:
I
^liiil
i;lj(|(||lf|pi;i:.
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/mediumruneffectsOOdufl
Dewey
Technology
Department of Economics
Massachusetts
Institute ot
Working Paper Series
The Medium Run Effects of Education Expansion: Evidence
from a Large School Construction
Program in Indonesia
Esther Duflo,
MIT
Working Paper 01 -46
November
2001
Roonn E52-251
50 iVJemoriai Drive
Cannbriclge, MA 02142
paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http://papers.ssrn Com/abstract=29431
This
\\\
Technology
Department of Economics
Working Paper Series
Massachusetts
Institute of
The Medium Run Effects of Education Expansion: Evidence
from a Large School Construction
Program in Indonesia
Esther Duflo,
MIT
Working Paper 01 -46
November 2001
Room
E52-251
50 Memorial Drive
Cambridge,
MA 02142
This paper can be downloaded without charge from the
Social Science Research Network Paper Collection at
http: / /papers. ssrn. com/paper. taf?abstractjd=294311
MASSACHUSEHS
INSTITUTE
OF TECHNOLOGY
MAR
7 2002
LIBRARIES
The Medium Run
Effects of Educational Expansion:
Evidence from a Large School Construction Program
Esther Duflo
in
Indonesia
*
November 2001
Abstract
This paper studies the
of
human
capital in a developing country.
built over 61,000
education
medium run consequences of an
among
From 1974
increase in the rate of accumulation
to 1978, the Indonesian government
primary schools. The school construction program
individuals
who were young enough
led to
an increase
in
to attend primary school after 1974,
but not among the older cohorts. 2SLS estimates suggest that an increase of 10 percentage
points in the proportion of primary school graduates in the labor force reduced the wages of
the older cohorts by 3.8% to
to 7%.
I
10% and
increased their formal labor force participation by
propose a two-sector model as a framework to interpret these findings. The results
suggest that physical capital did not adjust to the faster increase in
*I
for
thank participants at the conference "New Research on Education
I
in
for his discussion of the paper.
thank Lucia Breierova
I
also
human
for
comments, and particular to
thank two referees and the editor
for excellent research assistance,
capital.
Developing Countries" at the Center
Research on Economic Development and Policy Reform of Stanford University
Hanan Jacoby
4%
for very useful
comments.
Daron Acemoglu, Joshua Angrist, Abhijit Banerjee,
Robert Barro, Ricardo Caballero, David Card, Michael Kremer, Emmanuel Saez, and Jaume Ventura
helpful discussions,
and Guido Lorenzoni
for his insights
about transitional dynamics.
for
very
Introduction
1
Evaluations of social programs in developing economies tend to focus on the short run and
and do not try to assess
"partial equilibrium" effects of these programs,
consequences.
their
macroeconomic
Empirical studies of the determinants of economic growth form a largely inde-
pendent subfield that uses predominantly cross-country data
While aggregate cross-country data
is
readily available
ing conclusions, either because aggregate data
is
of
sets.
This division
and simple to
use,
it
is
unfortunate.
can lead to mislead-
poor quality (Krueger and Lindahl (1999)
Atkinson and Brandolini (1999)) or because regressions are mis-specified (Banerjee and Duflo
(2000)).
Conversely, policy recommendations based on "partial equilibrium" analysis can be
misleading,
if
the "general equilibrium" effects undo the direct effects of the policy (Heckman,
Lochner and Taber (1998)).
Moreover, the aggregate response to large programs
is
of independent interest,
and can be
a fruitful source of identification of macroeconomic relationships. In particular, large programs
are well identified shocks.
Studying the economy's aggregate response to these shocks
occasion to understand the process of adjustment.
is
the objective of macroeconomic studies of the
The adjustment
of an
"medium run" (Solow
macroeconomists and labor economists have long been interested
economy
(2000)).
in labor
is
to shocks
In particular,
supply shocks, such
as changes in cohort sizes (Welch (1979)), the level of education of the labor force (Katz
Murphy
(1992)), changes in level of education
for
effects of
and
by cohorts (Card and Lemieux (2001)), or adverse
labor supply shocks in Europe (Blanchard (1997)).
important dimensions of the
an
The speed and
efficiency of
a range of economic policies.
adjustment are
Trade pohcy analysis,
example, often assumes immediate adjustment of the production decisions, which could be
extremely misleading.
The
long term effects of economic crises, as well as the appropriate policy
response, are closely linked to whether they lead to efficient or inefficient restructuring, which
is
linked to the ability of the
(2000)).
economy
to allocate factors efficiently (Caballero
and Hammour
Thus, studying the aggregate consequences of large programs can inform economic
policy beyond the specific
program considered.
Most
studies of the
medium run consequences
of labor supply shocks focus
Europe. Yet, the response of the economy to a shock
is
closely related to its
on the U.S. or on
market
institutions.^
In developing countries, because market institutions (credit market, contractual enforcement,
labor market regulations) are less effective, one might expect the adjustment process to be par-
and
ticularly sluggish. Caballero
Hammour
lead to mis-allocation of resources
development. The evidence on
and
(2000) argue that institutional failures, because they
inefficiently slow restructuring,
medium term adjustment
in
is
at the root of under-
developing countries
however,
is,
extremely limited.
This paper studies the effects of a dramatic policy change that differentially affected different
cohorts and different regions of Indonesia on the allocation of the labor force across sectors and
on wages. In 1973, the Indonesian government launched a major school construction program,
the Sekolah Dasar
schools were built.
INPRES
program. Between 1973-74 and 1978-79, more than 61,000 primary
In earlier work (Duflo (2000)),
I
showed that the program had an impact
on the education and wages of the cohorts exposed to
wage
rates
and formal labor
it.
force participation of those
This paper studies the behavior of
who were
not directly exposed to the
program, from 1986, 12 years after the school construction program was initiated (and when
the
first
generation exposed to the program just entered the labor force) to 1999.
therefore a study of the
"medium" run aggregate
effects of the
This
program. Until 1997,
this
is
was
a period of rapid growth for the Indonesian economy: between 1986 and 1999, the economy
grew by over 50%, and the share of the labor force
in
6%
manufacturing doubled (from
to 13%).
Industriahzation occm-red throughout Java, and in concentrated pockets in the other Islands
Miguel, Gertler and Levine (2001).
I first
show that the program
in the regions
similar to that
where
it
led to faster increases in the fraction of primary school graduates
was more important, between 1986 and 1999. This increase
which would have been predicted
in the absence of
any migration.
I
is
strikingly
then proceed
See Blanchard and Wolfers (1999) for a comparison of the reaction to the labor supply shocks in the 1970s
across
European countries with
different labor
market
institutions.
to look at the effect of the
program on the wages and the formal labor
the cohorts that did not directly benefit from
it,
force participation of
because they were already out of school when
the program started. This allows us to look at the impact of the increase in education on factor
returns, for a population
less rapidly
whose
from year to year
skill level is
not affected.
in regions that received
It
more
turns out that wages increased
schools.
controlling for the factors that determined the initial allocation, and
This holds even after
may
have caused different
growth trajectories across these regions.
Using interactions between the survey year and the number of
INPRES
schools per 1,000
children as instruments for the fraction of educated workers in the region therefore suggests a
negative effect of the proportion of primary school graduates on individual wages, keeping the
individuals'
own
skill level
workers seems to cause an increase
in the
On
constant.
formal labor market.
The
in
the other hand, an increase in the fraction of educated
the participation of both educated and uneducated workers
negative impact of average education on individual wages does
not seem to be explained by selection bias caused either by selective migration or by selective
entry into the formal labor market.
I
propose a simple two sector model as a framework to interpret these
magnitude. Individuals can work either in the informal or
sector, they are self
employed and labor
(skilled
factor. In the formal sector, labor (skilled
earn a wage.
The production
and human
capital combined.
to a
movement
an unskilled)
is
is
combined with land, a
combined with
The
fact that the increase in the share of
and individuals
educated workers led
of workers from the informal to the formal sector indicates that the elasticity
and land
in the informal sector
substitution between labor and capital in the formal sector.
The
is
smaller than the elasticity of
elasticity of the
model.
I
compare two polar versions
of the model.
supply of capital
effect of
the program on
The benchmark
version assumes
with respect to the share of educated labor determines the predicted
in the
capital,
fixed
function in the formal sector exhibits constant returns to physical
of substitution between labor
wages
and their
formal sector. In the informal
in the
and unskilled)
effects
costless adjustment of the capital stock. In this case, in the period
we study (1986
to 1999, 12 to
25 years after the program was initiated), physical and
rate
and there should be no
relative fall in
wages
human
in regions
capital should grow at the
where human capital grows
same
faster.
This holds in a closed economy model as well as in an open economy model where capital
accumulated nationally and
efficiently allocated across regions.
compares the empirical estimates
I
The second
is
version, in contrast,
obtain to the parameters predicted by the model in the
absence of any adjustment of capital to the increase in education. These empirical estimates
are close to
what
this version of the
model would
predict.
This suggests that physical capital
human
did not adjust to the regional differences in the rate of accumulation of
capital induced
by the program.
The remainder
program and
of this
is
organized as follows. In section
on average education. In section
its effects
effects of average
paper
3, I
2,
I
describe the
INPRES
discuss the identiiication of the
education on individual wages and derive the empirical specifications. Section
4 presents the results. Section 5 presents the model that organizes and explains the findings,
and compares the estimates to what the two polar versions
(costless
adjustment of capital or no
adjustment of capital) of the model would predict.
2
The program and
2.1
its effects
on average education
The Sekolah Dasar INPRES program
In 1974, the Indonesian government initiated a large primary school construction program, the
Sekolah Dasar
INPRES
structed, doubling the
where
initial
sity of the
program.
number
Between 1974 and 1978, 61,807 new buildings were con-
of available schools per capita.
More
schools were put in regions
enrollment rates were low, which caused important regional variations in the inten-
program. Using a large household survey conducted in 1995 (the
to data on the
number
SUPAS
1995) hnked
of schools constructed in each individual's region of birth, Duflo (2001)
showed that the growth
in
education between cohorts unexposed to the program and cohorts
exposed to the program was faster in regions that received more
INPRES
schools. This differ-
ence can be attributed to the program with a reasonable level of confidence, because no similar
pattern
present
is
if
we compare cohorts that were not exposed
to the program.
In addition,
the program affected mostly primary school completion, whereas omitted factors would have
affected other levels of schooling as well.
Each point on the
from Duflo (2001).
This pattern
solid line
is
summarized
summarizes the
in figure 1,
effect
reproduced
one more school
per 1,000 children had on the average education of children born in each cohort.^
in
built
Children
Indonesia normally go to primary school until age 12 (although delay at school entry and
uncommon),
repetition are not
children
who
reached 12 before 1974,
increasing. This
If
therefore one would expect the effect of the
is
when
first
for
schools were built, and to be progressively
what the picture shows.
exactly
when comparing the evolution
should see the average education (and
Data and empirical
for this
more
program enter the labor market, one
specification
to 1999.
These surveys are repeated cross
sections, of approximately 60,000
The surveys contain information on province and
in the last
week, and wages for individuals
pation. Using this data,
worked on
wage
I
district of residence (but not of
These are the
An
individual
primary occupation.
coefficients
is
for a
wage
in their
of hoiurs
primary occu-
considered as part of the formal sector
The survey
if
he works
questions and definitions are homogenous
obtained by regressing years of education on the the interactions between the number
of schools built per capita in the individual's region of birth
and region of birth
who work
employment, number
construct the average hourly wage as weekly wage divided by hours
this occupation.
in his
primary school graduates)
schools.
birth), education level achieved, labor force participation, t3rpe of
worked
adults over the years
paper comes primarily from the annual Indonesian Labor Force Survey (SAK-
ERNAS), from 1986
households.
to the
among
in particular the fraction of
increase faster in the regions that received
The data
of average education
As the generations exposed
in the different regions.
for a
to be
migration flows were either small or not affected by the program, one would expect to see a
similar pattern
2.2
the
program
fixed effects.
and year of birth dummies,
after controlHng for year
between 1986 and 1999.
I
restrict the
sample to males aged 20 to
Descriptive statistics are presented in table
1.
The
60,
and
exclude Jakarta.
I
fraction of individuals born after 1962,
and
therefore theoretically exposed to the program, in the age groups 20-40 and 20-60, increases
progressively over the years.
gion in each year. There
is
I
consider the proportion of primary school graduates in each re-
a total of 3,826 district-year
observations in each cell in the
data
set,
and each
cell is
sample.^
full
cells,
with an average of 287 individual
All regressions are performed
on
this aggregate
weighted by the number of observations used to construct
it.
Consider comparing two regions in 1986 and 1999, one of which received a large number
of
INPRES
schools per capita, while the other one received a small
fraction of people
in
1999 than
who were young enough
in 1986.
We know
relative to the older ones,
number
of schools.
INPRES program
to be exposed to the
is
The
bigger
that the gains in years of education of these younger cohorts,
were bigger
in the regions
that received
more
schools. If the effect of the
program was not undone by migration, one would expect the average education
(in particular,
the proportion of primary school graduates) to have grown faster between 1986 and 1999 in
the region that received more schools.
attainments between 1999 and 1986
This suggests comparing the difference in educational
in these
two regions. More generally,
this suggests that,
one runs a regression of the difference between average educational attainment
on the number
summary,
In
this suggests the following specification:
S^-^t +
i/j+
1999
Yl i^i*Pjhn+ Y.
fit
is
is
the proportion of primary school graduates
a survey year fixed effect and Uj
is
a region fixed
schools built between 1974 and 1978 in district
t
=
I,
and
otherwise).
Cj
is
i^i * (^j)^n
+
ejt
(1)
(=1987
(=1987
Sjt
coefficient.
any year-to-year difference.
1999
where
1999 and 1986
one should see a positive
of schools per capita built in each region,
Clearly, this reasoning also applies to
in
if
j,
and
among
effect,
A; is
adults in year
Pj
is
the
t
number
a survey year
in region j,
of
dummy
INPRES
(A;
=
1 if
a vector of initial conditions that are introduced as control
There are on average 185 observations per
cell
of individuals
born before 1962, including 61 with wage data.
variables. In particular,
may
it
be important to control
for the
enrollment rate in 1971, since
was a determinant of the placement of the program. Note that the
enrollment rate in 1971
all
is
first-order eflFect of a higher
a difference in level of education, which should affect
survey years identically, and therefore be absorbed by the region fixed
which children attend school
in the rates at
in a region will lead to
it
cohorts, and
all
effect.
Only a change
a change in the rate at which
average education increases from year to year. Therefore, controlling for the enrollment rate in
1971, interacted with year
dummies,
enrollment rates are correlated with
is
important only to the extent that we think changes
levels.
I
also control for the
number
in
of children in 1971.
Results
2.3
Since the generations exposed to the program have already started entering the sample in 1986,
one would expect
year
dummies
the coefficients of the interactions between program intensity and survey
all
to be positive
Columns
and increasing.
1,
2,
4
and
5 in table 2
show these
coefficients for the specification that includes enrollment rates as a control, for adults
aged
20 to 40 and for adults aged 20 to 60, in the whole sample and in a sample that excludes
urban
The
districts.''
increases faster than
larger
and more
among
significant in the former group.
The
in the
,
and
The group
of individuals aged 20 to 60, however,
will therefore
coefficients are increasing for
as expected, they are larger in the
from 1991
individuals aged 20 to 40
those aged 20 to 60, and one would expect the coefficients to be
corresponds better to "the labor market"
of this analysis.
among
fraction of "young" (or exposed) people
be important
in the
both groups, they are jointly
group aged 20 to
40.
They become
second stage
significant, and,
individually significant
20-40 group and from 1996 in the 20-60 group.
This pattern could have been caused by other factors than the increase in education due to
INPRES,
PRES
for
example by migration of educated workers into
schools. If
we had data from
earher years,
it
districts that received
would be possible to use "pre-program" data
to test the identification assumption that the increase in education levels over time
The
specification without enrollment rates as a control
more IN-
is
very similar, and
is
therefore omitted.
would not
have been systematically different in regions where a different number of schools was
in
the absence of the program.
No comparable
was due to something other than the
survey was realized before 1986.
effect of the
mated a
to the
program (individuals born
specification similar to equation
1,
in
in table
2.
subsample
1962 or before). To check
of those
this, I esti-
with the fraction of primary school graduates among
individuals born in 1962 or before as the dependent variable.
columns 3 and 6
the pattern
program on education, however, one would
see a faster (or slower) increase in the education over the years, even in the
who were not exposed
If
even
built,
Figure 2 gives a graphical
The
summary
coefficients are presented in
of these estimates:
It
shows
the coefficients (and their confidence intervals) for the entire group aged 20 to 40, and for the
group of individuals born before 1962. There
The
before 1962.
coefficients in the
is
no systematic increase among the group born
two equations are
This indicates that the increase in average education
is
significantly different
likely
from each other.
due to the program, rather than
to other factors.
Does migration undo the
2.4
effect of local infrastructure
Although migration flows were not very important
negligible.
live in their
In 1995,
if
one excludes Jakarta,
province of birth, and
the urban districts).
and 25% did not
Among
24%
12%
development?
in Indonesia over the period,
of individuals in the
SUPAS
did not live in their district of birth
individual born in 1962,
live in their district of birth.
It is
13%
they are
far
from
sample did not
(17%
if
one excludes
did not five in their province of birth,
therefore interesting to study whether out
migration of educated workers dampened the effect of the program on average education in the
labor market.
The
program
results in the previous section already suggest a partial
affected the average education of adults,
undone by migration.
We
can, however,
make
answer to this question. The
which indicates that
this result
more
its effects
precise
were not totally
by comparing the
effect
the program should have had, in the absence of any off-setting effect of migration, with the effect
it
had
in reality.
The
results
from
this exercise are
important
in the context of
an increasing focus
on decentralization, notably
in Indonesia. If local
governments consider that their communities
are not getting any benefits from investment in education because educated people migrate
(with their
human
may
capital), decentralization of school finance
lead the public financing of
education to decline.^
To
get at this question,
I
INPRES
estimated the effect of the number of
first
schools con-
structed per capita in an individual's district of birth on the probability that an individual
completed primary school,
SUPAS
each cohort.
for
(Intercensal survey of Indonesia)
tative at the district level.
Indonesia.
The survey
It is
is
To
this end,
I
used the
SUPAS
1995 data.
a sample of over 200,000 household.
It is
The
represen-
conducted every 10 years by the Central Bureau of Statistics of
collects the
same information
as the
SAKERNAS
1995), as well as
more information about household members,
district of birth.
The sample
for this analysis is
(which
in particular their
men born between
it
replaces in
province and
1950 and 1972 (there are
152,989 individuals in the sample).
Using
this data,
I
regressed a
dummy
indicating whether an individual completed primary
school on a set of district of birth fixed effects, cohort of birth fixed effects, and interactions
between the number of schools constructed
The equation estimated
is
primary school completion
on the cohort born
school graduates
district before 1962.
denote by
(pkt
and year of birth dummies.
identical to equation (11) in Duflo (2001), except that
as the
in year k as
among
in one's district of birth
dependent variable. Denote the estimated
ifk-
I
effect of the
this
used the
program
used the 1995 data to compute the proportion of primary
those aged 20 to 40 in each year (from 1986 to 1999)
Denote
it
average by Sqj. For each survey year
the share of those aged 20 to 40
who were born
in
year k.^
t
bom
and year
We
in each
of birth k,
can then compute
the proportion of primary school graduates predicted by the program in each district and each
year
as:
Bound, Kzedi and Turner (2000) ask the same question
All the individuals
who
for college
are born in 1962 or before are in the
education in the U.S.
same cohort
k.
(t-20
E
\
""^kM
Note that
Sjt-
The
first
does not contain any information specific both to the district
this predicted value
and the year considered. There
observation
is
is
(2)
/
fc=1962
therefore no source of mechanical relationship between Sjt and
that Sjt and Sjt are strongly correlated.
The
regression of the actual
share of primary school graduates on the predicted share leads to a coefficient of 0.84, with a
t-statistic of 77). This,
is
is
not very informative, because a large part of this correlation
driven by those born before 1962, and would therefore
had migrated out
I
however,
of the high
program
run the same specification as in equation
education,
Sjt-
These
coefficients indicate
affected in each region in the absence of
7i;
The
districts.
1,
but
how
I
1
3
is
even
if all
the educated young
is
more informative.
use as the dependent variable the predicted
the average education of adults would have been
any
offsetting effect of migration.
The
coefficients
along with the coefficients obtained
3,
with the actual average as the dependent variable. The two sets
of coefficients are surprisingly close to each other.
predicted effect
exist
following experiment
obtained in this specification are plotted in figure
when estimating equation
still
In particular, there
is
no evidence that the
bigger than the actual effect.
Identifying the effect of a change in average education
Conceptual framework
3.1
Consider an economy with two sectors.
Assume
that there are only two types of workers,
educated (with a primary education or more) and uneducated (no primary education)
.
Assume
that the formal sector employs educated labor, uneducated labor, and capital, and the informal
sector employs educated and uneducated labor and land.^ Individuals are self-employed in the
informal sector, and receive a wage in the formal sector.
We
could allow land and capital to present in both sectors, but
and land between
sector, for
which we have
little
data.
10
Consistent with Harris and Todaro
we would have
to
model migration of
capita]
(1970) and other dual
economy models
of development,
we view the
potential of growth of the
informal sector limited while the economic growth happens as the formal sector expands. This
is
reflected in the
in the
assumption that land
is
The share
of the labor force
employed
formal sector and their wages are the two variables of interest.
The production
functions in the formal and informal sectors are given by /(Ap, Ep, Up,
and g{Ai,Ej,Uj,T) respectively, where
Ep
a fixed factor.
and
Up
K
and
T
are the stock of capital
denote educated and uneducated labor employed
E[ and Uj denote educated and uneducated labor employed
We
Aj are "productivity" parameters.
will treat the total
affected by allowing steady population growth).
and uneducated workers who are working
as a function of the
number
of educated
therefore write the
the formal sector, respectively,
and
Ap and
population as a constant (nothing
is
as well as the fraction of educated
in each sector, are
determined jointly
and uneducated workers
1,
respectively,
in the informal sector,
The wages,
the stock of capital, the stock of land, and the parameters
Normalizing the entire labor force to
in
and land
K)
Ap
in
equilibrium
the economy as a whole),
(in
and Ap-
and denoting S the share of educated workers, we can
wage and formal employment functions
as:®
\n{wE)
=
4>e{Ap,Aj,K{S),S,T),
\n{w^)^<l>u{Ap,Ai,K[S),S,T),Ep^^EiAF,Aj,K{S),S,T),andUp = iv{Ap,Aj,K{S),S,T).
K
is
explicitly written as a function of 5, to reflect the fact that a
change
in the
proportion
of educated workers has. a direct effect (the effect of the share of educated workers on the wage),
and an
The
indirect effect
due to the accumulation of physical capital
elasticity of physical capital
in response to this increase.
with respect to the share of educated labor
is
an empirical
question: In the long run, one might expect the physical capital to adjust to a change in the
fraction of educated workers, while in the very short run, adjustment will be
In the
"medium run" the speed
,
of the production function
of adjustment of the capital stock will
and the
~ MApuAiuKtiSt =
Writing the wage function directly
in
depend on the
availability of finance for the installation of
Consider a Taylor expansion of the wage function around
MAFt,Art,KtiSt),St,Tt)
much more hmited.
0),St
logarithm, for convenience.
11
5=
=
new
flexibility
capital.
0.
0,Tt)
+
St^ + St^^.
Denoting Kt{St
=
0)
by Kt,
InK,) =
We,
can be rewritten:
this
(^ + ^^)St + 4>is{AFuAnX,Tt)
therefore, seek to estimate the coefficient a^ in the expression:
:^
\n{wst)
asSt
+ (t>\s{AFt, Ajt, Kt.Tt)
In this expression, as reflects the direct effect of
5 on
(3)
the wage, as well as any indirect effect
due to the response of the stock of physical capital to the stock of human
Us
is
not determined a priori.
capital
and labor combined, as
in the share of
If capital
If capital
will
adjusts slowly, or
if
capital.
sign of
there are diminishing returns to
tend to be negative, reflecting the fact that the increase
educated workers increase the quantity of labor (measured
adjusts rapidly and there
The
is
no
fixed factor, a^
in efficiency units).
would be zero and even positive
the presence of an external effect of education (as in Lucas (1988)) or
if
an increase
in
in the
share of educated workers led to a more than off'setting increase in the stock of physical capital
(Acemoglu (1996)).
Consider running a regression of the wage of the uneducated workers on the share of educated
workers in the economy, without controlling for the stock of capital.
Clearly,
correlation between the level of capital (and the productivity of the formal
and the share
Since there are
districts.
many
Using equation
districts
3,
likely to
any
sector)
be the
cannot really be interpreted.
the growth of the log wage between two periods
+
estimate this relationship using an
is
given by:
4>is{AFt,An,Kt.Tt)-(t>is{AFt~i,Ait-i,Kt-i.Tt-i)
(4)
OLS regression, and we omit the term 4>is{AFt,Ajt,Kt-, Tt)-
4>is{AFt-\,Ait-i,Kt-i,Tt-\), the coefficient a^ will be biased
physical capital accumulation and
is
is
and years, we could instead compare wage growth across
\n{wst)-\n{wst-\)c:^as{St-St-i)
If we
there
and informal
of educated people, this relationship will be biased. Since this
case, coefficients obtained in such a regression
if
human
if
there
is
a correlation between
capital accumulation. In almost
any model of human
and physical capital accumulation based upon individual maximization, the increase
12
in the share
and the rate of physical capital accumulation
of educated workers
both are determined by the discount rate
as consistently,
we
in the
suppose
first
in particular,
therefore need an instrument, correlated with the increase in the share of
—
potential instrument for {St
structed by the
be related:
economy. In order to estimate the parameter
educated workers, but not with the evolution in the other factors
A
will
INPRES
St^i) in our setting
is
the
in the
number
economy.^
of primary schools con-
program. To understand how the instrument works and
its
limitations,
that the government allocated the schools randomly. Each school reduces the
fective cost of schooling,
and therefore increases the enrollment rate among
generations (but not that of the older generations).
The
increase in the
all
ef-
the future young
number
of schools
com-
bined with the fact that the young generations progressively enter the labor market starting
m
the late 1980's changes the rate of growth of
growth
is
number
In practice, however, the
rates at the
primary school
of
INPRES
the program before
instrument for {St
level
—
it
was
St-i)
if it is
the schools had been
schools would form an ideal instrument.
in regions
were lower. The evidence presented
initiated.
if
in the rate of
human
capital
in this
where enrollment
paper and
in
Duflo
was not systematically correlated with
Nevertheless, the level of the program will not be a valid
correlated with the rate of capital accumulation. This would
educational attainments in 1971 were correlated with capital accumulation between
1986 and 1999.
Regions with a lower
level of educational
could therefore have been growing faster,
if
product per capita until 1996
will control for
(Hill (1996)).
enrollment rate
in 1971.
attainment tend to be poorer, and
there had been a tendency for Indonesian regions to
converge. In practice, Indonesian provinces exhibited very
we
The modification
government allocated more schools
(2001) suggests that the rate of growth of
if
over time.
a function of the number of schools built, which suggests that
allocated randomly, the
happen
S
little
convergence in gross provincial
Nevertheless, to control for possible convergence,
We
will also present the results in the rural
For example, Moretti (1999), proposed to instrument
for {St
—
St-i) with the share of young people in the
base year, on the ground that education will grow faster in regions which have more young people.
remains, however, that the share of young people in the base year
accumulation as
well.
13
sample
is
The problem
very likely to influence physical capital
and omit the years 1998 and 1999 to allow
separately,
hit
wages
in richer regions,
and
much more than
in particular cities,
Thomas and Beegle
rural areas (Prankenberg,
for the fact that
(1999)), causing
in
the Indonesian
crisis
poorer regions and
some convergence
of
wage
in
rates
between regions.
Empirical specifications
3.2
The wage
of individual
observed in district j in year
i
\n{wijt)
where
Si
is
a
dummy
=
S^{ln{wEjt)
-
t is
given by:
+ M'UJUjt) +
^n{wujt))
v^jt,
(5)
indicating whether the individual has graduated from primary school.
error term Vijt reflects
all
The
the other factors that determine wage, besides individual and average
education.
Substituting the expression for \n.{wEjt) from equation 3 and including
which we do not measure
in the error
term,
individual education level, and regional
\^i\{w^jt)
wheiebjt
^
=
S^hjt
human
+
{[n.{wEjt)-'i.n{wujt)) (the skill
As we have
we obtain a
relationship
+
ejt
+
Mt
premium) and
,
by
OLS
fit
+
OLS
of bjt
estimate of
is
ay
will
that, even
if
there
be a biased estimate of the
is
+
t.
^ijt,
(6)
+ iyj+ejt =
4)iu{AFjt, Aijt, Kjt,Tj))
(treating
could be very misleading, because of the correlation between
Acemoglu and Angrist (1999) show
t^j
the variables
between individual wage,
capital in the district at date
OLuSjt
seen, estimating this equation
all
bjt
ejt,iit
or Uj
no omitted
effect of Sjt
as a
random
and
Sjt.
coefficient)
In addition,
district level variable, the
on \n{wijt)
if
the estimate
biased for any reason (such as measurement error in the education variable, or the
endogeneity of education)
.
They propose
to instrument for both Si and Sjt
could instrument Sjt with a variable that does not
the approach
we take
here.
14
aflfect Si,
.
Alternatively, one
the individual education, which
is
The nature
individuals
of the
INPRES program suggests
who were born
in
1962 or before were not affected by the program (we have verified
in the previous section that the
average education in this group did not grow faster from year
more INPRES
to year in the districts that received
affected the average education
the intensity of the
does not
To
the following instrumental variable strategy. All
by
On
schools).
the other hand, the program
affecting the education of those born after 1962. Therefore,
INPRES program
affect individual education,
a potential instrument for the average education which
is
we
if
sample to those born before 1962.
restrict the
derive the empirical specification, take the average of equation 6 for
all
individuals born
before 1962:
.^
\n{w^jt)
where
Sjto
year
in district j.
t
is
=
Sjtobjt
+
Sjtau
+
l^k
the proportion of primary school graduates
If
we take the
=
(Sjto-Sjt-io)bjt-i
first
+
l^j
+Vijt,
among
(7)
the old (born before 1962) in
and rearrange the terms, we
difference of this equation
obtain:
\Tl{w^Jt) -\r\{w,j,_i)
The
number
evolution of the
skill
premium
+ Sjt-io{bjt-bjt-i) + {Sjt-Sjt-i)o:u + l-k +
{bjt
—
bjt-i)
is itself
'^j
+ ^jt +
v-^ijt
a function of the evolution in the
of educated workers in the region, so that the effect of the evolution of
graduates on the average wages of the individual born before 1962
is
primary school
finally given
by the expres-
sion:
ln{w,jt)
-
\n{w^Jt-l)
=
[Sjt
-
Sjt-i)a
+
/xj
+
t/j
+
e^j
+ v,jt,
(8)
Subject to the caveats discussed in the previous sub-section, we can use the number of
INPRES
schools {Pj) as an instrument for Sjt
for variables
that Pj
is
—
Sjt-\ in equation
such as the enrollment rate and the wage
uncorrelated with {Sjto
—
Sjt-io),
which
15
is
in
8,
possibly after controlling
1986 (a vector Cj).
now included
in
We
have verified
the error term.
A
joint test of the validity of the strategy
Acemoglu and Angrist (1999)
regression of individual wages
is
and the seriousness of the problem suggested by
to use as dependent variable the average of the residual of a
on individual education.
If this
equation
should lead to the similar, but more precise estimate (since {Sjto
—
is
correctly specified,
it
Sjt-io) will not be part of
the error term any more).
Thus the reduced form with two years
\n{wjt)
-
of data would be written:
=
ln{wjt-i)
^it
+ l2Pj + hCj + ^jt
This equation can be generalized to incorporate
equation similar to eqviation
all
It/jf
=
/it
+
I/j-
+
1999
^
(A;*Pj)72(+
(=1987
where
and
fit
Equations
7^ are year
1
and
Yl, i^l
(=1987
+
* (^j)^2l
(9)
^3t,
district fixed effects, respectively.
and 9 form respectively the
variables strategy to estimate equation
The same
form
1:
1999
In
available years, leading to a reduced
stage and the reduced form of an instrumental
first
7.
reasoning applies to formal labor force participation, and the same specification can
be estimated with formal labor force participation instead of wages. Finally we can also estimate
equations similar to equation
variables
likely to
we consider here
7,
using the average
(wages, education,
skill
skill
premium
as
dependent variable. The
premium, formal labor force participation) are
be auto-correlated over time. Bertrand and Mullainathan (2001) show that
in large understatement of the standard errors in these types of setting.
errors in
all
I
it
can result
thus correct standard
equations using a generalization of the White variance formula which allows for a
flexible auto-correlation process
within any states.
Since the sample of individuals not affected by the program that
different every year, this specification
may
we
include in the analysis
from sample selection.
suffer
First, the
is
program may
have induced selective migration by the old people, potentially correlated with their productivity,
and therefore with
their wages. Second,
we
will
show that the program
16
affected the proportion
of old people
who work
for
a wage:
it
also
opens some room
that workers with the lowest productivity switched to the
for selection bias, since
wage
it is
possible
sector. Section 4.5 will present
additional evidence (using two other data sets) on whether these two possibilities for sample
selection affected the results.
4
Results
Summary
statistics for the
year, are presented in table
sample of people born before 1962, and aged 60 or
3.
sample belong to
fact that
primary school graduates among them increased
(.this
reflects the fact that individuals present in the
later cohorts in later years).
We
someone receives a wage. This fraction
terms, increased by about
50% between
the survey
of
The proportion
from 59% to 74% between 1986 and 1989
less in
use as an indicator of formal employment the
is
a
little
over 30%.
The average wage,
1986 and 1997, and declined by
22% between
in real
1997 and
1999.
4.1
Reduced form
The reduced form
results
results (the estimates of the coefficients 72; in equation 9) are presented in table
4 and in figures 4 A and 4B. These two figures summarize the reduced form effects on wages and
on formal employment. Although none of these
from
zero, the
coefficients
individually significantly different
reduced form coefficients in the wage equation are declining
coeflacients of average education,
probability of working for a
wage
which are increasing). The reduced form
are increasing.
In the
rural areas, the coefficients increase from 1997 to 1999,
impact of the
is
crisis.
contrast to the
coefficients
on the
sample that includes both urban and
which probably
reflects the differential
In the rural sample, they are monotonically declining.
17
(in
The
4.2
effects of
The main sample
I
will consider
average education on wage rates
for the anatysis
the individuals aged 20 to 60
is all
two independent variables.
First, the fraction of
who were born
primary school graduates
sample aged 20 to 60 (a reasonable approximation of the average education
second, the fraction of primary school graduates
among the 20-40
sample.
The
identification.^^
among
males.
The INPRES program
on the share
of primary school graduates
Using instead the share of primary school graduates among males and females
Table 5 presents
OLS
identical results.
estimates for equation
wage and the average of the
residual
wage
7,
where the dependent variable
(after controlling for individual
is
the average
education and age).
panel does not include district fixed effects, while the second one does. As expected,
first
from specifications that do not control
results obtained
They lump
and the second panel
much
for individual
together the "social" and the "private" returns.
coefficients of average
are
The former was
Focusing on the 20-40 variable puts more accent on the source of
results presented here focus
combined leads to almost
The
in the
in the labor market);
directly affected the latter (since the older affected people were 37 in 1999).
affected as a consequence:
before 1962.^°
We
education are bigger.
should, therefore, focus on the
education in the wage residual equation. The difference between the
illustrates the
remark we made
in the previous section: the
OLS
first
estimates
bigger than the corresponding fixed effects estimates, which suggests that they are
very strongly upward biased: educational attainments are higher in regions where wages are
higher, but this
is
as likely to
come from a
from the opposite relationship.^^
It
means that when we change
specifications in a
relationship running from income to education as
The OLS
year, there
is
results are all positive
both a cohort
effect sind
and
an age
sample which maintains a constant cohort composition, and the
In addition, the
first
stage
is
significant, while the
effect.
results
I
have run
were very
all
the
similar.
stronger for this variable, which minimizes the problems that can arise from
using weakly correlated instruments.
One would expect
the results with the 20-60 average to be a scaled up
version of the results obtained with the 20-40 average.
There are many reasons, besides those emphasized
upwards.
effect in
First, there
may be
a wealth effect
in
here,
education:
which would lead
Glewwe and Jacoby
OLS
coefficient to
(2000) find an important wealth
Vietnam. Second, with economic growth, expected returns to education improve, and
18
be biased
this
may
lead to a
OLS
results with fixed effect are positive, but significant only in the specification that has the
proportion of primary school graduates
among
These point estimates suggest small positive
among
share of primary school graduates
0.8%
in the
an increase of 10 percentage points
effects:
those aged 20 to 60
is
in
the
associated with an increase of
wages, after controlling for individual education. These coefficients are
one tenth of those estimated by Moretti (1999)
in the
those aged 20 to 60 as the dependent variable.
less
than
impact of the share of college graduates
for the
US.
Table 6 presents the instrumental variables
The
results.
results are presented for the entire
sample, and for a sample that excludes the years 1998 and 1999, since the
regions differently.
the estimates
The
first line
crisis hit different
of each panel presents the results on wages. In the full sample,
become more negative when the
affect the estimates in the rural areas.
crisis
The second
years are taken, while the estimates in rural
line presents the results
wages as the dependent variable. None of the estimates
is
significant.
The
using the residual
estimates obtained
using the residual wage or the actual wages as the dependent variable are very similar, which
is
reassuring: Since the
INPRES
instruments affects only average education, and not individual
education, controlling for individual education should not affect the estimate, which
find here.
The estimates using
the residual wage are
an increase of 10 percentage points
in
sample of rural areas. Without using the
The
precise.
the share of primary school graduates
year old led to a decrease of 3.8% in wages in the
-4.4% and -9%.
somewhat more
last
full
is
They suggest
among
that
the 20 to 60
sample, and to a decrease of 9.9
two years
what we
%
in the
of data, the coefficients are respectively
negative coefficients are significant (at the
10%
level) in
the rural sample
only.
If
the
we
first
focus on the share of primary school graduates
stage has
more power), the story
is
among
the same: an increase of 10 percentage points in
the share of primary school graduates leads to a decrease of 2.9
higher
demand
revolution,
and
for
education (see Foster and Rosenzweig (1996)
Bils
and Klenow (1998)
for
the 20 to 40 year old (for which
for
%
in
the wage of the old in the
micro-economic evidence of the Indian green
a re-interpretation of the cross-country evidence along these
19
lines.
sample, and to a decrease of 6.3% in the rural sample.
full
4.3
Skill
The
premium
third line in panel
A
B
and
of table 5 presents the results of estimating
without district dummies) an equation similar to equation
7,
by
is
negative, large (around -0.45), and significant.
the estimates are negative, but
to be biased
The
of the
upwards
much
(with and
but where the dependent variable
Without
the difference between the average wages of educated and uneducated workers.
dummies, the estimate
OLS
With
smaller (around -0.09) and insignificant.
district
OLS
is
district
dummies,
seems again
(in absolute value).
third line in panels
A
and B of table 6 presents the instrumental variables estimates
same equation. The IV estimates
of the effect of the share of primary school graduates
on the primary education premium are either negative or positive, and always insignificant.
The education premium does not seem
primary school graduates.
It
to have
been affected by the increase
in the
number
of
suggests that, in at least one sector of the economy, educated and
uneducated workers are close substitutes.
4.4
The
Formal labor force participation
fourth line in panels
A
and
B
of table 6 presents the instrumental variables results for formal
labor force participation (corresponding
variable
is
is
the fraction of people
OLS
who work
results are presented in table 5).
for a
The dependent
wage. The 2SLS estimates suggest that there
a strong positive effect of the fraction of primary school graduates on the probabihty that
someone works
for a
wage.
In the rural
proportion of primary school graduates
in the probability of
graduates
among
working
for a
and urban sample combined, a 10% increase
among
wage.
the 20 to 40 year old leads to a 4.5% increase
A 10%
increase in the proportion of primary school
the 20 to 60 year old leads to a 6.6% to 7.5% increase.
significant in all of the specifications,
in the
and are very similar across
20
The
specifications.
coefficients are
Sample selection
4.5
There are two possible sources of sample
selection. First, there
Second, since the program affected the proportion of people
probability of selection in the sample
is
for
might be selective migration.
whom we
by the instruments.
affected
observe wages, the
In particular, one can
imagine a situation where the program pushed the "marginal" self-employed into the formal
labor force, and these marginal employees receive a lower wage.
Moreover, new entrants into
the labor force have less experience, which should lower their wages.
Figure 2 (and columns 3 and 6 in table 4) suggest that the average education of individuals
born before 1962
in the
sample was not affected by the program: along
the sample remains comparable over time.
the individuals born before 1962 and
intensity
who earn
is
1.03),
when
I
regress in the education level of
wage on the interactions between the program
a
and the survey year dummies, there
statistic of the interactions
least,
Likewise,
this observable dimension,
is
no distinct pattern
in this regression (the
F
which indicates that, along observable characteristics at
the composition of the formal labor force did not change as a result of the program.
However, there
may
have been selective migration along unobserved dimensions
ductivity old people are attracted by the
program regions
old people leave the region), which will cause a
wages.
The
SAKERNAS
downward
for
example, or
who
is
low pro-
high productivity
bias in the effect of the
data does not indicate whether an individual
not have any income measure for individuals
if
(if
program on
a migrant, and
are not working for a wage.
We
we do
thus need
other sources of information to shed light on this issue.
First, the
earlier)
SUPAS
data
set (the
1995 intercensal survey of Indonesia which we described
has the individual's region of birth as well as his region of residence.
whether there are differences
related with the
of the hourly
INPRES
wage
in productivity
program,
of the migrants
currently residing in the district).
I
form
for
To
investigate
between migrants and non-migrants that are
cor-
each district the difference between the logarithm
and that of the non-migrants (among those born before 1962
Column
1
of table 7 presents a regression of this variable on
21
number
the
schools
of
INPRES
schools built per capita in the region.
The
coefficient of the
actually positive (but insignificant), which suggest that there
is
column
selection bias. In
2, I
who
staid.
This difference
is
of
no downward sample
who migrated
construct the difference between the wage of those
out of their region of birth and those
the program. There
is
number
unrelated to the level of
thus no evidence that selective migration was likely to bias the results
is
downward.
Second, we use the
SUSENAS
data (a nationally representative survey of about 50,000
households, which has an income and a consumption supplements once every 5 years).
the incomes modules from the
household income of
The
ratio
column
(3))
employed to household
less in 1993.
on the
INPRES program
On
from 1987 and 1993 and computed the
of
wage earners the income
very stable between 1987 and 1993: self-employed earn 17.85%
and 18.5%
in 1987,
7,
is
self
SUSENAS
balance,
it
is
I
level of the
of
less
I
ratio of the
wage
earners.
than employed
then regressed the difference in the log of this ratio (table
program, and found no relationship (the
0.0021, with an absolute
t
coefficient of the
statistic of 0.130).
therefore appears that the relative
wage
loss of the old generation in regions
where the program increased the supply of primary school graduates cannot be attributed
composition
We
A
can summarize the results from this section as follows.
decline in the
point estimate
although
it is
•
No change
•
An
is
An
increase in the share of the
to:
wage of older workers, whose
large (as large as the skill
significant only in
in the skill
some
level of
premium
education did not change.
itself,
specifications.
premium among
point estimates are large (a
an increase of at
least
4%
10%
The
or even larger in rural areas),
older workers.
increase in the share of the labor force employed in the formal sector,
The
to a
effect.
educated workers leads
•
used
among
the old.
increase in the share of educated workers leads to
in the share of old workers
22
employed
in the
formal sector) and
significant.
•
No change
the diflFerence between formal and informal sector earnings.
in
In the next section,
we
build a
model which can explain these
and serve
effects,
as a
framework
to interpret their magnitude.
5
Model and interpretation
What do
these results
tell
us about the the response of the
education of the labor force? In this section,
to interpret these effects
study the
effect of
and
their
use a simple two-sector model as a framework
magnitude. In the
first
subsection,
for the
I
use this model to compare
accumulation of capital. In the
its
human
miphcation
Below,
The
I
for
I
set
up the model and
prediction to the data.
first case,
there
In the second case, physical capital accumulation does not adjust at
the rate of
to an increase in the
education on wage and the allocation of labor across sectors, taking capital as
given. In the second subsection,
two polar cases
I
economy
capital accumulation. In this simple model,
all
is
I
specify
no adjustment
cost.
to the modification in
both assumptions have the same
formal labor force participation, but very different implications for wage rates.
provide some justification for the specific assumptions of the model.
fact that the skill
premium was not
affected suggests that, at least in one sector, educated
and uneducated workers
are very strong substitutes.
combined, while they are
self
employed
in the
Since in the formal sector, workers are
informal sector (we can think about this sector as
small scale agriculture), we will take as a starting point that educated and uneducated workers
are perfect substitutes in the informal sector. Suppose that the informal sector combines land
and human
sector
is
capital,
and the formal sector combines physical and human
characterized by a
long run (because land
is
downward
sloping
demand
capital.
The informal
for effective units of labor,
even in the
a fixed factor). In the formal sector, the slope of the labor
depends on how capital adjusts to changes
in the
composition of the labor
not adjust to an increase in effective labor supply, labor
23
demand
will
demand
force. If capital
does
be downward sloping.
It
would be
up
if
fiat if this
increase led to an oflfsetting increase in the supply of capital, or even slope
there were technological externalities, or
human
capital accumulation (as in
Acemoglu
if
faster capital
accumulation more than
(1996)).
Consider a simple competitive model, where wages are equalized
sector (for a given level of
induced by the
INPRES
skill).
offset
in the
formal and informal
Figure 5 illustrates the effect of the increase in educated workers
reform on uneducated old workers.
The
total
number
of old
uneducated
workers and their allocation between the formal and the informal sector are depicted on the
Y
axis.
The
wage
in the informal sector.
left
axis presents the
The
wage
in
the formal sector, and the right
number
increase in the
the labor supply expressed in efficiency unit, which
demand
sector.
expressed in
If
the labor
number
demand
in the formal sector
akin to a shift to the right of the labor
effect of
an increase
and a
downward
is
axis presents the
of educated workers leads to a shift in
of bodies in the informal sector,
unambiguously negative. The
allocation of labor
is
Y
X
in the
shift to the left in the
formal
slopping, the effect on wages
number
of educated workers
between formal and informal sector depends on the ratio of the
is
on the
elasticity of
substitution between capital and labor in the formal sector and that between land and labor
in the informal sector.
led to a shift
The observation
from the informal to the formal sector indicates that the
between labor and capital
We
will
that the increase in the proportion of educated worker
is
elasticity of substitution
bigger than the elasticity of substitution between labor and land.
capture this with the simplifying assumption that the informal sector uses a Leontieff
technology.
5.1
Model
This section builds a specific model along the
lines
described above.
We
start
by describing
the allocation of labor and the determination of wages, taking capital as given. In the second
subsection,
we
will
compare the predictions
of this
model to the data, using two polar cases
for
the adjustment of capital to the increase of the share of educated workers: no adjustment cost,
or no adjustment whatsoever.
24
There
is
a mass
1
of workers in the
economy, out
of
which a fraction
S
are educated. There
are two sectors, an informal (small scale agriculture) and a formal sector (industry).
Suppose that the informal sector
is
characterized by a Leontieff production function in terms
of efficiency units:
Yi
where Hj
is
the
number
= Mm{T,Hi),
of efficiency units
mahzed) amount of available
employed
in the
informal sector, and
T
is
the (nor-
land.
assume that educated and uneducated workers are perfect substitutes
In addition,
in the
informal sector:
Hi^Uj + hEi,
where h
is
the relative efficiency of educated workers.
Land
is
distributed optimally between
educated and uneducated workers, and both earn their marginal productivity, so that the ratio
of the educated to the
The production
uneducated wage
is h.
function in the formal sector
with constant returns to
capital
Wages
is
in the
Cobb-Douglas
in
human and
physical capital,
scale:
Yf =
Human
is
ApK'^Hy
a Cobb-Douglas aggregate of educated and uneducated workers:
formal sector are given by the marginal productivity of each factor:
and
wvF =
In equilibrium, the ratio
^^^^^
(1
- a)AFK'='Hp''{l -
must be equal to
h.
P)El,Up'^.
Therefore the ratio of uneducated to
educated labor in the formal sector must satisfy the relationship:
25
1-^
Together with the relationships
Ep + Ej —
Up +
S,
U]
=1
— 5 and hEi + Uj =
T, this
determines the employment of each category of workers in each sector:
Ef =
+ 5(/i-l)-T]
^[i
(11)
Ej
= ^[T-l-S{h-l) +
Up
=
Ui
= T-f3[T-l-S{h-l) +
{l-p)[l
^S]
(12)
+ S{h-l)^T]
(13)
^S]
(14)
Production in the formal sector can be expressed as a function of Ep:
Yp^ApUh^'^'
The wage
of
an educated worker
ln(u;£;)
\a{wE)
where
C
and
K^E'f^
therefore given by:
= C+aln(is:)-aln(EF)
C + a \n{K) - a ln(l + S[h -
C are functions of a,
The same equations
(with different
(15)
,
1)
- T)
(16)
,
/?,
Ap
and
h.
C
and
C)
describe the wage of an uneducated worker.
Calibration
5.2
5.2.1
The
=
is
'')
Effect of education
ratio of educated to
on formal employment
imeducated workers
in the formal sector, divided
by the
should be constant, according to our model, should and allow us to calculate
ratio of the
number
over time (table
1,
of uneducated workers to the
column
5),
while the
skill
number
premium
26
is
skill
(3.
of educated workers
stable (table
1,
column
premium,
In fact, the
is
6).
decreasing
The
value
of
(3
obtained from equation 10
is
account, such as the adoption of
(3 is
may
therefore increasing. This
more skill-complementary
reflect factors
technologies.
not taken into
The average
value of
0.62.
Equations 11 and 13 indicate how the employment of educated and uneducated workers
the formal sector responds to a change in S. Suppose that
"bom"
into the informal sector,
all
the newly educated workers are
and that young and old workers
formal sector to restore the correct proportion between
in
Ep and
shift
from the informal to the
When S
Up.
increases by 1%,
the probability that an educated worker born before 1962 will be employed in the formal sector
increases by ^
employed
workers
g
,
and the probability that an uneducated worker born before 1962
in the informal sector increases
among
(1
—
(3)
j
J5
,
where So
the share of educated
is
the old.
Values of the parameters
(the skill
by
be
will
premium)
is
/?,
h and So are shown
close to 0.40.
The
therefore predict that an increase in
S
in tables
1
average value of So
and
is
of one percentage point
3.
The average
around
0.65.
value of
/i
—
1
The model would
would make the educated and
uneducated born before 1962, respectively, 0.29 percent and 0.41 percent more
likely to
work
in
the formal sector.
The
coefficients
we have estimated suggest that an
to an increase in the probability of working for a
increase in
wage
S
of one percentage point leads
of 0.43 (in the
combined sample) and
0.71 (in the rural sample) for the educated workers. For the uneducated workers, the estimated
effects are respectively 0.51 (in the
combined sample) and
0.31 (in the rural sample). Therefore,
the model does a reasonable job in predicting the shift from the informal to the formal sector
for the
uneducated, and under- predicts the
shift of the
educated from the formal to the informal
sector, especially in the rural sample.
5.2.2
The
Effect of an increase in average education
effect of the increase in
on wage rates
average education on the wages (at a given level of
depends on the extent to which physical capital adjusts to the increase
27
in
human
human
capital)
capital.
We
conticist the implications of
two polar
the modification induced by
INPRES: we
closed
economy with
across regions.
differentially affected.
deliver the results that
INPRES
case, capital adjusts costlessly to
consider the situation where each district
first
In the second case, there
stock in response to the
first
and then a situation where capital
local accumulation,
Both models
In the
cases.
is
is
freely
is
a
mobile
wage growth across regions should not be
no adjustment
at all of the physical capital
program, over the period we consider.
• Costless capital stock adjustment, local capital accumulation
To study
the consequences of the
INPRES model
we
in the case of costless adjustment,
need to specify how human capital and physical capital are accumulated. Following Barro and
Sala-I-Martin (1995) (chapter
and physical
Output
capital.
5),
is
consider a one-sector endogenous growth model, with
human
divided between consumption and investment in the two forms
of capital.
Y=
where
7/^'
and
The changes
/// are
in
AfK'^H^f"''^
+ T = C + Ik +
the gross investment in physical and
Th,
human
capital, respectively.
the capital stocks are given by:
k = Ik- sk,
and
Hf =
There are two
differences
from the
L between
set
Vlnit
up
in
-L)-
5Hf.
Barro and Sala-I-Martin (1995).
First, there
is
a
investment in
human
capital
active: this reflects the fact that children
who go
to school today will enter the job market only
lag of duration
after a lag. Second,
a unit of
increase
human
H
77
and the date at which
represents the ratio at which each unit of investment
capital. ^^
Note that
H
could not accumulate indefinitely
were to increase the proportion of primary school graduates.
In the dual economy model,
77
reflects
is
We
it
will
become
transformed into
way
to
should think of
H
if
the only
both the cost of producing educated workers, and the rate at which
the educated workers are participating in the formal labor market.
28
as a
Cobb-Douglas aggregate of
levels of education:
all
we model
the beginning of the process
of development, where only two levels of education have been achieved.
The
solution to this
problem
obtained by setting up the Hamiltonian expression for the
is
household choice between consumption and investment
order conditions with respect to C,
K
7k,
///,
in
human
or physical capital, taking first
and Hp- The steady state growth rate
is
equal
to:
°"'
1
The steady
state value of -§-
is
L
capital. It
S—aAp
k* the value of
can model the
investment
is
to accumulate
can think about
it
as a
state, the ratio k* will
=
human
human
the transition, consumption,
in
in
(a
in
growth rate
state
human
is
given by:
-6-p]
as
(18)
an increase
INPRES program
new
in
77,
the rate at which
school makes
will last long
permanent increase
in rj)}'^
it
cheaper
enough that we
In the
new steady
the old steady state, to maintain the equality between
and physical capital
capital
(see equation 17).
18).
and physical capital
In the
new steady
in physical capital,
would not be easy to distinguish
accumulation was more rapid
The steady
the levels of wages (keeping constant
be exactly compensated by growth
it
(17)
l-a
capital accumulation: each
capital
steady state rate (given by equation
For our purpose,
[Apak*^--'^
Suppose that the
be lower than
should therefore cause a drop
will
^
permanent program
net returns to investment in
fK_Y
INPRES program
effect of the
capital.
the solution to the following implicit equation:
given by this equation.
-ff-
transformed into
human
"
\HfJ
7*
We
is
i^r
qe
Denote
^"^
I
pinned down by the condition of equality between net returns
human
to investment in physical and
A
it
the regions that received
29
will
state,
human
grow
growth
and wage growth
The program
capital).
at the
in
After
same new
human
capital
will therefore
be
from a temporary program, since the the rate of capital
more schools
for the entire period
we
consider.
human
unaffected by growth in
time,
^=
and note that
The
transition to the
To
capital.
see this, differentiate equation 15 with respect to
so that the two terms cancel.
-jj^,
new steady
-^
In this model, however, the adjustment in the ratio
decline in wages.
program was
place right after the
starts to accelerate.
To
initiated,
and well
see this, imagine that the
consider the household decision the
^,
state involves indeed a decline of
first
day the schools are
human
H
actually
a steady state, and
initially at
built:
should have taken
growth of
before the rate of
economy was
and therefore a
capital investment has
suddenly become cheaper, while physical capital investment has the same price and the same
returns as before (since the
growth
rate).
human
Net returns to investing
takes the form of
human
capital,
investing in physical capital
of this
model show that the
initiated (in one year
capital,
ratio
and
-^
capital available today keeps growing at the old steady state
in less
if
there
if
human
and the
become
ratio
in
ratio
capital are therefore higher,
jf-
declines rapidly until
it
all
is
the investment
low enough that
profitable again. Simulations of the transitional
-^
reaches
dynamics
steady state value very fast after the program
its
household are allowed to transform physical capital into
than 3 years otherwise). ^^
After reaching
its
new steady
human
state value, the
stay almost constant. -^^
This very fast adjustment rate
we have made
may seem
(costless capital adjustment)
but
unrealistic,
is
this
is
because the assumption
not realistic either, even though
implicitly or explicitly, the effort to identify a
human
it
underlies,
capital externality (in which case the
share of educated worker could have a positive effect on wages.
The important
point
is
to note
that in this model, the share of educated workers should not affect wages from 1986 to 1999.
^
For example,
every year.
if
71
was
initially
3.5% a
With reasonable parameters
assuming that the
(i.e.
is
effect of
year,
assuming a value of 0.05
values for q,
5, p,
and B
each school on the rate of growth of
S we
each school causes an increase of 0.5% in the growth rate),
built (per 1000 children),
2.5 years, jp- should
which would have been achieved
have
fallen
enough
in the
=
for S,
p
observe in the data
is
-^
0.3, i
should
in 1.25 years.
On
average region to reach the
30
=
would have declined by
0.05,
{ol
fall
=
0.02, 9
=
8%
3), and
the steady state effect
by about 10%
for
each school
average, 2 schools were built, so in
new steady
In the simulations, there are small oscillations around the steady state
dampening.
-^
A
vsilues,
state value.
and they are progressively
full
adjustment model similar to this one underlies. The data does not seem to support this
adjustment model
regions were
however: during the entire period, wages grow more slowly in
for Indonesia,
human
full
capital grows faster (see figure 4B).
• Costless capital stock adjustment, national capital
accumulation and capital
freely
is
mobile
across regions.
If
capital
is
freely
mobile across regions and there are no adjustment costs, the capital stock
simply adjusts to the increase in
human
regions. This determines the physical
capital to equalize the returns to physical capital across
capital/human capital ratio
which
k,
in
turns determines
the wage: thus (absent positive externalities), the evolution of wages should not be different in
regions that received
Once
more schools
again, the data does not
wages may of course be related to the program).
(the level of
seem
to support this model.
There were more than 10 years between the start of the program and the entry into the labor
force of the generations that
(there
Thus, even
it.
if
capital accumulates with a
gap
"time to build"), the conclusions of such a model would be unchanged.
is
Thus,
it
seems that, to account
frictions in the
•
were exposed to
pattern present of the data,
adjustment of the capital stock.
No adjustment
We
it is
necessary to introduce
consider this an extreme case of frictions.
of the capital stock in response to the increase in education
The other extreme
is
mulation to the increase
exogenously.
for the
a case where there
in the stock of
In this case,
if
the
is
absolutely no response of physical capital accu-
educated workers. The stock of capital
INPRES program
is
is
a valid instrument (that
thus evolving
is,
if it
is
not
correlated with the exogenous rate of capital accumulation), the stock of capital enters the error
term
in
equation 15 and
whether our model
instead of
\n(EF)
is
Uj
fits
S}^ Figure
is
is
uncorrelated with the level of the program.
the facts
6
is
to use ln(£'f ) or ln(l
+ 5(/i — 1)— T)
The
as the
best
way
to check
dependent variable
shows that these variables are also influenced by the program. The
the proportion of the labor force that
calculated using the district and year-specific
is
educated and works
skill
premium.
+ h*Ei.
31
T
is
in
the formal sector. ln(l
+ 5(/i— 1) — T)
calculated using its definition:
T = Hj =
first
stage coefficients for these variables look similar to each other, which
figure also looks similar to figure 2. In table 8,
The
7,
with ln(£'f ) or ln(l
1)
— T)
The
5 on
wages. In practice,
endogenous regressor of
as the
estimate
I
interest.
coefficients of a subset of the specifications estimated in table 6 are presented in table 8.
For \n{Ep) (columns
(1)
and
(2)
are negative, but insignificant.
rural sample.
The
+ S{h —
reassuring. ^^
present the results of estimating equation 15,
I
using the same strategy outlined for estimating the effect of
equation
is
The
),
the pattern
The
-I-
S{h —
1)
is
~ T)
what we found
similar to
point estimate
latter estimate of -0.22
results using ln(l
is
is
for S: the coefficients
standard errors
as the independent variable are
obtained
in the
in principle
-0.40, a.nd are all significant at the
for auto-correlation over
10%
These estimates are
They
The
clear cut.
The
estimates range
time within regions). These estimates (especially those
I
obtained using ]n{EF) (they should
are very close to (and not significantly different from) -0.3.
and
fairly precise
more
level of confidence (after correcting the
combined sample) are higher than what
be the same).
in the
close to -0.3, the conventional guess for a.
point estimates are similar in the combined sample and in the rural sample.
between -0.34 and
and -0.22
-0.07 in the urban sample,
their order of
magnitude
is
They
reasonable.
confidence in the ability of the model to describe the response of the
economy
reinforce the
to the education
expansion.
Note that the model predicts that the
in the formal
and
data on income
table
7,
in the informal sector.
for those
who do
which we commented on
effect of the
program on wages should be the same
As we already noted, the
SAKERNAS
does not have
not work for a wage. However, the regression in column 3 of
in section 4.5, suggests that the ratio of the
income of the wage
earners to that of non-wage earners was not affected by the program; the difference between the
value of this ratio in 1993 and in 1987
is
not correlated at
all
with the intensity of the program.
This set of estimates therefore suggests that the accumulation of physical capital happens
essentially independently
I
from the accumulation of human
estimate the same specification as in equation
1,
but
I
use
the dependent variables. Full results are presented in table Al.
32
capital.
InEp and
ln(l 4-
Even 25 years
S(h
—
1)
—
after the
T), respectively, as
program was
new
initiated, physical capital
efficiency units of labor created
does not seem to have been accumulated to employ the
This conclusion seems consistent with
by the program.
other evidence on the pattern of industrialization in Indonesia during the period.
the Indonesian Government cut
and small businesses
Hill (1996).
dustrialization across regions
period
in
all
regions where
previously existing tax incentives to geographical dispersion
Miguel
show that there was divergence
et al. (2001)
between 1985 and 1995: industrialization was
its level
After 1984,
was higher
in 1985.
uncorrelated with school construction (not only
Change
INPRES
in in-
faster during this
in the rate of industrialization
schools, but
seem
primary and junior
all
high schools) in the previous decade, as well as with other infrastructure variables (roads or
electricity).
Conclusion
6
This paper argued that the
INPRES
by the Indonesian government
in
program, a large school construction program undertaken
the 1970's, constitutes a good case study to empirically examine
the impact of average primary schooling on the wages of older cohorts. This program modified
the enrollment rates of the young generations, thus inducing a long-lasting change in the rate
of
human
capital accumulation in the regions
it
affected most.
The impact
of this shock on the
supply of educated workers can be studied on an "old" generation that did not directly benefit
from
it.
It
provides a natural solution to the identification problems inherent to any attempt to
identify the effect of the average of a regressor while trying at the
The instrumental
same time
to control for
it.
variables estimates presented in this paper suggest that the effect of average
education on individual wages (keeping the
skill level
constant)
is
negative:
in places
where
average educational attainments grew faster because of the program, wages grew more slowly.
This does not seem to be due to sample selection bias due to migration or increases
force participation. This
and to the
the
OLS
is
in
sharp contrast to the
fixed effect estimates,
OLS
in labor
estimates, which are strongly positive,
which are closer to zero but
still
positive.
Such a strong bias
in
estimates suggests that the cross-country relationship between output per capita and
33
education
likely to
is
be affected by the same upward
the participation in the formal labor market
Both
the
sets of estimates are
number
by the
shown
of efficiency units that
is,
entirely absorbed
is
Wages decrease
capital.
I
effects of average
education on
however, positive.
to be consistent with a simple dual-economy model, where
can be productively employed
availability of a fixed factor (land, for example).
labor force
The
bias.
by the formal
sector,
The
in the informal sector
is
increase in the productivity of the
which explains the increase
in participation.
to the extent that physical capital does not fully adjust to the increase in
contrast three versions of this model. In a closed
with costless adjustment of the capital stock, the
limited
human
economy endogenous growth version
faster increase in
human
capital after 1986
should be matched with a corresponding increase in the stock of physical capital, and wages
should be unaffected. In an open economy model where capital moves
freely,
ratio adjusts to maintain equal returns to physical capital across regions,
and the program should
have no
effect
on wages.
If
capital accumulation
is
unresponsive to the increase in
the model makes a prediction of what the coefficient on the
The
extent to which wages
fall
little
in educational opportunities in the regions
obtain are reasonable
What
if
this is the right
absent in this paper (and
is
stock did not adjust.
human
human
capital,
capital variable should be.
human
in response to the increase in
formal sector suggests that there was
the capital-labor
capital available in the
or no reaction of the physical capital to the increase
where
INPRES
built
more
schools.
The
estimates
model of the world.
left for
future work)
is
an explanation of why the capital
The program was pubhcly announced, and
the increase in the stock of
primary school graduates occurred gradually over 10 years after the program started.
a puzzle that 25 years after the program was
run" labor
demand
what extent
this
is
curve.
I
initiated, the labor
demand
Future work (and, probably, more data)
due to "myopic" behavior of investors, who
fail
is
It is
looks like a "short
needed to determine to
to recognize the increase in
the education level of the labor force, to very large adjustment costs of the capital stock, to a
financing constraint, or to a combination of the three.
The
results in this
paper are important because, contrary to what
34
is
often assumed (on the
basis of the experience of South-Ecist Asian countries), acceleration in the rate of accumulation of
human
capital
in particular)
Thomson
not necessarily accompanied by economic growth. Several countries (in Africa,
is
had very rapid expansion
(1998)).
Models of
important to understand
It is
but dismal economic growth (Kremer and
in education,
credit constraints (Banerjee
and
why
this
Newman
can be the case.
(1993), Galor
and Zeira (1993), Aghion
and Bolton (1997)) could be combined with models of costly adjustment of technology
to study
the effects of education on economic growth given the actual constraints faced by developing
economies.
The work
model combines
of Caballero
and
Hammour
(1998, 2000)
comes
closest to doing this. Their
costly adjustment with credit constraints but does not
therefore be directly applied to the question of
capital increases.
Once
built,
model growth.
what happens when the growth
It
cannot
rate of
human
such a model could then be compared to actual evolutions, in
exercises similar to Blanchard (1997) analysis of the
"medium
run".
References
Acemoglu, Daron (1996) 'A microfoundation
for social increasing returns in
human
cpital accu-
mulation.' Quarterly Journal of Economics 111(3), 779-804
Acemoglu, Daron, and Joshua Angrist (1999) 'How large are the social returns to education?
Evidence from compulsory schooling laws.' Mimeo,
MIT
Aghion, Phihppe, and Patrick Bolton (1997) 'A trickle-down theory of growth and development
with debt overhang.' Review of Economic Studies 64(2), 151-72
Atkinson, A.B., and A. Brandolini (1999) 'Promise and pitfalls in the use of "secondary" datasets:
Income inequahty
Banerjee, Abhijit, and
in
OECD
countries.'
Andrew Newman
opment.' Journal of Political
Mimeo,
Nuffield College, Oxford
(1993) 'Occupational choice and the process of devel-
Economy
101(2),
35
274-298
Banerjee, Abhijit, and Esther Duflo (2000) 'Inequality and growth:
Working Paper 7793, National Bureau
of
What
Economic Research, July
Barro, Robert, and Xavier Sala-I-Martin (1995) Economic Growth
(Mc Graw
Bertrand, Marianne Esther Duflo, and Sendhil Mullainathan (2001) 'How
differences in differences estimates.'
Bils,
can the data say?'
Mimeo,
Hill)
much should we
trust
MIT
Mark, and Peter Klenow (1998) 'Does schooling cause growth or the other way around?'
Working Paper 6393, National Bureau
Blanchard, Olivier (1997) 'The
medium
of
run.'
Economic Research
Brookings Papers on Economic Activity
2,
Blanchard, Olivier, and Justin Wolfers (1999) 'The role of shocks and institutions in the
89-158
rise of
european unemployment: The aggregate evidence.' Working Paper 7282, National Bureau
of
Economic Research
Bound, John, Gabor Kzedi, and Sarah Turner (2000) 'Trade
in university training:
variation in the production and use of college educated labor.'
MIMEO,
Cross state
University of
Michigan
Caballero, Ricardo, and
of Political
Mohamad Hammour
Economy
(1998) 'The macroeconomics of specificity. Journal
'
106(4), 724-767
(2000) 'Creative destruction and development:
Working Paper 7849, National Bureau
Card, David, and
crises,
and restructuring.'
Economic Research, August
Thomas
Lemieirx (2001) 'Can falling supply explain the rising return to college
A
cohort-based analysis.' Quarterly Journal of Economics 116(2), 705-
younger men?
for
of
Institutions,
746
Duflo, Esther (2000) 'Schooling and labor market consequences of school construction in Indonesia:
of
Evidence from an unusual policy experiment.' Working Paper 7860, National Bureau
Economic Research, August
36
(2001) 'Schooling and labor market consequences of school construction in Indonesia: Evi-
dence from an unusual policy experiment.' Am.encan Economic Review 91(4), 795-813
Foster,
Andrew
and Mark R. Rosenzweig (1996) 'Technical change and human-capital
D.,
re-
turns and investments; Evidence from the green revolution.' American Economic Review
931-953
86(4),
Frankenberg, Elizabeth, Duncan Thomas, and Kathleen Beegle (1999) 'The real costs of Indonesia's
economic
Paper 99-04,
crisis:
RAND
Preliminary findings from the Indonesia family
life
surveys.'
Working
Labor and Population Program
Galor, Oded, and Joseph Zeira (1993) 'Income distribution and macroeconomics.' Review of
Macroeconomic Studies
60,
35-52
Glewwe, Paul, and Hanan Jacoby (2000) 'Economic growth and the demand
there a wealth effect?'
Harris,
J.,
American Economic Revieio
Heckman, James
effects:
J.,
A
Is
MIMEO, World Bank
and M. Todaro (1970) 'Migration, unemployment and development:
analysis.'
ment
for education:
40.
A
two-sector
126-142
Lance Lochner, and Christopher Taber (1998) 'General equilibrium
study of tuition
policy.'
Working Paper 6426. National Bureau
of
treat-
Economic
Research
Hill,
Hal (1996) The Indonesian Economy Since 1966: Southeast Asia's Emerging Giant (Cambridge:
Cambridge University Press)
Katz, Lawrence F., and Kevin M.
and demand
Murphy
factors.' Quarterly
(1992) 'Changes in relative wages, 1963-1987: Supply
Journal of Economics 107(1), 35-78
Kremer, Michael, and James Thomson (1998) 'Why
isn't
convergence instantaneous?
workers, old workers, and gradual adjustment.' Journal of Economic Growth
Krueger, Alan, and Mikael Lindahl (1999) 'Education for growth:
37
Why
and
for
3,
young
5-28
whom?' Mimeo
Lucas, Robert (1988) 'On the mechanics of economic development.' Journal of Monetary Eco-
nomics
22, 3-42
Miguel, Edward, Paul Gertler, and David Levine (2001) 'Did industrialization destroy social
capital in Indonesia.'
Mimeo, U.C. Berkeley
Moretti, Enrico (1999) 'Estimating the social return to education:
cross-sectional
Solow, Robert
M.
and longitudinal
data.'
Working Paper
(2000) 'Toward a macroeconomics of the
22,
UC
Evidence from repeated
Berkeley
medium
run.'
Journal of Economic
Perspectives 14(1), 151-158
Welch, Finis (1979) 'Effects of cohort size on earnings: The baby
Journal of Political
Economy
87(5), S65-S97. part 2
38
boom
babies' financial bust.'
-
oooooooooooooo
E
3
OOOOOOOOOOOOOO
a O
o
r
(1)
fTi
-1
—
-n
u-
f
>
o
U
o
re
1
o
mmrNjrNifNifNfN—'rjf^lr-i — — —
C>0
CO
0'OOOOC>00>0:000
rl
-a
o
3 c
3
—O o
r-l
^^o
m ^
tn oo
—
r-
^—
^ ^ ^
oooooooooooooo
m:)
r-~
s
c
a.
"o
^ \0 ON On
^ ^^ m
r~ r~
o o o o o o o
On
5
o
r--
r--
r-~-
t-~
—
ON
^
oo r~ oo oo oo oo
o o o o o o
"v^
r--
c
o
IX
\D
ON
O
^
mNO'— moo — <NNOoofNi^NO^-
oooooociooooooo
E
3 pa
C
> t) o
rt j=:
c
o
w
•fi
O
c
o
o
— r-tN
rN)(Nr*-iro'^'rfrtun>r-i^ONOr-r--oo
<N<0(NNO(NlNOOs'/-100(Nr-
oooooooooooooo
NOr-ooONO
— (-Niro'^'ONOr-oooN
OOOOOOOOOnOnOnOnOnOnOnOnOnOn
On On on
On
On On On On On On On
On On
H,
?
^>^
T3
OJ
£
op
a>
so '5
>
<
On
•u
5
.2
C3
W5
C
o
-§
><
t)
u
00
.
Table 2: First stage regressions
program on the proportion of individuals completing primary school or more.
Coefficients of interactions of the intensity of the program and survey year dummies.
Effect of tine
Sample: Urban and
1
Sample: rural areas only
rural areas
Bom
Sample:
1986
Bom
20-40
20-60
before 1962
20-40
20-60
before 1962
(1)
(2)
(3)
(4)
(5)
(6)
omitted
omitted
omitted
omitted
omitted
0.0028
-0.0004
-0.0026
0.0009
-0.0015
-0.0023
(.008)
(.0075)
(.0077)
(.0099)
(.009)
(.0091)
0.0015
-0.0028
-0.0050
-0.0032
-0.0079
-0.0088
omitted
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
(.0064)
(.0058)
(.0059)
(.0078)
(.0069)
(.0069)
0.0014
-0.0012
-0.0048
-0.0010
-0.0043
-0.0068
(.0053)
(.0047)
(.005)
(.0066)
(.0056)
(.0061)
0.0027
-0.0042
-0.0106
-0.0002
-0.0071
-0.0126
(.0061)
(.0056)
(.0059)
(.0072)
(.0065)
(.007)
0.0151
0.0074
-0.0002
0.0110
0.0032
-0.0034
(.0058)
(.0055)
(.0055)
(.0067)
(.0062)
0.0124
0.0006
-0.0074
0.0082
-0.0058
(.0065)
(.0061)
(.0061)
(.0076)
(.0068)
(.007)
0.0205
0.0088
-0.0017
0.0143
0.0012
-0.0082
(.0065)
(.0062)
(.0059)
(.0072)
(.0067)
(.0067)
0.0195
0.0076
-0.0044
0.0145
0.0005
-0.0103
(.0068)
(.0068)
(.0069)
(.0074)
(.0072)
(.0078)
0.0202
0.0074
-0.0064
0.0161
0.0003
-0.0127
(.0064)
(.0063)
(.0061)
(.0072)
(.0068)
(.007)
0.0260
0.0112
-0.0027
0.0207
0.0016
-0.0122
(.0064)
(.0066)
(.0069)
(.007)
(.0066)
(.0075)
0.0252
0.0149
0.0039
0.0190
0.0070
-0.0010
(.0071)
(.0072)
(.0076)
(.0077)
(.0076)
(.0086)
0.0340
0.0190
0.0044
0.0300
0.0110
-0.0040
(.0077)
(.0074)
(.0072)
(.0073)
(.0066)
(.0076)
0.0290
0.0154
0.0045
0.0253
0.0076
-0.0029
(.0082)
(.0082)
(.0083)
(.0078)
(.0073)
(.0086)
3826
3826
3826
3140
3140
3140
7.23
3.40
1.29
5.02
1.68
0.95
(.0065)
•
-0.0126
Number of
cells
F. statistic
Notes:
1. The program intensity is
number of children in 97 1
the
number of INPRES schools
built
between 1974 and 1978, divided by the
1
2.
Survey year dummies, region dummies, interactions between survey year dummies and the enrollment
rate in 1971,
and interactions between suvey year dummies and the number of children are included
in the regressions.
3.
Regression
mn using kabupaten-year
kabupaten-year
4.
5.
averages, weighted
by
the
number of observations
cell.
The F statistic is for the hypothesis that the set of interactions is jointly insignificant.
The standard errors are corrected for auto-correlation within kabupaten.
in
each
Table 3: Descriptive Statistics
Sample: individuals aged less than 60 and born before 1962
Individuals
Survey year
% primary school
graduates
bom
for
(1)
before 1962, aged less than 60
% working
waige
(2)
average
S.D. of
skill
wage
wage
premium
(3)
(4)
(5)
1986
0.59
0.31
6.56
0.71
0.44
1987
0.60
0.32
6.58
0.66
0.48
1988
0.60
0.31
6,55
0.65
0.40
1989
0.61
0.33
6.61
0.66
0.46
1990
0.62
0.33
6.66
0.67
0.46
1991
0.65
0.34
6.70
0.66
0.44
1992
0.64
0.33
6.77
0.67
0.47
1993
0.65
0.34
6.81
0.73
0.51
1994
0.66
0.36
6.88
0.71
0.47
1995
0.64
0.36
6.93
0.72
0.57
1996
0.71
0.35
6.98
0.68
0,50
1997
0.69
0.36
7.08
0.72
0.54
1998
0.72
0.33
6.82
0.69
0,54
1999
0.74
0.33
6.86
0.70
0.57
.
Effect of the
Table 4: Reduced form regressions
program on wages, residual wages, and formal employment among individuals born before 1962.
Coefficients of interactions of the intensity of the program and survey year dummies
Sample: Urban and
1987
1988
Sample Rural areas only
rural areas
:
old,
Wages
Residual
Formal
wage
employment
(1)
(2)
(3)
omitted
omitted
omitted
;
1986
bom
20-60 yean
before
1
20-60 years old,
962
Wages
bom
before 1962
Residual
Formal
wage
employment
(4)
(5)
(6)
omitted
omitted
omitted
-0.0059
0.0025
-0.0064
-0.0087
0.0209
0.0094
(.0187)
(.014)
(.0056)
(.018)
(.0139)
(.0056)
-0.0041
-0.0051
-0.0040
0.0108
0.0103
-0.0047
(.0152)
(.0141)
(.004)
(.0139)
(.0136)
(.0042)
1989
0.0022
-0.0011
-0.0074
0.0155
0.0095
-0.0081
(.0161)
(.0139)
(.0047)
(.0159)
(.0141)
(.0044)
1990
-0.0140
-0.0111
-0.0027
0.0047
0.0056
-0.0030
(.0155)
(.0133)
(.0054)
(.0165)
(.0143)
(.0054)
1991
-0,0061
-0.0079
0.0035
0.0082
0.0047
0.0039
(.0187)
(.0146)
(.0051)
(.0185)
(.0149)
(.0049)
1992
-0.0173
-0.0116
0.0019
-0.0055
0.0021
-0.0002
(.0163)
(.0141)
(.0055)
(.0186)
(.0145)
(.0055)
1993
-0.0166
-0.0132
0.0069
-0.0097
-0.0038
0.0074
(.0156)
(.0137)
(.0055)
(.0203)
(.017)
(.0054)
-0.0222
-0.0189
0.0103
-0.0149
-0.0170
0.0068
(.0165)
(.0136)
(.007)
(.0224)
(.0175)
(.0069)
1994
1995
0.0031
-0.0010
0.0071
0.0044
0.0079
0.0058
(.0145)
(.0134)
(.0056)
(.0188)
(.015)
(.0051)
1996
-0.0190
-0.0173
-0.0035
-0.0158
-0.0116
-0.0051
(.0176)
(.0143)
(.0071)
(.0251)
(.0178)
(.0062)
1997
-0.0192
-0.0155
0.0061
-0.0119
-0.0166
0.0042
(.0174)
(.0169)
(.0068)
(.0232)
(.0183)
(.0066)
-0.0079
-0.0141
0.0083
-0.0190
-0.0072
0.0105
(.0189)
(.0157)
(.0074)
(.0219)
(.0164)
(.0072)
0.0034
-0.0104
0.0078
-0.0188
-0.0189
0.0074
(.019)
(.0159)
(.0071)
(.0238)
(.0185)
(.0076)
3804
3804
3804
3119
3119
3119
1998
1999
Number of
cells
Notes:
1
The program intensity is the number of INPRES schools
number of children in 1971.
2.
built
between
1
974 and 1 978, divided by the
Survey year dummies, region dummies, interactions between survey year dummies and the enrollment
rate in 1971,
and interactions between suvey year dummies and the number of children are included
in the regressions.
3.
Regression
mn using kabupaten-year averages,
kabupaten-year
4.
The standard
weighted by the number of observations in each
cell.
errors are corrected for auto-correlation within kabupaten.
Table
5:
OLS
estimates of the impact of average education on individual wages.
Men aged 20-60 and
born before 1962.
Independent variable:
% of primary
Independent variable:
school graduates in the 20-40 sample
Sample: Rural
PANEL
and urban areas
areas only
(1)
(2)
(3)
(4)
A: OLS, without region fixed effects (1986-1999)
premium
Formal employment
Formal employment among educated workers
Formal employment among uneducated workers
B:
OLS, with region
0.852
0.754
0.881
0.875
(0.035)
(0.041)
(0.030)
(0.037)
0.195
0.226
0.228
0.312
(0.024)
(0.028)
(0.021)
(0.027)
-0.458
-0.531
-0.423
-0.533
(0.041)
(0.047)
(0.038)
(0.045)
0.431
0.162
0.459
0.170
(0.016)
(0.015)
(0.015)
(0.015)
0.095
-0.036
0.108
-0.035
(0.015)
(0.015)
(0.013)
(0.015)
0.035
0.0046
0.045
0.0088
(0.011)
(0.011)
(0.011)
(0.011)
fixed effects (1986-1999)
Log(wage)
Log(wage) residual
Skill
premium
Formal employment
Formal employment among educated workers
Formal employment among uneducated workers
0.409
0.436
0.537
0.594
(0.041)
(0.046)
(0.041)
(0.047)
0.040
0.060
0.086
0.121
(0.032)
(0.035)
(0.032)
(0.037)
-0.082
-0.092
-0.094
-0.106
(0.073)
(0.080)
(0.077)
(0.085)
0.160
0.134
0.228
0.201
(0.016)
(0.017)
(0.016)
(0.018)
-0.0019
0.0079
0.017
0.031
(0.020)
(0.022)
(0.021)
(0.022)
0.012
(0.016)
.
0.015
0.031
0.033
(0.017)
(0.018)
(0.018)
Notes:
1.
Survey year dummies, region dummies, interactions between survey year dummies and the enrollment rate
and interactions between suvey year dummies and the number of children are included
2.
Sample: Rural
areas only
Log(wage) residual
PANEL
Sample: Rural
Sample: Rural
% of primary
m the 20-60 sample
and urban areas
Log(wage)
Skill
school graduates
in
1971,
in the regressions.
Regression run using kabupaten-year averages, weighted by the number of observations in each kabupaten-year
cell.
Table
6:
2SLS
estimates of the impact of average education on individual wages.
Men aged 20-60 and
born before 1962.
Independent variable:
% of primary
school graduates in the 20-40 sample
PANEL
A:
YEARS
Independent variable:
school graduates
% of primary
in the
20-60 sample
Sample: Rural
Sample: Rural
Sample: Rural
Sample: Rural
and urban areas
areas only
and urban areas
areas only
(1)
(2)
(3)
(4)
1986-1999
Log(wage)
-0.204
-0.834
-0.208
(.443)
(.701)
(.615)
(.837)
Log(wage) residual
-0.292
-0.633
-0,379
-0.994
(.355)
(.431)
(.512)
(.556)
premium
-0.434
-0.982
-0.596
-0.636
(.916)
(1.408)
(1.197)
(1.645)
0.441
0.454
0.661
0.745
(.159)
(.203)
(.238)
(.352)
Skill
Formal employment
Formal employment among educated workers
-0.871
0.432
0.501
0.543
0.713
(.197)
(.259)
(.264)
(.406)
0.379
0.409
0.510
0.318
(.203)
(.232)
(.354)
(.318)
Log(wage)
-0.358
-0.710
-0.451
-0.480
(.493)
(.821)
(.716)
(.801)
Log(wage) residual
-0.330
-0.588
-0.437
-0.902
(.412)
(.529)
(.618)
(.602)
premium
-0.225
-0.635
-0.291
0.536
(1.033)
(1.461)
(1.488)
(1.576)
0.463
0.442
0.716
0.694
(.282)
(.379)
Formal employment among uneducated workers
PANELS: YEARS
Skill
1986-1997
Formal employment
Formal employment among educated workers
Formal employment among uneducated workers
i
(.183)
(.233)
0.428
0.473
0.530
0.622
(.229)
(.301)
(.317)
(.479)
0.478
0.449
0.624
0.263
(.249)
(.277)
(.415)
(.319)
Notes:
Survey year dummies, region dummies, interactions between survey year dummies and the enrollment rate
and interactions between suvey year dummies and the number of children are included in the regressions.
1.
2.
Regression run using kabupaten-year averages, weighted by the number of observations
3.
The instruments
4.
The standard
are interactions
between survey year dummies and the program intensity
errors are corrected for auto-correlation within kabupaten.
in
in 1971,
each kabupaten-year
cell.
1
Table
7:
Additional evidence on
Log(wage migrant
in)
-log(wage stayer)
Log(wage migrant
out)
Formal sector premium 1993
-Formal sector premium 1987
-log(wage stayer)
11)
Program
sample selection
(3)
(2)
0.0229
0.011
-0.00207
(0.0159)
(0.0164)
(0.0160)
285
288
272
intensity
Number of cells
Note:
1
-
The program
intensity
number of children
in
1
2.
The formal
3.
The eiuollment
4.
The
data for columns
5.
The
data for column (3)
sector
the
is
premium
rate
number of INPRES schools
built
between
1
974 and
1
978, divided by the
97
is
defined as
:
log(income of wage eamer/income of non wage earners)
and the number of children
( 1 )
and
is
the
in
1971 are introduced as controls
SUPAS 1995
SAKERNAS, 1987 and
(2)
is
the
1993.
in the regressions
—
-
O
"ca
3
o
o
(N
T3
00
_>.
C
O
"H-
E
oj
u
rt
1/5
KJ
—
"H
W)
(U
O)
CO
CO
g
'5.
g
D
a S
o 3
i
E
£
D
f~
M
o
o
CO
Q.
E
?^
E
CJN
m
w
;:'
1>
i^
cd
T3
E
.S
TO
r-
^
>n2
E
o
""
3
73
C/I
o
Cfl
<u
1-
3
Ki
cn
CD
CD
o
^
o
E
CO
H
CD
—
W
-co
+
(/)
^o
V
03
B.
M—
,.
CN
1-1
s
CO
OS
00
CO
C
u^
uo
c (0
o o
c
ro
Q.
•o
MD rro
c^
1^o
CO
'"^
CO
-I
c
<^
m
03
O
(0
-d
J
x>
.3
c
C3
>
C
(U
oc
tu
rt
>
c
<u
T3
e
(!>
CL
<U
3
_>,
Cd
c
o
jj
CO
C3
E
u.
00
ca
cd
00
o
O-
p
3 2
^ 3
C
O
E
cd
sa
U-
3
X
Cvl
^&0
cd
|D
ii
X3
3 ^
-a
5
o ^
cd
C
'p
c
?:
3
5 s ^^
a -a
c/l
Di
D.
I.
cd
"
^
:^^
u
c
cd
E
00
p
C
C
3
S
3
cd
U
u
=
^
C
2
e ^
CO
3
cd
C
2
C
o
C
V
O
h
-a
>
S
C
cd
0\
OS
On
-
(0
C/i
Pi
<
1
-o
13
tii
<
0-,
w
<
a,
^
(n
c
o
g
to
s
3
O
C/^
Z ^
T3
cd
cd
T3
CO
CO
OJ
-C
Xi
Di
H H
csi
r<i
^'
c
U
-o
2
CO
C
c
CO
U
en
o
<
(/)
U)
F
E
n
a
I
m
a)
>^
>.
U)
r-^
CO
in
to
tl)
3
r U
m m
T)
>.
m
o
c o
o
a>
to
E
>^
ro
(Tl
m
E
o •^
n la
o o
U)
c:
o^
o m
o n
03
m
(U
1cu
x.
to
>
c
ai
-n
o
tf)
r
(U
o
e
O
O
CD
cm'
ni
13
CD
U.
n
(B
Q
!n
U)
lil
m
3
r T3
(i>
F
m
en
ro
C5>
n
o
Q. -C
<)
n
m
L
in
£ro
O F
(
>
CTi
<l)
c
u.
o
ss
a)
r
di
J3
o
(/)
ro
c ^
m
0)
<u
o
o
XI
CD
Dr
<u
n.
0)
Q
't
o
o
LD
CO
O
O
CO
o
o
in
CVJ
o
o
CM
o
o
in
o
o
.,—
o
o
in
o
o
o
'
'
^
T
en
m
en
^
-
o
«D
""-,
to
-
C
--._
1
(0
^
«
u>
-
a^
i^
"'3
Oi
ft
<D
.3^
en
E
">
" - .
T
1^
TJ
u.
CO
(U
CSJ
CO
a>
>> <>< Q)
'
CD
en
en
1^
^
^-
-
'
- '
,
- .
_
E
o
S
to _Q
•o
>• J2
CO
to
c
C 3 o
<D "O
to
.
*
en
^
"
-
/
<^
.*-'
ro
- - .
/
en
E c CO
(0 —
^
O) c D
O (D "D
a E C
M- >> CO
o o c
m o. CO
c E -Q
o 05 1"o
"
-
/
c
c
CD
F
><
C)
r\l
en
en
,
^—
Vr
o m
A
-•
CO
OJ
O)
CO
o.
c
in
E
To
(X)
§
!3)
CO
£
E
o
a
CO
T3
.
05
O
'
^-
.05
C3)
(D
CO
""
\
o g
/V-
^U.
-o
<
"*
.
s
..
<u
1^
00
''
5^
"
.''
N^
^
-rt'"
*
<y>
.i>l
U.
C
CO
>
c
"
CO
CO
05
0)
' >
^
w/
^
ac
a>
Q.
/
0)
Q
>^
">
A'
in
o
C3
o
C3
o
o
ci
^
00
1
m
o
o
o
y
Q)
CO
E
_l
a
f
II
^
1" 'oT
St:
a.
1
,—
o
o
1
ir>
o
o
1
CNJ
o
o
^
in
CM
o
o
o
to
a>
A
T
A
T
C7>
Oi
!
/
/
/
CO
en
<\
^,^
t—
'
«
>^
^^
CO
en
^
J»
.
,.--''
'
,
/"^
^-.'
,
-
'
'
in
'
CT)
'
* -
,
5f
O)
CT)
*
c
g
*
_J
'ro
0)
'
'
CO
cn
O)
J
Ul
-
>.
3
,^^
-|
to
CM
k
E
en
§,
O)
TO
CO
o
_I
•
S
.-'
\
a>
CO
>
1
«
«
«
»
»
'
>
CO
CO
/
i
T—
\
'
^
.
'
'
.
t^
:.'
\
' " .
• - .
,
" " ^
r
1
in
CM
o
o
CM
o
o
m
o
o
1
^_
o
o
i
m
o
o
o
^'
CO
35
-
u
_
J
LL
'
C3>
.'
5
O)
C7^
1
1
1
m
o
o
CD
'
'
o
o
in
CM
o
C)
»
Figure 5
Effect of an Increase in the Share of Educated Workers on
Wages and Formal Employment
^"
1
W
WuF
labor
labor
demand
demand
in the
in the
N
formal sector
/
.
w
w"^UF2
^-
/
/
/i~~~~~~~..,,^^/'
/'
'
,^
\
Uf2=1-U,2
Up,= l-U„
;
/
y'
—
i
/
^-^^^
.>.,^.
^UI
informal sector
Wu„
'Wu,2
UJ
en
O
I
H
c/)
+
C3)
O
oo
o
o
o
o
o
o
CN
o
o
CN
O
O
o
CO
o
Table A1: Alternative
first
stage: Effect of
Sample:
and
1986
rural areas
iog^tf)
and log(1+s(h-1)-T).
Independent variable::log(l+s(h-l)-T)
Independent variable: log(Ef)
Sample: Urban
program on
tine
Men aged 20-60
Sample: Rural
areas only
Sample: Urban
and
rural areas
Sample: Rural
areas only
(1)
(2)
(1)
(2)
omitted
omitted
omitted
omitted
-0.0244
-0.0344
-0.0320
1987
-0.0242
(.0429)
(.0521)
(.0546)
(.0663)
1988
-0.0279
-0.0502
-0.0131
-0.0248
(.022)
(.0259)
(.0304)
(.0361)
1989
-0.0109
-0.0331
-0.0409
-0.0697
(.0285)
(.0349)
(.0319)
(.0373)
0.0185
0.0033
-0.0057
-0.0135
(.0313)
(.0345)
(.0324)
(.0387)
0.0540
0.0436
0.0153
0.0058
(.0292)
(.0313)
(.03)
(.0321)
1990
1991
1992
1993
1994
1995
.
1996
1997
1998
1999
Number
0.0320
0.0109
0.0102
-0.0176
(.0307)
(.0325)
(.0306)
(.0338)
0.0518
0.0339
0.0250
0.0008
(.0308)
(.0332)
(.0335)
(.0373)
0.0534
0.0238
0.0548
0.0220
(.0345)
(.0392)
(.0354)
(.0404)
0.0681
0.0515
0.0266
0.0198
(.0307)
(.0328)
(.0352)
(.0363)
0.0459
0.0231
0.0313
0.0060
(.0344)
(.0343)
(.0363)
(.0401)
0.0703
0.0510
0.0522
0.0327
(.0375)
(.0414)
(.0404)
(.0477)
0.0807
0.0708
0.0417
0.0137
(.0351)
(.0371)
(.0517)
(.0615)
0.0691
0.0598
0.0918
0.1131
(.0357)
(.0383)
(.0476)
(.063)
3808
3122
3336
2762
3.28
2.49
1.69
1.30
of
cells
F. statistic
Notes:
1. The program intensity is the number of INPRES schools
number of children in 1971.
2.
built
between 1974 and 1978, divided by
Survey year dummies, region dummies, interactions between survey year dummies and the enroUr
rate in
1
97 1 and interactions between suvey year dummies and the number of children are included
,
in the regressions.
3.
Regression run using kabupaten-year averages, weighted by the number of observations in each
kabupaten-year
cell.
4.
The F
5.
See text for variables definition
6.
The standard
statistic is for
the hypothesis that the set of interactions
is
jointly insignificant.
errors are corrected for auto-correlation within kabupaten.
Lib-26-67
MIT LIBRARIES
mil
III
III
iiiiiiiiiiii|
iiipi
I
3 9080 02527 8893