a dynamic multidimensional measure of poverty...originating from foster, greer and thorbecke (1984)...
TRANSCRIPT
1
DEPARTMENT OF ECONOMICS
ISSN 1441-5429
DISCUSSION PAPER 25/13
A Dynamic Multidimensional Measure of Poverty
Aaron Nicholas*, Ranjan Ray and Kompal Sinha
Abstract This paper unites two strands of the literature on subgroup decomposable poverty measurement
originating from Foster, Greer and Thorbecke (1984) by incorporating information on both multiple
dimensions and multiple periods. This generalises the Alkire and Foster (2011a) measure into a
dynamic setting. In doing so, it introduces two variants of the ‘transfer’ axiom: one that gives
increasing weight to individuals whose deprivations are concentrated as repeated dimensions in a
specific period (what we term ‘breadth’) versus one that gives increasing weight to individuals
whose deprivations are concentrated as repeated periods in a specific dimension (‘depth’). The
measure is able to differentiate between both aspects of poverty as well as quantify the relative
contribution of each aspect towards overall poverty. This makes it well suited to make comparison
across subgroups when individual longitudinal data is available as well as allowing policy-makers
to quickly identify if the breadth or depth aspect requires more attention since the policy
prescriptions for each may differ. We apply the proposed measure to longitudinal data from China
and Indonesia where we find that the depth aspect of poverty is a large contributor to individual
level poverty. The existing static framework of multidimensional poverty in the literature may
therefore be inappropriate for subgroup comparisons when information over time is available.
Keywords: Multidimensional Poverty and Deprivation; Duration and Dynamics of Poverty;
Dimensional and Dynamic Transfer Axioms; Subgroup Decomposability.
JEL classification: I31, I32
*corresponding author: [email protected] Graduate School of Business Deakin University, Victoria 3170,
Australia
Ranjan Ray Department of Economics Monash University, Victoria 3800 Australia
Kompal Sinha Department of Econometrics & Business Statistics Monash University, Victoria 3800 Australia
© 2013 Aaron Nicholas, Ranjan Ray and Kompal Sinha
All rights reserved. No part of this paper may be reproduced in any form, or stored in a retrieval system, without the prior written
permission of the author.
3
1. Introduction
Traditional measures of poverty based on income and expenditure have been extended along two
major directions: the broadening of the measures to incorporate a wider set of dimensions that
together give a more accurate representation of welfare; and the deepening of the measures to
incorporate information that spans over several periods of observations.
Extensions along the first direction, largely influenced by the writings of Sen (1985), move away
from unidimensional measures and into a multidimensional approach based on the individual‟s lack
of access to a wide set of dimensions that include both market and nonmarket goods. Sen (1976)‟s
pioneering contribution introduced the axiomatic approach to the measurement of poverty and
provided the basis for the recent axiomatic approach to the multidimensional measurement of
poverty – examples include Tsui (2002), Bourguignon and Chakravarty (2003), and Alkire and
Foster (2011a) [henceforth AF].1
Extensions along the second direction move away from static measures into a dynamic approach
where repeated observations (or „spells‟) of poverty are treated differently to cases where poverty is
temporary. Examples of such extensions include Foster (2007), Calvo and Dercon (2009), Hojman
and Kast (2009), Duclos et.al. (2010), Hoy and Zheng (2011), Bossert et. al. (2012) and Gradin et
al.(2012).
Despite the usefulness provided by extensions along both directions, the literature has largely
considered both extensions independent of each other, retaining either the unidimensional or static
property of traditional measures. The principal motivation of this paper is to provide a generalised
framework for jointly incorporating both aspects – multidimensionality and dynamics – with a
particular emphasis on the interaction between both aspects and the information gained by
considering them jointly.
Recently, there have been attempts to unite both strands of the literature: Nicholas and Ray (2012)
[NR], and Bossert, Ceriani, Chakravarty and D‟Ambrosio (2012) [BCCD], consider a class of
subgroup decomposable poverty measures based on the „count‟ of an individual‟s deprivations
(Atkinson, 2003). The unique contribution of these papers is to define each deprivation as belonging
to a particular period of time and a particular dimension – an individual‟s poverty score is then
simply a function of the double-sum of these deprivations over time and over dimensions. However,
such a method leads to „path-independence‟ of deprivations, meaning that the measure is invariant
to whether the deprivations were summed first over periods of time or over dimensions. The
1 See, also, Chakravarty and D‟Ambrosio (2006), Bossert et. al. (2007), and Jayaraj and Subramanian (2010) for closely
related work on the measurement of multidimensional deprivation.
4
consequence of this is that the measure is unable to differentiate individuals for whom we observe
deprivations in repeated periods of time for specific dimensions from individuals for whom we
observe deprivations over multiple dimensions for specific periods of time. Nor can any of these
two groups of individuals be differentiated from individuals whose deprivations are scattered across
both dimensions and periods of time.
Consider a simple example, where an individual can be deprived in up to three dimensions, and that
we have observations for three periods of time for each dimension. This means an individual can
experience a maximum of nine deprivations. Consider an individual deprived for all three periods in
the first dimension, versus an individual deprived for one period in the first dimension, one period
in the second dimension and one period in the third dimension. While both individuals share the
same count of deprivations (three), the distribution of the deprivations differ, and indeed, we argue
that the first individual should count as more deprived.
In the spirit of this distinction, this paper proposes a measure that quantifies the contribution, to the
overall poverty measure, of the concentration of deprivation in specific dimensions relative to that
of the concentration of deprivation in specific periods of time. The measure of the relative
magnitude of the two contributors to poverty – introduced in this paper as the „Breadth to Depth
ratio‟ – is a useful concept from a policy viewpoint since it allows policy-makers to identify
whether their efforts should be concentrated on specific dimensions of poverty (e.g. access to
hospitals), or on specific periods of time that have been the main cause of poverty (e.g. large
negative macroeconomic shocks such as the Asian or the Global Financial Crisis).
In incorporating these additional properties, we retain the subgroup decomposability property from
Foster, Greer and Thorbecke (1984) [FGT], the dimensional decomposability property from Alkire
and Foster (2011a) and the dynamic decomposability property from Foster (2007).
We follow the framework established in AF by using a „dual-cutoff‟ approach in our axiomatisation
and application whereby non-poor individuals are ignored by the measure and „poor‟ can be defined
by the intersection, intermediate or union methods (Tsui, 2002). Our generalisation of AF allows us
to include new definitions and perspectives on the „transfer‟ axiom.2
Another significant feature of this study is the application of the proposed dynamic
multidimensional poverty measure to the data sets of developing countries. Both the recent
applications of dynamic multidimensional deprivation measures (NR and BCCD) have been on
panel data sets from developed countries. The absence of any similar previous application in a
2 The transfer axiom originated with Pigou (1912), Dalton (1920), with its modern characterisation in Sen (1976).
5
developing country context is largely due to the absence of information on a panel of households in
developing countries for a sufficiently long time period containing information on a reasonably
wide set of dimensions to make such a study of dynamic multidimensional deprivation possible.3
Such panel data sets are rare even in the context of developed countries and, until recently, almost
non-existent in the case of developing countries.4 The recent availability of such information for
China and Indonesia that we utilize in this study makes this study distinctive.5 The time interval
spanned by the data sets of both countries included the Asian Financial Crisis which affected
Indonesia particularly badly, much more than China.
The rest of the paper is organised as follows. The proposed dynamic multidimensional poverty
measure is introduced in Section 2 along with a discussion of its principal properties. The data sets
are described in Section 3 along with some summary features of the data that are relevant for this
study. The empirical results are presented in Section 4 while Section 5 concludes.
2. Analytical Framework
2.1. Notation
Assume we observe, for N individuals in a population of interest, different dimensions of
deprivation and T equally-spaced periods of time. We say that an individual n is deprived in
dimension j at time t when , where { }, { } { }
is individual n‟s achievement in dimension j at time t, and is a cut-off point that determines
whether or not an individual is considered deprived in a particular dimension at a particular time.
For example, in the dimension „health‟, may be the individual‟s Body Mass Index, in which case
would be some threshold below which the individual would be considered underweight and
therefore deprived in the health dimension.
Each individual can be said to have an individual deprivation profile, which is a matrix
.
/ where {(
)
{ } & { }.
3 As BCCD report, the problem of missing information at the household level on material deprivation in several
individual dimensions is present in the data sets of developed countries as well. This forces one to adopt either the
unsatisfactory practice of treating missing information as the household having access to the dimensions concerned as
done in BCCD, or simply not including such households as done in the present study. 4 There is now increasing availability of such panel data in developing countries. Besides the CHNS data set from China
and the IFLS data set from Indonesia that have been used here, an example of such a data set is that contained in the
household surveys conducted by the Research Centre on Rural Economy (RCRE) in Beijing, and used in the above
cited study by Duclos, et al (2010). 5 Mishra and Ray (2012) have recently compared multidimensional deprivation in the static framework between China
and India, but their study was not on panel data and, consequently, did not incorporate any of the dynamic elements of
the present study.
6
is a sensitivity parameter along the lines of the poverty measure due to FGT. Call
deprivation inputs. When observed achievement levels are discrete or ordinal, it is common to
restrict such that { }.
The population deprivation profile is a vector ,.... . Define the identification vector
where takes the value 1 if the individual is considered poor, and 0 otherwise.
An individual is considered poor if he has at least different deprivations; this can be based on a
minimum number of periods, or dimensions, or a combination of both. A union method of
identification would set while the intersection method would use . The poverty
index is a function .
2.2. The breadth versus depth of deprivation
Unlike the simpler unidimensional case in AF, each deprivation input in the dynamic
multidimensional case contributes to both the „breadth‟ of deprivation (number of dimensions
deprived at a particular time t) as well as the „depth‟ of deprivation (number of periods deprived in a
particular dimension j). In applications such as NR and BCCD, an individual‟s deprivation score is
simply the sum of deprivation inputs regardless of which dimension or period they belong to.
Therefore, the poverty index is insensitive to permutations of the individual‟s deprivation profile.
Consider the example below of three individuals with different deprivation profiles with
and .
{
} {
} {
}
A measure that takes as its inputs only the double sum of the deprivation inputs would conclude that
all three individuals are equally deprived since the „count‟ of deprivations would be six in each.
Path-independence with respect to deprivation inputs makes the measure insensitive to any
permutation of the deprivation inputs. However, one may argue that is more deprived than if
we adopt the view that adding one dimension of deprivation to a period that has many dimensions
of deprivation cannot be offset by removing one dimension of deprivation from another period that
has equal or less periods of deprivation.6 This can be seen as analogous to FGT‟s transfer axiom,
but defined over counts of deprivation dimensions rather than income/expenditure.
Similarly is more deprived than if we adopt the view that adding one period of deprivation
to a dimension that has many periods of deprivation cannot be offset by removing one period of
6 can be derived from by adding one deprivation to and removing one deprivation from .
7
deprivation from a dimension that has equal or less periods of deprivation.7 This can be seen as
analogous to FGT‟s transfer axiom, but defined over counts of deprivation periods rather than
income/expenditure. We define both of these properties formally in the next subsection.
Notice that in the dynamic multidimensional case, a permutation of the deprivation inputs could
increase the breadth of deprivation at a particular period while having no effect on the depth in a
particular dimension (e.g. from to ); or increase the depth of deprivation in a particular
dimension, while having no effect on the breadth of deprivation at a particular time period (e.g.
from to ). It is therefore important to distinguish between the two.
To illustrate the importance of our distinction in the policy context, consider another example:
{
} {
} {
}
We argue that and are more deprived than because in the deprivations occur in time
periods and in dimensions that are independent from each other, whereas has deprivation in
three periods for one dimension, and has deprivation in three dimensions for one period. One
may argue it is a subjective matter as to which of the three profiles are the most detrimental to
individual wellbeing; even so, from a policy perspective, and may be given more importance
simply due to the relative ease of targeted poverty-reduction. In , for example, poverty-reduction
policies would target dimension one, while in , the policies would try to understand what was
unique in period one and focus on those factors.
The highlight of a measure that differentiates all three profiles is therefore twofold. Firstly it is able
to better utilise panel data on individuals by allowing for sensitivity of the index towards
deprivation concentrated in certain periods and certain dimensions. This is useful for comparing
different subgroups of a population. Secondly, it is able to highlight whether it is better for poverty-
reduction strategies to be focussed on certain dimensions, certain time periods, or both.
2.3. Dimensional and Dynamic Transfer Axioms
We now state the unique properties of our measure formally by firstly defining two types of
transfers.
Define as the column vector of individual ‟s deprivation profile for period . A dimensional
permutation occurs if there is a rearrangement of any two deprivation inputs and
7 can be derived from by adding one deprivation to and removing one deprivation from .
8
along . Furthermore,
prior to rearrangement and ∑
remains the
same after rearrangement for all .
Define as the row vector of individual ‟s deprivation profile for dimension . A dynamic
permutation occurs if there is a rearrangement of any two deprivation inputs and
along . Furthermore,
prior to rearrangement and ∑
remains the same
after rearrangement for all .
(Axiom 1): Dimensional Transfer
if is obtained from by a dynamic permutation of individual ‟s deprivation
profile in dimension and ∑
∑
where and individual n is
poor.
The Dimensional Transfer axiom requires that the poverty measure should register an increase if
deprivation was transferred to a dimension where there are more periods of deprivation.
(Axiom 2): Dynamic Transfer
if is obtained from by a dimensional permutation of individual ‟s
deprivation profile at period and ∑
∑
where and
individual n is poor.
The Dynamic Transfer axiom requires that the measure should register an increase if deprivation
was transferred to a period where there are more dimensions of deprivation.
2.4. The Proposed Dynamic Measure of Multidimensional Poverty
Consider the following explicit form for :
{
[∑ ((∑ ∑
)
) ] ⁄
[∑
] ⁄
(1a)
where ; , ∑ ∑
.
Like NR and BCCD, the measure continues to be a double-sum across time and dimensions.
However, the double-sum is now defined over a new term,
.
9
{
[ (∑
)
(
∑
)
]
(1b)
is therefore a transformation of deprivation inputs that weights each input according to its
contribution to the breadth (first term) and depth (second term) of deprivation. Increasing the
parameter increases the sensitivity of the measure to both the breadth and depth aspects. This
would therefore give increasing weight to individuals whose deprivations are located along the
same period/dimension relative to those who deprivations are unrelated by time or dimensions. The
parameter gives more weight to either the breadth or depth aspects. For most applications, it
seems appropriate to weight them equally and set ; however, for some applications it may
be useful to increase the weight to either the breadth or depth depending on policy interest. For
example, some policy-makers may not be overly concerned with cases where there is a
concentration of deprivation in a single period of time (breadth) since this is largely retrospective
information (e.g. a financial crisis in the past) with little directions for targeted poverty-reduction.
By setting and both dynamic and dimensional transfer are satisfied. When
only dynamic transfer (sensitivity to depth) is satisfied and when only dimensional transfer
(sensitivity to breadth) is satisfied (proof in Appendix A).
Consistent with the class of subgroup decomposable measures, has a convenient interpretation
as the average poverty score for individuals in the population of interest. Two special cases of the
measure allow unique interpretations: when , gives us the poverty headcount ratio,
while with , , gives us the average share of deprivations for individuals in
the population of interest. Increasing the parameter essentially increases the sensitivity of the
index to the count of deprivations. We discuss the other properties that emerge with different
choices of in subsection 2.7.
NR can be seen as a special case of when and . AK can be seen as a special case
of when while Foster (2007) can be seen as a special case of when . Figure
1 summarises the special cases in the literature that arises from different parametric choices for
.
10
Figure 1: Generalisation-tree of the class of subgroup decomposable poverty measures^
^: Note that the strands of literature referred to in the figure do not necessarily represent the origin of the extensions; for
example, multidimensional extensions had appeared as early as Bourguignon and Chakravarty (2003). We simply refer
to measures that are direct special cases of .
2.5. The Breadth-to-Depth ratio
The Breadth-to-Depth ratio (BTD) allows us to identify the relative contribution to the poverty
index of repeated periods in a few dimensions (the depth aspect) versus many dimensions
concentrated in a few periods (the breadth aspect).
The numerator of the BTD is the deprivation index when (dimensional transfer only) while
the denominator is the deprivation index when (dynamic transfer only).
Dimensional and
dynamic
differentiation and
sensitivity
Dimensional and dynamic
combination
𝛀𝜶𝜷𝜹
Nicholas and Ray, 2012
(𝛽 𝑧
Foster, 2007
(𝛽 𝐽 𝛿
Alkire & Foster, 2011a
(𝛽 𝑇 𝛿
Foster, Greer & Thorbecke, 1984
(𝛽 𝑇 𝐽 𝛿
11
∑
(
∑ ∑ 0(
∑
)
1
)
∑
(
∑ ∑ [(
∑
)
]
)
A BTD above 1 indicates that deprivation is more likely to be concentrated within specific periods
of time in the form of multiple dimensions („breadth‟). A ratio below 1 indicates that deprivation is
more likely to be concentrated within specific dimensions in the form of multiple periods („depth‟).
The BTD can therefore serve as an indicator of whether a policy-maker or researcher should be
concentrating their attention on specific periods of time (BTD>1) or concentrating their attention on
specific dimensions (BTD<1).
Note for any individual, the breadth and depth aspects of deprivation can only differ when the count
of deprivations, ∑ ∑
, is greater than zero and less than ( – i.e., the theoretical maximum
number of deprivations. When the count of deprivations is ( , the deprivation profile is „full‟
and therefore both breadth and depth aspects are at their respective maxima. For values of
∑ ∑
that lie between zero and , the differentiation of breadth and depth becomes the
most meaningful and the BDT can take on different values depending on the deprivation profile.
Consider an example where . Imagine that we observe ∑ ∑
. Consider the
following two possible permutations of individual A‟s deprivation profile:
{
} {
}
and show permutations of a deprivation profile that yield the minimum and
maximum value of BTD (respectively) given a fixed deprivation count ∑ ∑
(in this case, 3).
Now consider the poverty index measure, for individual A. Assume that ; .
will yield a BTD of 0.33, while will yield a BTD of 3. This represents the range for
the BTD for all possible permutations of the deprivation profile given that ∑ ∑
. The
larger the BTD range, the more informative the BTD is since it becomes harder to infer the value of
12
BTD by looking at ∑ ∑
alone. When , the BTD range is largest for ∑ ∑
.
For the BTD range is maximised for values of ∑ ∑
lying between and (proof in
Appendix B). In applications, the average deprivation count ∑ ∑
can be used as an
indicator of the usefulness of distinguishing depth from breadth. Figure 2 gives us the BTD range
for all possible values of ∑ ∑
based on our example above.
Figure 2: Breadth-to-Depth Ratio Range as a function of Count of Deprivation Inputs
2.6. Dimensional and Dynamic Decomposability
One of the attractive features of path-independence is the ability to decompose the measure
according to the individual contribution of each dimension, or the individual contribution of each
time period. This is useful when utilised jointly with the BDT, since it allows one to identify the
dimensions associated with the most depth of deprivation, or the periods associated with the most
breadth of deprivation.
While NR and BCCD have dimensional and dynamic decomposability, decomposition across time
in their measure is equivalent to calculating a static multidimensional measure (as per AF) for each
time period separately. The decomposition therefore only uses information on deprivations
associated with the particular period of time. Similarly, decomposition across dimensions is
equivalent to calculating a dynamic unidimensional measure (as per Foster, 2007) for each
dimension separately. The decomposition therefore only uses information on deprivations
associated with the particular dimension. Our measure has the advantage of allowing the
decompositions to be sensitive to information outside the period or dimension being decomposed
since each deprivation input is transformed to be a function of the entire deprivation profile.
0
0.5
1
1.5
2
2.5
3
3.5
1 2 3 4 5 6 7 8 9
Bre
adth
-to
-De
pth
Rat
io
Count of Deprivation Inputs
Max BTD
Min BTD
13
(Axiom 3): Dimensional Decomposability
(
) (
) (
), where
are exogenously
imposed dimensional weights, and (
) is the deprivation index calculated using only the
row vector of each individual‟s deprivation profile.
Dimensional decomposability is useful since it allows the policy maker to calculate the percentage
contribution of each dimension towards the overall deprivation index.
(Axiom 4): Dynamic Decomposability
(
) (
) (
), where
are exogenously
imposed dynamic weights, and (
) is the deprivation index calculated using only the
column vector of each individual‟s deprivation profile.
Dynamic decomposability allows the policy maker to calculate the percentage contribution of each
period towards the overall deprivation index.
Recall that the parameter governs the sensitivity of the index to the count of deprivations. If
, then is insensitive to the order of summation of
(i.e. we have path-independence
with respect to
). We therefore get:
[∑(∑ ∑
)
] ⁄ ∑([∑(∑
)
] ⁄ )
⁄
∑([∑(∑
)
] ⁄ )
⁄
The second and third terms are simply the definitions of dimensional and dynamic decomposability
respectively, proving that satisfies both properties.
When 0, gives us the headcount ratio of poverty: dimensional and dynamic
decompositions are then meaningless since an increase in the number of deprivation inputs has no
effect on the measure if the person is already classified as poor When 1, if decomposition is
desirable it seems reasonable to assume that the decomposition should be made as if 1. This is
because is simply a positive monotonic transformation of each individual‟s poverty score – while
it increases the weight given to certain individuals relative to other individuals, it does not change
14
the weights given to certain dimensions (periods) relative to other dimensions (periods).8 Increased
weight to dimensions that have more periods/dimensions of deprivation associated with them is
already captured by . Since dimensional/dynamic decomposability is satisfied for any value of
we recommend using for the purpose of decomposition according to dimensions.
2.7. Other Properties
also satisfies the following properties, which are based on AF but extended to the dynamic
framework. The difference with the following over the axioms in NR is the compatibility with the
poverty focus axiom (i.e. allowing the possibility of non-union identification).
(Axiom 5): Subgroup Decomposability
Define and as two subgroups of the population such that . Decomposability
requires
and holds for any number of subgroups where
is the deprivation index calculated over only a particular subgroup.
The axiom allows deprivation to be decomposed into a subgroup‟s percentage contribution towards
aggregate level deprivation.
(Axiom 6): Replication Invariance
Define a single replication of as ; of as ; of as .
Replication invariance implies for any number of replications of N, and
so long as the number of replications are the same for N, and .
This ensures that larger populations do not automatically count as more deprived.
(Axiom 7): Symmetry (anonymity)
where is any permutation of the vector .
Therefore, all individuals are identical except for their deprivation profiles.
(Axiom 8): Normalisation and Nontriviality
achieves at least two distinct values: a minimum value of 0 and a maximum value of 1.
8 leads to two properties (Deprivation Transfer and Deprivation Transfer Sensitivity) that are only useful in the
context of subgroup comparisons. We discuss these in subsection 2.7: „Other Properties‟.
15
(Axiom 9): Poverty Focus
if is obtained from by having a non-poor individual experience an
achievement increase in any deprivation.
Therefore, if an individual is not considered poor, then improvements in that individual‟s
achievements are irrelevant to the measure, similar to a Rawlsian social welfare function.
(Axiom 10): Deprivation Focus
if is obtained from by an increase in the achievement in a deprivation
where an individual is not considered deprived.
Therefore, if an individual is not deprived in a particular period and dimension, then improvements
in that particular deprivation input are irrelevant to the measure.
Proposition 1: satisfies the following core properties: Subgroup Decomposability,
Replication Invariance, Symmetry, Nontriviality, Normalisation, Poverty Focus, Deprivation Focus.
Proof: Since individual scores are simply summed and averaged over individuals, always
satisfies Subgroup Decomposability, Replication Invariance, Symmetry, Normalisation and
Nontriviality. It also satisfies Poverty Focus based on the cutoff and Deprivation Focus based on
the cutoff
Define a deprivation decrement as a decrease in any one of an individual‟s deprivation inputs.
(Axiom 11): Deprivation Monotonicity
if is obtained from by a deprivation decrement among the poor.
Proposition 2: The poverty index satisfies the core properties (axioms 5 to 10), and in
addition, Dynamic Transfer, Dimensional Transfer, Dynamic Decomposability, Dimensional
Decomposability, Deprivation Monotonicity when .
Proof: Dynamic and Dimensional Transfer have been proved in subsection 2.3. Dynamic
Decomposability, Dimensional Decomposability have been proved in subsection 2.6. Deprivation
Monotonicity is proven in Appendix C.
Define an averaging of deprivations as a transfer of deprivations between A and B such that
[ + ] and
[ + ] where . More
16
generally, let be a bistochastic matrix whose elements are unity for every non poor person in
. Then an averaging of achievements can be represented as: .
(Axiom 12): Deprivation Transfer9
where is obtained from by an averaging of deprivations among any two
poor individuals A and B where ≠ ; and is derived by a deprivation decrement of .
A reduction in inequality between poor individuals must reduce poverty.
(Axiom 13): Deprivation Transfer Sensitivity10
where is obtained from by an averaging of deprivations
between A and B, while is obtained by an averaging of deprivations between C and D.
≠ are the deprivation profiles of person A, B, C and D, all four whom which are
poor. is derived by a deprivation decrement of , is derived by a deprivation decrement of
, is derived by a deprivation decrement of
That is, a reduction in inequality between poor individuals must reduce poverty by more than an
equivalent reduction in inequality between relatively less poor individuals.
Table 1: Parameter choices and corresponding properties
>0 >0 >0 ≥1
>0 >0 >1 >2
=1 =0 >0, <1 >0, <1
Dimensional
Transfer √ √ √
Dynamic Transfer √ √ √
Deprivation
Transfer √ √
Deprivation
Transfer
Sensitivity
√
Increasing above 1 allows Deprivation Transfer to be satisfied and increasing it above 2 allows
Deprivation Transfer Sensitivity to be satisfied in addition (proof in Appendix C). However, both of
these come at the cost of Dynamic and Dimensional Decomposability. However, as previously
9 Unlike AF, the Transfer axiom is here defined over deprivation inputs rather than achievements. This distinction is
useful, since it allows a variant of the Transfer axiom to be satisfied even when (which is common in most
multidimensional measurement applications). When changes in achievements have no effect on deprivation
inputs unless they bring the individual above/below a deprivation threshold. 10
While not explicitly stated in AF, Deprivation Transfer Sensitivity is satisfied in their measure when their parameter
.
17
mentioned, since simply changes inter- rather than intra-individual weights, in cases where
it would be appropriate to do dynamic and dimensional decomposition as if
Table 1 summarises key properties that are satisfied with various combinations of the parameters.
We discuss the use of alternate weighting schemes and other extensions to the measure in Appendix
D.
3. Data sets and summary features
In keeping with the motivation of this study to apply the proposed new dynamic measure of
multidimensional poverty in the context of developing countries and show their empirical
usefulness, we have chosen panel data sets from China and Indonesia over roughly
contemporaneous time periods during the 1990s and the first decade of the 21st century. The
empirical evidence is of particular interest since, besides their developmental focus, they cover
periods that include the Asian Financial Crisis and (for China) the global financial crisis. We do not
compare multidimensional poverty between China and Indonesia due to differences between the
two countries in the dimensions on which the required information is available, as well as
differences in the representativeness of the two data sets.
In many cases, the variable denoting deprivation takes the binary form. In other cases, the available
information is cardinal so that one has information on the magnitude of the shortfall from the
deprivation cut-off. The latter allows the measure to better distinguish between deprived individuals
based on the magnitude of their deprivation. We consider both types of data in the application.
3.1 China
The Chinese data came from the China Health and Nutrition Survey (CHNS). This is an ongoing
international project between the Carolina Population Center at the University of North Carolina at
Chapel Hill and the National Institute of Nutrition and Food Safety at the Chinese Centre for
Disease Control and Prevention. This project was designed to examine the effects of health,
nutrition and family planning policies and programs implemented by the national and local
governments and to see how the social and economic transformation is affecting the health and
nutritional status of the population. A detailed description of the CHNS data base has been
presented in Popkin, Du, Zhai and Zhang (2010). The surveys took place over a three day period
using a multi-stage, random cluster process to draw a sample of over 4000 households in nine
provinces that vary substantially in geography, economic development, public resources and health
indicators. We converted household level information to the individual level by assuming that the
18
household‟s access to a facility such as drinking water or electricity is the same for all individuals in
that household. In a departure from previous applications on CHNS data, we supplemented the
individual and household level information with the community level information available in the
CHNS community identifiers. The community questionnaire (filled out for each of the primary
sampling units) collected information from a knowledgeable respondent on community
infrastructure (water, transport, electricity, communications, and so on), services (family planning,
health facilities, retail outlets), population, prevailing wages, and related variables.
The dimensions at the household, individual and community levels considered in this study are
described in Appendix E1. Appendix E2 describes the three different samples used for China.
Appendix E2 also provides the cut offs used in the quantitative dimensions (years of education,
BMI, BP, for both countries and, additionally for Indonesia, expenditure) in defining the “poor”.
While the CHNS data set is longitudinal, there was a trade-off between the length of the time
interval and the number of dimensions on which information is available for the panel of
individuals. The first two samples consider different combinations of time and dimensions through
the use of exclusively „qualitative‟ dimensions where an individual is either „poor‟ or „not poor‟ in a
dimension – there is no information on gradations of poverty.11
In contrast, sample three contains
only dimensions that are „quantitative‟, meaning there is information on the size of the poverty gap.
This means that setting in will now have an effect on the measure.
For the analysis, we focus initially on sample one given the breadth of dimensions it covers:
Appendix E3 presents the summary statistics, year and dimension wise, of the poverty rates in
China for the panel of individuals in sample one. Only individuals aged 18 years and above in the
first year of the panel were included in construction of the balanced panel. The sample statistics for
the panel in the other two samples have not been presented here for space reasons, but are available
on request. A couple of features are worth noting: (a) in case of dimensions that are common to the
three samples or panels, there is hardly any difference in the summary statistics between the three
samples; (b) while some dimensions such as electricity, drink water, radio/TV recorded large
improvements over the period, this is not consistently true; for example, the proportion of
individuals with abnormal blood pressure increased sharply while morbidity (in the form of illness)
and income poverty did not show a pronounced trend in any direction; (c) the third panel shows that
there was a reduction in income poverty in China during 2006-9 that not only confirms the widely
11
Note that though in case of some of these dimensions, for example, BMI, blood pressure and years of schooling, the
quantitative information is available, we converted them into qualitative dimensions for consistency with the others,
such as access to toilet, fuel, etc. However, the quantitative information was used in the third sample which considered
the depth in the poverty in these dimensions and their sensitivity to the weights attached to the depth via the parameter,
α.
19
held view that the global financial crisis bypassed China but that the country witnessed a growth in
income in sharp contrast to the experience of US and the European countries.
3.2 Indonesia
The Indonesian data came from the Indonesia Family Life Survey (IFLS). The Indonesian Family
Life Survey (IFLS) is a longitudinal survey in Indonesia conducted over 1993-2007.12
The sample
is representative of about 83% of the Indonesian population and contains over 30,000 individuals
living in 13 of the 27 provinces in the country. The first wave of the IFLS (IFLS1) was conducted
in 1993/94 by RAND in collaboration with Lembaga Demografi, University of Indonesia. IFLS2
and IFLS2+ were conducted in 1997 and 1998, respectively, by RAND in collaboration with UCLA
and Lembaga Demografi, University of Indonesia. IFLS2+ covered a 25% sub-sample of the IFLS
households. IFLS3, which was fielded in 2000 and covered the full sample, was conducted by
RAND in collaboration with the Population Research Center, University of Gadjah Mada. The
fourth wave of the IFLS (IFLS4), fielded in 2007/2008 covering the full sample, was conducted by
RAND, the Center for Population and Policy Studies (CPPS) of the University of Gadjah Mada and
Survey METRE. The present study considered all the four waves of the IFLS.
Like the third sample for China, the Indonesian data provided information on the quantitative
dimensions and so allowed a closer examination of the sensitivity of the poverty estimates to the
poverty gap sensitivity parameter, α. The quantitative dimensions that were considered in this study
are (i) body mass index (BMI), (ii) blood pressure (BP), (iii) years of education of each individual,
and (iv) per capita household expenditure. A balanced panel dataset of individuals aged 18 years
and above in the first year with information available in all the four dimensions was created. An
individual is considered „poor‟ on BMI if her/his BMI is less than 18.5; deprivation on account of
BP is defined as having a diastolic measure not within the 60-90mm (millimetres of mercury) range
and the systolic measure not within the 90-140 mm range.13
The poverty cut-off for education is
having completed primary education14
. The poverty cut-off for expenditure is half sample median.
Three of these, namely, BMI, education and income/expenditure were considered by AF in their
calculations on the IFLS data set.
Table E4 presents the summary statistics, year and dimension wise, on the poverty rates in
Indonesia. The two health dimensions provide mixed messages, with BMI showing steady
improvement, and BP showing no clear pattern. It is worth noting that in 1997, around the time of
12
See http://www.rand.org/labor/FLS/IFLS.html for a detailed description of the IFLS data set. 13
This information is taken from the blood pressure chart at http://www.vaughns-1-pagers.com/medicine/blood-
pressure.htm 14
Primary Education in Indonesia is 6 years.
20
the Asian Financial crisis, everyone in the sample had a BP outside the normal range. There was a
sharp improvement over the next three years, but there has been a sharp increase in the proportion
of individuals outside the normal range since then. In contrast, the non-health dimensions generally
agree that there has been a steady improvement in Indonesia in the deprivation scenario over the
decade, 1997-2007, considered here. Note, however, that even by the end of the sample period,
there was considerable deprivation in Indonesia with respect to primary education with nearly 40 %
of individuals in 2007 not having completed primary education.
4. Results
This section provides empirical evidence on the usefulness and interpretation of the proposed
measure, , through the use of its properties of dynamic, dimensional and subgroup
decomposition as well as reporting the sensitivity of the decomposition to changes in , which is
the parameter that governs sensitivity towards the breadth and depth of deprivation. In the case of
China, the results are reported for two samples, namely, the qualitative (Sample 1) and the
quantitative (Sample 3), while for Indonesia only the quantitative information was used. In all our
calculations we assume (that is, the breadth and depth aspects are given equal weight) and
that (in order to preserve subgroup decomposability). The results presented in Tables 2-7
consider three models each. Model 1 is the base case where (this leads to the measure used in
NR). Model 2 sets , therefore satisfying Dimensional Transfer and Dynamic Transfer
properties. Model 3 increases to 2, thereby giving additional weight to deprivations repeated in
the same dimension or period of time.
4.1. Multidimensional Poverty in China
The results for the qualitative sample for China (sample one) are presented in Tables 2 and 3, while
those for the quantitative sample are presented in tables 4 and 5. In the quantitative sample we set
to take advantage of the additional information.
We turn firstly to the results on subgroup decompositions (Tables 2 & 4). We define subgroups
according to three different groupings: 1) male/female; 2) province; 3) rural/urban. The reported
calculations present three numbers: the first is the percentage share of overall poverty contributed
by each population subgroup; the second calculation is the poverty index calculated for each
subgroup; and lastly, the rank of each subgroup (from least to most deprived) according to .
Note that the percentage share calculation is sensitive to the size of the subgroup while is not,
21
by virtue of being an average across individuals (as per the replication invariance axiom).
However, is not comparable across models because of changes in the choice parameters.
We can see that on average, rural deprivation exceeds urban deprivation, and females are more
deprived than males. In terms of provinces, the Guizhou province consistently comes up as the most
deprived from both subgroup contribution as well as . The ranking of the other provinces vary
across the two tables due to differences in the choice of dimensions and the use of quantitative
information in sample three. While increasing generally increases the weight given to subgroups
who are more deprived, the magnitude of the effect is considerably different when comparing the
qualitative application (Table 2) to the quantitative one (Table 4). While the deprivation ranking
remains the same in the former, the rankings for Jiangsu, Shandong and Hubei differ as increases
in the latter. Indeed, the deterioration of the Jiangsu province as increases suggests a stronger
concentration of dimensions along the breadth/depth aspects, even though the count of deprivations
may be smaller than Shandong or Hubei.
We also include BTD calculations for each subgroup which is useful on two fronts: it allows us to
quickly identify whether differences between groups can be attributable to differences in
depth/breadth concentration; and it allows us to identify at the overall level whether we should be
paying more attention to individual dimensions (through dimensional decomposition) or to
individual periods (through dynamic decomposition).15
Since the BTD calculations presented in Tables 2 and 4 are all less than 1, it suggests that the
breadth aspect is less significant than the depth aspect for China; i.e. it is more likely to observe
deprivations occurring over multiple periods of time in the same dimension rather than over
multiple dimensions in the same period of time. The reciprocal of the BTD in the data suggests that
the depth aspect contributes around twice the percentage to deprivation relative to the breadth
aspect when .
Like the subgroup poverty scores, the BTD changes considerably more in the quantitative
application. Recall that the Jiangsu province experienced a decline in rankings as was increased
in Table 4. The BTD for Jiangsu in fact reveals that it has the highest depth concentration (i.e.
lowest BTD) among the provinces, which explains the driving factor behind its high sensitivity to
changes to .
The BTD being less than 1 suggests that the policy-maker or poverty researcher should be focussed
on certain dimensions where deprivation occurs repeatedly over time. We examine this further by
15
Note that BTD cannot be identified when . Similarly, becomes irrelevant in such a case.
22
considering dimensional decomposition of the measure in Tables 3 & 5. Here the ranks depict
percentage contribution to overall deprivation (from least to most).
Table 3 shows that lack of access to toilet, fuel, drink water, road and transport are the more
significant contributors to deprivation, along with blood pressure not being in the normal range.
These dimensions, taken together, contribute well over 50 % of total deprivation and their shares
increase with . In contrast, access to electricity and health (measured by BMI and frequency of
illness) are success stories in China with relatively minor contributions to deprivation and whose
contributions decline with . On the other hand, Table 5 suggests that the most important dimension
is an individual‟s education. Unlike Table 3, in Table 5 the number of years below completing
primary education is taken into account and we can see the measure suggesting that as much as 95%
of deprivation is coming from the dimension when This may be due to the nature of
education – individuals who have chosen not to complete an education remain „uneducated‟ for the
rest of their lives and are therefore considered to be „deep‟ in deprivation for that dimension.
Tables 3 & 5 show that, under dynamic decomposition, the percentage contribution of each period
of time to overall deprivation in China is nearly uniform, and that this picture is quite robust to β.
The share of period 3 (2000) is noticeably lower in the qualitative sample (Table 3) than in the
quantitative sample (Table 5). The BTD in Tables 2 & 4 suggests the breadth aspect should not
have a strong role in the data – Figure 3 confirms this graphically by considering the relationship
between the time period shares and : as increases we see the shares converging towards 0.20;
i.e. „equal‟ shares among the five periods. This is because, as indicated by the BTD, the overall
measure is dominated by the depth aspect, therefore in relative terms, increasing increasingly
discounts the contribution of the breadth relative to the depth aspect.
4.2. Multidimensional Poverty in Indonesia
The multidimensional poverty estimates of Indonesia are presented in Tables 6 and 7. Unlike in the
case of China, the Indonesian calculations are based on only one sample which contains the
quantitative information on the dimensions (see the bottom half of Appendix Table E2).
As in the case of China, Table 6 presents the subgroup decompositions by male/female and
rural/urban grouping. The rankings of the subgroups are the same as in China and the BTD ratio is
also below 1, suggesting the dominance of the depth aspect. Dimensional decomposition in Table 7
also supports the education dimension (i.e. individuals below primary education) as the primary
contributor to deprivation measured quantitatively, although, increasing from 0 to 1 changes the
23
rank of expenditure from the second least to third least contributor, suggesting that after education,
expenditure exhibits the most deprivations over repeated periods.
In the case of dynamic decomposition (Table 7), we see that the largest share of deprivation is due
to 1997. It is well known that Indonesia was one of the worst affected countries in the 1997 Asian
financial crisis. However, like China we also find that as increases the shares converge towards
0.33 (by virtue of there being only three periods here). This is confirmed by Figure 4 which depicts
the relationship in Indonesia between the various sub periods‟ share of poverty and β. The
discounting of 1997‟s contribution as increases can be interpreted as follows: while 1997 clearly
has the most count of deprivations, because most of these deprivations appear only in 1997, they
have very low depth and since depth contributes more to the overall measure, 1997‟s contribution is
increasingly discounted. This highlights an important property of decomposition when :
information from other periods is used when calculating the contribution of each individual period;
similarly, information from other dimensions is used when calculating the contribution of each
individual dimension.
5. Discussion and Conclusion
There has been a large and rapidly proliferating literature on the multidimensional measurement of
poverty and its application to household survey data. The increasing availability of micro datasets
containing a wealth of household and individual information has helped in the development of
increasingly sophisticated measures that take into account the richness of the available information.
We have proposed a dynamic multidimensional measure of poverty that unites two strands of a
literature that finds its origins in Foster, Greer and Thorbecke (1984). Our measure allows
population subgroups to be compared when information regarding both multiple different
dimensions of deprivation and multiple periods of time are available. Following the axiomatic
framework in Alkire and Foster (2011a) we lay out two unique properties of our measure that are
only identifiable when such panel data is available. We argue that the separate identification and
weighting of the breadth and depth aspects of poverty are useful to both policy-makers and
researchers due to the general idea that a concentration of deprivations (whether defined over time
or over dimensions) should be given as much, if not more, attention than the simple count of
deprivations. Furthermore, policy prescriptions for poverty may vary depending on the type and
location of such concentration of deprivation. In our empirical application for Indonesia, for
example, we found that while the breadth of deprivation present during the financial crisis in 1997
24
had a large effect on poverty, poverty over the 1997-2007 is better explained by long-lasting
deprivation manifest in dimensions such as a lack of education. While such a conclusion is clearly
dependent on our choice of deprivation dimensions, our purpose has been to illustrate the usefulness
of the proposed measure, rather than provide conclusive evidence for China and Indonesia. The
choice of deprivation dimensions, cut-offs, as well as the appropriate weights to attach to them
remains a fertile research agenda beyond the scope of this paper [for a brief discussion, see Alkire
and Foster (2011b)].
The empirical application has highlighted that our measure allocates subgroup, dimensional, and
dynamic shares that are different from other models in the literature that exist as special cases of our
measure. This is especially true in the case where quantitative/cardinal information is available, in
which case increasing the sensitivity of the measure to breadth and depth aspects changes the
measure to the extent of switching the deprivation rank of dimension and subgroup shares. Another
point illustrated by our application has been the importance of the depth aspect relative to the
breadth of poverty in both China and Indonesia, suggesting that for several dimensions, deprivation
remains chronic. While the Alkire and Foster (2011a) method was designed primarily to keep track
of changes to multidimensional poverty over different periods of time by calculating a separate
measure for each period, the importance of the depth component revealed here suggests that when
comparing different groups of individuals, information regarding individual deprivation over time is
crucial in accurately characterising individual poverty.
The generality and flexibility of our measure comes at a cost since any researcher or policy-maker
will have to make choices with regards to multiple parameters. While we have provided a guide
based on the desirable properties of the measure, the degrees of freedom available remains
considerable. However, it is important to highlight that the additional flexibility introduced in our
measure can never be detrimental relative to pre-existing measures that are generalised here. A
notable example is the case of a policy-maker/researcher interested in comparing poverty profiles
over time, rather than over subgroups. If so, the special case of the Alkire and Foster (2011a)
measure is appropriate since the concept of depth is not required. However, further assume the
policy-maker/researcher only has qualitative information on deprivations; he may then wish to set
to allow the measure to be increasingly sensitive to the count of deprivations (as per the
deprivation transfer axiom) since changing would have no effect; indeed, defaulting entirely to
the original Alkire and Foster (2011a) would inappropriately amount to assuming that .
When faced with a large degree of undecidedness with regards to parameter choices, we would
advocate calculating the results for various combinations of the parameters. Indeed, understanding
25
how and why poverty measurement changes with the various parameters is arguably more
informative than poverty conveyed by a single „definitive‟ number.
Table 2: Multidimensional Deprivation and its Subgroup Decomposition for
China based on CHNS Qualitative Sample One
Ωαβδ (Hg) BTD(Hg)
MODEL 1 MODEL 2 MODEL 3
MODEL
1 MODEL 2
MODEL
3
α 0 0 0 0 0 0
β 0 1 2 0 1 2
δ 1 1 1 1 1 1
γ irrelevant 0.5 0.5
Ωαβδ 0.3597 0.2155 0.1552 N/A 0.5704 0.3225
(Subgroup Proportion, Ωαβδ, Rank)
Male (0.4381, 0.3520, 1) (0.4386, 0.2092, 1) (0.4331, 0.1501, 1) N/A 0.5612 0.3134
Female (0.5619, 0.3659, 2) (0.5654, 0.2207, 2) (0.5669, 0.1593, 2) N/A 0.5775 0.3295
Province
Jiangsu (0.1539, 0.3425, 3) (0.1492, 0.1988, 3) (0.1453, 0.1394, 3) N/A 0.5690 0.3230
Shandong (0.1327, 0.4010, 6) (0.1403, 0.2541, 6) (0.1457, 0.1899, 6) N/A 0.5517 0.3008
Henan (0.1063, 0.4003, 5) (0.1114, 0.2514. 5) (0.1151, 0.1871, 5) N/A 0.5636 0.3122
Hubei (0.1621, 0.3655, 4) (0.1633, 0.2205, 4) (0.1637, 0.1592, 4) N/A 0.5718 0.3229
Hunan (0.1176, 0.3195, 2) (0.1095, 0.1783, 2) (0.1047, 0.1228, 2) N/A 0.5687 0.3221
Guangxi (0.1068, 0.2600, 1) (0.0923, 0.1346, 1) (0.0856, 0.0898, 1) N/A 0.5193 0.2689
Guizhou (0.2205, 0.4308, 7) (0.2341, 0.2741, 7) (0.2399, 0.2022, 7) N/A 0.6074 0.3613
Rural (0.8119, 0.3912, 2) (0.8316, 0.2401, 2) (0.8384, 0.1743, 2) N/A 0.5834 0.3342
Urban (0.1881, 0.2669, 1) (0.1684, 0.1432, 1) (0.1616, 0.0989, 1) N/A 0.5091 0.2647
26
Table 3: Dynamic and Dimensional Decomposition for China (CHNS Qualitative Sample
One) Model 1 Model 2 Model 3
α 0 0 0
β 0 1 2
δ 1 1 1
γ Irrelevant 0.5 0.5
Dynamic Decomposition
Period Contribution* Rank Contribution* Rank Contribution* Rank
Period 1 (1993) 0.2176 5 0.2141 5 0.2097 5
Period 2 (1997) 0.2025 4 0.2046 4 0.2048 4
Period 3 (2000) 0.1967 3 0.2011 3 0.2031 3
Period 4 (2004) 0.1917 2 0.1921 2 0.1941 2
Period 5 (2006) 0.1915 1 0.1880 1 0.1884 1
Total 1 1 1
Dimensional Decomposition
Contribution** Rank Contribution** Rank Contribution** Rank
No access to toilet 0.1240 11 0.1352 12 0.1437 12
No access to fuel 0.1568 13 0.1699 13 0.1830 13
No access to electricity 0.0017 1 0.0013 1 0.0010 1
No access to drink water 0.1165 10 0.1274 10 0.1373 11
No access to at least one
vehicle type 0.0549 5 0.0525 5 0.0502 6
No access to radio/TV 0.0595 6 0.0547 6 0.0478 5
BMI not in normal range 0.0212 2 0.0181 2 0.0160 2
Illness in the last four
weeks 0.0375 3 0.0250 3 0.0164 3
Blood pressure not normal 0.1339 12 0.1289 11 0.1269 10
Individual below primary
education 0.0826 8 0.0940 9 0.1050 9
No access to road in
community 0.0939 9 0.0867 8 0.0772 8
No access to bus station in
community 0.0776 7 0.0753 7 0.0716 7
No access to school in
community 0.0399 4 0.0313 4 0.0237 4
Total 1 1 1
*Contribution is the ratio ( ) where ∑ ((∑
) )
**Contribution is the ratio ( / ) where ∑ ((∑
) )
27
Table 4: Multidimensional Deprivation and its Subgroup Decomposition for China based on CHNS
Quantitative Sample Three
Ωαβδ (Hg) BTD(Hg)
MODEL 1 MODEL 2 MODEL 3
MODEL
1
MODEL
2
MODEL
3
α 1 1 1 1 1 1
β 0 1 2 0 1 2
δ 1 1 1 1 1 1
γ irrelevant 0.5 0.5
Ωαβδ 0.1396 0.0572 0.0369 N/A 0.4752 0.1987
(Subgroup Proportion, Ωαβδ, Rank)
MALE (0.3713, 0.0980, 1) (0.2812, 0.0305, 1) (0.2348, 0.0164, 1) N/A 0.5106 0.2259
FEMALE (0.6687, 0.1766, 2) (0.7490, 0.0811, 2) (0.7905, 0.0552, 2) N/A 0.4638 0.1917
REGION
Heilongjiang (0.0973, 0.0907, 1) (0.0680, 0.0259, 1) (0.0561, 0.0138, 1) N/A 0.4786 0.2013
Jiangsu (0.1421, 0.1418, 4) (0.1568, 0.0641, 5) (0.1678, 0.0443, 6) N/A 0.4516 0.1818
Shandong (0.1314, 0.1488, 6) (0.1408, 0.0654, 6) (0.1447, 0.0433, 5) N/A 0.4631 0.1897
Henan (0.0913, 0.1590, 7) (0.0991, 0.0708, 7) (0.1030, 0.0474, 7) N/A 0.4684 0.1942
Hubei (0.1255, 0.1430, 5) (0.1256, 0.0588, 4) (0.1256, 0.0379, 4) N/A 0.4824 0.2056
Hunan (0.0975, 0.1130, 2) (0.0798, 0.0379, 2) (0.0693, 0.0212, 2) N/A 0.4928 0.2152
Guangxi (0.0980, 0.1179, 3) (0.0792, 0.0391, 3) (0.0703, 0.0224, 3) N/A 0.4995 0.2115
Guizhou (0.2168, 0.2043, 8) (0.2963, 0.1971, 8) (0.2632, 0.1982, 8) N/A 0.4826 0.2050
RURAL (0.7262, 0.1467, 2) (0.7359, 0.0610, 2) (0.7383, 0.0395, 2) N/A 0.4806 0.2025
URBAN (0.2738, 0.1236, 1) (0.2641, 0.0489, 1) (0.2617, 0.0312, 1) N/A 0.4609 0.1882
28
Table 5: Dynamic and Dimensional Decomposition for China (CHNS Quantitative Sample
Three)
Model 1 Model 2 Model 3
α 1 1 1
β 0 1 2
δ 1 1 1
γ irrelevant 0.5 0.5
Dynamic Decomposition
Period Contribution* Rank Contribution* Rank Contribution* Rank
Period 1 (1997) 0.2160 5 0.2135 5 0.2110 5
Period 2 (2000) 0.2035 3 0.2030 3 0.2037 3
Period 3 (2004) 0.1784 1 0.1785 1 0.1845 1
Period 4 (2006) 0.2088 4 0.2083 4 0.2041 4
Period 5 (2009) 0.1933 2 0.1967 2 0.1966 2
Total 1 1 1
Dimensional Decomposition
Contribution** Rank Contribution** Rank Contribution** Rank
BMI not in normal range 0.0071 1 0.0024 1 0.0010 1
Blood pressure not normal 0.3329 2 0.1351 2 0.0481 2
Individual below primary
education 0.6600 3 0.8625 3 0.9509 3
Total 1 1 1
*Contribution is the ratio ( ) where ∑ ((∑
) )
**Contribution is the ratio ( / ) where ∑ ((∑
) )
29
Table 6: Multidimensional Deprivation and Subgroup Decomposition for Indonesia: IFLS
(Quantitative Sample)
(Proportion, Ωαβδ , Rank) BTD(Hg)
MODEL 1 MODEL 2 MODEL 3
MODEL
1
MODEL
2
MODEL
3
α 1 1 1 1 1 1
β 0 1 2 0 1 2
δ 1 1 1 1 1 1
γ irrelevant 0.5 0.5
Ωαβδ 0.0941 0.0377 0.0236 N/A 0.3512 0.1110
MALE (0.3890, 0.0823, 1) (0.3790, 0.0321, 1) (0.3706, 0.0197, 1) N/A 0.3561 0.1138
FEMALE (0.6110, 0.1035, 2) (0.6211, 0.0421, 2) (0.6296, 0.0268, 2) N/A 0.3484 0.1094
RURAL (0.6127, 0.1193, 2) (0.6422, 0.0501, 2) (0.6543, 0.0320, 2) N/A 0.3401 0.1054
URBAN (0.4018, 0.0731, 1) (0.3740, 0.0273, 1) (0.3640, 0.0167, 1) N/A 0.3583 0.1126
Table 7: Dynamic and dimensional decomposition for Indonesia: IFLS (Quantitative
Sample)
Model 1 Model 2 Model 3
α 1 1 1
β 0 1 2
δ 1 1 1
γ irrelevant 0.5 0.5
Ωαβδ 0.0941 0.0377 0.0236
Dynamic Decomposition
Period Contribution* Rank Contribution* Rank Contribution* Rank
Period 1 (1997) 0.4474 3 0.4285 3 0.4035 3
Period 2 (2000) 0.2783 2 0.2857 1 0.2991 2
Period 3 (2007) 0.2743 1 0.2858 2 0.2969 1
Total 1 1 1
Dimensional Decomposition
Contribution** Rank Contribution** Rank Contribution** Rank
BMI not in normal range 0.0211 1 0.0067 1 0.0022 1
Blood pressure not normal 0.0968 3 0.0259 2 0.0072 2
Individual below primary
education 0.8225 4 0.9368 4 0.9773 4
Income 0.0595 2 0.0306 3 0.0133 3
Total 1 1 1
*Contribution is the ratio ( ) where ∑ ((∑
) )
**Contribution is the ratio ( / ) where ∑ ((∑
) )
30
Fig 3: Variation in the Temporal Shares of Deprivation in China with β (for α=1)
Fig 4: Variation in the Temporal Shares of Deprivation in Indonesia with β (for α=1)
31
REFERENCES
Alkire, S. and Foster, J. (2011a), “Counting and multidimensional poverty measurement”,
Journal of Public Economics, 95, 476-487.
Alkire, S. and Foster, J. (2011b), “Understandings and misunderstandings of multidimensional
poverty measurement”, Journal of Economic Inequality, 9: 289-314.
Atkinson, A. (2003), “Multidimensional deprivation: contrasting social welfare and counting
approaches”, Journal of Economic Inequality, 1, 51-65.
Bossert, W., Chakravarty, S., and C. D‟Ambrosio (2013), “Multidimensional Poverty and Material
Deprivation with Discrete Data”, Review of Income and Wealth 59(1), 29-43.
Bossert, W., Chakravarty, S., and C. D‟Ambrosio (2012), “Poverty and Time”, Journal of
Economic Inequality, 10, 145-162.
Bossert, W., Ceriani, L., Chakravarty, S., and C.D‟Ambrosio (2012), “Inter temporal Material
Deprivation”, mimeographed.
Bossert, W., D‟Ambrosio,C., and V. Peragine (2007), “Deprivation and Social Exclusion”,
Economica, 74 (296), 777-803.
Bourguignon, F. and S. Chakravarty (2003), “The Measurement of Multidimensional Poverty”,
Journal of Economic Inequality, 1, 25-49.
Calvo, C. and S. Dercon (2009), “Chronic poverty and all that: the measurement of poverty over
time” in T.Addison, D.Hulme and R.Kanbur (eds), Poverty Dynamics: interdisciplinary
perspectives, 29-58. Oxford University Press, Oxford.
Chakravarty, S. and C. D‟Ambrosio (2006), “The Measurement of Social Exclusion”, Review of
Income and Wealth, 52, 377-398.
Dalton, H., (1920), “The Measurement of the Inequality of Incomes”, Economic Journal, 30: 348-
61.
Duclos. J, Araar, A. and J. Giles (2010), “Chronic and transient poverty: Measurement and
estimation, with evidence from China”, Journal of Development Economics, 91, 266-277.
Foster, J. (2007), “A class of chronic poverty measures” in T. Addison, D. Hulme and R. Kanbur
(eds), Poverty Dynamics: interdisciplinary perspectives, 59-76. Oxford University Press,
Oxford.
Foster, J., Greer, J. and E. Thorbecke (1984), “A class of decomposable poverty indices”,
Econometrica, 52, 761-766.
Gradin, C., del Rio, C. and O. Canto (2012), “Measuring poverty accounting for time”, Review of
Income and Wealth, 58, 330-354.
Hojman, D. and F. Kast (2009), “On the measurement of poverty dynamics”, Faculty Research
Working Paper Series, RWP 09-35, Harvard Kennedy School.
Hoy, M. and B. Zheng (2011), “Measuring lifetime poverty”, Journal of Economic Theory, 146,
2544-2562.
32
Jayaraj, D. and S. Subramanian (2010), “A Chakravarty-D‟Ambrosio View of Multidimensional
Deprivation: Some Estimates for India”, Economic and Political Weekly, XLV, 53-65.
Kahneman, D, and A. Tversky (1979), "Prospect theory: An analysis of decision under
risk." Econometrica: 263-291.
Mishra, A. and R. Ray (2012), “Multidimensional deprivation in the awakening giants: A
comparison of China and India on micro data” Journal of Asian Economics, 23(4), 454-465.
Nicholas, A. and R. Ray (2012), “Duration and Persistence in Multidimensional Deprivation:
Methodology and Australian Application”, Economic Record, 88, 106-126.
Pigou, A.F., (1912), Wealth and Welfare, Macmillan, London.
Popkin,B.M., Du S., Zhai,F. and Zhang, B. (2010), “Cohort Profile: The China Health and Nutrition
Survey- monitoring and understanding socio-economic and health change in China, 1989-
2011”, International Journal of Epidemiology, 39, 1435-1440.
Sen, A. (1976), “Poverty: an ordinal approach to measurement”, Econometrica, 44, 291-231.
Sen, A. (1985), Commodities and Capabilities. North Holland, Amsterdam.
Tsui, K. (2002),”Multidimensional poverty indices”, Social Choice and Welfare, 19, 69-93.
33
APPENDIX A: Dynamic and dimensional transfer
Proposition: satisfies Dynamic Transfer when , and .
Proof:
We restrict our attention to individual ‟s deprivation profile since that is the only thing that
changes. A dimensional permutation ensures that ∑
remains the same after the permutation,
for all t. It also ensures that ∑ ∑
remains the same after the permutation. Therefore a measure
that is path-independent in terms of (as in NR) will not change with a dimensional
permutation. This means that when , dynamic transfer is clearly not satisfied.
First, define
and
.
Dimensional permutation results in
and
where is any non-negative real
number. We need to show individual deprivation score has increased with the
permutation, which is equivalent to showing
when ∑
∑
where .
∑ ∑
∑ ∑
Consider the first term on the right.
∑ ∑
[ (∑ ∑
)
]
Equation (4b) is positive when . This is a necessary condition for the proof since it ensures
that increases in ∑ ∑
translate into increases in the individual‟s deprivation score.
Now consider the second term on the right of equation (4a). Note, ∑ ∑
∑ ∑
Recall that
[ (∑
)
(
∑
)
]
Since ∑
does not change with dimensional permutation, we can ignore its influence on
and instead define
(∑
)
.
(∑
)
(
∑
)
34
(∑
)
(
∑
)
We now need to show that
is a positive function of where
is any where
but . If so this automatically implies
by virtue of ∑
∑
. However, because appears in the denominator as well it is unclear whether its
influence on
is positive.
Differentiating (4c) by yields:
( ) (
) (
∑
)
,
∑
-
Equation (4d) is positive for any and .
Proposition: satisfies Dimensional Transfer when , and .
Following the proof for dynamic transfer, define
(∑
)
( ) (
) (
∑
)
{
∑
}
where is any
where but . Equation (4e) is positive for any and
.
35
APPENDIX B: Breadth-to-Depth Ratio
Proposition: When , the BTD range is maximised when the deprivation count ∑ ∑
. When the BTD range is maximised at some value of ∑ ∑
lying between and
.
Proof:
We first consider the simple case where and deprivation profiles are therefore square
matrices. For simplicity, we restrict our attention to a single individual‟s deprivation score, .
The BTD range can be represented by the gap between the BTDs obtained from and .
Because we have a square matrix, the deprivation count that maximises BTD for will also
automatically minimise the BTD for since will simply be a 90 degree rotation of the
deprivation profile in . This means that finding the deprivation count that maximises the
BTD for will be sufficient proof for the first part of the proposition.
The theoretically maximum any individual deprivation input can contribute to the numerator
term of the BTD is when the deprivation input belongs to a with ∑
. The theoretically
minimum any individual deprivation input can contribute to the denominator term of the BTD
is when the deprivation input belongs to a j with ∑
; that is,
is the only
deprivation input in . Both these conditions can only be simultaneously satisfied with a deprivation
profile where there is only one column with ∑
and each other column is empty. This is
clearly only satisfied when ∑ ∑
, which therefore gives us the deprivation count for
which achieves a maximum.
Given the square matrix assumption, rotating the profile gives us a profile where there is one row of
∑
and each other row is empty. This is also clearly only possible when ∑ ∑
,
which gives us the deprivation count for which achieves a minimum. Since ∑ ∑
allows and to achieve a maximum and minimum BDT respectively,
∑ ∑
maximises the BDT range.
Now consider the case where we have a non-square matrix, . Since the matrix is no longer
square, the deprivation profiles can no longer be rotated and a given deprivation count that
36
maximises the BTD for will not minimise the BTD for . Similarly, a deprivation
count that minimises the BTD for will not maximise the BTD for .
Note that the BTD converges to 1 when ∑ ∑
and tends towards 1 when ∑ ∑
is close to zero. Therefore deprivation counts that are further from the value that maximises the
BTD for will result in a lower BTD while deprivation counts that are further from the value
that minimises the BTD for will result in a higher BTD. Consider an example when by
some integer a. In such a case, setting ∑ ∑
by some integer b results in being b
counts away from a maximum BTD and being a+b counts away from a minimum BTD.
Whereas setting ∑ ∑
results in achieving a maximum BTD and being a
counts away from a minimum BTD.
Similarly, setting ∑ ∑
by some integer c results in being a+c counts away from a
maximum BTD and being c counts away from a minimum BTD. Whereas setting
∑ ∑
results in achieving a minimum BTD and being a counts away from
a maximum BTD.
For deprivation counts ∑ ∑
that lie between and , however, changing the count trades off
the distance between achieving a minimum versus a maximum .
We therefore conclude that the BTD measure is the most informative for deprivation counts that lie
between and . Simulations with and suggest that the BTD range is in
fact maximised when ∑ ∑
.
Because ∑ ∑
is not a choice variable establishing a general proof for the precise point at
which the BTD range is maximised is beyond the scope of this paper.
37
APPENDIX C: Conditions for Deprivation Monotonicity,
Transfer and Transfer Sensitivity
For the following, we restrict ourselves to individual deprivation score (∑ ∑
)
since is simply the average across all individual. Define define
Deprivation Monotonicity
Deprivation Monotonicity requires that
∑ ∑
∑ ∑
.
∑ ∑
[ (∑ ∑
)
] when
∑ ∑
∑∑
[ (∑
)
(
∑
) (∑
)
(
∑
)] for
when (by definition of
)
Deprivation monotonicity therefore holds for when .
Deprivation Transfer
Transfer requires that
( ) (
) where
is any not equal to
.
( ) (
)
∑ ∑
∑ ∑
( ) (
)
∑ ∑
∑ ∑
(
)
Assume that Monotonicity holds. Then,
∑ ∑
and ∑ ∑
>0.
TERM (1)
∑ ∑
( ) (
) ∑ ∑
( ) (
)
. We know from the proof for dimensional transfer that
( ) (
) for and but only for
that share the same dimension or
same period as . Otherwise,
( ) (
)
TERM (2)
∑ ∑
(
) [ (
∑ ∑
)
] for .
38
Since the Term (2) result holds regardless of the data, setting ensures that the index satisfies
the transfer property.
Deprivation Transfer sensitivity
Transfer requires that
( ) (
)
( ) (
)
∑ ∑
( ) (
)
(∑ ∑
) (
)
∑ ∑
( )
∑ ∑
(
)
∑ ∑
∑ ∑
( ) (
)
Assume that Transfer and Monotonicity holds. Then
∑ ∑
, ∑ ∑
,
(∑ ∑
) (
) >0
TERM (1)
Term (1) is positive if Term (1) for Transfer holds, and zero otherwise.
TERM (2)
(∑ ∑
) (
) [ (
∑ ∑
)
] for .
TERM (3)
When does not share the same dimension or period as
then, ∑ ∑
( ) (
)
∑ ∑
( ) (
)
If it shares the same dimension then:
( ) (
) (
∑
)
,
∑
-
( ) (
) for and
( ) (
) for .
Therefore, to ensure Transfer Sensitivity holds, set and .
39
APPENDIX D: Weights and Extensions
Because our measure is a generalisation of the class of measures in NR and BC, it can incorporate
the extensions used there, namely in terms of „persistence‟ (Gradin et al, 2012; Bossert et al 2012)
and „loss-aversion‟ (Hojman and Kast, 2009). Given our focus on the interaction between „depth‟
and „breadth‟ of deprivations, we avoided the use of such extensions in our empirical applications
as they would make it harder to identify the effect of depth versus breadth given the additional
trade-offs introduced with such extensions. However, we briefly describe how these can be
incorporated into our measure – the interested reader is advised to refer to the aforementioned
papers directly for more details regarding their properties.
0∑.(
∑ ∑
)
/
1 ⁄
is a generic weight associated with each
(the transformed deprivation input). These
weights can be defined in a variety of ways, though they clearly change the interpretation of the
measure. „Persistence‟, which gives increasing weight to deprivations that occur in consecutive
periods, for example, can be incorporated by defining ( ) where is the length of
the deprivation spell associated with a particular
. Another way of defining persistence (BCCD)
could be where the weights are increasing in the number of consecutive periods
that an individual experiences breath of deprivation above a predetermined cutoff.
Persistence may not always be relevant to every dimension of deprivation. One can, for example,
imagine that being unemployed for three consecutive periods and then being employed for three
consecutive periods is superior to alternating in and out of employment for six periods since one
incurs an „adjustment cost‟ when changing states. This can be captured with the concept of loss-
aversion (Kahneman and Tversky, 1979; applied to poverty indices in Hojman and Kast, 2009).
Consider, for example:
*
(
∑
∑
) (
)+ ∑
∑
*
(
∑
∑
) (
)+
40
This weighting scheme has two features: 1) it allows a particular period to be weighted heavier if it
is proceeded by more dimensions of deprivation and to be weighted by less if it is proceeded by less
dimensions of deprivation. 2) it captures the concept of loss-aversion by weighting increases in the
breadth of deprivation in the subsequent period more heavily than an equivalent decrease in the
breadth of deprivation in the subsequent period (in absolute terms).
Another potential use of the general weights is to assign different importance to different
dimensions. As suggested by Atkinson (2003), it is not unreasonable in most applications to start
with the case where
; that is, where each dimension is weighted equally. With additional
information, however, a researcher or policy-maker may assign different weights (so long as
∑ ). Reasons behind changing the weights include „double counting‟ in the sense that some
dimensions may essentially be capturing the same aspect of deprivation, which justifies the
discounting of the importance of the associated dimensions; another reason is that individuals may
actually put very low importance on certain dimensions; e.g. in Bossert et al (2013) dimensions are
weighted based on the views of society regarding the importance of those dimensions -- “consensus
weighting”.
41
APPENDIX E1: Description of Dimensions used for analysis
Type Variable Description
Chin
a Indonesia
Multidimensional Deprivation Dimension
Household
Access to
Toilet (D1)
Individual‟s household has access to improved
toilet facility as per UN norms. UN defines
improved facility as having own flush toilet,
own pit toilet, traditional pit toilet, ventilated
improved pit latrine, pit-latrine with slab, flush
toilet, and composting toilet X X
Access to
fuel (D2)
Fuel used for cooking by individual‟s
household is kerosene, electricity, LPG or
biogas X X
Access to
electricity
(D3)
Individual‟s household has access to electricity X X
Access to
drink water
(D4)
Individual‟s household has access to improved
drinking water source. UN defines improved
drinking water source as piped water into
dwelling, plot or yard, public tab/standpipe,
tube well, borehole, protected dug well,
protected spring and rainwater.
X X
Access to
atleast one
vehicle type
(D5)
Individual‟s household owns atleast one of the
following mode of transport: a bicycle,
motorcycle or car.
X X
Access to
radio/TV
(D6)
Individual‟s household owns atleast a radio,
b/w Television or a colour Television X X
Individual
BMI<18.5 or
BMI>30
(D7)
If BMI of the individual is not normal (i.e.,
either greater than 30 or less than 18.5.)
X X
Illness in the
last four
weeks (D8)
whether individual suffered illness in the last
four weeks.
X
Blood
pressure not
normal (D9)
If the blood pressure is normal (i.e., 90=<
systolic< 120 and 60<= diastolic<80)
X X
Individual
atleast
primary
educated
(D10)
Individual is educated upto primary ( year 6). X X
Community
Access to
road (D11)
If the village/neighbourhood has a stone/gravel
or paved roads.
X
Access to
bus/train
If there is a bus stop in the
village/neighbourhood.
X
42
station (D12)
Access to
school (D13)
If there is a primary/middle/upper middle
school in the village/neighbourhood.
X
APPENDIX E2: Description of balanced samples used for analysis
China Analysis: Sample Description
Sample 1 Balanced panel Sample of 1874 individuals over five years (1993-1997-2000-2004-2006)
covering 7 provinces.
Dimensions D1-D2-D3-D4-D5-D6-D7-D8-D9-D10-D11-D12-D13
Description This sample comprises of household, individual and community dimensions.
All dimensions are qualitative.
Sample 2 Balanced panel Sample of 1934 individuals across six years (1991-1993-1997-2000-2004-2006)
covering 8 provinces.
Dimensions D1-D2-D3-D4-D5-D6-D7-D8-D9-D10
Description This sample comprises of household and individual dimensions only. All
dimensions are qualitative.
Sample 3 Balanced panel Sample of 2458 individuals across five years (1997-2000-2004-2006-2009)
Dimensions 1. Years of education of individual (cut off is primary education i.e., 6 years)
2. Individual's Body Mass Index (cut off is 18.5)
3. Individual's Blood Pressure (systolic/diastolic).i.e., more than 120/80 then
high BP.
Description All dimensions in this sample are quantitative
Indonesia Analysis: Sample Description
Balanced
panel Individuals across three years 1997-2000-2007
Dimensions 1. Years of education of individual (cut off is primary education i.e., 6 years)
2. Individual's Body Mass Index (cut off is BMI > 18.5)
3. Individual's Blood Pressure (systolic/diastolic).i.e., more than 120/80 then
high BP.
4. Expenditure. ( cut off is half median)
Description This sample comprises of household and individual dimensions. All dimensions
are quantitative.
43
Appendix E3: Dimension wide poverty rates for China (Sample 1, Qualitative – balanced panel of 6 years and 13 dimensions)
CHNS
wave
No A
cces
s to
Toil
et
No A
cces
s to
fuel
No a
cces
s to
ele
ctri
city
No a
cces
s to
dri
nk w
ater
No a
cces
s to
atl
east
one
veh
icle
type
No a
cces
s to
rad
io/T
V
BM
I not
in n
orm
al r
ange
Illn
ess
in t
he
last
four
wee
ks
Blo
od p
ress
ure
not
norm
al
Indiv
idual
bel
ow
pri
mar
y
educa
ted
No a
cces
s to
road
in
com
munit
y
No a
cces
s to
bus/
trai
n
stat
ion i
n c
om
munit
y
No a
cces
s to
sch
ool
in
com
munit
y
Inco
mea l
ess
than
0.5
*M
edia
n
1991 0.6665 0.9189 0.0523 0.6526 0.1827 0.4823 0.1185 0.2391 0.4726 0.4454 0.8538 0.4885 0.2560 0.2344
1993 0.6573 0.8769 0.0164 0.6660 0.1637 0.4330 0.0944 0.1596 0.5239 0.4228 0.8964 0.4495 0.1837 0.2344
1997 0.6147 0.7414 0.0036 0.5557 0.2376 0.3740 0.0867 0.1585 0.6039 0.3982 0.7542 0.3217 0.1611 0.2488
2000 0.5952 0.6952 0.0092 0.5485 0.2550 0.2950 0.0841 0.1827 0.6485 0.3833 0.8481 0.3638 0.1432 0.2904
2004 0.5428 0.7086 0.0056 0.5141 0.2996 0.1652 0.1067 0.3479 0.6963 0.3366 0.9389 0.3628 0.2365 0.2776
2006 0.4772 0.6321 0.0056 0.4962 0.3258 0.1113 0.1042 0.2986 0.7045 0.4310 0.9656 0.4228 0.2817 0.3074
aIncome is used as a benchmark for comparison and not a dimension in itself.
44
Appendix E4: Dimension wise Poverty rates in Indonesia (for Qualitative Sample)
Indonesi
a
wave
Not
norm
al
BP
BM
I<18.5
Ed
uca
tion
(B
elow
Pri
mary
)
Hou
seh
old
per
cap
ita
exp
end
itu
re <
0.5
*M
edia
n
1997 1.000 0.124 0.579 0.114
2000 0.297 0.119 0.370 0.102
2007 0.443 0.107 0.379 0.094