!

!
!
!
!
!
!
!
!
!
THE WORLD DISTRIBUTION OF INCOME AND ITS INEQUALITY, 1970-2009
Paolo Liberati
!
!
"##$!!%%&'()'*)!!"#$%&'(!)*)+$,!!
-.&)*$/&0+'/#!1&!23#'#0&*!4'&5+$,&/6!1+(7&!,/81&!9#0*!:$+;!-#'7&'+;!
!
!
"#$%&'(!<*)+$!'=!>?@!AB>A!
!
!!C! "#$%&'(! <*)+$,! 1+7! .&)*$/&0+'/#! 1&! 23#'#0&*! ,5#7(#'#! 7*! D8'E&#'+! 1&! 1&587(*$+!
/+0)+,/&5*0+'/+F!&'!D#$0*!1+D&'&/&5*!#!)$#55&,#$&*F!&!$&,87/*/&!1&!$&3+$3G+!,3&+'/&D&3G+!#$&(&'*7&H!I*!
7#$#!)8JJ7&3*E&#'+!K!,#((+//*!*77L*))$#5*E&#'+!1+7!M#0&/*/#!N3&+'/&D&3#H!!
!
!!<+$!3&*,38'*!)8JJ7&3*E&#'+!5+'(#'#!,#11&,D*//&!(7&!#JJ7&(G&!)$+5&,/&!1*77L*$/H!7!1+7!.HIHIH!@>HOH>PQRF!
'H!??B!+!,833+,,&5+!0#1&D&3G+H!!
!
M#)&+!1+77*!)$+,+'/+!)8JJ7&3*E&#'+!)#,,#'#!+,,+$+!$&3G&+,/+!*77*!9+1*E&#'+H!!
!
+,+0)7*$+!D8#$&!3#00+$3&#!
*&!,+',&!1+77*!7+((+!>Q!*)$&7+!ABBQ!'H>B?
!
!
!
!
!
!
!
!
!
!
!
REDAZIONE:
Dipartimento di Economia
Università degli Studi Roma Tre
Via Silvio D'Amico, 77 - 00145 Roma
Tel. 0039-06-57335655 fax 0039-06-57335771
E-mail: [email protected]!
!
!
!
!
!
!
!
!
!
!
!
THE WORLD DISTRIBUTION OF INCOME AND ITS INEQUALITY, 1970-2009
Paolo Liberati
!
!
!
!
!
!
!
!
!
!"#$%&%"'()$*+%$,$)"-'
'
.&/0$1$"'2*'.$3$44$5'
6++&'7$8+%&'
9&"3"':&11&0&'
:"0*%%&';&5%0"*+$'
($3<$&'=*01$'
!
!
!
The world distribution of income
and its inequality, 1970-2009
Paolo Liberati ( * )
Abstract
This paper provides for the first time a full decomposition of world inequality, as
measured by the Gini coefficient, in the period 1970-2009. In particular, using the
Analysis of Gini (ANOGI), the paper describes the evolution of between inequality,
within inequality and the impact of overlapping on both factors. While there is evidence
that between inequality in the last decade significantly declines due to the rapid Chinese
growth, within inequality and overlapping go in the opposite direction. Furthermore, if
one makes exception for some Asian countries, the rest of the world does not move
significantly. As a result, world inequality remains high by any standard.
JEL Classification: I310; H000
Keywords: World Inequality; Gini Coefficient; ANOGI; Lognormal.
Università Roma Tre, Department of Economics, CEFIP. I wish to thank Claudio Mazziotta, Pasquale Tridico
and Shlomo Yitzhaki for their insightful comments. This paper was written during a visiting period at the IMT
Institute for Advanced Studies at Lucca (Italy), whose hospitality is gratefully acknowledged.
*
1. Introduction
The issue of world inequality is at the forefront of the economic research since many
years. On the one hand, after the pioneering work of Theil (1979), the main stimulus to
the investigation of world inequality has originated from the publication of the Penn
World Tables (Summers and Heston, 1988 and 1991), which provide comparable
information on per capita GDP for a large number of countries. On the other hand, the
scarcity of data on national income distributions has for some time confined the
analysis to the between component of world inequality (Theil, 1979; Podder, 1993;
Theil and Seale, 1994; Theil, 1996; Melchior et al., 2000; Melchior, 2001; Sala-iMartin, 2006), without any attempt to estimate a world income distribution or to
calculate inequality within countries.
This shortcoming has been later addressed in a number of ways. One approach has
been to variously mix information from national accounts and survey data. For
example, Berry et al. (1983, 1989) computed a world income distribution by
apportioning the per capita GDP to income shares provided by countries’ national
surveys or estimated by a regression analysis. Similar techniques have been later used
by Grosh and Nafziger (1986), Korzeniewicz and Moran (1997), Schultz (1998),
Firebaugh (1999) and Bourguignon and Morrisson (2002), where national surveys are
always used to get the cumulative share of income received at specific quantiles of the
income distribution. More recently, Sala-i-Martin (2006) has also integrated national
accounts with micro-data to measure the dispersion of the distribution around the
mean and then used a non-parametric approach to estimate the world income
distribution.
A second approach consists in recovering the world income distribution by some
known parameters of a specific functional form. In this vein, Chotikapanich et al. (1997)
calculated the world distribution of income using information on mean incomes and
Gini coefficients under the assumption that income is log-normally distributed (see
also Dikhanov and Ward, 2001; Quah, 2002). This method has been recently applied
by Pinkovskiy and Sala-i-Martin (2009), that assume a parametric log-normal
distribution, and by van Zanden et al. (2011), that use a log-normal assumption to
calculate the Gini coefficients.
2
A third approach exclusively relies on the use of survey data. Earlier works using
this method, in a poverty context, are by Ravallion et al. (1991), Chen et al. (1994), and
Ravallion and Chen (1997). More recently, Milanovic (2002) and Milanovic and
Yitzhaki (2002) have also undertaken studies of the world income distribution using
survey data. In the first case, data refer to 1988 and 1993; in the second case, results
are replicated for 1993 only using the analysis of Gini (ANOGI) to decompose world
inequality. The scant availability of survey data, however, has often confined these
studies to specific years, preventing a comprehensive investigation of the dynamics of
world inequality.
All methods have shortcomings. In the first approach, survey data are often used to
apportion per capita GDP and not to build a ‘true’ world income distribution. In the
best case, when quantiles are missing, the most common practice is to estimate missing
values with a regression analysis or other estimation techniques (Berry et al., 1983;
Korzeniewicz and Moran, 1997; Sala-i-Martin, 2006) or to approximate missing
income distributions in some countries with the income distribution of one or more
countries in the same group (Bourguignon and Morrisson, 2002). In the worst case,
countries with missing data are not treated, weakening the generality of results. In
both cases, the distribution of income within each quantile remains unknown and
assumed to be stable.
The second approach usually entails that all distributions follow the same
parametric pattern, which may be a debatable assumption. A robustness analysis is
thus usually required to check the impact of other assumptions about the form of the
income distribution (e.g., Pinkovskiy and Sala-i-Martin, 2009).
The third approach assumes that countries’ surveys are a better representation of
the true income. Notwithstanding this belief, it is known that there are unsatisfactory
features of national surveys: their quality widely differ across countries; some income
sources are very imperfectly captured (self-employment income, financial incomes,
rents, etc.); the incomes of the very rich and those of the very poor are usually
significantly underreported (Pinkovskiy and Sala-i-Martin, 2009); individuals do not
always reveal their true income when interviewed, which means that the different
extension of the black economy may affect the size of the reported income. Thus, even
if all national surveys were available, a great number of adjustments would in principle
be required to achieve something close to the true world income distribution.
3
The superiority of one of these approaches is therefore hardly sustainable on a mere
list of advantages and disadvantages. Rather, it would be more profitable to consider
them as complementary methods, for example to set inequality bounds or benchmarks,
and to choose between them according to a criterion of appropriateness to the
investigation undertaken, rather than searching for an unavailable perfect
information.1
For these reasons, this paper does not propose an entirely new method to analyse
world inequality. Rather, it expands over existing studies by fundamentally merging
two approaches. Firstly, it follows Chotikapanich et al. (1997) and Pinkovskiy and Salai-Martin (2009) in assuming that income in all countries is log-normally distributed.
Secondly, the paper proposes a forty-year decomposition of inequality using the
Analysis Of Gini (ANOGI), as in Milanovic and Yitzhaki (2002), to investigate the
path of between inequality, within inequality and the impact of overlapping on both.
The main characteristics of this paper are two. First, it provides a unique and
consistent framework to analyse world inequality in a long-term perspective, instead
of dealing with survey- country- and year-specific biases that originate from
independent (and sometimes unclear) country practices. Second, it provides for the
first time a full decomposition of the Gini coefficient for the whole period using
ANOGI, to follow the evolution of between, within and overlapping components by
world’s regions.
The paper is organised as follows. Section 2 deals with the more consolidated issue
of between inequality. Section 3 will propose a method to complement the previous
information with an estimation of the within inequality under various hypotheses.
Section 4 will implement a more comprehensive approach to derive a world income
distribution and to describe its inequality using ANOGI. Section 5 will provide some
sensitivity analyses and comparisons with other studies. Section 6 concludes.
For example, there may be good reasons to use per capita GDP, as it includes education and health services as
well as non-disbursed corporate profits that may be invested, items that are hardly reflected in survey data. Poor
countries that have grown faster may have expanded their education and health system; income measures that
disregard these items may also underestimate forces toward convergence (Melchior, 2001; 104).
1
4
2. Between-country inequality
Between-country inequality has been a natural dimension of investigation of world
inequality in many previous contributions. The availability of data for a large number
of countries and years has indeed stimulated a series of empirical works on this topic,
that have made recourse either to per capita GDP at current prices or, more
frequently, to a PPP converted measure of per capita GDP. Following most of the
literature on the same issue, the PPP converted per capita GDP at 2005 international
dollars (henceforth PGDP) available in the Penn World Table 7.0 (PWT) will be
taken here as the relevant variable, and a Gini coefficient of its distribution across
countries calculated.2 Following another consolidated issue in empirical works, the
between inequality measure will be population-weighted, which means that individuals,
and not countries, are implicitly given equal weight (Firebaugh, 1999; Sala-i-Martin,
2006).
2.1. Drivers of between inequality: some descriptive issues
The path of Gini coefficients of PPP converted per capita GDP from 1970 to 2009 is
illustrated in figure 1 by the bold line. Several periods can be distinguished. First, from
1970 to 1990, Gini inequality has remained almost flat. It is not an easy task to detect
a single factor shaping this behaviour, which is also beyond the scope of our analysis,
but a number of concomitant factors may be in place. First, the oil boom of the first
half of the Seventies caused inequality to decline in some oil-producing states, but it
also causes inequality to rise in some oil-consuming and populous countries (e.g. the
United States, +1.3 Gini points). At the same time, no recorded change in inequality
involved China, rather – according to the available estimates – the Chinese Gini
coefficient slightly declined in this period. Second, in the Eighties (until 1987) debt
crises were the cause of rising inequality in many countries in South America, in
Europe and in Oceania, but they still did not involve India and China, where inequality
2005 is used as a reference year also in PWT 6.3 and in the 146-country benchmark 2005 International
Comparison Program (ICP). It must be noted that PWT 6.1 had 1996 as the base year. The revision of the base
year, often entailing a revision of national aggregates in the past (e.g., GDP), may lead to significantly different
levels of inequality (Milanovic, 2010), an issue that will be discussed in section 5. PWT 7.0 now covers most of the
years between 1950 and 2009 for a large number of countries.
2
5
slightly declined or stayed invariant. Third, the gap of the Chinese per capita GDP
with the world average slightly narrowed in this period, as well as that of other poor
and populous countries (Indonesia, India, Vietnam, Pakistan); at the same time, the
richest countries took additional distance from the world average. As a balance, the
levels of per capita GDP shifted upward, but their dispersion remained more or less
the same. If one makes exception for the peak in between inequality occurred in 1989
(0.624), where the large 1988 recession in China gave rise to a short-term increase of
income inequality (Sala-i-Martin, 2006), these two decades can be thought of as the age
of high and flat inequality.3 Yet, as we will see below, excluding China from the analysis
will give a slightly increasing profile of inequality over the period, which is amplified
when all Asian countries are excluded.
After 1990, we can instead identify two ages of declining inequality. The first age
(moderate decline) is from 1990 to 1995. This decline, however, stopped in the second
half of the Nineties; the persistent economic misery in sub-Saharan Africa, the
increasing inequality in the republics of the FSU, the modest economic growth in
Latin American countries, and some unfavourable economic performances in South
Asian countries all contributed to counterbalance the Chinese convergence force in
this period (Melchior, 2001).
The second age (rapid decline) extends from 2000 onwards, with a sudden and
unambiguous decline of between inequality from 0.603 in 2000 to 0.531 in 2009. This
trend appears more marked than those identified by previous studies reporting data
until around 2000. For example, Sala-i-Martin (2006) claimed that between 1970 and
2000, global inequality declined by about 2.4 per cent. Our measure of global
inequality, confirms this decline (-2.8 per cent) but it also identifies a larger structural
decline occurring from 2000 to 2009 (about 12 per cent or 7.2 Gini points). Thus, the
reduction in between inequality seems now more pronounced than that already
identified by previous studies covering varying time periods until 2000 (Sprout and
Weaver, 1992; Melchior, 2001; Sala-i-Martin, 2006).
In order to capture the determinants of this trend, it is worth analysing the
behaviour of between inequality under alternative hypotheses. The first, and most
obvious, line of investigation is to explore what happens to inequality when excluding
In defining this period as an age of high inequality we take 1970 as the starting point, disregarding the empirical
evidence that suggests that inequality increased in the ‘50s and in the ‘60s and followed an opposite trend
afterwards.
3
6
China. The outcome is again reported in figure 1 by the line with empty triangles. It is
expected that a fast-growing country that is also the most populous country in the
world, may have caused significant effects on the estimation of a population-weighted
measure of global inequality. Indeed, the graph neatly reveals that when China is
excluded, inequality, with few exceptions, has increased from 1970 to 2000.4
Within this period, two sub-periods can be identified. In the first one (until around
1995), world between inequality is always higher than world between inequality
measured excluding China. This means that China contributed to larger world
inequality, for the basic reason that its per capita GDP was well below the world
average. Yet, its inclusion makes the increasing trend to disappear, because the
convergence entailed by its even modest growth compensates the divergence that
originates elsewhere in the world. In other words, when including China, the
contribution to convergence (the fact that China is mounting faster toward average
world income) is stronger than the contribution to divergence (the fact that it is still a
relatively poor country). However, after 1995 (the second period), the two lines cross
and the inclusion of China yields a lower between inequality as the result of its
accelerating growth. But unlike what has happened until 2000, the inclusion of China
is not fundamental to mark a decline in world inequality. Obviously, when including
China, the level of the Gini inequality is lower, but the trend is declining even when
China is excluded (even though the decline of the Gini coefficient is narrowed to 3.9
points instead of 7.2).5
A second line of investigation is to analyse what happens to inequality when
excluding India, another populous and recently fast-growing country. Now, the line
with the white circles in figure 1, shows that, unlike in the case of China, the exclusion
of India leads to a smaller world inequality in the whole period without altering its
profile. This is to say that India has less power in shaping inequality trends than China
has had and still has. Furthermore, while China has been a high-inequality-driver only
from 1970 to 1995, India is still a high-inequality-driver, i.e., its inclusion increases
inequality, because it has been and still is a relatively poor country. Yet, its
contribution to larger inequality has becoming narrower in the last years, as shown by
The line has a downward shift in 1990, when data of the FSU Republics became available with a lower than
average inequality. Since inequality suddenly increased also in those countries, the same trend is observed after the
jump.
5 On this point, see also Milanovic (2010).
4
7
the smaller distance between the two lines in figure 1, the main reason being that India
is also reducing gaps with the richest countries.
A third line of investigation moves from the observation that it is the bulk of Asian
countries that have more effectively caught up to the richest part of the world since the
Eighties. The impact of this growth on inequality is analogously measured in figure 1
by excluding all Asian countries from the calculation of between inequality.6 Some
points are worth noting. Until 2002 between inequality would have been much lower
without Asian countries, which means that when adding Asian countries one adds poor
and populous countries contributing to a greater dispersion of per capita incomes. It is
indeed worth remembering that Vietnam, Indonesia, Pakistan and Philippines belong
to this area. On the other hand, China explained only a small part of this gap when it
was a high-inequality-driver (until 1995); thus, other Asian countries were overall
responsible for a greater world inequality more than China was in the same period. But
since Asian countries were for a long period invariably poor and have always had a
large population weight, their inclusion has had the effect of stabilising world
inequality until around year 2000. By excluding Asian countries, inequality in the
same period is greatly lower, but increased faster, which means that per capita GDP
must diverge among the rest of non-Asian countries. After 2000, the exclusion of
Asian countries makes the downward trend of inequality much less pronounced, which
means that there are countries elsewhere in the world that display weaker or even
opposite stimulus to the convergence that originates from the Asian continent.7
Finally, it is of some interest by itself to investigate the impact of the exclusion of
the United States, one of the richest country in the world. As reported in figure 2, US
is a higher-inequality-driver along the whole period (its exclusion would give a lower
measured inequality), with its contribution slightly increasing over time (the difference
between the two lines becomes larger when moving to the most recent years). Yet, its
exclusion does not affect the downward trend of inequality observed in the last decade.
At this stage, this means that the declining trend of inequality is something that may
partially overcome the impact of China, India and the US (which together account for
Note that the exclusion of Asian countries makes more visible the downward jump of inequality in 1990 due to the
availability of data for FSU republics, with lower than average inequality.
7 On average, from 2000 to 2004, East Asian countries and South Central Asian ones have grown by 3.3 and 3.8 per
cent, respectively. The corresponding growth in North America and North Europe was 1.6 and 2.0 per cent. From
2005 to 2009, again on average, the differential was even higher, with about 3.7 per cent for all Asian countries and
less than 1 per cent for both North America and North Europe.
6
8
about 42 per cent of the world population in 2009). Indeed, even excluding the US and
all Asian countries and even though less pronounced, the downward trend of
inequality weakly prevails from 2000 onwards (the first time in the whole period
analysed), which means that this process is fed also by other countries.
Nevertheless, the role of China remains fundamental in shaping inequality. This can
be best captured in figure 3, that reports the path of each country PGDP for selected
countries, normalised on the population weighted world average (represented by the
straight bold line at 1.0 in the graph). From 2000 to 2009, Germany, US, UK, and
Japan converge toward the world average from above, falling back to the levels of the
Seventies; at the same time, China rapidly converge to the average from below. This
explain why China contributes to inequality reduction. On the other hand, China is
moving much faster than India (the second largest country in the world) since the
Nineties, which may contribute to increase between inequality.
2.2. Contributions to global inequality
The previous results may be more effectively integrated by the analysis of the
normalised gaps of PGDP between countries and their contributions to the Gini
coefficients. This information is particularly useful to understand the role of China as a
force of convergence and divergence at the same time. The first panel of table 1 reports
some descriptive variables in selected years (the PPP converted GDP per capita, the
same variable normalised on the world average, and the population share of each
country). Comparing the normalised PGDP in 2000 and 2009, one has the clear
evidence that some previously poor countries have experienced an unprecedented
reduction of the distance from the world average (Korea, Indonesia, China, India,
Vietnam, Egypt), while the richest countries have deteriorated their positions.
To understand the consequences of these gaps, the second panel of table 1 reports
the interactions of both China and the US with both the richest and the poorest
countries, considering, in each pair, the normalised gap of PGDP, the contribution of
the specific gap to global inequality, and the change in Gini points attributable to each
specific interaction. Note that the richest country appears in the first position in each
pair.
9
Looking first at the triangle US-China-India (the first three rows of the second
panel of table 1), one can see in numbers what has been depicted in figure 3. While the
normalised gap between China and India dramatically increases from 2000 to 2009
(from 0.17 to 0.42), which means that China is moving faster than India, the same gap
between US and China and US and India declines, denoting a convergence between
each pair (but faster in the case of China). Even though the contribution of this triangle
to the level of the Gini coefficient is above 14 per cent both in 2000 and 2009 (the sum
of the contributions between each pair), the contribution of the triangle to the change of
the Gini coefficient in the same period goes in an opposite direction. While a narrow
gap between US and China and US and India contributes to inequality reduction (-1.29
and -0.59 Gini points, respectively), the faster growth of China with respect to India
contributes to greater inequality (+0.83 Gini points), as China is leaving a populous
and poor country behind. Thus, over a total reduction of inequality of 7.17 Gini points
between 2000 and 2009, the convergence between US and China and US and India can
explain 1.9 Gini points. On the other hand, inequality has increased by 0.83 Gini
points because China is growing faster than India. This clarifies why and how China is
a force of convergence and divergence at the same time.
In order to capture this contradiction, one can consider the impact of China on
inequality when it interacts with a set of richer countries (France, Germany, UK, Italy,
Spain, Japan, Brazil, Korea, Mexico and Russia). In the last decade, China is moving
faster than all these countries, as the normalised gap declines. The counterpart of this
faster growth is a large contribution to inequality reduction in 2009, of which a total of
about 1.8 Gini points is due to the convergence of China towards the levels of France,
Germany, Spain, Italy, UK and Japan together. Moving faster than Brazil, Korea,
Mexico and Russia also contributes to a further decline of 0.5 Gini points. Overall,
including the effect of the interaction with the US, the rapid growth of China between
2000 and 2009 explains a reduction of 3.6 Gini points. These numbers give now a
more precise content to the role of China as a force of convergence.
On the other hand, China is playing a different role when it interacts with poorer
(and large) countries. It is now easily seen that the normalised gaps with Indonesia,
Nigeria, Vietnam, Ethiopia, Bangladesh, Pakistan, and Egypt (accounting for 14.6 per
cent of the world population) are all increasing between 2000 and 2009, suggesting
that China is taking increasing distance from the poorest part of the world. Thus, in
10
this case, the growth of China must contribute to growing inequality. This is captured
by the last column of the second panel of table 1. The distance from Indonesia
accounts for a larger inequality of about 0.21 Gini points, while the distance from
Pakistan and Bangladesh account, respectively, for 0.18 and 0.15 Gini points. Overall,
the interactions of China with poorer countries (including India) contributes to 1.7
Gini points of additional inequality. Analogously to the previous case, these numbers
now give a precise measure of the impact of China as a force of divergence. However,
its contribution to a reduction of inequality presently overcomes the contribution to
inequality growth, leaving its net impact at -1.9 Gini points.
Consider now the impact of the United States, one of the richest and
contemporaneously populous countries. In this case, with respect to both relatively
richer and relatively poorer countries, the normalised gap declines from 2000 to 2009
(with the possible exception of Italy), which means that the PPP converted per capita
GDP of the US and those of the selected countries of table 1 are closer in 2009 than
they were in 2000. As a consequence, the slower growth of the US must also
contribute to inequality reduction. This contribution is however almost entirely due to
the faster growth of poorer countries (-1.36 Gini points, including India and excluding
China), with the convergence of relatively richer countries accounting only for about 0.1 Gini points.
Putting all things together, between 2000 and 2009, one can observe that the
interactions of China and the US are responsible for an inequality change of -3.35 Gini
points (the sum of the two net impacts in table 1), which is about 46 per cent of the
overall change of Gini points in global between inequality (about 69 per cent if one
considers only the impact of both countries on inequality reduction). This contribution
is at the same time significant but not sufficient to explain the overall decline in
inequality, which is another way to confirm what has been already observed in figure
2, that between 2000 and 2009 global inequality follows a slightly declining trend even
when the US and all Asian countries are excluded from the calculations. Thus, while
the fact remains that China is a fundamental player for both the convergence and the
divergence of between inequality, the decade between 2000 and 2009 marks an
unprecedented reduction of between inequality as measured by PGDP. Even though
India and China are two fundamental engines of this decline (Milanovic, 2010), for the
first time in the period analysed between inequality follows a weak downward trend
11
even when these two countries are excluded, even though the levels of between
inequality remain remarkably high.
3. Within-country inequality
3.1. Imputation of missing Gini coefficients
The results discussed in the previous section are exclusively based on the concept of
population-weighted between-country inequality. The observed fact that this
inequality is rapidly declining over the last decade can tell us nothing about what
happens to within-country inequality and about its possible counterbalancing effect. On
the other hand, to have a reliable approximation of within inequality and of its changes
over time, that is also consistent with the measurement of between inequality, Gini
coefficients must be available for all years and for all countries used in the previous
analysis.
Unfortunately, this series of Gini coefficients is not available. One of the most
common database containing information on Gini coefficients is the revised wave of
the World Income Inequality Database (WIID2) developed by the World Institute for
Development Economics Research (WIDER). This dataset originates from Deininger
and Squire (1996) (henceforth DS), that was first used to generate WIID1 in 2000,
covering the period 1950-1998, and then updated. In the present version, WIID2
reports two different values of Gini coefficients. The first is calculated from quantile
shares, where available. The second (reported Gini in the database) is either reported by
the source or calculated by WIDER or by DS through parametric extrapolation. In
some countries, multiple Gini coefficients are also reported in the same years, based on
different income definitions. More recently, alternative or complementary datasets
have been developed. On the one hand, Milanovic (2010) has updated the All the Ginis
database, a collection of Gini coefficients retrieved from the Luxembourg Income
Study (LIS), the Socio-Economic Database for Latin America and the Caribbean
(SEDLAC), the World Income Distribution (WYD) dataset, the World Bank East and
Central Europe (ECA) database and the WIDER dataset. As a result, this new
database gives Gini coefficients for 1,541 country-year pairs, variously based on
12
income or consumption, on net and gross incomes and on whether the recipient unit is
household or individual. On the other hand, Solt (2009) has developed a Standardized
World Income Inequality Database (SWIID), to enlarge the comparability of Gini
coefficients of gross and net incomes, using a combination of WIDER and LIS, this
latter being used as a benchmark in the calculation of Gini coefficients. SWIID gives
4,459 Gini coefficients for 173 countries in (irregular) intervals between 1960 and
2009. In all cases, the datasets are far from being comprehensive collections of Gini
coefficients for a long-run analysis of inequality. Thus, to various extent, the analysis
is either limited by the availability of data or required to impute missing information.
Our first choice to address the issue of within inequality is the WIDER dataset with
4,333 country-year observations, despite its heterogeneity and a number of
shortcomings (Atkinson and Brandolini, 2001; Galbraith and Kum, 2004). The
heterogeneity of the dataset relates to different dimensions. In particular, age coverage
is “All” for 4,199 observations; area coverage is “All” for 3,555 observations;
population coverage is “All” for 3,464 observations; and the unit of analysis is “Person”
for 3,326 observations. Keeping the observations with all these characteristics at the
same time would reduce the dataset to 2,051 observations that are still largely
heterogeneous with respect to the income definition and to the equivalence scale used.
For example, 1,418 observations use household per capita, 256 the modified OECD
equivalence scale, while the other observations are dispersed on ad hoc equivalence
scales. Furthermore, 757 observations refer to disposable income, 319 to consumption,
265 to gross incomes, 242 to disposable monetary incomes and 126 to some not
specified definitions of income (disregarding other cases with a smaller number of
observations).
The heterogeneity of the dataset also extends to the geographical distribution of
these definitions. As reported in table 2, Europe and America make preferential
recourse to income-based measures (disposable income, gross income or earnings),
while Africa, and to a less extent Asia, mostly use expenditure-based definitions. This
may be less a cogent problem at least in those countries where the share of direct taxes
is minimal; but it is worth noting that there is a huge amount of observations whose
“income” definition cannot be classified among the previous ones. Thus, discarding
heterogeneous observations would be impracticable, if one wants to cover the entire
world, an obstacle that is common to other studies on the world distribution of income
13
(see for example Milanovic and Yitzhaki, 2002, who mixed incomes and expenditures).
Indeed, the choice to consistently use an homogeneous segment of data will severely
limit the possibility to discuss global trends in income inequality (Heshmati, 2006).
Alternatively, one can argue, as in Pinkovskiy and Sala-i-Martin (2009), that the
problems of heterogeneity may be less serious when observing inequality changes or
trends rather than levels.
For the practical impossibility of having a totally homogeneous dataset, we first
choose to take Gini coefficients as they are reported in the WIDER dataset and then to
propose a sensitivity analysis of the results with alternative datasets. A first step is to
classify countries according to the availability of Gini coefficients. Countries fall into
one of the following cases: 1) countries for which one or more data points on Gini
coefficients are available; 2) countries as in 1) but where multiple Gini coefficients for
the same years are available, possibly based on different income definitions; 3)
countries with no data on Gini coefficients.
In the first case, Gini coefficients are taken as given regardless of the income
definition. In the second case, Gini coefficients based on different definitions of income
for the same years have been averaged, an issue that involves 539 country-year pairs.
In the third case, nothing has been done at an initial stage. After this step, we are left
with countries having one or more single data points on Gini coefficients between
1970 and 2009 and countries with no data on Gini coefficients.8 Note, however, that on
average, the countries for which Gini coefficients are not available account for about
6.6 per cent of the total world population in the period; thus, their inclusion or
exclusion should not have a dramatic impact on world inequality measurement.
The second step is to integrate the missing values of Gini coefficients in those
countries where some coefficients are already available for some years and in those
where the vector of Gini coefficients is empty. The simplest way to perform this task is
to exploit as much information as possible with a regression analysis.9 In particular,
the regression is based on the relationship between inequality and per capita incomes
firstly proposed by Kuznets (1955). It is worth noting that the focus of our method is
Countries with no data are: Afghanistan, Angola, Antigua and Barbuda, Bahrain, Belize, Bermuda, Bhutan, Brunei,
Cape Verde, Chad, Comoros, Congo, Dominica, Equatorial Guinea, Eritrea, Gambia, Grenada, Kiribati, Kuwait,
Kyrgyzstan, Laos, Lebanon, Libya, Macao, Macedonia, Maldives, Marshall Islands, Micronesia, Montenegro,
Oman, Palau, Qatar, Russia, Samoa, Sao Tome and Principe, Saudi Arabia, Solomon Islands, St. Kitts & Nevis, St.
Lucia, Sudan, Syria, Timor-Leste, Togo, Tonga, Trinidad & Tobago, United Arab Emirates, Vanuatu, Yemen.
9 The use of regression techniques is not new. Berry et al. (1983), for example, performed a regression analysis to
estimate missing quantiles needed to identify the distribution of income within countries.
8
14
not to dispute or to validate the existence of a Kuznets curve, which is not the main
aim of the analysis, but to derive a profile of the Gini coefficient in each country, on the
basis of an assumed link between inequality and development.
A panel analysis is performed on which the Gini coefficient is regressed on per
capita incomes and their inverse (Anand and Kanbur, 1993; Deutsch and Silber, 2001)
controlling for the degree of openness of the economy at constant prices (openk in Penn
World Tables), and for different income definitions and equivalence scales by
introducing dummy variables.10 In symbols:
(1)
where G is the Gini index, Y is the log of per capita income, Open approximates the
openness of the economy by the sum of imports and exports over GDP (Anand and
Kanbur, 1993) and X is a vector of dummy variables controlling for different
definitions of income and equivalence scales.11 The panel analysis is performed with a
fixed effect estimator to capture country-specific effects and the possibility that
different countries may lie on Kuznets curves that have the same shape but different
intercepts.12 Since we are interested in inequality within countries – and to the overall
determinants of inequality – fixed effects are a convenient estimator to capture intracountry variation.
Results are reported in table 3, where it is shown that the coefficients of the log of
per capita income and its inverse are statistically significant. Of particular importance
is also the expected downward shift of the Gini coefficient when measured by
consumption (about -2.5 Gini points) and earnings (about -1.2 Gini points) and its
upward shift when Gini is averaged (+1.9 Gini points) and measured without
adjustments for equivalence scales (+1.4 Gini points).13 The predicted values of
equation (1),
, now give a vector of Gini coefficients for all country-year pairs, with
For example, consumption inequality is usually lower than income inequality, and inequality of non-adjusted
incomes is usually higher than inequality of equivalent incomes. The set of control variables is severely limited by
the availability of data for all countries and years.
11 Similar control variables have been used by Schultz (1998) for the analysis of intra-country inequality.
12 The basic regression is run without time dummies. The inclusion of time dummies has been experimented and it
does not affect the results, as strong and systematic significance levels are shown only in the period from 1977 to
1988. After 1994, time dummies are not significant.
13 Just to recall that the baseline is Gini coefficients calculated on the basis of definitions that cannot be classified as
gross income, consumption and earnings. The lack of statistical significance of the dummy gross income suggests
that these definitions are much closer to gross incomes than to other cases.
10
15
estimated values ranging from 0.174 to 0.776, which are sensible values. In principle,
these estimations would allow to draw a profile of inequality over time for each
country, regardless of the already available Gini coefficients. However, we first choose
to superimpose the estimated profile to the available Gini coefficients and fill the gaps,
instead of using the full vector of predicted values.
Thus, for those countries where the vector of the available Gini coefficients is
empty, the estimated vector of
is fully imputed. For the other countries, available
Gini coefficients provide anchors to fill the missing values according to the estimated
profile. This amounts to extrapolate the values of the missing Gini coefficients from
the vector of
. To capture the essence of this method, assume, for country j, that the
fixed effects estimation gives a series of five simulated Gini coefficients in five periods
identified as
and
to
. Assume also that the available Gini coefficients are only
, so that the other three coefficients must be recovered from the simulated
ones. In this case, for example, we would have that
, where the term in
square brackets gives the trend of the estimated profile (upward if the ratio is less than
one; downward in the opposite case), which is then anchored to the nearest Gini
coefficient available ahead (
). If no Gini is available ahead, the benchmark becomes
the first Gini coefficient back. Assume an upward trend, so that
. Thus,
in the same proportion.
The final result of this process is a series of Gini coefficients anchored to the
original data but with gaps filled imposing the same profile of
. This method is
particularly useful as it may take into account trends of inequality if they are
particularly strong, at the same time avoiding sudden jumps of the Gini coefficients
from one year to another. As a robustness check, however, the attempt will be made to
use the full vector of
to calculate within inequality, without taking into account the
anchors.
16
3.2. The trend of within inequality
Once the Gini coefficients are available for all countries and for all years, within
inequality can be estimated as a part of the standard Gini decomposition (Pyatt, 1976),
by which
, where
coefficient of country i,
is within inequality,
is its population share and
is now the Gini
is its income share. As it will
become clearer in the next paragraphs, standard within inequality as defined by
cannot include any impact of overlapping, which are instead fully
contained in the residual term
, which is still not recoverable at this stage.
Figure 4 reports the outcome of this estimation by continents. Some points are
worth noting. First, total within inequality (the bold continuous line) has been almost
stable until the mid-Nineties and has increased steadily since then. This marks a
similarity and a difference with the path of between inequality. The similarity concerns
the long age of stability from 1970 to 1995.14 The difference concerns the trend
followed from 1995 onwards, which is now positive.
Second, the trend of within inequality is totally driven by Asia and, in particular, by
China. Indeed, within inequality calculated in the Asian continent (the thin continuous
line) has the same profile of the overall within inequality. In other words, it is Asian
countries that dictate the trend, while the other countries just play a shifting effect.
Within Asia, however, it is China that shapes the trend. If one excludes China (the dot
line with the cross), the marked upward trend of inequality within Asian countries
disappears and the levels significantly fall. In all other continents, within inequality is
low and flat. We estimate that the change in the Chinese Gini coefficient from 2004 to
2009 (+5.1 points, about 11 per cent more than the Gini in 2004) entails an increase of
within inequality of about 0.15 points, which is about 6 per cent more than the level of
within inequality measured when assuming the invariance of the Chinese Gini
coefficient from 2004 to 2009. Thus, it seems that large increases of the Gini
coefficients do not affect the within component significantly, unless China and India
will become countries with a large world income share (see also Milanovic, 2002) and
with a Gini coefficient greater than the mean Gini in the world. The same result, only
slightly weaker, is still visible in the case where the simulated Gini coefficients ( ) are
14
The stability of within inequality for a smaller number of countries (49) has been observed also by Li et al. (1998).
17
used instead of anchoring the estimated profile to the available Gini coefficients (the
line with white boxes). The main reason of the distance is that the vector of
for
China underestimates the Gini coefficients projected on the basis of the available
anchors.
Third, within inequality, in absolute values, is small, the main reason being that any
given Gini coefficient is weighted by the product of population and income shares.
Large weights would emerge in richer and populous countries, while small weights
would be attached to poorer and small countries. But usually, large countries are
poorer and smaller countries are richer. The composite weight is thus on average
small, which makes the magnitude of within inequality dwarfed by the change in per
capita incomes (e.g. Quah, 2000). In our dataset, the highest weight is observed in
China 2009 (0.029) as the product of a population share of 19.8 per cent and an income
share of 14.7 per cent. The same weight in 1970 was 0.01, which means that, in the
calculation of within inequality, a given Gini coefficient in China counts three times as
much today than forty years ago. On the other hand, the highest US weight was in
1970 (0.015), as a product of a population share of 6.1 per cent and an income share of
25.3 per cent. This weight, in 2009, has declined to just around 0.01, which is the same
weight achieved by India in the same year, with a much larger population and a much
lower income share.
Overall, within and between inequality have thus followed an opposite path. This is
especially true in the last decade. The change in between inequality is always negative
from 2001 onwards, while the change in within inequality, in the same period, is
always positive (in both hypotheses of Gini coefficients). Even though the magnitude
of the change of between inequality is larger, the reduction of total inequality may
appear lower than it seems by looking only at the between inequality. However, given
the imperfect decomposition of the Gini coefficient, the treatment of overlapping on
both within and between component must be addressed, which is part of the analysis of
the next section.
18
4. The world distribution of income and the analysis of Gini
4.1. Recovering the world income distribution
The methodology adopted in the previous section leaves us with a full country-year set
of observations on mean incomes (per capita GDP) and Gini coefficients. This makes
possible to draw a long-run profile of at least two major components of the Gini
disaggregation, between and within inequality, and follow their trends and relative size
over time. Yet, nothing can be said about the underlying income distribution, which is
in fact unknown, and on the impact of the overlapping term, which may also be
significant when total inequality is measured by the Gini coefficient. In this section, a
complementary method to recover the world income distribution and to provide a full
decomposition of Gini inequality in a long-run perspective is proposed.
The starting point of the approach is to assume that income X is log-normally
distributed (as, for example, in Pinkovskiy and Sala-i-Martin, 2009). Accordingly,
is normally distributed. By the properties of the lognormal distribution, we
know that any income X of the original income distribution can be calculated by using:
(2)
where the mean and the standard deviation in the round brackets refer to the
distribution of log incomes Y and Z assumes the values of a standard normal
distribution. This means that if
and
were known, the entire income distribution
X could in principle be recovered. Aitchison and Brown (1957), showed that if income
is log-normally distributed, the Gini coefficient could be obtained by
where
,
is the value of the cumulative standard normal distribution.15 In our case,
the Gini coefficient is known, while the unknown parameter is
. Inverting the
previous formula, one can get:
Just recall that the assumption that income is log-normally distributed in each country does not entail that world
income (as a sum of lognormal distribution) is also log-normally distributed.
15
19
(3)
where now
is the value of the inverse of the cumulative standard normal
distribution. Furthermore, in a log-normal distribution,
. This means
that once the standard deviation is estimated by (3), the mean of Y can also be
estimated, if
(as in our case) is a known parameter. Having both
and
, the
full underlying income distribution X can be recovered from (2).
This procedure, for each country, would allow to recover the income distribution on
the basis of the anchors provided by the available Gini coefficients and per capita
incomes. Chotikapanich et al. (1997) used this method to calculate population and
income shares, while Pinkovskiy and Sala-i-Martin (2009) estimated the variance using
least squares regressions on the quintile shares reported in the surveys. It is worth
remembering, for the sake of clarity, that the anchors provided by the Gini coefficients
are those estimated using equation (1) and thus possible imperfections in their
estimates are transplanted to the whole income distribution. On the other hand,
without a full set of Gini coefficients, this method would be applicable only to those
cases where Gini coefficients were available, but in this case one would be back to the
issue of insufficient information for the long-run analysis of world inequality.
In order to assign the appropriate weight to each country in the world income
distribution, the Chinese income distribution is built by imposing five hundred
thousands observations, while other countries are assigned a number of observations
that is proportional to the ratio of their population to the Chinese one. This method
yields a dataset of more than 2 millions of observations in any year. Its potentialities,
however, should not conceal possible shortcomings, as raised in Milanovic (2002; 5354). On the other hand, the shortcomings of this approach should not be magnified
compared with alternative methods of analysis of world inequality (Pinkovskiy and
Sala-i-Martin, 2009; 4-6). Comparisons with other results and the sensitivity analysis
of section 5 will help understand the complementarity of this method with alternative
options.
20
4.2. The ANOGI decomposition: conceptual issues
The availability of the world distribution of income makes possible to investigate in
more detail inequality issues and refine the investigation of the previous two sections,
by decomposing the Gini coefficient according to the Analysis Of Gini (ANOGI) as
developed by Yitzhaki (1994). This would extend over the work by Milanovic and
Yitzhaki (2002) where ANOGI was applied only to the income distribution of 1993,
with data provided by countries’ household surveys.
As well known, one problematic issue of the analysis of the world income
distribution with Gini coefficients was indeed related to its imperfect decomposition
(Pyatt, 1976), as the Gini coefficient cannot be fully interpreted as the sum of a between
and a within component. A residual term (overlapping) completes the decomposition
that is not recoverable unless the full income distribution is available. The results of
the previous section, while suggesting that between and within inequality may follow
opposite directions, can tell nothing about the intensity of the overlapping term and its
impact on total inequality. This is the reason why all empirical studies, at the best,
provide an overall value of the Gini coefficient, while inequality decomposition is
usually performed with a Theil index or similar decomposable inequality measures. In
what follows, we try to fill this gap applying ANOGI, and providing a full
decomposition of the Gini coefficient in between inequality, intra-group inequality and
overlapping terms for the whole period.
In particular, let
defining the world income distribution given
by the union of the income distribution of different n countries, and denote G as the
Gini coefficient of the world income distribution. According to ANOGI, the Gini
coefficient can be decomposed as follows:
(4)
where
is the overlapping index of country i with the world distribution,
is the
is the between-Gini index of the Pyatt’s
Gini coefficient of between-inequality, and
(1976) decomposition.
21
The overlapping term is of particular relevance in this decomposition. Formally, it
, where the numerator is the covariance between
can be defined as
incomes of country i and their ranking in the world income distribution
, while
the denominator is the covariance between the same incomes and their rankings within
each country. This means that
if the incomes of a country i have the same
ranking as in the world income distribution, i.e. if the two distributions perfectly
overlap. More generally,
when the scatter of the ranks of a given country is
narrower than that of the total population; analogously,
when the scatter of the
ranks of a country is larger than that of the total population. It is worth noting that by
the same definition of the overlapping term, one can interpret the overlapping of
country i with respect to total population as the weighted sum of overlapping of
country i with all other countries j. In symbols:
(5)
where
is the share of population of country j and
, where the
denominator is as before and the numerator is the covariance between the income of
country i and their ranking in the population formed by the union of country i and j. If
no member of the j distribution lies in the range of distribution i, country j is a perfect
. In this case,
stratum and
, i.e. total overlapping is exactly equal to the
share of population i. When the two distributions are identical, instead,
and
.
By the properties of the overlapping term, the interpretation of
becomes easier, as it embodies the impact of overlapping (
inequality (
) on intra-group
). This characteristic is easily shown if one considers what happens
when the distribution of incomes of different countries form a perfect stratum, i.e.
when they do not overlap. In this case, we know that
, thus
, which is the definition of within inequality
22
in the standard Gini decomposition, where the absence of overlapping would make the
Gini coefficient exactly decomposable. However, in the general case
, which
means that IGO provides either a negative or a positive revision to intra-group
variability IG, respectively for
and
. This explains how overlapping may
affect the within-component.
Consider now the definition of the between Gini coefficient. In equation (4), a
distinction is made between
defines
and
. The original Pyatt’s (1976) decomposition
, by which between inequality is measured considering the covariance
between mean incomes and their rank among the distribution of mean incomes of all
countries. Thus, in the Pyatt’s decomposition, it is the rank of the mean income of the
country that defines between inequality. Alternatively, Yitzhaki and Lerman (1991),
provide a version of between inequality based on the covariance between mean
incomes and the mean rank of individuals according to their incomes in the country.
and
More formally, one can define
, where
is the mean rank of individual incomes in country i in the world income
distribution and
is the rank of the mean income of country i among the
distribution of mean incomes of all countries. When countries are perfectly stratified,
for example in the case where all poor individuals live in a country i and all rich
individuals live in a country j, there will be no overlapping among distributions. Thus,
for each country, the mean rank of individual incomes is equal to the rank of its mean
income in the total distribution, which means
and then
. In this
special case, the Pyatt’s Gini decomposition and the ANOGI decomposition gives the
same between inequality. In the general case where some incomes overlap, the two
ranks differ. This implies that the correlation between the rank of mean incomes and
the average rank of incomes is less than 1. In this case,
alternatively,
, or,
.16 In this latter form, the ratio can be used as an indicator of the
reduction of between inequality caused by overlapping of incomes across countries
, as in the case where one distribution has a low mean rank of
It is also worth noting that one can have
individuals incomes (e.g., there are many poor individuals), but at the same time, it has a higher mean income
because of the presence of few very rich individuals. In this case, the covariance between mean incomes and the
mean rank can be negative. See Frick et al. (2006).
16
23
(Milanovic and Yitzhaki, 2002). Thus, for a given Gini coefficient of total inequality, a
reduction of
must be associated with an increase of overlapping, embodied in the
last term of the r.h.s of equation (4), as shown in Frick et al. (2006).
4.3. The inequality of the world income distribution
Equipped with these techniques, we now proceed to discuss the ANOGI decomposition
for each year with individual countries as benchmark units (table 4). Some points are
worth noting. First, total inequality declines, with few exceptions, even though the
downward trend is particularly accentuated in the last decade (column (1)). Yet, a Gini
coefficient of 0.650 in 2009 is still high by any standard. To get the implications of this
number, it is worth recalling that this Gini coefficient would almost correspond to an
income distribution where 66 per cent of the population had zero incomes and all
incomes were equally divided among the rest of the population (Milanovic and
Yitzhaki, 2002).
Second, as reported in figure 5 where all series are normalised to the corresponding
means, the declining trend of total inequality (the bold line) is the outcome of two
different behaviours. On the one hand, between inequality pushes total inequality down,
especially since 2000 onwards, an issue already discussed in section 117; on the other
hand, within inequality (sGO in column (2)) partially compensates this decline, pushing
inequality upward. In absolute values, it is confirmed that the decline of between
inequality is larger than the increase of within inequality including the impact of the
overlapping term. This is the reason why total inequality declines; yet, it is clear from
figure 5 that the mounting distance of within inequality from its mean is larger than
the negative distance of between inequality from its mean, which implies that within
inequality grows at a faster pace than how between inequality declines.
Third, when within inequality is further decomposed to isolate the standard within
inequality and the impact of overlapping, one can note that overlapping has a nonnegligible impact on total inequality. To capture the essence of this impact, it is
It is worth recalling that the between-Gini of the ANOGI decomposition reported in column (3) is different from
the between-Gini calculated in section 1 (which is that in column (6)), for the reasons above discussed. Even though
differently calculated, the declining profile of between inequality is however preserved.
17
24
convenient to combine equations (4) and (5) to get
.
The previous expression suggests that overall within inequality (the left term), can be
split in standard within inequality without overlapping (the first term on the right hand
side) plus the impact of overlapping on within inequality (the far right term).18 This latter
term is calculated by the sum of the contribution of each country i to intra-group
inequality times the sum of its overlapping with each other country j weighted by the
population of country j. If all countries were perfect strata (i.e.
), then the
, which is the exact measure of
previous expression would become
within inequality provided by the standard Gini decomposition. On the other hand, if
income distributions in all countries would perfectly overlap (i.e.
) then:
Table 4 (columns (11) and (12)) clearly shows that the two terms have moved in the
same direction. As noted before, standard within inequality (column (11)) has increased
from 0.015 in 1970 to 0.026 in 2009, with a rapid growth occurred again in the last
decade. On the other hand, the most important contribution to total within inequality
comes from the increasing impact of overlapping (column (12)). Note, however, that in
the last decade, standard within inequality is moving faster than overlapping in
shaping within inequality. From 1970 to 2000, indeed, standard within inequality was
about 11 per cent of the total within inequality as measured by ANOGI. In the last
decade, the percentage has first risen to 11.5 per cent and then peaked at 12.4 per cent
in 2009. By simply extrapolating this trend would mean that within inequality is likely
to play a more prominent role in total inequality in the near future.
It is worth noting that what we call here within inequality without overlapping is the standard within inequality
term in the classical Gini decomposition where overlapping is entirely contained in the residual term of that
decomposition.
18
25
4.4. The ANOGI decomposition by world regions
The same analysis as before is now repeated using world regions as reference units in
selected years. Results are reported in table 5. It is worth starting the comments from
column (7), the mean rank. Just recall that the mean rank is the expected rank in the
world income distribution of an individual living in a given region. This column
clearly depicts the persistence of several worlds. Firstly, considering East Asia (which
includes China), one can note that in 1970 the mean rank of individuals living in that
region was 39.3rd percentile, i.e. well below the median in the world income
distribution. In the following years, the mean rank of East Asian people has climbed,
rising to the 53.5th percentile in 2009, mainly due to the rapid Chinese growth. The
same effect, for example, cannot be noticed for South Central Asia, where a number of
populous and still poor countries belong to (Bangladesh, India, Pakistan and then
Bhutan, Nepal, Sri Lanka and Maldives). In this case, the average rank of people in the
area was 34th percentile in 1970 and still 34.9th percentile in 2009 after a fall to around
the 30th percentile in all other years. Thus, even within the fast-growing Asia, there
are countries that do not move significantly in forty years relatively to other parts of
the world.
Secondly, looking at the various partitions of African countries (Central West
Africa, East Africa, North Africa and South Africa), there is the clear impression that
the continent is losing positions and it has not benefited from the rapid growth
occurred elsewhere. In all cases, the average rank of individuals living in Africa (with
the exception of North Africa) is lower in 2009 than in 1970. According to our
estimates, the rank of Central West African countries would fall from the 33.2nd
percentile in 1970 to the 17.3rd percentile in 2009, which means that on average all
Central West Africans are relatively poorer now than forty years ago. South African
and East African countries are not performing better, while North African countries
have recently fallen back below the median. Quite interestingly, South American and
Central American countries, that were, on average, well above the median in 1970,
have either converged to the median rank (South America) or even fallen below it
(Central America), deteriorating their position in the world income distribution.
26
Thirdly, as a world apart, the mean rank of Europe, North America and Oceania,
with only one exception in 2000, was always above the 70th percentile of the total
ranking, with North European always above the 80th percentile.
Thus, with very few exceptions, the dynamics of the world income distribution is
slower than it seems, if one makes an exception for China. The same impression can be
get by looking at the income shares of different areas over time (column (8)). East
Asian countries had 18.5 per cent of total world income in 1970 and have 28 per cent
of the same income in 2009. Correspondingly, even the shares of the richest countries
(North Americans and North Europeans) have significantly fallen over the period.
Given the importance of China in the world income distribution, it is not a case that
the Gini coefficient of the East Asian area is always very close to the total Gini
(column (1)).
A further way of looking at the nature of the regions is to analyse the overlapping
index. East Asian countries have overlapping indices that indicate that their
distribution mimics the world income distribution. More recently, however, the
overlapping index is below 1, which means that these countries form more a stratum
now than in the past. This is due to the fact that the Chinese growth partly closes the
gap with the richest part of the world and then leaves behind the bulk of poorer
countries. One can indeed note from the overlapping column that the richest regions of
the world have very low overlapping indices (below 0.5). Especially in North Europe,
the overlapping index is extremely low (0.209) which means that, even though not a
perfect stratum (as
), North European countries are very far from representing
the typical world income distribution. On the other hand, the overlapping index of the
South African region, which is always and increasingly above 1, indicates that this
region is rather heterogeneous with respect to the world, being more characterised by
two separate strata, one richer and the other poorer than the rest of the world. This
means that the scatter of ranks of people in South African regions is larger than the
scatter of ranks of the rest of the population in the world. In turn, this implies a high
Gini index (ranging from 0.65 to above 0.7).
Four worlds seems thus to emerge from the analysis. Some Asian regions, involved
in a rapid growth; almost all African regions, worsening their conditions; Central and
South American regions as well as South Central Asia to some extent preserving their
living standards or slightly deteriorating it; North America, North Europe and
27
Oceania in their ‘splendid isolation’. In any case, it is rather impressive that in forty
years, almost all regions are stuck where they were in terms of relative positions, with
very few exceptions.
4.5. The overlapping matrix
As discussed above, the general overlapping index can be obtained by the sum of
overlapping indices among pairs of countries. A useful product of the ANOGI
decomposition is then the investigation of the matrix
. Table 6 reports this
information for 1990 and 2009. It is worth recalling that the matrix is not symmetric.
Rows represent the region whose distribution is used as the base distribution (region
i). Several things can be noted that characterise the existence of different worlds.
Consider the matrix for 2009. First, when the richest part of the world is used as a
baseline (North America, North Europe and Oceania), one can see that African regions
have nothing in common with advanced economies; rather they form almost a perfect
stratum with respect to the income distribution of the richest countries. When African
regions are used as the baseline, richest countries also form a stratum in many cases,
with the caveat that the overlapping indices are usually higher. For example,
, where region 5 is Central West Africa and region 10 is North America.
The interpretation of these differences is that usually there are relatively more (poor)
citizens of the “richest world” in the range of Africa’s distribution, than there are
Africans in the range of the income distribution of the richest regions. This is also
more clearly seen by comparing the European (region 11) and the East Asian regions
(region 1). In this case,
=0.187, which means that there are only a few percent of
East Asian people that fall into the income range of European countries. On the other
hand,
=0.898, which means that more Europeans are within the income range of
the East Asian distribution. Particularly interesting is then the case where
. It is
worth recalling that in this case, the scatter of ranks of the income distribution of the
base country is larger than the scatter of ranks of the other distribution. In other
words, the base country forms two strata, one poorer and one richer than the country
whose distribution is compared. It is worth noting that this occurs significantly for
South Africa compared with most of the income distribution of other regions, which is
28
another way to capture the large inequality of the income distribution of this region (a
Gini coefficient of 0.707).
What has changed compared to 1990? Almost nothing, as can be seen in table 6.
When the richest part of the world is used as a base, African regions were already a
world apart, and South Africa had the same characteristic of having two strata with
respect to many regions in the world. That the characteristics of overlapping have not
significantly changed across regions, is indirectly supported by a very high correlation
coefficient between the columns of the matrix in 1990 and 2009. Interestingly, the
lowest correlation appears in the case of the East Asian region, where China is
included and where the most significant changes in the income distribution have
occurred in the last decade, again a support to the idea that, with very few exceptions,
world’s regions are moving slowly along the world income distribution.
4.6. The ranking matrix
Table 7 finally shows the average ranking of members of one region in terms of the
other, which means that the main diagonal is 0.5 for all regions. Thus, a value greater
than 0.5 means that, on average, people in the base region i are richer compared with
people in other regions. The opposite occurs for values lower than 0.5. In 2009, it is
striking to note, for example, that – on average – people living in North America and
in North Europe would rank, respectively, around or above the 95th percentile of the
, most African people would rank
distributions of all African regions. Since
in the lowest decile of the distribution of the richest world. In particular, compared
with North America, South African people would rank at the 5.7th percentile (thus, in
the middle of the lowest decile), while compared with North Europe, they would rank
at the 1.8th percentile of the corresponding income distribution. The same argument
holds for all African regions, and even more for Central West and East Africa. It is
also interesting to note that, despite the rapid growth of the Chinese economy, the
average rank of an East Asian individual in the North European income distribution
would be at the 7.7th percentile, which implies that the average rank of a North
European would be at the 92.3rd percentile. Some African regions would instead
perform better compared with Asian and Latin America regions (Central and South
29
America). In particular, the average rank of a North African in the East Asian
distribution would be at the 45.2nd percentile, which increases to the 70.5th percentile
when considering the distribution of the South Central Asian region. The main reason
is that even though the average income of this region is higher than the average
income in North Africa, in South Central Asia there are masses of people in India,
Bangladesh and Pakistan that have a very low rank in the world income distribution.
This same rank would be 53.9th percentile in Central America and 41.9th percentile in
South America.
Quite strikingly, the situation was again almost the same in 1990. The position of
North American and North European people with respect to all African regions were
already above the 95th percentile. The Chinese growth, however, has had an impact, as
in 1990 the average rank of an East Asian individual in the North European
distribution was at the 4.6th percentile; while North African people have slightly
climbed positions compared to Latin America (Central and South America). Again, the
fact that average ranks have not changed significantly across regions, is indirectly
supported by a very high correlation coefficient between the columns of the matrix in
1990 and 2009. Thus, in the last decade, China moves fast but the rest of the world
does not move significantly.
4.7. The concentration of world income
All indicators go in the direction of suggesting a large concentration of world income
in the hands of few people. As a final synthetic outcome, figure 6 shows the Lorenz
curve of the world distribution in the initial and in the final year of our analysis (1970
and 2009). Two things are worth noting. The first is that world income is slightly less
concentrated now than it was in 1970. The second is that income is still largely
concentrated in the hands of few people, which may incidentally opens the debate on
how the alleged benefits of the globalization wave have spread across countries or
world’s regions. The coordinates of the Lorenz curve reveals that in 1970, 80 per cent
of the population owned 23.7 per cent of income. After forty years, the same
percentage of population still owns less than 31 per cent of total income. As a
consequence, the top 10 per cent of the population still owns about 50 per cent of the
30
world income. Even worse, the bottom 20 per cent of the population has also slightly
worsen its position, as in 2009 the corresponding share of income (1.48 per cent) is less
than the share they owned in 1970 (1.51 per cent).
To this purpose, the first panel of table 8 shows the evolution of the income shares
at the bottom and at the top decile from 1970 to 2009 at five-year intervals. It is clear
that the share of the top decile has been slightly eroded, a fact that is partially
responsible for the dominance of the 2009 Lorenz curve in figure 6. On the other hand,
the share of the bottom decile has also been eroded. In percentage terms, the income
share of the poorest people has declined by more than 21 per cent, while the income
share of the richest people has only declined by less than 10 per cent. As argued above,
world income remains largely concentrated, as Gini coefficients above 0.65 would be
considered intolerable in any single country.
Furthermore, the second panel of table 8 shows the frequency of incomes in the
bottom and in the top decile for each continent between 1970 and 2009. In the bottom
decile, a great composition change has taken place between African and Asian people.
In 1970, 68.5 per cent of people in the bottom decile were from Asian countries; in
2009, this percentage has fallen to 35.8. Conversely, in 2009, the share of African
people falling in the bottom decile has more than doubled (60.1 per cent against 28.3
per cent). However, while Asian people have doubled its presence in the top decile
(from 13.6 to 27 per cent), African people are still virtually absent.
5. Robustness of results
5.1. Comparisons with other studies
How reliable are our results? As any study that involves estimation procedures (at
least partially) it is useful to compare its performance with other available results.
Figure 7 proposes this comparison with both studies showing a sufficient series of data
and studies with scattered evidence on Gini coefficients. The top graph of figure 7
reports the comparison between our estimates (for both total and between inequality)
and those studies with long time series (Sala-i-Martin, 2006; Pinkovskiy and Sala-iMartin, 2009). With regard to total inequality, the recent paper by Pinkovskiy and
31
Sala-i-Martin (2009) (henceforth PS) gives the profile identified by the white box until
2006. As clearly visible, our estimates (the continuous line) give a higher level of total
inequality, yet the declining profile strongly mimics the PS estimations, with the
possible exception of the second half of the Nineties. We think that the main reason for
this discrepancy is that our estimates are based on PWT 7.0, at 2005 international
dollars, while PS estimations are based on PWT 6.2, evaluated at 2000 international
dollars. As argued by Milanovic (2010), the recent revision of PPP, which has changed
the GDP estimates for China and India, may entail substantially higher inequalities
than previously thought. By a different strategy, our results confirm this impression,
even though our Gini coefficients for 2002 and 2005 are slightly lower than those
measured by Milanovic (2010; table 4).
It is worth noting that changing the reference year for the calculation of PPP
converted per capita GDP may cause discrepancies also in the past. In figure 7, around
year 2000, the between inequality estimated by Sala-i-Martin (2006), on the basis of
1996 international dollars, is higher than total inequality estimated in PS at 2000
international dollars. Levels are thus important in this kind of analysis, but the
analysis of trends may give relatively more reliable outcomes, given that revisions of
data are frequent. For example, with regard to between inequality, our estimates shows
lower values compared with Sala-i-Martin (2006), but again the profile until 2000 is
almost identical. The different number of countries covered (138 countries in Sala-iMartin, 2006) does not significantly affect the levels of between-country inequality.19
In the bottom graph of figure 7, the comparison is made among our estimates (the
continuous lines) and the available scattered points in other studies (Firebaugh, 1999;
Korzeniewicz and Moran, 1997; Dikhanov and Ward, 2001; Berry et al., 1983;
Chotikapanich et al., 1997). Given the small number of points, it is hard to identify any
reasonable profile; yet, it seems that the only divergent series, compared to our
estimates, is that reported by Korzeniewicz and Moran (1997), which is however
divergent also with respect to other studies.
Unfortunately, no comparable series of within inequality are available that may
validate our estimates of both the standard within inequality and the overlapping
factor. Two point estimates of within inequality are available in Milanovic (2002),
where within inequality is calculated at 0.013 both in 1988 and in 1993, against our
Data are not reported in the graph, but the calculation repeated with the same countries as in Sala-i-Martin
(2006) does not alter the results.
19
32
almost identical 0.014 in 1988 and 0.013 in 1993. It is worth noting that this similarity
is however achieved with a different number of countries (91 in Milanovic, 2002; 164
in 1988 and 186 in 1993 in the present study). This can be explained by the fact that
despite the number of missing countries in Milanovic (2002), they may represent a
smaller share of both total population and income, which means that their combined
weight and contribution to within inequality is small.
5.2. Sensitivity of results to different values of Gini coefficients
A second important sensitivity test concerns the WIDER Gini coefficients used as
anchors. A natural question to ask is whether a change of the anchors may change the
measurement of within inequality. This issue has been already partially addressed
above, where within inequality was measured once by the series of Gini coefficients
estimated on the basis of the anchors, and a second time by the full series of simulated
Gini coefficients estimated by (1), without any significant change of the inequality
profile. Here, we investigate whether a change of the dataset on Gini coefficients may
significantly affect the results. To this regard, some alternatives can be explored. First,
the Texas Inequality Project Database (TIPD) is taken, where the Gini coefficients
refer to gross household income inequality computed from a regression analysis
between DS inequality measures and the UTIP-UNIDO pay inequality measures,
controlling for the source characteristics in DS and for the share of manufacturing in
total employment. TIPD uses international datasets for global comparisons (e.g.,
UNIDO’s Industrial Statistics) as well as regional and national datasets for Europe,
Russia, China, India, and the US, and provides 3200 country-year pairs, more dense
and consistent that in the DS dataset, and reported to be homogeneous for Europe,
North America and South America, even though highly heterogeneous for Asia.
Available TIPD Gini coefficients, according to the methodology proposed in section
2, have been included in the fixed effect estimation of equation (1) and the full analysis
replicated. Missing Gini coefficients, as before, are filled with the predictions of the
regression. Figure 8 reports the path of the estimated within inequality with TIPD,
compared with the previous estimation with WIDER and the set of fully simulated
Gini coefficients. Even though at a different scale, within inequality with TIPD Gini
33
coefficients follows the same increasing trend, especially in the last decade. The fact
that TIPD Gini coefficients are based on global pay, which is expected to be just one of
the many possible income sources, may partially explain some divergence of this
estimation compared with the use of the WIDER dataset. In any case, the increasing
trend does not depend on the specific source of data, and this confirms the robustness
of our procedure, at least to track the evolution of inequality over time.
As a further robustness test, the SWIID database by Solt (2009) has been taken as a
reference for Gini coefficients, and the results drawn in figure 9 by world’s regions for
the period 1975-2005 (the time span where full comparability is assured).20 The
comparison between figures 4 and 9 shows negligible differences in levels and no
differences in trend. Note that when using the SWIID database, no imputation of
missing Gini coefficients is done; thus, the estimation of within inequality is not
affected by the possible weaknesses of the regression analysis. Yet, results are fully
comparable with those obtained by our methodology.
5.3. Income shares by quantiles
As a further sensitivity analysis of our estimation, one can take the comparison
between the decile shares as reported in WIDER and the decile shares resulting from
our estimated world income distribution. Unfortunately, decile shares are not
frequently available; among the most recent years, a meaningful comparison can be
made for year 2000, yet with a limited number of countries (56). Figure 10 reports, for
each available country, the graph of the difference between the income shares actually
imputed to each decile and the income shares estimated with our procedure. A positive
bar means that WIDER income shares are greater than those resulting from our
estimates; the opposite is true with a negative bar. As can be easily seen, all diagrams
report very small differences, if one makes exception for specific countries at the top of
the income distribution. To some extent, this is an expected outcome, as the lognormal
distribution fits the actual income distribution more satisfactorily at lower and central
levels, while it is less precise at the upper tails (Majumder and Chakravarty, 1990).
After 2005, the Gini coefficient of China is not available in SWIID, which makes meaningless any investigation of
within inequality.
20
34
Provided that there is no certainty that the WIDER income shares give the correct
information, Chile (CHL), Colombia (COL), Mexico (MEX) and South Africa (ZAF)
are cases where our world income distribution underestimates the income share of the
top decile; Romania (ROM), Austria (AUT), Hungary (HUN) and Ireland (IRL) are
instead cases of overestimation, which means that the error is not systematic. In all
cases, these countries represent
a small share of the world population; thus,
discrepancies at the top decile are not likely to significantly affect the general outcome.
Furthermore, the fact that all countries’ distribution are in this paper estimated with
the same homogeneous method makes the results overall more stable, compared with
the case where income shares are captured by national surveys or estimated.
Support to the quality but also caveats for our assumption also arise from the
comparisons reported in table 9. For example, our simulated world income distribution
fits pretty well with the results obtained by Milanovic (2010) for 2005 using national
household surveys. The only notable discrepancy is the different distribution of income
shares between the eightieth and the tenth decile. The underestimation of this latter
decile is partially spread over the lower deciles, with the effect of possibly
underestimate total inequality. However, our estimation of the income share of the top
ventile is not far from Milanovic’s one, and the aggregate income share of the top
three decile is underestimated by just 2 percentage points (82.1 instead of 84.1).
Comparisons with other studies are also described in table 9. Discrepancies among
these studies and our estimates are not huge, especially with more recent studies
(Morrisson and Murtin, 2011; Ortiz and Cummins, 2011) and surely not greater than
discrepancies that sometimes arise among different studies in the same year, as in
Chotikapanich et al. (1997) and Korzeniewicz and Moran (1997), for both 1980 and
1990. To some extent, discrepancies arise because methods are different, the quality of
data is imperfect in all cases, and it is not always clear whether the distribution by
quantiles is obtained by referring to the PPP converted per capita GDP or to a
measure of GDP valued at market exchange rates.
Beyond these discrepancies, all data in table 9 converge to show an impressive
invariance of the world income distribution, where about 75 per cent of income is in
the hands of 30 per cent of the population and more than 50 per cent of world income
is in the hands of the top 10 per cent, forty years ago as well as in more recent times,
35
despite the recent and rapid growth of some Asian countries and the alleged benefits of
the globalisation wave.
5.4. Purchasing power parity converted GDP and GDP at current prices
It is well known that the use of PPP converted per capita GDP with the GearyKhamis (GK) method introduces a systematic substitution bias, as it values the GDP of
each country at average international prices by ignoring local consumers’ ability to
shift towards local cheaper goods (Dowrick and Akmal, 2005). Since price and
quantities structures may be very different across countries – especially in the poorest
countries – the use of PPP converted GDP (PGDP) may underestimate inequality
when prices become more dissimilar across countries. On the other hand, the most
used alternative based on per capita GDP at current prices and market exchange rates
(CGDP), may understate the real incomes of the poorest economies. Indeed, market
exchange rates are more likely to equate prices in the traded sector, introducing a nontraded sector bias. Since this latter sector is likely to be more important in less
developed countries, the use of current prices undervalues the domestic purchasing
power of poor countries, which overstates inequality. Measuring between inequality
with PGDP and with CGDP may thus lead to opposite conclusions. By hypothesis, if
the price structure across countries would become more dissimilar, PGDP would show
lower inequality, while CGDP would show higher inequality (Dowrick and Akmal,
2005; 204), which may lead to contradictory outcomes. The issue of what series should
be used is basically unresolved, but there is the impression that the overestimation of
inequality using CGDP can be much stronger than the underestimation of inequality
implied by PGDP, which is an argument for sticking to PGPD data (e.g., Melchior,
2001).
For the sake of completeness, figure 11 reports the two series of betweeninequality. The PGDP series is that already used in section 1. The fact that CGDP lies
over PGDP in the whole period and slightly diverges from the beginning of the
Eighties until around 2000, could be a signal that, in that period, the price structure
has become more dissimilar. This divergence confirms the intuition by Dowrick and
Akmal (2000) that the use of different series may make inequality to appear increasing
36
or declining, giving contradictory outcomes. It is also consistent with the findings of
Melchior et al. (2000), where the unadjusted series shows a greater level of inequality
than the PGDP series, with diverging trend from 1980 to around 1993 (see also
Korzeniewicz and Moran, 1997). In our calculations, this effect seems to disappear
after 2000, supporting the conjecture that the similarity (or dissimilarity) in the price
structure is not fundamental to shape inequality trends in the last decade. Again, in the
last decade, levels may differ, but – according to the series used – the trend is common.
5.5. Population structure
The declining trend of inequality in the last decade so far observed may actually be
caused either by a realignment of per capita GDP or by a faster population growth in
richer countries (compared with the poorer countries) or by both factors. Thus, a
natural question to ask is whether the population growth from 1970 to 2009 may have
to some extent affected our estimates. Figure 12 reports the estimated between
inequality with actual population (the continuous line), the estimated between
inequality by assuming that all countries have in all years the population of 1970 (the
line with black circles) and the estimated between inequality by assuming that all
countries have in all years the population of 2009 (the line with white triangles). The
outcome suggests to exclude that a different distribution of the population among
countries may have significantly affected the path of between inequality. In particular,
the close behaviour of all lines reveals that the changed population distribution
between richer and poorer countries did not affect inequality (see also Milanovic,
2005), which means that the declining trend of inequality can be almost fully imputed
to the realignment of PPP converted per capita GDP.
5.6. Rural and urban China
Last but not least, it is worth addressing the role of China in shaping world inequality.
As the biggest country, China has wide differences among rural and urban areas, in
terms of both mean income and income growth. This issue is not without potential
37
consequences for both between and within inequality. On the one hand, if rural China
has a lower mean income than the country average, this can contribute to a greater
between world inequality. On the other hand, if the income distribution of each area is
relatively more homogeneous, their Gini coefficients should be lower than the Gini
coefficient of the overall country, which means that measured within inequality could
be lower.
In order to take into account this problem, according to our methodology, one must
have information on the distribution of population among areas as well as on the size
of the corresponding mean incomes. This is enough to recalculate between inequality.
With regard to within inequality, Gini coefficients of rural and urban areas are also
required. Table 10 reports the essential data for this analysis, which now spans from
1981 to 2009, and are mostly based on the series of mean incomes reported in Chen
and Ravallion (2007) until 2001, on Chow (2006) for 2002 and 2003, on Yang et al.
(2010) for 2004 and 2005 and on other official data from 2006 to 2009, on the series of
rural and urban Gini coefficients as calculated in Chen et al. (2010), and on
extrapolations of Gini coefficients from 2007 to 2009. With this information, the
previous procedure can be replicated assuming that urban China and rural China are
two different countries, in order to obtain two separate distributions. As a test for the
correspondence of these distributions with other empirical evidence, the far right panel
of table 10 reports data on the cumulative percentages of income in both urban and
rural areas available in Gustafsson et al. (2008) for 2002 – columns (9) and (11) – and
according to our estimates in the same year – columns (10) and (12). The similarities
between available data and our estimates are striking, which is further support to the
power of the methodology.
Figure 13 reports the outcome of the analysis for both between and within
inequality. In the top graph, as expected, measured between inequality is slightly
higher when China is split in two parts, and the distance with respect to between
inequality when China is a single country is more pronounced in recent times, which is
mostly due to the enlarged gap between rural and urban mean incomes. Indeed, rural
China includes a mass of people that is poorer than the average country; urban China,
instead, includes a mass of people whose income grow even faster than that of
relatively poor countries. Overall, however, the differences are not dramatic and the
trend of inequality is basically the same.
38
In the bottom graph of figure 13, one can instead note that by splitting rural and
urban China, within inequality maintains its increasing trend, but the level is
significantly lower. This is easily explained by the fact that both urban and rural China
have Gini coefficients that are below the Gini coefficient of the whole country. Thus,
when weighted by population and income shares, their contribution to within
inequality declines. In any case, it is worth remarking again that even after controlling
for rural and urban China, between inequality declines, while within inequality
increases.
6. Conclusions
Our analysis shows some insightful facts. First, between inequality is experimenting
an unprecedented decline in the last decade. Even though the bulk of this decline is due
to the performance of China and other Asian countries, we have shown that a (weaker)
declining trend survives even when these countries are excluded from the analysis.
Second, within inequality, in the last decade, is increasing and almost all of its growing
relevance is led by the increased level of inequality within China. As a consequence,
total inequality receives two contrasting inputs. On the one hand, the powerful
convergence force associated to the reduction of the Chinese gap with richer countries;
on the other hand, the powerful divergence force associated both to the enlargement of
the Chinese gap with poorer countries and to the greater weight of the Chinese within
inequality. Third, apart from this fundamental role played by China, the world
distribution of income forty years ago does not appear fundamentally changed in most
recent times. African countries had and still have nothing in common with advanced
economies. In 1970 as well as in 2009, they form almost a perfect stratum with respect
to the distribution of income of the richest part of the world and most African people
would rank in the lowest decile of the distribution of the richest world now as well as
forty years ago. Fourth, the fact that total inequality has slightly declined is not
necessarily an indicator that world resources are better shared. In 2009, the Gini
coefficient of world income is still measured around 0.65, which is a level of inequality
that would probably be intolerable in any single country. Despite the rapid growth of
few Asian countries (that incidentally are the most populous countries in the world),
39
the rest of the world is not significantly moving; South American and South Central
Asian regions are more or less on the same relative position they occupied forty years
ago; African regions, in relative terms, are in some cases worse now than before; while
Europe, North America and Oceania perpetuate their ‘splendid isolation’. Since from
1970 to 2009 a significant and persistent globalisation wave has characterised the
functioning of the economic systems, one may cast some doubts about how the
potential benefits of this structural change have spread across countries.
40
References
Aitchison J., Brown J.A.C. (1957), The Lognormal Distribution, Cambridge University
Press, Cambridge.
Anand S., Kanbur R. (1993), “Inequality and Development: A Critique”, Journal of
Development Economics, 41, 19-43.
Atkinson A.B., Brandolini A. (2001), “Promise and Pitfalls in the Use of ‘Secondary’
Data-Sets: Income Inequality in OECD countries as a Case Study”, Journal of
Economic Literature, XXXIX, 771-800.
Berry A., Bourguignon F., Morrison C. (1983), “Changes in the World Distribution of
Income between 1950 and 1977”, Economic Journal, 93, 331-350.
Berry A., Bourguignon F., Morrison C. (1989), "The World Distribution of Income:
Evolution over the Recent Period and Effects of Population Growth," DELTA
Working Papers 89-13, DELTA.
Bourguignon F., Morrisson C. (2002), “Inequality Among World Citizens: 1820-1992”,
American Economic Review, XCII, 727-744.
Chen J., Dai D., Pu M., Hou W., Feng Q. (2010), The Trend of Gini coefficient in China,
BWPI Working Paper 109, Brooks World Poverty Institute, Manchester.
Chen S., Datt G., Ravallion M. (1994), “Is Poverty Increasing in the Developing
World?”, Review of Income and Wealth, 40, 359-376.
Chen S., Ravallion M. (2007), “China’s (Uneven) Progress Against Poverty”, Journal of
Development Economics, 82, 1-42.
Chotikapanich D., Valenzuela R., Prasada Rao D.S. (1997), “Global and Regional
Inequality in the Distribution of Income: Estimation with Limited and Incomplete
Data”, Empirical Economics, 22, 533-546.
Chow G. (2006), Rural Poverty in China – Problem and Policy, manuscript.
Deininger K., Squire L. (1996), “A New Data Set Measuring Income Inequality”, World
Bank Economic Review, X, 565-591.
Deutsch J., Silber J. (2001), The Kuznets Curve and the Impact of Various Income Sources
on the Link Between Inequality and Development, WP 2001-03, Department of
Economics, Bar-Ilan University.
Dikhanov Y., Ward M. (2002), Evolution of the Global Distribution of Income in 1970-99,
International Workshop on Income Distribution and Welfare, Milan.
Dowrick S., Akmal M. (2005), “Contradictory Trends In Global Income Inequality: A
Tale Of Two Biases”, Review of Income and Wealth, 51, 201-229.
Firebaugh G. (1999), “Empirics of World Income Inequality”, American Journal of
Sociology, 104, 1597-1630.
Frick J.R., Goebel J., Schechtman E., Wagner G.G., Yitzhaki S. (2006), “Using
Analysis of Gini (ANOGI) for Detecting Whether Two Subsamples Represent the
Same Universe”, Sociological Methods & Research, 34, 427-468.
Galbraith J.K., Kum H. (2004), Estimating the Inequality of Household Incomes: A
Statistical Approach to the Creation of a Dense and Consistent Global Data Set, UTIP
Working Paper No. 22.
Grosh M.E., Nafziger E.W. (1986), “The Computation of World Income Distribution”,
Economic Development and Cultural Change, 35, 347-359.
Gustafsson B., Shi L., Sicular T., Ximing Y. (2008), “Income Inequality and Spatial
Differences in China, 1988, 1995, and 2002”, in Gustafsson B., Shi L., Sicular T.
(eds.), Inequality and Public Policy in China, Cambridge University Press, New York,
pp. 35-60.
41
Heshmati A. (2006), “The World Distribution of Income and Income Inequality: A
Review of the Economics Literature”, Journal of World-Systems Research, XII, 61107.
Korzeniewicz R.P., Moran T.P. (1997), “World-Economic Trends in the Distribution
of Income, 1965-1992”, American Journal of Sociology, 102, 1000-1039.
Kuznets S. (1955), “Economic Growth and Income Inequality”, American Economic
Review, 45, 1-28.
Li H., Squire L., Zou H. (1998), “Explaining International and Intertemporal Variation
in Income Inequality”, Economic Journal, 108, 26-43.
Majumder A., Chakravarty S.R. (1990), "Distribution of Personal Income:
Development of a New Model and Its Application to U.S. Income Data", Journal of
Applied Econometrics, 5, 189-96.
Melchior A., (2001), “Global Income Inequality”, World Economics, 2, 87-108.
Melchior A., Telle K., Wiig H. (2000), Globalisation and Income Inequality – World
Income Distribution and Living Standards, 1960-1998, Royal Norwegian Ministry of
Foreign Affairs Studies on Foreign Policy Issues, Report 6B.
Milanovic B. (2002), “True World Income Distribution, 1988 and 1993: First
Calculations Based on Household Surveys Alone”, Economic Journal, 112, 51-92.
Milanovic B. (2010), “Global Inequality Recalculated and Updated: the Effect of New
PPP Estimates on Global Inequality and 2005 Estimates”, Journal of Economic
Inequality, 10, 1-18.
Milanovic B., Yitzhaki S. (2002), “Decomposing World Income Distribution: Does the
World Have a Middle Class?”, Review of Income and Wealth, 2, 155-178.
Morrisson C., Murtin F. (2011), Internal Income Inequality and Global Inequality, WP 26,
Fondation pour les Etudes et Recherches sur le Developpement International.
Ortiz I., Cummins M. (2011), Global Inequality: Beyond the Bottom Billion, UNICEF
Social and Economic Policy Working Paper, April.
Pinkovskiy M., Sala-i-Martin X. (2009), Parametric Estimations of the World Distribution
of Income, NBER Working Paper 15433.
Podder N. (1993), “A Profile of International Inequality”, Journal of Income Distribution,
3, 300-314.
Pyatt G. (1976), “On the Interpretation and Disaggregation of Gini Coefficient”,
Economic Journal, 86, 243-255.
Quah D. (2000), Cross-Country Growth Comparison: Theory to Empirics, CEPR
Discussion Paper No. 442, London.
Quah D. (2002), “One Third of the World’s Growth and Inequality”, CEPR Discussion
Paper, No. 3316, CEPR, England.
Ravallion M., Chen S. (1997), “What Can New Survey Data Tell Us About Recent
Changes in Distribution and Poverty?”, World Bank Economic Review, 11, 357-382.
Ravallion M., Datt G., van der Walle D. (1991), “Quantifying Absolute Poverty in the
Developing World”, Review of Income and Wealth, 37, 345-361.
Sala-i-Martin X. (2006), “The World Distribution of Income: Falling Poverty and …
Convergence, Period”, The Quarterly Journal of Economics, CXXI, 351-397.
Schultz T.P. (1998), “Inequality and the Distribution of Personal Income in the World:
How Is It Changing and Why”, Journal of Population Economics, XI, 307-344.
Solt, Frederick (2009), “Standardizing the World Income Inequality Database”, Social
Science Quarterly, 90, 231-242.
Sprout R.V.A., Weaver J.H. (1992), “International Distribution of Income: 1960-1987”,
Kyklos, 45, 237-258.
42
Summers R., Heston A. (1988), “A New Set of International Comparisons of Real
Product and Price Levels: Estimates for 130 Countries, 1950-1985”, Review of
Income and Wealth, 1, 1-26.
Summers R., Heston A. (1991), “Penn World Table (Mark 5): An Expanded Set of
International Comparisons, 1950-1988”, Quarterly Journal of Economics, 106, 327368.
Theil H. (1979), “World Income Inequality”, Economics Letters, 2, 99-102.
Theil H. (1996), Studies in Global Econometrics, Kluwer Academic Publishers,
Amsterdam.
Theil H., Seale J.L. (1994), “The Geographic Distribution of World Income, 1950-90”,
De Economist, 4, 387-419.
Van Zanden J.L., Baten J., Földvari P., van Leeuwen B. (2011), The Changing Shape of
Global Inequality 1820-2000: Exploring a New Dataset, CGEH Working Paper Series
No.1, Universiteit Utrecht.
Yang J., Yan Y., He B. (2010), Growing Unequally? – A Study on the Relationship Between
Income Inequality and Economic Growth in China, School of Sustainable Development
of Society and Technology.
Yitzhaki S. (1994), “Economic Distance and Overlapping of Distributions”, Journal of
Econometrics, 61, 147-159.
Yitzhaki S., Lerman R. (1991), “Income Stratification and Income Inequality”, Review
of Income and Wealth, 37, 313-329.
43
Table 1 – Normalised gaps and contributions to Gini coefficients
(first panel) – PPP per capita GDP in US dollars
Source: Author’s calculation on PWT 7.0 data
44
Table 1 – Normalised gaps and contributions to Gini coefficients
(second panel)
Source: Author’s calculation on PWT 7.0 data
45
Table 2 – Definitions used to calculate Gini coefficients in WIDER (by geographical areas)
Source: Author’s calculation on the WIID2 dataset
46
Table 3 – Gini coefficients and per capita incomes, fixed effects estimation
Source: Author’s calculation on the WIID2 dataset
47
Table 4 – ANOGI decomposition, by countries
Source: Author’s calculations on the basis of Gini coefficients and PPP converted per capita GDP
48
Table 5 – ANOGI decomposition, by regions
49
Table 5 (cntd)
Source: Author’s calculations on the basis of Gini coefficients and PPP converted per capita GDP
50
Table 6 – The overlapping matrix, by regions
Source: Author’s calculations
51
Table 7 – The ranking matrix, by regions
Source: Author’s calculations
52
Table 8 – Income shares and frequencies at the bottom and at the top decile
=============================
Source: Author’s calculations
53
Table 9 – Income shares: comparisons with other studies
Source: Author’s calculations
54
Table 10 – Data for urban and rural China
55
Figure 1 – Gini coefficient of PPP converted per capita GDP (1970-2009, population weighted)
Source: Author’s calculation on PWT 7.0 data
56
Figure 2 – Gini coefficient of PPP converted per capita GDP (1970-2009, population weighted)
Source: Author’s calculation on PWT 7.0 data
57
Figure 3 – PPP converted per capita GDP normalised on world average
Source: Author’s calculation on PWT 7.0 data
58
Figure 4 – Within inequality, total and by continents
Source: Author’s calculation on WIID2 dataset
59
Figure 5 – Total, between and within inequality, normalised on their means
Source: Author’s calculations
60
Figure 6 – The concentration of world income
Source: Author’s calculations
61
Figure 7 – Total and between inequality: comparisons with other studies
Source: Author’s calculations
62
Figure 8 – Within inequality with Gini coefficients from alternative sources
Source: Author’s calculations
63
Figure 9 – Within inequality with Gini coefficients from SWIID, by world’s regions
Source: Author’s calculations
64
Figure 10 – Differences in income shares, by deciles, year 2000
Source: Author’s calculations
65
Figure 11 – Between inequality with CGDP and PGDP series
Source: Author’s calculations
66
Figure 12 – Between inequality and the population structure
Source: Author’s calculations
67
Figure 13 – The impact of rural and urban China on between and within inequality
Source: Author’s calculations
68