Download From James Monroe and the Quing Dynasty to George W

How democracy and dictatorship affect
economic growth: Evidence from James
Monroe and the Quing Dynasty to George W.
Bush and the Communist Party.
Abstract
This article provides solid evidence for the hypothesis that democracy is generally more
conducive to economic growth than dictatorship. At present, there is no consensus among
academics and policy makers on whether and how democracy affects economic growth. The
article reviews and evaluates some of the most important theoretical arguments on why
democracy and dictatorship might affect economic growth. Thereafter, the article presents the
most extensive statistical study conducted on the effect from democracy on economic growth,
with data from 1820 to 2003 on a global sample. Different panel data methods, 2SLS and
matching techniques are used to study the relationship. The estimated effect from democracy
on growth is positive, significant and relatively robust. Barro’s (1991, 1997) claim that semidemocracies perform better than democracies is also tested, and rejected.
1
The question and the answers
Is democracy in general better at generating economic prosperity than dictatorship?1 Can
people have political freedom and a growing economy at the same time? Some argue that this
best of all possible worlds-scenario is unrealistic. The “Lee-thesis” (Sen, 1999:15), credited to
former Singaporean PM Lee Kuan Yew, postulates that particularly in developing countries, a
strong authoritarian regime is necessary to push through economic development. The East
Asian Tiger-states, Pinochet’s Chile and present-day China are viewed as decisive empirical
evidence for this assertion. Democracies, with their perceived political conflicts, gridlocks
and electorally induced conspicuous consumption, are claimed unable to generate economic
development. However, several analysts have to the contrary argued that democracy is
beneficial for economic growth. There is a “democracy advantage” (Halperin et al., 2005).
Democracies for example have institutions that provide vertical and horizontal accountability
so that predatory rulers cannot extract resources in their own self-interest, and the open flow
of information in democracies improves abilities for correcting policy mistakes and for
diffusion of technology. However, rigorous statistical evidence for a “democracy advantage”
in economic growth is still needed.
Although there are adherents to both the Lee-thesis and the democracy advantage view, the
“skeptics” are the most prominent among academics studying the economic consequences of
democracy. Two seminal contributions are Przeworski and Limongi (1993) and Przeworski et
al. (2000). The first surveys earlier statistical studies, and finds that studies claiming positive,
negative or no effect from democracy on economic growth are about equal in numbers. The
1
The theoretical and empirical literature on how different forms of democracies (see e.g Persson and Tabellini, 2003) and
different forms of dictatorship (see e.g. Besley and Kudamatsu, 2007) influence the economy indicate that there is much to
gain from differentiating political regimes more than what is done in this study. However, whether democracy “on average”
produces better or worse growth rates than dictatorships is an academically interesting research question in itself, as well as
extremely relevant for policy debates, especially in developing countries.
2
second conducts a thorough statistical study of its own on the topic, and the general bottomline is that there is no significant effect from democracy on economic growth. These results
are accepted by many political scientists. Larry Diamond (2008:96), for example, asserts that
the “evidence is murky” when it comes to the claim that democracies produce more rapid
economic development, while George Tsebelis reports it as a surprising fact that there is no
evidence of democracies producing superior economic outcomes (2002:70). However, this
article does provide empirical evidence for the hypothesis that democracies on average
produce higher economic growth rates than dictatorships.
When analyzing the most
comprehensive data-set available, incorporating data from 1820 - when James Monroe was
the President of the US and the Quing-dynasty ruled China - to 2003, a clear and significant
effect emerges. Not only are the adherents of the Lee-thesis wrong, the “skeptics” should also
reevaluate their beliefs. According to the analysis below, democracy is good for economic
growth.
Theoretical arguments
In their seminal study, Przeworski and Limongi (1993) evaluated four theoretical arguments
on the relationship between political regime types and economic growth. These four
arguments are only a subset of the existing arguments in the literature on democracy,
dictatorship and effects on growth, but they are among the most important. The arguments
highlight how regime type might matter for I) property rights, II) capital investment, III)
autonomy of the state and IV) checks on predatory rulers. I will take a fresh look at these
arguments, and include some new theoretical insight as well as relevant empirical findings. I
also add a fifth argument on democracy and technological change. As Przeworski and
Limongi (1993), I score the arguments after whether they indicate that democracy increases
3
(for democracy) or decreases (against democracy) economic growth relative to dictatorship.
However, my scorecard differs from Przeworski and Limongi’s.
I) Democracy and the protection of property rights – For democracy
Consider a hypothetical country where the median citizen’s property entitlement, pm, is below
the average property entitlement, pa. Property can only be redistributed progressively. There
are aggregate economic costs related to redistributing property, r, given by the function c(r),
and these costs can be interpreted for example as economic loss due to distortion or
transaction costs. The costs are equal for all citizens, and c(r) is expressed in per capita terms,
with c’(r) and c’’(r)>0. Under democracy and the assumption of one-dimensional politics, the
median voter’s preferred outcome would be to redistribute property until the marginal
personal gain of redistribution is equal to c’(r). If we now compare the democracy to a rightwing authoritarian regime, where the median member in the regime’s group of backers, s, has
a property entitlement ps≥paverage, there will be no property redistribution since c’(r) is strictly
increasing. If income y is given by y = y0 - c(r), then dy/dr<0. Income will be lower under
democracy in this stylistic economy, and the argument could easily be remodeled slightly to
show that economic growth would also suffer from redistribution within democracy. The
argument that democracy will lead to extensive redistribution to the poor, with subsequent
negative effects for aggregate production is old, and was shared by for example John Stuart
Mill, David Ricardo and Karl Marx. Przeworski and Limongi (1993) asses the debate on the
economic consequences of democracy from the nineteenth century, and claim that the right to
vote and freedom of organization were widely perceived to have adverse effects on private
property rights protection, and further economic growth.
If we relax the assumption that property can only be redistributed progressively, there are
strong counterarguments to the claim that democracy weakens property rights protection.
4
Democracy is associated with power distribution, both horizontal and vertical, and these
features provide checks against violations of property rights. In principle, any form of
government implies concentration of coercive power and therefore the possibility of state-led
confiscation of property. A political position within a state apparatus “provides the
opportunity for individuals with superior coercive power to enforce the rules to their
advantage, regardless of their effects on efficiency. That is, rules will be devised and enforced
on behalf of the interests of the politically advantaged” (North, 2000:50). However,
democracies will have certain advantages over dictatorships. First, in democracies the
politically advantaged will constitute a larger segment of the population. In Olson’s model
(2003), a larger group will internalize more of the indirect negative incentive effects of
property rights violation on the overall economy, even if they gain directly from redistributive
activity: “[T]hough both the majority and the autocrat have an encompassing interest in the
society because they control tax collections, the majority in addition earns a significant share
of the market income of the society, and this gives it a more encompassing interest in the
productivity of the society. The majority’s interest in its market earnings induces it to
redistribute less to itself than an autocrat redistributes to himself” (Olson, 2003:122). Second,
there is a larger degree of power dispersion in democracies, also between different state
institutions, which makes the room for single actors enforcing their will at the cost of others
more unlikely. Because of lack of protection of individual rights, poor political accountability
and concentration of power, property rights will be less protected in dictatorships than in
democracies, since these features will allow dictatorial elites to confiscate property.
Przeworksi and Limongi recognize the multiplicity of arguments on the matter and the
different empirical implications. Their overall assessment is therefore that “[W]hile everyone
seems to agree that secure property rights foster growth, it is controversial whether
5
democracies or dictatorships better secure these rights” (Przeworski and Limongi, 1993:51),
and they further conclude that “[T]he idea that democracy protects property rights is a recent
invention, and we think a far-fetched one” (Przeworski and Limongi, 1993:52).
I disagree with Przeworski and Limongi in their conclusion. The fact that different arguments
have different implications does not mean that the arguments are equally valid. I will not go
into a theoretical evaluation here, but I have done so elsewhere (Knutsen, 2007). The most
compelling argument for refuting Przeworski and Limongi’s conclusion is in any case
empirical. Chile, Singapore and South Korea are the only recent empirical experiences of
authoritarian countries that were not hostile to property rights (Przeworski et al., 2000:211).
Boix finds that democracies empirically have a lower degree of expropriation risk than
dictatorships (Boix, 2003). In Knutsen (2007), I used different proxies for property rights
protection, and generally found that democracy significantly increased the protection of
property rights, even when controlling for an extensive set of variables. The effect was robust
over different specifications when using OLS with panel corrected standard errors, Random
Effects and Fixed Effects. The effect was not equally robust when using 2SLS, trying to
incorporate the plausible possibility of endogeneity of regime type, but several specifications
yielded a significant and positive effect. The conclusion is that democracies most likely
protect property rights better “on average” than dictatorships.
II) Dictatorships and investments – Either way, but for democracy when
human capital included
The highest average investment/GDP ratio in the period 1970-2000 belonged to Singapore.
An estimated 45% of GDP went to capital investment over this period (Knutsen, 2006:411).
The investment ratio for the Soviet Union in the 1930’s was equally high, and today China is
6
often racking up annual ratios over 40%. Could these high investment rates have been viable
under democratic regimes? Probably not! Dictatorships have several policy means that allow
them to drive up investment rates, thereby increasing medium to long run growth rates
(Solow, 1956, Romer, 1990). First, dictatorships often suppress freedom of association, thus
crippling independent organization of unions. In the absence of unions, it is plausible to
assume that wages are lower, and that capital owners take a larger share of the pie. When
combined with the assumption that savings rates increase with income, this yields the
prediction that aggregate savings and thereby probably investment rates will be higher in
dictatorships. Political accountability is also lower under dictatorship, among others because
of the lack of free and fair elections. This reduces the pressure on political rulers to channel
resources to immediate public consumption over the national budget. Instead, dictators can
channel resources to investments, independent of the desires of “short-sighted electorates”. A
similar political logic underlies the argument that dictatorial governments need not provide as
much social security to the population in general. The response of rational citizens living
under dictatorial rule is to save privately in order to self-insure for the future.
The argument that dictatorships are better able to generate higher savings and investment rates
is therefore founded on solid theoretical reasoning. However, political mechanisms that allow
dictatorships to have a higher investment rate, should the regime want so, does not imply that
most dictatorial governments have incentives to generate high investment rates. The empirical
evidence often cited in favor of the argument above seems prone to selection bias. The
relatively few historical dictatorships with extremely high savings rates are used as examples,
and general inferences on high savings rates under dictatorship are thereafter drawn, giving
rise to a systematic bias stemming from selecting cases on a specific value (high savings and
investment rates) on their dependent variable (Geddes, 2005 and King et al., 1994). A quick
7
growth accounting exercise on a global sample in the 1990’s showed that the estimated
average effect from dictatorship on capital stock-growth was negative, albeit insignificant
(Knutsen, 2006:348-350). Most dictatorships do not generate very high savings and
investment rates, and there are several reasons for this. First of all, self-interested dictators
might not see it in their interest to follow investment-induced growth, as will become clear
from the argument below on predatory dictators. A second related point is that investment,
and particularly foreign direct investment, is sensitive to the protection of property rights,
where democracies were argued to have an advantage. Third, a high degree of corruption
deters investment, and dictatorship probably increases corruption (Hegre and Fjelde, 2008,
Knutsen, 2008a). Therefore, even if it is theoretically plausible that certain dictatorships might
generate extremely high investment rates due to the large scope of possible policies under
limited political accountability, most empirical dictatorships do not produce high investment
rates.
Mankiw et al (1991) modified the traditional Solow-model of economic growth. They
substituted the traditional Solow-model, which claims that income Y is a function of
technology, A, labor, L, and physical capital, K, with a model where human capital, H, also
enters. In more mathematical terms they go from Y = F(AL,K) to Y=F(AL,K,H). If we stretch
the capital concept to include human capital, democracies have an extra advantage over
dictatorships, since it is hypothesized that democracies invest more in schooling and health.
Mankiw et al. (1991:417-8) estimated that human capital is at least equally important as an
input to the economy as traditional physical capital. Even if Mankiw et al.’s estimation
procedures have been criticized (Klenow and Rodriguez, 1997), human capital is widely
agreed among economists to be an important type of investment.
8
Education and health are highly valued by most people. This means that one would expect
more and better education and health care for most citizens in democracies, which are
assumed more responsive to citizens’ preferences than dictatorships. According to Acemoglu
and Robinson (2006:64), Lindert (2000) finds a strong and positive effect from
democratization on educational expansion in Western Europe. Engerman et al. (1998) find the
same effect in Latin America. Stasavage (2005) finds that democracy has a positive effect on
primary education spending in Africa. More generally, Baum and Lake (2003) find a positive
indirect effect from democracy on growth via human capital. Universal schooling and health
care provision are likely to be redistributive. Some economists have appreciated this point and
produced theoretical models where egalitarian pressures in democracies affect economic
growth through increasing public education, which is a kind of redistribution (Saint-Paul and
Verdier, 1993). This argument is generalized by Acemoglu and Robinson (2006:219-220),
who claim that democracy might force the provision of public goods towards a more socially
efficient level than in dictatorships.
In conclusion, it is not obvious that dictatorship on average increases investment in physical
capital. However, it seems likely that democracy increases the accumulation of human capital.
If we apply a broad definition of capital, one might therefore argue that democracy on average
increases investment, thus contributing to increased growth rates.
III) Dictatorships and autonomy of the state – Against democracy
Scholars studying East Asia have, as Przeworski and Limongi (1993) noted, often linked the
fantastic economic performances of some Asian dictatorships to the autonomy of the
dictatorial state. “In this view, the key to the superior economic performance of the Asian
"tigers" is "state autonomy," defined as a combination of the "capacity" of the state to pursue
developmentalist policies with its "insulation" from particularistic pressures, particularly
9
those originating from large firms or unions. This argument takes two steps: "state autonomy"
favors growth, and "state autonomy" is possible only under authoritarianism” (Przeworski and
Limongi, 1993:56). Olson (1982) argues that democracies are prone to capture from special
interest groups. This might lead to policies that are incoherent with the interests of the broader
populace. Economic growth might be sacrificed for the protection of specific business sectors
or pivotal voting blocs whose interest is not aligned with economic growth. In any case, the
lobbying process will be associated with wasteful rent-seeking, which will detract financial
resources and focus from more productive ventures. The argument is that politicians and
bureaucrats are insulated from such pressures under authoritarianism and are therefore better
able to conduct “proper” policies. One important special case is that of economic reforms.
Certain microeconomic reforms for example, which allow prices to adjust to a level where
they better reflect marginal costs and utilities, improve the efficiency of resource allocation.
However, an adjustment process towards an efficient equilibrium might be painful, and
certain earlier “privileged” groups might lose out. Under democracy, the potential losers
might be important political groups, so-called “veto players” (Tsebelis, 2002), and might act
to block reform. Trade liberalization might be a particularly fitting example, where protected
industries might block liberalization, even if the expected result is an increase in national
GDP. Even if a reform might carry a positive expected value with it for a group, group
members might be risk averse and prefer the status quo even if their best guess is that they
will gain from reform. Under dictatorship, the dictator is assumed to have the means, and in
some instances also the incentives, to carry out “painful” reform. Reform is also assumed to
be conducted more speedily under a dictatorial regime, since many of the procedural steps
needed in democracy as well as complex and time-consuming negotiation can be skipped.
10
There are counterarguments that can modify the picture painted above. First, more than state
autonomy alone is required for successful political decision making. Political and bureaucratic
processes need to be “embedded in a concrete set of social ties that binds the state to society
and provides institutionalized channels for the negotiation and renegotiation of goals and
policies” (Evans, 1995:12). Such “embeddedness” might suffer under dictatorship because of
the insulation of the regime from the general populace and the lack of an organized civil
society with local knowledge. Local knowledge is important to efficient implementation of
political decisions, and dictators are likely to be at an information-disadvantage (Sen, 1999).
When it comes to the speed of reforms, Lijphart (1999:259) argues that speedy reform
processes and speedy implementation might have negative consequences, especially if there is
large uncertainty about a reform’s effects.
Most importantly, it is questionable to assume that dictators are indeed as autonomous as
described above. Even if there is lack of free and fair elections linking the regime to the
broader electorate, no dictator could survive without backing from specific groups, whether
the party, the landlord elite or the military. Bueno de Mesquita et al. recognize this, and claim
that every leader “answers to some group that retains her in power: her winning coalition”
(2003:7). The winning coalition is again drawn from a “selectorate”, the group of actors that
can influence the selection of leaders. The difference between democracies and dictatorships
viewed in this framework is therefore not the degree of autonomy of the regime, but the sizes
and nature of the underlying winning coalitions and selectorates. Kim Jong Il might be
heavily dependent on his backers for political survival, but his backers are few, the high
ranking military officers, and possibly also some high ranking party members. The question
of whether a small underlying winning coalition is conducive to growth is different from the
question of whether more state autonomy is conducive to growth. This might lead us to
11
rethink the effects of dictatorial insulation from the general populace. Besley and Kudamatsu
(2007) argue that dictatorship produces good economic outcomes when the selectorate’s hold
on power is not too dependent on a specific leader, because a strong selectorate can hold the
dictator accountable. Such a situation is however not present in all empirical dictatorships.
IV) Democracies and constraints on predatory rulers – For democracy
Legge inn Bueno de m argument om coalition size her?relateres til vekst og kleptocracy,
accountability..også inn kudam og besley om når autoc verst?
According to Evans (1995:45), if autonomy is defined as not having goals shaped by social
forces, then Mobutu’s regime in Zaire was a very autonomous regime. Mobutu’s Zaire is an
infamous prime example of a “predatory regime”, where the dictator and his inner clique used
their powers for enriching themselves and securing their own continuation in office. Why
would self-interested dictators not use their vast powers to promote policies to their own
benefit, even in instances where the population in general would suffer economically? The
historical abundance of dictators that have used their power for following personal goals with
disastrous macro-consequences are numerous, ranging from Louis XIV’s Versailles-project to
Pol Pot’s decision to kill Cambodians with education or glasses. The most clear-cut examples
come from rulers that steal or confiscate socially productive resources for their own material
benefit, predatory rulers in the true sense of the word. However, the point is more general;
rulers might use strategies that are well-designed for achieving personal goals, but which
might reduce economic growth. One special case emerges when dictators want to minimize
the probability of being thrown out of office. If a dictator largely faces internal threats to his
position, like for example an emerging democracy movement or a guerrilla group, the dictator
might rationally conduct strategies that not only hurt his opponents, but also reduces the
economy’s growth potential (Knutsen, 2008b). If the dictator for example believes that
12
modernization theory is correct, with economic growth and industrialization leading to a
strong middle class and calls for democracy, the dictator will be better off not industrializing.
Another rational strategy could be to spend excessive amounts on a repressive apparatus
instead of using resources for productive investments. In democracies, leaders who try to
engage in predatory activities are more likely to be detected because of freedom of media,
more likely to be checked by other institutions like the legislature and courts, and more likely
to be thrown out of office in the next election. These institutional features provide checks on
predatory behavior.
Not all dictatorships are predatory. One reason is that dictatorships vary in terms of
institutionalized checks and balances (Przeworski et al., 2000). Some dictatorships have
legislatures and parties that play at least some political role, also when it comes to
constraining predatory behavior. Another point is that in some contexts, rational dictators
might not see it in their long term interest to act predatorily. Olson (2003) argues that dynastic
regimes might refrain from predatory activities. These regimes have a relatively long time
horizon and do not want to diminish the overall size of the future economy, which is the
future tax-base. Robinson’s model (2001) indicates that the willingness to engage in predatory
behavior depends on how the dictators’ survival probability is affected by the predatory
activity.
V) Democracy and technological change – For democracy
The question of what drives economic growth has been intensively scrutinized by economists,
and technological change is commonly accepted as the most crucial aspect underlying longterm economic growth (Solow, 1957, Romer, 1990, Helpman, 2004). Thus, if there is a link
between democracy and technological innovation and diffusion, there is a link between
democracy and growth. Sah and Stiglits (1986) showed that polyarchical organizations have
13
higher probabilities of accepting good, novel projects under uncertainty than hierarchical
organizations. This might have relevance for the democracy-dictatorship comparison, as
democracies exhibit greater dispersion of authority. Halperin et al. also claim that
democracies “realize superior developmental performance because they tend to be more
adaptable” (2005:14). They view democracies as “learning organizations”, where individuals
are engaged in the gathering of new information, debate, adjusting positions and revising preexisting knowledge. Evaluating and changing old ways of doing things as well as achieving
progress by trial and error are important for political and economic dynamism. Civil liberties,
inherent characteristics of democracy (Beetham, 1999), are especially relevant for these
processes. Free and open debate is instrumental for eliminating unfounded knowledge and for
opening up to new ideas. John Stuart Mill, referring to political suppression of “wrong” ideas,
noted that “the opinion which it is attempted to suppress by authority may possibly be true.
Those who desire to suppress it, of course, deny its truth; but they are not infallible”
(1974:77).
Economics and politics are not separate, and freedom of expression and debate, norms related
to competition between alternative views and to acceptance of change and dynamism might
spill over from the political to the economic sphere. In the economic sphere the
abovementioned norms will likely contribute to “creative destruction” (Schumpeter, 1976),
thus speeding up the process of technological change and thereby increasing economic
growth. Openness to new and alternative domestic and international information and broad
acceptance of trying out alternative methods of doing things generate a dynamic economy.
Freedom of speech contributes to actors “assessing and disseminating ideas from abroad,
discourages insular thinking and stimulates vigorous debate” (Halperin et al., 2005:13).
However, information and learning in the marketplace is not identical with openness of
14
political debate and the freedom to voice political opinion. Imitation of technologies can of
course also be done in a society where political debate is limited. The Asian Tigers and
Communist China might be empirical examples. However, Chinese control and censorship of
certain websites, out of political reasons, might hinder information flow and use of
communication technologies that could have given economic benefits. Bans or strict
restrictions on cell-phones and the internet have been exercised in several dictatorships, from
Cuba to Turkmenistan. This hurts not only national democracy movements, but also national
economies.
If effective techniques diffuse into the economy through adoption and learning, diversity of
ideas can do nothing but good in the long run. Neo-Schumpeterian economists have stressed
the importance of introduction of novel ideas into the economy. According to Verspagen the
evolution that characterizes a dynamic economy “is the outcome of a constant interaction
between variety and selection” (2005:496). Selection reduces variety since more efficient
techniques are adopted through learning or through “victory in the marketplace”, thus
competing out more inefficient methods of production. In order to keep up variety, one needs
a steady introduction of novel ideas. Freedom of speech and open idea-exchange under
democracy enhance both variety and selection, as the introduction of new ideas, either from
abroad or from local entrepreneurs, and learning processes in the economic sphere, rely on
the possibility of collecting and processing information in a relatively unrestricted manner. I
concur that the links between freedom of speech and information and technological and
organizational change are in need of further specification. However, given the importance of
technological and organizational change for long term growth, these links are most likely very
important as channels through which democracy affects economic growth positively.
*
15
The empirical analyses below will not test the more specific mechanisms identified in the
arguments. It will rather investigate the aggregate effect of these and other mechanisms, as it
tests whether there is an overall, average effect from democracy on economic growth rates.
The overall evaluation of the theoretical arguments, indicate that the deficiencies of
democracy when it comes to restricting growth are not as severe as some authors have
suggested. I sum up Przeworski and Limongi’s (1993) evaluation of the arguments and
compare them with my evaluation in Table 1. In my judgment, there is a lot more going for
democracy than what Przeworski and Limongi suggested and these authors are again more
“optimistic” than other academics in the field. In addition, the proposed effect from
democracy on technological innovation and diffusion point to an extra economic advantage
for democracy. Thus, the best a priori guess is an average positive effect from democracy on
growth.
Table 1: Theoretical arguments and comparative evaluations
Argument
Przeworski and Limongi's
conclusions
My conclusions
Democracy and property rights
Either way
For democracy
Dictatorship and investment
Against democracy
Either way. For democracy
when adding human capital
Dictatorship and state
autonomy
Against democracy
Against democracy, but with
qualifications
Autonomous rulers are
predatory
For democracy
For democracy
Democracy and technological
change
-
For democracy
16
Data
Przeworski and Limongi (1993) point out that empirical results on whether democracy
increases or decreases growth rates depend both on the data and methodology used. I contend
that this is a reason to use the most extensive data set available in order to include all possible
relevant information and to use the most proper methodological techniques at hand.
Chang (2006:145) labels 1960 as “year zero” in statistical work on economic development
issues. Short time series have often forced quantitative social scientists to leave out the main
part of modern history, conceived by historians to have started with the “dual revolutions”:
the industrial revolution in Great Britain and the political revolutions in eighteenth century
United States and France. Luckily, thanks to the Polity Project and economic historian Angus
Maddison, such historical data is actually possible to utilize. This study is so far the most
extensive study on democracy and growth, incorporating data all the way from 1820 (for
some countries) to 2003. The most extensive analysis includes more than 9400 observations
(country-years). Several countries which are often left out of studies of democracy and growth
because of lacking GDP-data, like North Korea, are included in this study. This is important
not only because of pure sample-size considerations, but also because it is most often the
(assumed) economically worst performing dictatorships that are left out of analyses (Knutsen,
2008a). Such systematic lack of data generates biases for studies not recognizing this “simple,
though frequently overlooked recording problem” (Halperin et al., 2005:32). More
specifically, it leads to underestimation of the effect from democracy on economic growth
rates.
The “Polity Index” from the Polity IV dataset is used as the measure for democracy. No other
democracy measure has equally long time series. This index goes from -10 (highly dictatorial)
to 10 (highly democratic). The dimensions which enter the index are competitiveness of
17
executive recruitment, openness of executive recruitment, constraints on the chief executive,
competitiveness of political participation and regulation of political participation.2 The
measure does not include civil liberties. I leave out country-years that are experiencing
interregnum-periods (Marshall and Jaggers, 2005:17). These are mainly periods of internal
anarchy or civil war. I also leave out country-years that are coded with “foreign interruption”,
which basically means foreign occupation.
Maddison’s data-set includes historic data on GDP and population. Estimating GDP in years
when the national accounting system was not even invented is a heroic task, and Maddison’s
estimates are bound to contain large errors.3 One particularly troublesome aspect of
Maddison’s estimation procedure is that movements in GDP per capita for some countries
with lacking data are estimated by assuming similarity to neighboring “core-countries” with
more abundant historical data. Maddison utilizes PPP-adjusted GDP-data (US 1990$), which
takes into account differing local price levels. This poses an extra dimension of uncertainty, as
local price levels need to be estimated. Econometric theory indicates that unsystematic
measurement errors in independent variables can generate attenuation biases, both for the
coefficient of the variable with measurement error, and for coefficients for other independent
variables correlated with the error-prone variable. GDP per capita and population are
independent variables with errors, but I have no a priori suggestions on how unsystematic
errors in these variables will affect the democracy-coefficient. If there is unsystematic
measurement error in the Polity-index however, the democracy-coefficient will be biased
towards zero. The dependent variable, growth in GDP per capita, is also measured with an
error. Econometric theory suggests that if this error is unsystematic, it will be more difficult to
2
See Marshall and Jaggers (2002:12-15).
3
See Maddison (2006:169-228) for a description of the diverse sources and procedures used to estimate population and GDPdata.
18
find significant regression coefficients, for example for democracy. Even though the data
problems are immense, Maddison claims that “[F]or the epoch of capitalist economic growth
back to 1820, quantitative economic historians have made great progress in measuring growth
performance ... There is still a need to fill gaps and crosscheck existing estimates, but the
broad contours of world development in this period are not under serious challenge”
(Maddison, 2007:294).
In what I have called the interpolated sample, I include 832 country-years where data on GDP
per capita growth or level or population level is constructed by interpolation. Several of the
time series for individual countries are interrupted in the Maddison data-set. I only
constructed interpolated data where at least 90% of the years missing between two
observation-years have the same score on the Polity-variable. I assign the average growth rate
of the period to all the country-years, where I interpolate. I will get artificially low variation in
the estimated GDP per capita growth rates, but the average growth rate is correct by
construction. There will be “construction errors” in the estimates of GDP-levels and
population levels at particular points in time, but the probable influence of this error on the
estimated effect of democracy on growth is likely to be tiny.
Empirical analysis
OLS with Panel Corrected Standard Errors (PCSE)
I use a version of OLS with PCSE which takes into account heteroskedastic panels,
autocorrelation within panels (AR1) and contemporaneous correlation. The technique allows
us to base inferences on both cross-sectional and temporal variation. The number of
observations utilized in this analysis by far supersedes that of earlier studies on the subject.
There is data from 154 existing and previous countries, and the number of observations in the
19
“small” sample is 8933. In the interpolated sample, the number is 9438. Along with the polity
variable, I enter several control variables that might affect both the nature of regime and
economic growth rates:4 I enter log of regime duration (+1) as a proxy for political stability. I
control for log of GDP per capita and log of population size. I also enter geographical regiondummies, time controls and plurality religion-dummies.5 The plurality religion-dummies are
based on newer data (Knutsen, 2007:67-8), and there might be errors for earlier years if the
plurality religion in a country has changed with time. When it comes to the time controls, I
use both a relatively extensive version, with dummies for each decade (the 1990’s being
extended until 2003), and a less extensive version where I add four dummy variables for the
periods 1820-69, 1870-1913, 1914-45 and 1946-72. I lagged all independent variables with
two years, in order to reduce endogeneity problems. The regression equation is given by:
GRi,t = β0+β1polityi,t-2+β2ln(dur+1)i,t-2+β3ln(gdp/cap)i,t-2+β4ln(pop)i,t-2+Σβjreligi+Σβkregi+Σβktime
As Table 2 shows, there is a robust, positive and significant effect from democracy on
economic growth. Independent of type of time control and sample chosen, the estimated
Polity-coefficient, β1, is always significant at the 1%-level. If we hold all the other variables
constant, our model predicts that going from most dictatorial to most democratic on the
Polity-scale, increases annual growth in GDP per capita with about one percentage point.
4
As a general principle, one should only control for variables that are causally prior to both regime type and growth. Several
earlier studies have controlled for variables such as investment and education levels, which as seen in the theoretical section
are expected to be causal channels through which regime type affects growth (see Tavares and Wacziarg (2001) and Baum
and Lake (2003)). This operation biases estimates of total causal effects as important indirect effects are controlled away. It is
however likely that for example education is both cause and effect of regime type, and thereby should have been controlled
for to reduce possibility of omitted variable bias. I have three responses. First, there are no long time series of for example
education data globally. Second, GDP per capita correlates strongly with several plausible control variables without data, like
education variables, urbanization and more. I thereby indirectly reduce omitted variable bias through controlling for GDP per
capita. Third, in the appendix I show OLS with PCSE results on shorter time series where several extra variables are
controlled for, and the results are largely unchanged.
5
The region dummies are: Western Europe plus North America, Australia and New Zealand, Eastern Europe plus the Soviet
Union (and resulting states), Asia and the rest of Oceania, Middle East and North Africa, Africa South of the Sahara, Latin
America. The religion dummies are: Catholicism, Protestantism with Anglicanism and Baptism, Orthodox Christianity, Shia
Islam, Sunni Islam, Buddhism together with Shintoism, Taoism and Confucianism, as well as a dummy for “traditional”
religions.
20
Assume that we have two identical countries in terms of scores on the control variables in
1820. The only difference between them is that one country is a perfect democracy and the
other a harsh dictatorship. If our model is correct, and we forget that we should have
controlled for increases in GDP along the way, the democratic country will be more than six
times richer than the dictatorship in 2003, starting out equally rich in 1820. The analysis
indicates that in the long run, regime type matters a lot for economic welfare.
Table 2: Results from OLS with PCSE-analyses6
Democracy
Sample
Time controls
coefficient
T-value
Observations
Small
1870-1914-46-73
0.047
3.25***
8933
Small
Decades
0.048
3.54***
8933
Interpolated
1870-1914-46-73
0.048
3.42***
9438
Interpolated
Decades
0.049
3.68***
9438
Fixed- and Random Effects
What if there are country-specific factors not included in the OLS with PCSE-analyses that
determine both growth rate and degree of democracy? If so, the methodology used above is
inappropriate, as we should have controlled for country-specific effects. One solution is to run
a Fixed-Effects regression, incorporating country dummies. Thereby one will only use intranational variation for inference. The resulting analysis is therefore very restrictive. Beck and
Katz (2001:490) claim that it is an “essentially nihilist position that any cross-sectional
variation could be the result of idiosyncratic factors”, when discussing FE-analysis, and I
agree on this point. FE does not incorporate information on economic effects from comparing
6
In all tables, *** denotes significant at the 1%-level, ** significant at the 5%-level and * significant at the 10%-level.
21
regimes in different countries. Nevertheless, the benefits of incorporating dummies for both
time-periods and countries relate to the reduction of omitted variable bias, even if we are left
with less information from which to draw our inferences.
Table 3: Results from Fixed Effects-analyses (robust s.e.)
Democracy
Sample
Time controls
coefficient
T-value
Observations
Small
1870-1914-46-73
0.036
2.20**
8933
Small
Decades
0.021
1.22
8933
Interpolated
1870-1914-46-73
0.035
2.26**
9438
Interpolated
Decades
0.020
1.20
9438
The FE-results from Table 3 are not as clear-cut as the OLS with PCSE-results. Nevertheless,
even if we control for country-specific effects and time-specific effects along with logs of
GDP per capita, population and regime duration, we obtain some statistically significant
results. When the four time period dummies are entered as time controls, the estimated effect
from democracy is only about 25% smaller than the estimated effect from OLS with PCSE,
and the effect is significant at the 5%-level. However, when we enter decade-dummies
instead, the effect is no longer significant at conventional levels. The effect is therefore not
completely robust to choice of econometric specification. Another indication of this is that
none of the models show significant effects when we use standard errors that are not robust
but incorporate the possibility of AR1 autocorrelation instead. The estimated effects are
however still positive.
22
FE-analysis assumes that each country has its own specific intercept in the regression.
Random-Effects (RE) analysis moderates this assumption. RE, like FE, generates a different
intercept for each cross-section unit, “but it interprets these differing intercepts in a novel
way. This procedure views the different intercepts as having been drawn from a bowl of
possible intercepts, so they may be interpreted as random … and treated as though they were
part of the error term” (Kennedy, 2003:304). Under the assumption that the intercept is
randomly selected, RE gives increased efficiency to the estimates when compared to FE.
However, RE will be biased if the error term is correlated with any of the independent
variables.
Table 4: Results from Random Effects-analyses (GLS and robust s.e.)
Democracy
Sample
Time controls
coefficient
T-value
Observations
Small
1870-1914-46-73
0.045
3.12***
8933
Small
Decades
0.044
2.93***
8933
Interpolated
1870-1914-46-73
0.043
3.12***
9438
Interpolated
Decades
0.042
2.87***
9438
The RE-analyses in Table 4 show much clearer results than the FE-analyses, in the sense that
all the model-specifications yield significant results at the 1%-level. Both the estimated
effects and the t-values closely resemble those found when using OLS with PCSE, with a one
unit increase on the Polity-scale giving a predicted 0,04-0,05% extra annual growth. When we
enter standard errors that are no longer robust, but which capture autocorrelation (AR1), the
results are qualitatively the same for the RE-analyses.
23
The main conclusion from the panel-data analyses is that there is a positive and relatively,
albeit not completely, robust effect from democracy on economic growth. Even if the
different analyses build on diverse assumptions regarding incorporation of cross-sectional
information and error-specifications, we most often find an effect that is statistically
significant at least at the 5%-level, and all models show a positive estimated effect.
2SLS
Above, we have assumed democracy to be exogenous. However, Przeworski and Limongi
(1993) argue that regime type might be endogenous to economic growth, and that coefficients
from regression analysis might be biased. One solution is to use modeling techniques that
incorporate the selection-bias, like the one suggested by Heckman (1978), but as Przeworski
and Limongi note, these are not robust (1993:64). The analyses above are based on a panel
data structure, and GDP per capita is controlled for at each point in time. This reduces the
endogeneity problem when it comes to an effect from income level on regime, but there might
be other systematic effects from growth in GDP on regime type. One simple way to
investigate endogeneity is to conduct a Granger-test. The simplest of all Granger-tests is to
control for the dependent variable in t-1 and investigate the effect from the independent
variable of interest. Democracy in t-1 was estimated to have a positive effect on growth in t,
and the result was significant at the 1%-level. When democracy in t was entered as a function
of democracy in t-1 and growth in t-1, the estimated effect from growth was actually negative
and significant at the 5%-level. According to this analysis, one should actually expect that the
coefficients in the analyses above would underestimate a positive causal effect from
democracy on economic growth.
However, there are other and better ways of dealing with the endogeneity problem than
Granger-tests. One proposed solution in the econometric literature is to find so-called
24
instrumental variables for endogenous independent variables. There are two requirements for
a variable to be an efficient and consistent instrument. First, it should be at least moderately
correlated with the independent variable. Second, it should not be directly related to the
dependent variable. The intuition behind the procedure is that we only utilize the “exogenous”
part of the variation in the independent variable, and thereby get a better estimate of the causal
effect from the independent variable on the dependent. The technique utilized below is often
referred to as Two Stage Least Squares (2SLS), and the estimates produced are consistent.
However, 2SLS produces relatively inefficient estimates, especially if the correlation with the
instrument is low. It is therefore often difficult to find significant results.
Figure 1: Causal structure underlying IV/2SLS analysis
Instrument
Democracy
Economic growth
Helliwell (1994) uses past value of democracy as an instrument for present values, but there
might be reasons to believe that the political history of a country affects its present growth
rates. I suggest a more proper instrument. This instrument draws on Samuel Huntington’s
(1991) observation that the share of democratic regimes globally has followed systematic
temporal trends: Democratization has historically proceeded in three temporal waves.
According to Huntington’s categorization, the first long wave started in 1828 and ended in
1926, and the second wave lasted from 1943 to 1962 (Huntington, 1991:16). The third wave
of democratization “began, implausibly and unwittingly, at twenty-five minutes after
midnight, Thursday, April 25, 1974, in Lisbon, Portugal” (Huntington, 1991:3). So-called
reverse waves have come between the three waves, and in these reverse waves
authoritarianism has triumphed relatively. Since Huntington wrote his book in the early
1990’s, there is no concrete suggestion to when the third wave ended, if it ever has. However,
25
Freedom House Executive Director Jennifer Windsor stated that “[A]lthough the past 30 years
have seen significant gains for political freedom around the world, the number of Free
countries has remained largely unchanged since the high point in 1998”. I will therefore
assign 1998 as the end-year of the third wave of democratization.7
The timing and clustering of democratization experiences can be attributed to factors that are
considered as exogenous to national politics, namely large geo-political events and contagion
effects from neighbours (Knutsen, 2007:54-5). If democratization waves and reverse waves
represent such exogenous factors that affect probability of democratization and that are not
directly related to economic factors, whether regime transition for a country occurred within
one of the waves or alternatively within reverse waves (dummy variable) can serve as a valid
instrument for democracy. I employ the same approach as Persson and Tabellini (2003), when
they propose instruments for type of constitutional arrangement and electoral system.8 I use
the starting year of the political regime, according to Polity, as a point of departure, and score
countries after whether their existing regime (at a particular point in time) originated in a
reverse wave or not. I could also have categorized after regime change in a wave, since
Huntington’s categorizations are somewhat ambiguous, with overlapping waves and reverse
waves. Huntington stated that the reverse waves lasted from 1922-1942 and from 1958-1975
(Huntington, 1991: 16). Earlier analysis has shown that the correlation between democracy
and the reverse wave-instrument is stronger than with the wave-instrument (Knutsen, 2007).
According to the econometric theory of instrumental variables the preferred instrument is the
one with the higher correlation with the endogenous independent variable, since it will
produce more efficient estimates. Therefore, country-years where the current regime
7
See Huntington (1991) for an explanation of the driving forces underlying the different waves.
8
Persson and Tabellini (2003) have no underlying theory for their scoring-procedure, but use relative frequencies of
emergence of specific constitutional types and electoral systems globally as their criteria for categorizing time-periods.
26
originated in the periods (,1827], [1922,1942], [1958,1975], [1998,2003] are given a 1 on
the dummy-instrument.9
The reverse wave instrument for democracy is relatively strong, as the absolute t-values in the
first stage regressions are above 35. The decade-dummy model generated coefficients that
were positive, and actually larger than those generated by the RE-models.10 Nevertheless, the
2SLS standard errors were so large that the p-values for the small and interpolated samples
were 0.13 and 0.09 respectively. I performed Hausman tests, comparing the coefficients from
the RE and 2SLS models, and the hypothesis of similar coefficients could not be rejected at
any conventional significance-level, with both t-values below 1. One interpretation of the
Hausman tests is that the endogeneity problem is not severe, and that we can go with the more
efficient RE-results, which show a statistically significant effect from democracy. However,
one can increase the efficiency of the 2SLS estimator by adding more instruments. I added
absolute latitude from Hall and Jones’ dataset as an instrument in the first stage. Hall and
Jones (1999) used latitude as one of four instruments for institutional structure in their
seminal paper. When running the RE-version of 2SLS with “reverse wave” and “absolute
latitude” as instruments, the democracy-coefficient was now significant at the 5%-level in
both samples, with p-values of 0,02 and 0,01. Even if the latitude-instrument might not be a
very adequate instrument for institutional structure, and in my case democracy (Acemoglu et
al., 2001), this result gives a very strong indication that there is a positive causal effect from
9
The US regime, originating from before 1827 according to Polity, is not scored a 1, given its explicit categorization in the
first wave by Huntington (1991:16).
10
In this section, I only refer to the decade-dummy models, which have the most extensive time controls. The model with
less extensive time controls yielded somewhat weaker, albeit qualitatively similar, results.
27
democracy on economic growth.11 The estimated effect from democracy holds up, even when
we try to incorporate the possibility of endogeneity of regime type.
Table 5: G2SLS – random effects IV-estimation results, models with decade dummies
Instruments
Sample
Coefficient T-value Observations
Reverse wave
Small
0.059
1.51
8933
Reverse wave
Interpolated
0.063
1.70*
9438
Small
0.084
2.28**
8354
Interpolated
0.087
2.47**
8859
Reverse wave,
Absolute latitude
Reverse wave,
Absolute latitude
Matching
Regression-based techniques assume linear effects. The probable complexity of the
relationship between democracy and growth means that imposing a linearity restriction might
be a too strong assumption to make. Recently, there has been some interest in matchingtechniques among researchers studying political-economic topics (Persson and Tabellini,
2003), but the usage is still far from widespread. Matching is a non-parametric estimation
technique, where we relax assumptions of functional form, and where we do not have to
assume that the effect is independent of values on contextual variables.
Matching-techniques draw on experimental logic: “The central idea in matching is to
approach the evaluation of causal effects as one would in an experiment. If we are willing to
make a conditional independence assumption, we can largely re-create the conditions of a
Acemoglu et al.’s (2001) settler mortality instrument, although plausible as an instrument in this setting, was not used since
this instrument is only defined for ex-colonies.
11
28
randomized experiment, even though we have access only to observational data” (Persson and
Tabellini, 2003:138). The main underlying idea is that we split our independent variables into
two groups, the control variables, which here consist of all independent variables except for
regime type, and the treatment variable, regime type. Further, we need to dichotomize the
treatment variable, and assume so-called conditional independence; that is, we need to assume
that the selection of democracy is random, after we have controlled for the other independent
variables. This assumption might not be strictly warranted in the case of regime type and
growth (Przeworski and Limongi, 1993).
The technique is based on the underlying idea that we should compare most similar units. We
make “local” comparisons over units that are relatively similar on all but the treatment
variable. The estimated effect of democracy is computed for each of these comparisons, and
then an average of these effects is calculated, the average treatment effect. I use robust
standard errors when I estimate results, but the problem of autocorrelation of residuals might
still pose a problem to the analysis, because of the panel structure in the data. The unit of
analysis is country-year. In practice, we therefore incorporate variation that is due to a country
going from non-democratic to democratic or vice-versa in the estimating procedure, as well as
information from comparisons between relatively similar countries. We have to dichotomize
the Polity-variable. I choose two different cut-off points to check the robustness of the results.
The first specification scores all countries with a score of 0 or better as democracies, and the
second uses a score of 4 or more as the cut-off for democracy. I also explore how robust the
results are by allowing for a different number of matches in different analyses. Replacement is
allowed in the estimation procedure, which means that a unit can be used as a match more
than once.
29
Table 6: Average treatment-effects when going from dictatorship to democracy from the
matching analyses
Number Average
of
treatment
matches
effect
Sample
Polity
cut-off
T-value
Number of
observations
Small
0
1
2.19
3.75***
8933
Small
0
3
1.29
3.55***
8933
Small
0
5
0.98
3.49***
8933
Small
0
10
0.68
3.22***
8933
Small
4
1
1.32
2.89***
8933
Small
4
3
0.78
2.50**
8933
Small
4
5
0.67
2.73***
8933
Small
4
10
0.58
3.10***
8933
Interpolated
0
1
2.21
4.01***
9438
9438
Interpolated
0
3
1.37
3.99***
Interpolated
0
5
1.05
3.98***
Interpolated
0
10
0.74
3.73***
Interpolated
4
1
1.39
3.23***
Interpolated
4
3
0.88
3.00***
Interpolated
4
5
0.77
3.26***
Interpolated
4
10
0.65
3.58***
9438
9438
9438
9438
9438
9438
The results from the matching-analyses triangulate well with the earlier findings. All the
democracy-coefficients from the 16 analyses reported in Table 5 are positive and significant
at the 5%-level, and 15 coefficients are significant at the 1%-level. The average treatment
effect from democracy is positive and substantial in size. Some of the results above indicate a
far larger effect of going from relatively dictatorial to relatively democratic than results found
in the linear models. However, the largest estimates come from analyses with a small number
of matches, and the standard errors are also generally large in these specifications. The
30
models with a larger number of matches generate lower point-estimates, but also lower
standard errors, and I would argue that these estimates are more credible. Generally, the
models where fewer country-years are accepted into the democracy category generate lower
estimates, but independent of classification, democracy is found to have a positive and
significant effect on economic growth. The estimated effect of changing regime type to
democracy from dictatorship varies from 2,19% extra annual growth to 0,65% extra annual
growth.
The inverse u-curve?
Robert Barro (1991, 1997) found no significant linear relationship between democracy and
economic growth, but claimed that the relationship was inversely u-shaped, indicating that an
intermediary level of democracy is conducive to economic growth. Barro offers an
interpretation: “[I]n the worst dictatorships, an increase in political rights tends to increase
growth and investment because the benefit from limitations on governmental political power
is the key matter. But in places that have already achieved a moderate amount of democracy,
a further increase in political rights impairs growth and investment because the dominant
effect comes from the intensified concern with income redistribution” (Barro, 1997:59). A
combination of two differently signed mechanisms can lead to an inversely u-shaped overall
effect, for example if the two mechanisms are concave, but the theoretical rationale for
concavity here is not very strong. One relevant analogy comes from the literature on
determinants of civil war. Hegre et al (2001) found that intermediate levels of democracy
produced a higher probability of civil war, and it is plausible that such “incoherent regimes”
might also be bad for economic growth. Incoherent regimes might be prone to political
instability, and might not have a sufficient level of democracy to secure property rights or
provide a functioning human capital system. On the other hand, these regimes might not be
autonomous enough to conduct tough economic reforms or generate extreme investment
31
levels. I transform the Polity variable so that it ranges from 0 to 20, square it and add it to the
OLS with PCSE-models. All specifications, regardless of sample and time-dummies, show an
estimated positive coefficient for both the linear and squared polity-variables. Figure 2 shows
the estimated marginal effect of democracy as a function of level of democracy. It is always
positive and increasing in level of democracy. However, the squared term is not significant at
any conventional level, and we can therefore not reject a linear specification. In any case,
Barro’s claim of an inverse u-relationship between democracy and growth, based on OLS on
cross-sectional averages, does not hold up to empirical scrutiny when more data and better
methods are used.
Figure 2: Estimated marginal effect of democracy on economic growth
Interpolated sample
Marginal
effect
Small
sample
Polity-score
Conclusion
Scholars studying East Asian politics have often concluded that authoritarianism is better than
democracy at generating economic growth, and that democracy as a political regime is to be
32
viewed as some kind of luxury, affordable once a country becomes rich. Sen (1999, 15)
observes that “many people in different countries of the world are systematically denied
political liberty and basic civil rights. It is sometimes claimed that the denial of these rights
helps to stimulate economic growth and is “good” for economic development”. Authoritarian
regimes are perceived as necessary to increase savings and investments, and for the state to
have some autonomy from partial interest groups. And who other than strong autocrats could
push through painful reforms? The South Korean, Singaporean and now recent Chinese
histories seem to give the ultimate empirical evidence. But as Przeworski et al. (2000)
rightfully claim: In order “to assess the impacts of political regimes, we must examine their
full record, not just the best performers” (Przeworski et al., 2000:4). When we do so, the
proposed trade-off between freedom and bread is illusory.
The statistical analyses conducted in this article, based on the most extensive data-sample
used in the literature, do not find any evidence for the claim that authoritarianism produces
higher economic growth in general. On the contrary, democracy is generally found to have a
positive and significant impact on economic growth. In addition to extending the evidence
back to long before WWII, this study utilizes statistical methodologies that are much sounder
than the previously often used OLS on averages over long time spans. Our best guess should
therefore no longer be that democracy does not systematically influence economic growth, as
many prominent researchers have proposed. Democracy on average seems to enhance growth
rates in GDP per capita. Many academics and policy makers should now start reevaluating
their beliefs about the economic effects of democracy. The claim induced by Barro’s analysis
(1997) that partial democracy is economically superior to a system with more extensive
freedoms and rights should also be shed in light of the new evidence. The research on the
topic should move on to look at more particular mechanisms and contingencies of effects. The
33
theoretical arguments presented in this article point to some potential causal mechanisms, but
theoretical work should be further developed, especially along the lines of creating coherent
models that incorporate different mechanisms and explicate the relations between them. When
it comes to the empirical work, much has been done on the economic effects of different types
of democracies (Persson and Tabellini, 2003). However, the empirical variation in growth
rates among dictatorships is far larger (Rodrik, 2000 and Knutsen, 2008b), and more research
is needed to understand why this is so.
34
Literature
Acemoglu, Daron and James A. Robinson (2006). Economic Origins of Dictatorship and Democracy.
New York: Cambridge University Press.
Acemoglu, Daron, Simon Johnson and James A. Robinson (2001). “The Colonial Origins of
Comparative Development: An Empirical Investigation”. American Economic Review 91: 1369-1401
Barro, Robert J. (1991). “Economic Growth in a Cross Section of Countries”. The Quarterly Journal of
Economics 106 (2): 407-443.
Barro, Robert J. (1997). Determinants of Economic Growth. A Cross-Country Empirical Study.
Cambridge, MA.: The MIT Press.
Baum, Matthew A. and David A. Lake (2003). “The Political Economy of Growth: Democracy and
Human Capital”. American Journal of Political Science 47(2): 333-347.
Beck, Nathaniel and Jonathan Katz (2001). “Throwing Out the Baby with the Bath Water: A Comment
on Green, Kim, and Yoon”, International Organization 55(2): 487-495.
Beetham, David (1999). Democracy and Human Rights. Cambridge: Polity Press.
Besley, Timothy and Masayuki Kudamatsu (2007). “Making Autocracy Work”. Working Paper.
Boix, Carles (2003). Democracy and Redistribution. New York: Cambridge University Press.
Bueno de Mesquita, Bruce, Alistair Smith, Randolph M. Siverson and James D. Morrow (2003). The
Logic of Political Survival. Cambridge, MA: MIT Press.
Chang, Ha-Joon (2006). The East Asian Development Experience – The Miracle, the Crisis and the
Future. London: Zed Books.
Diamond, Larry (2008). The Spirit of Democracy. New York: Times Books.
Engerman Stanley L., Elisa Mariscal and Kenneth L. Sokoloff (1998). “Schooling, Suffrage and the
Persistence of Inequality in the Americas, 1800-1945”. Los Angeles: UCLA, Department of Economics.
Working Paper
Evans, Peter (1995). Embedded Autonomy. States & Industrial Transformation. Princeton: Princeton
University Press.
Geddes, Barbra (2003). Paradigms and Sand Castles – Theory Building and Research Design in
Comparative Politics. Ann Arbor: The University of Michigan Press.
Hall, R.E. and Charles I. Jones (1999). “Why Do Some Countries Produce So Much More Output Per
Worker Than Others?” Quarterly Journal of Economics 114: 83-116.
Halperin, Morton H., Joseph T. Siegle and Michael M. Weinstein (2005). The Democracy Advantage.
How Democracies Promote Prosperity and Peace. New York: Routledge.
Hegre, Håvard and Hanne Fjelde (2008). ”Democracy Depraved: Political Corruption and Institutional
Transitions”. Paper presented at the Norwegian National Political Science Conference,Tromsø, 28th 30th April, 2008.
35
Hegre, Håvard, Tanja Ellingsen, Scott Gates and Nils Petter Gleditsch (2001). “Toward a Democratic
Civil Peace? Democracy, Political Change and Civil War, 1816-1992”. American Political Science
Review 95: 33-48.
Heckman, James J. (1978). "Dummy Endogenous Variables in a Simultaneous Equation System".
Econometrics 46: 931-59.
Helpman, Elhanan (2004). The Mystery of Economic Growth. Cambridge, MA.: The Belknap Press of
Harvard University Press.
Helliwell, John F. (1994). “Empirical Linkages between Democracy and Economic Growth”, British
Journal of Political Science 24: 225-248.
Huntington, Samuel P. (1968). Political Order in Changing Societies. New Haven: Yale University
Press.
Kennedy, Peter (2003). A Guide to Econometrics. Fifth edition. Cambridge, MA: The MIT Press.
King, Gary, Robert O. Keohane and Sidney Verba (1994). Designing Social Enquiry – Scientific
Inference in Qualitative Research. Princeton: Princeton University Press.
Klenow, Peter and Andres Rodriguez-Clare (1997): “The Neoclassical Revival in Growth Economics:
Has it gone too Far?” in B.S. Bernanke and J.J. Rotemberg [eds.] NBER Macroeconomics Annual 1997.
Cambridge: MIT Press.
Knutsen, Carl Henrik (2006). “Political regime types and economic growth: Are democracies better at
increasing prosperity”. Oslo: University of Oslo, Departement of Political Science. Master thesis.
Knutsen, Carl Henrik (2007). “Democracy and Property Rights - A theoretical and empirical analysis
on the effects of political regime type on property rights arrangements?” Oslo: University of Oslo,
Department of Economics. Master Thesis.
Knutsen, Carl Henrik (2008a). “The Economic Effects of Democracy – An Empirical Analysis”. Oslo:
University of Oslo, Department of Political Science. Working Paper.
Knutsen, Carl Henrik (2008b). “The Diverging Economic Dynamics of Dictatorships: Enlightened Rule
and Tyranny?” Oslo: University of Oslo, Department of Political Science. Working Paper.
Lijphart, Arend (1999). Patterns of Democracy. New Haven: Yale University-Press.
Lindert, Peter (2000). “The Political Economy of Comparative Mass Education in Europe before
1914”. Davis: University of California at Davis, Agricultural History Center. Working Paper.
Maddison, Angus (2006). The World Economy. Paris: OECD Publishing
Maddison, Angus (2007). Contours of the World Economy, 1-2030 AD. Oxford: Oxford University
Press.
Mankiw, Gregory N., David Romer and David N. Weil (1992). “A contribution to the Empirics of
Economic Growth”. Quarterly Journal of Economics 107: 407-437.
Marshall, Monthy G. and Keith Jaggers (2002). “Polity IV Project – Dataset Users’ Manual”. College
Park: Program Center for International Development and Conflict Management, University of
Maryland. Manual.
36
Mill, John Stuart (1974). On Liberty. Harmondsworth: Penguin Classics.
North, Douglass C. (2000). “Institutions and Economic Growth: A Historical Introduction” in Jeffrey a
Frieden and David Lake [eds.] International Political Economy. Perspectives on Global Power and
Wealth. Boston: Bedford/St. Martin’s.
Olson, Mancur (1982). The Rise and Decline of Nations. Economic Growth, Stagflation and Social
Rigidities. New Haven: Yale University Press.
Olson, Mancur (2003). “Dictatorship, Democracy and Development”. Chapter 6 in Stephen Knack
[ed.] Democracy, Governance and Growth. Ann Arbor: The University of Michigan Press.
Persson, Torsten and Guido Tabellini (2003) The Economic Effects of Constitutions. Cambridge, MA.:
The MIT Press.
Przeworski, Adam and Limongi, Fernando (1993). “Political Regimes and Economic Growth”. The
Journal of Economic Perspectives 7 (3): 51-69.
Przeworski, Adam, Michael A. Alvarez, José Antonio Cheibub and Fernando Limongi (2000).
Democracy and Development: Political Institutions and Well-being in the World, 1950-1990. New
York: Cambridge University Press.
Robinson, James A. (2001). “When is a State Predatory?”. Cambridge, MA: Harvard University.
Working Paper.
Rodrik, Dani (2000). Institutions for High-Quality Growth: What They Are and How to Acquire Them.
Cambridge MA.: National Bureau of Economic Research. NBER Working Paper no 7540.
Romer, Paul (1990). “Endogenous Technological Change”. Journal of Political Economy 98: 71-102.
Sah, Raaj Kumar and Joseph E. Stiglitz (1986). “The Architecture of Economic Systems: Hierarchies
and Polyarchies”. The American Economic Review 76: 716-727.
Saint-Paul, Gilles and Thierry Verdier (1993). “Education, democracy and growth”. Journal of
Development Economics 42: 399-407.
Schumpeter, Joseph A. (1976) Capitalism, Socialism and Democracy. New York: Harper Perennial.
Sen, Amartya (1999). Development as Freedom. New York: Anchor Books.
Solow, Robert M. (1956). “A Contribution to the Theory of Economic Growth”. The Quarterly Journal
of Economics 70: 65-94.
Solow, Robert M. (1957). “Technical Change and the Aggregate Production Function”. Review of
Economics and Statistics 39: 312-320.
Stasavage, David (2005). “Democracy and Education Spending in Africa”. American Journal of Political
Science 49(2): 343-358.
Tavares, José and Romain Wacziarg (2001). “How democracy affects growth”. European Economic
Review 45: 1341-1378.
Tsebelis, George (2002). Veto Players – How Political Institutions Work. Princeton: Princeton
University Press.
37
Verspagen, Bart (2005). “Innovation and Economic Growth” in Jan Fagerberg, David C. Mowery and
Richard R. Nelson [eds.] The Oxford Handbook of Innovation. Oxford: Oxford University Press.
38
Appendix 1: Robustness of the results: The post-war experience12
There is always a possibility that inferred effects might be due to properties of the sample, the
specific operationalization of variables or choice of control variables. In the analysis above,
the extensive time period chosen limited the possibility of using different operationalizations
and adding control variables, since data are scarce before WWII. However, I conducted
several analyses on shorter panels, and in general the main result of a positive and statistically
significant effect from democracy on economic growth withstood the test. The dependent
variable is now operationalized by using data on PPP-converted GDP per capita from the
World Development Indicators, and the reliability of the GDP-numbers should be higher than
in the previous analysis using the data from Maddison.
Polity is not the only available democracy indicator if we shorten the time span. I used the
average of the political rights and civil liberties indexes from Freedom House, to calculate the
aggregate seven-point scaled Freedom House Index (FHI), and the dichotomous ACLPmeasure of democracy, which scores countries after the existence of free and competitive
elections. The analysis starts in 1972 when the FHI is used as democracy-indicator and in
1960 when ACLP and Polity are used. The FHI can be labeled a more inclusive democracymeasure and the ACLP-measure a more restrictive one than Polity. In general, the FHI
provides an even larger estimated effect from democracy on economic growth than Polity,
when going all the way from most autocratic to most democratic. I used a regression model
that in addition to the democracy-indicator incorporated GDP per capita, the logarithm of
population, colonizer-, region- and religion dummies, energy production per unit of GDP, the
logarithm of regime duration and the share of the population living in urbanized areas. In
addition, a linear time trend was added in all the analyses, except for Random- and Fixed
12
All the results are described in a more detailed fashion in Knutsen (2008a), where these analyses were originally
conducted.
39
Effects, which relies heavily on intra-country variation when producing estimates. Table 7
below shows the results from different analyses using the various democracy-indicators. As
we see from the table, the only specification not yielding a significant coefficient at the 10%level is the Polity-specification when using OLS with panel corrected standard errors. The
FHI-specification for the same analysis yields a coefficient that is significant at the 5%level.13 The estimated effect of going from most autocratic to most democratic in this
analysis, when keeping all the controls constant is approximately 1,6% extra annual growth
when using the FHI and 0,8% when using Polity. The estimated effects are even stronger
when using Fixed Effects, where all the democracy-indicators are significant even at the 1%
level. The estimated effect of going from most autocratic to most democratic is here 2% extra
annual growth in GDP per capita for Polity and even more for the FHI. Random Effects gave
somewhat less strong results than Fixed Effects, but the FHI and Polity-specifications were
significant at the 5%-level. The ACLP-measure however, was only significant at the 10%level.
Table 7: Results from the post-WWII analysis
Democracy- Coefficient
Method
measure
democracy T-value
13
Observations Countries
I added some possibly (they might be viewed as endogenous to regime type!) relevant variables to the extended FHI-model
and checked the results: The estimated effect was still significant at the 5%-level, and the absolute size of the estimate
actually increased when I controlled for trade as a percentage of GDP, final government consumption expenditure as a
percentage of GDP, percentage of GDP in agriculture and the latest assigned Gini-coefficients (assigned also for the five
years before first score; many years therefore have identical scores). I have argued that investment level (investment level as
share of GDP) and human capital (here proxied by gross enrollment in secondary schooling) are to be viewed as causal
channels from democracy on growth. However, if I check the robustness of my statistical findings by including these
variables as well, the FHI-measure is now actually significant at the 1%-level. If I also add ores and metal export as a
percentage of merchandise export as an extra control, the FHI-coefficient is still significant at the 5%-level but not at the
1%-level. These inclusions of extra control variables to my extended model indicate that at least in the case of PCSTS using
FHI as operationalization, the effect from democracy on growth is robust. When I included the ores- and metal variable in the
model without investment and human capital however, the FHI-coefficient was not significant at the 5%-level.
40
OLS with PCSE
FHI
0,26
2,29**
3016
118
OLS with PCSE
Polity
0,04
1,57
2927
117
OLS with PCSE
ACLP
0,56
1,66*
2783
118
FE
FHI
0,37
3,09***
3016
118
FE
Polity
0,10
3,42***
2927
117
FE
ACLP
1,01
2,58***
2783
118
RE, ML
FHI
0,30
3,26***
3016
118
RE, ML
Polity
0,06
2,72***
2927
117
RE, ML
ACLP
0,57
-1,79*
2783
118
Finally, I conducted Matching analyses on the post-war sample. I used the FHI, with two
different cut-off points for being considered democratic, 2.5 and 3.5, and I used the
dichotomized ACLP-measure. The standard errors are robust, replacement is allowed and I
used different numbers of matches to check the robustness of the results. In general, the
effects were statistically significant at conventional levels when the number of matches were
higher than one. The results clearly indicate that there is indeed a positive and significant
effect from democracy on economic growth
Table 8: Average treatment-effects when going from dictatorship to democracy from the
matching analyses (FHI)
FHI cut-off
3,5
2,5
3,5
2,5
3,5
2,5
3,5
2,5
Matches
1
1
2
2
4
4
10
10
Coefficient
0,48
0,30
0,59
0,60
0,69
0,59
0,86
0,83
41
t-value
1,06
1,00
1,98**
2,59***
2,88***
2,51**
4,09***
4,41***
Observations
3016
3016
3016
3016
3016
3016
3016
3016
Table 9: Average treatment-effects when going from dictatorship to democracy from the
matching analyses (ACLP)
Matches
Coefficient
t-value
Observations
1
0,45
-1,25
2783
2
0,77
-2,56**
2783
4
0,71
-2,82***
2783
10
0,60
-2,69***
2783
42