The Impact of Education on Economic Growth
Theory, Findings, and Policy Implications
Brian G. Dahlin
Duke University 2
I. Introduction
In June of 2002, President Bush announced a doubling of funds for the African
Education Initiative. Total U.S. spending on basic education in Africa will total $630
million over the next five years. Motivation for such an increase lies in the belief that the
education of children in developing countries “is key to future economic growth and
lasting democracy, leading to greater stability and improved standards of living.”
1
Many
growth models include education and offer predictions as to the implications of education
policy changes on macroeconomic performance. Some empirical analyses of the growth
rate of real per capita GDP in the U.S. suggest that years of secondary and higher
schooling contribute positively toward economic growth.
2
Such research is of particular
importance as developed nations continue taking a more active role in the development of


Source: Michaelowa, Katharina. (2000) “Returns to Education in Low Income Countries: Evidence for Africa.”
http://www.hwwa.de/Projects/Res_Programmes/RP/Development_Processes/VfS_EL_2000_Rev2.pdf
Direct and indirect effects of education are shown in the above diagram. Key
assumptions underlying the diagram are: 1) education results in learning – it is not merely
a “signal” of worker quality (see section V for more on signaling); 2) demand within the
economy is sufficient to consume higher levels of output resulting from productivity
gains; 3) monetary and fiscal policy are sufficiently responsive to meet the demands of a
growing economy (to prevent deflation, the money supply grows at a rate equal to the
growth rate of GDP).
Direct effects of education such as increased individual wages follow from the
assumption that education results in learning that increases a worker’s productivity. If
workers are paid the value of their marginal product, it follows that better-educated
workers should earn higher wages.
Externalities and other indirect effects related to education,
health, and population growth:
 higher educ. attainment and achievement of children
 better health and lower mortality of children
 better individual health
 lower number of births
Lower population
growth and better
health of population
(and labor force)
Education
Increased earnings
(higher productivity)

enables more parental involvement in each child’s education (as parents’ time is scarce).
Increased parental involvement in a child’s education may enable the child to perform
better in school and encourage him or her to pursue additional years of education.
An individual’s choice to pursue further education may improve the earnings of
his or her neighbors. Michaelowa offers the example of an educated farmer who
implements new agricultural techniques. Neighbors may observe the new methods used
by the educated farmer and imitate them. Learning through observation is a mechanism
by which such educational benefits may be spread within a community.
6

To quantify the private rate of return to education, we may regress individuals’
incomes on their level of education and other characteristics.
7
Linking a nation’s growth

3
For additional examples of externalities related to education beyond those mentioned here, we suggest:
Heckman, James and Klenow, Peter. (1997) “Human Capital Policy.”
http://www.klenow.com/HumanCapital.pdf

4
Michaelowa, Katharina. (2000) “Returns to Education in Low Income Countries, Evidence for Africa.”
http://www.hwwa.de/Projects/Res_Programmes/RP/Development_Processes/VfS_EL_2000_Rev2.pdf
Michaelowa references the following studies supporting positive correlations between parental education
and children’s health: Glewwe (1999), Schultz (1993), Hobcraft (1993), and Thomas, Strauss and
Henriques (1991).
5
Ibid. Michaelowa references the following studies with regard to the impact of education on family
planning: Wolfe and Behrman (1984), Schultz (1989), and Behrman (1990).
6

education, individuals’ years of schooling is frequently used as an independent variable.
This method has advantages in that such data are readily available in developed countries,

8
“Human capital” has many interpretations and is discussed in greater detail in section IV.
9
Positive effects were found in the following studies: Mankiw, Romer, and Weil (1992), Levine and Renelt
(1992), Barro (1991). Insignificant effects were found in the following studies: Pritchett (1997), Islam
(1995), Caselli, Esquivel, and Lefort (1996).
10
The existence of an industry focused on standardized test preparation, racial disparities in test scores, and
concerns over test-retest reliability have led to criticism of the use of standardized tests in recent years. For
further information, see:
Gordon, Edmund. (1995) “Toward an Equitable System of Educational Assessment.” Journal of Negro
Education, Vol. 64, No. 3, pp. 360-372.

7
but it does not account for differences in the quality or type of education received.
Alternatively, individuals may be classified by highest degree completed. This measure
also has problems; for example, an individual nearly finished with college is counted as a
high school graduate.
In macroeconomic analysis, economists often include a variable for human
capital. Because human capital encompasses a range of characteristics such as education,
work experience, and health, it is extremely difficult to directly measure human capital.
11

Any measure of a country’s aggregate human capital must have the following
characteristics: 1) it must be comparable across countries; 2) it must address the broad
range of criteria that comprise human capital; 3) it must include elements of human
capital for which data are available or estimable.

educational quality. For example, a recent study found that per-pupil spending is a poor
proxy for and index of school quality.
14
Alone, none of these measures provides much
insight into the quality of education – a low student-faculty ratio, for instance, says
nothing about faculty’s ability to teach.
Techniques used to measure the education of individuals and the aggregate human
capital of an economy are imperfect. Disagreement among researchers as to the “best”
measure of various aspects of education and human capital makes it more difficult to
compare the findings of empirical studies to determine the true impact of education on
individuals’ incomes and economies’ growth rates.

V. Microeconomic Theory

Microeconomic analysis attempts to determine the effect of education on an
individual’s wage. People invest in education up to the point where the marginal cost of
additional education equals its marginal benefit. As an investment in human capital, a
year of schooling produces a financial return by raising an individual’s income once he or

13
Conrad, Clifton and Pratt, Anne. (1985) “Designing for Quality.” Journal of Higher Education, Vol. 56,
Issue 6. pp. 601-622.
14
Hanushek, Eric. (1996) “Measuring Investment in Education.” The Journal of Economic Perspectives,
Vol. 10, Issue 4. pp. 9-30.

9
she enters the workforce. Following is a model that considers education to be an
investment in human capital.


r
PV
11
1
)(
1
τ
τ
τ
τ
τ
γ
(1)
Subject to:
Ms Msg s
τ
τ
() () ( )
=
−
(2)
The interest rate is denoted as r. The objective function represents the present value of
lifetime income. The first term in the objective function captures the present value of an 15
Krueger, Alan and Lindahl, Mikael. (December 2001) “Education for Growth: Why and for Whom?”
Journal of Economic Literature, Vol XXXIX pp. 1101-1136.
16
Wagstaff, Adam. (2001) “Deriving the Mincerian Earnings Function.” University of Sussex.

() ()






+
−
++
+
+
+
+
=
−sLs
r
sLg
r
g
r
g
r
sM
PV
1
)(

1
)2(

=
+
≡−
sL
i
i
r
ig
rsLG
1
)1(
)(
,
17
We follow the derivation outlined in Wagstaff (see previous footnote) that offers an excellent, though
more technical discussion of the Mincerian wage equation.

11
),(
)1(
)(
rsLG
r
sM
PV
s
−

minimal. Thus,
()
rsLG ,ln − may be regarded as a constant and we may rewrite the
relationship expressed in (7) as below.
+
−
≈ rssMPV )(lnln constant (8)
Individuals with higher ability may have a higher marginal benefit to additional schooling
in terms of generating income. Ability differences can cause the present value of lifetime
income to vary across individuals (since investment decisions are made at the margin, we
expect higher-ability individuals to invest more in education). Solving (8) for )(ln sM
and replacing
PVln with ε (an error term that captures individual differences in the
present value of lifetime income that are consequences of ability differences) we obtain:
ε
+
+
= rsconstantsM )(ln (9)
Equation (9) relates an individual’s starting salary (that is, upon entering the workforce
with no work experience) to his or her years of schooling and ability. We may also 18
The approximation rr =+ )1ln( increases in accuracy as 0→r . It remains a close approximation for
values of r such that
2.0<r (interest rates beyond this magnitude are unusual).

12
develop a wage equation that relates a worker’s wage in any period τ to his or her years
of schooling, ability, and work experience. Equation (2) expresses the wage in period τ

19
Spence’s model assumes that
education adds nothing to an individual’s human capital; rather, the educational system
serves as a filter through which the most able students pass. As a result, the possession of
more education “signals” a worker’s quality in the job market. While there are various 19
Spence, Michael. (Aug. 1973) “Job Market Signaling.” The Quarterly Journal of Economics, Vol. 87,
No. 3. pp. 355-374.

13
interpretations of education’s effect on an individual’s human capital, Krueger and
Lindahl note that “definitive answers to these questions are not available, although the
weight of the evidence clearly suggests that education is not merely a proxy for
unobserved ability.”
20

Most researchers agree that Mincerian estimates of the return to investment in
education tend to underestimate (or at the very least not overestimate) its true value. This
tendency toward downward-biased estimates is in part the result of two sources of
simultaneity bias within the Mincerian model.
21

First, since the error term reflects individual ability, it is positively correlated with
an individual’s choice of years of schooling. Second, individuals make their choice of
schooling based on the knowledge of the earnings function.
22
Both cases are violations
of the OLS assumption that the independent variable (years of schooling) is exogenously

Mincerian model to the data of 61 countries.
25
His methodology does not account for
simultaneity bias; rather, standard OLS is applied (likely underestimating the rate of
return to education, as discussed above). However, if we are only interested in the
relative differences between rates of return to education across countries, not the explicit
values of the returns themselves, Psacharopoulos’ results remain valuable.
Psacharopoulos’ findings are summarized in the following statements:
26

• The rate of return tends to be higher in low-income countries.
• Primary education makes the most valuable contribution to an individual’s
expected income in developing countries.
• The rate of return declines with the level of schooling and the country’s per capita
income.
• Investment in girls’ education tends to yield a higher rate of return than
investment in boys’ education.
• Among those in the labor force, the return to educated people is generally higher
in the private, competitive sectors than in the public sector. 24
For further reference, see Card (1993), Butcher and Case (1994), Ashenfelter and Krueger (1994), and
Ashenfelter and Zimmerman (1993). Variations of the IV technique are applied, the conclusions of which
support Harmon and Walker’s suggestion that the OLS estimate is downward biased.
25
Psacharopoulos, George. (1985) “Returns to Education: A Further International Update and
Implications.” The Journal of Human Resources, Vol. 20, No. 4. pp. 583-604.
(Psacharopoulos updated his study using more recent data in 1994, the results of which are consistent with
those of his previous research.)

27
Psacharopoulos, George. (1985) “Returns to Education: A Further International Update and
Implications.” The Journal of Human Resources, Vol. 20, No. 4. pp. 583-604.
28
United Nations Department of Economic and Social Affairs
http://millenniumindicators.un.org/unsd/mi/mi_results.asp?row_id=611
Girl-to-boy ratios are more disparate at secondary and higher levels of education in developing nations.

16
services;
29
2) faster rates of innovation within industry; 3) more informed voting choices
among the electorate.
30

Private rates of return to education are still useful in the policymaking arena.
Consider a local government seeking ways to improve the economic status of its
constituents through education.
31
Assuming a low initial level of educational
achievement within this community, government would be wise to focus its spending on
raising the number of children that complete primary school. This is not to say that
secondary and higher education should be ignored; rather, that the greater individual rate
of return to primary schooling is a more fruitful investment with regard to individuals’
incomes. Subpopulations within the community exhibiting lower average levels of
educational attainment should receive more education, again due to the higher rates of
return to education within such groups.
32

In idealized examples such as that of the small community described above,

earnings. We have shown that empirical analyses utilizing the model produce findings
largely consistent with intuition. The Mincerian wage equation’s focus on private returns
to investment in education renders it of limited use in the policymaking arena. Attempts
have been made to generalize the Mincerian equation to estimate an economy’s
geometric mean wage as a function of the labor force’s mean education.
33
As we turn to
macroeconomic literature and its assessment of the relationship between education and
economic growth, we shall examine the results of such “macro-Mincer” models prior to
current endogenous growth models that incorporate human capital.

VI. Macroeconomic Theory
Macroeconomic analysis of growth considers the rate of change of per capita
GDP. Using aggregate data to examine the relationship between education and growth in
a macroeconomic framework, we can better grasp the effects of human capital
externalities that affect growth.
34
These externalities are not evident in individual
estimates of the wage equation; however, in the aggregate, their net impact may be more
apparent. Determining the effects of human capital externalities on growth motivated
Heckman and Klenow’s recent estimate of the “macro-Mincer” wage equation that we
shall discuss shortly. 33
Krueger, Alan and Lindahl, Mikael. (Dec. 2001) “Education for Growth: Why and for Whom?” Journal
of Economic Literature, Vol XXXIX pp. 1101-1136.
34
Recall Michaelowa’s diagram in figure 1 and the subsequent discussion of growth-related externalities of
education.

ln (12)
Equation (12) expresses the log of the geometric mean wage (
g
Y
τ
) as a function of mean
worker education (
τ
S ).
38
Observations are made annually (denoted by subscript τ).
By aggregating individual characteristics through the use of the macro-Mincer
model, Heckman and Klenow seek the impact of human capital externalities on per-35
Barro, Robert J. (2002), “Education as a Determinant of Economic Growth.” Edward P. Lazear (ed.)
Education in the Twenty-first Century, Palo Alto, The Hoover Institution, pp. 9-24.
36
Ibid.
37
Krueger, Alan and Lindahl, Mikael. (Dec. 2001) “Education for Growth: Why and for Whom?” Journal
of Economic Literature, Vol XXXIX pp. 1101-1136.
38
The definition of mean worker education is subject to various interpretations as outlined in section IV.

19
capita GDP growth. As “most economies … subsidize human capital investments
substantially,” the objective of Heckman and Klenow’s application of the macro-Mincer
equation is to determine whether economies’ human capital investment decisions are


39
Heckman, James and Klenow, Peter. (1997) “Human Capital Policy.”
http://www.klenow.com/HumanCapital.pdf

40
Ibid.
41
Ibid.
42
Ibid.

20
Psacharopoulos’ cross-country study.
43
Heckman and Klenow interpret these results as
providing no evidence either for or against human capital externalities.
44

Heckman and Klenow estimate that current levels of U.S. government subsidies
for higher education are efficient if the total rate of return to education (social plus
private) is roughly 30% greater than the private rate of return.
45
Taking Psacharopoulos’
9.9% return to schooling as an estimate for the private rate of return to education, they
estimate that the total rate of return to education should be 12 to 13%. As their highest
estimate is short of this range, the argument may be made that the U.S. government
overly subsidizes higher education.
The results of Heckman and Klenow’s study suggest that education serves more
than a signaling purpose. This conclusion is drawn from the fact that the macro-Mincer

In Uzawa’s model, published several years later, investment in
human capital induces technological progress.
48

We return to new growth theory, which shall be the remaining focus of our
discussion due to its popularity and the significant attention it has received in the
literature. First we look at a simplified example of a new growth model that does not
distinguish between physical and human capital. We shall discuss (but not derive) the
more general case of the model incorporating human capital.
49
Assuming (for simplicity)
a Cobb-Douglas production function, output at time τ may be represented as:
50

α
ττ
α
ττ
−
−−=
1
])1([])1[( LaAKaY
LK
10
<
<
α
(13)
In the above equation,
K

Part 2): S71-S102.
50
The assumption of a Cobb-Douglas production function is a simplification of the original model
presented in Romer (see previous footnote). This simplification does not change the model’s main
implications.

22
Since the level of technology is determined within the model, we need to consider
how the economy’s allocation of inputs (labor and capital) affects the growth rate of
technology. The Cobb-Douglas nature of this model allows us to write the time
derivative of
τ
A as follows:
51

θ
τ
γ
τ
β
ττ
ALaKaBA
LK
)()(=
&
(14)
,0>B

,0≥
β

23
discovery at separate locations over the same period of time – as each party is oblivious
to the existence of the other’s research, twice the necessary amount of resources (labor
and capital) are exhausted toward the pursuit of the same goal. By the opposite
argument, increasing returns to scale in the production of technology is possible through
high levels of interaction between researchers. The sharing of ideas may have a
synergistic effect on the production of technology. Romer focuses his discussion of non-
constant returns to scale on the excludability of technological advances. He cites patent
laws (and their enforcement) as an institution through which knowledge becomes
excludable – lessening the returns to scale in the production of technology. If technology
can be patented, other firms may invest resources to reverse engineer the patented
technology in order to enter the market.
53
This wastes resources, as it amounts to re-
inventing the wheel.
54

A key difference between new growth theory and the neoclassical growth model
is that increasing or decreasing returns to scale within the production function for
technology allows net increasing, decreasing or constant returns to scale of the produced
factors (capital and technology), within the production of goods.
55
The neoclassical
model assumes constant returns to scale in production functions.
We make several additional assumptions regarding this model for the sake of
simplicity: 1) the savings rate is exogenous and fixed; 2) capital does not depreciate; 3)
the rate of population growth is exogenous. These assumptions are exhibited in the time
derivatives for capital and labor:
56
In the general case of the model there are four basic inputs: 1) capital; 2)
labor; 3) human capital; 4) a technology index.
57
Romer’s general model treats human
capital as “a distinct measure of the cumulative effects of activities such as formal
education and on-the-job training.”
58

Romer distinguishes human capital (
H) from A, the technology parameter, in that
H represents knowledge that is rival, while A indexes an economy’s nonrival technology.
The research sector in this model is analogous to the example given in equation (14) save
for the addition of human capital as an input into the production function for new
knowledge. To simplify the dynamic analysis of the model and rule out “an analysis of
fertility, labor force participation, or variation in hours worked per worker,” Romer
assumes that the supply of labor is constant.
59
Romer also assumes the aggregate supply 56
Romer discusses this simplified model in greater detail. For further reference, see:
Romer, David. Advanced Macroeconomics. New York: McGraw-Hill, 2001. pp.107-114
57
Romer, Paul. (1990) “Endogenous Technological Change.” Journal of Political Economy, 98 (October,
Part 2): S71-S102.
58
Ibid.
59

endogenous accumulation of human capital, Romer postulates a similar result in that
individuals would invest less in human capital, leading to a shortage of human capital in 60
Romer, Paul. (1990) “Endogenous Technological Change.” Journal of Political Economy, 98 (October,
Part 2): S71-S102.
61
Ibid. Examples of more recent models incorporating endogenously determined levels of human capital
are presented shortly; see footnote 64.
62
Ibid.