Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium
Member
Libraries
http://www.archive.org/details/fromeducationtodOOacem
fiBWB,
HB31
M415
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
From Education
to
Democracy?
Daron Acemoglu, Simon Johnson,
James A. Robinson,
Pierre Yared
Working Paper 05-04
December 15, 2004
RoomE52-251
50 Memorial Drive
Cambridge,
This
MA 021 42
paper can be downloaded without charge from the
Network Paper Collection at
Social Science Research
http://ssrn.com/abstract=668542
MASSACHUSETTS INSTITUTE
OF TECHNOLOGY
|
j
MASSACHUSETTS INSTITUTE
OF TECHNOLOGY
LIBRARIES
\
From Education
Daron Acemoglu^
to Democracy?*
Simon Johnson
James A. Robinson^
1
Pierre Yared^
This Version: December 2004.
Abstract
The conventional wisdom views high
levels of education as a prerequisite for
democracy. This paper shows that existing evidence for
this
view
sectional correlations,
which disappear once we look
other words, there
no evidence that countries that increase
more hkely
to
is
based on cross-
their education are
become democratic.
Keywords: democracy,
JEL
is
at within- country variation. In
education, institutions, political development.
Classification: P16, O10.
for the American Economic Association Papers and Proceedings 2005.
^Department of Economics, Massachusetts Institute of Technology, 50 Memorial Drive, MA02139.
e-mail: daron@mit.edu.
*Sloan School of Management and International Monetary Fund, Massachusetts Institute of Technology, 50 Memorial Drive, MA02139. e-mail: sjohnson@mit.edu, sjohnson@imf.org.
^Harvard University, Department of Government, Littauer, 1875 Cambridge St., Cambridge MA02138;
'Prepared
e-mail: jrobinson@gov.harvard.edu.
^Department of Economics, Massachusetts
e-mail: yared@mit.edu.
Institute of Technology, 50
Memorial Drive, MA02139.
.
Introduction
1
The
conventional wisdom, since at least the writings of John
levels of educational
to promote democracy both because
because
it
The most
(1916), views high
Education
is
argued
enables a "culture of democracy" to develop, and
is
also
celebrated version of this argument
is
thought to cause
political
development.
modernization theory, popularized by
Lipset (1959), which emphasizes the role of education as well as economic
growth in promoting
for
it
leads to greater prosperity, which
Seymour Martin
Dewey
attainment as a prerequisite for democracy.
political
development
in general
and democracy
in particular. Lipset,
example, argues that
"Education presumably broadens men's outlooks, enables them to understand the need for norms of tolerance, restrains them from adhering to extremist and monistic doctrinies, and increases their capacity to
make
rational
electoral choices." (p. 79),
and concludes
we cannot say
"If
for
that a "high" level of education
is
democracy, the available evidence does suggest that
a necessary condition"
Recent empirical work,
it
a
sufficient condition
comes
close to being
(p. 80)
for
example, by Robert Barro (1999) and
Adam
Przeworski,
Michael Alvarez, Jose A. Cheibub and Fernando Limongi (2000), provides evidence consistent
with this view. Edward Glaeser, Rafael La Porta, Florencio Lopez-de-Silanes and
Andrei Shleifer (2004) go further and argue that differences in schooling are a major
causal factor explaining not only differences in democracy, but
institutions,
The high
and provide evidence consistent with
correlation
cornerstone of this view.
more generally
in political
this view.
between schooling and democracy, depicted
This figure shows the most
in Figure 1,
common measure
is
the
of democracy,
the Freedom House index of political rights, against the average years of schooling of
the population in the 1990s (see below for data details). Correlation does not establish
causation, however.
Existing literature looks at the cross-sectional correlation between education and
democracy rather than at the within
tentially driven
run.
A
variation.
Hence existing inferences may be po-
by omitted factors influencing both education and democracy in the long
causal link between education and democracy suggests that
we should
also see a
relationship between changes in education
and changes
should ask whether a given country (with
its
likely to
become more democratic
that the answer to this question
is
in
democracy. In other words, we
is
more
We
show
other characteristics held constant)
as its population
becomes more educated.
no. Figure 2 illustrates this
by plotting the change
in
the Freedom House democracy score between 1970 and 1995 versus the change in average
years of schooling during the
same time
period.
Countries that become more educated
show no greater tendency to become more democratic. In
ure
1
seems to be driven by some
common
this light, the pattern in Fig-
omitted factors affecting both education and
democracy.
We
further investigate these issues econometrically.
relationship between schooling
We
show that the
and democracy disappears when country
cross-sectional
fixed effects are
included in the regression. Although fixed effects regressions are not a panacea against
biases arising in pooled
OLS
long-run determinants of both education and democracy.
lack of a relationship between education
and democracy
We
is
all
removing the potential
regressions, they are very useful in
also
document that the
highly robust to different
econometric techniques, to estimation in various different samples, and to the inclusion
of different sets of covariates.
The recent paper
Glaeser, et
al.
(2004) also exploits the time-series variation in democ-
racy and education, and presents evidence that changes in schooling predict changes in
democracy and other
political institutions.
stems from their omission of time
increase in education
include year
dummies
However, we document below that this result
effects in the regressions, so
and democracy at the world
in their regressions, the
pears entirely. Motivated by the Glaeser, et
no
effect of
al.
reflects the over-time
35 years. Once we
impact of education on democracy disap(2004) paper,
we
also
show that there
literature
is
related to the large political
on the creation and consolidation of democracy, which we do not have
enough space to discuss here
(see
Daron Acemoglu and James A. Robinson, 2004,
discussion of this literature).
It is
also related to our
mon
for a
companion paper by Acemoglu,
Si-
Johnson, Robinson and Pierre Yared (2004), which investigates the other basic tenet
of the modernization hypothesis, that
In that paper, using both fixed effects
that there
We
is
education on other measures of political institutions.
In addition to the studies mentioned above, our paper
economy
it
level over the past
is little
also offer
income (and economic growth) causes democracy.
OLS and instrumental variable regressions, we show
evidence in favor of a causal effect from income to democracy either.
a theory for the differences in long-run factors causing the joint evolution of
education, income, and democracy, and
we provide supporting evidence
for this theory.
The
rest of the
we
In Section 3
results. Section 5
measures of
paper
is
we
organized as follows. In Section 2
present our
main
shows that there
is
briefly describe
the data.
Section 4 documents the robustness of these
results.
no evidence of a causal
effect of
education on other
political institutions. Section 6 concludes.
Data
2
We follow the existing literature in economics and measure democracy using the Freedom
House
7,
Index (see Freedom House, 2004). This index ranges from
Political Rights
with 7 representing the least amount of
country gets a score of
list
political
come
1 if political rights
freedom and
1
to
the most freedom.
A
closest to the ideals suggested
of questions, beginning with whether there are free
who
1
and
by a check-
whether those
fair elections,
are elected rule, whether there are competitive parties or other political groupings,
whether the opposition plays an important role and has actual power, and whether minority
groups have reasonable self-government or can participate in the government through
we supplement
informal consensus. Following Barro (1999),
Kenneth Bollen (1990, 2001)
variable from
both indices so that they
lie
between
and
1,
for 1955, 1960,
with
1
this index with the related
and 1965, and we transform
corresponding to the most democratic
set of institutions.
We also show that our results are robust to using the other two popular measures in the
composite Polity index, and the dichotomous democracy index developed
literature, the
by Przeworski,
which are
all
we do not
details,
(2000) and extended by Carles Boix and Sebastian Rosato (2001)
et al.
normalized between
and
1 for
comparison. Because of space restrictions,
describe these data here and refer the reader to
and
also for descriptive statistics
Our main right-hand
and above,
intervals
between 1960 and 2000. The value of
al.
(2004) for
available in five year
sample ranges
4.65.
is
a five-yearly panel, where we take the democracy score for each
fifth year.
This results in an unbalanced panel of 108 countries spanning the
basic dataset
country every
is
this variable in our base
from 0.04 to 12.18 years of schooling with a mean of
Our
et
side variable, average years of schooling in the total population
from Barro and Jong-Wha Lee (2000) and
of age 25
is
Acemoglu
on the key variables.
period between 1965 and 2000, with a total of 765 observations with countries included
they have been independent for at least
using the
CIA World Factbook
(2004).
five years,
We
where independence year
is
if
determined
prefer using the observations every fifth year
to averaging the five-yearly data, since averaging introduces additional serial correlation
'as
we document below.
We
Nevertheless, our results are robust to using five-year averages.
also report robustness checks using 10-year
subsamples that exclude former and current
data between 1970 and 2000 and using
socialist countries,
Sub-Saharan Africa, and
predominantly Muslim countries.
Results
3
Table
provides our
1
pooled
OLS
main
results using the
relationship between education
Freedom House data. Column
shows the
1
and democracy by estimating the following
model:
dit
where da
variable
is
dynamics
in
some equilibrium value
_i
i
+ [i + v
capture
(1)
it
t
in period
democracy
t.
The lagged
value of this
common
effect
(common
shocks to
an error term, capturing
Column
1
shows a
on democracy.
fi t
the tendency of the democracy
(i.e.,
for the country)
the lagged value of average years of schooling.
whether education has an
is
it
included on the right hand side to capture persistence in democracy and also
score to return to
Vn
+ 7S
adit-x
the democracy score of country
is
potentially mean-reverting
Sit-i,
=
The main
.
variable of interest
The parameter 7
denotes a
full set of
is
therefore measures
time
trends in) the democracy score of
all
effects,
which
countries,
and
other omitted factors.
all
statistically significant correlation
between of education and democ-
racy.
The estimate
of
1%}
If causal, this
estimate would imply that an additional year of schooling increases
7
0.027 with a standard error of 0.004, which
is
the "steady-state" value of democracy by 0.093
effect is calculated as
mean
of
democracy
7/
(rs
0.027/(1
—
0.709),
significant at
where the long-run
This
is
a reasonably large magnitude relative to the
sample which
is
0.57. Notice that this estimate includes
(1
in the
—
is
a)).
both
the direct and the indirect effect of education on democracy working through income
(since greater education corresponds to greater income,
which might also lead to more
democracy). Below we also report models that estimate the separate
effect of
education
and income.
Equation
(1) is similar to
the regressions in the existing literature in that
control for country fixed effects.
Thus the
it
does not
entire long-run differences across countries are
used to estimate the effect of education on democracy. As a result, omitted factors that
influence both
democracy and education
in the long
run
will lead to spurious positive
estimates of 7.
1
All the standard errors are robust for arbitrary heteroscedasticity
and
clustering at the country level.
The
alternative
to allow for the presence of such omitted factors (that are not
is
time- varying) by including a country fixed effects, by estimating a model of the form
dit
which only
The
identical to
different,
from
differs
Table
rest of
column
however.
1
1
=
adi t _i
+ jsu-i + fi + Si + u
t
because of the
(1)
full set
reports estimates of
except for the fixed country
Now 7
is
effects,
the
dummies, the
5j's.
Column
similar to (2).
The
Jj's.
2
is
results are radically
estimated to be -0.005 with a standard error of 0.019, thus
highly insignificant and has the opposite sign to that predicted by the modernization
it is
OLS
hypothesis (and to that found in the pooled
In the regression in
with u is for s
<
t,
column
2,
(e.g.,
Jeffery
because the regressor da-i
(2) to
column
W. Anderson and Cheng Hsiao
dit-2 is uncorrelated
We
with
Au
3, is
to use the
t
lt
,
11).
Our
first
strategy
methodology proposed by
un
(i.e.,
t
-i+AiJ,t +Auit. In the absence
no second order
serial correlation in
so can be used as instrument for
find that this procedure leads to even
example, in our basic specification, 7
0.021),
mechanically correlated
(1982). This involves time differencing equation
of serial correlation in the original residual,
consistent estimates.
is
1).
not consistent in panels with a
Ada = aAdi -i+jAsi
eliminate the 5^s, which yields
Aun),
is
M. Wooldridge, 2002, chapter
to deal with this problem, adopted in
Theodore
column
regression of
the standard fixed effect estimation
short time dimension
for
of country
7 from models
(2)
it ,
is
and shows no evidence of a positive
now
Ada-i
more negative
to obtain
estimates,
estimated to be -0.018 (standard error
effect of
=
education on democracy.
Although the instrumental variable estimator of Anderson and Hsiao (1982) leads to
consistent estimates,
correlation in
it is
u it not only
,
not
efficient, since,
dj t -2j bvrt all
under the assumption of no further
further lags of d it are uncorrelated with
can also be used as additional instruments. Manuel Arellano and Stephen R.
develop a Generalized Method-of- Moments Estimator
conditions.
in
column
We use this
3,
GMM estimator in column 4.
-0.017 (standard error
=
0.022).
(GMM)
using
The estimate
The presence
all
for
7
Bond
of these
is
serial
Au#, and
(1991)
moment
similar to that
of multiple instruments in the
GMM procedure allows us to investigate whether the assumption of no serial correlation
in
u
it
can be rejected and also to
test for overidentifying restrictions.
The AH(2)
the Hansen J test, reported at the bottom of this column, indicate that there
serial correlation
Columns
5
and the overidentifying
and 6 repeat the
But now the AR(2)
test
test
and
no further
restrictions are not rejected.
fixed effects
OLS and
GMM
regressions
2 and 4 using five-year averages of democracy, with very similar (and
estimates.
is
shows residual
serial correlation
from columns
more negative)
and the Hansen J
because of the additional
test rejects the overidentifying restrictions
serial correlation
introduced by averaging. This motivates our focus on the dataset using the observations
every
fifth
year rather than averaging the five-yearly observations.
Columns 7 and
results
(though
8 repeat these regressions with ten-year data, again with very similar
now
the estimate of 7
is
positive
and
with the fixed
insignificant
effects
OLS).
These results therefore cast considerable doubt on the causal
education on
effect of
democracy.
Robustness
4
Table 2 documents the robustness of these findings to alternative samples.
the fixed effects
OLS and
One concern
is
We only report
GMM regressions to save space.
whether the presence of
socialist countries,
where the persistence and
the later collapse of non-democratic regimes might have very different causes,
these results. This
may be
driving
is
a valid concern, since some of these former socialist countries
already had high levels of education in the 1980s and did not experience any marked
increase in education during or immediately prior to transition.
Columns
1
and
2 exclude
former and current socialist countries, and show that these countries had no effect on
The
the results.
estimates of
estimate with fixed effects
7 are very
OLS
is
similar to those in Table
-0.003 (standard error
=
1.
For example, the
and with
.019)
GMM -0.015
(standard error =0.022).
Another concern
is
whether results are a due to the unstable
political
dynamics
sub-Saharan Africa, or potentially driven by Muslim countries, which have lower
schooling and have been slower to democratize.
is
and
more than 40%
of the
Muslim. The results are very similar to the baseline in both cases.
Finally, another concern
House data.
levels of
Columns 3 and 4 exclude Sub-Saharan
African countries and columns 5 and 6 exclude countries where
population
in
To show
is
whether the
results are
due to the use of the Freedom
that our results are robust across different datasets, columns 7
8 report our basic fixed effects
OLS and
and columns 9 and 10 report our basic
GMM regressions using the Polity dataset,
OLS and GMM regressions using the
fixed effects
Boix-Rosato dataset (which extends Przeworski et
al.'s
data to the present). The results
are again similar.
In order to address the concern that the omission of certain variables are causing our
result,
Table 3 shows that controlling for a range of important covariates leaves these
basic pattern unchanged.
GMM
OLS and
effects
We focus
on the Freedom House data and report the basic fixed
Columns
regressions.
1
and
2 control for the age structure
population by including the fractions of the population in
five different
and
age ranges, the
median age of the population, and the logarithm of total population. 2 These variables are
correlated with average educational attainment of the population,
effect,
making
it
find that the age structure variables are jointly significant at
GMM,
not
and might have a direct
impossible for us to identify the influence of education on democracy.
while log population
is
10%
We
using fixed effects but
not significant. The effect of education on democracy
continues to be highly insignificant in both cases.
Columns 3 and 4
and has no
effect
add income per
and
control for the investment share of
on the sign or magnitude of the education
capita.
interestingly,
Education
is still
income per capita
is
itself is insignificant
variable.
Columns
5
and
is
with a negative
coefficient.
3
columns 7 and 8 control
6
The
the other basic tenet of the moderniza-
therefore also not robust to controlling for country fixed effects.
investigate this issue in greater detail in Acemoglu, et
Finally,
insignificant
is itself
insignificant (and has a negative coefficient),
causal effect of income on democracy, which
tion hypothesis,
GDP, which
al.
We
(2004).
for all of these covariates simultaneously,
again with
similar results.
From Education
5
The
recent paper by Glaeser, et
to Institutions?
al.
on institutions. They substantiate
(2),
(2004) argues that there
this
is
a causal
Why
Polity, the
in
effect of
education on
replicate their results, which use the constraint
on executive
are their results different from ours?
In Table 4 panel A,
from
education
by reporting regressions similar to our model
but with very different results, in particular showing a positive
democracy.
effect of
we
autocracy score from Polity, the democracy score from Polity, and the
autocracy score from Przeworski et
al.
4
These columns exactly match their
regressions,
but are different from our corresponding regressions, because they do not include time
effects,
2
the
/z 's
The ranges
t
in equations (1)
and
(2).
In the absence of time effects, the parameter
and 60 and above. Age ranges and median age
and population is from World Bank (2002).
3
Investment Share of GDP and GDP per Capita are from the Penn World Tables dataset in Alan
Heston, Robert Summers, and Bettina Aten (2002).
4
The only difference from their results is that, instead of using autocracy scores like they do, we
transform everything to democracy, so that all coefficients have the same sign.
Note that for these
results, the index of institutions is not normalized between
and 1.
are
to 15, 15 to 30, 30 to 45, 45 to 60,
are from United Nations Population Division (2003)
7
is
identified
from the over-time variation,
in this context, the world-level increase in
education and democracy. This clearly does not correspond to any causal
Panels
time
B and C
report estimates with
and without income per
our basic specifications. In
effects as in
all
cases education
the incorrect sign as in our basic results. Moreover, in
1%
effects are jointly significant at
10% (and
or less,
interestingly, in that case, as
without time
and
all
in that
effect.
capita, but including
is
insignificant
and has
columns except one, the time
one case they are significant at
column 4 shows, education
is
insignificant even
effects).
The evidence
in
Table 4 therefore shows that there seems to be no
on democracy or on other
effect of
education
political institutions.
Concluding Remarks
6
A common
view clearly articulated by the modernization theory claims that high
levels
of schooling are both a prerequisite for democracy and a major cause of democratization.
The evidence
in favor of this
view
sectional regressions. This paper
fixed effects
is
largely based on cross-sectional or pooled cross-
documents that
and exploiting the within-country
this evidence is not robust to including
variation. This strongly suggests that the
cross-sectional relationship
between education and democracy
influencing both education
and democracy rather than a causal
is
driven by omitted factors
relationship.
This evidence poses two important questions:
1.
Is
there no long-run causal relationship between education and democracy?
It is
We
have
important to emphasize that our paper does not answer this question.
exploited the five
and ten yearly variation
changes in education have very long run
in the
affects,
postwar
not manifest themselves in the shorter timeframe that
2.
What
fixed effects?
tion of economic
and
We conjecture that these
political
development ("the
our companion paper, Acemoglu, et
this conjecture.
We document that
al.
(2004)
It is
we have looked
are the omitted factors influencing both education
by the country
era.
possible that
say over 50 or 100 years, that do
at.
and democracy, captured
are related to the joint evolu-
historical
development paths"). In
we provide evidence
consistent with
the fixed effects for the former European colonies
are very highly correlated with the historical, potentially-exogenous determinants of
institutional
development
in this sample, in particular, the mortality rates faced
by
the European settlers and the density of the indigenous populations (see, Acemoglu,
Johnson and Robinson, 2001 and 2002) as well as early experiences with democracy.
Bibliography
7
Acemoglu, Daron, Simon Johnson and James A. Robinson (2001) "The
Origins of Comparative Development:
An
Colonial
Empirical Investigation," American Economic
Review, December, 91, 1369-1401.
Acemoglu, Daron, Simon Johnson and James A. Robinson (2002)
of Fortune:
Geography and
Institutions in the
Making
of the
"Reversal
Modern World Income
Distribution," Quarterly Journal of Economics, 118, 1231-1294.
Acemoglu, Daron, Simon Johnson, James A. Robinson and Pierre Yared
(2004) "Income and Democracy," mimeo.
Acemoglu, Daron and James A. Robinson (2004) Economic
torship
Anderson, Theodore
of
Origins of Dicta-
and Democracy, Cambridge University Press, forthcoming.
W. and Cheng Hsiao
Dynamic Models using Panel Data," Journal
Arellano,
(1982) "Formulation and Estimation
of Econometrics, 18, 67-82.
Manuel and Stephen R. Bond (1991) "Some
Specification tests for
Panel Data: Monte Carlo Evidence and an Application to Employment Equations," Review of Economic Studies, 58, 277-298.
Barro, Robert
J.
(1999) "The Determinants of Democracy," Journal of
Political
Economy, 107, S158-S183.
Barro, Robert J. and
Jong-Wha Lee
Attainment: Updates and Implications,"
(2000) "International Data on Educational
CID Working Paper #42.
Boix, Carles and Sebastian Rosato (2001) "A Complete Data Set of
cal
Politi-
Regimes, 1800-1999," Unpublished, Department of Political Science, University of
Chicago.
Bollen,
Kenneth A. (1990)
"Political
Democracy: Conceptual and Measurement
Traps," Studies in Comparative International Development, 25, 7-24.
Bollen,
Kenneth A. (2001) "Cross-National Indicators of Liberal Democracy, 1950file]. 2nd ICPSR version. Chapel Hill, NC: University of North Carolina
1990" [Computer
[producer], 1998. Inter-university
Consortium
for Political
and
Social Research [distribu-
tor].
Central Intelligence Agency (2004) CIA World Eactbook, Website and Book.
Washington, DC.
Dewey, John (1916) Democracy and
Education,
New
York;
The Macmillan Com-
pany.
Freedom House (2004) Freedom
in the World, Website
10
and Book. Washington,
DC.
Glaeser,
Edward
L.,
Rafael La Porta, Florencio Lopez-de-Silanes, and An-
drei Shleifer (2004) "Do Institutions Cause Growth?"
NBER Working
Paper #10568.
Heston, Alan, Robert Summers, and Bettina Aten (2002) Penn World
Version
Tables
Center for International Comparisons at the University of Pennsylvania
6.1.
(CICUP).
La Porta,
W. Vishny
Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer
and Robert
(1999) "The Quality of Government," Journal of Law Economics and Or-
ganization, 15, 222-279.
Lipset,
Seymour Martin (1959) "Some
Development and
Przeworski,
Political Legitimacy,"
Social Requisites of
American
Democracy: Economic
Political Science Review, 53, 69-105.
Adam, Michael Alvarez, Jose A. Cheibub and Fernando Limongi
(2000) Democracy and Development: Political Institutions and material well-being in the
world, 1950-1990,
Cambridge University Press,
New York NY.
United Nations Population Division (2003) World Population
2002 Revision, Disk.
New
Wooldridge, Jeffery M. (2002) Econometric Analysis
Data,
MIT
of Cross Section
The
and Panel
Press; Cambridge.
World Bank (2002) World Development Indicators, CD-ROM and
ton,
Prospects:
York.
DC.
11
Book. Washing-
<N
CD
(D
_l
j3
O
(^
L—
CD
&
00
00^
<
"o
^
o
o
o
CO
CDm-
o
to
CD
<D
>
CD
_^ O)
CO
l_
d)
>
<
N K
bi>
D)
_C
.c
"*
>.
u
>
o
Ci-
w
o
o
s
T3 g^
tlH
E
£
>§
& g
-- c
c
g
S
J3
-a
o
-
I
8"
9'
xapu| sii|6!y leojinod
i
V
i
Z
esnoH wopsejj
r
>
u
n i
9cJP
i
Si
E 2
ti^
I
i
<*
I
•a
CM
i
°^
K
o
§
1
>
1
>
CD
LO
a>
a
6
o>
CO
CD
*—
o
D)
O
o
o
c
o
o
w
co
(D
o
^
"q
il
?
o
w
1—
CO
CD
>
•O ON
CD
D)
cm
ro
1
CD
>
<
CD
o
CD
CO
c
o
o
E
CO
O
a)
Q
-
-
1
I
I
9"
xepu| sjL)6iy leojjipd
I
o
I
9esnoH ujopeejj
lui
a6ueqo
§
£
'
1
ss
s
o
9 o
<=
m
s
in ^t
o
o
PQ >5
o
s
On
£
-a
1
£
_ o
S
*
o
to o ^ o
—'
o o
e
2
t-~
Ti-
'•o
—
'
--.
°
b
O
—
J=
SCI
c
a
-o
—c
t~-
-*
13
g
2
CQ
o
"2
S o E
2
1 o
w
SS
c
*
•*'
JJ
a
w
—
.3
e
S
™
>-»
E
c
•£
c s 1
ft3
>
c
-
a. "E.
Q
|
g
s
"a.
c
a.
1
C
s
s
—
"2
-a
-a -a
|
73
q- "S
-a S
o o-
.£3
o o
o
p-
•g -3
o
1
3
C
=
| |
~
a
_3
>
i
.5
o
—
§ i
JL
"
c
i
S
<_
do
u
W
[i
„
u
H
uS
W
c
H
u
1
H
c
o
o
-3
n
1
B? -O
o
K < O U
_s
.c
U
££l
13
>
3
a/
T3
>
of o XI
+-"
iJ
*°
C
>
o-
c
J2
— b
i=
.1
1
X
5f "S
a
£
t3
ci
£
'5b
_
o
£ s
vo
I
-s -s
£
— 3
a.
-a
.5
X
o\ c> o^ ^n
no
oo
—
o
o o ^ —
aO
5
<
2
T3
O
03
T3
r-
£
§
§
J2
|
§
T
on
N
—
bO
C
3
o
o
£
u
< a
°
^
S
Q
a
u
ID
n.
o
E
o
U
_rt
O
<
5
<^ <n
s °
ro c> o
O
O O ^ ^
u
in
a
a
Ic
'C
<
-M
3 S
"--
M
x:
o
OO ID <N
°°
— O
_g
s
.2
jo
id .£
Os
M
3
^
o
c o
fi
§ o
n Q £
-°
3 3 22
E a u
u
-5
«
W
'w
4-
Ic §
.?
-a
no
"I
>
§
d "3
Q.
E
ed
00
u
$
o
+2
Q
T3
.0
<
-g
9
ID
u
J3
00
e3
^
J> .a
a'
I
S
O
=*
—
P o
o o
«n
o
'
.3
s e I
=
< 5
£
—
e
oj
o
U
«
Q. 03
E
3
3
03
43
'5
a
a
3
s
—w
a. t-
a
5
o
%
1
ca
NO °°
u
a
JS
-
o^
=>
S
.S
ID
£J
a-
00 -O
Hh "O
cd
§
—
"
s
a
s
o
as"o
>
o
o
cs
s-
~
3 -3 "S
g > i
>>
o
w
U
c^
>
D. J;
OT
.2
3
_u
-O
.2
i
•
CO
^
o
o
o
rJ
if
.2
-*"
o £> a
s
Q.
E o
3
GO
*
no I* JL
u u £
3 -O Q
o
„
o
o
o
(N
ON
5
-g no
3
S
c
g
O
_^
°°
ID °°
O
no
tf3
<
g
§£
-S 23 -a
I a
CO
1 fs
«
r
£— 1 o
c a -a o
_2
no
o\
u
c
00
*,
JJ u.
CL
£
CO
3
o
03 -g
U
1
O
S
cd r- rn
nD
O O
2 to
o "
S
o>
ng
CQ
rt
+rf
c
H
f~
•n
Q 2
-a
.3 vo
-f
c "2 c
S
u ™
a «*
tn
.
•"
t
3
u
"-
1
rt
a.
T
'
K
1
M I
g
'
„
t3 _r
3
>-
§
.2
" " o c
9
g S I
i-S'S
u
O g i> 1
1/1
no
**
d
O
H ^
H
.32
a
a o "
<OUtti
Bi
-a
ac =
2 S
-k
•-
sp "M.
-°
o
_
"5
is
I
I
5
z
£
£
.5
s
S
g
j<
on
on
~
H,
C-
X
o
o
p|S
-a
3
U
o
„
S
1
1
J2
^ o\
o d
O
oo C!
>n
c
»
•s
oo
T3
5
-5
-a
o
§
1
i-:5
in
o —
O
a
- 3
§ s
9 ©
CO
'?
,
-- y: in .3
m Os
-J
si
H ^
JS
32
"3 8
O
J3
u £ —
o
GU
o
T3
VO
O
"^
1}
—
cd
'
u
'd .2
S
Jj ^o
T-
S
-a
§
S
P o
2S
J3
Ml
so
«
00
o
§T
i,-,"
JL
«
g.|
"7
"
o &c
°
JS
°
•3
5
<
u o u
"O
§
c3
2:
•gill
00
T3
iS
^
^ ©
r--
00
*0 o^
O
"3
—
J2
2 *
"
—
OV
IN OO
d d
o
(Nl
>/-)
^o
u
«•>
o
'S
-2
S
—
o
—
'i
-e
I
'5
5"
J?
-a
O
o
B
is
— x
c is
2 -3
o
-S
1
A 3
£*
•O
«
"
d 1
Q
a
00
O
J
H
PJ
c
U
s
§
H
;
g
'C
=
E<0
Pd J3
3
O
U
CS
§
of
1
at.
5"
=
T3 t3
s r
Table 4
Fixed Effects Results: Education, Democracy, and Political Institutions
Glaeser
(2004) Sample, 1965-2000
et al.
5-year data
Fixed Effects
OLS
Fixed Effects
Panel A: No Time Effects
Institutions
,_)
GDP per Capita
,.,
R-squared
Education
Log
GDP per Capita
Time
,_,
Effects F-test
R-squared
-0.547
-0.515
-0.864
(0.072)
(0.068)
(0.065)
(0.103)
Autocracy
0.498
0.909
0.700
0.096
(0.119)
(0.179)
(0.180)
(0.071)
0.038
-0.508
0.292
0.267
(0.403)
(0.630)
(0.606)
(0.202)
0.30
0.47
0.32
is
Change
in Institutions
A utocracy
Exec.
Autocracy
Democracy
-0.618
-0.616
-0.580
-0.897
(0.073)
(0.071)
(0.067)
(0.106)
-0.163
-0.318
-0.432
-0.137
(0.192)
(0.267)
(0.298)
(0.126)
0.168
-0.317
0.477
0.292
(0.360)
(0.550)
(0.561)
(0.192)
[0.01]
[0.00]
[0.00]
[0.08]
0.37
0.50
0.40
0.39
Dependent Variable
,.]
in Institutions
-0.572
Panel C: Including Time Effects
Institutions
Change
Democracy
Dependent Variable
t_,
is
is
Change
in Institutions
Exec.
Autocracy
Democracy
-0.617
-0.615
-0.579
-0.891
(0.073)
(0.071)
(0.068)
(0.107)
Autocracy
-0.125
-0.389
-0.324
-0.088
(0.182)
(0.229)
(0.289)
(0.125)
[0.01]
[0.00]
[0.00]
[0.07]
R-squared
0.39
0.40
0.37
0.49
Observations
499
499
499
Education,.,
Time
Effects F-test
Fixed effects
OLS
regressions in
dummies
are included in panels
column
is
1
change
in
all
B and
column
Autocracy Index from Przeworski
where
in
parentheses. Year
C, and the time effects F-test gives the p-value for their joint significance. Dependent variable in
3
is
change
et al. (2000).
the start date of the panel refers to the
and sources.
349
columns with country dummies and robust standard errors clustered by country
Constraint on the Executive from Polity. Dependent variable in column 2
Polity. Dependent variable in
OLS
(4)
A utocracy
0.33
,.]
Fixed Effects
Exec.
Panel B: Including Time Effects
Institutions
OLS
Fixed Effects
(3)
Dependent Variable
Education ,.,
Log
OLS
(2)
(1)
in
Democracy Index from
Base sample
in all
dependent variable
columns
(i.e.,
Polity.
is
is
change
Dependent variable
in
in
negative Autocracy Index from
column 4
is
change
in negative
an unbalanced panel, 1965-2000, with data at 5-year intervals
t=1965, so
t-
1=1 960). See Glaeser et
al.
(2004) for data definitions
1#
Date Due
MIT LIBRARIES
3 9080 02618 2755