Inequity within countries: individuals and groups 2

Across the world, individuals and groupsface highly unequal opportunities to betterthemselves economically and socially.Inequalities, as such, might not be of particu-lar concern if outcomes varied for reasonsthat had to do mainly with individual efforts.But, taking our cue from the first chapter, weare concerned here with systematic differ-ences in opportunities for individuals andgroups who differ only in skin color, caste,gender, or place of residence, predeterminedcharacteristics that can be argued to be“morally irrelevant.” As illustrated in focus 1,on the Indian village of Palanpur, when suchinequalities of opportunity are pronounced,they are often reproduced over time and notonly affect welfare directly but also act to sti-fle human development and economicgrowth.

On the basis of what predeterminedcharacteristics should groups be definedsuch that we would not want to see sys-tematic differences in their opportunities?Clearly there is no single answer. Roemer(1998) argues that society has to make thischoice through some kind of ethical andpolitical process. The circumstances couldinclude social origin variables outside anindividual’s control, such as sex, race, eth-nicity, caste, parental education and occu-pation, wealth, or place of birth. Cogneau(2005) notes that a society’s choice ofcircumstances establishes a direct linkbetween equality of opportunities and theintergenerational transmission of out-comes. In this chapter, we are largely com-pelled to let data availability dictate thegroup definitions we consider. We canthus present only a partial, and often rudi-mentary, picture of the full range ofinequity that might exist in a country.Because we wish not only to look within a

country but also to compare across coun-tries, we use group definitions of broadrelevance.

Although economic inequalities areclearly part of the story, this chapter goesbeyond incomes to emphasize inequalitiesin key dimensions of opportunity, such ashealth, education, and the freedom andcapacity of people to participate in andshape society. There is a special concernwith inequalities that tend to perpetuatedifferences across individuals and groupsover time, within and across generations.These result in “inequality traps” that arepervasive in many countries. Such inequal-ity traps reinforce our concern with equityon intrinsic grounds, but they can also beparticularly detrimental to the developmentprocess, because they act to curtail eco-nomic dynamism.

A key objective here is to show howinequalities combine, interact, and arereproduced through interlinked economic,political, and sociocultural processes. Indi-viduals and groups differ markedly in theirpower to influence these processes; indeed,they differ even in their capacity to aspireto such influence. The report emphasizesthat such “agency” is a dimension ofopportunity, alongside education, health,and wealth. And inequalities of agency arecentral in explaining how inequalities ofopportunity are transmitted over time(box 2.1).

This chapter presents evidence of ahigh degree of inequality of opportunityin many developing countries—inequali-ties manifest in a variety of dimensions,such as health, education, and income. Itthen focuses on the specific dimension ofinequality of power, or agency. Through-out the chapter, we emphasize that

28

Inequity within countries:individuals and groups

2c h a p t e r

(c) The International Bank for Reconstruction and Development / The World Bank

inequalities in different dimensions caninteract with, and reinforce, one anotherover time. To highlight these connections,we end by focusing on the specific case ofgender inequity.

Inequalities in healthAlongside the intrinsic importance ofhealth as a dimension of welfare, poorhealth can directly influence an individ-ual’s opportunities—his or her earningscapacity, performance at school, ability tocare for children, participation in commu-nity activities, and so on. This importantinstrumental function of health impliesthat inequalities in health often translateinto inequalities in other dimensions ofwelfare. And these inequalities are repro-duced over time. We focus here on chil-dren, while recognizing that differences insocial status, wealth, and health also mat-ter for adults.

Demographic and Health Survey (DHS)data indicate that health status variessharply across population groups. To whatextent does it vary across population groupsdefined by characteristics that are predeter-mined and arguably have no moral rele-vance? We draw on DHS data from 60countries to examine how the health ofchildren varies across population groupsdefined by mother’s education, rural orurban residence, and parent’s economic sta-tus, proxied by an index of household own-ership of consumer durables. (We look fur-ther at cross-country differences in healthin chapter 3.)

Infant mortality. For these countries,infant mortality rates vary markedly—from a low of around 25 per 1,000 livebirths in Colombia and Jordan, to morethan 125 in Mali, Niger, and Mozambique(figure 2.1). But even where overall infantmortality rates are high, the figures forchildren whose mothers have a secondaryeducation or higher are dramaticallylower. The risk of death among childrenwith well-educated mothers in Mali, forexample, is about the same as that for theaverage child in Indonesia. And while theoverall infant mortality rate in Brazil lies

below 50 (estimates from 1996), the ratefor children whose mothers have not beeneducated is roughly twice as high. Furtheranalysis, not reported here, indicates thatinfant mortality rates are also sharply differ-entiated across population groups definedby rural-urban residence and economic sta-tus, proxied by asset ownership.

Stunting. Another dimension of health,extreme stunting (with height-for-agebelow three standard deviations from thereference population), also varies markedlyacross countries. Overall rates are as high as 30 percent in Pakistan and the Rep-ublic of Yemen, but negligible in Trinidadand Tobago and very low in Jordan, Arme-nia, Brazil, and Kazakhstan (figure 2.2).

Inequity within countries: individuals and groups 29

As a prelude to the themes in this chapter,we describe one attempt to quantify thelevel and persistence of inequalities ofopportunity in Brazil, based on nationallyrepresentative household survey data. Brazilwas selected for a reason.With a Gini coeffi-cient of per capita incomes just below 0.6and persistent over time, it is generally per-ceived to be one of the world’s mostunequal countries.*

Brazil’s main household survey, thePesquisa Nacional por Amostra de Domicilios(PNAD), included in 1996 a set ofsupplemental questions on the parents ofrespondents.This permitted an analysis ofthe intergenerational persistence ininequalities. Using four circumstance vari-ables (parental schooling, father’s occupa-tion, race, and region of birth), Bourguignon,Ferreira, and Menendez (2005) investigatedhow inequalities of opportunity generateinequality in current earnings across differ-ent cohorts of adult individuals. Applying aconceptual framework closely related tothat in chapter 1, they decomposed earn-ings inequality into a lower bound compo-nent attributable to the inequality of oppor-tunity—to the effect of the four observedcircumstance variables—and a residualcomponent, which would account for per-sonal effort, luck, measurement error, transi-tory income, and other unobservable char-acteristics.They found that the fourvariables accounted for more than a fifth ofthe total earnings inequality within gender

cohorts. Of the four, family background wasmost important.

This distribution of certain opportuni-ties and outcomes has persisted across gen-erations.When the authors estimatedeconometrically the relationship betweenschooling and race, region of origin,parental education, and father’s occupation,only the coefficient on parental educationseems to have fallen across cohorts. In otherwords, race, region of origin, and father’soccupation continue to predict an individ-ual’s education level. And even for educa-tion, mechanisms are at work to reproduceschooling levels across generations, espe-cially at the lower end of the distribution.

Brazil underscores the need to look at arange of outcomes (of which incomes areonly one, with education, health and servicesalso of great concern). It also underscores theneed to look at a range of processes—ofwhich income and economic wealth-basedmechanisms form only part, and for whichgroup-based interactions are as central ashousehold and individual conditions, behav-iors, and characteristics.

Source: Bourguignon, Ferreira, and Menendez(2005).* The perception of particularly high inequalityin Brazil may to some extent be a result of theway income is measured there. Alternativeapproaches to measuring inequality, based onother welfare indicators, indicate that Brazilmay be less of an outlier in Latin America thanpreviously believed. See box 2.5 and also DeFerranti and others (2004).

B O X 2 . 1 Unequal opportunities persist across generationsin Brazil


30

Infant mortality rate per 1,000 live births160

140

120

100

80

60

40

20

0

No education

Secondary or higher

Colo

mbi

aJo

rdan

Sri L

anka

Viet

nam

Para

guay

Philip

pine

sTh

aila

ndBo

tsw

ana

Sou

th A

frica

Peru

Nica

ragu

aBr

azil

Turk

eyGu

atem

ala

Indo

nesia

Egyp

tTu

nisia

Zim

babw

eGh

ana

Mor

occo

Sene

gal

Keny

aNi

geria

El S

alva

dor

Turk

men

istan

Indi

aBo

livia

Eritr

eaSu

dan

Nepa

lBa

ngla

desh

Cam

eroo

nTo

goCo

mor

osHa

itiUg

anda

Yem

enCa

mbo

dia

Zam

bia

Pakis

tan

Beni

nM

adag

asca

r

Cent

ral A

frica

n Re

publ

icGu

inea

Burk

ina

Faso

Chad

Côte

d’Iv

oire

Mal

awi

Ethi

opia

Rwan

daM

ali

Nige

rM

ozam

biqu

e

Figure 2.1 Infant mortality varies across countries but also by mother’s education within countries

Source: Authors’ calculations from Demographic Health Survey (DHS) data.Note: The continuous dark line represents the mean infant mortality rate in each country, while the endpoints of the whiskers indicate the infant mortality rates by different levels of mother’s education.

Rural

Urban

35

30

25

20

15

10

5

0

Percentage of children severely stunted (z-score < 3)

Trin

idad

and

Tob

ago

(*)

Jord

anAr

men

iaBr

azil

Kaza

khst

an (*

)Co

lom

bia

Dom

inic

an R

epub

licPa

ragu

ayTu

nisia

Turk

eyEg

ypt

Gabo

nTu

rkm

enist

anHa

itiSr

i Lan

kaNa

mib

iaM

oroc

coNi

cara

gua

Ghan

aBo

livia

Zim

babw

eSe

nega

lGu

inea

Beni

nUg

anda

Côte

d’Iv

oire

Mau

ritan

iaTa

nzan

iaBu

rkin

a Fa

soBa

ngla

desh

Rwan

daM

ali

Cam

bodi

aCh

adGu

atem

ala

Nepa

lNi

geria

Zam

bia

Mal

awi

Ethi

opia

Yem

enPa

kista

n

Figure 2.2 Stunting levels of children born in rural versus urban areas are far from the same

Source: Authors’ calculations from Demographic Health Survey (DHS) data.Note: The continuous dark line represents the percentage of severely stunted children in each country, while the endpoints of the whiskers indicate the percentages for urban and rural areas.* Indicates stunting level in urban areas are higher than in rural areas.


The difference between children born inrural and urban areas can be dramatic,particularly at higher overall stunting lev-els. In Guatemala, stunting rates for chil-dren in urban areas are around 10 percent,but in rural areas they are as much as threetimes higher. Children in Guatemala clearlyhave no choice in deciding whether theyare born in the countryside or the city, buttheir opportunities to achieve good healthare clearly much less assured in rural thanin urban areas. As for infant mortalityrates, stunting among children is alsosharply differentiated by mother’s educa-tion and household economic status.

Access to immunization. Children born infamilies whose asset ownership places themin the top quintile of the distribution ofeconomic status have a high probability of

access to health services, proxied here ashaving received at least one of three keychildhood vaccinations—bacille Calmette-Guérin; diptheria, pertussis, and tetanus; ormeasles (figure 2.3). This is so even incountries where the overall percentage ofchildren without any coverage is as high as40 percent. Conversely, children whose par-ents are in the bottom quintile are muchmore likely to lack access to such basichealth care. In Morocco, where roughly 5percent of children have not received evenone of these three vaccinations, the propor-tion for children in the poorest quintile iswell above 15 percent.

High-impact health services. The WorldBank (2003j), drawing on DHS data from30 low- and middle-income countries, findsthat the poor are considerably less likely


Percentage not covered

70

60

50

40

30

20

10

0

Poorest

Wealthiest

Egyp

tJo

rdan

(*)

Colo

mbi

aRw

anda

Peru

Sout

h Af

rica

Ken

yaM

alaw

iBr

azil

Zam

bia

(*)

Viet

nam

Turk

eyGu

atem

ala

Tanz

ania

Indo

nesia

Turk

men

istan

(*)

Mor

occo

Ghan

aBe

nin

Philip

pine

sBa

ngla

desh

Com

oros

Boliv

iaPa

ragu

ayKa

zakh

stan

(*)

Yem

enBu

rkin

a Fa

soCa

mer

oon

Ugan

daIn

dia

Mau

ritan

iaHa

itiTo

goEt

hiop

ia

Cent

ral A

frica

n Re

publ

icM

adag

asca

rM

ozam

biqu

eGu

inea

Mal

iCa

mbo

dia

Pakis

tan

Eritr

eaNi

ger

Chad

Figure 2.3 Access to childhood immunization services depends on parents’ economic status

Source: Authors’ calculations from Demographic Health Survey (DHS) data.Note: The continuous dark line represents the percentage of children without access to a basic immunization package in each country, while the endpoints of the whiskers indicate thepercentages for the top and the bottom quintile of the asset ownership distribution.* Indicates that the poorest quintile have higher access to childhood immunization services than the wealthiest quintile.


than the non-poor to have access to high-impact health services, such as skilled deliv-ery care, antenatal care, and complemen-tary feeding. Similarly, Wodon (2005) drawson household survey data from 15 Africancountries to indicate that, while virtually allurban households are within one hour’stravel time to a health center, the propor-tion in rural areas is generally only aroundhalf, and as low as 35–38 percent in Nigerand Ethiopia.

Disability. Data from a number of coun-tries suggest that disabled people aremuch more likely to be poor. Hoogeveen(2003) reports that in Uganda the proba-bility of poverty for urban dwellers livingin a household with a disabled head is 38percent higher than for those who live in ahousehold with an able-bodied head. TheSerbian Poverty Reduction Strategyreports that 70 percent of disabled peopleare unemployed. In a study drawing on 10 household surveys in eight countries,self-reported disability was found to bemore correlated with nonattendance atschool than other characteristics, includ-ing gender or rural residence.1 Sen (2004)emphasizes that the disabled face not onlyan “earnings handicap,” associated with alower probability of employment andlower compensation for their work, butalso a “conversion handicap.” By this hemeans that a physically disabled personrequires more income than an able-bodied person to achieve the same livingstandard.

Social inequalities damaging health. Notonly are health outcomes correlated withinequalities in other dimensions, but suchsocial inequalities can be argued to bedetrimental to individual health out-comes.2 In his comprehensive review ofthe literature, Deaton (2003) argues that,while it is certainly plausible that variousinequalities (such as those in power) causebad health, it is not clear that inequality ofincome is the main culprit. He providesevidence suggesting that, after controllingfor an individual’s income, income in-equality at the group level does not matterindependently for individual health. Thus,

the main inequalities that affect healthmay not be in the income space. He citesexamples of other key dimensions ofinequality: land ownership, women’sagency (health and fertility in India), anddemocratic rights (in England in the1870s and in the U.S. South in the 1960s).In general, an individual’s rank in the rele-vant hierarchy has been found to be im-portant to health in animal and humanexperiments. Repeated stress associatedwith insults and the lack of control thatcomes from low rank has a well-developedbiochemical basis.3

The consequences of poor health arereflected in education achievements, eco-nomic prosperity, and future generations.Consider the plight of AIDS orphans insouthern Africa, the stark inequalities ofopportunity they face, and the possible rolefor public action (box 2.2).

DHS data (figures 2.1–2.3) providedetailed insights into the relationshipbetween inequalities in health and somekey circumstance variables. But they arenot particularly well suited to capturingthe contribution of detailed spatial factors,such as place of birth, in overall inequality,because of the limited sample size. In oneattempt to get around this problem, childheight in Cambodia was estimated at thecommune level based on a statistical pro-cedure to combine DHS data with popula-tion census data.4 The study documentsconsiderable heterogeneity across Cambo-dia’s more than 1,600 communes in theprevalence of stunting and being under-weight among children under the age offive (figure 2.4). The analysis providesclear evidence that in Cambodia a child’sopportunities for good health have astrong spatial dimension to them. Yetclearly, no child is able to determine inwhich locality he or she is born.

TrendsAverage health in most countries improvedin the twentieth century (chapter 3).Deaton (2004) documents that improve-ments in health are likely to have accom-panied economic growth, but he alsoemphasizes the globalization of knowl-edge, facilitated by local political, eco-

32 WORLD DEVELOPMENT REPORT 2006


nomic, and educational conditions. In the1980s and 1990s, however, progressslowed—a result of the worldwide HIV/AIDS epidemic and rises in cardiovascularmortality in Eastern Europe and formerSoviet Union countries.

How have inequalities in health evolvedwithin countries? Data from DHS providesome clues. For a subset of countries, mul-tiple rounds of DHS data are available todocument changes in infant mortality overtime. Of some 36 “spells” of health changethat could be identified, roughly 25 corre-sponded to improved health outcomes inthe form of lower infant mortality rates.Although overall health improved in these25 cases, the gaps between urban and ruralareas, between groups defined by mother’seducation, and between groups defined bydurable asset ownership did not univer-sally decline alongside the overall declines


It is hard to imagine people with fewer assets,through absolutely no fault of their own,thanAIDS orphans.Left to fend for themselves on thedeath of one or both parents from a progressivelydebilitating,heavily stigmatized,and costly-to-treat disease, their plight would be of concerneven if they numbered but a few. In southernAfrica,however, the United Nations Children’s Fund(UNICEF) estimates that there were 12.3 millionAIDS orphans in 2003,a veritable demographicgroup in their own right.By 2010,UNICEF projectsthat there will be 1.5 million AIDS orphans inSouth Africa; by 2014,1 million in Zambia.

An entire generation of Africans is emergingwho will have been raised, if they are lucky, bygrandparents or extended family members(themselves likely to be impoverished,overwhelmed, and suffering from the disease).At worst, they will grow up in child-headedhouseholds or in situations in which their basicrights to food, clothing, shelter, and adequatecare are routinely denied.

Wills and schoolingBeginning to overcome the huge disadvantagesthat AIDS orphans start life with requires specialattention on numerous fronts (box 7.11 consid-ers a variety of policy options). From a legalstandpoint, parents who know their death isimminent and who have young children need tobe encouraged (even if they are illiterate) to pre-pare enforceable wills that will protect the inher-itance rights of their children to ensure that sur-viving adults do not just forcibly take their land,

savings, or other valuables. From an educationstandpoint, it is vital to keep children in school,where the acquisition of even basic skills cangive them some viable prospect of being able tomove out of poverty.Where a child is the head ofa household and perhaps its sole income earner,however, the pressures to drop out of school areenormous. Numerous studies document signifi-cantly higher dropout rates of AIDS orphans. InKenya, one extreme example found that “52 per-cent of the children orphaned by AIDS were notin school, compared to 2 percent of non-orphans”(UNAIDS 2002, 135).

Attending school is also important from acivic perspective: it socializes children into thenorms and mores of society, and gives them theconfidence and capacity to participate morefully in it.Without such socialization, vulnerableyoung children are easy targets for those offer-ing them security and status through member-ship in a street gang, criminal network, or militiamovement. If AIDS orphans continue to stayfrom school at their current rate, comments onesenior U.N. official,“you will have a society wherekids haven’t been to school and therefore can’tfulfill even basic jobs...a society where a largeproportion can have antisocial instincts becausetheir lives have been so hard.You [will] have ageneration of children who will be more vulnera-ble to exploitation and to disease because theywon’t have the same sense of self-worth”(citedin Fleshman 2001, 1). Such children face the dis-mal prospect of failing to accumulate assetsbecause of the extreme burdens thrust on them

in their most formative years and the paucity ofopportunities available to them thereafter.

Avoiding infectionThe most immediate priority,however, is ensuringthat AIDS orphans do not themselves becomeinfected with the disease,thereby increasing thelikelihood that they will perpetuate the cycle.AIDSorphans face precisely such a risk,however,because the stigma of HIV/AIDS means that peopleoften assume that the children of parents who diedfrom AIDS must be infected,shunning,shaming,orexploiting them accordingly.Some AIDS orphanshave even been denied access to schools andhealth clinics because of the fear their verypresence generates.Children grieving the loss of aparent are also vulnerable to the sexual predationsof those putatively claiming to offer them comfort.Indeed,the desperation and apparent hopeless-ness of their circumstances—all the more so if itcoincides with a natural disaster such as drought—can drive AIDS orphans into prostitution.

The plight of AIDS orphans provides agraphic illustration of how cycles of disadvantagecan perpetuate themselves, and how social isola-tion and exclusion (especially at a young age) canpreclude the acquisition of assets and underminethe capacity to sustain participation in the insti-tutions that provide the best path out of poverty.

Sources: Avert.org (2004) http://www.avert.org/aidsorphans.htm. Accessed December 14, 2004.Fleshman (2001). Hargreaves and Glynn (2002),Lewis (2003), UNAIDS (2002), UNICEF (2003), USAID,UNAIDS, and UNICEF (2004).

B O X 2 . 2 Unequal assets, unequal opportunities: AIDS orphans in Southern Africa

Stunting and underweight(number of communes)

Stunting low/underweight low (260)Stunting low/underweight high (189)Stunting high/underweight low (365)Stunting high/underweight high (780)Incomplete data

2000 100 200

kilometers

Figure 2.4 Stunting and underweight in Cambodia

Source: Fujii (2005).


http://www.avert.org/aidsorphans.htm

http://www.avert.org/aidsorphans.htm

in infant mortality.5 The improvements inhealth were not necessarily shared across allgroups in the population.

As Cornia and Menchini (2005) note,mortality differentials across groups tendto narrow with an improvement of theaverage only if policies focus explicitly onequity. Without such a focus, improve-ments in the average may not translate todeclining group differences. For example,in the United States between the 1950s and

1990s, the overall decline in the infantmortality rate to 7.9 in 1994 was accompa-nied by an increase in the ratio of black towhite infant mortality rates from 1.6 in1950 to 2.2 in 1991. Inequality in healthdoes not inevitably fall as overall healthimproves, but such a virtuous process ispossible (box 2.3).

Inequalities in educationEducation is of great intrinsic importancewhen assessing inequalities of opportunity.It is also an important determinant of indi-viduals’ income, health (and that of theirchildren), and capacity to interact and com-municate with others. Inequalities in educa-tion thus contribute to inequalities in otherimportant dimensions of well-being.

Measuring inequality in education isnot easy. Census and survey data in mostcountries can generally yield statistics on,for example, years of schooling. But suchinformation does not capture well thequality of education and how that mightvary across individuals. Nor is it easy tocompare years of schooling across coun-tries, because those years might meansomething quite different from country tocountry.

Test results. Despite the measurement diffi-culties, there is considerable evidence ofinequalities of opportunity in education inthe developing world. Consider the differ-ences in test performance among Ecuado-rian children ages three to six years acrosspopulation groups defined by parentaleducation, region of residence, and wealth(box 2.4).

Test results among very young childrencapture well the inequality in opportunityin education, but such data are not readilyavailable for large numbers of developingcountries. So we look instead at the percent-age of household heads with no educationby gender and by urban-rural residence.

Male and female household heads. Theoverall percentage of household headswithout any education varies dramaticallyacross our sample of 60-odd countries(figure 2.5). In the high-income countries,the percentage rates are negligible. But at


Paxson and Schady (2004), drawing on mul-tiple rounds of DHS data, document thedeclining infant mortality rate in Perubetween the late 1970s and late 1990s. Ageneral downward trend exhibited a sharpsetback during the major economic crisisbetween 1988 and 1992, but resumed afterthe crisis.The downward trend remainedevident even after adjusting for age ofmother, recall period, education, and urbanstatus—indicating that the overall trenddecline in infant mortality was not attributa-ble only to general improvements in educa-tion, an aging population, or urbanization. Inaddition, the fact that infant mortality rosesharply around 1990, even after theseadjustments, supports the notion that thedecrease in household income and the col-lapse of public expenditures on health as aresult of the crisis were important.

Infant mortality rates in Peru variedmarkedly with the education level of themother in late 1970s and the 1980s (see fig-ure below). During the economic crisis,increases in mortality were largest amonginfants born to women with less education.After the crisis, the gap between infant mor-tality rates associated with different mater-nal education levels declined steadily, sug-gesting an overall decline in inequality inmortality alongside the decline in overallmortality rates.

There is some support for the view thatchanges in the amount and composition ofpublic expenditures on social programsdrove these improvements. Real totalexpenditures increased two and a half timesbetween 1991 and 2000, and such publicspending did not bypass the poor.

B O X 2 . 3 Health improvements and greater health equityin Peru

0.15

0.12

0.09

0.06

0.03

0.00

90th percentile or 11 years

75th percentile or 10 years50th percentile or 5 years25th percentile or 2 years10th percentile or 0 yearYears of maternal education

199819941990198619821978

Probability of dying in the first year

Adjusted infant mortality rates by maternal education

Source: Paxson and Schady (2004).


the other extreme, in Burkina Faso andMali, for example, the overall percentage ismore than 80 percent. What is similarlystriking is that, in most countries, the likeli-hood that the household head is uneducatedis dramatically higher than average whenshe is a woman. In the Laos People’s Demo-cratic Republic, for example, although theoverall percentage of household heads withno education is about 20 percent, the rate iscloser to 70 percent for female householdheads.

Rural and urban household heads. Simi-lar patterns can be observed for rural andurban areas (figure 2.6). In general, house-hold heads are far more likely to have noeducation when they are based in ruralareas than in urban areas. Even in coun-tries where the overall percentage withouteducation is very high, the rate in urbanareas can be dramatically lower. For exam-ple, in Burundi, the percentage of house-hold heads with no education in urbanareas compares with the national averagein Mexico, the Dominican Republic, andBrazil.

Access to teachers. A recent study of primaryschools and health clinics in Bangladesh,Ecuador, India, Indonesia, Peru, and Ugandahas identified teacher absenteeism as animportant, common, problem. The studyfound that higher income areas generallyhave lower teacher absentee rates thanpoorer areas.6 It also found that higher paidteachers, generally more educated and expe-rienced, appear equally or more likely to beabsent than contract or less remuneratedinstructors, perhaps because these instruc-tors sense a lower risk of being fired for theirabsence. And although salaries in rural areaswere often higher than in urban areas,teacher attendance in these areas was typi-cally lower than in urban areas. In most sur-veyed countries, the quality of infrastructureand the frequency of monitoring appeared tocontribute to lower absenteeism.

TrendsAnother way to assess inequalities ofopportunity in education is to calculate anoverall index of inequality for years of edu-

cation and to assess how much overallinequality of education can be attributed tomean differences between “morally irrele-vant” groups. Araujo, Ferreira, and Schady(2004) find that the inequality of adulteducation, measured by years of schoolingfor 124 countries, can be pronounced. Theyalso find that it is strongly (and inversely)correlated with mean years of schoolingacross countries.7

The data assembled by these authors alsoindicate that the inequality of education forspecific subgroups of the population canchange. While female schooling achieve-ments relative to male achievements weredramatically lower among the oldest cohorts,particularly in Sub-Saharan Africa, South


That education achievements varymarkedly by population groups—and thatthis can have profound implications—isbrought out forcefully in a recent study byPaxson and Schady (2005).They show thatcognitive development of Ecuadorian chil-dren ages three to six years, as measured bya test of vocabulary recognition (TVIP),varies significantly depending on thewealth of their household, their place of res-idence, the education of their mother, andthat of their father.The extent to whichthese circumstance variables are associatedwith performance on cognitive tests is typi-cally more pronounced for the olderchildren in their sample.

These socioeconomic characteristicsare significantly associated with cognitivedevelopment even after controlling forchild health and home environment. Theresearchers point to the striking evidencethat, in Ecuador, the youngest children,irrespective of wealth quintile or educa-tion of their parents, perform broadly aswell as their comparators. But as childrenin Ecuador get older, their cognitive devel-opment, relative to this benchmark, falterssignificantly. Only children in the top halfof the wealth distribution and with highlyeducated parents maintain their perform-ance relative to their comparators. By thetime they are six years old, most childrenin the sample are so far behind in theircognitive development that it is uncertainwhether and how they could ever catchup.

B O X 2 . 4 Child test scores in Ecuador: the role of wealth,parental education, and place of residence

40 60 7050

100

90

80

70

60

110

Wealthiest 25%

Median score

Median score

Poorest 25th%

40 60 7050

Age in months

Age in months

Wealthiest and poorest quartiles

Maternal education

100

90

80

70

60

110

12 or more years

0–5 years

Source: Paxson and Schady (2005).


Asia, and to a lesser extent the Middle Eastand North Africa, these disparities arenoticeably lower for the younger cohorts,particularly in Sub-Saharan Africa (figure2.7). Additionally, disparities in years ofschooling between urban and rural areashave been falling in some regions, moststrikingly in the Middle East and NorthAfrica and in Eastern Europe and CentralAsia. But in Sub-Saharan Africa there hasbeen little, if any, change. The (urban-rural)between-group contribution to inequalityin this region has hovered at around 30 per-cent across all the cohorts examined.

Economic inequalitiesAn individual’s consumption, his or herincome, or his or her wealth have all beenused as indicators of the command of an

individual over goods and services that canbe purchased in the market and that con-tribute directly to well-being. It is clear too,that individuals’ economic status can deter-mine and shape in many ways the opportu-nities they face to improve their situations.Economic well-being can also contribute toimproved education outcomes and betterhealth care. In turn, good health and goodeducation are typically important determi-nants of economic status.

An ideal measure of economic well-being for assessing inequality will capturean individual’s long-term economic status.But it is difficult to produce such a compre-hensive indicator accurately. In practice, it iscommon to work with measures of currentincome or consumption compiled fromhousehold survey data. While consumption


Percentage of household heads with no education

100

80

60

40

20

0

Unite

d Ki

ngdo

mCa

nada

Norw

ayUn

ited

Stat

es

Finla

ndGe

rman

ySw

eden

Kyrg

yzst

anKa

zakh

stan

Jam

aica

(*)

Tajik

istan

Rom

ania

Isra

elCh

ileM

oldo

vaPe

ruPa

ragu

aySr

i Lan

kaIta

lyTh

aila

ndGe

orgi

a (*

)Vi

etna

mAl

bani

a

Bosn

ia a

nd H

erze

govin

aCo

sta

Rica

Turk

eyIn

done

siaEc

uado

rPa

nam

aBo

livia

Colo

mbi

a

Repú

blic

a Bo

livar

iana

de

Vene

zuel

aEl

Sal

vado

r

Mex

ico

Dom

inic

an R

epub

licBr

azil

Lao

PDR

Pola

ndM

adag

asca

rTa

nzan

iaJo

rdan

Keny

aHo

ndur

asCa

mbo

dia

Nica

ragu

aCa

mer

oon

(*)

Guat

emal

aPa

kista

nYe

men

Côte

d’Iv

oire

Ghan

aHa

itiBa

ngla

desh

Tim

orBe

nin

Buru

ndi

Mor

occo

Guin

eaEt

hiop

iaM

aurit

ania

Mal

i

Burk

ina

Faso

(*)

Female

Male

Figure 2.5 Education levels vary across countries, but they also depend on gender of household head

Source: Authors’ calculations from household survey data.Note: The continuous dark line represents the percentage of household heads with no education in each country, while the endpoints of the whiskers indicate the percentages for male andfemale-headed households. * Indicates that female-headed households have higher average levels of education than male-headed households.



Percentage of household heads with no education

100

80

60

40

20

0

Cana

daUn

ited

Stat

es(*

)

Finla

ndSw

eden

Kaza

khst

anJa

mai

caTa

jikist

anRo

man

iaIs

rael

(*)

Chile

Mol

dova

Peru

Para

guay

Sri L

anka

(*)

Italy

Thai

land

Geor

gia

Viet

nam

Alba

nia

Bosn

ia a

nd H

erze

govin

aCo

sta

Rica

Turk

ey(*

)In

done

siaEc

uado

rPa

nam

aBo

livia

Colo

mbi

a

Repú

blic

a Bo

livar

iana

de

Vene

zuel

aEl

Sal

vado

r

Mex

ico

Dom

inic

an R

epub

licBr

azil

Lao

PDR

Pola

ndM

adag

asca

rTa

nzan

iaJo

rdan

Keny

aHo

ndur

asCa

mbo

dia

Nica

ragu

aCa

mer

oon

Guat

emal

aPa

kista

nYe

men

Côte

d’Iv

oire

Ghan

aHa

itiBa

ngla

desh

Tim

orBe

nin

Buru

ndi

Mor

occo

Guin

eaEt

hiop

iaM

aurit

ania

Mal

iBu

rkin

a Fa

so

Rural

Urban

Germ

any

Unite

d Ki

ngdo

m(*

)No

rway

(*)

Kyrg

yzst

an

Figure 2.6 Education levels vary by country and between rural and urban sectors

Source: Authors’ calculations from household survey data.Note: The continuous dark line represents the percentage of household heads with no education in each country, while the endpoints of the whiskers indicate the percentages for urban andrural households. * Indicates that rural households have higher average levels of education than urban households.

0.4

0.3

0.2

0.1

0.0

Between-group contribution to total inequality(proportion)

Other

East Asia

Europe andCentral Asia

South Asia

Latin America and Caribbean

Middle East andNorth Africa

Sub-Saharan Africa

1975–91970–41965–91960–41955–9Year of birth

1950–41945–91940–41935–9

Figure 2.7 The share of inequality in years of schooling attributable to differences betweenmales and females has been declining

Source: Araujo, Ferreira, and Schady (2004).

and income inequality are expected to cor-relate reasonably well with long-term well-being, it is unclear exactly how well theyactually do. And different measures of eco-nomic welfare—based on income, con-sumption, or wealth—can yield quite dif-ferent assessments of inequality (see alsobox 2.5).

For example, Sudjana and Mishra(2004), drawing on evidence produced byClaessens, Djankov, and Lang (2000), arguethat wealth inequality in Indonesia is farmore concentrated than suggested by com-parable figures based on consumption (fig-ure 2.8). In 1996 more than 57 percent ofthe stock market capitalization in Indonesiawas controlled by 10 families. This is instark contrast to neighboring countries,such as Singapore and Malaysia, but it is


only marginally higher than the figure forthe Philippines. More generally, Davies andShorrocks (2005) report estimates pub-lished by Merrill Lynch and Forbes thatsome 20 percent of the world’s millionairescome from the developing world. Similarly,Morck, Stangeland, and Yeung (2000) finda higher ratio of billionaire wealth to grossdomestic product (GDP) in Latin Americaand the Caribbean, and East Asia, but notIndia and South Africa (see chapters 6 and9 for further discussion). These figuresimply that the distribution of wealth may,on average, be more concentrated in devel-oping countries than in the developed.When wealth is associated with politicalinfluence, such inequalities also translateinto political capture and can provide awindow on this added dimension ofopportunity.

Bearing in mind the warnings offered inbox 2.5, figure 2.9 provides an approximate

picture of how economic inequality is dis-tributed across countries. The highest levelsof recorded inequality occur in Africa, thesecond highest in Latin America. But in-equality measures for Latin America comelargely from income data, while those inother regions, such as South Asia, comemainly from consumption data. As box 2.5illustrated, income data tend to producehigher measured inequality. Within regions,the data suggest that inequality can varymarkedly between countries: consumptioninequality in South Africa is extremely high,while in Mauritius it is lower even than inOECD countries.

How much overall economic inequalitywithin countries is attributable to differ-ences across population groups? Unlikehealth and education inequalities, the sys-tematic decomposition of income inequal-ity by population groups has long been sub-ject to analysis in the economics literature.


Because countries differ in their data collectionsystems, cross-country data on economicinequality are generally based on a variety ofindicators that are treated interchangeably.Thelack of a uniform basis for measuring economicinequality in different countries has seriousimplications for comparability.

One of the main sources of noncomparabil-ity of inequality is that some countries usehousehold income as indicator of well-beingwhile others use consumption expenditures(Atkinson and Brandolini 2001).These two indi-cators capture different aspects of economicwelfare, with the former perhaps seen better asa measure of welfare opportunity and the latteras a measure of welfare achievement. In mostcountries, measured inequality based onincome is higher than if it is based on consump-tion. But this is not inevitable, and the degree towhich the two indicators disagree varies fromcountry to country (see table to the right).

The problem of comparability is notconfined to the choice of welfare indicator. Animportant but underappreciated additionalissue is that, even for a given indicator, its defini-tion varies considerably across countries andeven within countries over time. Consumptioninequality based on different definitions of con-sumption can vary markedly, and will dependon a variety of factors, including the following:

• The length of the recall period over whichconsumption is recorded.

• The degree of disaggregation of consump-tion items.

• The methods for imputation of housing anddurables consumption.

Similarly, income inequality can vary dependingon whether income—

• Is intended to capture pre- or post-tax income,

• Includes actual and implicit transfers, and

• Refers to full income or earnings only.

Additional factors confounding comparabil-ity include differences in survey nonresponserates across countries (which are likely to affectmeasured inequality—see Korinek, Mistiaen,and Ravallion forthcoming). Differences acrosscountries in the availability of spatial priceindexes can also affect conclusions.Thomas(1987) demonstrates that adjusting for spatialprice variation can affect conclusions about thedegree of income or consumption inequality.Across countries there tends to be little unifor-mity in whether, and how, spatial price variationis accommodated.

Cross-country datasets on economicinequality generally incorporate some attemptsto improve comparability, but they typically fallfar short of achieving strict comparability.With-out a concerted effort to harmonize data collec-tion across countries, it is unlikely that suchglobal databases can be relied on to providemore than a tentative picture of differences ininequality across countries.

Inequality: summary measures in a selectionof countries: consumption versus income

Gini coefficient

Year Consumption Income

Panama 1997 0.468 0.621

Brazil 1996 0.497 0.596

Thailand 2000 0.428 0.523

Nicaragua 1998 0.417 0.534

Peru 1994 0.446 0.523

Morocco 1998 0.390 0.586

Vietnam 1998 0.362 0.489

Nepal 1996 0.366 0.513

Albania 1996 0.252 0.392

Bulgaria 1995 0.274 0.392

Russian Federation 1997 0.474 0.478

Bangladesh 2000 0.334 0.392

Source: Authors’ creation.

B O X 2 . 5 Beware of intercountry comparisons of inequality!

IndonesiaPhilippines

ThailandHong Kong, ChinaRepublic of Korea

SingaporeMalaysia

Taiwan, ChinaJapan

57.7%52.5%

46.2%32.1%

26.8%26.6%

24.8%18.4%

2.4%

Figure 2.8 Market capitalizationcontrolled by the top 10 families inselected countries, 1996

Source: Claessens, Djankov, and Lang (2000).


39

0.0 0.2 0.4 0.6 0.8

TurkmenistanGeorgiaEstoniaTurkey

FYR MacedoniaLatvia

MoldovaLithuaniaTajikistan

RussiaCroatiaAlbaniaPoland

KazakhstanKyrgyzstan

SloveniaBelarus

BulgariaRomania

SerbiaUzbekistan

ArmeniaUkraine

SlovakiaCzech Republic

Bosnia and HerzegovinaHungary

Azerbaijan

MalaysiaPapua New Guinea

SingaporePhilippinesCambodia

Thailand

China

Lao PDRMongoliaIndonesia

Rep. of KoreaVietnam

Japan

NamibiaBotswana

Central African RepublicSwaziland

LesothoSouth Africa

ZambiaMalawi

Gambia, TheZimbabwe

MadagascarCôte d’Ivoire

KenyaUganda

CameroonBurundiNigeria

Burkina FasoAngola

SenegalMozambique

MaliGhana

GuineaMauritania

BeninTanzania

NigerEthiopia

Mauritius Based on consumption

Sub-Saharan Africa

High-Income Economies

Latin America and the Caribbean

Middle East and North Africa

South Asia

East Asia and Pacific

Europe and Central Asia

0.0 0.2 0.4 0.6 0.8

Sri LankaNepalIndia

BangladeshPakistan

IranTunisia

MoroccoJordanAlgeria

IsraelEgypt

Yemen

BrazilBolivia

ParaguayColombiaEcuador

ChileArgentinaHonduras

El SalvadorMexico

Dominican RepublicPanama

GuatemalaCosta Rica

GuyanaSt. LuciaUruguay

R.B. de VenezuelaJamaica

NicaraguaTrinidad and Tobago

Peru

Hong Kong, ChinaPortugal

United StatesNew Zealand

GreeceUnited Kingdom

SpainCanada

AustraliaIreland

ItalySwitzerland

FranceLuxembourgNetherlands

AustriaGermanyNorway

DenmarkBelgiumSwedenFinland

Taiwan, China

Based on incomeBased on consumptionBased on income

Figure 2.9 Africa and Latin America have the world’s highest levels of inequalityIncome and expenditure Gini coefficients

Source: Authors’ calculations from household survey data.


These decomposition exercises seek tounderstand what share of inequality can beattributed to differences between groupsand what to inequality within groups. Thereare several attractions to studying certainpopulation groups in this way and to com-paring findings across countries.

Our interest here is to define groups bycircumstances we might consider “morallyirrelevant,” thereby gaining a window onthe importance of inequality of opportu-nity in the economic sphere. Additionally,decomposition results generally are far lesssensitive to differences in definitions ofunderlying welfare indicators than are

measured levels of inequality. In that sense,some of the difficulties with cross-countrycomparisons described in box 2.5 are atten-uated by subgroup decompositions.

Between-group shares of total inequalityWhile the “between-group” share of overallinequality is an appealing indicator of thesalience of differences across groups in theoverall assessment of inequality, there areconcerns about its interpretation.8 In par-ticular, empirical measures of between-group shares are generally found to bequite low (see figures 2.10 and 2.11).9 Theconventional presentation of between-group inequality is relative to total inequal-ity. Elbers and others (2005), however, notethat total inequality can be viewed as thebetween-group inequality that would beobserved if every household in the popula-tion constituted a separate group. Clearly,against such a benchmark, one wouldrarely observe a high share of between-group inequality.

Elbers and his colleagues propose analternative, comparing the actual between-group inequality with the maximum possi-ble inequality that would be obtained bykeeping the number of groups and their sizesat actual levels. For example, an assessmentof the contribution of gender differences toinequality compares actual between-genderinequality with the hypothetical between-gender inequality that would be obtained bysorting the income distribution so that allmales appeared at one end of the distribu-tion and all females at the other. This ratioprovides a measure of how far actualbetween-group inequality lies below themaximum between-group inequality that isfeasible given the existing configuration ofgroups.

Economic inequality can be decom-posed in a large sample of countries basedon several population breakdowns, two ofwhich are presented in figures 2.10 and2.11: social group and education of house-hold head. Such decompositions can followthe conventional decomposition method-ology, complemented by the Elbers andothers (2005) measure of feasible groupdecomposition.


0.10 0.2 0.3Proportion

0.4 0.5 0.6

ConventionalFeasible

NepalSri Lanka

Bangladesh

IsraelJordan

ParaguayGuatemala

BoliviaPanama

PeruBrazil

GuyanaNicaragua

St. Lucia

United StatesGermany

FranceLuxembourg

United KingdomCanadaBelgium

SwitzerlandAustralia

IrelandNorwaySwedenAustria

Finland

KyrgyzstanRomania

Vietnam

South Africa Madagascar

BeninCôte d’Ivoire

NigerGuinea

Sub-Saharan Africa

East Asia and Pacific


High-Income Economies



South Asia

Figure 2.10 Between-group inequality decompositions: social group of the household head



Different population breakdowns con-tribute to differing extents to overallinequality. In general, the conventional cal-culation of the between-group contribu-tion points to a fairly low share attributableto between-group differences. But in somecountries even the conventional share is

high. For example, in Paraguay, wheninequality is decomposed between groupsby language spoken at home, the conven-tional between-group share is approxi-mately 30 percent (figure 2.10). Andwhen inequality is decomposed for fivebroad education groups in Guatemala,


Proportion

Proportion

0.00 0.15 0.30 0.45

AzerbaijanTajikistan

ArmeniaUkraine

KazakhstanRussia

Bosnia & HerzegovinaMoldovaGeorgia

KyrgyzstanEstonia

MacedoniaAlbania

HungaryTurkeyPolandSerbia

RomaniaLithuania

VietnamLao PDR

PhilippinesEast Timor

Papua New GuineaIndonesia

Thailand

NigerEthiopia

MauritaniaNigeria

Burkina FasoTanzania

KenyaGuinea

MaliBenin

Côte d’IvoireUgandaBurundiSenegal

MadagascarCameroon

0.00 0.15 0.30 0.45

NepalPakistan

Sri LankaBangladesh

YemenIsrael

JordanMorocco

SurinameGuyana

St. LuciaTrinidad Tobago

R.B. de VenezuelaEl Salvador

Dominican RepublicUruguay

Costa RicaHonduras

BoliviaHaiti

JamaicaEcuador

ColombiaMexico

ParaguayChile

ArgentinaNicaragua

PeruPanama

BrazilGuatemala

BelgiumAustralia

AustriaCanadaNorwaySweden

SwitzerlandNetherlands

FinlandGermany

ItalyUnited Kingdom

IrelandUnited States

Taiwan, ChinaFrance

Luxembourg

Sub-Saharan Africa High-Income Economies



South Asia

East Asia and the Pacific




Figure 2.11 Between-group inequality decompositions: education of the household head



the between-group contribution is above40 percent (figure 2.11).

In most countries, the between-groupshare is noticeably higher for decomposi-tions based on the alternative, “feasible” cal-culation. Based on this approach, observedbetween-group differences are indeed sub-stantial in many countries—for the groupdefinitions here. To the extent that these cir-cumstances are judged “morally irrelevant,”the findings suggest that in economic life,just as in health and education, a substantialportion of observed inequality in manydeveloping countries can be linked toinequalities of opportunity.

Spatial differencesAs with inequalities in health, conventionalsurvey data cannot say much about the con-tribution of finely detailed spatial hetero-geneity to overall inequality—because of thelimited sample size. In an exercise analogousto that for health in Cambodia (figure 2.4), avariety of studies have applied statisticaltechniques to combine survey data withpopulation census data to produce tentativeestimates of inequality at the communityand district levels. Elbers and others (2004)document the contribution to overall esti-mated inequality of differences in meanconsumption for subdistricts in Ecuador,Madagascar, and Mozambique. Theydemonstrate that the between-subdistrictcontribution to total estimated inequalityranges from a low of 22 percent in Mozam-bique to more than 40 percent in Ecuador(table 2.1). Based on a similar approach,World Bank (2004e) reports between-com-mune differences in Morocco, accountingfor 40 percent of overall estimated con-sumption inequality. The general impres-sion is that spatial differences across locali-

ties account for a larger share of totalinequality as the number of localitiesincreases. The analysis confirms that forsome countries the spatial dimension ofinequality is of considerable importance.This conclusion carries over even morepowerfully at the global level, where thebetween-country contribution to globalinequality is dramatic (chapter 3).

Other studies and methodologies cor-roborate the finding that spatial differenceswithin countries are important. Using farm-household data for rural China, Jalan andRavallion (1997) identify “spatial povertytraps,” where poorer areas have lower provi-sions of essential public goods (such asroads) and, as a result, households in thearea experience lower productivity on theirinvestments. Various studies find spatialeffects on living standards, even after con-trolling for nongeographic household char-acteristics. Ravallion and Wodon (1999)demonstrate that place of residence is animportant determinant of poverty inBangladesh. They also note that importantspatial differences can be discerned evenwithin urban areas—households in the dis-trict of Dhaka are markedly better off thantheir counterparts in other urban districts.

Many studies suggest that spatial differ-ences in incomes are driven by policy. InChina, Kanbur and Zhang (2001) find ameasurable polarization between inland andcoastal regions where factors unrelated tophysical geography—development of heavyindustry in certain provinces, trade open-ness, and government investment in coastalregions—are associated with wideninginterregional inequality. Escobal and Torero(2003) compare coastal Peru with the high-lands and find that average per capita expen-ditures vary markedly and that this varianceis associated with fewer and weaker infra-structure services in the highlands.

The role of infrastructure is thus central.Although it is not disputed that physical geog-raphy can also influence poverty directly, theassociation between geographic variation inpoverty and geographic variation in infra-structure access is typically strong. Accord-ingly, it is argued that the influence ofregional geographic location on inequalitywill diminish as access to transport and


Table 2.1 Decomposition of inequality between and within communities

Level of Number of Within-group inequality Between-group inequality decomposition communities (percent) (percent)

Ecuador 1,579 58.8 41.2

Madagascar 1,248 74.6 25.4

Mozambique 424 78.0 22.0

Source: Elbers and others (2004).Note: Our communities in Ecuador are zonas in urban areas and parroquias in rural areas. Communities in Madagas-car are firiasana (communes) and in Mozambique they are administrative posts. The decompositions are performedusing the conventional methodology.


communications services improve; beinggeographically isolated will matter lessbecause infrastructure improvements willhelp compensate for distance.10

The relationship between groupdifferences and inequalityAs is clear from the discussion here, ourinterest in the contribution of group differ-ences to total inequality extends beyondnormative considerations of fairness andjustice. Differences between groups are alsothought to explain overall inequality out-comes, particularly the reproduction ofinequalities over time. The basic idea is thatbetween-group differences in incomeinequality, for example, will tend also to bemirrored in between-group differences inhealth and education inequalities—and inthe agency of groups in influencing theircircumstances (see below). These group dif-ferences will then reinforce one another.Group differences in education, for exam-ple, will translate into differences inincomes and in political voice and partici-pation. These inequalities will, in turn,affect health inequalities between groups,which are passed on to education inequali-ties and so on. “Inequality traps” are theresult. A corollary of this idea is that effortsto moderate overall inequality levels mightrequire a focus on reducing between-groupdifferences.

It is difficult to systematically documentthis instrumental role of group differences.Figure 2.12 illustrates one attempt. Overallinequality is correlated with the between-group share for the sample of countries infigures 2.10 and 2.11, controlling for regionand whether the underlying welfare indica-tor is income or consumption. Nothing inthe mechanics of the calculation forcesoverall inequality to be correlated with theshare attributable to between-group differ-ences. Yet, for this sample of countries,higher overall inequality is associated with alarger between-group share of overallinequality, which is attributable to therural-urban breakdown, to differencesacross social groups, to differences in edu-cation, and (weakly) to differences in broadoccupation class of the household head.11

One interpretation of these findings isthat between-group differences account for,and possibly explain, a non-negligible por-tion of overall inequality. This is consistentwith the broader theme of this report: thatgroup differences reinforce one anotherand in this way contribute to the replica-tion of inequality over time. But these sim-ple correlations, while suggestive, couldalso be pointing to other processes and ontheir own cannot exclude other competingexplanations.

Inequality and growth, economicstructure, and tradeSystematic exploration of the impact ofbetween-group shares on overall inequalityhas not, to date, been a major topic ofempirical investigation. A longer-standingquestion in economics has been howinequality evolves with economic growthmore generally. Pioneering work by Kuznetsin the 1950s launched an enormousamount of empirical work on this question,stimulating much debate. There is still no


Urban-rural

Overall inequality

Occupation of the household head

Between-group share

Overall inequality

Social group of the household head

Overall inequality

Education of the household head

Between-group share

Between-group share Between-group share

Overall inequality

Figure 2.12 Location, education, and social groups can make a difference: regressions of totalinequality on shares of between-group inequality of different household characteristics

Source: Authors’ calculations from household survey data.Note: Regressions include as controls (X) regional area dummies and a welfare measure (Y/C) dummy. The shares ofthe between component of inequality across gender and age of the household head, and regions within the countrywere not significant.


consensus on a systematic relationshipbetween the long-term growth processes ofindustrialization and urbanization—andoverall inequality (box 2.6).

Cross-country studies have also analyzedthe relationship between inequality and eco-nomic structure. Bourguignon and Morri-son (1990), for example, argue that “develop-ing countries which are comparativelyendowed with mineral resources and land(climate) tend to be less egalitarian than oth-ers, although the effect of the agriculturalcomparative advantage may be offset by thedistribution of land.” They also find that thelabor productivity difference between agri-culture and the rest of the economy is a pow-erful explanatory factor for differences inincome inequality in a number of developingcountries in the 1970s and 1980s.12

A large body of literature has alsoexplored the relationship between tradeopenness and inequality but has not reacheda consensus. For example, Dollar and Kraay(2002) and Dollar and Kraay (2004) find noeffect of trade openness on inequality, butLundberg and Squire (2003) do find suchan effect. Ravallion (2001) and Milanovic(2002) report that at low incomes opennessmay be inequality-increasing, but that thiseffect reverses at higher incomes.

TrendsThe discussion above highlights the manymechanisms for hypothesizing how aggre-gate economic growth, and the evolution ofdifferent sectors of the economy, can influ-ence economic inequality. Popular lines ofargument have emphasized Lewis-Kuznets


The starting point of the literature linking eco-nomic development and income inequalitydates to the well-known works of two NobelPrize winners, W. Arthur Lewis (1954) and SimonKuznets (1955). Lewis, in his classic 1954 article“Economic Development with Unlimited Sup-plies of Labor,” developed a theoretical model inwhich growth and accumulation in a dual econ-omy would start in the modern industrial sector,where capitalists would hire at a given wageand reinvest a share of their profits.The numberof traditional agricultural laborers willing tomove to this high-productivity, high-wage sec-tor was assumed to be unlimited. In this processof development, and as long as these assump-tions would prevail, inequality in thedistribution of income would increase as aver-age incomes rose.There would be a turningpoint after which inequality would fall again asthe surplus labor phase ends and the dualisticeconomy becomes a single-sector, fully industri-alized economy.

Although Kuznets did not explicitly modelthe intersectoral shifts of population as part ofthe development process, he did build on themto articulate his basic idea of an inverted-U rela-tionship between economic growth and incomeinequality (the “Kuznets curve”). In his presiden-tial address at the Annual Meeting of the Ameri-can Economic Association in 1954, he hypothe-sized that in the process of growth andindustrialization, inequality would first increase,because of the shift from agriculture and thecountryside to industry and the city, and thendecrease as returns across sectors equalized.Thedata Kuznets used to make this statement came

from a long-run series of inequality indicatorsfor England, Germany, and the United States,and from a single observation in time for threedeveloping countries—India, Ceylon (Sri Lankatoday), and Puerto Rico.These were the dataavailable at that time, and Kuznets was wellaware of the limitations of the empirical backingof his argument, in his own words, on “5 percentof empirical information and 95 percent specu-lation, some of it possibly tainted by wishfulthinking.”

Kuznets based his speculation primarily onlongitudinal data and called for in-depth casestudies of the economic growth of nations. Butmany subsequent studies simply usedaggregate cross-country data (often of not par-ticularly high quality) and reduced-form modelsto explore and support the hypothesis of aninevitable tradeoff between development andequality.The Kuznets curve became one of themost quoted stylized facts of the study ofincome distribution for nearly four decades.

Cross-country data can be misleading for dynamic processesWith the development of much larger data sets,such as the Deininger and Squire (1996) interna-tional inequality database (following on fromFields 1989), empirical “tests” of the Kuznetscurve were widely conducted. But it has becomeunderstood that the use of cross-country datato analyze what are essentially dynamicprocesses can be strongly misleading. Moreover,numerous studies have shown that theevidence in favor of the Kuznets curve is not atall robust to econometric specifications, sample

composition, and period of observation. See,among others, Bourguignon and Morrisson(1989), Fields and Jakubson (1994), Deiningerand Squire (1998), and Bruno, Ravallion, andSquire (1998). Bruno, Ravallion, and Squire(1998), while drawing in part on cross-countrydata, also analyzed one country—India—forwhich relatively long time-series data hadbecome available, and again found no sign thatgrowth increased inequality.

Why the Kuznets curve does not hold in prac-tice probably has to do with the fact that devel-oping countries do not generally satisfy theassumptions on migration processes and sectoraldevelopment underlying the Kuznets hypothesis.To explain international differences in inequalityof incomes, it is important that the link betweeneconomic inequalities and other factors, such aseconomic dualism, land, education, and regionaldifferences, be more carefully analyzed.

No straightforward relationship betweenincome and inequalityTo conclude, there is today something of a con-sensus that no straightforward relation betweenincome and inequality can be established. Asargued by Kanbur (2000) in his exhaustivereview of the Kuznets curve literature in theHandbook of Income Distribution: “it seems to usfar better to focus directly on policies, or combi-nation of policies, which will generate growthwithout adverse distributional effects, ratherthan rely on the existence or nonexistence of anaggregative, reduced form relationship betweenper capita income and inequality.”

Source: Authors’ creation.

B O X 2 . 6 Revisiting the Kuznets hypothesis for economic growth and inequality


type processes, the race between relative sup-ply and demand for skills along with house-hold adjustments to participation, educa-tion, and fertility; the transitions fromcontrolled to market-oriented economicsystems; and various forms of power andbargaining-related views of the world. Inthe end, and perhaps not surprisingly, it isdifficult to identify a single overarchingexplanation. Until recently, this did notseem to matter much because there was ageneral perception that inequality does notvary markedly over short periods.13 In ear-lier studies, few countries having data oninequality over multiple time periods indi-cated sharp changes.

For countries and regions. Empirical investi-gation of how inequality evolves in a countryis subject to concerns similar to those forcomparisons of levels (see box 2.5). But thereis a growing sense that the impression of sta-ble, unchanging income inequality may wellbe misleading. A few recent examples ofchanging inequality bear mentioning. First,careful work by Atkinson (2003) has docu-mented the evolution of inequality in OECDcountries during the second half the twenti-eth century. He finds that inequality in theUnited States has been rising steadily sincethe early 1970s (after seeing little change, andpossibly some decline, in the precedingdecades) and has risen dramatically in theUnited Kingdom since 1980. Between 1984and 1990, the Gini coefficient in the UnitedKingdom rose by 10 percentage points (butthen did not increase further)—an unprece-dented increase over such a short time. Else-where in the OECD, inequality changes havebeen less marked. But to the extent that theearly and middle decades of the twentiethcentury were associated with declininginequality in these countries, this trendseems to have halted by the century’s laterdecades.

Second, inequality in China wasmarkedly higher at the end of the 1990sthan it had been in the early part of the1980s. In general, the recent evidence inEast Asia suggests that inequality has risenfaster in the second round of high growthAsian economies—such as China and Viet-nam—than had been observed in the first

round—Hong Kong (China), Republic ofKorea, Malaysia, Singapore, and Taiwan(China). A complete picture of the factorsbehind this process is as yet unclear.Although it is likely that at least part of thestory is linked to intersectoral transfers, asemphasized by Lewis (box 2.6), Ravallionand Chen (2004) indicate that inequality inChina grew fastest during periods wheneconomic growth and poverty reductionwere slow. They argue that China provideslittle support for the view that risinginequality is inevitable with rapid economicgrowth and poverty reduction.

Third, South Asia has generally beenperceived as a region with relatively lowinequality. This probably is due, in part, toinequality being measured by consump-tion. In this region, too, the prevailing viewhas been that inequality changes little overtime. But the stylized fact of low and stableinequality in South Asia has also been chal-lenged. In India, the largest country in theregion, some uncertainty remains over howinequality has evolved, because of well-publicized issues concerning data compa-rability over time.14 The best available esti-mates suggest that inequality in India hasbeen rising, but with no solid assessment ofby how much.15

In Bangladesh, Nepal, and Sri Lanka,however, recent and reliable data show verylarge increases of inequality in the late1980s and 1990s. In Bangladesh, incomeinequality (as opposed to consumptioninequality) has been documented to haverisen from a Gini of 0.30 to 0.41 between1991 and 2000.16 In Sri Lanka, the increasein consumption inequality has been verysimilar, from 0.32 to 0.40 between 1990 and2002.17 And, in Nepal, the Planning Com-mission has produced estimates suggestingthat consumption inequality rose from 0.34to 0.39 between 1995–6 and 2003–4.18 Onlyin Pakistan is the evolution of inequalitynot clear, because of difficulties with datacomparability.

In other regions of the world, the recentpicture on inequality trends is more diffi-cult to summarize. For Latin America, DeFerranti and others (2004) indicate thatinequality increased in most countries, by asizable margin, during the “lost decade” of



the 1980s. But during the 1990s, inequalitycontinued to rise in only about half of thecountries in the region, and less rapidly. Theauthors note that, in Argentina, inequalityhas risen sharply in the growth period andduring the crisis years. In Brazil and Mex-ico, the 1990s witnessed some smalldeclines. In Eastern Europe and CentralAsia, changes in inequality during the early1990s, associated with the transition to themarket economy, have been difficult to doc-ument systematically because of data prob-lems, according to World Bank (2000c).Between 1998 and 2003, consumptioninequality declined in the former SovietUnion countries (with the exception ofGeorgia and Tajikistan), while there was noclear trend in eastern and southern Euro-pean countries (World Bank, 2005a). InAfrica and the Middle East, it is difficult topoint to broad trends, largely because ofconcerns with data comparability overtime.

To what extent does our examination oflevels and trends in income inequality bearon the themes of this report? This report ismost concerned about changes in inequali-ties in incomes, and other specific dimen-sions, if these dimensions are associatedwith changes in underlying inequalities ofopportunities. Rising income inequality inRussia during the 1990s, for example, is ofconcern precisely because of its strong asso-ciation with rising political influence andstate capture.

But this is not inevitably the case. Arecent study of income distribution dynam-ics in six East Asian and Latin Americancountries by Bourguignon, Ferreira, andLustig (2005) decomposes income distri-bution dynamics into the underlying driv-ing forces. They show that complex andcountry-specific interactions betweenpowerful underlying social and economicphenomena imply that distributional expe-riences must be assessed country by country.For example, improvements in education(equalizing opportunities) may be associ-ated in one case with falling income inequal-ity—Brazil or Taiwan, China—but inanother country with rising inequality—Indonesia or Mexico. Our assessment ofthe equity implications of changes in

income inequality will thus differ acrosscountries.

Across generations. Our assessment will alsodepend on the degree to which inequalitiesare transmitted across generations. Thestudy of intergenerational transmission ofwelfare is not straightforward, because ofthe scarcity of datasets containing informa-tion on various generations of adults in thesame family. Data from long panels are rare,and questions about family background ofindividuals are not always asked in surveys(the Brazil data described in box 2.1 are arare exception). Information about educa-tion or occupation for various generationscan be captured relatively easily in recallquestionnaires. But information aboutother dimensions, such as the incomes,earnings, or even health status of earliergenerations, is not easily remembered byindividuals (not least because they oftenchange during a lifetime). The scarcity ofintergenerational data is particularly strik-ing in developing countries. Even thoughthe persistence of inequalities across gener-ations is often thought to be much moreacute in developing countries, studies onintergenerational mobility in the develop-ing world remain few and far between.

Even when the data exist, differences inmethodologies and data often limit thescope for comparisons across countries.The most widespread measure of intergen-erational mobility in the economics litera-ture is the intergenerational earnings elas-ticity, or the elasticity of sons’ earningswith the earnings of their parents. Thismeasure generally comes from a log-linearregression of sons’ earnings (although itcould also be income or years of schooling)on fathers’ observed earnings (or its pre-dicted value using such other informationas education or occupation). The closer theelasticity is to zero, the more mobile thesociety is supposed to be. This elasticity hasbeen widely used in the U.S. literature,where longitudinal data are relativelyabundant. And for comparability, it hasalso been calculated in most other coun-tries’ recent studies.19

Until recently, estimates of the intergener-ational elasticity of earnings were thought to



be around 0.4 in the United States, suggest-ing a reasonably mobile society in incomes.20

More recently, however, Mazumder (2005)uses new data and recent econometric tech-niques to correct for transitory fluctuationsin earnings—he shows that the previous esti-mates of intergenerational elasticity werebiased downward by about 30 percent. Heargues that the true estimate is somewherearound 0.6 for the United States.

An intergenerational elasticity of 0.6 com-pared to 0.4 paints a dramatically differentpicture of mobility in American society. Forexample, it implies that a family whose earn-ings are half the national average wouldrequire five generations instead of threebefore it substantially closed the gap. Obvi-ously a difference of two generations, orabout fifty years, is quite substantial and sug-gests the need to examine policies that fostergreater mobility.21

In parallel analyses, estimates of inter-generational mobility in Canada, Finland,or Sweden, among others, have tended toreport elasticities closer to 0.2 or lower, sug-gesting that these societies are considerablymore mobile than the United States. A rela-tively early study of mobility in the UnitedKingdom (Atkinson, Maynard, and Trinder1983) reports an elasticity of 0.43, while amore recent study by Dearden, Machin, andReed (1997) estimates an elasticity of 0.57.These studies indicate that people in theUnited Kingdom are about as mobile asthose in the United States. Because of thedata limitations, only a few exceptionalstudies on intergenerational earnings elas-ticities for less-developed countries havebeen carried out. These provide evidence ofrelatively low mobility.22

In another literature review of cross-country differences in intergenerationalearnings mobility, Solon (2002) askswhether there is any link between cross-sectional inequality within a generationand the intergenerational transmission ofinequality. Although there is greater cross-sectional inequality in the United Statesand the United Kingdom than in Swedenor Finland, Canada also has relatively highinequality. The evidence needed to pro-vide a clear answer to this question istherefore still fragmentary, and only “con-

tinuing research (on international evi-dence of intergenerational mobility) willimprove our understanding of why theintergenerational transmission of eco-nomic status is strong in some countriesand weak in others.”23

The intergenerational transmission mech-anisms of inequalities will differ across coun-tries and within countries across differentpopulation groups. As described above,Mazumder (2005) points to rather low lev-els of intergenerational mobility in theUnited States. He also highlights an impor-tant racial dimension to this limited mobil-ity and finds evidence of substantial immo-bility at the ends of the distribution. Heshows that of the individuals whose fatherswere in the bottom decile of the earningsdistribution, 50 percent will be below thethirtieth percentile and 80 percent belowthe sixtieth percentile. He finds the evidenceto be consistent with the hypothesis thatsuch immobility “might be due to theinability of families to invest in their chil-dren’s human capital due to the lack ofresources.” By contrast, more than 50 per-cent of the individuals whose parents werein the top decile will remain above theeightieth percentile and two-thirds will beabove the median.

In another U.S. study, Hertz (2005) con-firms the findings of Mazumder (and oth-ers) on the size of the intergenerationalelasticity. He then shows evidence that it islargely driven by the especially low rate ofmobility of black families from the bottomof the income distribution. While only 17percent of whites born to the bottom decileof family income remain there as adults, thecorresponding figure is 42 percent forblacks. He also finds that “rags-to-riches”transitions from the bottom quartile to thetop were less than half as likely for black asfor white families. He further provides evi-dence that the black-white mobility gap isnot “appreciably altered by controlling forparents’ years of schooling.” Last, he pro-vides evidence that the incomes of blackchildren are unresponsive to small changesin parents’ incomes at the bottom of thedistribution.

To recap, summary measures suggest thateven in such developed countries as the



United States and United Kingdom there israther limited intergenerational mobilityacross generations. Research in these coun-tries has highlighted important hetero-geneities in the patterns of reproduction ofdifferent inequalities across populationsgroups. For most developing countries, rela-tively little is known about intergenerationalincome mobility. But given the acute group-based inequalities in many developing coun-tries, there appears to be little basis forexpecting much intergenerational mobility.

Agency and equity:inequalities of powerThe foregoing discussion has raised explic-itly the question of how inequalities aredetermined and reproduced. It has pointedto the potentially important role of groupdifferences in this process. This focus onprocess and the factors that account for thepersistence of inequality over time puts thespotlight on how much inequality is rootedin deeper institutions in society—institu-tions of governance, access to land, controlof labor, market regulation. Chapter 6 dealswith the emergence and effects of suchinstitutions in more detail. Here we turn todifferent kinds of evidence—and tradi-tions of analysis—to discuss the unequalcapacity of people to influence the formtaken by these institutions and the conse-quences of unequal institutions for contin-uing inequality in such capacities. For

poverty the inequalities in capacity to forgethe institution or society can be as impor-tant as inequalities in health, income, andeducation.24

A recent study of inequalities in gover-nance in four slums of Delhi found thataccess to formal government by slumdwellers is more available to the better offand to those who have good contact net-works.25 Community leaders in these slumsfacilitate access primarily to their castemembers, and slum dwellers are more likelyto delegate custodianship of their intereststo better-educated community leaders. Thestudy concludes that because access tobureaucracy and political representation forslum dwellers in Delhi is largely the pre-serve of the better off and better connected,decisions of formal policymakers do notseek to represent slum interests as a whole,producing interventions that do not targetthose in most need. The lack of broadly dis-tributed “voice” thus results in patterns ofresource allocation, and income generation,that are far from egalitarian.

The nature of this unequal capacity canbe captured through the sociological con-cept of agency. Agency refers to people’scapacity to transform or reproduce suchsocietal institutions. Some of this capacity isconscious—for example, when interestgroups lobby for a change in land tenurelegislation, or when women refuse to acceptlaws around marriage that systematicallydisadvantage them. Some of it is uncon-


A recently completed Human DevelopmentReport for Pakistan provides rich documenta-tion of the skewed distributional impact of cor-ruption (United Nations DevelopmentProgramme 2003).The report notes that corrup-tion raises the costs of getting things done—forsetting up a new business, for crossing borders,for obtaining a driver’s license. In Pakistan thesecosts fall most heavily on those least able toafford them: the poor. According to the PakistanHuman Development Report, 16.7 percent ofthe extremely poor reported paying a bribe torun their business enterprise, handing over anaverage of 6,800 rupees. Only 6.7 percent of thenon-poor paid a bribe, of 9,300 rupees. In ruralareas, the contrast is even starker: 20 percent of

the extremely poor paid a bribe, while only 4.3percent of the non-poor had to do so. In urbanareas, the extremely poor paid on average 8,700rupees in bribes, while the non-poor paid only1,200 rupees.

Similar patterns emerge for mediating dis-putes.The extremely poor not only pay a higherprice to seek a resolution than the non-poor, butalso they are less likely to receive a satisfactoryoutcome (38.5 percent versus 80.8 percent).Indeed, the fee the extremely poor must pay isoften higher than their annual householdincome, leaving many to choose to suffer theconsequences of a dispute even when they areclearly in the right. In addition, the extremelypoor receive less assistance from the police (the

most immediate representatives of the formaljustice system), who are involved in only 1 per-cent of their disputes but nearly 5 percent of thedisputes of the non-poor.The poor perceive thatthe police will be slow and inefficient inhandling their cases, and they frequently experi-ence outright harassment and intimidation.Even to register a case of kidnapping with thepolice requires paying a bribe. In thesesituations, it is hardly surprising that the poorfind it more expedient to take the law into theirown hands, creating in many urban areas a hostof new problems related to gang violence andvigilantism.

Source: United Nations Development Programme(2003).

B O X 2 . 7 Inequitable agencies and institutions in Pakistan


scious—for example, when people engagein land transactions without questioningthem, they reproduce the institutions ofland tenure and the markets in land. Whena disadvantaged group accepts its disadvan-tage as “taken for granted,” the effect is toallow the continuing existence of the rela-tionships that create such disadvantage.

The internalization of disadvantageleads to pernicious forms of agency thatperpetuate inequalities. From inequalitiesin agency come inequalities in power, voice,and self-confidence—a major part of ourstory (box 2.7). Inequalities of agency are asmuch products of dominant institutions assources of those institutional arrangements.Maintaining these arrangements bothreflects and produces the distribution ofpower among people. As for health, educa-tion, and income, though, this distributioncan change—and it has. Indeed, it has oftenchanged in relation to changes in theseother distributions.

Internalization of disadvantage and inequalities of agencyRecent work on urban slum dwellers inIndia26 (and elsewhere)27 suggests that a keyform of powerlessness for the poor involvesliving with “negative terms of recognition.”This concept highlights the conditions andconstraints under which the poor negotiate

the very norms that frame their social lives.Being so routinely treated with contempt bygovernment officials, employers, and fellowcitizens—and encountering such enormousobstacles to advancement—means thatexcluded groups can, over time, come tosubscribe to norms about themselves andtheir situation “whose social effect is to fur-ther diminish their dignity, exacerbate theirinequality, and deepen their lack of accessto material goods and services.”28

In these circumstances, the poor are notonly persistently and overtly discriminatedagainst. Their problems are further com-pounded and consolidated by their apparentcomplicity in it, their revealed “adaptivepreference”29 for menial occupations andascription to norms and subservient behav-iors that only legitimize and perpetuate theirpowerlessness. Dire material circumstances,rational expectations about their limitedprospects for upward mobility, and strongbeliefs about the legitimacy and immutabil-ity of their situation conspire to create avicious circle from which it may be very dif-ficult for the poor to escape (see box 2.8).30

Inequality traps may cause crime andviolence. First, people who perceive theirpoverty as permanent may be driven byhostile impulses rather than rational pur-suit of their interests. Second, sensitivityto inequality, especially by those feeling


The Batwa, who are described in many parts aspygmy peoples, live in Eastern Uganda, easternDemocratic Republic of Congo, and Rwanda.Batwa have been subject to negative stereotypessince at least 1751, when Edward Tysonconcluded that pygmies were not human butrather apes or monkeys.They suffer multipleasset depletion and wide ranging forms of dis-crimination, a situation that public actions haveat times made worse.Though longstanding forestdwellers, the British sought to expel them to cre-ate forest reserves in the 1930s. In 1991, theUganda National Park authorities increasedefforts to enforce this exclusion from forest areas.Although the World Bank—which was fundingsome of the park authorities’work—required thatthe government assess the impact on indigenouscommunities and follow defined compensationprocedures, these did not take sufficient account

of power differences among Batwa and otheraffected groups, nor consider Batwa preferences.All communities were viewed as uniform, a prac-tice that the authorities later recognized “did nottake into account Batwa realities and left themwith nothing”(Zaninka (2003), 170).

Non-Batwa locals have resisted efforts to pro-vide more appropriate compensation to theBatwa. A Participatory Poverty Assessment (PPA)highlighted persistent discrimination, describingthe Batwa as a “group of people who aredespised”and who “have no means of productionsuch as land, credit and training.They areregarded by other ethnic groups in Kisoro as apeople with no rights.”This leads to everyday andinstitutionalized forms of exclusion, with theBatwa suffering discrimination in access to bothpublic spaces and services.While some Batwarespond to this by organizing themselves, others

respond in ways that—however rational and self-protecting—often reproduce the extent to whichthey are excluded.The same PPA reports someBatwa children saying that they did not attendschool because it was so unfriendly to them.When asked what they wanted to do upon com-pleting school, one child replied that she wishedto be “a cleaner.”Discrimination and prejudicediminish the capacity to aspire to and imagine adifferent future.

Repudiation and discrimination can alsolead the Batwa to self-exclude from the publicsphere.The PPA notes that no Batwa attendedPPA exercises. Non-Batwa locals explained:“Batwa would never come to such meetings, sothere is no point in mobilizing them.”

Source: Moncrieffe (2005), citing ParticipatoryPoverty Assessment reports.

B O X 2 . 8 Legacies of discrimination and the reproduction of inequalities and poverty among theBatwa in Uganda


trapped at the bottom, may lead to higher-risk tactics like crime, when the expectedpayoffs from socially legitimate activitiesare poor. Third, people may be particularlysensitive to group-based inequalities. If, forexample, racial heterogeneity and incomeinequality are correlated and consolidatestatus distinctions in a society, this couldspell potential for violence. Finally, as Mer-ton (1938) elegantly states,

. . . when a system of cultural values empha-sizes, virtually above all else, certain commonsymbols of success for the population at largewhile its social structure rigorously restrictsor completely eliminates access to approvedmodes of acquiring these symbols for a con-siderable part of the same population, . . . anti-social behavior ensues on a considerablescale. (italics reflect original emphasis)

A lack of upward mobility in a society, com-bined with a high premium on economicaffluence, results in anomie—a breakdown ofstandards and values.31

Changing between-group inequalitiesof agency and institutional powerInequality of agency often leads to institu-tions that reproduce such inequality. Butthese relationships are not immutable.There are ample cases in which interven-tions—by civil society, reformist publicofficials, external actors, religious institu-tions, and others—have given more self-confidence and assertiveness to disadvantagedgroups, worked against the internalization ofdisadvantage, and created new channels forexcluded groups to exercise voice with greatereffect. These changes improve the terms ofrecognition for the powerless: they becomerecognized by more powerful groups whootherwise would not acknowledge them at all,leading to empowerment of disadvantagedgroups in economic, social, and politicalrealms.

Empowerment can occur in many ways.32

Change typically occurs through the interac-tion between the opportunities for actioncreated by dominant political structures andthe capacity of poorer or middle groups toengage. The “political opportunity struc-ture”—that shapes the possibilities foraction—is itself a function of the openness of political institution, the coherence and

positions of elites, and the effectiveness ofgovernments to implement approved coursesof action. The capacity of subordinate groupsis influenced by their “economic” capital—their education and economic resources—their “capacity to aspire,” and the closely asso-ciated capacity to organize.33

In Indonesia, the Kecamatan Develop-ment Project (KDP) illustrates changeoccurring through action from above andbelow: it aims to improve the terms of recog-nition and the political agency of marginalgroups, and to create new institutions forgreater agency to lead to material changes inpatterns of public investment. Consistentwith the ongoing process of democratizationin Indonesia, the source of change comesfrom public policy rather than nongovern-mental action, allowing the project to oper-ate on a large scale (see also focus 4 for exam-ples of change occurring at the local level).

A recent study34 of the efficacy of theKDP on challenging and changing theterms of recognition of participants sug-gests that it does provide villagers with a setof deliberative routines for more equitablymanaging the conflicts it inevitably trig-gers.35 These routines introduce marginalactors to more equitable spaces of engage-ment with more organized and influentialactors. But building this conflict manage-ment capacity among marginal groupsdepends on more than just forging collabo-rative routines. It also requires a set ofrules—defined by the KDP—that limit theunfair exercise of power by dominantgroups. With the KDP cultivating collabo-ration and tangible points of politicalpower for marginalized groups, the resultsinclude a well-functioning school or med-ical clinic but equally important a style ofgroup (re)definition and defense.

Changes in the agency of indigenouspeoples in Ecuador since the 1960s provideanother example in which mobilizationfrom below came to change national andlocal structures. These changes are clear atboth local and national levels. In the 1960sin the Andean province of Chimborazo,the indigenous Quichua people sufferedmultiple deprivations. They were subjectto everyday forms of violence and to dom-ination and racism in their interactions



with other ethnic groups and with author-ities.36 Power was concentrated in the tri-umvirate of landowner, priest, and localgovernment authority. Much indigenouslabor was tied to large rural estates onwhich labor relations were sometimes vio-lent and returns to labor manifestly unfair.Life expectancy was short, alcoholismsevere, and children’s access to educationand health acutely constrained.

At the start of this twenty-first century,indigenous people now occupy severalcounty mayorships and have a majority ofcouncilors in several counties. The provin-cial prefect is also Quichua. Similarly at anational level, former leaders of nationalindigenous people’s organizations are nowministers. And the national Confederationof Indigenous Nationalities of Ecuador hascontrol of the directorate of bilingual edu-cation, the indigenous development coun-cil, and the office of indigenous health. Italso played a big part in negotiating andadministering a World Bank and Interna-tional Fund for Agricultural Develop-ment–supported national Program for theDevelopment of Indigenous and Afro-Ecuadorian Peoples. By any calculation,power relationships have changed inEcuador, becoming more equitable, withindigenous people participating morecompletely (and more equitably) in theirsociety.

The inequality trap for womenUnequal opportunities in health, education,economic welfare, and political agency canbe readily observed in most developingcountries. The preceding sections haveemphasized that these different manifesta-tions of inequity are not generally inde-pendent from one another and that thisinterdependence can replicate inequalitiesover time. This interrelationship can bevividly illustrated by examining the natureand implications of the inequality that trapsmany women in developing countries.

Men and women around the world havestarkly different access to assets and opportu-nities, reinforced by unequal norms andsocial structures, perpetuating gender differ-ences over centuries. Gender inequity directlyaffects the well-being of women and deci-

sions in the home, affecting investments inchildren and household welfare (box 2.9).

Gender inequity is the archetypical“inequality trap.” Most societies have normsthat preserve the prevalent social order, delin-eating different roles and spheres of influencefor men and women. The male sphere is typi-cally outside the home in market work andsocial interactions that enhance the family’s


Gender inequity causes many societies todisplay some preference for male children.But the “son preference” is strong enoughto result in substantial excess female childmortality in parts of East and South Asia—leading to the phenomenon of whatAmartya Sen calls “missing women.” (Sen,1990). In China and India the practice offemale infanticide was noted at least acentury ago, and in the Republic of Koreaand India high juvenile sex ratios (the pro-portion of male to female children belowthe age of 4) have been documented sincethe first modern censuses were taken. Bycontrast there seems to be little son pref-erence in Southeast Asia or in most otherparts of the developing world.

The reasons for this seem to stem fromrigid patrilineal inheritance systems.Whilemost societies deny women inheritancerights, in other parts of the world there issome flexibility in these rules. In peasantEurope and Japan, for instance, womencould inherit land if their parents had nosons. Despite egalitarian laws, customarypractices in China, the Republic of Korea,and northwest India permit a man, if hedoes not have sons, to adopt one fromother male kin. In the past, it would alsohave been possible to take another wife.The driving motivation is to use whatevermeans possible to continue the male fam-ily line.Thus, girl children are undervalued.

During pregnancy, sex-selection maylead to aborting female fetuses, reflected insex ratios at birth that are more masculinethan the biological rate of 105 boys forevery 100 girls. Sex-selection can also hap-pen through infanticide, although the datamake it difficult to distinguish betweenselective abortion and infanticide.The third,and most common, mechanism is the neg-lect and other practices that result in highermortality rates for girls than boys duringearly childhood.37

In China, intense efforts by the govern-ment resulted in a brief improvement in thesex ratio during 1953–64 (see figure to theright). But since the 1980s it has steadily

risen. In the Republic of Korea stark declineshave become only apparent in the lastdecade—perhaps because ofimprovements in labor market opportuni-ties for women. India, as a whole, does nothave juvenile sex ratios that are far differentfrom many other parts of the world. Butnorthwest India has seen some particularlyworrying trends, with sex ratios sharply ris-ing between 1981 and 2001, much attribut-able to the higher incidence of sex-selection in abortion. Other parts of India,especially the south, have more equitablelabor markets and fewer restrictions onwomen’s mobility and inheritance.

Source: Das Gupta and others (2003).

B O X 2 . 9 Sex ratios and “missing women”

1.00 1.05 1.10 1.15 1.20 1.25

1953 1964 1982 1990 1995 2000

1949 1960 1970 1980 1990 1995 2000

1951 1961 1971 1981 1991 2001

1951 1961 1971 1981 1991 2001

China

Republic of Korea

Punjab and Haryana (India)

India

Juvenile (newborn to four years old) sex ratios inChina, the Republic of Korea, India, and Punjaband Haryana, 1950–2000


status and power. The female sphere is usuallyinside the home—looking after householdwork, rearing children, and contributing tothe stability of the household. So, women’sactivities serve primarily as inputs into thehousehold’s collective well-being, while menare ostensibly at its center—its breadwinnersand its link to the larger world where eco-nomic and social status are determined.

Marriage and kinship systems preservethese structures of patriarchy. Most societiesare “patrilocal,” with women moving fromtheir parents to their husband’s home aftermarriage. Marriage can therefore be thoughtof as a framework that serves to exchangewomen between households, and marriagedecisions are made with a view toward ensur-ing that this exchange of women promises themaximum gain to both households. Theman’s household is the point of reference—while the woman is simply an input into theprocesses for households controlled by mento generate economic and social returns.38

Inheritance tends to be consistent withthis pattern. Most societies are not justpatrilocal—they are also patrilineal, withinheritance and property rights primarilypassed on to men. The majority of countries,outside of Europe and Central Asia and LatinAmerica and Caribbean, restrict inheritancerights to women.39 Some countries have leg-islation that guarantees equality in inheri-tance laws. But these laws often are notenforced, and real authority over decisionson inheritance rests in the hands of villageelders and chiefs, who follow customarypractices that discriminate against women.

Most countries that have unequal inheri-tance laws also have unequal property rightsregimes.40 Indeed, the vast majority of landowners are men.41 Many societies compoundthis by denying women the right to divorce.This inequality in property rights regimespersists even in countries where agriculturalproduction depends heavily on women’slabor, such as many in Sub-Saharan Africa.In Cameroon, women make up more than51 percent of the population and do morethan 75 percent of the agricultural work, butthey are estimated to hold fewer than 10 per-cent of all land certificates.42 So, if womenwork on farms, they are usually working onfarms owned by men.

In addition to being denied inheritanceand property rights, women in many soci-eties face restrictions on their mobility. Forexample, in the state of Uttar Pradesh innorthern India close to 80 percent ofwomen require their husband’s permissionto visit a health center, and 60 percent haveto seek permission before stepping outsidetheir house.43 These mobility restrictionsmay be socially imposed, as with gunghatamong Hindus—or have religious sanc-tions, as with purdah among Muslims. Suchpractices are not just socially enforced, theycan be internalized by women who treatthem as marks of honorable behavior. Thesenorms are transmitted by parents to theirchildren, ensuring their continuity over gen-erations; in many societies, they are enforcedby older women in the community.44

Restrictions on mobility and rules of kin-ship and inheritance help shape social per-ceptions about women’s roles. If women aresocially and economically directed to focustheir attention and energy on activities in thehome, this is not just what men expect ofthem—it is also what other women expect ofthem. In much of the developing world,women’s participation in the labor market ismore a function of adversity than activechoice—because husbands cannot earn anadequate income or because of an unantici-pated shock, such as a child’s illness.Bangladeshi women described it this way,“Men work to support their families, womenwork because of need.”45 Women around theworld participate in a fair amount of market-based activity for a wage, but they have to con-tinue to perform most household chores (fig-ure 2.13). They thus face a time squeeze,spending more time at work, both in and outof the home, than men do.

Because social and economic factorsdetermine women’s life chances more in mar-riage than in labor markets, parents invest lessin their human capital. Throughout thedeveloping world, women are much lesslikely to be enrolled in secondary school oruniversity than men.46 So, they typically workin less lucrative occupations. Moreover, labormarkets may themselves be discriminatory,paying women less than men for the samework. For these reasons, even when womenparticipate in the labor market, they earn less



than men. Low earnings are a further disin-centive for women to enter the labor market,perpetuating traditional social roles.

Inequality in the homeFor a long time, economists did not ade-quately recognize that gender inequity has animpact in the home, and models of thehousehold assumed that decisions were takenby one person—with no room for differentchoices across spouses. The consequence ofthis world view is not just academic. It sug-gests, for instance, that policy interventionsthat attempt to alleviate poverty should notbother with targeting by gender—or suggeststhat taxes on a household will not affect theallocation of resources within it.

Economists now question this view,developing models of household decisionmaking that allow for inequality betweenspouses. The new models start with theassumption that households are efficient, inthe sense that they make decisions that max-imize the use of the household’s resources.With this assumption, the models show thata spouse’s share in household resources isdetermined by two factors. The first is thefallback option for the spouse in the event ofdivorce—laws of inheritance, property, anddivorce would matter here. Second is the rel-ative size of the spouse’s contribution to thehousehold’s income, which is determined bytheir opportunities in the labor market.47 Ifhusbands and wives have different prefer-ences, an increase in a woman’s outsideoptions or in her labor market opportuni-ties should reflect consumption choicesmore in line with her preferences.

Econometric work confirms that an in-crease in a woman’s relative worth and animprovement in her fallback options haveeffects on consumption patterns.48 Thehealth of Brazilian children improves whenadditional nonlabor income is in the hands ofwomen.49 In the United Kingdom, when leg-islation ensured that child support paymentswere made directly to mothers, expenditureson children’s clothing tended to rise.50 InBangladesh and South Africa, women bring-ing more assets into the marriage increasehousehold expenditures on children’s educa-tion.51 The patterns seem to indicate that,when women are better off, children seem to

benefit more than when men are better off.The most obvious way to explain bargainingand sharing is to assume that women intrin-sically care more about children than men do,but this risks being tautological.

Perhaps the explanation can benefit fromunderstanding that social and economic dif-ferences outside the household can matternot only for determining bargaining powerbut also for socially determined perceptionsof what men and women consider impor-tant. If men and women occupy different“outside” and “inside” spheres of influence, itseems to make sense that improvements inwomen’s incomes would have a greaterimpact on investments in the household.Improvements in the income of men, bycontrast, are more likely to result in socializ-ing activities outside the home and in pur-chases that reflect social status.

Another consequence of this separationbetween inside and outside roles is that


0 400400Minutes a day

Australia

Marketactivity

Marketactivity

Women Men

Nonmarketactivity

Nonmarketactivity

R.B. de VenezuelaUnited States

United KingdomPhilippines

NorwayNetherlandsNepal, urban

Nepal, ruralKenya, urban

Kenya, ruralItaly

IsraelIndonesia

GuatemalaGermany

FranceFinland

DenmarkColombia

CanadaBangladesh

Austria

Figure 2.13 Women work longer hours than do men

Source: United Nations Development Programme (1995).Note: Data refers to rural Bangladesh in 1990, urban Colombia 1983, rural Guatemala 1977, urban Indonesia 1992, ruralKenya 1988, urban Kenya 1986, rural Nepal 1978, urban Nepal 1978, rural Philippines 1975–77, urban Venezuela 1983,Australia 1992, Austria 1992, Canada 1992, Denmark 1987, Finland 1987-88, France 1985–86, Germany 1991–92, Israel1991–92, Italy 1988–89, the Netherlands 1987, Norway 1990–91, the United Kingdom 1985, and the United States 1985.


inequalities in the home are also manifestedin differences in access to information,which can be used to manipulate intra-household bargaining. In an ethnographicstudy of Bangladeshi garment workers,Kabeer (1997) found that men and womentried to control information about theirincomes from their spouses so that theycould make purchases without consultingthem. Women may also hesitate to shareinformation with their husbands, or to col-laborate efficiently in farming their plots ofland, to retain control over their property. Instudying the farms owned by men and thoseowned by women in Ghana, Udry (1996)found, keeping other things constant, thatwomen-owned farms were less productivethan those owned by men. When wives andhusbands are not sharing information, ormanipulating the flow of information, theyclearly are not using their resources opti-mally. In other words, intrahouseholdbehavior is not efficient—contradicting animportant assumption in economic models.

The widespread domestic violence in thefamily is another type of inefficiency. RecentWorld Health Organization (WHO) datashow that both physical and sexual violenceare widespread in diverse parts of the world(table 2.2). An important reason for domes-

tic violence is that it allows husbands toinstitute a regime of terror to control theirwives’ behavior. In India, Bloch and Rao(2002) find that husbands systematically useviolence as a means of extracting a largerdowry from their wives. This “instrumental”use of violence has widespread acceptanceamong both men and women. Surveys havefound that large percentages of respondentsin developing countries report that menhave the right to beat their wives when theyanswer back or disobey them.52

Gender inequity is thus the result of anoverlapping set of economic, social, cultural,and political inequalities that reinforce eachother. They cause women to have less access toproperty rights, wealth, and education—andlimit their access to labor markets and tospheres of activity outside the home. This, inturn, constrains their ability to influencehousehold decisions. Also limiting this influ-ence are asymmetries of information in thehousehold and the use of violence to controlwomen’s behavior. All of this maintains a cleardemarcation between the roles of women andmen, readily reproduced across generations.

There are some signs that changes inlabor markets and interventions by the statecan break this inequality trap. The develop-ment of the garment industry in Bangladeshhas resulted in a sharp and visible increase inwomen’s access to a lucrative labor market,expanding their ability to influence house-hold choices.53 Higher wages for womenseem to compensate for restrictive practices,such as purdah, by reducing limits onwomen’s physical mobility, and increasingtheir say in household decision making.54

Globalization has expanded opportunitiesfor women in Mumbai and increased theiraccess to schooling.55 A comparative study ofthe Philippines, Sumatra, and Ghana foundthat patterns of land inheritance and invest-ments in schooling have became more egali-tarian because of changes in labor marketopportunities for women.56 And althoughChina, Republic of Korea, and India startedout with similar discriminatory social struc-tures, intervention by the state has improvedgender equity much more in China than inRepublic of Korea or India.57


Table 2.2 Percentage of women who have everexperienced physical or sexual violence by anintimate partner

Physical violence Sexual violence

Bangladesh, rural 42

Brazil, urban 27 10

Ethiopia, rural 49 59

Namibia, urban 31 17

Peru, rural 62 47

Samoa 41 20

Serbia and Montenegro 23 6

Tanzania, urban 33 23

Thailand, rural 34 29

Source: Unpublished data from the WHO Multi-Country Study onWomen’s Health and Domestic Violence Against Women obtainedfrom a presentation by Claudia Garcia-Moreno at the World Bank’sConference on Gender-Based Violence. The final published com-parative report is forthcoming.Note: Data refer to different time periods. Brazil, Peru, and Thailandrefer to 2000. Reference period for Bangladesh, Ethiopia, Namibia,Samoa, Serbia and Montenegro, and Tanzania are unknown.


Documents

Inequity within countries: individuals and groups 2