The Democratic Transition
1870-2000 *
Fabrice Murtin – Romain Wacziarg
OECD – UCLA, NBER and CEPR
March 2010
PRELIMINARY AND INCOMPLETE – PLEASE DO NOT QUOTE
*
Murtin acknowledges financial support from the Mellon Foundation when he was hosted by the Stanford
Centre for the Study of Poverty and Inequality. The findings, interpretations, and conclusions expressed in
this paper are entirely those of the authors and do not necessarily represent the views of the OECD.
1 Introduction
In its “Bill for the more general diffusion of knowledge” (1779), Thomas Jefferson
envisioned that education would be a cornerstone of democracy, as he argued that “the
most effectual means of preventing [tyranny] would be, to illuminate, as far as
practicable, the minds of people at large, and more especially to give them knowledge of
those facts, which history exhibiteth, that, possessed thereby of the experience of other
ages and countries, they may be enabled to know ambition under all its shapes, and
prompt to exert their natural powers to defeat its purposes”. This vision has been largely
shared by other observers throughout history. For instance, Tocqueville (1835) noted that
the vigor of democracy in the United States was resting partly on mass education, which
was already a leading country in terms of educational attainment at the time (see Lindert
(2006) and Morrisson-Murtin (2009)). He argued that “the education of the people
powerfully contributes to the maintenance of the democratic republic. That will always
be so, in my view, wherever education to enlighten the mind is not separated from that
responsible for teaching morality” 1 .
Is there any robust empirical foundation to the argument developed by Jefferson,
Tocqueville and others? Is the later argument specific to the United States or does it
describe a general experience shared by other countries? Can the expansion of education
be seen as a trigger of the democratic transition, and what is the magnitude of its potential
effect, if any? Is the direction of causality from education to democracy unidirectional or
is there a mutually beneficial relationship between the two?
1
Volume 1 chapter 9, “The main causes that tend to maintain a democratic republic in the United States”.
This paper focuses on the two-way relationship between education and democracy using
a large panel of countries starting in 1870. On a first step, it addresses the link from
economic development to democracy, while concentrating on two dimensions of socioeconomic standards, namely GDP per capita and average years of schooling among the
adult population. On a second stage, the paper looks at the potential effect of democracy
upon child education as measured by average years of schooling completed by young
cohorts.
The issues raised by this paper relate to a large and controversial literature on the dual
relationship between economic development and democracy, which can be organized as
follows. A first body of literature investigates the relationship from economic
development to democracy. Continuing a stream of idea that goes back in time as far as
Aristotle, Lipset (1959) argued in a seminal article that improvements in economic
standards would ultimately lead to democratization, or, quote, that “the more well-to-do a
nation, the greater the chances that it will sustain democracy”. Lipset’s “modernization
hypothesis” has then received empirical support from Barro (1999), who isolated among
other variables GDP per capita and primary schooling as positive determinants of
democracy in a large sample of countries spanning from 1960 to 1995. However, in a
series of studies Acemoglu et al. (2005, 2007, 2008) have revisited and contradicted the
empirical significance of the later two major determinants of democracy. Once country
fixed-effects are accounted for, GDP per capita is no longer a significant determinant of
democracy. More recently, Bobba and Coviello (2007) as well as Castello-Climent (2008)
have re-examined the evidence. The former isolated a significant effect of GDP per capita
while using a SYS-GMM estimator, while the later found a significant effect of education
attained by the majority of population, even after controlling for fixed-effects.
Conversely, a second strand of literature has looked at the consequences of
democratization on economic performance, and in particular, on the expansion of
schooling. Generally speaking, democratization entails mixed economic outcomes. For
instance, Barro (1997) finds a nonlinear effect of democracy on growth, with an initial
increase in growth and a negative relation once some modest level of democracy has been
reached. Tavares and Wacziarg (2001) assess a positive effect of democracy on human
capital accumulation and a negative one on physical capital accumulation, while Rodrick
and Wacziarg (2005) find a positive short-term effect of democratization on the flow of
investments and a decrease in growth volatility. More specifically related to education,
Lindert (2004) documents how the extension of the franchise in rising European
democracies has gradually led to the introduction of public funding for education over the
XIXth and XXth centuries. In contrast, Murtin and Viarengo (2010) do not find any
effect of democracy on the amount of compulsory years of schooling in Western
European countries after 1950.
Adopting a larger perspective, the two-way relationship between education and
democracy is at the core of an important debate on the ultimate sources of economic
prosperity. Does education help raising the quality of institutions as well as productivity,
or is an efficient institutional framework a pre-requisite condition for the blooming of
education and economic growth in general? The direction of causality between education
and democracy has ignited a debate between the advocates of institutions as the prime
engine of growth (Acemoglu and Robinson (2000), Acemoglu et al. (2001, 2002, 2005)),
and the tenants of the opposite thesis that views human capital as the origin of economic
development (Glaeser et al. (2004, 2007), Glaeser-Campante (2009)). Deciphering the
respective effects of democracy and education upon each other would therefore shed
some light on this larger and crucial debate.
The contributions of this paper are threefold. Firstly, taking stock of Morrisson and
Murtin (2009) data set on educational attainment in 74 countries since 1870, the paper
finds strong empirical support for the modernization hypothesis over the XXth century. It
shows that primary schooling, rather than GDP per capita, has been the major trigger of
global democratization over that period. Then, the paper addresses the reverse causality
issue and finds no robust evidence of the effect of democracy upon child education.
Next section describes the empirical facts related to democracy and development over the
XXth century. Section three tests the Modernization Hypothesis in light of new
educational data, while Section four addresses the reverse causality issue. Last section
concludes.
2 Democracy and Development 1870-2000
The following section describes the sample of countries and takes a bird’s-eye view on
the main structural transformations that have marked global economic history over the
twentieth century. As differences across countries are central to our analysis, we focus on
convergence effects and disentangle three particular periods: 1870-1910, 1910-1960, and
1960-2000. This corresponds roughly to the two globalization periods as well as the
deglobalization period of the interwar period. In practice, the full sample is composed of
seventy countries spanning over all continents, while there are nineteen countries forming
a balanced panel that starts in 1870. Table 1 provides elementary descriptive statistics
that we comment below.
2.1 Democracy
Our main measure of democracy is the Polity IV index developed by Marshall and
Jaggers (2004), which is rescalled between 0 (pure autocracy) and 1 (pure democracy).
This index captures consists of six component measures that record key qualities of
executive recruitment, constraints on executive authority, and political competition. It
also records changes in the institutionalized qualities of governing authority. We plot the
world distribution of scores of democracy on Figure 1 (unbalanced panel) and Figure 2
(balanced panel). Boxplots represent the 25th and 75th percentiles as well as the median of
the global distribution. These two Figures illustrate an overall increase in democratic
attainment between 1870 and 1920, followed by a marked decrease until Second World
War. The immediate postwar period witnessed a new rise in democracy that has lasted
until the end of our period in the balanced panel. In the global panel assembling all
observations, democracy has stagnated or even regressed between 1960 and 1980 but has
rised again since then. The world distribution of democracy has widened from 1930
onwards, while at the end of the period, cross-country differences in democracy have
been dramatically reduced.
The later aspect is related to global inequality in democracy: Has democracy converged
across countries over time? The full sample shows that democracy has indeed converged
within any of the three periods described above, and at almost the same pace. Indeed,
when calculated on an annual basis 2 , the speed of convergence of democracy happens to
be equal to 4.7% over the 1870-1910 period, 3.4% in 1910-1960 and 3.7% in 19602000. 3 The first panel of graphs at the top of Figure 3 illustrates those calculations.
How to reconcile convergence in democracy with the increase in cross-country
democracy inequality observed between 1930 and the mid-1980s? Convergence entails
inequality reduction if shocks affecting democracy are stationary over time. But as made
clear by Figure 1 and Table 1, the occurrence of two World Wars and the associated
political turmoil have significantly enlarged the distribution of democracy. Hence, the
distribution of democracy has been framed by a convergence effect, but heteroskedastic,
time-specific, shocks have considerably widened it.
2.2 Income and Schooling
Economic development can be proxied either by the level of GDP per capita or by
average years of schooling among the adult population, the later two variables being
highly correlated across countries and over time. In this study, GDP per capita is taken
from Maddison (2006) while average years of schooling is provided by Morrisson-Murtin
(2009).
There is an extensive literature that describes the global evolution of key aggregate socioeconomic variables across countries over the XXth century, and it is obviously much
2
The annual convergence rate is given by –log(ρ)/T where ρ is the estimated coefficient of initial
democracy in an absolute convergence regression, and T is the length of the period over which the
difference in democracy is calculated.
3
There are respectively 34, 39 and 59 countries involved in the latter computation. In the balanced panel,
the results are almost identical: a convergence process has taken place in the 1910-1960 and 1960-2000
periods at an annual rate of respectively 3.2% and 3.6%. In the first period, convergence occurred but only
among a club of advanced democracies.
beyond the scope of this paper to describe and comment these transformations. 4 Let us
simply recall a few facts, mirrored by elementary statistics provided by Table 1: in broad
terms, modern economic growth has taken off in Western Europe and offshoots in the
first half of the XIXth century, and has spread to Asia and Latin America in the eve of the
XXth century as well as to Africa after Second World War. In terms of education, the
Western world had the quasi-monopoly of schooling in 1870, but Eastern Europe as well
as some fast-developing Asian countries, most notably Japan, have caught up along the
XXth century. A polarized schooling distribution in 1870 has been transformed in a
substantially heterogeneous distribution in 2000: On top, high-income countries reaching
mass scolarization in primary and secondary schooling with growing enrolment in
tertiary education; in the middle of that distribution, Latin America, the Middle-East and
North-Africa as well as most of developing Asian countries achieving mass scolarization
in primary and partly in secondary school; at the bottom, South Asia and Sub-Saharan
barely achieving mass scolarization in primary schooling.
In terms of differences across countries, economic development does not reflect the
convergence in democracy described above. As well-known, income inequality across
countries has increased throughout the XXth century, except among a convergence club
of relatively high-income countries before First World War and after Second World War.
Similarly, there has not been any convergence in average education at any period, except
among two clubs of advanced countries during the first and second globalization periods.
4
A shortlist of references could include Bourguignon-Morrisson (2002), who describe the world income
distribution since 1820. Galor and Weil (1999, 2000) and Galor (2005) analyze the joint variations of
income and population over the long run as well as the structural forces that have triggered the Industrial
Revolution. O’Roorke and Williamson (1999) and Hatton and Williamson (2005) focus on the effect of
globalization on economic performance. Morrisson-Murtin (2008, 2009) describe the spread of education at
a global level. Murtin (2009) investigates the determinants of the demographic transition over the XXth
century.
These two facts are clearly illustrated by the second and third panels of graphs on Figure
3. As a sum, democracy and economic development do not exhibit any common growth
pattern at any time, as the former has always converged across countries while the latter
has displayed dissimilar time variations across levels of wealth. Next, we examine the
correlation between the two variables in a more quantitative fashion.
2.3 The Joint Distribution of Democracy and Development
Figures 4 presents a kernel estimation of the bivariate distribution of democracy and
average years of schooling within four different sub-periods: 1870-1970, 1920-1940,
1950-2000 and 1980-2000. 5 As similar graph can be obtained with GDP per capita
instead of education, and results are qualitatively identical. Some interesting facts emerge:
at the eve of the twentieth century, the bivariate distribution of democracy and schooling
was clearly bi-modal, splitting rich democracies on one hand from poor autocracies on
the other. Over the next sub-period, a large share of initial autocracies witnessed
economic growth and joined the club of democracies at the same time, ensuring a strong
positive correlation between democracy and income. After 1950, and even most visibly
after 1980, there have been two distinct groups of unequal size, poor autocracies on one
hand and rich democracies on the other. Among the latter, the correlation between
democracy and income has flattened out as countries had already reached high levels of
democracy in 1950 while experiencing continuous economic growth.
5
We have used an Epanchenikov kernel with bandwidth adjusted to the finite sample size.
4 Testing the Modernization Hypothesis
4.1 Democracy and Income
In his original framework, Lipset (1959) hypothesized that economic development would
lead to the emergence of democracy. As a convenient starting point, a natural
econometric framework that tests this hypothesis can be written as follows
D(i,t) = a(i) + b(t) + c X(i,t-1) + u(i,t)
(1)
where D(i,t) is an index of democracy in country i at time t rescaled between 0 and 1, a(i)
and b(t) are respectively country-specific and time-specific effects, X a variable that
proxies economic development such as log GDP per capita or average education, and u a
residual. We introduce the first lag of this variable in the regression, and as a benchmark,
consider a 10-year time span between subsequent observations.
Alternatively, one may also include the lagged value of the dependent variable inside the
regression. A rationale for this is that the political structure of a country changes slowly
over time, as the introduction or the modification of laws, constitutions and the changes
in political regimes meet resistance. Consequently, the score in democracy is likely to be
a time persistent variable. Classically, autocorrelation in the dependent variable is taken
into account within a dynamic panel model
D(i,t) = a(i) + b(t) + ρ D(i,t-1) + c X(i,t-1) + u(i,t), |ρ|<1
(2)
which can be estimated with GMM methods developed respectively by Arellano-Bond
(1995) (henceforth AB) and Blundell-Bond(1998) (henceforth BB).
In Table 2 we focus on log GDP per capita as an explanatory variable and we disclose the
estimates of (1) and (2) over the whole period as well as over three sub-periods, 19602000, 1930-2000 and 1870-2000, which enable us to gauge the influence of time-specific
sample selection. As regards the choice of countries, we use first the whole sample of
countries, then a sample that excludes countries already at the maximum level of
democracy at initial year, and finally the balanced panel of countries. Thus, Table 2
examines the relationships between democracy and development using a full range of
econometric procedures, time-periods, and country samples.
The overall conclusion is straightforward: among 36 specifications, there are 27 in which
lagged income has a positive and significant sign. Let us now detail the results by type of
sample.
In the first part of Table 2, we use the full sample of countries. Using a fixed-effects
estimator (henceforth FE), lagged income is not significant over the 1960-2000 period as
emphasized by Acemoglu et al. (2008), but is significant in other sub-periods and highly
significant over the whole period. Most likely, a weak instruments problem affects AB
estimates, as none of them is significant. This is the reason why BB techniques have been
introduced in the literature, and BB estimates are all highly significant and identical in
magnitude across all sub-periods. This confirms a result found by Bobba and Coviello
(2007). Moreover, the BB estimation passes all robustness tests even if a limited number
of instruments has been chosen in order to avoid instruments proliferation problems. 6
A feature that could explain the non-significance of some estimates in the previous
exercise is sample selection. Countries already at the “democratic frontier” at initial date
are unlikely to experience large decreases in democracy, even if this fact has been
6
As described by Anderson and Sorenson (1996), Bowsher (2002) and Roodman (2009),
instruments proliferation generate implausibly low values of Hansen tests of instruments
exogeneity. This is because the size of the variance matrix of the moments is too large to
be estimated accurately within a finite sample. Limiting the number of instruments
therefore limits that problem. Besides, we use Windmeijer (2005) finite sample
correction of standard errors in order to increase robustness.
observed several times in history, notably after First World War. 7 To gauge the influence
of this group of countries, it is convenient to run the former regressions while excluding
the countries that were already at the maximum level of democracy at initial date. As
shown by the second part of Table 2, excluding these countries from the analysis has a
large influence over the results, as the 1960-2000 fixed-effects estimates and the two
most recent AB estimates become significant.
Then, sample composition could further affect our results as the sample does not remain
fixed over time. In practice, countries join the sample at their date of independence, but
many young countries have experienced erratic political processes, including declines in
democracy in some cases. Their inclusion in the sample could create compositional
effects and spuriously affect the long-term interpretation of our results. In the last part of
Table 2, we therefore restrict the sample to a balanced panel of nineteen countries
observed since 1870. Within this sample, lagged GDP per capita appears to be a strongly
significant determinant of democracy in ten specifications over twelve. In the FE
regression pertaining to the 1870-2000 period, the coefficient of lagged income appears
to be equal to 0.23. This implies that doubling GDP per capita entails closing 16 percent
of the gap between the worst dictatorship and the best democracy. 8 Recalling that
countries from the balanced panel had an average GDP per capita of about $ 1500 in
1870 and $ 14 600 in 2000, started at a level of democracy equal to 0.44 and ended up at
a level equal to 0.95, this means that economic development accounts for virtually all of
7
In statistical terms, this raises the issue of the measurement of democracy, which is proxied by a bounded
variable. Even if some countries have already converged towards the maximum reported level of
democracy at initial date, institutions have most likely kept on evolving, most likely on improving, within
these countries.
8
As 0.23 x log(2)=0.16
the progress in democracy within that sample. 9 However, this interpretation should be
taken cautiously as point-estimates differ across econometric procedures and because
specification tests of AB and BB estimations are not fully satisfactory.
Our findings differ from Acemoglu et al. (2008) who show that the relationship between
democracy and lagged income is statistically insignificant when calculated over the XXth
century, and turns significant only over the very long term, say between 1500 and 2000.
Figure 5 best illustrates their finding. It shows the seemingly null correlation between the
change in the Polity IV score of democracy and growth in GDP per capita between 1900
and 2000.
This divergence in results can be firstly explained by the difference in the time span used
across the two analysis. Acemoglu et al. (2008) consider a 25 years time span, while the
present study focuses on a decennial time span. Suppose that we were using a larger one,
say 30 years, then we would also find that lagged income is insignificant. Note that this
finding would be mostly explained by a smaller sample size. Indeed, regressing
democracy on log income lagged 30 years, but using a 10 years time span as in our
study 10 , we would still find a highly significant coefficient for lagged income 11 . Whether
a 10 years or 25 years time span is more appropriate is left at the appreciation of the
reader, although simple econometric intuition would plead for a larger sample size.
Furthermore, the non-significant correlation depicted on Figure 5 can be easily explained
by the omission of the initial level of democracy from the analysis. Indeed, the pairwise
correlation between growth in GDP per capita and the change in democracy can be
9
As calculations show that 0.23xlog(14600/1500)/(0.95-0.44)=1.02
Therefore using a sample of 456 observations and 67 countries rather than a sample of 156 observations
and 61 countries.
11
Using log GDP per capita lagged 10 years while using a 30 years time span, we also find a significant
coefficient for lagged income at a 10 percent confidence level.
10
spuriously contaminated by unobserved variables. Among those, the initial score of
democracy is a potential candidate, as convergence in democracy has taken place over all
periods as described earlier. The negative correlation between the change and the initial
level of democracy might therefore contaminate the relationship between the change in
democracy and per capita GDP growth. Table 3 presents a simple set of regressions
where the dependent variable is the change in democracy between 1900 and 2000, while
explanatory variables include per capita GDP growth (column I), the latter variable and
initial democracy (column II), the latter two variables and initial log GDP per capita
(column III). As made clear from these simple cross-country regressions, per capita GDP
growth is highly significant once the initial level of democracy is included inside the
sample as in columns II and III.
This suggests that the non-significance of per capita GDP growth on column I is in fact
an omitted variable bias arising from the exclusion of the initial level of democracy.
Interestingly, both the initial level and the growth in GDP per capita have a significant,
positive, association with the change in democracy over the period, suggesting both
accumulation and level effects 12 .
Let us now examine our findings obtained with a different proxy for economic
development, namely education.
4.2 Democracy and Education
In this section, we run the same analysis while replacing GDP per capita by average years
of schooling of the adult population (population older than 15 years).
12
On a different issue, Aghion, Howitt and Murtin (2010) find positive causal effects of both the change
and the initial level of life expectancy on per capita GDP growth, which correspond respectively to Lucas
and Nelson-Phelps interpretations of the effect of human capital on economic growth.
First, the second part of Table 3 describes the same cross-sectional regressions of change
in democracy between 1900 and 2000 on the corresponding change in mean years of
schooling among the adult population, initial democracy and initial average years of
schooling. We find qualitatively the same results. The change in schooling is not
significant on column IV – although the p-value equals 0.11 – and is significant at a 1
percent confidence level on columns V and VI once other explanatory variables are
added. Controlling for the whole set of income and education-related variables as in
column VII, we find positive and significant coefficients for the changes in log GDP per
capita and average schooling, as well as for initial levels of log GDP per capita and
democracy.
Coming back to a higher data frequency that might ensure more robustness, we report on
Table 4 some results that can be compared to those depicted in Table 2. In this setting, we
estimate equations (1) and (2) using various samples and econometric methods, replacing
log GDP per capita by average years of schooling.
Overall, we find mixed evidence as education is significantly associated with democracy
in only 15 specifications over 36. As regards the full sample of countries, education
seems to be weakly associated with democracy, as found by Acemoglu et al. (2005),
when using FE methods. It is indeed only significant at a 10% level over the 1870-2000
and 1900-2000 periods. Also, it is never significant with the AB estimator, but strongly
significant with BB.
As regards the following parts of Table 4, the former results are not modified by the
exclusion of initial democracies from the sample. Over a balanced panel, though, few
coefficients remain significant, and when so, they are only weakly significant or display
implausibly high values such as the AB estimates. At first sight, the relationship between
democracy and education seems therefore weaker than that prevailing between
democracy and income.
However, this result can be explained quite easily. Large progress in democratic
achievement is often achieved when countries are still at their infancy in terms of
educational development. In other words, increased political participation might rather
involve the transition between illiteracy and literacy than further developments of
secondary schooling and higher education, which take place in already mature societies.
Consequently, what matters for democracy may rather be the average number of years of
primary schooling of the adult population than its average total number of years of
schooling. Table 5 investigates this issue, first controlling for average years of primary,
secondary and tertiary schooling independently, then controlling for all three variables.
As in former Tables 2 and 4, most of Hansen specification tests failed for AB regressions,
we rule out the AB estimator in the remaining sections of the paper, and focus on both the
FE and BB estimators.
Firstly, Table 5 shows that average years of primary schooling is a strongly significant
determinant of democracy in both FE and BB regressions, and across all periods. All
specification tests are valid and point-estimates of primary schooling are consistent
across all regressions. Secondary and tertiary schooling deliver more mixed results when
taken individually. FE and BB estimates are strongly significant over the 1930-2000 and
1960-1960 periods, but not over the whole period. Secondary schooling delivers better
results than tertiary schooling as at least BB estimates are significant over any period
with satisfactory specification tests. Plausibly, as higher education has reached mass
enrolment rates in high-income countries over the last quarter of the XXth century where
democracy was already well-established, higher education can hardly explain the
emergence of democracy.
Last part of Table 5 presents one of the main findings of the paper. When disaggregating
average years of schooling into its three main components, it turns out that across all
specifications and time-periods, primary education is highly significant with a constant
coefficient close to 0.05 and valid specification tests. In contrast, secondary and higher
education are almost never significant in any regression.
4.3 Channels of Democratization: Income or Primary Education?
The former section has given empirical flesh to two highly discussed issues of the
political economy and political science literatures: Quantitative evidence demonstrates
that economic development, more precisely log GDP per capita and average years of
primary schooling, are strongly significant determinants of democracy. This result is
obtained after controlling for fixed-effects, persistence in the dependent variable and after
instrumenting by a reduced number of lagged explanatory variables.
Table 6 investigates which of the two variables log GDP per capita and average years of
primary schooling has the highest explanatory power using the full sample of countries.
The results are not fully conclusive, but we tend to think that primary education is a more
robust determinant. Indeed, it is significant in 7 specifications over 8, while lagged log
GDP per capita is significant in 5 regressions over 8. Focusing on FE regressions, lagged
income and primary education are both highly significant over any period, except in the
1960-2000 period for income. Then, focusing on a BB estimator, we find satisfactory
specification tests in all cases, and a significant coefficient for primary schooling over
any period except during the 1930-2000 period. In contrast, lagged GDP per capita is
significant in BB analysis only in periods starting after 1930. Over the whole period in
particular, it fails to reach significance contrary to primary schooling. The coefficient of
the latter variable is equal to 0.06, entailing an increase in 0.06x6 = 0.36 points of the
democracy score when the country shifts from complete illiteracy to full literacy
(assuming a duration of 6 years for primary schooling). As in the balanced sample, the
average score of democracy has increased by 0.51 points and average primary schooling
by 3.5 years between 1870 and 2000, this means that 0.06*3.5/0.51 = 40% of the average
variation in democracy in high-income countries between 1870 and 2000 can be
explained by the increase in primary schooling and associated literacy achievements.
4.4 Robustness Analysis
In the following, we provide some robustness analysis. Do our main results still hold over
the two globalization periods (1870-1910 and 1960-2000) and the deglobalization period
(1910-1960)? Table 7 focuses on the benchmark BB analysis and clearly shows that
primary education has been the key determinant of democracy in all sub-periods. In terms
of magnitude of the effect, it is comparable across all three sub-periods. Interestingly,
income has not been a significant determinant of democracy until the postwar period.
Furthermore, in unreported results we found that our results are unchanged when using a
different democracy index, namely the Freedom House indices of political rights and
civil liberties over the period 1960-2000.
Next section focuses on the reverse causality from democracy to schooling.
5 From democracy to schooling?
In theory, it is possible that education and schooling influence each other without
hampering the identification of this two-way relationship. Indeed, one direction of
causality involves the stock of schooling among adult population as a determinant of
democracy, while the other direction involves the effect of democracy on the flow of
schooling, namely schooling enrolment among younger cohorts. As the education
variables differ across the two analyzes, inference remain possible.
In this section we explore the impact of democracy on schooling enrolment. We relate the
level of democracy at date t to average years of schooling of cohorts of pupils aged
between 5 and 14 at the same date. This group starts with pupils at the beginning of
primary schooling and closes with pupils at the middle of secondary schooling. Ten years
later, this cohort gathers pupils aged between 15 and 24 years, which are possibly at midsecondary or at the end of tertiary schooling. In other words, the level of democracy
taking place between two subsequent observations of our panel affects the whole
spectrum of pupils at school, which is well captured by the cohort aged between 5 and 14
at initial date.
In practice, average years of schooling of the latter cohort is estimated as the value of the
cohort aged between 25 and 34 years twenty years later. This value is taken from data on
education by age borrowed from the same sources than those underlying MorrissonMurtin (2009) for the period 1870-1960 and Cohen-Soto (2007) after 1960. We explain
in details in annex 1 the construction of this variable.
Other variables than democracy may affect enrolment at school. In particular, past
enrolments might capture a sum of persistence factors ranging from religion, cultural
traits, public schooling policies or imitative behavior. Then, variables capturing the level
of development, such as average schooling of the adult population or log GDP per capita,
are expected to be positive determinants of the level of enrolment, for several reasons: a
simple income effect with wealthier parents investing more in child education,
substitution between quantity and quality of children in the course of economic
development (Becker (1981), Galor and Weil (1996, 2000), Murtin (2009)), higher life
expectancy that increases the lifetime benefits reaped from schooling investments
(Kalemzi-Ozcan et al. (2000), Soares (2005), Cervellatti-Sunde (2005)) and so on.
Thus, accounting for the later factors we run the dynamic panel model
E(i,t) = a(i) + b(t) + ρ E(i,t-1) + α D(i,t) + β S(i,t) + γ y(i,t) + u(i,t), |ρ|<1
(3)
where E(i,t) is average years of schooling completed by the cohort aged between 25 and
34 twenty years later, D(i,t) the score of democracy, S(i,t) average years of schooling of
population older than 15, y(i,t) log GDP per capita, a(i) and b(t) respectively country and
time fixed-effects. In particular, this specification allows to test whether α > 0, namely
whether improvement in democracy leads to increases in educational attainment of young
cohorts, conditional on past increases, global trends or improvements in living standards.
Table 8 presents the results of OLS estimates (imposing a(i) = 0 and ρ = 0), panel fixed
effects (ρ = 0) and BB estimation, controlling or not for the level of development. Two
sets of regressions are run, one spanning over the whole period, the other focusing on
1960-2000. In order to eliminate some persistence in residuals, all variables have been
first-differenced in the BB regressions, and as a result, Arellano-Bond autocorrelation
tests on residuals and the Hansen test of joint exogeneity of instruments are satisfied in
all cases.
Overall, mixed and inconclusive results arise regarding the potential role of democracy
for human capital. Over the 1870-2000 period, democracy is a seemingly significant
determinant of average years of schooling attained by young cohorts in OLS and FE
regressions in columns I and III, but it is no longer significant once the level of
development is controlled for as in columns II and IV. In BB regressions, democracy is
never significant (columns V and VI). Over the 1960-2000 period, democracy is
significant in OLS regressions but switches sign when other variables are introduced. In
BB estimation, it is significant and positive in all cases (columns XI and XII). As a sum,
democracy is a positive and significant determinant of schooling attained by young
cohorts in 5 specifications over 12. Hence, we do not find strong and robust evidence that
democracy leads to a larger quantity of years of schooling.
6 Conclusion
Schumpeter (1943, p.296) argued that “modern democracy rose along with capitalism,
and in causal connection with it (…). Democracy in the sense of our theory of
competitive leadership presided over the process of political and institutional change by
which the bourgeoisie reshaped, and from its own point of view rationalized, the social
and political structure that preceded its ascendancy (…). Modern democracy is a product
of the capitalist process”. This paper provides empirical support to this view, as it clearly
establishes an empirical link from the level of development, be it the level of GDP per
capita or average years of primary schooling among the adult population, to democracy.
Investigating the reverse causality from democracy to the educational attainment of
young cohorts, the paper finds mixed evidence of a positive and significant link. Over the
long term, this relationship does not hold empirically. However, a case might be made
for public spending in education across types of political regimes. Democracies may not
entail larger quantity of schooling among young cohorts, but higher quality, say, in the
form of larger teacher-pupils ratio, or bigger expenditure per pupil at each stage of
education. We leave that question opened for future research.
7 References (TO BE COMPLETED)
Acemoglu, D., Johnson, S., Robinson, J.A. and P. Yared (2005). From Education to
Democracy?. American Economic Review Papers and Proceedings, vol. 95, no2, pp. 4449.
Acemoglu, D., Johnson, S., Robinson, J.A. and P. Yared (2008). Income and Democracy.
American Economic Review, vol. 98, no3, pp. 808-842.
Anderson, T.G., and B.E. S_renson. 1996. GMM estimation of a stochastic volatility
model: a Monte Carlo study. Journal of Business and Economic Statistics 14: 328-52.
Arellano, 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.
Blundell, Richard & Bond, Stephen, 1998. "Initial conditions and moment restrictions in
dynamic panel data models," Journal of Econometrics, Elsevier, vol. 87(1), pages 115143, August.
Bowsher, C.G. 2002. On testing overidentifying restrictions in dynamic panel data
models. Economics Letters 77: 211-220.
Cervellati, M. and U. Sunde (2005). Human Capital Formation, Life Expectancy, and the
Process of Development. American Economic Review, vol. 95(5), pages 1653-1672.
Galor, O. (2005). The Transition from Stagnation to Growth: Unified Growth Theory.
Handbook of Economic Growth, North Holland, 2005, 171-293
Galor Oded and David N. Weil, "The Gender Gap, Fertility, and Growth," American
Economic Review, 86, 374-387.
Galor, O. and D. Weil (2000). Population, Technology, and Growth: From the
Malthusian Regime to the Demographic Transition and Beyond. American Economic
Review, 90, 806-828.
Glaeser, E. and F. Campante (2009), Yet Another Tale of Two Cities: Buenos Aires and
Chicago, NBER Working Paper.
Glaeser, E., Ponzetto, G. and A. Shleifer (2007). Why does democracy need education?.
Journal of Economic Growth, vol. 12(2), pp. 77-99.
Glaeser, E., La Porta, R., Lopez-de-Silanes, R. and A. Shleifer (2004). Do Institutions
Cause Growth?. Journal of Economic Growth, vol. 9(3), pp.271-303.
Hatton, T.J. and J.G. Williamson (2005). Global Migration and the World Economy: Two
Centuries of Policy and Performance. MIT Press.
Kalemli-Ozcan, Sebnem & Ryder, Harl E. & Weil, David N., 2000. "Mortality decline,
human capital investment, and economic growth," Journal of Development Economics,
Elsevier, vol. 62(1), pages 1-23.
Maddison, A. (2006). The World Economy. OECD, Paris.
Morrisson, C. and F. Murtin (2008). Education Inequalities and the Kuznets Curves.
Working Paper PSE.
Morrisson, C. and F. Murtin (2009). The Century of Education. Journal of Human Capital,
Vol. 3 (1), pp.1-42
O’Rourke, K., and J. G. Williamson (1999). Globalization and History: The Evolution of
a 19th Century Atlantic Economy. Cambridge, Mass.: MIT Press.
Roodman, D.M. 2009. A note on the theme of too many instruments. Oxford Bulletin of
Economics and Statistics, vol. 71(1), pp.135-158.
Windmeijer, F. 2005. A _nite sample correction for the variance of linear e_cient twostep GMM estimators. Journal of Econometrics, vol. 126, pp. 25-51.
Annex
10
8
6
Democracy Index
4
2
0
−2
−4
−6
−8
−10
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Figure 1 - Distribution of Democracy Index Over Time (Unbalanced Panel)
10
8
6
Democracy Index
4
2
0
−2
−4
−6
−8
−10
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Figure 2 - Distribution of Democracy Index Over Time (Balanced Panel)
Figure 3 - Convergence in Democracy, GDP per capita and Average Years of Schooling by Period
10
1920−1940
Democracy Score
Democracy Score
1870−1910
5
0
−5
−10
10
5
0
−5
−10
10
5
10
15
Average Years of Schooling
1980−2000
Democracy Score
Democracy Score
5
10
15
Average Years of Schooling
1950−2000
5
0
−5
−10
10
5
0
−5
−10
5
10
15
Average Years of Schooling
5
10
15
Average Years of Schooling
Figure 4 - The Joint Distribution of Democracy and Schooling by Period - Kernel Estimation
1
THA
Change in Democracy Score 1900−2000
0
.5
BGR
MEX
TUR
HUN
IRN
AUT SWE
DNK
PRT
NLD
BRA
ITA
VEN
JPN
ARG
CHL
BEL
ESP
GBR
PER
−.5
FRA
CAN
1
1.5
2
2.5
Change in Log GDP per capita 1900−2000
3
1
Figure 5 – Change in Democracy and Economic Growth 1900-2000
SWE
DNK
NLD
AUT
BGR
MEX
TUR
HUN
IRN
PRT
NIC
ITA
BRA
PRY
VEN
JPN
SLV
ARG
CHL
ESP
GBR
BEL
PER
HND
CAN
FRA
ETH
−.5
Change in Democracy Score 1900−2000
0
.5
THA
GTM
0
1
2
3
4
5
Change in Average Years of Primary Schooling Among Population 15+ 1900−2000
Figure 6 – Change in Democracy and in Average Years of Primary Schooling 1900-2000
Table 1 - Descriptive Statistics
Full sample
1870 1910 1960 2000
Balanced sample
1870 1910 1960 2000
0.42
0.65
0.25
0.60
0.44
0.73
0.25
0.70
Democracy
average
coefficient of variation
p25
p75
0.58
0.54
0.35
0.80
0.58
0.64
0.15
1.00
0.77
0.36
0.65
1.00
0.61
0.56
0.30
0.95
0.64
0.63
0.15
1.00
0.95
0.09
0.90
1.00
Income
average
coefficient of variation
p25
p75
1520 2910 3810 8470
0.66 0.48 0.82 0.98
740 1690 1390 1430
2000 4060 6070 16010
1490 2950 6110 14600
0.53 0.50 0.56 0.48
720 1790 3070 7220
1880 4200 8750 20320
2.27
0.86
3.40
0.73
4.13
0.73
7.29
0.45
2.53
0.75
4.09
0.64
5.95
0.46
9.51
0.25
1.77
0.85
2.60
0.68
3.04
0.64
4.47
0.33
2.02
0.73
3.18
0.58
4.24
0.35
5.49
0.12
0.48
1.17
0.77
1.06
1.00
1.10
2.44
0.68
0.49
1.02
0.87
0.99
1.57
0.83
3.43
0.46
0.02
1.22
0.04
1.02
0.09
1.15
0.38
0.89
0.02
1.09
0.04
0.90
0.14
0.96
0.59
0.65
27
26
59
69
19
19
19
19
Average Years of Schooling
Total
average
coefficient of variation
Primary
average
coefficient of variation
Secondary
average
coefficient of variation
Tertiary
average
coefficient of variation
Number of countries
Table 2 – Democracy and GDP per Capita 1870-2000
FE
1870-2000
AB
BB
FE
1900-2000
AB
BB
FE
1930-2000
AB
BB
FE
1960-2000
AB
BB
0.08
(0.06)
0.26
(0.10)**
0.11
(0.16)
0.32
(0.10)***
0.19
(0.03)***
275
68
45
0
0.71
0.04
307
69
56
0
0.58
0.23
Dependent variable is standardized Polity IV index of democracy
Full sample
D(-1)
Log y(-1)
N
N countries
N instruments
AB1 p-value
AB2 p-value
Hansen p-value
Excluding initial
democracies
D(-1)
Log y(-1)
N
N countries
N instruments
AB1 p-value
AB2 p-value
Hansen p-value
Balanced Panel
D(-1)
Log y(-1)
N
N countries
N instruments
AB1 p-value
AB2 p-value
Hansen p-value
0.27
(0.11)**
0.14
-0.09
(0.04)*** (0.17)
0.41
(0.10)***
0.15
(0.03)***
0.25
(0.12)**
0.16
-0.06
(0.04)*** (0.19)
0.36
(0.11)***
0.16
(0.03)***
0.19
(0.10)**
0.15
0.14
(0.05)*** (0.14)
0.32
(0.11)***
0.18
(0.03)***
567
69
560
69
83
0
0.28
0.75
512
69
505
69
78
0
0.3
0.66
424
69
417
69
73
0
0.22
0.42
309
69
0.25
(0.11)**
0.15
-0.09
(0.04)*** (0.18)
0.41
(0.10)***
0.16
(0.03)***
0.26
(0.11)**
0.17
0.00
(0.05)*** (0.23)
0.38
(0.11)***
0.16
(0.03)***
0.17
(0.11)
0.20
0.40
(0.06)*** (0.18)**
0.31
(0.10)***
0.19
(0.03)***
0.15
(0.09)*
0.26
0.41
(0.08)*** (0.15)***
0.29
(0.10)**
0.21
(0.03)***
541
67
534
67
83
0
0.27
0.81
450
63
444
63
78
0
0.37
0.81
309
54
306
54
73
0
0.36
0.93
216
50
215
50
56
0
0.71
0.63
-0.02
(0.23)
0.23
0.21
(0.07)*** (0.58)
0.08
(0.25)
0.09
(0.02)***
-0.01
(0.26)
0.30
0.27
(0.08)*** (0.60)
0.09
(0.24)
0.38
(0.63)
0.41
(0.22)*
0.36
0.87
(0.10)*** (0.90)
0.41
(0.29)
0.49
(0.68)
0.31
(0.14)**
0.33
0.67
(0.12)*** (0.34)**
0.38
(0.20)*
0.23
(0.13)*
247
19
247
19
83
0.17
0.94
1
209
19
209
19
78
0.17
0.91
1
152
19
152
19
73
0.05
0.64
1
95
19
95
19
56
0.06
0.24
1
483
69
58
0
0.34
0.08
459
67
58
0
0.35
0.12
228
19
58
0.24
0.77
1
456
69
55
0
0.36
0.07
397
63
55
0
0.44
0.13
209
19
55
0.27
0.72
1
372
69
56
0
0.29
0.18
267
54
56
0.01
0.47
0.65
152
19
56
0.04
0.78
1
190
49
45
0
0.80
0.27
95
19
45
0.08
0.64
1
Note: second and third lags (respectively second to fourth and second to fifth) of democracy and log income as instruments over periods 1870-2000 and 1900-2000
(resp. 1930-2000 and 1960-2000). Robust standard errors and Windjmeier correction for standard errors in BB.
Table 3 – OLS results using long-differences between 1900 and 2000
I
II
III
IV
V
VI
VII
Dependent variable: change in score of democracy between 1900 and 2000
change in log GDP per capita
0.07
(0.13)
0.17
(0.10)*
initial level of log GDP per capita
initial level of the score of democracy
-0.84
(0.06)***
0.23
(0.09)**
0.19
(0.07)**
0.18
(0.06)***
0.15
(0.06)**
-1.03
(0.05)***
change in average years of schooling
0.06
(0.03)*
-0.84
(0.06)***
-0.97
(0.06)***
-1.04
(0.05)***
0.06
(0.02)**
0.05
(0.02)**
0.05
(0.03)
0.03
(0.01)***
0.01
(0.01)
0.88
0.92
37
30
initial level of average years of schooling
R2
0.01
0.83
0.89
0.07
0.85
N
30
30
30
37
37
note: robust standard errors. * (respectively ** and ***) stand for significance at 10 (resp. 5 and 1) percent.
Table 4 – Democracy and Average Years of Schooling 1870-2000
FE
1870-2000
AB
BB
FE
1900-2000
AB
BB
FE
1930-2000
AB
BB
FE
1960-2000
AB
BB
Dependent variable is standardized Polity IV index of democracy
Full sample
D(-1)
0.41
0.51
(0.11)*** (0.08)***
0.37
0.49
(0.12)*** (0.09)***
0.20
(0.11)
0.36
(0.10)***
0.20
(0.10)**
0.31
(0.11)***
E(-1)
0.02
(0.01)*
-0.02
(0.03)
0.04
(0.01)***
0.02
(0.01)*
-0.02
(0.03)
0.04
(0.01)***
0.01
(0.02)
-0.05
(0.04)
0.04
(0.01)***
-0.02
(0.03)
-0.09
(0.05)*
0.05
(0.01)***
N
N countries
N instruments
AB1 p-value
AB2 p-value
Hansen p-value
657
70
534
70
58
0
0.2
0.14
613
70
83
0
0.18
0.83
583
70
498
70
55
0
0.21
0.11
540
70
78
0
0.18
0.71
465
70
389
70
56
0
0.22
0.16
427
70
73
0
0.16
0.51
335
70
278
69
45
0
0.91
0.12
310
70
56
0
0.76
0.13
0.23
(0.12)**
0.41
(0.10)***
0.18
(0.11)*
0.31
(0.10)***
Excluding initial
democracies
D(-1)
0.39
0.51
(0.11)*** (0.09)***
0.37
0.48
(0.12)*** (0.09)***
E(-1)
0.02
(0.01)*
-0.00
(0.04)
0.04
(0.01)***
0.03
(0.02)*
-0.01
(0.04)
0.04
(0.01)***
0.00
(0.02)
-0.02
(0.05)
0.04
(0.01)***
0.01
(0.03)
-0.04
(0.04)
0.05
(0.01)***
N
N countries
N instruments
AB1 p-value
AB2 p-value
Hansen p-value
631
68
510
68
58
0
0.2
0.15
587
68
83
0
0.17
0.9
518
64
435
64
55
0
0.27
0.2
476
64
78
0
0.24
0.86
349
55
283
55
56
0
0.32
0.60
316
55
73
0
0.23
0.97
240
51
193
50
45
0
0.95
0.41
218
51
56
0
0.82
0.82
0.03
(0.80)
0.37
(0.17)**
0.38
(0.39)
0.06
(0.21)
-0.11
(0.67)
0.33
(0.18)*
0.27
(0.44)
0.34
(0.57)
-0.25
(0.41)
0.14
(0.12)
0.28
(0.41)
-0.06
(0.12)
0.28
(0.23)
-0.00
(0.15)
0.31
(0.41)
0.04
(0.06)
228
19
58
0.71
0.9
1
247
19
57
0.13
0.66
1
209
19
55
0.77
0.7
1
209
19
78
0.31
0.93
1
152
19
56
0.71
0.54
1
152
19
73
0.03
0.48
1
95
19
45
0.15
0.18
1
95
19
56
0.30
0.12
1
Balanced Panel
D(-1)
E(-1)
-0.02
(0.03)
N
N countries
N instruments
AB1 p-value
AB2 p-value
Hansen p-value
247
19
0.01
(0.04)
209
19
-0.00
(0.05)
152
19
-0.01
(0.07)
95
19
Note: second and third lags (respectively second to fourth and second to fifth) of democracy and education as instruments over periods 1870-2000 and 1900-2000
(resp. 1930-2000 and 1960-2000). Robust standard errors and Windjmeier correction for standard errors in BB.
Table 4 – Democracy and Average Years of Primary, Secondary and Tertiary Schooling
1870-2000
FE
BB
1900-2000
FE
BB
1930-2000
FE
BB
1960-2000
FE
BB
Dependent variable is standardized Polity IV index of democracy
Primary
D(-1)
0.51
(0.09)***
0.48
(0.09)***
0.38
(0.11)***
0.31
(0.10)***
P(-1)
0.05
0.06
(0.02)*** (0.01)***
0.06
0.07
(0.02)*** (0.01)***
0.07
0.07
(0.02)*** (0.02)***
0.08
0.09
(0.03)** (0.02)***
N
N countries
N instruments
AB1 statistics
AB2 statistics
Hansen statistics
657
70
583
70
465
70
335
70
Secondary
D(-1)
613
70
83
0
0.18
0.87
0.60
(0.08)***
540
70
78
0
0.19
0.67
427
70
73
0
0.16
0.50
310
70
56
0
0.80
0.19
0.56
(0.09)***
0.40
(0.11)***
0.40
(0.11)***
S(-1)
-0.00
(0.02)
0.07
(0.03)***
-0.01
(0.02)
0.08
(0.02)***
-0.06
0.08
(0.03)** (0.03)***
-0.12
0.09
(0.04)*** (0.03)***
N
N countries
N instruments
AB1 statistics
AB2 statistics
Hansen statistics
657
70
613
70
83
0
0.16
0.87
583
70
540
70
78
0
0.16
0.75
465
70
335
70
Tertiary
D(-1)
0.70
(0.08)***
0.67
(0.08)***
427
70
73
0
0.14
0.38
0.51
(0.10)***
310
70
56
0
59
0.08
0.52
(0.11)***
H(-1)
-0.15
(0.10)
0.12
(0.15)
-0.20
(0.10)*
0.17
(0.14)
-0.30
0.20
(0.11)*** (0.13)
-0.32
0.23
(0.14)** (0.18)
N
N countries
N instruments
AB1 statistics
AB2 statistics
Hansen statistics
657
70
613
70
83
0
0.14
0.80
583
70
540
70
78
0
0.15
0.69
465
70
335
70
All
D(-1)
0.52
(0.09)***
0.49
(0.10)***
427
70
73
0
0.11
0.43
0.37
(0.11)***
310
70
56
0
0.47
0.06
0.32
(0.11)***
P(-1)
0.04
0.07
(0.02)** (0.02)***
0.05
0.07
(0.02)*** (0.03)**
0.05
0.07
(0.03)** (0.03)**
0.06
(0.04)*
0.09
(0.03)***
S(-1)
0.02
(0.03)
-0.00
(0.05)
0.01
(0.03)
0.00
(0.05)
-0.03
(0.04)
0.02
(0.06)
-0.10
0.03
(0.05)** (0.06)
H(-1)
-0.14
(0.13)
-0.06
(0.21)
-0.15
(0.13)
-0.09
(0.21)
-0.16
(0.15)
-0.09
(0.23)
-0.01
(0.18)
-0.15
(0.26)
N
N countries
N instruments
AB1 statistics
AB2 statistics
Hansen statistics
657
70
613
70
83
0
0.19
0.93
583
70
540
70
78
0
0.19
0.66
465
70
427
70
73
0
0.16
0.48
335
70
310
70
56
0
0.83
0.17
Note: second and third lags (respectively second to fourth and second to fifth) of democracy and education as instruments over periods 1870-2000 and
1900-2000 (resp. 1930-2000 and 1960-2000). Robust standard errors and Windjmeier correction for standard errors in BB. Comparable results are
obtained with BB estimator when excluding initial democracies.
Table 5 – Democracy, GDP per Capita and Average Years of Primary Schooling
1870-2000
FE
BB
1900-2000
FE
BB
1930-2000
FE
BB
1960-2000
FE
BB
Dependent variable is standardized Polity IV index of democracy
D(-1)
0.67
(0.08)***
Log y(-1)
0.14
0.01
(0.04)*** (0.04)
P(-1)
0.04
(0.02)**
567
69
0.65
(0.08)***
0.15
0.01
(0.04)*** (0.04)
0.68
(0.12)***
0.14
-0.01
(0.05)*** (0.04)
0.05
0.07
0.05
0.08
0.06
(0.02)**
(0.02)*** (0.02)**
(0.03)*** (0.02)***
N
560
512
505
424
381
N countries
69
69
69
69
69
N instruments
79
78
50
AB1 statistics
0
0
0
AB2 statistics
0.2
0.2
0.18
Hansen statistics
0.73
0.73
0.12
Note: third lags of democracy, primary education and log income as instruments.
0.76
(0.16)***
0.07
(0.06)
-0.09
(0.07)
0.09
(0.04)**
309
69
0.11
(0.04)***
307
69
36
0
0.69
0.21
Table 6 - Effect of Democracy on Average Years of Schooling of Young Cohorts
OLS
I
II
1870-2000
FE
IV
III
BB 2
V
OLS
VI
VII
VIII
1960-2000
FE
X
IX
Dependent variable is proxied average years of schooling of population aged 5-14
Full sample
Lagged dependent variable
Democracy index
5.31
0.07
(0.33)*** (0.16)
0.42
-0.21
(0.21)** (0.19)
-0.01
(0.11)
0.29
(0.45)
0.06
(0.12)
0.21
(0.46)
5.35
-0.44
(0.58)*** (0.27)*
-0.24
(0.38)
BB 2
XI
XII
1
-0.23
(0.37)
-0.03
(0.11)
0.80
(0.44)*
0.01
(0.18)
1.21
(0.54)**
Average Years of Schooling of
population 15+
0.99
(0.03)***
0.53
(0.06)***
-0.13
(0.28)
0.95
(0.05)***
-0.25
(0.21)
-0.60
(0.46)
log GDP per capita
0.67
(0.11)***
0.84
(0.19)***
1.52
(0.46)***
0.73
(0.17)***
1.33
(0.43)***
2.84
(1.01)***
0.51
0.93
0.28
0.85
0.35
0.92
0.02
0.17
R2
N
497
446
497
446
364
325
181
181
181
181
137
137
N countries
64
64
64
64
56
56
64
64
64
64
56
56
N instruments
66
66
34
34
AB1 p-value
0
0
0
0
AB2 p-value
0.72
0.99
0.39
0.35
Hansen p-value
0.95
0.98
0.48
0.45
Note: second and third lags (respectively second to fifth) of democracy and average schooling of population 15+ as instruments over periods 1870-2000 (resp. 1960-2000).
Robust standard errors and Windjmeier correction for standard errors in BB.
1
Proxy is average years of schooling of population aged 25-34 observed 20 years later.
2
Differentiated dependent and explanatory variables.