Poverty Comparisons Over Time and Across
Countries in Africa
DAVID E. SAHN and DAVID C. STIFEL
*Cornell University, Ithaca, NY, USA
Summary. ÐWe use Demographic and Health Surveys (DHS) to compare ``poverty'' at two or more points in time within and between African countries. Our welfare measure is an index resulting from a factor analysis of various household characteristics, durables, and household headsÕ education. An advantage of this measure is that for intertemporal and intraregional comparisons, we need not rely on suspect price de¯ators and currency conversion factors. The wide availability and similarity of questionnaires of the DHS facilitate comparisons over both time and countries. Our results generally show declines in poverty during the previous decade, largely due to improvements in rural areas. Ó 2000 Published by Elsevier Science Ltd.
Key words ÐAfrica, asset index, factor analysis, poverty, stochastic dominance, welfare measures
1. INTRODUCTION
The contentious debate on the eectiveness of economic and social policy in Africa over the past decade continues largely unresolved. One reason for the disparate views on the role of reform in alleviating poverty is that we remain largely ignorant about the basic question of what has happened to poverty during the last 10 years. Addressing this issue is a pre-requisite to improving our understanding of the under-lying social and economic processes that have contributed to changes in economic well-being. A new generation of nationally representa-tive household income and expenditure surveys has helped to provide a better understanding of living standards in Africa. 1These surveys have
been very useful in our analysis of the level and characteristics of poverty on the continent. They have de®ned welfare and the corre-sponding notion of poverty based on the use of consumption expenditures (including the imputed value of home consumption), generally regarded as the preferred money metric of utility.2 Much of the available household survey data that have been used to measure poverty are both recent, done within the past 10 years, and in the form of one-time cross-sec-tions. Thus, while we have learned a great deal about poverty at a particular point in time in many African countries, the view remains a snapshot. In the vast majority of African countries, we remain unable to make
inter-temporal comparisons of poverty due the unavailability of data. Where survey data are available at more than one point in time, the determination of changes has proven prob-lematic. First, survey designs change. It is now well established that dierences in recall peri-ods,3 changes in the survey instrument (e.g., the number and choice of item codes listed), 4 and even the nature of interviewer training, can have large systematic eects on the measure-ment of household expenditures. Compounding this problem, intertemporal comparisons of money-metric welfare are only as precise as the de¯ators used. Consumer price indices are often suspect in Africa, due to weaknesses in data collection and related analytical proce-dures. Thus, relying on ocial CPIs is often precarious, at best. 5 Alternatives such as deriving price indexes from unit values, where quantity and expenditure data are collected, also have some serious drawbacks.6
In combination, these factors have limited what we know about changes in poverty, and the reliability of the relatively few estimates
2000 Published by Elsevier Science Ltd. Printed in Great Britain 0305-750X/00/$ - see front matter PII: S0305-750X(00)00075-9
www.elsevier.com/locate/worlddev
*The authors would like to thank an anonymous referee, Stephen Younger and George Jakubson for invaluable comments. They are also indebted to Macro International Inc., for supplying the data, and in particular, Bridget James for her assistance and prompt responses to queries. Final revision accepted: 5 May 2000.
that are available. This motivates our use of the Demographic and Health Surveys (DHS) as an alternative instrument for assessing changes in poverty, relying on an asset index as an alter-native metric of welfare.
The DHS have been collected in a large number of African countries, and in many cases, at more than one point in time. 7 The surveys were not designed for econometric (or even economic) analysis. Instead, the purpose of the surveys was to assist governments and private agencies in developing countries to better evaluate population, health and nutrition programs. Consequently, there are no data on income or expenditures, the standard money metric measures of well-being. Despite this important drawback, the DHS do contain information on household assets that can be employed to represent an alternative to a money metric utility approach to welfare measurement. 8 The DHS also have two distinct advantages: they are available at two or more points in time for a large number of countries in Africa, 11 to be precise, and key survey instruments are standardized for all countries. Therefore, we can con®dently compare living standards, across time periods, within a given country, and also across coun-tries for many of our poverty measures.
In the absence of income or expenditure measures, we derive a welfare index constructed from the households' asset information avail-able in the survey. This is the outcome of a factor analysis of various household charac-teristics (water source, toilet facilities, and construction materials) and durables (owner-ship of radio, television, refrigerator, bicycle, motorcycle and/or car) as well as education of the household head. We assume that there is a common factor, ``welfare,'' behind the owner-ship of these assets, and allow the factor anal-ysis to de®ne that factor as a weighted sum of the individual assets.9 One of the advantages of this measure is that for intertemporal and intraregional comparisons, we need not rely on what are often tenuous and suspect price de¯ators that are used to compare money metric measures of welfare.10
In this paper, we compare ``poverty'' as measured by our welfare index over time.11We do this by comparing percentages of families whose welfare falls below a certain level in the asset index distribution. We also compare the distributions of our asset welfare measure at the two (or more) points in time when the DHS data were collected, using standard tests for
welfare dominance (Ravallion, 1991; Ravallion, 1994; Davidson & Duclos, 1998). That is, we try to identify distributions that will show less poverty regardless of the poverty line or poverty measure used. Our next approach is to decompose poverty measures regionally (as in Ravallion & Huppi, 1991). This allows us to see whether overall changes in poverty are due to changes in one or more particular regions, or movements between regions with dierent poverty levels. Finally, we use the asset index to make cross-country comparisons of poverty.
Before presenting our results, we discuss in some more detail the methods employed, and the data we use. We conclude with a summary of our ®ndings.
2. METHODOLOGY
(a) Asset index
To construct an index of the household assets recorded in the DHS survey requires selecting a set of weights for each asset. That is, we want an index of the form
Ai^c1ai1 c^KaiK;
whereAiis the asset index for householdi, the
aik's are the individual assets,k, recorded in the
survey, and the c's are the weights, which we must estimate. Because neither the quantity nor the quality of all assets is collected, nor are prices available in the data, the natural welf-arist choice of prices as weights is not possible. Rather than imposing arbitrary weights as in Montgomery, Burk, and Paredes (1997), we let the data determine them directly. Hammer (1998) and Filmer and Pritchett (1998) use a similar method that employs principal compo-nent analysis to construct an asset index. The weights for their indices are the standardized ®rst principal component of the variance-co-variance matrix of the observed household assets. We use factor analysis instead of prin-cipal component analysis because the latter forces all of the components to accurately and completely explain the correlation structure between the assets. Factor analysis, on the other hand, accounts for the covariance of the assets in terms of a much smaller number of hypothetical common variates, or factors (Lawley & Maxwell, 1971). In addition, it allows for asset-speci®c in¯uences to explain the variances. In other words, all of the common factors are not forced to explain the
entire covariance matrix. In our case, we assume that the one common factor that explains the variance in the ownership of the set of assets is a measure of economic status, or ``welfare.'' Finally, the assumptions necessary to identify the model using factor analysis are stated explicitly and provide guidance in determining which assets should or should not be included in the index.12
Unlike with principal component analysis, we must explicitly impose structure from the outset. The structural model includes only one factor:
aikbkciuik
fori1;. . .;N households fork1;. . .;K household assets:
1
The ownership of each observed asset (k) for each household (i), represented by the variable aik, is a linear function of an unobserved
common factor for each household, ci, which
we label ``household welfare.''13Note that the relationship between the asset and the unob-served common factor,bk, as well as the noise
component (``unique element''), uik, are also
unobserved and must be estimated.14
To identify the model, we make the following assumptions:
(A1): Households are distributediid. (A2):E uijci 0
Kx1.
(A3):V ui Diagfr21;. . .;r2Kg.
Structure can now be imposed on the vari-ance-covariance of the observed assets. To see what these restrictions are, ®rst rewrite the set ofkeqn. (1) in vector form,
aibciui; 1a
where b b1;. . .;bK. Assumption (A3)
implies that once the common factor accounts for a portion of the variance in the ownership of assets, the remainder of the variance, the disturbance terms (``unique elements''), should be uncorrelated across assets. Note that these errors are not constrained to be identically distributed. This gives us the variance-covari-ance matrix of the unique disturbvariance-covari-ances E uiu0i Diagfr
2 1;. . .;r
2 Kg W:
Without loss of generality, we assume that the mean of the common factor (wealth) is zero, thus the variance of the common factor is E cic0i r
2 c:
Orthogonality of the common factor and the disturbance (A2) permits us to write the vari-ance of the assets as
E aia0i E bciui bciui 0
;
which gives us
Xbb0r2cW: 2
Note that identi®cation requires the normalization of one of the parameters, and typically it is the variance of the unobserved factor (r2
c1). Although this normalization
makes it dicult to interpret the coecients on the common factor b, we shall do so anyway since none of the statistical packages that provide factor analysis procedures have options for other normalizations and since interpreta-tion of these parameters is not crucial to the analysis.15
If we assume multivariate normality ofciand
ui, we can estimate b and W using maximum
likelihood techniques (Lawley & Maxwell, 1971). Once these parameters have been esti-mated, the common factor (asset index) can be estimated for each household, by de®ning the asset index as the projection of unobserved household wealth (ci) on the observed
house-hold assets:
E cijai c1ai1 cKaiK; where
cv ai ÿ1
cov ai;ci: 3
Given the normalization, r2
c1, it is
reason-ably straightforward to show that cov ai;ci b, and thuscXÿ1b. Finally, the
estimate of the asset index for household i is de®ned as:
Ai^c1ai1 c^KaiK; where
^
cX^ÿ1b^r^2
c: 3a
The assets included in the index can be placed into two categories: household durables and household characteristics. The household durables consist of ownership of a radio, TV, refrigerator, bicycle, and motorized transpor-tation (a motorcycle or a car). The household characteristics include source of drinking water (piped or surface water relative to well water), toilet facilities (¯ush or no facilities relative to pit or latrine facilities), and ¯oor material (low quality relative to higher quality). We also include the years of education of the household head to account for householdÕs stock of human capital.16 Since we want to compare
separate samples to estimate the wealth indexes for each of the households. 17
(b) Stochastic tests of welfare dominance
We employ standard tests of welfare domi-nance to compare distributions of our asset index over time. The idea is to make ordinal judgments on how poverty changes for a wide class of poverty measures over a range of poverty lines. We explain brie¯y how to esti-mate the orderings and to perform statistical inference on them. The discussion follows Davidson and Duclos (1998) closely.
Consider two distributions of welfare indi-cators with cumulative distribution functions, FAandFB, with support in the nonnegative real
numbers. 18Let
for any integer sP2. Now distribution A is said to (strictly) dominate distribution B at ordersifDs
A x6 <D s
B x, for allx2 0;zmax,
wherezmaxis the maximum acceptable poverty
line.19
Davidson and Duclos (1998) show thatDs x
can be equivalently expressed as
Ds
Further, if we have a random sample of N
independent observations on the welfare vari-able, yi, from a population, then a natural
estimator ofDs xis
where F^ is the empirical cumulative distribu-tion funcdistribu-tion of the sample, and I is an indicator function, which is equal to one when itÕs argument is true, and equal to zero when false.
Because we apply this estimator to two independent samples of asset indexes for each country,
iidvariables. Simpletstatistics are constructed to test the null hypothesis,
H0: ^DsA x ÿD^ s B x 0;
for a series of test points up to an arbitrarily de®ned highest reasonable poverty line. In cases where the null hypothesis is rejected for each test point, and the signs on all of thetstatistics are the same, then dominance of order s is declared. The tests were conducted up tos3, after which ``no dominance'' is declared.20
Stochastic dominance tests are conducted for the distributions of household asset indexes up to two relative poverty lines determined sepa-rately for each country. For a given country, the lower (upper) poverty line is simply the 25th (40th) percentile of the distribution in the ®rst survey. Because the weights are constant across surveys for a country, applying this poverty line to the second survey is consistent.
Since the cumulative distribution functions are de®ned over supports in the nonnegative real numbers, and because shifting all of the distributions of asset indexes by the same constant does not change any of the informa-tion, we added values of 10 to each household index to conduct the tests so that all asset index values would be positive. Note that sinceDs x
is not normalized by the poverty linex(i.e. the magnitude of the ``poverty gap'' (xÿy) is all that matters in the estimates of Ds x and
var Ds x), shifts in both the indicator and the
maximum poverty line do not aect the outcome of the tests.
(c) Regional decompositions
The DHS surveys are relatively short on regressors that might help explain changes in our welfare variable, but we can begin to scratch the surface with simple regional decompositions. Here we concern ourselves with how aggregate changes in poverty, as measured by the asset index, follow from the relative gains or losses of the poor within speci®c sectors as opposed to population shifts between sectors.
We shall illustrate this decomposition, proposed by Ravallion and Huppi (1991), for two sectors (ufor urban, andrfor rural). If we have Foster, Greer, and Thorbecke (1984) poverty measures (Pa, for aP0) for two
distributions (AandB) of indicators, then
PB Change in urban poverty at survey A population share
PB Change in rural poverty at survey A population share
X
Change in poverty arising from population shifts changes and population shifts
wherePt
ajis the poverty measured in sectorjfor
distribution (or time)t, andnt
jis the population
share of sector j for distribution t. This decomposition follows directly from the addi-tively separable nature of the FGT class of poverty measures.
The ®rst two components, the urban and rural intrasectoral eects, show how changes in poverty in each of the sectors contribute to the aggregate change in poverty. The third component is the contribution of changes in the distribution of the population across the two sectors. Ravallion and Huppi (1991) note that the ®nal component can be interpreted as a measure of the correlation between population shifts and changes in poverty within the sectors. This method of decomposing the changes in poverty is applied at the urban±rural and regional levels for each of the 11 countries using the asset index.
As with the dominance tests, in order to calculate poverty rates, the distributions of assets and nutrition outcomes and the poverty lines must be shifted rightward to eliminate all negative values. Although the size of the shift can be arbitrarily large, the magnitude of the FGT measures foraP1 depends on the size of the shift. These measures will change by a factor of z=zshifta. But since the poverty lines applied to the asset index are constant over the course of time for a given country (i.e.
zAzBz), the 1=zshift
a
term drops out of the decomposition, leaving the relative results unchanged.
(d) Pooling across countries for cross-country comparisons
To consistently compare asset poverty across countries, the latest surveys from each of the countries are pooled to estimate asset weights and construct household asset indexes. Five additional sub-Saharan African countries for which recent DHS data are available are added to broaden the pool of countries.21Once the asset indexes are estimated, upper and lower poverty lines are chosen as the 25th and 40th percentiles of the pooled distribution of indexes, respectively. Poverty measures are then applied to each of the national distributions separately, and the rankings of the countries are then compared to non-DHS price-depen-dent indicators of well-being. Finally, stochas-tic dominance tests are carried out for each pair of country distributions using the 50th percen-tile of the pooled distribution as the maximum poverty line.
3. DATA
The Demographic and Health Survey (DHS) program has conducted over 70 nationally representative household surveys in more than 50 countries since 1984. With funding from USAID, the program is implemented by Macro International Inc. For our purposes, 11 sub-Saharan African countries have cross-sectional surveys available for two or more periods.22 The DHS surveys are conducted in single rounds with two main survey instruments: a household schedule and an individual ques-tionnaire for women of reproductive age (15±49). The household schedule collects a list of household members and basic household demographic information and is used primarily to select respondents eligible for the individual survey. The individual survey, inter alia, provides information on household assets, reproductive histories, and the health and nutrition status of the womenÕs young children. The quality of the data is generally good with improvements made over successive rounds.
clusters. This practice was changed in the later waves (DHS II and III) to have a nationally representative sample of men, by interviewing all men age 15±49 living in every third or fourth household.
Although the designs of the surveys are not entirely uniform temporally and across coun-tries, eorts were made to standardize them so that in most cases they are reasonably compa-rable. 23 The DHS program is designed for typical self-weighted national samples of 5,000 to 6,000 women between the age of 15 and 49. In some cases the sample sizes are considerably larger, and some areas are over/under sampled.24 Household sampling weights are used to account for over- and under-sampling in various regions within surveys. For all of the countries in this study, except Uganda, the surveys are nationally representative. Districts in northern Uganda were not included in the 1988 survey because of armed con¯ict. For consistency purposes, those regions included in the 1995 Uganda survey that were not in the 1988 survey have been dropped.
4. RESULTS
(a) Asset index weights
The weights for the asset index from the factor analysis procedure appear in Table 1. The signs are all as expected, with positive weights on all but the assets that are de®ned relative to left out variables that indicate greater wealth (i.e. surface drinking water, no toilet facilities and low ¯oor quality). The magnitudes across the 12 countries are surprisingly stable. We ®nd large positive weights placed on ownership of a television and a radio, as well as piped drinking water and ¯ush toilet. Relatively large negative weights are also assigned to low-quality household ¯oor material. Zimbabwe is the exceptional country with two assets receiving the bulk of the weights: ¯ush toilet facilities and piped drink-ing water.
Weights are missing for motorized transpor-tation for Kenya and for ownership of a bicycle for Cameroon, Madagascar, Mali, Senegal and Zambia. The former is due to absence of the variable in the data. The latter was dropped because the identifying assumptions were clearly violated. After estimating the parame-ters and constructing the weights, the variance-covariance matrix of the errors (unique
elements),W, was estimated. Ideally this should be a diagonal matrix because orthogonality of the error terms is required for identi®cation (see assumption A3). Visual inspection indicates where gross violations occur, and elimination of the bicycle variables solves the problem. This led to the dropping of bicycle ownership in the case of Cameroon, Madagascar, Mali, Senegal and Zambia. Floor construction material was also dropped from the Madagascar asset index because the quality of the data on this house-hold characteristic was put into question by implausible changes over the course of the two surveys. For this reason, Madagascar was also dropped from the expanded pooled sample for crosscountry analysis.
All of the household asset indexes used in the analysis are calculated on a per household basis. The implicit assumption of doing so is that economies of scale of the assets within the household are in®nite. Asset indexes were also calculated for assets per capita (no economies of scale), and for assets divided by the square root of the number of household members, to determine if our results are sensitive to this assumption. 25 The ®ndings are robust to the
choice of equivalence scales as illustrated in Appendix A.
(b) Changes in welfare indicators over time
We summarize our results on the analysis of the asset index in Table 2, both in terms of percentage changes in the headcount ratios and our tests of stochastic dominance. Using our asset index as an alternative proxy for wealth, and employing a relative poverty line set at the 25th percentile of the wealth distribution at the time of the ®rst survey (timet), we ®nd that of the nine countries with only two surveys, all but Zimbabwe witnessed statistically signi®cant reductions in poverty.26 For the countries where poverty declined, we show ®rst order dominance in all but Togo and Uganda, where we observe second order dominance. In the cases of Kenya and Senegal, poverty worsened slightly in Senegal during the ®rst two periods, and remained unchanged in Kenya. There is improvement during survey periods two and three in both cases, with the Kenya data indi-cating second order, and the Senegal data showing ®rst order dominance. This second order dominance for Kenya, however, is limited to only the rural areas.
Between-period reductions in the headcount ratio was greatest in Ghana. In Zimbabwe the
Assets Cameroon Ghana Kenya Madagascar Mali Senegal Tanzania Togo Uganda Zambia Zimbabwe Pooled
Durables
Radio 0.095 0.103 0.075 0.123 0.082 0.052 0.161 0.099 0.121 0.086 0.062 0.098
TV 0.249 0.340 0.196 0.266 0.312 0.312 0.169 0.410 0.202 0.127 0.105 0.297
Refrigerator 0.208 0.350 0.142 0.125 0.183 0.274 0.216 0.197 0.129 0.086 0.087 0.212
Bicycle 0.023 0.008 0.024 0.020 0.011 0.009
Motorized transport 0.082 0.073 0.132 0.126 0.095 0.160 0.152 0.035 0.042 0.049 0.049
Characteristics
Piped drinking water 0.190 0.132 0.225 0.253 0.172 0.131 0.149 0.132 0.243 0.242 0.256 0.189
Surface drinking water )0.056 )0.098 )0.154 )0.143 )0.010 )0.014 )0.093 )0.057 )0.067 )0.061 )0.031 )0.074
Flush toilet 0.169 0.117 0.259 0.201 0.066 0.146 0.134 0.433 0.180 0.199 0.459 0.205
No toilet facilities )0.038 )0.020 )0.064 )0.148 )0.068 )0.100 )0.058 )0.130 )0.055 )0.080 )0.089 )0.075
FloorÐlow quality )0.148 )0.060 )0.220 )0.234 )0.099 )0.247 )0.037 )0.311 )0.272 )0.073 )0.168
Education of head 0.144 0.056 0.040 0.064a 0.142 0.124a 0.149 0.127 0.118 0.123 0.039 0.054a
aDummy variable for household head with some education.
POVERTY
COMPAR
ISONS
Table 2. Summary of asset index (25th percentile)a
Country Orders of dominance in
stochastic dominanc tests
``Poverty'' headcount P0
``+'' (``)'') indicates
improve-ment (worsening)
National Urban Rural
National Urban Rural 1st year 2nd year Change 1st year 2nd year Change 1st year 2nd year Change
Poverty line is 25th percentile of 1st year country Africa
Cameroon (1991, 1997) 1+ 1+ 1+ 27.11 24.48 )2.63 9.56 6.55 )3.01 37.76 33.60 )4.16
Ghana (1988, 1993) 1+ 1+ 1+ 24.97 8.54 )16.44 3.77 0.82 )2.96 34.72 13.91 )20.81
Kenya (1988, 1993) ND ND ND 25.45 25.09 )0.36 1.72 1.11 )0.61 30.79 30.15 )0.64
Kenya (1993, 1998) 2+ 3) 2+ 25.09 21.37 )3.72 1.11 1.20 0.09 30.15 27.18 )2.97
Kenya (1988, 1998) 2+ ND 2+ 25.45 21.37 )4.08 1.72 1.20 )0.52 30.79 27.18 )3.61
Madagascar (1992, 1997) 1+ 2) 1+ 25.47 12.50 )12.97 2.93 5.81 2.87 29.86 14.68 )15.18
Mali (1987, 1995) 1+ ND 1+ 23.02 16.02 )7.01 3.12 2.10 )1.02 29.46 22.03 )7.43
Senegal (1986, 1992) 1) ND 1) 24.58 28.80 4.21 2.80 2.66 )0.13 36.78 47.61 10.83
Senegal (1992, 1997) 1+ 1+ 1+ 28.80 24.67 )4.13 2.66 1.59 )1.07 47.61 40.29 )7.33
Senegal (1986, 1997) 2) 1+ 2) 24.58 24.67 0.08 2.80 1.59 )1.21 36.78 40.29 3.50
Tanzania (1991, 1996) ND 1+ ND 22.60 19.13 )3.48 3.73 2.13 )1.60 28.65 24.01 )4.64
Togo (1988, 1998) 2+ 2+ 2+ 25.04 20.15 )4.89 4.90 1.71 )3.18 34.94 29.87 )5.07
Uganda (1988, 1995) 2+ ND 2+ 26.77 24.35 )2.42 3.12 3.30 0.18 29.59 27.75 )1.84
Zambia (1992, 1996) 1+ 2+ 1+ 24.87 18.21 )6.66 1.18 0.73 )0.45 44.02 28.49 )15.53
Zimbabwe (1988, 1994) 1) 1) 1) 23.33 30.11 6.78 0.21 1.01 0.80 34.51 43.58 9.07
Poverty line is 40th percentile of 1st year country
Cameroon (1991, 1997) 1+ 1+ 1+ 40.28 37.28 )2.99 13.34 9.04 )4.30 55.90 49.96 )5.94
Ghana (1988, 1993) 1+ 1+ 1+ 39.90 26.54 )13.37 8.30 5.00 )3.30 54.43 41.52 )12.91
Kenya (1988, 1993) ND ND ND 40.24 35.44 )4.80 3.11 1.97 )1.14+ 48.60 42.51 )6.09
Kenya (1993, 1998) 2+ 3) 2+ 35.44 35.49 0.05 1.97 2.99 1.02+ 42.51 44.86 2.35
Kenya (1988, 1998) 2+ ND 2+ 40.24 35.49 )4.75 3.11 2.99
)0.12 48.60 44.86 )3.74
Madagascar (1992, 1997) 1+ ND 1+ 36.91 31.07 )5.83 3.85 13.52 9.66 43.35 36.80 )6.56
Mali (1987, 1995) 1+ ND 1+ 43.28 30.71 )12.57 10.30 8.27 )2.03 53.94 40.39 )13.55
Senegal (1986, 1992) 2) ND 1) 44.34 34.34 )10.00 8.96 4.58 )4.38 64.16 53.78 )10.38
Senegal (1992, 1997) 1+ 1+ 1+ 34.34 29.90 )4.44 4.58 2.34 )2.24 53.78 48.56 )5.22
Senegal (1986, 1997) 2) 1+ 2) 44.34 29.90 )14.44 8.96 2.34 )6.62 64.16 48.56 )15.60
Tanzania (1991, 1996) ND 1+ ND 39.58 33.07 )6.51 10.06 5.94 )4.13 49.02 40.86 )8.16
Togo (1988, 1998) 2+ 2+ 2+ 40.83 35.26 )5.57 11.98 6.01 )5.98 55.00 50.68 )4.32
Uganda (1988, 1995) 2+ ND 2+ 38.51 35.19 )3.32 4.31 5.34 1.03 42.58 40.00 )2.58
Zambia (1992, 1996) 2+ 3+ 1+ 39.93 39.32 )0.61 2.84 3.05 0.21 69.90 60.66 )9.25
Zimbabwe (1988, 1994) 1) 1) 1) 39.74 45.53 5.79 0.83 3.26 2.43 58.55 65.11 6.56
aND indicates that there was no stochastic dominance up to order 3.*Statistical signi®cance at the 90% level of con®dence.**Statistical signi®cance at the 95% level of
con®dence.***Statistical signi®cance at the 99% level of con®dence.
WORLD
DEVELO
PMENT
percentage of the population below the poverty line increased, corresponding to the unambig-uous increase in poverty as indicated by the 1988 distribution of assets ®rst-order dominat-ing that of 1994. Of note is that the changes in rural poverty incidence were far greater than in urban areas. This re¯ects the far lower initial level of poverty in the cities. We will return to a discussion of the regional dimensions of these changes when presenting the decompositions in the next section.
While the large changes in percentage poor measured with the asset index for Ghana may seem unrealistic, they are consistent with at least two other studies of the change in poverty over time. Using LSMS data, Demery (1995) ®nds that prior to adjusting for changes in the survey instruments, the headcount ratio for Ghana estimated using expenditures changed from 36.9 in 1988, to 41.8 in 1989, to 31.4 in 1992. After making adjustments, Demery and Mehra (1996) estimated headcount ratios of 26.1 in 1988, 31.9 in 1989 and 27.4 in 1992. The asset index estimates of 39.9 in 1988 and 26.5 in 1993 do not look that unrealistic in this context. McCulloch and BaulchÕs (1999) ®nd-ings for Zambia for 1991±96 are also consistent with those from the DHS data. Their plots of cumulative distribution functions of per adult equivalent expenditures estimated from house-hold surveys show large drops in the headcount ratio (from 0.25 to approximately 0.12) when the 25th percentile from the 1991 survey is used as the poverty line. The change is also much smaller at the higher 40th percentile poverty line, with the distributions crossing close to the 50th percentile.
To get a better grasp on what assets are driving the large changes in poverty in Ghana, simulations were run allowing individual assets to change one at a time, leaving the remaining assets unchanged. Since identical weights calculated from pooled data are applied to the assets for each survey within a country, the only source of change for the distribution of asset indexes is the ownership of the assets themselves. The method used to break down the ownership of these assets is described in Bourguignon, Fournier, and Gurgand (1998), and requires mapping changes from one cross-section to another by quantiles of the asset being changed. For Ghana, where the asset index headcount measured at the lower 25th percentile poverty line dropped from 24.97% in 1988 to 8.54% in 1993, the changes in the assets were relatively evenly distributed. The
improvements in access to quality drinking water led to the largest drop in the asset index headcount to 21.94 in 1993, followed by increases in the education of household head (22.15) and declines in the number of house-holds with low quality ¯oor material (23.30).
(c) Decompositions
The decompositions of the asset index head-count ratio suggest that intrarural eects accounted for most of the changes (Tables 3± 13). In those cases where there is a substantial drop in poverty (e.g., Ghana and Mali), migration also contributed to a decline in the headcount, generally on the order of 20%. In both of these cases, the contribution of declin-ing poverty in urban areas is small, around 5%. In the case of Zimbabwe, where the headcount increased by a signi®cant amount, it was also driven by changes in the rural areas, with only small migration and urban eects. In a few countries where we witnessed small declines in poverty (e.g., Kenya during 1988±92, Senegal during 1992±97, Tanzania, and Zambia), we also see that migration worked in the opposite direction of the intraregional eects. In these cases, the explanation for migration contribut-ing to worsencontribut-ing poverty is found in the increasing population shares in rural areas (either due to migration, higher fertility, or a combination of both).
The regional decompositions also paint a picture of dierent contributions to the change in headcount poverty levels (Tables 14±23). Particularly noteworthy is the case of Ghana where the Upper West, Upper East and Northern regions, in combination referred to as the Savannah region, played a large role in the overall decline in rural poverty. To a lesser extent this is true for Brong Ahafo, a more prosperous forest zone region. In another example of how the regional decompositions inform the regional aspects of changes in welfare, the West and Manicaland regions made particularly large contributions to the increases in our headcount measure for Zimbabwe.
(d) Cross-country analysis
Table 3. Cameroon: decomposition of changes in ``poverty'' between 1991 and 1997
Poverty Total change
Intrasectoral eects
1991 1997 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1991
Headcount 27.11 24.48 )2.63 )1.11 )2.75 1.29 )0.06 Poverty gap 0.28 0.24 )0.04 )0.01 )0.05 0.02 0.00 Poverty gap
squared
0.005 0.004 )0.001 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.42 1.04 )0.49 0.02
Poverty gap 1.00 0.18 1.16 )0.40 0.06
Poverty gap squared
1.00 0.12 1.16 )0.35 0.07
Poverty line is 40th percentile in 1991
Headcount 40.28 37.28 )2.99 )1.36 )3.53 1.99 )0.09 Poverty gap 0.85 0.76 )0.09 )0.03 )0.10 0.04 0.00 Poverty gap
squared
0.023 0.020 )0.003 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.45 1.18 )0.66 0.03
Poverty gap 1.00 0.33 1.12 )0.49 0.04
Poverty gap squared
1.00 0.23 1.14 )0.43 0.05
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Table 4. Ghana: decomposition of changes in ``poverty'' between 1988 and 1993
Poverty Total change
Intrasectoral eects
1988 1993 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1988
Headcount 24.97 8.54 )16.44 )0.93 )14.26 )2.95 1.70 Poverty gap 0.31 0.08 )0.23 )0.01 )0.21 )0.04 0.03 Poverty gap
squared
0.005 0.001 )0.004 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.06 0.87 0.18 )0.10
Poverty gap 1.00 0.03 0.92 0.17 )0.12
Poverty gap squared
1.00 0.02 0.95 0.16 )0.13
Poverty line is 40th percentile in 1988
Headcount 39.90 26.54 )13.37 )1.04 )8.85 )4.40 0.92 Poverty gap 0.62 0.22 )0.40 )0.02 )0.35 )0.07 0.04 Poverty gap
squared
0.014 0.004 )0.010 0.00 )0.01 0.00 0.00
Share of total change
Headcount 1.00 0.08 0.66 0.33 )0.07
Poverty gap 1.00 0.05 0.87 0.19 )0.11
Poverty gap squared
1.00 0.03 0.92 0.17 )0.12
*
Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
Table 5. Kenya: decomposition of changes in ``poverty'' between 1988±1993, and 1993±1997
Poverty Total change
Intrasectoral eects
1988 1992 Urban Rural Migration Interaction
1988±1993
Poverty line is 25th percentile in 1988
Headcount 25.45 25.09 )0.36 )0.11 )0.52 0.27 )0.00 Poverty gap 2.65 2.65 )0.00 )0.01 )0.02 0.03 0.00 Poverty gap
squared
0.277 0.280 0.002 )0.001 )0.000 0.003 0.000
Share of total change
Headcount 1.00 0.31 1.44 )0.75 0.00
Poverty gap 1.00 3.56 12.87 )15.39 )0.03
Poverty gap squared
1.00 )0.36 )0.20 1.54 0.02
Poverty line is 40th percentile in 1988
Headcount 40.24 35.44 )4.80 )0.21 )4.98 0.43 )0.05 Poverty gap 6.31 6.14 )0.18 )0.03 )0.22 0.07 )0.00 Poverty gap
squared
0.869 0.861 )0.008 )0.003 )0.015 0.010 )0.000
Share of total change
Headcount 1.00 0.04 1.04 )0.09 0.01
Poverty gap 1.00 0.15 1.24 )0.39 0.01
Poverty gap squared
1.00 0.36 1.86 )1.22 0.00
Poverty Total change
Intrasectoral eects
1993 1998 Urban Rural Migration Interaction 1993±1998
Headcount 25.09 21.37 )3.72 0.01 )2.46 )1.44 0.15 Poverty gap 2.65 2.31 )0.34 0.02 )0.22 )0.16 0.02 Poverty gap
squared
0.280 0.231 )0.049 0.003 )0.038 )0.017 0.003
Share of total change
Headcount 1.00 )0.00 0.66 0.39 )0.04
Poverty gap 1.00 )0.05 0.64 0.46 )0.05
Poverty gap squared
1.00 )0.05 0.77 0.34 )0.06
Poverty line is 40th percentile in 1988
Headcount 35.44 35.49 0.06 0.18 1.95 )2.00 )0.07 Poverty gap 6.14 5.45 )0.68 0.02 )0.38 )0.36 0.03 Poverty gap
squared
0.861 0.742 )0.119 0.005 )0.080 )0.051 0.006
Share of total change
Headcount 1.00 3.12 33.98 )34.95 )1.15
Poverty gap 1.00 )0.03 0.55 0.52 )0.04
Poverty gap squared
1.00 )0.04 0.67 0.43 )0.05
*Signi®cance at the 95% levels of con®dence. **
Table 6. Madagascar: decomposition of changes in ``poverty'' between 1992 and 1997
Poverty Total change
Intrasectoral eects
1992 1997 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1992
Headcount 34.79 31.20 )3.59 1.56 )3.21 )3.05 1.11 Poverty gap 0.19 0.17 )0.02 0.01 )0.02 )0.02 0.01 Poverty gap
squared
0.003 0.003 )0.0003 0.000 )0.000 )0.000 0.000
Share of total change
Headcount 1.00 )0.43 0.89 0.85 )0.31
Poverty gap 1.00 )0.44 0.84 0.90 )0.30
Poverty gap squared
1.00 )0.44 0.84 0.90 )0.30
Poverty line is 40th percentile in 1992
Headcount 43.72 40.71 )3.01 2.01 )2.60 )3.70 1.28 Poverty gap 0.73 0.65 )0.08 0.03 )0.07 )0.06 0.02 Poverty gap
squared
0.017 0.015 )0.002 0.001 )0.001 )0.001 0.001
Share of total change
Headcount 1.00 )0.67 0.86 1.23 )0.42
Poverty gap 1.00 )0.42 0.89 0.84 )0.30
Poverty gap squared
1.00 )0.43 0.87 0.87 )0.31
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Table 7. Mali: decomposition of changes in ``poverty'' between 1987 and 1995
Poverty Total change
Intrasectoral eects
1987 1995 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1987
Headcount 23.02 16.02 )7.01 )0.25 )5.62 )1.51 0.37 Poverty gap 0.29 0.24 )0.05 )0.00 )0.03 )0.02 0.00 Poverty gap
squared
0.004 0.004 )0.001 )0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.04 0.80 0.22 )0.05
Poverty gap 1.00 0.02 0.64 0.38 )0.04
Poverty gap squared
1.00 0.02 0.48 0.53 )0.03
Poverty line is 45th percentile in 1987
Headcount 43.28 30.71 )12.57 )0.50 )10.24 )2.50 0.66 Poverty gap 0.36 0.29 )0.07 )0.00 )0.05 )0.02 0.00 Poverty gap
squared
0.006 0.005 )0.001
)0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.04 0.81 0.20 )0.05
Poverty gap 1.00 0.03 0.69 0.33 )0.05
Poverty gap squared
1.00 0.02 0.53 0.48 )0.04
*
Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
Table 8. Senegal: decomposition of changes in ``poverty'' between 1986±1992, and 1992±1997
Poverty Total change
Intrasectoral eects
1986 1992 Urban Rural Migration Interaction
1986±1992
Poverty line is 25th percentile in 1986
Headcount 24.58 28.80 4.21
)0.05 6.94 )2.03 )0.65 Poverty gap 0.36 0.56 0.20 0.00 0.25
)0.03 )0.02 Poverty gap
squared
0.007 0.013 0.006 0.00 0.01
)0.00 )0.00
Share of total change
Headcount 1.00 )0.01 1.65 )0.48 )0.16
Poverty gap 1.00 0.01 1.26 )0.15 )0.12
Poverty gap squared
1.00 0.02 1.18 )0.10 )0.11
Poverty line is 40th percentile in 1986
Headcount 44.34 34.34 )10.00 )1.57 )5.37 )3.29 0.24 Poverty gap 0.53 0.74 0.21
)0.01 0.29 )0.04 )0.03 Poverty gap
squared
0.012 0.021 0.008 0.00 0.01
)0.00 )0.00
Share of total change
Headcount 1.00 0.16 0.54 0.33 )0.02
Poverty gap 1.00 )0.04 1.37 )0.20 )0.13
Poverty gap squared
1.00 0.01 1.22 )0.12 )0.11
Poverty Total change
Intrasectoral eects
1992 1997 Urban Rural Migration Interaction
1992±1997
Poverty line is 25th percentile in 1986
Headcount 28.80 24.67 )4.13 )0.45 )4.26 0.67 )0.09 Poverty gap 0.56 0.45 )0.11 )0.01 )0.11 0.01 )0.00 Poverty gap
squared
0.013 0.010 )0.003 )0.00 )0.00 0.00 )0.00
Share of total change
Headcount 1.00 0.11 1.03 )0.16 0.02
Poverty gap 1.00 0.07 1.03 )0.12 0.02
Poverty gap squared
1.00 0.06 1.03 )0.11 0.02
Poverty line is 40th percentile in 1986
Headcount 34.34 29.90 )4.44 )0.94 )4.20 0.77 )0.07 Poverty gap 0.74 0.61 )0.13 )0.01 )0.13 0.02 )0.00 Poverty gap
squared
0.021 0.016 )0.004 )0.00 )0.00 0.00 )0.00
Share of total change
Headcount 1.00 0.21 0.94 )0.17 0.02
Poverty gap 1.00 0.09 1.02 )0.13 0.02
Poverty gap squared
1.00 0.07 1.02 )0.12 0.02
*Signi®cance at the 95% levels of con®dence. **
Table 9. Tanzania: decomposition of changes in ``poverty'' between 1991 and 1996
Poverty Total change
Intrasectoral eects
1991 1996 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1991
Headcount 22.60 19.13 )3.48 )0.39 )3.51 0.48 )0.06 Poverty gap 0.42 0.36 )0.06 )0.01 )0.06 0.01 )0.00 Poverty gap
squared
0.010 0.009 )0.00 )0.00 )0.00 0.00 )0.00
Share of total change
Headcount 1.00 0.11 1.01 )0.14 0.02
Poverty gap 1.00 0.17 0.97 )0.15 0.01
Poverty gap squared
1.00 0.30 0.94 )0.23 0.00
Poverty line is 40th percentile in 1991
Headcount 39.58 33.07 )6.51 )1.00 )6.18 0.76 )0.08 Poverty gap 0.91 0.76 )0.15 )0.02 )0.15 0.02 )0.00 Poverty gap
squared
0.029 0.025 )0.00 )0.00 )0.00 0.00 )0.00
Share of total change
Headcount 1.00 0.15 0.95 )0.12 0.01
Poverty gap 1.00 0.14 0.97 )0.12 0.01
Poverty gap squared
1.00 0.18 0.96 )0.15 0.01
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Table 10. Togo: decomposition of changes in ``poverty'' between 1988 and 1998
Poverty Total change
Intrasectoral eects
1988 1998 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1988
Headcount 25.04 20.15 )4.89 )1.05 )3.40 )0.47 0.03 Poverty gap 0.28 0.22 )0.06 )0.01 )0.05 )0.01 0.00 Poverty gap
squared
0.004 0.003 )0.001 )0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.21 0.69 0.10
)0.01
Poverty gap 1.00 0.12 0.79 0.10 )0.01
Poverty gap squared
1.00 0.05 0.86 0.11 )0.02
Poverty line is 40th percentile in 1988
Headcount 40.83 35.26 )5.57 )1.97 )2.90 )0.68 )0.03 Poverty gap 0.76 0.65 )0.11 )0.02 )0.08 )0.01 0.00 Poverty gap
squared
0.019 0.016 )0.003
)0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.35 0.52 0.12 0.00
Poverty gap 1.00 0.15 0.72 0.14 )0.01
Poverty gap squared
1.00 0.12 0.78 0.12 )0.01
*
Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
Table 11. Uganda: decomposition of changes in ``poverty'' between 1988 and 1996
Poverty Total change
Intrasectoral eects
1988 1996 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1988
Headcount 26.77 24.35 )2.42 0.02 )1.64 )0.86 0.07 Poverty gap 0.32 0.30 )0.02 0.00 )0.01 )0.01 0.00 Poverty gap
squared
0.006 0.005 )0.000 0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 )0.01 0.68 0.36 )0.03
Poverty gap 1.00 )0.03 0.47 0.59 )0.03
Poverty gap squared
1.00 )0.04 0.66 0.42 )0.04
Poverty line is 40th percentile in 1988
Headcount 38.51 35.19 )3.32 0.11 )2.30 )1.25 0.12 Poverty gap 0.54 0.51 )0.04 0.00 )0.02 )0.02 0.00 Poverty gap
squared
0.012 0.011 )0.001 0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 )0.03 0.69 0.37 )0.04
Poverty gap 1.00 )0.02 0.57 0.47 )0.03
Poverty gap squared
1.00 )0.03 0.60 0.46 )0.03
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Table 12. Zambia: decomposition of changes in ``poverty'' between 1992 and 1996
Poverty Total change
Intrasectoral eects
1992 1996 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1992
Headcount 24.87 18.21 )6.66 )0.20 )8.59 3.28 )1.15 Poverty gap 0.41 0.25 )0.16 )0.00 )0.18 0.05 )0.02 Poverty gap
squared
0.009 0.004 )0.004 )0.00 )0.00 0.00 )0.00
Share of total change
Headcount 1.00 0.03 1.29
)0.49 0.17
Poverty gap 1.00 0.03 1.15 )0.34 0.15
Poverty gap squared
1.00 0.03 1.09 )0.27 0.15
Poverty line is 40th percentile in 1992
Headcount 39.93 39.32 )0.61 0.10 )5.11 5.13 )0.72 Poverty gap 1.12 0.90 )0.23 )0.01 )0.32 0.15 )0.04 Poverty gap
squared
0.039 0.027 )0.013 )0.00 )0.02 0.01 )0.00
Share of total change
Headcount 1.00 )0.16 8.40 )8.44 1.19
Poverty gap 1.00 0.03 1.43 )0.65 0.19
Poverty gap squared
1.00 0.03 1.22 )0.41 0.16
*
Signi®cance at the 95% levels of con®dence.
Table 14. Cameroon: decomposition of changes in ``poverty'' between 1991 and 1997
Poverty Total change
Intrasectoral eects
1991 1997 Yaounde/ Douala
N & Adam
Cent, S, & E
West & Littoral
NW &SW
Migra-tion
Interac-tion
Poverty line is 25th percentile in 1991
Headcount 27.11 24.48 )2.63 0.04 1.76 )0.16 )1.75 )1.26 )0.78 )0.48 Poverty gap 0.28 0.24 )0.04 )0.00 0.02 )0.00 )0.03 )0.02 )0.00 )0.01 Poverty gap
squared
0.005 0.004 )0.001
)0.00 0.00 )0.00 )0.00 )0.00 )0.00 )0.00
Share of total change
Headcount 1.00
)0.01 )0.67 0.06 0.67 0.48 0.29 0.18 Poverty gap 1.00 0.01 )0.45 0.03 0.75 0.46 0.07 0.13 Poverty gap
squared
1.00 0.01 )0.49 0.12 0.75 0.45 0.01 0.15
Poverty line is 40th percentile in 1991
Headcount 40.28 37.28 )2.99 )0.05 1.75 )0.42 )2.07 )1.27 )0.45 )0.49 Poverty gap 0.85 0.76 )0.09 )0.00 0.05 )0.00 )0.07 )0.04 )0.01 )0.01 Poverty gap
squared
0.023 0.020 )0.003 )0.00 0.00 )0.00 )0.00 )0.00 )0.00 )0.00
Share of total change
Headcount 1.00 0.02 )0.59 0.14 0.69 0.42 0.15 0.16 Poverty gap 1.00 0.01 )0.54 0.05 0.72 0.45 0.16 0.15 Poverty gap
squared
1.00 0.01 )0.53 0.06 0.74 0.46 0.10 0.15
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Table 13. Zimbabwe: decomposition of changes in ``poverty'' between 1988 and 1994
Poverty Total change
Intrasectoral eects
1988 1994 Urban Rural Migration Interaction
Poverty line is 25th percentile in 1988
Headcount 23.33 30.11 6.78 0.26 6.12 0.32 0.08
Poverty gap 0.33 0.48 0.15 0.00 0.14 0.00 0.00
Poverty gap squared
0.007 0.010 0.003 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.04 0.90 0.05 0.01
Poverty gap 1.00 0.01 0.94 0.03 0.01
Poverty gap squared
1.00 0.00 0.95 0.03 0.01
Poverty line is 40th percentile in 1988
Headcount 39.74 45.53 5.79 0.79 4.42 0.54 0.04
Poverty gap 0.93 1.19 0.26 0.01 0.24 0.01 0.00
Poverty gap squared
0.029 0.040 0.011 0.00 0.01 0.00 0.00
Share of total change
Headcount 1.00 0.14 0.76 0.09 0.01
Poverty gap 1.00 0.05 0.90 0.05 0.01
Poverty gap squared
1.00 0.02 0.93 0.04 0.01
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Poverty Total change
Intrasectoral eects
1988 1993 Western Central Greater
Accra
Eastern Volta Ashanti Brong
Ahafo
Upper W, E & N
Migration Interac-tion
Poverty Line is 25th Percentile in 1988
Headcount 24.97 8.54 )16.44 0.14 )1.03 )0.48 )1.65 )2.08 )2.93 )3.01 )5.38 )2.34 2.33
Poverty gap 0.31 0.08 )0.23 0.00 )0.02 )0.00 )0.02 )0.02 )0.04 )0.05 )0.08 )0.04 0.04
Poverty gap squared
0.005 0.001 )0.004 0.00 )0.00 )0.00 )0.00 )0.00 )0.00 )0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00
)0.01 0.06 0.03 0.10 0.13 0.18 0.18 0.33 0.14 )0.14
Poverty gap 1.00 )0.01 0.07 0.02 0.10 0.10 0.16 0.23 0.33 0.16 )0.16
Poverty gap squared
1.00 )0.00 0.08 0.01 0.09 0.09 0.14 0.25 0.35 0.18 )0.18
Poverty line is 40th percentile in 1988
Headcount 39.90 26.54 )13.37 0.66 )1.13 )0.55 )1.43 )1.31 )1.26 )2.42 )5.84 )2.87 2.79
Poverty gap 0.62 0.22 )0.40 0.00 )0.03 )0.01 )0.04 )0.04 )0.06 )0.08 )0.13 )0.06 0.06
Poverty gap squared
0.014 0.004 )0.010 0.00 )0.00 )0.00 )0.00 )0.00 )0.00 )0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 )0.05 0.08 0.04 0.11 0.10 0.09 0.18 0.44 0.21 )0.21
Poverty gap 1.00 )0.01 0.08 0.03 0.10 0.10 0.16 0.21 0.33 0.16 )0.15
Poverty gap squared
1.00 )0.01 0.08 0.02 0.10 0.10 0.15 0.23 0.33 0.16 )0.16
*Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
POVERTY
COMPAR
ISONS
Table 16. Kenya: decomposition of changes in ``poverty'' between 1988±1993, and 1993±1997
Poverty Total
change
Intrasectoral eects
1988 1993 Nairobi Central Coast Eastern Nyanza Rift Valley Western Migration
Interac-tion
1988±1993
Poverty line is 25th percentile in 1988
Headcount 25.45 25.09 )0.36 0.06 0.47 )0.18 1.20 1.05 )0.70 )2.25 0.22 )0.24
Poverty gap 2.65 2.65 )0.00 0.00 0.01 )0.10 0.10 0.10 0.05 )0.17 0.04 )0.03
Poverty gap squared 0.277 0.280 0.002 0.000 0.001 )0.017 0.008 0.009 0.015 )0.015 0.005 )0.004
Share of total change
Headcount 1.00 )0.17 )1.30 0.50 )3.29 )2.89 1.92 6.18 )0.61 0.65
Poverty gap 1.00 )0.27 )6.29 49.63 )50.97 )51.49 )24.26 89.19 )20.04 15.51
Poverty gap squared 1.00 0.00 0.57 )8.37 3.98 4.48 7.35 )7.50 2.50 )2.02
Poverty line is 40th percentile in 1988
Headcount 40.24 35.44 )4.80 0.03 )0.79 )0.31 0.55 1.58 )2.61 )3.17 0.24 )0.32
Poverty gap 6.31 6.14 )0.18 0.01 0.03 )0.12 0.24 0.23 )0.08 )0.51 0.07 )0.06
Poverty gap squared 0.869 0.861 )0.008 0.001 0.005 )0.031 0.031 0.031 0.013 )0.060 0.012 )0.010
Share of total change
Headcount 1.00 )0.01 0.16 0.06 )0.11 )0.33 0.54 0.66 )0.05 0.07
Poverty gap 1.00 )0.05 )0.19 0.70 )1.39 )1.34 0.44 2.88 )0.41 0.36
Poverty gap squared 1.00 )0.08 )0.60 3.99 )4.00 )3.97 )1.69 7.61 )1.56 1.30
continued opposite
WORLD
DEVELO
PMENT
Poverty Total change
Intrasectoral eects
1993 1998 Nairobi Central Coast Eastern Nyanza Rift Valley Western Migration
Interac-tion
1993±1997
Poverty line is 25th percentile in 1988
Headcount 25.09 21.37 )3.72 )0.08 )0.59 )0.07 )1.16 )1.54 )1.35 0.62 0.85 )0.40
Poverty gap 2.65 2.31 )0.34 )0.00 0.00 0.01 0.03 )0.16 )0.27 0.03 0.10 )0.07
Poverty gap squared 0.280 0.231 )0.049 )0.000 0.000 0.003 0.007 )0.018 )0.043 0.001 0.010 )0.008
Share of total change
Headcount 1.00 0.02 0.16 0.02 0.31 0.41 0.36 )0.17 )0.23 0.11
Poverty gap 1.00 0.00 )0.00 )0.03 )0.08 0.49 0.81 )0.09 )0.30 0.20
Poverty gap squared 1.00 0.00 )0.00 )0.06 )0.14 0.37 0.89 )0.03 )0.21 0.17
Poverty line is 40th percentile in 1988
Headcount 35.44 35.49 0.06 )0.08 )0.26 )0.03 )0.73 )0.65 )0.90 2.07 1.04 )0.40
Poverty gap 6.14 5.45 )0.68 )0.01 )0.04 0.01 )0.11 )0.34 )0.44 0.16 0.22 )0.13
Poverty gap squared 0.861 0.742 )0.119 )0.001 )0.003 0.004 0.001 )0.052 )0.091 0.014 0.031 )0.021
Share of total change
Headcount 1.00 )1.31 )4.60 )0.52 )12.79 )11.37 )15.70 36.07 18.23 )7.01
Poverty gap 1.00 0.02 0.06 )0.02 0.16 0.50 0.65 )0.24 )0.32 0.19
Poverty gap squared 1.00 0.01 0.03 )0.03 )0.01 0.44 0.77 )0.12 )0.26 0.18
*
Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
POVERTY
COMPAR
ISONS
Table 17. Madagascar: decomposition of changes in ``poverty'' between 1992 and 1997
Poverty Total
change
Intrasectoral eects
1992 1997
Antananar-ivo
Fian-arantsoa
Toamasina Mahajanga Toliary Antsirana Migration Interaction
Poverty line is 25th percentile in 1992
Headcount 34.79 31.20 )3.59 )0.46 )0.07 )1.49 )0.69 0.87 )1.07 )0.86 0.18
Poverty gap 0.19 0.17 )0.02 )0.01 )0.00 )0.01 )0.01 0.01 )0.00 )0.00 0.00
Poverty gap squared
0.003 0.003 )0.0003 )0.000 )0.000 )0.000 )0.000 0.000 )0.000 )0.000 0.000
Share of total change
Headcount 1.00 0.13 0.02 0.41 0.19 )0.24 0.30 0.24 )0.05
Poverty gap 1.00 0.32 0.19 0.62 0.28 )0.58 0.07 0.18 )0.09
Poverty gap squared
1.00 0.32 0.19 0.62 0.28 )0.58 0.07 0.18 )0.09
Poverty line is 40th percentile in 1992
Headcount 43.72 40.71 )3.01 )0.48 0.21 )1.66 )1.49 1.44 )0.77 )0.54 0.28
Poverty gap 0.73 0.65 )0.08 )0.01 )0.01 )0.03 )0.02 0.02 )0.02 )0.02 0.00
Poverty gap squared
0.017 0.015 )0.002 )0.000 )0.000 )0.001 )0.000 0.001 )0.000 )0.000 0.000
Share of total change
Headcount 1.00 0.16 )0.07 0.55 0.49 )0.48 0.26 0.18 )0.09
Poverty gap 1.00 0.17 0.08 0.44 0.24 )0.32 0.23 0.22 )0.06
Poverty gap squared
1.00 0.22 0.11 0.51 0.24 )0.41 0.18 0.21 )0.07
*Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
WORLD
DEVELO
PMENT
Poverty Total change Intrasectoral eects
1987 1995 Kayes,
Koulikoro
Sikasso, Segou
Mopti, Gao, Timbuctou
Bamako Migration Interaction
Poverty line is 25th percentile in 1987
Headcount 23.02 16.02 )7.01 0.05
)3.73 )2.28 )0.05 )1.55 0.54
Poverty gap 0.29 0.24 )0.05 0.01 )0.04 )0.00 )0.00 )0.02 0.00
Poverty gap squared
0.004 0.004 )0.001 0.00 )0.00 0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 )0.01 0.53 0.32 0.01 0.22 )0.08
Poverty gap 1.00 )0.18 0.82 0.05 0.00 0.39 )0.08
Poverty gap squared
1.00 )0.31 1.10 )0.26 0.00 0.54 )0.08
Poverty line is 45th percentile in 1987
Headcount 43.28 30.71 )12.57 )2.21 )5.39 )3.28 )0.08 )2.27 0.65
Poverty gap 0.36 0.29 )0.07 0.01 )0.05 )0.01 )0.00 )0.02 0.01
Poverty gap squared
0.006 0.005 )0.001 0.00 )0.00 0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.18 0.43 0.26 0.01 0.18 )0.05
Poverty gap 1.00 )0.08 0.71 0.11 0.00 0.33 )0.07
Poverty gap squared
1.00 )0.26 1.01 )0.16 0.00 0.49 )0.08
*
Signi®cance at the 95% levels of con®dence.
**Signi®cance at the 99% levels of con®dence.
POVERTY
COMPAR
ISONS
Table 19. Senegal: decomposition of changes in ``poverty'' between 1986±1992, and 1992±1997
Poverty Total change
Intrasectoral eects
1986 1992 West Central South North East
Migration Interaction
1986±1992
Poverty Line is 25th Percentile in 1986
Headcount 24.58 28.80 4.21 0.84
)1.50 1.54 2.72 0.11 0.50 Poverty gap 0.36 0.56 0.20 0.03 0.02 0.05 0.09
)0.00 0.01 Poverty gap
squared
0.007 0.013 0.006 0.00 0.00 0.00 0.00
)0.00 0.00
Share of total change
Headcount 1.00 0.20 )0.36 0.36 0.65 0.03 0.12 Poverty gap 1.00 0.15 0.10 0.23 0.45 )0.00 0.07 Poverty gap
squared
1.00 0.14 0.16 0.21 0.42 )0.00 0.07
Poverty line is 40th percentile in 1986
Headcount 44.34 34.34 )10.00 3.09 5.10 )7.50 )8.02 )1.13 )1.55 Poverty gap 0.53 0.74 0.21 0.02 0.01 0.06 0.10 0.00 0.02
Poverty gap squared
0.012 0.021 0.008 0.00 0.00 0.00 0.00
)0.00 0.00
Share of total change
Headcount 1.00 )0.31 )0.51 0.75 0.80 0.11 0.15 Poverty gap 1.00 0.12 0.05 0.27 0.47 0.01 0.08 Poverty gap
squared
1.00 0.14 0.14 0.22 0.43 )0.00 0.07
Poverty Total change
Intrasectoral eects 1992 1997 West Central South North
East
Migration Interaction
1992±1997
Poverty line is 25th percentile in 1986
Headcount 28.80 24.67 )4.13 )1.42 )1.30 0.88 )2.60 )0.33 0.65 Poverty gap 0.56 0.45 )0.11 )0.03 )0.04 0.03 )0.08 )0.01 0.02 Poverty gap
squared
0.013 0.010 )0.003 )0.00 )0.00 0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.34 0.32 )0.21 0.63 0.08 )0.16 Poverty gap 1.00 0.27 0.36 )0.32 0.78 0.10 )0.20 Poverty gap
squared
1.00 0.23 0.37 )0.36 0.88 0.12 )0.23
Poverty line is 40th percentile in 1986
Headcount 34.34 29.90 )4.44 )1.87 )1.18 1.43 )3.30 )0.42 0.90 Poverty gap 0.74 0.61 )0.13 )0.04 )0.05 0.04 )0.10 )0.01 0.03 Poverty gap
squared
0.021 0.016 )0.004 )0.00 )0.00 0.00 )0.00 )0.00 0.00
Share of total change
Headcount 1.00 0.42 0.27 )0.32 0.74 0.10 )0.20 Poverty gap 1.00 0.30 0.35 )0.32 0.78 0.10 )0.20 Poverty gap
squared
1.00 0.25 0.36 )0.35 0.84 0.11 )0.22
*Signi®cance at the 95% levels of con®dence. **Signi®cance at the 99% levels of con®dence.
Poverty Total change
Intrasectoral eects
1988 1998 Maritime Plateau Central Kara Savane Migration Interaction
Poverty line is 25th percentile in 1988
Headcount 25.04 20.15 )4.89 0.51 )0.05 )1.76 )0.76 )2.43 )0.75 0.33
Poverty gap 0.28 0.22 )0.06 0.01 0.01 )0.02 )0.01 )0.04 )0.01 0.01
Poverty gap squared 0.004 0.003 )0.001 0.000 0.000 )0.000 )0.000 )0.001 )0.000 0.000
Share of total change
Headcount 1.00
)0.11 0.01 0.36 0.15 0.50 0.15 )0.07
Poverty gap 1.00 )0.23 )0.18 0.41 0.21 0.73 0.19 )0.12
Poverty gap squared 1.00 )0.31 )0.33 0.45 0.19 0.93 0.23 )0.16
Poverty line is 40th percentile in 1988
Headcount 40.83 35.26 )5.57 )0.66 0.15 )1.62 )1.33 )1.72 )0.68 0.29
Poverty gap 0.76 0.65 )0.11 0.02 0.01 )0.05 )0.02 )0.07 )0.02 0.01
Poverty gap squared 0.019 0.016 )0.003 0.001 0.001 )0.001 )0.001 )0.002 )0.001 0.000
Share of total change
Headcount 1.00 0.12 )0.03 0.29 0.24 0.31 0.12 )0.05
Poverty gap 1.00 )0.22 )0.09 0.43 0.19 0.60 0.19 )0.09
Poverty gap squared 1.00 )0.25 )0.17 0.43 0.19 0.72 0.20 )0.12
*Signi®cance at the 95% levels of con®dence.
**
Signi®cance at the 99% levels of con®dence.
POVERTY
COMPAR
ISONS
Table 21. Uganda: decomposition of changes in ``poverty'' between 1988 and 1996a
Poverty Total
change
Intrasectoral eects
1988 1996 West Nile East Central West South
West
Kampala Migration Interaction
Poverty line is 25th percentile in 1988
Headcount 26.77 24.35 )2.42 0.74 )2.64 0.40 )0.92 0.11 )0.03 0.02 )0.10
Poverty gap 0.32 0.30 )0.02 0.00 )0.01 0.00 )0.01 )0.01 0.00 0.00 0.00
Poverty gap squared 0.006 0.005 0.000 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 )0.30 1.09 )0.17 0.38 )0.05 0.01 )0.01 0.04
Poverty gap 1.00 )0.24 0.54 0.10 0.28 0.43 0.00 )0.04 )0.06
Poverty gap squared 1.00 )0.23 0.40 0.10 0.31 0.52 0.00 )0.08 )0.02
Poverty line is 40th percentile in 1988
Headcount 38.51 35.19 )3.32 0.15 )1.52 )0.86 )0.21 )0.83 0.02 )0.46 0.39
Poverty gap 0.54 0.51 )0.04 0.01 )0.02 )0.01 )0.01 )0.01 0.00 0.00 0.00
Poverty gap squared 0.012 0.011 )0.001 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 )0.05 0.46 0.26 0.06 0.25 )0.01 0.14 )0.12
Poverty gap 1.00 )0.14 0.50 0.13 0.19 0.37 0.00 0.03 )0.07
Poverty gap squared 1.00 )0.21 0.46 0.10 0.27 0.46 0.00 )0.04 )0.05
aandindicate signi®cance at the 95% and 99% levels of con®dence, respectively.
WORLD
DEVELO
PMENT
Poverty Total change
Intrasectoral eects
1992 1996 Central
Copper-belt
Eastern Luapula Lusaka
North-ern
N-West-ern
South-ern
Western
Migra-tion
Interac-tion
Poverty line is 25th percentile in 1992
Headcount 24.87 18.21 )6.66
)0.53 0.11 )1.48 )1.66 0.15 )3.02 )0.52 )0.29 )0.73 2.97 )1.67
Poverty gap 0.41 0.25 )0.16 )0.01 0.00 )0.04 )0.03 0.00 )0.06 )0.01 )0.01 )0.01 0.05 )0.03
Poverty gap squared 0.009 0.004 )0.004 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.08 )0.02 0.22 0.25 )0.02 0.45 0.08 0.04 0.11 )0.45 0.25
Poverty gap 1.00 0.07 )0.01 0.28 0.16 0.00 0.41 0.05 0.06 0.07 )0.32 0.22
Poverty gap squared 1.00 0.07 0.00 0.31 0.12 0.01 0.38 0.04 0.07 0.06 )0.25 0.20
Poverty line is 40th percentile in 1992
Headcount 39.93 39.32 )0.61 )0.75 0.56 )0.76 )1.21 0.32 )2.01 )0.44 1.00 )0.44 4.62 )1.50
Poverty gap 1.12 0.90 )0.23 )0.02 0.01 )0.07 )0.06 0.01 )0.12 )0.02 0.00 )0.02 0.13 )0.07
Poverty gap squared 0.039 0.027 )0.013 0.00 0.00 0.00 0.00 0.00 )0.01 0.00 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 1.24 )0.92 1.25 1.99 )0.53 3.30 0.72 )1.65 0.72 )7.58 2.46
Poverty gap 1.00 0.11 )0.04 0.29 0.25 )0.03 0.52 0.09 )0.01 0.11 )0.59 0.31
Poverty gap squared 1.00 0.08 )0.01 0.29 0.18 )0.01 0.43 0.06 0.05 0.08 )0.38 0.24
aandindicate signi®cance at the 95% and 99% levels of con®dence, respectively.
POVERTY
COMPAR
ISONS
Table 23.Zimbabwe: decomposition of changes in ``poverty'' between 1998 and 1994a
Poverty Total
change
Intrasectoral eects
1998 1994
Man-icaland
Mashonaland Matabeleland
Mid-lands Mas-vingo
Harrare/Chi-tungwiza
Bul-awayo
Migra-tion
Interac-tion
Central East West North South
Poverty line is 25th percentile in 1998
Headcount 23.33 30.11 6.78 1.95 0.91 1.61 2.94 0.05 0.05 1.18 0.35 0.08
)0.05 )1.08 )0.81
Poverty gap squared 0.33 0.48 0.15 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Poverty gap 0.007 0.010 0.003 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.29 0.13 0.24 0.43 )0.05 0.01 0.17 0.05 0.01 )0.01 )0.16 )0.12
Poverty gap 1.00 0.24 0.18 0.19 0.39 )0.06 )0.05 0.26 0.05 0.00 0.00 )0.08 )0.12
Poverty gap squared 1.00 0.24 0.21 0.17 0.41 )0.08 )0.10 0.28 0.04 0.00 )0.01 )0.07 )0.11
Intrasectoral eects
Poverty line is 40th percentile in 1998
Headcount 39.74 45.53 5.79 2.94 0.42 1.10 2.73
)0.22 0.30 0.84 0.34 0.26 0.04 )2.53 )0.43
Poverty gap squared 0.93 1.19 0.26 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Poverty gap 0.029 0.040 0.011 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Share of total change
Headcount 1.00 0.51 0.07 0.19 0.47 )0.04 0.05 0.14 0.06 0.04 0.01 )0.44 )0.07
Poverty gap 1.00 0.32 0.14 0.19 0.42 )0.05 )0.02 0.20 0.06 0.01 0.00 )0.18 )0.10
Poverty gap squared 1.00 0.27 0.17 0.19 0.41 )0.06 )0.05 0.24 0.06 0.01 0.00 )0.12 )0.11
aandindicate signi®cance at the 95% and 99% levels of con®dence, respectively.
WORLD
DEVELO
PMENT
their level of poverty. A number of countries where only one DHS survey has been conduc-ted are included in the table, as well as the most recent survey for those countries for which we examined the changes in poverty over time. While the years of the surveys vary, they all fall within a span of ®ve years, 1993±98.
Based on the 25th percentile poverty line headcount ratios, our rankings suggest that the country with the lowest poverty incidence is Ghana, followed by Senegal and Comoros. Poverty is highest in Mozambique, with Central African Republic in 1994 and Uganda in 1995 also showing high head-counts. When we use the 40th percentile as the poverty line, the ordering changes some-what, but those at the bottom end of the distribution maintain their position. While Ghana and Senegal remain among the top three in terms of lowest poverty incidence, Cote d^ ÕIvoire moves from its rank of ®fth to a rank of second for the higher poverty line. The biggest shift in the rankings occurs for Mali, which moves from the six to the 10th position, and Zimbabwe, which moves from the 11th to the ®fth position.
The 40th percentile asset index rankings are compared to those based on GDP per capita and GNP per capita using purchasing power parity for years equivalent to those of the DHS survey for each country (Figures 1(a) and (b), respectively).27 Since the GDP and GNP ®gures do not measure poverty, and hence do not take into account distributional consider-ations, we naturally expect to ®nd substantial
dierences across the three sets of measures. Nonetheless, we note that the asset index rankings are broadly consistent with the other two sets of rankings. There are a couple of exceptions, however. While the DHS ranks Uganda 14th and Cameroon 12th based on the 40th percentile cut-o, they are ranked ®fth and third, respectively, by the GDP per capita ®gures. For Uganda, however, the DHS rank-ing is much closer to the PPP rankrank-ing of 11th. Another large divergence is the case of Tanza-nia, which is ranked ninth and 15th, respec-tively, according to the DHS and GDP criteria. Once again, the PPP ®gures are slightly closer to the DHS, ranking Tanzania 14th. And ®nally the number one ranking given to Ghana by the DHS is far better than the number six ranking of the GDP numbers, but once again, much closer to the number two ranking according to PPP GNP. One ®nal country-speci®c result that warrants some comment is that of Zimbabwe. ItÕs poverty ranking is ®fth according to the DHS, while based on GDP and PPP GNP, it ranks as the wealthiest country. This divergence can be partly explained by the large degree of income inequality in Zimbabwe, which next to South Africa, is the highest in Africa and among the highest in the world. Nonetheless, we remain somewhat puzzled by ZimbabweÕs poor performance relative to other African coun-tries.
In Figure 1(c), we also show how the poverty ranking from the DHS compare with the six countries for which Chen, Datt, and Ravallion
Table 24. Asset index poverty by country (weights calculated from pooled samples)a
Poverty line is 25th percentile of pooled distribution of asset indexes
Poverty line is 40th percentile of pooled distribution of asset indexes
Headcount Rank Headcount Rank
Benin (1996) 29.29 12 42.62 11
Cameroon (1997) 20.97 10 46.27 12
C.A.R. (1994) 39.57 14 69.97 13
Comoros (1996) 12.68 3 34.66 6
C^ote d'Ivoire (1994) 16.33 5 25.66 2
Ghana (1993) 8.64 1 17.46 1
Kenya (1998) 18.36 7 39.05 7
Mali (1995) 16.84 6 42.47 10
Mozambique (1997) 51.87 15 73.94 15
Senegal (1997) 10.51 2 26.33 3
Tanzania (1996) 20.13 8 41.02 9
Togo (1998) 15.68 4 29.56 4
Uganda (1995) 38.01 13 70.91 14
Zambia (1996) 20.29 9 40.09 8
Zimbabwe (1994) 21.06 11 33.82 5
(1994) have calculated poverty headcounts based on the US$1 per day poverty line. The results are also quite similar. The only devia-tion of greater than one place is the case of Senegal, where the DHS data suggests that there is less poverty than indicated by the US$1 per day estimates.
Table 25 presents the results of cross-country dominance tests. 28 Recall, that rejecting the
null of nondominance is indeed based on a very demanding criteria, requiring for all points along the cumulative distribution up to a maximum poverty line, that the values in one curve statistically dominate the values in the other. The results indicate that Ghana, for example, dominates all countries except
Comoros and Senegal, while conversely, poverty is worse in Mozambique than in all other countries, with the exception of the Central African Republic. But the other important ®nding that emerges from the dominance results is that while Zimbabwe ranked lower than expected, its asset index distribution is only statistically dominated up to the 50th percentile by those of Comoros, Ghana and Senegal. Furthermore, only in the case of Ghana, do we have ®rst order domi-nance. Similarly, while Mali ranks higher than expected, we reject the null and conclude that MaliÕs asset index distribution is everywhere below only those of Benin, Central African Republic and Mozambique.
Figure 1.Country rankings.
Gha Sen Com Zbwe Zam C.I. Ken Togo Mali Tanz Cam Uga Ben CAR Moz
Ghana (1993) ± ND ND 1 1 1 1 1 1 1 1 1 1 1 1
Senegal (1997) ± ND 3 2 2 2 2 1 2 2 1 1 1 1
Comoros (1996) ± ND 3 3 3 2 1 2 1 1 1 1 1
Zimbabwe (1994) ± ND 3 3 2 3 2 2 1 1 1 1
Zambia (1996) ± 3 3 3 3 2 2 1 1 1 1
Cote d'Ivoire (1994)^ ± ND ND ND ND ND 1 1 1 1
Kenya (1998) ± ND ND ND ND 2 1 1 1
Togo (1998) ± ND ND ND 2 1 1 1
Mali (1995) ± ND ND ND 1 1 1
Tanzania (1996) ± ND ND 1 1 1
Cameroon (1997) ± ND 2 1 1
Uganda (1995) ± 3 1 1
Benin (1996) ± 1 1
C.A.R. (1994) ± ND
Mozambique (1997) ±
aThe numbers indicate the order at which the row distribution dominates the column distribution. ND indicates dominance not achieved up to order 3.
POVERTY
COMPAR
ISONS
