Review of Income and Wealth

Series 48, Number 3, September 2002

EDUCATION AND INCOME INEQUALITY: NEW EVIDENCE FROM

CROSS-COUNTRY DATA

BY JOSE DE GREGORIO
Central Bank of Chile, and Center for Applied Economics, Uniersidad de Chile

AND

JONG-WHA LEE*
Economics Department, Korea Uniersity

This paper presents empirical evidence on how education is related to income distribution in a panel
data set covering a broad range of countries for the period between 1960 and 1990. The findings
indicate that educational factorshigher educational attainment and more equal distribution of edu-
cationplay a significant role in making income distribution more equal. The results also confirm
the Kuznets inverted-U curve for the relationship between income level and income inequality. We
also find that government social expenditure contributes to more equal distribution of income. How-
ever, a significant proportion of cross-country variation in income inequality remains unexplained.

1. INTRODUCTION
Income distribution has long been a topic of interest among economists.
Initially, attention was focused on whether inequality is necessary for accumula-
tion and how income distribution changes with economic growth. In recent years,
motivated by the availability of new data sets and advances in the theories of
economic growth and development, there has been renewed interest in under-
standing the determinants and the dynamics of income distribution.1 It is no
surprise, then, that the title of Anthony Atkinsons presidential address at the
Royal Economic Society in 1996 was Bringing Income Distribution in from the
Cold.
The literature emphasizes education as one of the major factors affecting
the degree of income inequality. Policymakers usually justify higher educational
spending as a highly effective tool for reducing income inequality. However,
theoretical studies suggest that the relation between education and income
inequality is not always clear. For instance, the human capital model of income
distribution, stemming from the work of Schultz, Becker, and Mincer, implies

The authors wish to thank two anonymous referees and seminar participants at the Harvard
Institute for International Development, Latin American Econometric Society Meetings, Universidad
de Chile, and Kyung-Hee University for their helpful comments. Claudio Bravo provided superb
research assistance. Financial support provided by grants from FONDECYT (Chile) Grants 1980161
and 1990063, and Korea University are gratefully acknowledged.
*Correspondence to: Jong-Wha Lee, Department of Economics, Korea University, Sungbuk-ku,
Anam-Dong 1, Seoul 136701, South Korea ([email protected]).
1
See Atkinson (1997), Acemoglu (1997), Becker and Tomes (1986), Benabou (1994), Deininger
and Squire (1996, 1998), Durlauf (1996), Galor and Zeira (1993), Galor and Tsiddon (1997), and
Gottschalk and Smeeding (1997), among others.

395
that the distribution of earnings (or income) is determined by the level and the
distribution of schooling across the population. While the model predicts an
unambiguously positive association between educational inequality, as measured
by the variance of schooling, and income inequality, the effect of increased aver-
age schooling on income inequality may be either positive or negative, depending
on the evolution of rates of return to education.
In the literature on development economics, Knight and Sabot (1983) also
emphasize the complicated effect of human capital accumulation on income dis-
tribution due to composition and wage compression in a dual economy.
They argue that an expansion of education has two different effects on the earn-
ings distribution. The composition effect increases the relative size of the group
with more education and tends initially to raise income inequality, but eventually
to lower it. On the other hand, the wage compression effect decreases the pre-
mium on education as the relative supply of educated workers increases, thereby
lowering income inequality. Consequently, the effect of increased education on
the dispersion of income is ambiguous.
The purpose of this paper is to investigate the relationship between education
and income distribution. Considering the ambiguous theoretical predictions about
the relation between education and income distribution, we look for empirical
evidence based on a cross-country data set. A number of empirical studies have
investigated the relationship between education and income equality (see the sur-
vey by Psacharopoulos and Woodhall, 1985, pp. 26470; Ram, 1989). Earlier
work shows a close relation between education and income distribution in devel-
oped countries. Becker and Chiswick (1966) show that, across regions in the
United States, income inequality is positively correlated with inequality in school-
ing and negatively correlated with the average level of schooling. Chiswick (1971),
using cross-sectional data from nine countries, suggests that earnings inequality
increases with educational inequality. Subsequent studies have been based on a
slightly larger sample of countries. Most of them find that a higher level of school-
ing reduces income inequality, while inequality of educational attainment
increases it.2 On the other hand, Ram (1984, 1989) finds, with slightly different
specifications and data, that mean schooling and schooling inequality have no
statistically significant effects on income inequality.
Previous cross-national studies are very much hampered by a lack of inter-
nationally comparable data, and they therefore end up with a few data points
from heterogeneous sources. The quality of data on income distribution and
schooling has always been questioned. Unlike previous empirical studies, our
paper utilizes a newly constructed panel data set of internationally comparable
human capital and income distribution for a broad number of countries measured
at five-year intervals from 1960 to 1990. Data on income distribution have been
compiled by Deininger and Squire (1996) and in the form of a time series, at
irregular frequencies, for more than 100 countries since 1960. Measures of the
level of schooling and its inequality are constructed from the Barro and Lee
(1996) educational attainment data set for more than 100 countries at five-year

2
Adelman and Morris (1973), Chenery and Syrquin (1975), Ahluwalia (1976), Marin and Psachar-
opoulos (1976), and Winegarden (1979).

396
intervals from 1960. Using this unbalanced panel data set, we investigate both
cross-national and intertemporal relationships between education and income
distribution.
In the next section we discuss the data and present the results of estimating
the effects of education and income on income inequality. In section 3 we analyze
the determinants of educational level and its inequality across countries, based
on past values of educational variables and income. In section 4 we assess our
results by decomposing the cross-sectional differences in income inequality, and
we provide some quantitative estimates of the long-term effects of changes in
education on income distribution. Section 5 presents our conclusions.

2. THE EFFECTS OF EDUCATION ON INCOME INEQUALITY

Income distribution is related to the populations average schooling and its
dispersion. Income inequality increases with education inequality. In contrast, for
a given distribution of education, an increase in average schooling has an ambigu-
ous effect on income distribution.
To illustrate this, traditional models of human capital theory would suggest
the following expression for the level of earnings (Y) of an individual with S years
of schooling:
s
log Ys Glog YoC log (1Cr j)Cu,
j G1

where rj is the rate of return to the j-th year of schooling and u reflects other
factors that influence earnings independent of education. The function can be
approximated by
log Ys Glog YoCrSCu.
Using a bar over a variable to denote its mean, we can write the distribution
of earnings as
Var(log Ys)Gr 2 Var(S )CSr 2 Var(r)C2rSr Cov(r, S )CVar(u).
Hence, an increase in educational inequality (Var(S )) leads unambiguously
to greater income inequality, with other variables held constant. If the rate of
return (r) and schooling level (S) are independent, an increase in the level of
schooling will also lead unambiguously to a more unequal income distribution.
If, however, the covariance between the return to education and the level of edu-
cation is negative (as evidenced by a number of studies by Psacharopoulos), an
increase in schooling can reduce income inequality. For example, we can think of
an economy where improved access to education may allow people with high
ability to earn more income than people with low ability, even when all of them
have the same level of education.3 In this case, as education expands, income
distribution may become more unequal. This may be particularly important in
economies with very low levels of education. However, as more people receive
education, the return on education will decline, reducing income inequality.
3
See, for example, De Gregorio and Kim (2000).

397
 In the rest of this section we look at the relationship among income
inequality, the level and dispersion of education, and the level of income across
countries. We focus on the issue of whether countries with higher educational
levels or less dispersion of education among the population, or with higher levels
of development, have a more equal or less equal income distribution.
We estimate the following regression4,5 :
(1) G j,t Ga0,tCa1 Ej,tCa2 E j,tCa3 log y j,tCa4 [log y j,t]2CaD D jC j,t ,
where G is the Gini coefficient, a measure of the distribution of income. We
constructed Gini coefficients from the raw data on income share by quintile
reported in Deininger and Squire (1996). The variable E is the dispersion of
educational attainment in the population. It is calculated as the standard devi-
ation of schooling for a given year and is based on the data constructed by Barro
and Lee (1996). We refer to this variable as the distribution of education, the
dispersion of education, or the standard deviation of education, and we use it as
a proxy for educational inequality. E is the average years of school attainment
for the population aged 15 and over, from Barro and Lee (1996). The variable y
is GDP per capita. Finally, D is a set of dummy variables that distinguish certain
data characteristics and regions to which countries belong. The subscripts j and t
index countries and periods, respectively. The data set is an unbalanced panel for
which data are available at five-year intervals from 1960 to 1990.6 The number
of observations for each period t depends on the number of countries with all the
data available.
To measure educational inequality, we constructed a standard deviation of
educational distribution for the total population over 15 years of age. The Barro
and Lee (1996) data set on educational attainment provides panel information on
the distribution of population by level of educational attainment in seven categor-
ies: no formal education, incomplete primary, complete primary, first cycle of
secondary, second cycle of secondary, incomplete higher, and complete higher.
This information enables us to construct the standard deviation of educational
attainment for each country at five-year intervals from 1960 to 1990.6
We present a summary of statistics in Table 1. On average, since the 1960s,
educational attainment has increased throughout the world. Even in regions with
greater inequality, such as Latin America, we see that there has been a substantial
increase in average years of education. In contrast, overall inequality has
increased slightly, especially in Africa. However, one has to be careful when inter-
preting these results, since for the whole world we have only 23 countries with
data available in 1965, whereas we have 71 in 1990. So it may be that new data
4
One concern in the empirical specification is that any effect of education (as well as of income)
on income distribution may reflect reverse causation, that is, income inequality affecting educational
attainment. We try to control for this endogeneity problem by using lagged rather than contempor-
aneous values of the independent variables in the estimation of equation (1). See the results of
Regression 2.3 in Table 2.
5
We also test whether the level and the dispersion of education have any nonlinear relationship
with income inequality. However, the squares of educational attainment and inequality turned out to
be statistically insignificant in the regressions for income inequality.
6
The standard deviation is computed by assuming that each person has an educational attainment
of log(1Cyears of schooling). Thus a person with no formal schooling is assumed to have one (effec-
tive) year of educational human capital.

398
 TABLE 1
SUMMARY OF DATA

Standard Standard
Average E Deviation sE Average E Deviation sE
School Attainment 1965 1965 1990 1990
All countries
Mean 3.86 0.73 5.58 0.81
Standard deviation 2.58 0.19 2.83 0.19
Maximum 9.91 1.10 11.7 1.19
Minimum 0.17 0.32 0.65 0.36
Africa
Mean 1.61 0.74 2.81 0.88
Standard deviation 1.06 0.18 1.40 0.16
Asia
Mean 2.96 0.85 5.24 0.89
Standard deviation 1.87 0.19 2.56 0.13
Latin America
Mean 3.51 0.85 5.43 0.81
Standard deviation 1.29 0.19 1.59 0.13
OECD
Mean 6.66 0.55 8.44 0.59
Standard deviation 2.25 0.15 1.98 0.15
Average Average
Gini Coefficient 1965 1990
All countries
Mean 0.368 0.411
Standard deviation 0.090 0.101
Maximum 0.560 0.623
Minimum 0.229 0.233
Africa*
Mean 0.462 0.460
Standard deviation 0.127 0.097
Asia
Mean 0.379 0.367
Standard deviation 0.079 0.075
Latin America*
Mean 0.517 0.498
Standard deviation 0.063 0.067
OECD
Mean 0.354 0.327
Standard deviation 0.088 0.050
Note: *The first period is 1970 because of data availability.

becoming available tend to be from countries with greater inequality than the
previous average.7 Another interesting observation is that the standard deviation
of educational attainment across countries has increased in all regions, which
may be a factor offsetting the potential equalizing effect of increased educational
attainment.
The regional dummies in the regressions are included to capture differences
in income distribution that are not accounted for by education or income. For
7
This does not happen with data on education, since the panel of countries is relatively balanced.

399
example, it has been argued that the distribution of landholding and the distri-
bution of natural resources are important factors explaining income distribution
in Latin American countries (Londono, 1996), and, as is apparent from Table 1,
there may be factors other than education that explain the differences between
Latin America and Asia. We include dummies for African, Asian, and Latin
American countries, since our results indicate that there is no evidence of any
other significant regional dummy in equation (1).
We also include some dummies to control for data characteristics. The data-
base prepared by Deininger and Squire (1996) uses different sources to compute
Gini coefficients, depending on the data available in each country. There are three
major differences. The first is whether the unit of analysis is a household or an indi-
vidual. If, as is usually the case, poor households have more members than more
affluent households, the distribution of income at the household level will be more
equal than that computed at the individual level. Therefore, in our regression
analysis we expect to find that the Gini coefficients are greater (income distri-
bution is more unequal) in countries that report data at the individual level. The
second issue is whether income data refer to income before or after tax. Provided
the tax system is progressive, countries that collect data on gross (before-tax)
income will probably have a higher Gini coefficient than countries that report
data on net income, and hence the dummy is presumed to have a positive coef-
ficient. Finally, some countries measure the distribution of income, while others
measure the distribution of expenditure, which is measured on the basis of net
income. In addition, given that high-income households presumably save a larger
proportion of their income than poor households, it is expected that countries
that use income rather than expenditure will have higher Gini coefficients.
Before going into the details of the results, it is useful to look at the simple
cross-correlation between income inequality and the educational variables. Figure 1
plots average years of schooling (also referred to as educational attainment) in 1990
against the Gini coefficient. The relationship is negative, indicating that increases in
education reduce inequality. On the other hand, Figure 2 shows that there is a
positive relationship between income and educational inequality. Although these
figures are suggestive, further statistical analysis is required to examine their
robustness and obtain orders of magnitude for the importance of educational
factors in explaining differences in income distribution across countries.
We estimate equation (1) from the panel data set on education and income
distribution that we have compiled. The panel consists of six equations: for 1965,
1970, 1975, 1980, 1985, and 1990.8 The system is estimated by seemingly-
unrelated-regression (SUR) techniques.9 This procedure allows for different
8
We use data from 1960 as one-period-lagged explanatory variables in the regressions.
9
The SUR estimation identifies relationships using both cross-country variations and within-
country time-series variation. Since there is little within-country time-series variation in measured
income inequality, the estimated results are mostly driven by cross-sectional relationships. The panel
estimation including country fixed effects (not reported here) confirms that there is no significant
relationship between educational and income inequality. The relationship between educational
inequality and income inequality also becomes insignificant if we use balanced panel data, mainly
because the sample size of the balanced panel data becomes very small, covering only 15 countries in
each period of the panel from 1965 to 1990. By contrast, the cross-sectional relationship between
educational inequality and income inequality for each period of the panel turns out to be statistically
significant for most single periods (results available from the authors on request).

400
 0.7

0.6

0.5
Gini, 1990

0.4

0.3

0.2
0 5 10 15

Average years of schooling, 1990

Figure 1. Educational Attainment and Income Distribution, 1990

0.7

0.6
Income Gini, 1990

0.5

0.4

0.3

0.2
0.2 0.4 0.6 0.8 1.0 1.2 1.4

Education Dispersion, 1990

Figure 2. Education Dispersion and Income Distribution, 1990

401
 TABLE 2
PANEL REGRESSIONS FOR INCOME INEQUALITY

Gini Coefficient
Dependent Variable 2.1 2.2 2.3 2.4
Educational inequality ( E) 0.097 0.058 0.047 0.014
(0.027) (0.028) (0.027) (0.032)
Educational attainment (E) 0.009 0.008 0.008 0.006
(0.002) (0.003) (0.003) (0.003)
Log of GDP per capita 0.284 0.270 0.454
(0.085) (0.082) (0.096)
Square of log of GDP per capita 0.018 0.018 0.029
(0.005) (0.005) (0.006)
Social expenditureGDP (tA1) 0.0021
(0.0012)
Data dummies
Individual (vs. household) 0.032 0.034 0.037 0.027
(0.008) (0.008) (0.008) (0.009)
Gross income (vs. net) 0.041 0.041 0.046 0.037
(0.010) (0.010) (0.010) (0.011)
Income (vs. expenditure) 0.019 0.016 0.018 0.023
(0.013) (0.013) (0.013) (0.014)
Regional dummies
Africa 0.095 0.105 0.105 0.090
(0.015) (0.017) (0.018) (0.019)
Asia 0.021 0.021 0.028 0.032
(0.013) (0.015) (0.015) (0.016)
Latin America 0.103 0.093 0.089 0.074
(0.012) (0.012) (0.012) (0.014)
R2 (number of obs.) 0.52 (22) 0.49 (21) 0.49 (21) 0.48 (18)
0.74 (39) 0.74 (39) 0.80 (37) 0.71 (31)
0.70 (46) 0.73 (46) 0.71 (45) 0.74 (36)
0.76 (48) 0.76 (48) 0.75 (48) 0.61 (42)
0.62 (54) 0.63 (54) 0.64 (54) 0.77 (48)
0.58 (65) 0.59 (63) 0.59 (65) 0.49 (64)
Note: The system has six equations, where the dependent variable is the Gini coefficient at five-
year intervals from 1965 to 1990. The system of equations was estimated by the seemingly unrelated
regression (SUR) technique. The estimation allows for different error variances in each equation and
for correlation of these errors across equations. Different constant terms (not reported) are included
in each equation. Standard errors are reported in parentheses. The R2 and number of observations
apply to each equation. Regression 2.3 uses one-period-lagged values for the educational attainment,
educational inequality, and income per capita variables.

error variances in each equation and for correlation of these errors across equa-
tions. We allow for different constant terms in each equation, but we assume that
the slope coefficients are the same for each variable across equations.10 The
regressions apply to a total of 274 observations.
The results are presented in Table 2. Regression 2.1 is a basic regression
without income variables. Using only contemporaneous education and other con-
trolling variables, the regression explains about 70 percent of the variance of
10
We test the restrictions on the slope coefficients for each explanatory variable across equations
in Regression 2.2, reported in Table 2. A test of equality for all six coefficients for educational
inequality is accepted by a Wald test at a p-value of 0.57. The p-value for the test of equality of the
coefficient on educational attainment is 0.07. For income per capita and its square terms, the joint
test of equality of estimated coefficients is accepted at a p-value of 0.13.

402
income distribution, except for the latter period, where its explanatory power
drops below 50 percent.
The results show the role of education in income inequality. Countries with
higher educational attainment also have a more equal income distribution. Con-
sidering that the standard deviation of the cross section of educational attainment
is between 2.5 and 2.9 years, the coefficients (0.009) suggest that an increase of
one standard deviation in educational attainment reduces the Gini coefficient by
about 0.03 (that is, 3 percentage points), which may account for about 30 percent
of the standard deviation of the Gini coefficient. From a country viewpoint, how-
ever, this coefficient is relatively small. The average increase in educational attain-
ment is about 2.5 years over the past 30 years. In addition, as long as educational
attainment is related to the level of income and educational inequality, changes
may be quantitatively more important. We discuss these issues later in the paper.
Inequality of schooling, measured as the standard deviation of educational
attainment of the population, has a significantly positive effect on income
inequality.11 The estimated coefficient indicates that a reduction in educational
dispersion by one standard deviation, about 0.2, reduces income inequality by
0.02. In our sample of countries, the dispersion of education has increased on
average by 0.08 over the 30-year period.
The dummy variables that distinguish certain data characteristics show that,
when income is measured before taxes, it is more unequally distributed. This is an
indication that taxes are indeed progressive. On average, taxes reduce inequality
and decrease the Gini coefficient by 0.04. The dummy that distinguishes whether
distribution refers to income rather than expenditure is in general not statistically
significant, although it has a positive point estimate, which indicates that expendi-
ture is more equally distributed than income by 0.02. The puzzling result regarding
the dummy variables is that expenditure appears to be distributed more equally
in countries where income distribution is measured at the individual level. For a
sample of countries where data on income distribution are available at the indi-
vidual and the household levels, Deininger and Squire (1996) show that income
is indeed more equally distributed at the household level because poor households
tend to be larger. Our result comes from a different sample of countries.
With respect to regional dummies, Latin America and Africa appear to have
greater income inequality than the world average, by about 0.1 of the Gini co-
efficient. In contrast, the Asian countries as a group have a more equal income
distribution, by about 0.02 of the Gini coefficient. There exists, therefore, a large
inequality gap between Latin America and Africa with respect to other countries,
particularly Asia.
Regression 2.2 in Table 2 adds the log of income per capita and its square
in order to capture the inverted-U curve proposed by Kuznets for the relationship
between income distribution and the level of income (Kuznets, 1995). The results
confirm that a Kuznets curve also exists in these panel data. We use the log of
income to estimate this relationship, because no relationship was found when the
measured level of income per capita was used as a regressor. The Kuznets curve
11
We also used other indicators of educational inequality. In particular, we constructed a Gini
coefficient index for educational inequality, and we used an index of wage dispersion. The results were
similar to those reported, although they were in general less significant.

403
resulting from the regressions in Table 2 indicates that income distribution
becomes more unequal with higher levels of income up to a range of income
between US$1,800 and US$2,700 (in purchasing power parity-adjusted dollars at
1985 international prices) and then starts to equalize.
Another specification for the Kuznets curve has been proposed by Anand
and Kanbur (1993) and estimated by Deininger and Squire (1998). It includes
income in the regression as y and 1y. Our results show that for different specifi-
cations the nonlinearity in the relationship between income and its distribution is
significant, and it also holds when estimated for each period.12,13
We can conclude, along lines similar to Deininger and Squire (1998), that
with cross-sectional data there is evidence of a Kuznets curve, but this result does
not mean that the relationship is strong or holds over time. In Figure 3 we show
the orthogonal component of the Gini coefficient to educational variables against
GDP per capita. That is, we estimate Gini coefficients against educational
inequality and attainment, and the residual is plotted against the log of income
per capita. The figure shows the data for 1970 and 1990 as an example. As can
be seen, there is no clear evidence of a strong relationship.14 Therefore, although
the Kuznets relationship is captured in the regressions, it is rather weak.15
Regression 2.3 includes all the independent variables lagged by one period
to control for the possible endogeneity problem. The results are the same as for
Regression 2.2 using contemporaneous variables; that is, higher educational
attainment and less inequality of schooling lead to a more equal income distri-
bution. This equation can be used to predict future income distribution from
current data.
We also examine the effect of government social expenditure on income dis-
tribution. This is an important variable whose omission could bias the results on
education if it is correlated with educational variables. Hence, we run Regression
2.1 including the ratio of social expenditure to GDP averaged over the previous
five years.16 The result, reported in column 2.4 of Table 2, shows that government
12
To verify that the results do not depend on the variables included in the regression, or on the
period, we regressed the Gini coefficients on the income variables and time-specific effects only, omit-
ting all dummy variables. In Table A.1 of the Appendix we report the results assuming that the
relationship is the same for all periods (Regression 1.1) and, alternatively, that it is different across
periods of the panel (Regressions 1.2 to 1.8). The regressions show that, with the exception of the
quadratic specification using the level of income, there is evidence of a Kuznets curve.
13
When we perform regressions for quintile sharesthe share of the highest quintile, the share
of the lowest quintile, and the share of the middle three quintilesGDP per capita has a highly
significant relation to each income-share measure in nonlinear form, thus confirming the Kuznets
curve. As GDP per capita increases, the highest-quintile share first increases and then declines, while
the shares of both the lowest quintile and the middle three quintiles decrease initially and then rise.
We also find that the average income of the poorest quintile rises with average household income. All
of those results are reported in the working paper version of this paper (De Gregorio and Lee, 1999).
Gallop, Radelet, and Warner (1998) show that the positive relation between growth in income per
capita and the poorest groups income is empirically robust.
14
The figures are very similar for all periods and when, instead of using the orthogonal component
of the Gini coefficients, we use the level of the Gini coefficients.
15
For more discussion on the Kuznets curve, see Sarel (1997) and Barro (1999).
16
Government social expenditure is measured by the average ratio of general government social
security and welfare expenditure to GDP over five-year subperiods: 197074, 197579, 198084, and
198589. The data, available from the early 1970s, were constructed from International Monetary
Fund, Goernment Finance Statistics Yearbook. We use the average ratio of 197074 in the regressions
for Gini coefficients of 1965 and 1970.

404
 (A)
0.2

Gini 1970, orth. comp. to education

0.1

0.0

-0.1

-0.2
5 6 7 8 9 10

log of per capita GDP, 1970

(B)
0.2
Gini 1990, orth. comp. to education

0.1

0.0

-0.1

-0.2
5 6 7 8 9 10

log of per capita GDP, 1990

Figure 3. Kuznets Curves

social expenditure reduces income inequality. The estimated coefficient, 0.002 (t-
statisticG1.8), implies that a 1-percentage-point increase in the social expendi-
ture-GDP ratio lowers the Gini coefficient by about 0.2 percentage points. This
positive contribution of government social expenditure to income equality may
405
occur through two mechanisms. The first is that part of social expenditure consists
of direct transfers to the poor, increasing their income and redistributing income
from rich to poor.17 The second is that social expenditure may promote access
for the poor to education and other human-capital-enhancing activities, such as
health care, thereby contributing to future income equality, especially when credit
markets are imperfect.

3. THE DETERMINANTS OF EDUCATION

In the previous section we examined how distribution is related to the aver-
age level of educational attainment, as well as its dispersion and the level of
development of the economy as measured by income per capita. We showed in
that section that not only the level of education, but also how it is distributed
among the population, is an important determinant of income distribution. Fur-
thermore, an increase in the average level of education has an equalizing effect
on income distribution.
In this section we go one step further by analyzing the evolution of edu-
cation. In particular, we examine how the initial level of development, together
with educational characteristics, affects current educational attainment and its
dispersion. This will allow us to examine the dynamics of income distribution.18
In the specifications presented in this section, we focus on past income and edu-
cational variables to estimate future educational variables.19

3.1. Eolution of Educational Attainment

In order to examine the evolution of educational attainment, we start by
specifying the following regression:
(2) E j,t Gb0,tCb1 E j,tA1Cb2 Ej,tA1Cb3 log y j,tA1Cb4 [log y j,tA1]2CbD D jC j,t .
The regression includes one-period-lagged values of educational inequality
and income per capita and the lagged dependent variable as explanatory vari-
ables. We also include income per capita squared as a regressor.20 As is clear from
Table 1, there are strong regional differences in educational attainment, so we
also include all regional dummies. The equations were also estimated using the
SUR technique allowing for different time intercepts. The data are those in the
five-year panel from 1965 to 1990, and because we use only income and edu-
cational data, our sample contains about 90 countries.

17
The regressions for quintile shares show that higher social expenditure lowers the share of the
wealthiest, while increasing the shares of the poorest and the middle-income groups.
18
We do not fully investigate the factors that determine the level and dispersion of educational
attainment, because our main interest is in explaining the interaction between education and income
variables. We control for the possible endogeneity problem by using one-period-lagged independent
variables in the regressions.
19
We exclude the income inequality variable so as to maximize the size of the sample. In 1965,
for example, we have data on income inequality for 22 countries and data on educational variables
for 92. In fact, we have estimated the effects of income inequality on educational variables in a smaller
sample, and the results are not significantly different from those reported here. However, they are
much more sensitive to changes in specification.
20
The square of log income per capita turns out to be statistically insignificant, so it is excluded.

406
 TABLE 3
REGRESSIONS FOR EDUCATIONAL ATTAINMENT

Average Years of Schooling

Dependent Variable 3.1 3.2 3.3 3.4
Educational attainment (E)(tA1) 0.710 0.948
(0.254) (0.011)
Educational inequality ( E)(tA1) 0.085 0.222 0.278 0.337
(0.305) (0.083) (0.353) (0.106)
Log of GDP per capita (tA1) 3.830 0.710 2.421 0.658
(0.910) (0.254) (1.099) (0.314)
Square of log of GDP per capita (tA1) 0.184 0.038 0.060 0.033
(0.057) (0.016) (0.069) (0.021)
Social expenditureGDP (tA1) 0.108 0.001
(0.022) (0.005)
Regional dummies
Africa 2.251 0.126 0.758 0.084
(0.331) (0.054) (0.334) (0.068)
Asia 0.237 0.082 1.096 0.087
(0.452) (0.054) (0.438) (0.063)
Latin America 0.794 0.122 0.066 0.147
(0.399) (0.042) (0.300) (0.050)
R2 (number of obs.) 0.50 (92) 0.99 (92) 0.53 (65) 0.99 (65)
0.51 (93) 0.95 (93) 0.56 (66) 0.94 (66)
0.54 (98) 0.98 (98) 0.58 (68) 0.98 (68)
0.54 (103) 0.96 (103) 0.68 (90) 0.95 (90)
0.58 (105) 0.99 (105) 0.71 (94) 0.99 (94)
0.60 (106) 0.99 (106) 0.69 (80) 0.98 (80)
Note: Standard errors are in parentheses. See the notes to Table 2.

Regression 3.1 of Table 3 presents the results of estimating equation (2) with-
out the lagged dependent variable. We find that a regional dummy for African
countries has significant negative intercepts, implying that Africa is the continent
with the least education, by about 2.3 years per person, even after controlling for
its low GDP per capita. However, the other regional dummies are not statistically
significant. Past educational inequality also does not help to explain the level of
education.
Regarding income, the log of income per capita and its square turn out
to be statistically significant. The estimated coefficients show an inverted-U-type
nonlinear relationship between income and educational attainment, implying that
educational attainment decreases when the log of income per capita is over 10.4.
In the sample range, however, educational attainment always increases with
income, although at a declining rate. The coefficients imply that, at the mean
value of log income per capita (8.1), the net marginal impact of log income per
capita on educational attainment is 0.84 (3.8320.1848.1). Hence, a country
that is twice as wealthy as another one will have about 0.6 year (0.84 ln(2)) more
schooling per person than the less wealthy country.
The results of Regression 3.2, which includes past levels of educational
attainment, show that educational attainment is highly persistent, and in this
regression the coefficient on past educational inequality turns out to be strongly
positive. Greater educational inequality tends to promote educational attainment
407
by the population. This relationship may be consistent with the theoretical pre-
sumption that a higher dispersion of education, given its level, results in high
returns to education, and hence generates stronger incentives to invest in
education.
Regression 3.3 repeats Regression 3.1 with government social expenditure
included as an additional explanatory variable. Social expenditure enters with a
significantly positive sign, implying that greater social expenditure raises the aver-
age educational attainment of the population. However, the effect of social expen-
diture on educational level is insignificant in Regression 3.4, where the past level
of educational attainment is also included. Accordingly, government social expen-
diture seems to be correlated more closely with cross-country variations in edu-
cational attainment than with variations over time.

3.2. Eolution of Educational Inequality

The last building block in our empirical investigation is to look at the deter-
minants of educational inequality across countries and over time. In particular,
we want to examine the relationship between past values of education and
income, on the one hand, and current educational attainment, on the other. The
equation is specified as follows21 :
(3) Ej,t Gb0,tCb1 Ej,tA1Cb2 E j,tA1Cb3 E 2j,tA1Cb3 log y j,tA1CbD D jC j,t .
The specification incorporates a nonlinear relationship between educational
attainment and educational inequality. Indeed, the dispersion of education would
be expected either to rise or to decline with an increase in average education,
depending on its initial level and distribution. Two extreme cases can be imagined:
the first consists of an economy with no education at all, where an expansion of
educational attainment (from zero) will mean that some people start receiving
education. In this case the average level of educational attainment of the popu-
lation will increase, as will dispersion. The other extreme would be an economy
where most people attend primary and secondary school but only some pursue
higher education. An increase in the level of educational attainment among the
labor force is the result of more people attending postsecondary school, so that
the dispersion of education is likely to decline.22

21
The square terms of educational inequality and per capita income are not included because they
turn out to be statistically insignificant.
22
To illustrate the nonlinear relationship between the level and the distribution of education more
formally, consider that each person can have an educational level between 0 and N. However, actual
education is uniformly distributed between n1 and n2, where 0Fn1 F n2 F N. Therefore n2 is the years
of school attainment of the most educated part of the population, while n1 is the minimum years of
school attainment in this economy. Given the distributional assumption, the mean of educational
attainment, E, is equal to (n2Cn1)2, and the standard deviation of educational attainment, E, is
(n2An1)120.5. Clearly, regardless of whether n1 or n2 increases, E unambiguously increases. However,
an increase in n1 reduces the standard deviation of school attainment, while an increase in n2 raises
it. An (additive) increase of the same magnitude in both, n1 and n2, leaves E unchanged. Which case
is more likely? We think that when n1 and n2 are both low, in the limiting case equal to zero, expansion
of education occurs because of increases in n2, therefore increasing educational dispersion. When n2
reaches N, further increases in average education, in our example, are produced by increases in n1,
and therefore educational inequality declines.

408
 Accordingly, the impact of an expansion of education on its dispersion must
depend on its initial level. In equation (3), the impact of a marginal increase in
EtA1 on E is b1C2b2EtA1, and we expect this to be positive (negative) for low
(high) values of E; thus we presume that b1H0 and b2F0. Indeed, an initial
confirmation of this presumption is presented in Figure 4, where panels (A) and
(B) plot the dispersion of education against average educational attainment for
the two extreme periods in our panel data. It is clear from the figure that, at low

(A)
1.2
Education Dispersion, 1965

1.0

0.8

0.6

0.4

0.2
0 2 4 6 8 10 12

Education Attainment, 1960

(B)
1.4

1.2
Education Dispersion, 1990

1.0

0.8

0.6

0.4

0.2
0 5 10 15

Education Attainment, 1985

Figure 4. Education Inequality and Education Attainment

409
levels of education, its expansion causes an increase in dispersion, whereas for
higher levels the relationship is reversed. The figures show that this reversal occurs
between 3 and 5 years of schooling.23
TABLE 4
REGRESSIONS FORSCHOOLING INEQUALITY

Education Inequality
Dependent Variable 4.1 4.2 4.3 4.4
Educational inequality ( )(tA1)
E
0.898 0.883
(0.014) (0.020)
Educational attainment (E)(tA1) 0.066 0.025 0.061 0.024
(0.010) (0.004) (0.011) (0.004)
Square of educational attainment (E2) 0.008 0.0011 0.007 0.0010
(tA1) (0.001) (0.0003) (0.001) (0.0003)
Log of GDP per capita (tA1) 0.014 0.006 0.018 0.007
(0.013) (0.004) (0.016) (0.005)
Social expenditureGDP (tA1) 0.011 0.0009
(0.002) (0.0006)
Regional dummies
Africa 0.066 0.031 0.066 0.028
(0.032) (0.008) (0.033) (0.008)
Asia 0.130 0.017 0.019 0.021
(0.038) (0.008) (0.036) (0.008)
Latin America 0.029 0.019 0.060 0.023
(0.032) (0.006) (0.028) (0.007)
R2 (number of obs.) 0.34 (92) 0.95 (92) 0.40 (65) 0.96 (65)
0.31 (93) 0.89 (93) 0.39 (66) 0.88 (66)
0.38 (98) 0.94 (98) 0.52 (68) 0.94 (68)
0.39 (103) 0.90 (103) 0.46 (90) 0.89 (90)
0.46 (105) 0.96 (105) 0.57 (94) 0.96 (94)
0.48 (106) 0.96 (106) 0.61 (80) 0.96 (80)
Note: Standard errors are in parentheses. See the notes to Table 2.

The regression results for equation (3) are presented in Table 4. Regression
4.1 confirms the nonlinear relationship between educational attainment and edu-
cational inequality. The estimated coefficients on educational attainment and its
square are consistent with the evidence of Figure 4. According to the coefficient
estimates, the point at which education begins to reduce dispersion is 4.2 years.24
The regression result also shows that initial GDP appears to have a positive
relationship with educational inequality, but the coefficient on income per capita
becomes insignificantly different from zero.
Regression 4.2 uses the same specification but adds a lagged dependent vari-
able. The result shows that the dispersion of education is highly persistent over
time. In this case the coefficients on educational attainment and its dispersion
change sign, but this has no important implications in the relevant range, as they
imply that educational dispersion decreases with attainment up to 12 years, which
is outside the values in our sample. The coefficients on the regional dummy vari-
ables for Africa, Asia, and Latin America also change sign across specifications.
Their inclusion does not affect the main results but does increase the R2.
23
The graphs are very similar when current educational attainment is used in the horizontal axis.
24
In a previous version we used a different specification for nonlinearity, including an interaction
term between E and its lagged value, and the results showed a cutoff at about 3.5 years.

410
 Regressions 4.3 and 4.4 include government social expenditure as an explana-
tory variable. The coefficient on social expenditure is negative, implying that
higher social expenditure helps to decrease the inequality of schooling. However,
the estimated coefficient is statistically insignificant in Regression 4.4, which
includes in addition the inequality of past schooling. In summary, we find that
government social expenditure helps to explain cross-country differences in
income inequality, level of educational attainment, and dispersion of education.

4. SOURCES OF INCOME INEQUALITY

In the previous section we estimated the determinants of education and
income distribution. In this section we use our estimates to explain cross-country
differences in income inequality. We use the empirical result of Regression 2.4 to
determine the relative contribution of each of the explanatory variables to income
inequality in 1990 for 49 countries for which all necessary data are available.25
Table A.2 in the Appendix presents the detailed results in a framework of
sources of income inequality to show how the explanatory variables account
for each countrys income inequality relative to the mean value for all the sample
countries. Thus this exercise looks at cross-sectional differences in inequality of
income. The 10 most equal countries in the upper quintile of the income distri-
bution include five OECD countries (Spain, Finland, Belgium, Canada, and the
Netherlands), four Asian countries (Bangladesh, Sri Lanka, Taiwan, and Pakis-
tan), and one AfricanMiddle Eastern country (Egypt). The ten most unequal
countries include five African countries (Zimbabwe, Guinea-Bissau, Kenya, Mali,
and South Africa) and five Latin American countries (Chile, Panama, Brazil,
Mexico, and Guatemala). Thus income inequality differs systematically across
regions, and this is why regional dummies appear very significantly in the
regressions. Nonetheless we find that education and income factors played a mod-
erate role in explaining the variations across countries. For example, in the five
most equal OECD countries as listed in Appendix Table A.2, the mean Gini
coefficient was below the world average by 0.137; 0.020 of this gap was explained
by educational factors (including educational attainment and educational
inequality) and 0.040 by income factors. Social expenditure also contributed to
lower income inequality by 0.020, which leaves an unexplained gap of 0.057. For
the five most unequal African countries, the mean Gini coefficient was 0.158
higher than the world average; educational factors accounted for 0.020 of this
gap, while income and social expenditure explained 0.006 and 0.008, respectively.
Thus educational factors, government social expenditure, and income cannot
explain the bulk of the inequality in the most unequal countries. As income level
has a nonlinear relationship with income inequality, the lower income in the
African countries does not contribute much to income inequality, compared with
countries in other developing regions.

25
This exercise adopts a form of growth accounting, which is often used in the literature of cross-
country growth regressions (see Barro and Lee, 1994, for example). The predicted value of the dependent
variable from the regression can be broken down into the contributions of each of the explanatory
variables. The contribution of each explanatory variable is calculated by multiplying the mean value of
the variable by its estimated coefficient from the regression.

411
 TABLE 5
EXPLAINING CROSS-COUNTRY DIFFERENCES IN INCOME INEQUALITY IN 1990*

Africa Asia Latin America OECD

Gini (1990)**
Actual 0.062 0.033 0.096 0.081
Predicted 0.061 0.030 0.087 0.081
Explained by
Education 0.019 0.001 0.005 0.020
Educational attainment 0.018 0.000 0.004 0.017
Educational inequality 0.001 0.001 0.001 0.003
Income 0.003 0.015 0.032 0.041
Social expenditure 0.007 0.009 0.005 0.015
Other factors*** 0.032 0.055 0.044 0.005
Notes:
*We use the empirical result of Regression 2.4 to determine the relative contribution of each
of the explanatory variables to income inequality in 1990 for 49 countries for which all necessary data
are available. The list of the countries and the same type of accounting for each individual country
are shown in Table A.2 of the Appendix.
**Differences from the mean value of Gini coefficients for a sample of 49 countries in 1990.
***Other factors include data characteristics and regional dummy.

The cross-sectional accounting of national differences in income inequality

grouped by regions is presented in Table 5. The average Gini coefficient for
African countries was 0.062 higher than the world average in 1990, and the
regression explains a difference of 0.061. Because African countries had greater
educational inequality and, more importantly, a lower level of education, their
Gini coefficient was predicted to be higher by 0.019. Lower income and lower
social expenditure in Africa also contributed to greater income inequality by 0.003
and 0.007, respectively.
The effect of income levels on regional differences in inequality is more vis-
ible in the Latin American and OECD countries. In Latin America, which has a
Gini coefficient 0.096 higher than the average, 0.005 of this difference is attributed
to the difference in educational factors, 0.032 to the difference in income, and
0.005 to the difference in social expenditure. In OECD countries educational and
income factors accounted for 0.020 and 0.041, respectively, out of a total differ-
ence of 0.081. Higher social expenditure in OECD countries also explained about
0.015 of the difference. Between regions, in Asia neither education nor income
significantly explains the relative equality of income distribution.
The main conclusion from this exercise is that income and educational factors
are important in explaining the cross-sectional differences in income inequality,
although we cannot explain the bulk of the differences by these factors alone.

5. CONCLUDING REMARKS
In searching for explanations of inequality around the world, one interesting
issue is why inequality differs so much across countries and between regions,
and how some countries have been able to reduce it. By understanding the main
determinants of income inequality, we can also find policy implications that
would help to reduce inequality. This paper looks closely at one of the main
412
determinants of income distribution, namely, education. Policymakers usually
argue that efforts in the educational field reduce income inequality.
This paper provides empirical evidence on how education and income relate
to inequality in a panel data set covering a broad range of countries for the period
from 1960 to 1990. We have also analyzed the effects of social expenditure. The
findings indicate that educational factorshigher attainment and more equal dis-
tribution of educationplay some role in changing income distribution. We also
find the Kuznets inverted-U relationship between income level and income
inequality, and we find a positive contribution of government social expenditure
to a more equal distribution of income.
However, we should emphasize that a significant proportion of the variation
in income inequality across countries and over time remains unexplained. The
significance of the regional dummies for Africa and Latin America indicates that
income distribution in the countries of these regions has been systematically less
equal than in those of other regions. Some studies have examined the effects of
macroeconomic factors on income distribution (De Gregorio, 1995; Sarel, 1997;
Bulr, 1998; Li, Squire, and Zou, 1998), but they do not take account of all
the educational factors we have examined in this paper. We plan to explore the
connections among income distribution, education, macroeconomic factors, and
government policy in subsequent research.
The small quantitative effects of educational expansion on income distri-
bution are due in part to the impact of educational expansion on the inequality
of educational attainment in the population. Therefore a policy to expand edu-
cation needs to focus closely on the inequality of education if the aim is to reduce
income inequality.

APPENDIX
TABLE A.1
REGRESSIONS FOR KUZNETS CURVE (DEPENDENT VARIABLE : GINI COEFFICIENT)

Regression Number and Period

Independent 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
Variables 196590 1960 1965 1970 1975 1980 1985 1990
I. Log specification
log gdp* 0.447 1.304 0.488 0.798 0.714 0.616 0.529 0.236
(0.113) (0.416) (0.341) (0.260) (0.203) (0.208) (0.192) (0.168)
[log gdp]2 0.030 0.085 0.033 0.052 0.047 0.040 0.034 0.016
(0.007) (0.026) (0.021) (0.016) (0.013) (0.013) (0.012) (0.010)
II. Square specification
gdp** 0.98 8.85 5.43 1.32 0.89 0.66 0.89 0.26
(0.49) (25.4) (16.9) (1.56) (1.04) (9.20) (0.80) (0.81)
2
[gdp] *** 1.29 27.8 5.15 22.1 16.2 2.76 0.57 2.53
(3.33) (27.2) (16.1) (13.6) (8.08) (6.36) (5.29) (4.99)
III. AnandKanbur specification
gdp** 1.11 3.52 1.55 2.02 1.73 1.45 1.18 0.73
(0.20) (0.96) (0.70) (0.54) (0.37) (0.32) (0.29) (0.27)
1gdp 40.9 159.8 48.3 88.5 77.3 71.0 70.0 7.78
(19.1) (63.5) (51.8) (38.7) (34.4) (43.7) (39.6) (29.8)
Notes: Standard errors are in parentheses.
*gdp corresponds to GDP per capita.
**Coefficient is multiplied by 105.
***Coefficient is multiplied by 1010.

413
 TABLE A.2
SOURCES OF CROSS-COUNTRY DIFFERENCES IN INCOME DISTRIBUTION, 1990

Explained by
Actual Predicted Social
Country Gini Gini Education Income Expenditure Other
Spain 0.150 0.074 0.005 0.012 0.017 0.039
Finland 0.148 0.099 0.026 0.046 0.010 0.016
Belgium 0.141 0.108 0.019 0.040 0.032 0.016
Canada 0.133 0.083 0.031 0.067 0.006 0.021
Bangladesh 0.120 0.034 0.026 0.030 0.009 0.098
Netherlands 0.114 0.105 0.018 0.039 0.032 0.016
Sri Lanka 0.108 0.053 0.002 0.039 0.005 0.098
Taiwan 0.108 0.081 0.010 0.000 0.004 0.075
Pakistan 0.097 0.043 0.017 0.030 0.009 0.098
Egypt 0.089 0.012 0.016 0.038 0.000 0.066
U.K. 0.086 0.116 0.018 0.040 0.015 0.043
Italy 0.084 0.072 0.001 0.035 0.020 0.016
Sweden 0.084 0.118 0.022 0.051 0.028 0.016
Uganda 0.079 0.017 0.027 0.026 0.008 0.024
Indonesia 0.078 0.039 0.010 0.038 0.011 0.098
Denmark 0.077 0.079 0.032 0.045 0.022 0.021
Norway 0.076 0.098 0.014 0.052 0.016 0.016
Korea, South (R) 0.073 0.014 0.023 0.012 0.009 0.011
Japan 0.059 0.045 0.021 0.048 0.004 0.021
Ghana 0.055 0.062 0.019 0.010 0.010 0.024
Mauritius 0.042 0.050 0.005 0.019 0.003 0.024
Portugal 0.041 0.008 0.013 0.005 0.010 0.016
U.S. 0.031 0.094 0.038 0.073 0.004 0.021
Singapore 0.019 0.026 0.004 0.029 0.010 0.011
Gambia 0.019 0.064 0.028 0.002 0.010 0.024
New Zealand 0.007 0.058 0.035 0.027 0.016 0.021
Guyana 0.007 0.031 0.001 0.020 0.002 0.008
Tunisia 0.006 0.007 0.017 0.039 0.003 0.066
Australia 0.008 0.063 0.029 0.049 0.005 0.021
Bolivia 0.012 0.062 0.011 0.035 0.009 0.008
Zambia 0.026 0.036 0.012 0.009 0.010 0.024
Venezuela 0.032 0.102 0.008 0.017 0.009 0.068
Peru 0.040 0.059 0.001 0.039 0.011 0.008
Philippines 0.041 0.007 0.005 0.036 0.011 0.048
Hong Kong 0.041 0.070 0.016 0.052 0.009 0.011
Costa Rica 0.052 0.112 0.004 0.037 0.004 0.068
Malaysia 0.075 0.002 0.003 0.025 0.009 0.038
Thailand 0.079 0.039 0.004 0.036 0.010 0.011
Dominican Republic 0.096 0.093 0.014 0.039 0.009 0.031
Mali 0.131 0.031 0.030 0.030 0.007 0.024
Kenya 0.135 0.062 0.017 0.010 0.011 0.024
Mexico 0.141 0.091 0.002 0.019 0.006 0.068
Guinea-Bissau 0.152 0.055 0.030 0.009 0.010 0.024
Panama 0.156 0.099 0.010 0.039 0.002 0.068
Zimbabwe 0.159 0.073 0.018 0.023 0.008 0.024
Chile 0.170 0.043 0.004 0.031 0.014 0.031
Guatemala 0.182 0.138 0.021 0.039 0.010 0.068
Brazil 0.187 0.126 0.014 0.033 0.011 0.068
South Africa 0.214 0.163 0.007 0.038 0.008 0.111
Notes:
Gini is the difference of Gini coefficients from the mean value of all 49 countries.
Other factors include data characteristics and regional dummies.

REFERENCES
Acemoglu, Daron, Matching, Heterogeneity, and the Evolution of Income Distribution, Journal of
Economic Growth, 2, 6192, March, 1997.

414
Adelman, Irma and Cynthia T. Morris, Economic Growth and Social Equity in Deeloping Countries,
Stanford University Press, Stanford, California, 1973.
Ahluwalia, Montek S., Income Distribution and Development: Some Stylized Facts, American
Economic Reiew, Papers and Proceedings, 66, 12835, May, 1976.
Anand, Sudhir and Ravi Kanbur, The Kuznets Process and the Inequality-Development Relation-
ship, Journal of Deelopment Economics, 40, 2552, February, 1993.
Atkinson, Anthony B., Bringing Income Distribution in from the Cold, Economic Journal, 107,
297321, March, 1997.
Barro, Robert, Inequality, Growth, and Investment, mimeo, Harvard University, Cambridge, Mas-
sachusetts, 1999.
Barro, Robert and Jong-Wha Lee, Sources of Economic Growth, Carnegie-Rochester Series on
Public Policy 40, 146, June, 1994.
, International Measures of Schooling Years and Schooling Quality, American Economic
Reiew, Papers and Proceedings, 86, 21823, May, 1996.
Becker, Gary S. and Barry R. Chiswick, Education and the Distribution of Earnings, American
Economic Reiew, Papers and Proceedings, 56, 35869, May, 1966.
Becker, Gary S. and Nigel Tomes, Human Capital and the Rise and Fall of Families, Journal of
Labor Economics, 4, S139, July, 1986.
Benabou, Roland, Education, Income Distribution, and Growth: The Local Connection, NBER
Working Paper 4798, National Bureau of Economic Research, Cambridge, Massachusetts, 1994.
, Inequality and Growth, NBER Macroeconomics Annual, 1192, 1996.
Bulr, Ales, Income Inequality: Does Inflation Matter? IMF Working Paper WP987, Inter-
national Monetary Fund, Washington, 1998.
Chenery, Hollis B. and Moshe Syrquin, Patterns of Deelopment, 19501970, Oxford University Press
for the World Bank, London, 1975.
Chiswick, Barry R., Earnings Inequality and Economic Development, Quarterly Journal of Eco-
nomics, 85, 2139, 1971.
De Gregorio, Jose, Comment, in R. Dornbusch and S. Edwards (eds), Stabilization, Economic
Reform and Growth, University of Chicago Press, Chicago, 1995.
De Gregorio, Jose and Se-Jik Kim, Credit Markets with Differences in Abilities: Education, Distri-
bution and Growth, International Economic Reiew, 41, 579607, August, 2000.
De Gregorio, Jose and Jong-Wha Lee, Education and Income Distribution: New Evidence From
Cross-Country Data, Development Discussion Papers No. 714, Harvard Institute for Inter-
national Development, Cambridge, Massachusetts (www.hiid.harvard.edupubddps.html),
1999.
Deininger, Klaus W. and Lyn Squire, A New Data Set Measuring Income Inequality, World Bank
Economic Reiew, 10, 56591, September, 1996.
, New Ways of Looking at Old Issues: Inequality and Growth, Journal of Deelopment
Economics, 57, 25987, December, 1998.
Durlauf, Steven N., A Theory of Persistent Income Inequality, Journal of Economic Growth, 1, 75
93, March, 1996.
Gallop, J., Steven Radelet, and Andrew Warner, Economic Growth and Income of the Poor,
mimeo, Harvard Institute for International Development, Cambridge, Massachusetts, 1998.
Galor, Oded and Daniel Tsiddon, The Distribution of Human Capital and Economic Growth,
Journal of Economic Growth, 2, 93124, March, 1997.
Galor, Oded and Joseph Zeira, Income Distribution and Macroeconomics, Reiew of Economic
Studies, 60, 3553, January, 1993.
Gottschalk, Peter and Timothy M. Smeeding, Cross-National Comparisons of Earnings and Income
Inequality, Journal of Economic Literature, 35, 63387, June, 1997.
Knight, John B. and Richard H. Sabot, Educational Expansion and the Kuznets Effect, American
Economic Reiew, 73, 11326, December, 1983.
Kuznets, Simon, Economic Growth and Income Inequality, American Economic Reiew, 45, 128,
March, 1955.
Li, Hongyi, Lyn Squire, and Heng-fu Zou, Explaining International and Intertemporal Variations
in Income Inequality, Economic Journal, 108, 2643, January, 1998.
Londono, Juan Luis, Poverty, Inequality, and Human Capital Development in Latin America, 1950
2025, World Bank Latin American and Caribbean Studies 2, World Bank, Washington, 1996.
Marin, A. and George Psacharopoulos, Schooling and Income Distribution, Reiew of Economics
and Statistics, 58, 3328, 1976.
Psacharopoulos, George, Unequal Access to Education and Income Distribution: An International
Comparison, De Economist, 125, 38392, 1977.
Psacharopoulos, George and Maureen Woodhall, Education for Deelopment: An Analysis of Inest-
ment Choices, Oxford University Press, New York, 1985.

415
Ram, Rati, Population Increase, Economic Growth, Educational Inequality, and Income Distri-
bution: Some Recent Evidence, Journal of Deelopment Economics, 14, 41928, April, 1984.
, Can Educational Expansion Reduce Income Inequality in Less-Developed Countries?
Economics of Education Reiew, 8(2), 1859, 1989.
Sarel, Michael, How Macroeconomic Factors Affect Income Distribution: The Cross-Country Evi-
dence, IMF Working Paper WP97152, International Monetary Fund, Washington, 1997.
Winegarden, C. R., Schooling and Income Distribution: Evidence from International Data, Eco-
nomica, 46, 837, 1979.

416