HUMAN CAPITAL: GROWTH, HISTORY, AND POLICY—

A SESSION TO HONOR STANLEY ENGERMAN†

Human Capital and Growth

By ROBERT J. BARRO*

Since the late 1980’s, much of the attention very different levels of economic development.
of macroeconomists has focused on the deter- Table 1 shows panel regression estimates for the
minants of long-term economic growth. This determination of the growth rate of real per
paper emphasizes the role of education. The capita GDP.1 (Henceforth, the designation GDP
analysis distinguishes the quantity of education, refers to real per capita GDP.) The growth rate
measured by years of school attainment, from is measured over three ten-year periods, 1965–
the quality, as gauged by scores on internation- 1975, 1975–1985, and 1985–1995. Estimation
ally comparable examinations. is by three-stage least squares, using lags of the
independent variables as instruments (see the
I. Basic Empirical Results on Growth notes to Table 1). The effects of the variable y
show up in the level and square of log(GDP) at
The empirical framework, derived from an the start of each period. The other regressors are
extended version of the neoclassical growth measures of government consumption, rule of
model and summarized in Barro (1997), can be law, international openness, the inflation rate,
described by the fertility rate, the ratio of investment to GDP,
the terms of trade, and the quantity and quality
(1) Dy ⫽ F共y, y*兲 of schooling. Before focusing on education, I
summarize the results for the other variables.
where y is per capita product, y* is the long-run
level of y, and Dy is the growth rate. In the The Level of GDP.—As is well known, the
neoclassical model, Dy is inversely related to y simple relation between growth rates and initial
and positively related to y*. The value y* de- levels of GDP is virtually nil. However, when
pends on government policies and institutions the other independent variables shown in Table
and on the character of the national population. 1 are held constant, there is a strong relation
For example, better enforcement of property between growth rate and level. The estimated
rights, fewer market distortions, and a greater coefficients are significantly positive for log-
willingness to save tend to raise y*. In a setting (GDP) and negative for the square of log(GDP).
that includes human capital, y would be gener- These coefficients imply the partial relation be-
alized to encompass the levels of physical and tween growth rate and level as shown in Figure
human capital. In some theories, Dy rises with 1. For the poorest countries (with GDP below
the ratio of human to physical capital. $580 in 1985 prices), the marginal effect of
The empirical analysis applies to roughly 100 log(GDP) on the growth rate is small and may
countries and therefore includes countries at be positive. For the richest countries, the mar-
ginal effect is strongly negative. For example,
for the United States, which in 1995 had the
†
Discussants: Stanley Engerman, University of Roches- second-largest GDP ($18,951 in 1985 prices),
ter; Robert Margo, Vanderbilt University; Clayne Pope,
Brigham Young University.
1
* Economics Department, Harvard University, Cam- The GDP figures in 1985 prices are the purchasing-
bridge, MA 02138. This research has been supported, in power-parity-adjusted, chain-weighted values from the
part, by the National Science Foundation. I appreciate the Penn World Table of Robert Summers and Alan Heston,
assistance with the education data provided by my frequent version 5.6. These data are available on the Internet from
coauthor, Jong-Wha Lee. the National Bureau of Economic Research 具nber.org典.
12
VOL. 91 NO. 2 HUMAN CAPITAL: GROWTH, HISTORY, AND POLICY 13

TABLE 1—PANEL REGRESSION FOR GROWTH RATE

Independent variable Coefficient

Log(per capita GDP) 0.107
(0.025)
Log(per capita GDP) squared ⫺0.0084
(0.0016)
Male secondary and higher schooling 0.0044
(0.0018)
Govt. consumption/GDP ⫺0.157
(0.022)
Rule-of-law index 0.0138
(0.0056)
Openness ratio 0.133
(0.041)
(Openness ratio) ⫻ log(GDP) ⫺0.0142
(0.0048)
FIGURE 1. GROWTH RATE VERSUS LOG(GDP)
Inflation rate ⫺0.0137
(0.0090) Notes: The variable on the vertical axis is the growth rate
net of the estimated effect of all explanatory variables aside
Log(total fertility rate) ⫺0.0275 from log(GDP) and its square. The value plotted was nor-
(0.0050) malized to make its mean value zero.
Investment/GDP 0.033
(0.026)
Growth rate of terms of trade 0.110
(0.030) the estimated marginal effect is ⫺0.058. This
convergence coefficient implies that an increase
Numbers of observations: 81, 84, 81
R 2: 0.62, 0.50, 0.47
in GDP by 10 percent lowers the growth rate on
impact by 0.6 percent per year.
Notes: The dependent variable is the growth rate of real per Government Consumption.—The ratio of
capita GDP for each of the periods 1965–1975, 1975–1985,
and 1985–1995. Individual constants are included in each government consumption to GDP, G/Y, is in-
panel for each period. The log of real per capita GDP and the tended to measure public outlays that do not
average years of school attainment are measured at the begin- directly enhance productivity. The estimated ef-
ning of each period. Government consumption is measured fect on growth is significantly negative: an in-
exclusively of spending on education and defense. The open-
ness ratio is the ratio of exports plus imports to GDP, filtered
crease in G/Y by 10 percentage points is
for the estimated relation of this ratio to country size, as estimated to reduce the growth rate on impact
measured by the logs of land area and population. The gov- by 1.6 percent per year.
ernment consumption ratio, the openness ratio, the ratio of
investment (private plus public) to GDP, the inflation rate (for Rule of Law.—Many analysts believe that se-
consumer prices), the total fertility rate, and the growth rate of
the terms of trade (export over import prices) are period aver- cure property rights and a strong legal system are
ages. (For the last period, the government and investment central for economic growth.2 These factors have
ratios are for 1985–1992.) The variable openness ratio ⫻ been assessed subjectively by a number of inter-
log(GDP) is the openness ratio multiplied by the log of per national consulting companies, including Political
capita GDP at the start of the period. The rule-of-law index is
the earliest value available (for 1982 or 1985) in the first two
Risk Services in its publication International
equations and the period average for the third equation.
Estimation is by three-stage least squares. Instruments
are the actual values of the schooling, openness, and terms-
2
of-trade variables, and lagged values of the other variables. In previous analyses, I also looked for effects of de-
The earliest value available for the rule-of-law index (for mocracy, measured by political rights or civil liberties.
1982 or 1985) is included as an instrument for the first two Results using subjective data from Freedom House (see
equations, and the 1985 value is included for the third Raymond D. Gastil, 1982–1983) indicate that these mea-
equation. Standard errors are shown in parentheses. The R 2 sures have little explanatory power for growth, once the
values apply to each period separately. explanatory variables shown in Table 1 are held constant.
14 AEA PAPERS AND PROCEEDINGS MAY 2001

Country Risk Guide.3 The variable used in Table reason to believe that the estimated relation
1 is an index for overall maintenance of the rule reflects effects of greater investment on the
of law. This index is measured on a 0 –1 scale, growth rate, rather than the reverse.
with 0 indicating the poorest maintenance of the
rule of law and 1 the best. The results indicate Terms of Trade.—The estimates indicate that
that an increase in the rule of law has a positive improvements in the terms of trade (a higher
and statistically significant effect on growth. An growth rate of the ratio of export prices to
improvement by one category among the seven import prices) enhance economic growth.
used by Political Risk Services (i.e., an increase
by 0.17) is estimated to raise the growth rate on II. Effects of Education
impact by 0.2 percent per year.
Given the level of GDP, a higher initial stock
International Openness.—The measure of of human capital signifies a higher ratio of hu-
openness is the ratio of exports plus imports man to physical capital. This higher ratio tends
to GDP, filtered for the estimated relation of to generate higher growth through at least two
this ratio to country size (as gauged by pop- channels. First, more human capital facilitates
ulation and area). This openness variable has the absorption of superior technologies from
a significantly positive effect on growth. leading countries. This channel is likely to be
However, the negative effect of the interac- especially important for schooling at the sec-
tion term with log(GDP) means that the effect ondary and higher levels. Second, human capi-
of openness diminishes as a country gets tal tends to be more difficult to adjust than
richer. The estimates imply that the effect of physical capital. Therefore, a country that starts
openness on growth would reach zero at a with a high ratio of human to physical capital
GDP of $11,700 (1985 U.S. dollars), a value (such as in the aftermath of a war that destroys
below the GDP’s of the richest countries, primarily physical capital) tends to grow rapidly
such as the United States. by adjusting upward the quantity of physical
capital.
Inflation Rate.—Table 1 shows a marginally The first set of empirical results measures
significant, negative effect of inflation on eco- human capital by the quantity of education, in
nomic growth. The estimated coefficient im- the sense of the value at the start of each period
plies that an increase in the average rate of of the years of school attainment of a population
inflation by 10 percent per year would lower the group aged 25 and older. (Results are similar
growth rate on impact by 0.14 percent per year. for persons aged 15 and older.) The school-
attainment data are discussed in Barro and Jong-
Fertility Rate.—The estimates indicate that Wha Lee (2001).
economic growth is significantly negatively re- In Table 1, the school-attainment variable refers
lated to the total fertility rate. Thus, the choice to males at the secondary and higher levels. The
to have more children per adult (and, hence, in estimated coefficient is positive and statistically
the long run, to have a higher rate of population significant, and Figure 2 depicts the partial rela-
growth) comes at the expense of growth in tionship with growth. The estimates imply that an
output per person. additional year of schooling (roughly a one-
standard-deviation change) raises the growth rate
Investment Ratio.—The results show that the on impact by 0.44 percent per year. (This estimate
growth rate depends positively and marginally can be shown to imply a social rate of return to
significantly on the investment ratio. Since the male secondary and higher education of around 7
instrument lists include lagged values of the percent per year.)
investment ratio, but not values that are contem- Other measures of school attainment were
poraneous with the growth rate, there is some added one at a time to the system shown in
Table 1. Female attainment at the secondary and
higher levels of education lacks significant ex-
3
These data were introduced to economists by Stephen planatory power: the estimated coefficient is
Knack and Philip Keefer (1995). ⫺0.0011 (SE ⫽ 0.0040). One possible interpre-
VOL. 91 NO. 2 HUMAN CAPITAL: GROWTH, HISTORY, AND POLICY 15

These variables were entered into the system

from Table 1. One shortcoming with this ap-
proach is that later values of test scores are
allowed to influence earlier values of economic
growth. However, the results turn out to be
similar if the instrument lists omit the test-score
variables and include instead only prior values
of variables that have predictive content for test
scores. These variables, suggested by Barro and
Lee (1997), are the total years of schooling of
the adult population (a proxy for the education
of parents), pupil–teacher ratios, and school
dropout rates.
The results for the growth effects of test scores
are in Table 2. Note that sample sizes are less than
half of those from Table 1 because of the limited
availability of the data on examinations. Science
FIGURE 2. GROWTH RATE VERSUS SCHOOLING scores have a statistically significantly positive
effect on growth, as shown in column (i) of Ta-
ble 2. With this score variable included, the esti-
tation is that many countries follow discrim- mated coefficient of male upper-level attainment
inatory practices that prevent the efficient is still positive, but only marginally significant.
exploitation of well-educated females in the The estimated coefficient on the science scores,
formal labor market. Male primary schooling 0.13 (SE ⫽ 0.02), implies that a one-standard-
is also insignificant for growth: the esti- deviation increase in scores (by 0.08) would raise
mated coefficient is 0.0011 (SE ⫽ 0.0013). the growth rate on impact by 1.0 percent per year.
(Primary schooling is, however, critical as a In contrast, the estimated coefficient for the school
prerequisite for secondary education.) Female attainment variable, 0.002 (SE ⫽ 0.001), implies
primary schooling is positive (coefficient ⫽ that a one-standard-deviation rise in attainment
0.0019; SE ⫽ 0.0013) but statistically insig- would increase the growth rate on impact by only
nificant. Note, however, that the estimation 0.2 percent per year. Thus, the results suggest
holds fixed the fertility rate. If fertility is not that the quality and quantity of schooling both
held constant, then the estimated coefficient matter for growth but that quality is much more
on female primary schooling becomes signif- important.
icantly positive: 0.0039 (SE ⫽ 0.0013). Mathematics scores are also significantly
Hence, female primary education likely pro- positive in column (ii), but less significant
motes growth indirectly by encouraging lower than the science scores. Column (iv) includes
fertility. the two scores together, and the results indi-
Many researchers argue that the quality of cate that the science scores are somewhat
schooling is more important than the quantity. more predictive of economic growth. Reading
For example, Erik Hanushek and Dennis Kimko scores are puzzlingly negative in column (iii).
(2000) find that scores on international exami- However, the reading coefficient becomes
nations (indicators of the quality of schooling positive when this variable is entered with the
capital) matter more than years of attainment for science or mathematics scores in columns
subsequent economic growth. Information on (v)–(vii).
student test scores in science, mathematics, and To increase the sample size, I constructed a
reading is available for 43 of the countries in my test-scores variable that was based on science
sample. Unfortunately, the data apply to differ- scores, where available, and then filled in some
ent years and are most plentiful in the 1990’s. missing observations by using the reading
The available data, discussed in Barro and Lee scores. (The mathematics scores turned out not
(1997), were used to construct a single cross to yield any additional observations.) The re-
section of test scores in the three subject areas. sults, shown in column (viii), are similar to
16 AEA PAPERS AND PROCEEDINGS MAY 2001

TABLE 2—PANEL REGRESSIONS FOR GROWTH RATE:

EFFECTS OF TEST SCORES

Regression
Independent
variable (i) (ii) (iii) (iv)

Science score 0.129 — — 0.064

(0.022) (0.037)

Mathematics — 0.076 — 0.036

score (0.022) (0.029)

Reading score — — ⫺0.025 —

(0.040)

Overall test — — — —
score

Male secondary 0.0019 0.0019 0.0013 0.0020

and higher (0.0011) (0.0013) (0.0018) (0.0012)
schooling

Numbers of 37, 37, 36 34, 34, 33 32, 32, 32 34, 34, 33

observations:
R 2: 0.72, 0.45, 0.68, 0.52, 0.72, 0.39, 0.69, 0.52,
0.28 0.55 0.53 0.51
FIGURE 3. GROWTH RATE VERSUS TEST SCORES

Regression
Independent III. Summary of Major Results
variable (v) (vi) (vii) (viii)

Science score 0.060 — 0.034 — The growth effects of education were ana-
(0.021) (0.027)
lyzed in a panel of around 100 countries
Mathematics — ⫺0.001 ⫺0.017 —
score (0.027) (0.029) observed from 1965 to 1995. Growth is pos-
Reading score 0.034 0.074 0.067 —
itively related to the starting level of average
(0.026) (0.028) (0.028) years of school attainment of adult males at
Overall test — — — 0.125 the secondary and higher levels. Since work-
score (0.029) ers with this educational background would
Male secondary 0.0000 0.0010 0.0009 0.0017 be complementary with new technologies, the
and higher (0.0009) (0.0009) (0.0009) (0.0015)
schooling results suggest an important role for the dif-
fusion of technology. Growth is insignifi-
Numbers of 26, 26, 26 23, 23, 23 23, 23, 23 43, 43, 42 cantly related to years of school attainment of
observations:
R 2: 0.82, 0.29, 0.74, 0.36, 0.76, 0.33, 0.65, 0.59, females at the secondary and higher levels.
0.53 0.55 0.54 0.37 This result suggests that highly educated
women are not well utilized in the labor mar-
Notes: Test scores from science, mathematics, and reading
examinations are measured as percentage correct. The data kets of many countries. Growth is insignifi-
used are a cross section, consisting of only one average score cantly related to male schooling at the
in each field per country. The overall test score, used in primary level. However, this schooling is a
regression (viii), equals the science score, where available, and prerequisite for secondary schooling and
uses the reading score, adjusted for differences in average
levels from the science scores, to fill in some additional ob-
would therefore affect growth through this
servations. The test-score variables were entered into the sys- channel. Education of women at the primary
tem described in Table 1. The test-score variables are included level stimulates growth indirectly by inducing
in the instrument lists for each equation. For the other explan- a lower fertility rate.
atory variables in the system, the estimated coefficient of the Data on students’ scores on internationally
variable for male secondary and higher school attainment is
shown, but the other coefficients are not shown. See the comparable examinations in science, mathe-
notes to Table 1 for additional information. matics, and reading were used to measure the
quality of schooling. Scores on science tests
have a particularly strong positive relation
those found in column (i). Figure 3 shows with growth. Given the quality of education,
graphically the partial relation between growth as represented by the test scores, the quantity
and the overall test-score variable. of schooling, measured by average years of
VOL. 91 NO. 2 HUMAN CAPITAL: GROWTH, HISTORY, AND POLICY 17

attainment of adult males at the secondary . “International Data on Educational At-

and higher levels, is still positively related to tainment—Updates and Implications.” Ox-
subsequent growth. However, the effect of ford Economic Papers, 2001 (forthcoming).
school quality is quantitatively much more Gastil, Raymond D. Freedom in the world. West-
important. port, CT: Greenwood Press, various years.
[Recent editions are published by Freedom
House.]
REFERENCES Hanushek, Eric and Kimko, Dennis D. “School-
ing, Labor-Force Quality, and the Growth of
Barro, Robert J. Determinants of economic Nations.” American Economic Review, De-
growth: A cross-country empirical study. cember 2000, 90(5), pp. 1184 –1208.
Cambridge, MA: MIT Press, 1997. Knack, Stephen and Keefer, Philip. “Institutions
Barro, Robert J. and Lee, Jong-Wha. “Determi- and Economic Performance: Cross-Country
nants of Schooling Quality.” Unpublished Tests Using Alternative Institutional Mea-
manuscript, Harvard University, July 1997; sures.” Economics and Politics, November
Economica (forthcoming). 1995, 7(3), pp. 207–27.