Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/incomedemocracyOOacem
OBWBY
HB31
.M415
m.05'Oi
Massachusetts Institute of Technology
Department of Economics
Working Paper Series
'
INCOME AND DEMOCRACY?
Daron Acemoglu, Simon Johnson,
James A. Robinson,
Pierre Yared
Working Paper 05-05
February 2005
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://ssrn.com/abstract=675066
MAR
1
2005
irlBRARIES
Income and Democracy*
Daron Acemoglu^
Simon Johnson^
James A. Robinson^
Pierre Yared^
This Version: February 2005.
Abstract
We
revisit
one of the central empirical findings of the pohtical economy
litera-
ture that higher hicome per capita causes democracy. Existing studies estabUsh a
strong cross-country correlation between income and democracy, but do not
cally control for factors that simultaneously affect
both variables.
We
tjqai-
show that
by including country fixed effects removes the statistibetween income per capita and various measures of democracy. We
also present instrumental-variables estimates using two different strategies. These
estimates also show no causal effect of income on democracy. Furthermore, we reccontrolling for such factors
cal association
oncile the positive cross-country correlation
between income and democracy with
the absence of a causal effect of income on democracy by showing that the long-run
evolution of income and democracy
this,
is
related to historical factors. Consistent with
the positive correlation between income and democracy disappears, even with-
out fixed
effects,
when we
control for the historical determinants of economic
and
pohtical development in a sample of former European colonies.
Keywords: democracy, economic
JEL
*
growth, institutions, political development.
Classification: P16, OlO.
We thank David Autor,
Robert Barro, Jason Seawright, Sebastidn Mazzuca, and seminar participants
de Colombia, Boston University, the Canadian Institute for Advanced Research, the CEPR annual conference on transition economics in Hanoi, MIT, and Harvard for comments.
^Department of Economics, Massachusetts Institute of Technology, 50 Memorial Drive, MA02139.
e-mail: daron@mit.edu.
at the
'^
Banco de
la Repiiblica
Sloan School of Management and International Monetary Fund, Massachusetts Institute of Technol-
Memorial Drive, MA02139. e-maU: sjohnson@mit.edu, sjohnson@imf.org.
^Department of Government, Harvard University, Littauer, 1875 Cambridge St., Cambridge MA02138;
ogy, 50
e-mail: jrobinson@gov.harvard.edu.
^Department of E)conomics, Massachusetts
e-mail: yared@mit.edu.
Institute of Technology, 50
Memorial Drive, MA02139.
Introduction
1
One
of the
most notable empirical
regularities in political
between income per capita and democracy. Today
many
while
all
economy
OECD
Saharan Africa and Southeast Asia. This positive relationship
Most
is
example sub-
not only confined to a
countries were nondemocratic before the
process took off at the begimiing of the 19th century.
modern growth
Democratization came together
growth. Barro (1999, S160), for example, summarizes the findings from his detailed
study
in
the relationship
of the nondemocracies are in the poor parts of the world, for
cross-country comparison.
•with
is
countries are democratic,
as:
"increases in various measures of the standard of hving forecast a gradual rise
democracy. In contrast, democracies that arise without prior economic development
tend not to
This statistical association between income and democracy
influential
is
the cornerstone of the
modernization theory, which sees a direct causal link between economic growth
and democracy. According to this
theory, economic
growth engenders "a culture of democ-
racy" and provides the foundations for democratic political institutions.
clearly articulated in Lipset (1959),
relatively
who argued
This thesis
is
that "only in a wealthy society in which
few citizens lived in real poverty could a situation exist in which the mass of the
population could intelligently participate
m
politics
and could develop the
self-restraint
necessary to avoid succvunbing to the appeals of irresponsible demagogues"
is
...
last."^
also reproduced in
all
the major works on democracy
(e.g.,
(p.
75).
It
Dahl, 1971, Huntington,
1991).
In this paper,
Our
we
starting point
re\'isit
establish causation.
the relationsliip between income per capita and democracy.
that existing work, based on cross-country relationships, does not
is
First, there is the issue of reverse causality;
causes income rather than the other
the potential for omitted variable bias.
of the political regime
first
strategy
is
Some
and the potential
We utihze two strategies
for
other factor
may
determine both the nature
income on democracy. Our
to control for country-specific factors affecting
effects.
is
economic growth.
to investigate the causal effect of
racy by including coimtry fixed
perhaps democracy
way round. Second, and more important, there
While
both income and democ-
fixed effect regressions are not a
panacea
against omitted variable biases,^ they are weU-suited to the investigation of the relation^ Also see, among others,
Lipset (1959), Londregan and Poole (1996), Przeworski and Limongi (1997),
Barro (1997), Przeworski, Alvarez, Cheibub, and Limongi (2000), and Papaioannou and Siourounis
(2004).
^
effects would not help inference if there are time- varying omitted factors affecting the dependent
and correlated with the right-hand side variables (see the discussion below). They may also make
Fixed
variable
ship between income
and democracy. Major sources of potential bias
democracy on income per capita are country-specific,
political
and economic development.
If
a regression of
these omitted characteristics are, to a
imation, time-invariant, the inclusion of fixed effects will remove
bias.
in
historical factors influencing
them and
first
both
approx-
this source of
Consider, for example, the comparison of the United States and Colombia.
United States
is
as well as the existing empirical strategies in the literature
effects,
would suggest that there
idea of fixed effects
variation";
i.e.,
The
both richer and more democratic, so a simple cross-country comparison,
is
to
is
move beyond
this
relative to the
for fixed
comparison and investigate the "within-country
whether as Colombia becomes relatively
more democratic
which do not control
a relationship between democracy and income. The
United States.
richer,
it
also tends to
become
In addition to improving inference on
the causal effect of income on democracy, this approach
is
also
more
closely related to
modernization theory as articulated by Lipset (1959), which claims that countries should
become more democratic
as they
become
richer,
not simply that rich countries should be
more democratic.
Our main
finding from this strategy
positive relationship
appears.
is
that once fixed effects are introduced, the
between income per capita and various measures of democracy
measures of democracy, the Freedom House and Polity scores (see below
for
each country between 1970 and 1995 against the change in
same
dis-
Figures 1 and 2 show this diagrammatically by plotting changes in our two
period.
GDP
There appears to be no relationship between changes
for
data
details),
per capita over the
in
income per capita
and democracy.
This basic finding holds with various indicators for democracy, with different econometric specifications and estimation techniques, in different subsamples, and
the inclusion of additional covariates.
standard errors. In
many
of income on democracy,
is
robust to
Moreover, these results are not driven by large
cases, two-standard error
bands include only very small
and often exclude the OLS estimates. These
shed considerable doubt on the claim that there
is
a strong causal
effects
results therefore
effect of
income on
democracy.^
While the
fixed effects estimation
is
useful in
removing the influence of long-run deter-
problems of measurement error worse because they remove a significant portion of the variation in the
right-hand side variables. Consequently, fixed effects are certainly no substitute for using an instrumentalvariables approach with a valid instrument.
'It remains true that over time there is a general tendency towards greater incomes and greater
democracy across the world. In our regressions, time effects capture these general (world-level) tendencies.
Our estimates suggest that these world-level movements in democracy are unlikely to be driven by the
causal effect of income on democracy.
minants of both democracy and income,
of
An
income on democracy.
it
does not necessarily estimate the causal
would be a superior approach, but
it is
difficult to find
could not affect democracy through other channels.^
We
gressions.
the
for
first
second
instrument
is
come
to use
IV
re-
to use past savings
The argument
effect
on democracy. The second instrument, which we believe
and constructs predicted
in-
each country using a trade-share-weighted average income of other countries.
for
show that
capita,
is
that variations in past savings rates affect income per capita,
of independent interest, creates a matrix of trade shares,
We
first is
income that
for
strategy
to use changes in the incomes of trading partners.
is
but should have no direct
is
vahd instruments
Our second
experiment with two potential instruments. The
rates, while the
effect
instrumental- variables (IV) strategy with a valid instrument
this predicted
and argue that
it
income has considerable explanatory power
for
income per
should have no direct effect on democracy. Both IV strategies
confirm our basic findings and show no evidence of a causal effect of income on democracy.
We
recognize that neither instrimaent
is
perfect, since there are
scenarios in which our exclusion restrictions could be violated
(e.g.,
some reasonable
saving rates might
be correlated with future anticipated regime changes; or democracy scores of a country's
trading partners, which are correlated with their income levels, might have a direct effect
on
its
democracy). To alleviate concerns about the validity of the instruments, we show
that the most likely sources of correlation between our instruments and the error term in
the second stage are not present.
These
results naturally raise the following important question:
the cross-sectional correlation between income and democracy?
democratic today?
democracy, but
is,
it
One
possible explanation
is
that there
is
what
Why
a causal
is
the source of
are rich countries
effect of
income on
works at much longer horizons than the existing literature posited, that
over 50 or even 100 years rather than 10 or 20 years. Another hypothesis, suggested by
approaches that emphasize the importance of historical factors in long-run development,^
is
that the cross-sectional relationship reflects the persistent influence of these historical
factors.
political
Put
differently, events
during certain crucial junctures impact the economic and
"development path" of a
influences
on economic and
society, leading to persistent,
political
within-correlation between income
*
as
A recent creative attempt is by Miguel,
an instrument
for
income
though not permanent,
outcomes. Both of these hypotheses suggest that the
and democracy should be stronger when we look
Satyanath and Sergenti (2004),
impact of income on
in Africa for investigating the
at
who use the weather conditions
civil
wars. Unfortunately,
weather conditions are only a good instrument for relatively short-rim changes in income, thus not
necessarily ideal to study the relationship between income and democracy.
^See, among others, North and Thomas (1973), North (1981), Jones (1981), Engerman and Sokoloff
(1997), Acemoglu, Johnson and Robinson (2001, 2002).
longer horizons.
We
for
an
by looking at the relationship between income and
investigate this possibihty
democracy over the past 160
effect
years,
and over the past 500
of income on democracy
in
We
years.
over the past 500 years there seems to be a very strong correlation.
is
that this pattern
is
find little evidence
samples that span 100 or 160 years. In contrast,
Our
interpretation
consistent with the second hypothesis, because the 500 years in
question spans the period of divergence of national development paths
of constitutional monarchies, the rise of the
modern nation
(e.g.
,
the emergence
and
state, industrialization,
the colonial experience).
In addition,
we
also provide direct evidence consistent with the second hypothesis
looking at the sample of former European colonies. This sample
us to exploit the quasi-natural experiment provided by the colonization of
societies
by European powers
after 1492,
economic and
led to significant divergence in the
eties (see, e.g.,
We
1997).
where differences
political
enables
many
diverse
in the colonization experience
development paths of these
soci-
Acemoglu, Johnson and Robinson, 2001, 2002, and Engerman and Sokoloff,
document that
in this sample, the fixed effects in the
are closely linked to the potential determinants of
ticular,
by
it
useful since
is
democracy regressions
European colonization strategy
(in par-
the density of the indigenous population at the time of colonization and potential
mortality rates of European settlers), date of independence and measures of institutions
independence
in the early
era.
The
positive correlation between
disappears, even without fixed effects,
when we
income and democracy
control for these historical determinants.
This evidence further supports the second hypothesis.
Finally,
in the
we
we document
that there are
some income-related determinants
of
democracy
postwar sample. In particular, contrary to the implications of modernization theory,
find that economic crises lead to democracy.
by the
fact that dictatorships are
than democracies are
more
likely to revert
The paper proceeds
as follows.
discuss our econometric approach
We show that
this result is driven entirely
likely to collapse in the face of
economic
crises
back to dictatorship.^
In Section 2
we
describe the data.
and present the basic
results. Section
results. Section 5 discusses potential interpretations for these results.
interpretations. Section 6 investigates the longer-run relationship
In Section 3
we
4 presents our IV
Motivated by these
between income and
democracy, and Section 7 looks at the historical determinants of economic and political
*^In
passing,
we
also
show that income per capita does not appear to have a causal
effect
when we
separately at transitions to and from democracy, contrary to the findings in Przeworski et
Since
work.
we do not have space
in this paper,
we
leave a
more detailed
al.
look
(2000).
investigation of transitions to future
development in the sample of former European
crises
colonies. Section 8 looks at the effect of
on democracy. Section 9 concludes. The Appendix contains some additional
and further information on the construction of the instruments used
Data and Descriptive
2
results
in Section 4.
Statistics
We follow much of the existing research in this area in adopting a Schumpeterian definition
based on a number of institutional conditions.' Our
A
the Freedom House Political Rights Index.
is
political rights
come
closest to the ideals suggested
with whether there are free and
fair elections,
and main measure of democracy
first
country receives the highest score
by a
if
checklist of questions, beginning
whether those who 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
government or can participate
in the
self-
government tlu'ough informal consensus.^ Following
Barro (1999), we supplement this index with the related variable from Bollen (1990, 2001)
for 1950, 1955, 1960,
they
lie
and 1965. As
and
between
1,
The Freedom House
with
as
The
when augmented with
Polity
a check on our main measure, we
IV
dataset,
to 10
and
is
(see
Bollen's data, only enables us
on the other hand, provides informa-
Both
for pre- 1950 events
also look at the other widely-used
composite Polity index, which
Autocracy indices
from
Barro (1999), we transform both indices so that
countries since independence starting in 1800.
all
racy, the
in
corresponding to the most democratic set of institutions.
index, even
to look at the postwar era.
tion for
1
is
and
measure of democ-
the difference between Polity's Democracy and
Marshall and Jaggers, 2004).
The
Polity
Democracy Index ranges
derived from coding the competitiveness of political participation, the
openness and competitiveness of executive recruitment and constraints on the chief executive.
way
The
Polity Autocracy Index also ranges from
to the
to 10
and
is
constructed in a similar
democracy score based on scoring countries according to competitiveness of
political participation, the regulation of participation, the
of executive recruitment and constraints
on the
openness and competitiveness
chief executive.
To
facilitate
compari-
''Schumpeter (1950, p. 250) argued that democracy was: "the institutional arrangement for arriving
at pohtical decisions in which individuals acquire the power to decide by means of a competitive struggle
for the people's vote."
''The
main
checklist includes 3 questions
on the
electoral process, 4 questions
on the extent of political
pluralism and participation, and 3 questions on the functioning of government. For each checkhst question,
to 4 points are added, depending
on the comparative
rights
and
liberties present (0 represents
4 represents the most) and these scores are combined to form the index.
http://www.freedomhouse.org/research/freeworld/2003/methodology.htm
the
least,
See FVeedom House (2004),
.
son with the Freedom House score, we also normahze the composite Pohty index to he
and
between
1.
Using the R-eedom House and the Polity data, we construct
and annual
We
panels.
five-yearly, ten-yearly,
For the five-year panels, we take the observation every
fifth year.
prefer this procedure to averaging the five-yearly data, since averaging introduces
additional serial correlation,
making inference and estimation more
12). For the ten- yearly panels,
we
difficult (see
footnote
take the observation every tenth year for similar reasons.
For the Freedom House data which begiirs in 1972, we follow Barro (1999) and assign the
1972 score to 1970 for the purpose of the five-year and ten-year regressions.
The
GDP
per capita (in
PPP) and
savings rate data for the postwar period are from
Heston, Summers, and Atten (2002), and
longer sample are from
Maddison
Other variables we use
1
Appendix Table Al
for
and sources)
It
European
colonies, the
shows that there
is
sample we focus on
less
some
for
of the historical
significant variation in all the variables for
sample and the former colonies sample. Countries
somewhat
is
contains descriptive statistics for the key vai'iables both for the whole world
for former
regressions.
entire
trade- weighted world income instrument
in the analysis are discussed later (see also
detailed data definitions
and
The
data from International Monetary Fund Direction of Trade Statistics (2005).
built using
Table
(2003).
GDP per capita (in constant 1990 dollars) for the
in the
both the
former colonies sample are
democratic and substantially (about 30 percent) poorer than the average
country in the whole sample.
Main Results
3
Basic Specifications and Interpretation
3.1
Our
basic regression
model
dit
where da
is
=
is:
adu-i
+ iVit-i + x';_i/3
the democracy score of country
on the right hand side
is
in period
t.
H-
(^^
+ uu,
(i.e.,
The lagged
of log income per capita.
The main
The parameter 7
variable of interest
is
denote a
full set
yu-i, the lagged value
therefore measures whether income has an
on democracy. All other potential covariates are included
5,'s
value of this variable
the tendency of the democracy score to return to some
equilibrium value for the country).
addition, the
(1)
included to capture persistence in democracy and also potentially
mean-reverting dynamics
effect
i
-I- /ij
of country
dummies and the
in the vector Xu-i/Xj's
denote a
full set
In
of
time
of
effects, wfiich
all
Un
comitries.
for all
i
and
common
capture
is
The sample period
t.
(common
sliocks to
an error term, capturing
is
trends
in) ttie
democracy score
other omitted factors, with
all
E {u^t) =
1960-2000 and time periods correspond to five-year
interv^als.
The standard
which
is
regression in the hterature, for example, Barro (1999),
identical to (1) except for the omission of the fixed effects,
dummies capture any
these country
is
pooled OLS,
In our framework,
Si^s.
time-invariant country characteristic that affect the
equiUbriiun democracy level.
As
is
well
known, when the true model
yit-i or Xit_i, then pooled
if
OLS
either Cov(yit_i, 5i -f uu) 7^
will
be inconsistent (where
is
given by (1) and the
or Cov(a;^j_j, 5i
xj^^-^
+ uu)
7^
for
some
component
refers to the jth.
covariances refer the population covariances). In contrast, even
nonzero, the fixed effects estimator will be consistent
for all j (as
^ 00, see below).
T
J^'s
axe correlated wdth
More
estimates are biased and inconsistent.
if
j,
the
specifically,
OLS
estimator
when these covariances
Cov(yjt_i, Ujt) =Cov(a:^^_;^, Uitj
This structure of correlation
and
of the vector Xit_i,
are
=
particularly relevant in
is
the context of the relationship between income and democracy because of the possibihty
of underlying political
and
both equihbrium poHtical institutions and
social forces shaping
the potential for economic growth.
Nevertheless, there should be no presumption that
fixed effects regressions will necessarily estimate the causal effect of
To
illustrate this point
and as a preparation
income on democracy.
for the discussion in Section 5, consider
a
simplified version of (1), without the lagged dependent variable and the other covariates
and with contemporaneous income per capita on the
another error component,
7]fj,
which admits a imit
where
Moreover suppose that the
right
hand
side.
Let us also add
root, such that:
d^t
=
lyu
77f,
=
rj^.^+vi.
+ S'l + vi + ui,
statistical process for
(2)
income per capita
also admits a unit
root,
where
While
Jj
countries,
and
rjf^
^
on democracy.
=
s^
rjl
=
ril_^+v\.
+ v^t + y^u
(3)
correspond to fixed difierences in levels of democracy and income across
and
across countries.
y^t
T]\^
As
capture
fax;tors affecting
before, the parameter
the evolution of democracy and income
7 represents the
Denote the variance of v\ by
a^^y
causal effect of income
and of u^ by
al^y.
Assume
that
=
CoY{ui,ul^,^) =Cov{vl,vl_^,^) =Cov(7jfj,i;fj^fc)
ine that
we have data
for
for all
estimator 7^^^ in a panel with only two periods
plim7
=
7
7
and
and
+
Var
{-nl
^
fc
Now
0.
imag-
limit of the fixed effects
is:
-
+ ul -
ril_-^
ul_^]
+
where the second equality uses the assimiptions on the
definitions in (2)
z
two time periods. Then the probability
Uj's
and the
fj's
together with the
(3).
This expression shows that the fixed effects estimator will lead to consistent estimates
only
if ufj
and
v^^
are orthogonal,
i.e.,
if
there are no correlated shocks influencing the
evolution of income and democracy. Nevertheless,
if,
plausibly, Cov(vf^, ff^)
>
so that
such shocks axe positively correlated, the fixed effects estimator will be biased upwards
and
will
provide an upper bound on the causal effect of income on democracy.
In addition to the conceptual issues, there
the estimation of
(1).
The
regressor du-i
so the standard fixed effect estimator
However,
11).
as the
number
it
is
is
is
an econometric problem involved in
mechanically correlated with
not consistent
can be shown that the fixed
effects
this
(e.g.,
OLS
of time periods in the sample increases
implement a number of strategies to deal with
3.2
also
(i.e.,
Uis for s
<
t,
Wooldridge, 2002, chapter
estimator becomes consistent
as
T—
>
00).
We
discuss
and
problem below.
Results
Table 2 uses the Freedom House data and Table 3 uses the Polity data, in both cases
for
our entire (base) sample, over the period 1960-2000. All standard errors in the pa-
per (unless indicated otherwise) are robust against arbitrary heteroscedasticity in the
variance-covariance matrix, and allow for clustering at the country
We
start
the democracy score on
is
level.^
with a column showing the most parsimonious pooled
its (five-year)
lag
and
log
OLS
regression of
income per capita. Lagged democracy
highly significant, and shows a considerable degree of persistence (mean reversion) in
democracy. Log income per capita
positive relationship
^Clustering
is
is
also significant
and
illustrates the
between income and democracy. Though
well-documented
statistically significant, the
a simple strategy to correct the standard errors for potential correlation across obserand within the same time period. See for example Moulton (1986) or Bertrand,
vations both over time
Duflo and Mullainathan (2004).
index takes discrete values.
The
heteroscedasticity correction takes care of fact that the democrax;y
income
effect of
=
error
is
0.010) in
quantitatively small. For example, the coefficient of 0.072 (standard
is
column
1
of Table 2 implies that a 10 percent increase in
associated with an increase in the
small (for
this
Freedom House score
comparison, the gap between the United States and Colombia today
pooled cross-section regression identified the causal
then the long-run
right
GDP per capita
of less than 0.007, which
hand
side
effect
would be
larger
than
this,
effect of
is
increase in
because the lag of democracy on the
would be increasing over time, causing a further increase
GDP
per capita would be
If
income on democracy,
in the
democracy
Since lagged democracy has a coefficient of 0.706, the long-run effect of a
score.
very
is 0.5).
0. 007/(1-0. 706)~0. 024,
wliich
is still
10%
quantitatively
small.
The remainder
of Table 2 presents our basic results with fixed effects.
Column
shows that the relationship between income and democracy disappears once fixed
are included.
it
Now
the estimate of 7
With the
highly insignificant.
wrong
is
0.010
wth
2
effects
a standard error of 0.035, which makes
Polity data in Table 3, the estimates have in fact the
(negative) sign, -0.006 (standard error=0.039).
One might be
worried that the lack of relationship in the fixed effects regressions
is
a consequence of the imprecision of the estimates resulting from the inclusion of fixed
effects.
OLS
This does not seem to be the case. Although, as pomted out above, the pooled
estimate of 7
effects estimates
is
quantitatively small, the two standard error bands of the fixed
almost exclude
it.
More
specifically,
two standard error bands exclude short-run
with the Freedom House estimate,
effects greater
than 0.008 and long-run
effects
greater than 0.013 on the democracy index (the implied long-run effect of 0.024 in the
pooled cross-sectional regression
on lagged democracy
That these
is
is
comfortably outside this interval because the coefficient
smaller wdth fixed effects).
results are not driven
feature of the data
is
by some econometric problems or some unusual
further shoviTi in Figures 1
and 2 above, which plot the change
in
the Freedom House and Polity score for each coimtry between 1970 and 1995 against the
change
in
GDP
per capita over the same period.
These scatterplots correspond to the
estimation of the fixed effects equation (1) with contemporaneous income as the right-
hand
side regressor, without
1995.^''
They show
and changes
in
any covariates and using only two data points, 1970 and
clearly that there
democracy over
is
no strong relationship between income growth
this period.
"These two dates are chosen to maximize sample size. The regression of the change in Freedom House
score between 1970 and 1995 on change in log income per capita between 1970 and 1995 yields a coefficient
of 0.032, with a standard error of 0.058, while the same regression with Polity data gives a coefficient
estimate of -0.024, with a standard error of 0.063.
These
initial results
show that once we allow
for fixed effects, per capita
income
a major determinant of democracy. The remaining columns of the tables consider
is
not
alter-
native estimation strategies to deal with the potential biases introduced by the presence
of the lagged dependent variable discussed above.
Our
first strategy,
adopted in column
derson and Hsiao (1982), which
Adii
=
is
aAd,i_i
where the fixed country
+ lAy^-i + Ax'^^.^^a + A^^ + Au,*,
removed by time
effects are
cannot be estimated consistently by OLS,
(4)
original residual,
with
no second order
(i.e.,
in the
differencing.
absence of
(4)
Although equation
serial correlation in
the
serial correlation in Au^j), dit-2 is uncorrelated
can be used as instrument for Adjj_i to obtain consistent estimates and
Aiiit, so
similarly,
un
by An-
to use the methodology proposed
3, is
to time difference equation (1), to obtain
yu-i
is
used as an instrument for
negative estimates
and shows no evidence of a
We
Ai/it-i-
-0.104, standard error
(e.g.,
positive effect of
=
find that this procedure leads to
0.107 with the Freedom House data),
income on democracy.
Although the instrumental variable estimator of Anderson and Hsiao (1982) leads to
consistent estimates,
it is
not
correlation in Ua, not only
and can
also
all
but
all
serial
Au^,
these
moment
(GMM)
estimator using
conditions are valid, this
We
the Anderson and Hsiao's (1982) estimator.
The
Arellano and
be used as additional instruments.
Generalized Method-of-Moments
Wlien
under the assumption of no further
further lags of du are uncorrelated with
efficient, since,
dit-2-,
coefficients are
now even more
-0.129 (standard error
now comfortably
=
With
investigate whether the assumption of
OLS
no
is
estimate of 7 (which,
GMM
recall,
no further
was
0.072).
procedure allows us to
u^ can be rejected and
With the Freedom House
test indicate that there is
and Bover (1995)
two standard error bands
serial correlation in
overidentifying restrictions are not rejected.
'^In addition, Arellano
conditions.
tliis
this estimate, the
also to test for overidentifying restrictions.
(1).
moment
negative and more precisely estimated, for example
0.076).^^
exclude the corresponding
and the Hansen J
of these
(1991) develop a
GMM estimator more efficient than
GMM estimator in column 4.
use
In addition, the presence of multiple instruments in the
test
all
Bond
data, the Ail(2)
serial correlation
and the
^^
also use time-differenced instruments for the level equation,
Nevertheless, these instruments would only be valid
if
the time-differenced instruments are orthogonal
(e.g., five-year income growth is unlikely to
be orthogonal to the democracy country fixed effect), we do not include these additional instruments.
'^We also checked the results with five-year averaged data rather than our data set which uses only
the democracy information every fifth year. The results are very similar, but in this case, the AR(2) test
shows evidence for additional serial correlation, which is not surprising given the serial correlation that
averaging introduces. This motivates our reliance on the five-yearly or annual data sets.
to the fixed effect. Since this
is
not appealing in this context
10
With the
PoHt}- data, both the
Anderson and Hsiao (1982) and Arellano and Bond
(1991) procedures lead to more negative (and statistically significant) estimates. However,
in this case,
though there continues to be no
we need
test is rejected, so
to
serial correlation in ua,
be more cautious
the overidentification
in interpreting the results
wath the Polity
data.
Column
of
a
5
shows a simpler specification in which lagged democracy
Freedom House or Pohtj^ measure of democracy there
either the
significant effect of
estimate
democracy, which
Column
fixed effect
large.
income on democracy, and
easily outside the
is
is
two standard
OLS
OLS
is
dropped. With
in this case, the corresponding
error bands (the
not sho^Ti in the table,
6 estimates (1) with
is
again no e\'idence
is
OLS
OLS
estimate without lagged
0.235 with a standard error of 0.012).
using annual obseri-ations. This
is
useful since the
estimator becomes consistent as the number of obser\-ations becomes
With annual
observations,
we have
a reasonably large time dimension. However,
estimating the same model on annual data with a single lag would induce significant
serial correlation (since
our results so far indicate that five-year lags of democracy predict
now
changes in democracy). For this reason, we
log
GDP per capita in these annual regressions.
for
the joint significance of these
positive effect of
as
was the case
Finally,
Tlie table reports the
results
p
\'alue of
show no evidence
income on democracy (while democracy
is
an F-test
of a significant
strongly predicted by
its lags,
in earher columns).
columns 7 and 8 present regressions using a dataset consisting of ten-year
observ'ations.
democracy
The
\'ariables.
include five lags of both democracy and
This
will
is
useful to investigate whether the relationship betw^een
income and
be stronger with lower-frequency data. The results are similar to those
with five-year observations and to the patterns in Figures
of a positive association between changes in income
1
and
2,
which show no evidence
and democracy between 1970 and
1995.
Overall, the inclusion of fixed effects proxjing for time-in\-ariant country specific characteristics
results
removes the cross-country' correlation betn'een income and democracy. These
shed considerable doubt on the conventional wisdom that income has a strong
causal effect on democracy.
3.3
Robustness
Table 4 investigates the robustness of these results
we only
report the robustness checks for the
axe similar
and are
available
upon
request)
.
in alternative samples.
Freedom House data (the
Columns
11
1-3
show the
To save
results
space,
with Polity
regressions correspond-
ing to columns
This
is
2,
4 and 6 of Table 2 for a balanced sample of countries from 1970 to 2000.
useful to check whether entry
and
2 and 3 might be affecting the results.
from the base sample of Tables
exit of countries
All three
columns provide very similar
results.
For example, using the balanced sample of Freedom House data and the fixed effects
specification, the estimate of
error
bands now exclude the
Colunms
7
is
OLS
-0.031 (standard error=
.049),
OLS
and the two standard
estimate.
4-6 exclude sub-Saharan Africa, where
many
countries
became democratic
immediately after independence and later lapsed into nondemocracy. The results in this
sample are also similar and show no evidence of a significant positive
democracy
countries
in
any of the
and former
specifications.
Columns 7-12 report
socialist countries, again
effect of
income on
regressions excluding
with very similax
Muslim
results.
Table 5 investigates the influence of various covariates on the relationship between
income and democracy. To save space, we again report
We
House data.
start with the pooled
includes log population
and age
OLS
structure,
results only with the
Freedom
Columns
regressions for comparison.
and columns 4-6 add education.
Columns
7-9 include the full set of covariates from Barro's (1999) baseline specification.^^
cases, there
is
is
smaller than the baseline estimate in column
between income and democracy when fixed
1
of Table
affect
The
2.
all
included. Education
is itseff
insignificant with a negative
education on democracy, which
is
Age
coefficient.
structure variables
when education
The
is
causal effect of
We
investigate this
Acemoglu, Johnson, Robinson, and Yared (2005).
In addition, in regressions not reported here,
effects of
the table
the other basic tenet of the modernization hypothesis,
therefore also not robust to controlling for country fixed effects.
issue in greater detail in
rest of
the (lack of) relationship
effects are included.
are significant in the specification that excludes education, but not
monotonic
In
a positive and significant estimate of 7 in the pooled cross section, which
shows that the presence of these covariates does not
is
1-3
income on democracy and
we checked
for non-linear
and non-
for potential non-linear interactions be-
tween income and other variables, and found no evidence of such relationships.^*
'^Age structure variables are from United Nations Population Division (2003) and include median age
and variables corresponding to the fraction of the population in the following four age groups: 0-15, 15-30,
30-45, and 45-60. Total population is from World Bank (2002). In our regressions we measure education
as total years of schooling in the population aged 25 and above. In columns where we add covariates
from Barro (1999), we follow Barro's strategy by measuring education as primary years of schooling in
the population aged 25 and above. Both education variables are from Barro and Lee (2000). Additional
covariates from Barro (1999)'s regression are urbanization rate, male-female education gap, and a dummy
for major oil producer (used in the pooled cross-section only). For detailed definitions and sources see
Appendix Table Al.
''The only subsample where we find a positive association between income per capita and democracy
12
Instrumental Variable Estimates
4
As discussed above,
fixed effects estimators
do not necessarily
income on democraxiy. The estimation of such causal
of exogenous variation.
While we do not have an
are two promising potential instruments
effects requires us to exploit
a source
ideal source of exogenous variation, there
and we now present IV
results using these.
The Savings Rate Instrument
4.1
The
identify the causal effect of
instrument
first
The corresponding
y^t-i
where
the savuigs rate in the previous five-year period, denoted by Sn.
=
stage for log income per capita, yit-i, in regression (1)
7r^Sj£-2
+
a^d,t_i
+ x-j_i/3-^ + fj(_i +
The
identification restriction
is
that Cav{sit-2,Uit
the residual error term in the second-stage regression,
We
Sf
is
+ u[t_^,
(5)
the variables are the same as defined above, and the only excluded instrument
all
is Sit-2-
is
first
Xj;_i,/if,(5i)
\
=
0,
naturally expect the savings rate to influence income in the future.
While we do not have a
excludability?
where u^
is
(1).
What about
precise theory suggesting that the savings rate
should have no direct effect on democracy,
it
seems plausible to expect that changes
in
the savings rate over periods of 5-10 years should have no direct effect on the culture of
democracy, the structure of political institutions or the nature of pohtical conffict within
society.
Nevertheless, there are a
number
of channels through which savings rates could be
correlated with the error term in the second-stage equation, un. First, the savings rate
might be influenced by the current
itself
correlated with
u^
if all
political regime, for
example,
dit-2,
and could be
the necessary lags of democracy are not included in the system.
Second, the savings rate could be correlated with changes in the distribution of income or
composition of assets, which might have direct effects on political equilibria. Below, we
provide evidence that these concerns are unlikely to be important in practice.
With
in
IV
these caveats in mind. Table 6 looks at the effect of
regressions using past savings rates as instruments
(results using
is
Pohty data are
in
GDP per capita on democracy
and the Freedom House data
Appendix Table A2 and are
similar).
The
savings rate
defined as nominal income minus consumption minus government expenditure divided
conditional
on
fixed effects
is
the postwar sample with 18 West European countries.
when we look
at
However,
this
Freedom House data, and not with the Polity data, and also disappears
a longer sample than the postwar period alone. Details are available upon request.
relationship holds only with the
13
by nominal income.^^
We
report a
number
and with or without
of different specifications, with or without a lag of democracy,
GMM.
The
show the OLS estimates
three columns
first
democracy on the
cross section, the fixed effects estimates without lagged
and the
fixed effects, there
present.
column
right
1 is
hand
first
t-statistic of
democracy
almost
The
side.
(standard error
=
With
look at
IV
and
specifications,
The 2SLS estimate
first
stage
=
does not include lagged
is
no longer
and the bottom panel shows
Column
0.081).
significant at
between income and the savings
0.094).
6 uses the
Column
5%. These
GMM
Now
rate,
of the effect of income per capita on
very similar, and
is
rate as the excluded instrument for income.
negative,
it
fixed effects, this relationship
first-stage relationship
5.
-0.035 (standard error
is
hand
side,
Without
side.
stages.
Column 4 shows a strong
right
stronger than before because
side).
The remaining columns
the corresponding
hand
right
pooled
hand
a strong association between income per capita and democracy
is
(the relationship in
democracy on the
with a
democracy on the
fixed effects estimates with lagged
in the
right
5 adds lagged
now
democracy on the
the estimate of
7
is
-0.020
procedure, again with the savings
the estimate of 7
results, therefore,
relatively large
is
show no evidence
and
of a positive
causal effect of income on democracy.
The remaining columns mvestigate the robustness
of our exclusion restriction.
Column 7 shows a very
of this fuiding
Mrican countries are excluded. Column 8 adds labor share
as
for our results. ^^
The
first
Column
9 includes further lags of
differences in savings rates
the results.
The estimate
this case. Finally,
is
useful since
it
regressor,
and inequality might
stage shows no significant effect of labor share
on income per capita, and the 2SLS estimate of 7
labor share.
plausibility
when sub-Saharan
an additional
to check whether a potential correlation between the savings rate
be responsible
and the
similar estimate
is
similar to the estimate without the
democracy to check whether systematic
between democracies and dictatorships might have an
of 7
is
similar to before and,
column 10 adds a further
if
anything, a
little
lag of the savings rate as
effect
more negative
on
in
an instrument. This
enables a test of the overidentifying restriction (namely, a test of whether
the savings rate at t-3
is
a valid instrument conditional on the savings rate at t-2 being
a valid instrument). The 2SLS estimate of 7
is
again similar and the overidentification
PPP, numbers from the Penn World Tables. The first stage
is weaker and the second stage has a larger standard error if we use PPP data. The first and second
results are similar if we use an "investment rate" which is this measure of savings minus net exports.
""This is the labor share of gross value added from Rodrik (1999). We use these data rather than the
standard Gini indices, because they are available for a larger sample of countries. The results with Gini
coefficients are very similar and are available upon request.
^^We
calculate savings using nominal, not
14
restriction
is
accepted comfortably (the x^-statistic for a Hausman, 1978, test takes the
value of 0.00, which
is
accepted at the p- value of 1.00).
The Trade- Weighted World Income Instrument
4.2
Our second instrument
let ft
=
cjij is
In practice,
(which
the share of trade between country
we use two measures
of Q.
N
denote the
[uJij]-
trade shares between countries in our sample, where
Namely,
The
i
first is
x
A'^ is
N
matrix of (time-invariant)
number
the total
and country j
in the
is
details
The transmission
of business cycles
income of a country as
i.
Romer
(1999).
The
from one country to another through trade
(e.g.,
on data sources and construction.
Baxter, 1995, Kraay and Ventiu-a, 2001) implies that
for
GDP of country
a measure of predicted average
trade shares from a standard gravity equation used in Prankel and
Appendix provides
of countries.
actual trade shares between 1980-1989
chosen to maximize coverage). The second
is
To
exploits the existence of trade relationships across countries.
develop this instrument,
we can think
of a statistical model
follows:
N
^it-1
for
aU
i
=
1,
..., A'',
=
C
(^ijyjt-l
/_^
+ ^it-l:
where Yu-i denotes log income, so
the log population oiiatt
— 1. The parameter
yit^i
=
C measures the
(6)
Yu-i
—
Pit-i
where Pu-i
is
effect of the trade- weighted
world income on the income of each comitry.
Given equation
(6),
as follows: the error
and
if so,
the identification problem in the estimation of (1) can be restated
term
£it-i in (6) is potentially correlated
with u^ in equation
(1),
the estimates of the effect of income on democracy, 7, wiU be inconsistent. The
idea of the approach in this section
is
to purge Ya-i, and hence yu-i, from ea-i to achieve
we
N
consistent estimation of 7. For this purpose,
to use as an instrument for yu-i- Here Yu-i
is
construct
a weighted
sum
of world income for each
country, with weights varying across countries depending on their trade pattern. Given
Yit-i,
we can
oi^dit-i
+
consider a model for income per capita of the form:
x'jj_j/3
+
fi[_^
+
Sf
+
iij^_i.
Substituting for
relationship:
N
15
(7),
ya-i
=
we obtain our
Tt^Yu-i
+
first-stage
where the parameter
TT^).
The
vr^
assumption
identification
A sufficient
corresponds to Ctt^ (we do not need separate estimates of ^ and
condition for this
is
for this strategy
is
for
Y^-i to be orthogonal to un.
that Yjt-i be orthogonal to uu for
There are two problems with
all
this strategy, however. First, there
reasons for this identification assumption to be violated.
correlated with democracy in country j at time
t,
djt,
j
^
i.
may be
which
may
influence
Although we have no way
other, political, social or cultural channels.
"economic"
For example, Yjt-i
dit
may be
through
of ruling this
out a priori, we test for this in our empirical specifications below by controlling for the
democracy of trading partners, and
direct effect of the
find
no evidence to support such
a channel.
Second, there
is
an econometric problem, arising from the general equilibrium nature
of equation (6).^^ Since this equation also applies for country j, the disturbance term eu-i,
which determines Ya-i,
and
£it-i,
be correlated with
and
y,_i
=
let St-i
— C^)~
yt_i
]
sarily zero, £it-i will
tor for C, C
(i.e.,
~
be the
+ £,_i =
-^1
>
i-e.,
Nx
+ e,_i =
-
[[(/- c^)"']
the diagonal elements of [[(/
= Q,
be the TV x
Then
be mechanically correlated
with plim^
this, let Yt_i
vector of errors in (6).
1
cf^y,_i
inducing a correlation between Yjt-i
Yjt-i,
and thus between Ya-i and eu-i. To see
log incomes,
Since [[(/
will
]
~
we had a
vector of
+ st-i-
/] £t-i
— C^)~
Avith Yu-i. If
1
-^1
)
^^^ 1^°^ neces-
consistent estima-
then by implication we would also have plim£i_i
(where the probability limit applies as
A''
—
>
oo).
=
£(_i
This would enable us to construct an
adjusted instrument y^^"^, such that
[i-cny'
Using the Continuous Mapping Theorem
2.3), Y^^Bi^
would be uncorrelated with
remove the
indirect effect of ea-i
(see, for
e^t-i.
ht-i.
(9)
example, van der Vaart, 1998, Theorem
In other words, this transformation would
on yu-i working through the general equilibrium
inter-
actions across countries as well as the direct effect in (6). Obtaining a consistent estimate
of
C,
is
not straightforward, however.'-^ Here
this problem. First,
we take a number
of approaches to deal with
under some regularity conditions, the problem disappears as A'
so appealing to asymptotics on the
number
of countries, our first strategy
is
in
(2002).
^^This problem
is
>
oo,
it would result from an equilibrium model of crossBaxter (1995), Kraay and Ventura (2001), or Acemoglu and Ventura
''^We refer to this as "general equilibrium", since
country income determination as
—
to ignore this
investigated in current work, Acemoglu, Kursteiner
16
and Yared
(2005).
problem and use
Our second
(7).
(9)
(more details on
(7)
with lagged values of
and
Uit in
equation
strategy
this are given in the
to estimate C
is
and perform the adjustment
Appendix). Our third strategy
is
which also removes the source of correlation between
Yjt-i-,
(1) if eit-i's are serially uncorrelated.
in
to construct
Yjt-i
All three strategies give very
similar results.
The main
results using the
Freedom House data are presented
the Polity data are in Appendix Table A2)
Similar to Table
effects,
6,
the
first
.
In the
three columns report
in
Table 7 (results using
bottom panel we report the
OLS
and the patterns are similar to those presented
before.
(7)
with a
When we
-0.213 (standard error= 0.150).
is
slightly less negative
httle
more
precise with
Column 7 shows a
Africa.
an
Colunm
effect, influencing
in the first
and more
dit
5.
The 2SLS estimate
in
precise, -0.120 (standard error
GMM in column 6, -0.133
similar,
almost
add lag democracy
though
=
and second
is
and becomes a
0.077).
without sub-Saharan
find that
We
construct a world
as in equation (7),
democracy index, du
We
because of the differences in weights.
=
of 7
the estimate
democracy of trading partners might have
same trade shares
stages. This
5,
0.105),
(standard error
inference with this instrumental variable.
using the
column
slightly less precise, estimate
8 investigates whether the
democracy index,
both
t-statistic of
basic
constructed as in
is
The bottom panel shows
using the actual average trade shares between 1980 and 1989.
first-stage relationship
first stage.
-without fixed
Column 4 shows our
2SLS estimate with the trade-weighted instrument. The instrument
a strong
and
regressions with
,
d^ has no
and include
this
also varies across countries
effect either in
the
or
first
the second stages, consistent with our identification assumption that Ya-i should have no
effect
on democracy
Column
in country
except through
i
its
influence on
i/it-i.
9 uses Yit-2 instead of Ya-i on the right-hand side of (7) as an alternative
strategy to remove the mechanical correlation between Yu-i and Su-i. Finally, column 10
performs an overidentification test similar to that in column 10
both
Yit-i
and
Yit-2-
The estimate
and the overidentifying
the
first
of
7
is
in
restriction that the twice-lagged instrument
instrimient being valid
test takes the value of 0.14,
is
which
Table 6 by including
column
4,
valid conditional
on
similar to the baseline estimate in
is
easily accepted (the x^-statistic for a
is
Hausman,
1978,
accepted at the p- value of 1.00).
Table 8 presents further robustness checks. Columns 1-3 exclude Singapore, which
an
outlier in the first stage.
and the second-stage
Yit^i using (9),
The
coefficient
which has
first-stage relationship is
weaker but
still
is
significant,
remains negative and insignificant. Columns 4-6 adjust
little effect
on the estimates. Columns 7-9 present specifications
using the gravity equation to construct Q, which yield similar results to those in Table
17
7.
Overall, our
two IV strategies give
and indicate that there
is
no evidence
with the fixed
results consistent
a strong causal
for
effect of
effects estimates
income on democracy.
Interpretation
5
The above
results raise the following question:
if
there
is
no causal
effect of
income on
democracy, what explains the strong cross-sectional relationship between the two
This strong cross-country relationship
ables?
House data and
is
shown
is
vari-
Freedom
in Figure 3 using the
the source of the positive and significant estimates in the pooled cross-
sectional regressions.
Two
Hypothesis
5.1
The
hypotheses naturally present themselves.
first possibility is
1:
Long Lags
that there
is
a causal effect of income on democracy, but
with such long and variable lags that
postwar data. Although the issue of lags
is
a natural extension of
may
his thesis, since
it
it
works
impossible to detect this causal channel in the
it is
not discussed explicitly in Lipset (1959), this
is
take a long time for a culture of democracy
to emerge, or because political institutions change only slowly.
According to this hypothesis, we need to look at longer time spans. The key question
is
what the
right horizon should be. Figures 1
and
2
show that there
look at a 25-year period (1970-95) during the recent past.
version of this hypothesis should predict
some
effect
is
no
We believe that
when we
effect
when we
any reasonable
look at a 50-year horizon
or perhaps, at the longest, a 100-yeax horizon.
We
next investigate the longer-run relationship between income and democracy. But
before doing this,
5.2
we
discuss an alternative explanation for our results.
Hypothesis
The second
2:
Divergent Development Paths
possible explanation, which
we
find
more appealing than hypothesis
1, is
that
the positive cross-sectional relationship reflects relatively time-invariant, historical factors
that have influenced the economic and political development path of societies. There are
reasons to expect that economic and political development processes should be related.
For example, economic institutions encouraging economic growth are unlikely to develop
and endure
exercise
it
in societies
where
political
power
to further their interests, even
if
growth.
18
is
in the
this
is
hands of a small
elite,
at the cost of aggregate
who can
economic
Moreover, events during certain
fects
critical
on income and democracy because of
junctures
may have
their influence
large
development paths of societies. Consequently, part of the variation
poUtical institutions that
we observe today may be the
and
persistent ef-
on the economic and
result of
in
some
political
both prosperity and
societies
having em-
barked on a path of development based on relatively democratic political and economic
institutions encouraging growth, while other societies
ended up with repressive regimes
controlled by narrow elites, which engendered neither prosperity nor democracy.
The
contrast between European colonies in Latin America, such as Peru and Bohvia,
and the North American colonies
illustrates the potential effects of
critical junctures. All of these societies
key events during
were colonized by Europeans after 1492, but the
a great deal. While smallholder societies, with the
exax:t colonization strategy differed
majority of the population enjoying access to land, secure property rights, and economic
opportunities soon developed in the North American colonies, a liighly coercive society,
based on the exploitation and enslavement of the indigenous peoples, emerged in Peru and
Bohvia (potential reasons
in Section 7).
for these differences in the colonization experience are discussed
Moreover, these early differences in economic and pohtical institutions have
shown a considerable degree
of persistence.
While the United States and Canada have
remained democratic and grown rapidly over the past 300 years, the
record of Peru and Bohvia has been
much more
political
and economic
checkered. Hypothesis 2 implies that to
understand the relationship between income and democracy, we need to look at the events
and factors influencing
histitutional equilibria at critical junctures, in this case during the
early phases of coloniahsm.
5.3
Implications and Discussion
Which
of these
two hypotheses
is
closer to the truth?
We
are unable to answer this
question definitively at the moment, but two empirical exercises are informative.
strategy
is
to look at the longer-run relationship between income
the second strategy
is
first
to look directly at potential determinants of divergent historical
development paths, which we turn to
in Section 7.
Both hypotheses suggest that the longer-run
for very different reasons,
and potentially
relationship should be
is
political
positive, but
the events during critical junctures,
so the time interval should include dates spanning these junctures.
modern economic and
more
at very different horizons. In particular, accord-
ing to hypothesis 2, the source of the correlation
the
The
and democracy, while
development experience,
this
In the context of
means that the
positive
long-run relationship should emerge (or become more pronounced) in a data set covering
19
window
the era before the 19th century rather than a
of 50 or 100 years during the recent
past.
To
clarify the
main
issues, let us return to the statistical
model
in equations (2)
and
(3).
In the context of these equations, our emphasis on political and economic development
paths diverging at some
some
v^f at
t
=
,
some common date
Now
t
then have a persistent
will
because of the unit root in
at
juncture corresponds to large and correlated shocks
critical
T* which
=
and
rjf^
T*
differencing equations (2)
-
diT
and
diT-s
for
(3),
is
= 7 {ViT -
happen
only for simplicity).
two time periods again,
we
and
vf^
on democracy and prosperity
(the implicit assumption that these shocks
rj\
for all countries
imagine we have data
effect
=T—S
t
and
t
—
T. Time-
obtain:
ViT-s)
+ rjfr - ^frs + ^fr -
(10)
'^fr-s^
and
Recall that the variances ofv^ and u^ are denoted by a^y and a^y, and
0,
=
Cov{vl,vl^,^)
Cov(vfj^,,v^j.,^
affecting
this,
=
0,
cr|..
be positive and
both economic and
political
=
suppose that Cov(vf^,v^f)
positive,
for all
i
and
and
in particular
fc
7^ 0.
Now
=
and Cov{vi,vft^,^)
much
for all
and k
i
a^x* for
t
smaller than
at this critical juncture.
^
T*, which
Cov(uf^, u^n+k)
Moreover,
0.
To
we presume
contrast with
to be small
Suppose also that Cov(vf^,v^f^i.^
a"^,.
let
importance of a major event
large, capturing the
outcomes
we have
7^
and
=
consider the fixed effects estimator 7^"^, where the time span
is
given by S. Standard arguments imply that the probability limit of this estimator using
these two data points
is:
^
plim7r = 7 +
^
7
+
+
X
,
airaiy+2aiylS
{ol,-ol^.)/s+a'l^,
aiy+2aiy/S
where the second equality exploits the
fact that Vi^s
Equation (11) has three important implications.
fixed effects estimator
racy increases (and as
less
-
5, T]
e[T-
S, T]
if
r* ^ [T
if
T*
and
(11)
'
Uj's are serially uncorrelated.
First, as
S
increases, the bias of the
and the predicted positive relationship between income and democ-
S becomes
smaller, plim7f'^ approaches 7). This
increases, the non-persistent shocks to income,
become
\^
^
is
because, as
which are uncorrelated with democracy,
important relative to the component correlated with democracy,
20
S
v^^,
and
this
reduces the denominator through the term 2a\y/ S. Second, and more important,
the time span from
=T—S
t
to
t
= T
mcludes the
critical
stronger positive relationship between income and democracy, since
is
relevant in interpreting
why we
if
we
— cr^j,. >
see a positive relationship between these
during some horizons but not others. Finally, under hypothesis
be reduced
ctI^.
2,
when
we expect a
juncture T*,
0.
This
two variables
the bias of (11) will
are able to control variables that proxy for or are correlated with the
common component
in
and
t'fj..
v^j,.
(in practice, historical
determinants of divergent
development paths).
Democracy and Income
6
6.1
Democracy and Income Over the past 160 Years
Although
historical
data are typically
less reliable
than postwar data, the Polity IV dataset
extends back to the beginning of the 19th centm-y
dent),
we
for
that were then indepen-
many
countries from aromrd
construct a 5-yearly and a 10-yearly dataset between 1840 and
Countries that gained and maintained independence before 1900 and have more
than 5 observations
result
(for coimtries
and Maddison (2003) gives estimates of income
1820. Using these data,
1940.
Long Run
in the
is
in the
1840 to 1940 data period are included in this dataset. The
an unbalanced panel with a coimtry entering when there are observations from
both Polity and Maddison.^^
Table
9,
A
Panel
reports
OLS
pooled cross-sectional
some
basic regressions with this dataset.
show the conventional
regressions again
per capita having a positive and significant sign.
to our results above, the coefficient estimate
Column
3 uses
GMM
Column
result,
In
1,
with income
2 adds fixed effects
on income per capita becomes
colmnn
and
similar
insignificant.
and column 4 excludes lagged democracy, again with
similarly
insignificant results.
Columns
5
and 6 repeat the regressions from columns 2 and 3 using ten-year instead of
five-year intervals.
We
income on democracy
use this strategy to check whether the lack of a significant effect of
is
caused by measurement error or country noise in democracy over
the five-year horizon, and also to mvestigate whether there could be an effect of income
on derriocracy at lower frequencies than over 5 years. The basic results are identical to
those in the
first
four columns of the table.
'®The countries in this section are Argentina, Austria, Belgium, Brazil, Bulgaria, Canada, Chile, China,
Colombia, Denmark, France, Germany, Greece, Hungary, Italy, Japan, Mexico, Netherlands, Norway,
Peni, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States, Uruguay, Venezuela.
21
The
conclusion from this investigation
Europe and Latin America
control for fixed effects, there
that the pre- 1940 evolution of countries in
is
similar to the results
is
is
no
from the post- 1960 sample. Once we
significant relationship
between income per capita and
C
by taking the same sample of
democracy.
We
B
further investigate these ideas in Panels
and
countries and extending the dataset to 2000 (thus constructing a 160-year panel with 28
countries).
with
GDP
log
Without
gi'eater
fixed effects, higher per capita
income
democracy. In this 160-year sample, there
per capita
when we
robust to estimation via
is
again strongly associated
also a positive coefficient
control for fixed effects, though this relationship
is
on
not
GMM in the five-year sample.
Figure 4 depicts the change in the
GDP
per capita between 1870 and 1995
Polity composite index versus the change in log
(dates chosen to
is
maximize sample
size)
,
and shows a positive but
insignificant relationship
between these dates.^"
Further investigation suggests that the results in the 1840-2000 panel are partly driven
by one country, Venezuela, especially the postwar correlation between income per capita
and democracy in Venezuela.^^ Further
the Venezuelan
revenues
GDP
investigation suggests that the correlation between
and democracy over the relevant period
is
related to increases in oil
—not the type of income variation generally thought to promote democratization.
In particular, there
is
a close association between
oil
income and democracy
in Venezuela,
but not between non-oil income and democracy (details are available upon request). Panel
C
B
therefore repeats the regressions in Panel
without Venezuela. In this case, though
the estimate of the coefficient on income per capita remains positive,
it
is
no longer
statistically significant.
6.2
Democracy and Income Over the past 500
Next we push the reasoning
suggests that
if
in equation (11) further.
we could construct a
divergent development paths,
existing dataset spans
discussion of hypothesis 2
longer dataset, spanning the critical junctures for
we should obtain a
more than 160 years
estimates of income per capita for almost
The corresponding
Our
Yeairs
stronger relationship.
Although no
of political development, there exist rough
all
areas of the world in 1500. Moreover,
we
regression yields the coefficients of 0.076, with a standard error of 0.123. Parallel
with our treatment of the very long run below, we also looked at the relationship over the period 18701995 assigning the lowest Polity score to countries without Polity data. We found only a small and
marginally significant coefficient on income in the full sample, and a small and insignificant coefficient
for
former European colonies, which we discuss further below. Details are available upon request.
In Tables 2-8, Venezuela did not have a disproportionate effect on the results because these regressions
^^
included a considerably larger set of countries than Table
22
9.
also have information about the variation in poUtical institutions
around the turn of the
16th century. While no country was fully "democratic" according to current definitions,
there were significant differences in the political institutions of countries around the world
even at this date. In particular, most countries outside Europe were ruled by absolutist
regimes while some European countries had developed certain constraints on the behavior
of their monarchs.
Acemoglu, Johnson and Robinson (2004b) provide a coding of constraint on the executive for
European countries (based on the Polity
various sources.
It also
definition) going
back to 1500 from
appears reasonable to assume that constraint on the executive for
non-Etnopean countries and the other components of the Polity index (competitiveness
of executive recruitment, openness of executive recruitment,
both
ical participation)
score in 1500.
for
for
Based on
and competitiveness of polit-
European and non-Em"opean countries should take the lowest
this information,
we can
construct the Polity Composite index
1500 (details available upon request). Combining these data with estimates of income
per capita,
we can
get a glimpse of the relationship between income
and democracy over
this 500-year interval.
This
1500.^^
is
done
The
and changes
in
figure
in
Figure 5 using Maddison's (2003) estimates of income per capita in
shows a strong positive relationship between changes
income
for 143 countries (1995 is
Table 10 to maximize sample
size).
in
democracy
used as the end date in the plot and in
This plot corresponds to a fixed
from
effect regression
1500 to 1995 with only two time periods, of the form:
Adi,i995_i500
This regression
is
=
reported in column
To
+ 7Ayt,1995-1500 +
1 of
Table
significant estimate of 7, 0.139 (standard error
10,
=
and
.033).
yields a large
Column
relationship this time assigning the lowest
democracy score
shows a very similar estimate. The
Table 10
rest of
Overall, the very long-run evidence
income, but this correlation
years or so.
We
is
to
statistically
2 estimates the
same
in 1500,
and
be discussed below.
will
shows an association between democracy and
not present or quite weak
effect
and
aU countries
when we look
interpret this evidence as consistent with hypothesis
would have suggested a strong
^
Azi^.
even in the 100- or 140-year data
at samples of 100
2.
Hypothesis
1
sets. In contrast.
For this exercise, when Maddison (2003) provides estimates for individual countries, we use these
When he provides estimates for broad geographic areas, we assign this estimate to all countries
estimates.
now occupying
in the right
these territories without country-level data.
hand
side variable
We
by clustering the standard errors
country.
23
take into account the cross-correlation
at the
data aggregation
level for
each
we
hypothesis 2 would suggest the stronger relationship should emerge once
a time interval
sufficiently long to
development paths. In
span the
this case, the fact that
at a 500-year period, but not
critical
consider
junctures leading to the divergent
a strong relationship emerges when we look
when we only go back
to the mid-19th century,
is
consistent
with this view.
Understanding the Fixed Effects
7
We now
If
directly investigate the potential determinants of divergent
we can pinpoint
between income and democracy,
this
over, as suggested in Section 5,
if
would be evidence supporting hypothesis
we can
their proxies, the positive correlation
7.1
development paths.
these potential determinants and their influence on the relationship
2.
More-
directly control for these historical factors or
between income and democracy should weaken.
Divergent Development Paths
Among
the Colonies
Acemoglu, Johnson and Robinson (2001, 2002) document that factors affecting the profitability of different institutional structures for
on early
institutions,
and on subsequent
European
political
colonizers
had a major impact
and economic development. They em-
phasize that two factors affecting colonial strategies, and therefore the subsequent devel-
opment paths, were the mortality
rate faced
by potential European
lation density of indigenous peoples before colonization (in practice
mortality rates discouraged Europeans from settling, and
more
and the popu-
around 1500). Higher
made an
associated with coercive and non-participatory institutions,
settled areas also discouraged
settlers
extractive strategy,
likely.
More
densely-
European settlements, and even conditional on settlements,
encouraged the establishment of coercive institutions designed to control the indigenous
population and to transfer resources from them.
We
next use these ideas in the sample of former European colonies, where European
intervention created a potentially exogenous source of divergence in political and economic
development paths.
We examine the effects of both
tion in 1500 (population density in 1500, for short)
countries with high rates of settler mortality
in
the density of the indigenous popula-
and
of settler mortality.^'^
We
expect
and higher indigenous population density
1500 to have experienced greater extraction of resources and repression by Europeans,
^^Population density in 1500
McEvedy and Jones
is
calculated by dividing the historical measures of population from
(1975) by the area of arable land (see Acemoglu, Johnson and Robinson, 2002).
data on settler mortality are from Acemoglu, Johnson and Robinson (2001), who constructed it
based on research by Philip Curtin and other historians (e.g., Curtin, 1989, 1998, and Gutierrez, 1986).
Finally,
24
and consequently to be
and European
less
democratic today. However, both population density in 1500
settler mortality rates are subject to
a large amount of measurement error,
and are only some of the influences on the ultimate development path. For example,
various reasons, Europeans opted for extractive institutions in
many
for
areas, such as Brazil,
with low population density. Therefore, a direct measure of institutions immediately after
the end of the colonial period
on current outcomes. For
tive from the Polity
IV
is
also useful to gauge the effect of the historical conditions
this reason,
we look
measured as the average score during the
closest variable
this score to
at the
measure of constraint on the execu-
dataset right at (soon after) independence for each former colony,
we have
first
ten years after independence.^'* This
to a measure of institutions during colonialism.
We
is
to 1 scale like democracy, with 1 representing the highest constraint
a
the
normalize
on
the executive.
we
Finally,
also control for the date of independence. This
on the executive
at different dates of
independence
important to control for the date of independence.
is
may mean
useful because constraint
different things, so
it is
In addition and potentially more
important, countries where Europeans settled and developed secure property rights and
more democratic
institutions typically gained their independence earlier
with extractive institutions.
Another important
effect of
than colonies
the date of independence on
and economic development might be that former colonies undergo a
political
relatively
lengthy period of instability after independence, adversely affecting both growth prospects
and democracy.
7.2
Historical Variables
and Fixed
Effects
Figures 6-9 show that the fixed effects in the democracy regressions are closely hnked to
the historical determinant and proxies for divergent development paths
European
effects,
colonies. ^^
and
In particular, they
the former
show a strong association between the
fixed
respectively, settler mortality rates, the density of the population in 1500,
constraint on the executive at (shortly after) independence,
These
among
and year of independence.
figures suggest that the fixed effects are indeed related to the conditions
which
contributed to the divergent development paths of these countries.
^'Data on date of independence are from the CIA World Factbook (2004). For detailed data definitions
and sources see Appendix Table Al. The data on constraint on the executive from Polity begins in 1800
or at the date of independence. In our former colonies sample only one country, the United States became
independent before 1800. The United States broke with Britain in 1776 and was recognized as a new
nation following the Treaty of Paris in 1783.
^^These fixed
eff'ects
We
code the U.S. date of independence as 1800.
column 2 of Table
are from the regression for the former colonies sample in
25
11.
The
Tables 11 further documents this point using regression evidence.
OLS and
columns show the pooled
fixed effects relationship
racy in the sample of former European colonies. ^^
As
first
two
between income and democ-
in the other samples, there
is
a
strong cross sectional relationship between income and democracy, but no evidence of a
causal effect with fixed effects in the sample of former European colonies.
The remaining columns
of Tables 11
show that when the four proxies
for the evolution
of the development path are included in regressions without fixed effects, the influence of
income per capita on democracy
disappears. For example,
when
is
all
when
and independence year are included
only the last three variables are included, there
income per capita on democracy.^'' Table A3
the Polity data
These
specifications,
it
four of settler mortality, population density in 1500,
constraint on executive at independence
or
weakened and in many
substantially
in the
is
no longer a
Appendix shows
in
column
6,
significant effect of
similar results using
set.
results suggest that the fixed effects,
lationship between income
and democracy, are
sample of former European
colonies.
As
which account for the cross-sectional
re-
closely related to colonial history in the
such, they lend support to hypothesis
2,
which
emphasizes the importance of divergent economic and political development paths.
7.3
Revisiting the Very
we now
In the light of these findings,
sample shown
in Table 10.
with hypothesis
2,
since
it
Long Run
We
return to the positive relationship in the 500-year
have so
far
argued that this relationship was consistent
was precisely during
divergent development paths.
A
this period that countries
embarked upon
further check on hypothesis 2 would be that
when we
include the historical variables emphasized in the previous section, the 500-year relationship between income and democracy in the former European colonies sample should again
^''We adopt the definition of former European colonies used in Acemoglu, Johnson and Robinson (2001,
European powers during
motivated by our interest in former colonies as a sample where the
process of institutional development, in particular during the 19th century and earlier, was shaped by
European intervention (see Acemoglu, Johnson and Robinson, 2002). The results are robust to including
the Middle Eastern countries.
^^We also investigated whether other time-invariant characteristics that have been emphasized in the
2002), which excludes the Middle Eastern countries that were briefly colonized by
the 20th century. Tliis definition
is
previous literature, in particular, identity of the colonizer, legal origin, religion or latitude, have explanatory power for democracy once the fixed effects are removed in the sample of former colonies.
We
found
on democracy when the historical determinants emphasized in this
section are included in the regressions. Furthermore, we added more lags of democracy on the right hand
side, and constraint on the executive at independence remained significant, suggesting that this variable
does not capture some slow dynamics in democracy, but is instead related to the political and economic
development path of the former colonies. Results are available upon request.
that none of
them have any
effect
26
weaken
or even disappear. This
Column
is
investigated in the rest of Table 10.
3 shows a similar relationship between change in
racy to that showTi in Figure 5 and in column
The remaining colmnns show
1
among
mcome and change in democ-
the former European colonies.
the historical variables emphasized in the pre\'ious section.
when
all
when we
that the positive relationship disappears
include
For example, in colunm 7
four of population density in 1500, settler mortality, constraint on executive at
independence and independence year are included, the coefficient on
essentially zero (-0.006, standard error
illustrated in Figure 10
=
0.054) instead of 0.126 in
becomes
Ayi.500,1995
column
which depicts the relationship between changes
in
3.
This
changes in income in former colonies conditional on these four historical factors.
evidence also supports hypothesis
is
democracy and
This
2.
Economic Crises and Democracy
8
Our
analysis so far shows that
time variation (at the
In this section,
five- or
income seems to have
little
predictive power for the over-
ten-year frequencies) in democracy in the postwar sample.
we document that there
are
some predictable movements
in changes in
democracy.
A
number
of theories, including
Haggard and Kaufman (1995) and Acemoglu and
Robinson (2004), emphasize economic
crises as events destabilizing
both democratic and
nondemocratic regimes, and leading to regime transitions. The overall
democracy
is
ambiguous, and depends on whether the destabilizing
or nondemocracies
We
is
investigate this issue in Table 12.
time
any year
in
t
—
1 i£
We
define
five years ago.
an economic
More
crisis as
level of
the change in the five-yearly average growth rate of
in the five-year period
between t —
1
and t —
a sudden and
specifically, there is
2
is less
— Ay.^^ We choose the threshold Ay
as 3 percent (columns 1
3 and 4) or 5 percent (colmnns 5 and
6).
^®
on the
on democracies
greater.^*
sharp decline in growth relative to
crisis at
effect
effect
an economic
GDP
per capita
than a certain threshold
and
2),
4 percent (columns
These thresholds are motivated by the fact that
Various empirical papers in the political science literature, for instance Gasiorowski (1995) and Przeal. (2000) have investigated the impact of crises on regime transitions. However, this research
worski et
has not attempted to control for omitted variables or for the endogeneity of income.
^^ Recall that the time period, t, refers to five year intervals. The years s between i — 2 and
[5t
— 10, 5t — 6]. We denote
the forward average five-year growth rate in year s by Ays
f
—1
are s
£
=
X^ -^ Ays+j/5,
This means that we need 15 years
and define a crisis during year s as occuring if Ay^ — Ay^-s < —Ay.
of data on income per capita preceding the democracy observation, and our sample
1965.
27
in
Table 12 starts in
one standard deviation of this variable
The dependent
in Panel
crises
A show that,
make democracy more
This result at
likely.
B and C show why
looks paradoxical.
first
How could crises
economic
this is so:
on
transitions
crises
have a small, and typically not
away from democracy, but a
large effect
away from nondemocracy to democracy. In other words, economic
dictatorships to
crises
thus increasing the likelihood of democracy. To docmnent
fall,
results
likely?
statistically significant, effect
transitions
The
again contrary to the predictions of modernization theory, economic
make democracy more
Panels
equal to 4 percent.
is
A is the Freedom House democracy score.
variable in Panel
on
cause
this,
we
take a simple approach and create two variables measuring movements towards and away
from democracy. In particular, equation
dft
for transitions to
ad^t-i
modified as follows:
+ n'Vit-i + x-i_i/3 + pj + S^ + uu
(12)
+ IVit-i + At-\P +
(13)
democracy and
dTt
for transitions
=
(1) is
=
oidit-i
from democracy, where
d'l^
=
iJ,t
+ 5, + uu
max{djt,d,j_i} and
d~^
=
mm{dit,dit-i}.^^
This procedure implies that for d^ we only consider upward movements in the democracy
,
score
and ignore declines
democracy
A
score.
full
in democracy,
(OLS
or
d^^
only considers deterioration in the
analysis of democratic transitions
paper. For our purposes here,
fixed effects
whereas
GMM)
it is
finds
sufficient to
is
beyond the scope of
this
note that estimation of these equations with
no evidence that income per capita has a causal
eflFect
on either transitions to or transitions away from democracy. However, these estimates
suggest that economic crises which enter into
x'jj_j^
do have an
effect
on transitions to
democracy.^^
Using these definitions, in Panels
B and
C,
we
investigate the relationship
between
economic
crises
and transitions to and away from democraxjy. The estimates show that
economic
crises
make
much
dictatorships
smaller effect on transitions
together explain
why economic
more
likely to transition to
democracy, but have a
away from democracy. These two
crises are
results
combined
found to lead to greater democracy in Panel
A.32
Although (12) and (13) are non-hnear equations, the fixed effects, the Si's, enter additively and can
be differenced out to achieve consistent estimation.
'^
We did not find any robust results using this definition of crises on the 160 year sample discussed in
Section
6.
^^A natural question
is
whether growth accelerations have a similar
28
effect
on regime change.
A
recent
Conclusion
9
The conventional wisdom
a causal
effect
in the pohtical
on democracy. In
racy are positively correlated, there
most probably
historical, factors
is
we argue
that,
though income and democ-
no evidence of a causal
effect.
Instead, omitted,
appear to have shaped the divergent political and eco-
nomic development paths of various
societies, leading to
democracy and economic performance.
fixed effects
economy Hterature that income per capita has
this paper,
the positive association between
Consequently, regressions that include country
and/or instrumental variable regressions show no evidence of a causal
income on democracy. These
results
effect of
shed considerable doubt on the conventional wisdom
both in the academic Hterature and in the popular press that income causes democracy,
and a general increase
in
income per capita wiU bring improvements
This result immediately leads to an important question:
democratic today?
why
in institutions.
are richer countries
more
We provided evidence that this is hkely to be because the political and
economic development paths are interwoven. Some countries appear to have embarked
upon a development path associated with democracy and economic growth, while others
pursued a path based on cUctatorship, repression and more limited growth. Consistent
with
this, in
the sample of former European colonies where
the historical sources of variation in development paths,
we have good measures
we showed
of
that the fixed effects
indeed captine the impact of these historical differences.
In emphasizing the importance of historical development paths,
a historical determinism
suggest that there
is
in the regressions,
and the development paths
we do not want
in political institutions.
The
in theory, create a tendency,
other factors influence equilibrium political institutions. In the last section,
how
severe economic crises lead to the collapse of dictatorships,
hkely.
The most important
effect of these time- varying
area for future research
and human
factors
is
to
fixed effects
but
many
we showed
making democracy more
a further investigation of the
on the evolution of equilibrium
political
institutions.
paper by Hausmann, Pritchett and Rodrik (2004) emphasizes the importance of growth accelerations.
We also investigated the effect of growth accelerations in a symmetric way to economic crises by creating
a dummy for acceleration if the change in the five-yearly average growth rate of GDP per capita in any
year between f — 2 and i — 1 is above a certain threshold Ay (again 3 percent, 4 percent or 5 percent).
None of these specifications showed any effect of growth accelerations on democracy.
29
,
Appendix
10
This Appendix describes the construction of the trade- weighted world income instrument
used in Section
Second,
we
we
4. First,
we
describe the instrument constructed from actual trade shares.
describe the instrument derived from gravity-predicted trade shares. Finally,
describe our
method
rium nature of equation
of adjusting for the potential bias caused
we measure the matrix
First,
=
il
[uij]
.
Let Xijs denote the total trade flow between
and exports from
to j
j to
year s for which both flows from
measured using either
are available,
we
using actual trade shares between 1980 and
.
Fund Direction
the International Monetary
i
equilib-
maximize coverage. Bilateral trade data are from from
1989. These dates are chosen to
from
by the general
(6).
FOB
i
and j
We
in year s.
exports from
to j or
i
take the average, and otherwise
US
All trade data are deflated into 1983
calculate Xij^ for
i
CD-ROM.
meaning the sum of exports
in year s,
and from j to
to j
i
(DoT) (2005)
of Trade Statistics
i
all
country pairs in
are available. These flows can be
GIF imports by
j from
i.
we use whichever measure
dollars using the
US CPI from
Wlaen both
is
available.
International
Financial Statistics (2004).
Let
denote the total
Y^l
GDP
of country
Heston, Summers, and Aten (2002), and
for
which
bilateral
data between
i
and j are
i
in year s in 1983
US
dollars obtained
from
be the number of years between 1980 and 1989
lij
available.
Our main measure
of
Jl
=
[w^
]
.
.
is:
1989
T
u;„
where Xus
Since
follows.
=
by
=
^
^"
1
v-^
/
/
Y^
1980 ^
^
Af
'^JS
^*
definition.
we have an unbalanced
Define Ijt-i
=
panel,
we construct our instrument
defined in (7) as
{0, 1} as an indicator for Yjt-\ being available in the dataset.
Then
y.-.
where
Yjt-\
is
log
=
c
f
E
-,/,._.y,._.')
income as before.
The
f ^F^-y"'^'
third term in (14) ensures that the
the weights Uij are the same across time for a given country
term
is
equal to
1
in a balanced panel.
We
i,
and
this
djt for l^t-i
and
let Ijt-\
sum
of
adjustment
measure trade-weighted democracy
analogous fashion using (14), where we substitute
an indicator referring to the
(14)
)
now
dit
in
an
represent
availability of the variable djf
Second, we generate an alternate measure of
30
fi,
Vt^
,
from the following gravity equation
and Romer
as in Frankel
(1999):
=Po + Pilfj + P2Zij + Pzlpij +
log {u,j)
where Z^
1 if
a vector of country-pair level covariates,
is
the two countries share a land border, and e^
includes the following covariates:
between
log distance
population of
we
construct
and
i
dummy
if i is
area of
i,
j, log
We estimate (15)
QP = \ujfA where:
j.^^
.
L
can use
ojf^
in
a
dummy variable which equals
an error term where
landlocked,
dummy
if
£" (e^ )
j
is
With our estimates
of
/Jg, /5i,
=
Zy
0.
landlocked,
i,
and
log
^2^
^^^
i^zi
log area of j, log population of
using OLS.
.
'J J Zjj
+ Pjfj + %Z,, + /?3/^^»,)
exp (do
We
is
/j| is
(15)
ez,-,
an analogous fashion to build an alternate instrument Y^^j now using
gravity estimated weights:
y.f-.
=
c
E
(
-S''.-.
I
(
3°"f
""'
(16)
I
where second and third terms of
(16) represent a gravity- weighted average of world
where the sum of the weights
actual trade openness 'Yli=ij^i'^ir^'^
Finally, for the
which includes a
specification
sample
in the
is
adjustment in
we estimate
(9),
full set
of time
second stage in Table 7 column
(6)
income
using actual trade shares in a
and country dummies and using the same
4:
N
Fii_i
=
C
^
Wy-y^j-i -F£i(_i,
This provides us wdth an estimate of
instrument. In inverting the matrix
whenever
Uij
is
unavailable for any
(
where
C,
and eu-i-
/
—
($!
,
j
eif_i
We
=/i°_i
then use
we drop the
ith
-1-5°
(9)
+ u°_i.
to create an adjusted
row and
zth
column of
Q.
j.
^^As in Frankel and Romer (1999), we constrain the coefficients on the two landlocked dummies to
be the same. For this exercise, area is from CIA (2004). Population is the average log population from
1980 to 1989 from WDI (2002). Landlocked, contiguity, and distance are from Click and Rose (2002).
Detailed results are available upon request.
^"^
Results are similar
if
the
sum
of the weights
is
predicted openness. Results are available upon request.
31
11
Bibliography
Acemoglu, Daron, Simon Johnson
Origins of Comparative Development:
ctnd
An
James A. Robinson (2001) "The
Colonial
Empirical Investigation," American Economic
Review, December, 91, 1369-1401.
Acemoglu, Daron, Simon Johnson and James A. Robinson (2002)
Geography and Institutions
of Fortune:
"Reversal
Making of the Modern World Income
in the
Distribution," Quarterly Journal of Economics, 118, 1231-1294.
Acemoglu, Daron, Simon Johnson and James A. Robinson (2004a)
tions as the
Fundamental Cause of Long-Run Growth,"
NBER
"Institu-
Working Paper #10481,
forthcoming in Philippe Aghion and Steven Durlauf eds. Handbook of Economic Growth,
Amsterdam; North-Holland.
Acemoglu, Daron, Simon Johnson and James A. Robinson (2004b)
of Europe: Atlantic Trade, Institutional
"Rise
Change and Economic Growth" American Eco-
nomic Review, forthcoming.
Acemoglu, Daron, Simon Johnson, James A. Robinson and Pierre Yared
(2005) "From Education to Democracy?" American Economic Review Papers and Proceedings, forthcoming.
Acemoglu, Daron
torship
cind
James A. Robinson (2004) Economic
Origins of Dicta-
and Democracy, Cambridge University Press, forthcoming.
Acemoglu, Daron and Jaume Ventura (2002) "The World Income
Distribution,"
Quarterly Journal of Economics, 117, 659-694.
Anderson, Theodore
of
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.
Arellcino,
Manuel and Olympia Bover (1995) "Another Look at the Instrumental
Variables Estimation of Error-Component Models," Journal of Econometrics, 68, 29-51.
Barro, Robert
J.
(1997) The Determinants of Economic Growth:
Empirical Study, Cambridge;
Robert
Bcirro,
Economy,
J.
MIT
A
Cross-Country
Press.
(1999) "The Determinants of Democracy," Journal of
Political
107, S158-S183.
Bcirro,
Robert
J. £ind
Jong-Wha Lee
Attainment: Updates and Imphcations,"
(2000) "International Data on Educational
CID Working Paper #42.
32
Baxter, Marianne (1995) "International Trade and Business Cycles,"
Grossman and Kenneth Rogoff
eds.
Gene M.
in
Handbook of International Economics, Amsterdam;
North-Holland.
Bertrand, Marianne, Esther Duiio and Sendhil Mullainathan (2004) "How
Much
Shoiild
we Trust
nomics, 119, 249
Kenneth A. (1979)
Bollen,
American
Differences-in-Differences Estimates?" Quarterly Journal of Eco-
275.
-
"Political
Democracy and the Timing
of Development,"
Sociological Review, 44, 572-587.
Kenneth A. (1990)
Bollen,
"PoUtical Democracy: Conceptual
and Measurement
Traps," Studies in Comparative International Development, 25, 7-24.
Kenneth A. (2001)
Bollen,
1990," [Computer
file].
2nd
"Cross-National Indicators of Liberal Democracy, 1950-
ICPSR version. Chapel Hill, NC:
[producer], 1998. Inter-university
Consortium
for Political
University of North CaroUna
and Social Research
[distribu-
tor].
Central Intelligence Agency (2004) CIA World Factbook, Website and Book.
Washington, DC.
Curtin, Pliilip D. (1989) Death by Migration: Europe's Encounter with the Tropical
World
in the Nineteenth Century,
Cambridge University
Press,
New York NY.
Curtin, Philip D. (1998) Disease and Empire: The Health of European Troops in
the Conquest of Africa,
Cambridge University
Robert A. (1971)
Delhi,
Press,
New York NY.
Polyarchy: Participation and Opposition,
New Haven;
Yale
University Press.
Engerman, Stanley
Institutions,
H. Haber ed.
and
L.
Differential
How
and Kenneth
L. Sokoloff (1997) "Factor Endowments,
Paths of Growth among
New World
Economies," in Stephen
Latin America Fell Behind, Stanford; Stanford University Press.
Frankel, Jeffrey A.
and David Romer (1999) "Does Trade Cause Growth?,"
American Economic Review,
89, 379-399.
Freedom House (2004) Freedom
in the World, Website
and Book. Washington,
DC.
Gasiorowski,
Mark
J. (1995) "ISconomic Crises
Event History Analysis," American
Click,
Trade?:
Reuven
ctnd
and Pohtical Regime Change:
An
Political Science Review, 89, 882-897.
Andrew K. Rose
(2002) "Does a Currency Union Affect
The Time-Series Evidence," European Economic Review,
46, 1125-1151.
Gutierrez, Hector (1986) "La Mortalite des Eveques Latino- Americains aux XVIIe
et
XVIII
Siecles,"
Annales de Demographic Histonque, 29-39.
33
Haggard, Stephan and Robert R. Kaufman (1995) The
Political
Economy
of
Democratic Transitions, Princeton University Press, Princeton NJ.
Heston, Alan, Robert Summers, and Bettina Aten (2002) Penn World Tables
Version
Center for International Comparisons at the University of Pennsylvania
6.1.
(CICUP).
Hausman, Jerry (1978)
"Specification Tests in Econometrics," Econometrica, 46,
1251-1271.
Hausmann, Ricardo, Lant Pritchett and Dani Rodrik (2004) "Growth
erations,"
NBER Working
Paper #10566.
Huntington, Samuel P. (1991) The Third Wave: Democratization
Twentieth Century, University of Oklahoma Press,
International
Browser
CD-ROM.
International
and Browser
Accel-
Monetary Fund (2005)
in the Late
Norman OK.
Direction of Trade Statistics, Database and
Washington, D.C.
Monetary Fund
CD-ROM.
(2004) International Financial
Statistics,
Database
Wasliington, D.C.
Jones, Eric L. (1981) The European Miracle: Environments, Economies, and Geopolitics in
the History of Europe
Kraay, Aart and
Jaume
Section of Business Cycles,"
Lipset,
and Asia, Cambridge University
NBER Working
Seymour M. (1959) "Some
Development and
Press,
New
York.
Ventiu-a (2001) "Comparative Advantage and the Cross-
Political Legitimacy,"
Paper #8104.
Social Requisites of
American
Democracy:
Elconomic
Political Science Review, 53, 69-105.
Londregan, John B. and Keith T. Poole (1996) "Does High Income Promote
Democracy?" World
Politics, 49, 1-30.
Maddison, Angus (2003) The World Economy:
Historical Statistics,
Development
Centre of the Organisation for Ek:onomic Cooperation and Development, Paris, Prance.
Mcirshall,
Monty G. and Keith Jaggers
(2004)
"Political
Regime Characteristics
and Transitions, 1800-2002," Polity IV Project. University of Maryland.
McEvedy, Colin and Richard Jones (1975)
Facts on File, New York.
Atlas of World Population History,
Miguel, Edward, Shcinker Satycinath, and Ernest Sergenti (2004) "Economic
Shocks and Civil Conflict:
Economy,
An
Instrumental Variables Approach," Journal of Political
112, 725-753.
Moulton, Brent R. (1986) "Random Group
Effects
and the Precision of Regression
Estimates," Journal of Econometrics XXXII, 385-397.
North, Douglass C. (1981) Structure and Change
34
in
Economic
History,
New York;
\\'.\\'.
Xorton
k
Co.
North, Douglass C. and Robert P.
World:
A
New Economic
History.
Thomas
Cambridge
(1973) The Rise of the Western
Uiiiversit\- Press,
Cambridge UK.
Papaioannou, Ehas and Gregorios Siourounis (2004) "Economic and
Factors Driving the Third
PrzeTvorski.
Adam,
Wave
of Democratization"
IMichael Al\'arez, Jose A.
Cheibub and Fernando Limongi
(2000) Democracy and Development: Political Institutions and material well-being
world, 1950-1990,
Przeworski,
Facts," World
Cambridge
Adam and
Universitv- Press,
Social
mimeo, London Business School.
in the
Xew York NY.
Fernando Limongi (1997)
"Modernization: Theor}' and
Politics, 49, 155-183.
Rodrik, Dani (1999) "Democracies Pay Higher Wages," Quarterly Journal of Economics, 114, 707-738.
Schumpeter, Joseph A. (1950)
Harper
&;
Capitalism, Socialism
and Democracy,
Xew
York,
Brothers.
United Nations Population Division (2003) World Population
2002 Revision, Disk.
Vaart, A.
New
W. van
Prospects:
The
York.
der (1998) Asymptotic
Statistics,
Cambridge
University- Press;
Cambridge.
Wooldridge, Jeffery
AI. (2002) Econometric Analysis of Cross Section
and Panel
Data, IMIT Press; Cambridge.
World Bank (2002) World Development Indicators, CD-ROM and
ton,
DC.
35
Book. Washing-
.
Table
1
Descriptive Statistics
Former
High Income
Low Income
All countries
Colonies
Countries
Countries
(1)
(2)
(3)
(4)
Freedom House Measure
of Democracy,
Polity
0.78
0.36
(0.30)
(0.30)
0.57
0.50
0.80
0.36
(0.38)
(0.35)
(0.30)
(0.31)
8.16
7.85
9.02
7.30
(1.02)
(0.89)
(0.56)
(0.53)
GDP per Capita ^
(Chain Weighted 1996 Prices)
Log
0.47
(0.34)
Measure
of Democracy,
Log
0.57
(0.36)
GDP per Capita
Savings Rate
7.92
7.62
8.84
7.08
(1.05)
(0.93)
(0.61)
(0.55)
,_]
(1990 dollars)
0.17
0.16
0.22
0.11
(0.13)
(0.14)
(0.10)
(0.14)
,.2
Trade-Weighted World
Log
GDP
,.,
11.61
11.09
13.02
10.28
(8.43)
(9.30)
(9.83)
(6.60)
Constraint on the Executive
at
0.38
Independence
(0.35)
Independence Year
1911
(65)
Log Population Density
in
1500
0.41
(1.60)
Log
Settler Mortality
4.75
(1.20)
Observations
945
democracy, lag democracy, and lag income
A
in
in
472
473
541
Values are averages during sample period, with standard deviations
parentheses. All countries are those for
which
five-year intervals are available at least once during 1960-2000, for the
Column 1 refers to the sample in
Columns 3 and 4 split the sample in
column by the median income (from Penn World Tables 6.1) in the sample of column 1. Freedom House Measure of
Democracy is the Political Rights Index, augmented following Barro (1999). Polity Measure of Democracy is Democracy
Index minus Autocracy Index from Polity IV. GDP per capita in 1996 prices with PPP adjustment is from the Penn World
Freedom House measure of democracy.
the regression in Table 2,
column
1
.
country must be independent for
Column
2 refers to the sample in Table
at least 5 years.
1
1
column
1
.
1
Tables
6.1
;
GDP per capita in
World Tables
6.
1
and
is
1990 Geary-Khamis dollars
income (not PPP). Trade- Weighted World log
Statistics
(2005) and Penn World Tables
independence
(2002).
Log
is
is
from Maddison (2003). Nominal Savings Rate
is
from Penn
defined as nominal income minus consumption minus government expenditure divided by nominal
6.1.
GDP
is
constructed as in equation (7) using data from
Constraint on the Executive at Independence
from the CIA World Factbook. Log Population Density
Settler Mortality is
by European powers before
1
in
1500
is
is
from
IMF
Polity.
Direction of Trade
Year of
from Acemoglu, Johnson, and Robinson
from Acemoglu, Johnson, and Robinson (2001). Former colonies
900. For detailed definitions and sources, see Appendix Table
A
1
is
the subsample colonized
'
,
-o Ji
uL
J2
r^ ^
0^
O
o o
O
oo
oo
f-r^
,
'"'^
o
>» -o
—
—
= 1
3
o
O
g
r-'
-o
VO
-o
P
o
"5
.E
-a
ca
OJ
"u
?
E
a
s
bb
n
o
0^
O
£^
TD
iS
T3
ca
bO
XT,
ca
c
o
•g
<
E
o
c
T3
£
XI
aj
w^
"^
^a
OJ
u,
,o
^
c
1
1
E u^
;>^
c
>
—
K3
o
o
=
o
o
c
13
'E
60
c3
^
in
iS
ea
-T3
jl_
'w
o
'o
o"
-o
oo
M)
4
O
^
-g
r ^
>
,c
ca
Q.
T3
-p -a
vn
_5
n
s
c
'5
a
"
-S
o
o
o ^
3
o c
ov
b >
"u
1
t2
JL
3
c
u
u.
c
o
a
_3
Ills
S3
^ "*
o o
l/^
<~^
^ O
.2
^ —
—
I
E
&
Q
i2
•2
E 2 «
oo t^
rn tN
oo
s
S E o
«
O
u
o
«
S2
o
-s
-
S.
g
^
"
s
>
E
-a
3
E
n
bO VO
5
ca
m
X
a
ca
K
n
o
_S
IS
a
E
"o
1:
~
1
0CT\
T3
g
O
^
ov
—
'
(3s
o
-J"
X —
E
o
o
o
fcu
c
<
"o
J
o
S
=
-o
s
s
a
4-
C
a o
« o
S
>
03
c 3 -a E
lU
3
E
O
3
;J
< o u
Di
0. -a
ca
Sop
<
]5
_2 h2
o"
^
o
^
E
ca
^
•d-
3
c
e
3
s
_C
E
o
3 1
c
a>
n.
a.
<
oc
3
H
"ca
3
ca
'i
>
c
>
c
"O
-a
-a
3 c
oo
•g
3
a
X
S '3
O "B.
E
C
o E
~ 3
5
*^
CQ T3
JL
•o
00
i
-
^
•a
3
GO
I
n.
Q
c
n.
^
-a
~
ca
J^
u
en
o
V S
c 3
" O
3
i
""
T)
OO c2
o C
S
C
o
1
o m o o
"
d d>
£:>
-a
•a 'u
'^^
^ o
j3
g
"I
^
-g
is
S
g
-a
=
=«
.s
£
S
1-
3
E
E 12
•2
b?3
•
.S ?^
oo
c 2;
.9
^
T3
O
S °
*-
c/3
J
°
U-
c
<^
a
S
D-
^
iS
s
c^
<^
-.;
p
a\
r^
rt
^ ^
„ «
ON
T3
O
M
I
?
°
o
-g
S E 3
S
~i
Is
-a
.5
J
°
en
—
g S
VO
OT
'^
f~-
Q.
W J
i
ii S
XI -S
.a u
a ?
—
o 1
3
^
*=!
ON ^O
^
u
,"
^
S
2
«
-a
II
">
§
s
^
-t;
g
I.
f~.
-C
o
S
—
E E
>
2
S!
'5
^ X i
I
J^
c
c
a
-5;
3
S
8
0\
Qi
m
E
^r
-a
= " o
li
."
^
^
>
^
o o
5
S « §
°13
C T3 H
-a
.—
D a
^ o
< o U
pS'
a
E
—
c
13
CN
—
E S
'
00
^
t^
O
I"
^ E
< o i
l-l
>/^
S U
=3
<
c
CO
S —
o ^
o"
2 -a
n a ^
3 <
E I S c
_i 00
„
E S
_^
3
O
u
J3
tJJ
O
o
o
o
>o
CnI
o\
00 ,^
_J
o
CO"'
?
(U
't^
c
3
3
bO
CI.
u
s
u t:
5. 3
s s
=
o
Si
CQ
"O
OS O
d O ^ -^
<N
2
O
s
£
c:
CT\
^^ 1
3
^
ctT
u
VO'
rn (N
c
g
3
r-
•T3
>. _3 "o
.A 3^
ca
GO
X
(U
c"
J
^
C g
t-
B
c3
S
CQ
e
T3
|
S
&o
3
O
a:
E
o
-a
ID
tj
I/D
-J
c
_
3
O
u
o
o
c
CN
O
ON
<
_u
~"
o
•a
I
* ^
"
l3
"f
>^ -r
^
S
1J
o o 5
-a c
u,
,0
H
3
3
.S
=
-ir;
nl
^
^ B ^ 3
s i u
(U
'-
c
a
"d.
s
C
*w
3
GO
a>
"2
US
X)
3
•-
3^ Si
&G
E
CO
3
C
S
2
'S
§
5
O
-a
-5
rOn
^
(D
c
<a
cu
T3
<u
u
c
«
CO
ca
0
ra
o
CQ
S
*
t~~
\o
r:
O
E E
c3
(U
eg
ly^
^
3
o
-1
-
.S
DO
"S
">
-a
Ji
ex
S
-2
S
g
S
T3 -^
r-
i2
H
.a
"
O O
ci
tS
_
00
a
CQ
2
13
-5
C
>•!=
T3 -a
a 3 C
•
,
_2
O
<
°
—
U.
c>
^
-
c
'J
u
V —
_c:
c
5
a
.52
o
a.
'-n
^
(u
ca
1
^
m
o —
^
_3
O
to
C
O
1^
C
o
W J
O
2
o
wo
•a
§
a c
o
2
o
O 3
^ ^ a.
ob
2
DO
gu
S
"3
_2
X
=
-a
_g
.g is
On
>^ -^
—
^i
o
a
g
-a
j5
O
<
§
c
tr,
3
O
I ^
1
o
3
s
.S
s 1
O "S >
>
c
^ ac
P-1
T3
C
1
-a
o
t:
a
1
a.
<
a
1o ^u
— H
"2
^CL
-a
'B
a.
§
o
c
52
.2
o
Q. -§ '2
IS
3
o
•a
la
a
•a
C
,o
OJ
o- OS '3
1-^
1
o c
e
o
00 _3
-
=
B
H
1
a
—
-a
«
1 c 3
-a T3
o
s
-o
c
J
O
o
;
<
"ca
3
rt
<
ca
.2
o
00
a
bJ)
^B
0\
OS
<u
gg
_o
o
-a
00
3
O
o
-a
c
o
&•
_a
T3
^
03
o
S
"c5
:i
C
o
^ qo 3 a.
ON
-^
m
T3
-o
«
e
o"
G
a
-c
iX"
U o
c
S
3
C
a
o
t-^
CL
-a
^
[
>N o<
t2
1
J2
S-
^
:s
I
= i as
- c
00
4-.
II
^ I
I
^
-5
«
=
_^
D
rt
O
I-
1-
s-
—
o
,-^
_2
-f-
g
g
-i
U
cu
Q
O
bX)
3
O
Oh
tuO
CQ
O U
Qi
o
M
a
o
m3 t—
<
3
o
t~~
c
^,
fS
S
^
>0
r^
CN
^
°^
ca
3
l
00
^
;z.
cN
;:^
CN
ro C^
00 -o
IZ)
J
O
t
T3
c
-o X)
_3
c
o
1
to
-a -a
c _D
(U
i2
i
£
c
ca
_n
3
r;^
^
£
3
"K
^
Er
"H
1
^
^
CT\
J-
tj
TD
1
=1
-D
o
-a
13
-o
O
r-"
j:2
c
c
DO
nl
t
_3
"c
S
i.
aj
<
1 1
-a •5
f J
M^
1 u
T3
1
1 ^ oC c
o -C
o.
u s ^ ^
c
Tt
Q
O
.2
b -oc
C
5
a.
a. "Y
G
-C -a
w
C
^5
1?
"o
£
S
5
^ U
o
CL
.c
.ti
'5.
-O -a
£
£
;^
c
E
U
g
.§
>.
c
3
12
1
-2
n
>^
i
1
Q
5 J " S >^
^
S S 2
- M
IN a. -5 _.
^ H
(vT
c 2"
X
<:
C
i2
w
:z
c
=
t;;
"
I
I
a. -B
ai.
:E
cd
i.
<u
^
O
>,
U <
'
9 o
_
fN
—
t
U
-a
C ^
£ S
On
"
C
->
-a
S
m
c ^"
=
:§
-§ =?
&
o (N (N
^ pj
^ lO =»
—
o o
a\
M
VO
—
—
'
^ <
•*
S H —
1 i IS
^ = H
I
« c
"
si"
111
?l
"S
>
:S
E
<
5 S
a o
fN
"Al
^^
rsl
"
5
Si!
to
P3
^i
^
^ ^ ^
E S '^
y.
°-
5 £ S
S "
^ "
J2
5
II
5
ov
=
<
o
o
r^
-S i^
O
g
^
,?
I
J
"^ XI
I
-^
^ ? f ^ ^ I
rj
o
—
m o in
(N
CO —
o o
°1
^
3
J
?^
S
J^
o
S
2i
5
«
>ri
2
s^ E^
E u
||
S
u
—
_5J
I I
I i|
"I
O M_
o J <
-J
g
3 !^ -^
/-;
ci
(N
M-
OO
—
^
m
s.? ? «
:;
On
iC
ON (N <^
oo --^
s
o
a
T3 >.
.2
_!
^- :S
"i
I
:
I
i
^
-i
o g
E
c
o w
-c
.2^
u:
C
3
=
u
H 3>
r7 C
?
x>
u
li
-ii
E
'.
S
o M
8
5 5
ffl
*-
-^
Table 9
Polity
Measure of Democracy
in the
Long Run
'
5-year data
Arellano-Bond
Fixed Effects
Fixed Effects
Arellano-Bond
OLS
GMM
OLS
OLS
GMM
(2)
(3)
(4)
(5)
(6)
OLS
Pooled
(1)
Dependent Variable
Panel A: All Countries
Democracy
Log
,.,
GDP per Capita
Hansen
10-year data
Fixed Effects
t.i
is
Democracy, 1840-1940
0.841
0.652
0.433
0.455
0.230
(0.039)
(0.089)
(0.121)
(0.136)
(0.238)
0.072
-0.076
-0.183
-0.109
-0.035
-0.092
(0.017)
(0.064)
(0.090)
(0.107)
(0.135)
(0.287)
J Test
[1.00]
AR(2) Test
[1.00]
[0.40]
[0.05]
358
358
297
370
188
142
Countries
28
28
27
28
28
28
R-squared
0.78
0.81
0.69
0.75
0.772
0.682
0.443
0.407
0.287
(0.043)
(0.070)
(0.096)
(0.082)
(0.104)
Observations
Dependent Variable
Panel B: All Countries
Democracy,.!
Log
GDP per Capita
t.j
is
Democracy, 1840-2000
0.081
0.066
-0.041
0.135
0.134
0.176
(0.015)
(0.032)
(0.033)
(0.072)
(0.070)
(0.112)
Hansen J Test
AR(2) Test
[1.00]
[1.00]
[0.83]
[0.93]
Observations
656
656
578
682
342
Countries
28
28
28
28
28
R-squared
0.79
0.80
0.63
0.70
0.771
0.673
0.427
0.400
0.244
(0.045)
(0.074)
(0.099)
(0.085)
(0.105)
Depen\dent Variable
Panel C: Without Venezuela
Democracy
Log
,.]
GDP per Capita
,.)
is
285
,
28
Democracy, 1840-2000
0.078
0.041
-0.058
0.104
0.076
0.099
(0.015)
(0.030)
(0.034)
(0.079)
(0.068)
(0.147)
Hansen J Test
AR(2) Test
[1.00]
[1.00]
[0.91]
[0.96]
637
637
562
662
331
276
Countries
27
27
27
27
27
27
R-squared
0.78
0.80
0.63
0.69
Observations
Pooled cross-sectional
columns
2, 4,
and
5,
OLS
regression in
column
I
,
(1991), with robust standard errors; in both methods
variable
is
Polity
with robust standard errors clustered by country
we
instrument for income using a double lag.
Composite Democracy Index. Base sample
is all
country must be independent for 5 years before
and
(i.e.,
C use
1
it
enters the panel.
the start date
A
1
.
OLS
regressions in
Column 3 and 6 use GMM of Arellano and Bond
Year dummies are included in all regressions. Dependent
Columns
of the panel refers
5 and 6 use
t=1840, so t-l=1830); a country must be independent for 10 years before
840-2000. Panel
Appendix Table
parentheses. Fixed effects
countries with at least 5 observations in the 5-year data between
which are independent by 1900. Columns 1-4 use 5-year data where
variable
in
with country dummies and robust standard errors clustered by country in parentheses.
C drops Venezuela. GDP per capita is
it
1
to the
dependent variable
0-year data where the
enters the panel. Panel
start
(i.e.,
1
840 and 1900 and
t= 1840, so
t-l
= 1835);
a
date of the panel refers to the dependent
A uses a long sample,
1840-1940, Panels
from Maddison (2003). For detailed data definitions and sources see Table
1
and
B
age
> i ^
P o
9 o
u:
X
(N
9 o
o S^
9 o
P o
KJ
*?
tJ)
c3
—
m
— ro
1
ly-i
=
C
i:
o
J^
o
"^
o
9 o
5^
J^
•^
9 O
9 o
c
c
cd
>>
O,
C3
to
U
c
Q
a.
c
0. .o
Q
a
o
12
^
x"
T3
_C
^
t3
g3
o
3 O
jO
O
c
D.
S
'm
Q
M
O
y
-a _g
.2
.
.•^
g 3
" -a
o
-a
<
o
•a
£
—
>•
-
c
.S
"
"o
o
=
15
S
£
t2
-c
-a
c
E
o
*
c
^
c
"
^
o
u
O
Di
U
°
•
'^
Q
=
S
i
-
18
5
- o
.£
-S
'
o
if:
o
m
^ o
\o
^
(-S,
--
^
—
o
vo
^^
"
C
O
U
C
_
S
°
S
C ^ ^ ON
s i ^ "
-a
^
E
?-
'-t;
— —
,
^
r~^
^
^
m
,^
"^
o
a
S ^
o
(U
(U
j:^
C^
OS
j= -^
C/5
O
U^
o
°
"
cj
ji
en
r
J3
-
Q
C
R
c 3
O
c
O
"5
.
xc "
^
C
-a
u
3
S
3
'3
k;
"tS
P
3
0)
_C3
S
"*
n!
«-l
c3
g
r
ca
=
n
o
>
\£i
T3
w
M
-s
C
u G
3 c
TS tr .P "3
E
a
n
^
U 2
W
CD
t~^
T3
>
u
ca
hn >-
a
-§
s
g
ID
i
^
§ ^
lO "^
o
2i
o
O
c
^
.5 "^
2 "=
U
J" ^
C3
«
i-.
Q %
Q
U
,
o
^
C
-a
^
XI
cd
C O
u .5
>
S w
3 E
5
U
c
3
-DO.
O U
Pi
o
o
fl.
£ —
3 o
w
ll
_U
>-
>
>. '^
-° R
n.
OJ
-a a, -a
H
^
O
u
o
<
11
Table 12
Fixed Effects Results for Crises, using Freedom House Measure of Democracy
Base Sample 1965-2000
Fixed Effects
Arellano-Bond
Fixed Effects
Arellano-Bond
Fixed Effects
Arellano-Bond
OLS
GMM
OLS
GMM
OLS
GMM
(1)
(2)
(3)
(4)
(5)
(6)
Crisis tlireshold
-3%
is
Crisis threshold is
Panel A
Dependent Variable
Democracy
Log
,_]
GDP per Capita,.,
Crisis
,.1
0.493
0.356
0.495
0.354
0.496
(0.090)
(0.056)
(0.089)
(0.055)
(0.089)
-0.003
-0.141
0.003
-0.148
0.009
-0.155
(0.047)
(0.088)
(0.047)
(0.091)
(0.047)
(0.092)
0.044
0.057
0.059
0.063
0.074
0.096
(0.019)
(0.024)
(0.021)
(0.026)
(0.025)
(0.029)
[0.41]
[0.46]
[0.58]
[0.73]
[0.57]
[0.74]
0.79
Log
GDP per Capita
Crisis
,.1
,_!
Hansen
R-squared
0.697
0.609
0.698
0.607
0.699
(0.039)
(0.062)
(0.038)
(0.062)
-0.004
-0.067
0.001
-0.071
0.005
-0.079
(0.034)
(0.062)
(0.034)
(0.064)
(0.034)
(0.066)
0.038
0.045
0.045
0.051
0.049
0.067
(0.014)
(0.016)
(0.016)
(0.018)
(0.020)
(0.021)
Log
,.,
Hansen
[0.04]
[0.04]
[0.23]
[0.35]
[0.18]
0.89
Dependent Variable
GDP per Capita,.,
Crisis
[0.05]
0.89
(.]
is
0.89
Transition
away from Democracy
0.744
0.796
0.747
0.796
0.747
0.797
(0.038)
(0.054)
(0.038)
(0.054)
(0.038)
(0.053)
0.001
-0.074
0.002
-0.076
0.003
-0.076
(0.023)
(0.057)
(0.022)
(0.058)
(0.022)
(0.058)
0.006
0.012
0.014
0.013
0.025
0.029
(0.011)
(0.014)
(0.011)
(0.015)
(0.011)
(0.017)
Test
J
Democracy
(0.062)
C
Democracy
Transition toward
0.602
AR(2) Test
Panel
is
(0.038)
Test
J
0.79
0.79
Dependent Variable
t-i
-5%
Democracy
0.347
Panel B
Democracy
Crisis tlireshold is
(0.056)
Hansen J Test
AR(2) Test
R-squared
is
-4%
AR(2) Test
[0.14]
[0.16]
[0.51]
[0.47]
[0.21]
[0.48]
R-squared
0.93
Observations
739
659
739
659
739
659
Countries
124
122
124
122
124
122
Fixed effects
2, 4,
and
6,
regressions.
OLS
with country
0.93
dummies and
robust standard errors clustered
with robust standard errors; in this method
The dependent
variable in Panel
A
calculated as max((^,„rfj,.|) from this Index, and
sample
(i.e.,
is
in
the
in the
t-l
we
at
columns 5 and 6 the threshold
is
-5%. See
In
in
I,
3,
and
C
Transition
Political Rights Index, in
away from Democracy,
5-year intervals in levels where the
5;
1
and 2 the threshold
is
lag.
Year dummies
are included in all
calculated as min((f jy^i,.,) from this Index.
start
-3%,
GMM of Arellano-Bond in columns
Panel B, Transition towards Democracy,
in
Base
date of the panel refers to the dependent variable
dummy
for 15 years before they enter the panel. Crisis is a
columns
text for details.
columns
instrument for income and for cnses using a double
= 1960). Countries must be independent
growth rate by more than a threshold.
by country
augmented Freedom House
Panel
an unbalanced panel, 1965-2000, with data
t=1965, so
change
is
0.93
columns
3
variable corresponding to a
and 4 the threshold
For detailed data definitions and sources see Table
1
is
-A%, and
and Appendix Table Al.
in
-
)
^
PJ
.0
J
^
LJ
(J
5
3 5
O
j:
CJ
G
T3
03
^
C
ON
u
^
.2
S
3
2
Q. ^
s >
•rl
bn
"5b T"
<
(LI
r
)
Tt
<r
c^
(1)
5
rON
o
O
<
O
H
R
3
1S
bj)
03
o
-g
w
g
C8
O
> M
I
^
s
g
i!
B
SP
o
O
o
o
in
o
Q
S
U
S
03
S-^J
bO
Sg
5
n.
o
(^
r^
O
~
CQ
2 ^
12
o
5
o
10
03
i!
CQ
P
:S
Q
i2
c«
3
"hn
ID
03
Qi
J=
u
5> °
•:=
'a:
Q
o
Cu
E
u
Q
n
a
(U
>:
<u
03
S)
T3
S
2
03
S
§
•-
U
0)
(U
_2
r-
—
CTv
c
o
bD
C
(D
ON
03
0)
o
s
.s
t»:J
mum-
.-^
K
ON j:
^
i^
<J
.23
.u^
'^^
Q ^
i—
>-.
t:
03
< ^ cd. .5^ Q
^
g
°
^ U
03
03
'
D
9
o
1^
^
s
r o a
O
<U
:?
—
>
3 «
w c
^
Q
jr:
Q.
l2
T3
Cu
'
r
^
b
^ I
=
o
0NO
a. ON
On
ID
s
r-i
."
.S2
V 1
9 s
u
^-^
Q
T3
.
CO
T3
H
0)
c
-G
X
ID
U-
ON
T3
3
^
Q.
g
°
CQ
.^
?3
^'-
"S
==>
^~
13
lU
J
"5
3
(D
'^
ly-i
t
3
o.
o
S3
u
^1
3
(D
(N
_
Cn)
P
p
"
•5^
ON
•-^
-a
_2
13
-o
^
ti;
S -a
c
(L>
I
m
<
5
<
>
^
IO §O
ail
c
PJ
hJ
1)
d
(1)
of On
.2
.5
03
13
U
^""
S
ra
03
(U
bO
^"
e
•=
>
•13
03
03
:§
-a
C3
P
oj
>,
>
c o o
r^
CQ
o
OJ
i;
.>
C
y^
O
T3
u
u
-S
03
c
>
O
JS
t\j
•—
>-
JIZ
^
0)
4
?
s
:
'
CQ
s
c
Q
Q
H
U
^
t4^
(U
o
S ^
]>l
T3
O
o
u5
ly.
t/)
CU
U
3: -^
=«
sis?.
a
c^
.-s
G.
•2
[S
a
T3
o
o
-a
o
o
C
^
t"
OS
t^
^
l2
?J
.S?
<u
-?^
^C ^
£ ^
-C
C M
O aj
13
oo
a>
•-
5
5i
n
^
e
^
"^
vo
H
^
<=>
—
c«
(U
2
^
^
r
^ 2 o
1^
3 b
'*-
«
i2
-
^
(V)
U O
(D
-O
b/l
O o
to
"S
p o
o -a
~
C _.
E
H—
^•-
^
53
tun
on
-^
>.
c3
^
2 -a
'*C (U
o c ^ £
u .2 -^ °
-a
1= "S
2
<
o
C
u
c
—c
c
ts
^:>
2 ^
fij
£2
-S
oi
-u
^
£2
C
^
U
:C
i
.S <o
T3
3
Ki
O
i:
O
tj
*-:
rt
«
-S
§
-55
«
C
I
C
oj
><
jj
X^
•+-
o
CO
c
o n. S
2
-^ o
5 13 o
==
.
?^
&H
>
£
.S
—
--£-!'"
_
td
M
2
—
H
.5 v^
3
X)
H
en
?^
-^
c
tQ
m
<
(V
<•
i:
^
ca
-:;;
o
*^
S
j2
I
o- i
'i^
•-a
^
5
p
5
ON
oo
o
S
Q < U « —
ca
^
=s
O
C
.S
~
° ^
o S c
-^
o
^
y
?
ra
r;
_- ^U
^
o -5a
Si
o
"^
1.5
£;
t/l
rti
I
3
X
c .2
JS .2 "O
u-
^
>
-a
C3
H Q
2
f-
0.
Q
O
<"
•^ -5
£
n)
Q <
.2?
c
''K
UJ
K
T3
i1>
>i
1^
ca
cu
C3
i=
'
^
n
Q
H O
t:;
-73
!^
r-
o C
S ^
O ^
U
MO ^
S ^ -u
^ - ii
tJjJ
>
u.
C«
t<
3
(D
(1J
tJG
W
T3
.S "o
''3
-^
t5
U
-s
3
>
"t;
c3
CO
G
o
CO
>-
o
— N
oo
O,
•
'
CN oo rn
o
m
^
o
r~~
<-;
rq
(N
O
m o
rn ^
~
o o
rn
_,
-^
C
CTs
_D
°
g
"
—
»^
-O
—
2
•?
-.
CQ"
o
t-^
^
w J
in
^ O
00
m
—
= o JJ
o "- 5
o
-a
(X
•
5
-s
=;
u
I>
ra
-;
- oO
o „S
-3
2
~
P
^
c/i
u
hJ
p
-o
>-
(U
-^
„
S
,S
-^
u
CO
.5
u
g
CO
*-.
n.
P3
^
ca
^
~
X-
E
^
T3
o
CI
-o
m
(U
F-^
O
00 -H
"S
.a
tin
5
p 8
E
§
* ^ S I b
O o -E p
M ="
P
0)
^
';^
p
c
S
S
ON
-a
(N
00
O
(N
-^
00
-E
.5
ca
o
^o|
Q oi
2
g
E
oo
—
JJ
1-
-a
P
-p
«
E
dj
^
w
E
^
00
-J
Q
O
QC
o
J
U
en
1—5
K
c
S3
s'
X)
a,
Q
ffi
O
P
>
—3O
^
^
™
3
5
'n
p
c3
C3
1)
ffl
M"
cd
2
g
"
<;
H
;^
5
c^
P
•-
D.
§" XI
°?
'^
CL
o
S
—P
J"
I
"
<
•=
°
^
"
5;
w
u
-a
o
B
O
H
°
u
-g
Q
S
3
J s
s o
o
S
>o^
—
ca
•
E
.2
>-
" ^
i3
.H
u
^
O
X < o U
'S
r
nf
-o
p
H
13
JL
p
5;
::;
U
^
°;^'l
:5 -^
E
p
00
R
S^s
-^
O
^
c/T
-^
-E X!
g
f^
—
-«
S "S
s
^ ^
^
°
P
""
P
o
S, §
(X '§
>. g Q
3
a " "-^ "K
o ^ SP p
O A? 9 °
—
m
m —
oo ^^
0^
C
^
-P
0-
m
—
§
P
u y
a.
"O
"^
-fi
—bo
^
>,
(L>
Si
G
3
(U
g
OJ
f^
W
I
< O U
E
o E
o D
o E
"
>.
O. T3 XJ
c
(-1
nr
m <
-a
rt
I
J-
u
o
8
£:
03
^C S
^
Oh
E
Q
:2h
o
ac
.
"
£:-
3
u
-a
c
IS
fc
T3
S
ex
<:
-a
ca
s
a.
i2
E
U
>.
2;
H
E
3
—
13
;::;:
S
-
ao -O
',^
-„
O
-"
"O
-73
15
00
° o
o o
m
n.
-
.^
-
=2
—
OS
f^
u
C
„-=*-«
u
^
E
ii :S
u
?i
=
w
Z3
CO
^
o
3
ca
5
U rS
.S iS
c
^
^
-ii
o
E •=
>0
^
vo
=5
t^
t:
>'
O ,0
C3
Q
,X
'^3
ca
J
5j
U t S ^
Z o ^
&>
o
S
-a
-S
^m
3 3
-iS
E
"5!
T3
^
1
-a
u
,0 3 c
ca
3 ^
,0 C3 c
^•
"
3
,
^
« •"
i
^c ^ ^C
tn"
-?;
.0
hJ
•g
"ca
"vi
c
U
>-
c
K3
-3
ca
>.
lA
3
o
u
O U
Di
E
5
c
a.
,
•a
OS
o>
>.
C
u
-a
c
ca
aJ
.2
t;
>
S-M
XJ
.S
U
—
.
5 <
Appendix Table
Codes Used
Andorra
Code
ADO
Ghana
Afghanistan
AFG
Angola
AGO
Albania
A4
Represent Countries in Figures
to
Code
GHA
Country
Code
Netherlands
NLD
Guinea
GIN
Norway
NOR
Gambia, The
GMB
Nepal
Guinea-Bissau
New Zealand
Oman
Argentina
ARG
Greece
Armenia
ARM
Grenada
Antigua
Guatemala
Azerbaijan
ATG
AUS
AUT
AZE
Croatia
GNB
GNQ
GRC
GRD
GTM
GUY
HND
HRV
NPL
NZL
United Arab Emirates
ALB
ARE
Burundi
BDI
Haiti
HTI
Poland
Belgium
BEL
Hungary
HUN
Benin
BEN
BFA
Indonesia
India
IDN
IND
Ireland
IRL
QAT
Iran
ROM
Iraq
IRN
IRQ
Romania
Bahrain
BGD
BGR
BHR
Qatar
Russia
RUS
Bahamas
BHS
Iceland
ISL
Rwanda
RWA
Bosnia and Herzegovina
BIH
Israel
ISR
Saudi Arabia
Belarus
BLR
BLZ
BOL
Italy
ITA
Sudan
Jamaica
JAM
Senegal
SAU
SDN
SEN
Jordan
JOR
Singapore
Japan
JPN
Solomon
Kazakhstan
Kyrgyz Republic
KAZ
KEN
KGZ
Sierra
Bhutan
BRA
BRB
BRN
BTN
Botswana
BWA
Cambodia
KHM
Sao
Central African Republic
CAF
Kiribati
KIR
Suriname
Canada
Slovakia
SUR
SVK
Korea, Rep.
KNA
KOR
Slovenia
SVN-
Kuwait
KWT
Sweden
China
CAN
CHE
CHL
CHN
Lao
Swaziland
SWE
SWZ
Cote d'lvoire
CIV
Lebanon
Cameroon
CMR
Liberia
Congo, Rep.
Colombia
COG
COL
LAO
LBN
LBR
LBY
LCA
Togo
Comoros
COM
LIE
Thailand
TGO
THA
Cape Verde
CPV
Sri
LKA
Tajikistan
TJK
Costa Rica
CRI
Lesotho
Turkmenistan
TKM
Cuba
Lithuania
Tonga
Cyprus
CUB
CYP
LSO
LTU
Luxembourg
Trinidad and Tobago
Czech Republic
CZE
Latvia
LUX
LVA
Germany
DEU
Morocco
Turkey
Djibouti
DJI
Moldova
Taiwan
TWN
Dominica
DMA
Madagascar
Denmark
DNK
Maldives
MAR
MDA
MDG
MDV
TON
TTO
TUN
TUR
Dominican Republic
DOM
Mexico
MEX
Ukraine
Algeria
DZA
ECU
EGY
Macedonia,
Spain
Country
Australia
Austria
Burkina Faso
Bangladesh
Bulgaria
Belize
Bolivia
Brazil
Barbados
Brunei
Switzerland
Chile
Country
Equatorial Guinea
Guyana
Honduras
Kenya
St. Kitts
and Nevis
FDR
Libya
St.
Lucia
Liechtenstein
Lanka
OMN
Pakistan-post- 1 972
PAK
Pakistan-pre-1972
PAK_1
Panama
Korea, Dam. Rep.
PAN
PER
PHL
PNG
POL
PRK
Portugal
PRT
Peru
Philippines
Papua
New
Guinea
PRY
Paraguay
El Salvador
SGP
SLB
SLE
SLV
Somalia
SOM
Islands
Leone
Tome
and Principe
Seychelles
Syrian Arab Republic
Chad
Tunisia
STP
SYC
SYR
TCD
Tanzania
TZA
Uganda
UGA
UKR
URY
MKD
Uruguay
Mali
MLI
United States
Malta
MLT
Uzbekistan
ERI
ESP
Myanmar
MMR
Mongolia
MNG
Estonia
EST
Mozambique
Ethiopia
ETH
Mauritania
ETM
Mauritius
Finland
FIN
Malawi
Fiji
FJI
Malaysia
MOZ
MRT
MUS
MWI
MYS
France
FRA
Namibia
NAM
South Africa
ZAF
Gabon
GAB
GBR
GEO
Niger
NER
NGA
Congo, Dem. Rep.
ZAR
Nigeria
Zambia
Nicaragua
NIC
Zimbabwe
ZMB
ZWE
Ecuador
Egypt, Arab Rep.
Eritrea
East
Timor
United Kingdom
Georgia
FYR
USA
UZB
St. Vincent and the Grenadines VCT
Venezuela, RB
VEN
Vietnam
VNM
Vanuatu
VUT
Western Samoa
WSM
Yemen
YEM
YUG
Yugoslavia
-
post 1991
CM
in
w
o
I
(1)
JQ
T3
CO
1
1-
.1
u
"S
3
T3
1-
2
o
^^
c
c
I
T-
D)
u
S 2^
0\
Q^
-o
TO
"q.
o
O
O
""
c
M
CL
c:
w
u
&
>-
CO
D) -o
LL
c
OJ
•
r--
=J
"S
O
U £
Q)
ro
>^
1 H
-D
(U
^
Q.
CL
£
s|
c
iJ
Is
"
o
2 —
CO
i—
o
o
E
u>
o
_l
_c
a
"
J-
c §
s
^
"3
Q)
C
CD
^
o
C
ii
s
*"
'5
CD
^s
i N
H
o
1 1 ?
^
1"»^
O
^—
1
5
n
I
AoBJoouuaa
1
p
ajnsB9|/\|
9asnoH Luopaaj j
r
^
<
-g
:i
s
^Bue^Q^
--'
C
.
3
II
z
e s?
^
^
T3 C
I
u;
2
g
o o
h|
1
i
9"
i3
•o
u
a-a
(^ S
< f
°
I
2
^
"
^S
«
"g
CD
CD
1—
CD
CD
IT3
C
C
Qi
Q.
Q)
Cvl
D)
C
TO
CO
o
O
T3
i_
o
C
Q.
(0
Q_
o
(0
l_
o
o
£
O
Q
0)
C
(D
O
c
CD
sz
O
I
S"
AoBJOOLUsa io
98jnse8|/\| A;!|Od uj
l-
sBueqo
w
J)
SI
CD
h-
c
c
CD
Q.
„
00
E
Q.
CD
O
Q.
Q.
Q
O
D)
O
hr--"
- CD
1
~i
I
I
I
8'
AoBJOoLusQ
9"
p
9jnsB9|/\|
\
p-
Z
ssnoH Luopsajj
r
>
2
3 E
o a
in
00
IT)
G>
o
B|"«i
- 00
00
c
o
" B
'^
C/)
D
^
111
lO.^
S^f
T3
CO
^
S
c
1 1
w
o
a
c
CNJD.
CD
(D
s=5
O
m
L_
0)
^
O
Q.
Q_
.2^ -D
Q
O
CM Q)
O
O
_l
CD
i_
o
o
£
3
S
£
O
D.
O
C
ii
^
c
1 1
fa
1
S
J^
-o
g
3
•a
<2
^-
a
u
£ ^
t4_
t3
C
1)
^ c
Ri
.9
-c
(U
':
s
^
<
"
S
D3
C
Q
CD
SL
inO
"is-
C
-a
~~1
1
1
I
8"
9"
AoBJOOLuea \o
p-
T"
Z
ajnsea|/\| Aiijod uj
96ueqo
o
t^
l\l
- -^
co.^
Q.
ro
S o °
ti2
^
a.
Q S
O ^
-S
XI
-13
"^
o
O
.>
c
-^
u
li
:§
u
O
i_
0)
Q.
Q.
Q
O
CM ro
O
_l
c
D)
C
05
•^
(S
o
-
i
2i
c -
^
U
tS
C
-a
o
ES
i3
rt
o
ob
.£
u
u
^ S
i-i
ij
a.
O
< £ ^ I
2-
p-
9"
AoBJOOLueQ
ui
96ueqo
Z
•
(f)
CD
O)
o
CD
QJ
_J
Dc
CD
•4—
O
LJ_
s
W
>+-
LU
O)
o
T3
Q)
X
L
o
ns
i_
o
o
Q
a <
CN
f-
E
"
^
paj^3 psxy Aoejooaiaa
J'
o
o
in
-
•'d-
w
c
Q
c
g
o
o
_ro
D
Q.
O
w
c
Q.
Q)
13
LJ_
(U
Q
D)
O
c
o
_1
DC
o
:3
Q.
o
o
CL
^Z
LU
N
2
O
•D
0)
X
L
^
O
o
£
(D
Q
<
_ CM
0)
o
c
0)
Dc
<u
Q.
Q)
-a
c
>
T3
C
o
<D
X
Q.
LU
T3
C
0)
CO
(U
=3
TO
;—
LL
W
c
O
o
X
LU
o
•
T3
c
03
+-•
o
i
LU
T3
- CM
X
L
CD
i_
O
o
Q
pajj3 paxy AoejooLuaa
c
o
c
5
w
c
o
o
o
CM
03
<U
>in
0)
o
c
oc
o
o
0)
Q.
<U
(D
ac
uc
CO
>0)
n3
OT
5-
o
C
(D
T3
LL
c
t:
LU
0)
Q.
T3
0)
(D
T3
X
C
Ll
o
03
o
o
E
CO
i
<D
Q
00
paj43 paxy AoejooLuaa
1
-
CM
a>
S s
s
in
>-
o
o
in
I
O
(U
o 2
o o
—C LLCO
c _
CD
^ O
m
.ti
S)
^
ro
u
0)
ro
ON
—
a.o
T3
_
Q-
1—
>
o
"^
M
o c
Ou.
5?,
Q
^
o
w
oOi
D o
O)
Ll
CO,
J2
O
Q-O
CO
Q < O
D)m-
,-&
"C
o o
zz
LUQ
ro
rt
8
%
:2
_5
^-
§ §
s >
III
^'
X)
-o
c
_!*-
j3
.eS
ill
o
Is-
g
^
c
2 ^
-
c
s
n
g
-5
CD
i_
o
o
E
^
CO
B ^
£
S
«3
z
111
<D
Q
O)
c
CD
O
- CM
9"
sjopej leouojsjH
AoBJooujaQ
P luapuedapui
U!
sBueqo
3
I £-
a
*^
Date Due
Lib-26-67
MIT LIBRARIES
3 9080 02618 2722