Full text of DOI:10.1111/1467-7679.00098

Atta ining the Inte rna tiona l De ve lop m e nt Ta rg e ts:
Will Growth Be Enoug h?
Luc ia Ha nm e r a nd Fe lix Na sc ho ld *
During the late 1980s and early 1990s improvements in health and education and
progress in poverty reduction ground to a halt, or began to be eroded, in many
developing regions and in the transition countries of Eastern Europe and Central
Asia. The DAC International Development Targets (IDTs) reflect the concern of
development agencies and their developing country partners about these trends
(see Table 1 in the Overview article in this issue). The present article considers
whether these targets are attainable.
The structure of the article is as follows. The following two sections review the
economic literature and examine what data from the 1980s and 1990s can tell us
about the determinants of changes in poverty and human development over this
period. The next two sections outline the methods of the poverty projections and
give estimates of poverty in 2015 under various scenarios, followed by projections
for the human development targets with estimates for infant, under-five and
maternal mortality and primary school enrolment rates. The final section draws
together the key conclusions regarding the attainability of the DAC targets.
Modelling poverty and human development
Earlier attempts to assess the attainability of the DAC targets (Hanmer et al.,
1999; Demery and Walton, 1998) are based on the premise that poverty reduction
and human development improvement can be specified as functions of income
growth alone. However, this is a simplification that could produce misleading
results, as research shows that other variables are also important determinants.
Poverty elasticity and income inequality
The relationship between income growth and poverty reduction is given by the
poverty elasticity, namely the percentage change in the poverty headcount
*Research Fellow and Research Assistant, respectively, Overseas Development Institute,
London.
Development Policy Review Vol. 18 (2000), 11–36
© Overseas Development Institute 2000. Published by Blackwell Publishers, 108 Cowley Road,
Oxford OX4 1JF, UK, and 350 Main Street, Malden, MA 02148, USA.

12
Development Policy Review
incidence that results from a 1% change in per capita GDP.1 It is impossible to
predict in abstract how overall income inequality will affect the poverty elasticity,
as the outcome will depend on how the income distribution varies over time, and the
specific properties of the poverty measure (Ravallion, 1997). One finding to emerge
as robust is that the size of the poverty elasticity varies systematically with income
inequality (Ravallion and Sen, 1996; Ravallion, 1997; and Hanmer et al., 1999).
Ravallion (1997) tests the hypothesis that the poverty elasticity falls as inequality
rises, using a general model of the rate of poverty reduction that can be tested for the
restrictions that leave growth alone, as opposed to a distribution-corrected growth
rate. His results confirm that higher initial levels of inequality are associated with
lower rates of poverty reduction at any given positive rate of growth.
Poverty reduction and the growth path
While growth is important for poverty reduction it has also been recognised for
a long time that the type of growth, that is, the particular processes and sectors
that generate growth, matters. Here we investigate the proposition that broad-
based (labour-intensive) growth reduces poverty more effectively than other types
of growth.
It is frequently argued that the agricultural sector is particularly important as
a source of broad-based growth as it often possesses the characteristics that can
stimulate the sort of growth that reduces poverty rapidly (see Lipton, 1977;
Stewart, 1978; and Ranis, 1979 for early statements of these views). Recent
studies of poverty reduction in India support this view (Datt and Ravallion, 1990,
1996a and b and 1998). One finding is that poverty reduction is the result of
growth within agriculture — growth of agricultural output and small-scale
enterprises and services related to the rural economy — and not due to a shift
of labour and capital out of agriculture into large-scale industry.2 Thorbecke and
Jung (1996) come to similar conclusions for Indonesia. The significance of
agriculture for poverty reduction is also confirmed by results from cross-section
data sets. Timmer (1997) finds that manufacturing reduces poverty directly due
to an increase in the income of employed workers, but it also worsens the
distribution of income, thus reducing the overall benefits to the poor, in contrast
to agricultural growth.3 Bourgignon and Morrisson (1998), using a sample of 38
1. There are two approaches to poverty elasticity estimation – the analytic and the econometric
(see Hanmer et al., 1999 for a discussion of their relative merits).
2. i.e. the effect of resource reallocation from low to high productivity sectors, the sectoral shift
that is associated with Kuznets’ inverted ‘U’ trajectory.
3. The Timmer sample of the Deininger and Squire dataset covers 3.3 billion people in 1995 or
two-thirds of the population in low- and middle-income countries, with agriculture being a
quarter of GDP and employing half the labour force.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
13
small and medium-sized developing countries, find that growth in agriculture
and in basic services reduces poverty more than expanding industrial output.
The conclusion that agricultural growth is best for poverty reduction is
conditional, however, on equitable land distribution. Mellor (1999) argues that
agricultural growth reduces poverty so effectively because, in addition to
generating income for poor farmers, it generates demand for goods and services
that can easily be produced by the poor. However, if land and income distribution
is highly skewed, consumption patterns of landowners are skewed towards
imported or capital-intensive consumer goods rather than the products of labour-
intensive domestic manufacturing and services. De Janvry and Sadoulet’s (1996)
analysis of poverty reduction in 11 countries in Latin America and the Caribbean
confirms this conclusion. They find that due to the highly unequal pattern of land
ownership agricultural growth increases overall income inequality. The inequality
effect is so strong that agricultural growth produced greater poverty in Latin
America and the Caribbean between 1970 and 1994.
Their study has important implications for poverty elasticity estimation and
projections based on estimated future growth rates. They find that the qualitative
and structural features of the growth path4 have strong effects on poverty and
inequality reduction. In fact, once these variables are included in the poverty
reduction model the role of income growth is very small and its total effect is
often perverse, the latter resulting from the fact that late growth (in the 1980s)
was usually accompanied by rising urban poverty and rising inequality, though
it reduced rural poverty. Their findings suggest that simple bi-variate models of
poverty reduction (e.g. Hanmer et al., 1999; Demery and Walton, 1998) may be
mis-specified as they fail to control for variations in the poverty reduction
capacity of different types of growth paths.
To sum up, recent analysis of the determinants of the poverty reduction leads
to the conclusion that using ‘blanket’ elasticities derived from a bi-variate
regression model of per capita GDP growth on poverty to produce future
projections is likely to be highly misleading. Estimators derived from such a
model will be biased, as relevant variables such as labour productivity growth
(real labour income growth), the volume of employment creation and the
sectoral origin of economic growth have been omitted from the model.
Furthermore, it may be important to control for initial levels of income
4. Qualitative features are: predicted growth of the Gini coefficient; length of growth or
recession sequence; difference in growth of value added between agriculture and
manufacturing: difference in growth of value added between agriculture and services;
hyperinflation; real exchange rate growth; coefficient of variation of GDP per capita around its
trend; migration rate; urban minimum wage. Structural features are: initial GDP per capita;
share of agriculture in GDP; initial level of inequality; initial urban/rural poverty; natural
growth rate of urban/rural poverty; initial share of urban/rural in total population (de Janvry and
Sadoulet, 1996).

14
Development Policy Review
inequality when estimating the poverty-reducing effects of growth.
Finally, research has also established the importance of the government’s
policy stance in creating the environment for sustained growth and poverty
reduction (e.g. Burnside and Dollar, 1998; Wade, 1990). Maintaining
competitive exchange rates and an open trade regime, fostering domestic price
stability and controlling balance-of-payments and budget deficits are generally
considered to be important preconditions for economic growth.
Modelling HDIs
Hanmer et al. (1999) established that improvements in human development
indicators are highly correlated with the rate of growth of real per capita GDP
and with technological progress, which is independent of national rates of per
capita GDP growth.5 However, HDIs imperfectly correlated with per capita
income and the explanatory power of models can usually be improved with the
addition of further socio-economic variables (see for example, UNDP, 1996).
But it is important to establish that explanatory variables included in specified
models are robust, that is, their coefficients have a significant effect of
approximately the same magnitude across various model specifications.6 We
therefore base our choice of explanatory variables for the infant and under-five
mortality rates (IMR and U5MR) on Hanmer and White’s (1999) identification
of robust determinants.7 This found that, in addition to per capita GDP, the
coefficients of the following variables are robust across model specifications: the
availability of health services (as measured by an output variable rather than an
expenditure variable);8 immunisation rates; education; gender inequality. In many
developing countries the HIV/AIDS epidemic has had significant effects on
progress in reducing infant and child mortality. In some countries in sub-Saharan
Africa infant and child mortality has begun to increase again after years of steady
5. This study found that between 1970 and 1990 the intercept of the regressions of real GDP
per capita on life expectancy and adult literacy respectively shifted upwards. Hence at the same
level of per capita GDP life expectancy (literacy) was higher in 1990 than in it was in 1970.
6. There are a large number of empirical studies of the determinants of infant and under-five
mortality rates from the perspective of various disciplines including medical science,
demography and economics. The choice of independent variable is for the most part contingent
on the disciplinary approach.
7. We know of no similar study that identifies robust variables for primary school enrolment
rates.
8. Health sector expenditures are not necessarily representative of the quantity and availability
of health services. Money can be inefficiently spent and some health care systems are more
expensive than others and so the cost of the same health services varies widely between
countries. An output rather than an input indicator of health services can better capture the
effect of increasing health expenditures.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
15
improvement (see for example UNICEF, 1999; Hanmer and White, 1999).
Whether the HIV/AIDS virus could mean that the infant and child mortality target
is unattainable is obviously an important question. New data published by
UNAIDS (1998) make it possible to include its effects in our model.
Accounting for changes in the 1980s and 1990s
In order to produce projections of poverty and HDIs in the future we estimate
base-run models using past observations and cross-section data for one or more
points in time for individual countries. The base-run models thus show how, on
average, changes in poverty and human development are related to economic
growth and the other independent variables. The final model presented below
shows the relationship between the dependent variable and a set of independent
variables selected on the basis of the results of model specification tests, and is
then used to make projections for the future value of poverty and HDIs.
Poverty
The World Bank (1999) and Chen et al. (1994) have produced estimates of the
percentage of people living below a poverty line of $1 (1985 PPP) a day for 58
developing countries. We used poverty observations between 1985 and 1995 (a
total of 121 observations) as the dependent variable in the poverty model. We
considered the following independent variables: GDP per capita (GDP); openness
(Sachs Warner dummy) (OPEN); ratio of value added per worker in modern
sector to value added per worker in agriculture (VA); incremental labour:capital
ratio (dL/dK) and incremental capital:output ratio (ICOR), where the latter four
variables are qualitative variables that seek to capture the characteristics of the
growth path. The Sachs Warner dummy gives a broad indication of the
government’s policy stance. It is equal to one if the black market exchange-rate
premium is larger than 20% for at least ten years, and/ or there is a public sector
export monopoly for crops, and/or the country is classified as socialist, and/ or
more than 40% of customs import code lines are affected by some sort of
quantitative restrictions. Otherwise it equals zero. Hence government policy is
‘good’ if the dummy equals zero and ‘bad’ if the dummy equals one. The VA
variable is introduced to capture the sectoral origins of growth. The incremental
labour: capital ratio aims to capture the employment-creating capacity of the
growth path, the argument being that new investment has to keep pace with the
annual increase in the labour force coming onto the market if labour-intensive
growth is to be poverty-reducing. The ICOR is included as a broad proxy for
labour productivity growth, on the basis that the more efficiently capital is used
the more likely it is that labour productivity will grow too.
Table 1 shows that, in addition to economic growth, several of the growth path

16
Development Policy Review
variables and the policy variable are significant determinants of the poverty
headcount, showing bivariate models are indeed misspecified. Our results also
show that the poverty elasticity was systematically related to income inequality,
hence Table 1 presents the regression for results for two groups of countries, one
with low-income inequality (average Gini = 0.34) and one with high-income
inequality (average Gini = 0.55).9
Table 1
Poverty regression model:
Dependent variable: poverty headcounta
Sample
Independent variablesa
R2b
n
Constant OPEN
dL/dK
0.16
ICOR
0.016
GDP
-0.93
Low
Gini
10.77
0.25
0.61
49
(7.31)*
(0.85)
(0.97)
(2.00)* (-2.66)*
High
Gini
8.54
(9.74)*
0.88
(3.12)*
0.35
(2.07)*
-0.34
(-1.73)**
0.6
55
Notes: (a) Dependent variable and independent variables dL/dK and GDP in logs. (b) adjusted.
T statistics in brackets: * t statistic significant at 5% confidence level or above; ** t statistic
significant at 10% confidence level: n is the number of observations.
What is immediately striking is the large difference in the poverty elasticity. In
low inequality countries, 10% economic growth reduces the poverty headcount
by around 9%, but in high inequality countries only by about 3%.10 In other
words, low inequality countries will be more effective in reducing the poverty
headcount than are high inequality countries with a given predicted rate of growth.
The absolute values of the poverty elasticities are lower than those derived from
the bi-variate specification used in Hanmer et al. (1999) which is consistent with
de Janvry and Sadoulet’s findings for Latin America.
Turning now to the effects of the qualitative variables we can identify the
following effects:
• The openness variable is significant and its sign shows that, for a given level
of GDP, poverty is higher if government policies fit the Sachs and Warner
9. Chow tests show that the sample should be divided into two groups containing countries with
Gini coefficients of above and below 0.43 respectively. Model specification tests showed that
VA could be dropped from the model.
10. The systematic difference in poverty elasticities holds at the regional level for sub-Saharan
Africa, Latin America and the Caribbean. Numbers of observations limited testing for
systematic differences in poverty elasticities in other regions.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
definition of ‘bad’ policies.
17
• The incremental labour:capital ratio is negatively correlated11 with the poverty
headcount, implying that the higher is investment relative to the rate of growth
of the labour force, the lower is the poverty headcount.
• The incremental capital:output ratio is positively correlated to the poverty
headcount, meaning that poverty is lower the more efficiently capital is used.
The significance of dL/dK and ICOR is consistent with the view that
investment growth can play a key role in poverty reduction if it contributes to a
growth strategy characterised by growing labour productivity and increasing real
wages. The values of both coefficients of the variables are lowest (most
favourable) for the East Asia and Pacific region where long-run growth has been
associated with these characteristics.
The ratio of modern to agricultural productivity has a significant effect in the
regression models for the sub-Saharan African and South Asian country sub-
samples but not for the total sample. In sub-Saharan Africa and South Asia the
higher rates of productivity per worker in agriculture relative to the modern sector
productivity are associated with lower poverty headcounts. In other words, the
performance of the agricultural sector is critical for poverty reduction in these two
regions.12
The total effect of the qualitative variables on poverty reduction is quite large,
as we show later on. Hence, if the growth path is broad-based and policies are
‘good’, poverty reduction will be greater than is implied by the coefficient on the
GDP term alone. This means that success in reducing poverty depends critically
on the type of growth that occurs; high growth alone does not guarantee rapid
poverty reduction.
Human development
Base-run models were estimated for the HDIs relating to the DAC targets for
infant and child mortality, maternal mortality, primary school education and
gender equality.
Infant and under-five mortality are modelled as functions of GDP per capita,
education, access to health services and HIV/AIDS. All the data (apart from HIV
prevalence) are drawn from the 1998 World Development Indicators CD-ROM.
11. The logarithm of a ratio < 1 is negative and all observations for dL/dK <1.
12. For all developing countries, however, this variable was not significant and small sample
size for each of the regions meant that it was better to use the parameters derived from the
model using all developing countries for projections.

18
Development Policy Review
The mortality rates used are from the 1990s as these are based on Demographic
and Health Surveys and other cited surveys and censuses. The HIV prevalence
data are drawn from the UNAIDS report on the Global HIV/AIDS Epidemic
(1998).
We considered the following independent variables:13 GDP per capita (GDP);
prevalence of HIV/AIDS women attending ante-natal clinics; physicians per
thousand population; births attended by trained personnel; primary school
enrolment ratio. Physician per capita is taken as a proxy for the quantity of health
services available. Births attended by trained personnel can be expected to have
a direct effect on child survival prospects and also indicate access to and
availability of primary health care services.
Table 2
IMR and U5MR regression results
Dependent Variable
IMR
Independent Variables
Adj R2
n
Constant
197.92
(3.99)*
GDP
-20.42
HIV
3.49
PHY
-11.28 0.81
32
32
(-3.62)* (1.64)** (-3.5)*
-37.86 3.26 -26.33 0.73
(-2.87)* (0.65) (-3.59)
U5MR
347.48
(2.87)*
Notes: all independent variables are logged; t statistics in parentheses; * t statistics significant
at 0.05 confidence level or above; ** t statistics significant at 0.10 confidence level; n number
of observations.
Table 2 shows the regression results. We found that HIV/AIDS is positively
correlated and GDP per capita and the quantity and availability of health
services are negatively correlated with infant and child mortality. In other
words, high HIV/AIDS prevalence levels result in higher infant mortality rates,
and higher average income levels and greater availability of health services
increase the prospects for child survival. Model restriction tests showed that
births attended by trained personnel and primary school enrolment could be
dropped from the model with no loss of explanatory power. This challenges the
commonly held belief that education in general, and female education in
particular, has an effect over and above the effects of a higher standard of
13. We could not include immunisation rates and gender inequality in the model as the number
of countries which had a complete set of observations for all the independent variables once
these two are included is less than 20.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
19
living.14 Table 2 thus shows the final models that are used to make projections.15
Maternal mortality Data on maternal mortality rates are scarce. We use data
from the 1990s and specify maternal mortality as a function of per capita GDP,
access to health services, adult literacy and HIV/AIDS. The percentage of births
attended by trained personnel is used as a measure of access to, and the quality
of, health services.16 We use the reported HIV prevalence rate of women
attending antenatal clinics for the HIV/AIDS variable.
Table 3
Maternal mortality regression results
Dependent Variable
Constant
Independent Variables
Birth Att Ln HIV Lit
-0.03 0.15 -0.008
(3.98)* (-2.43)*
Adj R2
n
MatMR
8.03
0.88
2
7
(35.00)* (-8.79)*
Notes: * t statistics significant at 0.05 confidence level or above; ** t statistics significant at
0.10 confidence level; n number of observations.
Table 3 shows that all signs are as expected. HIV/AIDS is strongly and
positively correlated with levels of maternal mortality. Higher adult literacy and
percentages of births attended by skilled health personnel (doctors, nurses and
trained midwives) are associated with lower maternal mortality. The level of
GDP per capita has little influence on maternal mortality once the effects of
HIV/AIDS and access to health services and adult literacy are controlled for,
and model restriction tests show that the variable can be dropped from the
model. In other words maternal mortality rates can be explained by access to
health services (and, presumably health service quality) and the HIV prevalence
rates. This suggests that maternal mortality could be reduced in many countries
in the absence of increases in per capita income if maternal health services were
improved and HIV prevalence reduced.
Primary school enrolment rates. The education target is universal primary
14. The finding that education is insignificant in this model does not mean it is not a policy
relevant variable. Eeducation is highly correlated with per capita GDP and the other
independent variables and so no inferences about its effect can be drawn from the model.
15. The HIV/AIDS data are for the early, not the mid, 1990s. They therefore, represent the
mortality rates towards the beginning of the epidemic, which do not yet fully reflect the impact
of HIV/AIDS on mortality. A 1990 observation of U5MR would include children born in 1985
– before the epidemic had become widespread. We therefore include HIV/AIDS as an
explanatory variable in the U5MR model, despite the fact that it is insignificant in the model.
16. This would probably be a better measure of access to, and the quality of, health services for
the IMR model as well but it was not used as its inclusion restricted the sample size to less than
20 observations.

20
Development Policy Review
education by 2015. The indicator chosen to proxy this is the primary school
enrolment ratio. Two key factors are likely to determine this. First, the family
needs to be able to afford to send its children to school. Even if there are no
school fees to be paid, money is still needed for books, other materials and
uniforms. Secondly, schools have to be accessible to the whole population. We
use GDP per capita and school expenditure per student as explanatory variables
in our model.
Table 4
Primary school enrolment regression results
Dependent Variable
Net primary school
enrolment
Constant
-36.59
(-1.73)**
Ln GDP
14.55
(5.47)*
Adj R2
0.38
n
48
Notes: * t statistics significant at 0.05 confidence level or above; ** t statistics
significant at 0.10 confidence level; n number of observations.
Table 4 shows that primary school enrolment is positively related to per
capita income. However, overall the model’s explanatory power is weak, as
only 38% of the variation in the data is explained by the level of per capita
GDP. School expenditure per student, on the other hand, is not a significant
determinant. However, it is probably a very imperfect measure of the
accessibility of schools to the whole population. Projections are therefore based
only on the forecasts of future GDP per capita and we have much less
confidence in these results than the projections for the other IDTs.
Gender equality in primary (PSER) and secondary (SSER) school
enrolment rates. We use the ratio of female to male primary and secondary
enrolment as the dependent variable and consider the relationship between
gender inequality, on the one hand, and the level of GDP per capita and income
inequality, on the other. Table 5 shows only a weak correlation between them.
Income inequality is positively and significantly correlated with the ratio of
female to male PSER; the higher the Gini the lower the level of gender
inequality in SERs.17 However, the overall explanatory power of the model is
weak (R 2 = 0.20). No significance can therefore be attributed to the effect that
income inequality or per capita income has on gender equality and hence the
model cannot be used to predict whether the gender target will be met, given
projected future growth.
17. The dependent variable is a ratio less than one for almost all observations; hence higher
ginis are correlated with higher values of the ratio, i.e. it increasing to 1 or above. This result
may be driven by East European and Central Asian economies that frequently have high levels
of inequality and higher female than male SERs.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
21
Table 5
Gender equality regression model results
Dep Var
Constant
Ln GDP
Gini
Adj R2
n
Ratio of female to
male PSER
Ratio of female to
male SSER
46.8
3.18
(1.27)
9.26
0.52
(2.86)*
1.00
0.21
29
(2.24)*
-20.35
(-0.41)
0.20
29
(1.7)**
(2.5)*
Notes: * t statistics significant at 0.05 confidence level or above; ** t statistics
significant at 0.10 confidence level; n number of observations.
This result confirms both the country studies’ conclusions (see the article on
Uganda in this issue) and the findings of the gender and development literature
that the role of women, and hence gender equality, depends very much on local
culture and customs. The country studies argue that policies to promote women’s
economic and political empowerment, rather than economic growth, are the
critical determinants of advances in gender equality. Such policies need to address
the overall position of women in society if the incentive system is to be changed
so that parents want to send their girl children to school and that girls want to
remain there and have the opportunity to succeed when they do.
Poverty projections
Country sample
The poverty regression models specified in Table 1 are combined with economic
growth rate forecasts to produce projections of poverty incidence for 2015. The
projections for high and low inequality countries are run separately and then
combined in a population-weighted regional average. The projections are based
on the countries which were used to produce the base-run model which, as Table
6 shows, contain 66%, 88% and 98% of the populations of sub-Saharan Africa,
East Asia and the Pacific and South Asia respectively. The World Bank’s
estimates of $1 a day poverty at 1985 PPPs are shown in the table, with 1990
poverty incidence highest in sub-Saharan Africa and South Asia at 44 and 47%
respectively, and the number of people living in poverty highest in East Asia and
the Pacific and South Asia at 453 and 517 million respectively.

Table 6
Poverty regression model: sample country characteristics, 1990
No. of Population
countries (m.)
% of
Headcount (% of No. of people in
Av. Gini
Av. GDP per
population pop. under $1/day)
in sample
poverty (m.) coefficient capita (US$ PPP)
19
11
8
336
129
207
66
44.1
43.0
44.7
148
56
93
43.7
56.2
35.9
1267
2198
686
Sub-Saharan Africa
High inequality countries
Low inequality countries
Middle East and North
Africa
High inequality countries
Low inequality countries
4
0
4
60
60
24
88
98
2.5
2.5
2
2
39.2
39.2
3228
3228
5
2
3
1449
73
1376
31.2
4.7
32.7
453
3
449
35.9
44.9
35.5
1621
4084
1490
East Asia and Pacific
High inequality countries
Low inequality countries
5
0
5
1108
1108
46.7
46.7
517
517
31.7
31.7
962
962
South Asia
High inequality countries
Low inequality countries

Latin America and
Caribbean
High inequality countries
Low inequality countries
15
12
3
369
350
19
84
27.9
28.2
22.5
103
99
4
55.4
56.1
42.5
4768
4859
3111
Eastern Europe and
Central Asia
High inequality countries
Low inequality countries
10
0
10
110
110
27
78
9.3
9.3
10
10
26.8
26.8
4023
4023
All developing countries
High inequality countries
Low inequality countries
58
25
33
3433
552
2881
35.9
28.6
37.3
1233
158
1075
37.2
54.6
33.9
1817
4134
1373
Notes: all data from World Development Indicators 1999 except Gini coefficients which are from Deininger and Squire (1996) and World Development Report
1998; the percentage of the population covered in our sample cannot be calculated separately for the high and low inequality categories as denominators are
unknown; Eastern Europe and Central Asia data are unreliable (see Mosley and Kalyuzhnova article in this issue).

24
Development Policy Review
Growth rates
The projections in Table 7 use two economic growth scenarios, one based on the
latest forecasts available and the other on the rates of growth achieved between
1965 and 1997.
Table 7
Growth rate assumptions
Annual growth in GDP per
capita
Global Economic
Prospects
Projections 2001–8
Historic Growth
Rates
1965–97
0
Sub-Saharan Africa
Middle East and North Africa
East Asia and Pacific
1.4
1.4
5.6
3.7
3.0
4.8
4.0
0.1
5.4
2.3
1.3
3.2
3.0
South Asia
Latin America and Caribbean
Eastern Europe and Central Asia
All Developing Countries
Note: The Global Economic Prospects (World Bank, 1998) projections are for 2001-8. We
assume they remain the same between 2008 and 2015.
Future scenarios
For the poverty projections we consider four scenarios:
• ‘No change and high growth’ combines World Bank (1998) growth forecasts
with the base-run model, assuming there is no change in the qualitative
features of the growth path up to 2015.18
• ‘No change and low growth’ combines economic growth at historic averages
between 1965 and 1997 (World Bank, 1999) with the base-run model, on the
same assumption.
• ‘Broader-based and high growth’ combines the optimistic growth forecasts
with the base-run model, assuming that policies improve the qualitative
variables having an increased poverty-reducing effect up to 2015.
18. The values of all the independent variables, apart from per capita GDP, are the same in 2015
as they were in 1990.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
25
• ‘Broader-based and low growth’ combines economic growth at historic
averages between 1965 and 1997 with the base-run model, on the same
assumption.
The broader-based growth scenario makes the following assumptions: Sachs
Warner openness dummy (OPEN): all regions become open by 2015; incremental
capital output ratio (ICOR): sub-Saharan Africa, South Asia, and Latin America
and Caribbean reach the East Asian and Pacific 1990 ICOR by 2015; East Asia
Pacific converges to the lowest 1990 ICOR in the region by 2015; and the
incremental labour capital ratio (dL/dK) improves by 25% by 2015.
These broader-based scenario assumptions have the effect of increasing the
absolute value of the poverty elasticity to 1.5% for low inequality and 1.45% for
high inequality developing countries.
Meeting the income poverty target
Table 8 presents our forecasts of the poverty headcount in 2015 for developing
regions overall. It shows that there is a good chance that poverty will be half its
present level by 2015, as long as policies are in place to induce a broader-based
growth path.19
Table 8
Poverty incidence (%) in 1990 and 2015 in developing countries (for a
range of $1 a day at 1985 PPP estimates)
Scenario
1990
2015
High Growth Low Growth
Poverty incidence
No change
Broader based
-
18
13
36
36
20
14
Poverty incidence as % of 1990 poverty
No change
Broader based
56
40
50
35
Note: Bold numbers indicate the target has been attained.
Table 9 shows how income inequality affects the attainability of the income-
poverty target. High inequality developing countries had lower levels of poverty
in 1990 than low inequality countries. However, they are less likely to attain the
19. Sensitivity analysis to changes in the poverty line (to simulate measurement errors) showed
the results to be robust.

26
Development Policy Review
poverty target than low inequality countries. Good policies and broader-based
growth improve poverty reduction prospects in high and low-income inequality
countries alike. However, in high-income inequality countries it is not certain
that even broader-based high growth will be enough to meet the target, as
poverty incidence only falls to 49% of its 1990 level by 2015.
Table 9
Poverty in 1990 and 2015 – the effect of income inequality under no
change and broader growth paths in the high growth scenario
Scenario
1990
2015
Broader based
No Change
Poverty incidence
High inequality countries
Low inequality countries
Poverty incidence as % of 1990
poverty
29
38
19
18
14
13
High inequality countries
Low inequality countries
68
47
49
33
Note: Bold numbers indicate the target has been attained.
Table 10 shows whether the poverty incidence target is attained at the
regional level. We present the upper and lower bounds of the projection results:
i.e., the worst-case scenario assumes low growth and no policy change, and the
best-case scenario assumes high and broader-based growth.
Low inequality East Asia and Pacific meets the target in all scenarios.20
Eastern Europe and South Asia are likely to meet it in the best-case scenario.
Reaching the target in Latin America and the Caribbean is possible but less
certain even in the best-case scenario. Prospects are worst for sub-Saharan
Africa; in the worst case the poverty in Africa in 2015 is almost as widespread
as in 1990. Overall, the attainment of the DAC income-poverty target is
strongly influenced by the performance of China and India. Their large
population size drives the result of the low-income inequality sample, and that
of the overall developing countries.
20. The high inequality sample for East Asia and Pacific (consisting only of Thailand and
Malaysia) is too small to use for projections.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
27
Table 10
Poverty incidence in 2015 as a % of poverty incidence in 1990
1990
2015 headcount as % of 1990
headcount
High broader
No change, low
based growth
growth
Best Case
Worst Case
44
43
44
56
57
56
95
87
99
Sub-Saharan Africa
High inequality
Low inequality
Middle East and North
Africa
3
51
75
High inequality
Low inequality
East Asia & Pacific
High inequality
Low inequality
South Asia
3
31
5
33
47
51
27
75
38
27
35
38
64
High inequality
Low inequality
Latin America and
Caribbean
High inequality
Low inequality
Eastern Europe and
Central Asia
47
64
35
28
28
23
68
68
81
44
44
44
9
55
32
High inequality
Low inequality
Developing Countries
High inequality
Low inequality
9
55
56
73
53
32
33
49
33
36
29
38
Notes: Bold numbers indicate the target has been attained. The samples for the Middle
East and North Africa, South Asia and Eastern Europe and Central Asia do not contain
any high inequality countries. For East Asia and Pacific see Note 20.
Growth rates required to halve poverty by 2015
Forecast growth is sufficient to halve poverty by 2015 for developing countries
as a whole as long as the growth path is broad-based. Without change in the
growth path, per capita growth of 7.1% in high inequality and 3.7% in low
inequality countries is needed to halve poverty by 2015. For Latin America and
the Caribbean the difference in the poverty-reducing capacity of the two

28
Development Policy Review
scenarios is particularly marked, especially for high inequality countries as
Table 11 shows.
Table 11 also shows that growth of 2.4% under the broad-based growth
scenario and 5.9% under the no-change scenario is needed to halve poverty in
sub-Saharan Africa. The latter rate is well above the forecast rate and far in
excess of growth rates achieved in the region since the 1960s.
Table 11
Forecast growth and growth required to halve poverty by 2015 (average
annual real growth in GDP per capita)
Forecast
growtha
Growth required to halve
poverty by 2015
Broader based
No change
1.4
3.0
2.4
3.5
2.1
5.9
10.4
4.6
Sub-Saharan Africa
High inequality
Low inequality
Latin America and
Caribbean
High inequality
Low inequality
0.6
0.5
2.1
7.0
7.0
4.5
(a) Global Economic Prospects (World Bank, 1998) projections are for 2001–8
Thus, without policies in place that alter the growth path to favour the poor,
the growth rates required to halve poverty are high enough to be considered
unfeasible in these regions. Between 1960 and 1990 the high performing Asian
economies grew at about 5.5% per annum — less than the minimum growth
rates of 6 and 7% required to halve poverty in sub-Saharan Africa and Latin
America and the Caribbean respectively, under the no-change scenario. The
equality of income distribution also has a significant impact on the growth rate
required to halve poverty. Under both scenarios high inequality developing
countries need per capita income growth which is roughly twice as high as that
of low inequality countries. Hence policies that reduce inequality in high
inequality developing countries and stabilise it in low inequality ones are
another route to ensuring that the DAC targets are achieved.
Attaining the human development targets
Meeting the infant and under-five mortality rate targets

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
29
The World Bank (1998) growth rate was combined with assumptions about
changes in the other independent variables and changes in the constant term —
the technology effect — to produce three future scenarios.21
• the Better Health scenario, keeping HIV/AIDS infection rates at their present
level and assuming that the availability of health services per capita increases
and technological progress continues to have effects.
• the No Health Gains scenario, keeping HIV/AIDS infection rates at their
present level and assuming that the availability of health services per capita
does not increase, but that technological progress continues to have effects.
• the AIDS Pandemic scenario, projecting HIV/AIDS infection rates spreading
throughout all developing regions at the rate predicted by epidemiological
models currently used in developing countries. All regions have adult
HIV/AIDS prevalence rates of 22.5% by 201522 and the benefits of
technological progress are assumed to be wiped out.
Table 12 shows that, for developing countries as a whole, attaining a
reduction of two-thirds the 1990 level of infant mortality is possible in the better
health scenario. However, without improvements in health services and greater
access to them the target is unlikely to be met. The course of the HIV/AIDS
epidemic is a crucial factor, as IMR remains at 76% of its 1990 level in the
AIDS pandemic scenario. The corresponding rate of 42 per 1000 live births is
more than twice as high as the projection for the better health scenario.
The regional breakdown shows that, even with the AIDS pandemic, infant
mortality is expected to fall in all regions apart from Middle East and North
Africa. East Asia and Pacific is the only region that is likely to reduce it by two-
thirds under any circumstances. Latin America and Eastern Europe need to
control the spread of AIDS to meet the target, while the Middle East and South
Asia, in addition, have to improve health services.
21. Hanmer et al. (1999) found that some improvements in infant and child mortality over time
could be attributed to autonomous effects of improved technology and knowledge. We
incorporate this finding in our model.
22. Epidemiological models (e.g. Stover, 1997) plot the course of the AIDS epidemic as a ‘S’
shaped curve. Infection spreads very slowly in the initial years and then, after prevalence levels
of about 5% are reached, the epidemic spreads rapidly, stabilising after about 15 years at a
maximum value of 22.5%. This may, of course, not actually happen. In Western Europe and
North America the rapid spread of AIDS has to date only affected particular high risk sub-sets
of the population.

30
Development Policy Review
Table 12
Infant mortality rates (IMR) per thousand in 1990 and in 2015 as
% of 1990 levels
IMR in 2015 as % of 1990 level
1990 IMR
(‘000)
93
Better
Health
56
No Health
Gains
65
AIDS
Pandemic
94
Sub-Saharan Africa
Middle East and North
Africa
East Asia and Pacific
South Asia
51
36
83
34
14
44
110
29
79
18
14
34
Latin America and
Caribbean
43
74
12
12
Eastern Europe and
Central Asia
All developing countries
21
56
89
76
24
33
24
40
Note: Bold figures indicate that target (reducing by 2/3, which is 33% of the 1990 level) is
met. The constant figures for East Asia, Latin America and Eastern Europe in the better health
and no health gain scenarios reflect the lower bound of five per thousand imposed by the
projection model.
Table 13
Under-five mortality rates (U5MR) in 1990 and in 2015 as
% of 1990 levels
U5MR as a percentage of its 1990 level
U5MR
1990
149
Better
Health
64
No Health
Gains
77
AIDS
Pandemic
96
Sub-Saharan Africa
Middle East and North
Africa
East Asia and Pacific
South Asia
62
42
105
41
12
46
107
30
79
15
12
33
Latin America and
Caribbean
49
70
10
10
Eastern Europe and
Central Asia
All developing countries
26
72
29
36
19
47
26
77
Note: Bold figures indicate that target (reducing by 2/3, which is 33% of the 1990 level)
is met. The constant figures for East Asia, Latin America and Eastern Europe in the better
health and no health gain scenarios reflect the lower bound of 5 per thousand imposed by
the projection model.

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
31
Table 13 shows that, for developing countries as a whole, under-five
mortality is likely to fall to 36% of its 1990 level (72 per thousand) if better
health policies increase access to health services and contain the spread of
HIV/AIDS. In the AIDS pandemic scenario it decreases slowly compared to
past trends, falling by less than 20 per thousand over the fifteen-year period and
remaining at about three-quarters of its 1990 level. Although developing
countries overall do not reach the target , all regions, apart from sub-Saharan
Africa, do so under the better health scenario.
Meeting the maternal mortality rate target
The estimated rates of growth of adult literacy23 are combined with the
assumptions about access to reproductive health care and the spread of
HIV/AIDS to create three scenarios:
• the Better Health Services scenario, keeping HIV/AIDS infection rates at
their present level and assuming that the 80% of births in 2015 are attended
by skilled health personnel;24
• the Slower Health Gains scenario, keeping HIV/AIDS infection rates at their
present level and assuming that improvements in births attended by skilled
health personnel take place at only half the rate necessary to meet the births
target (and that around 70% of births are attended in 2015);
• the High AIDS scenario, making the same births attended assumption as the
Slower Health Gains scenario, but also assuming that HIV/AIDS infection
rates spread throughout all developing regions at the rate predicted by
epidemiological models. All regions have adult HIV/AIDS prevalence of
22.5% by 2015.
Maternal mortality data are available for very few countries. Table 14
therefore shows the country that has the largest weight in the result for each
region, in the case where the regional country sample is small.
None of the regions (or countries) under any of the scenarios reaches the
target of reducing maternal mortality rates by three-quarters by 2015, which
would imply a value of 25% in 2015. Starting from a very high level sub-
Saharan Africa gets close to reaching the target, but that is contingent on
achieving a level of 80% births attended by skilled personnel and on halting the
23. A model of adult literacy as a function of GDP per capita is used to estimate future adult
literacy rates given the growth rates predicted by the World Bank.
24. A subsidiary IDT is that 80% of births are attended by trained personnel by 2015.

32
Development Policy Review
spread of HIV/AIDS at current rates of infection. In terms of percentage change
the best-case outlook for Africa is better than for other regions, as currently the
ratio of births attended is very low and hence there are bigger potential
improvements to be made. If the health gains take place at only half the rate
necessary to hit the attended births target, sub-Saharan African maternal
mortality in 2015 falls only to 60% of its 1990 level.
Table 14
Maternal mortality rates (MMR) per 100,000 in 1990 and in 2015 as %
of 1990 levels
2015 MMR
as a percentage of 1990 levels
1990 MMR
(per 100,000)
Better
Health
Services
30
Slower
Health
Gains
60
High AIDS
Sub-Saharan Africa
Middle East and
North Africa
Vietnama
768
68
221
1356
263
50
45
60
72
65
75
153
101
96
Pakistanb
Latin America and
Caribbean
98
109
45
72
68
73
86
129
Polandc
All developing
countries
694
44
65
94
Note: Vietnam, Pakistan and Poland represent more than 75% of the population in their
regions for which data are available. (a) plus Mongolia, Cambodia and some Pacific
Islands; (b)plus Afghanistan, Bhutan, Maldives and Nepal; (c) plus Moldova and Georgia.
Comparing the Slower Health Gains and the High AIDS scenarios, Table 14
shows that for Africa the effect of AIDS on maternal mortality is smaller than
the effect due to births being attended. This is because HIV/AIDS infection
rates in the region are already very high, nearing the epidemiological maximum.
The situation is reversed for a region like the Middle East and North Africa,
which currently has low HIV/AIDS infection rates but higher attended births
statistics. Though the relative importance of the births-attended and the AIDS
effects differs by region, it is clear that to improve maternal mortality rates in
developing countries it is crucial to enhance health services and contain the
spread of the AIDS epidemic.
Universal primary education

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
33
Table 15 shows the 1990 levels and 2015 projections of the net primary school
enrolment rate for the high and low growth scenarios by region.
Table 15
Net primary school enrolment rates in 2015, by region
1990
2015
High growth
Low growth
Sub-Saharan Africa
Middle East and North Africa
East Asia & Pacific
54
84
99
62
90
96
82
68
98
100*
83
100*
100*
100*
65
96
100*
79
100*
100*
98
South Asia
Latin America and Caribbean
Eastern Europe and Central Asia
Developing Countries
Note: ‘100*’ indicates full net primary school enrolment.
Only three regions, East Asia and Pacific, Latin America and Caribbean and
Eastern Europe and Central Asia will have universal enrolment by 2015. The
other regions are unlikely to reach the target. The Middle East and North Africa
gets very close, and South Asia is projected to reach around 80% at least.
Forecasts for sub-Saharan Africa suggest that even in 2015 more than three out
of ten children will not attend primary school.
In the efforts to improve education an important caveat must be kept in mind.
Enrolment does not ensure education. Our country case study, Uganda, shows
that quality concerns can be critical. Teacher pupil ratios have soared to 70:1 in
government schools with the Universal Primary Education policy, and the
challenge facing the government is to ensure that all pupils leave primary school
literate and numerate and that the overall standard of education does not fall.
The Tanzania study shows that deteriorating standards of education can lead to
education becoming highly price-elastic. The recent imposition of a charge of
TSh1000 per year (about £1) led to falling enrolments as parents would not pay
‘something for nothing’ (see the article on Tanzania in this issue).
Conclusions
This article set out to examine whether the DAC targets are achievable. It
sought to improve on models that predicted their achievability on the basis of
growth alone. It found that, although economic growth is an important
determinant of both poverty and human development, other independent
variables also play an important role in determining poverty reduction and
human development improvements.

34
Development Policy Review
Halving income poverty by 2015 is possible, so long as policies are in place
to induce broader-based growth paths. And growth does not have to be as high
as forecast if the growth path becomes broader-based; developing countries
overall will meet the target if they achieve the 3% per capita real average growth
rates that they had between 1965 and 1997. Growth is important but attaining
the poverty targets depends crucially on action being taken by developing
country governments and donor agencies to influence the qualitative nature of
the growth path, including policies to ensure that growth is rooted in sectors of
the economy in which the poor are able to participate, in order to create
conditions for rising labour productivity by increasing the rate and efficient use
of investment and policies to improve openness. Our results also point to the
critical role of income inequality. The prospects for reducing poverty are much
better in low-income inequality countries than high inequality countries. The
latter are unlikely to attain the target even if growth is rapid and broad-based;
they need growth rates about twice as high as low inequality countries to meet
the income-poverty target.
Despite the good prospects for reducing poverty in developing countries
overall there are some sobering findings. Sub-Saharan Africa is unlikely to
reach the target, given the forecast rates of growth. If the growth path were
broader-based, then growth of about 2.5 per capita per annum would be needed.
Other regions have achieved sustained growth of this magnitude, but it will take
a major effort by the international community and African governments alike
to launch sub-Saharan Africa on such a growth path. Contrary to widespread
assumptions, it contains a large number of high inequality countries, and there
is therefore scope for increasing the rate of poverty reduction for a given growth
rate if policies that induce high growth and falling income inequality are
devised. Further research is needed to establish exactly what these policies
should be but, if the experience of Latin America is anything to go by, ensuring
that land tenure reforms have the net effect of increasing equality — including
gender equality — seems likely to be a critical ingredient in many African
countries (see Killick et al., 1999).
There are good prospects for meeting the human development targets too, as
long as policies are in place that support the gains that higher income levels can
bring. Reducing infant and under-five mortality rates by two-thirds will depend
on policy interventions that can: halt the spread of HIV/AIDS; increase the
capacity of health sectors to deliver more health services; and ensure that
technological progress spills over to benefit the developing world. Making
maximum progress towards the maternal mortality rate target requires
investment in health services that increase women’s access to maternal health
services and halting the spread of HIV/AIDS. In developing regions as a whole
primary school enrolment can become universal. However, South Asia and
particularly sub-Saharan Africa are unlikely to reach the enrolment target.
Moreover, attaining universal primary education is conditional on maintaining

Hanmer and Naschold, Attaining the IDTs: Will Growth Be Enough?
35
the quality of teaching. Finally, we find no evidence that gender equality in
enrolment in primary and secondary education increases automatically as
development proceeds. This suggests that the achievement of this target depends
critically on pro-active policy measures that promote the cultural, political and
economic empowerment of women.
References
Bourguignon, F. and Morrisson, C. (1998) ‘Inequality and Development: the
Role of Dualism’, Journal of Development Economics 57:233–57.
Burnside, C. and Dollar, D. (1998) Aid, the Incentive Regime and Poverty
Reduction. Policy Research Department, Working Paper No.1937.
Washington, DC: World Bank.
Chen, S., Datt, G. and Ravallion, M. (1994) ‘Is Poverty Increasing in the
Developing World?’, Review of Income and Wealth 40 (4): 359–76.
Datt, G. and Ravallion, M. (1998) ‘Farm Productivity and Rural Poverty in
India’, Journal of Development Studies 34(4).
Datt, G. and Ravallion, M. (1996a) Why Have Some Indian States Done Better
than Others at Reducing Rural Poverty? Washington, DC: World Bank.
Datt, G. and Ravallion, M. (1996b) ‘How Important to India’s Poor is the
Sectoral Composition of Economic Growth?’, The World Bank Economic
Review 10(1).
Datt, G. and Ravallion, M. (1994) How Important to India’s Poor is the Urban-
Rural Composition of Growth? Washington, DC: World Bank, Policy
Research Department.
Datt, G. and Ravallion, M. (1990) Regional Disparities, Targeting and Poverty
in India. Washington, DC: World Bank.
Deininger, K. and Squire, L. (1996) ‘A New Data Set Measuring Income
Inequality’, The World Bank Economic Review, 10(3): 565–91.
Deininger, K. and Squire, L. A new data set measuring income inequality,
http://www.worldbank.org/html/prdmg/grthweb/ddeisqu.htm
De Janvry, A. and Sadoulet, E. (1996) Growth, Inequality and Poverty in Latin
America: A Causal Analysis, 1970-94. Working Paper No. 784. Department
of Agricultural and Resource Economics, University of California at
Berkeley.
Demery, L. and Walton, M. (1998) ‘Are Poverty and Social Goals for the 21st
Century Attainable?’. Paper presented at seminar on ‘What can be done about
poverty?’. Institute of Development Studies, Sussex University, Brighton.
Hanmer, L., de Jong, N., Kurian, R. and Mooij, J. (1999) ‘Are the DAC Targets
Achievable? Poverty and Human Development in the Year 2015’, Journal of
International Development 11:547–63.
Hanmer, L. and White, H. (1999) ‘The Impact of HIV/AIDS on Under-Five
Mortality in Zambia and Zimbabwe’ in Human Development in Sub-Saharan

36
Development Policy Review
Africa: The Determinants of Under-Five Mortality. The Hague: ISSAS,
January.
Killick, T., Kayizzi-Mugerwa, S., Savane, M.-A. and White, H. (1999) ‘Africa
Status Report 1999’. Report prepared for the SPA, August (mimeo).
Lipton, M. (1977) Why Poor People Stay Poor: Urban Bias in World
Development. London: Temple Smith.
Mellor, J. (1999) ‘Poverty Reduction – Sequences and Priorities’. Abt
Associates Inc., Bethesda, MD, 29 August (mimeo).
Ranis, G. (1979) ‘Appropriate Technology in the Dual Economy: Reflections
on Philippine and Taiwanese Experience’ in A. Robinson (ed.) Appropriate
Technologies for Third World Development. New York: Macmillan.
Ravallion, M. (1997) Can High-Inequality Developing Countries Escape
Absolute Poverty? Washington, DC: World Bank.
Ravallion,M., and Sen, B. (1996) ‘When Method Matters: Monitoring Poverty
in Bangladesh’, Economic Development and Cultural Change 44(4): 761–92.
Sachs, J. and Warner, A.M. (1995) Economic Convergence and Economic
Policies, National Bureau of Economic Research, Inc. Working Paper Series
5039. New York: NBER.
Stewart, F. (1978) Technology and Underdevelopment. London: Macmillan.
Stover, J. (1997) AIDSproj Spreadsheet Model to Project HIV, AIDS and TB.
Glastonbury: The Futures Group International, The POLICY Project.
Thorbecke, E, and Jung, H.S. (1996) ‘A Multiplier Decomposition Method to
Analyse Poverty Alleviation’, Journal of Development Economics 48 (2):
279–300.
Timmer, C.P. (1997) How Well Do the Poor Connect to the Growth Process?
CAER Discussion Paper No. 178. Cambridge, MA.: Harvard Institute for
International Development.
UNAIDS (1998) ‘Report on the Global HIV/AIDS Epidemic June 1998’,
http://www.who.int/emc-hiv/global_report/.
UNDP (various years) Human Development Report. New York: UN.
UNICEF Eastern and Southern Africa Regional Office (1999) ‘Progress in the
Social Sectors in Eastern and Southern Africa in the 1990’s: A UNICEF
Perspective’ (mimeo).
UNICEF (1998) State of the World’s Children 1998. New York: UNICEF.
Wade, R. (1990) Governing the Market: Economic Theory and the Role of
Government in East Asian Industrialization. Princeton, NJ: Princeton
University Press.
World Bank (1998) Global Economic Prospects for Developing Countries
1998/99: Beyond Financial Crisis. Washington, DC: World Bank, December.
World Bank (1999) World Development Indicators. Washington, DC: World
Bank.