Democracy and Growth: A Dynamic Panel Data Study1
Jeffry Jacob†
Department of Business and Economics
Bethel University
Thomas Osang
Department of Economics
Southern Methodist University
February 2015
ABSTRACT
In this paper we investigate the idea whether democracy can have a direct effect on economic growth. We
use a system GMM framework that allows us to model the dynamic aspects of the growth process and
control for the endogenous nature of many explanatory variables. In contrast to the growth effects of
institutions, regime stability, openness and macro-economic policy variables, we find that measures of
democracy matter little, if at all, for the economic growth process.
JEL: O1, F1
Keywords: Democracy, Growth, Institutions, Openness, Dynamic Panel
1
We would like to thank Chris Barrett and seminar participants at Bethel University and the Conference on Economics of
Global Poverty, Gordon College, for useful comments and suggestions.
†
Department of Business and Economics, Bethel University, 3900 Bethel Dr., St. Paul, MN 55112;
E-mail: [email protected]: Tel: (651) 638-6715

Department of Economics, Southern Methodist University, 3300 Dyer, Dallas, TX 75275; E-mail: [email protected]; Tel:
(214) 768-4398; Fax: (214) 768-1821
1. Introduction
Over the past several decades, and in particular after the fall of communism in Eastern Europe and
Western Asia, the world has seen a steady rise in the number of countries that became democracies. In 1960,
only 39% percent of the world’s countries had a democratically elected government, compared to 61% percent
in 20102. While this dramatic rise in the world’s level of democratization has undoubtedly affected the
political and social context of people’s lives, economists have been examining the question whether more
democracy had any discernable impact on people’s material well-being, that is, economic growth and
development. Of crucial importance for this inquiry is an observation that dates back to the seminal work of
Huntington (1968) who was one of the first to claim that the democracy-growth link may be tangential. In
particular, Huntington argued that it is the effectiveness and stability of the policy-making process rather than
a country’s level of democracy that matters primarily for economic progress.3 In this paper, we re-examine the
empirical democracy-growth nexus in the spirit of Huntington. To do so, we distinguish between measures of
democracy - variables that measure the extent to which governments are elected democratically -, and
measures of the quality of public institutional – variables that measure the effectiveness and stability of the
political decision making process. These two sets of measures can be quite distinct. While democratic states
are generally linked to higher degrees of political stability, there are plenty of examples of unstable and
ineffective democracies. Furthermore, as Huntington pointed out, political stability and effective governance
can be found in non-democratic societies such as the countries of Eastern Europe during the cold war era.
The economics and political science literature considers several reasons why democracies should have
a positive impact on economic growth. First and foremost, democratic countries grant its people the right to
remove a bad government from office, a luxury residents living under autocratic or dictatorial rules can only
dream of. The option of removing a bad or dysfunctional government opens the possibility that the
government voted into office will do a better job at governing the country thereby lowering the cost of doing
business for firms and individuals alike. This increase in efficiency in turn improves the country’s economic
output and raises its growth rate. Another argument in favor of the connection between democracy and
prosperity stresses the importance of the predictable transfer of power in democracies which in turn increases
2
3
The numbers are based on the Polity2 democracy variable from the Polity IV database. See the data section for further details.
“The United States, Great Britain and the Soviet Union have different forms of government, but in all three systems the
government governs” (Huntington, 1968).
stability and hence growth. Also, stronger redistribution of income in democracies can increase political
stability and reduce the harmful external effects of extreme poverty, both of which are growth enhancing.
Democracies may however not always produce better economic outcomes. In effect several arguments
have been advances why democracies may be detrimental for economic development. First, democratic
elections can be manipulated by corrupt government officials. A similar argument points out that
democratically elected officials can turn into dictators. Second, democracies can suffer from political gridlock
and the fight between contending parties over a country’s resources. Third, the existence of checks and
balances in a democracy may delay difficult policy choices. Fourth, the preoccupation of policy makers with
their re-election may cause higher levels of government consumption compared to non-democracies. Fifth, the
possibility of democratic countries to achieve higher levels of income redistribution may lead to substantial
efficiency losses due to high levels of taxation. Finally, a higher government turnover rate corresponds to a
rise in political uncertainty.
When economic theories point to opposing outcomes, the question needs to be settled by empirical
research. Unfortunately, the empirical literature too is inconclusive when it comes to the impact of
democracy on growth. Some studies find a positive effect of democracy on growth, some report a
negative impact, and a third group finds that the results are inconclusive.4
While the empirical literature examining the role of democracy on economic growth and
development is extensive, there are only a handful of papers that employ the cross-section time-series data
approach used in our study. One of the first empirical studies was conducted by Barro (1996) who found a
weakly negative impact of democracy on economic growth. Although Barro’s study is based on repeated
observations across countries, the estimation approach did not explicitly control for country or time fixed
effects. Controlling for country specific heterogeneity in a panel data set up, Rodrik & Wacziarg (2005)
found that countries undergoing democratic transitions, on an average, experience higher economic
growth in the subsequent periods. Complementing this finding, Persson & Tabellini (2008) show that a
country’s transitioning out of democracy results in a substantial negative impact on growth. Investigating
4
The early empirical literature on democracy and growth is surveyed by Przeworski & Limongi (1993) and Brunetti & Weder
(1995). While a number of studies find a positive impact of democracy on growth (Acemoglu et al., 2014; Acemoglu, 2008;
Burkhart & Lewisbeck, 1994; Epstein, Bates, Goldstone, Kristensen, & O'Halloran, 2006; Londregan & Poole, 1996;
Scully, 1988, for example), other papers report a negative democracy effect (see, for example, Barro, 1996; Helliwell, 1994;
Landau, 1986; Tavares & Wacziarg, 2001). A third group of studies report inconclusive results, among them (Alesina, Ozler,
Roubini, & Swagel, 1996; Barro & Lee, 1994; Barro, 1991).
the connection between the sequencing of economic and political reforms, Giavazzi & Tabellini (2005)
concluded that countries transitioning into democracies have higher economic growth if democratization
is preceded by economic reforms. At a regional level, Rock (2009) found that in a sample of Asian
countries, while democracy by itself does not exert a statistically significant impact on economic growth,
democracy interacted with a measure of state capacity (constructed from bureaucratic quality and rule of
law) had a positive impact on growth. In another regional study, Bates, Fayad, & Hoeffler (2012) find that
the recent wave of democratization in the Africa countries has had a positive impact on growth.
Measuring growth volatility as the standard deviation of GDP growth over five year intervals, Yang
(2008) finds no statistically significant relationship between growth volatility and democracy. Using
annual data for a large number of countries over 40 years, Acemoglu, Robinson, Naidu, & Restrepo
(2014) find a statistically significant positive role of democracy.
We extend the empirical democracy and growth literature in several ways. First, we use several
measures of democracy to account for differences in definitions and methodologies used to construct
these measures. Second, we control for the quality of public institutions in the spirit of Huntington and
North. Third, we control for the impact of economic integration and geography on economic
development. Fourth, our use of systems GMM approach (Arellano & Bover, 1995; Blundell & Bond,
1998) enables us to correctly model the dynamics of growth, account for possible simultaneity bias by
using instruments from within the model and also obtain estimates for the time invariant geography
measures. Fourth, and very importantly, a recent study by Bazzi & Clemens (2013) has shown that most
of the studies using dynamic panel data in the economic growth literature may suffer from
underidentifiation and weak instrumentation. We conduct a variety of diagnostic tests to check the
appropriateness of estimation results in this regard. Finally, we extend our analysis by including several
policy variables introduced in the influential study by Dollar & Kraay (2003).
Using data from over 120 countries over the 1961 to 2010 period, we find that the various
democracy measures do not have a significant impact on economic growth. However, the variables that
measure the quality of institutions and political stability exert an appreciably positive and statistically
significant effect on growth. These results are confirmed by a variety of robustness checks, as well as the
extended model estimates that includes several policy controls.
The rest of the paper is organized as follows. Section 2 contains the empirical model and a
discussion of the estimation methodology. We describe the data in Section 3. In Section 4, we derive and
interpret the estimation results as well as robustness checks and extensions. Section 5 concludes.
Appendix A consists of variable definitions and summary statistics. The estimation results are given in
Appendix B.
2. Empirical Model
2.1 Structural Model
A study of the relationship between democracy and growth needs to account for other
determinants of economic growth. Over the last decade and a half, a lot of attention has been given to the
study of the “deeper” determinants of economic development as coined by Rodrik, Subramanian, &
Trebbi (2004). According to this approach, factors which affect economic development can be classified
into two tiers. While inputs in the production function such as labor, physical and human capital directly
affect income and thus economic development, they themselves are determined by deeper and more
fundamental factors. And although it remains an open question what exactly constitutes a “deeper”
determinant of development, three broad categories have emerged in the literature: geography,
institutions, and international trade (integration).5 Hence, we embed our investigation of the role of
democracy into the deep determinants framework:
Income = f (Democracy, Institutions, Integration, Geography)
Geographical factors typically characterize the physical location of a nation such as distance from
the equator, access to sea, agro-climatic zone, disease environment, soil type, and natural resources.
Geography may matter for development through its impact on transaction costs. For example, a country’s
size, access to sea and topography can crucially affect transport costs and the extent of its integration with
the world market. Latitude and climate are also related to disease environment, which directly impacts
labor productivity and life expectancy, among others. Geography can also impact economic development
through institutions. Climate and soil affect the types of crops planted. This, along with the availability of
natural resources, can dictate whether the early institutions were extractive or productive. In fact, some
authors like Gallup, Sachs, & Mellinger (1999) and, more recently, Sachs (2003) argue that geography is
5
Easterly & Levine (2003) provide a good overview of studies analyzing the three determinants. In this paper geography is
defined as physical geography, as opposed to economic geography as discussed in Redding & Venables (2004).
the most important variable of interest for development, even after controlling for the quality of
institutions and other factors.
The argument for economic integration as a fundamental determinant of development is based on
the gains from trade argument. Next to the classic case of comparative advantage gains are more modern
approaches that stress the importance of trade in the transfer of new technologies and ideas, which in turn
enhance productivity. Moreover, supplying to a larger international market allows higher degrees of
specialization and thus entails productivity gains. There are many empirical studies on the link between
international trade or integration and economic development. One of the more influential ones is Sachs,
Warner, Åslund, & Fischer (1995) who constructed an index of openness and found that greater openness
leads to higher growth. As with institutions, trade variables are likely to be endogenous with regard to
income. Frankel & Romer (1999) examine this issue in detail. They construct a predicted trade share for
countries based on their geographic and size characteristics. This variable is used as an instrument for
actual share and their findings point to a positive link between integration and income.
In addition to Huntington, the importance of the quality of public institutions for the development
process was emphasized in the works of North (1993, 1994a, 1994b, 1994c). His motivation to consider
institutions can be linked to his view that the neo-classical theory is unable to explain widespread
differences in economic performances across countries. If only factor accumulation led to progress, then
all countries would do so, provided there was a high-enough payoff involved. Differences in income thus
require differences in “payoffs” which is where institutions come in (North, 1994a). Institutions are the
rules of game which a society lays down for itself and which determine the incentives people face and
thus the choices they make. Another way of looking at institutions is through their impact on transaction
costs. Well defined rules and their smooth enforcement, i.e. better institutional quality, greatly reduce
transaction costs economic agents face and thus lead to more efficient outcomes (North 1993, 1994b).
Hall & Jones (1999) was one of the first empirical studies to examine the impact of institutions on
economic development. The importance of institutions was further developed by Clague, Keefer, Knack,
& Olson (1999), Acemoglu, Johnson, & Robinson (2001) and Acemoglu & Johnson (2005) who show
that good institutions, by fostering productive investments lead to favorable economic outcomes.
Following, Caselli, Esquivel, & Lefort (1996) and Dollar & Kraay (2003) among others, the above
structural model can be specified as a reduced form dynamic panel data model:
git = b0 + b1yi,t-1 + b 3 Xit + b 4 Zi + n i + g t + mi,t
(1)
where, gi,t is growth rate in GDP per capita, yit-1 is the previous period’s GDP per capita capturing
the idea of convergence, Xit is the set of time varying variables democracy, institution and trade, Z i is the
set of time-invariant exogenous such as geographical measures,  i is the country specific fixed effect, g t
is the time dummy and
mit is the idiosyncratic error term. Using the definition of growth rate as
yi,t - yi,t-1 and rearranging variable, this model can be rewritten in levels of GDP per capita as:
yit = b0 + b1 yi,t-1 + b 3 Xit + b 4 Zi + n i + g t + mi,t
(2)
where b1 = 1+ b1 .
Estimation of (2) poses a number of difficulties that need to be addressed. Endogeneity of the
lagged dependent variable and the possible endogeneity of institution, democracy and trade measures, due
to measurement error, and/or reverse causality6 are typically addressed by using instrumental variable
estimation methods. Finding suitable instruments in the context of economic development- variables
correlated with the endogenous variables but not directly with income- is no easy task7. External
instruments are hard to find and those that have been used successfully like settler mortality as an
instrument of institutions, are time invariant and thus not useful in the panel data context. As an
alternative, the use of internal instruments as in the popular GMM estimator has gained momentum.
2.2 GMM Dynamic Panel Data Estimation
The standard approach is to estimate the above model in first differences, using previous lags of
the explanatory variables as instruments (Caselli et al., 1996; Dollar & Kraay, 2003)8. Specifically, the
estimating model thus becomes:
yi,t - yi,t-1 = b%1 (yi,t-1 - yi,t-2 ) + b 3 (Xit - Xi,t-1 ) + g%t + ( mi,t - mi,t-1 )
6
(4)
See Frankel & Romer (1999), Hall & Jones (1999) , Acemoglu et al. (2001), and Baier & Bergstrand (2007).
Durlauf, Kourtellos, & Tan (2008)
8
For a more recent application of the dynamic panel data method in the context of corruption and growth, see Swaleheen
(2011)
7
Though first differencing eliminates the individual fixed effects, there is still endogeneity in the
model since E[(yi,t - yi,t-1 )(mi,t - mi,t-1 )] ¹ 0 . The dynamic panel data estimator developed by (Holtz-Eakin,
Newey, & Rosen (1988) and Arellano & Bond (1991) addresses this issue by using two period or more
lags of the explanatory variables as instruments for the differenced variables:
E[yi,t-s (mi,t - mi,t-1 )] = 0, for t = 3,4...T and s ³ 2
Blundell & Bond (1998) show that this difference estimator may not perform well when there is
persistence in the lagged dependent variable and that the systems GMM, initially proposed by Arellano &
Bover (1995) may be better suited. The systems GMM is based on the idea that additional moment
conditions can be introduced by adding the level equation to the differenced equation and using lagged
differences of the explanatory variables as instruments for the level equation, that is:
E[ mit (yit-1 - yi,t-2 )] = 0 for t = 3,4...T
In most of our specifications, the estimated value of b1 was close to unity, making the Systems
GMM estimator more suitable. Another advantage of the systems estimator is that while the ArellanoBond estimator purged all time invariant measures from the estimating equation, Roodman (2009) shows
that time-invariant exogenous variables (which are orthogonal to the individual fixed-effects) can easily
be included in the systems-GMM model. In our context, this means that the economic performance
impact of both time varying and time-invariant measures of the three deep determinants along with
democracy can be estimated directly.
Another advantage of the systems GMM approach is that the endogeneity of the explanatory
variables in X can be addressed by using appropriate lags of these variables as instruments. For example,
if E[xit mis ] ¹ 0 for s £ t but E[xit mis ] = 0 for s > t , two or more lags of X could be used since,
E[xi,t-2 (mit - mi,t-1 )] = 0 . Thus, one does not need to use external instruments to control for potential
endogeneity in the model. There are two other possible causes of endogeneity- measurement error and
omitted variable bias. We address the former by using alternative measures of democracy including an
index that draws upon ten underlying democracy variables. Finally, as indicated in equation 1, we control
for the three deep determinants in all specifications and, in addition, for macroeconomic policy variables
in the extended model to minimize the potential omitted variable bias.
2.3 Specification tests for the dynamic panel data model
To test the validity of our Systems GMM estimates, we perform a battery of tests. First, since lagged
values are used as instruments, unbiased estimation requires the absence of second-order serial correlation
in the error term (see Arellano and Bond, 1991). To test this requirement, we perform the Arellano-Bond
AR(2) test. A p-value of greater than 0.05 implies the absence of second-order autocorrelation. In that
case, the systems GMM can be applied without any adjustments to the instrument set. A p-value of less
than 0.05 indicates the presence of an MA error term of order one or higher. In this case, the model needs
to be re-estimated with the instrument set lagged by an additional period (Cameron & Trivedi, 2005). To
test whether the modified Systems GMM estimator has the correct error structure, we test for the absence
of a third-order autocorrelation using the Arellano-Bond AR(3) test. Failure to reject the null (p-value of
greater than 0.05) indicates the absence of higher order serial correlation.
Second, to test the validity of the exclusion restrictions, we perform the Hansen J-test. Under the
null hypothesis, the instruments are correctly excluded from the model. Since we use Systems GMM, we
report a second test of the exclusion restrictions known as the difference-in-Hansen test. This test checks
the validity of the additional exclusions restrictions that arise from the level equations of the Systems
GMM model (see Roodman, 2009b).
Our final battery of tests is motivated by the issues of instrument proliferation, underidentification
and weak instruments in the Systems GMM estimator. Roodman (2009) shows that having numerous
instruments, which usually is the case in GMM estimation, can result in an over-fitting of the model. This
in turn can fail to rid the explanatory variables of their endogenous components, potentially leading to
biased estimates. In this case both Hansen tests may produce very high p-values, often close to 1. If this
happens, the instrument set should be reduced by either restricting the number of lags (Beck, DemirgucKunt, & Levine, 2000) or by “collapsing” the instrument set into a smaller dimension matrix (Roodman,
9
2009a; Vieira, MacDonald, & Damasceno, 2012) . Hence, our systems GMM estimates shown below on
the restricted number of lags of the instrument sets.
As Bazzi and Clemens (2013) point out, GMM estimators can suffer from underidentification or
weak instruments or both, making it tenuous to conduct any meaningful hypothesis testing. They also
show that these problem may persist even after reducing the number of instruments. Currently, there are
no formal tests to tackle underidentification/weak instrument issues in the dynamic panel data context.
However, Bazzi and Clemens (2013) suggest a number of ad-hoc tests as a second best solution. First, to
test the null hypothesis of underidentification, we use the Kleibergen-Paap (K-P) rk Wald test to the first
stage of both the level and the difference model. Second, even if underidentification is rejected, there may
be a weak instruments problem, in the sense that the correlation between the endogenous variables and the
instruments could be quite small.
To test for the presence weak identification under the assumption of iid erros, Stock & Yogo
(2005) proposed using the Cragg Donald F stat and plotted the critical values for different
weak
intrument biases of the two stage least square relative to OLS. However, there are two issues with their
approach. First, the iid assumption may be violated. Second, Stock and Yogo (2005) construct the critical
values for their procedure for only three endogenous variables, which limits its use in settings where the
number of endogenous variables is large. For both reasons, we report the K-P rk Wald F stat instead.
Finally, we report two-step robust standard errors corrected for finite sample (Windmeijer, 2005)
throughout.
9
æ
ç
ç
ç
ç
ç
çè
For the difference equation, the restricted (with the lag length equal to two) and the collapsed instrument sets take the form:
0
0
0
yi1
0
0
0
0
0
0
0
0
0
0
yi 2
yi1
0
0
0
0
0
0
yi 3
yi 2
M
M
M
M
M
M
æ 0
¼ ö
ç y
¼ ÷
ç i1
÷
¼ ÷ and ç yi 2
ç
÷
¼ ÷
ç yi 3
çè
÷
O ø
0
0
0
0
yi1
0
yi 2
yi1
¼ ö
¼ ÷
÷
¼ ÷ , respectively.
÷
¼ ÷
÷ø
For the level equation the corresponding instrument sets are:
æ 0
ç Dy
i2
ç
ç 0
ç
ç 0
çè
0
0
0
0
Dyi 3
0
0
Dyi 4
æ 0
¼ ö
÷
ç Dy
¼
i2
÷
ç
¼ ÷ and ç Dyi 3
÷
ç
¼ ÷
ç Dyi 4
÷ø
çè
ö
÷
÷
÷.
÷
÷
÷ø
3. Data
The data set covers the five decades from 1961 to 2010. Following Islam (1995) and Caselli et al.
(1996), we use five-year averages of all our variables from 1960-2010, giving us a maximum of 10 time
periods. Taking averages ensures that, to a large extent, short-term fluctuations in these variables resulting
from changes in business cycles can be smoothed out and that we are capturing the longer-term impact of
our explanatory variables on economic growth. The cross-section dimension varies between model
specification, ranging from N=100 to N=165.
We use four different democracy measures, all of which are explained in detail in Table A1 below.
Our first, measure is the institutionalized democracy score from the Polity IV dataset, P-IV Democracy.
Our second democracy measure, Polity Score, also comes from the Polity IV dataset. As a third
democracy measure, FH Index, we use the average of political rights and civil liberties scores from the
Freedom House dataset. Our final measure is the Unified Democracy Score from Pemstein, Meserve, &
Melton (2010), a measure that is constructed from ten different democracy measures, including the tree
previously discussed measures.
To capture the extent of regime turnover, we use Regime Instability, taken from the Database of
Political Institution (Beck, et al, 2011). This measure is used as a proxy for the extent of uncertainty of
the political decision making process and has been widely used in the literature (Bornea at al, 20,
Huntinglton, 1967)
Our preferred measure of institutions is contract intensive money (CIM), defined as the ratio of
non-currency money to total money in an economy, as proposed by Clague et al. (1999). The basic
argument for such a measure stems from the fact that in societies where the property and contract rights
are well defined, even transactions which heavily rely on outside enforcement can be advantageous10.
Currency in this setting is used only in small transactions. Agents are increasingly able to invest their
money in financial intermediaries and exploit several economic gains. Thus, stronger institutions would
be manifested by a greater share of CIM.
10
A similar point is made by Acemoglu & Johnson (2005) who examine both, “contracting institutions” and “property rights
institutions”. The former govern contracting relationships between private parties and the latter govern the relationship between
private citizens and those with political power. They show that while contracting institutions may be useful for financial
intermediation, it is the property rights institutions that have a positive impact on economic growth and development.
But is CIM indeed a measure of institutional quality and not just a proxy for financial
development? Based on a case study of seven countries, Clague et al. (1999) show that CIM tracks major
political developments that have little direct impact on the financial sector. In fact Clauge et al. (1999)
characterize CIM as the “contract-intensive money indicator of property rights enforcement (pp. 186)”. In
this regard, CIM can be seen to be a good proxy for both Acemoglu & Johnson (2005) property rights
measures as well as North’s (1994) definition of institutions. Finally, an additional benefit of the CIM
measure is that it is more objective and more precisely measured than most organization measures, which
are often survey-based and thus suffer from biases and measurement errors.
Another important measure of institutions is the “number of veto players” (Veto Players), which
captures the extent of checks and balances within the government and is obtained from the World Bank’s
Data on Political Institutions database (Thorsten Beck, Clarke, Groff, Keefer, & Walsh, 2001). The
motivation here is that countries with multiple decision makers offer greater protection to individuals and
minorities from arbitrary government action (Keefer & Stasavage, 2003). This measure counts the number
of decision makers in the government, taking into consideration whether they are independent from each
other. We use constraints on the executive, Constraints on Exec, as our final institution measure
The remaining explanatory variables, all of which are explained in detail in Table A1 in the
Appendix, fall into three categories: Openness to international trade (Trade Share and Real Openness),
geography (Dist. Equator and Malaria Ecology) and macroeconomic policy variables (BMP, Inflation
rate, Govt. spending/GDP).
Tables 1a and 1b present the summary statistics and simple correlation coefficients of the time
averaged variables.
(Insert Table 1)
4. Empirical Results
4.1 Baseline Model Results
Our benchmark model is presented in Table 2. Both of our trade measures and CIM are used in log
form implying that the coefficients can be interpreted as long-term elasticity estimates of GDP per capita
with respect to these variables. We consider all time-varying right hand side variables to be predetermined
except for the trade and institutions, which we treat as endogenous. The two time-invariant geography
measures are considered to be exogenous. To check whether instrument proliferation is an issue, we first
estimate the model with the full instrument set (columns 1-3) followed by the estimation with the
collapsed instrument sets (columns 4-6) and the restricted instrument set (columns 7-9).
(Insert Table 2 here)
In cols 1 and 2, Real Openness is used as a measure of trade while in column 3, it is replaced
with Trade Share. We also use Distance Equator and Malaria Ecology as two alternative measures of
geography. Both, CIM and the two trade measures have a positive and significant impact on long term
growth, with the average elasticity of CIM in columns 1-3 being larger than that of the two openness
measures by a factor of three. Dist Eq. as a measure of geography also has a positive and significant
impact on growth as well. The latter result is similar to Spolaore & Wacziarg (2013) who found that
among the different measures of geography they considered, a country’s latitude had a significant impact
on
income.
Malaria
Ecology
in
column
2
does
have
the
expected
sign
but
is
not
significant. The implied coefficient on the lagged dependent variable indicates a fairly slow rate
of convergence, which is similar to the results in Dollar & Kraay (2003). Most importantly, the
coefficient estimate of our democracy measure, P-IV Democracy, is quite small and negative throughout.
However, it is not statistically significant in any of the specifications, a result that is reminiscent of the
finding in Barro (1996).
In terms of diagnostic tests, we first examine the serial correlation of the error structure. The low
AR(2) test p-values indicated the presence of MA(1) error terms in all models necessitating the need for
lagging the instrument set. The high p-value of AR(3) tests implies the absence of serial correlation in the
error structure. Regarding the first stage diagnostics, the Hansen Test indicates that the instruments in the
system GMM are correctly excluded. Though system GMM does perform better in the presence of
persistence of series (coefficient of lagged dependent variable is close to one) it may suffer from a
drawback in that the additional instruments brought about in the process may overfit the model and
may not be able to remove the endogenous component from the estimation (Roodman, 2009b). We
also report the instrument count for each of the equations of the systems GMM. With the full instrument
set, there are 150 instruments compared to 146 countries. While the Hansen test for overall exclusion of
the instrument set does perform well, the high p-value of the difference-in-Hansen test, which tests the
exclusion of the additional instruments in the levels equation, does indicate a potential problems with
identification. We report the Kleibergen-Paap (K-P) rank Wald Test for underidentification, one for the
difference and one for the levels equation. While the difference model does not suffer
from underidentification, the problem does exist for the levels model, as the high p-values do not allow us
to reject the null of underidentification. The K-P rk Wald F values of testing for weak instruments are the
low side, with substantially higher value for the difference equation than the level equation. In any case
we cannot rule out a potential weak instrument bias in our estimates. Since this problem primarily stems
from instrument proliferation, one solution is to restrict the instrument count, either by limiting the
number of lags of the endogenous variables entering the instrument set or by collapsing the instrument
set.
Columns 4-6 present the results with the collapsed instrument set, which reduces the number of
instruments to 38. Like the previous estimation, past GDP and CIM continue to be statistically significant
and exert a stronger impact of growth than Real Open or geography. In addition, the latter measures are
not statistically significant. As before, the P-IV Democracy coefficient estimates are negative and
statistically insignificant. The p-value of the Hansen test rejects the null of correct exclusion restriction in
all three specifications, casting doubts on the validity of the results.
As an alternative to both the full and collapsed instrument sets we employ a hybrid approach
where we restrict the instrument set to second and third lag for trade and institution, third and fourth
for lagged GDP and first lag for the democracy variable (columns 7-9). This produces a set of 93
instruments. Lagged GDP and CIM continue to have strong and significant coefficient estimates. In
addition, all openness measures and the latitude measures have the expected sign and are statistically
significant. Interestingly, P-IV Democracy is not only negative as before but it is also statistically
significant in two cases. In terms of diagnostics, both Hansen tests point to the validity of the exclusion
restrictions. Furthermore, the unlike the full instrument set results, the low enough p-values of difference-
in-Hansen test indicate that instrument proliferation is not an issue. As in the case of the previous two set
of results, while the difference model does not suffer from underidentification, the problem does exist for
the levels model, as the high p-values again do not allow us to reject the null of underidentification. Also,
the two K-P F stats indicate that weak-instruments is likely to be an issue.
As a result of the comparison between the full, collapsed and restricted instrument set reported in
Table 2, we will employ the restricted instrument set approach throughout the remainder of the paper. We
will thus focus on the restricted model the remainder of the paper.
In the next subsection, we check the robustness of the baseline results. Specifically, we use
alternative measures of democracy (Section 4.2.1), alternative model specifications (Section 4.2.2) and
yearly data in contrast to five year averages (Section 4.2.3).
4.2 Robustness Checks
4.2.1 Alternative measures of democracy and institutions
(Insert Table 3 here)
In Table 3, we introduce alternative measures of democracy in columns 1 through 8. As with P-IV
Democracy, we treat these measures are pre-determined. Our first alternative measure of democracy is,
Polity Score. We classify a country as democratic for a given five year period if it was democratic for
three or more years in that period. This measure is not statistically significant in our estimation (columns
1 and 2). The second alternative measure of democracy, the average of the scores of civil liberty and
political rights from the Freedom House dataset, FH Index also does not have a statistically significant
impact on economic growth (columns 3 and 4). In columns 5 and 6, we use a comprehensive unified
democracy score, a measure that is constructed from ten different democracy measures used in the
literature. The parameter estimate of this measure is also statistically insignificant. In addition to
these three broad democracy measures, we also consider tangential measure of democracy-Regime
Instability. This measure has a strong adverse and statistically impact on growth (column 7 and 8). In all
these regressions, CIM and the two openness measures (except for columns 1 and 2) continue to be
positive and statistically significant. The geography measure is statistically significant in only three out of
these eight specifications. These results suggest that controlling for institutions and openness, broad
democracy measures do not have much impact on economic growth.
In columns 9 and 10, we include two broad democracy measures (Polity Score and FH Index) and
instability of tenure along with the measures of institutions, trade and geography. The results for lagged
income, institutions, trade and geography have the expected signs and are statically significant. Among
the three democracy measures, only Regime Instability is statistically significant and with the expected
negative sign in both regressions.
With regard to the diagnostic tests, we evaluate the various models using three criteria. First, we
make sure that there is no serial correlation in the differenced error terms. Next, we check for the correct
exclusion of the instrument set. Finally, we check for the absence of underidentification in both the
difference and level structural equations of the system GMM estimator. With all p values less than 0.05
for the AR(2) test, we reject the null hypothesis of serially uncorrelated error terms. Thus all results in
Table 3 have the instrument set lagged by an additional time period (to incorporate MA(1) error terms).
The two Hansen tests have p-values greater than 0.05 indicating that we cannot reject the null hypothesis
that the instruments are correctly excluded. None of these values are close to unity either, suggesting that
instrument proliferation might not be an issue in these regressions. In most cases, The K-P tests of the
first stage structural equations reject the null of underidentification for both, the levels and the difference
equations. The low F stats for the K-P Weak id tests point to the possibility of weak instrument bias and
does remain a concern.
There is also a possibility that lagged democracy may have a positive impact on current economic
growth. To test this, we re-estimated Table 3 with replacing current democracy variables with their five
year averaged lagged values. The results, which are not reported here, stay essentially the same. One
exception is that coefficient on the lagged Regime Instability is also statistically insignificant.
(Insert Table 4 here)
Next, we check the robustness of our results to alternative measures of institutions (Table 4). In
columns 1-4, we use Veto Players as our institutional measure along with Trade Share as the measure of
openness. In three out of these four specifications, Veto Players is statistically significant with the
expected positive sign. Trade Share also has a positive impact of economic growth. Among the various
democracy measures, only Regime Instability is statistically significant and has an adverse impact on
economic growth. In columns 5-7, Trade Share is replaced by Real Openness as a measure of openness.
Again, while Polity Score and FH Index have the right sign, they are not statically significant, Regime
Instability continues to be negative and significant. Our third measure of intuitions, Constrains on Exec is
used in columns 8-11. It is statistically significant in only one specification. Overall, the results in Table 4
also suggest that institutions and trade have the strongest impact on economic growth while most
democracy variables, except Regime Instability, do not have a statically significant impact.
4.2.2 Alternative model specifications
We estimate different model specifications incorporating interactions between institutions and
democracy. We also test the non-exchangeability hypothesis discussed in Brock & Durlauf (2001) and
Durlauf, Kourtellos, & Tan (2008).
Interaction between democracy and state capacity were first analyzed in Rock (2009) who found
that the interaction between state capacity (a measure capturing the quality of institutions) and measures
of democracy has a positive impact on economic growth. Further, Giavazzi & Tabellini (2005) found a
positive impact of democratization on economic growth if democracy was immediately preceded by
economic reforms. To this extent, we interact current democracy variables with previous period’s CIM.
The results are presented in Table 5. Alternating between each of the five democracy measures, columns
1-5 consist of these measures along with their interaction with lagged CIM, while Real Openness is used
to capture openness. Only two democracy variables- FH Index and Regime Instability- are statistically
significant along with their respective interaction terms. In columns 6-10, we keep the same set up but
replace Real Openness with Trade Share. Four democracy measures are significant but P-IV Democ and
Polity Score have a negative sign. Thus there is some evidence that interaction of democracy with
institutional quality does have a positive impact on economic growth. However, when we re-estimated
Table 5 by including CIM in all these specifications, neither the democracy measures nor the interaction
terms were statistically significant, while, as before, CIM was positive and statistically significant. This
seems to suggest that the institutional quality is a stronger driver of economic growth than the sequencing
of economic and democratic reform. Regarding the diagnostics, based on the AR(2) and AR(3) tests, all
specifications are estimated incorporating MA(1) error terms. The two Hansen tests cannot reject the null
of exclusion restrictions of the instruments and do not point towards instrument proliferation. Further, the
K-P Wald test rejects the null of under-identification in almost all difference models and in four of the ten
levels models.
(Insert Table 5 here)
Some papers have cast doubt on the validity of the results from growth regressions using a cross
section of countries (see Brock & Durlauf, 2001 and Durlauf, Kourtellos, & Tan, 2008 for a discussion).
Their concerns broadly revolve around three categories. First, most growth empirics have involved crosssection estimation. Cross section regressions however assume parameter homogeneity, an assumption that
is not likely to be true. As Brock and Durlauf (2001) note, panel data corrects a part of this problem by
allowing for country fixed effects. However, identification of cross section heterogeneity may be difficult
if the panel data does not include a sufficient number of time periods. This may not be an issue in our
estimation since even after taking five-year averages, our sample includes upto ten time periods. The
second problem with growth empirics is the issue of endogeneity and the difficulty of finding valid
instruments. As discussed above, we have addressed this issue by using a systems GMM approach and
reporting extensive first stage diagnostic tests to check the underlying estimation assumptions. Third,
Brock and Durlauf (2001) raise the question whether the growth empirics from a larger sample are
applicable to specific sub-samples of countries. For example, the sample of developing countries may be
on a different regression surface than the sample of developed countries. To address the third concern, we
re-estimate our baseline model including only developing countries. These results are presented in Table
6.
(Insert Table 6 here)
In columns 1 to 5 we use CIM and Trade Share as our measures of institutions and openness
respectively while alternating between all the five democracy variables. These results are consistent with
the previous findings. In all five regressions, trade and institutions are statistically significant and have a
positive impact on growth. Dist Equator is also significant in four of the five regressions. None of the
four democracy measures is statistically significant, while Regime Instability is statistically significant
and has an adverse impact on growth. In column 6, Real Openness replaces Trade Share from column 5.
Regime Instability continues to be statistically significant. In columns 7-9 we use Malaria Ecology as our
measure of geography and alternate between P-IV Democracy, Polity Score and Regime Instability.
Again, only Regime Instability is statistically significant among these three measures (column 9).
Institutions and trade measures are significant in all three with average elasticities of 0.22 and 0.10,
respectively. Malaria Ecology is statistically significant only in column 7.
The model diagnostic tests perform well with the two Hansen tests indicating the validity of
exclusion restrictions and the K-P test rejecting the null of underidentification in all of the difference
models (though only in one levels model).
4.2.3: Five- year averages versus yearly data
A recent paper by Acemoglu et al. (2014) finds a positive and statistically significant relationship
between democracy and economic growth. There are several notable differences between their estimation
approach and the one presented here. First, they use annual data, which is unusual given that most papers
use either five or ten year averages to smoothen business cycle fluctuations. Second, it is well known that
any single measure of democracy is fraught with measurement error. Acemoglu et al. (2014) address this
problem by constructing a dichotomous democracy variable that essentially combines P-IV Democracy
with the FH Index. In contrast, we use an even broader democracy measure, Unified Democracy Score composed of ten democracy measures- that was constructed by Pemstein et al. (2010) specifically to
tackle the measurement error problem. Fourth, they address the potential endogeneity of democracy by
using external instrument: Regional waves of democratization and reversals. As Baazi and Clemens
(2013) point out, while the use of external instruments delivers stronger performance with regard to
underidentification / weak identification tests, external instruments often fail the exclusion restrictions
either empirically or on theoretical grounds. For example, Ndulu & O’Connell (1999) use waves of
democratization as an explanatory variable in studying economic growth in Africa and find them to be
statically significant. Finally, as noted by Acemoglu et al. (2014), their baseline specification, which only
includes the democracy index and the four lags of GDP is likely to suffer from omitted variable bias
problem. To show how crucial their results depend on the use of annual data, we compare estimation
results from annual and five year averaged data using our sample.
(Insert Table 7)
In Table 7, column 1, we estimate a model similar to their baseline specification with three
democracy measures. In all specifications we include four lags of GDP per capita and time and estimate
the model using one-step difference GMM estimator used by Acemoglu et al (2014). Similar to their
results, we find that all measures of democracy (P-IV Democracy, FH Index and Unified Democracy
Score) have a positive impact on economic growth and are statically. The perfect p value of 1 for the
Hansen exclusion tests suggest the problem of instrument proliferation, indicating the need to either
collapse or restrict the instrument set. This can also be seen from the fact that the number of instruments
far exceeds the number of panel units. In column 2, we restrict the instrument set for the lagged GDP
variables11. Now, only the FH Index is statistically significant though all three retain their positive impact
on economic growth.
11
Acemoglu et al. (2014) re-estimate their baseline model with alternative moment conditions in Table A6, Cols. 2-5 but the
number of instruments remain much higher than the number of countries included.
The above regressions assumed democracy to be exogenous. The next two columns (3 and 4)
report results from instrumenting democracy, using both, full and restricted instrument sets respectively.
Similar to column 2, only FH Index is statistically significant in column 3. Restricting the instrument set
(column 4) produces negative coefficient estimates for both PIV Democracy and Unified democracy
Score but only the first is statistically significant.
While we do not given much credence to the negative effect of democracy on economic growth,
given that the model is clearly mis-specified, the results from Table 7 suggest that the positive impact of
democracy on growth postulated in Acemoglu et al. (2014) is only short lived and can not be sustained
over the long term.
4.3 Inclusion of Macroeconomic Policy Variables
In this section, we combine important features of Dollar and Kraay’s (2003) seminal growth
model with our baseline setup. In particular, we include three policy variables, the black market premium,
the inflation rate and the share of government spending in total GDP, as additional explanatory variables
in our empirical model. These results are presented in Table 8.
(Insert Table 8 here)
Columns 1 and 2 present the basic extended model. P-IV Democracy is not statistically significant
in both cases. CIM has a statistically significant and positive impact on growth (its five year average
elasticity is 0.188 and 0.148, respectively). Trade Share in column 1 and Real Openness in column 2 also
have a positive though, a slightly smaller impact on growth than institutions (the coefficients are 0.084
and 0.068, respectively). Being further away from the equator has a positive impact on growth.
With respect to the policy variables, while all three have the same negative signs as in Dollar and
Kraay, only two are statistically significant. Government spending appears to have the strongest negative
impact (its coefficient is -0.168 and -0.111 in columns 1 and 2 respectively). Typically, government
consumption is counter-cyclical so we would expect government consumption to go up during periods of
weak GDP growth, implying a negative coefficient estimate of the variable. We control for this
simultaneity by instrumenting government consumption with previous periods lags and differences. Even
after this the elasticity of this variable is statistically significant and negative. The coefficients on
inflation rate is significant only in column 1 with its coefficient being -0.024. An interesting finding is
that the total positive impact of trade and institutions dominates the total negative impact on economic
growth of the three policy variables.
In the remainder of the table, we include the various measures of democracy in the extended
model. As before, we first introduce the three additional broad measures of democracy, Polity Score, FH
Index and Unified Democracy Index (columns 3 - 7). In columns 4 and 5, the three deep determinants
continue to be statistically significant with the expected signs (except for Real Openness in column 4)
while the Polity Score is negative, as in Table 3, and not significant. All policy variables are negative, as
before, with only Govt Spending being statistically significant in both columns while BMP is negative and
significant in 3 and inflation in 4. When we replace Polity Score with FH Index (columns 5 and 6), the
results for the deep determinants and the policy variables are very similar to the previous columns.
Interestingly, the FH Index coefficient estimate is positive though not statistically significant. Switching
to the Unified Democracy Score in columns 7 does not change much. Institutions, trade, geography, and
Govt Spending continue to have the expected sign and are statistically significant but BMP and Inflation
are not. The Unified Democracy Score has a negative coefficient and, like the previous two democracy
measures, is statistically insignificant.
While the broad democracy measures do not seem to have any statistically significant impact on
economic growth, results for the tangential measure of democracy that we consider, Regime Instability,
points to the importance of stable democracies for economic growth (columns 8 and 9). In column 10 we
include Regime Instability along with P-IV Democracy and with Polity Score in column 11. While CIM
and Dist Equator continue to be positive and significant, the openness measures are not significant. None
of the democracy measures are now significant.
For the diagnostic tests, with all p values in excess of 0.05 for the AR(2) test, we cannot reject the
null hypothesis of serially uncorrelated error terms. While, the two Hansen tests have p-values greater
than 0.05 indicating that we cannot reject the null hypothesis that the instruments are correctly excluded,
in quite a few instances the values are close to unity. Consequently, instrument proliferation, which can
result in underidentification is a possibility. The K-P tests of the first stage structural equations, however,
reject the null of underidentification for both, the levels and the difference equations. As in Tables 2 and
3, the low F stats for the K-P Weak id tests point to the possibility of weak instrument bias. However,
given the quality of our diagnostic results taken together, we believe that the estimated models in Table 4
are correctly identified and justify a causal interpretation of the various growth determinants.
5. Conclusions
Employing cross section and time series data for a sample of more than 160 countries over a
period of fifty years, we do not find a statistically significant impact of democracy on economic growth.
In contrast, variables that measure the quality of public institutions and the stability of the political regime
exert a statistically significant and positive impact on growth. These findings pass a number of robustness
checks as well as the inclusion of macroeconomic policy measures used in Dollar and Kraay (2003).
Addressing the issues of instrument proliferation and underidentification that often confound the
results from dynamic panel data studies (Bazzi and Clemens, 2013), we report a battery of tests that show
that our instruments are relevant and do not suffer from underidentification, while the ad hoc tests for
weak identification are mixed. Our results indicate that in addition to public institutions and regime
stability, trade and geography are all also statistically significant and robust determinants of economic
growth. The explanatory variables vary, however, in terms of their economic impact. The quality of
institutions, especially when measured as the extent of contract intensive money, exhibits the strongest
impact. Regime stability as well as the two trade measures are economically important as well, but to a
lesser extent than the institution measures. The economic impact of geography, in particular when
measured by relative distance from the equator, cannot be ignored but is weaker than measures of trade
and institutions. The policy variables first introduced by Dollar and Kraay (2003) are not only significant
but exert a sizeable economic impact.
Our results capture the tension between the various roles of the government in the economy. While
the choice of the political regime (democracy versus non-democracy) appears to matter little for economic
development, a governments’ focus on improving the strength of the public institutions as well as
maintaining political stability has a distinctly positive impact on economic growth. As other studies have
found, increased government consumption relative to the GDP and high levels of inflation impede
economic growth, with inflation having a weaker impact on growth than government consumption
expenditure. We should note that looking at government consumption as an aggregate, instead of
analyzing its different components could be misleading since it is well known that some government
expenditures, such as infrastructure investments, are growth enhancing.
Finally, it should be noted that the impact of democracy on growth may occur through indirect
channels rather than the direct effect estimated in this paper. For example, political institutions critical to
economic development may be more likely to exist and function effectively under democratic rule. Barro
(1996), Tavares and Wasziarg (2001), and others have investigated indirect effects of democracy and
identified a number of potential channels such as educational attainment, inequality, government
spending, and capital accumulation. Their analysis shows that some channels are positive while others are
negative. If the growth-enhancing and growth-reducing indirect effects of democracy more or less offset
each other, the aggregate (i.e., direct) effect of democracy on growth will be minimal, which is precisely
the finding of this paper.
Bibliography
Acemoglu, D. (2008). Oligarchic versus democratic societies. Journal of the European Economic
Association, 6, 1–44. doi:10.1162/JEEA.2008.6.1.1
Acemoglu, D., & Johnson, S. (2005). Unbundling Institutions. Journal of Political Economy.
doi:10.1086/432166
Acemoglu, D., Johnson, S., & Robinson, J. A. (2001). The Colonial Origins of Comparative
Development: An Empirical Investigation. American Economic Review. doi:10.1257/aer.91.5.1369
Acemoglu, D., Robinson, J. A., Naidu, S., & Restrepo, P. (2014). Democracy Does Cause Growth. MIT
Department of Economics Working Paper, 14-09. Retrieved from
http://www.nber.org/papers/w20004
Alesina, A., Ozler, S., Roubini, N., & Swagel, P. (1996). Political Instability and Economic Growth.
Journal of Economic Growth, 1, 189–211. doi:10.1007/BF00138862
Arellano, M., & Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence and
an Application to Employment Equations. The Review of Economic Studies, 58(2), 277–297.
Arellano, M., & Bover, O. (1995). Another look at the instrumental variable estimation of errorcomponents models. Journal of Econometrics. doi:10.1016/0304-4076(94)01642-D
Baier, S. L., & Bergstrand, J. H. (2007). Do free trade agreements actually increase members’
international trade? Journal of International Economics, 71(1), 72–95.
Barro, R. J. (1991). Economic Growth in a Cross-Section of Countries. The Quarterly Journal of
Economics, 106, 407–443.
Barro, R. J. (1996). Democracy and growth. Journal of Economic Growth. doi:10.1007/BF00163340
Barro, R. J., & Lee, J.-W. (1994). Sources of economic growth. Carnegie-Rochester Conference Series on
Public Policy. doi:10.1016/0167-2231(94)90002-7
Bates, R. H., Fayad, G., & Hoeffler, a. (2012). The state of democracy in Sub-Saharan Africa.
International Area Studies Review, 15, 323–338. doi:10.1177/2233865912462373
Bazzi, S., & Clemens, M. A. (2013). Blunt Instruments: Avoiding Common Pitfalls in Identifying the
Causes of Economic Growth. American Economic Journal: Macroeconomics, 5(2), 152–186.
doi:10.1257/mac.5.2.152
Beck, T., Clarke, G., Groff, A., Keefer, P., & Walsh, P. (2001). New Tools in Comparative Political
Economy: The Database of Political Institutions. The World Bank Economic Review, 15(1), 165–
176. doi:10.1093/wber/15.1.165
Beck, T., Demirguc-Kunt, a., & Levine, R. (2000). A New Database on the Structure and Development of
the Financial Sector. The World Bank Economic Review, 14(3), 597–605.
Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models.
Journal of Econometrics. doi:10.1016/S0304-4076(98)00009-8
Brock, W., & Durlauf, S. (2001). Growth Empirics and Reality. The World Bank Economic Review, 15,
229–272.
Brunetti, Y., & Weder, B. (1995). Political sources of growth: A critical note on measurement. Public
Choice, 82, 125–134. doi:10.1007/BF01047733
Burkhart, R. E., & Lewisbeck, M. S. (1994). Comparative Democracy - the Economic-Development
Thesis. American Political Science Review, 88, 903–910. doi:Doi 10.2307/2082715
Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics: methods and applications. Cambridge
University Press.
Caselli, F., Esquivel, G., & Lefort, F. (1996). Reopening the convergence debate: A new look at crosscountry growth empirics. Journal of Economic Growth. doi:10.1007/BF00141044
Clague, C., Keefer, P., Knack, S., & Olson, M. (1999). Contract-intensive money: Contract enforcement,
property rights, and economic performance. Journal of Economic Growth, 4(2), 185–211.
doi:10.1023/A:1009854405184
Dollar, D., & Kraay, A. (2003). Institutions, trade, and growth. Journal of Monetary Economics.
Durlauf, S. N., Kourtellos, A., & Tan, C. M. (2008). Are any growth theories robust? In Economic
Journal (Vol. 118, pp. 329–346). doi:10.1111/j.1468-0297.2007.02123.x
Easterly, W., & Levine, R. (2003). Tropics, germs, and crops: How endowments influence economic
development. Journal of Monetary Economics.
Epstein, D. L., Bates, R., Goldstone, J., Kristensen, I., & O'Halloran, S. (2006). Democratic
transitions. American Journal of Political Science. doi:10.1111/j.1540-5907.2006.00201.x
Frankel, J. A., & Romer, D. (1999). Does Trade Cause Growth? American Economic Review.
doi:10.1257/aer.89.3.379
Gallup, J. L., Sachs, J. D., & Mellinger, A. D. (1999). Geography and Economic Development.
International Regional Science Review. doi:10.1177/016001799761012334
Giavazzi, F., & Tabellini, G. (2005). Economic and political liberalizations. Journal of Monetary
Economics, 52, 1297–1330. doi:10.1016/j.jmoneco.2005.05.002
Hall, R. E., & Jones, C. I. (1999). Why Do Some Countries Produce So Much More Output Per Worker
Than Others? The Quarterly Journal of Economics, 114(1), 83–116. doi:10.1162/003355399555954
Helliwell, J. F. (1994). Empirical linkages between democracy and economic growth. British Journal of
Political Science, 24(02), 225–248.
Holtz-Eakin, D., Newey, W., & Rosen, H. S. (1988). Estimating Vector Autoregressions with Panel Data.
Econometrica, 56, 1371–1395. doi:10.2307/1913103
Huntington, S. P. (1968). Political Order in Changing Societies. New Haven (p. 1). New Haven.
Islam, N. (1995). Growth empirics: A panel data approach. The Quarterly Journal of Economics, 110,
1127–1170. doi:10.2307/2946651
Keefer, P., & Stasavage, D. (2003). The Limits of Delegation: Veto Players, Central Bank Independence,
and the Credibility of Monetary Policy. American Political Science Review, 97(03), 407–423.
Landau, D. (1986). Government and economic growth in the less developed countries: an empirical study
for 1960-1980. Economic Development and Cultural Change, 35–75.
Londregan, J. B., & Poole, K. T. (1996). Does High Income Promote Democracy? World Politics.
doi:10.1353/wp.1996.0024
Ndulu, B. J., & O’Connell, S. A. (1999). Governance and growth in sub-Saharan Africa. The Journal of
Economic Perspectives, 41–66.
North, D. C. (1993). Five propositions about institutional change. Economics Working Paper Archive at
WUSTL. Retrieved from
/citations?view_op=view_citation&continue=/scholar?hl=en&as_sdt=0,24&scilib=1&scioq=%E2%8
0%9CFive+Propositions+about+Institutional+Change,%E2%80%9D+&citilm=1&citation_for_view
=BbOFiFcAAAAJ:u5HHmVD_uO8C&hl=en&oi=p
North, D. C. (1994a). Institutional Change: A framework for Analysis (No. 9412001). St. Louis.
North, D. C. (1994b). Institutions matter. Economic History. Retrieved from
/citations?view_op=view_citation&continue=/scholar?hl=en&as_sdt=0,24&scilib=1&scioq=%E2%8
0%9CInstitutions+Matter%E2%80%9D++north&citilm=1&citation_for_view=BbOFiFcAAAAJ:ux6o8ySG0sC&hl=en&oi=p
North, D. C. (1994c). Transaction Cost Through Time (No. 9411006). St. Louis.
Pemstein, D., Meserve, S. A., & Melton, J. (2010). Democratic compromise: A latent variable analysis of
ten measures of regime type. Political Analysis, 18(4), 426–449.
Persson, T., & Tabellini, G. (2008). The Growth Effect of Democracy : Is it Heterogenous and How Can
it be Estimated ? In Institutions and Economic Performance (pp. 544–585). doi:10.3386/w13150
Przeworski, A., & Limongi, F. (1993). Political Regimes and Economic Growth. Journal of Economic
Perspectives. doi:10.1257/jep.7.3.51
Redding, S., & Venables, A. J. (2004). Economic geography and international inequality. Journal of
International Economics, 62, 53–82. doi:10.1016/j.jinteco.2003.07.001
Rock, M. T. (2009). Has Democracy Slowed Growth in Asia? World Development, 37, 941–952.
doi:10.1016/j.worlddev.2008.09.002
Rodrik, D., Subramanian, A., & Trebbi, F. (2004). Institutions rule: The primacy of institutions over
geography and integration in economic development. Journal of Economic Growth, 9(2), 131–165.
Rodrik, D., & Wacziarg, R. (2005). Do Democratic Transitions Produce Bad Economic Outcomes?
American Economic Review. doi:10.1257/000282805774670059
Roodman, D. (2009a). “How to do xtabond2: An introduction to difference and system GMM in Stata,.”
Stata Journal, 9(1), 86–136.
Roodman, D. (2009b). Practitioners’ corner: A note on the theme of too many instruments. Oxford
Bulletin of Economics and Statistics, 71(1), 135–158.
Sachs, J. D. (2003). Institutions Don’t Rule: Direct Effects of Geography on Per Capita Income. SSRN
eLibrary.
Sachs, J. D., Warner, A., Åslund, A., & Fischer, S. (1995). Economic reform and the process of global
integration. Brookings Papers on Economic Activity, 1995(1), 1–118. doi:10.2307/2534573
Scully, G. W. (1988). The Institutional Framework and Economic Development. Journal of Potitical
Economy, 96, 652–662.
Spolaore, E., & Wacziarg, R. (2013). How Deep Are the Roots of Economic Development? Journal of
Economic Literature, 51(2), 325–369. doi:10.1257/jel.51.2.325
Stock, J. H., & Yogo, M. (2005). Testing for weak instruments in linear IV regression. Identification and
Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, 1.
Swaleheen, M. (2011). Economic growth with endogenous corruption: An empirical study. Public
Choice, 146(1), 23–41.
Tavares, J., & Wacziarg, R. (2001). How democracy affects growth. European Economic Review, 45(8),
1341–1378.
Vieira, F., MacDonald, R., & Damasceno, A. (2012). The role of institutions in cross-section income and
panel data growth models: A deeper investigation on the weakness and proliferation of instruments.
Journal of Comparative Economics, 40(1), 127–140.
Windmeijer, F. (2005). A finite sample correction for the variance of linear efficient two-step GMM
estimators. Journal of Econometrics, 126(1), 25–51.
Yang, B. (2008). Does democracy lower growth volatility? A dynamic panel analysis. Journal of
Macroeconomics, 30, 562–574. doi:10.1016/j.jmacro.2007.02.005
Appendix A: Variable Definitions and Summary Statistics
Table A1: Variable Definitions and Data Sources
Name
GDP per capita
CIM
Veto Players
Definition and Source(s)
PPP converted gross domestic product per capita. From Penn World Tables v7.1 (Heston et al, 2012)
Contract Intensive Money: Defined as the ratio of non-currency to total money (M2). Author calculations from IFS-IMF
(2014) (ratio of the sum of lines 15, 24 and 25 to the sum of 14, 15, 24 and 25).
Number of Veto Players: This variable counts the number of veto players in a political system, adjusting for whether these
veto players are independent of each other, as determined by the level of electoral competitiveness in a system, their
respective party affiliations, and the electoral rules. Veto players are defined as the president, largest party in the
legislature, for a presidential system; and as the prime minister and the parties in the government coalition in a
parliamentary system. (Also see Keefer, 2002). From DPI2000 (Beck at al., 2001), where it is coded as CHECKS.
Constraints on Exec Constraints on Executive: This variable refers to the extent of institutionalized constraints on the decision-making powers
of chief executives, whether individuals or collectivities. (1) unlimited authority (2) intermediate category (3) Slight or
moderate limitation on executive authority (4) intermediate category (5) Substantial limitations of executive authority (6)
intermediate category (7) Executive parity or subordination. From Polity IV dataset (Jaggers and Marshal, 2000), where it
is coded as XCONST
P-IV Democracy
Institutionalized Democracy: Democracy is conceived as three essential, interdependent elements. One is the presence of
institutions and procedures through which citizens can express effective preferences about alternative policies and leaders.
Second is the existence of institutionalized constraints on the exercise of power by the executive. Third is the guarantee of
civil liberties to all citizens in their daily lives and in acts of political participation. The Democracy indicator is an
additive eleven-point scale (0-10). From Polity IV dataset (Marshall et al, 2002), where it is coded as DEMOC.
Polity Score
This democracy measure is derived from the POLITY2 variable in the Polity-IV dataset (Marshall et al 2002). It is
essentially the sum of a country’s democracy and autocracy score It ranges from -10 to 10. Following the previous
literature (Rodrik & Wacziarg, 2005; Rock, 2009) a country is classified as democratic if this score is positive. For the
five year averages, a country is coded as a democracy if it has positive POLITY2 score in three of the five years.
FH Democracy Index Average of Political Rights and Civil Liberty; both indicators from Freedom House (FH), (2004)
Unified Democracy
Score
Regime Instability
A cumulative democracy score constructed from ten underlying democracy variables. From Pemstein et al (2010)
Trade Share
Real Openness
Imports plus exports relative to GDP; From PWT Mark 7.1 (Heston et al., 2012)
Imports plus exports in exchange rate US$ relative to GDP in purchasing power parity US$. Author calculations from
Penn World Table , 7.1 (Heston et al, 2012), following Alcala and Ciccone, 2004))
Relative Distance from the equator: Calculated as distance from the equator, divided by 90. From Gallup et al. (1998) and
Hall and Jones (1999)
A measure of malaria incidence that combines temperature, mosquito abundance and vector specificity. The underlying
index is measured on a highly disaggregated sub-national level, and then is averaged for the entire country. Because ME
is built upon climatological and vector conditions on a country-by-country basis, it is exogenous to public health
interventions and economic conditions. From Sachs (2003)
Dist Equator
Malaria Ecology
BMP
Inflation Rate
Government
Spending / GDP
Regime Instability: Measure of government stability that captures the extent of turnover in any one year of a government’s
key decision makers. It is calculated by dividing the number of exits between year t and t+1 by the total number of veto
players in year t. The variables are therefore on a 0-1 scale, with zero representing no exits and one representing the exit
and replacement of all veto players. From DPI 2000 (Beck et al., 2001) where it is coded as STABS.
Black Market premium. From World Bank, Global Development Network Database
Annual percentage change in the consumer price index. From the World Development Indicators, World Bank.
General government final consumption expenditure (% of GDP). From the World Development Indicators, World Bank.
Table 1a: Summary Statistics
Variable
Mean
Ln GDP per capita
CIM
Veto Players
Constraints on the Executive
P-IV Democracy
Polity Score
FH Index
Unified Democracy Score
Regime Instability
Trade Share
Real Openness
Distance from the Equator
Malaria Ecology
BMP
Inflation rate
Share of Govt Spending in GDP
8.249
75.931
2.461
4.101
4.159
0.489
3.881
0.502
0.114
71.988
51.246
0.290
3.811
569.941
36.484
15.724
Standard Deviation
Observations
overall
between
within
1.309
1.258
0.353
1455
19.223
16.406
10.263
1277
1.588
1.303
0.933
1182
2.283
1.900
1.286
1315
4.083
3.478
2.145
1315
0.500
0.392
0.322
1350
1.999
1.803
0.880
1186
0.228
0.203
0.103
1462
0.126
0.069
0.107
1063
45.627
40.993
20.249
1456
84.684
61.697
67.156
1456
0.185
0.186
0.000
1630
6.655
6.675
0.000
1570
13489.750
5756.256
12500.730
925
254.461
109.217
233.724
1168
6.426
5.411
3.879
1327
Table 1b: Correlation coefficients of main time varying explanatory variablss
Constraints
Ln GDP
Veto
on the
P-IV
Polity
per capita CIM
Players
Executive Democracy Score
Ln GDP per
capita
CIM
Veto Players
Constraints on
the Executive
P-IV
Democracy
Polity Score
FH Index
Unified
Democracy
Score
Regime
Instability
Trade Share
Real Openness
FH Index
Unified
Democracy Regime
Score
Instability
1.00
0.42
0.37
1.00
0.27
1.00
0.46
0.32
0.70
1.00
0.51
0.36
-0.55
0.32
0.29
-0.36
0.71
0.68
-0.69
0.96
0.88
-0.88
1.00
0.91
-0.92
1.00
-0.80
1.00
0.54
0.31
0.73
0.90
0.93
0.81
-0.95
1.00
0.04
0.26
0.12
0.06
0.17
0.00
0.14
0.03
0.01
0.26
0.03
0.03
0.27
0.03
0.04
0.25
0.01
0.03
-0.21
-0.05
-0.03
0.22
0.03
0.05
1.00
-0.09
-0.06
Countries
163
156
163
151
151
153
160
163
163
163
163
163
157
158
157
160
Trade
Share
Real
Openness
1.00
0.28
1.00
Table 2: Basic Dynamic Panel Data Model
1
2
Full Instrument Set
3
4
0.984***
(0.016)
0.223***
(0.058)
-0.006
(0.004)
0.045**
(0.022)
0.993***
(0.016)
0.170***
(0.055)
-0.005
(0.004)
0.059***
(0.023)
0.969***
(0.017)
0.256***
(0.071)
-0.005
(0.004)
0.943***
(0.053)
0.576***
(0.182)
-0.009
(0.006)
0.050
(0.056)
Dep. Variable: Log GDP p.c.
Ln GDPpc (t-1)
Ln CIM
P-IV Democracy
Ln Real Open.
0.101***
(0.029)
0.206***
(0.068)
Ln Trade Share
Dist. Equator
0.166***
(0.061)
5
6
Collapsed Instrument Set
1,027
145
150
8
9
Restricted Instrument Set
0.999***
(0.020)
0.161**
(0.073)
-0.009**
(0.005)
0.057*
(0.030)
1,035
146
38
1,027
145
38
0.967***
(0.020)
0.263**
(0.109)
-0.004
(0.004)
0.089**
(0.042)
0.220**
(0.086)
0.135*
(0.074)
-0.003
(0.005)
1,035
146
150
1.012***
(0.023)
0.117
(0.073)
-0.008*
(0.004)
0.063*
(0.036)
-0.003
(0.003)
1,035
146
93
1,027
145
93
1,035
146
93
0.43
0.30
0.24
0.86
0.60
0.72
0.32
AR(3) pval
0.51
0.40
0.31
0.04
0.03
0.01
0.10
Hansen Test pval
0.97
0.77
0.99
0.08
0.06
0.14
0.14
Difference in Hansen test overid pval
No. of Instruments in difference model
116
116
116
32
32
32
59
Kleibergen-Paap underid difference
pval
0.00
0.00
0.00
0.00
0.00
0.00
0.00
No. of Instruments in levels model
41
41
41
14
14
14
41
Kleibergen-Paap underid levels pval
0.76
0.55
0.55
0.23
0.42
0.30
0.76
Kleibergen-Paap Weak id difference F
2.16
2.14
1.58
1.97
1.97
1.91
2.24
Kleibergen-Paap Weak id levels F
0.70
0.82
0.82
0.35
0.16
0.26
0.70
All models estimated using Blundell-Bond two step system GMM estimator with robust bias corrected standard errors (Windmeijer, 2005)
Hansen test: tests the exclusion restrictions in the systems GMM, p-values reported
Diff in Hansen test for level instruments: tests the validity of instruments in the levels equation
Kleibergen - Paap Tests: Test the null that the first stage regression is underidentified, reported separately for the difference and level models (p val of
Wald test reported).
AR(2) and AR(3): p-value of Arellano and Bond test for autocorrelation of order 2 and 3, respectively.
*/**/***: Significant at 10%, 5% and 1%, respectively
All regressions include time dummies.
0.25
0.14
0.16
0.28
0.04
0.52
59
59
0.00
41
0.55
2.24
0.82
0.00
41
0.55
1.81
0.82
Observations
Countries
No. of Instruments in systems GMM
1,035
146
150
0.920***
(0.042)
0.641***
(0.228)
-0.005
(0.008)
0.164**
(0.076)
0.275
(0.169)
0.226
(0.190)
-0.003
(0.002)
Malaria Ecology
0.957***
(0.049)
0.472***
(0.175)
-0.010
(0.006)
0.050
(0.056)
7
1,035
146
38
Table 3: Dynamic Panel Data Regressions- basic model with alternative democracy variables included (restricted instrument set)
Dep Var: Ln GDP pc
1
2
3
4
5
6
Ln GDPpc (t-1)
Ln CIM
Polity Score
0.997***
(0.020)
0.200**
(0.092)
-0.009
(0.027)
1.000***
(0.021)
0.236***
(0.090)
-0.011
(0.032)
FH Index
0.973***
(0.021)
0.225**
(0.093)
0.962***
(0.028)
0.334***
(0.074)
0.008
(0.011)
0.001
(0.011)
Unified Democracy Score
0.990***
(0.025)
0.218**
(0.105)
-0.122
(0.114)
0.974***
(0.027)
0.335***
(0.075)
7
8
9
10
0.940***
(0.035)
0.267***
(0.087)
0.963***
(0.029)
0.342***
(0.087)
0.986***
(0.019)
0.197***
(0.072)
0.986***
(0.017)
0.249***
(0.086)
0.005
(0.016)
0.005
(0.020)
-0.002
(0.006)
-0.113*
(0.062)
0.046
(0.030)
-0.001
(0.007)
-0.159**
(0.071)
-0.042
(0.117)
P-IV Democracy
Regime Instability
Ln Real Open.
0.003
(0.032)
Ln Trade Share
0.066*
(0.039)
Dist. Equator
0.100
(0.078)
0.020
(0.046)
0.073
(0.086)
Observations
Countries
No. of Inst. in systems GMM
AR(2) pval
AR(3) pval
Hansen test pval
Diff in Hansen test for level inst.
No. of Inst. in difference model
K-P underid for diff model pval
No. of Inst. in levels model
K-P underid levels model pval
K-P Weak id difference F
K-P Weak id levels F
Notes: Same as Table 2
1,042
146
87
0.0007
0.726
0.1316
0.2087
55
0.00
39
0.00
2.1283
1.8486
1,042
146
87
0.0009
0.7717
0.0438
0.1366
55
0.07
39
0.06
1.1769
1.252
-0.167***
(0.064)
0.123**
(0.056)
0.046
(0.037)
0.144*
(0.074)
0.118**
(0.052)
0.125
(0.084)
1,063
160
83
0.0183
0.5484
0.0903
0.2828
53
0.00
37
0.00
2.1333
1.599
1,063
160
83
0.018
0.483
0.0771
0.4813
53
0.00
37
0.03
1.4978
1.3916
0.140*
(0.083)
0.098**
(0.049)
0.109
(0.097)
1,155
161
87
0.0122
0.8883
0.0569
0.6014
55
0.00
39
0.12
2.0879
1.1536
1,155
161
87
0.0144
0.7891
0.0218
0.6447
55
0.00
39
0.07
1.5582
1.2512
-0.228***
(0.064)
0.175*
(0.100)
0.119**
(0.055)
0.106
(0.091)
0.145*
(0.077)
0.080*
(0.048)
0.130*
(0.067)
922
155
77
0.042
0.139
0.0549
0.0138
48
0.00
35
0.40
1.8342
0.8956
922
155
77
0.028
0.1518
0.1662
0.108
48
0.09
35
0.53
1.1462
0.8112
844
145
103
0.0011
0.5026
0.1261
0.381
60
0.00
49
0.00
1.4183
1.6385
844
145
103
0.0011
0.5973
0.0595
0.3109
60
0.37
49
0.36
0.874
0.8812
Table 4: Dynamic Panel Data Regressions- basic model with alternative public institutional variables included (restricted instrument set)
Dep Var: Ln GDP pc
1
2
3
4
5
6
7
8
9
Ln GDPpc (t-1)
Veto Players
0.988***
(0.019)
0.035**
(0.015)
0.999***
(0.016)
0.021*
(0.011)
0.980***
(0.022)
0.019
(0.012)
0.987***
(0.019)
0.024**
(0.010)
0.996***
(0.018)
0.015
(0.012)
0.985***
(0.021)
0.019*
(0.012)
0.996***
(0.024)
0.022*
(0.012)
Constraints on Exec
P-IV Democracy
-0.005
(0.006)
Polity Score
0.021
(0.033)
FH Index
Regime Instability
Ln Trade Share
0.159***
(0.051)
0.122***
(0.038)
0.198***
(0.046)
Dist. Equator
Observations
Countries
No. of Inst. in systems GMM
AR(2) pval
Hansen test pval
Diff Hansen test for level inst.
No. of Inst. in difference
K-P underid for diff model p
No. of Inst. in levels model
K-P underid levels model p
K-P Weak id difference F
K-P Weak id levels F
Notes: Same as Table 2
12
0.978***
(0.016)
0.020*
(0.012)
0.994***
(0.016)
0.988***
(0.018)
0.982***
(0.020)
0.000
(0.028)
0.003
(0.017)
0.018
(0.017)
0.007
(0.014)
0.020**
(0.009)
-0.009
(0.007)
-0.026
(0.055)
-0.011
(0.013)
-0.142**
(0.061)
0.196***
(0.049)
Ln Real Open.
11
0.994***
(0.017)
0.024
(0.034)
-0.018
(0.014)
10
-0.014
(0.018)
-0.130**
(0.066)
0.059
(0.093)
0.016
(0.097)
0.006
(0.099)
0.042
(0.091)
0.038
(0.024)
0.077
(0.086)
1,033
151
85
0.34
0.01
0.08
53
0.00
39
0.00
1.93
1.60
1,059
153
85
0.21
0.01
0.07
53
0.00
39
0.00
2.04
1.97
1,120
160
84
0.06
0.01
0.07
53
0.00
38
0.00
2.35
1.77
1,030
163
75
0.09
0.03
0.14
47
0.00
34
0.00
2.60
1.29
1,059
153
85
0.18
0.01
0.05
53
0.01
39
0.00
1.82
1.58
0.125***
(0.037)
0.116***
(0.035)
0.125***
(0.039)
-0.144**
(0.070)
0.155***
(0.044)
-0.021
(0.016)
-0.050
(0.063)
0.101**
(0.040)
0.072**
(0.035)
0.026
(0.098)
0.075**
(0.036)
0.023
(0.106)
0.093
(0.087)
0.063
(0.089)
0.042
(0.088)
0.032
(0.097)
0.125*
(0.072)
1,120
160
84
0.04
0.00
0.03
53
0.00
38
0.00
2.25
1.61
1,030
163
75
0.07
0.01
0.03
47
0.00
34
0.50
2.45
0.83
1,155
151
96
0.33
0.02
0.59
61
0.02
42
0.00
1.28
1.76
1,155
151
96
0.32
0.02
0.60
61
0.00
42
0.00
2.66
1.54
1,053
150
89
0.34
0.02
0.55
56
0.00
40
0.08
2.23
1.22
934
151
77
0.50
0.00
0.10
48
0.00
35
0.16
3.35
1.21
923
150
101
0.53
0.04
0.31
59
0.00
48
0.02
1.37
1.32
Table 5: Dynamic Panel Data Regressions- inclusion of alternative democracy variables and their interaction with lagged CIM (restricted instrument set)
Dep Var: Ln GDP pc
1
2
3
4
5
6
7
8
9
Ln GDPpc (t-1)
Dem* LnCIM (t-1)†
Polity Score
0.987***
(0.019)
0.154
(0.125)
-0.656
(0.527)
P-IV Democracy
0.995***
(0.016)
0.014
(0.015)
0.981***
(0.019)
0.029***
(0.011)
0.989***
(0.020)
0.998***
(0.358)
0.979***
(0.018)
0.023*
(0.014)
0.970***
(0.021)
0.026*
(0.014)
0.979***
(0.018)
0.747**
(0.344)
-0.121***
(0.043)
-0.117**
(0.057)
-0.442
(0.866)
Regime Instability
0.042**
(0.021)
0.045*
(0.023)
0.090***
(0.030)
-4.431***
(1.544)
0.092***
(0.025)
0.129**
(0.056)
0.102
(0.075)
0.035
(0.076)
-1.313
(1.036)
-3.348**
(1.484)
0.071***
(0.020)
Ln Trade Share
0.148**
(0.060)
0.964***
(0.027)
0.316
(0.239)
-0.100*
(0.060)
Unified Democracy Score
Dist. Equator
0.982***
(0.019)
0.265**
(0.118)
-1.128**
(0.501)
-0.059
(0.065)
FH Index
Ln Real Open.
0.995***
(0.020)
0.103
(0.198)
10
0.093
(0.074)
0.118***
(0.030)
0.166**
(0.075)
0.128***
(0.038)
0.182**
(0.077)
0.174***
(0.042)
0.152
(0.093)
Observations
1,016
1,010
1,036
903
1,126
1,016
1,010
1,036
Countries
146
146
160
155
161
146
146
160
No. of Inst. in systems GMM
90
90
82
72
90
90
90
82
AR(2) pval
0.00
0.00
0.01
0.13
0.01
0.00
0.00
0.01
AR(3) pval
0.93
0.37
0.38
0.36
0.84
0.90
0.22
0.43
Hansen test pval
0.10
0.13
0.06
0.18
0.09
0.09
0.08
0.02
Diff in Hansen test for level inst.
0.54
0.45
0.04
0.06
0.30
0.26
0.21
0.06
No. of Inst. in difference model
57
57
52
45
57
57
57
52
K-P underid for diff model pval
0.14
0.00
0.00
0.01
0.00
0.31
0.00
0.01
No. of Inst. in levels model
40
40
37
33
40
40
40
37
K-P underid levels model pval
0.00
0.00
0.11
0.39
0.08
0.77
0.05
0.16
K-P Weak id difference F
1.07
1.45
1.78
1.39
2.09
0.96
1.47
1.39
K-P Weak id levels F
1.52
1.68
1.16
0.90
1.22
0.68
1.30
1.10
Notes: Same as Table 2
†: Dem refers to the specific democracy variable used in the model. Interaction between the democracy measure and Ln CIM(t-1)
0.196***
(0.047)
0.106
(0.081)
0.167***
(0.037)
0.135
(0.092)
903
155
72
0.07
0.21
0.16
0.05
45
0.02
33
0.30
1.37
0.97
1,126
161
90
0.01
0.89
0.01
0.13
57
0.00
40
0.35
1.79
0.94
Table 6: Dynamic Panel Data Regressions- sampleof only developing countries, basic model with alternative democracy variables included (restricted instrument set)
Dep Var: Ln GDP pc
1
2
3
4
5
6
7
8
9
Ln GDPpc (t-1)
Ln CIM
P-IV Democracy
0.963***
(0.025)
0.269***
(0.092)
-0.006
(0.005)
Polity Score
0.971***
(0.022)
0.248**
(0.097)
0.956***
(0.032)
0.371***
(0.130)
0.958***
(0.035)
0.366**
(0.156)
0.978***
(0.024)
0.174**
(0.077)
0.943***
(0.032)
0.295***
(0.103)
0.001
(0.010)
0.033
(0.135)
Regime Instability
0.081**
(0.034)
0.082**
(0.035)
0.129***
(0.045)
0.122***
(0.041)
-0.228***
(0.086)
0.100**
(0.041)
Ln Real Open.
0.205**
(0.095)
0.199**
(0.090)
0.203**
(0.100)
0.170
(0.109)
0.241***
(0.091)
-0.163**
(0.068)
760
106
93
0.02
0.16
0.20
0.12
59
0.00
41
0.08
1.66
1.19
767
106
93
0.01
0.84
0.19
0.12
59
0.00
41
0.17
1.73
1.06
757
112
87
0.02
0.80
0.32
0.24
55
0.00
39
0.12
1.60
1.13
821
113
93
0.01
0.97
0.13
0.04
59
0.00
41
0.40
1.76
0.90
661
109
77
0.01
0.20
0.25
0.11
48
0.04
35
0.20
1.20
1.04
0.086**
(0.036)
0.087**
(0.035)
-0.218**
(0.090)
0.124***
(0.037)
-0.004*
(0.002)
-0.003
(0.002)
-0.002
(0.003)
760
106
93
0.02
0.10
0.26
0.14
59
0
41
0.3
1.66
0.96
767
106
93
0.01
0.99
0.16
0.06
59
0
41
0.21
1.73
1.03
648
107
77
0.01
0.20
0.23
0.08
48
0.02
35
0.28
1.29
0.97
0.050
(0.056)
0.218**
(0.100)
Malaria Eco.
Observations
Countries
No. of Inst. in systems GMM
AR(2) pval
AR(3) pval
Hansen test pval
Diff in Hansen test for level inst.
No. of Inst. in difference model
K-P underid for diff model pval
No. of Inst. in levels model
K-P underid levels model pval
K-P Weak id difference F
K-P Weak id levels F
Notes: Same as Table 2
0.971***
(0.028)
0.191**
(0.075)
-0.008
(0.005)
-0.011
(0.031)
Unified Democracy Score
Dist. Equator
0.986***
(0.047)
0.214**
(0.099)
0.003
(0.028)
FH Index
Ln Trade Share
0.953***
(0.030)
0.329***
(0.095)
661
109
77
0.01
0.23
0.16
0.03
48
0.00
35
0.90
1.99
0.56
Table 7: Yearly and Five year averaged Dynamic Panel Data Regressions with Democracy as the only additional explanatory variable
1
2
3
4
5
6
7
Yearly data#
5 yr average
Dep. Variable: Log
GDP p.c.
Full Inst Set
Dem Exog
Rest. Inst Set Full Inst Set
Dem Exog
Dem IV
Rest. Inst Set Full Inst Set
Dem IV
Dem Exog
Rest. Inst Set Full Inst Set
Dem Exog
Dem IV
8
Rest. Inst Set
Dem IV
Panel A: Polity IV Democarcy Measure
Ln GDPpc (t-1)
Ln GDPpc (t-2)
P-IV Democracy
Observations
Countries
No. of Instruments
AR(2) pval
Hansen Test pval
0.917***
(0.044)
-0.054*
(0.032)
0.002*
(0.001)
5,390
151
1216
0.77
1.00
0.420***
(0.122)
-0.013
(0.030)
0.001
(0.001)
5,390
151
136
0.30
0.02
0.954***
(0.036)
-0.060*
(0.032)
0.001
(0.001)
5,390
151
2430
0.76
1.00
0.649***
(0.093)
-0.040
(0.031)
-0.006*
(0.003)
5,390
151
270
0.24
1.00
0.666***
(0.067)
0.563***
(0.132)
0.574***
(0.139)
0.347
(0.244)
-0.005**
(0.002)
999
151
45
0.72
0.03
-0.006***
(0.002)
999
151
24
0.99
0.12
-0.025**
(0.011)
999
151
17
0.83
0.00
-0.028**
(0.014)
999
151
11
0.37
0.00
Panel B: Freedom house democracy index
Ln GDPpc (t-1)
Ln GDPpc (t-2)*
FH Index
Observations
Countries
No. of Instruments
AR(2) pval
Hansen Test pval
0.940***
(0.067)
-0.072**
(0.036)
-0.009**
(0.004)
5,317
167
1103
0.76
1.00
0.474***
(0.116)
-0.003
(0.029)
-0.006*
(0.003)
5,317
167
109
0.22
0.00
0.948***
(0.060)
-0.074**
(0.036)
-0.010**
(0.004)
5,317
167
1788
0.75
1.00
0.585***
(0.090)
-0.022
(0.029)
-0.000
(0.013)
5,317
167
211
0.31
0.92
0.609***
(0.070)
0.490***
(0.157)
0.339*
(0.185)
0.189
(0.283)
-0.001
(0.007)
1,033
167
43
0.23
0.00
0.001
(0.007)
1,033
167
22
0.39
0.13
0.086**
(0.035)
1,033
167
16
0.86
0.00
0.082*
(0.045)
1,033
167
10
0.50
0.15
Panel C: Unified Democracy Score
Ln GDPpc (t-1)
0.997***
0.533***
1.012***
0.689***
0.604***
0.521***
0.463***
0.334
(0.053)
(0.122)
(0.051)
(0.079)
(0.073)
(0.143)
(0.145)
(0.204)
Ln GDPpc (t-2)*
-0.072**
-0.027
-0.074**
-0.050*
(0.035)
(0.023)
(0.036)
(0.028)
Unified Dem. Score
0.037*
0.033
0.020
-0.104
-0.019
-0.061
-0.609**
-0.593**
(0.022)
(0.031)
(0.019)
(0.067)
(0.056)
(0.056)
(0.247)
(0.293)
Observations
6,031
6,031
6,031
6,031
1,150
1,150
1,150
1,150
Countries
169
169
169
169
169
169
169
169
No. of Instruments
1119
130
2236
258
45
24
17
11
AR(2) pval
0.27
0.46
0.26
0.10
0.19
0.32
0.71
0.78
Hansen Test pval
1.00
0.01
1.00
1.00
0.02
0.14
0.00
0.04
#: All regression with yearly data also include third and fourth lags of LnGDP. Only in one case were these significant and thus are not
reported to conserve space.
Dem Exog/Dem IV: Democracy measure treated as exogenous/ endogenous (instrumented with lagged values)
Notes: Same as Table 2
Table 8: Dynamic Panel Data Regressions- inclusion of policy variables (restricted instrument set)
Dep Var: Ln GDP pc
1
2
3
4
5
6
Ln GDPpc (t-1)
0.981***
0.992***
0.982***
0.998***
0.976***
0.984***
(0.021)
(0.024)
(0.025)
(0.019)
(0.021)
(0.023)
Ln CIM
0.188**
0.148*
0.204**
0.170**
0.228***
0.226**
(0.085)
(0.078)
(0.083)
(0.076)
(0.084)
(0.091)
P-IV Democracy
-0.002
-0.002
(0.004)
(0.003)
Polity Score
-0.011
-0.009
(0.024)
(0.024)
FH Index
0.002
0.002
(0.011)
(0.009)
Unified Democracy Score
7
0.984***
(0.025)
0.228***
(0.086)
0.084***
(0.031)
Ln Real Open.
Ln BMP
Ln Inflation rate
Ln (Govt Spending/GDP)
Dist. Equator
-0.013
(0.009)
-0.024**
(0.012)
-0.168***
(0.056)
0.290***
(0.092)
0.077***
(0.027)
0.068**
(0.033)
-0.017
(0.011)
-0.018
(0.013)
-0.111*
(0.067)
0.236**
(0.096)
-0.017
(0.011)
-0.021*
(0.012)
-0.165***
(0.058)
0.277***
(0.094)
0.088**
(0.036)
0.057
(0.037)
-0.017*
(0.010)
-0.018
(0.012)
-0.127**
(0.057)
0.219***
(0.081)
-0.014
(0.009)
-0.018
(0.011)
-0.139**
(0.071)
0.233**
(0.096)
0.080**
(0.032)
-0.018**
(0.007)
-0.012
(0.010)
-0.144***
(0.052)
0.187**
(0.089)
422
109
111
0.39
0.98
1.00
63
0.00
50
0.06
1.18
1.05
422
109
111
0.29
0.98
1.00
63
0.00
50
0.54
1.25
1.01
423
109
111
0.43
0.98
1.00
63
0.00
50
0.02
1.17
1.16
423
109
111
0.29
0.96
1.00
63
0.00
50
0.03
1.21
1.12
404
115
98
0.20
0.62
0.95
55
0.13
48
0.07
0.87
1.03
404
115
98
0.19
0.71
0.71
55
0.12
48
0.01
0.88
1.22
10
0.951***
(0.023)
0.344***
(0.127)
-0.002
(0.009)
11
0.968***
(0.022)
0.280**
(0.112)
-0.005
(0.018)
-0.018
(0.039)
-0.000
(0.015)
0.087***
(0.032)
-0.013
(0.010)
-0.019
(0.012)
-0.183***
(0.062)
0.252***
(0.087)
-0.115
(0.071)
0.082**
(0.032)
-0.178***
(0.066)
0.084***
(0.028)
-0.109
(0.075)
0.033
(0.038)
-0.048
(0.072)
0.038
(0.043)
-0.011
(0.009)
-0.024**
(0.011)
-0.135**
(0.056)
0.249***
(0.081)
-0.014
(0.010)
-0.017*
(0.010)
-0.061
(0.049)
-0.015
(0.014)
-0.021*
(0.012)
-0.099
(0.070)
0.311***
(0.115)
-0.015
(0.013)
-0.019
(0.016)
-0.099
(0.068)
0.266***
(0.102)
342
108
71
0.10
0.30
0.53
52
0.00
50
0.31
1.10
0.77
343
108
71
0.10
0.32
0.65
52
0.01
50
0.52
1.00
0.68
Malaria Ecology
Observations
Countries
Inst. in systems GMM
AR(2) pval
Hansen test pval
Diff in Hansen test
Inst in difference model
K-P underid diff. pval
No. of Inst in levels model
K-P underid levels pval
K-P Weak id difference F
K-P Weak id levels F
Notes: Same as Table 2
9
0.988***
(0.019)
0.191**
(0.076)
-0.033
(0.091)
Regime Instability
Ln Trade Share
8
0.973***
(0.017)
0.253***
(0.087)
443
116
111
0.30
0.93
0.99
63
0.01
50
0.03
1.06
1.13
362
115
80
0.09
0.65
0.94
44
0.01
40
0.31
1.18
0.81
-0.003
(0.002)
353
112
80
0.10
0.23
0.66
44
0.01
40
0.02
1.41
1.19