Other Summary Measures of Poverty and Inequality

Sen Index
Sen (1976) proposed an index that seeks to combine the effects of the number of poor, the
depth of their poverty, and the distribution of poverty within the group. The index is given
by:
 UP

PS  P 1  1  G P 
Z 

0

where P0 is the headcount index, μP is the mean income (or expenditure) of the poor, and GP

is the Gini coefficient of inequality among the poor. The Gini coefficient ranges from 0

(perfect equality) to 1 (perfect inequality). The Sen Index can also be written as the average
of the headcount and poverty gap measures, weighted by the Gini coefficient of the poor,
giving:

PS  P0 G P  P1 1  G P 
It can be shown (Xu and Osberg 2002) that the Sen Index may also be written as:
PS  P0 P1P  1  G PP 

where GPP is the Gini coefficient of the poverty gap ratios of only the poor and is the poverty
gap index calculated over poor individuals only.
1. The Sen Index has the virtue of taking the income distribution among the poor into
account.
2. However, the index is almost never used outside of the academic literature, since
because it lacks the intuitive appeal of some of the simpler measures of poverty,
3. It “cannot be used to decompose poverty into contributions from different subgroups”
(Deaton 1997, 147).

The Sen-Shorrocks-Thon Index

The Sen Index has been modified by others, and one of the more attractive versions is the
Sen-Shorrocks-Thon (SST) index, defined as:
 

PSST  P0 P1P 
1  G
P


 

which is the product of the headcount index, the poverty gap index (applied to the poor only),
and the Gini coefficient of the poverty gap ratios (that is, of the Gn’s) for the whole

1
population. This Gini coefficient is close to 1, indicating great inequality in the incidence of
poverty gaps.

Example: In 1996, 12.4 percent of the population of Quebec province (Canada) was in
poverty. The poverty gap index, applied to the poor only, stood at 0.272. And the Gini
coefficient of the poverty gap ratios was 0.924.
Thus the SST index was 0.065 = (0.124 × 0.272 × (1 + 0.924)).
1. One of the strengths of the SST index is that it can help give a good sense of the
sources of change in poverty over time. This is because the index can be decomposed
into
LnPSST  LnP0  LnP1P  Ln 1  G P  
which may be interpreted as, percentage change in SST index = percentage change in
headcount index + percentage change in poverty gap index (among poor) + percentage
change in (1 + Gini coefficient of poverty gaps). This allows us to decompose poverty into
three aspects: Are there more poor people? Are the poor poorer? And is there higher
inequality among the poor?

The Watts Index

The first distribution-sensitive poverty measure was proposed in 1968 by Watts, and takes the
form:
1 q
1 q
Z
W 
N
  Ln Z   Ln Yi    N
 Ln Y 
i 1 i 1  i

where the N individuals in the population are indexed in ascending order of income (or

expenditure), and the sum is taken over the q individuals whose income (or expenditure) yi

falls below the poverty line z.

1. Although the Watts Index is not a particularly intuitive measure, it is increasingly
used by researchers because it satisfies all the theoretical properties that one would
want in a poverty index.
2. Ravallion and Chen (2001) argue that three axioms are essential to any good measure
of poverty. Under the focus axiom, the measure should not vary if the income of the
non-poor varies; under the monotonicity axiom, any income gain for the poor should
reduce poverty; and under the transfer axiom, inequality-reducing transfers among

2
 the poor should reduce poverty. The Watts index satisfies these three axioms, but the
headcount (P0) and poverty severity (P1) measures do not.

Other Measures
There are other additive poverty measures that are distribution-sensitive. Following Atkinson
(1987), one can characterize a general class of additive measures, encompassing the Watts
index, the FGT class of measures, and some other measures (such as the second measure
proposed by Clark, Hemming, and Ulph [1981]), as taking the following form:
N
1
P
N
 P Z , Y 
i 1
i

where p(z, yi) is the individual poverty measure, taking the value zero for the non poor (yi >

z) and some positive number for the poor, the value of which is a function of both the poverty
line and the individual living standard, non decreasing in the former and non increasing in the
latter.

Inequality Measures
Inequality is a broader concept than poverty in that it is defined over the entire population,
not just for the portion of the population below a certain poverty line. Most inequality
measures do not depend on the mean of the distribution; this property of mean independence
is considered to be a desirable feature of an inequality measure.
Inequality measures are calculated for distributions other than expenditure—for instance, for
income, land, assets, tax payments, and many other continuous and cardinal variables.
The simplest way to measure inequality is by dividing the population into fifths (quintiles)
from poorest to richest, and reporting the levels or proportions of income (or expenditure)
that accrue to each level.
Quintile information is easy to understand, although sometimes a summary measure is
needed rather than a whole table of figures.

Commonly Used Summary Measures of Inequality

Several summary measures of inequality have been developed over the years.

Decile Dispersion Ratio

3
A simple and popular measure of inequality is the decile dispersion ratio, which presents the
ratio of the average consumption (or income) of the richest 10 percent of the population to the
average consumption (or income) of the poorest 10 percent.
This ratio can also be calculated for other percentiles (for instance, dividing the average
consumption of the richest 5 percent, the 95th percentile, by that of the poorest 5 percent, the
5th percentile).
The decile dispersion ratio is readily interpretable, by expressing the income of the top 10
percent (the “rich”) as a multiple of that of those in the poorest decile (the “poor”). However,
it ignores information about incomes in the middle of the income distribution, and does not
even use information about the distribution of income within the top and bottom deciles.

Gini Coefficient of Inequality

The most widely used single measure of inequality is the Gini coefficient. It is based on the
Lorenz curve, a cumulative frequency curve that compares the distribution of a specific
variable (for example, income) with the uniform distribution that represents equality. To
construct the Gini coefficient, graph the cumulative percentage of households (from poor to
rich) on the horizontal axis and the cumulative percentage of expenditure (or income) on the
vertical axis. The Lorenz curve is shown in the figure below.

Lorenz Curve
Cumulative % 100
of Expenditure

0 Cumulative % of Population 100

The diagonal line represents perfect equality. The Gini coefficient is defined as A/(A + B),
where A and B are the areas shown in the figure. If A = 0, the Gini coefficient becomes 0,
which means perfect equality, whereas if B = 0, the Gini coefficient becomes 1, which means

4
complete inequality. In this example, the Gini coefficient is about 0.35. Some users, including
the World Bank, multiply this number by 100, in which case it would be reported as 35.

Formally, let xi be a point on the x-axis, and yi a point on the y-axis. Then
N
Gini  1    X i  X i 1  Yi  Yi 1 
i 1

When there are N equal intervals on the x-axis. The above equation can be written as:
N
1
Gini  1 
N
 Y
i 1
i  Yi 1 

The Gini coefficient is not entirely satisfactory. To see this, consider the criteria that make a
good measure of income inequality:
 Mean independence. If all incomes were doubled, the measure would not change.
The Gini satisfies this.
 Population size independence. If the population were to change, the measure of
inequality should not change, all else equal. The Gini satisfies this, too.
 Symmetry. If any two people swap incomes, there should be no change in the measure
of inequality. The Gini satisfies this.
 Pigou-Dalton Transfer sensitivity. Under this criterion, the transfer of income from
rich to poor reduces measured inequality. The Gini satisfies this, too.
It is also desirable to have
 Decomposability. Inequality may be broken down by population groups or income
sources or in other dimensions. The Gini index is not easily decomposable or additive
across groups. That is, the total Gini of society is not equal to the sum of the Gini
coefficients of its subgroups.
 Statistical testability. One should be able to test for the significance of changes in the
index over time. This is less of a problem than it used to be because confidence
intervals can typically be generated using bootstrap techniques.

Generalized Entropy Measures

There are a number of measures of inequality that satisfy all six criteria. Among the most
widely used are the Theil indexes and the mean log deviation measure. Both belong to the
family of generalized entropy (GE) inequality measures. The general formula is given by:

5
 1 1 N
 Yi 


GE ( )  
 (  1)  N
 
i 1  Y 
  1


where y– is the mean income per person (or expenditure per capita). The values of GE
measures vary between zero and infinity, with zero representing an equal distribution and
higher values representing higher levels of inequality. The parameter α in the GE class
represents the weight given to distances between incomes at different parts of the income
distribution, and can take any real value. For lower values of α, GE is more sensitive to
changes in the lower tail of the distribution, and for higher values GE is more sensitive to
changes that affect the upper tail. The most common values of α used are 0, 1, and 2.

GE (1) is Theil’s T index, which may be written as:

1 N
Yi  Yi 
GE (1) 
N
 Y Ln Y 
i 1

GE (0), also known as Theil’s L, and sometimes referred to as the mean log deviation
measure, is given by: ¯

N  ¯ 
1 Y 
GE (0) 
N

i 1
Ln
 Yi 
 

Atkinson’s Inequality Measures

Atkinson (1970) has proposed another class of inequality measures that are used from time to
time. This class also has a weighting parameter ε (which measures aversion to inequality).
The Atkinson class, is defined as:
1 / 1
1 N
 Yi 
1

A  1        1
 N i 1  Y  

 Y 
N
(1 / N )
i
=
1 i 1
 1
Y
One caveat is in order: income is more unequally distributed than expenditure. This is a
consequence of household efforts to smooth consumption over time. It follows that when
comparing inequality across countries it is important to compare either Gini coefficients
based on expenditure, or Gini coefficients based on income, but do not mix the two.