Does more democracy lead to greater economic freedom? New

European Journal of Political Economy
Vol. 19 (2003) 547 – 563
www.elsevier.com/locate/econbase
Does more democracy lead to greater
economic freedom? New evidence for
developing countries
Jakob de Haan a,b,*, Jan-Egbert Sturm c,d
a
Department of Economics, University of Groningen, PO Box 800, 9700 AV Groningen, The Netherlands
b
CESifo, Munich, Germany
c
Center for Economic Studies, University of Munich, Schackstrasse 4, 80539 Munich, Germany
d
Ifo Institute for Economic Research, Poschingerstrasse 5, 81679 Munich, Germany
Received 4 March 2002; received in revised form 5 May 2002; accepted 17 May 2002
Abstract
This paper examines the relationship between economic and political freedom, focusing on
developing countries. We conclude that increases in economic freedom between 1975 and 1990 are
to some extent caused by the level of political freedom. This result shows up for all measures of
political freedom that we employ. Our conclusion also holds for the sample without outliers. These
outlying observations are identified using so-called robust estimation techniques.
D 2003 Elsevier B.V. All rights reserved.
JEL classification: P51
Keywords: Economic freedom; Liberalization; Democracy; Political freedom
1. Introduction
The growth of liberal democracy, together with its companion, economic liberalism,
has been the most remarkable macro political phenomenon of the last four hundred
years.
(Fukuyama, 1992, p. 48)
* Corresponding author. Department of Economics, University of Groningen, PO Box 800, 9700 AV
Groningen, The Netherlands. Tel.: +31-50-3633706; fax: +31-50-3633720.
E-mail address: [email protected] (J. de Haan).
0176-2680/03/$ - see front matter D 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0176-2680(03)00013-2
548
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
Does more democracy lead to more economic freedom?1 Or is an autocratic regime in a
better position to introduce liberalization measures?2 Economic theory does not give a
clear answer to this question. As pointed out by Dethier et al. (1999), there exist arguments
for both views. According to some observers, only an authoritarian government is in a
position to introduce liberalization measures that initially may involve massive layoffs and
cuts in entitlements. An authoritarian regime may be needed at the beginning of
liberalization, because electorates often turn down economic reform even when it is
known in the end that they would benefit a majority of voters. Fernandez and Rodrik
(1991) show that uncertainty about the impact of reform at the level of the individual could
lead a rational electorate to vote against reform.3 In other words, policies that would be
popular ex post are often not implemented in a democratic regime. Supporters of this view
often refer to the experience of countries such as Chile, South Korea and Taiwan, which
only introduced democracy after economic reform was successfully implemented (see
Edwards, 1991). Because of the predominantly adverse effects of economic liberalization
in its early stages, the reform can be halted or reversed due to political backlash that it
generates. Postponement of democracy until the reform starts delivering its positive effects
may help bridge this period (Fidrmuc, 2000).
Arguments why democracy may lead to more economic freedom are generally similar
to the arguments as to why democracy may foster economic growth (see Przeworski and
Limongi, 1993; De Haan and Siermann, 1996, for surveys). First, only governments with
some legitimacy will be able to implement and sustain policies that may bear high shortterm costs. Second, many of the institutional characteristics of a democracy, like an
independent legal system, are also required for a successful liberalization. As North (1993)
puts it, ‘‘well specified and enforced property rights, a necessary condition for economic
growth, are only secure when political and civil rights are secure; otherwise arbitrary
confiscation is always a threat.’’ Third, democratization may limit rent seeking due to its
system of checks and balances. Recently, Rodrik (2000) has argued that democratic
institutions—political parties, elected representatives, free speech, and the like—can be
viewed as the ultimate institutions of conflict management, as they allow for differences
among social groups to be resolved in a predictable, inclusive, and participatory manner.
As liberalization may lead to distributional conflicts, this view implies that democracies
should be better able to liberalize their economies than nondemocracies.4
1
According to Gwartney et al. (1996), individuals have economic freedom when (a) property they acquire
without the use of force, fraud, or theft is protected from physical invasions by others, and (b) they are free to use,
exchange, or give their property to another as long as their actions do not violate the identical rights of others.
2
The answer to this question, although perhaps interesting in it own right, has become even more relevant as
a number of recent studies suggest that (an increase in) economic freedom stimulates economic growth. See, e.g.
Easton and Walker (1997), Wu and Davis (1999), Gwartney et al. (1999), De Haan and Sturm (2000), Sturm and
De Haan (2001a).
3
Similarly, Alesina and Drazen (1991) show that efficiency-enhancing reform can be delayed due to a war of
attrition and asymmetric distribution of payoffs.
4
Recently, La Porta et al. (2002) differentiated between institutions of English freedom, which guarantee the
independence of judges from political interference in the administration of justice from those of American freedom,
which allow judges to restrain law-making powers of the sovereign through constitutional review. They found that
English institutions of judicial independence are strong predictors of economic freedom (proxied by the security of
property rights and the number of steps that an entrepreneur must take in order to legally open a business).
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
549
Clague et al. (1996) have put forward a somewhat different view. They argue that the
quality of economic policies and institutions depends partly on the incentives and constraints
faced by policymakers. These incentives and constraints vary from one autocracy to another
and from one democracy to another. According to these authors they vary so much within
these two types of regime that empirical tests that merely distinguish between governments
as democratic or autocratic are bound to be misspecified.5 Clague et al. (1996) hypothesize
that in autocracies it is the time horizon of the individual autocrat that is the main determinant
of property and contract rights, whereas in democracies these rights depend upon whether
the democratic system is durable.
There is some scant empirical evidence, especially for countries in transition, on the role
of democracy in fostering liberalization. Using data for 26 transition countries, De Melo et
al. (1996) found a correlation of 0.8 between their index of economic liberalization and the
Freedom House index of political freedom. De Melo et al. (1997) use panel data to estimate
the determinants of economic liberalization in transition countries. Political freedom appears
with a positive and highly significant coefficient in their equation. Similarly, using panel data
from 25 post-communist countries of Central and Eastern Europe and the former Soviet
Union between 1992 and 1997, Dethier et al. (1999) conclude that democracy has facilitated
economic liberalization. These results suggest that political liberalization encourages
economic liberalization. Accordingly, the effect of political liberalization on economic
performance is indirect, via its effect on economic liberalization. In contrast, Fidrmuc
(2000), who focuses on performance instead of liberalization per se in the countries in
transition, reports that democracy has a U-shaped effect on economic performance, at least
during the early part of the transition (contraction). Thus, either no democracy or full
democracy leads to better economic performance than moderate democracy.
Clague et al. (1996) report support for their hypotheses. Autocrats who had been in
power for some time are associated with better property and contract rights than autocrats
who were in power for a short period. However, they also find that, in general,
democracies provide greater security of property and contractual rights than autocracies.
Still, these rights were often poor in democracies that had lasted only a short time.
Dawson (1998) found that for a sample of 92 (OECD and developing) countries the
level of economic freedom—measured in the same way as in the present paper—in 1990 is
significantly related to political and civil freedom at the beginning of the estimation period
(i.e. 1975). There are, however, serious questions about causality in Dawson’s model as he
also includes economic growth over the same period as explanatory variable.6 De Haan
and Sturm (2000) and Sturm and De Haan (2001a) have shown that the change in
economic freedom is robustly related to economic growth. Another drawback of Dawson’s
study is that his dependent variable is the end-of-period level of freedom. Furthermore, he
does not distinguish properly between developing and industrial countries. This distinction
should be made, as the level of political freedom hardly changed in the industrial
countries, in contrast to developing countries.
5
Likewise, Wintrobe (1998) distinguishes between two sorts of authoritarian regimes: totalitarians and
tinpots. Whereas totalitarians derive utility from power as such and try to maximize it, tinpots choose the level of
power that secures their remaining in office.
6
See Section 3 for a further discussion.
550
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
The purpose of this paper is to examine the relationship between economic and political
freedom, focusing on developing countries. We conclude that the level of political freedom
is positively related to increases in economic freedom between 1975 and 1995. This result
shows up for all measures of democracy that we employ. Our conclusion also holds for the
sample without outliers. These outlying observations are identified using so-called robust
estimation techniques.
The remainder of this paper is organised as follows. Section 2 presents the model and
the data. Section 3 presents the results. Section 4 explains our robustness analysis and
Section 5 contains our robust estimation results. Section 6 shows the outcomes of panel
regressions. The final section offers some concluding comments.
2. The model and the data
Our dependent variable is the (change in the) economic freedom indicator as
constructed by Gwartney et al. (1996) over the period 1975 –1995. An index of economic
freedom should measure the extent to which rightly acquired property is protected and
individuals are free to engage in voluntary transactions. In an economically free society,
the fundamental function of government is the protection of private property and the
enforcement of contracts. When a government fails to protect private property, takes
property itself without full compensation or establishes restrictions that limit voluntary
exchange, it violates the economic freedom of its citizens. Institutional arrangements that
restrain trade, increase transaction costs, weaken property rights, and create uncertainty
will reduce the realization of gains from trade and also the incentive of individuals to
engage in productive activities.
Gwartney et al. (1996) choose 17 measures and rate a high number of countries on each of
these measures on a scale of 0 –10, in which 0 means that a country is completely unfree and
10 means it is completely free. The measures are in four broad areas: money and inflation;
government operations and regulations; ‘‘takings’’ and discriminatory taxation; and international exchange. The components in the monetary area reflect the availability of sound
money to the citizenry. The components in the economic structure are indicators on reliance
on markets rather than the political process to allocate resources. In the takings area, the
index is designed to measure the degree to which governments treat individuals equally
rather than engage in transfer activities and impose taxes. Finally, the components in the
international area are designed to measure the presence of policies consistent with free trade
(Gwartney et al., 1999).
These 17 measures are combined in three ways in aggregated rankings. In the first
Index (Ie) each component is assigned a weight equal to the inverse of its standard
deviation, while in Index Is1 the importance of the components is based on a survey
under experts in the field of economic freedom. Finally, in Index Is2 the weighing is also
based on a survey, but this time the survey was held under a number of country experts.
As they are available for a large number of countries for quite a long period, we have
used the indicators of Gwartney et al. (1996) of alternative indicators of economic
freedom (see De Haan and Sturm, 2000 for a comparison of various indicators of
economic freedom).
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
551
We use various indicators of democracy. First, we employ Gastil’s ranking of countries
with respect to their political rights and civil liberty over 1973 –1975 (Pol75 and Civ75,
respectively). The survey’s regular publication by Freedom House provides useful and
consistent time series. Gastil has created two measures of liberty: civil liberty and political
rights. Both are ranked from 1 (the highest degree of liberty) to 7 (the lowest). The civil rights
rankings purport to measure the rights of the individual (e.g. independence of the judiciary,
freedom of the press, freedom of assembly and demonstration, freedom of political
organization, free trade unions, free religious institutions). The political rights rankings
are based on the degree to which individuals in a state have control over those who govern.
The measures are available from 1973 onwards.
As an alternative, we also employ Gasiorowski’s (1993) data set on political regime
change in which countries are characterized on an annual basis. A country is classified as a
democracy if there is a regime in which (i) meaningful and extensive competition exists
among individuals and organized groups for all effective positions of government power, at
regular intervals and excluding the use of force; (ii) no major (adult) social group is excluded
from these competitions; and (iii) a sufficient level of civil and political liberties exists to
ensure the integrity of political competition and participation. Based on this definition,
historical sources were used by Gasiorowski to classify non-OECD countries for their entire
post-independence or modern-state period. On the basis of this data set we have constructed
a variable (Gasdem) showing the relative number of years that a country falls in the category
democracy since 1961 (or since independence). So Argentina has a score of 0.31 for this
variable, as the country is categorized as a democracy during 10 years over the period 1961 –
1992.7
The dependent variable in our empirical analysis is the change in economic freedom
between 1975 and 1995 using the Ie index.8 We start in 1975 as the Gastil democracy
variables are only available from 1973 onwards. Apart from proxies for democracy, the
following explanatory variables are subsequently added to our cross-section model:
the level of economic freedom in 1975 (Free75)
(the log of) per capita income in 1975 ( Y75)
aid received (Aid7175: average aid as a percentage of GDP during 1971– 1975)
openness (Open: import and export as a share of GDP during 1970– 1990)
dummies for regions (East Asia) and dummies for colonial history (Spanish colony:
Spain)
economic growth in the period 1960– 1975 (Growth6075)
Barro-Lee’s indicator for political instability over the period 1960 – 1974.
The first two variables are included in most previous studies as explanatory variables.
They all refer to the beginning of our estimation period, so there is no problem of reverse
causality. Development aid given may be conditional on economic reform and therefore we
use aid received as control variable. Openness is taken up as more open countries are
7
The number of years of democracy (10) is divided by the total number of years for which observations are
available (32), which yields 0.31.
8
We have also used the Is1 and Is2 indices. This gave basically the same outcomes. Results are available on
request.
552
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
probably more confronted with competition from abroad. Economic growth in the previous
period is included as it is possible that countries that become more free grow faster which
may lead them to increase their economic freedom even more in the future (Gwartney et al.,
1999).9 Finally, political instability (measured by the frequency of coups and revolutions) is
taken into consideration as it seems likely that unstable regimes may have more difficulties
reforming the economy than stable ones. Appendix A reports details about sources of the
variables used.
3. Results
We start with a very simple graphical representation of the relationship between the
change in economic freedom and our proxies for democracy. Figs. 1 and 2 show the partial
leverage plots between initial democracy (Dem75), measured by either the civil liberty
index or the political rights index, and the change in economic freedom over 1975– 1995
(DFree) of the most simple model we consider, i.e. a model in which besides a constant
only one control variable—the initial level of economic freedom (Free75)—is included:10
DFree ¼ constant þ a Free75 þ b Dem75
These figures suggest that there is a clear relationship between the change in economic
freedom and the level of democracy. It is quite interesting that, especially countries with an
above-average level of democracy are characterized by a substantial increase in economic
freedom. In the group of countries with below-average levels of democracy we can
observe much more diversity: some countries increased their economic freedom, while
others reduced it. The largest outlier in Figs. 1 and 2 is Venezuela, which had a relatively
high level of democracy, but had a stark reduction in economic freedom.
To examine whether the positive association between democracy and the change in
economic freedom survives the addition of control variables, we turn to simple cross-section
OLS regressions. Table 1 presents the results using the civil liberty index (Civ75) as the proxy
for democracy. It follows that only the coefficient of some control variables are significantly
different from zero (notably aid received, the regional dummies and openness). The
coefficient of the level of economic freedom in 1975 (Free75), although always significantly
different from zero, is rather unstable.11 Interestingly, the coefficient of our proxy for
democracy is always significantly different from zero; it ranges between 0.28 and 0.44.
9
Gwartney et al. (1999) conclude that economic growth in the past had no effect on economic freedom.
Partial-regression leverage plot—also called added-variable plot—is a graphical device which shows the
relationship between the dependent variable and one explanatory variable after both have been corrected for the
information available in the other explanatory variables. To be more precise, the residuals that result from regressing
the dependent variable on all other explanatory variables and the residuals from regressing the explanatory variable
of interest on the remaining explanatory variables are shown in a scatter plot. As is well known, the regression
coefficient of the variable of interest in a multiple regression can also be determined from the bivariate linear
relationship between these two residual. The simple correlation between the two residuals equals the partial
correlation between the dependent and the explanatory variable of interest in the multiple regression.
11
Limited data availability of additional explanatory variables reduces the number of observations.
Restricting the observations to those available in the most comprehensive specification in all regressions does not
affect the conclusions.
10
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
553
Fig. 1. Partial leverage plot of the change in economic freedom and civil liberties.
Table 2 presents the same regressions using the average political rights index of Gastil
over 1973 – 1975 (Pol75) as the proxy for democracy. Again, the coefficient of the proxy
for democracy is always significantly different from zero. The results for the control
variables are very similar to those in Table 1.
As the democracy index has only an ordinal meaning (see De Haan and Siermann,
1996), interpreting the magnitude of the coefficient is not so easy. We therefore also explore
the relationship between economic freedom and democracy by introducing a dummy
Fig. 2. Partial leverage plot of the change in economic freedom and political rights.
554
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
Table 1
Change in economic freedom (1975 – 1995): cross-country estimates, Gastil civil liberty index
Civ75
Free75
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.37**
( 4.37)
0.59**
( 3.83)
0.32**
( 3.10)
0.63**
( 4.36)
0.27
(1.26)
0.44**
( 3.89)
0.76**
( 4.44)
0.20
(0.91)
0.02
(1.09)
0.42**
( 3.81)
0.78**
( 4.64)
0.17
(0.70)
0.01
(0.48)
0.01
(1.09)
0.41**
( 3.64)
0.84**
( 4.75)
0.27
(1.18)
0.03
(1.50)
0.00
(0.29)
1.55**
(3.76)
0.28**
( 3.09)
1.26**
( 8.29)
0.05
( 0.24)
0.03+
(1.91)
0.01*
(2.14)
2.38**
(5.40)
1.94**
(6.02)
0.28**
( 3.08)
1.26**
( 8.19)
0.01
( 0.06)
0.04+
(1.95)
0.01*
(2.22)
2.39**
(5.41)
1.91**
(5.77)
0.26
( 0.48)
0.32
68
0.42
57
0.41
57
0.49
57
0.63
57
0.62
57
0.30**
( 3.11)
1.25**
( 8.04)
0.03
( 0.15)
0.03+
(1.79)
0.01*
(2.15)
2.35**
(5.34)
1.84**
(5.17)
0.40
( 0.66)
0.44
(0.40)
0.61
55
Y75
Aid7175
Open
East Asia
Spain
Growth6075
Political Instability
R2 (adj.)
0.31
Number of observations 68
A constant is included in all regressions. White t-statistics in parentheses. * and ** denote significance at the 5%
and 1%, respectively.
Table 2
Change in economic freedom (1975 – 1995): cross-country estimates, Gastil political rights index
Pol75
Free75
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.30**
( 4.98)
0.55**
( 3.73)
0.27**
( 3.60)
0.58**
( 4.13)
0.21
(0.99)
0.32**
( 3.75)
0.72**
( 4.30)
0.16
(0.63)
0.01
(0.31)
0.30**
( 3.71)
0.74**
( 4.54)
0.12
(0.44)
0.01
( 0.25)
0.01
(1.17)
0.28**
( 3.40)
0.80**
( 4.60)
0.22
(0.88)
0.01
(0.69)
0.00
(0.45)
1.47**
(3.45)
0.24**
( 3.81)
1.27**
( 8.23)
0.16
( 0.82)
0.03+
(1.76)
0.01*
(2.50)
2.41**
(4.92)
2.18**
(6.44)
0.24**
( 3.80)
1.27**
( 8.23)
0.17
( 0.80)
0.03+
(1.76)
0.01*
(2.47)
2.40**
(4.92)
2.19**
(6.33)
0.08
(0.15)
0.31
68
0.38
57
0.38
57
0.45
57
0.64
57
0.63
57
0.24**
( 3.73)
1.27**
( 8.19)
0.19
( 0.87)
0.03
(1.57)
0.01*
(2.40)
2.39**
(4.90)
2.16**
(6.01)
0.04
(0.07)
0.39
(0.34)
0.62
55
Y75
Aid7175
Open
East Asia
Spain
Growth6075
Political Instability
R2 (adj.)
0.31
Number of observations 68
A constant is included in all regressions. White t-statistics in parentheses. * and ** denote significance at the 5%
and 1%, respectively.
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
555
Table 3
Change in economic freedom (1975 – 1995): cross-country estimates, freedom dummy
Freedum
Free75
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
1.04**
(3.80)
0.53**
( 3.64)
0.85**
(2.70)
0.58**
( 4.23)
0.34
(1.58)
1.14**
(3.45)
0.75**
( 4.76)
0.25
(0.95)
0.01
( 0.31)
1.09**
(3.22)
0.76**
( 4.86)
0.23
(0.84)
0.01
( 0.49)
0.00
(0.51)
1.07**
(3.31)
0.82**
( 4.92)
0.33
(1.26)
0.01
(0.49)
0.00
( 0.23)
1.59**
(3.90)
0.87**
(3.12)
1.29**
( 8.40)
0.06
( 0.30)
0.02
(1.48)
0.01+
(1.79)
2.51**
(5.27)
2.18**
(6.01)
0.87**
(3.10)
1.29**
( 8.34)
0.03
( 0.15)
0.03
(1.52)
0.01+
(1.87)
2.52**
(5.28)
2.16**
(5.86)
0.23
( 0.39)
0.28
68
0.36
57
0.35
57
0.43
57
0.62
57
0.61
57
0.86**
(3.03)
1.29**
( 8.31)
0.04
( 0.16)
0.02
(1.48)
0.01+
(1.86)
2.51**
(5.25)
2.15**
(5.56)
0.28
( 0.44)
0.09
(0.08)
0.60
55
Y75
Aid7175
Open
East Asia
Spain
Growth6075
Political Instability
R2 (adj.)
0.26
Number of observations 68
A constant is included in all regressions. White t-statistics in parentheses. * and ** denote significance at the 5%
and 1%, respectively.
Table 4
Change in economic freedom (1975 – 1995): cross-country estimates, Gasiorowski’s variable
Gasdem
Free75
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
1.43**
(3.45)
0.41**
( 2.85)
1.17*
(2.41)
0.47**
( 3.32)
0.27
(1.15)
1.54**
(3.18)
0.66**
( 3.87)
0.14
(0.51)
0.00
( 0.18)
1.47**
(3.07)
0.68**
( 4.07)
0.11
(0.40)
0.01
( 0.58)
0.01
(0.90)
1.41**
(3.12)
0.76**
( 4.38)
0.25
(0.94)
0.01
(0.39)
0.00
(0.57)
1.78**
(5.35)
1.10**
(3.00)
1.27**
( 8.02)
0.11
( 0.59)
0.02
(1.52)
0.02**
(2.87)
2.84**
(8.53)
2.26**
(6.39)
1.11**
(3.10)
1.27**
( 8.08)
0.13
( 0.63)
0.02
(1.50)
0.01**
(2.88)
2.84**
(8.75)
2.27**
(6.18)
0.12
(0.20)
0.29
65
0.36
55
0.36
55
0.44
55
0.64
55
0.64
55
1.14**
(2.90)
1.27**
( 7.89)
0.16
( 0.75)
0.02
(1.27)
0.02**
(2.79)
2.81**
(8.43)
2.23**
(5.69)
0.04
(0.07)
0.51
(0.37)
0.63
53
Y75
Aid7175
Open
East Asia
Spain
Growth6075
Political Instability
R2 (adj.)
0.28
Number of observations 65
A constant is included in all regressions. White t-statistics in parentheses. * and ** denote significance at the 5%
and 1%, respectively.
556
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
variable. The Freedom House divides all countries into three groups (free, partly free, and
not free). We focus on democratic countries and therefore construct our dummy Freedum to
equal 1 if the country is considered free according to the Freedom House definition and 0
otherwise. Note that as a consequence of this definition the expected sign of this proxy for
democracy is positive. Table 3 shows the results. Again, the coefficient of this proxy for
democracy is always significantly different from zero. In comparison to Table 1, the aid
variable (Aid7175) loses significance.
Finally, Table 4 presents the outcomes employing the Gasiorowski index as the proxy
for democracy (Gasdem). This proxy is very much in line with the argument of Clague et
al. (1996) that especially the duration of a democracy matters for economic policy. Note
that the expected sign of this variable is positive. Again, we find that (duration of)
democracy is positively associated with economic freedom. As in Table 3, the coefficient
of the aid variable is not significantly different from zero.
4. Robustness analysis
It is well known that one or a few outlying observations may dominate regression
results. Clague et al. (1996) find, for instance, that some of their results are driven by
just three observations. In this section we therefore discuss so-called robust estimators,
which can be used to examine the extent to which our results are sensitive to outlying
observations. Following Barnett and Lewis (1994, p. 316) we define an outlier as an
observation ‘lying outside’ the typical relationship between the dependent and
explanatory variables revealed by the remaining data. For instance, point A in Fig.
3(a) is clearly an outlier. Outliers in the dependent variable—i.e. in the y-direction—
often possess large positive or large negative residuals, which are easy to detect by
plotting the residuals.12 Observations may be outlying for several reasons. The most
obvious ones involve problems with the quality of the data, and nonlinearities in the
data that—by definition—cannot be captured by a linear regression model. Outliers in
the explanatory variables are even more likely as the number of explanatory variables
is usually greater than 1, and hence there are more opportunities for something to go
wrong. As Fig. 3(b) shows, an unusual observation in the x-direction (B) can actually
tilt the OLS regression line. In such a case we call the outlier a (bad) leverage point.
Note that looking at the OLS residuals cannot discover bad leverage points. If a
leverage point tilts the regression line, deleting the points with the largest OLS
residuals implies that some ‘good’ points would be deleted instead of the ‘bad’
leverage point.
Basically, there are two ways to deal with outliers: regressions diagnostics and
robust estimation. Diagnostics are certain statistics mostly computed from the OLS
regression estimates with the purpose of pinpointing outliers and leverage points.13
Often the unusual observations are then removed or corrected after which an OLS
12
Note, however, that if xi is near the centre of the set of explanatory observations, as is the case in Fig. 3(a),
it will mainly affect the constant and hardly alter the slope.
13
See, for instance, Belsey et al. (1980) and Chatterjee and Hadi (1988).
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
557
Fig. 3. Outlying observations and bad leverage points. The solid lines represent the OLS estimates including the
unusual observation(s). The dotted lines represent the OLS estimates without the unusual observations (A, B, or
C). The dashed line represents the OLS estimate without observations (C and D).
analysis on the remaining observations follows. When there is only one unusual
observation, some of these methods work quite well. However, single-case diagnostics
are well known to be inadequate in the presence of multiple outliers or leverage points
(Temple, 2000).
Take, for instance, Fig. 3(c). Deleting either of the two outliers will have little effect on
the regression outcome and will therefore not be spotted by the single-case diagnostics.
The potential effect of one outlying observation is clearly masked by the presence of the
other. Therefore we prefer so-called robust regression techniques that employ estimators
that are not strongly affected by (groups of) outliers.
Two closely related methods are the least median of squares (LMS) and least
trimmed squares (LTS) introduced by Rousseeuw (1984). LMS minimizes the median of
the squared residuals. LTS typically minimizes the sum of squares over half the
observations, the chosen half being the combination, which gives the smallest residual
sum of squares.14 According to Temple (2000), LTS is generally thought preferable to
LMS.15
Robust estimators are not only a good way to deal with outliers. As pointed out by
Temple (2000), the use of robust estimators is also useful for dealing with the related
(and—likewise—often neglected) problem of parameter heterogeneity. We can illustrate
14
Note that the choice of minimising the half the observations can be altered, and the proportion increased if
required (Temple, 2000).
15
Robust estimation techniques are increasingly being used in practice, e.g. in finance, chemistry, electrical
engineering, process control, and computer vision (Meer et al., 1991). For a survey of robust methods and some
applications, see Rousseeuw (1997).
558
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
this by the following example. Fig. 4 shows a data set for two variables (x and y) in
which 40% of the (50) observations follow a different distribution than the rest of the
observations. Assume that a researcher would not know this and would simply expect
and, hence, estimate a linear relationship between x and y. As the OLS model assumes
all observations are drawn from one single distribution, the OLS regression line
estimated by this researcher as shown in Fig. 4 will not reveal any valuable
information. In contrast, the robust regression line looks for a linear relationship,
which fits the majority of the data. As 60% of the data set follows a linear
relationship, LTS will reveal that relationship. The remaining 40% of the observations
will have large negative residuals, which are easily depicted by graphing the standardised residuals. As this example shows, that does not mean that those observations
should be ignored and simply thrown away. Those observations reveal that the linear
model is not adequate for the entire data set as not all observations follow the same
distribution. So robust estimation may yield indications for differences in the
parameter estimates for certain groups of observations. It may make sense in such
a situation to split the sample and run regressions for sub-samples, provided that the
various groups contain enough observations.
As shown by Rousseeuw (1984), robust estimators have an abnormally slow
convergence rate and, hence, perform poorly from the point of view of asymptotic
efficiency. Because of its low finite-sample efficiency, LTS is not suited for inference.
As proposed by Rousseeuw (1984), this can be resolved by using reweighted least
squares (RLS). A simple, but effective, way is to put weight zero if the observation is
an outlier and weight one otherwise. The resulting estimator is more efficient and
yields all the usual inferential output such as t-statistics and R2.
Fig. 4. Hypothetical example. Hypothetical data set for series x and y of 50 observations. The first 30 observations
have the following characteristics: series x is uniformly distributed between 0.5 and 4.5. Series y is distributed
around the line y = 2 + x + e, where e is normally distributed with mean zero and standard deviation 0.2. The
remaining 20 observations follow a normal distribution: x f N(6, 0.5) and y f N(2, 0.5). The solid line represents
the LTS estimate, whereas the dotted line represents the OLS estimate.
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
559
5. Results of robustness analysis
In this section we report the main results if we redo all the regressions shown in
Section 3 except that we now exclude the outliers identified following the methodology as outlined in the previous section. To save space, we only report the outcomes
for the basic model and the model with all control variables included (all other results
are available on request). We would like to stress that we do not think that the
information content of the outliers is simply thrown away. The results in this section
should be regarded as some diagnostic test to see whether the results as reported in
Section 3 are mainly driven by some outlying observations. As follows from Table 5
this is clearly not the case. Quite to the contrary even. The coefficients of our proxies
for democracy generally have a higher level of significance. This is quite a remarkable
outcome as much of our previous research using robust estimation techniques has
taught us that the results of many empirical political-economy models are driven by
outliers (see e.g. Sturm and De Haan, 2001b).
Table 5
Change in economic freedom (1975 – 1995): cross-country estimates, robust estimations
(1)
Civ75
(2)
Pol75
(3)
(4)
Freedum Gasdem
(5)
Civ75
(6)
Pol75
0.40** 0.34** 1.19**
1.64**
0.44** 0.22**
( 5.52) ( 6.59) (5.77)
(6.40)
( 5.89) ( 4.61)
Free75
0.40** 0.31** 0.21+ 0.21* 1.24** 0.93**
( 3.10) ( 2.93) ( 1.80) ( 2.23) ( 12.55) ( 8.97)
0.09
0.32*
Y75
(0.51)
(2.06)
Aid7175
0.05**
0.03+
(1.87)
(3.18)
Open
0.01**
0.00
(2.82)
(1.59)
East Asia
2.76**
1.80**
(14.27)
(4.17)
Spain
1.74**
1.42**
(6.27)
(4.33)
Growth6075
0.92*
0.48
( 2.33) ( 0.99)
Political Instability
0.50
0.68
(0.75)
(0.81)
R2 (adj.)
0.31
0.35
0.24
0.31
0.77
0.69
Number of
65
64
62
61
50
50
observations
Outliers
IRN,
CHL,
ARG,
CHL,
NIC,
DZA,
NIC,
IRN,
CHL,
IRN,
FJI,
IRN,
NIC,
VEN
NIC,
HND,
NIC,
PAN,
VEN
IRN,
VEN
URY,
VEN,
NIC,
VEN
SYR
VEN
Democracy
* and ** denote significance at the 5% and 1%, respectively.
(7)
Freedum
(8)
Gasdem
0.89**
(4.36)
1.00**
( 9.92)
0.35*
(2.10)
0.03+
(1.89)
0.01
(1.52)
2.09**
(4.65)
1.71**
(5.01)
1.07*
( 2.00)
0.26
( 0.27)
0.63
52
1.17**
(3.25)
1.38**
( 8.91)
0.51*
( 2.38)
0.01
( 0.17)
0.03**
(3.96)
2.82**
(9.05)
2.31**
(7.05)
0.13
( 0.21)
2.17
(1.45)
0.68
50
IRN,
NIC,
VEN
MEX,
TGO,
URY
560
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
The number of outliers identified by LTS is generally around three. We have examined
which countries are outlying. They are shown in the bottom row of Table 5. Nicaragua,
Iran, Venezuela, Honduras, and Algeria experienced the largest decline in economic
freedom. Similarly, Syria and Panama have places 8 and 10 on the top-10 list of the
countries, which had a decline in their level of economic freedom. In contrast, Argentina
and Chile have experienced the largest increase in economic freedom. This suggests that
our linear model has difficulty explaining large changes in economic freedom. Uruguay,
Togo, Mexico and Fiji do not show extreme changes in their level of economic freedom.
These countries are outliers because something is going on there with respect to the other
explanatory variables. Partial leverage plots might reveal which explanatory variable
causes a country being an outlier. In column 8 of Table 5, Mexico seems to be an outlier
because its relationship between political instability and the change of economic freedom
does not follow the main pattern. Togo is probably an outlier due to the openness variable.
In column 5, Fiji is outlying because of the Aid variable.
6. Panel estimates
As a further check on the robustness of our results, we have also estimated panel
versions of the regressions reported in Tables 1 – 4. For this purpose, we have split
Table 6
Change in economic freedom (1975 – 1995): panel estimates
Democracy
Free75/85
(1)
Civ75
(2)
Pol75
(3)
Freedum
(4)
Gasdem
(5)
Civ75
(6)
Pol75
(7)
Freedum
(8)
Gasdem
0.22**
( 4.20)
0.33**
( 4.29)
0.17**
( 4.18)
0.32**
( 4.28)
0.61**
(3.55)
0.30**
( 4.03)
0.78**
(3.56)
0.24**
( 3.42)
0.18
145
0.17
145
0.16
134
0.24**
( 3.95)
0.54**
( 5.79)
0.10
( 0.25)
0.02
(0.86)
0.00+
(1.87)
1.21**
(4.18)
0.53+
(1.67)
0.29
( 0.19)
0.88*
(2.00)
0.30
122
0.16**
( 3.42)
0.56**
( 5.88)
0.13
( 0.30)
0.02
(0.82)
0.00+
(1.87)
1.26**
(4.43)
0.70*
(2.20)
0.48
( 0.31)
0.70+
(1.67)
0.28
122
0.66**
(3.31)
0.56**
( 5.86)
0.01
(0.02)
0.01
(0.52)
0.00
(1.42)
1.38**
(4.67)
0.66*
(2.03)
0.22
( 0.14)
0.82+
(1.96)
0.29
122
0.92**
(3.90)
0.55**
( 6.07)
0.19
( 0.44)
0.01
(0.46)
0.00*
(2.30)
1.51**
(4.44)
0.75*
(2.40)
0.47
( 0.27)
0.80+
(1.81)
0.30
118
Y75/85
Aid7175/8185
Open74/84
East Asia
Spain
Growth7074/8084
Political
Instability7074/8084
R2 (adj.)
0.19
Number of
145
observations
Time dummies are included in all regressions. White t-statistics in parentheses. +, * and ** denote significance at
the 10%, 5% and 1%, respectively.
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
561
our sample period into two periods of equal length. We use exactly the same models
as before, except that a dummy for both periods is included as well. We test whether
the change in economic freedom measured over 10-year periods is related to our
proxies for democracy, using the same set of control variables as before (see
Appendix A for further details). In terms of qualitative conclusions, the results for
the panel regressions are the same as for the cross-section model. Again, we only
report the results for the basic model and the model with all control variables
included (all other results are available on request). The quantitative effect of
democracy is somewhat lower in the panel model. Take, for instance, column 1 in
Table 6, which is the equivalent of column 1 in Table 1. The coefficient of the civil
liberty index is somewhat lower than in Table 1, but is again significantly different
from zero. Basically the same results are found for the other specifications. With
respect to the significance of the control variables, the main difference in comparison
to the cross-country regressions is that the coefficient of the income variable becomes
significant in some regressions.
7. Concluding comments
It is widely believed that market liberalization should be a crucial element in any
structural reform program in developing countries. Having said that, this paper has
analysed to what extent liberalization is furthered by the democratic nature of a
country. Although it is sometimes suggested that democracy may hamper liberalization, our results clearly suggest otherwise. Focusing on the relationship between the
change in the economic freedom indicators of Gwartney et al. (1996) and various
indicators for democracy, we find that in our sample of developing countries
increases in economic freedom between 1975 and 1990 are to some extent caused
by the level of political freedom. This result shows up for all measures of democracy
that we employ, both in our cross section and panel estimates, even if we control for
various other potential determinants of liberalization. Our conclusion also holds for
the sample without outliers. These outlying observations are identified using so-called
robust estimation techniques. The latter result is reassuring as the outcomes of many
relationships in econometric models are often found to be very sensitive to the
inclusion of just a few observations. Our findings are in line with those of Dethier
et al. (1999) for transition countries and Pitlik and Wirth (2003) for developing
countries.
Acknowledgements
We would like to thank the participants of the SOM workshop on Economic Freedom,
Groningen, December 2001, and two referees for their comments on a previous version
of this paper.
562
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
Appendix A . Data
Variable
Description
Source
Free75
Fraser Institute
Y1975
Level of economic freedom
in 1975 (equal impact indicator)
Civil liberties indicator in 1973 – 1975
Political rights indicator in 1973 – 1975
Dummy indicating whether country is
free based on civil liberties and
political rights
Gasiorowski’s indicator for duration
of democracy
Log of GDP per capita in 1975
Growth6075
Growth in real GDP over 1960 – 1975
Aid7175
Average aid as percentage of GDP
during 1971 – 1975
Share of import and export in GDP
during 1970 – 1990
Dummy for East Asian countries
Dummy for former Spanish colonies
Assassinations and revolutions
1960 – 1974
Civ75
Pol75
Freedum
Gasdem
Openness
East Asia
Spain
Political Instability
Freedomhouse
Freedomhouse
Freedomhouse
Gasiorowski
Penn World Tables
(Mark 5.6a)
Penn World Tables
(Mark 5.6a)
IMF
Penn World Tables
(Mark 5.6a)
Barro-Lee data set
In the panel estimates the data refer to the following years or periods:
Period
1
2
DFree
Free
Civ/Pol
Y
Aid
Openness
Growth
Political instability
1985 – 1975
1975
1975
1974
avg (1971 – 1975)
1974
1970 – 1974
1970 – 1974
1995 – 1985
1985
1985
1984
avg (1981 – 1985)
1984
1980 – 1984
1980 – 1984
References
Alesina, A., Drazen, A., 1991. Why are stabilizations delayed? American Economic Review 81, 1170 – 1188.
Barnett, V., Lewis, T., 1994. Outliers in Statistical Data, 3rd ed. Wiley, New York.
Belsey, D.A., Kuh, E., Welsch, R.E., 1980. Regression Diagnostics. Wiley, New York.
Chatterjee, S., Hadi, A.S., 1988. Sensitivity Analysis in Linear Regression. Wiley, New York.
Clague, C., Keefer, P., Knack, S., Olson, M., 1996. Property and contract rights in autocracies and democracies.
Journal of Economic Growth 1, 243 – 276.
J. de Haan, J.-E. Sturm / European Journal of Political Economy 19 (2003) 547–563
563
Dawson, J.W., 1998. Institutions, investment, and growth: new cross-country and panel data evidence. Economic
Inquiry 36, 603 – 619.
De Haan, J., Siermann, C.L.J., 1996. New evidence on the relationship between democracy and economic
growth. Public Choice 86, 175 – 198.
De Haan, J., Sturm, J.-E., 2000. On the relationship between economic freedom and economic growth. European
Journal of Political Economy 16, 215 – 241.
De Melo, M., Denier, C., Gelb, A., 1996. From plan to market: patterns of transition. World Bank Economic
Review 10, 397 – 424.
De Melo, M., Denier, C., Gelb, A., Tenet, S., 1997. Circumstances and choice: the role of initial conditions and
policies in transition economies. Policy Research Working Paper 1866. The World Bank, Washington, DC.
Dethier, J.-J., Ghanem, H., Zoli, E., 1999. Does democracy facilitate economic transition? An empirical study of
Central and Eastern Europe and the former Soviet Union. Journal for Institutional Innovation, Development
and Transition 3, 15 – 30.
Easton, S.T., Walker, M.A., 1997. Income, growth, and economic freedom. American Economic Review 87,
328 – 332.
Edwards, S., 1991. Stabilization and liberalization policies in Central and Eastern Europe: lessons from Latin
America. NBER Working Paper 3816. National Bureau of Economic Research, Cambridge, MA.
Fernandez, R., Rodrik, D., 1991. Resistance to reform: status quo bias in the presence of individual-specific
uncertainty. American Economic Review 81, 1146 – 1155.
Fidrmuc, J., 2000. Liberalization, democracy and economic performance during transition. ZEI Working Paper
B52000. ZEI, Bonn.
Fukuyama, F., 1992. The End of History and The Last Man. Penguin, London.
Gasiorowski, M.J., 1993. The Political Regime Change Data Set. Louisiana State University, Baton Rouge.
Gwartney, J., Lawson, R., Block, W., 1996. Economic Freedom in the World, 1975 – 1995. Fraser Institute,
Vancouver.
Gwartney, J., Lawson, R., Holcombe, R.G., 1999. Economic freedom and the environment for economic growth.
Journal of Institutional and Theoretical Economics 155, 643 – 663.
La Porta, R., Lopes-de-Silanes, F., Pop-Eleches, C., Shleifer, A., 2002. The guarantees of freedom. NBER
Working Paper 8759. National Bureau of Economic Research, Cambridge, MA.
Meer, P., Mintz, D., Rosenfeld, A., Kim, D., 1991. Robust regression methods in computer vision: a review.
International Journal of Computer Vision 6, 59 – 70.
North, D.C., 1993. The Paradox of the West. Economics Working Paper Archive. Washington University-St.
Louis, Missouri.
Pitlik, H., Wirth, S., 2003. Do crises promote the extent of economic liberalization? An empirical test. European
Journal of Political Economy 19 (3), this issue.
Przeworski, A., Limongi, F., 1993. Political regimes and economic growth. Journal of Economic Perspectives 7,
51 – 69.
Rodrik, D., 2000. Where did all the growth go to? External shocks, social conflict, and growth collapses. Journal
of Economic Growth 4, 385 – 412.
Rousseeuw, P.J., 1984. Least median of squares regression. Journal of the American Statistical Association 79,
871 – 880.
Rousseeuw, P.J., 1997. Introduction to positive-breakdown methods. In: Maddala, G.S., Rao, C.R. (Eds.), Handbook of Statistics. Elsevier, Amsterdam, pp. 101 – 121.
Sturm, J.-E., De Haan, J., 2001a. How robust is the relationship between economic freedom and economic
growth really? Applied Economics 33, 839 – 844.
Sturm, J.-E., De Haan, J., 2001b. Central bank independence and inflation in developing countries: the role of
high inflation observations. Unpublished paper, Department of Economics, University of Groningen.
Temple, J., 2000. Growth regressions and what the textbooks don’t tell you. Bulletin of Economic Research 52,
181 – 205.
Wintrobe, R., 1998. The Political Economy of Dictatorship. Cambridge Univ. Press, Cambridge.
Wu, W., Davis, O.A., 1999. The two freedoms, economic growth and development: an empirical study. Public
Choice 100, 39 – 64.