Multidimensional Poverty in Bhutan: Estimates and Policy
Implications
Maria Emma Santos* and Karma Ura"
Abstract
This paper estimates multidimensional poverty in Bhutan
applying a recently developed methodology by Alkire and
Foster (2007) using the 2007 Bhutan Living Standard Survey
data. Five dimensions are considered for estimations in both
rural and urban areas (income, education, room availability,
access to electricity and access to drinking water) and two
additional dimensions are considered for estimates in rural
areas only (access to roads and land ownership). Also, two
alternative weighting systems are used: a baseline using equal
weights for every dimension and another one using weights
derived from the Gross National Happiness Survey. Estimates
are decomposed into rural and urban areas, by dimension and
between districts. It was found that multidimensional poverty
is mainly a rural phenomenon, although urban areas present
non-depreciable levels of deprivation in room availability and
education. Within rural areas, it was found that poverty in
education, electricity, room availability, income and access to
roads, contribute in similar shares to overall multidimensional
poverty, while poverty in land ownership and water have a
relatively smaller contributions. The districts of Samtse,
Mongar, Chukha, Trashigang and Samdrup Jongkhar are
identified as giving the highest contribution to overall
multidimensional poverty. The methodology is suggested as a
potential formula for national poverty measurement and for
budget allocation among the districts and sectors.
' Oxford Poverty and Human Development Initiative (OPHI), Oxford
University and Consejo Nacional de Investigaciones Cientificas y
Tecnicas (CONICET)-Universidad Nacional del Sur, Argentina.
** President, The Centre for Bhutan Studies, Thimphu.
Journal of Bhutan Studies
1. Introduction
Fostered by Sen's (1985, 1990, 1999) pioneering 'capabtiity
approach', there is now an increasing consensus that poverty
is an intrinsicaUy multidimensional phenomenon. This has
led scholars to propose different multidimensional poverty
measures. However, some of the proposed measures seem to
have incorporated a multi-dimensional perspective at the cost
of giving up the simplicity and intuition that characterise the
unidimensional measures. Departing from this, Alkire and
Foster (2007) propose a new family of multidimensional
poverty measures which is a variant of the extensively used
Foster, Greer and Thorbecke's (1984) class of one-dimension
poverty measures (FGT from now on). The dimension adjusted
FGT measures keep the simple structure of the one-
dimension case and satisfy a set of convenient properties,
among which decomposabitity across population subgroups
and the possibtiity to break it down by dimension are useful
for policy purposes.
In this paper, the mentioned new class of measures is applied
to estimate multidimensional poverty in Bhutan. Bhutan
constitutes an extremely interesting example of how a country
can define development goals, tailor its policies to these goals,
and see them materialized. Since 1961, the country
implemented coordinated efforts towards development
through consecutive five-years-plans. In particular, the
country has made significant progress in extending the
access to safe drinking water and sanitation, protecting and
managing the country's natural resources, providing basic
health care and increasing the access to primary education.
However, more can sttil be done in some of the mentioned
areas as weU as in others. Within this development agenda,
the MiUennium Development Goals play a key role since
Bhutan is seriously committed to contribute to the realisation
ofthe Millennium Declaration.
In this context, this paper intends not only to present
estimates of multidimensional poverty in Bhutan, which
would complement the income poverty estimates performed
Multidimensional Poverty in Bhutan
by the National Statistics Bureau, but also to suggest the
applied methodology as a potential formula for budget
aUocation among the twenty districts, and within each
district, among the different gewogs, the lowest
administration units.
The data used in this paper correspond to the 2007 Bhutan
Living Standard Survey. It constitutes a unique data source of
this country, representative both at the national and district
levels. Estimations are performed for rural and urban areas
considering five dimensions and also for rural areas
exclusively, with two additional dimensions. Each measure is
also estimated at the district level, and in aU cases, using two
alternative weighting structures: a baseline of equal weights
and another one with weights derived from the ranking of
'sources of happiness' identified through the Gross National
Happiness Survey.
Results confirm that, indeed, income deprivation should not
be the only considered dimension. Deprivation in other
dimensions such as education, access to electricity and room
availabUity in the house, are significant both in rural and
urban areas, and not necessarily related to deprivation in
income. AdditionaUy, deprivation in access to roads is a
significant component of multidimensional poverty in the
rural areas. Land ownership in the rural areas and access to
drinking water in both rural and urban areas, seem to be
relatively less important. It was also found that
multidimensional poverty is mainly a rural problem, which is
particularly important given that 74% of the population in
Bhutan live in rural areas. When analysing at the district
level, it is found that Samtse, Mongar, Chukha, Trashigang
and Samdrup Jongkhar are the five districts with the highest
contributions to aggregate multidimensional poverty.
However, even in the other districts with lower contributions,
improvements in the mentioned dimensions are sttil
important.
Journal of Bhutan Studies
The rest of the paper is organised as follows. Section 2 briefly
revises the literature on multidimensional poverty measures.
Section 3 presents the methodology used in the paper
(measures estimated, data-set used, selected dimensions,
deprivation cutoff values and weighting structures). Section 4
presents the estimation results. Finally, Section 5 contains
the concluding remarks.
2. Literature review
Since Sen (1976), the measurement of poverty has been
conceptualised as foUowing two main steps: identification and
aggregation. In the unidimensional space, the identification
step is relatively an easy one. Even when it is recognised that
the concept of a poverty line-as a threshold that dichotomises
the population into the poor and the non-poor- is somehow
artificial, it is agreed to be necessary. Greater consideration is
given to the properties that should be satisfied by the poverty
index that will aggregate individuals' data into an overall
indicator. However, in the multidimensional context, the
identification step is more complex. Given a set of
dimensions, each of which has an associated deprivation
cutoff or poverty line, it is possible to identify for each person
whether he/she is deprived or not in each dimension.
However, the difficult task is to decide who is to be considered
multidimensionaUy poor.
One proposed approach has been to aggregate achievements
in each dimension into a single cardinal index of weU-being
and set a deprivation cutoff value for the weU-being measure
rather than for each specific dimension to identify the
multidimensionaUy poor. This approach has some practical
drawbacks, in particular, in that it is based on a number of
restrictive assumptions, such as the existence of prices for aU
dimensions. Moreover, it does not agree with the conceptual
framework of the capability approach which considers each
dimension to be intrinsically important. Then, each
dimension with its corresponding deprivation cutoff value
needs to be considered at the identification step of the
multidimensionaUy poor.
Multidimensional Poverty in Bhutan
In this perspective, two extreme approaches have been
traditionaUy used. On the one hand, there is the intersection
approach, which requires the person to be poor in every
dimension under consideration so as to be identified as
multidimensionaUy poor. Clearly, this is a demanding
identification criterion, by which the set of the poor is
reduced as the number of dimensions considered increases,
and may exclude people that are indeed deprived in several
important dimensions. On the other hand there is the union
approach, which requires the person to be poor in at least one
of the considered dimensions. Clearly, with this criterion, the
set of poor increases as the number of dimensions does, and
it may include people that many would not considered to be
multidimensionaUy poor (Alkire and Foster, 2007, pp.8). The
union approach has received important support both in the
theoretical and empirical literature. In particular, Tsui (2002)
and Bourguignon and Chakravarty (2003) adopt it for the
measures they propose.
Tsui (2002) develops an axiomatic framework for
multidimensional poverty measurement (which includes
subgroup consistency) and derives two relative
multidimensional poverty measures, one of which is a
generalization of Chakravarty's (1983) one-dimensional class
of poverty indices, and the other is a generalization of Watt's
(1968) poverty index. He also derives two absolute
multidimensional poverty measures, i
Bourguignon and Chakravarty (2003) distinguish two groups
of multidimensional poverty indices, depending on whether
they consider dimensions to be independent or to have some
substitutability or complementarity. Those that consider
i The distinction between relative and absolute poverty indices is
due to Blackorby and Donaldson (1980). Relative poverty indices are
invariant to changes in scale, such as a doubling of the poverty line
and all incomes, while absolute indices are invariant to translations
or additions of the same absolute amount to each income and to the
poverty line (Foster and Shorrocks, 1991). In practice, relative
poverty indices are the ones that have been most frequently used.
Journal of Bhutan Studies
attributes to be independent satisfy what they call the One
Dimensional Transfer Principle, by which poverty decreases
whenever there is a Pigou-Dalton progressive transfer of the
achievement in some dimension between two poor people. The
progressive nature of the transfer is judged by the
achievements of the two poor people in that specific
dimension, independently of the achievements in the other
dimensions. These indices are additively decomposable. The
second group of indices are non-additive -ie. non
decomposable- and by choosing appropriate values of the
parameters they can reflect either a substitutabitity or a
complementarity relationship between the dimensions. For
both groups of indices, extensions of the FGT class are
proposed.
On a more practice-based perspective, the Unsatisfied Basic
Needs Approach, widely used in Latin America, also uses a
union criterion, identifying as households with unsatisfied
basic needs those that are deprived in one or more of the
selected indicators.
In view of the two prevailing extreme criteria to identify the
multidrmensionally poor, Alkire and Foster (2007) propose a
new identification methodology which, whtie containing the
two extremes, also aUows for intermediate options. Assume
that there are k = l, , d considered dimensions, and that
ci represents the number of dimensions in which individual
/' = 1, ,nis deprived, then an individual is considered to be
multidimensionaUy poor if c. > k . When k = 1, the approach
coincides with the union approach, whereas when k = d, it is
the intersection approach. For 1 < k < d , the identification
criterion lies somewhere in the middle between the two
extremes. Then, for the aggregation step, they use the weU-
known FGT class of poverty indices. The resulting farmly of
measures satisfies a set of convenient properties including
decomposabitity by population subgroups and the possibtiity
of being broken down by dimensions. These last properties
Multidimensional Poverty in Bhutan
make it particularly suitable for policy targeting. Additionally,
the class includes measures that can be used with ordinal
data, which is very common in a multidimensional context. A
detailed description of this class of measures is presented in
Section 3.2.
A final note must acknowledge the probably most popular
multidimensional poverty measure, which is the Human
Poverty Index (HPI), developed by Anand and Sen (1997),
companion index of the Human Development Index (HDI).
Both indices are periodically estimated by the United Nations
Development Programme for all countries to monitor the level
of deprivation and development correspondingly with a
broader perspective than income. The components of the HPI
are survival deprivation (measured by the probability at birth
of not surviving to age 40), deprivation of education and
knowledge (measured by the adult literacy rate) and economic
deprivation (measured by the average of the percentage of
population without access to an improved water source and
chtidren under weight for age). In developed countries the
indicators for each of the components are specified according
to the higher living standards.2 An important advantage of the
HPI is that it only requires macro-data, which can be
especiaUy important for countries in which micro-data
coUection is sttil at its beginnings and its quality is not
assured. However, it has some disadvantages. Clearly, the
three selected dimensions can be argued to be arbitrary as
well as the weighting system used to calculate the measure.
When micro-data sets are available more informative
measures can be calculated, with a higher number of
dimensions and alternative weighting systems.
2 In particular, the survival deprivation is estimated as the
probability at birth of not surviving to age 60, the deprivation of
education and knowledge is defined as adults lacking functional
skills, the economic deprivation is defined as the percentage of
population below 50% of he median adjusted disposable income, and
a social exclusion component is also added, defined as the rate of
long-term unemployment (lasting 12 months or more).
7
Journal of Bhutan Studies
3. Methodology
3.1 Data
The dataset used is the 2007 Bhutan Living Standard Survey
(BLSS) conducted by the National Statistics Bureau (NSB).
There are 9798 households in the sample and 49165 people.
This is the second BLSS performed; the previous one was
done in 2003. Both surveys have foUowed the Living Standard
Measurement Study methodology developed by the World
Bank. However, the 2007 survey has more than doubled the
2003 sample size and it has also extended the coverage, so
that the sample is representative both nationaUy and at each
of the 20 Bhutanese districts (Dzongkhags), in rural and
urban areas.
The unit of analysis to identify the poor is the household.
However, households are weighted by their size (as weU as by
their sample weights), so that results are presented in
population terms. Table A. 1 in the Appendix presents the
composition ofthe sample.
3.2 Multidimensional poverty measures
The poverty measure applied in this paper corresponds to
Alkire and Foster's (2007) family of multidimensional poverty
measures. Before introducing it, it is convenient to clarify
notation in the first place.
Let M"' denote the set of all nxd matrices, and interpret a
typical element y e M"' as the matrix of achievements of n
people in d different dimensions. For every i = l,2,...,n and
J' = 1,2,...,d, the typical entry y.. of y is individual i's
achievement in dimension j. The row vector
yt =(yi\,yi2,----,yidS) contains individual i's achievements in
the different dimensions; the column vector
y ■ = (y1 -,y2 ■,....,y„,-)' gives the distribution of achievements
8
Multidimensional Poverty in Bhutan
in dimension /' across individuals. Let z, > 0 be the
J j
deprivation cutoff value (or poverty line) in dimension j.
FoUowing Alkire and Foster (2007)'s notation, the sum of
entries in any given vector or matrix v is denoted by \v\,
whtie /u(v) is used to represent the mean of v (or | v\ divided
by the number of entries in v).
For any matrix y, it is possible to define a matrix of
deprivations g = [g. ], whose typical element g.. is defined
by gtj = 1 when ytj < z ■, and gtj = 0 when ytj > z ■. That is,
the ij entry of the matrix is 1 when person i is deprived in
dimension j, and 0 when he/she is not. From this matrix,
define a column vector of deprivation counts, whose ith entry
c. =| gt | represents the number of deprivations suffered by
person i. If the variables in y are cardinal, then a matrix of
normalised gaps g = [g. ] can be defined, where the typical
element g]}.= (z}.- yi}.) / z}. when yij 0,
whose typical element g" is the normalised poverty gap
raised to the a-power.
The methodology to identify the multidimensionaUy poor
proposed by Alkire and Foster (2007) compares the number of
deprivations with a cutoff level k When each selected
dimension has the same weight, the possible values of k go in
the range of k— \ ,d. However, the methodology also
aUows other weighting systems, which will be explained at the
end of the section. In general, for any weighting system, let
pk be the identification method such that pk (yi, z) = 1 when
Journal of Bhutan Studies
c. > k , and pk (yt, z) = 0 when ct < k . That means that an
individual is identified as multidimensionaUy poor if he/she is
deprived in at least k dimensions. This methodology is said to
be a dual cutoff method, because it uses the within dimension
cutoffs z ■ to determine whether an individual is deprived or
not in each dimension, and the across dimensions cutoff k to
determine who is to be considered multidimensionaUy poor. It
is also presented as a counting approach, since it identifies
the poor based on the number of dimensions in which they
are deprived. When equal weights are used, when k — \ , the
identification criterion corresponds to the union approach,
whereas when k — d, the identification criterion corresponds
to the intersection approach. This identification criterion
defines the set of the multidrmensionally poor people as
Zk — {/ : pk {yi; z) = 1} . Once identification is applied, a
censored matrix g (k) can be obtained from g by replacing
the ith row with a vector of zeros whenever pk {yi, z) = 0.
Matrix ga (k) can be defined analogously for a > 0, with its
typical entry g" (k) = g" if i is such that c. > k, whtie
g^ (k) = 0 if z is such that c. < k .
A first natural measure to consider is the percentage of people
that are multidimensionaUy poor: the multidimensional
Headcount Ratio H = H(y,z) defined byH = ql n , where q is
the number of people in set Zk. This measure is the
analogous to the unidimensional Headcount Ratio, and it has
the advantages that it is easy to compute and understand,
and that it can be calculated with ordinal data. However, it
suffers from the disadvantages first pointed by Watts (1969)
and Sen (1976) for the one-dimensional case, namely, being
insensitive to the depth and distribution of poverty, violating
monotonicity and the transfer axiom. Moreover, in the
multidimensional context, it also violates what Alkire and
10
Multidimensional Poverty in Bhutan
Foster (2007) call dimensional monotonicity: if a poor person
becomes deprived in an additional dimension (in which
he/she was not previously deprived), Hdoes not change.
Considering this, Alkire and Foster (2007) propose the
dimension adjusted FGT measures, given by
Ma (y; z) = ju(ga (k)) for a > 0 . When a = 0, the measure is
the Adjusted Headcount Ratio, given by
M0 = ju(g (k)) = HA, which is the total number of
deprivations experienced by the poor (| c(^) |=|