e., the midpoint when all people are ranked in ascending
order of income) was somewhat lower at $790. This difference reflects the typically
asymmetric distribution of income where a relatively small number of people have
relatively very high household incomes, and a large number of people have relatively
lower household incomes.’’
Thus, from our findings it appears that the larger the city, the larger the growth of income
in the highest income categories. This would be a good and prosperous scenario if there were
a large number of people in the highest income categories. But, the ABS census statistics also
shows, that nationally, there are few people in the highest income categories commanding a
large portion of the income, with a very large number of people commanding a much smaller
portion of the income. The scaling relationship in this paper, therefore, reveals a scaling
where inequalities seem to become more pronounced with population size. It would be useful
to test the scaling of income distributions on data from other countries around the world,
especially countries that are reportedly more unequal than Australia.
Finally, computation of real income and expenditures (as opposed to total reported
income) in the income categories will be performed in our future work. Because of the
unique condition that a few urban areas house most of the population as well as all the
economic opportunities, the costs of living (especially the largest proportion of expenditure,
housing costs) are extremely high, most notably in Sydney and Melbourne. This leads to a
situation where the poor spend a substantially high proportion of their income in housing
and travel costs, whereas the rich spend, comparatively, a much lower proportion of their
income for the same goods and services. Thus, it is not immediately clear that the scaling of
post-tax or any measure of real or residual incomes (incomes minus expenditures on basics
18
Environment and Planning B: Planning and Design 0(0)
of housing and transportation) will necessarily be more equally distributed than the scaling
of pre-tax or gross measures of income. Computation of real, disposable, or residual incomes
in the income brackets may change the scaling behavior reported in this paper, and it will be
interesting to see whether real income computation intensifies the scaling observed in the
gross case, or takes it towards increasing equity. Similarly, the scaling of other forms
economic inequality, separate from income inequality, may also be quantified and studied
in the same manner. Thus, the approach overall would be nonetheless useful in studying the
scaling of income or other forms of economic inequality for a country.
The empirical findings reported in this paper are useful in advancing the understanding of
urban growth: they show that the differential scaling of income is one of the important socioeconomic effects that scales with city size, thereby calling into question the ideas around the
desirability of unbounded growth of large urban agglomerations, their management, and
planning. The findings are also useful for informing urban planning and economic policy,
particularly insights into how the size of an urban system may be an implicit driver for
inequality. Future research will try to examine some of the urban and economic processes
that are driving the findings we have observed in this research. Several hypotheses can be
made on why we observe the relationships between larger cities being more unequal by way
of agglomeration of the super-rich in these areas. First, big, global and international
businesses or technical and creative industries concentrate in large urban areas, and
therefore there exists a wage premium for the super-skilled jobs. Second, as Joseph
Stiglitz notes (Stiglitz, 2012), in recent years CEO salaries have seen unprecedented and
sharply accelerated levels of increase. Again, larger businesses and CEOs have a tendency
to agglomerate disproportionately in larger cities. Third, the world’s financial centers
concentrating in the largest cities could induce global flows of wealth between the
‘‘dragon king’’ cities (Arcaute et al., 2015), but this could also have local or regional
consequences in terms of the dependence of cities lower in the hierarchy on these largest
cities. In other words, the SUAs around Sydney and Melbourne, for example, occur in
conurbation-like/geographic bands around Sydney or Melbourne. This could mean that
global flows of wealth into Sydney and Melbourne could be causing agglomeration effects
and the regional city systems to grow in ways dependent on Sydney and Melbourne.
Consequently, patterns of incomes in this city hierarchy would then be affected by these
patterns of growth. Fourthly, due to high living costs and high prices of housing and
transport in the largest cities, partly fueled by the incomes in the highest brackets, people
on lower wages might actually be pushed out of the biggest cities (if they have a choice), or
be forced to access housing and transport at ever increasing costs (if they don’t). Both these
causes would intensify the gap between the richest and the poorest in different ways. The full
set of social and economic processes that could explain our observations are complex,
interconnected and varied, and need to be examined in detail in future work. Only then
will it be possible to provide some specific policy advice about how we can grow our cities in
ways which reduce the levels of inequality.
Acknowledgement
The authors thank PA Robinson and Andy Dong for helpful discussions and comments.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or
publication of this article.
Sarkar et al.
19
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or
publication of this article: University of Sydney Henry Halloran Trust Blue Sky Grant and a Research
Incubator Fellowship.
References
Australian Bureau of Statistic (ABS) (2011a) Australian Statistical Geography Standard (ASGS),
Australian Bureau of Statistics. Available at: http://www.abs.gov.au/websitedbs/d3310114.nsf/
home/australianþstatisticalþgeographyþstandardþ%28asgs%29 (accessed 25 April 2015).
Australian Bureau of Statistic (ABS) (2011b) Income Data in the Census, Australian Bureau of
Statistics. Available at: http://www.abs.gov.au/websitedbs/censushome.nsf/home/factsheetsuid?
opendocument&navpos¼450 (accessed 15 July 2015).
Australian Bureau of Statistic (ABS) (2011c) Table Builder, Australian Bureau of Statistics. Available
at: http://www.abs.gov.au/websitedbs/censushome.nsf/home/tablebuilder (accessed 15 July 2015).
Arcaute E, Hatna E, Ferguson P, et al. (2015) Constructing cities, deconstructing scaling laws. Journal
of the Royal Society, Interface 12: 20140745.
Atkinson A (2015) Inequality: What can be Done? Cambridge, MA: Harvard University Press.
Barthelemy M, Bordin P, Berestycki H, et al. (2013) Self-organization versus top-down planning in the
evolution of a city. Nature Scientific Reports 3: Article no. 2153.
Batty M (2007) The size, scale, and shape of cities. Science 319(5864): 769–771.
Batty M (2013a) The New Science of Cities. Cambridge, MA: MIT Press.
Batty M (2013b) A theory of city size. Science 340(6139): 1418–1419.
Bettencourt LMA, Lobo J, Helbing D, et al. (2007a) Growth, innovation, scaling, and the pace of life
in cities. Proceedings of the National Academy of Sciences of the United States of America 104(17):
7301–7306.
Bettencourt LMA, Lobo J and Strumsky D (2007b) Invention in the city: Increasing returns to
patenting as a scaling function of metropolitan size. Research Policy 36(1): 107–120.
Bettencourt LMA and West G B (2010) A unified theory of urban living. Nature 467: 912–913.
Fragikas M, Lobo J, Sturmsky D, et al. (2013) Does size matter? Scaling of CO2 emissions and the US.
PLoS One 8: e64727.
Louf R and Barthelemy M (2014) Scaling: Lost in the smog. Environment and Planning B: Planning and
Design 41.
Oliveira EA, Andrade JS and Makse HA (2014) Large cities are less green. Scientific Reports 4: 4235.
Piketty T (2014) Capital in the 21st Century. Cambridge, MA: Harvard University Press.
Pink B (2011) Australian Standard Geographic Classification (ASGC). Australian Bureau of Statistics
ABS Catalogue No., 1216.0.
Rybski D, Reusser DE, Winz A-L, et al. (2014) Cities as nuclei of sustainability? Available at: http://
arxiv.org/abs/1304.4406 (accessed 26 October 2016).
Sen A (1973) On Economic Inequality. Oxford: Clarendon Press.
Sen A (1997) From income inequality to economic inequality. Southern Economic Journal 64(2):
384–401.
Stiglitz J (2012) The Price of Inequality. New York: W W Norton and Company.
Wilkinson R and Pickett K (2009) The Spirit Level: Why Equal Societies almost always do Better.
Bristol, UK: Allen Lane.
Somwrita Sarkar is a Lecturer of Design and Computation at the Design Lab, University of
Sydney, where she leads a research program in the quantitative modeling of urban systems.
Somwrita’s research looks at complex systems and networks in urban, design, technological
and biological domains, and developing mathematical and computational models and
20
Environment and Planning B: Planning and Design 0(0)
methods to better understand and predict them. In her work in city science research and
urban computing, she has developed novel data mining and computational methods for
solving problems in urban design and planning research, especially focused around big
data and the social and economic dynamics of housing markets and transportation in the
city. The Henry Halloran Trust has awarded her a Blue Sky and a Research Incubator grant
and her PhD in Design Theory and Methodology in 2010 was awarded the Faculty of
Architecture, Design, and Planning Exceptional PhD Award and was a finalist for the
Rita and John Cornforth University PhD Medal. She was a postdoctoral fellow at the
School of Physics, University of Sydney.
Peter Phibbs is a geographer, planner, and social economist with extensive experience in
program evaluation, financial analysis, and cost benefit analysis. He has over 20 years’
experience undertaking housing research. Currently he is the Chair of Urban and
Regional Planning and Policy at the University of Sydney and also Director of the Henry
Halloran Trust at the same University.
Roderick Simpson is the inaugural Environment Commissioner of the Greater Sydney
Commission that commenced operations at the end of January 2016. Until that time he
was Director of the Urban Design and Master of Urbanism Programs in the Faculty of
Architecture, Design and Planning at the University of Sydney and principal of
Simpson þ Wilson whose work ranges across architecture, urban design, and strategic
planning.
Sachin Wasnik is a PhD candidate at the Design Lab, University of Sydney. His PhD work
focuses on modeling the role of social, online and news media in the information dynamics
of the housing market.