Review of Income and Wealth

Series 46, Number 3, September 2000

GROWTH, POVERTY, AND INEQUALITY IN LATIN AMERICA:

A CAUSAL ANALYSIS, 1970-94

University of Califormia at Berkeley

Taking advantage of consistent poverty and income inequality data for 12 Latin American countries
between 1970 and 1994, we analyze the determinants of changes in the incidence of urban and rural
poverty and in Gini coefficients over spells of years, stressing in particular the role of aggregate income
growth. We find that income growth reduces urban and rural poverty but not inequality. We also
find that income growth is more effective in reducing urban poverty if the levels of inequality and
poverty are lower, and the levels of secondary education higher. We show that there is an asymmetry
in the impact of growth on poverty and inequality, with recession having strong negative effects on
both poverty and inequality. Since growth does not reduce inequality, economic cycles create ratchet
effects on the level of inequality. However, post-structural adjustment growth is quite effective at
reducing poverty, particularly if inequality is low.

The Latin American region has exceptionally high levels of inequality and
an "excess" incidence of poverty compared to other regions at similar levels of
average per capita income (IDB, 1998). The high costs that poverty and inequality
entail have been brought to public attention by the derailing of economic recovery
in Mexico in part as a consequence of social exclusion, and by the threats of
social backlash to the structural adjustment reforms in many other Latin Ameri-
can countries as the distribution of the benefits from growth is perceived to be
excessively unfair (The Economist, 1996; Berry, 1997). Public sensitivity is exacer-
bated by the fact that the serious losses in purchasing power during the years of
crisis and adjustment have increased impatience in sharing the benefits of recov-
ery, leaving little room for further postponement. Successful transitions to democ-
racy or improvements in democratic representation, and decentralization of
governance throughout the continent, have given the poor new channels of access
to the political process, making their demands for participation to the gains from
growth more difficult to ignore. The strength of these demands has been signifi-
cantly enhanced by proliferation of grassroots organizations and social move-
ments which act as advocates for the poor. Finally, lessons derived from the
Asian experience (Stiglitz, 1996) and from empirical studies of endogenous growth
(Benabou, 1996; Aghion and Howitt, 1998) have made it increasingly well-known
that income inequality can have a negative influence on income growth, opening
to question the economic wisdom of continued high levels of inequality in Latin
America.
Note: We are indebted to many who gave us critical comments as this paper was being developed,
particularly Nora Lustig, Miguel SzCkely, Martin Ravallion, Patrick and Sylviane Guillaumont,
Franqois Bourguignon, Samuel Morley, and Douglas Marcouiller.
 There is hence widespread recognition that the existing levels of poverty and
inequality in Latin America need to be reduced (IDB, 1998). With countries
emerging from severe recessions and resuming growth in the context of liberalized
economies, the question as to how effective will aggregate income growth be in
reducing poverty and inequality is a central issue for policy making: can income
growth be relied upon to significantly reduce poverty and inequality, or is it a
weak force that needs to be complemented by other policy interventions? This is
the issue that we analyze in this paper.
Due to lack of comparable data over time and across nations, there have been
few systematic causal analyses of the roles of economic growth and other variables
in explaining poverty and inequality in Latin America. However, recent efforts at
generating data made by Altimir (1995) and the Economic Commission for Latin
America and the Caribbean (ECLAC, 1996) allow such an analysis to be under-
taken. While data quality problems remain (Lustig, 1994), this information has
started to yield results on identifying the determinants of poverty and inequality.
For instance, inspection of these data by a number of analysts of Latin American
incomes such as Altimir (1995), Beccaria et al. (1992), Fields (1992), Lustig (1995),
Morley (1995), and Psacharopoulos et al. (1995) has led to the general conclusion
that poverty and inequality have been closely linked to the economic cycle, rising
during periods of recession and falling during recovery. Nevertheless, whether
future growth will be able to sufficiently reduce poverty and inequality to accom-
modate popular demands so they do not create threats to the sustainability of
recoveries remains a matter of debate. Opinions range from optimism (Morley,
1995), to calls on the need to complement the impact of growth with attention
to education and employn~entcreation (Psacharopoulos et al., 1995; Lustig and
Deutsch, 1998), and to advocacy of extensive interventions to reduce inequality
and target poverty (Beccaria et al., 1992; Fujii and Aguilar, 1995).
Quantitative analyses of the relation between growth, inequality, and poverty
and of the role of other causal factors remain insufficient for five reasons. First,
there have been few econometric analyses of the sources of change in poverty and
inequality. Most studies do two-way tabular classifications of changes in GDP per
capita (positive and negative) and changes in poverty or inequality across countries
(positive and negative), looking for most frequent correspondence as evidence of
a relationship. Recent exceptions are the work of Ravallion and Datt (1996),
Ravallion (1997), and Ravallion and Chen (1997) who proceed with econometric
analyses, though not for Latin America. Closest to what we do in this paper, the
IDB (1997) used regression analysis to explain the change in inequality between
1985 and 1995 in 13 Latin American countries, showing that the change in GDP per
capita (GDPpc) during the period and the initial level of education both helped
reduce inequality, while the standard deviation of years of schooling increased
inequality. Second, analysis of the role of aggregate income has in general been done
by observing changes in poverty and inequality during globally defined historical
periods broadly associated with growth and recession. Thus, data for the 1970s have
been associated with growth, for the early and mid-1980s with recession, and for
the late 1980s and the 1990s with growth recovery (see for example Londoiio and
Szekely, 1997). This is, however, a gross approximation since countries had highly
idiosyncratic phases of growth and recession, with some countries like Colombia
268
and Costa Rica largely avoiding recession, countries like Peru and Brazil still in
recession in the late 1980s and even the early 1990s, and countries like Honduras
and Venezuela relapsing into recession in the 1990s after temporary recoveries.
To account for this, we conduct instead a detailed analysis of spells of years where
episodes of growth and recession are specific to each country.' Third, if care is
not taken to separate periods of growth and recession, the overall negative
relation that has been established by analysts between changes in income and
changes in poverty or inequality may derive from growth, from recession, or from
both. If the strongest relation happens to occur during recession, the optimistic
predictions about the role of income growth in reducing poverty and/or
inequality will be dangerously fallacious. To avoid this error, we allow for separ-
ate effects of aggregate income growth on poverty and inequality during episodes
of growth and recession. Fourth, there are important qualitative differences in
the policy context of growth before and after the structural adjustment reforms
that may alter the capacity of growth to influence poverty and inequality. Before
the reforms, most Latin American economies were still implementing import sub-
stitution industrialization (ISI) strategies and accumulating debt, thus discrimi-
nating against agriculture and favoring growth in capital intensive industries.
With the reforms, the general prescriptions of the "Washington Consensus" (Wil-
liamson, 1990) were implemented across the region, with strict fiscal discipline,
restrictive monetary policies, competitive exchange rates, and trade liberalization,
potentially redefining the distribution of benefits from growth. To capture this
potential difference, we separate the effects of growth before and after the
reforms. And fifth, the particular structural context within which growth occurs
affects the ability of growth to influence poverty and inequality, calling on the
need to look at interactive effects between growth and context. Most important
among contextual features that affect the role of growth are the initial levels of
inequality (Ravallion, 1997), poverty, and secondary education (Psacharopoulos
et al., 1995; IDB, 1997).
In this paper, we start in Section 2 by reviewing recent analyses of the deter-
minants of poverty and inequality to motivate the specification of causal relations
for econometric analysis. We proceed in Section 3 to present a data base where
the unit of observation is change over spells of years between data points that
give measurements of urban and/or rural poverty and of inequality. In Section
4, we present econometric estimates of the causal determinants of urban and rural
poverty and of inequality across growth spells. Finally, in Section 5, we extract
policy implications regarding the role of growth and the identification of
additional instruments to reduce poverty and inequality.

The determinants of poverty and inequality that have been identified in pre-
vious studies can be classified in the following four categories.

'For clarification, the terminology we use in thls paper is as follows: Spell of years: sequence of
years between two consecutive data points. Episode: we distinguish between three growth episodes-
early growth (spells with positive Gross National Income per capita growth (GNIpc) before the struc-
tural adjustment reforms), recession (spells with negative GNIpc), and late growth (spells with positive
GNIpc after the reforms). The symbol x represents the growth rate of X.
2.1. Role of Per Capita Aggregate Income Growth
There is general agreement in the profession that aggregate income growth
is necessary to reduce poverty (World Bank, 1990). Using national-level data for
the Latin American countries in the 1980s, Morley (1995) and Psacharopoulos et
al. (1995) give evidence that poverty has mirrored the economic cycle, rising dur-
ing recession and falling during recovery. Using a compilation of 682 measure-
ments of the income of the poorest quintile in 108 countries, Deininger and Squire
(1996) show that, in periods of aggregate growth, the income of the poor
increased in 88 percent of the cases. Using a subset of 43 countries from the same
data base, Birdsall and Londoiio (1997) find that that aggregate growth increases
the income growth of the poor with an elasticity of 1.3. From state-level data,
Ravallion and Datt (1996) find that aggregate income growth reduced poverty in
India. And using household data from pairs of surveys in 42 countries, Ravallion
and Chen (1997) show that aggregate income growth reduced poverty. Hence,
there is little disagreement on the proposition that aggregate income growth helps
reduce poverty.
While agreement on the role of growth for poverty reduction is widely
shared, this is not the case for inequality. Psacharopoulos et al. (1995) show that
inequality in Latin America was, like poverty, anti-cyclical with growth. Other
studies are not so affirmative. For instance, Ravallion and Chen (1997) in their
42 countries study find no evidence that aggregate income growth helped reduce
inequality.
All changes in aggregate income may not be equally effective in affecting
poverty and inequality. First, there may exist an asymmetry in the relationship
between changes in poverty and inequality and changes in income associated with
growth and with recession. For instance, a 1 percent increase in Gross National
Income per capita (GNIpc) may have less effect on poverty or inequality than a
1 percent fall in GNIpc. Whether this difference exists or not has, however, not
been pursued in past empirical analyses. As examples, the Psacharopoulos et al.
(1995) and the Ravallion and Chen (1997) studies work with scatters of points
that relate changes in poverty and inequality to changes in income. The negative
relation between poverty/inequality and per capita income can come from growth
as well as from recession, and no attempt is made to see if the strength of this
relation differs as income rises or falls.
Second, the effect of income growth on poverty and inequality may also
differ by policy context. While most previous analyses of the Latin American
experience have focused on the 1980s, the effect of GNIpc growth on poverty
and inequality can be contrasted between the IS1 context that prevailed before
the structural adjustment reforms and the open economy context that was intro-
duced by the reforms. The 1990 World Development Report (World Bank, 1990)
stressed the importance of the labor intensity of growth in reducing poverty. With
transition from IS1 to more open economy industrialization, the labor intensity
of growths should have increased, and hence also the income elasticity of poverty.
2.2. Role of the Quality of Aggregate Income Growth
There are three other qualitative features of growth that have been associated
with the poverty reduction power of a given quota of aggregate income growth.
270
 (i) Instability of growth. This is measured by the coefficient of variation of
GNIpc around its trend. Datt and Ravallion (1998) show that fluctuations in
crop yields and in per capita non-farm output are detrimental for poverty
reduction after controlling for income growth. Since we are working with spells
of years which we characterize as episodes of growth or recession, the coefficient
of variation of GNIpc is important in capturing income instability within the
spell. Indeed, a spell may end up having positive or negative overall income
growth while hiding considerable intra-spell instability. The coefficient of vari-
ation will thus help qualify how steady growth was for a given intra-spell trend.
(ii) Sectoral composition of growth. Ravallion and Datt (1996) analyze the
relation between change in poverty and change in sectoral value added in the
primary, secondary, and tertiary sectors. Using data for India, they find that
growth in the primary and tertiary sectors has been effective in reducing both
rural and urban poverty, but that secondary sector growth has not been effective
due to the high capital intensity of industry. This stresses the roles that differential
growth of the agricultural and services sectors can play in reducing poverty.

2.3. Role of Macroeconomic Performance

(i) Inflation and hyperinflation. Inflation has been associated with rising
poverty by Datt and Ravallion (1998) and systematically mentioned in reviews of
the determinants of poverty in Latin America, particularly when reaching hyper-
inflation levels (Lustig and Deutsch, 1998).
(ii) Real exchange rate depreciation. Adjustment has usually been associated
with sharp depreciations of the real exchange rate. The poverty effects of
depreciation have been observed to vary across social sectors. With depreciation,
real wages in the formal sector tend to fall, increasing urban poverty, but rural
poverty may decline if small farmers produce tradables and there is passing
through from border to farm prices (Bourguignon, de Melo, and Morrisson,
1991). Inequality can also be reduced if the real incomes of the rural poor and of
the urban poor in the informal sector fare better than those of the modern sector
workers (Sahn, Dorosh, and Younger, 1996).

2.4. Role of the Structural Context and Initial Conditions

(i) Structural variables. There are a number of structural variables that have
been identified as affecting poverty. One is the share of agriculture in GDP (posi-
tive on urban poverty and negative on rural poverty), after controlling for the
level of GNIpc and the share of population in the rural sector. Demographic
features also matter, particularly the rate of population growth and the distri-
bution of population between rural and urban sectors (for the impact of the latter
on aggregate inequality, see Fields, 1980).
(ii) Education. Psacharopoulos et al. (1995) place much emphasis on how
the spread of education reduces both inequality and the incidence of poverty.
Higher school enrollment ratios equalize the distribution of education and sub-
sequently reduce income disparities. They also help reduce poverty directly
by increasing the productivity of the poor, improving their chances of access to
better-paid employment, and allowing vertical mobility for the children of the
poor. The role of secondary education in reducing inequality is observed by Bour-
guignon and Morrisson (1998). In general, the role of education in reducing pov-
erty and inequality is both direct (as above) and indirect in giving the poor greater
ability of benefiting from aggregate income growth, an interaction that we will
explore.
(iii) Initial level of GNIpc. Convergence theories tell us that lower initial
levels of GNIpc should be associated with higher growth rates, and hence poten-
tially with more decline in poverty (Sala-I-Martin, 1996). After controlling for
GNIpc growth, however, the relation is ambiguous. For instance, higher GNIpc
countries have better performing labor markets, allowing greater trickle down of
the benefits from growth and hence higher income elasticities of poverty.
(iv) Initial level of inequality. Here again, there are direct effects of the initial
level of inequality on changes in poverty and inequality and also indirect effects
through the qualification of income growth. Direct effects on inequality can be
either a pure numerical phenomenon (a given change in inequality AG, where G
is the Gini coefficient, gives a lower percentage change in inequality AG/Go when
initial inequality Go is high) or represent greater opportunities for change when
initial inequality is low, implying in both cases a negative sign between initial
inequality and growth rate in inequality. With inequality very high in Latin
America and differing substantially across countries (the Gini ranged from 0.62
in Brazil to 0.30 in Uruguay in 1993-94), the role of inequality in indirectly
reducing the benefits of growth for poverty alleviation is a significant issue. Raval-
lion (1997) analyzes 41 spells in 23 countries to show that a higher initial
inequality lowers the income elasticity of poverty, thus also sheltering the poor
during recessions. He estimates a relation P = P(1 - G ~ ) G D P ~where C , P is the
headcount ratio, Go the initial inequality index, and P(1 -Go) the income elas-
ticity of poverty, showing that inequality dampens the poverty reduction effects
of GDPpc growth. With higher inequality, the poor receive a lower share not
only of total income but also of its increments through growth. Psacharopoulos
et al. (1995) similarly find that high inequality reduces the poverty reduction effect
of aggregate income growth. And, more specifically, Birdsall and Londofio (1997)
show that it is land and educational inequality that contribute to lowering the
income growth of the poor
(v) Initial level of poverty. While this has not been analyzed quantitatively,
higher initial levels of poverty could reduce the income elasticity of poverty either
through numerical effects (as above for G) or through higher inequalizing effects
of growth.
In this paper, we analyze separately rural and urban poverty because they
have different determinants, and hence suggest differential policy interventions
for poverty reduction. It should be clear, however, that the overall incidence of
poverty could be falling with no change in either the rural or urban incidence of
poverty due to sectoral population shifts. Conversely, the relation that holds at
the aggregate level between growth, inequality, and poverty does not carry at the
sectoral poverty levels. In the results we present here, it is consequently possible
for sectoral poverty not to fall when GNIpc is rising and income distribution has
remained unchanged.
272
 The data characterize 48 spells in twelve Latin American countries for which
we have consistent information on changes in poverty and inequality for the
period 1970-94 as well as the set of exogenous variables needed for causal analy-
sis (several data points on poverty cannot be used due to lack of information on
the initial level of inequality for the ~ p e l l ) It
. ~ is evident that there is no perfect
data set and that considerable disagreements remain among users of Latin Ameri-
can income statistics. We tried to minimize discrepancies across data points by
remaining true to the data set on poverty assembled by Altimir and published by
the Economic Commission for Latin America and the Caribbean in Social Panor-
ama for Latin America (ECLAC, 1996). The analysis presented in this paper will
clearly need to be repeated and the results obtained confirmed with data sets
constructed by other analysts of Latin American income^.^
To identify (1) potential asymmetries in the relation between poverty/
inequality and income during episodes of growth and recession; and (2) the con-
trasted effects of growth on poverty and inequality under alternative development
strategies (IS1 before the structural adjustment reforms and open economy
growth afterwards), the spells are classified into three growth episodes as follows:
-Early growth: spells with positive GNIPC before the reforms.
-Recession: spells with negative GNIPC.
-Late growth: spells with positive GNIPC originating after the reforms.
For each country, the year of the reforms was identified from macroeconomic
indicators, looking in particular for a major upsurge in the real exchange rate.
These years are given in Table 1. The spells, the corresponding GNIpc and their
classification in three episodes are also given in Table 1.
The timing of the surveys from which observations are derived was not the
same across countries. Most countries have an observation in 1970. Most of the
subsequent observations are clustered around 1979-81, 1986, 1990, and 1994. A
majority of countries had growth during the 1970 to 1979-91 period, recession
between 1979-81 and 1986, and then different degrees of recovery or relapse in
the 1990s for which we have shorter spells. Thus, in general, the spells characterize
fairly well phases of growth or recession. However, this is only an approximation.
For this reason, we characterize each spell not only by the overall G N I ~ C during
the period, but also by the coefficient of variation of GNIpc around its trend. An
exact breakdown into spells of years of uninterrupted growth or recession would
require annual data, which we do not have.
The observed spells fall quite naturally before and after the reforms. The
only exception for which there is ambiguity is Colombia in 1980-86 since the
reforms occurred sometime toward the middle of the spell. For this reason, we do
sensitivity analysis by classifying this spell alternatively as early and late growth.
The indicators of poverty we use are the growth rates in the headcount ratios
for the rural (Pa and urban (Pi)populations, and the indicator of inequality is
' ~ e t a i l e dinformation on data sources and data adjustments, when needed, are available on the
authors' webpages at http://are.Berkeley.EDU/-alain/ or http://are.Berkeley.EDU/-sadoulet/.
3 ~ o n d o i i oand Sztkely (1997) for instance construct annual time series on poverty for the
Latin American countries starting from the "good quality" income inequality measures reported by
Deininger and Squire (1996). This approach is not free from strong assumptions.
 TABLE 1
GROWTHSPELLSAND GNI PER CAPITAANNUALGROWTHRATES, LATINAMERICA,1970-94
(Average annual growth rates in percent)

Growth Episode
Year of Early Growth Recession Late Growth
Countries the Reforms GNIuc > 0 GNIVC < 0 GNIVC > 0
Argentina 1976

Brazil 1980

Chile 1973

Colombia 1984

Costa Rica 1981

Guatemala 1986
Honduras 1988

Mexico 1992

Panama No reform

Peru 1977-78

Umguay 1983

Venezuela 1984

'Urban Poverty headcount ratio not available.

*Rural Poverty headcount ratio not available.
"Gini not available.
GNIpc not available for Panama 1970-79 and Venezuela 1970-81.
Data sources: See authors' webpages.

the growth rate in the Gini coefficient (G). Working with growth rates, however,
underestimates the difficulty of reducing poverty and inequality at high levels of
Ph,P:, and G. Consequently, the regression equations overrate the achievements
at low levels of poverty and inequality. An alternative set of indicators is to use
changes in the rural and urban incidence of poverty and in inequality over the
spells. In this case, the indicators underestimate the difficulty of reducing poverty
and inequality at low levels of P;, PI;, and G, and the regressions overrate the
achievements at high levels of poverty and inequality. In the empirical analysis,
we use both types of indicators of change to identify the robustness of the deter-
minants of poverty and inequality.
 In the econometric analysis that follows, each country is considered as a
separate experience and the data are consequently not weighted by population
size. However, they are weighted by years of duration of each spell since a longer
spell is equivalent to a repeat of observations of years of growth or recession
compared to a shorter spell. Spells were longer under early than late growth since
household expenditure surveys were not as frequent in the 1970s and early 1980s.
The endogenous variables in the regressions are consequently the average annual
growth rates of, and the average annual differences in, poverty and inequality
during the corresponding spells.
Countries have countless unobservable characteristics that also influence the
observed changes in poverty and inequality. Since we work with panel data, and
the endogenous variables are changes over spells, additive fixed effects are elimin-
ated by differencing. However, non-linear unobservable country effects may
remain. We consequently introduce fixed and random effects in the estimated
equations and test for the hypotheses that fixed effects are zero and random
effects have a zero variance. We use these effects whenever the tests reject the
corresponding hypotheses.

In the following analyses, we start by estimating the roles of aggregate income

growth, of the qualitative features of growth, of the macroeconomic performance
that accompanies growth, and of the structural context where growth occurs, on
changes in poverty and inequality. We then qualify the role of income growth by
showing how its effects on poverty and inequality are modified by the initial levels
of inequality, poverty, and education; by differences between episodes of early
growth, recession, and late growth; and by the sectoral composition of growth.

4.1. Urban Poverty

Role of Aggregate Income Growth
Results in Table 2 show the determinants of the annual growth rate in the
incidence of urban poverty P;. The hypotheses that fixed effects are zero and
random effects have a zero variance could not be rejected. These effects are conse-
quently not included in the results presented in Table 2.
Confirming what other analysts have observed (Fields, 1992; Morley, 1995;
Psacharopoulos et al., 1995; Altimir, 1995; Lustig, 1995), we find that GNIPC
reduces urban poverty, here with an elasticity of - 0.95. Hence, urban poverty is
indeed anti-cyclical, falling under GNIpc growth and rising in recession.
Among macroeconomic variables, depreciation of the real exchange rate
increases urban poverty. Adjustment policies have a short run cost on poverty.
By contrast, structural variables have no significant effects.
We can use the estimated equation in Table 2 to calculate the relative roles
of income growth and of all the other variables that explain poverty. At the
sample mean of all explanatory variables (except GNIPC) over the 48 spells for
 TABLE 2
DETERMINANTS
OF CHANGEIN URBANPOVERTY

Coefficient P-value
Aggregate income growth
GNIpc growth
Qualitative features of growth
Coefficient of variation of GNIpc
Macroeconomic performance
Real exchange rate growth
Hyperinflation dummy
Structural context at beginning of spell
GNIpc ('000 1995US$)
Share of agriculture in GDP
Population growth
Urban population share
Secondary education
Inequality
Incidence of urban poverty
Intercept
Number of observations
R2
Adiusted R 2

Endogenous variable: annual growth rate in urban poverty incidence by

spell.
Weighted OLS, with weights equal to the length of the spell. Hypotheses
that country fixed effects are zero and random effects have a zero variance
could not be rejected with an F-test and a Breusch and Pagan Lagrangian
multiplier test, respectively.

which there are complete data and for the whole 1970-94 period, the estimated
equation is:4

With GNIpc = 0.83 over the sample, the net upward effect on the rate of change
in the incidence of urban poverty of the combined 2.33 percent drift due to other
variables and the negative effect of income growth with an elasticity of - 0.95 is
1.54 percent. What this result shows is that, for growth to have held the incidence
of urban poverty in check, the average annual growth rate in GNIpc over the
period would have had to be not 0.83 percent but a much larger 2.45 percent.
Clearly, the growth performance over the period fell far short of this achievement
and the incidence of urban poverty rose.

Role of the Qualifiers of Income Growth

The qualifiers of income growth were introduced one at a time in separate
regressions by interacting GNIPCwith these qualifiers. Partial results are reported
in Table 3 when significant. The first three experiments show that GNIPC is only
effective in reducing the growth rate of urban poverty if (1) the level of inequality

4 ~ h regression
e is written as P: = Xfl - 0.95 G N I ~ Cwhere
, X represents all the covariates other
. the mean value 8 of these covariates, Xfi = 2.33.
than G N I ~ CAt
 TABLE 3
PARTIAL
RESULTSON THE DETERMINANTS
OF CHANGEIN URBANPOVERTY

Overall Effect of Growth

Adjusted
Coefficient P-value R~ Lowest Highest
In 1993-94
Interactive effect with initial inequality (Uruguay) (Brazil)
GNIpc growth -2.92 0.01 - 1.61** 0.23
Initial inequality *GNIpc growth 4.36 0.08
Interactive effect with initial poverty (Uruguay) (Honduras)
GNIpc growth - 2.24 0.00 - 1.98** 0.73
Initial urban poverty *GNIpc growth 0.04 0.01
Interactive effect with education (Honduras) (Uruguay)
GNIpc growth 0.03 0.96 - 0.58** - 1.49**
Secondary education *GNIpc growth -0.02 0.06
Growth episodes
GNIpc growth, early growth episode - 0.22 0.73
GNIpc growth, recession episode - 1.06 0.01
GNIpc growth, late growth episode - 1.05 0.00
Growth episodes and interactive effect with initial inequality (Uruguay 92) (Guatemala 86)
GNIpc growth, late growth episode - 5.53 0.04 - 2.45** 0.10
Initial inequality *GNIpc growth 10.24 0.09
Sectoral composition of growth
Agricultural growth 0.06 0.89
Industrial growth -0.06 0.75
Service growth -0.95 0.01
Sectoral composition of growth and interactive effect with initial inequality (Uruguay) (Brazil)
Service growth -4.54 0.03 0.63 - 1.95** 0.78
Initial inequality *Service growth 8.63 0.08

Endogenous variable: annual growth rate in urban poverty incidence by spell.

Qualitative features of growth, macroeconomic performance, and structural context variables as
in Table 2 not reported.
*(**)Significantly different from 0 at the 90% (95%) confidence level.

is not too high, (2) the level of poverty is not too high, and (3) the level of
secondary education is high enough. To illustrate the meaning of these results,
we predict the roles of the initial levels of inequality and poverty and of the level
of secondary education on the income elasticity of poverty by using their lowest
and highest observed values in the sample of countries in 1993-94. This gives the
following results. While with low inequality (a Gini of 0.30 as observed in Urug-
uay) the overall income elasticity of poverty is - 1.61, it is not significantly differ-
ent from zero with high inequality (a Gini of 0.62 as observed in Brazil). Hence,
as suggested by Ravallion (1997), high inequality does indeed erase the ability of
growth to reduce poverty. Put it another way, countries with high levels of
inequality cannot rely on growth to reduce poverty, but need to deal first with
inequality through a separate set of redistributive interventions if they want to
rely on growth as an anti-poverty strategy. Similarly, the overall income elasticity
of poverty is - 1.98 with low initial urban poverty (a headcount ratio of 6 percent
as observed in Uruguay) while it is not significantly different from zero with high
poverty (a headcount ratio of 70 percent as observed in Honduras). Hence,
growth is only effective in reducing poverty when poverty is not too high. Other-
wise, direct interventions to reduce poverty are first needed. Finally, a high level
of secondary school enrollment (83 percent in Uruguay) yields an income elas-
ticity of poverty of - 1.49, while it is only - 0.58 (but still significantly different
from zero at the 95 percent confidence level) with a low school enrollment
(33 percent in Honduras). Hence, a one percentage point of income growth is 2.6
times more effective in reducing poverty at a high than at a low level of secondary
education, confirming the key role that education plays in allowing growth to
reduce poverty as suggested by Psacharopoulos et al. (1995).
The next two experiments in Table 3 have the objective of testing for asym-
metries between the roles of early growth, recession, and late growth in affecting
poverty. We find that there is a strong asymmetry between early growth and
recession. Early growth was unable to reduce urban poverty. By contrast, a one
percentage point decline in GNIpc increases poverty by 1.06 percent. An econo-
metric analysis of the relationship between income growth and poverty using data
for early growth and recession without separating these episodes would thus have
been highly misleading: the estimated negative relation would have been estab-
lished by recession instead of growth. However, this asymmetry disappears with
late growth, as it reduces poverty with an elasticity of - 1.05. Late growth (open
economy) is thus much more effective to reduce poverty than was early growth
(ISI).'
The role of growth episodes is also conditional on the initial level of
inequality. We found that, overall, growth only reduces poverty when inequality
is low. We see that this relation is basically established during the late growth
period. For late growth, the overall elasticity is - 2.45 with low inequality (a Gini
of 0.30 as observed in Uruguay in 1992) but insignificant with high inequality (a
Gini of 0.55 as observed in Guatemala in 1986). Open economy growth is thus
particularly sensitive to inequality in its capacity to reduce poverty.
Finally, the last experiment in Table 3 looks at the importance of the sectoral
composition of growth. Results show that service sector growth is effective in
reducing urban poverty, while agricultural and industrial sector growth are not.
Interacting sectoral growth with inequality, a low inequality (a Gini of 0.30 as
observed in Uruguay) gives service sector growth an income elasticity of poverty
of - 1.95, while a high inequality (a Gini of 0.62 as observed in Brazil) erases the
role of service sector growth in poverty reduction. Growth of the service sector
is thus particularly sensitive to inequality in reducing poverty.

4.2. Rural Poverty

The data on rural poverty are not as complete as those on urban poverty,
and there are reservations about the validity of some of the data reported in
ECLAC (1998). In particular, the rural poverty data for Argentina, Colombia,
and Uruguay reported by ECLAC are not nationally representative. Dropping
these data reduces the number of useable spells to 34.

5 .
Since reforms in Colombia came somewhere around the middle of the 1980-86 spell, it is not a
priori evident whether this spell should be assigned to early or to late growth. We did a sensitivity
analysis by reassigning this spell to late growth and found that results are robust to this change. The
income elasticities of urban poverty for the three periods are - 0.19, - 1.05, and - 1.O8, respectively.
 TABLE 4
DETERMINANTS
OF CHANGEIN RURALPOVERTY

Coefficient P-value
Aggregate income growth
GNIpc growth
Qualitative features of growth
Coefficient of variation of GNIpc
Macroeconomic performance
Real exchange rate growth
Hyperinflation dummy
Structural context at beginning of spell
GNIpc ('000 1995US$)
Share of agriculture in GDP
Population growth
Rural population share
Secondary education
Inequality
Incidence of rural poverty
Intercept
Number of observations
Overall R2

Endogenous variable: annual growth rate in rural poverty incidence by

spell.
Weighted OLS, with weights equal to the length of the spell.
Random effects model.

Role of Aggregate Income Growth

In testing for fixed and random effects, we find that the former can be
rejected but not the latter. In Tables 4 and 5 , we consequently report results for
rural poverty equations with random effects. Results in Table 4 show that, like

TABLE 5
PARTIALRESULTSON THE DETERMINANTSOF CHANGEIN RURALPOVERTY

Overall Effect of Growth

Overall
Coefficient P-value R2 Lowest Highest
Interactive effect with initial poverty (Costa Rica) (Honduras)
GNIpc growth 0.01 0.69 - 1.18** - 0.29
Initial rural poverty *GNlpc growth 0.17
Growth episodes
GNIpc growth, early growth episode
GNIpc growth, recession episode
GNIpc growth, late growth episode
Sectoral composition
Agricultural growth
Industrial growth
Service growth

Endogenous variable: annual growth rate in rural poverty incidence by spell.

Qualitative features of growth, macroeconomic performance, and structural context variables as
in Table 4 not reported.
*(**) Significantly different from 0 at the 90% (95%) confidence level.
Random effects models.
in the urban sector, GNIpc is a significant force in reducing the incidence of rural
poverty. The income elasticity is - 0.75, compared to - 0.95 in the urban sector.
Hence, rural poverty is less responsive to changes in aggregate income: it responds
less to growth but also less to recession, sheltering the rural poor from booms
and busts.
At the sample mean of all variables else than G N I ~ Cfor
, the 34 spells over
the whole 1970-94 period, the estimated equation is:

With G N I ~ = C0.64 over the sample of 34 spells, the net effect is a downward drift
in the incidence of rural poverty of - 0.36 percent. The variables that contribute
significantly to this negative drift in rural poverty are the initial levels of GNIpc
and of secondary education. By contrast, the macroeconomic performance (real
exchange rate depreciation), which was a source of increase in urban poverty, has
no effect on rural poverty. Hence, rural poverty was also sheltered from macro
performance. Also, observed annual growth in per capita income (0.64 percent)
was greater than the minimum rate of growth needed to prevent an increase in
rural poverty, namely 0.16 percent, and the incidence of rural poverty fell.

Role of Qualifiers of Income Growth

The role of income growth in reducing rural poverty is qualified by the initial
level of poverty (Table 5). The income elasticity of poverty is - 1.18 at low levels
of rural poverty (Costa Rica with an incidence of rural poverty of 23 percent)
while it is not significantly different from zero at high levels of poverty (Honduras
with an incidence of rural poverty of 76 percent). Hence, similar to urban poverty,
the beneficial effect of growth on poverty is erased at high levels of poverty.
However, by contrast to urban poverty, secondary education does not help make
growth more effective in reducing rural poverty, a phenomenon that has been
observed by Lopez and Valdes (1997) across a number of Latin American
countries.
Similar to urban poverty, there is an asymmetry in the role of early growth
and recession, with a 1 percent decline in income during recession erasing the
poverty gains of 3.0 percentage points of early growth. However, this asymmetry
is to the benefit of growth under late growth, with a 1 percentage point of growth
erasing the poverty losses of 1.6 percentage points of recession.
Finally, as in the urban sector, the sectoral composition of growth matters
for rural poverty reduction, with service sector growth the most effective, with an
elasticity of - 0.70, while agricultural and industrial growth do not have signifi-
cant effects on rural poverty.

4.3. Inequality
Role of Aggregate Income Growth
Inequality, measured by the Gini coefficient, is harder to explain than pov-
erty (Table 6). There are also measurement problems for inequality since reported
Gini are sometimes national and sometimes only urban. We used national
inequality data from Altimir (1998), Morley (1995), Lustig and Deutsch (1988),
280
 TABLE 6
DETERMINANTS
OF CHANGEIN INEQUALITY
- -

Coefficient P-value
Aggregate income growth
GNIpc growth
Qualitative features of growth
Differential growth agriculture-non-agriculture
Coefficient of variation of GNIpc
Macroeconomic performance
Real exchange rate growth
Hyperinflation dummy
Structural context
GNIpc ('000 1987US$)
Share of agriculture in GDP
Population growth
Urban population share
Secondary education
Initial inequality
Intercept
Number of observations
R2
Adjusted R2

Endogenous variabIe: annual growth rate of Gini coefficient by spell.

Weighted OLS, with weights equal to the length of the spell.
Hypotheses that country fixed effects are zero and random effects have a zero
variance could not be rejected with an F-test and a Breusch and Pagan Lagrangian
multiplier test, respectively.

CEPAL (1993 and 1995), and INEGI (1993 and 1994) when available, and urban
inequality data from Social Panorama in Latin America (ECLAC, 1996) when
not.6 We show in the Appendix that there is no significant bias in combining
national and urban Gini in analyzing (1) the role of the initial level of inequality
(Go) on the growth of poverty and inequality, and (2) the determinants of growth
in inequality (G). Results are consequently robust to these definitional problems.
They indicate that income growth did not reduce inequality over the 1970-94
sample of spells at the 90 percent confidence level. This lack of influence is consist-
ent with the Ravallion and Chen (1997) finding. However, frequently made
assertions that growth has been inequalizing in Latin America are globally incor-
rect. The role of variables other than income growth implies a positive drift in
the annual growth rate of inequality of 0.58 percent. Among these variables, the
initial level of inequality contributes to reduce this drift, suggesting that it is
harder for inequality to rise when it is already high.

Role of Qualifiers of Income Growth

There is a strong asymmetry in the effects of growth and recession on
inequality. Results in Table 7 show that the association between income growth
and change in inequality exclusively comes from recession. Neither early nor late
growth are able to make a dent in inequality. Hence, estimating an overall relation

6See the authors' webpages for detailed information on data sources.

 TABLE 7
PARTIALRESULTSON THE DETERMINANTS
OF CHANGEIN INEQUALITY

Coefficient P-value Adjusted R'

Growth episodes
GNIpc growth, early growth episode 0.65 0.17 0.26
GNIpc growth, recession episode - 0.28 0.07
GNIpc growth, late growth episode - 0.02 0.92

Endogenous variable: annual growth rate of Gini coefficient by spell.

Qualitative features of growth, macroeconomic performance, and structural
context variables as in Table 6 not reported.

between income and inequality, as in Table 6, without distinguishing between

episodes of growth and recession is misleading: the policy implication is not that
income growth does not affect inequality, but that recession is devastating on
inequality, and growth ineffective in reducing it. Growth, even late growth which
we have seen to be powerful in reducing both urban and rural poverty, cannot
be relied upon as an equalizing force. If greater equality is deemed desirable by
governments, in particular to enhance the poverty reducing effect of growth, then
direct inequality-reducing interventions will be necessary.

We consider here two issues to check on the robustness of the results. One
has to do with the choice of endogenous variables and the other with measure-
ment errors.
As indicated above, using rates of change in poverty and inequality as
endogenous variables in the estimated equations gives more importance to
achievements in low poverty and low inequality countries, respectively. An alter-
native specification of the endogenous variables is to use points of change in the
indicators of poverty and inequality, again scaled on an annual basis within spells
since they are of unequal length. This specification gives more importance to
countries with high levels of poverty and inequality. Estimated coefficients from
the two sets of endogenous variables would thus bracket what an ideal indicator
would offer. We should recall, however, that in both cases we control for the
initial levels of poverty and inequality, thus reducing the difference in these indi-
cators on estimated coefficients. Results in Table 8 show that the role of aggregate
income growth remains the same with the two approaches. A one percent increase
in GNIpc lowers the incidence of urban poverty by 0.25 percentage point and the
incidence of rural poverty by 0.30 percentage point. Asymmetries between growth
episodes remain clear. Early growth had no impact on poverty reduction,
recession creates a sharp increase in poverty, and late growth is effective in reduc-
ing poverty. The overall effect of GNIpc growth on inequality has borderline
significance. Hence, one could say that aggregate income growth has a negative
effect on inequality. However, it is clear that this effect is exclusively due to
recession. Neither early nor late growth have an impact on inequality. The con-
clusion that crises create ratchet effects on inequality that growth cannot erase is
thus confirmed.
 TABLE 8
SENSITIVITY OF POVERTYAND INEQUALITY
ANALYSIS:INDICATORS

Coefficient P-value Goodness of Fit

Determinants of change in urban poverty Adjusted R*
GNIpc growth - 0.25 0.00 0.55
Growth episodes
GNIpc growth, early growth episode -0.18 0.28 0.53
GNIpc growth, recession episode - 0.28 0.00
GNIpc growth, late growth episode - 0.22 0.02
Determinants of change in rural poverty (random effects model) Overall R~
GNIpc growth - 0.30 0.00 0.64
Growth episodes
GNIpc growth, early growth episode - 0.06 0.85 0.66
GNIpc growth, recession episode - 0.30 0.03
GNIpc growth, late growth episode - 0.39 0.03
Determinants of change in inequality Adjusted
GNIpc growth - 0.0007 0.10 0.33
Growth episodes
GNIpc growth, early growth episode 0.0029 0.14 0.34
GNIpc growth, recession episode - 0.0013 0.05
GNIpc growth, late growth episode -0.0001 0.89

Endogenous variables: changes in urban and rural poverty incidence, and change in Gini.
Qualitative features of growth, macroeconomic performance, and structural context variables as
in Tables 2, 4, and 6 not reported.

Econometrically, the fact of using a rate of change as the endogenous vari-

able, while using the initial level of the same variable as exogenous, can create a
correlation between the exogenous variable and the error term if there are
measurement errors in the exogenous variable, resulting in an estimation bias.
This concern would apply to the inequality equation where the change in Gini is
explained by the initial level of inequality with a significant coefficient. Since one
can expect measurement errors on inequality to be large, the significant role of
initial inequality may be due to measurement errors. To verify robustness of the
results to this specification, we instrumentalize the initial level of inequality by
splitting the observations into three groups and using dummy variables for the
lowest and highest third, with the middle group serving as reference level. This
procedure averages measurement errors and thus reduces the correlation between
the exogenous variable and the residual. Results in Table 9 show that significance

TABLE 9
SENSITIVITY ERRORSIN INEQUALITY
ANALYSIS:MEASUREMENT

Coefficient P-value Adjusted R'

GNIpc growth -0.14 0.16 0.23
Initial inequality: lowest third -0.11 0.88
Initial inequality: highest third - 1.47 0.07
Endogenous variable: annual growth rate of Gini coefficient by spell.
Qualitative features of growth, macroeconomic performance, and structural
context variables as in Table 6 not reported.

283
of the initial inequality effect was not due to measurement errors. The income
elasticity of inequality also remains unaltered (-0.14) and non-significant.

While there is no agreement among analysts of Latin American incomes

about what the best data set is, we found that, with the information used here
that principally derives from the efforts of the ECLAC at constructing a consist-
ent data set, strong causal regularities emerge. Analysis of the determinants of
change in poverty and inequality over spells between 1970 and 1994 shows that
per capita aggregate income growth can lower the incidence of urban and rural
poverty with statistically significant elasticities of -0.95 and -0.75, respectively.
Growth has, by contrast no significant impact on inequality, although significance
can be borderline when using points of change in Gini instead of rates of change,
i.e., when using an indicator of change that gives more weight to countries with
high levels of inequality.
However, the reassuring result that growth can be relied upon as an instru-
ment to reduce poverty needs to be seriously qualified for the following four
reasons:
1. Growth only reduces urban poverty if the initial levels of inequality and
poverty are not too high. In the Latin American countries where this is not satis-
fied, growth is ineffective in reducing urban poverty. Once these thresholds have
been reached, the lower the initial levels of inequality and poverty are, the more
effective growth is in reducing poverty. For rural poverty, the same result holds
with respect to the initial level of poverty. While the level of secondary education
has no effect on the income elasticity of rural poverty, it needs to be sufficiently
high for growth to reduce urban poverty. Hence, current concerns that aggregate
income growth is not sufficiently powerful to attack urban poverty imply that
targeted efforts need to be made to reduce inequality and poverty and to increase
educational levels to make it more effective.
2. There exists an asymmetry in the effect of income change on poverty,
whereby a 1 percent decline in income can cancel the poverty reduction effects of
more than 1 percent of income growth. For urban and rural poverty, this was
true between early growth and recession. If asymmetry prevails, the overall
relation between income and poverty is misleading since it is principally obtained
through recession instead of through growth, leading to erroneous predictions
about the potential of growth to reduce poverty. A very encouraging result is that,
following the shift from import substitution to open economy industrialization,
asymmetry in poverty effects is no longer observed under late growth. Late
growth is a strong force in reducing the incidence of poverty, with elasticities of
-1.05 for urban poverty and -1.06 for rural poverty. Late growth is particularly
sensitive to inequality in helping reduce urban poverty. Hence, direct attacks on
inequality can help make late growth a more effective poverty reducing
instrument.
3. While late growth is effective in reducing poverty, this is not the case with
inequality. Results show that growth was always ineffective in reducing
inequality, under early as well as late growth. Hence, while it cannot be said that
growth is inequalizing, neither can reliance be placed on growth to reduce
inequality. Other instruments need to be used for this purpose, basically focusing
on enhancing control over assets for the poor and on equalizing opportunities in
accessing the markets, institutions, and public goods that determine the income
generation value of these assets. Recession creates a ratchet effect on inequality.
Strong asymmetry between growth and recession implies that avoiding the
inequalizing effects of recession should be an explicit policy concern.
4. Finally, differential growth of the services sector has been key in reducing
the growth of both urban and rural poverty. It has often been observed that the
informal sector, which includes a large service component, has indeed been
the most dynamic source of employment creation (PREALC, 1991). The urban
poverty reducing effect of services growth can be reinforced by lowering the level
of inequality.
We thus conclude that the current concerns with inequality in Latin America
are unlikely to be met with aggregate income growth. Inequality needs to be
attacked through direct policies designed for this purpose. Growth can be effec-
tive to reduce urban and rural poverty, but only beyond thresholds in the levels
of inequality and poverty. For urban poverty, growth is only effective if the level
of inequality is sufficiently low and that of secondary education sufficiently high.
Beyond these thresholds, the urban poverty reducing effects of aggregate income
growth can be significantly enhanced by direct interventions to reduce the initial
levels of inequality and poverty to and raise the coverage of secondary education.

To test whether there is a systematic bias in combining Gini measurements

at the national and urban levels in regression analyses that use rates of growth as
endogenous variables, all poverty regressions were run with a multiplicative
national dummy variable D, (taking the value of one when the Gini is measured
at the national level), and all inequality regressions were run with both level and
multiplicative national dummies in the inequality equations as follows:
Poverty equations: P = P'X+ (yo+ ylD,)Go
Inequality equations: G = ( a o+ a,D,) + P'X+ (yo+ y,D,)Go,
where Xis the set of other regressors in each equation.
In all regressions, the coefficient of the D, variable was not significant. Simi-
lar interactive and level dummies were introduced in all regressions with inter-
active terms between inequality and GNIpc growth, inequality and GNlpc growth
per episode, and inequality and sectoral growth, with no significant effects. We
thus conclude that there is no significant bias in combining the national and urban
Gini in analyzing (1) the role of the initial level of inequality (Go) on the growth
of poverty and inequality, and (2) the determinants of growth in inequality (G).

Aghion, P. and P. Howitt, Endogenous Growth Theory, MIT Press, Cambridge, MA, 1988.
Altimir, O., Changes in Inequality and Poverty in Latin America, Chile CEPAL, Santiago, 1995.
 , Inequality, Employment, and Poverty in Latin America, in V. Tokman and G. O'Donnell,
eds., Poverty and Inequality in Latin America, University of Notre Dame Press, Notre Dame, IN,
1998.
Beccaria, L., J. Boltvinik, J. C. Feres, 0 . Fresneda, A. Leon, and A. Sen, America Latina: El Reto de
la Pobreza, Caracteristicas, Evolucidn y Perspectivas, UNDP, Bogota, 1992.
BCnabou, R., Inequality and Growth, in B. Bernanke and J. Rotemberg, eds., NBER Macroeconomics
Annual 11, MIT Press, Cambridge, MA, 1996.
Berry, A,, The Income Distribution Threat in Latin America, Latin American Research Review, 3-40,
32(2), 1997.
Birdsall, N. and J. L. Londoiio, Asset Inequality Matters: An Assessment of the World Bank's
Approach to Poverty Reduction, American Economic Review, 32-7, 87(2), 1997.
Bourguignon, F., J. de Melo, and C. Morrisson, Poverty and Income Distribution During Adjust-
ment: Issues and Evidence from the OECD Project, World Development 1485-1508, 19(1I), 1991.
Bourguignon, F. and C. Morrisson, Inequality and Development: The Role of Dualism, Journal of
Development Economics, 233-57, 57(2), 1998.
CEPAL (Comision Economica para America Latina y el Caribe), La Pobreza en Chile en 1992,
CEPAL, Santiago, 1993.
,Situacion de la Pobreza en Chile: Encuesta CASEN 1994, CEPAL, Santiago, 1995.
Datt, G. and M. Ravallion, Why Have Some Indian States Done Better than Others at Reducing
Rural Poverty? Economica, 65, pp. 17-38, 1998.
Deininger, K. and L. Squire, A New Data Set Measuring Income Inequality, The World Bank Econ-
omic Review, 10(3), 565-91, 1996.
ECLAC (Economic Commission for Latin America and the Caribbean), Social Panorama of Latin
America, Economic Commission for Latin America and the Caribbean, Santiago, 1996.
Economist (The), Backlash in Latin America: Gestures Against Reform, November 30, 1996.
Fields, G., Poverty, Inequality, and Development, Cambridge, University Press, London, 1980.
,Changing Poverty and Inequality in Latin America, Public Finance, Vol. 47, pp. 59-76, 1992.
Fujii, G. and G. Aguilar, La distribucion del ingreso en Mtxico, 1984-1992: un estudio por compon-
entes, Comercio Exterior, Vol. 45, No. 8, pp. 609-14, 1995.
INEGI (Instituto Nacional de Estadistica Geografia e Informatica), Magnitud y Evolucidn de la
Pobreza en Mixico, 1984-1992. INEGI, Mexico City, 1993.
, Encuesta Nacional de Ingresos y Gastos de 10s Hogares 1994, INEGI, Mexico City, 1994.
IDB (Inter-American Development Bank), Latin America After a Decade of Reforms, Economic and
Social Progress in Latin America. 1997 Report, The Johns Hopkins University Press, Baltimore,
MD, 1997.
,Facing Up to Inequality in Latin mer rich, Economic and Social Progress in Latin America,
1998-1999 Report, The Johns Hopkins University Press, Baltimore, MD, 1998.
Londoiio, J. L. and M. Szekely, Persistent Poverty and Excess Inequality: Latin America, 1970-1995,
Working Paper Series 357, Inter-American Development Bank, Washington, DC, 1997.
Lopez, R. and A. Valdts, Rural Poverty in Latin America: Analytics, New Empirical Evidence, and
Policy, Technical Department, Latin America and the Caribbean Region, The World Bank,
Washington, DC, 1997.
Lustig, N., Measuring Poverty in Latin America. The Emperor Has No Clothes, The Brookings
Institution, Washington, DC, 1994.
,Introduction in Coping with Austerity: Poverty and Inequality in Latin America, N . Lustig,
ed., The Brookings Institution, Washington, DC, 1995.
Lustig, N. and R. Deutsch, The Inter-American Development Bank and Poverty Reduction: An Over-
view, Inter-American Development Bank, Washington, DC, 1998.
Morley, S., Poverty and Inequality in Latin America: The Impact of Adjustment and Recovery in the
1980s. The Johns Hopkins University Press, Baltimore, 1995a.
,Structural Adjustment and Determinants of Poverty in Latin America, in N. Lustig, ed.,
Coping with Austerity: Poverty and Inequality in Latin America, The Brookings Institution,
Washington, DC, 1995b.
PREALC (Programa Regional de Empleo para America Latina y el Caribe), Empleo y Equidad: El
Desajio de 10s 90, International Labor Office, PREALC, Santiago, 1991.
Psacharopoulos, G., S. Morley, A. Fiszbein, H. Lee, and W. Wood, Poverty and Income Inequality
in Latin America During the 1980s, Review of Income and Wealth, Series 41, No. 3, pp. 245-64,
September 1995.
Ravallion, M., Can High-Inequality Developing Countries Escape Absolute Poverty? Economic Let-
ters, 56, pp. 51-7, 1997.
Ravallion, M. and S. Chen, What Can New Survey Data Tell Us About Recent Changes in Distri-
bution and Poverty? The World Bank Economic Review, 11(2), 357-82, 1997.

286
Ravallion, M. and G. Datt, How Important to India's Poor Is the Sectoral Composition of Economic
Growth? The World Bank Economic Review 10(1), 1-25, 1996.
Sahn, D., P. Dorosh, and S. Younger, Exchange Rate, Fiscal, and Agricultural Policies in Africa:
Does Adjustment Hurt the Poor? World Development 24(4), 719-48, 1996.
Sala-I-Martin, X., The Classical Approach to Convergence Analysis, The Economic Journal 106(437),
1019-36, 1996.
Stiglitz, J., Some Lessons from the East Asian Miracle, World Bank Research Observer, 11, 151-77,
1996.
Williamson, J., Latin American Adjustment: How Much Has Happened? Institute for International
Economics, Washington, DC, 1990.
World Bank (The), World Development Report 1990: Poverty, Oxford University Press, Oxford, 1990.