Has the expansion of higher education led to greater economic growth?
There is an enduring belief by UK policymakers chat a large higher
education sector is an important driver of long-run economic growth,
which has been part of the narrative since the
then, there was plenty of
and assumption, but strikingly
little concrete evidence to support such a belief. This paper asks
whether the evidence base has strengthened in the 50 years since it was
published. It looks at a number of different growth equation
specifications and, using international education data, attempts to draw
out the contribution of both the number of, and the growth in, graduates
since the 1960s. There are three main findings. Firstly, many growth
relationships, including those estimated elsewhere in the literature,
are quite sensitive to the countries included–which often depends on
the variables used-and time period of analysis. I argue that, given
these issues, growth equations should always be treated with caution.
Secondly, and remembering this caveat, neither the increase nor the
initial level of higher education is found to have a statistically
significant relationship with growth races both in the
see Organization for Economic Cooperation and Development.
worldwide. This result is robust to numerous different specifications.
Thirdly, there is some evidence, consistent with the existing
literature, that levels of technical skills at the end of
matter. The employment of higher level technical skills
(proxied by the number of employed researchers in an economy) is also a
strong predictor of growth. This gives a possible mechanism linking the
output of (some) of the higher education sector with economic growth.
However, it does not imply that mass higher education necessarily leads
to higher growth. This depends on the skills produced by an expanding
, in the Roman Catholic Church, member of a third order. The third orders are chiefly supplements of the friars—Franciscans (the most numerous), Dominicans, and Carmelites.
sector and their utilisacion (or underutilisatlon) in the jobs
available to increasing numbers of graduates.
: Economic growth; higher education; skills
JEL Classifications: O4; I25; O38
There is an enduring belief by UK policymakers that a large higher
education sector is an important driver of long-run economic growth. The
rapid increase in participation rates throughout the 1990s was in part
related to the idea that meeting the rising demands for skills and
knowledge caused by technological progress would be key to economic
success. Prior to its 2004 higher education reforms, the previous Labour
government stated that:
“there is compelling evidence that education increases
productivity, and moreover that higher education is the most important
phase of education for economic growth in developed countries, with
increases in HE found to be positively and significantly related to
growth” (DES, 2004, p. 58).
The present Coalition government has scrapped Labour’s 50 per
cent participation target, but has in no way indicated it believes that
the sector is any less important for achieving growth. This was
emphasised in the Browne Review–“Higher education helps to produce
economic growth, which in turn contributes to national prosperity”
(Browne Review, 2010, p. 14) –and in policy documents produced in
support of the 2011 funding reforms:
“Higher Education is important to growth through
tr.v. e·quipped, e·quip·ping, e·quips
a. To supply with necessities such as tools or provisions.
individuals with skills that enhance their productivity in the
workplace, promoting the economy’s knowledge base and driving
innovation” (BIS, 2011, p. 21).
The roots of this belief can be seen as far back as the 1963
Robbins Report. One conclusion made by the Committee was that by looking
at comparisons with other developed countries, the UK could experience a
decline in its standing if it failed to increase the number of
university-educated workers in the labour force:
“in modern societies the skills and the versatilities required
are increasingly those
v. con·ferred, con·fer·ring, con·fers
1. To bestow (an honor, for example):
by higher education. Indeed, unless
this country is prepared to expand higher education on something like
the scale we recommend, continued economic growth on the scale of the
targets set by the National Economic Development Council is, in our
view, unlikely to be
v. at·tained, at·tain·ing, at·tains
1. To gain as an objective; achieve:
” (Robbins Report, p. 73).
However, while the correlation between higher education and
development are clear (“The communities that have paid most
attention to higher studies have in general been the most obviously
progressive in respect of income and wealth”), the Report also
notes that “[u]nfortunately–or at least unfortunately for our
present purposes–the increase in productivity arising from an increase
in educational expenditure does not lend itself to easy
measurement” and consequently
/cau·sal/ () pertaining to, involving, or indicating a cause.
relating to or emanating from cause.
connections rest largely on
“it is probably just at the higher level that the external
effects most relevant to growth are of the greatest consequence. The
capacity for systematic
, the capacity readily to perceive and
apply the results of scientific progress, and the capacity for
leadership both in the fields of organisation and in the transmission
and the sifting of ideas–such capacities, if they do not come solely
from education at the higher stages, certainly
1. To obtain or receive from a source.
2. To produce or obtain a chemical compound from another substance by chemical reaction.
in a large measure
from the existence of a sufficient proportion of persons educated to
this level and of institutions devoted to higher education and
research” (Robbins Report, p. 206).
This paper asks whether the evidence base for the supposed
connection between higher education and economic growth has been
strengthened substantially since the 1960s to justify the claims made by
policymakers. Looking at evidence in recent policy documents, it is not
immediately obvious that it has. The most obvious problem is that while
there are many studies on education and growth, the availability of good
quality data has meant that few have focused specifically on higher
education. In providing evidence for the
Inquiry into higher
education, Gemmell (1997) concludes:
“However the cross-section evidence for higher education
remains limited; recent results are more encouraging than earlier
studies suggested but the robustness of these results is uncertain”
It does not seem as though a huge number of studies have been
conducted since that review, and as a result, many references are now
becoming somewhat dated. For example, Gemmell’s own analysis of
higher education and growth between 1960 and 1985 (Gemmell, 1996) is
also the main supporting reference in both the DES (2004) and BIS
(2011), despite huge growth and change in the sector since the 1980s.
One exception here is
, Kaspar Friedrich 1733-1794.
German anatomist noted for his pioneering work in embryology. His chief work, Theoria Generationis (1759), refuted the theory of preformation, which held that the embryo is a fully formed miniature adult.
(2001), who, amongst a wide range of
estimations, finds relationship between university enrolment rates and
growth between 1950 and 1990, and between the change in enrolment rates
and growth between 1960 and 1990. However, he also argues that any
effects on enrolment rates are probably biased upwards due to reverse
causality–rising incomes encourage more people to go to university.
Moreover, he found little effect of the higher education attainment of
the current workforce (rather than the enrolment rate of the future
workforce) on growth. He concludes for growth, “a certain threshold
of schooling is required, but once beyond this level of social
capability, additional general education has little marginal return …
on measured productivity” (p. 757).
Given the lack of robust results arising from
Abbr. XC or X-C
1. Moving or directed across open country rather than following tracks, roads, or runs:
comparisons, it is perhaps surprising how strongly the growth-higher
education link is asserted. In this paper, I use the readily available
Barro-Lee data on average years of education in the workforce to explain
across both the developed and the developing
world. Directed by the theoretical literature, I follow a number of
different specifications. These specifications are not new to this
paper, and similar approaches are found throughout existing
1. Relying on or derived from observation or experiment.
2. Verifiable or provable by means of observation or experiment.
work. I extend these approaches to include variables for primary,
separately. I then look across all of
these findings to see whether there are any indicators that a larger
higher education sector has been associated with faster growth over the
past 50 years.
Section 2 summarises the main theoretical frameworks for discussing
higher education and economic growth, and highlights the key
implications arising from these theories. Section 3 sets out the
estimated regressions suggested by theory. Section 4 discusses the data
used in the regressions and section 5 presents the results. Section 6
discusses these results and offers some conclusions.
2. Higher education and economic growth
Higher education has at least three potential channels through
which to affect economic growth: the
of productive skills
and capabilities; the generation of new knowledge through innovation,
and enabling quicker adoption of existing cutting-edge technologies.
2.1 Neoclassical growth
Human capital theory posits that education increases the
productivity of an individual through the creation of skills and
capabilities. Therefore, at the national level, a country with a larger
human capital stock should have a larger national output than an
otherwise identical country with a smaller supply of human capital.
and Weil (1992) assume that the standard neoclassical
production function can be augmented to include the stock of human
capital in a way similar to the physical capital stock. Hence, output
(Y) is defined:
Y = [AK.sup.[alpha]] [L.sup.[beta]][H.sup.[gamma]] (1)
where [alpha] + [beta] + [gamma] = 1, K is physical capital, H is
human capital, (2) L is the labour force and A is total factor
productivity, which captures (amongst other features) the available
level of technology. From this, economic growth follows from (1)
accumulation of both sorts of capital and (2) increases in technology,
as in Solow’s (1956) growth model, except that there are two
accumulation processes for each sort of capital. Investment adds to the
capital stock (per worker) if it exceeds the replacement demand that
arises following population growth and depreciation (e.g. machines
wearing out, skills becoming
old-fashioned or obsolete
). Investment is assumed to be a
fixed proportion of output (based on a constant saving rate). However,
because there are
to capital but replacement demand
to the capital stock, then, as the economy grows,
investment becomes increasingly used to maintain capital stocks per
worker, and less is used to increase stocks. Eventually, the economy
reaches a steady state where all investment is used to maintain the
existing level of capital per worker.
Therefore, long-run growth does not depend on investments in
physical or human capital. Increasing the investment rate in human
capital (for example, by
v. di·vert·ed, di·vert·ing, di·verts
1. To turn aside from a course or direction:
more national resources into the
provision of university education) creates a short-term increase in
growth rates until the new steady state is reached. For instance, when
newly qualified university graduates are replacing equally well educated
retirees, the human capital stock will reach its steady state and growth
return to its long-run trend (driven by
technology). However, if university education adds to the human capital
stock, we should observe an increase in growth over a relatively lengthy
period while new graduates replace less well educated retirees.
Empirical approaches which test this model tend to come in two
forms. Firstly, there are the growth accounting approaches which break
down observed growth into that which can be attributed to increases in
the productive factors (through investment and human
). Secondly, the approach of Mankiw, Romer and Weil analyses
the dynamics of growth as convergence towards a common underlying rate
driven by technological progress. Countries that start far away from
their steady state should grow faster as investments lead to more new
capital and have much lower replacement requirements. Therefore, growth
can be regressed on initial income
(or, typically, its
) and the factors which determine the steady
state–savings rates for physical and human capital, depreciation rates
and population growth.
Combining the approaches is problematic from a theoretical
perspective. For example, it does not typically make sense to look at
how growth might be explained by both initial income level and factor
accumulation rates separately, as the countries with low initial income
levels are expected to grow faster because they
capital more quickly. Hence, these two sets of variables are measuring
the same thing. There might be an argument for doing so if the level of
adopted technology was not common across countries, as is assumed in the
Solow model. Initial
(guanosine diphosphate): see guanine.
might then be used to capture differences in
the take-up of available technology (or distance from the technological
in U.S. history, the border area of settlement of Europeans and their descendants; it was vital in the conquest of the land between the Atlantic and the Pacific.
), but this would depend on the countries used in the sample.
2.2 Higher education and
/en·dog·e·nous/ () produced within or caused by factors within the organism.
1. Originating or produced within an organism, tissue, or cell.
In the 1980s, a number of
proposed models of economic
growth which went beyond the Solow model (see Romer, 1994). In
Lucas’ (1988) model of economic growth and human capital, there are
spillovers from human capital accumulation, where more educated workers
pass on their knowledge and productive capabilities to other workers.
Therefore, the output of an economy shown in Equation (1) can be
adjusted as follows to include these spillovers:
Y = [AK.sup.[alpha]] [L.sup.[beta]] [H.sup.[gamma]] [h.sup.[delta]]
where [delta] < 1. Suppose, for simplicity, that [delta] =
[beta]. Then, output per capita would be given as:
= A[(K/L).sup.[alpha]] [h.sup.1-[alpha]] (3)
where h = H/L. This means that, instead of diminishing returns, the
production function now exhibits constant returns to factors that can be
v. ac·cu·mu·lat·ed, ac·cu·mu·lat·ing, ac·cu·mu·lates
To gather or pile up; amass. See Synonyms at gather.
To mount up; increase.
per worker–if human capital and physical capital per worker
doubles, then output per worker also doubles. Saving and investment per
worker would increase at the same rate, meaning that the shares of
investment used for replacement and new capital remain constant and
there is no steady-state level. Therefore, if we
as a measure
of the broad stock of capital (both physical and human), then the
production function becomes:
Y/L = A[??] (4)
This demonstrates why these models became known as endogenous
growth theories–the rate of growth is generated within the model by the
accumulation of factors of production. Increasing the stock of these
factors generates higher incomes, which leads to more investment, which
again leads to higher incomes, and so on. Therefore, the long-run growth
rate of the economy will be the rate at which this broad measure of
The second important feature of this model is how human capital is
accumulated. In the
, variant of
(1988) model, the human capital growth rate is
Being in due proportion; proportional.
tr.v. pro·por·tion·at·ed, pro·por·tion·at·ing, pro·por·tion·ates
To make proportionate.
to the fraction of time spent in education and training
activities, so that:
h/h = [
where 0 < s < 1 is the proportion of total working time spent
studying and training, and [theta] is a
which captures the
maximum growth rate if all time was allocated to education. The
of this equation is that human capital can increase
adj. forever, as in one’s right to keep the profits from the land in perpetuity.
. An increase in time spent studying permanently adds to the
human capital stock, even when the individuals who have received this
additional education are no longer working. In essence, new generations
‘inherit’ the human capital of the retiring generation, and
add to it as they themselves pass through education and training. Human
capital in this model has the characteristics of a non-rival good (and
could be thought of more like knowledge than particular learned skills),
which has implications for how we think about the effects of increasing
the number of university graduates. For example, an
probably starts to study from a point similar to that of earlier
generations and may go through a similar learning process during the
duration of the course. Similarly, more vocationally-focused higher
education courses teach similar sets of skills over time, updated only
by changes in technology and procedures.
Notice that both the neoclassical model and the Lucas model feature
growth driven by the accumulation of the factors of
production–increasing resources dedicated to the production of capital
(such as an expansion of higher education) creates extra growth. The key
differences non-rival human capital stocks and no steady state mean that
there should be no convergence in the Lucas model, so accumulation
affects the long-run growth rate and there should be no negative
relationship with initial GDP, providing differences in initial capital
and technology take-up are controlled for.
2.3 Innovation and adoption of new technologies
The role of knowledge and non-rival human capital is also
emphasised in Romer’s (1986) model of endogenous growth. In this
model, the spillovers which remove the existence of a steady
state–leading to a production function similar to that in equation
(4)–arise from the investment in physical capital. These investments
create new knowledge which can be shared amongst firms who have not made
the same physical capital investments. So, for example, installing a new
piece of machinery in a factory generates new skills as workers use it
and knowledge about how to produce a particular good more efficiently.
Therefore, ensuring individuals are suitably equipped to
To offset sell orders or a new security offering with buy orders.
knowledge in the workplace should lead to higher growth rates.
Consequently, numerous authors have placed an emphasis on mathematical
and scientific skills as being key to linking schooling with long-run
economic growth (see Hanushek and Kimko, 2000). Although such skills can
be developed in a number of ways, an increase through higher education
would, holding everything else constant, lead to more growth.
In other endogenous growth models (Romer 1990, Jones 1999),
knowledge is explicitly linked to research and development activities
(R&D), and long-run growth rates depend on the output of this
sector. These models also generate growth
1. Produced or growing from within.
2. Originating or produced within an organism, tissue, or cell:
, as investment in
R&D increases the stock of knowledge, which helps facilitate further
advances in a way similar to the human capital accumulation relationship
described in equation (5). (3) Higher education obviously plays a key
role in ensuring there are sufficient workers in the economy with the
ability to carry out R&D (this was noted in the Robbins Report,
quoted in the Introduction).
If the link between higher education and economic growth is seen
through the mechanisms of innovating and adoption, then the implication
is that countries with higher initial levels of education should grow
faster, either because they will have larger R&D sectors which place
the country on a higher growth path, or because they adopt new
technologies at a faster rate and remain closer to the technological
In the remainder of this paper, I estimate a series of models
v. de·rived, de·riv·ing, de·rives
1. To obtain or receive from a source.
from the theory discussed above using
in psychology: see defense mechanism.
In statistics, a process for determining a line or curve that best represents the general trend of a data set.
. The aim
here is to find any consistent results across numerous different
specifications which link higher education to economic growth. The
models examine, in turn: (1) convergence to a steady state; (2)
accumulation of human capital; (3) initial differences in human capital;
(4) differences in educational quality, skills and research intensity.
3.1 Steady state convergence
The first set of regression builds up Mankiw, Romer and Weil’s
(1992) estimation of convergence to a
steady state, as
predicted by a neoclassical type model. In this model, the steady state
depends on the saving rates which drive investment in physical and human
capital. Higher saving rates, for a given initial level of GDP, imply
faster growth. The growth equation I estimate is: (4)
[??]/y = a + [[phi].sub.k] ln[s.sub.k] + [[phi].sub.h] ln [s.sub.h]
– [lambda] ln [y.sub.1966] (6)
where [y.sub.1966] is the initial level of GDP per capita in 1966.
As in Mankiw, Romer and Weil, I distinguish between saving rate for
investment ([s.sub.k], proxied by investment share of GDP) from
[s.sub.h], the saving rate for human capital accumulation, or the share
of national resources put aside for investing in education in any given
year. (5) I use enrolment rates for primary, secondary and tertiary
education as a
for this. The constant term captures the effects on
steady-state growth of depreciation, population growth and exogenous
total factor productivity (
TfP Training for Peace
) growth. For simplicity in our
estimation, I assume that these variables are exogenous and uncorrelated
with the saving rate or initial level of
GDE Graphical Development Environment
GDE Generic Data Exemption
GDE Gimbal Drive Electronics
GDE General Dynamics Electronics Division
An alternative way of accounting for growth in the neoclassical
model is to look at changes in the stock of the factors of production.
In particular, the production function in equation (1) can be expressed
in log-linear terms, and then
v. dif·fer·en·ti·at·ed, dif·fer·en·ti·at·ing, dif·fer·en·ti·ates
1. To constitute the distinction between:
with respect to time.
[MATHEMATICAL EXPRESSION NOT
v. re·pro·duced, re·pro·duc·ing, re·pro·duc·es
1. To produce a counterpart, image, or copy of.
2. Biology To generate (offspring) by sexual or asexual means.
or a set of codes used to represent letters, numbers, a few symbols, and control characters. Originally designed for teletype operations, it has found wide application in computers.
Equation (7) can be easily estimated using OLS. If we first assume
human capital is rival and relates (
1. Of or relating to an exponent.
a. Containing, involving, or expressed as an exponent.
) to schooling years,
then the final term in equation (7) is the absolute change in the
average years of schooling, s, rather than its growth rate:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (8)
Estimation of equation (8) takes the growth of total factor
productivity A as constant across all countries.
3.3 Levels of schooling
The endogenous growth theories outlined suggest that convergence
might not occur and that total factor productivity growth may vary
across countries depending on other factors. In the Lucas model, growth
in human capital depends on the level of schooling (see equation 5). In
state of balance. When a body or a system is in equilibrium, there is no net tendency to change. In mechanics, equilibrium has to do with the forces acting on a body.
, capital grows at the same rate. Therefore, equation (7)
reduces down to:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (7′)
I follow Hanushek and Woessmann (2007), who include initial GDP in
their estimate equation (7′). The initial output variable
potentially captures the original distance from the technological
frontier, as countries with greater capacity for catching up should grow
faster. The measure might also capture differences in capital stock.
I then extend this estimation in three ways. Firstly, I
growth rates in a more general model that includes both increases in
years of education and initial levels of education. One way of
interpreting this estimation is as a test of neoclassical and endogenous
growth theories (see
and Krueger, 2001). It is also consistent
with an endogenous growth model where countries were not at equilibrium
initially. I also relax the equilibrium assumption that physical and
human capital grow at the same rate and include both separately in the
Secondly, I control for differences in the quality of schooling.
Measuring the variable s simply by the quantity of education assumes
that its quality is identical. Differences in the quality, or skills, of
students should matter in these models. Greater levels of skills allow,
for example, for the quicker adoption of new technology or the
generation of new ideas. Again following the approach of Hanushek and
Woessmann (2007), I re-estimate equation (7) and control for differences
in quality related educational outputs using international test data.
Finally, quality and skill measures using data from international
tests such as
, city (1991 pop. 98,928), capital of Pisa prov., Tuscany, N central Italy, on the Arno River. It is now c.6 mi (9.7 km) from the Tyrrhenian Sea, which once reached the city.
do not allow us to look at the impact of the quality
of higher education on growth as international tests are conducted for
pre-16 year-old students. There are, to my knowledge, no internationally
comparable tests for tertiary students. One indicator of higher
education quality is if university graduates end up working in jobs
which absolutely require high level technical skills. As discussed in
section 2, endogenous growth theories have emphasised R&D activities
as a way of generating technological progress and long-run growth–a
larger research sector leads to higher long-run growth rates. In
addition, some of the technical skills produced through higher education
that are needed to work in high end research would also be available to
other sectors, leading to faster growth through higher productivity and
in biology, has several meanings. It can mean the adjustment of living matter to environmental conditions and to other living things either in an organism’s lifetime (physiological adaptation) or in a population over many many generations (evolutionary
of new technologies. Therefore, I use a measure of
the number of individuals employed as researchers as a proxy for the
supply of high level technical skills. This measure is then included in
a re-estimate of equation (7).
For the purposes of estimating the above sets of growth equations,
I use the available World Bank data. My main dependent variable is the
annualised growth rate of real per capita GDP between 1966 and 2006. Per
capita GDP is given in constant $US, using the year 2000 as the base.
Growth rates for each country are given in column 1 of table A1, in the
Some studies use shorter time periods to measure growth; these
studies have a larger number of observations and may be better at
capturing the effects of structural and policy shifts during the time
period being studied. However, they are also more
1. readily affected or acted upon.
2. lacking immunity or resistance and thus at risk of infection.
side fluctuations. Figure A1 illustrates how fluctuations in demand
affect shorter-term measures of growth rates for a selection of
at primary, secondary and tertiary levels
is taken from
see United Nations Educational, Scientific, and Cultural Organization.
in full United Nations Educational, Scientific and Cultural Organization
data for 2006. Investment shares of GDP are taken
directly from World Bank national accounts data. For the saving rate for
physical capital during the time period of interest, I calculate the
average investment share between 1976 and 2006. I multiply investment
shares of GDP by per capita GDP to give per capita investment.
To calculate the growth rate of capital, I assume investment in
1966 represents a good proxy for all prior investment and define initial
capital stock as the sum of these investments,
at a constant
rate, [delta]: (6)
[k.sub.1966] = [I.sub.1965]/[delta] (9)
The growth rate of capital is equal to total investment during the
time period over the initial capital stock. I capture this as the ratio
of current investment rate, as a proxy for investment during the period,
to the initial investment rate, which from equation (9) is proportional
to the initial capital stock. The depreciation term is picked up in the
1. an expression of the change or effect produced by variation in certain factors, or of the ratio between two different quantities.
in the regression.
data, I largely rely on the most recent
which is readily available.
and Lee (2010) have
compiled international data on mean years of schooling, amongst other
variables, every five years from 1960. For initial levels of education,
I use available data in 1965. Factor accumulation measures look at
increases between 1965 and 2005. I take advantage of the
/de·com·po·si·tion/ () the separation of compound bodies into their constituent principles.
of years of schooling into primary, secondary and tertiary education to
investigate non-linearities in the relationship between more education
and economic growth. Table 1 shows the
Individual country measures are given in the appendix (table A1, columns
2 to 7).
Table 2 shows the means of these variables across different groups
of countries. The richest (and faster growing) countries had higher
educational levels in the 1960s, although there has been some catch-up
by middle income countries since then. The gap between rich and poor in
terms of educational attainment has widened overall, although gaps in
attainment at primary level have narrowed.
Correlations between these variables is given in table 3 for
countries with growth data (n = 90). Levels of education are all
positively correlated–this is particularly high for secondary and
tertiary education in 1965. This is discussed further in the analysis
Educational quality is proxied using an international test score.
In particular, I take the mean maths and science PISA score in 2006 as
our main measure of quality and the skills level of the workforce. This
measure correlates highly (r = 0.94) with the one used in Hanushek and
Woessman (2007). Finally, the quality and technical skill level of
higher education students is proxied here by the number of researchers
per million of the population. As noted before, this is a somewhat crude
proxy for technical skills, as it might also include professional
researchers not connected to scientific research (for example, it
includes arts and humanities PhDs). For the purpose of this paper, I
assume the available data correlate with the underlying supply of higher
Table 4 shows the
for the 29 countries with the
skill measures. PISA scores and researchers have a strong
First, I estimate a number of specifications for equation (6),
shown in table 5. In column (1) I estimate the convergence relationship
for a classic Solow model which is not augmented to include human
is significantly improved when enrolment
rates are included. Column (2) follows Mankiw, Romer and Weil and
includes only enrolment at the secondary level, which is a strongly
significantly predictor of the conditional steady state. Finally, column
(3) includes all three enrolment rates. This estimation shows that
Mankiw, Romer and Weil’s original specification was well
chosen–neither primary nor tertiary enrolment rates have a significant
effect on growth in this model. (7)
Wolff’s (2001) study of OECD countries estimates a similar
model to equation (6), for a smaller sample of countries. When higher
education enrolment rates were included as the sole measure of the
steady state level of education per worker, a strongly significant and
positive relationship was found. In contrast, my estimation includes
enrolment at all three levels together, and finds no relationship
between growth and the steady state level of higher education. This is
not affected by focusing solely on OECD countries.
5.2 Growth and factor accumulation
I next estimate equation (8). Table 6 shows these results.
The estimation in column (1) shows a significant relationship
between the growth and changes in average years of schooling of the
workforce over the time period. The estimation suggests that an increase
of one year in average schooling corresponds to a 0.2 per cent increase
in the growth rate. Controlling for the growth in the capital stock over
the same period lowers this estimate, but finds it is still
significantly related to growth. The estimations in columns (3) and (4)
break down average years of schooling into primary, secondary and
tertiary education. The results show that the increase in schooling
effect on growth can be clearly attributed to secondary education, with
a somewhat less clear contribution coming from the increase in tertiary
education–the estimate becomes significant at the 90 per cent level
once the growth rate of capital is accounted for. Increases in primary
have no effect on growth rates over the past 40 years.
5.3 Growth and initial schooling
I now estimate equation (7′) to explore the relationship
predicted by endogenous growth theories between real economic growth and
measures of human capital. Initially, I look at quantity measures of
schooling, using average years of schooling at the start of the period,
and focus only on OECD countries. I compare my results with Hanushek and
Woessmann (2007), who use the same educational data to look at growth
between 1960 and 2000. The results are reported in table 7.
The table shows strikingly different results between the two
datasets. Looking at the first two columns, years of schooling in 1960
enters positively into the growth equation, while years of schooling in
1966 does not. Both measures are highly
v. cor·re·lat·ed, cor·re·lat·ing, cor·re·lates
1. To put or bring into causal, complementary, parallel, or reciprocal relation.
(r = 0.98), so it is
not so much the quality of the
Serving or intended to explain:
variable as it is the
robustness of the relationship which we should question.
One potential explanation is that the countries included in the two
regressions are different. Hanushek and Woessmann’s data looked at
24 OECD countries, as does our data taken from the World Bank. However,
there is not a perfect
1. A part or portion of a structure that extends or projects over another.
2. The suturing of one layer of tissue above or under another layer to provide additional strength, often used in dental surgery.
between the two datasets; there are three
countries in each dataset that are not represented in the other. To
investigate the role of particular outliers found in one dataset rather
than the other I
the estimation to countries which are common
to both datasets. The final two columns show the results, and indicate
that the above conclusion continues to hold (the p-value on the initial
schooling term falls to 0.14 but remains outside of significance at the
10 per cent level). The results suggest that growth and education
regression results are sensitive to the selection of time period.
Next, I use the breakdown of the Barro-Lee education data into
primary, secondary and tertiary education to establish whether this
average years of schooling measure has non-linearities that correspond
to the level of education. Table 8 shows the results.
In section 41 noted that secondary and tertiary education are
highly correlated (r = 0.89). Columns (2) and (3) show that removing one
or the other variable does not alter the results, meaning the lack of
significance does not appear to be
secondary and tertiary education. (80
I extend the above analysis to include all the available countries
in the dataset, as well as distinction between high, middle and low
income countries as defined by the World Bank. Table 9 shows the
Including all 91 countries which have schooling and income data
reveals a strongly significant correlation between years of schooling
and growth between 1966 and 2006. When broken down by level of
education, this correlation remains positive and significant for initial
levels of primary and secondary schooling, but not significant for years
of tertiary education. Looking across different groups of country by
income, there are few significant results. This is probably because the
level of a country’s income is based on current incomes, so there
issues–countries that have grown faster and are now
I next combine both initial level and accumulation effects in a
single model. The combined model, shown in table 10, is consistent with
previous results–changes in secondary schooling and initial levels of
primary schooling are positively associated with growth rates.
Controlling for capital growth, as in equation (2), improves the model;
when I do this, initial years of secondary education becomes significant
at the 90 per cent level, but years of primary schooling drops out of
significance. Critically, neither tertiary education variable has any
significant effect on growth.
5.3 Growth, quality and skills
Hanushek and Woessmann (2007) show that quantity measures explain
little variation in growth rates once differences in quality of
education measures are included. In this section I re-estimate equation
(7′) to include quality measures of human capital differences
The quality data only capture differences at the level of primary
and secondary education. Therefore, the quality measure should not
affect the years of tertiary education variable. Indeed, it is possible
that controlling more precisely for human capital differences at the
primary and secondary level could increase the explanatory power of
tertiary education, on the assumption that there is a complementary
relationship between the quality of students leaving secondary schooling
and the effectiveness of higher education.
Table 11 reports the results of this analysis using our growth
data. The results show that once quality is controlled for, either via
the composite Hanushek and Woessmann (2007) measure or a simple PISA
average, then none of the years of schooling (to the extent that they
mattered before controlling for quality) enter the growth equation in a
The final column of table 11 includes the number of researchers per
million of the population as a proxy for
/qual·i·ta·tive/ () pertaining to quality. Cf. quantitative.
pertaining to observations of a categorical nature, e.g. breed, sex.
outcomes at the tertiary level that have a relevance for economic growth
The results show that countries with more researchers have grown more
over the past 40 years. Introducing these measures has effects on the
contribution of the other variables. Years of tertiary education becomes
negatively related to growth, once R&D activity has been taken into
account. Educational quality remains significant, although years of
secondary schooling also becomes significant again.
6. Discussion and conclusions
The above results fail to find a significant effect of the higher
education sector (either through its initial level or its expansion) on
economic growth during the past forty years, whereas positive and
significant relationships are found with primary and secondary
education, measures of technical skills, research activity and capital
One response to these results would be to argue that there is a
relationship between economic growth and higher education but, despite
casting a wide net in an attempt to find some evidence that countries
grow faster if they invest more in higher education, the true
relationship is not easily drawn out in the regression specifications
used above. As I have shown in this paper, estimations are sensitive to
the data that are included, both in terms of the countries used and the
time period. Moreover, measurement error might be a persistent problem.
Although data quality continues to improve with each new set of
cross-country comparisons, I cannot rule out the possibility that
mismeasuring the size and importance of the tertiary education sector is
not responsible for these results. These are real difficulties in using
cross-national data to understand anything about the differences in
economic growth. Of course, these same difficulties tend to be ignored
when more positive results are discovered, so the most important
implication of this analysis is that more caution should be used in this
As a result, this paper does not go so far as to argue that higher
education does not, or cannot lead to economic growth. What it does
argue is that there is a sizeable and worrying gap between what is
asserted and what the most readily available evidence has to say when
applied to the most standard of approaches. It may come as a surprise to
policymakers in the UK and elsewhere that cross-country comparisons fail
to show any effect of higher education on growth–as noted in the Browne
Review, the OECD contends very strongly that this is precisely what
cross-country comparisons do show.
What alternative evidence is there? The lack of robust
n. (used with a sing. verb)
The study of the overall aspects and workings of a national economy, such as income, output, and the interrelationship among diverse economic sectors.
evidence has sometimes been supplemented with
n. (used with a sing. verb)
The study of the operations of the components of a national economy, such as individual firms, households, and consumers.
analysis. In particular, there is very strong evidence
that graduates command a large and significant wage premium over
non-graduates (Pscharoupoulos, 2009). If real wages are taken as a proxy
for productivity, then this evidence implies graduates are more
productive than nongraduates. Moreover, if graduates are more productive
than non-graduates, then even without
growth mechanisms, increases in participation in tertiary education
should have led to higher growth over the past 40 years.
Taken at face value, the results of this paper clearly clash with
this line of thinking. As
Wolf and others have discussed (see
Wolf, 2002) one reason for this might be that having a degree acts as a
filter for employers recruiting into higher paying jobs. In the
traditional signalling model (
, or “pantry” were hung the carcasses of a sheep or ewe, and two cows lately slaughtered.
– Sir W. Scott.
, 1973), individuals have private
information about their own productivity, and use the completion of a
degree to convey this information to employers. However, it is not
actually important whether graduates have prior differences in
productivity, just that there are a number of jobs which are higher
paying and for which the employers have a preference for hiring
graduates. (9) The main implication of this model is that while it makes
a lot of sense for any individual to pursue higher education in order to
be recruited into higher paying jobs (especially if that is what their
peers are doing), at the national level this decision should make little
difference. A country with a similar structure of occupations and
available technology, but far fewer graduates, would have an identical
national income and growth performance.
Some recent papers have used more sophisticated
n. (used with a sing. verb)
Application of mathematical and statistical techniques to economics in the study of problems, the analysis of data, and the development and testing of theories and models.
techniques than those presented here to find specific channels through
which higher education has added to growth. For example, Aghion et al.
(2009) find that increases in patenting in the US can be attributed to
exogenous increases in spending on fouryear degree courses at research
intensive universities and subsequent economic growth. Vandenbussche et
al. (2006) find a link between five-year growth rates and higher
education once distance from the technological frontier is controlled
for, again indicating that higher education plays a role through
technological innovation. Similar to my much simpler measure of the
supply of high level technical skills, these papers emphasise that the
interaction between some high level skills and technology could be
However, this does not imply that mass higher education necessarily
leads to higher growth. In part, this depends on what skills are being
produced as the system grows. This paper does not have adequate data to
control for the many differences in higher education, but differences
between subject of study and the quality of higher education
institutions will matter. What we do know is that as countries like the
UK have expanded their higher education system, the sector has become
. The fastest growing subjects tend not to be in
technical subjects–leading to repeated claims of shortages of science,
technology, engineering and maths (STEM) graduates–but in areas such as
creative arts and design, business and administration, psychology,
and mass communication. (10) It is less clear how the
skills produced on these courses lead to innovation and economic growth.
It is also worth noting that higher education is not the only way
to develop the necessary technical skills required to adopt new
technology or improve production processes. The expansion of higher
education has also led to the creation of more vocational courses which
has meant less young people following
system of learning a craft or trade from one who is engaged in it and of paying for the instruction by a given number of years of work. The practice was known in ancient Babylon, Egypt, Greece, and Rome, as well as in modern Europe and to some extent
pathways. UK labour
force data suggest that between 1995 and 2008, the share of apprentices
without degrees in employment fell from 15.2 per cent to 9.1 per cent;
Holmes and Mayhew (2013) estimate that this largely reflects changes in
the educational composition of occupations (5.0 per cent) rather than
the decline in occupations which were dominated by apprentice-trained
workers, such as skilled trades (1.1 per cent). This shift may come with
a cost if those practical or technical skills become
tr.v. de·plet·ed, de·plet·ing, de·pletes
To decrease the fullness of; use up or empty out.
exchange for more general academic knowledge.
Finally, the link between a growing higher education sector and
future growth also depends on what work new graduates are able to find
when they enter the labour market. Much of the discussion around human
capital theory which drives the growth models in section 2 assumes that
the demand side of the labour market is either
1. Of a kind and gentle disposition.
2. Showing gentleness and mildness. See Synonyms at kind1.
accommodating. If that were the case, then the supply of skills should
directly determine the productivity of labour. However, the demand side
of the labour market and the distribution of jobs also matters. Taking
the UK as an example, the growth of graduate-level jobs has been far
slower than the growth in graduates over the past 20 years, leading to
many graduates working in non-graduate jobs. The untested assumption is
that these graduates will be more productive than the non-graduates they
replace, but evidence of upskilling is thin, particularly in the service
sector occupations many new graduates enter (
, 2002; Keep and
Mayhew, 2004). Without an accommodating demand side, improvements in the
supply of graduate skills will at best be a necessary, but not
sufficient, condition for higher long-run growth.
Unless a specific
A sample drawn from a larger sample.
tr.v. sub·sam·pled, sub·sam·pling, sub·sam·ples
To take a subsample from (a larger sample).
is specified, all regressions use as
many countries as are available from World Bank data. Sample sizes
change significantly when certain variables are included, and the paper
has attempted to make it clear that the sensitivity of analyses to
sample is one reason that growth regressions should always be treated
with caution. The table below summarises which countries were included
in each regression. Specific regressions are denoted by the table they
appear in followed by the column of the table in parentheses.
Regression Countries 5.1 (1) Algeria, Argentina, Australia, Austria, Barbados, Belgium, Bolivia, Botswana, Brazil, Burundi, Cameroon, Canada, Chile, China, Colombia, Dem. Rep. of Congo, Rep. of Congo, Costa Rica, Cote d'Ivoire, Denmark, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Finland, France, Gabon, Ghana, Greece, Guatemala, Guyana, Honduras, Hong Kong, Hungary, Iceland, India, Iran, Israel, Italy, Japan, Kenya, Korea, Lesotho, Luxembourg, Madagascar, Malawi, Malaysia, Mauritania, Mexico, Morocco, Nepal, Netherlands, Nicaragua, Norway, Pakistan, Papua New Guinea, Paraguay, Peru, Philippines, Portugal, Rwanda, Senegal, Seychelles, Singapore, South Africa, Spain, Sri Lanka, Sudan, Sweden, Syria, Thailand, Trinidad and Tobago, Tunisia, Turkey, United Kingdom, United States, Uruguay, Venezuela, Zambia, and Zimbabwe. 5.1 (2) As 5.1 (1), except: Algeria, Brazil, Dem. Rep. of Congo, Rep. of Congo, Cote d'Ivoire, Egypt, Gabon, Guyana, Papua New Guinea, Seychelles, Singapore, Sri Lanka, Trinidad and Tobago, Zambia, and Zimbabwe. 5.1 (3) As 5.1 (2), except: Barbados, Bolivia, Canada, Costa Rica, Dominican Republic, Ecuador, Fiji, Guatemala, Honduras, Kenya, Malaysia, Nepal, Nicaragua, Paraguay, South Africa, Sudan, Syria, Uruguay and Venezuela. 5.2(l) As 5.1 (1), except: Madagascar and Seychelles, and including: Bangladesh, Belize, Benin, Central African Republic, Gambia, Indonesia, Latvia, Liberia, Niger, Panama, Sierra Leone and Togo. 5.2(2) As 5.2(I), except: Algeria, Austria, Bangladesh, Belgium, Belize, Benin, Bolivia, Brazil, Cameroon, Central African Republic, Rep. of Congo, France, Gabon, Gambia, Ghana, Indonesia, Latvia, Lesotho, Liberia, Malawi, Nepal, Netherlands, Niger, Panama, Portugal, Sierra Leone, Spain, Sudan, Togo, Turkey, and Zambia. 5.2 (3) As 5.2(I), except: Malawi. 5.2(4) As 5.2(2). 5.3 (1) As 5.3 (3), including: Ireland, Switzerland, and New Zealand 5.3 (2) As 5.3 (3), including: Hungary, Israel, and Luxembourg. 5.3 (3) Australia, Austria, Belgium, Canada, Denmark, Finland, France, Greece, Iceland, Italy, Japan, Korea, Mexico, Netherlands, Norway, Portugal, Spain, Sweden, Turkey, United Kingdom, and United States. 5.3 (4) As 5.3 (3). 5.4(l) As 5.3 (3). 5.4(2) As 5.3 (3). 5.4(3) As 5.3 (3). 5.5 (I) As 5.2 (1). 5.5 (2) Australia, Austria, Barbados, Belgium, Canada, Denmark, Finland, France, Greece, Hong Kong, Hungary, Iceland, Israel, Italy, Japan, Korea, Luxembourg, Netherlands, Norway, Portugal, Singapore, Spain, Sweden, Trinidad and Tobago, United Kingdom, and United States. 5.5 (3) Algeria, Argentina, Belize, Bolivia, Botswana, Brazil, Cameroon, Chile, China, Colombia, Rep. of Congo, Costa Rica, Cote d'Ivoire, Dominican Republic, Ecuador, Egypt, El Salvador, Fiji, Gabon, Ghana, Guatemala, Guyana, Honduras, India, Indonesia, Iran, Latvia, Lesotho, Malaysia, Mexico, Morocco, Nicaragua, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Senegal, South Africa, Sri Lanka, Sudan, Syrian Arab Republic, Thailand, Tunisia, Turkey, Uruguay, Venezuela and Zambia 5.5 (4) Bangladesh, Benin, Burundi, Central African Republic, Dem Rep. of Congo, Gambia, Kenya, Liberia, Malawi, Mauritania, Nepal, Niger, Rwanda, Sierra Leone, Togo, and Zimbabwe. 5.5(5) As 5.5(l), except: Malawi. 5.5 (6) As 5.5 (2) 5.5 (7) As 5.5 (3) 5.5 (8) As 5.5(4), except: Malawi. 5.6(l) As 5.2(2). 5.6(2) As 5.2(2). 5.7(l) As 5.3 (3). 5.7(2) As 5.3 (3), including: Hungary, Israel and Luxembourg 5.7 (3) As 5.3 (3), including: Argentina, Brazil, Chile, Colombia, Hong Kong, Hungary, Indonesia, Israel, Latvia, Luxembourg, Thailand, Tunisia, Uruguay. 5.7(4) As 5.3 (3), including: Argentina, Brazil, Colombia, Hong Kong, Hungary, Latvia, Luxembourg, Tunisia.
Figure A.1 shows shorter-term growth rates over this time period
for a few key countries. The three high income, high education countries
have the most stable growth rates over the entire time period. However,
Swed. Sverige, officially Kingdom of Sweden, constitutional monarchy (2005 est. pop. 9,002,000), 173,648 sq mi (449,750 sq km), N Europe, occupying the eastern part of the Scandinavian peninsula.
is clearly affected by a large
shock in the
early 1990s, followed by a recovery in the late 1990s and 2000s.
Upper middle income countries
, officially Republic of Singapore, republic (2005 est. pop. 4,426,000), 240 sq mi (625 sq km).
, Port. Brasil, officially Federative Republic of Brazil, republic (2005 est. pop. 186,113,000), 3,286,470 sq mi (8,511,965 sq km), E South America.
both had high
growth rates in the 1960s, which have subsequently declined. However,
the decline in Brazil has been far more pronounced than in Singapore
(and other East Asian countries such as
, Mandarin Xianggang, special administrative region of China, formerly a British crown colony (2005 est. pop. 6,899,000), land area 422 sq mi (1,092 sq km), adjacent to Guangdong prov.
, Korean Hanguk or Choson, region and historic country (85,049 sq mi/220,277 sq km), E Asia.
below the long-run growth rates of higher income countries. Again,
demand side factors are evident here–first in the build-up of debt
during the 1970s, followed by a debt crisis and in the early 1980s a
tr.v. pro·longed, pro·long·ing, pro·longs
1. To lengthen in duration; protract.
2. To lengthen in extent.
period of weak demand, with only a slight recovery in the
, Arab. Misr, biblical Mizraim, officially Arab Republic of Egypt, republic (2005 est. pop. 77,506,000), 386,659 sq mi (1,001,449 sq km), NE Africa and SW Asia.
and China both had relatively low growth in the 1960s,
considering their scope for technological catch-up. China has undergone
huge economic reforms since the 1970s and has had sustained levels of
high growth since that point. Egypt’s growth has also been affected
by a debt crisis in the 1980s. However, in general it would appear that
the two countries have very different long-run growth rates.
Interestingly, they have both followed very similar policies on
education, increasing attainment at all levels. The main difference
appears to be in years of primary education at the start of the period,
consistent with results found in the main body of this paper.
To improve the robustness of the result in section 5.3, I use later
measures of educational attainment of the workforce–recorded in 1985
and 2005. Data might be more accurate in recent years, compared to what
has been derived by Barro and Lee for the 1960s. Moreover, later
measures might better reflect changes in policy during the time period.
I also look at shorter time periods. Table A2 shows these results.
Columns (1) and (2) show the results for the later measures. Columns (3)
to (6) show the results for ten-year time periods starting in 1966-76
and finishing in 1996-2006. There is no indication in these results that
higher education is related to economic growth.
[FIGURE A1 OMITTED]
Table A1. Country long-run growth rates and educational attainment, 1966-2006 Annualised Years of Years of growth primary secondary rate, education, education, 1966-2006 1966 1966 Algeria 0.016 0.80 0.26 Argentina 0.011 4.97 0.85 Australia 0.021 5.52 3.54 Austria 0.026 3.38 1.68 Bangladesh 0.013 0.95 0.28 Barbados 0.019 5.01 1.62 Belgium 0.024 5.42 1.57 Belize 0.033 6.44 0.90 Benin 0.006 0.60 0.16 Bolivia 0.001 2.46 1.04 Botswana 0.068 1.43 0.10 Brazil 0.024 1.69 0.64 Burundi 0.003 0.74 0.05 Cameroon 0.007 1.18 0.25 Canada 0.020 5.42 2.78 Central African Republic -0.010 0.51 0.11 Chile 0.027 4.22 1.34 China 0.070 2.34 0.41 Colombia 0.020 2.67 0.70 Dem. Rep. Congo -0.031 0.88 0.08 Rep. of Congo 0.015 1.26 0.36 Costa Rica 0.022 3.51 0.52 Cote d'Ivoire -0.005 0.73 0.27 Denmark 0.020 6.06 1.89 Dominican Republic 0.032 2.67 0.64 Ecuador 0.016 3.24 0.59 Egypt 0.029 0.70 0.28 El Salvador 0.009 1.99 0.34 Fiji 0.020 4.66 0.39 Finland 0.028 5.20 0.92 France 0.022 3.73 0.80 Gabon 0.010 1.02 0.35 Gambia 0.005 0.31 0.20 Ghana 0.004 1.44 0.78 Greece 0.026 4.49 1.83 Guatemala 0.012 1.28 0.23 Guyana 0.010 4.42 0.78 Honduras 0.012 1.81 0.31 Hong Kong 0.046 3.75 1.82 Hungary 0.028 7.06 0.43 Iceland 0.025 5.21 1.21 India 0.030 1.08 0.18 Indonesia 0.040 1.95 0.21 Iran 0.016 1.02 0.37 Israel 0.026 5.00 2.63 Italy 0.025 4.06 1.11 Japan 0.030 5.41 2.26 Kenya 0.011 1.55 0.15 Korea 0.058 4.04 1.31 Latvia 0.032 3.17 1.65 Lesotho 0.024 3.17 0.06 Liberia -0.030 0.66 0.23 Luxembourg 0.032 5.20 1.75 Malaysia 0.039 2.60 0.68 Mauritania 0.004 1.51 0.09 Mexico 0.018 2.64 0.46 Morocco 0.024 0.39 0.29 Nepal 0.012 0.17 0.09 Netherlands 0.023 5.72 1.47 Nicaragua -0.009 2.05 0.46 Niger -0.018 0.36 0.04 Norway 0.029 6.87 1.01 Pakistan 0.024 0.86 0.48 Panama 0.018 3.75 0.99 Papua New Guinea 0.003 0.72 0.23 Paraguay 0.016 3.10 0.54 Peru 0.006 3.06 0.78 Philippines 0.012 3.64 0.83 Portugal 0.032 2.97 0.45 Rwanda 0.012 0.84 0.08 Senegal -0.002 1.83 0.17 Sierra Leone -0.001 0.49 0.17 Singapore 0.059 2.90 1.36 South Africa 0.006 3.44 0.99 Spain 0.026 3.02 0.56 Sri Lanka 0.033 4.30 0.80 Sudan 0.014 0.46 0.12 Sweden 0.021 5.32 2.11 Syria 0.025 1.43 0.25 Thailand 0.045 3.48 0.28 Togo -0.002 0.55 0.05 Trinidad and Tobago 0.022 5.23 0.73 Tunisia 0.031 0.90 0.36 Turkey 0.025 1.67 0.38 United Kingdom 0.023 5.27 1.42 United States 0.020 5.57 3.92 Uruguay 0.014 4.07 1.00 Venezuela -0.003 2.75 0.54 Zambia -0.011 2.51 0.05 Zimbabwe -0.002 2.22 0.45 Change in Years of years of tertiary primary education, education, 1966 1966-2006 Algeria 0.01 3.72 Argentina 0.12 1.50 Australia 0.55 0.40 Austria 0.07 0.44 Bangladesh 0.01 2.09 Barbados 0.04 0.83 Belgium 0.19 0.14 Belize 0.19 0.89 Benin 0.01 1.61 Bolivia 0.10 3.75 Botswana 0.00 4.64 Brazil 0.05 3.45 Burundi 0.00 1.71 Cameroon 0.01 2.93 Canada 0.46 0.43 Central African Republic 0.00 1.96 Chile 0.09 1.18 China 0.03 2.58 Colombia 0.06 1.52 Dem. Rep. Congo 0.00 1.63 Rep. of Congo 0.02 2.52 Costa Rica 0.09 1.81 Cote d'Ivoire 0.03 2.21 Denmark 0.28 0.30 Dominican Republic 0.03 1.77 Ecuador 0.05 1.55 Egypt 0.07 2.98 El Salvador 0.03 3.72 Fiji 0.07 1.82 Finland 0.15 0.53 France 0.11 0.87 Gabon 0.04 3.57 Gambia 0.01 1.44 Ghana 0.01 2.46 Greece 0.10 1.21 Guatemala 0.02 1.81 Guyana 0.02 0.76 Honduras 0.02 3.16 Hong Kong 0.14 1.27 Hungary 0.12 0.86 Iceland 0.13 0.47 India 0.03 1.99 Indonesia 0.01 2.31 Iran 0.03 3.14 Israel 0.36 1.37 Italy 0.07 0.57 Japan 0.15 0.44 Kenya 0.01 4.31 Korea 0.12 1.66 Latvia 0.22 1.85 Lesotho 0.00 1.85 Liberia 0.03 1.95 Luxembourg 0.17 0.28 Malaysia 0.04 2.60 Mauritania 0.01 1.74 Mexico 0.06 2.44 Morocco 0.02 2.05 Nepal 0.01 2.14 Netherlands 0.13 0.00 Nicaragua 0.10 1.62 Niger 0.01 0.83 Norway 0.13 -0.30 Pakistan 0.03 1.81 Panama 0.10 1.61 Papua New Guinea 0.00 2.65 Paraguay 0.05 1.93 Peru 0.12 1.95 Philippines 0.27 1.73 Portugal 0.04 2.11 Rwanda 0.00 2.33 Senegal 0.01 1.84 Sierra Leone 0.01 1.91 Singapore 0.06 2.12 South Africa 0.07 2.26 Spain 0.07 2.14 Sri Lanka 0.01 1.88 Sudan 0.02 1.91 Sweden 0.20 0.52 Syria 0.03 2.39 Thailand 0.03 1.47 Togo 0.00 3.14 Trinidad and Tobago 0.03 1.37 Tunisia 0.03 3.14 Turkey 0.03 2.57 United Kingdom 0.14 0.33 United States 0.54 0.38 Uruguay 0.11 1.45 Venezuela 0.06 1.75 Zambia 0.01 2.77 Zimbabwe 0.06 3.07 Change in Change in years of years of secondary tertiary education, education, 1966-2006 1966-2006 Algeria 2.68 0.24 Argentina 1.55 0.14 Australia 1.40 0.46 Austria 3.36 0.38 Bangladesh 1.77 0.10 Barbados 1.81 0.05 Belgium 2.50 0.66 Belize 0.87 0.05 Benin 1.17 0.06 Bolivia 1.68 0.32 Botswana 2.97 0.10 Brazil 1.19 0.15 Burundi 0.34 0.02 Cameroon 1.33 0.05 Canada 2.18 0.86 Central African Republic 0.87 0.05 Chile 2.32 0.56 China 2.10 0.17 Colombia 1.88 0.22 Dem. Rep. Congo 0.85 0.03 Rep. of Congo 1.74 0.03 Costa Rica 1.74 0.38 Cote d'Ivoire 0.88 0.09 Denmark 1.05 0.29 Dominican Republic 1.56 0.34 Ecuador 1.82 0.40 Egypt 2.33 0.24 El Salvador 0.99 0.27 Fiji 2.16 0.27 Finland 2.37 0.60 France 3.93 0.44 Gabon 2.39 0.28 Gambia 1.07 0.04 Ghana 2.00 0.07 Greece 1.56 0.70 Guatemala 0.59 0.06 Guyana 2.62 -0.01 Honduras 1.37 0.16 Hong Kong 2.61 0.28 Hungary 2.66 0.36 Iceland 2.45 0.64 India 1.28 0.12 Indonesia 1.19 0.06 Iran 3.08 0.43 Israel 1.24 0.68 Italy 3.11 0.23 Japan 2.07 0.93 Kenya 1.01 0.07 Korea 3.45 0.89 Latvia 3.00 0.31 Lesotho 0.93 0.04 Liberia 1.13 0.24 Luxembourg 2.13 0.32 Malaysia 3.46 0.30 Mauritania 0.68 0.05 Mexico 2.38 0.42 Morocco 1.42 0.23 Nepal 0.90 0.07 Netherlands 2.89 0.60 Nicaragua 1.50 0.34 Niger 0.29 0.02 Norway 4.03 0.60 Pakistan 1.59 0.15 Panama 2.28 0.57 Papua New Guinea 0.28 0.04 Paraguay 1.94 0.05 Peru 2.23 0.54 Philippines 1.44 0.71 Portugal 1.80 0.23 Rwanda 0.33 0.03 Senegal 0.75 0.07 Sierra Leone 0.78 0.03 Singapore 1.54 0.49 South Africa 1.42 0.08 Spain 3.28 0.65 Sri Lanka 3.36 0.45 Sudan 0.49 0.05 Sweden 2.79 0.57 Syria 0.66 0.06 Thailand 1.26 0.30 Togo 1.62 0.06 Trinidad and Tobago 1.85 0.09 Tunisia 1.91 0.23 Turkey 1.62 0.21 United Kingdom 1.67 0.51 United States 1.55 0.95 Uruguay 1.22 0.15 Venezuela 0.96 0.33 Zambia 1.11 0.03 Zimbabwe 1.72 -0.04 Table A2. Economic growth and types of schooling in the OECD, alternative education measures (1) (2) (3) Initial GDP/capita -0.0010 -0.0011 ** -0.0007 (0.107) (0.044) (0.489) Years of schooling, 0.0045 *** 0.0036 ** 0.00426 primary (0.001) (0.016) (0.231) Years of schooling, 0.0076 ** 0.0071 0.0092 secondary (0.015) (0.001) (0.291) Years of schooling, -0.0211 -0.0020 -0.0475 tertiary (0.156) (0.805) (0.037) Constant 0.0044 0.0121 ** 0.0217 *** (0.324) (0.044) (0.000) N 91 91 91 [R.sup.2] 0.267 0.323 0.044 (4) (5) (6) Initial GDP/capita -0.0001 -0.0005 -0.0003 (0.862) (0.466) (0.439) Years of schooling, 0.0035 * 0.0072 ** 0.0014 primary (0.090) (0.016) (0.522) Years of schooling, 0.0106 * 0.0080 0.0040 secondary (0.089) (0.235) (0.293) Years of schooling, -0.0667 * -0.0298 -0.0012 tertiary (0.060) (0.334) (0.939) Constant -0.0065 0.0206 *** 0.0096 (0.264) (0.003) (0.206) N 91 91 91 [R.sup.2] 0.135 0.130 0.043 Source: World Bank data. Dependent variable: (1)-(2) real GDP growth 1966-2006; (3) real GDP growth 1966-1976; (4) real GDP growth 1976-1986; (5) real GDP growth 1986-1996; (6) real GDP growth 1996-2006. Average years of schooling: (1) 1985; (2) 2005; (3) 1965; (4) 1975; (5) 1985; (6) 1995. Note: GDP/capita is measured in 1000s. *, ** and denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table A3. Growth in graduates by subject, 1994/5 to 2010/11 Annualised growth rate of Subject qualifications obtained (%) Mass communication 9.7 Subjects allied to Medicine 7.2 Biological Sciences 6.5 Creative Arts & Design 6.3 Business & Administrative Studies 4.9 Veterinary Science 3.9 Social, Economic & Political Studies 3.8 Law 3.6 Computer Science 3.6 Medicine & Dentistry 3.5 Mathematical Sciences 3.4 Humanities 3.3 Languages 2.2 Agriculture & Related Subjects 2.0 Architecture, Building & Planning 1.7 Education 1.2 Physical Sciences 0.6 Engineering & Technology 0.2 All UK HE Institutions 2.8 Source: HESA.
Aghion, P., Boustan, L., Hoxby, C., Vandenbussche, J. (2009),
‘The causal impact of education on economic growth: evidence from
the United States’, accessed online 5
town (1990 pop. 12,329), Worcester co., E central Mass.; inc. 1732. A Shaker house and cemetery, a Native American museum, and a Harvard observatory are there.
Barro, R. and Lee, J. (2010), ‘A new data set of educational
attainment in the world, 1950-2010’,
NBER Nittany and Bald Eagle Railroad Company
Working Paper No. 15902.
BIS (2011), ‘Supporting analysis for the Higher Education
White Paper’, BIS Economics Paper No. 14. Browne Review (2010),
Securing a Sustainable Future for Higher Education,
DES (2004), The Future of Higher Education.
Gemmell, N. (1996), ‘Evaluating the impacts of human capital
stocks and accumulation on economic growth: some new evidence’,
Oxford Bulletin of Economics and Statistics, 58(I), pp. 9-28.
–(1997), ‘Externalities to higher education: a review of the
new growth literature’, Report to the National Committee of Inquiry
into Higher Education.
Hanuschek, E. and Kimko, D. (2000), ‘Schooling, labour force
quality, and the growth of nations’, American Economic Review, 90,
Hanushek, E. and Woessmann, L. (2007), ‘The role of education
quality for economic growth’, World Bank Working Paper No. 4122.
Holmes, C. and Mayhew, K. (2013), ‘Are universities the best
way to create high skills? The consequences of mass higher education in
the UK’, conference paper presented at the 31st International
Labour Process Conference,
main campus at New Brunswick, N.J.; land-grant and state supported; coeducational except for Douglass College; chartered 1766 as Queen’s College, opened 1771.
Campuses and Facilities
Rutgers maintains three campuses.
, 20 March.
Jones, C. (I 995), ‘R&D based models of economic
growth’, Journal of Political Economy, 103(4), pp. 759-84.
–(1999), ‘Growth: with or without scale effects?’,
American Economic Review Papers and Proceedings, 89, pp. 139-44.
Keep, E. and Mayhew, K. (2004), ‘The economic and
distributional implications of current policies on higher
Oxford Review of Economic Policy
, 20(2), pp. 298-313.
Lindahl, M. and Krueger, A.B. (2001), ‘Education and growth:
why and for whom?’,
Quarterly Journal of Economics
, 39(4) pp.
Lucas, R. (1988), ‘On the
branch of physics concerned with motion and the forces that tend to cause it; it includes study of the mechanical properties of matter, such as density, elasticity, and viscosity.
development’, Journal of Monetary Economics, 22, pp. 3-42
Mankiw, G., Romer, D. and Weil, D. (1992), ‘A contribution to
the empirics on economic growth’, Quarterly Journal of Economics,
107, pp. 407-37.
Mason, G. (2002), ‘High skills utilisation under mass higher
education: graduate employment in service industries in Britain’,
Journal of Education and Work, 15(4), pp. 427-56
Pscharoupoulos, G. (2009), ‘Returns to investment in higher
education: European Survey–a contribution to the Higher Education
Funding Reform Project CHEPS-led consortium for the European
Commission’, accessed online 4 April 2013:
Robbins Report (1963),
Education in England
: Higher Education,
Romer, P. (1986), ‘Increasing returns and long run economic
growth’, Journal of Political Economy, 94(5), pp. 1002-37.
–(1990), ‘Endogenous technological change’, Journal of
Political Economy, 98(5), pp. $71 – 102.
–(1994), ‘The origins of endogenous growth’, Journal of
Economic Perspectives, 8(I), pp. 3-22.
Sianesi, B. and Van Reenan, J. (2003), ‘The returns to
education: macroeconomics’, Journal of Economic Surveys, 17 (2),
Solow, R. (1956), ‘A contribution to the theory of economic
growth’, Quarterly Journal of Economics, 70(I), pp. 65-94.
Spence, M. (1973), ‘Job market signaling’, Quarterly
Journal of Economics, 87(3): pp. 355-74.
Thurow, L, (1976), Generating
in mathematics, statement that a mathematical expression is less than or greater than some other expression; an inequality is not as specific as an equation, but it does contain information about the expressions involved.
city (1991 pop. 303,165), SE Ont., Canada, on the Thames River. The site was chosen in 1792 by Governor Simcoe to be the capital of Upper Canada, but York was made capital instead. London was settled in 1826.
river, c.200 mi (320 km) long, rising in two main forks in the Selwyn Mts., E Yukon Territory, Canada, and flowing generally W to the Pelly River. It was an important route to the gold fields from c.1890 to 1900.
Vandenbussche, J., Aghion, A. and Meghir, C. (2006), ‘Growth,
distance from the frontier and composition of human capital’,
Journal of Economic Growth, II, pp. 97-107.
Wolf, A. (2002), Does Education Matter?,
originally the common name for the now extinct great auk of the N Atlantic and now used (since the 19th cent.) for the unrelated antarctic diving birds.
Wolff, E. (2001), ‘Productivity convergence and education:
evidence from OECD countries’, Industrial and Corporate Change,
10(3): pp. 735-59.
Young, A. (1998), ‘Growth without scale effects’, Journal
of Political Economy, 106(I), pp. 41-63.
(1) DES (2004) explicitly refers to the literature review by
Sianesi and Van Reenen (2003). However, the sections of this review
which deal with tertiary education largely draw on Gemmell (1996).
(2) As H is a stock variable, it can be thought of as the product
of the average level of human capital per worker (call this h) and the
labour force (L), where the former is controlled, at least in part,
through the provision of education and training to workers.
(3) The predicted relationships between characteristics of the
research sector and growth differ across models, given different
underlying assumptions. Romer’s (1990) model implies that economic
growth is proportional to the stock of human capital engaged in R&D
(rather than changes in the stock). This has some problematic
implications, as it suggests that, holding everything else constant
(including the share of resources allocated towards research), larger
countries should grow faster. It also implies that growth rates should
1. Recurring regularly or frequently:
accelerating in countries with population growth, which
is not observed in real life. Introducing diminishing returns to the
existing stock of knowledge, the R&D production function (Jones,
1995) removes this
and suggests long-run economic growth
depends solely on the growth rate of the number of workers engaged in
R&D. Consequently, policy levers, such as an increase in the
resources allocated towards research, only have short-term effects, as
diminishing returns to knowledge also have a
v. di·min·ished, di·min·ish·ing, di·min·ish·es
a. To make smaller or less or to cause to appear so.
economic growth rates. Similarly, multi-sector models (e.g. Young,
1998), where economic growth leads to a widening variety of products,
also deal with the scale effects in Romer (I 990), but retain the
long-run effect of the policy
simple machine consisting of a bar supported at some stationary point along its length and used to overcome resistance at a second point by application of force at a third point. The stationary point of a lever is known as its fulcrum.
(research intensity). Jones (1999)
provides a good summary of each class of model.
(4) See Mankiw, Romer and Well (I 992), equation (I 6) for the full
in grammar: see inflection.
(5) The original Solow model assumed a closed economy, so that the
saving rate and investment rate could be equated, as I do in this
estimation. In an open economy, the saving rate may be different from
the investment rate when there are international capital flows.
Following Mankiw, Romer and Well, and given the data that are available,
the assumption made in this analysis is that the investment share of GDP
is an unbiased proxy of the equilibrium saving rate.
(6) Depreciation is assumed to be an identical rate across all
(7) There are a number of potential biases in this estimation. As
noted in Wolff (200 I), causal effects from growth to enrolment may bias
the estimate on higher education upwards. Enrolment rates may not take
account of the export of higher education. Countries with high enrolment
rates, such as the UK and US, tend to have a significant number of
overseas students who may not add to their university country’s
supply of human capital once their studies are complete. This would have
a negative bias on enrolment rate effects.
(8) This same approach was taken in all other regressions. The
results are not presented here; however, no significant relationships
were found through performing this check.
(9) This is the key idea behind
n. 1. (Meteor.) A dry sirocco in the Madeira Islands.
model (Thurow, 1973).
(10) Based on an analysis of
HESA Higher Education Student Affairs
HESA Higher Education Statistics Agency Ltd.
data between 1994/5 and 2010/I I
data shown in Appendix table A2.
, Edward Gordon 1872-1966.
British theatrical producer, director, and designer whose innovative productions and simplified stage designs influenced modern theater.
Holmes, Oxford University. E-mail
email@example.com. I would like to thank Ken Mayhew for
his advice and suggestions. I’m also thankful for the helpful
comments of three anonymous referees that have greatly improved my
Table 1. Educational attainment N Mean s.d. Min Max Years of schooling, 1965 91 3.68 2.53 0.27 10.04 Years of schooling, primary, 1965 91 2.83 1.85 0.17 7.06 Years of schooling, secondary, 1965 91 0.77 0.77 0.04 3.92 Years of schooling, tertiary, 1965 90 0.08 0.11 0.00 0.55 Change in years of schooling, 1965-2005 91 3.86 1.21 1.14 7.71 Change in years of schooling, primary, 91 1.83 1.01 -0.30 4.64 1965-2005 Change in years of schooling, secondary 91 1.76 0.87 0.28 4.03 1965-2005 Change in years of schooling, tertiary, 90 0.28 0.24 -0.04 0.95 1965-2005 Table 2. Educational attainment, by country income group High High Upper Lower income income middle middle Low OECD non-OECD income income income Years of 6.86 5.67 3.37 2.60 1.00 schooling, 1965 Years of schooling, primary, 1965 5.00 4.22 2.66 2.15 0.84 Years of schooling, secondary, 1965 1.67 1.38 0.65 0.41 0.14 Years of schooling, tertiary, 1965 0.19 0.07 0.06 0.05 0.01 Change in years of schooling, 1965-2005 3.69 3.58 4.67 3.78 3.15 Change in years of schooling, primary, 1965-2005 0.69 1.40 2.33 2.22 2.16 Change in years of schooling, secondary 1965-2005 2.43 1.95 2.04 1.40 0.94 Change in years of schooling, tertiary, 1965-2005 0.57 0.23 0.30 0.15 0.06 Table 3. Correlation matrix, educational attainment Years of Years of Years of schooling, schooling, schooling, primary, secondary, tertiary, 1965 1965 1965 Years of schooling, 1.000 primary, 1965 Years of schooling, 0.706 1.000 secondary, 1965 Years of schooling, 0.643 0.884 1.000 tertiary, 1965 Change in years of -0.692 -0.586 -0.517 schooling, primary, 1965-2005 Change in years of 0.445 0.303 0.171 schooling, secondary 1965-2005 Change in years of 0.609 0.694 0.656 schooling, tertiary, 1965-2005 Change in Change in Change in years of years of years of schooling, schooling schooling, primary, secondary, tertiary, 1965-2005 1965-2005 1965-2005 Years of schooling, primary, 1965 Years of schooling, secondary, 1965 Years of schooling, tertiary, 1965 Change in years of 1.000 schooling, primary, 1965-2005 Change in years of -0.190 1.000 schooling, secondary 1965-2005 Change in years of -0.450 0.517 1.000 schooling, tertiary, 1965-2005 Table 4. Correlation matrix, educational attainment and quality Years of Years of Years of schooling, schooling, schooling, primary, secondary, tertiary, 1965 1965 1965 Years of schooling, 1.000 primary, 1965 Years of schooling, 0.479 1.000 secondary, 1965 Years of schooling, 0.504 0.894 1.000 tertiary, 1965 PISA 2006 scores 0.628 0.493 0.414 Number of researchers 0.605 0.447 0.401 Number of PISA researchers, 2006 per million scores population Years of schooling, primary, 1965 Years of schooling, secondary, 1965 Years of schooling, tertiary, 1965 PISA 2006 scores 1.000 Number of researchers 0.712 1.000 Table 5. Steady state convergence (1) (2) (3) In (Initial GDP/capita) -0.0036 * -0.0042 ** -0.0061 (0.054) (0.009) (0.003) In (Investment/GDP) 0.0308 *** 0.0284 *** 0.0203 ** (0.005) (0.000) (0.024) In (Primary enrolment -0.0142 rate) (0.426) In (Secondary enrolment 0.0171 *** 0.0206 *** rate) (0.001) (0.006) In (Tertiary enrolment 0.0010 rate) (0.753) Constant -0.0368 -0.1069 ** -0.0176 (0.347) (0.000) (0.849) N 81 66 47 [R.sup.2] 0.292 0.446 0.545 Source: World Bank. Dependent variable: real GDP growth 1966 2006. Note: *, ** and ** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table 6. Economic growth and accumulation of factors (1) (2) Change in average years of schooling 0.005 ** 0.003 ** (0.001) (0.031) Change in average years of schooling, primary Change in average years of schooling, secondary Change in average years of schooling, tertiary Growth rate of capital 0.003 ** (0.000) Constant 0.002 0.001 (0.768) (0.118) N 91 60 [R.sup.2] 0.126 0.606 (3) (4) Change in average years of schooling 0.001 0.000 (0.406) (0.946) Change in average years of 0.008 * 0.005 ** schooling, primary (0.000) (0.001) Change in average years of 0.013 0.012 * schooling, secondary (0.129) (0.057) Change in average years of 0.003 * schooling, tertiary (0.000) Growth rate of capital -0.003 -0.004 (0.621) (0.403) Constant 90 60 0.254 0.718 N [R.sup.2] Note: *, ** and *** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table 7. Economic growth and schooline in the OECD All available OECD data 1960-2000 1966-2006 Initial GDP/capita -0.0029 *** -0.0009 (0.000) (0.000) Initial years 0.0017 * 0.0006 of schooling (0.051) (0.566) Constant 0.0399 ** 0.0060 (0.000) (0.034) N 24 24 [R.sup.2] 0.596 0.166 OECD countries in both 1960-2000 1966-2006 Initial GDP/capita -0.0031 *** -0.0016 ** (0.000) (0.017) Initial years 0.0023 ** 0.0020 of schooling (0.028) (0.143) Constant 0.0383 *** 0.0285 *** 0.000 (0.000) N 21 21 [R.sup.2] 0.549 0.310 Source: Hanushek and Woessmann (2007), World Bank data. Dependent variable: real GDP growth 1966-2006. Columns (I) and (3) use GDP growth from 1966-2000 and columns (2) and (4) use GDP growth from 1966-2006. Note: GDP/capita is measured in 1000s. and *** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table 8. Economic growth and types of schooling in the OECD (1) (2) (3) Initial GDP/capita -0.0019 ** -0.0019 *** -0.0018 *** (0.007) (0.008) (0.010) Initial years of schooling, 0.0044 ** 0.0042 ** 0.0044 ** primary (0.044) (0.053) (0.042) Initial years of schooling, 0.0044 0.0003 secondary (0.283) (0.900) Initial years of schooling, -0.0289 -0.0068 tertiary (0.237) (0.698) Constant 0.0217 *** 0.0236 *** 0.0231 *** (0.003) (0.001) (0.001) N 21 21 21 [R.sup.2] 0.453 0.384 0.394 Source: World Bank data. Dependent variable: real GDP growth 1966-2006. Note: GDP/capita is measured in 1000s. and ** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table 9. Economic growth and schooling by country income High Middle Low All income income income Initial -0.001 -0.0009 *** -0.0049 ** -0.0517 ** GDP/capita (0.124) (0.104) (0.013) (0.035) Schooling 0.0039 *** -0.0012 0.0021 0.0053 (0.001) (0.367) (0.160) (0.343) Primary 0.0042 *** -0.0023 (0.004) (0.257) Secondary 0.0122 ** 0.0027 (0.031) (0.556) Tertiary -0.0657 * -0.0142 (0.057) (0.616) Constant 0.0068 ** 0.0447 *** 0.0191 *** 0.0096 (0.034) (0.000) (0.000) (0.285) N 91 26 49 16 [R.sup.2] 0.163 0.283 0.13 0.311 High Middle Low All income income income Initial -0.0010 -0.0010 * -0.0048 ** -0.0517 * GDP/capita (0.129) (0.087) (0.02) (0.094) Schooling Primary 0.0027 0.0054 (0.217) (0.523) Secondary 0.0032 0.0770 (0.755) (0.262) Tertiary -0.0293 -0.4939 (0.621) (0.339) Constant 0.0071 0.0479 *** 0.0186 *** 0.0044 (0.034) (0.000) (0.000) (0.711) N 90 26 49 15 [R.sup.2] 0.199 0.332 0.136 0.388 Source: World Bank data. Dependent variable: real GDP growth 1966-2006. Note: GDP/capita is measured in 1000s. *, ** and *** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table 10. Economic growth and schooling by country income (1) (2) Change in average years of schooling, 0.0055 ** 0.0034 * primary (0.016) (0.089) Change in average years of schooling, 0.0057 * 0.0045 ** secondary (0.028) (0.021) Change in average years of schooling, 0.0066 0.0061 tertiary (0.528) (0.400) Initial years of schooling, primary 0.0036 ** 0.0020 (0.020) (0.128) Initial years of schooling, secondary 0.0073 0.0074 * (0.145) (0.038) Initial years of schooling, tertiary -0.0482 -0.0341 (0.157) (0.138) Growth rate of capital 0.0029 ** (0.000) Constant -0.0 158 ** -0.0151 (0.029) (0.019) N 90 60 [R.sup.2] 0.328 0.756 Source: World Bank data. Dependent variable: real GDP growth 1966-2006. Note: *, ** and ** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses. Table 11. Economic growth, schooling and quality OECD Initial -0.0018 *** -0.0011 ** GDP/capita (0.002) (0.025) Primary 0.0020 0.0004 (0.270) (0.769) Secondary 0.0012 0.0048 (0.712) (0.209) Tertiary -0.0135 -0.0345 (0.502) (0.143) HW2007 0.0002 *** quality (0.007) PISA 2006 0.000 1 * (0.051) Researchers Constant -0.0412 * -0.0176 (0.066) (0.418) N 21 24 [R.sup.2] 0.659 0.417 All available Initial -0.0012 *** -0.0016 *** GDP/capita (0.010) (0.005) Primary -0.0009 -0.0005 (0.519) (0.660) Secondary -0.0037 0.0064 ** (0.334) (0.078) Tertiary -0.0292 -0.0409 ** (0.224) (0.058) HW2007 quality PISA 2006 0.000 1 *** 0.000 1 ** (0.002) (0.039) Researchers 0.0019 * (0.098) Constant -0.0166 -0.0090 (0.235) (0.554) N 34 29 [R.sup.2] 0.430 0.600 Source: Hanushek and Woessmann (2007), World Bank data. Dependent variable: real GDP growth 1966-2006. Note: GDP/capita is measured in 1000s. *, and *** denote significance at the 90%, 95% and 99% level respectively. P-values in parentheses.