Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The American Economic
Review.
http://www.jstor.org
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
Inequalityand Growth:
WhyDifferentialFertilityMatters
We develop a new theoretical link between inequality and growth. In our model,
fertility and education decisions are interdependent.Poor parents decide to have
many children and invest little in education. A mean-preserving spread in the
income distributionincreases thefertility differentialbetween the rich and the poor,
which implies that more weight gets placed onfamilies whoprovide little education.
Consequently,an increase in inequality lowers average education and, therefore,
growth. Wefind that thisfertility-differentialeffect accountsfor most of the empir-
ical relationship between inequalityand growth. (JEL J13, 040)
How does the income distributionof a coun- which lowers average education. The fertility
try affect its rateof economic growth?We argue differential,in turn, is a function of the income
that to answer this question, it is essential to distribution. If the differential increases with
account for the fertility differentialbetween the inequality,countrieswith higher inequalitywill
rich and the poor. Our argumentis simple: the accumulate less human capital, and therefore
fertility differential matters because it affects grow slower.
the accumulationof human capital. Assuming We develop a growth model which captures
that we identify human capital with education, this channel from inequality to growth. Our
future human capital is a weighted average of model is related to Gerhard Glomm and B.
the education of today's childrenfrom families Ravikumar(1992), who analyze the effects of
in different income groups, with the weights public versus private education on growth in a
given by income-specific fertility rates. Poor model with fixed fertility. We use a similar
parentstend to have many childrenand provide overlapping-generationsframework,but model
little education. If the fertility differential be- endogenousfertility decisions along the lines of
tween the' rich and the poor is large, more Gary S. Becker and Robert J. Barro (1988).
weight is put on children with little education, Both fertility and education are thus chosen
endogenously. Parents face a quality-quantity
trade-off in their decision on children, and we
show that education increases with the income
* de la Croix:NationalFundfor ScientificResearch(Bel- of a family (richer families can afford more
gium), IRES and CORE, UniversiteCatholiquede Louvain, education), while fertility decreases with in-
Place Montesquieu3, B-1348 Louvain-la-Neuve,Belgium (e- come (the time cost of child rearingis high for
mail: delacroix@ires.ucl.ac.be);
Doepke:Departmentof Eco-
nomics, Universityof California,405 HilgardAvenue, Los
rich parents). The aggregate behavior of the
Angeles, CA 90095 (e-mail: doepke@econ.ucla.edu).We model depends on the initial distributionof in-
thank Costas Azariadis, Harold Cole, Roger Farmer,Gary come. Other things being equal, we find that
Hansen,Lee Ohanian,MartinSchneider,and two anonymous economies with a less equitable income distri-
refereesfor helpfulcommentsand suggestions.We also ben-
efitedfromdiscussionswith seminarparticipantsat the Stock-
bution have higher fertility differentials, accu-
holm School of Economics,the SED Meeting in Stockholm, mulateless humancapital,and have a lower rate
USC, GREQAM,the Universityof Namur,and the European of economic growth.A calibratedversion of our
UniversityInstitute.de la Croix acknowledgesfinancialsup- model with endogenous fertility choice shows
portfromthe BelgianFrench-speaking community(GrantNo. that this effect is quantitativelyimportantand
ARC 99/04-235) and the Belgian FederalGovernment(Grant
No. PAI P5/10); Doepke had supportfrom the NationalSci- accounts for most of the empirical relationship
ence Foundation(GrantNo. SES-0217051) and the UCLA between inequality and growth. In contrast, if
AcademicSenate. we impose fertilityto be constantacross income
1091
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1092 THEAMERICANECONOMICREVIEW SEPTEMBER2003
groups, the effects of inequalityon humancap- does not allow for long-rungrowth,the analysis
ital and growth are small. concentrateson nonlineardynamics during the
We also analyze the dynamic properties of transition to the steady state. In Oded Galor
the model. Since in our dynastic frameworka and Hyoungsoo Zang (1997), inequalityaffects
period correspondsto one generation, the dy- growththroughits effect on overall fertility and
namics of the model are to be interpretedas human capital. Financial market imperfections
changes which occur over a horizonof a century play a crucial role in their analysis. Olivier F.
or more. Here we find thatthe predictionsof the Morand (1999) has a model of inequality and
model are similar to broadpatternsof develop- fertility in which the sole motive for fertility is
ment observedin industrializingcountriesin the old-age support.He concentrateson the possi-
nineteenthand twentiethcenturies.For realistic bility of poverty traps when the initial level of
parametervalues, the interactionof fertility and humancapital is too low. Ourpaperalso relates
education decisions gives rise to nonmonotone to empirical studies of the growth-inequality
behavior in inequality and fertility: both vari- relationshipsuch as RobertoPerotti(1996) and
ables rise initially, and laterstartto fall. In other Barro (2000). Both Perotti and Barro find that
words, the model generates both a Kuznets demographicvariablesare importantfor under-
curve and a demographic transition. Thus, in standingthe growtheffects of the income distri-
addition to accounting for the cross-sectional bution,but once again differentialfertilityis not
relationshipbetween inequalityand growth, the considereddirectly.
model generates plausible implications for the In the following section, we introduce the
dynamic interactionof inequality, fertility, and model. Section II presentstheoreticalresults on
growth over long time horizons. the quality-quantitytrade-off and the long-run
The relationshipsbetween inequality, differ- dynamics of the model. In Section III we cali-
ential fertility, and growth postulated by the brateand simulatethe model to assess the quan-
model are supportedby empirical results. Mi- titative importance of the differential-fertility
chael Kremerand Daniel Chen (2000) examine channel,and to examine the implicationsfor the
the relationshipbetween inequality and differ- dynamic evolution of inequality, fertility, and
ential fertility. Using cross-country data, they growth.Empiricalevidence is discussed in Sec-
find that more inequalitytends to be associated tion IV, and Section V concludes.
with larger fertility differentialswithin a coun-
try. This supportsthe first part of our hypothe- I. The ModelEconomy
sis, linking inequalityto differentialfertility.To
examine the second part of our hypothesis, the Consider an economy that is populated by
link from differentialfertilityto growth,we add overlappinggenerationsof people who live for
a differential-fertility variable to a standard three periods: childhood, adulthood, and old
growth regression and find large significantef- age. Time is discrete and goes from 0 to oo.All
fects of differential fertility on growth. In the decisions are made in the adult period of life.
same regressions,the direct effect of inequality People care aboutadultconsumptionct, old-age
as measured by Gini coefficients is insignifi- consumption dt + 1, their number of children nt,
cant, once differentialfertility is included. and the human capital of children ht+1. The
The majorityof the existing literatureon in- utility function is given by:
equality and growth concentrates on channels
where inequalityaffects growththroughthe ac- ln(c,) + 13ln(dt+) + y ln(n h, +).
cumulationof physical capital(see RolandBen-
abou, 1996). To our knowledge, Paul G. The parameterf3 > 0 is the psychological dis-
Althaus (1980) is the only existing model that count factor and 7y> 0 is the altruismfactor.
analyzes the effects of differential fertility on The role of old-age consumptionis to provide a
growth. However, in Althaus' model fertility motive for savings and therefore generate an
differentialsare exogenously given, and the role endogenous supply of capital. Notice that par-
of humancapitalis not considered.Endogenous ents care about both the quantity nt and the
fertility differentials arise in Momi Dahan and quality ht + 1 of their children.Raising one child
Daniel Tsiddon (1998), but since their model takes fraction GE(0, 1) of an adult's time. An
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL 93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1093
B, = B(1 + p)(l-
T
where wt is the wage per unit of humancapital. (4) K)t.
We assume that the average human capital of
teachersequals the averagehumancapitalin the The parameterssatisfy B, 0 > 0, and -1 E (0,
populationht, so thateducationcost per child is 1). The presence of 0 guarantees that parents
given by etwtht. The assumption that teachers have the option of not educatingtheir children,
instead of parents provide education is crucial because even with et = 0 futurehuman capital
for generating fertility differentials. It implies remains positive. As in Peter Rangazas (2000),
that the cost of education is fixed and does not equations (3)-(4) are compatible with endoge-
depend on the parent's wage. Education is nous growthfor K = 1 - T,and with exogenous
thereforerelatively expensive for poor parents.1 growth otherwise. We will later explore the
In contrast,since raising each child takes a fixed implications of exogenous versus endogenous
amountof the parent's time, having many chil- growth for the long-run behavior of the
dren is more costly for parents who have high economy.
wages. Parents with high human capital and Productionof the consumption good is car-
high wages thereforesubstitutechild quality for ried out by a single representativefirm which
child quantity and decide to have less children operates the technology:
with more education.
The only friction in the model is that chil- Y = AKaLt-a
dren cannot borrow to finance their own ed-
ucation. Instead, education has to be paid for where Kt is aggregate capital, Lt is aggregate
by the parents. This assumption is made in labor supply, A > 0, and a E (0, 1). Physical
most studies of the joint determination of capital completely depreciates in one period.
fertility and education. In the real world, chil- The firm chooses inputs by maximizing profits
dren generally do not finance their own edu- Yt- wtLt - RtKt.
cation (at least up to the secondary level). Human capital is distributedover the adult
The budgetconstraintfor the old-ageperiodis: population according to the distributionfunc-
tion Ft(ht). Total population Pt evolves over
- time according to:
(2) dt+1 R,+ st.
1
Alternatively,we could have assumed that educationis Here I(-) is an indicator function, and it is
providedby parents,but thatthe parents'teachingproductivity
increaseswith their own humancapital.In the model devel- understoodthatthe choice variablesnt and ht + 1
oped by OmerMoav (2001) this assumptionleads to poverty are functions of the individualstate ht. Average
trapscharacterizedby high fertilityand low education. human capital ht is given by:
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1094 THEAMERICANECONOMICREVIEW SEPTEMBER2003
etn,h, dF,(h,) .
0
This last conditionreflects the fact that the time For a household that has enough humancapital
devoted to teaching is not available for goods 0
production. We are now ready to define an such that the condition xt > -- holds, there is
equilibriumfor our economy. an interior solution for the optimal education
level, and the first-orderconditions imply:
Definition 1: Given an initial distribution of
human capital Fo(ho), an initial stock of phys-
ical capitalKo, and an initial populationsize Po, (10) St = 1+ 3+ 7 wth,,
an equilibriumconsists of sequences of prices
{wt, Rt}, aggregate quantities {Lt, Kt+1, ht,
Pt + }, distributionsFt + (ht + ), and decision nX, - 0
(11) et =
rules {ct, dt+ , st, nt, et, ht+1} such that: 1- n
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1095
limx,_on, 1
limx,oo,nt 1- (16) N, = nt dGt(x,),
0 o
This relationshipwill turn out to be helpful to
interpret the role of the parameter 7 and to
calibrateits value. 1 X
-
The results derived so far reflect the main (17) G,+ 1(x) = n,I(x,+I <- x) dG,(x,),
effects of inequality on growth that we are in- 'J o
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1096 THEAMERICANECONOMICREVIEW SEPTEMBER2003
L, 1f+ +
(1 (/)x if K = 1 -T
) dG,(x,)
Io (1 + x (endogenousgrowth),
g*,= , 1-q
1+p otherwise
'o
h, (exogenousgrowth),
J 6/
and the growthfactor of population is:
which leads to:
(1
( )1- y
L, 1+ ((>- 0)(1 + + 7)'
(22) + he
1 1+(
Ptht-
PROOF:
Using (8), (10), (19), and (22), the capitalstock See Appendix A.
evolves accordingto the followinglaw of motion:
Along this balanced growth path, there is no
longer any inequality among households. This
1 holds because we have assumedthathouseholds
(23) k,+ =1 +(23)
13 N A(1
- a)kta.
differ only in theirinitial level of humancapital.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITY
AND GROWTH 1097
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1098 THEAMERICANECONOMICREVIEW SEPTEMBER2003
3 Since
2
Note that beyond the bifurcationpoint the number of convergenceto the balancedgrowthpathis slow,
fixed points does not change, whereas in a saddle-node the model still allows for substantialpopulationgrowth for
bifurcationtwo fixed points either appearor disappear. long time periods.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 AND GROWTH
DE LA CROIXAND DOEPKE: INEQUALITY 1099
cost parameter 4)for having a child determines with respect to educationof 0.6 in the balanced
the overall opportunitycost of children. Evi- growthpath. This numberis within the range of
dence in Robert Haveman and BarbaraWolfe estimates of the elasticity of earnings with re-
(1995) and John Knowles (1999) suggests that spect to schooling. Specifically, estimatesof the
the opportunitycost of a child is equivalent to returnto an additionalyear of schooling range
about 15 percent of the parents' time endow- from 7.5 percentin JoshuaD. Angrist and Alan
ment. This cost only accruesas long as the child B. Krueger (1991) to 12 to 16 percent in the
is living with the parents. If we assume that study of twins by Orley Ashenfelter and
children live with parentsfor 15 years and that Krueger(1994). The surveys by George Psacha-
the adult period lasts for 30 years, the overall ropoulos (1994) and Krueger and Mikael Lin-
time cost should be 50 percent of the time cost dahl (2001) report estimates of the return to
per year with the child present.Accordingly,we schooling in developed countries of 8-10 per-
choose 4)= 0.075. The parameter4)also sets an cent, with higherestimatesfor developing coun-
upper limit on the numberof children a person tries and low levels of schooling. Assuming that
can have. With our choice, a person spendingall an additionalyear of schooling raises education
time on raising children would have a little expenditureby 20 percent, these returnstrans-
above 13 children. A family of two could have late into an earningselasticity of schooling be-
a little under 27 children.4 tween 0.4 and 0.8. The elasticity implied by our
The parameter71 influences the elasticity of parameterchoices is exactly in the middle of
humancapital with respect to education,as well this range.
as the maximum fertility differential in the The remaining parameters K and T do not
economy. Specifically, the maximum differen- influence individual decisions, but still have an
tial written as a ratio is given by 1/(1 - 1). In effect on growthrates.The elasticity Kof future
the data set that we use below, the highest human capital with respect to average human
fertility ratio between women at the lowest and capital h can be calibratedto evidence on the
the highest education levels is 2.74 (Brazil). effects of the quality of schooling. We interpret
This differential is achieved by choosing 71 = the education choice et as the quantity of
0.635. Our choice of 17guaranteesthat realized schooling (correspondingto years of schooling
fertility differentialsin the model never exceed in the data) while h measures the quality of
the maximum differentialobserved in the data, schooling, since it is the average humancapital
which ensures that the role of the differential- of teachers.Comparedto the quantityof school-
fertility channel is not inflated. At the same ing, the quality of schooling (such as spending
time, for evaluating the differential-fertility per pupil at a given level of education)has been
channel it is also importantthat the elasticity of shown to have smaller effects on earningswith
future human capital with respect to education an elasticity of around0.1; see David Cardand
is calibratedrealistically.In the model, this elas- Krueger (1996) and Krueger and Lindahl
ticity is determinedjointly by 17and 0. Since 0 (2001). In line with evidence on the effect of the
enters the education choice of parents,it deter- quality of education, we set K = 0.1. Alterna-
mines aggregateexpenditureson education.We tively, K could also be interpretedas a measure
choose 0 such that in the balanced growth path of human capital externalities. Existing evi-
total educationexpenditureas a fractionof GDP dence [see DaronAcemoglu and Angrist (2001)
matches the correspondingvalue in U.S. data, and Kruegerand Lindahl (2001)] suggests that
which is 7.3 percent.5 The implied parameter these externalities are small as well, i.e., the
value is 0 = 0.0119. Our combined choices for social returnto human capital accumulationis
17and 0 imply an elasticity of human capital only slightly largerthan the privatereturn,con-
firming our low choice of K. Our results are
robust with respect to the choice of this elastic-
4 If we also modeled a
goods cost of having children,the ity in the sense that K matters only for the
upper bound would be lower, and close to the maximum determination of the growth rate of average
human fertility levels observed so far.
5 This figure (Digest of Education Statistics, 1998, U.S. human capital. Individual decisions and the
Departmentof Education)does not include on-the-jobtrain- evolution of inequality,fertility,and differential
ing, since it is not partof the parentalinvestmentin children. fertility are independentof K.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1100 THEAMERICANECONOMICREVIEW SEPTEMBER2003
TABLE1-INmrrAL
GROWTH ENDOGENOUS
WITH ANDEXOGENOUS
FERTILITY
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITY
AND GROWTH 1101
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1102 THEAMERICANECONOMICREVIEW SEPTEMBER2003
between inequality and growth. Since the em- function for human capital, since for long-run
pirical estimates carry sizable standarderrors, analysis the assumptionthat the only source of
the findingdoes not rule out that other channels inequality is the initial dispersion in human
could also play a role, but clearlythe differential- capital is less attractive. We find that adding
fertility effect appearsto be important. shocks slows growth, but leaves the qualitative
behavior of the model intact. Thus our model
C. The Dynamics of Inequality,Differential turnsout to be successful at accountingfor both
Fertility, and Growth the cross-sectional relationship between in-
equality and growth, and the dynamic interac-
We now turn to the dynamic implications of tion between inequality, fertility, and growth
our model. So far, we have only analyzed the over long time horizons.
effects of inequalityon growthduringthe initial Figure 4 shows the evolution of the growth
period. Since in our dynasticmodel a periodhas rate of human capital, the population growth
a length of 30 years, even the initial growth rate, inequality,and differentialfertility for two
effect extends over a long horizon, and conse- different values of r. Growth rates are annual-
quently the dynamics of the model are to be ized. The initial distributionof humancapital is
interpretedas changes which occur over a ho- assumedlognormalwith o2 = 1, corresponding
rizon of a century or more. We thereforeeval- to an initial Gini coefficient of 0.5. With a low
uate the dynamicbehaviorof the model relative T of 0.05, fertility, inequality, and differential
to the evolution of income, fertility, and in- fertilityconverge monotonicallyto their steady-
equality in industrializingcountries in the last state values. Initially,inequalityreducesgrowth
200 years. A centralfeatureof the data for this below its balancedgrowthvalue for the reasons
period is that the behaviorof populationgrowth explained above, but subsequently the growth
and inequalityis nonmonotone.As a benchmark rate increases. Low initial growth implies that
case, consider England, the first country to in- the capital stock increases more slowly than in
dustrialize.Fertility rates increased until about the balanced growth path. Since productivity
1830 and started to decline rapidly only after growth is exogenous, the effective capital stock
1870 (Jean Claude Chesnais, 1992). Income falls, and the usual transitional dynamics in
inequality followed a similar pattern, with in- exogenous growth models result in higher sub-
creasing inequalityuntil about 1870 and a rapid sequent growth. If T is raised to 0.2, the model
decline afterwards (Jeffrey G. Williamson, generates the hump-shapepatternsthat charac-
1985). The growthrate of income per capita, in terize the data. Fertility and inequalityfirst rise
contrast, does not display a hump shape. and then decrease, and the growth rate remains
Growth rates were essentially zero before the below its balanced growth value for several
industrialrevolution and then increased slowly periods as long as inequality remains high. A
throughoutthe nineteenthcentury(Angus Mad- moderate degree of intergenerationalpersis-
dison, 2001). Similar patternscan be observed tence in human capital is thus essential for
for Western Europe as a whole and, startinga matching the long-run evolution of inequality
little later, in the United States. and fertility to data.
To evaluatehow our model performsrelative The nonmonotonebehaviorof inequalityand
to these facts, we simulate the model with the fertility is related to the comer solution for
baseline calibrationover a horizon of eight pe- education. A fraction of the people in the first
riods, correspondingto 240 years. The main period decides not to invest in education.Since
finding is that for plausible parameters, the this group has the highest fertility rates, their
model can generatea hump shape in inequality childrenmake up an even largerfraction of the
and the populationgrowth rate that looks very populationin the next period. If there is a suf-
similar to the data. This is a surprisingfinding, ficient degree of intergenerationalpersistence,
since unlike most existing theories the model these childrenwill be at the comer solution for
generates the patternwithout requiringany ex- education as well. This leads to an increase in
ogenous change to the economic environment. populationgrowth in the first periods and slow
We also investigatethe behaviorof the model if growth of human capital, since investment in
idiosyncraticshocks are addedto the production education is low. Because of exogenous tech-
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1103
g
0.025 N
* .*- .. * =0.05
0.0225 ..^ 0.012
0.02 0.01
l* *.
0.0175 0.008
0.015 0.006
0.0125 0.004
...
0.01 0.002
0.0075 . . . . t 2 3 4 5 6 7 8
2 3 4 5 6 7 8
D
3-
2.5
2-
1.5-
1-
*-
nological change and human capital externali- small, i.e., we need some degree of intergen-
ties, however, after a few periods even the erational persistence in human capital and
dynasties that initially did not invest in educa- earnings.8
tion find it optimal to starteducatingtheir chil- The main disparity between the model and
dren. From this point on, inequality and the data is that in the data, growth rates were
population growth fall. slowly increasing throughout the nineteenth
This nonmonotone behavior only occurs if century, whereas the model produces a hump
the initial distribution of human capital is shape with firstrapidlyincreasingand then fall-
such that some people choose not to invest in ing growth rates. This behavior of the model is
education. If everyone is above the threshold generated by transitionaldynamics due to the
initially, convergence to the steady state is exogenous growth assumption. We therefore
monotone even for T = 0.2. For parameters also explore how the model performs if we
that lead to an initial rise in fertility and
inequality, the time paths for inequality,
growth, and fertility look surprisingly similar 8A unique equilibrium exists even if we increase T
to the patterns of development in Western above the bifurcationvalue of i = 0.246, where we enter
Europe between 1800 and 2000 described by the region of the parameterspace in which the dynamics of
Galor and David N. Weil (2000) and others. individualhumancapitalare no longer stable. The returnsto
parental human capital in the education function are not
For example, let us assume that t = 1 in the sufficiently decreasing to compensate the centrifugalforce
graphs corresponds to the period 1760-1790 of the quality-quantitytrade-off. Computationsof the dis-
in England. Then population growth peaks tributionof human capital after a large number of periods
around 1790-1820, and inequality peaks 30 indicate that an ever-decreasing share of the population
accumulatesan ever-increasingshareof humancapital.The
years later. The timing of the subsequent fall mass of people with above-averagehuman capital tends to
is close to what is observed in the data. For zero, but the fraction of human capital accounted for by
the hump to occur, however, T cannot be too them tends to one.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1104 THEAMERICANECONOMICREVIEW SEPTEMBER2003
0.025
g
the model by introducingidiosyncratic ability
*
shocks in the production function for human
-
0.02
*
K=1-t
capital. With ability shocks, inequality persists
* K=0.1
even in the long run.
The mainoutcomeof oursimulationswith abil-
0.01
-A - data
ity shocksis thatinequalitydecreasesmoreslowly
0.005 &
t
andgrowthis reduced,which was to be expected,
6 3 47 5 8
2 3 4 5 6 7 8
but that otherwisethe qualitativefeaturesof the
VERSUS
5. EXOGENOUS
FIGURE GROWTH
ENDOGENOUS transitionremainintact.The quantitativeimpact
of abilityshocks is rathersmall as well, provided
thatthe shocksarecalibratedto reproducecurrent
increasethe humancapitalexternalityfrom K = levels of earnings inequality in the United
0.1 to K = 1- T = 0.8 to allow for endogenous States.10Moreover,as can be seen in Figure6, the
growth. As stressed above, individualdecisions differencematerializesonly afterseveralperiods,
and the evolution of inequality, fertility, and since the ability shocks have little additionalim-
differential fertility are independentof the as- pactas long as the distributionof humancapitalis
sumption on K. However, the dynamic pattern widely dispersed.
of the growth rate can be affected.9 Figure In summary,we find that our relatively sim-
5 comparesgrowthratesunderthe two different ple model is surprisinglysuccessful at reproduc-
regimes with U.K. data from Maddison(2001). ing key features of the long-run evolution of
While growth with K = 0.1 increases quickly inequality, fertility, and growth. Under a mod-
and soon exceeds its balanced growth value, erate degree of intergenerationalpersistence in
with K = 1 - r growth converges slowly and human capital (which could be generated by
monotonicallyto its long-runlevel. The pattern persistence in innate ability), the model gener-
under the endogenous growth assumption is ates a hump shape in inequalityand fertility. In
closer to the observedpatternin England,where order to account for slowly increasing growth
the average growth rate of 2 percent was rates of income per capita, it is helpful to intro-
reachedonly towardsthe beginningof the twen- duce an endogenous growth mechanism.While
tieth century.Notice, however, that to generate in our model endogenous growth is generated
endogenous growth we have to assume a much by human capital externalities, we conjecture
higher degree of externalitiesthan suggested by that similarresults could be obtainedwith other
empirical estimates in recent data. The result sources of endogenous growth.
thereforeindicatesthateitherexternalitiesplayed Our results complement existing theories of
a bigger role in the past, or that anotherendoge- long-run growth. Relative to the literature,the
nous growthmechanismwas at work which gen- main novelty is that we link the evolution of
eratedthe slowly increasinggrowthrates. growth and population to the income distri-
So far, we have assumed that the only source bution. In contrast, the models developed by
of inequality is the initial dispersion in human Galor and Weil (2000), Raouf Boucekkine et
capital. As a consequence, inequality is transi- al. (2002), and Gary D. Hansen and Edward
tory in the model and disappearsin the balanced C. Prescott (2002) abstractfrom distributional
growth path. While this abstraction is of no issues.11 We find that allowing for inequality
consequence for the initial growth effects, for
long-run dynamics idiosyncratic shocks would 10The
be expected to be more important.To check the ability shock is multiplicativeand is drawnfrom
robustnessof our results, we thereforeextended a uniform distributionover [0.9, 1.1], which generates a
long-term Gini coefficient of 0.35. We assume that the
ability shock is uncorrelatedwith parental human capital,
since the parameter T already captures intergenerational
9 To calibrate the endogenous growth version of the transmissionof skills.
"
model, we keep T at 0.2 and set K = 0.8. The overall Doepke (2001) has a model of the industrialrevolution
productivityB in the productionfunction for humancapital and the demographictransitionwhich does allow for inequal-
now governs the growth rate of outputper capita. We pick ity. However, since there are only two types of agents, the
B = 0.367 which ensures a long-term growth rate of 2 income distributionhas just two points.A humpin inequality
percent per year. arisesonly if thereare exogenouspolicy changes.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1105
g I
FIGURE
6. GROWTH
ANDINEQUALITY
WITHANDWITHOUT
ABILITY
SHOCKS
in human capital combined with endogenous evidence? The first part of our hypothesis,
fertility and education choice generates real- that income inequality leads to high fertility
istic predictions for inequality, fertility, and differentials, has been analyzed by Kremer
growth in a simple and naturalway. Galor and and Chen (2000). In line with our conjec-
Weil (2000) generate a hump in fertility by ture, they find that Gini coefficients have a
introducing a subsistence level of consump- significant and sizable positive correlation
tion, but the evolution of the income distri- with fertility differentials. In this section we
bution is not explained. In Hansen and examine the second part of our hypothe-
Prescott (2002) fertility is exogenous, so nei- sis, the link from fertility differentials to
ther the hump in fertility nor in inequality are growth. Our approach is to introduce a
accounted for. A limitation of our approachis differential-fertility variable into a standard
that we take the initial conditions at the start growth regression. The analysis is designed to
of the industrial revolution as given. For a full be comparable to recent empirical studies of
account of the evolution of the economy from inequality and growth. As predicted by our
pre-industrial stagnation to modem growth model, we find that differential fertility has a
we would have to add an element to the model negative effect on growth. Moreover, when
that generates the initial stagnation phase. We the differential-fertility variable is present,
suspect that this could be done along the lines the Gini variable is no longer significant in
of Galor and Weil (2000) or Hansen and the regression.
Prescott (2002), but the extension is beyond
the scope of the current paper. A. Data
A main implication of our dynamic analysis
for the inequality-growth relationship is that Our sample contains 68 countries for which
the relationship can be modified by transi- data on fertility differentials is available. The
tional dynamics. Since inequality first in- dependentvariable(GR) in all regressionsis the
creases and then decreases during the average annual growth rate of GDP per capita
transition, inequality does not map one-to-one over the periods 1960 to 1976 or 1976 to 1992
into growth rates. In the empirical analysis, it (the period depends on the availabilityof fertil-
is therefore important to control for transi- ity data).The GDP data is from the Penn World
tional dynamics to isolate the role of the Tables, and growthrates are continuouslycom-
differential-fertility channel. pounded and expressed as percentages.' Since
we are interestedin long-rungrowth, we chose
IV. EmpiricalEvidence
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1106 THEAMERICANECONOMICREVIEW SEPTEMBER2003
TABLE2-DESCRIPIIVE
STATISTICS
Standard
Sample Observations Variable Mean deviation Minimum Maximum
1960-1976 40 GR 1.95 3.65 -5.75 8.44
GINI 44.32 11.14 23.38 68.00
TFR 5.56 1.89 2.02 7.93
DTFR 2.23 1.56 0.22 5.30
1976-1992 43 GR 0.39 1.89 -3.46 4.97
GINI 45.91 9.56 28.90 69.00
TFR 6.06 1.08 3.37 8.00
DTFR 2.41 0.99 0.10 4.50
Total 83 GR 1.14 2.97 -5.75 8.44
GINI 45.14 10.32 23.38 69.00
TFR 5.82 1.54 2.02 8.00
DTFR 2.32 1.29 0.10 5.30
Notes: GR: Annualized growth rate of income per capita in percent. GINI: Initial income
inequality.TFR:Total fertilityrate.DTFR:Fertilitydifferentials.Detaileddefinitionsanddata
sources are in the main text.
the longest subsample periods available in the countries (AFR), and the initial total fertility
Penn World Tables. rate (TFR). I/GDP and G/GDP are from the
Following Kremer and Chen (2000), for Penn World Tables, the income Ginis are from
fertility differentials we rely on information Klaus Deininger and Lyn Squire (1996), and
from the World Fertility Survey and the De- total fertilityrates and life expectancy (which is
mographic and Health Surveys on total fertil- used as an instrument)are from the Barro-Lee
ity rates by women's educational attainment data set.13The inclusion of initial GDP and the
[see Elise F. Jones (1982); United Nations investmentratio is importantto controlfor tran-
(1987, 1995); Gora Mboup and Tulshi Saha sitional dynamics.
(1998)]. For countries that participated in the One shortcoming of the data set is that the
World Fertility Survey, the independent vari- observations on fertility differentials are close
able is growth in GDP per capita in the first to the end of the period over which we compute
period, and for countries that participated in growthrates. Since the fertilityobservationsare
the Demographic and Health Surveys the left- five-yearaveragesandresultfrom decisions and
hand-side variable is growth over the second actions taken even earlier, the endogeneity
period. Our differential fertility variable problem is not too severe. We correct for po-
DTFR is the difference in the total fertility tential endogeneity of the differentialsby using
rate between women with the highest and the instrumentalvariables.
lowest education level. For some countries we Table 2 provides descriptive statistics for
have two observations from the Demographic the main variables in our analysis. The two
and Health Surveys, in which case we aver- subsamples are similar, except that the aver-
aged the two resulting values. For 25 coun- age growth rate is much lower in the second
tries we have only observations for the first sample. Since we will allow for different con-
period, for 28 countries we have only obser-
vations for the second period, and for 15 13
Where possible, the Gini coefficients are from the
countries there are observations from both initial year; otherwise we used the closest available year.
periods. For a few countries, no data is available in Deininger and
The remainingindependentvariablesare ini- Squire (1996). For Benin, Burundi, Central Africa, and
tial GDP per capita (GDP), the average ratio of Namibia,we relied on the EconomicReporton Africa 1999:
investment to GDP (I/GDP), the average ratio The Challenges of Poverty Reduction and Sustainability,
United Nations (2000). For Haiti and Syria, only land Ginis
of government expenditure to GDP (G/GDP), from IdrissJazairyet al. (1992) are available.We regressed
the initial Gini coefficient for the income distri- the available income Ginis on the land Ginis (correlation:
bution (GINI), a dummy variable for African 0.61) and used the predictedvalues.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1107
TABLE3-GENERALIZEDMETHODOF MOMENTSESTIMATION
Regression
Independent
variable (1) (2) (3) (4)
ConstantA 12.35** (1.31) 12.79** (1.33) 15.30** (1.46) 13.92** (1.69)
ConstantB 10.41** (1.36) 10.98** (1.38) 13.40** (1.45) 12.18** (1.63)
ln(GDP) -1.33** (0.17) -1.21** (0.16) -1.37** (0.15) -1.55** (0.20)
I/GDP 0.14** (0.02) 0.13** (0.02) 0.07** (0.03) 0.08** (0.04)
G/GDP -0.08** (0.03) -0.07** (0.03) -0.05* (0.03) -0.05* (0.03)
AFR -1.75** (0.35) - 1.80** (0.35) - 1.95** (0.32) -2.41** (0.44)
GINI -0.03** (0.01) 0.02 (0.03) 0.06 (0.05)
ln(TFR) -1.84** (0.87) -1.01 (1.01)
ln(DTFR) - 1.22** (0.50)
J-test 17.71 [0.48] 17.11 [0.45] 16.79 [0.40] 9.58 [0.85]
LRI 5.53 [0.01]
LR2 2.08 [0.35]
Notes: The variablesare describedin more detail in the text. The dependentvariableis the growthrate of real per capitaGDP.
Estimationby GMM. The instrumentsare:constant,log of initial GDP per capita,log of initial GDP per capitasquared,initial
investment/GDPratio, initial governmentspending/GDPratio, initial fertility, initial fertility squared,initial life expectancy,
initial life expectancy squared,Africa dummy, and the tropics and access to the sea variablesof Jeffrey Sachs and Andrew
Warner(1997). Heteroskedasticity-consistentstandarderrorsare reportedin parentheses.J-test is the test for overidentifying
restrictionsof Lars P. Hansen (1982), asymptoticallyX2 distributedwith n degrees of freedom, where n is the number of
overidentifying restrictions. Correspondingp-values are reportedin brackets. LR, is a quasi-likelihood ratio test for the
absence of the differentialfertilityin the equation.LR2 is the test for the absence of both Gini and total fertility.The statistics
are computed as the normalized difference between the constrainedand the unconstrainedobjective function, where the
weighting matrix is provided by the unconstrainedestimation (see Donald Gallant, 1987). The correspondingp-values are
reportedin brackets.
* Significant at the 10-percentlevel.
** Significant at the 5-percent level.
stant terms in the two subsamples, this differ- reproduces standardresults in the growth re-
ence will not play a role in the results. These gression literature:the investment rate has a
will thus reflect cross-sectional differences positive effect, whereas the government share,
among countries, as well as variation over initial GDP, and the African dummy have neg-
time within countries. ative effects on growth.Regression (2) adds the
Gini coefficient. The estimated coefficient is
B. EstimationResults significantly negative, and its value is close to
Barro's estimate. The value of the parameter
Table 3 contains our estimationresults. In all implies that an increase in the Gini of 0.4
cases, the left-hand-sidevariable is the average (roughly the range of variation in the data)
growth rate of GDP per capita. The regression lowers growth by about 1.2 percent per year.
equation is estimated with the Generalized However, when the total fertility rate is in-
Method of Moments (GMM). For countriesthat cluded [regression (3)], the coefficient on the
are presentin both sample periods,we allow the Gini coefficient changes sign and becomes in-
error term to be correlatedacross the periods, significant, which is also in line with Barro
and we use instrumentalvariablesto correctfor (2000).
possible endogeneity of I/GDP, G/GDP, GINI, Regression(4) includesthe differential-fertility
and DTFR. We allow the constant to differ variable.The coefficient on differentialfertility
across the periods (Constant A for the early is significantlynegative. The point estimate im-
period and ConstantB for the late period). Our plies that an increase in the fertility differential
regressions are designed to be comparable to from one to two would lower growth by 0.8
Barro (2000), but we include fewer variable percent per year. With differential fertility in-
because of the small sample size. Regression (1) cluded, the coefficients on both Gini and TFR
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1108 THEAMERICANECONOMICREVIEW SEPTEMBER2003
4-GMM ESTIMATION
TABLE GDP ANDCROSSEFFECTS
wrTHSQUARED
Regression
Independent
variable (1) (2) (3) (4)
ConstantA 9.88 (10.1) -6.70 (17.5) -3.31 (18.8) 6.72 (19.9)
ConstantB 7.95 (10.1) -8.48 (17.4) -4.77 (18.6) 479 (19.7)
ln(GDP) -0.65 (2.73) 4.65 (4.12) 2.37 (4.09) 1.39 (4.26)
ln(GDP)2 -0.04 (0.18) -0.40 (0.24) -0.15 (0.20) -0.26 (0.22)
I/GDP 0.14** (0.02) 0.10** (0.02) 0.07* (0.04) 0.08** (0.04)
G/GDP -0.08** (0.03) -0.08** (0.03) -0.07** (0.04) -0.07** (0.03)
AFR -1.73** (0.36) -1.64** (0.46) -2.23** (0.43) -2.10** (0.52)
GINI -0.13 (0.22) 0.31 (0.27) -0.18 (0.37)
ln(GDP)*GINI/100 -0.01 (0.03) -0.04 (0.04) 0.03 (0.05)
In(TFR) -2.20* (1.13) 0.14 (1.51)
ln(DTFR) - 1.56** (0.68)
J-test 17.63 [0.41] 13.87 [0.54] 16.58 [0.28] 8.38 [0.82]
LRl 5.54 [0.02]
LR2 0.52 [0.77]
Notes: The variablesare describedin more detail in the text. The dependentvariableis the growthrate of real per capitaGDP.
Estimationby GMM. Instrumentsand test statistics as in Table 3. Heteroskedasticity-consistentstandarderrorsare reported
in parentheses.
* Significantat the 10-percentlevel.
** Significantat the 5-percent level.
are insignificant,and the point estimate on the Moreover, there is a large improvement in the
Gini is positive. value of the test when differential fertility is
Based on the results in Section III, our model introduced. The significance of the differen-
predictsthat Gini, total fertility, and differential tial fertility variable is verified both by its
fertility should all be equally negatively related t-statistic and by the quasi-likelihood ratio
to growth. It is therefore not clear why the test LR . The test LR2 of joint insignificance
coefficients on the Gini and the total fertility of GINI and TFR is not rejected.
rate become insignificantonce differentialfer- Barro (2000) argues that there are important
tility is introduced.One possibility is that in- nonlineareffects that relate the levels of devel-
equality and total fertility are influenced by opment and inequality to growth rates. He
other factors which do not affect growth, while shows that unless these nonlinear effects are
differentialfertility is observed with less noise. addressed,the effects of inequality and fertility
A second possibility is that total fertility and on growth are difficult to distinguish.To check
inequality have other effects on growth, which whetherour findings are robust with respect to
are not present in our model and do not work the inclusion of nonlinear terms, we add a
through differential fertility. If some of these squaredterm for GDP per capita and an inter-
effects on growth are positive and therefore action terminvolving GDP and inequalityto the
offset the negative effects, it would be plausible explanatoryvariables. Table 4 presents the re-
that the overall effect of total fertility and the sults. The new terms are never significant and
Gini becomes insignificantonce the differential- do not affect the significance of the J-tests.
fertility channel is controlled for. Given our relatively small number of observa-
Hansen's J-test measures how close the tions, the inclusion of additionalvariables low-
residuals are to being orthogonal to the in- ers the significance of the other variables. In
strument set. It can be seen as a global spec- particular, the Gini coefficient is now never
ification test. The degrees of freedom equal significant, but total fertility is significant in
the number of restrictions imposed by the regression (3). Our main conclusion remains:
orthogonality conditions. These restrictions when differentialfertility is added in regression
are never rejected at the 5-percent level. (4), it is significant, while total fertility is not.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL.93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1109
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1110 THEAMERICAN REVIEW
ECONOMIC SEPTEMBER
2003
1
APPENDIXA: PROOFOF PROPOSITION
l+p I 1 1- )
and:
- (
X3(+ 0)(1 + + y)A(l - a)1(1
(
k- -a)
Here g* and /* are the balanced growth path values. In the case K = 1 -T we replace g* by
B(q(o - 0)/(1 - 1))T. In the case K 0 1 - r, we replace g* by 1 + p and I* by {[BI(l +
p)][q(p _- 0)/(1 - q)]}1/(1 -K--T). Both substitutions lead to the same expression, i.e., equation
(24) of the main text. The expression can be rewrittenas:
-
0(1 - 7) 0qx-0-'l
I(x; T) = Xt () 0) + max 0 (- 0)) -
We first consider the limits of this function. We have: AP(0;T) = 0 and P'(0; T) = + oo, and:
lim P(x; T) =
XT+ - X,
' c- O-
x--
which implies:
Hence the function T starts from (0, 0) with an infinite slope and goes either to -oo or +00
depending on parametervalues.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL 93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1111
By definition of the aggregatebalanced growth path, x = 1 is a steady state and thus T(1; T) =
0. This steady state is locally stable if and only if x'(1; T) < 0, i.e.,
71
T+ --0 o 1 <0.
At the point:
T= 1-
the dynamics of individual capital described by: xt+ - xt = (xt; ) undergo a transcritical
bifurcation,as proved in Proposition2. There are thus two steady-stateequilibria, 1 and x, near (1,
iT)for each value of T smaller or larger than T. The equilibrium1 (resp. x) is stable (resp. unstable)
'
for r < and unstable (resp. stable) for T > T.
Another point of interest is xt = ?; If, at this point, the function T is negative, it crosses
the horizontalaxes between 0 and 0. The existence of this steady stateresultsfrom the infiniteslope
of T at 0 and from its continuity;uniquenessresultsfrom the concavity of the functionin the interval
(0, --). We evaluate:
P 7 T,) = (2 -))( ) -7
We are now able to fully characterizethe dynamics of x as a function of the parameterT. The
bifurcationdiagramis presentedin Figure 2.
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
1112 THE AMERICANECONOMICREVIEW SEPTEMBER2003
Card, David and Krueger,Alan B. "School Re- Economic Literature,December 1995, 33(4),
sources and Student Outcomes: An Overview pp. 1829-78.
of the Literature and New Evidence from Jazairy,Idriss; Alamgir,Mohiuddinand Panuc-
North and South Carolina."Journal of Eco- cio, Theresa.Thestate of world ruralpoverty:
nomic Perspectives,Autumn 1996, 10(4), pp. An inquiryinto its causes and consequences.
31-50. New York: New York University Press,
Chesnais,Jean Claude. The demographic tran- 1992.
sition. Oxford: Oxford University Press, Jones, Elise F. Socio-economic differentials in
1992. achieved fertility. World Fertility Survey
Dahan, Momi and Tsiddon, Daniel. "Demo- Comparative Studies No. 21. Voorburg,
graphic Transition, Income Distribution, and Netherlands:InternationalStatistics Institute,
Economic Growth." Journal of Economic 1982.
Growth,March 1998, 3(1), pp. 29-52. Kelley, Allen C. and Schmidt,Robert M. "Eco-
Deininger, Klaus and Squire, Lyn. "Measuring nomic and DemographicChange: A Synthe-
Income Inequality: A New Database." Har- sis of Models, Findings, and Perspectives."
vard Institute for International Development Duke University Departmentof Economics
(Cambridge, MA) Development Discussion Working Paper No. 99/01, 1999.
Paper No. 537, 1996. Knowles,John. "CanParentalDecisions Explain
Digest of Education Statistics, 1998. U.S. Depart- U.S. Income Inequality?"Mimeo, University
ment of Education, 1999. of Pennsylvania, 1999.
Doepke, Matthias. "Accounting for Fertility De- Kremer,Michaeland Chen,Daniel."IncomeDis-
cline During the Transition to Growth." tribution Dynamics with Endogenous Fer-
UCLA Department of Economics Working tility." National Bureau of Economic
Paper No. 804, 2001. Research (Cambridge,MA) Working Paper
Gallant, Donald. Nonlinear statistical models. No. 7530, 2000.
New York: John Wiley & Sons, 1987. Krueger,Alan B. and Lindahl,Mikael. "Educa-
Galor, Oded and Weil, David N. "Popula- tion and Growth:Why andfor Whom?"Jour-
tion, Technology, and Growth: From nal of Economic Literature,December 2001,
Malthusian Stagnation to the Demographic 39(4), pp. 1101-36.
Transition and Beyond." American Eco- Leibowitz,Arleen. "Home Investmentsin Chil-
nomic Review, September 2000, 90(4), pp. dren."Journal of Political Economy,March-
806-28. April 1974, 82(2), pp. S 111-31.
Galor, Oded and Zang, Hyoungsoo. "Fertility, Maddison,Angus. The world economy: A mil-
Income Distribution, and Economic Growth: lennial perspective. Paris: OECD, 2001.
Theory and Cross-CountryEvidence."Japan Mboup,Gora and Saha, Tulshi. Fertility levels,
and the WorldEconomy,May 1997, 9(2), pp. trends, and differentials. Demographic and
197-229. Health Surveys Comparative Studies No.
Glomm, Gerhard and Ravikumar, B. "Public 28. Calverton, MD: Macro International,
Versus Private Investment in Human Capital: 1998.
Endogenous Growth and Income Inequality." Moav, Omer. "Cheap Children and the Persis-
Journal of Political Economy, August 1992, tence of Poverty." CEPR Discussion Paper
100(4), pp. 818-34. No. 3059, November 2001.
Hansen, Gary D. and Prescott, Edward C. Morand, Olivier F. "EndogenousFertility, In-
"Malthusto Solow." AmericanEconomicRe- come Distribution,and Growth."Journal of
view, September2002, 92(4), pp. 1205-17. Economic Growth,September1999, 4(3), pp.
Hansen, Lars P. "Large Sample Properties of 331-49.
Generalized Method of Moments Estima- Perotti,Roberto."Growth,Income Distribution,
tors." Econometrica, July 1982, 50(4), pp. and Democracy: What the Data Say." Jour-
1029-54. nal of Economic Growth,June 1996, 1(2), pp.
Haveman,Robertand Wolfe,Barbara."TheDe- 149-87.
terminants of Children's Attainments: A Re- Psacharopoulos,George."Returnsto Investment
view of Methods and Findings." Journal of in Education:A Global Update." WorldDe-
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions
VOL 93 NO. 4 DE LA CROIXAND DOEPKE: INEQUALITYAND GROWTH 1113
This content downloaded from 168.176.5.118 on Fri, 19 Jun 2015 19:14:56 UTC
All use subject to JSTOR Terms and Conditions