Economic American Association
How Important Is Human Capital for Development? Evidence from Immigrant Earnings
Author(s): Lutz Hendricks
Reviewed work(s):
Source: The American Economic Review, Vol. 92, No. 1 (Mar., 2002), pp. 198-219
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/3083328 .
Accessed: 28/11/2011 08:55
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 [e-mail].
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
http://www.jstor.org
How Important Is Human Capital for Development?
Evidence from Immigrant Earnings
By LUTZ HENDRICKS*
Thispaper offers new evidence on the sources of cross-countryincome differences.
It exploits the idea that observing immigrantworkersfrom differentcountries in the
same labor marketprovides an opportunityto estimate their human-capitalendowments. These estimates suggest that humanand physical capital accountfor only a
fraction of cross-countryincomedifferences.For countriesbelow 40 percent of U.S.
outputper worker, less than half of the output gap relative to the United States is
attributedto human and physical capital. (JEL 015, 041, F22)
Cross-countrydifferences in per capita outputs are far larger than standardneoclassical
growth models predict. In response, some authorshave proposedto abandonthe neoclassical
frameworkin favor of theories where countries
differ in their total factor productivities,possibly due to technology gaps (e.g., Paul Romer,
1993; EdwardC. Prescott, 1998). An alternative
approach,pioneered by N. GregoryMankiw et
al. (1992), is to augmentthe neoclassical model
by adding human capital. Since the two approaches differ dramaticallyin their policy implications, it is important to determine the
o
relativecontributions f humancapital and total
factor productivity (TFP) to cross-country income differences.
Previous attemptsat resolving this issue have
encountered the problem of measuring countries' human-capital stocks. A common approach is to assume that workers of given age
and educationhave the same human-capital ne
dowments in all countries (e.g., Robert E. Hall
and Charles I. Jones, 1999). A difficulty with
this approachis that possible differences in unmeasured skills are not captured. Since mea* Departmentof Economics, Arizona State University,
P.O. Box 873806, Tempe, AZ 85287 (e-mail:
F
[e-mail]). or helpful comments I am grateful to Michele Boldrin, Boyan Jovanovic, Pete Klenow,
John McDowell, and especially Lee Ohanian.Seminarparticipants at Arizona State University, the University of
Arizona, the Midwest Macro Conference, and the NBER
Growth Conference provided valuable suggestions. I am
indebtedto an anonymousreferee for exceptionallydetailed
and helpful comments.
sured skills account for only a relatively small
fraction of earnings variationwithin countries,
this could be an importantomission. An alternative approachpostulatesa human-capital rop
duction function and constructs human-capital
stocks based on a perpetualinventory method
(e.g., Peter J. Klenow and Andres RodriguezClare, 1997a). A difficultywith this approachis
that the implications may be sensitive to the
human-capital production function chosen.
Whetherdifferencesin humancapital or in total
factor productivity account for the bulk of
cross-country income gaps remains therefore
controversial(RobertTopel, 1998).
This paper offers a new empirical strategy
which avoids these measurement roblems.The
p
idea is to estimatethe humancapital of workers
from differentcountriesby observingtheirearnings in the same labor market.Specifically, differences in labor earnings across U.S.
immigrants with identical measured skills are
used to infer their unmeasuredhuman-capital
endowments.' This approachhas the benefit of
capturingmeasuredas well as unmeasuredskill
differences without having to impose a humancapital productionfunction.
In order to quantitativelyexplore this idea, I
develop a neoclassical growth model that incorporates both human capital and productivity
gaps as sources of cross-countryincome differ' The idea of using internationalmigration as a natural
experimenthas been proposed a numberof times (Mancur
Olson, Jr., 1996 p. 16; Klenow and Rodrfguez-Clare, 997b
1
p. 612) but it has apparentlynever been studied in detail.
198
VOL 92 NO. ]
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
H
I
F
ences. The model decomposes these income
differences into the contributions of physical
capital, observed skills (such as education and
experience), unobserved skills as measured by
relative immigrantearnings, and a total factor
productivityresidual.
Immigrantearnings are estimated from U.S.
o
Censusdata(U.S. Department f Commerce,Bureau of the Census, 1990). A key observationis
a
thatthe gap between immigrant nd native earnings is less than25 percentfor most sourcecound
tries, suggestingthat cross-country ifferencesin
unobservedskills are much smaller than crosscountry income gaps. As a result, human- and
a
a
physical-capital ccumulation ccountfor only a
i
fractionof cross-countryncome differences.For
w
a sampleof low-incomecountries, hich produce
on average18 percentof U.S. outputper worker,
humanand physical capitalaccountfor a reduction in output to 53 percent of the U.S. level,
F
leaving a factorof 3 unexplained. or the poorest
five countriesin the sample,outputper workeris
overpredicted y a factorof 8.
b
A possible objection against this approachis
that immigrant self-selection could drive a
wedge between the unmeasuredhuman-capital
endowments of immigrantsand source country
workers. If immigrantsare positively selected,
then relative immigrantearnings overstate the
human-capitalendowments of the source countries. However, reasonable degrees of selfselection do not alter the qualitative findings.
Fully accountingfor observed cross-countryincome differences on the basis of human and
physical capital implies that immigrants from
poor source countries must possess several
times more human capital than source country
workers. Yet data on the earnings of emigrants
and returnmigrants suggest that self-selection
in unobserved skills is rathermodest. Even allowing for degrees of self-selection that are
largerthanthe data suggest implies thathumanand physical-capitaldifferences fail to account
for large partof observed cross-countryincome
gaps. For a sample of low-income countries,
humanand physical capital accountfor a reduction of output per worker to 36 percent of the
U.S. level, compared with 18 percent in the
data. For the poorest five countries in the sample, output per worker is overpredicted by a
factor of 5.
The paper then considers whether skill
199
complementaritiescould increase the ability of
humancapitalto accountfor large cross-country
income differences. Scarcity of skilled labor
may depress unskilled wages in poor countries.
This might help explain why migrants experience large earnings gains, even though unskilled workerspossess the same human-capital
endowments in all countries. I find that skill
complementaritiesmprovethe ability of human
i
and physical capital to account for crosscountry income differences, but output per
workerremainsoverpredictedby factors of 2 or
more for low-income countries.
I conclude that data on immigrant earnings are difficult to reconcile with the view
that differences in human and physical capital
account for the bulk of the observed crosscountry income dispersion. The data are more
consistent with Prescott's (1998) conclusion
that accounting for large income differences
across countriesrequiresa theory of total factor
productivity.
A number of other papers have recently offered empiricalcritiquesof augmentedneoclassical models (see Klenow and Rodriguez-Clare,
1997a; Prescott, 1998; Mark Bils and Klenow,
2000). The main advantage of the approach
taken here is that it requires few assumptions
beyond those maintained in virtually all versions of neoclassical growth models (most importantly, that factors are paid their marginal
products). It therefore avoids issues related to
the measurementof humancapital that underlie
the currentcontroversyabout the importanceof
human capital for cross-country income
differences.
The rest of the paperis organizedas follows.
Section I lays out the model and derives its
implications for cross-country earnings differences. Section II describes the data and discusses measurement issues. The empirical
findings are presentedin Section III. Section IV
concludes.
I. The Model
This section develops a model that encompasses the two competing hypotheses about
i
cross-country ncome differences:human-capital
gaps and productivitygaps. Outputper worker
depends on a country's stocks of human and
physical capital,as for example in Mankiwet al.
200
THEAMERICAN CONOMIC EVIEW
E
R
(1992). But it also depends on a country's level
of total factorproductivityas suggestedby Prescott (1998). A parameterizedversion of the
model is used below to investigate to what
extent capital accumulationcan account for the
large cross-country income differences observed in the data.
Each country, indexed by c, is inhabitedby
large numbersof workersindexed by i. Aggregate output is produced from physical capital
(KC)and labor (LC)using a Cobb-Douglasproduction function
(3)~~~wc,
MARCH2002
=
-GsKc t7-
LcH
L,L
AJL
G
' -'
L
3
where NC'sdenotes the numberof workerswith
skill s in c. Within a skill class, individual
earnings are thus proportionalto workers' endowments of labor efficiency units. In what
follows I shall assume that the labor aggregator
is of the constant elasticity of substitutiontype
G(LC,H,
LC,L) = (PHLC,H + PLLC,L),
with elasticity of substitutiona = (1 The labor weights are normalizedsuch that PH
(1)
YC=
K'(AcLc)'
+ PL = 1. I also consider the special case
where the skill types are perfect substitutes:
Labor input is an aggregate of skilled and unskilled labor inputs: LC = G(LC,H, LC,L). This
specification allows for complementarity between skilled and unskilled workers while retaining a constant capital share in national
income. All marketsare competitive.Firms rent
physical capital and labor services from households so as to maximize period profits given
factorprices. From the first-order ondition,the
c
rental price of labor of skill s is
= (I
(1A-)a-(Kc1LjOG5
(2)
G(LC,H,
LC,L) =
LC,H
+
LC,L.
In the empiricalimplementationhe skill types
t
will be identified itheducation evels. In orderto
w
l
account for earnings differences within skill
o
classes, for exampleby age, education, r sex, the
labor force is furthersubdividedinto J classes.
Workersin classes j belonging to the set Js are
endowed with hjgcj efficiency units of labor of
skill type s. The hj capturerelativelaborefficiencies across skill classes that are common across
countries,while the qcrjcapturethe efficiencyof
countryc workersrelativeto a referencecountry
withina skill class. This referencecountrywill be
the United States for which I normalize Tusj = 1.
where Gs denotes the derivative of G with respect to labor of skill s. Competitionin factor
markets ensures that effective capital-laborratios (KCILc) and thus capital-output ratios
a
(KC
KCIYC) re equalized across workers
within a country.2Hence, the wage ratesmay be
written as functions of labor inputs and the
capital-outputratio:
WCS
Ws(KC,
(1
-
LC,H LC,L; AC)
-)AcKI
(0)Gs(LcH,
LCL).
Earningsper workerin countryc are then given
by
2
A possible concernis thatbindingminimum-wagelaws
might drive a wedge between wage rates and marginal
productsfor immigrantsfrom poor source countries. However, in the data the bulk of immigrantsearns considerably
more than the minimum wage.
If the numberof class j workersin countryc is
denoted by N1,,j then countryc's labor endowments are given by
L,
= lj EjJ, Nc,jhjqc,
j
A numberof reasons why observationallyidentical workersmay differ in human-capital evels
l
across countries have been suggested in the
literature. Examples include differences in
school quality or in the humancapital of teachers (Bils and Klenow, 2000). I shall refer to
differences in nc1j as unmeasuredskill differences. In the empirical implementation,these
will be estimated from immigrantearnings.
The model nests the two competing hypotheses about cross-countryincome differences as
special cases. In order to capturetheir implications clearly, I define two versions of the model
meant to represent the two hypotheses. The
human-capitalmodel assumes that total factor
VOL.92 NO. 1
H
F
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
I
201
productivity does not differ across countries:
where LNs = Yjc J, NCjhj. The ratio W f/
AC = A. The total factor productivity (TFP)
wcCs
model assumes that unmeasured skills do not
differ across countries (mqcij 1).
=
A. A Decomposition of Cross-Country
Income Differences
This section presentsan empiricalframework
for quantifying the contributions of physical
and human capital to cross-countryincome differences. The approachis to choose parameters
of the productionfunction such that U.S. labor
and capitalinputsyield the earningsreceived by
U.S. native-bornworkers. The contributionsof
physical and human capital to cross-country
earnings differences are then quantifiedby sequentially replacing the U.S. estimates of the
capital-output atio (KC), the populationweights
r
(Nc j), and the unmeasuredskill levels (q,j)
with their source country counterpartsin the
production function. The implied sequence of
earnings per worker, calculated from (3), decomposes the gap between U.S. and source
country earnings per worker into the contributions of physical capital, measuredand unmeasured skills. Specifically, I define the following
earnings per worker concepts:
1. Earnings per worker of skill type s in the
United States are given by
W uSs'
=
wjs(KUS,
LUS,H' LUS,L;
Aus) LUS Nus,s
S
which can be estimated from U.S. Census
data.
2. Replacing the U.S. capital-labor ratio KuS
with its source countrycounterpart C yields
K
measures the contribution of observed
skills (education and experience) to earnings differences.
4. Using immigrants'unmeasuredskills to calculate labor endowments yields
WC,5=
(s
(KC, L2,H, L2,L; A US) Ls1,s,
J
where LC5 =
The
hj
s
unmeasured skills of immigrants differ
from source country efficiencies (rc,j) by
a factor of sC reflecting
self-selection.
The ratio we!s/wNs measures the contribution of unobserved skills to earnings differences before accounting for possible
self-selection of immigrants.
5. Using source country unmeasuredskills to
calculate labor endowments yields
WC'S=
()s(Kc,
where LCs =
LC,H, LC,L; Aus )L,s1N,s
Yj E J
NC,jhj1C,j are the
source country labor endowments. wPS is
the predicted level of earnings per worker
in country c according to the humancapital model.
6. Measuredearnings per worker in country c
are denoted by wc,s. The ratio wpslwc,s rep-
resentsthe human-capital odel's prediction
m
error.One interpretations that this residual
i
is due to differences in total factor productivities across countries. Parametersof the
productionfunction are chosen such that for
the United States the predictedand the measuredearningslevels coincide:wpu, = wu.
For each earnings concept I also define an
average over skill types. For example, mean
earningsper workerin the United States are
given by wus = Es wus}\NusslNus.
=
js (KC 9 LUS,H, LUS,L;
Aus) LUS /NUS,SS
The ratio WcKlwuss
measures the contribution of physical capital to cross-country
earnings differences.
3. Using source country population weights to
calculate labor endowments yields
(K
LNH LN
N
Strictly speaking, the predictions of the
model apply only immediately after arrival in
the host country.For earlierarrivalspostmigration human-capitalinvestments could breakthe
relationshipbetween source country and immigrant earnings. However, immigrant earnings
growth does not differ sufficiently from native
earnings growth to make a difference. George
J. Borjas (1988) and Darren Lubotsky (2000)
202
THEAMERICAN CONOMICREVIEW
E
estimate that immigrant earnings increase by
10-13 percentrelativeto native earningsduring
the first 20 years after migration. While these
earnings changes are large in absolute terms,
they are small comparedwith cross-countryincome differences of up to 30. It is therefore
unlikely thatpostmigrationskill investmentsinvalidate the predictions of the human-capital
model. In order to verify this conjecture, I restrict the sample to immigrantswho arrivedat
most 10 years ago and confirmthatthis does not
significantlyalter the findings reportedbelow.
A relatedconcern is that some skills may not
be fully transferable across borders (Rachel
Friedberg, 1996). Immigrant earnings would
then underestimate source country humancapital endowments.This would strengthenmy
main conclusion that human capital accounts
for only moderate fractions of cross-country
earnings gaps.
The TFP model matches cross-countryearnings differences by construction:the A, can be
chosen such that predictedand measuredearnings coincide. However, it makes a testable
prediction about immigrant earnings: Immigrants should earn the same as natives with
identical measuredskills for all source and host
countries.
II. Data and Empirical Implementation
This section provides an outline of the data
and empiricalprocedures.Appendix B provides
additional details as well as data for all countries contained in my sample (see Table B1).
The full sample consists of 67 countries for
which sufficientdata are available. I also report
results for a low-incomesample which contains
37 countries with real per capita GDP per
workerbelow 40 percent of the U.S. level. The
o
objects to be estimatedare the parameters f the
productionfunction (0, AC,PH' ,), the capitaloutputratios (KC), the relative labor efficiencies
of differentskill classes (hj, -n%), the degree of
immigrantself-selection with respect to unmeasured skills (sc), the population weights N,1j,
and source country mean earnings w, ,
Labor Efficiencies.-The relative labor efficiencies of workersfrom differentcountriesare
estimatedfrom the earningsof U.S. immigrants.
A sample of native- and foreign-bornworkersis
MARCH2002
drawnfrom the 1990 U.S. Census of Population
and Housing 5-percent State Sample data files.
Results for 1980 are similar. All results are
reported as averages over male and female
workers. The sample is restricted to full-time
workers between the ages of 20 and 69 who
report positive earnings and who are not selfemployed and do not live in group quarters.
Only immigrantswho arrivedat age 20 or later
are included so as to ensure that most schooling
was completed in the source countries.3Countries with fewer than 150 observationsfor each
sex are dropped. For the two skill case the
minimumnumberof observationsis 40 per sex
and skill class. The resulting sample consists of
106,263 immigrants.
For each sex and country of birth, workers
are sorted into J = 60 classes accordingto age
and education.The labor efficiency coefficients
are calculated as mean earnings per hour in
class j. For U.S. natives these represent hj,
while for immigrantsthey representthe products qc jhjsc. Small sample sizes make it difficult to estimate immigrant earnings precisely
for all J classes. I thereforeassume that -c j is
the same for all classes within a given skill type
(j E J1). Labor efficiencies are converted into
annualearningsper workerassumingthat mean
hours worked equal 2,100 per year for all
classes. A potentialdifficulty with the approach
is self-selection. The unmeasuredskills of immigrantsmay differ from those of source country natives. In terms of the model, the sc factors
could be differentfrom one. This problem will
be addressedin Section III, subsection B.
ProductionFunction Parameters.-The proA
ductivityparameter us is chosen to matchU.S.
mean earnings per worker using (3). For the
human-capitalmodel, Ac = AUS for all source
countries. For the TFP model, ACis chosen to
match predicted earnings per worker in the
source countries. The labor weight PH matches
the U.S. ratio of aggregate skilled to unskilled
earnings. A normalizationimplies PL = 1 PH. A number of alternative definitions of
skilled versus unskilled labor and a range of
substitutionelasticities are explored.
3 Unfortunately, he Census data do not identify when or
t
where schooling was completed.
VOL.92 NO. I
203
F
I
H
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
The capital share parameter 0 is set to a
standard alue of 0.33 for all countries.Douglas
v
Gollin (1997) finds that capital shares do not
systematicallyvary with per capita incomes. As
a result, the decomposition of cross-country
earnings gaps into the contributionsof capital
and TFP presentedbelow also holds for crosscountry income gaps. An alternativewould be
to assume that capital flows equalize rates of
returnacross countries, in which case 0 would
be country specific. However, the humancapital model would then imply that migration
has no effect on earnings, which is at variance
with evidence presentedbelow.
Source Country Statistics.-Data on source
countryreal GDP per workerare taken from the
Penn World Table Mark 5.6 for 1990. For five
countries, data for 1987 through 1989 are used
instead. Capital-outputratios are taken from
Ellen R. McGrattanand James A. Schmitz, Jr.
(1998). Lacking data on hours worked, mean
annual earnings per worker are calculated as
(1-0) times real GDP per worker. Mean earnings by skill class are the computed from the
identity w,L, = Es WCLc s together with ess
timates of the source country skill premia Wc,H/
The latter are calculated from source
Wc,L.
country Mincer regressions described in the
Technical Appendix (available upon request).
Source country populationweights are taken
from RobertBarro and Jong-WhaLee's (2000)
data on educational attainment together with
data on population age distributionsfrom the
U.S. Bureau of the Census InternationalData
Base. The joint distributionof age and educational attainmentis constructedfrom this data
togetherwith informationon educationalattainment by age taken from the Organizationfor
Economic Cooperation and Development's
(OECD) Education at a Glance 2001 database
using an algorithm described in Appendix B.
The findings change very little, if it is assumed
instead that educational attainmentis independent of age.
One limitationof Barroand Lee's data is that
educationalattainmentis available only for the
entire populationover age 25, whereas for estimating source country human-capital stocks
data on the educational composition of the
working populationwould be desirable.For the
United States this differenceis small, but it may
be larger for other countries. Average years of
schooling estimated from worker survey data
typically exceed Barro and Lee's estimates by
several years, even for rich countries(see Table
B 1). As a result,the contribution f educationto
o
cross-country income differences would be
smaller than reportedbelow, if educational attainmentwere taken from worker survey data.
III. Implications for Cross-Country
Earnings Differences
This section investigates to what extent human- and physical-capital accumulation account for observed cross-country earnings
differences. As a starting point, I consider a
version of the human-capital odel in which the
m
skill types are perfect substitutesand in which
immigrants do not differ from source country
natives in their unmeasured skills (sC
-
1).
Both assumptionwill be relaxed below.
A. One Skill Type-No Self-Selection
The model's implications for decomposing
cross-countryearningsdifferences into the contributionsof physical capital, measuredand unmeasured skills are shown in Table 1, which
shows the relative earningsconcepts defined in
Section I for all countries in the sample. For
example, mean earningsper workerin the Philippines equal wclwus = 0.13 of the U.S. level.
The lower Filipino capital-output atio accounts
r
for an earnings reduction of 6 percent (w'/
wus = 0.94).
The relative lack of measured
skills reduces earnings further to wNIwUs =
0.72. The fact that Filipino immigrants earn
about 17 percent less than U.S. natives implies
that lower unmeasuredskills reduce earningsin
the Philippines further to wl/wus
= 0.60 of the
U.S. level, leaving a factor of 4.6 unexplained.
The last rows of Table 1 show the geometric
means of these relative earnings concepts for
the full sample, the low-income sample, and for
the poorest five countries.
Figure 1 illustrates the data by plotting the
earnings concepts reported in Table 1 against
real source countryearningsper workerrelative
to the United States (wclwus). Panel (a) shows
the effect of capital-output ratios on source
country earnings, wlwus. For most countries,
physicalcapitalaccountsfor only a small fraction
MARCH2002
E
R
THEAMERICAN CONOMIC EVIEW
204
E
G
O
TABLE 1-DECOMPOSITION F CROSS-COUNTRYARNINGS APS
Country
Egypt
Ghana
Kenya
South Africa
Barbados
Canada
Costa Rica
Dominican Republic
El Salvador
Guatemala
Haiti
Honduras
Jamaica
Mexico
Nicaragua
Panama
Trinidadand Tobago
Argentina
Bolivia
Brazil
Chile
Colombia
Ecuador
Guyana
Peru
Uruguay
Venezuela
Bangladesh
China
Hong Kong
India
Indonesia
Iran
Iraq
Israel
Japan
Jordan
Korea, Republic of
Malaysia
Pakistan
Philippines
Sri Lanka
Syria
Taiwan
Thailand
Austria
Belgium
Denmark
France
Germany,West
Greece
Hungary
Ireland
Italy
Netherlands
Norway
Poland
Portugal
Spain
PWT
No.
14
18
22
41
52
54
55
57
58
60
61
62
63
64
65
66
71
73
74
75
76
77
78
79
81
83
84
86
88
89
90
91
92
93
94
95
96
97
100
105
106
110
111
112
113
116
117
121
123
125
126
127
129
130
133
134
135
136
138
w'/wUS
43.2
55.3
91.2
95.7
80.3
99.5
86.0
84.9
68.3
70.0
59.0
77.2
112.1
85.7
80.8
92.8
81.7
100.3
90.5
88.7
96.4
82.3
97.4
138.4
97.9
101.7
102.7
46.3
82.6
79.9
75.0
79.8
90.7
89.5
99.7
114.1
88.1
89.3
96.1
59.6
94.4
63.2
77.4
88.0
78.7
110.9
107.9
109.4
113.6
116.7
104.8
89.4
108.4
110.3
107.1
113.0
83.4
103.7
108.8
wc wus
WwWus
/
wcIwus
w?Iw.
31.2
37.4
57.8
66.5
58.4
91.0
62.4
59.7
47.3
47.9
38.9
52.7
76.1
61.0
56.5
70.9
56.9
74.7
65.5
62.1
72.4
58.0
73.9
93.2
73.9
76.5
73.5
31.2
58.3
62.5
52.1
54.2
62.6
61.7
81.3
98.0
65.6
72.3
66.7
41.1
72.1
43.6
56.1
66.5
55.8
89.0
85.1
94.1
87.8
93.8
81.0
70.3
85.0
83.0
86.1
97.6
64.2
74.2
78.7
30.4
29.5
57.2
83.8
61.0
110.4
55.0
48.8
35.9
36.4
30.6
39.3
73.7
47.3
38.8
66.6
55.8
75.7
52.1
58.0
66.2
49.3
61.6
88.4
59.5
74.1
67.1
24.5
47.3
63.8
51.8
51.8
57.6
56.1
88.4
122.3
59.8
59.2
65.2
36.4
59.5
42.4
59.7
66.4
48.1
105.3
104.7
120.1
105.2
104.2
84.1
73.0
100.2
96.2
93.1
123.0
60.0
79.2
79.8
18.7
5.1
5.1
26.1
40.0
93.5
27.3
18.8
14.9
20.2
5.4
12.1
14.0
46.3
11.3
21.8
54.1
36.5
14.5
30.0
32.2
27.5
24.6
8.1
18.6
32.2
47.4
13.0
6.0
62.1
8.8
13.7
31.0
32.3
64.7
61.5
34.4
43.6
34.1
12.6
13.0
15.6
43.2
50.1
18.4
72.6
86.3
67.9
82.6
80.3
48.2
29.4
65.4
83.8
85.0
79.5
20.3
45.2
71.7
1.6
5.8
11.3
3.2
1.5
1.2
2.0
2.6
2.4
1.8
5.6
3.2
5.3
1.0
3.4
3.1
1.0
2.1
3.6
1.9
2.1
1.8
2.5
10.9
3.2
2.3
1.4
1.9
7.9
1.0
5.9
3.8
1.9
1.7
1.4
2.0
1.7
1.4
1.9
2.9
4.6
2.7
1.4
1.3
2.6
1.4
1.2
1.8
1.3
1.3
1.7
2.5
1.5
1.1
1.1
1.5
3.0
1.8
1.1
VOL.92 NO. I
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
I
H
F
205
TABLE 1-Continued.
PWT
Country
No.
Sweden
Switzerland
Turkey
United Kingdom
Yugoslavia
Australia
Fiji
New Zealand
Subsample means:
Full sample
Low-income sample
Poorest five countries
139
140
141
142
144
145
146
147
wcfIwLUs
wNl/w
w'/wUS
Uw/ws
106.7
121.1
95.1
95.1
117.6
111.1
94.1
109.7
89.5
99.5
65.4
74.9
92.0
94.9
65.4
96.4
113.6
122.7
71.1
92.9
101.8
118.8
56.4
117.8
77.2
89.2
23.5
72.8
27.2
82.4
32.1
69.1
1.5
1.4
3.0
1.3
3.7
1.4
1.8
1.7
90.4
83.4
80.6
67.1
59.2
54.0
65.7
53.1
46.4
30.4
17.7
5.8
2.2
3.0
8.0
Notes: PWT No. denotes the Penn World Table 5.6 country number. Columns 2-4 show the cumulative effects of KIY,
measuredand unmeasuredskills on source countryearningsper worker.wc/wus denotes relative source country earningsin
the Penn World Tables. w"'/wcis the ratio of predictedto measured source country earnings.
of the observed earnings gap relative to the
United States. For example, in the low-income
sample, lack of physical capital reduces earnings by 17 percent, whereas measuredearnings
are only 18 percentof the U.S. level. The reason
is that even low-income countries often have
capital-output atios greaterthan one-half of the
r
U.S. level. According to (3), their earnings are
reduced at most by a factor of wlwL,
-
0.5. This is, of course, precisely the reason
why a Solow growth model cannot account for
large cross-countryoutput differences.
The effect of measuredskills on earnings,wIN
W', is shown in Panel (b). For all source countries,
educationalattainmentin the Barro-Lee(2000)
data is lower than in the United States. This reduces earningsper workerbetween20 percentfor
the richest and 40 percentfor the poorest countries.The combinedeffect of physicalcapitaland
measured kills [wcNlwus,
s
shownin Panel(c)], is to
reduceearningsin the full sampleto 67 percentof
the U.S. level. Forthe poorestfive countries, arne
ings are reducedto 54 percent.
Panel(d) shows the effect of unmeasuredkills,
s
e
wrcIlw'v,stimatedby the earningsof immigrants
relative to U.S. natives with identicalmeasured
skills (-qcsc). In the one skill model, source country
earningsare proportionalo unmeasured uman
t
h
capital.Hence,the ratiosplottedin Panel(d) equal
the -qsc factors.The key insightfrom this datais
that the gap between the earningsof immigrants
and U.S. natives with identicalskills is less than
25 percentfor most sourcecountries.Immigrants
from richer source countriesearn more, but the
i
relationships weak. A tenfoldincreaseof source
country earnings (or source country output per
worker)is associatedwith an immigrant arnings
e
improvement f only aroundone-third.4 ccounto
A
ing for this fact poses a challenge for models
whereproductivitys embodiedin workers,but it
i
arises naturally f productivitys countryspecific
i
i
as in the TFP model.
The human-capitalmodel's predicted source
country earnings per worker, w"7Yw are
us
shown in Panel (e). They represent the joint
effect of physical capital, measuredand unmeasured skills. In the full sample, human capital
accounts for a 27-percent reduction in output
per worker.The joint effect on humanand physical capitalis to reducerelativeearningsto 0.66,
leaving on average a factor of 2.2 unexplained.
Figure 2 shows the ratio of predicted to measured immigrantearnings [wY'/wc, or the ratio
of the data shown in Panel (e) of Figure 1 to the
45-degree line]. Consistent with theories that
attribute cross-country income differences to
total factor productivity,the degree of overprediction is larger for poorer countries. In the
low-income sample, earnings per worker implied by the model (0.53) overpredictmeasured
earnings by a factor of 3. For the poorest five
countries, predicted earnings of 0.46 exceed
measuredearnings by a factor of 8.
4 Borjas (1988 Table 5) has a similar finding.
MARCH2002
THEAMERICAN CONOMICREVIEW
E
206
(a) Effect of capital-output ratio on earnings
150
0
0~~~~~~~~~
c~oc0o0
t
j'Dq
~
io100 C
10
(~
0(5)L
v
900(3D0
Cb
ZAB0~0
-
00
~50
0
20
10
0
0
50
40
30
60
70
80
90
100
80
90
100
90
100
(b) Effect of measured skills on earnings
150100
-
50
C~~~~~~~
0
20
10
30
50
40
60
70
(c) Joint effect of capital and measured skills
150
;~~~d
10
0
+
ca2
'9
0~~~~
00
01
0
10
20
60
70
50
40
30
Relative source country earnings
80
O
D
E
OF
FIGURE1. DECOMPOSITION CROSS-COUNTRYARNINGS IFFERENCES:NE SKILLMODEL
OtherHost Countries.-There are reasons to
believe that similar findings are valid for a variety of host countries, not only for the United
t
States. In particular, he observationthat immigrants earn within 25 percent of natives with
identical measured skills is common in the literature. For example, Borjas (1988 Table
6.1) reportsthat the typical immigrantwith 12
years of schooling at age 50 earns 9 percent
more than observationallysimilarnatives in the
United States, 10 percentless in Canada,and 5
percent less in Australia [see also David E.
Bloom and Morley Gunderson (1991 Table
12.5) for Canada and see John J. Beggs and
Bruce J. Chapman(1991) for Australia]. Similar findingshold for the United Kingdom(Brian
D. Bell, 1997, especially Table 5), Italy (Alessandra Venturini and Claudia Villosio, 1998
Table 4.1), Denmark(Leif Husted et al., 2000),
Norway (JohnE. Hayfron, 1998), Sweden (PerAnders Edin et al., 2000), and Germany(Christoph M. Schmidt, 1997 Table 4).
Importantly,the observation remains valid
for poorer host countries as well. For Israel,
Friedberg(1996 Table 4) reports that, controlling for individual characteristics,earnings of
immigrantsfrom all regions are very similar to
those of natives. For example, with 12 years of
F
I
HENDRICKS:HOW IMPORTANT S HUMAN CAPITAL OR DEVELOPMENT?
VOL.92 NO. I
207
FIGURE1-Continued.
(d) Effect of unmeasured skills on earnings
:
160
140
120
0
10
20
80 -
0
0 0cbe8 00
0
0
0
40
50
60
?00
0
79
i6o0
60
40
20 0
0
40
30
20
10
50
60
70
80
90
100
90
100
(e) Predicted source country earnings
160140n20 -0
C
0
0
01000 0
?80 0
~0Q0
60 -o oRD&'~Poec
00 00
40 -o00(to&
20
0
0
10
20
70
60
50
40
30
Relativesource countryearnings
12
KEN
GUY
co
10
CD
8
O
CHN
6CD
D
JAM
PHL
0t
E4
YUG
L
CD
}-2
-
0
BGDa=G
10
20
JPN*A
70
80
40
50
60
30
Relative source country earnings
90
100
FIGuRE 2. RATIO OF PREDICTED TO MEASURED SOURCE
CoUNTRYARNINGS
E
schooling and 20 years of experience the ratio
of immigrantto native earningsis 99 percentfor
Western Europeans,96 percent for EasternEuropeans, 101 percentfor Soviets, and 86 percent
for Asians/Africans.
80
Discussion.-In orderto understand hy huw
man and physical capital fail to account for
large partsof cross-countryincome differences,
consider the model's implications for migrant
earningsgains. If thereis only one skill type and
if countries share identical production functions, migration affects earnings of a given
worker only by changing the capital-outputratio. Specifically, moving from country c to
country s increases earningsby a factor of (K,I
Kc) /(1 - 0) according to (3). For a capital share
of 0 = 1/3 and a source country with one-half
the U.S. capital-output ratio, migration increases earningsby 40 percent. Hence, the predicted earnings gains from migrationare much
smaller than observed cross-country earnings
differences for workerswith identical measured
skills, so that immigrantsshould earn much less
than U.S. natives. By contrast, in the data immigrantstypically receive at least 75 percent of
native earnings. Successfully accounting for
cross-country income differences without appealing to TFP gaps therefore either requires
larger earnings gains from migration or that
208
MARCH2002
THEAMERICAN CONOMIC EVIEW
R
E
migrantsdiffer in their unmeasuredskills from
nonmigrants.
An additional challenge for the humancapital model is to account for the variationof
immigrantearnings across host countries. The
fact that immigrantearningscluster aroundnative earnings in poor as well as in rich host
countriesarises naturallyin the TFP model, but
poses a problem for models in which productivity is embodied in workers. For example,
r
Israel's capital-output atiois close to thatof the
United States. Hence, the human-capitalmodel
does not provide any reasons why the earnings
of immigrants with identical measured skills
should differ between Israel and the United
States by a factor of 2.
The following subsections examine whether
two extensions help reconcile the model with
the data. These extensions are self-selection
of emigrants in terms of unmeasuredhumancapital and skill complementarities.5
Self-Selection Implied by the Model.-It is
instructive to ask what degree of immigrant
self-selection would be required to fully account for cross-countryincome gaps based on
human and physical capital alone. First, I calculate the predicted earnings of immigrantsat
source country skill prices from (3):
B. One Skill Type-Nonrandom
Selection of Migrants
The interpretationis that immigrants should
earn ERCtimes mean earnings, if they returned
to their source countries.The first ratio reflects
measuredskill differences. It can be calculated
directly from population weights in source
countries and of immigrants.The factor sc reflects self-selection in unmeasuredskills. If human and physical capital fully account for
cross-country earnings differences, then sC =
I
wN/wC. n otherwords, the self-selection factors
SCequal the unexplained earnings gaps shown
A possible defense of the human-capital
model is that migrantsare self-selected so that
they possess more unmeasuredhuman capital
than nonmigrants.The large earnings gap bei
tween U.S. immigrantsand nonmigrants n poor
source countries (Figure 2) would then reflect
strong self-selection instead of large earnings
gains. In other words, the reason why immigrantsearnseveraltimes more thanpredictedby
m
the human-capital odel would then not be that
migrationleads to large earningsgains, but that
immigrantspossess several times more unmeaw
suredhumancapital (s,) than nonmigrants ith
identical characteristics. However, several
pieces of evidence suggest that unmeasured
self-selection is likely much smaller than necessary for reconciling the model with large
cross-countryearnings gaps.6
5 Given the limited evidence, it is difficultto quantifythe
role of possible human-capital spillovers. See James E.
Rauch (1993) and Daron Acemoglu and Joshua Angrist
(2000) for attemptsat measuringsuch spillovers.
6 Section III, subsection D, builds on the evidence presented here to quantify the contributions of human and
physical capital to cross-countryincome differences in the
presence of plausible degrees of self-selection.
C
=
wO(Kc, LCH,
x,Nushj-q
j
LCL;
Aus)
j s c1/E
J
NC.
j
Hence the predictedratio of immigrantto mean
source country native earnings is given by
wm E NC hyz1/E NCuJ
.
J
(4)
ERC
w
NC =
jhjqCj1ENCSC.
in Figure 2.
These figures are easier to interpret when
expressed as the implied positions of immigrants in the source country earnings distribuo
tion. I assume thatthe distribution f earningsin
each country is lognormal with a standarddeviation that matches the quintile ratios reported
in Klaus Deininger and Lyn Squire's (1996)
data set of inequalitymeasures. The lognormal
distributionapproximatesthe earningsdistributions of several countries fairly closely, except
for the very highest earnings levels (John
Creedy, 1985). Using Deininger and Squire's
inequality measures likely understatesthe degree of migrant self-selection implied by the
model, as their estimates representincome inequality across the entire population which is
typically largerthan earningsdispersionamong
the labor force. Figure 3 shows the implied
F
H
I
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
VOL 92 NO. I
bc
100
I
-'.
*VAU~t
'
I
NIC PER
95-HNCDTHAE~
I
igrated to the United States. However, other
evidence suggests that self-selection in unmeasured skills is generally quite C-A
a)
A
JPN
~~~~~~SR weak.
I
95
MYS
SL91PAN
R
KOR
CRIAR
PRT
C
o
95
85 DNIAU-bWE BEL
~~~~~RC GBR
90
80 -
RLP
NDAAHE
AUS
VEN
CAN
BRB
U)
2
75
c
70 -
E
65
E
GTM
ITA
rro
HKG
60
55 -
MEX
50
0
10
20
30
40
50
60
70
80
209
90
100
Relative source country eamings
IN
O
P
FIGuRE3. PREDICTEDOSITION F IMMIGRANTS SOURCE
D
COUNTRY ARNINGS ISTRIBUTION
E
i
percentilepositions of inumigrantsn the source
countryearningsdistributions.For several lowincome countries the model predicts that the
typical immigrantshould be drawnfrom the top
1 percent of the observed earnings distribution
(which does not include expatriates).
This strong degree of self-selection appears
especially implausiblefor source countrieswith
large emigrant populations. The most striking
case is Jamaica,where 14.5 percent of the population resided in the United States in 1990, yet
the mean immigrantmust be drawnfrom the top
0.1 percent of the earnings distribution.Other
countries for which the fraction of the population residing in the United States exceeds the
predicteddegree of self-selection include Guyana (15.2 percent vs. 0.1 percent), Nicaragua
(4.7 percent vs. 4.4 percent), the Philippines
(1.6 percent vs. 0.6 percent), and Hungary(1.1
percent vs. 0.2 percent). Another case in which
strong self-selection is unlikely is El Salvador.
Edward Funkhouser (1992) estimates that 35
percent of households have family members
h
living abroad.Furthermore, e finds little selfselection of emigrants within households. Yet
the human-capitalmodel implies that the 9.1
percentof the populationresiding in the United
States in 1990 must be drawnfrom the top 11.1
percent of the earnings distribution. In these
countries,accountingfor cross-countryearnings
differencesbased on humanand physical capital
alone would require that the entire top of the
earningsdistribution(and only the top) has em-
Estimates of Emigrant Self-Selection.-The
most direct evidence suggesting weak selfselection in unmeasuredskills comes from studies that follow individual workers across
borders.Based on a sample of 490 recent U.S.
immigrants,GuillerminaJasso et al. (1998) find
that migrants on their last source countryjobs
earned 75 percent more than the mean source
country worker. However, this gap is largely
accountedfor by differences in measuredskills.
i
Immigrants n their sample possess almost eight
years more schooling than nonmigrants.With a
Mincerianreturnto schooling of 9.9 percentper
year, which is the average in the 56-country
sample of George Psacharopoulos(1994), immigrants' higher education accounts for more
than the entire earningsgap, leaving little room
for self-selection with respect to unmeasured
skills. For Egypt, RichardH. Adams, Jr. (1993)
finds that emigrants tend to be poorer than
nonmigrants.7
An indirect measure of self-selection can be
obtained from return migrants. If emigrants
were strongly self-selected, return migrants
should earn substantiallymore than never migrants. However, in the data the earnings of
both groups are very similar, suggesting that
self-selection is weak. For Hungary, Catherine
Y. Co et al. (1999) find that female returnmigrants earn slightly more than never migrants,
whereas the difference is insignificantfor male
workers.G. M. Arif (1998) estimates that Pakistani returnmigrants earn less than those who
never migrated. Alan Barrett and Philip J.
O'Connell (2000) find thatmale returnmigrants
in Ireland earn 10 percent more than never
migrants, although no wage premium is found
for women. For Puerto Rico, Fernando A.
Ramos (1992) finds that whether a worker is a
7 Earnings of political refugees provide another opportunity for observing immigrantearnings where nonrandom
selection appears unlikely. For Cuban refugees arriving
during the Mariel boatlift, David Card (1990) finds that,
after controlling for measured skills, their earnings differ
only by 18 percent from those of previous Cuban immigrantsin Miami. A more systematicinvestigationof refugee
earnings would be a useful task for future research.
210
THEAMERICAN CONOMIC EVIEW
E
R
return migrant or even whether a person was
born in the United States has little effect on
earnings.This patternis precisely what the TFP
model predicts: earnings are determined by
where a person works, not by place of birth.
Of course, the fact that return migrants do
not earn much more than those who never
left the source countriescould be due to the fact
that returnmigrantsare strongly negatively selected in terms of unmeasuredskills. However,
recent longitudinal studies of U.S. immigrants
find that this is not the case. Lubotsky (2000)
estimates that return migrants earn around 15
percent less than immigrants who stay in the
United States.
Additional evidence suggesting that selfselection is weak comes from estimates of migrantearningsgains. Except for some very poor
countries with unusually low capital-outputratios, the human-capitalmodel predicts that immigration should raise earnings by modest
amounts or even not at all. Yet empirical estimates of the earningsgains associated with migration are typically large. Jasso et al. (1998)
find that, controlling for purchasingpower differences, Chinese immigrantsearn three times
more in the United States than they did on their
last home countryjob. By contrast, the model
predictsearningsgains of only 21 percent.Similarly, Filipinos earn 2.5 times more in the
United States while the model predictsan earnings improvementof only 6 percent. The only
low-income source country for which Jasso et
al. (1998) do not find large earnings gains is
Mexico. However, their finding contrasts with
other studies, such as RichardW. Cuthbertand
Joe B. Stevens (1981) or Douglas S. Massey et
al. (1987), who find that Mexican immigrants
earn aroundsix times more in the United States
than in Mexico, comparedwith a predictedgain
of 17 percent.For PuertoRicans, Ramos (1992)
finds that,controllingfor migrantself-selection,
working in the United States doubles earnings,
whereas the model predicts an earningsgain of
only 8.4 percent. Large earnings gains due to
migrationare also found for Pakistaniworkers
in Saudi Arabia (around800 percent;Hafiz A.
Pasha and Mir Anjum Altaf, 1987) and in the
Middle East (Nadeem Ilahi and Saqib Jafarey,
1999), and for Egyptian emigrants (Adams,
1993). This evidence suggests that migrating
from poor to rich countries results in earnings
MARCH2002
gainsthatareconsiderably argerthanthehumanl
capital model predicts.
One possible way of reconciling the humancapital model with large immigrant earnings
gains is self-selection based on job matches.8If
migration occurs because workers have received attractivejob offers, then the earnings
gap between immigrants and source country
natives may in part reflect the high quality of
immigrant ob matchesinstead of differences in
j
human capital. Consistentwith this hypothesis,
Arnold De Silva (1997) finds that immigrants
who are admittedinto Canadawith prearranged
employment enjoy higher earnings. However,
the estimated earnings benefit of 17 percent
accounts for only a small fraction of the unexplained gap between immigrant and source
countryearnings.Moreover,only aroundone in
five Canadian immigrants arrive with prearranged employment.
A related concern is that skill-based admissions could induce strong immigrant selfselection. This could be a problemfor countries
such as Canada, which admits around 40
percent of immigrants based on skill or employment criteria. However, it is much less
of a concern for the United States where, since
the 1965 ImmigrationAct, the bulk of immigrants are admitted as relatives or family
members of U.S. residents. The fraction of
skill-based admissions never exceeded 16 percent during the period 1988-1998 and most of
these come from rich source countries (U.S.
Immigrationand NaturalizationService, 1998).
Moreover, while skill-based immigrantsenjoy
an initial earnings advantage over family migrants, the gap vanishes after some years of
U.S. experience (Jasso and Mark R. Rosenzweig, 1995; Harriet 0. Duleep and Mark C.
Regets, 1996).
A final reason to doubt the hypothesis of
strong self-selection is that the earnings of immigrants cluster around native earnings in all
host countries. This arises naturally in a TFP
model, where the earnings of immigrants and
natives benefit equally from the countryspecific productivitiesof the host countries. By
contrast,if productivityis embodiedin workers,
8
I am gratefulto an anonymousreferee for pointing out
this possibility.
VOL.92 NO. 1
H
I
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
F
the fact that immigrant earnings are close to
native earnings requires a specific pattern of
self-selection. In particular, immigrants from
poorer countries must be more positively selected (as measured in Figure 2), and immigrants in poorer host countries must be less
positively selected.
Accounting for Cross-Country Earnings
Gaps with Stronger Migrant Self-Selection.Taken together, this evidence suggests that migrant self-selection in unmeasured skills is
likely weak. If this is the case, then the estimates of Section III, subsection A, quantifythe
contributionsof human and physical capital to
cross-countryearningsgaps. It is, however, useful to examine the robustnessof these findings
against the possibility of strongerself-selection
in unmeasured skills. The evidence reviewed
earlier indicates that unmeasuredself-selection
is weakerthan measuredself-selection. In terms
of the model this means that the sc factors are
smaller than the ratio of measured skills of
immigrants relative to source country natives
[the first term in equation (4)]. Setting the sc
equal to these ratios should thereforeoverstate
unmeasured selection. It implies that immigrants are drawn from the top 5 percent of the
source country earnings distribution for eight
countries in the sample, and that immigrantsin
the low-income sample possess 2.3 times more
humancapital than nonmigrants.This degree of
self-selection is far greater than the evidence
presented earlier suggests. Still, large crosscountry income gaps remain unaccounted for.
Outputper worker is overpredictedby a factor
of 2 for the low-income sample and by a factor
of 5 for the poorest five countries. I conclude
that plausible degrees of unmeasured selfselection do not overturn the finding that
physical- and human-capitalaccumulationfail
to account for a large part of cross-country
income differences.
211
scarce may have low average earnings, even
though a typical unskilled workerpossesses the
same amount of human capital in all countries.
Migration then leads to large wage gains for
unskilled workersbecause they benefit from the
larger supply of skilled labor in the host
country.9
In order to empirically implement the model
with two skill types, it is necessary to define
which education classes belong to each skill.
Barro and Lee's (2000) data distinguish seven
education classes for the source countries (no
formal schooling; primary,secondary,or higher
schooling attained or completed). A common
approachis to count only college graduatesas
skilled. For this case Per Krusell et al. (2000)
estimate a substitutionelasticity between skilled
and unskilled labor of 1.67. However, when
applied across countries,these parameters ield
y
skill premiain poor countriesthat are up to ten
times largerthanin the United States. Obtaining
reasonable skill premia requiresa broaderdefinition of skill and a higher substitutionelasticity. In what follows I define workers with at
least completed secondaryeducation as skilled.
The substitutionelasticity is set to 5, so that the
model matches the mean skill premium in the
low-income sample. Experimentationwith alternativeskill definitionsor substitutionelasticities eitherreduces the explanatorypower of the
model or results in unreasonableskill premia.?0
The implications of the two skill model for
source country earnings are shown in Figure 4.11 Comparingthe findings with those of
Figure 1 for the one skill model reveals that
imperfect substitutability f skills improves the
o
model's ability to account for large crosscountry income differences. However, the gaps
between source countryearningsin the data and
the model predictions, shown in Figure 5, remain large. In the full sample, physical and
humancapitalaccountfor a reductionin relative
earningsto 0.58, leaving an unexplainedratioof
C. Multiple Skill Types
In neoclassical growth models, it is typically
assumed that workers of different skill levels
are perfect substitutes in production. Relaxing
this assumptionmight help reconcile the model
with the data. If skilled and unskilled labor are
poor substitutes,countrieswhere skilled laboris
9 I am grateful to Michele Boldrin for suggesting this
extension.
10The Technical Appendix (available upon request)
shows that the qualitative conclusions reportedhere carry
over to the case where skill types are not observed by the
econometrician.
"1The Technical Appendix (available upon request) reports the underlyingdata.
MARCH2002
E
THEAMERICAN CONOMICREVIEW
212
(a) Effect of capital-output ratio on earnings
150
0
0~~~~~~~
150
0
10
o
10
1
0
1000
10
80
90
100
0
70
80
50
60
30
40
s
c an
Rconfelativ ourcountry easrningskil
90
100
20
50
60
80
00
40
30
BGo0o
Ee
10
MoIINO
20
D
50
cai tal a
0
7
60
50
g
J
(b) teffect of
0c&~~~~~~~~~00
0~~~
FIGUR 4.D-
0
40
*
0
0
9
30
20
i
150 4
100
40
30
(b) Effectof measuredskillson earnings
0o 10
50
90
20
i
60
+
70
eas urnings
)
RS-ONR
1.9. As in the one skill case, the unexplained
gaps are largerfor poorercountries.In the lowincome sample,predictedearningsare 2.6 times
largerthanmeasuredearnings,and for the poorest five countries,the factorof overpredictionis
5.6.
The reasonfor the limited improvementis the
high substitution elasticity of skilled and unskilled labor. However, the skill premia predicted for poor countries are, on average, close
to those calculatedfrom source countryMincer
regressions. This indicates that reducing the
substitution elasticity substantially below 5
would lead to unreasonable skill premia. To
ANNSDFEECS
w
KL
OE
illustrate,reducing the substitutionelasticity to
3 raises skill premia roughly 50 percent above
those observed in the data.
As in the one skill model, it is useful to ask
whether stronger migrant self-selection increases the abilityof humanand physical capital
to accountfor cross-countryearningsgaps. As a
proxy for unmeasuredself-selection I again set
the sc factors equal the ratios of measuredskills
of immigrants relative to source country natives. Consistent with the findings for the one
skill case, the implied ratios of predicted to
measuredearningsper workerremainlarge. For
the low-income sample, output per worker is
VOL 92 NO. 1
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
H
I
F
213
FIGURE -Continued.
4
(d) Effect of unmeasured skills on earnings
160 -
140 120 -
o
00
~~Z10O
100 -
?QQP~~&
C-tcp000
80
o0 0000
0
0
0
60
0
20
10
30
40
50
60
70
80
90
100
(e) Predicted source country earnings
T
l
160
X
140 -
O
20~~~~~~~~~~
0
n100
~~~ -0
6
~80
0
IP
10
20
30
40
50
60
70
Relative source country earnings
10 -
80
90
100
tory power of the model, large income gaps
remain unaccountedfor.
GUY
9
0
0060
0
0
0
00~~~~~~~~
a)8
D. Comparisonwith Other Estimates
,7
0
?
CHN
5
GHA.D
2> 4O2
L"
o
0
10
20
P
Y
MYSB
Mari
N
'
U
U
30
40
50
60
70
80
Relative source country earnings
H
;CAN
90
100
FIGURE 5. RATIO OF PREDICTED TO MEASURED SOURCE
COUNTRY EARNINGS IN THE TWO SKILLL ODEL
M
overpredicted on average by a factor of 1.7,
while the correspondingfactor for the poorest
five countries is 3.6. I conclude that while imperfect skill substitutionimproves the explana-
This subsection compares the decomposition
of cross-countryincome gaps presented in this
article with estimates reported in the literature.12My findings are directly comparableto
those of Hall and Jones (1999). Except for the
method of measuring human capital, their accounting framework is the same as the one
described in Section I. As a result, their estimates of the contributionof physical capital to
cross-country output differences are close to
mine. Hall and Jones estimate human-capital
stocks based on Mincer regressions that are
common to all countries together with mean
years of schooling taken from an earlierversion
12
An interesting attempt at quantifyingthe role of human capital in an environmentwith skill-specific technologies is presentedin Acemoglu and FabrizioZilibotti (2001).
HUN
0.9 -
POL
KOR
a)
0o .7
C
~CHN
PHL
-~~~~
CHLpWY
O
.6
mGYZAF
JAM
&OL
0.
YUG
/
FJI
TH
_
MARCH2002
E
THEAMERICAN CONOMICREVIEW
214
A
K
A
w
N5
ND
GTM
PAK
0,4
0.3
0.4
045
0.5
0.55
0.6
0565
0.7
0.75
0.8
0.85
0.9
One skll model
FIGURE 6. CONTRIBUTION OF HUMAN CAPITAL TO OUTPUT
PER WORKER DIFFERENCES
Notes: The vertical axis shows source country output per
workerrelative to the United States implied by the humancapital stock estimates of Hall and Jones (1999). The horizontal axis shows the correspondingvalues predictedby the
one skill model.
of Barro and Lee's data. Figure 6 plots their
estimates of human capital per worker relative
to the United States against those derived here
for the one skill model. In spite of this very
different estimation method, the two estimates
are highly correlated(the correlationcoefficient
is 0.82). For countries in the low-income sample, human capital accounts for a reduction in
output per worker relative to the United States
of 44 percent according to Hall and Jones's
estimates, compared with 36 percent based on
the one skill model. It is easy to see why the two
estimates are close. Hall and Jones's estimation
frameworkassumes that the human-capitalendowments of workers with identical schooling
are the same in all countries. The approach
pursued here estimates these human-capital
endowments from immigrant earnings, but
reaches a similar conclusion.
My findings are also consistent with those of
Klenow and Rodriguez-Clare(1997a), who des
velop measuresof countryhuman-capital tocks
based on a Mincerian earnings function. The
contributionof human and physical capital to
cross-countryoutputdifferencesis measuredby
regressingthe logarithmof predictedoutputper
workeron the logarithmof observed outputper
worker.In my notation,the regressionequation
is given by ln(wr) =
B + I31ln(w,) + 8E,
where ECis a stochastic errorterm. One interpretation is that the fraction I31of the crosscountry variance in log output is accountedfor
by physical and human capital. Klenow and
Rodriguez-Clarefind that f31is likely less than
0.5, with the exact figuredependingon assumptions about the productionand measurementof
humancapital.Replicatingthis regressionin my
full sample yields an estimate of f31 = 0.38
e
(standard rror0.04) for the one skill model and
of f,3 = 0.45 (standarderror0.05) for the two
skill model. Consistentwith the findingthat the
unexplained output gaps are larger for poorer
countries,the low-income sampleyields smaller
estimates of f3l (0.27 in the one skill model and
0.38 in the two skill model). The fact that my
estimates are consistent with those of Hall and
Jones (1999) and Klenow and Rodriguez-Clare
(1997a) suggests that the implications for the
sources of cross-countryincome gaps are robust
against alternative methods of constructing
country human-capitalstocks.
IV. Conclusion
This paper offers new evidence on the
sources of cross-countryincome differences. It
exploits the idea that immigrant workers provide an opportunity to estimate the humancapital endowments of workers from a variety
of source countries based on earnings attained
in a common labor market.This approachcaptures both observed and unobservedskill differences without having to postulate a particular
human-capitalproduction function. Immigrant
earnings data suggest that cross-countrydifferences in unobserved skills are much smaller
thancross-countryincome gaps. As a result, my
estimates stronglyreject the hypothesis that human and physical capitalaccountfor the bulk of
cross-countryincome differences.
For a sample of low-income countries, human andphysical capitalaccountfor a reduction
in output per worker to one-half of the U.S.
level, comparedwith one-fifth in the data. Ala
lowing for skill complementarities nd stronger
immigrantself-selection still leaves an average
income gap of 1.7 unexplained.For the poorest
five countries in the sample, output per worker
implied by human- and physical-capitaldifferences is at least 3.6 times largerthanin the data.
This evidence is consistent with Prescott's
VOL 92 NO. I
I
F
HENDRICKS:HOW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
(1998) conclusion that accounting for crosscountryincome differentialsrequiresa theoryof
total factor productivity.
APPENDIX
A: CENSUS DATA
Census data are taken from the 1990 PUMS
5-percentState Sample data files. Individualsare
excludedfrom the sampleif they reside in group
quarters, re youngerthan 20 years or older than
a
69 years,do not work at least 30 hoursper week
and 40 weeks per year, or are self-employed.In
a
addition,observations redeletedif weekly hours
exceed 120 or annualearningsareless than$500.
Such cases are exceedingly rare and likely due
to measurement error. Immigrants are also
i
i
dropped f they arrived n the UnitedStatesbefore
age 20. This excludes immigrantswho attained
most of their educationin the United States. Increasingthe lowest anival age to 24 makes little
difference.The resultingsamplecontains2.2 million nativesand 178,000 immigrants. hen averW
aging over male and female workers, fixed
weightsof 0.6 and0.4 areused.This avoidscounting differencesin genderratios as differencesin
human capital. Observationsare sorted into the
following classes:
* Years of schooling: 0-4, 5-8, 9-11, 12, 1314, 15+.
* Age: 20-24, 25-29, 30-34, 35-39, 40-44,
45-49, 50-54, 55-59, 60-64, 64-69.
* Sex: Male or female.
* Birthplace:According to Penn World Table
country codes.
Annual work hours are calculated from
"weeks worked last year" and "hours usually
worked per week." Labor earnings are calculated from "wage and salary income," which
does not include self-employment income. The
variable"educationalattainment" ives years of
g
schooling in an intervallic format. Each person
is assigned its interval midpoint as years of
schooling.
APPENDIX
B:
SOURCE COUNTRY DATA
215
very similar if Penn World Table data are used
instead.For four countriescapital stock data are
not available (Belize, Dominica, Hungary, and
Poland). In these cases I impute a capital-output
ratio by regressing ln(capital stock per worker)
on ln(real GDP per worker).
For the two skill model it is necessary to
calculate source countryearningsby skill type.
Mincer regressions are used to calculate the
relative earningsof skilled to unskilled workers
in the source countries. Psacharopoulos(1994)
provides sources for a large number of countries, which are updated in Bils and Klenow
(2000). Only earnings regressions that do not
control for additionalvariables which might be
correlatedwith education/experienceare used.
Moreover,the underlyingsamples must be representative for a significant fraction of the
source country workforce. Additional detail is
provided in a Technical Appendix, which is
available from the author.
Educational Attainment.-In order to calculate the contributionof measuredskills to crosscountry income differences, it is necessary to
constructthe joint distributionof age and educational attainmentfor each country.Data from
three sources are used. Barro and Lee (2000)
report the population fractions, 7r, in each of
c = 1,..., 7 educational attainment classes.
The U.S. Bureau of the Census International
Data Base provides the populationfraction, t,a
in each of a = 1,..., 9 age classes. The
objective is to construct the joint distribution,
Pr(c, a), in such a way that the marginal distributionsirc and tla are respected.
The algorithmdraws on OECD data, which
provide conditional fractions Pr(j|a) for a subset of 39 countries,wherej = 1, ..., 3 indexes
education classes. Following Barro and Lee
(2000), I map the c classes into the j classes as
follows. Classj = 1 capturespersons with less
than upper secondary education and is mapped
into Barro-Leeclasses c = 1, ..., 4. Class j =
2 contains persons who completed upper secondary education and corresponds to c = 5.
Finally, j = 3 refers to tertiaryeducation and
corresponds to c = 6, 7.
Data on source country aggregates are generally taken from the Penn World Table Mark
5.6. Capital-output ratios are taken from
McGrattanand Schmitz (1998), but results are
The conditional fractions Pr(jla) are extended to the finer c classes according to
Pr(c|a) = Pr(jJa)7rcc/c=j sc where Jj denotes the set of c classes that are mapped into
216
THEAMERICAN CONOMICREVIEW
E
class j. This scales the marginal fractions ir,
such thatthe Pr(j|a) arerespected.Constructing
the joint distributionaccording to Pr(c, a) =
Pr(cla)ga would respect the age marginals[ta,
but not the education marginals
correction term AzT a
C
Therefore, a
Pr(c, a) is
i,.
added to each column of Pr(c, a), so that the
resultingjoint distributionrespects both Ia and
MARCH2002
t
iic while capturing he observationthat younger
persons have higher educational attainment
containedin the OECD data.Countriesnot contained in the OECD data set are divided into
three classes based on whether real GDP per
worker is below or above one-thirdof the U.S.
level. TheirPr(j|a) are replacedby the averages
across countries in their income class.
[Table B1 follows.]
TABLE
Country
Egypt
Ethiopia
Ghana
Kenya
Nigeria
South Africa
Barbados
Belize
Canada
Costa Rica
Dominica
Dominican Republic
El Salvador
Guatemala
Haiti
Honduras
Jamaica
Mexico
Nicaragua
Panama
Puerto Rico
Trinidadand Tobago
Argentina
Bolivia
Brazil
Chile
Colombia
Ecuador
Guyana
Peru
Uruguay
Venezuela
Bangladesh
China
Hong Kong
India
Indonesia
Iran
Iraq
Israel
Japan
B1
SOURCE COUNTRY AND IMMIGRANT CHARACTERISTICS
Years of Schooling
ImmigrantEarnings
PWT
No.
Relative
RGDPW
KIY
Unadjusted
Adjusted
BL
14
15
18
22
34
41
52
53
54
55
56
57
58
60
61
62
63
64
65
66
67
71
73
74
75
76
77
78
79
81
83
84
86
88
89
90
91
92
93
94
95
18.7
2.1
5.1
5.1
5.7
26.1
39.9
30.5
93.5
27.3
18.2
18.8
14.9
20.2
5.4
12.1
14.0
46.3
11.3
21.8
70.9
54.1
36.5
14.5
30.0
32.2
27.5
24.6
8.1
18.6
32.2
47.4
13.0
6.0
62.1
8.8
13.7
31.0
33.9
64.7
61.5
0.4
0.5
0.7
1.6
1.8
2.2
1.5
1.9
2.4
1.8
1.5
1.7
1.1
1.2
0.8
1.4
3.0
1.8
1.6
2.1
2.0
1.6
2.4
2.0
1.9
2.2
1.6
2.3
4.6
2.3
2.5
2.5
0.4
1.6
1.5
1.3
1.5
2.0
1.9
2.4
3.1
128.7
80.2
87.3
126.4
81.4
175.7
95.6
80.2
150.2
80.9
73.5
65.9
57.0
59.1
66.4
61.2
87.6
56.1
62.1
101.4
77.8
94.9
115.5
82.6
92.4
101.1
80.9
74.3
88.4
80.2
91.9
100.2
89.8
88.4
111.0
131.8
124.5
118.0
100.8
127.1
173.5
93.7
73.8
70.4
99.0
67.1
135.9
95.5
84.6
125.8
86.4
85.4
79.1
74.7
75.9
72.7
73.0
90.4
76.5
66.5
90.6
85.3
91.9
102.6
78.6
94.1
90.7
83.9
82.2
88.7
77.3
96.3
89.2
78.8
77.3
98.3
97.5
96.7
91.2
88.3
109.7
136.4
4.9
Mincer
4.2
3.8
5.3
8.2
10.6
5.5
4.5
3.8
3.0
3.3
3.8
4.2
6.3
3.5
7.2
6.7
7.8
5.6
3.8
7.2
4.3
6.4
5.4
6.6
6.5
5.0
3.2
6.8
9.2
5.0
4.1
4.4
4.4
9.4
9.7
6.4
8.4
7.0
4.0
6.1
8.6
-
8.7
9.5
5.0
8.2
8.1
9.7
10.2
8.7
7.9
-
8.0
12.6
Immigrant
N
15.5
13.8
14.9
15.6
15.8
15.4
11.7
11.4
13.7
11.7
10.4
9.7
8.7
9.2
10.9
10.3
11.5
7.5
11.6
13.3
10.8
12.2
13.1
12.9
12.8
13.2
11.8
10.9
11.9
12.7
11.3
14.3
14.3
12.3
13.9
15.6
14.7
15.3
13.1
14.2
15.4
812
303
242
124
694
297
314
173
3214
274
102
1575
3251
1615
1704
570
1890
25799
1069
463
4247
655
789
248
620
509
2069
1016
814
1199
195
235
248
4080
690
5374
340
1557
307
579
1908
VOL.92 NO. 1
I
HENDRICKS: OW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
H
F
217
TABLEB1-Continued.
Country
Jordan
Korea, Republic of
Malaysia
Pakistan
Philippines
Sri Lanka
Syria
Taiwan
Thailand
Austria
Belgium
Czechoslovakia
Denmark
France
Germany,West
Greece
Hungary
Ireland
Italy
Netherlands
Norway
Poland
Portugal
Romania
Spain
Sweden
Switzerland
Turkey
United Kingdom
U.S.S.R.
Yugoslavia
Australia
Fiji
New Zealand
Mean
Years of Schooling
ImmigrantEarnings
PWT
No.
Relative
RGDPW
KIY
Unadjusted
Adjusted
BL
96
97
100
105
106
110
111
112
113
116
117
120
121
123
125
126
127
129
130
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
34.4
43.6
34.1
12.6
13.0
15.6
43.2
50.1
18.4
72.6
86.3
21.0
67.9
82.6
80.3
48.2
29.4
65.4
83.8
85.0
79.5
20.3
45.2
11.2
71.7
77.2
89.2
23.5
72.8
41.6
27.2
82.4
32.1
69.1
38.1
1.4
1.9
2.2
0.8
2.1
0.9
1.4
1.9
1.5
3.0
2.8
3.1
2.9
3.1
3.3
2.6
1.9
2.8
2.9
2.8
3.1
1.7
2.6
2.5
2.9
2.7
3.5
2.2
2.2
3.7
3.3
3.0
2.1
2.9
2.1
98.2
95.1
106.1
98.1
94.8
131.7
118.1
132.9
101.6
164.0
162.2
128.2
160.4
153.1
142.4
102.6
125.9
127.4
112.6
140.8
166.5
98.0
87.2
114.0
113.2
165.6
169.7
127.0
165.1
118.9
114.5
162.5
73.3
148.9
109.9
91.3
77.6
93.5
81.9
76.4
100.0
106.2
99.4
83.0
126.3
126.5
100.5
131.4
126.5
117.0
102.6
100.4
119.3
119.1
110.2
131.0
92.3
109.4
97.8
105.5
129.2
131.4
107.0
130.5
93.0
111.4
131.3
81.4
126.2
97.6
6.7
10.5
6.7
3.5
7.1
5.7
5.9
8.6
5.8
9.3
8.9
Mincer
8.0
9.0
9.7
11.2
7.9
9.5
9.0
9.4
8.5
6.8
9.0
11.2
9.9
4.7
12.4
6.3
9.7
10.6
5.0
8.8
8.0
8.6
10.5
7.9
11.6
6.8
11.2
11.6
11.8
11.1
11.0
11.8
10.3
9.1
Immigrant
N
13.6
14.2
14.7
14.6
14.3
15.3
13.5
15.9
14.2
14.3
14.6
13.9
14.2
14.6
14.3
11.0
13.4
12.9
9.9
14.3
14.3
12.4
7.4
13.3
11.9
15.1
15.0
14.1
14.6
14.1
11.2
14.9
11.1
14.7
13.1
191
396
188
938
7737
155
266
1930
486
251
160
477
195
678
727
1092
568
1252
2790
508
165
2474
1494
710
541
257
319
363
4068
1157
976
306
144
140
1417
Notes: PWT No. is the Penn World Table 5.6 countrynumber.Relative RGDPW denotes real GDP per workerrelativeto the
United States in 1990. Data for 1990 are not availablefor five countries.In four of these cases, 1989 RGDPW is used instead.
For Iraq, 1987 RGDPWis reported.KIYis the capital-output atiotakenfrom McGrattan nd Schmitz (1998). Two definitions
r
a
of relative immigrantearningsare shown. Unadjustedrelative earningsdenote mean earningsper immigrantdivided by mean
earningsper native-bornworker.Adjustedrelative earningsare defined as the ratioof immigrantearningsper workerrelative
to native-bornworkerswith identicalage, education,and sex. Average years of schooling are takenfrom BarroandLee (2000;
column BL), from a sample of source countryMincer regressions (column Mincer;not available for all countries), and from
T
immigrantdata (Immigrant). he figures shown in column BL are calculatedby assigning respectively the values 0, 2.5, 6.5,
10, 12, 13, and 17 years of schooling to the seven education categories reportedin Barro and Lee (2000). N denotes the
number of immigrantsobserved in the Census sample.
REFERENCES
Acemoglu, Daron and Angrist, Joshua. "How
Large Are Human Capital Externalities?Evidence from Compulsory Schooling Laws,"
in Ben S. Bernanke and Kenneth Rogoff,
eds., NBER macroeconomics annual 1997.
Cambridge,MA: MIT Press, 2000, pp. 5-59.
Acemoglu, Daron and Zilibotti, Fabrizio. "Pro-
ductivity Differences." QuarterlyJournal of
Economics, May 2001, 116(2), pp. 563-606.
Adams, Richard H., Jr. "The Economic and DemographicDeterminantsof International iM
gration in Rural Egypt." Journal of
Development Studies, October 1993, 30(1),
pp. 146-67.
218
THEAMERICAN CONOMICREVIEW
E
Arif, G. M. "Reintegration f Pakistani Return
o
Migrantsfrom the Middle East in the Domestic Labour Market."Pakistan Development
Review, Summer 1998, 37(2), pp. 99-124.
Barrett, Alan and O'Connell, Philip J. "IsTherea
Wage Premium for Returning Irish Migrants?"IZA working paper, Bonn, 2000.
Barro, Robert and Lee, Jong-Wha. "International
Data on Educational Attainment: Updates
and Implications."National Bureau of Economic Research (Cambridge,MA) Working
Paper No. 7911, 2000.
Beggs, John J. and Chapman, Bruce J. "Male
ImmigrantWage and Unemployment Experience in Australia,"in John M. Abowd and
Richard B. Freeman, eds., Immigration,
trade, and the labor market. Chicago: University of Chicago Press, 1991, pp. 36984.
Bell, Brian D. "The Performanceof Immigrants
in the United Kingdom: Evidence from the
GHS." Economic Journal, March 1997,
107(441), pp. 333-44.
Bils, Mark and Klenow, Peter J. "Does Schooling
Cause Growth?" American Economic Re-
view, December 2000, 90(5), pp. 1160-83.
Bloom, David E. and Gunderson, Morley. "An
Analysis of the Earningsof CanadianImmigrants,"in John M. Abowd and Richard B.
Freeman, eds., Immigration, trade, and the
labor market. Chicago: University of Chicago Press, 1991, pp. 321-42.
Borjas,GeorgeJ. Internationaldifferencesin the
labor market performance of immigrants.
Kalamazoo, MI: W. E. Upjohn Institute of
EmploymentResearch, 1988.
Card, David. "TheImpactof the Mariel Boatlift
on the Miami Labor Market."Industrialand
Labor Relations Review, January 1990,
43(2), pp. 245-57.
Co, Catherine Y.; Gang, Ira N. and Yun, Myeong-
Su. "Returnsto Returning."Working paper,
Rutgers University, 1999.
Creedy,John. Dynamics of income distribution.
Oxford: Blackwell, 1985.
Cuthbert, Richard W. and Stevens, Joe B. "The
Net Economic Incentive for Illegal Mexican
Migration:A Case Study."InternationalMigration Review, Fall 1981, 15(3), pp. 54349.
Deininger, Klaus and Squire, Lyn. "A New Data
Set Measuring Income Inequality." World
MARCH2002
Bank Economic Review, September 1996,
10(3), pp. 565-91.
De Silva, Arnold. "Earnings of Immigrant
Classes in the Early 1980s in Canada: A
Reexamination." Canadian Public Policy,
June 1997, 23(2), pp. 179-202.
Duleep, Harriet 0. and Regets, Mark C. "Admis-
sion Criteria and Immigrant Earnings Profiles." International Migration Review,
Summer 1996, 30(2), pp. 571-90.
Edin, Per-Anders; LaLonde, Robert J. and As-
lund, Olof. "Emigrationof Immigrants and
Measures of Immigrant Assimilation: Evidence from Sweden." Working paper, University of Chicago, 2000.
Friedberg, Rachel. "You Can't Take It With
You? ImmigrantAssimilation and the Portability of HumanCapital."NationalBureauof
Economic Research(Cambridge,MA) Working Paper No. 5837, 1996.
Funkhouser, Edward. "Mass Emigration, Remit-
tances, and Economic Adjustment:The Case
of El Salvador in the 1980s," in George J.
Borjas and RichardB. Freeman,eds., Immigration and the workforce: Economic consequences for the United States and source
areas. Chicago: University of Chicago Press,
1992, pp. 135-75.
Gollin, Douglas. "Getting Income Shares
Right: Self-Employment, Unincorporated
Enterprise, and the Cobb-Douglas Hypothesis." Working paper, Williams College,
1997.
Hall, Robert E. and Jones, Charles I. "Why Do
Some CountriesProduceSo Much More Output Per Worker Than Others?" Quarterly
Journal of Economics, February 1999,
114(1), pp. 83-116.
Hayfron, John E. "The Performanceof Immigrants in the Norwegian Labor Market."
Journalof PopulationEconomics,May 1998,
11(2), pp. 293-303.
Husted, Leif; Nielsen, Helena S.; Rosholm, Michael and Smith, Nina. "Employment and
Wage Assimilation of Male First Generation
Immigrants in Denmark." IZA Discussion
Paper No. 101, Bonn, 2000.
Ilahi, Nadeem and Jafarey, Saqib. "Guestworker
Migration, Remittances and the Extended
Family: Evidence from Pakistan."Journal of
Development Economics, April 1999, 58(2),
pp. 485-512.
VOL.92 NO. I
HENDRICKS:HOW IMPORTANTS HUMAN CAPITAL OR DEVELOPMENT?
I
F
219
Jasso, Guillermina and Rosenzweig, Mark R. "Do
Pasha, Hafiz A. and Altaf, Mir Anjum. "Return
Immigrants Screened for Skills Do Better
than Family-ReunificationImmigrants?"International Migration Review, Spring 1995,
29(1), pp. 85-111.
Migration in a Life-Cycle Setting: An Exploratory Study of Pakistani Migrants in
Saudi Arabia."Pakistan Journal of Applied
Economics, Summer 1987, 6(1), pp. 1-21.
Jasso, Guillermina; Rosenzweig, Mark R. and
Smith, James P. "Determinants of Immi-
grants' Economic Gains from Immigration."
Working paper, University of Pennsylvania,
1998.
Klenow, Peter J. and Rodriguez-Clare, Andres.
"The Neoclassical Revival in Growth Economics: Has It Gone Too Far?" in Ben S.
Bernanke and Julio J. Rotemberg, eds.,
NBER macroeconomics annual 1997. Cambridge, MA: MIT Press, 1997a, pp. 73-103.
.__
"Economic Growth: A Review Essay."
Journal of Monetary Economics, December
1997b, 40(3), pp. 597-617.
Krusell, Per; Ohanian, Lee E.; Rios-Rull, JoseVictor and Violante, Giovanni. "Capital-Skill
Complementarityand Inequality: A Macroeconomic Analysis." Econometrica, September 2000, 68(5), pp. 1029-53.
Lubotsky, Darren. "Chutesor Ladders?A Longitudinal Analysis of ImmigrantEamings."
Working paper, PrincetonUniversity, 2000.
Mankiw, N. Gregory; Romer, David and Weil,
David N. "A Contributionto the Empirics of
Economic Growth." Quarterly Journal of
Economics, May 1992, 107(2), pp. 407-37.
Massey, Douglas S.; Alarcon, Rafael; Durand,
Jorge and Gonzalez, Humberto. Return to
Aztlan. The social process of international
migration from western Mexico. Berkeley,
CA: University of CaliforniaPress, 1987.
McGrattan, Ellen R. and Schmitz, James A., Jr.
"Explaining Cross-Country Income Differences." FederalReserve Bank of Minneapolis
Staff Report No. 250, July 1998.
Olson, Mancur, Jr. "Distinguished Lecture on
Economics in Government:Big Bills Left on
the Sidewalk: Why Some Nations Are Rich,
and OthersPoor."Journal of Economic Perspectives, Spring 1996, 10(2), pp. 3-24.
Organization for Economic Cooperation and Development. Educationat a glance: OECD in-
dicators. Paris: Centre for Educational
Research and Innovation,2001.
Prescott, Edward C. "Needed: A Theory of Total
FactorProductivity." nternationalEconomic
I
Review, August 1998, 39(3), pp. 525-51.
Psacharopoulos, George. "Returns to Investment
in Education:A Global Update." WorldDevelopment,September1994, 22(9), pp. 132543.
Ramos, Fernando A. "Out-Migration nd Return
a
Migration of Puerto Ricans," in George J.
Borjas and RichardB. Freeman, eds., Immigration and the workforce: Economic consequences for the United States and source
areas. Chicago: University of Chicago Press,
1992, pp. 49-66.
Rauch,JamesE. "Productivity ains from GeoG
graphicConcentration f HumanCapital:Evo
idence from the Cities." Journal of Urban
Economics, November 1993, 34(3), pp. 380400.
Romer, Paul. "Idea Gaps and Object Gaps in
Economic Development." Journal of Monetary Economics, December 1993, 32(3), pp.
543-73.
Schmidt, Christoph M. "Immigrant erformance
P
in Germany:Labor Earnings of Ethnic German Migrants and Foreign Guest-Workers."
QuarterlyReview of Economicsand Finance,
1997, Spec. Iss., 37, pp. 379-97.
Topel, Robert. "Labor Markets and Economic
Growth."Working paper, University of Chicago, 1998.
U.S. Department of Commerce, Bureau of the
Census. 1990 census of population and housing. Washington, DC: U.S. Government
PrintingOffice, 1990.
U.S. Immigration and Naturalization Service.
1998 statistical yearbook. Washington, DC:
U.S. GovernmentPrintingOffice, 1998.
Venturini, Alessandra and Villosio, Claudia.
"ForeignWorkersin Italy: Are They Assimilating to Natives? Are They Competing
Against Natives? An Analysis by the S.S.A.
Dataset."Working paper, University of Bergamo, 1998.