-
NiCE Working Paper 12-107
Version 2
December 2013
The International Wealth Index (IWI)
Jeroen Smits
Roel Steendijk
Nijmegen Center for Economics (NiCE)
Institute for Management Research
Radboud University Nijmegen
P.O. Box 9108, 6500 HK Nijmegen, The Netherlands
http://www.ru.nl/nice/workingpapers
-
1
Abstract In this paper we present the International Wealth Index
(IWI), the first strictly comparable asset based index for
household’s long-term economic status that can be used for all low
and middle income countries. IWI is similar to the widely used
wealth indices included in the Demographic and Health Surveys and
UNICEF MICS surveys, but adds the property of comparability across
place and time. IWI is based on data from 2.1 million households in
97 developing countries. With IWI we provide a stable and
understandable yardstick for evaluating and comparing the economic
situation of households, social groups and societies across all
regions of the developing world. A household’s ranking on IWI
indicates to what extent the household possesses a basic set of
assets, valued highly by people all across the globe. IWI is tested
thoroughly for reliability and validity. National IWI values are
highly correlated with the Human Development Index, life
expectancy, national income and educational outcomes and IWI-based
poverty measures are highly correlated with Poverty Headcount
Ratios. Jeroen Smits is director of the Global Data Lab
(www.globaldatalab.org) and associate professor Inequality and
Development at the Nijmegen Center for Economics. Roel Steendijk is
consultant at Steendijk Statistics (www.steendijk-statistics.nl).
We are grateful to MeasureDHS, the UNICEF MICS department, the Pan
Arabic Project for Family Health (PAPFAM), the Integrated Public
Use Microdata Series (IPUMS) department of the Minnesota Population
Center, the National Statistical Offices of Brazil, Chile, Costa
Rica, Sudan, Uruguay and Venezuela, the Statistical Information and
Monitoring Programme on Child Labour (SIMPOC) of ILO-IPEC, and the
Carolina Population Center at the University of North Carolina at
Chapel Hill for making the datasets available that have been used
in this project. Contact: Jeroen Smits, Global Data Lab, Nijmegen
Center for Economics, Radboud University Nijmegen. PO Box 9108,
6500HK Nijmegen, The Netherlands, phone +31 24 3612319/5890
[email protected], [email protected] Website International
Wealth Index: iwi.globaldatalab.org
-
2
1. Introduction Since the late 1990s, wealth indices have become
widely used instruments for measuring economic status of households
in low and middle income countries. Hundreds of research papers
have appeared in which wealth indices were used for studying
variation in health, mortality, poverty, education, work and other
outcomes in almost all countries of the developing world (e.g.
Gwatkin et al., 2007; Howe et al., 2008; Filmer & Scott, 2012;
Falkinham & Namazie, 2002). Wealth indices are considered
effective indicators of long-term socio-economic position, living
standard or material well-being of households (Filmer &
Pritchett, 1999, 2001; Sahn & Stifel, 2000, 2003; McKenzie,
2005; Howe et al., 2008). They often perform as well or better than
expenditure data in explaining variation in education, child
mortality, nutrition, fertility and health care use (Filmer &
Pritchett, 2001; Bollen et al., 2002; Sahn & Stiefel, 2003;
McKenzie, 2005; Filmer & Scott, 2012). Important reasons for
the success of these indices are their ease of computation,
intuitive appeal, and their wide availability in household surveys
for developing countries like the Demographic and Health Surveys
(DHS) and UNICEF MICS surveys. Also the fact that the required data
can be more reliably measured than those needed for computing
income or expenditure measures, the most obvious alternatives, has
contributed to their success (Sahn & Stifel, 2003; McKenzie,
2005; Filmer & Scott, 2012). In spite of these positive
properties, wealth indices suffer from one great disadvantage: they
are not comparable among countries and time points (McKenzie, 2005;
Gwatkin et al., 2007). For each survey usually a separate wealth
index is constructed on the basis of the assets available in the
survey data. Such a separate index is tailored completely towards
the specific wealth distribution in the survey year in the country
on which it is based. This means that it is a valid indicator of
wealth differences in that specific country-year combination, but
-- as the wealth distributions in other country-year combinations
generally will be different – cannot be used to study wealth
differences in other countries and years. The scores on
survey-specific wealth indices are therefore interpreted as
relative wealth levels (Rutstein & Johnson, 2004; Gwatkin et
al., 2007). In most applications the wealth distribution is divided
into quintiles, with the lowest 20 percent of the population
defined as the poor and the upper 20 percent as the rich. For
analyzing within country inequalities in education, health, or
other outcomes, comparing the lowest and highest wealth quintiles
does indeed make sense. However, cross-national or cross-temporal
comparisons of groups with similar levels of wealth or poverty are
not possible with these relative measures, as the average wealth
level of the wealth quintiles differs among countries and years. To
solve this problem, a general wealth index is needed that uses the
same criteria for rating households independent of country or year.
The International Wealth Index (IWI) is such a general wealth
index. Whereas other wealth indices are constructed on the basis of
data from one or a restricted number of household surveys, IWI is
based on data derived from 165 household surveys, held between 1996
and 2011 in 97 low and middle income countries. Together these
surveys included information on 2.1 million households, covering
all regions of the developing world. Using this broad database, IWI
was constructed in the same way as most other wealth indices.
Information on households’ possession of consumer durables, access
to basic services and housing characteristics was entered into a
principal component analysis (PCA), from which the asset weights of
the first component were derived. These asset weights were
subsequently brought together into the IWI formula, which
constitutes the basic instrument for providing households with an
IWI value.
-
3
2. The International Wealth Index (a) Material well-being The
central idea behind IWI is that households across the globe can be
placed on an underlying dimension of material need satisfaction (or
living standard) for which we will use the term “material
wellbeing”. This dimension runs from a situation in which a
household has no possessions at all that may help satisfy their
material needs, to a situation in which the household possesses all
assets that are broadly considered necessary for living an easy and
comfortable life. The kinds of assets that are most relevant for a
household’s material well-being depend on the household’s economic
situation. For very poor households, material well-being is
associated with the satisfaction of the basic needs of food,
clothing and safety/shelter, which have to be met to survive. One
step higher, material well-being refers to the possession of goods
and access to basic services that make life easier and more
comfortable. There are all kinds of relatively cheap utensils that
reduce the workload people have (pots, pans, plates, cutlery,
tools) or make it more comfortable (tables, chairs, carpets, beds).
A major step is made when the household gets access to electricity,
because this opens up infinite new possibilities for increasing
material well-being in relatively cheap ways. With electric light,
the time that can be spent on useful and leisure activities
increases considerably. A refrigerator reduces daily shopping time.
Electric tools and utensils reduce time spent on cooking and on
work around the home. If the household gets access to clean water,
the workload is reduced even more, as this may save an often
considerable amount of time spent on fetching water. The quality of
the house in which the household lives is another an important
aspect of material well-being. The kind of building and flooring
material determines how much maintenance there is to the house,
whether rain, wind and pests are kept outside well, and how
comfortable the house is. Having more than one room, a separate
kitchen and bathroom, and a decent in-house toilet facility greatly
enhances quality of living. Besides by technical equipment that
makes life easier, material wellbeing can also be improved by means
of transportation and communication equipment. With a bike, cart,
boat, motorbike or car transportation of heavy loads becomes easier
and travelling time is reduced. Radio and TV bring the world into
the home and phones, computers and the internet greatly enhance
communication and access to information. Given that everywhere in
the world households tend to buy the assets and ask for the basic
services mentioned above, it seems that there is some kind of
globally shared consensus about the material requirements needed
for living a decent life. IWI is meant to tap into this dimension
of material need satisfaction and to indicate the degree to which a
household’s material basic needs are met. (b) Measuring material
well-being Like other asset based wealth indices, IWI measures a
household’s level of material well-being by looking at the
household’s possession of durables, access to basic services, and
characteristics of the house in which it is living. Households that
own more expensive durables, have a better quality house, and have
access to basic services are considered to have a higher level of
material well-being than household with less expensive durables,
worse housing and no access to services. Any household for which
the required asset information is available can be given a value on
IWI and any household with the same combination of assets obtains
the same IWI score. The IWI scale is additive. If a household owns
a specific durable, has better access to public services, or has a
higher value on a housing characteristic, its IWI value is raised
by a specific amount (the re-scaled asset
-
4
weight). The IWI scale runs from 0 to 100. If a household has
all durables and highest quality housing and services, its IWI
value is 100. If it has none of the durables and lowest quality
housing and services, its IWI value is 0. Households with the same
value on the IWI scale are assumed to have reached the same level
of material need satisfaction. This does not mean that they own
exactly the same assets. Depending on individual preferences and
the context a household is living in, households may reach the same
level of material need satisfaction with different portfolios of
assets. Ownership of a phone increases a household’s value on the
IWI scale to the same extent as having a high quality instead of a
medium quality toilet facility. A household’s level of material
well-being is closely related to the household’s economic
situation. The assets required to satisfy material needs come at a
price, hence wealthier households have more possibilities to
satisfy these needs. This close relationship between material and
economic well-being has stimulated the strong growth in the use of
information on asset ownership to indicate the economic welfare of
households (e.g. Filmer & Pritchett, 1999; Sahn & Stiefel,
2003; Gwatkin et al., 2007; Howe et al., 2008). Table 1 presents an
overview of the assets on which IWI is based, their raw weights and
the coefficients to be used in the IWI formula. The assets include
seven consumer durables (possession of a TV, refrigerator, phone,
bicycle, car, cheap utensil and expensive utensil), access to two
public services (water and electricity) and three housing
characteristics (number of sleeping rooms, quality of floor
material and toilet facility). This set of assets was selected
because of its wide availability in household surveys and because
it differentiates well across the wealth range needed for a wealth
index covering the complete developing world. (c) Clumping and
truncation Differentiating well between wealthier and less wealthy
households is important to prevent clumping and truncation, two
problems of which asset based wealth indices may suffer (McKenzie,
2005; Vyas & Kumaranayake, 2006). Clumping (or heaping) means
that there are many households with the same asset combinations,
leading to a high percentage of cases in the same category.
Clumping can be prevented by including more assets in the index. As
IWI includes eight two-category (yes-no) items and four
three-category (low, middle, high) items, the total number of
possible combinations is over 20,000. Even though many of these
combinations are less likely (having a car and flush toilet, but a
floor of earth), the number of likely combinations is so big that
no clumping is expected. Truncation of the wealth distribution
means a lack of discriminative power at the top or bottom end of
the scale. This is a problem that cannot completely be prevented,
because the wealth range covered by the index is restricted by the
number and values of the included assets. However, by choosing the
included assets strategically, an index can be computed that allows
for enough differentiation at the top and bottom end of the scale
to prevent excessive truncation. For IWI this was done by including
both at the top and at the bottom of the distribution enough assets
to differentiate among households. Households with an IWI value of
100 have a TV, fridge, phone, car, a house with piped drinking
water, electricity, a flush toilet, good quality floor material and
three or more rooms. These households have reached a standard of
material well-being that, even from the perspective of a
high-income country, can be considered very reasonable. For an
index meant to measure household wealth in developing countries,
more discriminative power at the top of the scale does not seem
necessary.
-
5
Households with an IWI value of 0, on the other hand, own none
of the included items -- not even a cheap utensil like a chair,
watch or radio -- have a floor of earth or dung, have no or bad
quality toilet, no electricity, only one room, and water from an
unprotected source. From any reasonable perspective these
households are considered to be extremely poor. Differentiating
further within this group would probably be possible (e.g. by
including nutritional items), but their situation is already so
miserable that from a policy perspective it does not seem relevant
to subdivide them further. As this group falls below any reasonable
poverty line, policies should focus on improving the situation of
all of these households.
Table 1: Mean and standard deviations of asset indicators, raw
asset weights, and coefficients of IWI formula (N=2189221)
Consumer durables Mean Std. Deviation Raw indicator weight IWI
Formula
weight Television 54.25 49.82 0.798552 8.612657 Refrigerator
36.99 48.28 0.781531 8.429076 Phone 38.74 48.72 0.660869 7.127699
Car 11.68 32.12 0.431269 4.651382 Bicycle 29.12 45.43 0.171238
1.846860 Cheap utensils 74.48 43.60 0.381851 4.118394 Expensive
utensils 28.16 44.98 0.603345 6.507283 Housing characteristics
Floor material: Low quality 34.97 47.69 -0.700809 -7.558471 Medium
quality 36.08 48.02 0.113815 1.227531 High quality 28.95 45.35
0.566271 6.107428 Toilet facility: Low quality 40.13 49.02
-0.689810 -7.439841 Medium quality 17.57 38.06 -0.101100 -1.090393
High quality 42.29 49.40 0.754787 8.140637 Number of rooms: Zero or
one 38.44 48.65 -0.343028 -3.699681 Two 32.64 46.89 0.035609
0.384050 Three or more 28.92 45.34 0.319416 3.445009 Public
utilities Access to electricity 62.30 48.46 0.747001 8.056664 Water
source: Low quality 32.13 46.70 -0.584726 -6.306477 Medium quality
23.85 42.62 -0.213440 -2.302023 High quality 44.02 49.64 0.737338
7.952443 Constant 25.004470 Minimum value -2.318374 0 Maximum value
6.953466 100
-
6
The foregoing does not preclude that in very poor or very rich
countries the majority of households is concentrated at the lower
or upper end of the distribution. But that is precisely what can be
expected of a comparable indicator of the economic situation of
households and groups. In very poor countries, there are many
households at the bottom of the wealth distribution, independently
of whether their economic situation is measured by income,
expenditure, or an asset-based wealth index.
In wealthier countries, the reverse is true. However, there the
situation is somewhat more complex. Whereas monetary income can
increase infinitely, the number of assets that can be included in a
wealth index is limited. When the basic needs of a household are
met, the range of more luxurious goods and services on which
additional income can be spent is so wide that it becomes
practically impossible to include them all in a questionnaire. The
use of a wealth index is thus restricted to countries where for a
substantial number of households not all basic needs are met. In
practice this means that IWI can be used for all low income
countries and the majority of middle income countries.
3. Constructing IWI To be able to create a comparable wealth
index like IWI, several important choices have to be made. First,
the number and type of assets to be included in the index have to
be chosen. Second, the number of datasets and countries on the
basis of which the index will be computed has to be decided on.
Third, a choice has to be made on how these countries should be
weighed when constructing the index, as some of the countries are
much larger than others. Fourth, the method to be used for
computing asset weights has to be chosen. In the next sections,
these choices and their outcomes are discussed in detail.
(a) Number of assets and number of surveys
A major challenge in constructing a comparative wealth index is
to find a reasonable compromise between number of surveys and
number of assets. Because the number of asset questions used in
surveys is restricted, and the type of asset on which information
is collected varies among surveys, including more assets in the
index would mean that less surveys could be used. Nevertheless, a
reasonable set of assets had to be included to reduce the risk of
clumping and truncation. This asset set should preferable include
assets from different domains of household needs, like household
chores, transport, communication, access to basic services, and
hygiene.
The compromise we came upon was the use of a series of twelve
assets, including seven consumer durables, three housing
characteristics, and access to two public services. With this set
of assets, we could compute IWI on the basis of data from 165
national representative household surveys. The consumer durables
included are the possession of a TV, refrigerator, phone, bicycle,
car, a cheap utensil and an expensive utensil. The housing
characteristics are the number of sleeping rooms, quality of the
floor material and quality of the toilet facility. The basic
services are access to clean water and electricity.
(b) Measurement
The consumer durables included in the construction of IWI are
measured with two-category variables. These variables have value
‘1’ if the household or one of its members owns the durable and
value ‘0’
-
7
if this is not the case. A similar two-category variable is used
to indicate whether (1) or not (0) the household has access to
electricity.
Quality of water supply, of floor material and of toilet
facility are measured with three categories: (1) low quality, (2)
middle quality, and (3) high quality. For the number of sleeping
rooms also a three-category variable is used: (1) zero or one
sleeping rooms, (2) two sleeping rooms, and (3) three or more
sleeping rooms. Zero and one rooms are combined, because it is in a
substantial number of surveys not possible to distinguish between
households that have one sleeping room and households that only
have one room, and hence use the living room for sleeping.
The categories of the ‘quality’ variables need further
explanation. Floor material, water sources and toilet facilities
may differ among countries, depending on local availability and
traditions. In the survey data for different regions, therefore,
different categories may be used. For constructing a comparative
wealth index, however, it is necessary that a variable is measured
with the same categories in each survey. To solve this problem we
recoded the substantial categories used in the different surveys
into the three general quality categories. In doing so, the
following guiding principles were followed:
Water supply: - high quality is bottled water or water piped
into dwelling or premises;
- middle quality is public tap, protected well, tanker truck,
etc;
- low quality is unprotected well, borehole, spring, surface
water, etc.
Toilet facility: - high quality is any kind of private flush
toilet;
- middle quality is public toilet, improved pit latrine,
etc.;
- low quality is traditional pit latrine, hanging toilet, or no
toilet facility.
Floor quality: - high quality is finished floor with parquet,
carpet, tiles, ceramic etc.;
- middle quality is cement, concrete, raw wood, etc.;
- low quality is none, earth, dung etc.
The variables ‘cheap utensils’ and ‘expensive utensils’ need
further explanation, as these are constructed variables. Early
experience with the use of wealth indices revealed a lack of
discriminatory power at the lower end of the wealth distribution
(Rutstein, 2008). To be able to better differentiate among the
poorest groups, some cheaper assets, like having a chair, table,
clock, watch, water cooker, radio, fan or mixer were included in
later surveys. However, the kind and number of these cheap asset
questions varies considerably among surveys. It is therefore not
possible to include them as separate items in a comparative wealth
index.
As an alternative we have created a more general indicator
‘cheap utensils’ that is based on the information on any cheap
(roughly under 50 US Dollar) item that is present in the data. This
indicator can be created if information for one or more of such
items is available. Household owning one or more cheap utensils get
value ‘one’ and other households value ‘zero’ on this
indicator.
The indicator ‘expensive utensils’ is meant to create more
discriminatory power at the upper end of the wealth distribution.
It is constructed in a similar way as the cheap utensils variable,
but with respect to the possession of expensive (roughly over 300
US Dollar) items, like having a washer, dryer, computer, motorbike,
motorboat, airconditioner, or generator. If information on the
possession of at
-
8
least one of these items is available, the indicator ‘expensive
utensils’ can be created by giving households owning one or more
expensive utensils the value ‘one’ and other households the value
‘zero’.
The inclusion of the cheap and expensive utensil indicators in
the construction of IWI introduces some variation among countries
and time points. However, we consider the increased discriminatory
power at the upper and lower end of the scale more important than
this loss of generalizability. As our test analyses later will
show, the removal of either indicator has hardly any influence on
the overall performance of the index.
(c) Survey datasets
The major data sources used for the construction of IWI are the
Demographic and Health Surveys (DHS), funded by USAid and collected
under responsibility of Measure DHS (www.measuredhs.com), and the
Multiple Indicator Clusters Surveys (MICS) collected by UNICEF
(www.childinfo.org). Other data sources are World Health Surveys
(WHS) collected under supervision of the World Health Organization
(www.who.int/healthinfo/survey), the Integrated Public Use
Microdata Series (IPUMS) of the Minnesota Population Center
(international.ipums.org), the Pan Arabic Project for Family Health
(PAPFAM) surveys, with the League of Arab States (www.papfam.org)
as major sponsor, the Statistical Information and Monitoring
Programme on Child Labor (SIMPOC) surveys of ILO-IPEC
(www.ilo.org/ipec), and the 2004 Chinese Health and Nutrition
Survey (www.cpc.unc.edu/projects/china).
We aimed to use at least two surveys for each country. Our
preferred choice among the data sources were DHS and MICS surveys,
because these are large series of comparable datasets of high
quality, including a substantial number of assets variables. Only
if for a country none or only one DHS or MICS survey was available,
we opted for other sources. Given this preference, we ended up with
a database with combined data from 165 surveys held between 1996
and 2011 in 97 low and middle income countries. Of these surveys,
99 were DHS, 36 MICS, 16 WHS, 7 IPUMS, 3 PAPFAM, 3 SIMPOC and one
Chinese Health and Nutrition Survey. The total number of included
households was 2,189,221. Information on the data sources and
country-year combinations used is presented in Appendix A.
To get an as broad as possible coverage of the developing world,
in some cases datasets with a missing item were accepted. The
missing items were replaced by the values for another item, or by
zero or one, depending on what was most likely (e.g. electricity is
1 for relatively developed countries as Malaysia or Uruguay). Test
analyses revealed that this procedure had a negligible effect on
the final index. Information on where this happened is available in
Appendix A.
(d) Weighing the countries
An important issue that has to be decided on when constructing a
comparable wealth index is how to handle the differences in
population size among countries. The countries included in our
database range from small states like Belize, Suriname, and Sao
Tome Y Principe, with population sizes below one million, to India
and China with over one billion inhabitants. Weighing them by
population size
-
9
does not seem a good choice, as IWI then would be almost
completely determined by India and China. However, weighing the
countries equally, the most obvious alternative, also is
problematic, as it does not seem reasonable to let the smallest
countries have the same influence on IWI as countries with a
thousand times larger population.
To solve this issue, we have estimated two test versions of IWI,
one with the countries weighed by population size and one with the
countries weighed equally. It turned out that the wealth indices
derived from these opposite approaches were very much alike. The
Pearson correlation between the two indices was 0.999, thus
indicating that IWI is very robust to differences in the way the
countries are weighed.
Given that both extremes have their drawbacks and that there is
hardly any differences between the wealth indices computed with
each of them, we decided to opt for an intermediate position.
Instead of weighing the countries by population size or weighing
them equally, we have weighed them by the square root of the
population size. This means that the larger countries weigh heavier
in computing IWI, but that the difference is much smaller than when
absolute population weights would have been used. With our square
root weights, the difference in influence between the largest and
smallest country is a factor 78, whereas it was a factor 6000 for
absolute population weights. The correlations between the IWI
version based on the square root weight and the versions based on
population weight and equal weight are over .999.
If in the datasets case weights were available to create
representative country samples, these weights were also used when
constructing IWI.
(e) Computing the asset weights
The easiest way to compute an asset index is to add up the
number of assets a household owns (McKenzie, 2005). This method,
which has been used in some earlier studies (e.g. Montgomery et
al., 2000; Guiley & Jayne, 1997), has the disadvantage that it
weighs each item equally. This would imply that possession of a
table, a bicycle, a car, or a flush toilet would contribute equally
to a household’s wealth score. As this obviously is not realistic,
it is recommendable to use a more advanced method to determine the
relative weights of the assets included in a wealth index.
Since the landmark papers of Filmer an Pritchett (1999, 2001),
almost all asset based wealth indices have used principal component
analysis (PCA) for computing the asset weighs. There have been a
few attempts to use other techniques for this purpose, but the
outcomes differed very little from those using PCA (Booysen et al.,
2008, multiple correspondence analysis; Sahn and Stifel, 2000,
factor analysis). In line with the tradition in the field, we
therefore have chosen to use PCA for estimating the weights.
PCA is a multivariate statistical technique that can be used to
reduce the number of variables in a dataset by converting them into
a smaller number of components; each component being a linear
weighted combination of the initial variables (Vyas and Kumaranayka
2006). The first component, which explains the largest part of the
variation in the data, is chosen as the wealth index (Filmer and
Pritchett 2001, Sahn and Stifel 2003, McKenzie, 2005). For IWI,
this first factor explains 30 percent of the variation in the
assets, which is somewhat higher than the percentages generally
obtained using
-
10
country-specific indices (26% Filmer & Pritchett, 2001; 27%,
McKenzie, 2005; 24-27%, Cordova,2008 ).
PCA estimates a weight for each initial variable, and these
estimated weights form the basis for computing the wealth index.
The weights reflect the possibility that a household that owns one
specific asset also owns one of the other assets in the analysis.
The estimated indicator weights for IWI are presented in Column 3
of Table 1. We call them ‘raw’ indicator weights to distinguish
them from the rescaled weights used in the final IWI formula.
As can be seen in Table 1, more valuable assets do not necessary
have a higher weight than cheaper assets. This is because the
weight of an asset indicates the increase in the household’s IWI
value on top of the contribution of the other assets. Because
households owning a car also possess most of the other assets, even
with a relatively small weight for car ownership their IWI values
will be higher than those of households without a car.
Because there is some discussion in the literature about the use
of PCA for discrete data like our asset indicators (Sahn &
Stifel, 2000; Howe et al., 2008; Booysen et al., 2008), we have
repeated our analysis using categorical PCA (with SPSS CAPCA),
Multiple Correspondence Analysis, and Factor Analysis. These
analyses gave weights that were the same up to the 8th decimal.
(f) Scaling IWI
The raw indicator weights provided in Table 1 show the relative
contribution of each asset to a household’s wealth score. On the
basis of these weights we computed a raw wealth score, which we
subsequently rescaled to the 0-100 range. To obtain the raw wealth
score, the asset weights multiplied by the asset indicator
variables had to be summed up. This led to the following equation,
where is the raw wealth score, the estimated indicator weight of
the Th asset and the indicator variable of the Th asset.
∙
When applying this formula to our data, we obtained a household
wealth distribution with a minimum score of -2.318 and a maximum
score of 6.953. Households with the minimum score are those with
the lowest value on all assets items. They own none of the consumer
durables, have no electricity, lowest quality water supply, floor
material and toilet facility and no more than one sleeping room.
Households with the maximum score own all consumer durables, have
electricity, highest quality water supply, floor material and
toilet facility, and have three or more sleeping rooms.
To obtain a scale with a more intuitively understandable range,
we transformed the wealth distribution to the range 0-100, with an
IWI value of 0 for households having none of the durables, no
access to public services and lowest quality housing and the value
of 100 for households having all durables and highest quality
services and housing. This new scale was created by adding the
opposite of the lowest value (2.318) to each household score, to
put the minimum at ‘0’. The maximum value
-
11
then becomes 9.271 (=6.953+2.318). To put the maximum at 100,
the scale values are multiplied by 100 and divided by the (new)
maximum. Hence:
100 ∙ ∑ ∙ 2.3189.271 25.004′ ∙
Where ′ are the rescaled asset weights. These are obtained by
multiplying the original weights by 10.785. Together with the
constant 25.004, the rescaled asset weights make up the IWI
formula. The exact values (with six decimal places) of the constant
and the rescaled weights are presented in the fourth column of
Table 1.
Unlike income and consumption expenditure data, asset based
wealth indices are generally not adjusted for household size. The
reason is that the assets used for constructing these indices
consist almost completely of household public goods (Filmer &
Scott, 2012). Housing characteristics, access to basic services and
durables like a TV, fridge, clock or car tend to benefit all
household members (Filmer & Pritchett, 2001; Rutstein &
Johnson, 2004; McKenzie, 2005; Howe et al., 2008). Empirical
comparisons of unadjusted and size-adjusted wealth indices showed
either no substantial differences or less plausible results for the
adjusted indices (Sahn & Stifel, 2000; Rutstein & Johnson,
2004; Howe et al., 2008; Filmer & Scott, 2012). We therefore
have chosen not to adjust IWI for household size.
After computing a household’s IWI score, it is rounded to one
decimal place to prevent the number of IWI combinations from
becoming unrealistically large. Although the number of possible
asset combinations is over 20,000 (28+34), many of those
combinations are very unlikely (having a car and a flush toilet,
but a floor of earth). An index with 1000 unique values is rich
enough to address this variation. The IWI formula and help files to
compute IWI are made available at the IWI website
iwi.globaldatalab.org.
4. Testing IWI Now we have constructed IWI, we would like to see
what its distribution looks like and conduct a number of
performance tests. In those tests, we first determine to what
extent IWI depends on the inclusion of specific assets in the
index. This is done by comparing the IWI version based on all
assets with IWI versions with one or more of the assets removed.
Second, we test to what extend IWI depends on data from specific
parts of the world or time periods, by correlating the original IWI
with reduced IWI versions, computed on datasets with one of the
world regions removed, or for specific years. Third, we test for
households within a specific world region to what extent their IWI
values are correlated with their values on a wealth index computed
on data for only that world regions. Fourth, we compare IWI values
for households in the DHS datasets with the values of those
households on the original DHS wealth index.
-
12
(a) Distribution of IWI
The distribution of IWI is displayed in Figure 1. As can be
seen, the 2.1 million households on which IWI was constructed are –
except for some overrepresentation at the lower end of the
distribution -- more or less regularly distributed across the IWI
scale.
The clumping at the lower end of the distribution is a common
feature of all asset based wealth indices. It is caused by the fact
that the first and last steps on the scale cannot be smaller than
the lowest asset weight. In the case of IWI, this lowest asset
weight is the weight of 1.8 for having a bicycle. Hence after the
households with a value of 0, the first possible IWI score is 1.8.
The next possible score is 4.0, for households that have middle
quality water (protected public source) and none of the other
assets. Then comes the value 4.1, for households possessing only a
cheap utensil (like a watch, fan, or radio) and none of the other
assets. Together these households with value 4.0 and 4.1 create the
high spike of 107,103 respondents at the (rounded) 4 score in
Figure 1.
Above value 5, the number of possible combinations increases
rapidly and there are no large empty spaces any more until above
value 95, where we observe a similar phenomenon as between 0 and 5:
Below the households with an IWI value of 100 come the households
lacking only a bicycle. Given the bicycle weight of 1.8, these
households obtain an IWI score of 98.2. Next come the households
who possess all assets but have two instead of three sleeping
rooms, with an IWI score of 96.9, etc. This clustering at the
extremes, from which all asset-based indices suffer, is not very
problematic. Households with an IWI value of, say, under 10, are
extremely poor. In most practical applications, these households
therefore will be combined. The same is probably true for the
households with a very high IWI value.
Figure 1 also reveals that besides this clumping at the
extremes, the observations are rather evenly distributed across the
further range of the index (say between 10 and 90). As there is in
that range no intermediate category containing more than a few
percent of the households and there are neither at
0
20000
40000
60000
80000
100000
120000
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
100
Figure 1. Distribution of IWI (N= 2,189,221)
-
13
zero nor at 100 indications of truncation (only 3% of cases has
value zero and only one percent has value 100), we can conclude
that with regard to truncation and clumping IWI performs well.
(b) Dependency on specific items
To find out to what extent IWI depends on the inclusion of
specific assets in the index, we have computed twelve reduced IWI
versions, each with one of the assets removed from the index. For
this purpose, twelve new PCA analyses were conducted on our
database, leading to twelve reduced IWI formulas. The reduced
wealth indices created with these formulas were all scaled to run
from 0 to 100, in the same way as this was done for IWI. In Table
2, Pearson correlations between the reduced indices and the
original IWI are presented.
Table 2. Pearson correlations between IWI based on all 12 assets
and reduced IWIs based on 11 assets IWI without: Correlation -
water 0.986 - toilet 0.987 - rooms 0.996 - floor 0.991 -
electricity 0.996 - TV 0.996 - refrigerator 0.996 - phone 0.996 -
car 0.999 - bicycle 0.999 - cheap utensil 0.999 - expensive utensil
0.997
The correlations show that removing one item has not much effect
on the wealth index that is produced. The weakest correlations,
those for the three-category items ‘quality of water supply’ and
‘quality of toilet facility’, are .986 and .987. This is such a
strong correlation that there seems to be hardly any difference
between the original and reduced versions of IWI. The correlations
between the other reduced versions and IWI are with .99 and over
even higher. Hence we can conclude that IWI is robust against
removal of an asset. This also implies that for datasets in which
only eleven of the twelve assets are available, computing IWI on
the basis of those 11 assets will provide a good approximation of
IWI based on 12 assets.
We also have tested the effect of removing any combination of
two assets from the index. For the combination without water supply
and toilet facility, the correlation of the reduced index with IWI
was .961. Without water supply and floor material and without
toilet facility and floor material it was .972. Removing water
supply or toilet facility with any other asset gave correlations of
.975 or
-
14
higher. For any other combinations of two assets removed, the
correlation became in the order of .99 or higher.
We can thus conclude that even with two assets removed, the
reduced wealth indices rate households almost the same as IWI. This
means that even in situations where two assets are lacking in a
dataset, we still can approach the households’ real IWI score very
well by applying the formula for the reduced index on the data. The
approximation will be best if information on water supply and
toilet facility is available in the data, given that removal of
these assets has the largest influence on the IWI scores. The
reduced IWI formulas are made available through the IWI website:
iwi.globaldatalab.org.
(c) Dependency on specific regions or time period
To find out whether IWI depends strongly on the data for a
specific region of the developing world, we have created four
reduced versions of IWI on the basis of our database. For each of
these versions, the data for a specific global region – Latin
America, sub-Saharan Africa, Middle East and North Africa (MENA),
and Asia without the Middle East – were removed and a PCA analysis
was performed on the reduced database. The reduced IWI versions
were subsequently applied to all countries in our database. It
turned out that the reduced IWI versions were all very high (over
.999) correlated with the original IWI (Table 3, middle column).
This result makes clear that the data from none of these global
regions exert an unreasonable strong influence on IWI.
Table 3: Pearson correlations between IWI and wealth indices
with data for a global region removed and between IWI and
region-specific wealth indices
Global region Region removed Region-specific
index
Asia 0.999 0,998
Sub-Saharan Africa 0.999 0,996
Latin America 0.999 0,998
Middle East and North Africa (MENA) 0.999 0,999
We also wanted to know to what extent the IWI values for a
specific region of the developing world are similar to those of a
reduced wealth index, computed only on the data for that region. We
therefore have constructed four regional wealth indices, for Latin
America, sub-Saharan Africa, the MENA region and Asia (without
Middle East), by running PCA analyses on the data for these
regions. Table 3 (right column) shows that the regional wealth
indices constructed in this way were highly correlated with IWI for
these regions. Pearson correlations were .998 for Latin America and
Asia, .996 for sub-Sahara Africa and .999 for the MENA region.
Hence regional indices hardly perform better than IWI for their own
regions.
-
15
To test the degree to which IWI is influenced by the time period
for which data is used, we have split our database into three time
periods: 1996-2000, 2001-2005, 2006-2011. For each time period a
separate wealth index was constructed. The Pearson correlations
between these separate indices and IWI turned out to be .999, .999
and .997 respectively. We therefore can conclude that, within the
time range of our data, IWI is hardly influenced by the period for
which data is used.
(d) Association with national wealth indices
To test the performance of IWI further, we have compared the IWI
distributions within countries with those of the country-specific
wealth indices available in the DHS household surveys. Because IWI
in the first place is meant to be a comparable wealth index, it was
not developed specifically to fit the wealth distribution of an
individual country. Still, the principle behind the index – rating
households on the basis of their assets – is similar to that behind
the country-specific indices. Hence we would expect IWI to rate the
households within countries more or less in the same way as the
country-specific indices; the scores on both indices should be
positively correlated. As the country-specific indices are
generally based on all assets available in a dataset (Rutstein
& Johnson, 2004), they provide the best available wealth rating
of households in a country. We therefore would like IWI to be well
correlated with the country-specific indices.
Appendix C shows for the DHS surveys the correlations between
the IWI scores of households within a country and the same
household’s values on the DHS country-specific indices. We could
include 96 DHS surveys with a country-specific wealth index. In 73
(76%) of these surveys, the Pearson correlation between IWI and the
local wealth index was over .90 and in 92 (96%) of these surveys it
was over .80. The average correlation was .92. Hence we can
conclude that in the large majority of DHS surveys, IWI ranks the
households very similar to the country-specific wealth index.
5. Associations with welfare and poverty measures The final
proof of the pudding is the eating. Our test analyses have made
clear that IWI is a stable index, that does not depend much on the
inclusion of specific assets, nor on data for specific regions of
the developing world. IWI is also highly correlated with regional
wealth indices and with the country-specific wealth indices
available in the DHS surveys.
These are favorable test outcomes that indicate that IWI is a
reliable measuring instrument. However, being a reliable instrument
does not necessarily mean being also a useful instrument. Given the
fact that IWI was constructed on the basis of items related to
material need satisfaction, it seems likely that IWI will be a
reasonable indicator of a household’s level of material well-being.
However, in most practical applications, asset-based wealth indices
are used as – and were found to perform well as -- indicators of
the economic status of households (e.g. Filmer & Pritchett,
1999, 2001; Sahn & Stifel, 2000, 2003; McKenzie, 2005; Howe et
al., 2008; Filmer & Scott, 2012).
In this section we aim to test the performance of IWI as such an
economic indicator. Given that IWI is meant to be an index that is
comparable among countries, we will do so by performing a
cross-national analysis. If IWI is an effective indicator of the
economic status of households, we would expect national IWI values
to be a good indicator of level of economic development of a
country and
-
16
effective poverty measures. This is tested by comparing national
IWI values of developing countries with those countries’ values on
established indicators of economic development, human development,
education, health and poverty.
The national indicators with which IWI is compared were
downloaded from the UNDP Human Development Report website
(hdr.undp.org) and from the Worldbank website (data.worldbank.org).
From UNDP, we downloaded the development, education and health
indicators. The definitions of these indicators given below were
derived from the UNDP website or from UNDP (2011, p.130) Economic
development is measured by Gross National Income per capita (GNIc).
Human development is measured by the Human Development Index (HDI).
Education in measured by two indicators: Mean years of schooling
and expected years of schooling. Mean years of schooling is the
average number of years of education received by people aged 25 and
older, converted from education attainment levels using official
durations of each level. Expected years of schooling is the number
of years of schooling that a child of school entrance age can
expect to receive if prevailing patterns of age-specific enrolment
rates would persist throughout the child’s life. A country’s health
situation is measured by life expectancy at birth: the number of
years a newborn infant could expect to live if prevailing patterns
of age-specific mortality rates at the time of birth stay the same
throughout the infant’s life.
The poverty indicators were downloaded from the Worldbank
website (data.worldbank.org). We use the Poverty Headcount Ratios
at $1.25 and $2.00 a day (PPP), defined as the percentage of the
population living on less than $1.25 or $2.00 a day at 2005
international prices. From the Worldbank website we also derived
the Gini coefficient for income inequality, that is used to split
up the countries in high and low income countries.
The indicators of UNDP and Worldbank were not available for all
years for which we have an IWI value. To fill in the missing years
we used linear interpolation when possible. If interpolation was
not possible, we used the value from a nearby year. If the nearest
year was more than five years apart, the country/year combination
was removed from the data. For each country data for one year was
used for the analysis. This was generally the latest year for which
valid data was available, but priority was given to DHS and MICS
data. Appendix B presents the values of the indicators used in the
welfare and poverty analyses, together with information on the
country-year combinations that were used.
(a) Welfare measurement
The Pearson correlations between national IWI values for the
last available year and the HDI and its components are presented in
Table 4. The figures make clear that there are very strong positive
correlations with all of the indices. The correlation with HDI is
strongest, with a value of .899, but also the correlation with life
expectancy is with .841 impressive. IWI and national income are
correlated somewhat weaker with .788. The correlations with the
educational indicators are with .720 and .658 weakest, but
nevertheless still substantial. The correlations of IWI with HDI,
health and education are all stronger than those of national income
with these indicators.
-
17
Table 4. Pearson correlations of IWI with HDI and its
components
IWI
Life expectancya .841 Life exp
GNIca .788 .672 GNIc Exp. years of educationa
.720 .651 .682 Exp. edu
Mean years of educationb
.658 .559 .538 .728 Mean edu
HDIb .899 .870 .835 .833 .808 a N=87 b N=85
The finding that IWI is more strongly correlated with human
development, health and education than national income is
important. It suggests that IWI is a broader index than GNIc and
represents more than only the economic situation of households.
There are two likely reasons for this difference. First, IWI does
not necessarily rise when the income of the rich increases, as per
capita income does. The reason for this is that the questions on
ownership of consumer durables used for constructing IWI are yes/no
questions; in most surveys the households were asked whether they
owned at least one item of the durable. A household owning two or
more TV’s or cars therefore counts the same for IWI as a household
owning one TV or car. This is not the case with per capita income,
for which the prices of all TV’s and cars are added up. Compared to
per capita income, IWI thus gives a country a lower value in
situations of inequality.
Second, a household’s IWI value to a certain extent depends on
the provision of public goods -- like supply of water and
electricity -- in the area where the household lives and does
therefore not completely depend on the household’s income. The
increase in household’s welfare due to the access to public
services is thus better captured by IWI than by per capita income.
The fact that also HDI is less sensitive to inequality and better
captures access to public goods than national income (Stanton,
2006) may to a certain extent explain the high correlation between
IWI and HDI.
In Figure 2, the associations between IWI and the welfare
indices are depicted graphically. The plots confirm the picture
that was already given by the Pearson correlations: IWI is strongly
and positively correlated with all other measures. To find out
whether the difference between IWI and GNIc is to a certain extent
due to a different sensitivity to inequality, as suggested above,
in the top-right plot (B) the countries are split up into low and
high inequality countries. This was done on the basis of the Gini
coefficient for income inequality. The dark dots represent
countries with above average Gini and the light dots the countries
with below average Gini. In almost all cases, the dark dots are
situated above the light dots, thus indicating that for a given IWI
value the more unequal countries have higher levels of national
income than the more equal countries. This confirms the idea that
IWI differs more from national income in more unequal
countries.
-
18
Figure 2. Plots of national IWI values against national welfare
measures (A-F) and of IWI-based
poverty measures against Poverty Headcount ratios (G-H)
A B
C
E
G H
F
D
-
19
(b) Poverty measurement
A second important test of the usefulness of IWI involves its
performance in measuring poverty. To assess this performance, we
have defined the 20th, 30th, 40th, 50th and 60th percentiles of the
IWI distribution as IWI poverty lines. Table 5 presents Pearson
correlations between the national percentages of people with an IWI
value below these lines and the Poverty Headcount Ratios (PHR) at
$1.25 and $2.00 a day (PPP). Again we see strong correlations, all
above .8, which makes clear that IWI-based poverty measures perform
well in comparison with these established measures. The lines
differ not very much in the strength of the correlations, but the
IWI poverty line at the 30th percentile is most strongly correlated
with the PHR at $1.25 a day and the line at the 50th percentile
most strongly with the PHR at $2.00 a day. When using IWI for
poverty measurement, these percentiles therefore seem to be usable
IWI poverty lines. We call them the IWI-30 and IWI-50 poverty
lines.
Table 5. Pearson correlations of IWI-based poverty lines with
headcount ratios (N=76)
Headcount 1.25$ Headcount 2.00$ IWI-20 Poverty line .845 .839
IWI-30 Poverty line .875 .886 IWI-40 Poverty line .874 .906 IWI-50
Poverty line .860 .914 IWI-60 Poverty line .835 .906
In Panels G and H of Figure 2 the associations between these IWI
poverty lines and the headcount ratios are displayed graphically.
There are a few deviations from linearity, mostly due to countries
with more households underneath the IWI poverty lines than
underneath the headcount-based lines. In those countries thus being
above a dollar based poverty line does not always mean being able
to buy enough assets to cross the IWI-based lines. However, the
deviations are small and in particular the correlation of .914
between the IWI-50 and the PHR at $2.00 is impressive.
To test the performance of the IWI poverty lines further, Table
6 presents for the 2.1 million households in our data the
percentages of households below and above these lines that own a
specific asset. The poverty lines behave as could be expected. In
all cases, except for the bicycle and the cheap utensil, the
differences in asset ownership and quality of facilities is very
large. Of the households below the IWI-30 line only 5% owns a TV
and 0.3% owns a refrigerator, whereas this is the case for 84% and
59% of the households above this line. Of the households below the
IWI-30 line only 4% has high quality floor material, 2% a high
quality toilet facility, and 5% high quality water supply. Of the
households above this line, these figures are 44%, 67% and 68%
respectively. Similar, but less extreme, differences in asset
ownership can be observed between households below and above the
IWI-50 poverty line.
-
20
6. Conclusions Asset-based wealth indices are widely used
instruments for measuring the economic status of households and
studying inequality and poverty in low and middle income countries.
The indices used so far suffered however from one great problem,
they were not comparable among countries and time points. Although
there have been a few studies in which more general wealth indices
were created (e.g. Sahn & Stifel, 2000; Booysen et al., 2008),
the geographic coverage of these indices was restricted and they
were not turned into broadly usable instruments.
In this paper we introduce the International Wealth Index (IWI),
the first strictly comparable asset based wealth index that can be
used for all low and middle income countries. IWI is constructed by
applying Principal Component Analysis on data for over 2.1 million
households, derived from 165 household surveys held between 1996
and 2011 in 97 low and middle income countries. With IWI we provide
a stable and understandable yardstick for evaluating and comparing
the economic situation of households, social groups and societies
across all regions of the developing world.
A household’s position on IWI indicates to what extent the
household or its members own a basic set of assets that is valued
highly by people across the globe. These assets include consumer
durables, housing characteristics, and access to public utilities.
The IWI scale runs from 0 to 100, with 0 indicating that the
household owns none of the consumer durables, has lowest quality
housing and no connection to public utilities, and 100 indicating
that the household owns all included consumer durables, has highest
quality housing and good access to public utilities.
To assess the performance of IWI as a comparable indicator of
household wealth, a number of test analyses were conducted. On
these tests IWI performed very well. The 2.1 million households
were distributed rather evenly across the IWI scale, without
problematic clumping or truncation. Removing one or even two assets
from the index hardly influenced the rating of households; the same
was true for removing data from specific regions of the developing
world. Within DHS countries high correlations between IWI and
national DHS wealth indices were obtained.
Comparisons of IWI at the national level with established
welfare indices revealed very high correlations of .90 with the
Human Development Index (HDI), .84 with life expectancy, .79 with
Gross National Income per capita (GNIc), and .66 and .72 with
educational indices. IWI is more highly correlated with HDI than
with GNIc. This is probably due to the fact that -- just as HDI --
IWI is less affected by inequality and captures welfare effects of
access to public goods (services). Another reason might be found is
that asset based wealth indices like IWI are more than monetary or
expenditure based welfare measures indicators of longer-term, more
stable, aspects of household’s economic status (Sahn & Stifel,
2003; Howe et al., 2009; Filmer & Scott, 2012). The
correlations between IWI and educational indices (mean and expected
years of education) are somewhat lower than those with the other
indices, but still clearly higher than those between national
income and these educational indices (.54 and .68). Hence, IWI
seems to be a better predictor of human capital than national
income. Overall we can thus conclude that in comparison with
national welfare indices IWI performs very well.
To test the usefulness of IWI for poverty measurement, we
compared several IWI-based poverty measures with the Poverty
Headcount Ratios (PHR) at $1.25 and $2.00 a day (PPP). These
-
21
comparisons revealed high correlations. The percentage of
households with an IWI value below 30 was most strongly correlated
(.88) with the PHR at $1.25 and the percentage of households with
an IWI value below 50 showed a very high correlation (.91) with the
PHR at $2.00. These IWI-based poverty measures thus measure poverty
almost similarly to PHR, which means that in situations where these
other measures are not available, using IWI might constitute a good
alternative.
Given the excellent performance of indices derived from IWI as
welfare and poverty measures at the national level, it seems
plausible that such IWI-based indices will also perform well when
aggregated to the sub-national level. This implies that with the
introduction of IWI for the first time sub-national indicators can
be constructed for measuring in a comparable way the welfare level
and degree of poverty of sub-national areas across the developing
world.
References Bollen, K. A., Glanville, J. L., & Stecklov, G.
(2002). Economic status proxies in studies of fertility in
developing countries: Does the measure matter? Population
Studies, 56, 81–96
Booysen, van der Berg, Burger, Maltitz von, Rand du (2008) Using
an Asset Index to Assess Trends in Poverty in Seven Sub-Saharan
African Countries. World Development Vol. 36, 6, 1113 – 1130.
Cordoba, A. (2008). Methodological Note: Measuring Relative
Wealth using Household Asset Indicators. AmericasBarometer
Insights: No.6. Latin American Public Opinion Project.
Falkingham, Jane and Namazi, Ceema (2002). Measuring health and
p[overty: A review of approaches to identifying the poor. London:
DFID Health Systems Resource Centre.
Filmer, Deon and Pritchett, Lant. (1999) The Effect of Household
Wealth on Educational Attainment: Evidence from 35 Countries.
Population and Development Review, Vol. 25, No. 1, pp. 85 –
120.
Filmer, Deon and Pritchett, Lant. (2001) Estimating Wealth
Effect without Expenditure Data-or Tears: An Application to
Educational Enrollments in States in India. Demography Vol. 38, No.
1, 115 – 1
Filmer, Deon and Scott, N. (2012). Assessing Asset Indices.
Demography, 49, 359–392
Guiley D, Jayne S. (1997). Fertility transition in Zimbabwe:
determinants of contraceptive use and method choice. Population
Studies, 51(2), 173-190.
Gwatkin, D.R., Rutstein, S. Johnson, K., Suliman, E., Wagstaff,
A., Amouzou, A. (2007) Socio-economic Differences in Health,
Nutrition, and population within Developing Countries. HNP, The
World Bank.
Howe, Laura D., Hargreaves, James R. and Huttly, Sharon R.A.
(2008) Issues in the construction of wealth indices for the
measurement of socio-economic position in low-income countries.
Emerging Themes in Epidemiology, 5:3.
Howe, Laura D., Hargreaves, James R. , Gabrysch, S. and Huttly,
Sharon R.A. (2009) Is the wealth index a proxy for consumption
expenditure? A systematic review. Journal of Epidemiology and
Community Health, 63, 871-880.
McKenzie, David J. (2005). Measuring inequality with asset
indicators. Journal of Population Economics, 18, 229-260.
-
22
Montgomery, M., Gragnolati, M., Burke, K., & Paredes, E.
(2000). Measuring living standards with proxy variables.
Demography, 37, 155–174.
Rutstein, Shea and Johnson, K. (2004) The DHS Wealth Index, DHS
Comparative Report, No. 6, ORC Macro, Calverton, MD.
Rutstein, Shea O. (2008). The DHS Wealth Index: Approaches for
Rural and Urban Areas. DHS Working paper, 2008, 60, Measure
DHS.
Sahn, David E. and Stifel, David C. (2003) Exploring Alternative
Measures of Welfare in the Absence of Expenditure Data. Review of
Income and Wealth, Vol. 49, No. 4, 463-489.
Sahn, David .E. and Stifel, David C. (2000) Poverty Comparisons
over Time and Across Countries in Africa. World Development, Vol.
28, No. 12, pp. 2123 – 2155.
Stanton, Elizabeth (2006). Accounting for Inequality: A proposed
revision of the Human Development Index. PERI Working Paper 119.
Amherst: Political Economy Research Institute.
UNDP (2011). Human Development Report 2011. New York: United
Nations Development Programme.
Vyas, S., Kumaranayake, L. (2006) Constructing socio-economic
status indices: how to use principal components analysis. Oxford
University Press, DOI 10.1093/heapol/czl029.
-
23
Appendix A. Information on datasets used and average national
IWI values
Code Country Year Source N IWI value AFG Afghanistan 2010 DHS
21986 32.9 AGO Angola 2011 DHS 8028 33.4 AGO Angola 2000 MICS2 6244
20.2 ARM Armenia 2005 DHS 6562 78.1 ARM Armenia 2010 DHS 6653 77.2
AZE Azerbaijan 2000 MICS2 5859 56.6 AZE Azerbaijan 2006 DHS 7123
67.1 BDI Burundi 2005 MICS3 8150 10.6 BDI Burundi 2010 DHS 8517
15.8 BEN Benin 2001 DHS 5718 26.5 BEN Benin 2006 DHS 17330 28.7 BFA
Burkina Faso 1998 DHS 4741 15.6 BFA Burkina Faso 2003 DHS 9042 19.4
BGD Bangladesh 2006 MICS3 62127 25.0 BGD Bangladesh 2007 DHS 10381
24.8 BLZ Belize 2006 MICS3 1821 71.2 BOL Bolivia 2003 DHS 19100
48.2 BOL Bolivia 2008 DHS 19300 54.6 BRA Brazil 1996 DHS 13151 66.7
BRA Brazil 2000 IPUMS 50301 67.7 BTN Buthan 2010 MICS4 14670 56.1
CAF Central African Republic CAR 2006 MICS3 11655 15.8 CHL Chili
2002 IPUMS 41016 83.5 CHN China 2003 WHS 3962 72.5 CHN China 2004
CHNS 4044 64.1 CIV Cote d'Ivoire 1999 DHS 2101 31.0 CIV Cote
d'Ivoire 2006 MICS3 7514 41.6 CMR Cameroon 1998 DHS 4618 26.6 CMR
Cameroon 2004 DHS 10358 27.2 COD Congo Democratic Republic 2007 DHS
8748 19.4 COD Congo Democratic Republic 2010 MICS4 11258 15.7 COL
Colombia 2005 DHS 37211 72.6 COL Colombia 2010 DHS 51415 76.9 COM
Comoros 1996 DHS 2163 25.4 COM Comoros 2003 WHS 1640 37.7 CRI Costa
Rica 2000 IPUMS 28705 68.0 DOM Dominican Republic 1996 DHS 8772
56.4 DOM Dominican Republic 2002 DHS 26886 65.0 DOM Dominican
Republic 2007 DHS 32076 72.4 DZA Algeria 2002 PAPFAM 8228 76.8 ECU
Ecuador 2000 SIMPOC 14055 61.8 EGY Egypt 2000 DHS 16869 74.4 EGY
Egypt 2003 DHS 20128 80.4 EGY Egypt 2005 DHS 21810 78.3 EGY Egypt
2008 DHS 18838 77.7 ETH Ethiopia 2005 DHS 13607 11.5 ETH Ethiopia
2011 DHS 16612 15.3 GAB Gabon 2000 DHS 6068 45.4 GEO Georgia 2003
WHS 2723 71.3 GEO Georgia 2005 MICS3 11883 64.7 GHA Ghana 1998 DHS
5964 25.7 GHA Ghana 2006 MICS3 5909 35.1 GHA Ghana 2008 DHS 11669
43.0 GIN Guinea 2005 DHS 6172 16.9 GMB Gambia 2000 MICS2 4489 35.4
GMB Gambia 2006 MICS3 5978 42.8 GNB Guinea Bissau 2006 MICS3 4993
31.8 GTM Guatemala 1999 DHS 5434 44.4 GTM Guatemala 2003 WHS 4408
55.2
-
24
HND Honduras 2005 DHS 18636 56.5 HTI Haiti 2005 DHS 9899 27.1
IDN Indonesia 2003 DHS 32577 48.0 IDN Indonesia 2007 DHS 40131 48.7
IND India 1999 DHS 92306 31.5 IND India 2006 DHS 108714 37.3 IRQ
Iraq 2006 MICS3 17868 74.1 JOR Jordan 2002 DHS 7825 85.8 JOR Jordan
2007 DHS 14547 87.3 KAZ Kazakhstan 1999 DHS 5816 62.8 KAZ
Kazakhstan 2006 MICS3 14564 74.0 KEN Kenya 1998 DHS 8242 18.6 KEN
Kenya 2003 DHS 8480 21.1 KEN Kenya 2008 DHS 9018 27.7 KGZ
Kyrgyzstan 1997 DHS 3647 52.8 KGZ Kyrgyzstan 2006 MICS3 4893 64.9
KHM Cambodia 2005 DHS 14171 29.6 KHM Cambodia 2010 DHS 15622 40.6
LAO Laos 2003 WHS 4838 38.0 LBR Liberia 2007 DHS 6634 20.7 LKA Sri
Lanka 2003 WHS 5768 48.8 LSO Lesotho 2010 DHS 9385 30.0 MAR Morocco
2003 DHS 10964 65.0 MAR Morocco 2003 WHS 4696 61.7 MDG Madagascar
1997 DHS 7141 15.2 MDG Madagascar 2009 DHS 17744 22.1 MDV Maldives
2009 DHS 6402 80.0 MEX Mexico 2003 WHS 38537 79.8 MLI Mali 2006 DHS
12768 22.0 MMR Myanmar 2003 WHS 5880 43.0 MNG Mongolia 2000 MICS2
5981 42.0 MNG Mongolia 2005 MICS3 6219 46.7 MOZ Mozambique 1997 DHS
9052 13.2 MOZ Mozambique 2003 DHS 12249 13.7 MRT Mauritania 2007
MICS3 10095 28.6 MUS Mauritius 2003 WHS 3748 88.1 MWI Malawi 2004
DHS 13495 13.8 MWI Malawi 2006 MICS3 30290 12.7 MWI Malawi 2010 DHS
24612 16.4 MYS Malaysia 2003 WHS 5947 90.7 NAM Namibia 2000 DHS
6295 36.8 NAM Namibia 2006 DHS 9086 45.3 NER Niger 1998 DHS 5831
11.0 NER Niger 2006 DHS 7589 12.4 NGA Nigeria 1999 DHS 7254 24.8
NGA Nigeria 2003 DHS 7091 29.6 NGA Nigeria 2008 DHS 33621 36.0 NIC
Nicaragua 1998 DHS 11172 40.0 NIC Nicaragua 2001 DHS 11245 43.3 NPL
Nepal 2006 DHS 8697 26.5 NPL Nepal 2011 DHS 10820 41.6 PAK Pakistan
2003 WHS 6096 45.9 PAK Pakistan 2007 DHS 9150 52.9 PAN Panama 2000
SIMPOC 9177 55.8 PER Peru 2000 DHS 28671 47.6 PER Peru 2004-2008
DHS 45998 53.9 PHL Philippines 1998 DHS 12257 52.0 PHL Philippines
2008 DHS 12371 61.1 PRY Paraguay 2003 WHS 5072 62.3 RWA Rwanda 2010
DHS 12479 19.8 SDN Sudan 2000 MICS2 24791 22.6 SDN Sudan 2008 IPUMS
36856 18.4 SEN Senegal 1997 DHS 4722 29.7 SEN Senegal 2011 DHS 7902
49.3 SLE Sierra Leone 2005 MICS3 7054 18.0 SLE Sierra Leone 2008
DHS 7177 22.2
-
25
SLV El Salvador 2001 SIMPOC 11953 59.2 SOM Somalia 2006 MICS3
5707 18.6 SSD South Sudan (urban) 2000 MICS2 1553 14.1 SSD South
Sudan 2008 IPUMS 7442 11.4 STP Sao Tome & Principe 2000 MICS2
3252 31.3 STP Sao Tome & Principe 2009 DHS 3529 42.9 SUR
Suriname 2006 MICS3 5603 76.6 SWZ Swaziland 2000 MICS2 4309 37.2
SWZ Swaziland 2006 DHS 4806 40.9 SYR Syria 2006 MICS3 19006 82.3
TCD Chad 2004 DHS 5301 8.6 TGO Togo 2006 MICS3 6484 28.8 THA
Thailand 2006 MICS3 40483 77.5 TJK Tajikistan 2000 MICS2 3696 46.3
TJK Tajikistan 2005 MICS3 6684 51.2 TLS Timor Leste 2009 DHS 11454
31.9 TUN Tunisia 2001 PAPFAM 6048 72.6 TUN Tunisia 2003 WHS 4863
74.4 TUR Turkey 2003 DHS 10738 75.7 TZA Tanzania 2004 DHS 9660 15.3
TZA Tanzania 2010 DHS 9569 21.9 UGA Uganda 2006 DHS 8748 14.8 URY
Uruguay 2003 WHS 2938 89.7 URY Uruguay 2006 IPUMS 19954 80.1 UZB
Uzbekistan 1996 DHS 3687 53.3 UZB Uzbekistan 2005 MICS3 10127 62.4
VEN Venezuela 2001 IPUMS 26815 76.8 VNM Vietnam 1997 DHS 6998 33.0
VNM Vietnam 2002 DHS 6985 43.8 VNM Vietnam 2006 MICS3 8354 55.3 YEM
Yemen 1997 DHS 9669 35.1 YEM Yemen 2003 PAPFAM 11146 38.2 YEM Yemen
2006 MICS3 3562 48.4 ZAF South Africa 1998 DHS 11886 53.9 ZAF South
Africa 2003 WHS 2129 70.0 ZMB Zambia 2002 DHS 7072 18.8 ZMB Zambia
2007 DHS 7088 24.1 ZWE Zimbabwe 1999 DHS 6308 33.4 ZWE Zimbabwe
2006 DHS 9201 34.8 ZWE Zimbabwe 2011 DHS 9756 38.5
In the following cases missing asset variables were replaced by
other variables or by imputation of a value: floor replaced by
rooms for Brazil 2000, India 1999; phone replaced by refrigerator
for Angola 2000, Azerbaijan 2000, Brazil 1996, Gambia 2000,
Mongolia 2000, Sudan 2000, South Sudan 2000, Sao Tome Y Principe
2000, Swaziland 2000, Tajikistan 2000; bicycle replaced by car for
Brazil 1996, Brazil 2000; Costa Rica 2000, Jordan 2002, by 0 for
Jordan 2007, Venezuela 2001, by motorbike for Tunisia 2001, Uruguay
2006, by 1 for El Salvador 2001; rooms replaced by water for Benin
2001, Burkina Faso 2003, Guinea 2005, Indonesia 2003, Kazakhstan
1999, Zambia 2002, Zimbabwe 1999; cheap utensils put at 1 for China
2003, Georgia 2003, Mexico 2003, Mauritius 2003, Malaysia 2003,
Paraguay 2003, Tunisia 2003m Uruguay 2003; electricity put a t 1
for China 2004, Georgia 2003, Mauritius 2003, Malaysia 2003,
Uruguay 2003. In Brazil 1996 and 2000, Geaorgia 2003, Mauritius
2003, Malaysia 2003, and Uruguay 2003 two items were missing.
Data Sources: DHS Demographic and Health Survey
(www.measuredhs.com) MICS UNICEF Multiple Indicator Cluster Surveys
(www.childinfo.org). MICS2 is 2000 round, MICS3 is 2005-2006
round, MICS4 is 2010-2011 round WHS World Health Surveys
collected under supervision of the World Health Organization
(www.who.int/healthinfo/survey) PAPFAM Surveys of the Pan Arabic
Project for Family Health (PAPFAM), sponsored by among others the
League
of Arab States (www.papfam.org) IPUMS Minnesota Population
Center. Integrated Public Use Microdata Series, International:
Version 6.1 [Machine-
readable database]. Minneapolis: University of Minnesota, 2011
(international.ipums.org) SIMPOC Surveys of the Statistical
Information and Monitoring Programme on Child Labor (SIMPOC) of
ILO-IPEC
(www.ilo.org/ipec) CHNS Chinese Health and Nutrition Survey 2004
(www.cpc.unc.edu/projects/china).
-
26
Appendix B. Data used for computing associations between IWI and
welfare measures and between IWI-30 and IWI-50 and poverty
Headcount Ratios
ISO_code Year IWI-value HDI GNIc Life exp.
Exp. eduyrs
Mean eduyrs GINI
IWI-30
IWI-50
HR $1.25
HR $2.00
AFG 2010 32.9 0.4 1351 48.3 9.1 3.3 27.8 - - - - AGO 2000 - - -
- - - - 76.8 90.2 54.3 70.2 AGO 2011 33.4 0.5 4874 51.1 9.1 4.4 - -
- - - ARM 2010 77.2 0.7 5009 74.1 12.0 10.8 30.9 0.2 4.4 1.3 12.4
AZE 2006 67.1 - 3940 69.0 11.5 - 34.7 2.2 18.9 2.1 9.8 BDI 2010
15.8 0.3 359 50.0 10.5 2.7 - 90.7 96.1 81.3 93.5 BEN 2006 28.7 0.4
1311 54.3 9.2 3.0 38.6 62.3 81.9 47.3 75.3 BFA 2003 19.4 0.3 996
51.6 4.2 1.3 39.6 83.5 91.1 56.5 81.2 BGD 2007 24.9 0.5 1256 67.6
8.0 4.4 32.8 69.0 84.9 47.6 75.4 BLZ 2006 71.2 0.7 5765 74.8 12.6
7.8 - - - - - BOL 2008 54.6 0.7 4320 64.7 14.0 8.3 56.3 23.3 40.5
15.6 24.9 BRA 2000 67.7 0.7 7698 70.1 14.5 5.6 60.0 8.2 19.3 11.6
21.5 BTN 2010 56.1 0.5 5060 66.8 11.0 2.3 38.1 12.9 44.4 10.2 29.8
CAF 2006 15.8 0.3 660 44.4 5.8 3.2 56.3 89.8 96.8 62.8 80.8 CHL
2002 83.5 0.8 10483 77.6 13.4 9.0 54.6 1.7 4.8 2.1 5.2 CHN 2004
64.1 0.6 3832 71.9 10.5 7.0 42.5 5.0 26.3 20.3 41.7 CIV 2006 41.6
0.4 1492 52.1 6.3 3.1 43.8 39.8 63.4 23.6 46.5 CMR 2004 27.2 0.4
1866 49.5 8.6 5.3 39.7 62.1 82.5 10.2 31.4 COD 2010 15.8 0.3 270
48.1 8.2 3.5 - 84.4 91.4 87.7 95.2 COL 2010 76.9 0.7 8043 73.5 13.6
7.3 55.9 2.3 7.2 8.2 15.8 COM 1996 25.4 - 1118 56.9 7.8 - - - - - -
CRI 2000 68.0 0.7 7467 77.8 10.7 8.0 46.5 3.5 13.6 5.5 10.9 DOM
2007 72.4 0.7 6632 72.5 11.9 6.9 48.7 3.5 14.4 3.8 11.5 DZA 2002
76.8 0.6 6209 70.7 12.0 5.9 - - - - - ECU 2000 61.8 0.7 5005 73.4
12.9 6.9 56.6 9.8 25.6 20.7 37.7 EGY 2008 77.7 0.6 4917 72.4 11.0
6.0 30.8 0.6 3.1 1.7 15.4 ETH 2011 15.3 0.4 971 59.3 8.5 1.5 - 84.1
91.7 39.0 77.6 GEO 2005 64.7 0.7 3650 72.8 12.6 12.1 41.1 2.9 22.2
16.0 33.5 GHA 2008 43.0 0.5 1329 62.7 9.7 6.9 - 35.8 63.0 28.6 51.8
GIN 2005 16.9 0.3 860 51.1 7.5 1.6 39.8 81.0 91.3 49.8 75.2 GMB
2006 42.8 0.4 1075 56.9 8.4 2.4 - 34.4 67.5 33.6 55.9 GNB 2006 31.8
0.3 955 46.4 8.9 2.3 35.5 56.5 83.0 48.9 78.0 GTM 1999 44.4 0.5
3861 67.2 8.4 3.4 55.0 36.2 55.3 14.1 27.7 HND 2005 56.5 0.6 3120
71.4 10.9 5.9 59.7 20.7 42.7 26.4 40.1 HTI 2005 27.1 0.4 959 59.9
7.6 4.5 59.2 63.7 82.5 61.7 77.5 IDN 2007 48.7 0.6 3122 67.8 12.4
5.5 34.0 21.9 52.6 24.2 56.1 IND 2006 37.3 0.5 2474 63.7 10.0 4.1
33.4 48.4 66.7 39.8 74.2 IRQ 2006 74.1 0.6 2578 68.4 9.8 5.4 30.9
1.7 8.0 2.8 21.4 JOR 2007 87.3 0.7 4770 72.9 12.9 8.2 35.8 0.3 0.9
0.2 2.8 KAZ 2006 74.0 0.7 8264 65.5 14.9 10.2 30.8 0.2 7.7 0.4 3.3
KEN 2008 27.7 0.5 1407 55.2 10.4 6.8 - 62.3 84.3 43.4 67.2 KGZ 2006
65.0 0.6 1728 66.8 12.4 9.2 38.7 0.9 20.0 5.9 32.1 KHM 2010 40.6
0.5 1753 62.7 9.8 5.8 37.9 39.5 67.7 22.8 53.3 LBR 2007 20.7 0.3
254 53.7 11.0 3.6 38.2 76.1 93.6 83.8 94.9 LSO 2010 30.0 0.5 1643
47.6 9.9 5.9 - - - - - MAR 2003 65.0 0.5 3203 69.7 9.4 3.7 40.7
15.0 27.7 5.0 20.9 MDG 2009 22.1 0.5 846 66.0 10.4 5.2 44.1 80.2
93.0 78.6 92.0 MDV 2009 80.0 0.7 4828 76.1 12.4 5.6 - - - - - MLI
2006 22.0 0.3 978 49.4 6.9 1.8 39.0 76.1 89.1 51.4 77.1 MNG 2005
46.7 0.6 2550 66.0 12.6 8.2 34.7 - - - - MOZ 2003 13.8 0.3 571 47.8
7.2 1.0 47.1 90.3 95.3 74.7 90.0 MRT 2007 28.6 0.4 1762 57.7 7.8
3.5 40.5 58.9 78.6 23.9 48.9
-
27
MWI 2010 16.4 0.4 730 53.5 8.9 4.2 - 86.3 94.3 73.9 90.5 NAM
2006 45.3 0.6 5442 59.4 11.6 7.1 63.9 44.7 57.8 31.9 51.1 NER 2006
12.4 0.3 605 52.2 3.9 1.3 34.6 91.0 95.2 48.0 75.3 NGA 2008 36.0
0.4 1806 50.4 8.9 5.0 46.9 47.8 70.3 66.3 84.0 NIC 2001 43.3 0.5
2058 70.2 10.0 4.8 43.1 36.1 58.9 14.4 34.4 NPL 2011 41.6 0.5 1160
68.8 8.8 3.2 32.8 37.3 62.6 24.8 57.3 PAK 2007 52.9 0.5 2347 64.5
6.6 4.7 31.4 25.8 44.8 21.8 60.6 PAN 2000 55.9 0.7 7721 74.2 12.8
8.5 57.3 21.7 36.9 15.2 22.8 PER 2000 - - - - - - - 34.6 49.5 12.4
24.1 PER 2004/8 53.9 0.7 5803 72.1 12.9 8.1 - - - - - PHL 2008 61.1
0.6 3195 68.1 11.8 8.8 43.0 14.4 31.2 19.8 42.7 RWA 2010 19.8 0.4
1086 55.1 11.1 3.3 50.8 83.5 94.8 65.0 83.4 SDN 2008 18.4 0.4 1706
60.5 4.4 3.0 35.3 80.4 89.6 19.8 44.1 SEN 2011 49.3 0.5 1708 59.3
7.5 4.5 - 31.1 49.5 33.5 60.4 SLE 2008 22.2 0.3 680 46.2 7.2 2.8 -
74.9 90.8 53.4 76.1 SLV 2001 59.2 0.6 5153 70.1 11.5 5.9 53.6 20.1
38.0 14.4 23.0 SOM 2006 18.6 - - - - - - - - - - SSD 2008 11.4 - -
- - - 45.5 - - - - STP 2000 - - - - - - - 60.2 82.1 28.2 54.2 STP
2009 42.9 0.5 1380 63.3 10.2 4.2 - - - - - SUR 2006 76.6 0.7 6213
69.1 12.6 7.2 - - - - - SWZ 2006 40.9 0.5 4601 46.4 10.0 6.7 51.1
41.7 66.7 46.2 69.6 SYR 2006 82.3 0.6 3830 74.9 11.0 5.7 35.8 0.4
3.1 1.7 16.9 TCD 2004 8.6 0.3 1010 48.2 5.8 1.5 39.8 96.1 98.4 61.9
83.3 TGO 2006 28.8 0.4 766 55.6 9.7 4.9 34.4 64.5 82.9 38.7 69.3
THA 2006 77.5 0.7 6625 73.4 12.2 6.0 42.4 0.4 4.1 1.0 7.6 TJK 2005
51.2 0.6 1430 65.4 11.0 10.0 33.6 9.8 54.3 18.7 45.6 TLS 2009 31.9
0.5 2867 62.0 11.2 2.8 - - - - - TUN 2001 72.6 0.6 5371 72.8 13.4
5.0 40.9 5.5 15.0 2.3 11.9 TUR 2003 75.7 0.7 10208 71.1 10.8 5.9
43.4 2.2 10.6 2.5 10.0 TZA 2010 21.9 0.5 1272 57.4 9.1 5.1 - 76.6
88.9 67.9 87.9 UGA 2006 14.8 0.4 924 50.9 10.2 4.4 42.6 87.1 95.7
51.5 75.6 URY 2006 80.1 0.8 10051 76.0 15.3 8.0 47.2 0.5 4.0 0.7
3.6 UZB 2005 62.4 0.6 2000 67.2 11.5 10.0 36.7 - - - - VEN 2001
76.8 0.7 9449 72.4 10.5 5.9 47.2 3.0 9.0 9.6 20.8 VNM 2006 55.3 0.6
2214 74.0 10.4 5.0 35.8 9.7 46.8 21.4 48.1 YEM 2006 48.4 0.4 2025
63.2 8.6 1.9 37.7 31.8 49.0 17.5 46.6 ZAF 1998 53.9 0.6 7401 56.8
13.1 8.2 41.3 29.2 46.0 24.3 41.7 ZMB 2007 24.1 0.4 1117 46.0 7.9
6.4 54.6 71.5 82.8 68.5 82.6 ZWE 2011 38.5 0.4 376 51.4 9.9 7.2 - -
- - -
Meaning of abbreviations: HDI Human Development Index, source
hdr.undp.org GNIc Gross National Income per capita (PPP), source
hdr.undp.org Life exp Life expectancy at birth, source hdr.undp.org
Exp. eduyrs Expected years of schooling a child of school entrance
age can expect to receive, source hdr.undp.org Mean eduyrs Mean
years of education received by people aged 25 and older, source
hdr.undp.org GINI Gini coefficient for income inequality, source
data.worldbank.org IWI-30 Percentage of households with an IWI
value below 30 IWI-50 Percentage of households with an IWI value
below 50 HR $1.25 Poverty Headcount Ratio at $1.25 a day (PPP) ,
source data.worldbank.org HR $2.00 Poverty Headcount Ratio at $2.00
a day (PPP) , source data.worldbank.org Data sources are the UNDP
Human Development Report website (hdr.undp.org), for HDI, GNIc,
life expectancy, expected years of schooling, and mean years of
education, and he Worldbank website (data.worldbank.org), for GINI
and the Poverty Headcount Ratios. Websites were approached in
December 2012. The indicators of UNDP and Worldbank were not
available for all years for which we have an IWI value. Missing
years were filled in with linear interpolation when possible. If
interpolation was not possible, values from a nearby year were
used. If the nearest year was more than five years apart, the
country/year combination was removed from the data.
-
28
Appendix C. Pearson correlations between DHS wealth index and
IWI for DHS countries
ARM 2005 0.856 ARM 2010 0.784 AZE 2006 0.863 BDI 2010 0.939 BEN
2001 0.927 BEN 2006 0.914 BFA 2003 0.951 BFA 1998 0.923 BGD 2007
0.933 BOL 2003 0.958 BOL 2008 0.951 BRA 1996 0.925 CIV 1999 0.958
CMR 1998 0.968 CMR 2004 0.935 COD 2007 0.938 COL 2005 0.900 COL
2010 0.846 COM 1996 0.965 DOM 1996 0.908 DOM 2002 0.928 DOM 2007
0.873 EGY 2000 0.877 EGY 2003 0.862 EGY 2005 0.870 EGY 2008 0.829
ETH 2005 0.950 ETH 2011 0.961 GAB 2000 0.933 GHA 1998 0.958 GHA
2008 0.937 GIN 2005 0.959
GTM 1999 0.935 HND 2005 0.917 HTI 2005 0.940 IDN 2003 0.895 IDN
2007 0.915 IND 1999 0.935 IND 2006 0.944 JOR 2002 0.894 JOR 2007
0.700 KAZ 1999 0.920 KEN 1998 0.965 KEN 2003 0.891 KEN 2008 0.894
KGZ 1997 0.921 KHM 2005 0.914 KHM 2010 0.939 LBR 2007 0.915 LSO
2010 0.922 MAR 2003 0.950 MDG 1997 0.890 MDG 2009 0.880 MDV 2009
0.761 MLI 2006 0.843 MOZ 1997 0.953 MOZ 2003 0.950 MWI 2004 0.941
MWI 2010 0.936 NAM 2000 0.980 NAM 2006 0.972 NER 1998 0.970 NER
2006 0.968 NGA 1999 0.920
NGA 2003 0.929 NGA 2008 0.937 NIC 1998 0.967 NIC 2001 0.951 NPL
2006 0.924 NPL 2011 0.898 PAK 2007 0.925 PER 2000 0.962 PER 2004-8
0.952 PHL 1998 0.933 PHL 2008 0.935 RWA 2010 0.928 SEN 1997 0.943
SEN 2011 0.939 SLE 2008 0.938 STP 2009 0.921 SWZ 2006 0.941 TCD
2004 0.899 TLS 2009 0.898 TUR 2003 0.750 TZA 2004 0.899 TZA 2010
0.936 UGA 2006 0.934 UZB 1996 0.942 VNM 1997 0.967 YEM 1997 0.949
ZAF 1998 0.972 ZMB 2002 0.971 ZMB 2007 0.923 ZWE 1999 0.938 ZWE
2006 0.966 ZWE 2011 0.939