University of Groningen
Corruption and governance around the world
Seldadyo, H.
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to
cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2008
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Seldadyo, H. (2008). Corruption and governance around the world: An empirical investigation Enschede:
PrintPartners Ipskamp B.V., Enschede, The Netherlands
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the
author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the
number of authors shown on this cover page is limited to 10 maximum.
Download date: 18-06-2017
CORRUPTION AND GOVERNANCE
AROUND THE WORLD
An Empirical Investigation
Harry Seldadyo Gunardi
Publisher:
PPI Publishers, Enschede, The Netherlands
Printed by:
PrintPartners Ipskamp, Nederland
ISBN: book: 978–90–367–3299–4; ebook: 978–90–367–3300–7
c 2008 Harry Seldadyo Gunardi
Alle rechten voorbehouden. Niets uit deze uitgave mag worden verveelvoudigd, opgeslagen in een geautomatiseerd gegevensbestand, of openbaar gemaakt, in enige vorm of op
enige wijze, hetzij elektronisch, mechanisch, door fotkopieën, opnemen of enige andere
manier, zonder voorafgaande schriftelijke toestemming van de auteur.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system of any nature, or transmitted in any form or by any means, electronic, mechanical,
now known or hereafter invented, including photocopying or recording, without prior
written permission of the author.
RIJKSUNIVERSITEIT GRONINGEN
CORRUPTION AND GOVERNANCE
AROUND THE WORLD
An Empirical Investigation
Proefschrift
ter verkrijging van het doctoraat in de
Economie en Bedrijfskunde
aan de Rijksuniversiteit Groningen
op gezag van de
Rector Magnificus, dr. F. Zwarts,
in het openbaar te verdedigen op
donderdag 20 maart 2008
om 16.15 uur
door
Harry Seldadyo Gunardi
geboren op 8 januari 1966
te Jakarta, Indonesia
Promotor:
Prof. dr. J. de Haan
Beoordelingscommissie:
Prof. dr. N. Hermes
Prof. dr. M. Paldam
Prof. dr. T. J. Wansbeek
Acknowledgements
Ladies and Gentlemen. . . Now we have a winner. In the world corruption
league, the grand prize goes to. . . Mr. Soeharto, the former president of Indonesia! Let’s have a loud cheer for the most corrupt president in the globe!
(see, Tranparency International: The Global Corruption Report 2004 ).
That is my personal reason why I am now in the research area of corruption. But it would have never happened without Jakob de Haan. On
the very first day I formally became a part of the University of Groningen,
he asked me if I were interested in this issue. Since then till the day I completed this thesis, my days and nights were only about corruption. And, as
will be shown in this dissertation, corruption is a serious problem.
This thesis is the result of collaboration with many people. First of all,
I have to mention again Jakob de Haan, a person with many roles: supervisor, discussion partner, and co-author; without him I could not have
completed my PhD project. I also enjoyed working with Emmanuel Pandu
and Paul Elhorst. Two chapters in this dissertation were written with them.
Like Paul Elhorst, Maarten Allers inspired me to investigate spatial issues in
governance. Tom Wansbeek’s comments have very much improved the chapter on corruption persistence. Jan-Egbert Sturm and Richard Jong-A-Pin
introduced the RATS program enabling me to run millions of regressions.
Paramita Muljono of the UEA-UK kindly shared her data on the Global
Competitiveness Report. Ward Romp gave me his beautiful LATEXprogram.
Also, I am indebted to the reading committee—Niels Hermes (University of
Groningen), Tom Wansbeek (University of Amsterdam), and Martin Paldam (University of Aarhus, Denmark)—for their willingness to read the
final manuscript and for their valuable comments. Certainly, I thank the
ii
two paranymphs—Poppy and Jeroen—who have helped me a lot in arranging many things before and during the defence.
Some other names have to be mentioned. Anneke Toxopeus and the
other nice people at the ISD-RUG were always helpful whenever I had questions or came to their office with the blue envelope from the Belastingdienst.
I thank the management of the Ubbo Emmius Scholarship program who
made me possible to do my PhD project at this University. Astrid, Dity,
Ellen, Rina, and Martin from SOM who have helped me a lot.
Groningen, even the Netherlands, is geographically not a big place. Yet,
there are many friends who have supported my study, directly and indirectly: those in PPI, PD, and GBI. Also, the relatives in the Netherlands
have played their own roles. It is impossible to list all of them, but their
names are written in my heart.
I am thinking of my parents and parents in law in Indonesia. From a
distance, I believe, they pray for me. Finally, a very special thanks must
be given to Mia for her patience and support in days and nights, sadness
and happiness, summer and winter. She is a blessing from God. I cannot
imagine what I could do without her on my side. Now, as promised, I have
more time to spend with her.
Harry Seldadyo
Contents
1 Setting the Scene
1
1.1
Basic Story . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.2
Main Findings . . . . . . . . . . . . . . . . . . . . . . . . . .
6
1.2.1
Determinants of Corruption . . . . . . . . . . . . . . .
6
1.2.2
Corruption Persistence . . . . . . . . . . . . . . . . . .
7
1.2.3
Governance and Growth . . . . . . . . . . . . . . . . .
8
1.2.4
Spatial Dimension . . . . . . . . . . . . . . . . . . . .
9
2 Determinants of Corruption: A Survey
11
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2
Corruption: Definition and Measurement
2.3
2.4
. . . . . . . . . . . 12
2.2.1
Definition . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2
Measurement . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.3
Criticisms . . . . . . . . . . . . . . . . . . . . . . . . . 17
Determinants of Corruption . . . . . . . . . . . . . . . . . . . 20
2.3.1
Four Classes . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.2
Empirical Issues . . . . . . . . . . . . . . . . . . . . . 29
Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 31
3 On the Sensitivity of Corruption Determinants
33
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2
The Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3
Some First Results . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4
Effect of Observations . . . . . . . . . . . . . . . . . . . . . . 45
iv
Contents
3.5
Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 54
4 Is Corruption Really Persistent?
57
4.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2
Convergence: Preliminary Evidence . . . . . . . . . . . . . . . 59
4.3
A Closer Look
4.4
Markov Chain Analysis . . . . . . . . . . . . . . . . . . . . . 74
4.5
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
. . . . . . . . . . . . . . . . . . . . . . . . . . 70
5 Governance and Growth Revisited
81
5.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2
Governance: Concept and Construct . . . . . . . . . . . . . . 83
5.3
Governance and Growth . . . . . . . . . . . . . . . . . . . . . 88
5.4
5.3.1
Parsimonious Model . . . . . . . . . . . . . . . . . . . 88
5.3.2
Effect of Observations . . . . . . . . . . . . . . . . . . 89
5.3.3
Sensitivity Analysis . . . . . . . . . . . . . . . . . . . 90
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6 Geography and Governance: Does Space Matter?
95
6.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2
Spatial Dependence . . . . . . . . . . . . . . . . . . . . . . . . 97
6.3
6.4
6.2.1
Preliminary Evidence . . . . . . . . . . . . . . . . . . 97
6.2.2
Spatial Regression Models . . . . . . . . . . . . . . . . 100
6.2.3
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.2.4
Results . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Spatial Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . 108
6.3.1
Local Statistics . . . . . . . . . . . . . . . . . . . . . . 108
6.3.2
Geographically Weighted Regression . . . . . . . . . . 112
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7 Putting the Pieces Together
117
7.1
Conclusions and Limitations of the Research
. . . . . . . . . 117
7.2
Policy Implications . . . . . . . . . . . . . . . . . . . . . . . . 121
Contents
v
Bibliography
123
Appendix
139
Samenvatting (Summary in Dutch)
155
List of Tables
2.1
Indicators of Corruption, Correlation and Summary Statistics 18
3.1
Robustness Analysis (No F , Z up-to-3) . . . . . . . . . . . . 40
3.2
Robustness Analysis (F =0, Z up-to-4 and up-to-5) . . . . . . 43
3.3
Robustness Analysis (F =0, Z up-to-3, Various N ) . . . . . . 48
3.4
Robustness Analysis (F =0, Z up-to-4, Various N ) . . . . . . 51
3.5
Robustness Analysis (F =0, Z up-to-5, Various N ) . . . . . . 53
4.1
Losers and Winners 1984-2003 . . . . . . . . . . . . . . . . . 61
4.2
Corruption Correlation Over Time . . . . . . . . . . . . . . . 63
4.3
Ordered Logit Regression . . . . . . . . . . . . . . . . . . . . 65
4.4
Ordered Probit Regression . . . . . . . . . . . . . . . . . . . . 67
4.5
Tests of β-Convergence . . . . . . . . . . . . . . . . . . . . . . 69
4.6
Tests of σ-Convergence . . . . . . . . . . . . . . . . . . . . . . 70
4.7
Modality (Rescaled) ICRG data . . . . . . . . . . . . . . . . . 74
4.8
Markov Chain Estimates, 1984-2003 . . . . . . . . . . . . . . 78
5.1
Correlations among the ICRG Indicators of Governance . . . 85
5.2
CFA Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.3
Parsimonious Growth-Governance Regressions . . . . . . . . . 89
6.1
Moran’s I and Geary’s c for Governance . . . . . . . . . . . . 100
6.2
Spatial Governance using W 1 . . . . . . . . . . . . . . . . . . 105
6.3
Spatial Governance using W 2 . . . . . . . . . . . . . . . . . . 107
6.4
Normalization by Largest Eigenvalue (W 1∗ and W 2∗ ) . . . . 110
viii
List of Tables
6.5
Significance Tests . . . . . . . . . . . . . . . . . . . . . . . . . 115
List of Figures
1.1
10 Most Corrupt and Cleanest Countries 2005-2006 . . . . . .
3
3.1
Corruption (y-axis) and Two Robust Determinants (x-axis) . 50
4.1
Annual Kernel Density, 1984 (top-left) to 2003 (bottom-right) 73
5.1
Governance in Factor Analysis Model . . . . . . . . . . . . . 86
5.2
Regressions with Various Samples . . . . . . . . . . . . . . . . 91
5.3
Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . 92
6.1
Governance and Distance to Worst and Best Practices, 2005 . 99
6.2
Local Moran’s I Statistics (Matrices W 1 and W 2 ) . . . . . . 111
6.3
Local Geary’s c Statistics (Matrices W 1 and W 2 ) . . . . . . 111
6.4
Local Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . 114
Chapter 1
Setting the Scene
“How do I tell the story of trillions of dollars of dirty money
lodged around the world,
derived from many types of illegal activities,
shifting across some 200 countries,
affecting both economic and political affairs?”
————————————————————
Raymond W. Baker (2005)
1.1
Basic Story
Corruption is one of the unholy trinity of dirty money, together with criminal
and illegal commercial activities (Baker, 2005). It is a dark side of society,
claimed to cause people in corrupt countries to stay poor and illiterate, and
to suffer from high infant and child mortality rates, low birth-weight babies,
as well as high dropout rates in primary schools (Kaufmann et al., 1999;
Gupta et al., 2001). Not only that, corruption is also said to deteriorate
countries’ distribution of income (Li et al., 2000), quality of public infrastructure (Tanzi and Davoodi, 1997), and productivity (Lambsdorff, 2003a).
Clearly, it is a serious problem and nothing good can be expected from
corruption. As Tanzi (1997: 164-165) states:
“I must confess that I have little patience for those who try to find benefits
in corruption. These individuals are often addressing artificial or unusual
situations. Unfortunately, when corruption exists, it is often widespread.
It affects not just some decisions or sectors but many decisions and most
2
Chapter 1
sectors. Thus, in reality, corruption is likely to distort markets and to impose
major costs on the economy.”
As corruption is a world-wide phenomenon, not surprisingly it has attracted a lot of attention in both the policy and academic arenas. Some
decades ago, there were only a few studies on corruption. Since then, numerous academic articles have been published.1 This partly reflects an
increased public concern for the problems caused by corruption, and partly
explains the emergence of various perceived-corruption indexes that have
made corruption measurable.
Today, the world is still split in countries that are governed by clean
governments and those that are not.2 The group of clean countries mainly
consists of Western nations, while the group of dirty countries predominantly contains African, Asian, and Latin American nations. The International Country Risk Guide (ICRG) that produces a perceived-corruption
index reports that 79 per cent of the around 140 nations in the world (20052006) are run by corrupt bureaucrats. An aggregated corruption index of
Kaufmann et al. (2007) of the World Bank (WB) gives a lower figure: 59
per cent of 207 countries and territories are corrupt.3 Figure 1.1 shows the
10 most and least corrupt countries in 2005-2006 according to both indexes.
Finland is always regarded as the cleanest country by these indexes. Meanwhile, Somalia is the most corrupt country according to the WB index, and
Zimbabwe according to the ICRG data.
But, what is corruption? Corruption can be viewed from different angles. A commonly accepted view is Waterbury’s (1973: 533) definition
explaining corruption as “the abuse of public power and influence for private ends.” Jain (2001) gives a socio-political context of corruption, namely
discretionary power, economic rents, and a weak judicial system in defining
1
The Google Scholar search engine (12 November 2007) indicated some 157,000 articles
on corruption in social sciences, arts, and humanities.
2
Later, we will show that there is a tendency of convergence via upward and downward
dynamics.
3
The ICRG index is scaled between 0 and 6, while the WB index ranges between −2.5
and +2.5, where a higher score denotes less corruption. The division of the countries into
corrupt and clean countries and territories is based on the middle score of each index.
Zimbabwe
Kenya
Rep.Congo
Gabon
Iraq
Lebanon
Myanmar
Niger
N.Korea
PNG
Austria
Canada
Luxembourg
Netherlands
Norway
Sweden
Denmark
Iceland
N.Zealand
Finland
Somalia
Myanmar
Eq.Guinea
N.Korea
Afghanistan
Haiti
Rep.Congo
Iraq
N.Caledonia
Zimbabwe
Austria
Netherlands
Norway
Switzerland
Sweden
Singapore
N.Zealand
Denmark
Iceland
Finland
Setting the Scene
3
2.5
1.5
0.5
-0.5
-1.5
-2.5
(a) WB Index
6.0
5.0
4.0
3.0
2.0
1.0
0.0
(b) ICRG Index
Figure 1.1: 10 Most Corrupt and Cleanest Countries 2005-2006
4
Chapter 1
corruption. He (p. 73) defines corruption as an act “in which the power
of public office is used for personal gain in a manner that contravenes the
rules of the game.”
Corruption also reflects poor governance of government, i.e., public officials are not qualified, laws are not adhered to, and lack of public transparency and accountability. Government governance itself refers to the
process of decision-making by the government and the process by which
decisions are implemented (or not implemented). Poor governance is often
regarded as one of the root causes of all evil within societies. Although
definitions of governance differ, there seems to be broad consensus that
good governance has various characteristics: It is accountable, transparent, responsive, effective and efficient, and follows the rule of law, thereby
assuring that corruption is minimized.
Clearly, corruption is a failure of good governance or a manifestation
of poorly functioning state. How corruption reflects poor governance is
indicated by Hassan (2004: 32). In his words,
In the political realm, corruption undermines democracy and good governance by subverting formal processes and rules of conduct. It erodes the
institutional capacity of government as established procedures are disregarded, resources are siphoned off, and officials are assigned or promoted
without regard to performance. Corruption in elections usually elects the
wrong people, those who are parasites and put personal greed over national
interests. Corruption in legislative bodies undermines accountability and
representation in policy making. In the administrative realm, corruption
results in the unequal provision of services, which undermines the States
legitimacy and, in extreme cases, may render a country ungovernable and
lead to political instability and social conflict. Corruption in the judiciary
circumvents the rule of law, and justice is often delivered late or even denied.
This dissertation mainly deals with the issues of corruption and governance. It raises a general question about the causes and consequences
of corruption and governance. As the causes of corruption may be deeply
rooted in economic, political, and socio-cultural as well as geographical environment (Rose-Ackerman, 1999; Andvig et al., 2000; Jain, 2001), the first
research question that we deal with is what determines the cross-country
variation in corruption.
Setting the Scene
5
Apart from being widespread, many authors argue that corruption is
persistent. There is some theoretical support for this view (Basu et al.,
1992; Tirole, 1996; Mauro, 2004; Mishra, 2006) and some scant empirical
evidence as well (Herzfeld and Weiss, 2003; Alesina and Weder, 2002; Damania et al., 2004). However, like other institutions, corruption is a not
static phenomenon. Over a subtantial time span, we may observe a change
in corruption. Hence, the second research question is whether corruption is
really persistent.
As corruption is closely linked to governance, the third issue to investigate is the relationship between economic growth and government governance. Authors like Knack and Keefer (1995), Barro (1997), Keefer and
Knack (1997), and Chong and Calderon (2000) argue that the impact of
governance on growth is positive, but others report the opposite (Quibria,
2006). A deeper look at country cases (Qian, 2003; Pritchett, 2003) tends
to support the latter view. China and Vietnam, for example, are still far
away from good governance, but their growth rates are remarkable. Under
the kleptocratic regime run by Soeharto, Indonesia was one of Asian ‘tigers’.
Contributing to this debate, we re-examine the relationship between growth
and governance.
The idea to combat corruption via the promotion of good governance
has spread across the globe. At the same time, nations in the world are
not isolated from international economic and political interactions. Hence,
neighbors may matter in the formation of good governance to result in a
spatial dependence among that are close nations. This spatial dependence
may be due to spillovers and diffusion-adoption processes, as also found
in democracy, war and peace, or economic liberty (O’Loughin et al., 1998;
Ward and Gleditsch, 2002; Simmons and Elkins, 2004). It also can stem
from policy convergence (Mukand and Rodrik, 2005), interdependency of
policy decisions (Brueckner, 2003), or transmission of government forms
(Starr, 1991). This implies that countries may be geographically clustered
according to their quality of governance. At the same time, the effect of the
determinants of governance may not be homogenous over space, i.e., the
6
Chapter 1
determinants may have a different impact in different countries. These are
the final issues covered in this dissertation.
To summarize, the four issues discussed in this dissertation are: (1)
what determines corruption?; (2) is corruption persistent?; (3) what is the
relationship between governance and growth?; and (4) does space matter
in explaining cross-country variation in the quality of governance? The
following section summarizes our approach and main findings.
1.2
1.2.1
Main Findings
Determinants of Corruption
As shown in Chapter 2, the theoretical literature provides many view points
in explaining the existence of corruption. In the absence of a theory-based
consensus on what determines corruption, empirical researchers typically
experiment with a set of variables that may be correlated with corruption.
Others focus on a particular variable of interest, using a set of control variables. If one runs a regression using a particular combination of explanatory
variables, it is possible to find that one variable of interest is significant but
becomes insignificant when other combinations of explanatory variables are
used. The same holds for the sign of the estimate: the impact of a variable
may change if different variables are controlled for.
Using the Sensitivity Analysis (SA) of Sala-i-Martin (1997), we deal
with this issue in Chapter 3. The idea behind this approach is to generate a
series of estimates (β’s) of particular variables of interest using all possible
combinations of control variables drawn from a pool of variables. The pool
contains variables argued to be correlated with corruption. Our focus is to
examine whether the distribution of β lies on one side of the cumulative
distribution function. We employ 45 variables that previously have been
claimed to be significant in explaining cross-country variation in corruption.
We also experiment with different numbers of observations to examine the
robustness of the finding. We find that only two variables having stable
coefficients, namely government effectiveness and rule of law.
Setting the Scene
7
To come up with the list of variables, we first survey what researchers
report on the determinants of corruption. Updating surveys by Andvig et
al. (2000) and Jain (2001), Chapter 2 starts with a discussion of the definition and measurement of corruption. Although there is a widely accepted
definition, corruption is difficult to measure. Corruption can be indirectly
measured via the perception of respondents in surveys, but there are also
some studies trying to directly measure it in money metric terms. Nowadays, perceived-corruption indexes have been widely used. At the same
time, these indexes have been criticized. Some scholars criticize such indexes from a technical point of view (Galtung, 2006; Kurtz and Schrank,
2007), others from a more substantive perspective (Olken, 2006; Donchev
and Ujhelyi, 2007). Some of these criticisms, however, have limited validity.
In addition, this chapter discusses determinants of corruption as found in
the literature. While other categorizations are possible, we may group these
determinants into four broad classes. First, economic and demographic
determinants including economic institutions. Second, variables categorized
as political institutions. Third, those falling in the area of judicial and
bureaucracy determinants, and finally, geography and culture.
1.2.2
Corruption Persistence
Theoretical and empirical research on corruption generally concludes that
corruption is persistent. However, using ICRG data for the period 19842003, we find strong evidence that corruption changes over time. Many
corrupt countries saw their level of corruption decline, while many clean
countries became less clean over the same period. We also find that this
convergence process is not continuous: there has been an improvement in
world corruption in the first part of our sample period, but a worsening in
the second half.
We start with a simple check on the correlation between the current
and past levels of corruption, and find that the correlation shrinks when
the time lag increases. We also find evidence on β-convergence when we
8
Chapter 1
regress the change in corruption on its initial value. Examining the trend
as indicated by the standard deviation and coefficient of variation of crosscountry corruption over time, we discover σ-convergence. Also, the results
from a set of ordered logit and probit regressions give additional support to
the conclusion that corruption is not persistence.
These finding are confirmed in our further analysis of the dynamics of
the distribution of corruption data. There is a significant modality shift in
the corruption distribution over time. Using a Gaussian kernel function we
detect a transformation in the distribution of corruption from a bimodal to
a unimodal distribution. Finally, on the basis of a Markov chain analysis, we
also find interclass upward and downward shifts of countries. These issues
are discussed in Chapter 4.
1.2.3
Governance and Growth
The literature on the governance-growth relationship does not provide clear
cut conclusions about the relevance of governance for growth. There are
two issues in this debate: the robustness of the relationship and the way
governance is measured. We contribute to this debate by introducing our
governance index generated using Confirmatory Factor Analysis (CFA) on
ICRG governance indicators, namely democratic accountability, government
stability, bureaucracy quality, corruption, and law and order. CFA is a
latent variable approach that can be used to analyze proxies for theoretical
concept of governance. We discover that the five indicators of governance
can be combined into one single index with an impressive goodness of fit.
Using our CFA-based index of governance, we test the stability of the
growth-governance nexus via a set of parsimonious models, recursive regressions, and the Sensitivity Analysis of Sala-i-Martin mentioned previously.
We discover that our index is fairly robust in a series of experiments using
different control variables as well as different numbers and compositions of
observations. In the majority of cases, we find that good governance promotes economic growth significantly. Chapter 5 discusess this issue in more
Setting the Scene
9
details.
1.2.4
Spatial Dimension
Researchers widely recognize the role of geography in shaping countries’
quality of governance. Several proxies have been used for geography including latitude, climate, temperature, country size, climate-related diseases, or
dummies indicating that countries are landlocked, islands, or belong to a
particular region (Acemoglu et al., 2001, 2002; Easterly and Levine, 2003;
Rodrik et al., 2004; Olsson and Hibbs Jr., 2005). These proxies, however,
do not capture the spatial dimension of governance, whereas cross-country
data are generally characterized by spatial dimension.
We consider two spatial dimensions of governance, namely spatial dependence and spatial hetereogeneity. While the former refers to the degree
of dependence of a country’s governance on that of its neighbors’, the latter
deals with varying effects of the determinants of governance. Our preliminary analysis indicates that the closer the country’s distance to the best
(worst) practice of governance, the higher (lower) its level of governance.
We also discover a positive global spatial dependence, where poorly (well)
governed countries are geographically clustered with poorly (well) governed
countries.
As a further check we apply a spatial econometric method introduced by
Anselin (1988). This approach is superior to other techniques in capturing
the behavior of neighbors through a weight matrix. The weight matrix
does not only consist of physical-geographical distance, but also ‘political
distance’ where countries’ regime types are taken into account. We find that
governance in one country exhibits a positive relationship with governance
in neighboring countries.
Finally, we also run Geographically Weighted Regression (GWR) of
Brunsdon et al. (1999) that allows us to analyze the issue of spatial nonstationarity defined as “the variation in relationships and processes over
space” (Brunsdon et al., 1999: 497). We discover that the coefficients of
10
Chapter 1
the determinants of governance are not constant, but vary across observations. Chapter 6 discusses the spatial dimension of governance in more
detail. Chapter 7 offers the conclusions of this dissertation.
Chapter 2
Determinants of Corruption:
A Survey
“If government were a product, selling it would be illegal.”
————————————————————
P. J. O’Rourke (1993)
2.1
Introduction
Corruption is a world-wide phenomenon that is multi-faceted. While it
also exists in the private sector, corruption primarily involves government
officials. Hence, it is not surprising that corruption is labeled as “endemic
in all governments” (Nye, 1967: 417), where “... no region, and hardly any
country, has been immune” (Glynn et al., 1997: 7). Corruption is probably
as old as government itself.
Corruption affects almost all parts of society. Like a cancer, as argued
by Amundsen (1999: 1), corruption “eats into the cultural, political and
economic fabric of society, and destroys the functioning of vital organs.”
The World Bank (WB) has identified corruption as “the single greatest
obstacle to economic and social development. It undermines development
by distorting the rule of law and weakening the institutional foundation
Earlier versions of this chapter—joint work with Jakob de Haan—were presented at the
European Public Choice Society meetings in 2005 and 2006
12
Chapter 2
on which economic growth depends.”1 This explains why anti-corruption
measures rank high on the policy agenda of the World Bank and the United
Nations (2007) as exemplified by its recent Stolen Asset Recovery (StAR)
Initiative.
Corruption has also attracted the attention of researchers in the academic arena; not only in economics, but also in sociology, political science,
law, etc. Research in this area includes detailed descriptions of corruption scandals, case studies, and cross-country studies. It also ranges from
theoretical models to empirical investigations.
This chapter updates the surveys by Andvig et al. (2000) and Jain
(2001) and reviews empirical studies2 on the causes of corruption. Since corruption has a negative impact on economic outcomes (Mauro, 1995; Tanzi
and Davoodi, 1997; Gupta et al., 1998; Lambsdorff, 2001), it is important
to know which factors determine corruption.
The rest of this chapter is constructed as follows. In section 2.2 we
discuss the concept of corruption and its measurement. Section 2.3 explores
some empirical issues concerning the determinants of corruption. Section
2.4 concludes.
2.2
2.2.1
Corruption: Definition and Measurement
Definition
The Oxford Advanced Learner’s Dictionary (2000, p. 281) describes corruption as: (1) dishonest or illegal behaviour, especially of people in authority;
(2) the act or effect of making somebody change from moral to immoral
standards of behaviour. Here, corruption is linked to two important elements: authority and morality. Authors like Gould (1991: 468) explicitly
define corruption as a moral problem, i.e., it is “an immoral and unethical
phenomenon that contains a set of moral aberrations from moral standards
of society, causing loss of respect for and confidence in duly constituted au1
2
www1.worldbank.org/publicsector/anticorrupt/index.cfm.
An excellent survey on theoretical studies on corruption is Aidt (2003).
Determinants of Corruption: A Survey
13
thority.” This is in line with Dobel (1978: 960) who labels corruption as
“the moral incapacity to make disinterested moral commitments to actions,
symbols, and institutions which benefit the substantive common welfare.”
These normative definitions, however, are not without problems. Moral
norms differ from place to place and change from time to time. For example,
whose moral standard should be used, or what is the appropriate moral
benchmark if there is more than one standard? In African traditions ‘gift
giving’ is a common practice (de Sardan, 1999), but in Western cultures it is
often regarded as corruption (Qizilbash, 2001). Also, what was not regarded
in the past as corrupt acts, now may be labeled as corruption, and the other
way around. Moreover, viewing corruption merely as a moral problem tends
to individualize this social phenomenon and ignores the wider socio-political
context of corruption.
For corruption to exist, according to Jain (2001), three conditions should
be fulfilled: discretionary power, economic rents, and a weak judicial system.3 Discretionary power relates to authority to design and administer
regulations, which, in turn, is accompanied by the presence of extracted
rents associated with power. A weak judicial system implies a low probability of detection and lack of sanctions. Jain (p. 73) thus defines corruption
as an act “in which the power of public office is used for personal gain in a
manner that contravenes the rules of the game.” It brings us back to Waterbury’s (1973: 533) definition which is now widely accepted, i.e., corruption
is “the abuse of public power and influence for private ends.” Under this
definition, corruption takes many forms varying from the minor abuse of
influence to institutionalized bribery and systematic kleptocracy.
2.2.2
Measurement
The literature provides three ways of measuring corruption. First, direct
estimates in money metric terms that mainly capture corruption by politicians. For example, the Transparency International Global Corruption Re3
In the same vein, Klitgaard (2000) constructs an ‘equation’ saying that corruption
equals monopoly power plus discretion minus accountability.
14
Chapter 2
port 2004 estimates the size of stolen asset by the top 10 world kleptocrats,
ranging from 0.07-0.08 billion $US (Josep Estrada, the President of the
Philippine, 1998-2001) to 15-35 billion US$ (Mohamed Soeharto, the President of Indonesia, 1967-1998).4
Second, estimates based on micro level data. Some authors employ a
direct approach (e.g., Henderson and Kuncoro, 2004 and Kuncoro, 2004),
where the reported bribe payments to public officials is used to calculate
the magnitude of corruption at the micro level. Others employ an indirect
approach. A good example is the study of Olken (2006) who compares
Indonesian villagers’ beliefs about the likelihood of corruption in a roadbuilding project with a measure of ‘missing expenditures’ in the project.
Third, estimates based on the perception of the likelihood, frequency, or
level of corruption of respondents in surveys. There are more than 25 organizations around the world that collect and publish this type of corruption
data, either as poll-based data (primary source) or as poll-of-polls-based
data (secondary source).
A good example of the poll-based data is the International Country Risk
Guide (ICRG) data covering almost 150 countries since the beginning of the
1980s—making it the largest panel dataset available.5 The ICRG data are
expert-based assessments of political, economic, and financial risks. Corruption is one of the 12 political risk components, with scores ranging between
0-6, where a higher score means less corruption. Corruption is captured via
“actual or potential corruption in the form of excessive patronage, nepotism,
job reservations, ‘favor-for-favors’, secret party funding, and suspiciously
close ties between politics and business.”6
Another example is the World Economic Forum (WEF) dataset reported
in The Global Competitiveness Report. Released since 1979, it initially cov4
The list appeared also in the UN-WB report StAR Initiative 2007, but some of the
estimates have been actually published in Time Asia (24 May 1999) and The Guardian
(26 March 2004).
5
The complete list of primary sources can be found in the reports of Kaufmann et al.
(Governance Matters) and Lambsdorff (Framework Document).
6
www.prsgroup.com/ICRG Methodology.aspx
Determinants of Corruption: A Survey
15
ered only 16 countries, but over time has been expanded to almost 120
countries. It is based on an annual opinion survey that records the perspectives of business leaders around the world who compare their own operating
environment with global standards on a wide range of dimensions. Some of
these provide information on the perception of corruption. For instance, one
issue refers to ‘favoritism in decisions of government officials when deciding
upon policies and contracts.’ Other dimensions include ‘irregular payment’
in export and import, government procurement, tax collection, public contracts, and judicial decision as well as business cost of corruption. Each
of these variables is scaled between 1-7, where a higher score means less
corruption.
The third example of primary data is the one reported in the World
Competitiveness Yearbook of the Institute for Management Development
(WCY-IMD). The WCY has been published since 1987 and was initially a
joint initiative with the WEF. The IMD reports corruption on the basis of
a survey among thousands domestic and foreign firms operating in about
50 countries. There are various variables that are related to corruption, but
the most explicit one is an indicator labeled ‘bribery and corruption exist in
public sphere’. This indicator has a scale ranging from 0 (highly existing)
to 10 (not existing).
A well-known example of the poll-of-polls-based data is the Transparency
International (TI) index, called Corruption Perception Index (CPI).7 This
data covers about 130 countries. At least three primary surveys or sources
should be available for a country to be included in the calculation of CPI.
Computed by Lambsdorff on behalf of the TI since 1995, the CPI aggregates various perception-based indicators of corruption to a new index on
0-10 scale index where a higher score means less corruption. The index is
constructed via two steps. First, standardization of the primary data is done
by a two-sequential procedure: matching percentile and β-transformation.8
7
www.transparency.org/policy research/surveys indices/cpi/
Before 2002 the standardization was done by a simple mean and standard deviation
approach.
8
16
Chapter 2
The former is to tackle the differences in scaling system among the primary
sources as well as in the distribution of data, and also to ensure that the
resulting index lies between 0-10. The latter is to handle the tendency of
the standard deviation to attenuate over time. In the second step, the final index is computed as the unweighted average of the standardized and
transformed data (Lambsdorff, 2002, 2003b, 2004).
Due to the differences in the standardization and transformation and the
use of different sources over time, the CPI is not a consistent time series.
In Lambsdorff’s (2000: 4) words, “year-to-year changes may not only result
from a changing performance of a country . . . changes can result from the
different methodologies . . . not necessarily from actual changes.”9
The other main source of poll-of-polls corruption data is the World Bank
(Kaufmann and Kraay, 2002; Kaufmann, et al., 1999, 2006, and 2007).10
Corruption is one of the six components of the Kaufmann governance index.11 Reporting corruption data since 1996, this index covers almost 200
countries and territories and draws upon about 40 data sources produced by
more than 30 different organizations. To aggregate the various corruption
indicators, Kaufmann et al. (1999, 2002) use a latent variable approach in
which the observed perception indicators of corruption are expressed as a
linear function of the latent concept of corruption plus a disturbance capturing perception errors and sampling variation in each indicator. The resulting
index ranges between −2.5 (most corrupt) and +2.5 (least corrupt).
Table 2.1 reports the correlation among the indicators discussed above.
Both the Pearson and polychoric correlation coefficients demonstrate the
closeness of the corruption indicators. Apart from the TI, the World Bank,
and the ICRG indicators, the table also shows seven indicators of corrup9
In line with this, since it relies heavily on independently conducted surveys and expert
polls, the CPI is not available for a significant number of countries. Sometimes countries
are no longer included if the required minimum number of sources is missing. Galtung
(2006) therefore concludes that the CPI does not measure trends as it is a defective and
misleading benchmark of trends.
10
www.worldbank.org/wbi/governance/wp-governance.html
11
The other components are voice and accountability, political instability and violence,
government effectiveness, regulatory quality, and rule of law.
Determinants of Corruption: A Survey
17
tion drawn from the WEF, namely irregular payments in public utilities,
public contract, judicial decision (d1-d3), export-import, tax collection (d5d6), and business cost of corruption (d4), as well as favoritism in policy
making (d7). Using these WEF indicators, we also construct an aggregate
index of corruption using Principal Component Analysis (PCA).12 It is clear
that the correlation among the WEF indicators is very high, and so is the
correlation among the resulting PCA-based WEF index, the Lambsdorff,
Kaufmann, and ICRG indicators of corruption.13 In other words, although
the indicators are different, they proxy the same phenomenon: corruption.
2.2.3
Criticisms
Some of the critique on these indicators of corruption depart from the definition underlying these indicators. Galtung (2006), for example, argues that
the CPI does not explicitly distinguish between corruption in the civil service and political corruption. Likewise, Kurtz and Schrank (2007) criticize
that the measures of corruption that combine questions about the presence
of nepotism, cronyism, and bribe taking place in government with the intrusiveness of the bureaucracy or the amount of red tape may be problematic.
In their words (p. 543), “intrusiveness and red tape can be a sign of either
effective or ineffective governance, depending on the content of the policies
being enforced.”
In line with this is critique referring to the use of different sources to
come up with an indicator of corruption. Different sources may give different
definitions of corruption and different forms of corrupt act. Andvig (2005)
questions whether the different sources underlying the CPI cover the same
phenomenon. Also, Søreide (2003: 7) writes,
12
Other latent variable estimates such as Principal Factor (PF) and Maximum Likelihood
(ML) based Factor Analysis used to produce aggregate indexes give very similar results.
Each of the three techniques produces only one factor with a high eigenvalue and very
high loadings of the underlying indicators.
13
The correlation between the two aggregated indexes (WB and TI) and the three individual indexes are certainly high, because the former indexes contain the later indexes.
0.97
0.89
0.93
0.85
0.87
0.88
0.92
0.86
0.85
0.81
0.97
207
0.00
1.00
-1.68
2.49
WB
0.78
0.70
0.74
0.72
0.80
0.72
0.67
0.71
0.93
140
2.50
1.18
0.00
6.00
0.95
0.94
0.99
0.88
0.94
0.85
0.90
0.89
0.93
0.88
0.86
0.82
0.98
168
4.05
2.14
1.80
9.65
ICRG
TI
0.92
0.97
0.95
0.95
0.96
0.94
0.85
0.93
117
0.00
2.47
-4.69
4.96
WEF
(PCA)
0.93
0.95
0.87
0.88
0.82
0.82
0.91
0.92
0.67
0.84
117
5.04
1.09
2.10
6.80
WEF1
0.86
0.88
0.80
0.91
0.91
0.89
0.92
0.89
0.83
0.91
117
4.06
1.06
2.10
6.40
WEF2
0.89
0.92
0.84
0.98
0.86
WEF Indicators
WEF3 WEF4 WEF5
0.90
0.94
0.87
0.91
0.95
0.89
0.83
0.88
0.80
0.97
0.96
0.96
0.85
0.83
0.95
0.95
0.92
0.92
0.92
0.92
0.92
0.87
0.88
0.87
0.87
0.85
0.93
0.79
0.85
0.74
0.89
0.94
0.85
117
117
117
4.57
4.36
4.66
1.25
1.13
1.13
2.40
2.20
2.20
6.80
6.80
6.70
0.87
117
3.21
0.87
1.60
5.40
WEF7
0.84
0.85
0.80
0.88
0.69
0.87
0.82
0.90
0.74
0.68
53
4.63
2.62
0.51
9.38
0.97
0.98
0.93
0.93
0.84
0.91
0.89
0.94
0.85
0.82
0.87
IMD
WEF7 : favoritism in policy making.
WEF4 : business cost of corruption; WEF5 -WEF6 : irregular payments in export-import and tax collection;
PCA is based on indicators d1-d7. WEF1 -WEF3 : irregular payments in public utilities, public contract, and judicial decision;
0.68
0.82
117
4.88
1.19
2.20
6.80
WEF6
0.85
0.86
0.77
0.92
0.89
0.87
0.89
0.82
0.92
Correlation is based on the 2005-06 data. The lower (upper) diagonal is Pearson (polychoric) correlation.
Corruption
Indicators
a. WB
b. TI
c. ICRG
d. WEF
d1. WEF1
d2. WEF2
d3. WEF3
d4. WEF4
d5. WEF5
d6. WEF6
d7. WEF7
e. IMD
Observations
Mean
Std. Dev.
Minimum
Maximum
Table 2.1: Indicators of Corruption, Correlation and Summary Statistics
18
Chapter 2
Determinants of Corruption: A Survey
19
“Most of the polls and surveys ask for a general impression of the magnitude of the problem, which actually means people’s subjective intuitions
of the extent of a hidden activity. For the TI index, only one source asks
for people’s personal experiences with corruption. The quantification of the
problem is highly ambiguous. It is not clear to what extent the level of corruption reflects the frequency of corrupt acts, the severity to society, the size
of the bribes or the benefits obtained. Most of the surveys do not specify
what they mean by the word corruption. It can thus be quite difficult for
the respondents to answer when asked about a quantification of ‘the misuse
of public office for private or political party gain’ or when encouraged to
rate ‘the severity of corruption within the state’.”
These criticisms, however, have limited validity. As indicated in Table 2.1,
various measures of corruption are highly correlated, suggesting that they
measure the same phenomenon.
Another criticism points to the design of surveys, e.g., surveys designed
to capture perception may have a bias that affects the respondents’ interpretation of the questions in the surveys. As Bertrand and Mullainathan
(2001) argue, the sequence of questions may substantially affect the respondents as they tend to answer questions in line with their answers to previous
questions. Also, the prior questions may bring out certain memories which
influence the answers. Other technical problems raised by Bertrand and
Mullainathan include the time spent by the respondents to scrutinize each
question, scoring effects, and other cognitive issues.
A more substantive criticism is indicated by Olken (2006) and Donchev
and Ujhelyi (2007). They argue that what is believed by the respondents
may not reflect what actually happens. Fisman and Miguel (2006), however,
provide some evidence that beliefs about corruption and actual corruption
are closely related. They study the parking behaviour of UN diplomats who
have immunity from parking enforcement actions and compare the data
with perceived corruption. They find that diplomats from corrupt countries
(based on existing perception-based indexes) accumulate significantly more
unpaid parking violations demonstrating the abuse of power.
The choice of respondents is also criticized as it may potentially influence
the responses, i.e., different participants may give different opinions about
a country under review. Kurtz and Schrank (2007) argue that surveys on
20
Chapter 2
corruption may reflect the narrow interests of respondents, say, business
communities. A potential bias in sample selection may occur due to the exclusion of respondents “who did not succeed in the marketplace, or . . . who
were deterred from entering local markets by pervasive malgovernance or
corruption itself.” As a consequence, the sample is dominated by unrepresentative respondents, while those who “show up in the surveys . . . are the
beneficiaries of corruption and cronyism . . . therefore unlikely to report it
accurately.” Moreover, “malgovernance is effectively reported . . . because it
is not pervasive enough to create sufficiently strong distortions in firm-level
survival or investor behaviour to induce selection bias. And thus in such
contexts those who do not win from malfeasance can survive to report it!”
(p. 543).
This criticism may be true for individual surveys, but less so for surveys
aggregating such individual surveys. Kaufmann, et al. (2007), for example,
use a mixture of sources of data coming from various agents: four crosscountry surveys of firms, seven commercial risk-rating agencies, three crosscountry surveys of individuals, six sets of ratings produced by government
and multilateral organizations, and 11 data sources produced by a wide
range of non-governmental organizations.
Despite all these criticisms, some of which appear to be more valid than
others, perception-based corruption indicators have created the possibility
to study corruption. As a result, numerous studies have employed such
indexes. In the following, we review studies on the determinants of corruption.
2.3
2.3.1
Determinants of Corruption
Four Classes
We survey cross-country studies on the determinants of corruption published
over the last 10 years. All studies that we are aware of are summarized in
Appendix 1. While other categorizations are possible, we distinguish four
broad classes of the underlying causes of corruption, namely (1) economic
Determinants of Corruption: A Survey
21
and demographic determinants, (2) political institutions, (3) judicial and
bureaucracy environment, and (4) geography and culture.
Economic and Demographic Determinants
There is a wide range of economic variables that have been suggested to
cause corruption. Income is a commonly-used variable to explain corruption (Damania et al., 2004; Persson et al., 2003; and van Rijckeghem and
Weder, 1997; among others). Corruption can be seen as an inferior good,
where the demand falls as income rises. Also, along with an increase in
income, more resources are available to combat corruption. Mostly proxied
by GDP per capita, income is also used to control for structural differences
across countries. It is generally found that income has a negative and significant effect on corruption, even though Kaufmann et al. (1999) and Hall
and Jones (1999) question the causal relationship between corruption and
income. Two studies using panel data (Braun and Di Tella, 2004; Fréchette,
2006) deviate from this commonly-found result, as they report that higher
income increases corruption, especially when country fixed effects are considered. Braun and Di Tella (p. 93) argue that this is due to the pro-cyclical
nature of corruption, where “moral standards are lowered during booms, as
greed becomes the dominant force for economic decisions.”
Income distribution is also argued to affect corruption. As Paldam (2002:
224) puts it, “A skew income distribution may increase the temptation to
make illicit gains.” Proxied by the Gini coefficient, he claims that income
disparity significantly increases corruption. However, using the income share
of the top 20% population, Park (2003) does not find a statistically significant relationship. Similarly, Brown et al. (2006) find no evidence that a
greater income inequality increases corruption.
The size of government is also put forward as a determinant of corruption, but the causality may run in two directions. If countries exploit
economies of scale in the provision of public services—thus having a low
ratio of public services per capita—those who demand the services might
22
Chapter 2
be tempted to bribe bureaucrats ‘to get ahead of the queue’. On the other
hand, a large government sector may also create opportunities for corruption. Some empirical studies (like Fisman and Gatti (2002) and Bonaglia et
al. (2001)) find a negative impact of government spending on corruption,
while others (like Ali and Isse, 2003) report the opposite.
Another variable that, according to various authors, also explains corruption is the share of import in GDP. Herzfeld and Weiss (2003), Fisman and Gatti (2002), Fréchette (2006), and Treisman (2000) report that
a higher import share leads to less corruption. A high import share implies lower tariff and non-tariff import restrictions. The presence of such
restrictions—like the necessary licenses to import, for example—offers an
opportunity to bribe. Similarly, restrictions on foreign trade, foreign investment, and capital markets stimulate corruption; see, for instance, Knack
and Azfar (2003), and Fréchette (2006). Broadman and Recanatini (2001,
2002) show the existence of a positive relationship between entry barriers
and corruption; that is, the greater the barriers to entry and exit faced by
firms, the more widespread is corruption. Some other evidence on the link
between corruption and economic openness is based on the trade share that
is generally found to reduce corruption (Brunetti and Weder, 2003; Knack
and Azfar, 2003; Persson et al., 2003; Fisman and Gatti, 2002; Bonaglia et
al., 2001; Fréchette, 2006; Wei, 2000b; Ades and Di Tella, 1999; Laffont and
N’Guessan, 1999; Leite and Weidmann, 1997) and competitiveness that is
found to decrease corruption as well (Gurgur and Shah, 2005; Broadman
and Recanatini, 2001).
Economic freedom—measured by the indexes of the Heritage Foundation
and the Wall Street Journal, or the Fraser Institute—is also repeatedly
found to lessen corruption. Proponents of this view are Gurgur and Shah
(2005), Park (2003), and Treisman (2000), but Lederman et al. (2005) and
Paldam (2002) find mixed results.
Some other economic variables are less frequently considered as determinants of corruption. Inflation is argued to stimulate corruption as it
increases the monitoring costs (Braun and Di Tella, 2004; Paldam, 2002).
Determinants of Corruption: A Survey
23
Foreign aid is found to rise corruption since transferred resources with substantial discretion in the absence of accountability to the decision maker
may create rents to be extracted (Ali and Isse, 2003; Tavares, 2003), although Knack (2001) finds no significant link between aid and corruption.
Finally, structural reform is also found to reduce corruption as it includes a
reduction in the size of the public sector and, more importantly, a fundamental shift in the state’s role “from one implying owning or controlling most
productive resources to one that is more narrowly defined around essential
state functions” (Abed and Davoodi, 2000: 7).
Socio-demographic determinants associated with corruption include human capital (proxied by schooling) and population growth. Economies with
a high level of human capital have a low level of corruption as indicated
by Ali and Isse (2003), Alt and Lassen (2003), Brunetti and Weder (2003),
Persson et al. (2003), Rauch and Evan (2000), Ades and Di Tella (1997
and 1999), and van Rijckeghem and Weder (1997). Education increases
the ability of society to control government behavior and to judge their
performance. At the same time, educated society also play a role as an
external control on corruption in the administration (Brunetti and Weder,
2003; Ades and Di Tella, 1999)).
Knack and Azfar (2003) show that an increase in population is followed
by an increase in corruption—a finding also reported Fisman and Gatti
(2002). This may be explained from the view of economies of scale in governance. As argued by Knack and Azar (p. 4), “in large nations rulers can
extract significant resources from the country and pay off the constituencies
necessary for them to maintain power.”
Political Institutions
Most empirical studies on the political causes of corruption focus on political freedom. Using various proxies for political freedom—like civil liberty,
political rights, democracy and length of the democratic regime—these studies generally conclude that more freedom reduces corruption. A common
24
Chapter 2
argument for this is that political freedom ensures political competition
and checks-and-balances mechanisms. At the same time, it also enforces
transparency in public services. In other words, political freedom creates
incentives for politicians to be ‘clean’ and implies constraints on misconduct. The list of studies reporting support for this conclusion includes
Kunicová and Rose-Ackerman (2005), Lederman et al. (2005), Gurgur and
Shah (2005), Braun and Di Tella (2004), Chang and Golden (2007), Damania et al. (2004), Herzfeld and Weiss (2003), Knack and Azfar (2003),
Broadman and Recanatini (2001, 2002), Paldam (2002), Bonaglia et al.
(2001), Fréchette (2006), Swamy et al. (2001), Treisman (2000), Ades and
Di Tella (1997; 1999), Goldsmith (1999), Leite and Weidmann (1997), and
van Rijckeghem and Weder (1997). Also freedom of the press is reported to
reduce corruption (Brunetti and Weder, 2003). In line with these findings,
political participation and constraints on the chief executive are found to
reduce corrupt behaviour as well (Kunicová and Rose-Ackerman, 2005).
Some aspects of democratic elections may, however, create opportunities
for corruption. Selecting politicians through party lists, for example, can obscure the direct link between voters and politicians, thus degrading the ability of voters to hold politicians accountable (Kunicová and Rose-Ackerman,
2005; Persson et al., 2003). Chang and Golden (2007) find that corruption
increases (decreases) with district magnitude under open-list (closed-list)
political representation. Similar results are found by Persson et al. (2003).
Authors like Park (2003), who uses Alesina and Perotti’s (1996) PCAbased socio-political instability index, and Leite and Weidmann (1999), who
take the frequency of revolutions, coups, and political assassinations as indicators of political instability, find that corruption is high in unstable polities.
Shleifer and Vishny (1993) point out that the ephemeral nature of public
positions in unstable systems makes the officials irresponsible and get them
involved in illicit rent-seeking behavior. Some support on the role of political
stability is also given by Wei (2000a).
Finally, some studies report that presidential system promotes corruption (Brown, et al., 2005; Kunicová and Rose-Ackerman, 2005; Lederman
Determinants of Corruption: A Survey
25
et al., 2005; Chang and Golden, 2007) as the president tends to have extensive legislative and non-legislative powers. Also the number of political
parties (Chang and Golden, 2007) is reported to increase corruption. When
the number of parties increases it is more difficult for the public to monitor
the behaviour politicians. Yet, ideological polarization reduces corruption
because more ideologically polarized systems provide constraints related to
credible political commitments (Brown, et al., 2005). Likewise, corruption
seems to be lower under a plurality system, as monitoring of rent seekers is
more stringent than under a proportional representation system (Kunicová
and Rose-Ackerman (2005).
Judiciary System and Bureaucratic Environment
There is no question that the quality of the judicial system is important in
explaining corruption. A weak judicial system is fertile land for corruption
to grow. Many studies have employed various indicators to come up with
this conclusion. Damania et al. (2004), for example, have used the rule of
law index of Kaufmann et al. (1999) that measures the extent to which economic agents abide by the rules of society, perceptions of the effectiveness
and predictability of the judiciary, and the enforceability of contracts. Others (Brunetti and Weder, 2004; Ali and Isse, 2003; Herzfeld and Weiss, 2003;
Park 2003; and Leite and Weidmann, 1999) use the index of the ICRG that
reflects the degree to which the citizens of a country are willing to accept
the established institutions to make and implement laws and adjudicate disputes. This index also measures the extent to which countries have sound
political institutions, strong courts, and orderly succession of power. All
studies mentioned conclude that a strong rule of law reduces corruption.
The bureaucratic environment is also relevant in influencing corruption.
Van Rijckeghem and Weder (1997) argue that public sector wages are highly
correlated with the rule of law and the quality of the bureaucracy, which
may therefore may have an effect on corruption. In developing economies
civil servants often receive wages that are so low that they entice corrupt
26
Chapter 2
behaviour. Measured as the relative magnitude of government wages to
GDP, Herzfeld and Weiss (2003) identify that an increase in public sector
wages significantly reduces corruption. Likewise, van Rijckeghem and Weder
(1997) report that the level of government wages is negatively correlated
with corruption. Similar results are found by by Alt and Lassen (2003) and
Rauch and Evans (2000). However, other studies find that this relationship
is not always statistically significant (Gurgur and Shah, 2005; Treisman,
2000).
Gurgur and Shah (2005), Brunetti and Weder (2003), and van Rijckeghem and Weder (1997) report that the higher the quality of the bureaucracy, the lower corruption will be. Similarly, absence of meritocratic recruitment, promotion and professional training in the bureaucracy are also
found to be associated with lower corruption (Rauch and Evans, 2000).
Decentralization or federalism has also been argued to affect corruption.
Enhancing the autonomy of local governments may result in two different effects (Lederman et al., 2005). It may promote the ability of local
governments to compete against each other for citizens, but it can also provide authorities to amplify regulations in areas already covered by the central government. Hence, the effect of decentralization remains an empirical
question. Measuring decentralization as transfers from central government
to other levels of national government as a percentage of GDP, Lederman
et al. (2005) find that this variable reduces corruption significantly. Likewise, taking a binary variable of centralized unitary states and decentralized
federal systems, Ali and Isse (2003) report that decentralized government
lowers corruption.
Gurgur and Shah (2005) use the ratio of employment in non-central
government administration to general civilian government employment and
show that corruption is lower in both decentralized unitary and federal
states but the impact is higher in decentralized unitary systems. Fisman
and Gatti (2002) measure decentralization as the share of state and local
government spending in total government spending and find a negative effect
on corruption. In line with this, Abed and Davoodi (2000) indicate that
Determinants of Corruption: A Survey
27
corruption is higher in countries governed by a central planning system.
However, others fail to find a negative effect of decentralization. Kunicová and Rose-Ackerman (2005), who use a dummy for autonomous regions
with extensive taxing, spending and regulatory authority, even argue that
federalism increases corruption, holding other factors constant. Likewise,
using a dummy variable for the presence of a federal constitution, Damania
et al. (2004) and Treisman (2000) find that a federal structure is more conducive to corruption. The reason for this finding, according to Brown et al.
(2006: 13), is that “as the political pie is divided between a greater number of geographic entities, opportunities to generate political rents increase.”
Similarly, Goldsmith (1999) demonstrates that federalism is associated with
more perceived corruption.
Geography and Culture
Geography is a broad term that includes both human and physical geography. As it explains the interactions among the members of the society
that shape norms, traditions, cultures, and institutions, human geography
may also matter for corruption. Ethnolinguistic homogeneity, for example,
is found to be negatively related to corruption (Ali and Isse, 2003). In a
heterogeneous and fragmented society, the probability that economic agents
will be treated equally and fairly are lower. As a result, highly fragmented
communities are likely to be more corrupt than homogenous societies.
It is also often argued that countries with many Protestants tend to
have lower corruption levels (Chang and Golden, 2007; Bonaglia et al., 2001;
Treisman, 2000; La Porta et al., 1999). Protestant traditions are commonly
used to proxy for an egalitarian community. Compared to more hierarchical
societies, the likelihood of corruption to occur in egalitarian communities is
lower. Paldam (2001) reports that in countries dominated by two religions,
namely Reform Christianity (i.e., Protestants and Anglicans) and Tribal
religions, corruption tends to be lower compared to countries in which other
religions dominate.
28
Chapter 2
Another cultural variable used to explain corruption is colonial heritage
that captures “command and control habits and institutions and the divisive
nature of the society left behind by colonial masters” (Gurgur and Shah,
2005: 18). The evidence on the relevance of this variable is, however, mixed.
Countries that have been colonialized tend to suffer from corruption (Gurgur
and Shah, 2005; Herzfeld and Weiss, 2003; Tavares, 2003). On the other
hand, some studies indicate that former British colonies have lower levels of
corruption. Persson et al. (2003) measure the influence of colonial history by
partitioning all former colonies into three groups, namely British, SpanishPortuguese, and other colonial origin. They conclude that former British
colonies tend to be less corrupt. Similarly, La Porta et al. (1999) report
that in countries with German and Scandinavian legal origin corruption is
lower compared to those with a Socialist and French legal origin.
Women’s participation is also argued to correlate with corruption. Swamy
et al. (2001) indicate that the more women are involved in the public arena,
the lower corruption will be. They also find that a higher share of women in
parliament and in government reduces corruption. Following Gottfredson
and Hirshi (1990) and Paternoster and Simpson (1996), they (p. 52) provide four arguments to explain the finding. First, “women may be brought
up to be more honest or more risk averse than men, or even feel there is
a greater probability of being caught.” Second, “women, who are typically
more involved in raising children, may find they have to practice honesty in
order to teach their children the appropriate values.” Third, “women may
feel more than men—the physically stronger sex—that laws exist to protect
them and therefore be more willing to follow rules.” Lastly, “girls may be
brought up to have higher levels of self-control than boys which affects their
propensity to indulge in criminal behaviour.”
Countries situated far away from the equator—as measured as absolute latitude—tend to have lower corruption levels (La Porta et al., 1999).
This may be interpreted as colonial influence, as settlers preferred another
climate. Meanwhile, Ades and Di Tella (1999) argue that trade distance
correlates to corruption. Corruption is high in countries located far away to
Determinants of Corruption: A Survey
29
large exporting nations. This distance protects such countries from the penetration of foreign competitors due to high transport costs. As this ‘natural
protection’ rises, corruption also increases. Similarly, remoteness isolates
countries to have an anti-corruption regime (Bonaglia et al., 2001).
Finally, another form of physical geography is natural endowment. A
higher export share of raw materials, such as fuel, mineral, and ore, stimulates corruption (Leite and Weidmann, 1997). Natural endowments create
rents-related corruption, according to Tornell and Lane (1998). Aslaksen
(2007), through various panel model specifications, provides evidence that
corruption is high in countries with many natural resources, especially oil.
This is confirmed in cross-section studies by Herzfeld and Weiss (2003),
Tavares (2003), and Bonaglia et al., (2001).
2.3.2
Empirical Issues
In the absence of a theory-based consensus on which to base empirical research on the causes of corruption (Alt and Lassen, 2003), studies on the
determinants of corruption face several problems. First, researchers do not
know the ‘true’ determinants of corruption so that the choice of variables
to be included in the empirical model are open to question.
Second, a certain variable can be significant in a particular model, but
becomes insignificant once other variables are taken into account. Some
techniques, like the Extreme Bounds Analysis (EBA) of Leamer (1983, 1985)
as well as Levine and Renelt (1992) or Sala-i-Martin (1997), the Bayesian
Model Averaging (BAM) as introduced by Chatfield (1995) and Drapper
(1995), or the general-to-specific approach (Hendry and Krolzig, 2005) may
be used to examine the issue of model uncertainty. The significance of
variables may also be influenced by the choice of a particular sample and
the presence of outliers. The latter issue can be dealt with using robust
estimation (Temple, 1998).
Third, there may be a multicollinearity problem related to the previous issue; that is, the estimate of the influence of a particular determinant
30
Chapter 2
of corruption depends on the range of other determinants taken into account. Only in the case that a particular determinant is orthogonal to all
other determinants in the model, can we safely evaluate the effect of this
determinant on corruption independently of the specification used in the
estimation. Hence, including highly correlated variables in a model at the
same time might be problematic. Aggregating correlated variables via Factor Analysis can be a way to cope with this problem.
Fourth, there may also be a simultaneity problem as there could be a
feedback relationship between corruption and one or more of its determinants. The corruption-income nexus, for example, is widely known to suffer
from this problem. An increase in income may reduce corruption, while a
lower level of corruption may lead to an improvement in income. The same
applies to many other variables suggested to cause corruption, like openness,
political stability, government size, etc.
Fifth, there may be indirect and interaction effects. Empirical models
of corruption usually assume the existence of a direct link between corruption and its determinants. But there may be indirect effects, i.e., a variable
affects corruption through its effect on another variable that, in turn, influences corruption. In such a case, a structural model might be useful.
Similarly, corruption may be affected by the combined effect of some variables. Hence, an interaction between such variables should be considered.
Sixth, there may be parameter heterogeneity. In models explaining corruption, it is usually assumed that the magnitude of the effects of the determinants are country-invariant. One may question this assumption since it
implies that, for example, the impact of income in poor and rich countries,
or the effect of press freedom in democratic and non-democratic countries,
on corruption is the same. Quantile Regressions (QR; Koenker and Hallock,
2001) can be used to tackle such heterogeneity, since each estimated quantile depicts a particular segment of the conditional distribution to result in
a heterogeneous portrait of the relationship between per corruption and its
determinants. Meanwhile, viewing this as spatial heterogeneity, Geographically Weighted Regression (GWR; Brunsdon et al. (1996, 1999), or even
Determinants of Corruption: A Survey
31
Hierarchical Linear Modelling (HLM; Raudenbush and Bryk, 2002) can also
be considered to tackle the issue of parameter homogeneity.
Finally, there might be a spatial dimension.
Countries tend to be
clustered with neighbors having similar institutions. Diffusion-adoption of
knowledge, regional political and economic interactions, policy convergence,
or socio-cultural similarities are responsible for this clustering phenomenon.
Spatial Regressions (SR) can deal with these effects (Anselin, 1988).
2.4
Concluding Remarks
There is no doubt that corruption has a damaging impact on economic
outcomes. The first step in combating corruption is to know its underlying
determinants. This chapter has surveyed the extensive literature on the
determinants of corruption. The main conclusions can be summarized as
follows.
Corruption is not simply the deviation from a certain moral standard,
thus it should be viewed in a wider socio-political context. Under this view,
corruption takes a variety of forms, which in empirical studies are captured
by various indicators. We find that these indicators are highly correlated.
Or, in the language of latent variable analysis, the various measures of
‘private wealth-seeking behaviour misusing the entrusted public authority’
can be collapsed into a single latent concept of corruption. Therefore, the
criticisms against the indicators and measurements of corruption may not
have much empirical ground. In addition to this, the criticism on the bias in
the choice of observations is not entirely correct. The widely used corruption
indexes have incorporated a variety of surveys capturing a broad range of
respondents. Also, it is reasonable to argue that the perceived corruption
index may mirror the frequency, severity, and magnitude of corruption.
If it is possible to measure corruption, it is also possible to investigate
its causes. There are some empirical issues, however, in investigating these
variables. This is mainly due to the absence of a theoretical consensus
on which to base an empirical model. Likewise, questions on which and
32
Chapter 2
how many variables should be considered in an empirical model remain.
Some researchers may pay attention to a particular variable of interest, but
others explore the corruption determinants from a more general perspective.
This survey has reviewed these factors and broadly classifies them into four
classes, namely (1) economic and demographic factors, (2) political factors,
(3) bureaucracy and judicial system, and (4) geography and culture.
It seems save to conclude that corruption is severe in low income countries that are less integrated with the world economy and are densely populated. The political determinants generally follows the 1887 Lord Acton’s
dictum saying that “Power tends to corrupt, and absolute power corrupts
absolutely.” Hence, lack of democracy increases corruption. Corruption is
also high in countries suffering from a weak judicial system and low quality
of bureaucrats who earn low wages. Countries with a higher fraction of
Protestants, that are ruled by a system rooted in British legal origin, and
that are ethnically less fragmented have a low level of corruption. Lastly,
countries situated around the equator and that have many natural resources
are more likely to be corrupt.
Chapter 3
On the Sensitivity of
Corruption Determinants
“Let’s not mince words . . .
We need to deal with the causes of corruption.”
————————————————————
James Wolfenson (2005)
3.1
Introduction
As corruption is generally believed to have negative welfare effects1 it is important to find out what determines corruption. Many studies have searched
for empirical correlations between corruption and a variety of economic and
non-economic determinants. Two questions are usually addressed in these
studies. First, what variables might explain the cross-country variation in
corruption? Second, how robust are these variables in explaining crosscountry differences in corruption? Unfortunately, there is no consensus in
the literature on both issues. In fact, theory offers little guidance in specEarlier versions of this chapter—joint work with Jakob de Haan—were presented at the
European Public Choice Society meetings in 2005 and 2006
1
A series of studies investigating the impact of corruption include Kaufmann et al.
(1999) and Gupta et al. (2001) on income, infant and child mortality rates, low-birthweight babies, and dropout rates in primary schools, Mauro (1995) on growth, Li et
al. (2000) on distribution of income, Tanzi and Davoodi (1997) on the quality of public
infrastructure, and Lambsdorff (2003a) on productivity.
34
Chapter 3
ifying a proper regression model for corruption. As a consequence, various specifications incorporating a wide range of explanatory variables have
been used to explain corruption and to find the ‘true’ determinants. It
is commonly found, however, that a particular variable is significant in a
particular model, but becomes insignificant when some other variables are
incorporated. Therefore, it is still not clear what variables are really driving
corruption.
We employ around 45 variables that in previous studies have been found
to be correlated with corruption and examine whether they are robustly related to corruption. We use two sets of corruption measures: aggregated
and individual indexes. The indexes of Kaufmann et al. of the World Bank
(WB) and Lambsdorff of Transparency International (TI) are aggregated indexes, while those of the International Country Risk Guide (ICRG) and the
International Management Development World Competitiveness Yearbook
(IMD-WCY) are individual indexes. In addition, we generate an aggregated
corruption index using the indicators of the World Economic Forum Global
Competitiveness Report (WEF-GCR).
Using a variant of the Sensitivity Analysis (SA) as proposed by Sala-iMartin (1997) and checking for sample sensitivity, we find that only two variables are consistently related to corruption, namely government effectiveness
and rule of law. Other variables like income, regulatory quality, protestant fraction, absolute latitude, economic freedom, or even democracy—
commonly argued to be significant determinants— are not robustly correlated with corruption. The remainder of this chapter is organized as follows. Section 3.2 describes our methodology and data in some detail, while
sections 3.3 and 3.4 present our evidence. The final section offers some
concluding comments.
3.2
The Setup
Studies on the determinants of corruption are marked by three characteristics. First, different authors employ different specifications with different
On the Sensitivity of Corruption Determinants
35
variables to result in a model uncertainty of corruption determinants. The
setups of Treisman (2000), Paldam (2002), Ali and Isse (2003), and Park
(2003), for example, are not the same even though they all search for a common underlying factors of corruption. Second, it is also quite common that
investigators limit their attention to one or a small number of variables of interest. For instance, Graeff and Mehlkop (2003) investigate the link between
economic freedom and corruption. Likewise, Brunetti and Weder (2003) inspect the relationship between press freedom and corruption, while Swamy
et al. (2001) examine the link between corruption and gender. Finally, to
test for the reliability of their variable of interest, researchers usually vary
their specifications and examine how sensitive this particular variable is to
a certain range of control variables. However, most of the time, the varying
specifications are still limited to few variables and a narrow range of variation in model specification. Hence, one may question, how much confidence
should be given to the conclusions of previous studies.
Using the SA, we consider a large number of variables that have been
claimed to be related to corruption in previous studies. We prefer this tool
of analysis because it offers a systematic way to scrutinize what variables
are robustly related to corruption. It also allows us to explore a wide range
of possible model specifications. The SA departs from a simple setup as
follow:
C = αj + β f j F + βxj X + β zj Z + (3.1)
where C is some corruption index, F is a vector of ‘fixed’ explanatory
variables that are always included in the regression, but which may also
be zero; X is the variable of interest; Z is a vector of combinations of
up-to-M possible additional explanatory variables drawn from a pool of
variables which, according to the literature, may be related to the dependent
variable; and is an error term. It should be noted, however, that the
number of variables in F and Z that can be plugged into the model is
constrained by the degree of freedoms of the regression as well as by potential
36
Chapter 3
multicollinearity problems. Levine and Renelt (1992) include three variables
in F and all possible combinations of up-to-three variables in Z .
For each model j, X is explored using all possible linear combinations
of Z given F . The ‘robust’ correlation between X and C might be found
via Extreme Bound Analysis (EBA) of Leamer (1983, 1985) and Levine and
Renelt (1992) by constructing the upper and lower bounds for βxj defined
as
max
min βxj
± 2σβ2xj .
(3.2)
If within this range βxj does not change sign, the variable is considered
robust. If one finds that the sign of βxj changes within this range, the
corresponding variable is regarded as ‘fragile’.
Certainly, this approach is not without cost. As the bounds are really
extreme, Sala-i-Martin (1997) argues that this test is too strong for any
variable to pass it, thus “. . . nothing can be learned . . . ” (p. 179) from
such an approach. He suggests analyzing the entire distribution of the
estimates of the parameter of interest (βxj ) and examining the fraction of
the cumulative density function (CDF) lying on each side of zero.2 As zero
divides the area under the CDF into two, the larger of the two areas is
regarded as CDF(0), no matter whether it is above or below zero. Thus,
the CDF(0) is always a number between 0.5 and 1.0.
We use equation 3.1, but due to the absence of a commonly-agreed
theoretical framework, we set F to be zero. We experiment with a series of
models where Z contains several variables (m = 3, 4, 5) drawn from a pool
2
Sala-i-Martin et al. (2004) also propose another technique called Bayesian Averaging of
Classical Estimates (BACE) to check the robustness of different explanatory variables in
growth regressions. This approach builds upon the approach as suggested by Sala-i-Martin
(1997), in the sense that different specifications are estimated to check the sensitivity of the
coefficient estimate of the variable of interest. The major innovation of BACE as compared
to the Sala-i-Martin’s approach is that there is no set of fixed variables included and the
number of explanatory variables in the specifications is flexible. The biggest disadvantages
of the BACE approach are the need of having a balanced dataset, i.e., an equal number
of observations for all regressions (due to the chosen weighting scheme), the restriction of
limiting the list of potential variables to be less than the number of observations and the
computational burden.
On the Sensitivity of Corruption Determinants
37
of 45 selected variables. To check the robustness of a variable, we advocate
three decision rules. First, at least 95 per cent of the CDF(0) lies on one
side of zero; or simply CDF(0)> 0.95. This is different from Sala-i-Martin
who sets the CDF(0)> 0.90. We consider his benchmark too low given the
one-sidedness of the test. In addition, following Sturm and de Haan (2005),
for a variable to be considered robust, it should be significant at the 5 per
cent level in at least 90 per cent of all regressions. Finally, a variable must
pass the previous two benchmarks for three of the five corruption indexes
used in the regressions. To have a complete picture, however, we also show
the outcomes under the approaches suggested by Levine and Renelt as well
as Sala-i-Martin.
Two issues, however, remain, namely multicolinearity and simultaneity.
To deal with the former, we apply the following strategy. First, we exclude
individual variables produced by particular sources if they have been included into another variable by other sources. For example, variables such
as the Polity IV index, political rights, civil liberty, press freedom, freedom
of speech, have been dropped, because Kaufmann et al. (2007) have integrated them into a new variable called ‘voice and accountability’. The same
applies to the other variables of Kaufmann et al., representing regulatory
quality, government effectiveness, rule of law, and political stability.
Second, we chose a variable with the highest number of observations
in case we have more than one variable at hand proxying the same concept. For instance, we use the index of economic freedom of the Heritage
Foundation instead of that of the Fraser institute in view of its coverage of
countries. Third, for some cases we construct indexes for variables explaining the same concept using Principal Component Analysis (PCA). This is
done for variables capturing, decentralization (for instance, expenditures
and revenues of sub-national governments), human capital (like schooling
levels and literacy rate), income inequality (like the Gini coefficient, 10-20
per cent rich population), and women’s participation (women in economic,
social, and political arenas, female labor force).
Under this strategy, we have reduced the number of variables to be
38
Chapter 3
considered, from originally around 75 to 45 variables (Appendix 2). These
variables can be classified into four broad categories: (1) economic and
demographic factors, (2) political institutions, (3) judicial and bureaucratic
environment, and (4) geographic and cultural variables.
Meanwhile, to minimize simultaneity problems, corruption is measured
in 2005-2006, while the determinants refer to 2000.3 This also implies that
we allow a substantial time for the determinants to have effect on corruption.
We consider a set of corruption indicators as the dependent variable.
These variables differ along two dimensions: the methodological construction of the indexes and the number of observations available. The first two
indexes—i.e., the WB (scaled from −2.5 to +2.5) and the TI (0-10) corruption indexes, with 201 and 168 observations, respectively—are aggregated
indexes based on a variety of individual sources (poll-of-polls indexes). We
use also individual indexes including the ICRG (0-6) and the IMD (1-10)
indexes, with 140 and 53 observations, respectively. In addition, we exploit
PCA to generate an aggregated index that originates from seven indicators of the WEF (scale of 1-7) with 117 observations.4 The scales of these
indexes are adjusted so that a higher score means less corruption.
3.3
Some First Results
Table 3.1 displays the results of our robustness tests in case F is set to be
zero, while up-to-three variables are included in Z drawn from a pool of 45
variables5 using five different corruption indexes as dependent variable. In
columns 2-4, we report the two extreme bounds and the fraction of signifi3
Economic growth, however, is measured as the average of the 1990-2000 values, while
the income distribution is measured over 1993-2000.
4
The PCA produces only one component with a high eigenvalue (6.11) that accounts for
87.25 per cent of the combined variance. The ‘scoring coefficients’ of the indicators range
from 0.34 (favoritism) to 0.39 (irregular payments in public contracts), while business
costs of corruption as well irregular payments in public utilities, judicial decisions, exportimport, and in tax collection have ‘scoring coefficients’ of 0.38.
5
Under this setup, for each variable, we run (45−1)!
= 13, 244 regressions, or a total of
3!41!
595,980 regressions for all variables.
On the Sensitivity of Corruption Determinants
39
cant cases; in columns 5-7, the unweighted CDF(0)6 , the estimate of β, and
its standard deviation (σ 2 ) are displayed. To save space, we only report the
variables for which CDF(0)> 0.95.
It follows from Table 3.1 that there are 4-6 variables passing the first rule
(i.e., CDF(0)>0.95). However, only half of them pass the second and third
rules (90 per cent cases significant at the 5 per cent level and passing the
two banchmarks in three of five corruption indexes). The robust variables
are government effectiveness and rule of law. The CDF(0) of government
effectiveness is very close to one, while in almost all specifications government effectiveness has a significant impact on corruption. For the rule of
law variable, the CDF(0) ranges between 0.98-1.00, while the fraction of
the regressions in which the coefficient of this variable is significant ranges
between 93.92 and 99.98 per cent. Only in the case where the IMD index is
the dependent variable, is the rule of law variable slightly below the second
benchmark.7
Government effectiveness captures “the quality of public service provision, the quality of the bureaucracy, the competence of civil servants, the
independence of the civil service from political pressures, and the credibility
of the government’s commitment to policies”. Meanwhile, the rule of law
variable denotes mainly “the extent to which agents have confidence in and
abide by the rules of society” and “perceptions of the incidence of crime,
the effectiveness and predictability of the judiciary, and the enforceability
of contracts.” Therefore, rule of law measures “the success of a society in
developing an environment in which fair and predictable rules form the basis
for economic and social interactions, and importantly, the extent to which
property rights are protected” (Kaufmann and Kraay, 2002: 177-178). This
may explain why the two variables are robustly correlated with corruption.
6
Sala-i-Martin (1997) argues that the unweighted average is superior to the weighted
average in a situation with a spurious fit.
7
Notice that the IMD index covers only 53 countries, which is about one-fourth of the
WB country coverage.
-0.419
-0.245
-0.262
-0.540
Ln GDP per capita
Presidentialism
Aboslute Latitude
Voice-Accountability
-1.097
-0.662
Rule of Law
Ln GDP per capita
Aboslute Latitude
-0.684
-1.073
-0.166
-2.002
Government Effectiveness
Rule of Law
Polarization
Wage Bill per GDP
Corruption: ICRG (N = 140)
0.351
-1.424
Government Effectiveness
Corruption: TI (N = 168)
-0.326
Rule of Law
0.576
0.921
1.934
2.213
2.850
4.794
3.453
4.073
2.009
1.252
0.998
2.103
1.366
1.615
Bound
Bound
0.037
Upper
Lower
Government Effectiveness
Corruption: WB (N = 201)
Variable
79.047
96.957
98.671
99.071
82.520
90.939
99.894
100.000
87.149
81.297
83.902
90.562
99.977
100.000
Signif.
Fraction
Levine-Renelt
0.952
0.994
0.997
0.999
0.962
0.980
1.000
1.000
0.954
0.962
0.962
0.978
1.000
1.000
CDF(0)
-0.273
0.354
0.883
0.933
0.648
1.227
1.854
2.009
0.602
0.290
0.290
0.578
0.894
0.930
β
Unweigt.
0.116
0.090
0.108
0.111
0.159
0.150
0.125
0.122
0.069
0.070
0.070
0.067
0.051
0.048
σ2
Unweigt.
continued on next page . . .
Unweigt.
Sala-i-Martin
Table 3.1: Robustness Analysis (No F , Z up-to-3)
40
Chapter 3
-5.268
-3.067
-6.074
-1.723
Rule of Law
Protestant Fraction
Population
Economic Freedom
-0.628
-1.764
-2.923
-2.824
Government Effectiveness
Ln GDP per capita
Rule of Law
Regulatory Quality
Corruption: WEF (N = 117)
-1.610
Regulatory Quality
5.001
3.924
4.285
5.785
4.752
4.058
5.710
6.200
7.415
9.398
Bound
Bound
-3.736
Upper
Lower
Government Effectiveness
Corruption: IMD (N = 53)
Variable
88.206
97.584
99.071
99.985
88.183
77.076
91.000
93.922
95.417
99.343
Signif.
Fraction
Levine-Renelt
0.954
0.996
0.997
1.000
0.958
0.959
0.983
0.984
0.992
0.998
CDF(0)
Unweigt.
2.087
2.002
1.679
2.254
1.562
-1.075
0.965
2.391
2.852
2.681
β
Unweigt.
Sala-i-Martin
0.278
0.208
0.206
0.204
0.397
0.363
0.262
0.329
0.329
0.295
σ2
Unweigt.
On the Sensitivity of Corruption Determinants
41
42
Chapter 3
There is also another variable passing the three benchmarks: income.
Income passes the first two benchmarks as its CDF(0) is about 0.98-0.99
and the lowest fraction significant is 91 per cent. The two benchmarks are
achieved when corruption index is WB, TI, and WEF—the three aggregated
indexes. Hence, when F is set to be zero and Z is set up-to-three variables,
income appears as a robust variable according to the three rules.
The other variables do not consistently pass the three yardsticks. Regulatory quality, for example, only passes the tests when corruption is proxied
by the IMD index; the same holds for Protestantism. Absolute latitude
has a CDF(0) above 0.95 under the use of the WB and TI indexes, but
the fraction of significant regressions is still far from the second benchmark.
Democracy—captured by voice and accountability—is significant only in 87
per cent of the regressions, although it passes the CDF test for the use of
WB index. Economic freedom has similar figures.
Keeping F =0, we now experiment with Z containing up-to-four and
up-to-five variables8 where those passing the first yardstick are presented
in Table 3.2. The results reinforce our previous findings: government effectiveness and rule of law turn out to be robust variables. The CDF(0)
statistics of these determinants are very close to 1.00. At the same time,
we find that in 96-100 per cent of the regressions the impact of government
effectiveness is significant, while the corresponding figure for the rule of law
variable is about 96-99 per cent.9 Compared to the previous results, also
the standardized impact is very similar. Nevertheless, the results for income
are now different. It passes the first two benchmarks only when the WEF
index is used, both under up-to-four and five variables in Z . For the other
corruption indexes, income does not pass the tests. The other variables
cannot be regarded as robust variables according to the three rules.
For each variable, we run respectively (45−1)!
= 135, 751 and 1,086,008 regressions, or
4!40!
a total of 6,108,795 and 48,870,360 regressions for all variables.
9
Under the IMD index and Z containing up-to-five variables, the fraction of significant
regressions of rule of law is only slightly below the yardstick: 89.6 per cent.
8
86.465
Ln GDP per Capita
99.622
87.930
Rule of Law
Ln GDP per Capita
97.778
93.812
Rule of Law
Political Polarization
98.701
93.627
91.929
85.291
Government Effectiveness
Regulatory Quality
Rule of Law
Protestant Fraction
Corruption: IMD
97.761
Government Effectiveness
Corruption: ICRG
100.000
Government Effectiveness
Corruption: TI
99.912
Rule of Law
0.970
0.973
0.988
0.996
0.988
0.996
0.996
0.970
0.999
1.000
0.963
1.000
1.000
CDF(0)
Signif.
100.000
Unweigt.
Fraction
Government Effectiveness
Corruption: WB
Variable
0.901
2.303
2.758
2.665
0.305
0.864
0.918
1.117
1.805
1.999
0.523
0.878
0.924
β
Unweigt.
F =0, Z up-to-4
0.954
0.960
0.984
0.993
0.982
0.994
0.993
0.958
0.997
1.000
0.999
1.000
CDF(0)
Unweigt.
0.852
2.214
2.684
2.650
0.268
0.846
0.901
1.018
1.751
1.986
0.860
0.918
β
Unweigt.
continued on next page . . .
79.823
89.591
91.929
97.780
89.867
96.712
95.826
84.734
99.174
99.999
99.782
100.000
Signif.
Fraction
F =0, Z up-to-5
Table 3.2: Robustness Analysis (F =0, Z up-to-4 and up-to-5)
On the Sensitivity of Corruption Determinants
43
97.626
96.032
Ln GDP per Capita
Rule of Law
0.992
0.992
1.000
CDF(0)
Signif.
99.890
Unweigt.
Fraction
Government Effectiveness
Corruption: WEF
Variable
1.930
1.552
2.231
β
Unweigt.
F =0, Z up-to-4
94.281
95.398
99.629
Signif.
Fraction
0.988
0.987
0.999
CDF(0)
Unweigt.
1.861
1.438
2.208
β
Unweigt.
F =0, Z up-to-5
44
Chapter 3
On the Sensitivity of Corruption Determinants
45
Having two robust variables at hand, we treat government effectiveness
and rule of law as the fixed variables (F ) and rerun Model 3.1. Applying
the same decision rules, we do not find any variable that consistently has a
robust relationship to corruption. Some variables are able to pass the CDF
test, but most of them fail to pass the second one. None passes the third
test. In the following we narratively report the results.
Using the WB index, we find that population, fraction of population
belong to Hindu, GDP per capita growth, plurality, government expenditure, and fraction of Buddhists pass the first yardstick, but only population
can pass the second benchmark. This is different from the results drawn
from the TI index; here, the fraction of Hindu, voice and accountability,
and population are able to pass the first test, but only the first can pass the
second yardstick. Also, a different result is found when we use the ICRG
index as the dependent variable. Now, debt, wage, and polarization pass
the first test, while none but debt passes the second benchmark. Meanwhile, variables passing the first yardstick under the IMD index are voice
and accountability, fraction of Catholic, export of ores and metal, and human capital. Yet, none passes the second test. Finally, under the WEF
index, three variables—voice and accountability, fraction of Catholic, and
export of ores and metal—pass the first test, but none passes the second.
3.4
Effect of Observations
Up to this point, Z includes all observations (N ). Now we turn to an
experiment where N is varied. In this experiment, we order the observations
according to their corruption scores and estimate equation 3.1 employing
N = 100, 125, 150, 175 observations, or respectively 50, 62.5, 75, and 87.5
per cent of the total observations. We also compare these results if the
corruption scores are ordered from low to high. In the first ordering the first
50 per cent of the observations are dominated by corrupt nations, while the
reverse applies to the second ordering. In this part of the analysis, we use
only the WB index as it covers almost all countries over the world. The
46
Chapter 3
results are reported in Table 3.3 where Z is up to three variables.
As follows from Table 3.3, government effectiveness and rule of law again
appear as the robust variables correlated with corruption since they consistently pass the three benchmarks regardless of the number of observations.
Even, when the observations are reduced until 50 per cent, the CDF(0) and
fraction of significant are still far above the benchmark. The performance
of these variables is stable under both orderings. The other variables do
not consistently pass the tests; this holds true even for income and democracy (proxied by voice and accountability) that are commonly argued to be
significant in explaining corruption. Such variables are sensitive not only
to the existence of other (control) variables, but also to the number and
composition of the set of observations.
We again examine the sensitivity of our findings using up-to four and
five variables in the Z vector (Tables 3.4-3.5). Apart from government
effectiveness and rule of law, no variable is found to be robust according
to the three decision rules. Different Z , N , or ordering procedure do not
change the conclusion that these variables are robust. One, however, may
question why government effectiveness and rule of law are always robust in
their correlation with corruption in all circumstances. Figure 3.1 clarifies
it. As the plots scatter around their ‘means’, clearly no influencing outlier
is found in the figure. The figure shows that none of the nations with low
quality of government effectiveness and rule of law appears as a corruptionfree nation. Likewise, those regarded as clean countries always perform high
quality of government effectiveness and rule of law. This fact holds up using
a variety of measures of corruption.
Furthermore, there are three interesting features displayed in Tables 3.13.5 with respect to β. First, the CDF(0) increases as N increases, regardless
of the way of ordering. The same applies to fraction of significant regressions. This confirms the classical regression issue: a bigger sample gives
more convincing results. Second, the impact of government effectiveness is
always higher in all circumstances compared to that of rule of law. The
same holds true for both corrupt (the first 50 per cent ascending order) and
On the Sensitivity of Corruption Determinants
47
clean (the first 50 per cent descending order) countries.
Finally, the regressions based on the ascending-order produce much
lower β in the first 50 per cent of the observations, than the descendingorder regressions do. The coefficient of government effectiveness is about
0.4-0.5 under the former, and 0.8-0.9 under the latter. Similar results are
found for the rule of law variable. This implies that the impact on corruption of government effectiveness and rule of law are considerably different
in corrupt and non-corrupt regimes. In corrupt regimes the impact is about
50 per cent lower than in clean regimes, reflecting heterogeneity in the size
of the impact of corruption determinants.
Dependent Variable
(Ind. Var: WB Index)
N = 100
Government Effectiveness
Rule of Law
Regulatory Quality
Voice and Accountability
Political Stability
N = 125
Government Effectiveness
Rule of Law
Voice and Accountability
Regulatory Quality
Political Stability
Absolute Latitude
N = 150
Government Effectiveness
Rule of Law
GDP per Capita
Voice and Accountability
Area
Regulatory Quality
0.998
0.995
0.994
0.979
1.000
0.999
0.988
0.970
1.000
1.000
0.982
0.978
0.965
0.957
99.517
97.357
96.338
90.818
99.977
99.607
95.296
92.706
100.000
99.902
92.057
92.147
79.991
92.532
0.727
0.672
0.325
0.379
-0.173
0.482
0.593
0.538
0.309
0.353
0.464
0.436
0.290
0.229
Z=up-to-3; Ascending Order
% Sign. CDF(0) Beta-U
0.918
0.872
0.579
0.577
0.287
0.895
0.865
0.757
0.634
0.872
0.869
continued on next page . . .
1.000
1.000
0.971
0.965
0.954
90.335
79.364
99.992
99.962
89.671
1.000
0.999
0.951
0.983
90.169
92.170
99.841
99.857
1.000
0.997
99.706
98.014
Z=up-to-3; Descending Order
% Sign. CDF(0)
Beta-U
Table 3.3: Robustness Analysis (F =0, Z up-to-3, Various N )
48
Chapter 3
Dependent Variable
(Ind. Var: WB Index)
Foreign Debt
Presidentialism
Absolute Latitude
Political Stability
N = 175
Government Effectiveness
Rule of Law
Area
GDP per Capita
Voice and Accountability
Regulatory Quality
Presidentialism
Ethnic Division
Foreign Debt
0.973
0.985
0.955
0.954
90.645
92.683
78.232
81.931
0.316
-0.195
-0.139
0.565
0.919
0.868
1.000
1.000
100.000
99.977
0.810
0.761
-0.263
0.399
0.444
0.567
100.000
99.947
92.578
92.306
90.788
91.838
1.000
1.000
0.985
0.984
0.974
0.953
Z=up-to-3; Descending Order
% Sign. CDF(0)
Beta-U
86.477
0.968
-0.150
83.268
0.966
0.255
80.301
0.958
0.299
89.037
0.955
0.590
Z=up-to-3; Ascending Order
% Sign. CDF(0) Beta-U
On the Sensitivity of Corruption Determinants
49
50
Chapter 3
10
6
2
-2
WB
TI
ICRG
IMD
WEF
-6
-2.5
-1.5
-0.5
0.5
1.5
2.5
(a) Government Effectiveness
10
6
2
-2
WB
TI
ICRG
IMD
WEF
-6
-2.5
-1.5
-0.5
0.5
1.5
2.5
(b) Rule of Law
Figure 3.1: Corruption (y-axis) and Two Robust Determinants (x-axis)
Dependent Variable
(Ind. Var: WB Index)
N = 100
Government Effectiveness
Regulatory Quality
Rule of Law
Voice and Accountability
Political Stability
N = 125
Government Effectiveness
Rule of Law
Voice and Accountability
Regulatory Quality
Political Stability
N = 150
Government Effectiveness
Rule of Law
GDP per Capita
Voice and Accountability
Foreign Debt
N = 175
Government Effectiveness
0.994
0.988
0.987
0.957
1.000
0.999
0.976
0.962
1.000
0.999
0.972
0.962
1.000
98.703
93.882
95.571
84.007
99.888
99.470
90.871
90.289
99.993
99.719
88.520
86.947
99.998
0.806
0.727
0.668
0.303
0.359
0.594
0.537
0.294
0.349
0.459
0.285
0.434
0.210
Z=up-to-4; Ascending Order
% Sign. CDF(0) Beta-U
0.914
-0.145
0.914
0.851
0.530
0.522
0.894
0.851
0.571
0.872
0.858
continued on next page . . .
1.000
0.950
81.840
100.000
1.000
0.999
0.952
0.951
86.044
99.987
99.865
85.563
1.000
0.999
0.972
88.293
99.805
99.623
0.999
0.996
99.472
97.613
Z=up-to-4; Descending Order
% Sign. CDF(0)
Beta-U
Table 3.4: Robustness Analysis (F =0, Z up-to-4, Various N )
On the Sensitivity of Corruption Determinants
51
Dependent Variable
(Ind. Var: WB Index)
Rule of Law
GDP per Capita
Area
Voice and Accountability
Presidentialism
Z=up-to-4; Ascending Order
% Sign. CDF(0) Beta-U
99.801
1.000
0.750
88.505
0.973
0.370
85.358
0.970
-0.232
85.261
0.956
0.415
86.228
0.970
0.264
Z=up-to-4; Descending Order
% Sign. CDF(0)
Beta-U
99.907
1.000
0.850
86.863
0.958
0.515
52
Chapter 3
Dependent Variable
(Ind. Var: WB Index)
N = 100
Government Effectiveness
Regulatory Quality
Rule of Law
Political Stability
N = 125
Government Effectiveness
Rule of Law
Voice and Accountability
Regulatory Quality
N = 150
Government Effectiveness
Rule of Law
GDP per Capita
N = 175
Government Effectiveness
Rule of Law
GDP per Capita
Presidentialism
0.986
0.977
0.972
0.999
0.998
0.962
0.955
1.000
0.999
0.959
1.000
0.999
0.959
97.187
90.805
93.426
99.713
99.079
85.972
87.945
99.970
99.441
84.697
99.991
99.555
84.457
0.803
0.738
0.344
0.726
0.662
0.284
0.594
0.535
0.281
0.347
0.438
0.279
0.433
Z=up-to-5; Ascending Order
% Sign. CDF(0) Beta-U
1.000
0.999
0.950
78.751
1.000
0.998
100.000
99.763
99.985
99.675
1.000
0.997
0.993
0.957
96.744
83.815
99.762
99.248
0.999
99.166
0.221
0.909
0.830
0.909
0.828
0.893
0.835
0.845
0.514
0.871
Z=up-to-5; Descending Order
% Sign. CDF(0)
Beta-U
Table 3.5: Robustness Analysis (F =0, Z up-to-5, Various N )
On the Sensitivity of Corruption Determinants
53
54
Chapter 3
3.5
Concluding Remarks
Over the last two decades, the number of empirical studies on the determinants of corruption has rapidly increased. These studies employ a variety of
models using various economic and non-economic variables. Yet, there is no
‘true’ model due to the absence of an encompassing theory on the determinants of corruption. Still, many variables have been claimed to be significant
in explaining cross-country variation in corruption. Since the ‘true’ model
is far from known, the question which variables are really correlated with
corruption remains.
The literature offers various approaches in dealing with this issue including the EBA of Levine and Renelt (1992) and the SA of Sala-i-Martin
(1997). Slightly modifying these approaches, we advocate three sequential
decision rules to label a variable of interest ‘robust’ or ‘fragile’. We use 45
variables and search all possible linear combinations to check their robustness. We also experiment with up-to-three, four, and five variables in Z but
none in F .
We find that two variables that are consistently passing the three tests.
The two variables are government effectiveness and rule of law, drawn from
the governance dataset of Kaufmann et al. (2007). Government effectiveness
and rule of law repeatedly appear robust since their CDF(0) is always above
0.95 and in more than 90 per cent of the regressions these variables are found
to be significant at the 0.05 level. Also, the robustness of these variable does
not change if we use different corruption indexes. Two graphical expositions
demonstrating the closeness of the scattered data to their ‘means’ support
this finding. The other variables, however, are either found to pass the test
erratically or not pass the test at all across the use of different dependent
variables.
On the basis of these findings, we conclude that corruption is always
lower in countries governed by high quality officials and where the rule of
law is strongly enforced. This conclusion is not sensitive to the change in
the number and composition of observations as well as the ordering of ob-
On the Sensitivity of Corruption Determinants
55
servations. Also, these two variables always appear robust regardless of the
types of corruption index used in the regressions. The other variables do
not consistently pass the tests when different corruption indexes are employed, even though these indexes are highly correlated. In sum, of a bunch
of determinants claimed significant in explaining cross-country variation in
corruption, only government effectiveness and rule of law can be regarded
as robust variables.
Chapter 4
Is Corruption Really
Persistent?
“Change does not necessarily assure progress,
but progress implacably requires change.”
————————————————————
Henry Steele Commager (1902–1998)
4.1
Introduction
There is strong evidence that corruption has detrimental effects on determinants of economic growth, like investment and trade (see, for instance,
Wei (2000b), Anderson and Marcouiller (2002)). It is also widely believed
that, despite these negative consequences, corruption is highly persistent.
Theoretical models of corruption generally predict that corruption is immutable. Tirole (1996), for example, argues that once agents are corrupt,
they remain so due to reputation effects. Similarly, Mishra (2006) indicates
that when there are many corrupt individuals, it is optimal to be corrupt;
thus once corruption is widespread, it will be persistent. This is in line
with the view of Mauro (2004) that in countries where corruption is pervasive, there are no incentives for individuals to fight corruption, even though
Joint work with Jakob de Haan; the earlier version was presented at the First World
Public Choice Meetings 2007. We received enlightening comments from Tom Wansbeek.
58
Chapter 4
they would be better off if corruption were eradicated. Finally, Basu et al.
(1992) show that collusion of agents in a hierarchical structure also leads to
corruption persistence.1
Also on the basis of empirical research, some authors claim that corruption is persistent. Herzfeld and Weiss (2003), for instance, report that
the absence of law and order increases corruption, while more corruption
lowers enforcement. Thus, once trapped, it is difficult for a country with a
weak legal system to escape from corruption. This is in line with the view
of Jain (2001: 72), that “Once corrupted, the elite will attempt to reduce
the effectiveness of the legal and judicial systems through manipulation of
resource allocation and appointments to key positions. Reduced resources
will make it difficult for the legal system to combat corruption, thus allowing
corruption to spread even more.”2
However, we argue that corruption changes over time. Using the International Country Risk Guide (ICRG) data for the period 1984-2003, we find
strong evidence that corruption is not persistent. Many corrupt countries
were able to reduce their level of corruption, while many clean countries
have become less clean over the same period. Apart from some standard
regressions, we apply non-parametric methods to examine the distributional
dynamics of corruption in 101 countries. Specifically, we evaluate the modality of corruption data across countries and over time using a kernel density
function. In addition, we apply Markov chain analysis. All our results reject
the hypothesis that corruption is persistent.
The remainder of the chapter is organized as follows. Section 4.2 provides evidence on corruption convergence based on correlation. Section 4.3
discusses a Gaussian kernel density function, while section 4.4 presents a
Markov chain analysis. Section 4.5 concludes.
1
Other relevant theoretical papers on the persistence of corruption include Blackburn et
al. (2006), Damania et al. (2004) and Ventelou (2002). See also Bardhan (1997, especially
p. 1330-1334), for a survey of corruption persistence.
2
Also Alesina and Weder (2002) and Damania et al. (2004) argue on empirical grounds
that corruption appears to be persistent.
Is Corruption Really Persistent?
4.2
59
Convergence: Preliminary Evidence
We use data on corruption from ICRG covering 101 countries around the
world.3 These are the countries for which the ICRG corruption indicator
is available for all years in the sample 1984-2003. The ICRG data is based
on perceived corruption by a panel of experts. The level of corruption is
expressed on a scale between zero and six, but for the current analysis
the index is rescaled from one to seven where a higher score means less
corruption.4 We use the ICRG data because this is the only indicator of
corruption that is available for a long time period on a consistent basis.5
The ICRG data suggests that corruption changes over time. In some
countries corruption increases, while in others it declines. Table 4.1 displays
losers and winners in the world corruption league in which countries are
ranked according to their relative change in the level of corruption over
1984-2003. The first panel shows the losers, i.e., countries with an increase
in corruption between 1984-2003. Zimbabwe has the highest relative change
in this group. It drops from a relatively moderate level (3.92) to the lowest
level (1.00). The change in corruption does not only occur in countries
categorized as moderately or highly corrupt in 1984, but also in countries
that were initially labeled as clean. Canada, Singapore, Switzerland, the
United Kingdom, and Iceland fall into this category.
The second panel shows countries with no change in the level of corruption over the same period. There are 12 countries listed in this panel: they
come from almost all ranks in the corruption league in 1984. For instance,
Indonesia (one of the most corrupt countries in 1984), Jordan, and Colombia (from the mid ranks), as well as Austria (one of the cleanest countries)
3
http://www.prsgroup.com/ICRG Methodology.aspx.
The rescaling is required as we measure the relative change in corruption as discussed
later. The annual ICRG data is constructed on the basis of averaged monthly data. The
monthly data is measured on an ordinal scale, but the averaging arguably transforms the
data into continuous data. As we will show later, treating the data as ordinal does not
affect our main conclusions.
5
Other measures are problematic. The corruption index of the World Bank, for example, is only available for recent years, while the index of Transparency International is
constructed using a non-consistent methodology. See Section 2.2 for further details.
4
60
Chapter 4
are all in this group. Finland is also in this group, but this country has
reached the upper bound thus an increase in the ICRG index is not possible. As in the first panel, both developed and developing countries are in
this panel. This part of the table also shows that corruption persistence is
not a widespread phenomenon, as corruption remained at the same level in
a limited number of countries only.
Finally, the third panel shows the winners, i.e., countries with a declining
level of corruption over 20 years. There are 22 countries in this group; most
of them were in the low and mid positions in the corruption ranking in 1984.
Examples are Iraq, with the smallest positive relative change, as well as the
Philippines that has the biggest improvement. Also some African countries
saw their level of corruption decline.
To investigate more precisely how corruption has changed over time,
we calculate the correlation between the level of corruption in 1984—the
starting point of our sample period—and corruption in subsequent years.
In case of corruption persistence, the correlation coefficients should be stable
over time (ρ-persistence). Similarly, we also run a series of autoregressions
of the level of corruption on its initial level. If corruption is persistent, the
constant and the slope should be zero and one, respectively.
The first columns in Table 4.2 display the relationship between corruption in 1984 and corruption in subsequent years using Pearson, Spearman,
and polychoric correlation coefficients.6 The correlation between corruption in 1984 and corruption 1-4 years later is high, indicating that change
in corruption is not a short-term phenomenon. However, as the time-lag i
increases, the correlation coefficients gradually decrease to reach the lowest
level at time-lag i = 15.
6
Polychoric correlation is mostly used for ordinal variables (Olsson, 1979). Aish and
Jöreskog (1990) use this measure to compute the correlation among a set of perceptionbased indicators of political efficacy and responsiveness at two points in time (1974 and
1981).
Change
Country
Change
Country
-0.745
-0.524
-0.500
-0.500
-0.429
-0.417
-0.417
-0.417
-0.417
-0.400
-0.400
-0.400
-0.375
-0.375
-0.375
-0.354
-0.333
Zimbabwe
S. Africa
Lebanon
PNG
France
Brunei
Costa Rica
Malaysia
China
Albania
Bulgaria
Malawi
Algeria
India
Thailand
Ser.-Mont.
Gabon
Nigeria
Mexico
Libya
Kuwait
Japan
Ireland
Iran
Dom. Rep.
Bahrain
Angola
Argentina
Venezuela
Paraguay
Belgium
Italy
S. Arabia
Myanmar
-0.250
-0.250
-0.250
-0.250
-0.250
-0.250
-0.250
-0.250
-0.250
-0.250
-0.268
-0.268
-0.273
-0.286
-0.300
-0.308
-0.333
Jamaica
Hong Kong
Guatemala
Sudan
Spain
Taiwan
Hungary
Iceland
UK
Switzer.
Singapore
Canada
USA
Israel
UAE
Tunisia
Poland
Panel 1. Negative Change (Mean: -0.257; N =67)
Country
-0.167
-0.167
-0.167
-0.172
-0.182
-0.200
-0.200
-0.202
-0.214
-0.214
-0.214
-0.214
-0.221
-0.236
-0.250
-0.250
-0.250
Egypt
S. Korea
Peru
Sweden
NZ
Denmark
Australia
Senegal
Nicaragua
Greece
Norway
Nether.
Chile
Tanzania
Turkey
Togo
Country
-0.063
-0.065
-0.068
-0.071
-0.071
-0.071
-0.083
-0.125
-0.125
-0.125
-0.143
-0.143
-0.144
-0.163
-0.167
-0.167
Change
Continued on next page . . .
Change
Table 4.1: Losers and Winners 1984-2003
Is Corruption Really Persistent?
61
Change
Country
Change
0.000
0.000
Cameroon
Colombia
Guinea
Finland
Ecuador
0.000
0.000
0.000
0.059
0.097
0.132
0.167
0.167
Portugal
Romania
Brazil
El Salvador
Honduras
Ghana
Morocco
Pakistan
Zambia
Syria
Ivory Coast
0.355
0.333
0.250
0.241
0.241
0.231
The change is measured relative to the 1984 level
0.045
Iraq
Panel 3. Positive Change (Mean: 0.596; N =22)
0.000
Austria
Panel 2. Zero Change (Mean: 0.000; N =12)
Country
Bangladesh
Uganda
Kenya
Liberia
Bolivia
Panama
Jordan
Indonesia
Country
1.000
0.500
0.486
0.458
0.385
0.000
0.000
0.000
Change
Philippines
Mali
Haiti
Guyana
Congo DR
Uruguay
Trin.-Tob.
Sri Lanka
Country
1.917
1.000
1.000
1.000
1.000
0.000
0.000
0.000
Change
62
Chapter 4
Is Corruption Really Persistent?
63
The Pearson coefficient reduces from 0.98 to 0.74, the Spearman coefficient drops from more than 0.99 to 0.68, while the polychoric correlation
coefficient declines from 0.96 to 0.64, suggesting less than full persistence
(no ρ-persistence). Similar results are found for any other initial value of
corruption. Likewise, in the autoregressions the R2 drops from more than
0.96 to only 0.55, suggesting that the initial level of corruption loses much
of its power to explain the variation in the current levels of corruption.
The table also shows that the constant (τ ) and the slope (ω) of the autoregressions move away from zero and one, respectively; ω exhibits the same
pattern as the three ρ’s and R2 . This conclusion is supported by the result
of the joint test that τ = 0 and ω = 1 in the last column Table 4.2. In sum:
our evidence does not support corruption persistence.
Table 4.2: Corruption Correlation Over Time
i
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
ρ1
ρ2
ρ3
0.979
0.960
0.941
0.920
0.887
0.866
0.864
0.829
0.804
0.794
0.769
0.766
0.693
0.643
0.606
0.641
0.665
0.749
0.740
0.996
0.970
0.960
0.952
0.912
0.892
0.872
0.851
0.813
0.796
0.777
0.746
0.669
0.641
0.587
0.622
0.620
0.685
0.680
0.965
0.950
0.938
0.918
0.875
0.848
0.833
0.803
0.771
0.747
0.734
0.717
0.630
0.602
0.522
0.567
0.573
0.630
0.639
C1984+i = τ + ωC1984
R
τ
ω
F ∗]
0.958 0.272 0.959
6.660
0.921 0.400 0.928
5.880
0.886 0.586 0.884
8.370
0.846 0.757 0.849 10.710
0.786 0.819 0.825
8.880
0.749 0.944 0.807
9.950
0.746 1.078 0.786 13.560
0.688 1.388 0.746 20.150
0.647 1.690 0.695 29.170
0.631 1.888 0.649 37.590
0.591 2.018 0.608 39.720
0.587 1.912 0.607 34.300
0.480 2.039 0.542 33.340
0.413 2.001 0.516 30.790
0.368 1.989 0.490 32.480
0.411 1.851 0.510 34.040
0.442 1.635 0.527 38.600
0.561 1.122 0.568 72.540
0.548 1.266 0.551 68.830
2
ρ1 : Pearson, ρ2 : Spearman; ρ3 : Polychoric
∗]
Joint test F : τ =0 and ω=1
64
Chapter 4
As an alternative to the OLS regressions, we also estimate ordered logit
and probit regressions as perhaps not everyone might be convinced that our
corruption data can be treated as cardinal data. Tables 4.3 and 4.4 show
the results of the ordered logit and order probit estimations respectively. It
is clear that the longer the time lag, the lower is the influence of the initial
level of corruption (ω) on its current level. The estimate of ω reduces quickly
until i = 15, from 6.09 and 0.75 in the logit models and from 3.22 to 0.44
in the probit models. A similar pattern is found for the pseudo-R2 . The
ability of the initial level to explain the current level of corruption reduces if
the time lag increases. So, these results are fully in line with those reported
in Table 4.2.
Next, we adapt some terminology from the economic growth literature
(see, for instance, Barro and Sala-i-Martin, 1992, 1995). First, there is corruption convergence if a corrupt country improves more than a clean country, so that the corrupt country catches up with the clean one. This type
of convergence corresponds to the so-called β-convergence in the economic
growth literature.
The second type of convergence occurs if the cross-sectional dispersion
(measured by the cross-country standard deviation and the coefficient of
variation of the level of corruption) declines over time; this type of convergence is akin to σ-convergence in the economic growth literature.
To examine β-convergence, we estimate a simple linear relation between
the relative change in corruption over the period of 1984-2003—i.e., the
change in the level of corruption over 20 years corrected by its initial value—
and the initial level of corruption. We prefer the relative instead of the
absolute change in the level of corruption because one point improvement
(deterioration) for corrupt countries differs substantively from the same improvement (deterioration) for clean countries. In absolute terms, an increase
of the corruption index from 1 to 2, for example, is exactly the same as an
increase from 5 to 6, but it is completely different if the initial level of corruption is taken into account. Still, we also report estimates with the absolute
change in corruption as dependent variable and other possible measures.7
7
The correlations among these measures varies between 0.86 and 0.95.
10
9
8
7
6
5
4
3
2
1
i
ω
6.089
(0.890)
3.963
(0.465)
3.494
(0.411)
3.019
(0.356)
2.322
(0.266)
2.055
(0.238)
1.921
(0.221)
1.729
(0.205)
1.557
(0.193)
1.451
(0.183)
τ1
8.700
(1.935)
5.766
(1.163)
4.945
(1.063)
4.035
(0.956)
3.168
(0.806)
2.605
(0.758)
1.151
(0.892)
0.798
(0.877)
0.617
(0.880)
-0.356
(1.117)
Ordered Logit
τ2
τ3
τ4
13.023 20.102 26.105
(1.919) (2.895) (3.601)
7.905
13.134 17.627
(1.114) (1.627) (1.984)
6.743
11.380 15.654
(1.030) (1.432) (1.776)
5.431
9.637
13.644
(0.923) (1.244) (1.577)
4.223
7.500
10.484
(0.779) (0.971) (1.187)
3.388
6.508
9.136
(0.727) (0.879) (1.057)
3.322
5.997
8.358
(0.704) (0.818) (0.971)
2.796
4.980
7.175
(0.699) (0.745) (0.873)
2.043
3.895
6.404
(0.700) (0.693) (0.825)
1.586
3.360
5.990
(0.708) (0.659) (0.793)
τ6
39.428
(5.856)
25.594
(3.065)
23.490
(2.820)
20.775
(2.464)
15.992
(1.803)
14.216
(1.607)
13.147
(1.475)
11.883
(1.357)
10.801
(1.264)
10.152
(1.193)
0.273
0.288
0.320
0.358
0.382
0.423
0.521
0.579
0.626
Pseudo
R2
0.765
continued on next page . . .
τ5
33.418
(5.014)
21.697
(2.615)
19.199
(2.301)
16.184
(1.890)
12.711
-1.447
10.980
(1.251)
10.263
(1.164)
8.864
(1.020)
8.006
(0.957)
7.823
(0.951)
Table 4.3: Ordered Logit Regression
Is Corruption Really Persistent?
65
ω
1.315
(0.172)
1.247
(0.170)
0.970
(0.146)
0.846
(0.136)
0.754
(0.134)
0.819
(0.135)
0.877
(0.137)
1.086
(0.157)
1.058
(0.156)
τ1
-0.617
(1.113)
-0.669
(1.120)
0.738
(0.623)
0.610
(0.572)
0.490
(0.568)
0.605
(0.553)
-1.732
(1.088)
-1.265
(1.091)
-1.260
(1.099)
Ordered Logit
τ2
τ3
τ4
0.919
2.871
5.594
(0.713) (0.621) (0.760)
0.873
3.091
5.672
(0.728) (0.629) (0.776)
2.674
4.463
6.244
(0.580) (0.669) (0.823)
2.599
4.124
5.628
(0.561) (0.637) (0.751)
2.581
4.027
5.259
(0.575) (0.641) (0.730)
2.701
4.664
5.636
(0.560) (0.676) (0.749)
1.117
3.033
5.010
(0.544) (0.574) (0.698)
2.499
5.243
6.493
(0.595) (0.734) (0.849)
2.449
4.935
6.217
(0.600) (0.715) (0.822)
τ5
7.355
(0.912)
6.979
(0.899)
7.887
(0.966)
7.107
(0.884)
6.542
(0.858)
6.939
(0.874)
6.323
(0.811)
8.222
(1.011)
8.046
(1.001)
7: C > 6.5
3: 2.5 < C ≤ 3.5;
6: 5.5 < C ≤ 6.5;
2: 1.5 < C ≤ 2.5;
5: 4.5 < C ≤ 5.5;
1: C ≤ 1.5;
7.545
(0.931)
10.659
(1.409)
10.479
(1.402)
–
–
–
τ6
9.389
(1.102)
9.610
(1.140)
–
4: 3.5 < C ≤ 4.5;
The annual data is rescaled into an ordinal score (C is ICRG index):
Note: Figures in brackets are the standard errors
19
18
17
16
15
14
13
12
11
i
0.191
0.202
0.144
0.133
0.108
0.136
0.165
0.232
Pseudo
R2
0.249
66
Chapter 4
10
9
8
7
6
5
4
3
2
1
i
ω
3.217
(0.360)
2.057
(0.205)
1.823
(0.184)
1.605
(0.166)
1.235
(0.124)
1.122
(0.116)
1.055
(0.109)
0.969
(0.104)
0.870
(0.097)
0.821
(0.094)
τ1
4.623
(0.877)
3.099
(0.539)
2.655
(0.510)
2.212
(0.484)
1.726
(0.421)
1.485
(0.408)
0.728
(0.478)
0.480
(0.464)
0.369
(0.454)
-0.135
(0.564)
Ordered Probit
τ2
τ3
τ4
7.130
10.749 13.902
(0.872) (1.240) (1.475)
4.077
6.790
9.222
(0.525) (0.749) (0.895)
3.491
5.852
8.225
(0.496) (0.654) (0.807)
2.883
5.059
7.264
(0.468) (0.596) (0.742)
2.273
3.937
5.589
(0.406) (0.467) (0.559)
1.896
3.512
4.985
(0.392) (0.439) (0.515)
1.892
3.263
4.587
(0.382) (0.415) (0.480)
1.576
2.762
4.019
(0.380) (0.391) (0.443)
1.131
2.126
3.562
(0.373) (0.363) (0.414)
0.913
1.865
3.366
(0.376) (0.353) (0.404)
τ6
20.826
(2.385)
13.255
(1.366)
12.247
(1.259)
11.060
(1.156)
8.554
(0.849)
7.805
(0.787)
7.263
(0.731)
6.706
(0.687)
6.095
(0.636)
5.788
(0.612)
0.273
0.292
0.322
0.351
0.375
0.410
0.507
0.558
0.605
Pseudo
R2
0.761
continued on next page . . .
τ5
17.646
(2.050)
11.098
(1.117)
9.913
(0.997)
8.481
(0.842)
6.733
(0.662)
5.975
(0.597)
5.626
(0.564)
4.987
(0.514)
4.491
(0.479)
4.425
(0.479)
Table 4.4: Ordered Probit Regression
Is Corruption Really Persistent?
67
ω
0.757
(0.090)
0.715
(0.089)
0.570
(0.079)
0.510
(0.075)
0.442
(0.073)
0.488
(0.076)
0.509
(0.076)
0.615
(0.084)
0.603
(0.458)
τ1
-0.186
(0.538)
-0.235
(0.542)
0.489
(0.338)
0.443
(0.320)
0.303
(0.315)
0.447
(0.318)
-0.613
(0.455)
-0.368
(0.462)
-0.391
(0.329)
Ordered Probit
τ2
τ3
τ4
0.608
1.644
3.219
(0.372) (0.339) (0.398)
0.572
1.767
3.262
(0.377) (0.339) (0.404)
1.572
2.637
3.678
(0.322) (0.361) (0.432)
1.579
2.499
3.397
(0.319) (0.353) (0.407)
1.516
2.387
3.121
(0.321) (0.344) (0.388)
1.628
2.803
3.382
(0.324) (0.374) (0.411)
0.683
1.774
2.942
(0.315) (0.327) (0.379)
1.419
3.004
3.722
(0.334) (0.390) (0.442)
1.375
2.836
3.587
(0.377) (0.427) (0.527)
τ5
4.245
(0.469)
4.013
(0.460)
4.652
(0.518)
4.274
(0.486)
3.876
(0.464)
4.152
(0.486)
3.712
(0.437)
4.736
(0.535)
4.662
(0.689)
7: C > 6.5
3: 2.5 < C ≤ 3.5;
6: 5.5 < C ≤ 6.5;
2: 1.5 < C ≤ 2.5;
5: 4.5 < C ≤ 5.5;
1: C ≤ 1.5;
4: 3.5 < C ≤ 4.5;
The annual data is rescaled into an ordinal score (C is ICRG index):
Note: Figures in brackets are the standard errors
19
(0.082)
18
17
16
15
14
13
12
11
i
4.426
(0.510)
6.012
(0.696)
5.934
–
–
–
τ6
5.453
(0.575)
5.564
(0.597)
–
0.193
0.200
0.145
0.138
0.117
0.145
0.172
0.238
Pseudo
R2
0.253
68
Chapter 4
Is Corruption Really Persistent?
69
In all regressions the coefficients of the initial level of corruption are
negative and highly significant (p < 0.01, the t-values are in parentheses).
The initial level of corruption explains around 29-44% of the variation in the
change in corruption. Clearly, corrupt countries improve faster than clean
countries, supporting β-convergence. However, various clean countries have
also become less clean over the same period. In other words, Table 4.5
shows that there is not only a ‘bottom-up’ process, but also a ‘top-down’
process. This result is in line with our previous finding that there are both
losers and winners in the world corruption league (see Table 4.1).
Table 4.5: Tests of β-Convergence
Dep. Var
Relative Change
[1984 and 2003]
Absolute Change
[1984 and 2003]
Relative Change
[Averaged-Annual]
Absolute Change
[Averaged-Annual]
03
Log ( C
C84 )
Constant
0.525
(0.088)
1.266
(0.227)
0.039
(0.005)
0.067
(0.012)
0.364
(0.082)
β
-0.139
(0.020)
-0.449
(0.050)
-0.009
(0.001)
-0.024
(0.003)
-0.117
(0.018)
β-std
-0.582
R2 -Adj
0.330
-0.667
0.439
-0.635
0.398
-0.667
0.440
-0.543
0.287
Note: Figures in brackets are the standard errors
Absence of σ-convergence would imply that the standard deviation and
the coefficient of variation of cross-country corruption would be constant
over time. However, we find that the standard deviation (SD) shrinks by
more than 30%, from 1.63 to 1.22 (Table 4.6), suggesting σ-convergence.
Also the coefficient of variation (CV) demonstrates a long-run decreasing
trend over 1984-2003, although it reverses slightly after 1996. So also if
the mean level of corruption is taken into account, corruption dispersion
declines.
70
Chapter 4
Table 4.6: Tests of σ-Convergence
Year
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
4.3
Mean
4.211
4.310
4.309
4.308
4.331
4.292
4.342
4.387
4.529
4.616
SD
1.630
1.599
1.577
1.531
1.504
1.516
1.520
1.483
1.466
1.408
CV
0.387
0.371
0.366
0.355
0.347
0.353
0.350
0.338
0.324
0.305
Year
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
Mean
4.622
4.577
4.471
4.321
4.173
4.051
4.000
3.853
3.515
3.588
SD
1.332
1.288
1.292
1.275
1.308
1.316
1.298
1.291
1.237
1.215
CV
0.288
0.281
0.289
0.295
0.313
0.325
0.324
0.335
0.352
0.339
A Closer Look
In the previous section we have presented some evidence on the absence
of corruption persistence using some simple approaches borrowed from the
economic growth literature. However, these approaches ignore the crosscountry distribution of corruption and its dynamics over time. Using nonparametric approaches that are also borrowed from the economic growth
literature (Quah, 1993a, b, and c; and Fingleton, 1997 and 1999), we provide
some further evidence on corruption convergence.
We apply a kernel density estimation that provides information about
the modality of the distribution over time, i.e., the bump(s) in a kernel
density.8 We are therefore able to examine whether a particular distribution with, say, two classes of countries (bimodal)—e.g., corrupt and clean
countries—is transformed into a unimodal density.
For the dataset X1 , X2 , ..., Xn , a kernel density estimator is drawn from
8
According to three tests on normality of the data (D’Agostino et al. (1990), ShapiroWilk (1965), and Shapiro-Francia (1972)), normality of the distribution should be rejected
for years at the end of the sample period (1999-2003). The test on skewness captures a
positive skewness (1984-1989) that becomes negative in 1990-1994, and then turns positive
but close to zero. The kurtosis test detects that the distribution initially has thinner tails
compared to the normal distribution, but at the end of the sample period it tends to have
heavier tails.
Is Corruption Really Persistent?
71
an unknown probability density function, f (x):
n
X
1
x
−
X
i
fˆ(x) =
k
nhn
hn
(4.1)
i=1
where k is a bounded non-negative kernel function satisfying
R
k(x)dx = 1
and h is a sequence of positive numbers called the bandwidth. There are
R
several kernels satisfying the condition k(x)dx = 1. However, the choice
of a kernel function does not really matter, since the difference in efficiency
among kernel functions is very small (Silverman, 1986; Table 3.1). What
does matter is the choice of bandwidth h, because it determines the biasvariance trade-off in the density estimation.
To find an optimal bandwidth, one usually estimates the closeness of
the estimator fˆ to the true density f , where the estimate fˆ depends on the
data, the kernel, and the bandwidth. The most widely used approach is to
R
minimize the mean integrated square error, MISE(fˆ) = E [fˆ(x)−f (x)]2 dx.
Silverman (1986) argues that the optimal window width of a Gaussian kernel
1
can be computed via hopt = 1.06σn− 5 . We apply this to evaluate the 19842003 distributions of the level of corruption.
Figure 4.1 shows the evolution of the distribution of the level of corruption over 20 years. The vertical axis represents the estimated kernel
density, while the horizontal axis corresponds to the corruption level (i.e.,
the rescaled ICRG index). Several conclusions can be drawn. First, the
evolution is marked by a slow change in modality of the distribution from
bimodal to unimodal as it starts with two bumps in 1984 and ends with
a single bump in 2003, indicating convergence.9 In 1984, the first bump
is below a score of four of the ICRG index and around 0.25 of the kernel
estimate. It moves slowly to the right to an ICRG score above four by 1993
and about 0.30 of the kernel density (the very right graph on the second row
in Figure 1), but goes back to the left to a score around three for the ICRG
index and 0.40 of the density by 2003. This suggests that there has been an
9
As an experiment, we also checked the evolution of the distribution using mean-adjusted
data and a set of optimum bandwidths. We find similar results (available on request).
72
Chapter 4
improvement in world corruption in the first part of our sample period, but
a worsening in the second half. There is a decrease in the number of clean
countries over the same period as depicted by the dynamics of the second
bump. In 1984, the second bump starts with a very high level of the ICRG
index (about six), but it has dropped ten years later and vanishes in the
next ten years.
Second, a similar pattern can be discerned in the development of the
tails. During 1984-1990, the score of the ICRG index in the left tail of the
distribution is close to one, suggesting that there are a few highly corrupt
countries. Although the ICRG score of the left tail stays more or less at the
same level, there is a clear tendency for the kernel estimate to move to zero
during the years of 1991-1996. In the years 1997-2000, the kernel density
increases to 0.1, while the ICRG index moves to two.
Third, the right tail of the distribution lies above 0.1 of the kernel (1984)
and reaches its highest point in 1992 at about 0.15. However, it gradually
goes down to close to zero. This confirms our finding that there is an
increase in the number of clean countries over the first half of the period
and a decrease in the second half.
Table 4.7 shows that in 1984, the world corruption distribution is clustered around two peaks (3.915 and 6.462). This polarization continues to
be visible in most years until 1996; the distribution is marked by two convergence clubs—to borrow the terminology from the economic growth literature. In the first club, the corruption level moves from below to above
four (the ‘bottom-up’ process) over 1984-1995. The second club shows the
opposite pattern: the mode shrinks from above to below six (the ‘top-down’
process). It is interesting also to see that, within this period, the distance
between the two midpoints of the bumps reduces from 2.56 (1984) to 1.15
(1995). Before these bumps vanish, three bumps emerge in 1996. Between
1997 and the end of the sample period, only one bump is detected that
confirms the σ-convergence process.
Figure 4.1: Annual Kernel Density, 1984 (top-left) to 2003 (bottom-right)
Is Corruption Really Persistent?
73
74
Chapter 4
Table 4.7: Modality (Rescaled) ICRG data
4.4
Year
Bandwidth
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
0.687
0.672
0.664
0.645
0.633
0.639
0.640
0.625
0.618
0.593
0.561
0.542
0.544
0.537
0.551
0.554
0.547
0.544
0.521
0.512
Mode 1
3.915
3.915
3.916
3.916
3.916
3.916
3.916
3.916
4.000
4.085
4.000
4.000
4.000
3.940
3.938
3.937
3.868
3.845
3.016
3.104
Midpoint
Mode 2 Mode 3
6.462
5.429
5.760
5.511
5.679
5.590
5.253
5.671
5.530
5.236
5.148
5.395
5.641
-
Markov Chain Analysis
Now we turn to Markov chain analysis as an alternative way to examine
the dynamics of corruption over time. We may describe a Markov chain as
follows. In a set of states N = {1, 2, . . . , N }, the Markov process starts in
one of these states and moves from one state to another. The chain moves
from the current state i to the next step j with a (transition) probability
pij , where the probability does not depend on which states the chain was
in before the current state.10 Adopting Bickenbach and Bode (2003) who
investigate the dynamics of income per capita, here we also assume that corruption level follows a finite first-order Markov chain with stationary transi10
However, the process may remain in the state it is in, with probability pii .
Is Corruption Really Persistent?
75
tion probabilities. A finite first-order Markov process with stationary transition probabilities is a stochastic process {Xt , t ∈ T } with T = {0, 1, 2, . . .}
satisfying the Markov property (Bickenbach and Bode, 2003):
P [X(t) = j|X(t − 1) = i, X(t − 2) = it−2 , . . . , X(0) = i0 ] =
P [X(t) = j|X(t − 1) = i]
(4.2)
for all t and all possible states j, i, ik (k = 0, 1, . . . , t − 2).
If the transition probabilities pij (t) := P {X(t) = j|X(t − 1) = i} do
not depend on t (pij (t) = pij for all t ∈ T ), the Markov chain is said to be
time stationary. Here, a time-stationary Markov chain is determined by the
Markov transition matrix P,

Pt,t+s
p11
p12
...
p1N


 p21 p22 . . . p2N
=
 ..
..
..
..
 .
.
.
.

pN 1 pN 2 . . . pN N








∀i , 0 ≤ pij ≤ 1, and the initial distribution F (0) = (F1 (0), F2 (0), . . . , FN (0))
P
P
where
Fi (0) = 1. Since P is a right-row stochastic matrix, then j pij =
1.11 This transition matrix can be estimated via Maximum Likelihood (ML)
in which case the estimate is consistent with the relative frequency of the
transitions observed over the period of analysis.
The question now is what is the probability that after a certain transition
period the Markov chain is in a given state? For a given law of motion, as
determined by P, and given initial distribution F (0), the distribution after
the first transition period (t=1) is
F (1) = F (0) × Pt,t+S ,
(4.3)
11
We may say that pij is the probability for a country to move from corruption level i to
level j during the time span s.
76
Chapter 4
or, in general,
Fx,t+S = Fx,t × Pt,t+S .
(4.4)
The distribution after the m-th transition period is:
m
Fx,t+(m×s) = Fx,t × Pt,t+s
.
(4.5)
Provided the Markov chain is regular, the distribution converges toward the
limiting (ergodic) distribution is given by
m
F ∗ = lim F (0)Pt,t+s
.
m→∞
(4.6)
This can be obtained via a 1 × N row vector π such that
π = πP
(4.7)
as m tends to infinity. Comparing this limiting distribution to the initial
distribution of corruption can tell us whether countries converge or diverge.
Higher probabilities in the median-states of the limiting distribution, as
compared to the initial distribution, indicate convergence.
A critical issue here is to define the number of states as well as the
length of the interval between states. This is somewhat arbitrary and researchers usually consider the trade-off between the number of observations
in every state and the inter-state dynamics. Some researchers first divide
the observed countries evenly and determine the length of the interval later
(see, for instance, Bickenbach and Bode (2003), Hammond and Thompson
(2006)). Others determine the length of the interval first and then decide
upon the distribution—which is equally plausible (see, for example, Quah,
1993b; Fingleton, 1997)). In our case, due to the nature of the data, it
is difficult to generate a particular interval that results in an equal distribution. We, therefore, split the countries into six states according to
their level of corruption, where the distance from a state to another is the
Is Corruption Really Persistent?
77
same—consequently, the distribution is unequal.
Following Shorrocks (1978), we also measure the degree of mobility as:
Mobility =
N − tr(Pt,t+s )
,
N −1
where N is the number of states. This index ranges between zero and
(4.8)
N
N −1
where a higher value means more mobility across corruption classes.12
Using the same data as in the ordered logit and probit regressions, Table
4.8 gives the results of the Markov chain analysis: the transition matrix, the
initial distribution, the limiting distribution, and the mobility index. The
first panel displays the short-run dynamics in which the transition matrix
is measured as the average of one-year transition over every year from 19841985 to 2002-2003. Not surprisingly, as countries are concentrated in the
main diagonal of this matrix over a one-year horizon, there is a high degree
of corruption persistence. The first row shows that over the full sample (i.e.,
101 countries and 20 years) 41 observations fall in state 1. Of these, about
83 per cent remain in that state in the following year. Similarly, of the 165
observations in state 7, 91 per cent stay in that state in the following year.
In line with our previous findings, it turns out that few countries exhibit
a substantial change over this one-year horizon. For example, only 17 per
cent of the countries move from state 1 to states 2 and 3, while 11 per cent
jump from state 4 to states 2 and 3. The mobility index is very low (0.18),
confirming that one-year transitions do not indicate convergence.
However, comparing the limiting distribution with the initial one suggests that there is convergence. First, the share of the very upper tail (state
7) greatly reduces from nine per cent in the initial year to less than one per
cent in the limiting distribution. Second, the fraction of countries in state
2 almost doubles from seven to 14 per cent. Third, while no substantial
change is found for the shares of states 1 and 4, and a slight decrease occurs
12
As indicated by the trace of the matrix, an index of zero simply means that all observations lie on the main diagonal. A value of NN−1 implies that none of the observations
lies on the diagonal. Thus, in our case the index ranges between 0.00 and 1.20.
78
Chapter 4
for that of state 5, there is a substantial rise in the share of state 3, from
22 to 33 per cent. In short, these findings show that some countries become
less corrupt while others become less clean.
Table 4.8: Markov Chain Estimates, 1984-2003
Transition
State 1
State 2
State 1 (41)
State 2 (134)
State 3 (419)
State 4 (594)
State 5 (307)
State 6 (259)
State 7 (165)
Initial Dist.
1st Iteration
19th Iteration
Limiting Dist.
Mobility Idx.
0.829
0.030
0.000
0.000
0.000
0.000
0.000
0.021
0.020
0.020
0.025
0.179
0.122
0.813
0.069
0.002
0.000
0.000
0.000
0.070
0.075
0.120
0.141
State 1 (4)
State 2 (10)
State 3 (19)
State 4 (35)
State 5 (9)
State 6 (10)
State 7 (14)
Initial Dist.
1st Iteration
Limiting Dist.
Mobility Idx.
0.000
0.000
0.000
0.029
0.000
0.000
0.000
0.040
0.010
0.005
0.965
State 3 State 4 State 5
Annual Transition (1984-2003)
0.049
0.000
0.000
0.119
0.037
0.000
0.828
0.086
0.017
0.109
0.832
0.054
0.033
0.127
0.811
0.004
0.008
0.085
0.000
0.000
0.012
0.218
0.310
0.160
0.230
0.300
0.163
0.298
0.297
0.157
0.329
0.309
0.142
State 6
State 7
0.000
0.000
0.000
0.003
0.029
0.900
0.073
0.135
0.133
0.089
0.052
0.000
0.000
0.000
0.000
0.000
0.004
0.915
0.086
0.079
0.020
0.002
20-Year Transition (1984 and 2003)
0.750
0.250
0.000
0.000
0.000
0.300
0.500
0.200
0.000
0.000
0.474
0.368
0.158
0.000
0.000
0.143
0.514
0.257
0.057
0.000
0.000
0.667
0.222
0.111
0.000
0.000
0.300
0.000
0.600
0.100
0.000
0.071
0.071
0.357
0.429
0.099
0.188
0.347
0.089
0.099
0.198
0.406
0.168
0.139
0.069
0.346
0.445
0.191
0.012
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.071
0.139
0.010
0.000
The convergence phenomenon is much more visible when we use a 20year horizon of the transition matrix as shown in the second panel of Table
4.8. A direct indication of convergence is provided by the main diagonal:
only 37 per cent of the countries stay in the same state between two distant
years. Compared to the annual transition, the number of entries on the main
diagonal reduces sharply. For example, no single country is found to remain
in state 1. Rather, all countries that were in state 1 (in 1984) improve their
Is Corruption Really Persistent?
79
level to states 2 and 3 (in 2003). On the contrary, all countries that were in
state 4 drop down to states 1 (3 per cent), 2 (14 per cent), 3 (51 per cent),
or jump to state 5 (6 per cent). These upward and downward inter-state
shift dynamics are reflected in the very high value of the Shorrocks index of
0.96.
4.5
Conclusion
Is corruption really persistent? Using ICRG data over the period 1984-2003
we argue that it is not. We find that many countries see their level of corruption decrease or increase. There are more countries with a decline than
countries with a rise in their level of corruption. The correlation between
corruption in two years is high, but the larger the distance between the two
years becomes, the lower is the correlation, so there is no ρ-persistence. In a
very simple autoregressive model of corruption, the coefficient of the lagged
dependent variable and the R2 reduce over time. This applies also to the
similar ordered logit and probit models.
We find strong evidence for β-convergence and σ-convergence. Not only
do corrupt countries become less corrupt (catching-up), but clean countries tend to become less clean over time. This pattern is confirmed in our
analysis of the dynamics of the distribution of the data. There is a significant modality shift in the corruption distribution over time. Using a
Gaussian kernel function we detect a transformation in the corruption distribution from a bimodal to a unimodal distribution. Finally, on the basis
of a Markov chain analysis, we also find interclass upward and downward
shifts of countries.
Our analysis shows that research should not focus on the explanation
of corruption persistence, but should be redirected towards explaining the
process of corruption convergence.
Chapter 5
Governance and Growth
Revisited
“The care of human life and happiness, and not their destruction,
is the first and only legitimate object of good government.”
————————————————————
Henry Steele Commager (1809)
5.1
Introduction
There is a lively debate on the relation between governance and economic
growth. One group of studies presents empirical evidence that good governance stimulates economic growth (see, for instance, Knack and Keefer,
1995; Barro, 1997; Keefer and Knack, 1997; Chong and Calderon, 2000; and
Kaufmann and Kraay, 2002). Other studies, however, provide evidence for
a negative relationship between governance and economic development. For
instance, Quibria (2006) reports for 29 Asian countries that economies with
better governance fare worse than the ones with poor governance. Country case studies generally support this finding. A good example is China.
Even though its governance is below the world average, it has above-average
rates of growth (Qian, 2003). Likewise, the Philippines and Vietnam have
This chapter is based on joint work with Emmanuel Pandu Nugroho and Jakob de Haan,
published in Kyklos, 2007, 60(2): 279-290.
82
Chapter 5
more or less the same quality of governance, but Vietnam is booming out of
a poverty trap, while the economy of the Philippines stagnates (Pritchett,
2003).
The literature on the governance-growth nexus is plagued by various
shortcomings. First, most governance indicators are only available for recent years. This is certainly true for the governance index of Kaufmann et
al. (2006) which seems to have emerged as the industry standard (Quibria,
2006). Consequently, cross-country (panel) growth models can only identify
correlation. Second, as pointed out by Quibria (2006: 102), “governance remains a broad, multi-dimensional concept that lacks operational precision.
It has often been used as an umbrella concept to federate a whole assortment of different, albeit related, ideas.” Finally, most studies do not check
whether results are sensitive to model specification and sample selection.
In this chapter we construct a new index of governance applying Confirmatory Factor Analysis (CFA) to the International Country Risk Guide
(ICRG) dataset on indicators of governance, namely, democratic accountability, government stability, bureaucracy quality, corruption, and rule of
law. We focus on ICRG data as these are the only governance indicators
available for a large group of countries and a long time span. We use CFA
to analyze to what extent the different indicators of governance in the ICRG
dataset contain the same information. We find that the various dimensions
can be combined into one index. Next, we test whether our index is related
to growth for varying samples of countries and different sets of conditioning
variables. We find that our index is positively related to economic growth.
This result is fairly robust across different samples of countries and model
specifications.
The remainder of this chapter is organized as follows. In section 5.2, we
briefly review the concept of governance and explain how we use CFA to
construct our index. In section 5.3, we examine the relationship between
this index and economic growth. Section 5.4 concludes.
Governance and Growth Revisited
5.2
83
Governance: Concept and Construct
Governance is usually defined as the manner in which authority is exercised
in the management of a country’s socio-economic resources. Kaufmann et
al. (1999) have extended this definition by including three core dimensions
of governance, namely [1] the process by which those in authority are selected, monitored, and replaced, [2] the government’s capacity to effectively
manage its resources and implement sound policies, and [3] the respect of
citizens and the state for the country’s institutions. This definition is in line
with what Adserá et al. (2003: 446) call a ‘well-functioning government‘,
that is, “governments that abide by the rule of law, whose bureaucrats and
policy makers are not affected by graft practices, and whose administrative
machinery delivers goods and services in an efficient manner.”1
One crucial question is how to incorporate these various dimensions into
a measure of governance. Researchers typically use the average or summation of some ICRG indicators to construct a single index of governance (see,
for instance, Knack and Keefer, 1995; Knack, 1996; Hall and Jones, 1999).
There are some advantages and disadvantages of this approach. For example, the availability of ICRG data since the beginning of the 1980s makes
it possible to observe the governance-economic growth relationship over a
substantial time-span. However, simply averaging or aggregating such indicators to construct a governance index can be problematic. Governance
is a theoretical concept, which is not precisely measured by the ICRG indicators, which are imperfect measures of governance and likely to contain
measurement errors. In such a situation, governance should be treated as a
latent variable.2
In this chapter, we consider five variables of the ICRG dataset motivated
by the theoretical definitions described above, namely democratic accountability, government stability, bureaucracy quality, corruption, and rule of
1
Similar labels can also be found in Hall and Jones (1999), Wyatt (2003), and Borrmann
et al. (2006).
2
The dataset of Kaufmann et al. has been constructed on the basis of a latent variable
technique. Unfortunately, this dataset is available only for years since 1996.
84
Chapter 5
law, as indicators of governance on which we apply CFA yielding a new
index of governance.3 This latent variable technique transforms the ICRG
dataset into a cardinal index. As in the underlying ICRG variables, a higher
score means better governance. The ICRG variables can be described as follows:4
• Democratic Accountability (6-point scale) measures how responsive
government is to its people. The less responsive it is, the more likely
the government will fall, peacefully in a democratic society, but possibly violently in a non-democratic one.
• Government Stability (12-point scale) is an assessment of the government’s ability to carry out its declared program(s), and to stay in
office, assigned as the sum of three components: government unity,
legislative strength, and popular support.
• Bureaucracy Quality (4-point scale) gives a high score to countries
where the bureaucracy has the strength and expertise to govern without drastic changes in policy or interruptions in government services,
and where it has an established mechanism for recruitment and training.
• Corruption (6-point scale) proxies actual or potential corruption in
the form of excessive patronage, nepotism, job reservations, ‘favorfor-favors’, secret party funding, and suspiciously close ties between
politics and business.
• Law and Order (6-point scale) assesses the strength and impartiality
of the legal system (the law component) as well as popular observance
3
Confirmatory Factor Analysis seeks to determine if the number of factors and the
loadings of measured (indicator) variables on them conform to what is expected on the
basis of theory. Indicator variables are selected on the basis of prior theory and factor
analysis is used to see if they load as predicted on the expected number of factors. Usually
the researcher will posit expectations about which variables will load on which factors
(Kim and Mueller, 1978: 55) The researcher seeks to determine, for instance, if measures
created to represent a latent variable really belong together.
4
Drawn from http://www.icrgonline.com/
Governance and Growth Revisited
85
of the law (the order component).
These indicators are expected to be highly correlated. Table 5.1 indeed
shows that the correlation coefficients among these indicators range between
0.71 and 0.84; they are also all highly significant (p <0.01). Nevertheless, the
correlations among the governance indicators are lower than 1.00, suggesting
that these indicators are imperfect measures of the concept of governance.
Table 5.1: Correlations among the ICRG Indicators of Governance
Governance Indicators
Democratic Accountability (DA)
Government Stability (GS)
Bureaucracy Quality (BQ)
Corruption (C)
Law and Order (LO)
DA
1.000
0.716
0.725
0.759
0.735
GS
BQ
C
LO
1.000
0.786
0.736
0.762
1.000
0.836
0.821
1.000
0.821
1.00
In the following, we use a latent variables approach to evaluate to which
extent the governance indicators capture the same information. We use
ICRG data for 1984, the first year for which the data are available. More
specifically, we use CFA5 that can be expressed as
x = Λξ + δ
(5.1)
where x denotes a vector of observed variables drawn from the ICRG data
(democratic accountability, government stability, bureaucracy quality, corruption, and rule of law). These are the indicators of the exogenous latent
variable governance (ξ); Λ stands for the corresponding matrix of unknown
factor loadings that capture both the scale of indicators and the strength of
their relation to ξ; and δ denotes a vector of measurement errors. Moreover,
the basic assumptions accompanying equation 5.1 are E (δ)= 0 and E (ξδ’)=
0. By convention, variables in x and ξ are written as deviations from their
respective means. Figure 5.1 depicts the relationship between governance,
its indicators, and the error terms.
5
See Bollen (1989) as well as Wansbeek and Meijer (2000) for further details.
86
Chapter 5
From Equation 5.1, the population covariance matrix, Σ, follows as
Σ(θ) = E(xx0 )
= ΛΦΛ0 + Θ.
(5.2)
The last line of equation 5.2 shows that the covariance matrix of x is decomposed into the parameters of Λ; the covariance matrix of ξ, (Φ); and the
covariance matrix of δ, (Θ). This equation is estimated using the maximum
likelihood (ML) function,
FM L = log |Σ(θ)| + tr SΣ−1 (θ) − log |S| − q.
(5.3)
Democratic Accountability δ1
*
Governance H
@H
@HHH
@
HH
@
H
HH
@
j
H
@
@
@
@
@
R
@
Government Stability
δ2
Bureaucracy Quality
δ3
Corruption
δ4
Law and Order
δ5
Figure 5.1: Governance in Factor Analysis Model
Minimizing this fitting function means choosing the values for the unknown parameters that lead to the implied covariance matrix Σ(θ) as close
Governance and Growth Revisited
87
as possible to the sample covariance matrix (S); with q denotes the number
of indicators. On the basis of these parameters, the governance index can
be generated.
Table 5.2 reports the estimates of the parameters and corresponding
statistics that explain the resulting index of governance. This table shows
that all parameters are estimated with reasonable precision. The parameters
are statistically significantly different from zero as indicated by the high
values of the t-statistics. The value of χ2 , comparing the proposed model
to an unrestricted alternative (saturated) model, has a value of 4.70 with 5
degrees of freedom, which is not significant. The Normed Fit Index (NFI)
has a value of 0.99, while the Comparative Fit Index (CFI) has a value of
1.00. Likewise, the root mean squared error is zero. According to these
measures, the model fits very well.6 In addition, the last column of the
table shows the estimate of the reliability of each indicators; i.e., the squared
correlation with the latent variable governance. According to this measure,
democratic accountability performs the worst and bureaucracy quality the
best, respectively, with reliabilities of 0.67 and 0.84.
Table 5.2: CFA Estimates
Governance
Indicators
Democratic
Accountability
Government
Stability
Bureaucracy
Quality
Corruption
Law and
Order
RMSE
χ2
Estimates
(Standardized)
0.820
(10.092)
0.841
(10.503)
0.915
(12.088)
0.909
(11.931)
0.898
(11.704)
0.000
4.702
Error
Covariance
0.328
(6.401)
0.293
(6.252)
0.162
(5.118)
0.175
(5.298)
0.193
(5.523)
Reliability
0.672
0.707
0.838
0.825
0.807
Continued on next page
6
For a further discussion on the fit measures, see Wansbeek and Meijer (2000).
88
Chapter 5
Governance
Indicators
Normed Fit
Comparative Fit
Estimates
(Standardized)
0.989
1.000
Error
Covariance
Reliability
t-statistics in brackets
5.3
Governance and Growth
5.3.1
Parsimonious Model
We now examine the relationship between economic growth and governance.
In examining this relationship, we first run a set of parsimonious growth
models as follow:
G = F α + βξ
(5.4)
where G is a vector of average GDP per capita growth rates over 19842004 and ξ is the 1984 governance index we have constructed. Meanwhile,
F is a matrix consisting of three or four explanatory variables. At least
two possible sets of variables have been suggested in the empirical growth
literature. In Levine and Renelt (1992) and Mankiw et al. (1992), the
matrix F consists of the initial income, the school enrolment rate, and the
investment rate. Keefer and Knack (1997) and Beugelsdijk et al. (2004)
include initial level of GDP per capita, the school enrolment rate, and the
price level of investment in the matrix. We take the values of these variables
at the beginning of the growth period to minimize reverse-causation.
Table 5.3 displays the results of the parsimonious regressions. Our findings for income convergence and human capital confirm those of many other
growth studies. The inclusion of the price level of investment in model 4
yields a slightly higher R2 compared to the inclusion of the investment ratio
in model 3, but the coefficients of both variables differ significantly from
zero. However, once the governance index is included in the regressions, the
model with the price level of investment has more explanatory power than
the one with the investment ratio and the difference between the two R2 -s
Governance and Growth Revisited
89
becomes larger. Moreover, the investment ratio loses its significance as its
t-values drops from 2.31 to 1.35. Most importantly, the coefficients of our
governance index are very similar in both specifications and always positive
and significant.
Table 5.3: Parsimonious Growth-Governance Regressions
Explanatory
Variables
Constant
Income
Schooling
Inv. per GDP
Inv. Price
Governance
N
F
R2
Model 1
β
t-val.
5.50
2.61
-0.92 -2.69
4.06
3.40
6.73
2.31
Model 2
β
t-val.
5.52
2.59
-0.68 -2.12
4.56
3.84
-0.01
106
9.30
0.21
103
9.29
0.22
Model 3
β
t-val.
11.32
4.29
-1.56 -4.14
3.72
3.05
4.24
1.35
-2.46
0.94
82
8.53
0.31
3.25
Model 4
β
t-val.
12.31
4.67
-1.50 -4.14
4.22
3.58
-0.01
0.94
81
10.19
0.35
-2.53
3.36
Different N is due to list-wise deletion; Independent Variables: in 1984 values
Dependent Variable: Average growth rate of GDP per capita, 1984-2004
In the remainder of this section we will examine the robustness of our
finding. We test whether the coefficient of our governance indicator is sensitive to changes in [1] the number of observations and [2] model specification.
5.3.2
Effect of Observations
To examine the impact of the use of various observations on the stability of
our index, we apply recursive regressions. The idea behind this approach is
to use a different number of observations, starting from 50 per cent of the
sample to end up with the full set of 81 countries. For this purpose, we use
model 4 of Table 5.3 since it outperforms model 3.
First, we put observations in ascending order according to their governance scores, i.e., from the lowest to the highest scores. Starting with
50 per cent of the sample, we add one observation every time we run the
regression, and end up with the full sample. Thus, the first 50 per cent of
90
Chapter 5
the ascending-ordered sample are countries with poor governance. Second,
we follow the same procedure in the other way around, i.e., in a descending
order for the governance indicator. In other words, the first 50 per cent of
the descending-ordered sample are countries with good governance.
The results of the recursive regressions are displayed in Figure 5.2. Several conclusions can be drawn from this figure. The coefficients of the governance indicator computed on the basis of ascending and descending-ordered
samples fluctuate around 0.94. The largest deviations occur when the number of countries is between 42 and 48. The lowest coefficients range between
0.25 and 0.37, while the highest are in the 1.14-1.40 range. This implies that
the impact of our index is relatively stable, in the sense that it is not much
affected by a change in the number of observations. In sum, the effect of
governance on economic growth is always positive, no matter what sample
we take into consideration. However, the effect is bigger in poorly-governed
countries.
The graphs for the t-values tell a similar story. In general, the t-values
are high and they become higher when the number of observations increases
to converge at 3.36. In the governance-ascending-ordered countries, the tvalues are always significant, while significant t-values in the governancedescending-ordered countries are reached after the inclusion of 60 countries.
5.3.3
Sensitivity Analysis
Finally, we examine whether our results are sensitive to a change in model
specification. For this purpose, we extend the ‘base’ regression, equation
5.5, as follow:
G = F α + βξ + Zγ
(5.5)
where G, F, and ξ are defined as before; while Z is a matrix of up-to-three
variables drawn from a pool of M available variables. This is the setup we
use to identify the robustness of ξ adopted from Levine and Renelt (1992)
and Sala-i-Martin (1997). Levine and Renelt examine the robustness by
Governance and Growth Revisited
91
4.0
3.0
2.0
1.0
0.0
40
45
50
t-val.: asc.
55
t-val.: desc.
60
65
70
beta: asc.
75
80
beta: desc.
Figure 5.2: Regressions with Various Samples
running a series of regressions from all possible combinations in the matrix
Z (Appendix 3 for the detail list of variables). This results in a vector
of β̂ and its standard deviation, σβ̂ . According to Levine and Renelt, the
robustness of ξ is determined on the basis of two extreme bounds defined as
max β̂±2 σ .
min
β̂
They consider a variable robust if the lower and upper extremes
produce the same signs and remain significant; otherwise it is fragile, even
if there is only one regression for which the sign of the coefficient changes
or becomes insignificant.
Since this test is too strong for any variable to pass it, Sala-i-Martin
(1997) proposes to examine the entire distribution of β̂ and to run a test for
the cumulative distribution function (CDF). The CDF(0) test is based on
the fraction of the CDF lying on each side of zero; that is, the larger areas
under the density function regardless whether it is CDF(0) or 1-CDF(0).
So, CDF(0) will always be a number between 0.5 and 1.0.
Our judgment on the robustness of our governance index will be based
92
Chapter 5
on the test of Sala-i-Martin. Following Sturm and de Haan (2005), we apply
the criteria that CDF(0)> 0.95 and that the coefficient of the governance
index should be significantly different from zero in at least 90 per cent of
the regressions we run. Computation on the basis of 81 observations and 65
variables in the matrix Z —thus we run 43,680 regressions—indicates that
the CDF(0) test statistics equals 1.00. Moreover, we find that in 93 per cent
of the regressions, the coefficient of our index is significantly different from
zero at the 5 per cent level.
Finally, we redo the sensitivity analysis but with varying numbers of
observations. Like before, we order the sample in ascending and descending
ways. Starting with 50 per cent of the sample, in every step of the analysis
we add one observation to end up with the full sample. At this stage,
however, we focus on the CDF(0) test and run 3,581,760 regressions. The
results are displayed in Figure 5.3.
1.0
0.8
0.5
0.3
0.0
40
45
CDF-asc.
50
55
CDF-desc.
60
65
Frac. sign. asc.
Figure 5.3: Sensitivity Analysis
70
75
80
Frac. sign. desc.
Governance and Growth Revisited
93
In general, the results support our previous findings. The ascendingordered group starts with a CDF(0) of 0.95 and ends up with 1.00. The
descending-ordered group also passes the threshold of 0.95, but only after
the use of 53 observations. Starting with a CDF(0) of 0.71, the descendingordered group of countries catches up the other group to converge at a
CDF(0) of 1.00. In terms of the fraction of regression with 95 per cent
level of significance, however, the two groups perform less, although there
is a tendency to converge. In the ascending-ordered group, the 90 per centthreshold is reached after the use of 62 observations, but it is never achieved
by the descending-ordered group. Therefore, it is safe to conclude that our
index is fairly robust considering its performance in the CDF(0) test.
5.4
Conclusion
The literature on the governance-growth relationship does not provide clear
cut conclusions about the relevance of governance for growth. We contribute
to this debate by introducing our governance index generated using confirmatory factor analysis on ICRG governance indicators. We focus on ICRG
data as these are the only governance indicators available for a long timespan that include a large group of countries. Confirmatory factor analysis
is an ideal instrument to analyze to what extent the different dimensions of
governance as identified in the ICRG dataset contain the same information.
We find that the five dimensions of governance can be combined into one
single index.
Our parsimonious models indicate that this constructed index positively
and significantly explains economic growth. We also tests for the robustness of our results by applying recursive regressions and the sensitivity test
of Sala-i-Martin (1997). Furthermore, we apply the same test to varying
samples. On the basis of our results, it is safe to conclude that the impact
of our index on economic growth is fairly robust.
Chapter 6
Geography and Governance:
Does Space Matter?
“Everything is related to everything else,
but near things are more related than distant things.”
————————————————————
Tobler (1970)
6.1
Introduction
There is increasing evidence that government governance has an impact on
economic development. For instance, Rajkumar and Swaroop (2007) show
empirically that the differences in the efficacy of public spending can be
largely explained by the quality of governance. Major donors and international financial institutions make their aid and loans increasingly conditional
upon reforms that ensure ‘good governance’. Although definitions of governance differ, there seems to be broad consensus that good governance
means that the government concerned is accountable, transparent, responsive, effective and efficient, and follows the rule of law, thereby assuring that
corruption is minimized.
Various studies have examined cross-country differences in governance.
Geography is often taken into account in these studies. Variables like latiThis chapter is joint work with J. Paul Elhorst and Jakob de Haan.
96
Chapter 6
tude, climate, temperature, country size, climate-related diseases, or dummies indicating that countries are landlocked, islands, or belong to a particular region are commonly used as proxies for geography (Acemoglu et
al., 2001, 2002; Easterly and Levine, 2003; Rodrik et al., 2004; Olsson and
Hibbs Jr., 2005). However, these studies ignore that cross-country data are
generally characterized by spatial dependence. According to Anselin (2006:
901), “Spatial dependence is a special case of cross-sectional dependence, in
the sense that the structure of the correlation or covariance between observations at different locations is derived from a specific ordering, determined
by the relative position (distance, spatial arrangement) of the observations
in geographic space.”
There are several reasons why the geography-governance nexus may be
characterized by spatial dependence. The first reason is that survey based
indicators of governance may contain measurement errors. A bias may occur
when a respondent rates a country as poorly-governed because its neighbors
are badly-governed. Another systematic error may arise from the halo effect of other variables that are associated with governance through space.
Finally, an error may occur when missing values are imputed on the basis
of available information.
The second reason is more substantive. Like other political and economic
phenomena—democracy, war and peace, or economic liberty (O’Loughin
et al., 1998; Ward and Gleditsch, 2002; Simmons and Elkins, 2004)—
governance may have a spatial dimension due to spillovers and diffusionadoption processes. Other possible factors that may cause spatial dependence include policy convergence (Mukand and Rodrik, 2005), interdependence of policy decisions (Brueckner, 2003), or transmission of government
forms (Starr, 1991).
Another problem that may arise when data have a locational component
is that the parameters in such a model are not homogenous over space
but vary across different geographical locations. This is known as spatial
heterogeneity (Anselin, 1988). For example, the effect of the determinants
of governance may differ across countries due to differences in institutions,
Geography and Governance: Does Space Matter?
97
norms, or other country-specific characteristics.
Using the governance indicators of Kaufmann et al. (2006), we examine
the existence of both spatial dependence and spatial heterogeneity. Our
findings show that governance in one country exhibits a positive relationship with governance in neighboring countries, i.e., poorly (well) governed
countries are geographically clustered with other poorly (well) governed
countries. In a series of models explaining cross-country differences in governance, we find that this interaction effect is robust to different spatial
weights matrices measuring the spatial arrangement of the countries in the
sample. In addition, we find that the impact of the determinants of governance is different for different countries.
The remainder of the paper is organized as follows. Section 2 provides
some preliminary empirical evidence of geographical clustering and introduces two spatial econometric models to operationalize spatial dependence.
Section 3 examines the phenomenon of spatial heterogeneity, while section
4 concludes.
6.2
Spatial Dependence
6.2.1
Preliminary Evidence
To test for spatial dependence of governance among different countries, we
use the dataset of Kaufmann et al. (2006). We employ a governance index
that is constructed as the unweighted average of the various components
of governance, i.e., voice and accountability, political stability, government
effectiveness, regulatory quality, rule of law, and control of corruption in
2005 (see Appendix 4a for further details). Our index ranges from −2.5 to
+2.5, where a higher score reflects better governance.
Figure 6.1 displays every country’s governance index against the distance
to the country with the highest (Iceland) and the lowest governance index
(Somalia). The distance to these countries is measured as the kilometerconverted row-normalized great circle distance (explained below). The figure shows that the closer a country is located to the world’s best (worst)
98
Chapter 6
practice, the higher (lower) is its governance index.
As a more formal test, we calculate Moran’s I and Geary’s c statistics as
two common measures of spatial autocorrelation. The null hypothesis is no
spatial dependence. If I is greater (smaller) than its expected value, E(I),
the overall distribution of governance is characterized by positive (negative)
spatial dependence. If c is greater (smaller) than its expected value, E(c),
the overall distribution of governance is characterized by negative (positive)
spatial dependence. The statistical inference is computed on the basis of
z-statistics.
These two statictics are computed using two different spatial weights
matrices.1 The first spatial weights matrix that we use (W 1 ) is based on
the kilometer-converted great circle distance (dij ) between two countries (i
and j) on the sphere:
dij = arccos [(sin φi sin φj ) + (cos φi cos φj cos |δγ|)] ,
(6.1)
where φi and φj are the latitude of country i and country j, respectively, and
|δγ| denotes the absolute value of the difference in longitude between i and
j.2 To follow Tobler’s First Law of Geography, this distance is substituted
into a distance-decay function of the form:
1
wij
= (dij )−1 .
(6.2)
In the second matrix (W 2 ), we also take the type of political regime
1
A spatial weights matrix W is a N by N nonnegative matrix, which expresses for each
country (row) those countries (columns) that belong to its neighbourhood set as nonzero
elements. By convention, the diagonal elements of the weights matrix are set to zero,
since no country can be viewed as its own neighbour. For ease of interpretation, it is
common practice to normalize W such that the elements of each row sum to one. Since
W is nonnegative, this ensures that all weights can be interpreted as an averaging of
neighbouring values.
2
If simplified into a Cartesian space, latitude-longitude coordinates may correspond to
the vertical-horizontal axes. Using the coordinates of a country, we may construct a
contiguity structure by defining a ‘neighboring country’ as one lying within a particular
distance.
Geography and Governance: Does Space Matter?
99
2
Figure 6.1: Governance and Distance to Worst and Best Practices, 2005
FIN
DNK
NZL
NOR
SWENLDLUXCHE
CAN
AUS
AUT
IRL
SGP
GER
GBR
HKG
BEL
USA MLT
CHL
FRA PRT
BRB JPN
ESP BMU BHS LCA
VCT
KNA
EST
SVN
PRIANT
MAC
CYP
HUN
TWN
BWA
CZESVK
DMA
LTU
LVA
GRCATG CRI KOR
MUS
ITA
URY
WSM
POL
GRD ISR
ZAF
FSM
QAT
MYS
VUT
BRN
OMN
ARE
KWT
CPV
HRV BGR
KIR
BTN
BHR
TTO
BLZ
NAM
SYC
PAN
SUR
MDV
ROM TUN TURMNG
JOR GHA THA
MDG
MEX
LSO
JAM
FJI
SLV MLI BRASEN IND
TON
ARG
DOM ARM
STP
MAR
BFA
GUY
BEN
PHL
LKA
MOZ
MKD
SAU
SLB
GAB
MRT
UKR
NICGMB PER
TZA
MWI
SCG
ALB
BIH
LBN EGY
ZMB
CHN
VNM
HND COL
GEO
NER
MDA
SWZ
UGA
KAZ
GTM ECU BOLPRY
RUS DZA
IDN
KEN
PNGDJI
SLE
KHM
CMR
AZEGNB
LBY IRN
COM
CUB VEN
KGZ
SYR
BGD
YEMETHRWA
GIN PAK
BLR
LAOERI
TGO
NPL
TJK
COG AGO
NGA
TCD GNQ
CAF
BDI
LBR
TKM
PRK CIV
HTI
UZB
AFG
SDN
ZWE
ZAR
IRQ
-2
-1
0
1
ISL
SOM
0
.002
.004
.
.006
.008
2
(a) Distance to Best Practice
-2
-1
0
1
ISL
FIN
DNK
NZL
CHE
LUX
NOR
SWE
NLD
CAN
AUS AUT
IRL
SGP
GER GBR
HKG
BEL
USA
MLT
CHL
BRB BHS
JPNFRA
PRT
ESP
LCA
BMU
VCT
KNA
EST
SVN
PRI
ANT
MACTWNHUN
CYP
BWA
CZE
DMA
SVK
LTU
LVA
KOR
ATG
GRC
MUS
ITA
URY
CRI
WSM
POL
GRD
ZAF
ISR
FSM
QAT
MYS
VUT
BRN
OMN
ARE
KWT
CPV
HRV
KIR
BTNNAM BGR
BHR
TTO BLZ
SYC
PAN
MDV
SUR
THA
ROM
TUR
JOR
MNG
GHA
TUN
MDG
BRA
LSO
MEX
JAM
SEN
FJI
SLV
MLI
IND
TON ARG
DOM
STP BEN BFA
MAR
ARM
GUY
PHL
MOZ
MKD
SAUTZA LKA
SLB
GAB
MRT
UKR
NIC
GMB
PER
MWIEGY
ALB
SCG
BIH
COL HND
LBN GEO VNMNER
ZMB
CHN
MDA
SWZ
UGA
KAZ
DZA
BOL
RUS
IDN
KEN
ECU GTM
PRY
PNG
DJI
SLE
KHM
CMR
AZE
LBY
COM
GNB
VEN
CUB
KGZLAO
IRNPAK SYRBGD
RWA
YEM ETH
TGOBLR GIN
AGO
ERI
NPL
TJK
COG NGA
GNQ
BDI CAFTCD
LBR
TKM
PRK
HTI
CIV
SDN AFG ZWEUZB
ZAR
IRQ
SOM
0
.002
.004
.
(b) Distance to Worst Practice
.006
.008
100
Chapter 6
into account:
2
wij
= e−|ri −rj | (dij )−1 .
(6.3)
For this purpose, we use the regime indicator of Cheibub and Gandhi (2005).
This indicator r ranges from 0 to 5, where 0 represents a parliamentary
democracy, 1 a mixed democracy, 2 a presidential democracy, 3 a civilian
dictatorship, 4 a military dictatorship, and 5 a royal dictatorship. It is likely
that countries having a similar political regime are more sensitive to spillover effects and diffusion-adoption processes than countries having different
regimes.
Table 6.1 shows that Moran’s I statistic is greater than -0.005 with
highly positive z-values, while Geary’s c statistic is smaller than one with
highly negative z-values. These results indicate positive spatial dependence
of the governance index among countries. The correlation between the governance index g and the governance index weighted by the (political) distance to other countries (W 1 g or W 2 g) confirms this conclusion.
Table 6.1: Moran’s I and Geary’s c for Governance
Matrix
W1
W2
W1
W2
Correlation:
W 1 g vs g
W 2 g vs g
Moran’s I
0.170
0.430
Geary’s c
0.831
0.588
0.661
0.731
E(I)
-0.005
-0.005
E(c)
1.000
1.000
SD(I)
0.011
0.019
SD(c)
0.014
0.022
z-stat.
16.004
22.532
z-stat.
-12.058
-18.593
***
***
***
***
***
***
***: significant at 1%
6.2.2
Spatial Regression Models
In the previous section a series of evidence has indicated that governance is
spatially dependent. However, we are not certain about the type of spatial
dependence. In this section we further specify the type of spatial dependence. As pointed out by Anselin (1988, 2006), when specifying spatial de-
Geography and Governance: Does Space Matter?
101
pendence among the observations, the model may contain a spatially lagged
dependent variable, or the model may incorporate a spatially autoregressive
process in the error term. The first model is known as the spatial lag model
and the second as the spatial error model.
Formally, the spatial lag model is formulated as
g = ρW g + Xβ + ,
(6.4)
where g is an [n × 1] vector of the dependent variable (i.e., governance), X
is an [n × k] matrix consisting of the explanatory variables, ρ is the spatially
autoregressive parameter, β is the [k × 1] vector of parameters, and is an
[n × 1] vector of i.i.d. error terms.
There are two issues to be dealt with in estimating equation 6.4. First,
the presence of the spatially lagged dependent variable generates feedback
effects, because each country is also a neighbour of its neighbors. If gj
enters on the right-hand side of gi , gi also enters on the right-hand side of
gj . Second, the right-hand variable W g is correlated with the error term,
, as can be seen from a slight reformulation of model (4):
g = (I − ρW )−1 Xβ + (I − ρW )−1 .
(6.5)
Due to the spatial multiplier matrix (I −ρW )−1 , g in a particular country i not only depends on its own error term, but also on the error terms
of other countries. Hence, W g also depends on the error term of other
countries, as a result of which E[(W g)i i ] 6= 0. Estimating the spatial lag
model by OLS, called Spatial OLS (SOLS), will therefore not be consistent.
Franzese Jr. and Hays (2008) show that in case of positive (negative) spatial
dependence, SOLS will overestimate (underestimate) ρ and underestimate
(overestimate) β. By contrast, Maximum Likelihood (ML) estimation, taking into account the Jacobian term of the transformation from the error
term to the dependent variable |∂g/∂| = |I − ρW |, yields consistent and
efficient parameter estimates (Anselin, 1988, 2006).
102
Chapter 6
In the spatial error model, the error term of country i is taken to depend
on the error term of neighbouring countries according to the spatial weights
matrix W and an idiosyncratic component ξ, or formally:
g = Xβ + and = λW + ξ,
(6.6)
where λ is the spatial autocorrelation parameter and ξ is an [n × 1] vector
of i.i.d. error terms. This model is consistent with a situation where the
determinants of governance omitted from the model are spatially autocorrelated, and with a situation where unobserved shocks follow a spatial pattern.
Andrew (2003, 2005) calls these unobserved effects ‘common shocks’, covering a wide range of macroeconomic, political, and environmental shocks.
Although the OLS estimator of the response parameters of this model is unbiased, it is not efficient. ML estimation, taking into account the Jacobian
term of the transformation from the error term to the dependent variable
|∂g/∂| = |I − λW |, again solves this problem.3
6.2.3
Data
The explanatory variables of governance in our model have been selected on
the basis of the results of previous studies. Income per capita is included,
as higher incomes may increase demand for good governance (La Porta et
al., 1999; Kaufmann and Kraay, 2002). The data are drawn from the Penn
World Table (PWT) 6.2.4
The second variable is trade openness, defined as the log ratio of trade
and GDP, and is also drawn from the PWT. Several studies report that
openness influences governance (inter alia Bonaglia et al., 2001; Knack and
Azfar, 2003; Dollar and Kraay, 2003; Wei, 2000b). According to Rodrik
(2002: 4), “... a free trade regime is likely to reduce the corruption and
3
The spatial lag model and the spatial error model can also be estimated by generalized method-of-moments (GMM) (see Kelejian and Prucha, 1998; Anselin, 2006). Software packages to estimate spatial econometric models are Stata, Spacestat, Geoda (freely
downloadable) and Matlab (see www.spatial-econometrics.com).
4
http://pwt.econ.upenn.edu/php site/pwt index.php
Geography and Governance: Does Space Matter?
103
rent-seeking associated with trade interventions. Similarly, tariff bindings
... may generate greater predictability in incentives and solidify property
rights—two important attributes of a high-quality institutional framework.”
The third determinant of governance taken into account is legal origin.
Following La Porta et al. (1999), we consider three legal traditions: socialist, civil, and common law traditions that reflect the degree of state
involvement. Socialist Law is “a clear manifestation of the State’s intent
to create institutions to maintain its power and extract resources, without
much regard for protecting the economic interests or the liberties of the
population.” Civil Law tradition can be taken as “a proxy for an intent
to build institutions to further the power of the State, although not to the
same extent as in the socialist tradition.” Finally, Common Law tradition
can be taken as “a proxy for the intent to limit rather than strengthen the
State.” (La Porta et al., 1999: 231-232). We use an index ranging from 1
to 3, where 1 indicates the strongest state intervention (Communist Law),
3 the least state intervention (British Common Law), and Civil Law is in
between.
Fourth, we include the mostly-used geographic variable, namely absolute
latitude. This variable may explain cross-country variation in governance
for various reasons. Hall and Jones (1999: 101) argue that distance from
the equator is correlated with Western European influence which, in turn,
creates good institutions: “Western Europeans were more likely to settle in
areas that were broadly similar in climate to Western Europe, which again
points to regions far from the equator.”5 Other authors use this variable to
instrument for rule of law to disentangle the relationship between income,
institutions, integration, and geography (Dollar and Kraay, 2003; Rodrik et
al., 2004). The data are taken from the CIA World Factbook.6
Finally, we consider the fraction of the population with a Protestant religion drawn from Barro and McCleary (2005). A common view of Protestantism is that it is more egalitarian than other religious traditions. La
5
6
See also Acemoglu et al. (2001; 2002).
https://www.cia.gov/library/publications/the-world-factbook/
104
Chapter 6
Porta et al. (1997) find that more hierarchical religions are related to poor
governance. Similar results are reported by Treisman (2000) and Persson
and Tabellini (2003).
To enhance data availability and to avoid reverse causation problems,
data on these explanatory variables have been collected for the year 2000.
6.2.4
Results
This section reports our findings for cross-country differences in governance
in 2005.7 The first column of Table 6.2 shows the results of the OLS estimator applied to the model g = X β + without a spatially lagged dependent
variable or a spatially autocorrelated error term. This OLS model is used as
a benchmark. The classic indicators for a spatial lag model or for a spatial
error model are the Lagrange Multiplier (LM) tests, which may be calculated from the residuals of the OLS model and which follow a chi-squared
distribution with one degree of freedom (Anselin, 1988). Using these tests,
both the hypothesis of no spatially lagged dependent variable and the hypothesis of no spatially autocorrelated error term must be rejected at a one
per cent significance level.
The other columns of Table 6.2 show the results of different estimations
that all use the weights matrix W 1 . The second column of Table 2 shows
the results of the model extended to include a spatially lagged dependent
variable estimated by SOLS and the third column shows the results for
the model estimated by ML. The fourth column shows the results of the
model extended with a spatially autocorrelated error term estimated by ML.
Both the spatial autoregressive parameter ρ and the spatial autocorrelation
coefficient λ appear to be significant. According to the spatial lag model
estimated by ML, the quality of governance of a particular country increases
by 6 per cent if the quality of governance in surrounding countries increases
by 10 per cent. In the SOLS model the corresponding figure is about 7.5
per cent. Hence, as predicted, SOLS overestimates ρ by about 23 per cent
7
We also examined sub indexes of governance, but this did not change our overall conclusions. Results are available on request.
Geography and Governance: Does Space Matter?
105
compared to ML.
Table 6.2: Spatial Governance using W 1
Determinants
Ln GDP
per Capita
Ln Openness
Legal
Origin
Distance
to Equator
Fraction of
Protestant
Constant
OLS
0.449
(0.040)
0.241
(0.063)
0.226
(0.068)
0.011
(0.003)
0.908
(0.199)
-5.831
(0.343)
***
***
***
***
***
***
ρ
SOLS
0.384 ***
(0.043)
0.216 ***
(0.062)
0.251 ***
(0.067)
0.009 ***
(0.003)
0.707 ***
(0.202)
-4.903 ***
(0.430)
0.747 ***
(0.219)
MLSL
0.397
(0.041)
0.221
(0.061)
0.246
(0.06)
0.009
(0.003)
0.745
(0.196)
-5.231
(0.372)
0.606
(0.179)
λ
(Pseudo) R2
Log Likelihood
Spatial Lag:
LMρ
LMρr
Spatial Error:
LMλ
LMλr
0.719
0.736
12.358
4.081
***
**
10.475
2.198
***
Standard errors in brackets;
∗∗∗ , ∗∗ ,
0.730
-125.626
MLSE
0.435
(0.042)
0.235
(0.062)
0.243
(0.068)
0.012
(0.003)
0.797
(0.209)
-5.714
(0.390)
***
***
***
***
***
***
***
***
***
***
***
***
***
0.791
(0.183)
0.719
-126.478
***
and ∗ : significant at 1%, 5%, and 10%
To find out whether the spatial lag model or the spatial error model is
more appropriate to describe the data, we use the robust LM-tests, LMρr
and LMλr , proposed by Anselin et al. (1996). These tests are robust in
the sense that the existence of one type of spatial dependence does not
bias the test for the other type of spatial dependence. The results show
that the hypothesis of no spatially lagged dependent variable must still be
rejected at 5 per cent significance. However, the hypothesis of no spatially
autocorrelated error term can no longer be rejected. This indicates that
the spatial lag model is more appropriate. This finding is also consistent
106
Chapter 6
with the fact that the value of the log-likelihood of the spatial lag model is
greater than that of the spatial error model, −125.63 versus −126.48.
In line with previous studies, the coefficients of the explanatory variables
in the OLS model, the spatial lag model and the spatial error model are
significantly different from zero and have the expected signs. Higher income
and trade openness promote good governance. A legal system supporting
state intervention leads to poor governance. Countries situated far from
the equator tend to have better governance. Finally, a larger fraction of
Protestant population is related to good governance. Nevertheless, since
the spatial lag model is found to be more appropriate than the spatial error
model, the coefficients of the explanatory variables in the OLS model are
biased. The most affected variable is distance to the equator. In the spatial
lag model estimated by ML its coefficient is 0.009 and in the OLS model it
is 0.011. This means that the latter coefficient is overestimated by 25 per
cent. Similarly, the coefficient of GDP per capita is overestimated by 13 per
cent, of openness by 9 per cent, of Protestantism by 22 per cent, while the
coefficient of legal origin is underestimated by 8 per cent.
We now examine whether our conclusions are sensitive to the choice
of the spatial weights matrix. Table 6.3 reports the estimation results for
the spatial weights matrix W 2 that also captures similarity in the political
regime between countries. In general, the results are in line with those of
Table 6.2, the difference being that the spatial autoregressive parameter ρ
in the spatial lag model falls from 0.61 to 0.49.8 It shows that the degree of
interaction decreases if political regime similarity is made part of the spatial
weights matrix.
Up to this point, the spatial weights matrices have been row-normalized.
However, row normalization may be criticized. If, for example, an inverse
distance matrix is row normalized, it will lose its economic interpretation of
distance decay (Anselin, 1988). There are two reasons for this. First, due
8
We also ran regressions for an adjusted governance index excluding voice and accountability, since to some extent it is related to the political regime. Again, we found that the
spatial error model must be rejected in favour of the spatial lag model.
Geography and Governance: Does Space Matter?
107
to row normalization the spatial weights matrix may become asymmetric,
as a result of which the impact of country i on country j is not the same as
that of country j on country i. Second, due to row normalization remote
countries will have the same impact on all other countries in the sample as
core countries.
Table 6.3: Spatial Governance using W 2
Determinants
Ln GDP
per Capita
Ln Openess
Legal
Origin
Distance
to Equator
Fraction of
Protestant
Constant
ρ
SOLS
0.369 ***
(0.039)
0.198 ***
(0.058)
0.176 ***
(0.063)
0.006 ***
(0.003)
0.568 ***
(0.191)
-4.630 ***
(0.371)
0.523 ***
(0.087)
MLSL
0.374 ***
(0.038)
0.201 ***
(0.057)
0.179 ***
(0.062)
0.006 ***
(0.003)
0.590 ***
(0.187)
-4.710 ***
(0.360)
0.488 ***
(0.081)
λ
(Pseudo) R2
Log Likelihood
Spatial Error:
LMλ
LMλr
Spatial Lag:
LMρ
LMρr
0.719
0.760
-114.410
44.439
15.770
***
***
48.313
19.645
***
***
MLSE
0.402 ***
(0.041)
0.220 ***
(0.058)
0.221 ***
(0.067)
0.008 ***
(0.004)
0.640 ***
(0.206)
-5.225 ***
(0.417)
0.826
(0.104)
0.716
-116.216
***
Standard errors in brackets; ***, **, and * are significant at 1%, 5%, and 10%
LM statistics are based on OLS residuals.
Following Elhorst (2001), we therefore also consider a normalization
procedure where each element of W is divided by its largest eigenvalue
∗ = w /ω
ωmax : wij
ij
max . This has the effect that the eigenvalues of W are
also divided by ωmax , as a result of which the largest eigenvalue of the
108
Chapter 6
matrix W ∗ equals one, just like the largest eigenvalue of a row normalized
matrix, but without changing the mutual proportions between the elements
of W .
Table 6.4 reports the estimation results of the spatial lag model using
the largest-eigenvalue-corrected spatial weights matrices (W 1∗ and W 2∗ ).
The robust LM statistics based on the OLS residuals again indicate that the
spatial lag model is the preferred specification.9 The coefficient estimates
of the explanatory variables using the largest-eigenvalue-corrected spatial
weights matrices are also very close to those found for row-normalized spatial
weights matrices. However, the spatial autoregressive parameter ρ increases
from 0.61 to 0.68 when using W 1∗ and from 0.49 to 0.63 when using W 2∗ .
In sum, we may conclude that governance in one country is significantly
related to that in surrounding countries and that the degree of interaction is
approximately 0.63 to 0.68, depending on whether or not political similarity
is taken into account.
6.3
6.3.1
Spatial Heterogeneity
Local Statistics
In this section we test for spatial heterogeneity. We first provide local
indicators of spatial association, local Moran’s Ii and local Geary’s ci (i =
1, ..., n), in order to identify the contribution of specific locations to the
overall pattern of spatial dependence using the spatial weights matrices
W 1 and W 2 (Anselin, 1995). Figures 6.2 and 6.3 graph these statistics
for the observations arranged in alphabetical order (the first observation is
Afghanistan and the last one is Zimbabwe). Values for Ii greater (smaller)
than -0.005 and for ci smaller (greater) than 1 indicate positive (negative)
local spatial dependence, i.e., a clustering of countries with a (dis)similar
governance index around country i. Furthermore, values for Ii above (below)
the solid line and for ci below (above) the solid line indicate countries of
9
Also, the exclusion of voice and accountability from the overall governance index does
not change the conclusion that the spatial lag model is the preferred specification.
Geography and Governance: Does Space Matter?
109
which the degree of spatial dependence is greater (smaller) than its global
counterpart (see Table 6.1).
From Figures 6.2 and 6.3 it can be seen that local Moran’s I of a wide
range of countries points to positive spatial dependence: 71% in case of W 1
and 77% in case of W 2 . Similar results are obtained for local Geary’s c: 78%
in case of W 1 and 88% in case of W 2 . Furthermore, it can be seen that most
observations lie within the range of I¯ ± 2σ 2 (Ii ) and c̄ ± 2σ 2 (ci ). Statistical
outliers may be considered as ‘hotspots’. For example, 10 ‘hotspots’ with
a statistically significant contribution to global Moran’s I of 0.17 based
on W 1 are found in Western and Northern Europe, namely Luxembourg
(observation 101) with a local Moran’s I of 1.19, the Netherlands (120; 1.10),
Belgium (16; 0.97, Switzerland (163; 0.96), Germany (63; 0.96), Denmark
(46; 0.94), Norway (126; 0.92), Sweden (162; 0.89), the UK (178; 0.81), and
Austria (9; 0.77). The list of countries is provided in Appendix 4b.
***
***
***
20.130
8.145
14.630
2.645
Matrix
SOLS
0.384 ***
(0.042)
0.205 ***
(0.062)
0.252 ***
(0.066)
0.008 ***
(0.003)
0.704 ***
(0.197)
-5.049 ***
(0.380)
0.824 ***
(0.200)
0.743
W 1∗
MLSL
0.395 ***
(0.040)
0.211 ***
(0.060)
0.247 ***
(0.064)
0.009 ***
(0.003)
0.740 ***
(0.192)
-5.186 ***
(0.357)
0.680 ***
(0.161)
0.742
-123.339
39.069
14.506
44.232
19.670
***
***
***
***
Matrix
SOLS
0.381 ***
(0.039)
0.180 ***
(0.060)
0.202 ***
(0.063)
0.007 ***
(0.003)
0.620 ***
(0.191)
-4.799 ***
(0.364)
0.629 ***
(0.110)
0.762
Standard errors in brackets; ***, **, and * are significant at 1%, 5%, and 10%
LM statistics are based on OLS residuals.
(Pseudo) R2
Log Likelihood
Spatial Lag:
LMρ
LMρr
Spatial Error:
LMλ
LMλr
ρ
Legal
Origin
Distance
to Equator
Fraction of
Protestant
Constant
Ln GDP
per Capita
Ln Openness
Determinants
W 2∗
MLSL
0.386 ***
(0.038)
0.184 ***
(0.058)
0.203 ***
(0.062)
0.007 ***
(0.003)
0.639 ***
(0.186)
-4.870 ***
(0.352)
0.586 ***
(0.102)
0.762
-115.922
Table 6.4: Normalization by Largest Eigenvalue (W 1∗ and W 2∗ )
110
Chapter 6
-0.5
-1.0
-0.5
-1.0
4.0
3.0
2.0
1.0
0.0
4.0
3.0
2.0
1.0
0.0
187
5.0
5.0
125
6.0
6.0
63
7.0
7.0
1
0.0
0.0
1
63
1
2
125
1
63
125
Figure 6.3: Local Geary’s c Statistics (Matrices W and W )
0.5
0.5
187
1.0
1.0
125
1.5
1.5
63
2.0
2.0
1
2.5
2.5
Figure 6.2: Local Moran’s I Statistics (Matrices W 1 and W 2 )
187
187
Geography and Governance: Does Space Matter?
111
112
Chapter 6
6.3.2
Geographically Weighted Regression
In the models estimated in Section 6.2 it has been assumed that the coefficients of the determinants of governance are the same for all observations.
In this section we examine whether the determinants of governance vary
over space. In other words, we analyze the issue of spatial non-stationarity
defined as “the variation in relationships and processes over space” (Brunsdon et al., 1999: 497). For this purpose, we apply a geographically weighted
regression (GWR) technique of Brunsdon et al. (1996, 1999) that allows for
variability in the parameters (see also LeSage, 2004).
The model resembles the standard OLS model, but with varying parameter coefficients; that is
gi =
X
X ij βij (φi , `i ) + i ,
(i = 1, 2, . . . , N ),
(6.7)
j
where βij (φi , `i ) is the parameter of the j-th explanatory variable of observation i, located at (φi , `i ) in a geographical space.10 The inclusion of the
index i implies that equation 6.7 is not a single equation, but a set of n
equations where the dimensions of β are n × J. This results in a set of localized regression estimates, where each observation is given a certain weight
such that neighboring countries have more influence on the parameters than
those located farther away. If ϑik is defined as the weight of the k-th observation in predicting the i-th observation, constructed on the basis of the
Euclidean distance dik between i and k, and arranged as the k th diagonal
element of a diagonal matrix11 , we have
ϑik = e(−dik /h2 )
(6.8)
where h is the kernel bandwith calibrated by minimizing the cross-validated
10
In our case, the location is determined via the country’s longitude and latitude. We
locate the observations in such a way that all countries lie in the positive-positive (EastNorth) Cartesian quadrant.
11
Since local areas are relatively small, it is not necessary to use the great circle distance.
Geography and Governance: Does Space Matter?
113
sum of squared errors.
Figure 6.4 shows the local parameter estimates of the determinants of
governance where the vertical and horizontal axes are, respectively, the parameter estimates and the country code (see Appendix 4b). Although the
means of these local parameter estimates are close to their counterparts
reported in Tables 6.2, 6.3 and 6.4, there is substantial heterogeneity of
the local parameters across observations. Whereas the local parameters of
openness, legal origin and distance to the equator are still closely scattered
around their global parameters (the solid horizontal lines passing through
the plots), those of the constant term, income and Protestant religion are
widely dispersed.
A closer look at Figure 6.4 reveals which countries are responsible for
this result. The local coefficient of trade openness is found to deviate from
its global value for Samoa (observation 141 with coefficient 0.58) and Tonga
(170; 0.57), as well as for Fiji (57; -0.46), Kiribati (89; -0.16), New Zealand
(122; -0.70) and Vanuatu (182; -0.23). Interestingly, these countries are all
small islands situated in the ‘tip’ of the earth.
The same small island countries are also identified as spatial outliers for
the income variable. A highly positive income effect is found for Samoa
(141; 1.06) and Tonga (170; 1.05), while a highly negative income effect is
found for Fiji (57; 0.02), Vanuatu (182; 012), Kiribati (89; 0.13) and New
Zealand (122; -0.05). The same countries are also spatial outliers for the
local parameters of the other explanatory variables.
Table 6.5 reports the results of test statistics whether the local parameters are significantly different from their global values, based on 1000 Monte
Carlo simulations. In line with Figure 6.4, this table illustrates that the
parameters of three variables vary significantly over space at one per cent
significance, namely the constant term, income and Protestant religion. The
Monte Carlo simulation based bandwidth test indicates that GWR outperforms the global linear regression model with coefficients that are homogenous across space.
125
187
-0.02
-0.01
0.00
0.01
0.02
0.03
0.04
0.05
63
125
(d) Absolute Latitude
1
(a) Income
187
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
-2.0
-0.8
63
-0.2
1
-0.6
-0.4
-0.2
0.0
0.2
0.0
0.2
0.4
0.6
0.4
0.6
1.0
0.8
0.8
1.2
1
1
125
125
(e) Protestant
63
(b) Openess
63
187
187
Figure 6.4: Local Coefficients
-12.0
-10.0
-8.0
-6.0
-4.0
-2.0
0.0
2.0
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
1
1
125
125
(f) Constant
63
(c) Legal Origin
63
187
187
114
Chapter 6
Geography and Governance: Does Space Matter?
115
Table 6.5: Significance Tests
Independent
Variables
β̄j
βjmin
Significance Test for Non-Stationarity
Ln GDP per capita
0.432
-0.047
Ln Openness
0.287
-0.705
Legal Origin
0.246
-0.611
Absolute Latitude
0.013
-0.011
Fraction of Protestant
0.798
-1.301
Constant
-5.976 -10.617
Significance Test for Bandwidth
Bandwidth
37.603
Outliers are defined as those outside a range
βjmax
Non-Statry.Test
σβ2j
Sign.
1.055
0.577
1.136
0.045
2.962
0.681
0.147
0.134
0.178
0.007
0.785
1.520
***
***
***
Per cent
Outliers
3.21
3.74
4.28
4.81
3.74
4.28
***
of β¯j ± 2σβ2j
Summing up, the relationship between governance and its determinants
is not homogeneous over space but instead varies over different countries.
6.4
Conclusion
We have found a strong relationship between governance similarity and locational similarity. Countries are clustered according to their distance to the
world best (worst) practice in governance. Statistics used to test for global
spatial dependence confirm these results: well (poorly) governed countries
are found to be located near other well (poorly) governed countries.
Our econometric analysis of cross-country differences in governance, in
which we controlled for GDP per capita, openness, legal origin, distance to
the equator and Protestant religion, has also shown that the classic OLS
model must be rejected in favor of a model extended to include governance
observed in neighboring countries weighted by an inverse distance matrix.
This conclusion also holds when we employ a spatial weights matrix that
captures similarity in the political regime between countries or when the
spatial weights matrix is normalized by dividing all elements by its largest
eigenvalue rather than normalizing the elements such that the rows sum
to one. Overall, the degree of interaction of the governance index among
countries appeared to be approximately 0.63 to 0.68, depending on whether
116
Chapter 6
or not political similarity is taken into account.
Finally, our findings have shown that the relationship between governance and its determinants is not homogeneous over space but instead varies
over different countries. In other words, both global and local spatial dependencies systematically colour world governance.
Chapter 7
Putting the Pieces Together
“The world is round
and the place which may seem like the end
may also be the beginning.”
————————————————————
Ivy Baker Priest (1905–1975)
7.1
Conclusions and Limitations of the Research
In this thesis we have argued that corruption should not be seen from a normative perspective, but must be viewed from a wider socio-political context.
Thus, a definition of corruption must consider discretionary power, economic
rents, and the judicial system (Jain, 2001). Modifying Waterbury (1973)
and Jain (2001), we therefore may define corruption as private rent-seeking
behavior using the discretionary public power that contravenes the judicial
system. This definition opens a possibility for cross-country comparison.
Also, various forms of corruption can be identified, ranging from the minor
abuse of influence to institutionalized bribery and systematic kleptocracy.
Through corruption perception-based indexes, scholars are able to compare cross-country differences in corruption. Once such a comparison can
be made, the incidence of corruption can be also analyzed using a set of
economic and non-economic variables.
In the literature, corruption has been attributed to a low level of development —including a low level of income, an unequal income distribution,
and a low level of education—and to restricted economic openness, a large
118
Chapter 7
population size and a high rate of population growth. Scholars have also
identified a high correlation between corruption and aspects of the political
system in place, including the regime type, election procedures and the organization of political parties, political stability, and the governmental system.
The bureaucratic environment that includes remuneration, the recruitment
system, and the quality of the bureaucracy, is also considered as an underlying factor of corruption. Finally, the quality of law enforcement, religious
and cultural characteristics, as well as a country’s geographical position and
its natural resources have been claimed to be related to corruption.
From this long list of attributes, however, we only identified two variables that are not sensitive to changes in corruption measurement, model
specification, as well as the number and composition of observations used
in the regressions. The robust variables in this sense are government effectiveness and rule of law. To recall, the former represents “the quality of
public service provision, the quality of the bureaucracy, the competence of
civil servants, the independence of the civil service from political pressures,
and the credibility of the governments commitment to policies.” The later
captures mainly “the extent to which agents have confidence in and abide
by the rules of society, perceptions of the incidence of crime, the effectiveness and predictability of the judiciary, and the enforceability of contracts,
the success of a society in developing an environment in which fair and
predictable rules form the basis for economic and social interactions, and
importantly, the extent to which property rights are protected” (Kaufmann,
et al., 2002: 177-178). We find no other variable that can be regarded as
robust as these two variables.
This finding implies that corruption is deeply rooted in poor governance;
or, stated differently, corruption and poor governance are like two sides of
coin. The literature indicates that the cost of poor governance is very expensive, ranging from a low public infrastructure quality, high rates of infant
and child mortality, a high level of illiteracy, to income inequality. We study
the link between governance and one economic growth. We construct a new
index of governance applying factor analysis to the International Country
Putting the Pieces Together
119
Risk Guide (ICRG) dataset on indicators of governance, i.e. democratic
accountability, government stability, bureaucracy quality, corruption, and
rule of law. We focus on ICRG data as these are the only governance indicators available for a long time span and a large group of countries. In other
words, these data allow us to examine causality. Factor analysis is an ideal
instrument to analyse to what extent the different dimensions of governance
as identified in the ICRG data set contain the same information. We find
that the various dimensions can be combined into one governance index.
Next, we test whether our index is related to growth in varying samples
of countries and different sets of conditioning variables. We find that our
index is positively related to economic growth. This result is fairly robust
across different samples of countries and model specifications.
Today, the majority of nations are suffering from corruption. Various scholars pose that there is an association between the prevalence and
the persistence of corruption. As Mishra (2006) argues, when there are
many corrupt agents, it is optimal to be corrupt; thus once corruption is
widespread, it will be persistent. Mauro (2004) also indicates that in countries where corruption is pervasive, there is no incentive for individuals to
combat corruption and make it immutable.
This view, however, is not entirely correct. Using 20-year span of data,
we find that only few countries stagnated at the same level of corruption.
However, there are two groups of country signifying the change in countries’
corruption levels: those showing an improvement and those facing a deterioration in their levels of corruption. In other words, there is a tendency
toward a convergence through upward and downward dynamics. So in the
long run, corruption is not persistent. Our analysis shows that research
should not focus on the explanation of corruption persistence, but should
be redirected towards explaining the process of corruption convergence.
Finally, one may question whether there is a systematic spatial pattern
in the distribution of governance around the globe. In the final study of this
thesis we investigate governance from a spatial point of view. We discover a
positive spatial correlation where poorly (well) governed countries are spa-
120
Chapter 7
tially clustered with those performing the same quality of governance. A
further investigation indicates that a country’s quality of governance is determined not only by its domestic characteristics, but also its neighbors’
quality of governance. This conclusion remains whether we take only geographical distance or a combination of geographical and ‘political’ distance
into account.
Along with the systematic spatial pattern of governance, we also find
that the effect of governance determinants is not constant over space. In
some countries, the income effect of governance is high, but it is low in
others. The same applies to the fraction of population that is protestant (a
variable commonly argued to represent a society with an egalitarian structure). We also find that the model considering spatial heterogeneity outperforms the global linear regression model where the effects of governance
determinants are homogenous across space.
This thesis has some limitations. Our survey of the corruption determinants is based on a qualitative approach, whereas with the development of
meta analysis technique, a quantitative approach now is possible. Applying
this technique to the literature discussed in Chapter 2 is a topic on our
future research agenda.
As regards the sensitivity analysis, we ignore the possible interaction
effects among the long list of the corruption determinants. Also, we do not
pay further attention to the indirect mechanism among the determinants.
As we find that government effectiveness and rule of law are important in
explaining corruption, it would be interesting to investigate factors explaining these variables. Similar issues apply to our study on the growth impact
of governance.
In the study on corruption persistence we find that corruption changes
over time. It therefore is interesting to analyze factors that influence this
change. Finally, in the study of the spatial dimension of governance, we limit
our investigation to geographical distance and cross-country differences in
domestic politics. The influence of international politics, for example, is
still missing from our study.
Putting the Pieces Together
7.2
121
Policy Implications
Some policy implications can be derived from the findings in this thesis.
First, corruption is not a curse passing on generations. Our study discovers
that corruption changes over time. It is evident that even though many
clean countries are found to become less clean, many corrupt countries have
been able to reduce their levels of corruption. So our research is good news
for countries that want to reduce corruption: it takes time, but it is possible.
The question now is that where should we start? Two starting points
have been indicated in this thesis, namely an improvement in the quality of
government and in the quality of the rule of law. Our finding is that these
two variables are robustly correlated with corruption in all circumstances.
As a logical consequence, bureaucratic and judicial reforms can be seen as
the main items on the anti-corruption policy agenda. Bureaucratic reform
may start from an adjustment of the pay structure.1 In corrupt countries
public officials are paid less than their counterparts in the private sector.
Also, in these countries bureaucrats with discretionary power are paid at
the same rate with those without such power.2 Such a structure creates an
incentive for corruption to occur.
Yet, bureaucratic reform is only one step to combat corruption. The
other step is judicial reform with two related dimensions. First are measures
that reduce the possibility of corruption to occur and increase the likelihood
to detect corrupt acts. Second are measures to penalize corrupt acts not
as an ordinary crime but an extraordinary crime. The former refers to an
improvement in the rule of law system, the latter concerns the enforcement
of the law. In short, the main objective of bureaucratic and legal reforms is
1
Certainly, a bureaucratic reform covers a wide range of change in bureaucratic system.
Here we focus on the pay structure because of two reasons. First, corruption relates
to private wealth seeking behavior using discretionary power, thus the reform in pay
structure is directed to the heart of the problem. Second, the reform in pay structure may
be followed by other related reforms in the bureaucratic system such as the recruitment
system, career path, etc.
2
Pensions also still do not distinguish between those who retire in good order and those
who do not.
122
Chapter 7
to find a combination of carrot (a system of incentives) and stick (prevention
and sanctions) that minimizes the level of corruption.
Finally, as neighbors matter for good governance, regional cooperation
should also be put on the agenda to combat corruption. On the one hand,
good governance cannot be seen merely as a national issue with domesticspecific characteristics, because it has encompassed national borders to become a regional issue. On the other hand, some international policy issues
have been initiated world wide, but they may be too general and may ignore
regional-specific characteristics. Our study implies that special attention
has to be paid to policies and actions at the regional level. At this level
policies and actions may tackle some shortcomings that cannot be handled
at the national level (e.g., a bunch of domestic constraints faced by countries individually) and at the international level (i.e., a one-size-fits-for-all
prescription) as well.
Bibliography
Abed, George T. and Hamid R. Davoodi. 2000, ‘Corruption, Structural
Reforms, and Economic Performance in the Transition Economies’,
IMF Working Paper 132.
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001, ‘The
Colonial Origins of Comparative Development: An Empirical Investigation’, The American Economic Review, 91(5): 1369–1401.
Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2002, ‘Reversal of Fortune: Geography and Institutions in the Making of the
Modern World Income Distribution’, The Quarterly Journal of Economics, 117(4): 1231–1294.
Ades, Alberto and Rafael Di Tella. 1997. ‘The New Economics of Corruption: A Survey and Some New Results’, Political Studies, 45(3):
496–515.
Ades, Alberto and Rafael Di Tella. 1999, ‘Rents, Competition, and Corruption’, American Economic Review, 89(4): 982–992.
Adserá, Alicia, Carles Boix and Mark Payne. 2003, ‘Are You Being Served?
Political Accountability and Quality of Government’, Journal of Law,
Economics and Organization, 19(2): 445–490.
Aidt, Toke S. 2003, ‘Economic Analysis of Corruption: A Survey’, The
Economic Journal 113(491): F632F652.
Aish, Anne-Maria and Karl G. Jöreskog. 1990, ‘A Panel Model for Political Efficacy and Responsiveness: An Application of LISREL 7 with
Weighted Least Square’, Quality and Quantity, 24(4): 405–426.
124
Alesina, Alberto and Beatrice Weder. 2002, ‘Do Corrupt Governments
Receive Less Foreign Aid?’, American Economic Review, 92(4): 1126–
1137.
Alesina, Alberto and Roberto Perotti. 1996, ‘Income Distribution, Political
Instability, and Investment’, European Economic Review 40(6): 1203–
1228.
Ali, M. Abdiweli and Hodan Said Isse. 2003, ‘Determinants of Economic
Corruption: A Cross-Country Comparison’, Cato Journal, 22(3): 449–
466.
Alt, James E. and David Dreyer Lassen. 2003, ‘The Political Economy of
Corruption in American States’, Journal of Theoretical Politics, 15(3):
341–365.
Amundsen, Inge. 1999, ‘Political Corruption: An Introduction to the
Issues’, Chr. Michelsen Institute Development Studies and Human
Rights Working Paper 7.
Anderson, James E. and Douglas Marcouiller. 2002, ‘Insecurity and the
Pattern of Trade: An Empirical Investigation’, Review of Economics
and Statistics, 84(2): 342–352.
Andrews, Donald W. K. 2003, ‘Cross-section Regression with Common
Shocks’, Cowles Foundation Discussion Paper 1428.
Andrews, Donald W. K. 2005, ‘Cross-section Regression with Common
Shocks’, Econometrica, 73(5): 1551–1585.
Andvig, Jens Chr. 2005. ‘A House of Straw, Sticks or Bricks? Some Notes
on Corruption Empirics’, Paper presented at the Fourth Global Forum
on Fighting Corruption and Safeguarding Integrity.
Andvig, Jens Chr., Odd-Helge Fjeldstad, Inge Amundsen, Tone Sissener,
and Tina Søreide. 2000. ‘Research on Corruption: A Policy Oriented Survey’, Final Report, Chr. Michelsen Institute and Norwegian
Institute of International Affairs.
Anselin, Luc, Anil K. Bera, Raymond Florax, Maan J. Yoon. 1996, ‘Simple
Diagnostic Tests for Spatial Dependence’, Regional Science and Urban
Economics, 26(1): 77–104.
Anselin, Luc. 1988, Spatial Econometrics: Methods and Models, Kluwer
Academic Publishers.
Bibliography
125
Anselin, Luc. 1995, ‘Local Indicators of Spatial Association-LISA’, Geographical Analysis, 27(2): 93–115.
Anselin, Luc. 2006, ‘Spatial Econometrics’ in Palgrave Handbook of Econometrics, Vol. 1, Terence C. Mills and Kerry Patterson (eds.), Palgrave
McMillan, Basingstoke: 901–969.
Aslaksen, Silje. 2007. ‘Corruption and Oil: Evidence from Panel Data’,
Mimeo, Norwegian University of Science and Technology.
Baker, Raymond W. 2005, Capitalism’s Achilles Heel: Dirty Money and
How to Renew the Free-Market System, John Wiley and Sons, Inc.
Bardhan, Pranab. 1997, ‘Corruption and Development: A Review of Issues’, Journal of Economic Literature, 35(3): 1320–1346.
Barro, Robert J. 1997, Determinants of Economic Growth: A Cross–
Country Empirical Study., The MIT Press.
Barro, Robert J. and Rachel M. McCleary. 2005. Which Countries Have
State Religions?, Mimeo, Harvard University.
Barro, Robert J. and Xavier Sala-i-Martin. 1992, ‘Convergence’, Journal
of Political Economy, 100(2): 223–251.
Barro, Robert J. and Xavier Sala-i-Martin. 1995, Economic Growth, McGraw–Hill.
Basu, Kaushik, Sudipto Bhattacharya, and Ajit Mishra. 1992, ‘Notes on
Bribery and the Control of Corruption’, Journal of Public Economics,
48(3): 349–359.
Bertrand, Marianne and Sendhil Mullainathan. 2001, ‘Do People Mean
What They Say? Implications for Subjective Survey Data’, The American Economic Review Papers and Proceedings of the Hundred Thirteenth Annual Meeting of the American Economic Association, 91(2):
67–72.
Beugelsdijk, Sjoerd, Henri L. F. de Groot, and Anton B.T.M. van Schaik.
2004, ‘Trust and Economic Growth: A Robustness Analysis’, Oxford
Economic Papers, 56(1): 118–134.
Bickenbach, Frank and Eckhardt Bode. 2003, ‘Evaluating the Markov
Property in Studies of Economic Convergence’, International Regional
Science Review, 26(3): 363–392.
126
Blackburn, Keith, Niloy Bose, and M. Emranul Haque. 2006, ‘The Incidence and Persistence of Corruption in Economic Development’, Journal of Economic Dynamics and Control, 30(12): 2447–2467.
Bollen, Kenneth A. 1989, Structural Equations with Latent Variables, Wiley.
Bonaglia, Federico, Jorge Braga de Macedo, and Maurizio Bussolo. 2001,
‘How Globalization Improves Governance’, OECD Development Centre Working Paper 181.
Borrmann, Axel, Matthias Busse, and Silke Neuhaus. 2006, ‘Institutional
Quality and the Gains from Trade’, Kyklos, 59(3): 345–368.
Braun, Miguel and Rafael Di Tella. 2004, ‘Inflation, Inflation Variability,
and Corruption’, Economics and Politics, 16(1): 77–100.
Broadman, Harry G. and Francesca Recanatini. 2001, ‘Seeds of Corruption: Do Market Institutions Matter?’, MOCT-MOST: Economic Policy in Transitional Economies, 11(4): 359–392.
Broadman, Harry G. and Francesca Recanatini. 2002, ‘Corruption and
Policy: Back to the Roots’, The Journal of Policy Reform, 5(1): 37–
49.
Brown, David S., Michael Touchton, and Andrew B. Whitford. 2006, ‘Political Polarization as a Constraint on Government: Evidence from
Corruption’, Mimeo, University of Colorado.
Brueckner, Jan K. 2003, ‘Strategic Interaction among Governments: An
Overview of Empirical Studies’, International Regional Science Review, 26(2): 175–188.
Brunetti, Aymo and Beatrice Weder. 2003, ‘A Free Press is Bad News for
Corruption’. Journal of Public Economics 87(7): 1801–1824.
Brunsdon, Chris, Muraray Aitkin, Stewart Fotheringham, and Martin
Charlton. 1999, ‘A Comparison of Random Coefficient Modelling and
Geographically Weighted Regression for Spatially Non-Stationary Regression Problems’, Geographical and Environmental Modelling, 3(1):
47–62.
Brunsdon, Chris, Stewart Fotheringham, and Martin Charlton. 1996, ‘Geographically Weighted Regression: A Method for Exploring Spatial
Nonstationarity’, Geographical Analysis, 28(4): 281–298.
Bibliography
127
Chang, Eric C. C. and Miriam A. Golden. 2007, ‘Electoral Systems, District Magnitude and Corruption’, British Journal of Political Science,
37(1): 115–137.
Chatfield, C. 1995, ‘Model Uncertainty, Data Mining, and Statistical Inference’, Journal of the Royal Statistical Society Series A, 158(3): 419–
466.
Cheibub, José Antonio and Jennifer Gandhi. 2005, ‘Classifying Political Regimes: A Six-Fold Measure of Democracies and Dictatorships’,
Mimeo, Yale University.
Chong, Alberto and Cesar Calderon. 2000, ‘On the Causality and Feedback
between Institutional Measures and Economic Growth’, Economics
and Politics, 12(1): 69–82.
D’Agostino, Ralph B., Albert Belanger, and Ralph B. D’Agostino, Jr.
1990, ‘A Suggestion for Using Powerful and Informative Tests for Normality’, The American Statistician, 44(4): 316–321.
Damania, Richard, Per G. Fredriksson, and Muthukumara Mani. 2004,
‘The Persistence of Corruption and Regulatory Compliance Failures:
Theory and Evidence’, Public Choice, 121(3-4): 363–390.
de Sardan, J. P. Oliver. 1999, ‘A Moral Economy of Corruption in Africa?’,
The Journal of Modern African Studies, 37(1): 25–52.
Dobel, J. Patrick. 1978, ‘The Corruption of a State’, The American Political Science Review, 72(3): 958–973.
Dollar, David and Aart Kraay. 2003,‘Institutions, Trade, and Growth’,
Journal of Monetary Economics, 50(1): 133–162.
Donchev, Dilyan and Gergely Ujhelyi. 2007, ‘Do Corruption Indices Measure Corruption?’, Mimeo, Harvard University.
Draper, David. 1995, ‘Assessment and Propagation of Model Uncertainty’,
Journal of the Royal Statistical Society. Series B, 57(1): 45–97
Easterly, William and Ross Levine. 2003, ‘Tropics, Germs, and Crops:
How Endowments Influence Economic Development’, Journal of Monetary Economics, 50(1): 3–39.
Elhorst, J. Paul. 2001, ‘Dynamic Models in Space and Time’, Geographical
Analysis, 33(2): 119–140.
128
Fingleton, Bernard. 1997, ‘Specification and Testing of Markov Chain
Models: An Application to Convergence in the European Union’, Oxford Bulletin of Economics and Statistics, 59(3): 385–403.
Fingleton, Bernard. 1999, ‘Estimates of Time to Economic Convergence:
An Analysis of Regions of the European Union’, Regional Science
Review, 22(1): 5–34.
Fisman, Raymond and Edward Miguel. 2006, ‘Cultures of Corruption:
Evidence from Diplomatic Parking Tickets’, NBER Working Paper
12312.
Fisman, Raymond J. and Roberta Gatti. 2002, ‘Decentralization and Corruption: Evidence across Countries’, Journal of Public Economics
83(3): 325–345.
Fréchette, Guillaume R. 2006, ‘A Panel Data Analysis of the Time-Varying
Determinants of Corruption’, CIRANO Working Paper 28.
Franzese, Jr., Robert J., and Jude C. Hays. 2008, ‘Empirical Models of
Spatial Interdependence’ in The Oxford Handbook of Political Methodology, Janet M. Box-Steffensmeier, Henry E. Brady and David Collier
(eds.), Oxford University Press, forthcoming.
Galtung, Fredrik. 2006. ‘Measuring the Immeasurable: Boundaries and
Functions of (Macro) Corruption Indices’ in Measuring Corruption,
Charles Sampford, Arthur Shacklock, Carmel Connors, and Fredrik
Galtung (eds.), Ashgate Publishing Limited: 101-130.
Glynn, Patrick, Stephen J. Kobrin, and Moises Naim. 1997, ‘The Globalization of Corruption’ in Corruption and the Global Economy, Kimberly Ann Elliott (ed.), Institute for International Economics: 7-27.
Goldsmith, Arthur A. 1999, ’Slapping the Grasping Hand: Correlates of
Political Corruption in Emerging Market’, American Journal of Economics and Sociology 58(4): 865–883.
Gottfredson, Michael, and Travis Hirshi.
Crime, Stanford University Press.
1990, A General Theory of
Gould, David J. 1991, ‘Administrative Corruption: Incidence, Causes and
Remedial Strategies’ in Handbook of Comparative and Development
Public Administration, Ali Farazmand (ed.), Marcel Dekker Inc.: 761–
774.
Bibliography
129
Graeff, P. and G. Mehlkop. 2003, ‘The Impact of Economic Freedom on
Corruption: Different Patterns for Rich and Poor Countries’, European Journal of Political Economy, 19(3): 605620.
Gupta, Sanjeev, Hamid Davoodi, and Erwin Tiongson. 2001, ‘Corruption
and the Provision of Health Care and Education Services’ in The Political Economy of Corruption, Arvind K. Jain (ed.), Routledge Press:
111-141.
Gupta, Sanjeev, Hamid Davoodi, and Rosa Alonso-Terme. 1998, ‘Does
Corruption Affect Income Inequality and Poverty?’, IMF Working Paper 76.
Gurgur, Tugrul and Anwar Shah. 2005, ‘Localization and Corruption:
Panacea or Pandora’s Box’, WB Policy Research Working Paper 3486.
Hall, Robert E. and Charles I. Jones. 1999, ‘Why Do Some Countries Produce So Much More Output per Worker than Others?’ The Quarterly
Journal of Economics, 114(1): 83–116.
Hammond, George W. and Eric Thompson. 2006, ‘Convergence and Mobility: Personal Income Trends in US Metropolitan and Nonmetropolitan
Regions’, International Regional Science Review, 29(1): 35–63.
Hassan, Shaukat. 2004, ‘Corruption and the Development’, Journal of
Development Policy and Practice, 1(1): 25–41.
Henderson, J. Vernon and Ari Kuncoro. 2004. ‘Corruption in Indonesia’,
NBER Working Paper 10674.
Hendry, David F. and Hans-Martin Krolzig. 2005, ‘The Properties of Automatic GETS Modelling’, The Economic Journal, The Royal Economic
Society, 115(502): C32–C61.
Herzfeld, Thomas and Christoph Weiss. 2003, ‘Corruption and Legal
(In)Effective-ness: An Empirical Investigation’, European Journal of
Political Economy, 19(3): 621–632.
Hornby, Albert Sydney. 2000, The Oxford Advanced Learner’s Dictionary,
Oxford University Press.
Jain, Arvind K. 2001, ‘Corruption: A Review’, Journal of Economic Surveys, 15(1): 71–121.
Kaufmann, Daniel and Aart Kraay. 2002, ‘Growth without Governance’,
Economia, 3(1): 169–229.
130
Kaufmann, Daniel, Aart Kraay, and Pablo Zoido-Lobaton. 1999, ’Governance Matters’, World Bank Policy Research Working Paper No.
2196.
Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2006, Governance Matters V: Governance Indicators for 1996-2005. Mimeo, World
Bank.
Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2007, ‘Governance Matters VI: Aggregate and Individual Governance Indicators
for 1996-2006’, World Bank Working Paper Series 4280.
Keefer, Philip and Stephen Knack. 1997, ‘Why don’t Poor Countries Catch
Up? A Cross-National Test of an Institutional Explanation’, Economic Inquiry, 35(3): 590–602.
Kelejian Harry H. and Ingmar R. Prucha. 1998, ‘A Generalized Spatial
Two-Stage Least Squares Procedure for Estimating a Spatial Autoregressive Model with Autoregressive Disturbances’, Journal of Real Estate Finance and Economics, 17(1): 99-121.
Kim, Jae-On and Charles W. Mueller. 1978, Factor Analysis: Statistical
Methods and Practical Issues. Thousand Oaks: Sage Publications,
Quantitative Applications in the Social Sciences Series 14.
Klitgaard, Robert. 2000, ‘Subverting Corruption’, Finance and Development, 37(2): 2–5.
Knack, Stephen and Omar Azfar. 2003, ‘Trade Intensity, Country size,
and Corruption.’ Economics of Governance 4(1): 1–18.
Knack, Stephen and Philip Keefer. 1995, ‘Institutions and Economic Performance: Cross-country Tests using Alternative Institutional Measures’, Economics and Politics, 7(3): 207–227.
Knack, Stephen. 1996, ‘Institutions and the Convergence Hypothesis: The
Cross-National Evidence’, Public Choice, 87(3-4): 207–228.
Knack, Stephen. 2001, ‘Aid Dependence and the Quality of Governance:
Cross-Country Empirical Tests’, Southern Economic Journal, 68(2):
310–329.
Koenker, Roger and Kevin F. Hallock. 2001, ‘Quantile Regression’, Journal of Economic Perspectives, 15(4): 143–156.
Bibliography
131
Kuncoro, Ari. 2004, ‘Bribery in Indonesia: Some Evidence from MicroLevel Data’, Bulletin of Indonesian Economic Studies, 40(3): 329–354.
Kunicová, Jana and Susan Rose-Ackerman. 2005, ‘Electoral Rules and
Constitutional Structures as Constraints on Corruption’, British Journal of Political Science 35(4): 573–606.
Kurtz, Marcus J. and Andrew Schrank. 2007, ‘Growth and Governance:
Models, Measures, and Mechanisms’, The Journal of Politics, 69(2):
538–554.
La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert
Vishny. 1997, ‘Trust in Large Organizations’, The American Economic Review (Papers and Proceedings of the Hundred and Fourth
Annual Meeting of the American Economic Association), 87(2): 333–
338.
La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert
Vishny. 1999, ‘The Quality of Government’, Journal of Law, Economics and Organization, 15(1): 222–279.
Laffont, Jean J. and Tchetche N’Guessan. 1999, ’Competition and Corruption in an Agency Relationship’, Journal of Development Economics,
60(2): 271–295.
Lambsdorff, Johann Graf. 2000, ‘Background Paper to the 2000 Corruption Perceptions Index: The Precision and Regional Comparison
of Perceived Levels of Corruption–Interpreting the Results’, Transparency International and Göttingen University.
Lambsdorff, Johann Graf. 2001, ‘How Corruption in Government Affects
Public Welfare: A Review of Theories’, Center for Globalization and
Europeanization of the Economy Georg-August-Universität Göttingen
Discussion Paper 9.
Lambsdorff, Johann Graf. 2002, ‘Background Paper to the 2002 Corruption Perceptions Index: Framework Document’, Transparency International and Göttingen University.
Lambsdorff, Johann Graf. 2003a, ‘How Corruption Affects Productivity’,
Kyklos, 56(4): 457–474.
Lambsdorff, Johann Graf. 2003b, ‘Background Paper to the 2003 Corruption Perceptions Index: Framework Document’, Transparency International and University of Passau.
132
Lambsdorff, Johann Graf. 2004, ‘Background Paper to the 2004 Corruption Perceptions Index: Framework Document’, Transparency International and University of Passau.
Leamer, Edward E. 1983, ‘Let’s Take the Con Out of Econometrics’, The
American Economic Review, 73(1): 31–43.
Leamer, Edward E. 1985, ‘Sensitivity Analyses Would Help’, The American Economic Review, 75(3): 308-313.
Lederman, Daniel, Norman V. Loayza, and Rodrigo R. Soares. 2005,
‘Accountability and Corruption: Political Institutions Matter’, Economics and Politics, 17(1): 1-35.
Leite, Carlos A. and Jens Weidmann. 1999, ’Does Mother Nature Corrupt?
Natural Resources, Corruption, and Cconomic Growth’, IMF Working
Paper 85.
LeSage, James P. 2004, ‘A Family of Geographically Weighted Regression
Models’ in Advances in Spatial Econometrics, Luc Anselin, Raymond
J.G.M. Florax and Serge J. Rey (eds.), Springer.
Levine, Ross and David Renelt. 1992, ‘A Sensitivity Analysis of CrossCountry Growth Regressions’, American Economic Review, 82(4): 942–
963.
Li, Hongyi, Lixin Colin Xu, and Heng-Fu Zou. 2000, ‘Corruption, Income
Distribution, and Growth’‘ Economics and Politics, 12(2): 155–182.
Mankiw, N. Gregory, David Romer, and David N. Weil. 1992, ‘A Contribution to the Empirics of Economic Growth’ Quarterly Journal of
Economics, 107(2): 407–437.
Mauro, Paolo. 1995, ‘Corruption and Growth’, The Quarterly Journal of
Economics, 110(3): 681–712.
Mauro, Paulo. 2004, ‘The Persistence of Corruption and Slow Economic
Growth’, IMF Staff Papers, 51(1): 1–18.
Mishra, Ajit. 2006, ‘Persistence of Corruption: Some Theoretical Perspectives’, World Development, 34(2): 349–358.
Mukand, Sharun W. and Dani Rodrik. 2005, ‘In Search of the Holy Grail:
Policy Convergence, Experimentation, and Economic Performance’,
American Economic Review, 95(1): 374–383.
Bibliography
133
Nye, J. S. 1967, ‘Corruption and Political Development: A Cost-Benefit
Analysis’, American Political Science Review, 61(2): 417–427.
Olken, Benjamin A. 2006, ‘Corruption Perception vs Corruption Reality’,
NBER Working Paper 12428.
O’Loughlin, John, Michael D. Ward, Corey L. Lofdahl, Jordin S. Cohen,
David S. Brown, David Reilly, Kristian S. Gleditsch, and Michael
Shin. 1998, ‘The Diffusion of Democracy 1946-1994’, Annals of the
Association of American Geographers, 88(4): 545–574.
Olsson, Ola and Douglas A. Hibbs Jr. 2005, ‘Biogeography and Long-run
Economic Development’, European Economic Review, 49(4): 909–938.
Olsson, Ulf. 1979, ‘Maximum Likelihood Estimation of the Polychoric
Correlation Coefficient’, Psychometrika, 44(4): 443–460.
Paldam, Martin. 2001, ‘Corruption and Religion: Adding to the Economic
Model’, Kyklos 54(2-3): 383–414.
Paldam, Martin. 2002, ‘The Cross-Country Pattern of Corruption: Economics, Culture and the Seesaw Dynamics’, European Journal of Political Economy 18(2): 215–240.
Park, Hoon. 2003, ‘Determinants of Corruption: A Cross-National Analysis’, The Multinational Business Review, 11(2): 29–48.
Paternoster, Raymond and Sally Simpson. 1996, ‘Sanction Threats and
Appeals to Morality: Testing a Rational Choice Model of Corporate
Crime’, Law and Society Review, 30(3): 549–583.
Persson, Torsten and Guido Tabellini. 2003, The Economic Effect of Constitutions: What Do the Data Say?, MIT Press.
Persson, Torsten, Guido Tabellini, and Francesco Trebbi. 2003, ‘Electoral
Rules and Corruption’, Journal of the European Economic Association, 1(4): 958–989.
Pritchett, Lant. 2003, ‘A Toy Collection, a Socialist Star and a Democratic
Dud: Growth Theory, Vietnam, and the Philippines’ in In Search of
Prosperity: Analytical Narratives on Economic Growth., Dani Rodrick
(ed.), Princeton University Press: 123-151.
Qian, Yingyi. 2003, ‘How Reform Worked in China’ in In Search of Prosperity: Analytical Narratives on Economic Growth., Dani Rodrick
(ed.), Princeton University Press: 297–333.
134
Qizilbash, M. 2001, ‘Corruption and human development’, Oxford Development Studies, 29(3): 265–278.
Quah, Danny. 1993a, ‘Empirical Cross-Section Dynamics in Economic
Growth’, European Economic Review, 37(2-3): 426–434.
Quah, Danny, 1993b, ‘Empirics for Growth and Distribution: Stratification, Polarization, and Convergence Clubs’, Journal of Economic
Growth, 2(1): 27–59.
Quah, Danny. 1993c, ‘Galton’s Fallacy and Tests of the Convergence Hypothesis’, Scandinavian Journal of Economics, 95(4): 427–443.
Quibria, M. G. 2006, ‘Does Governance Matter? Yes, No or Maybe: Some
Evidence from Developing Asia’, Kyklos, 59(1): 99–114.
Rajkumar, Andrew Sunil and Vinaya Swaroop. 2007, ‘Public Spending
and Outcomes: Does Governance Matter?’, Journal of Development
Economics, forthcoming.
Rauch, James and Peter Evans. 2000, ‘Bureaucratic Structure and Bureaucratic Performance in Less Developed Countries’, Journal of Public
Economics 75(1): 49–71.
Raudenbush, Stephen W. and Anthony S. Bryk. 2002, Hierarchical Linear
Models: Applications and Data Analysis Methods, Sage Publication,
Inc. Advanced Quantitative Techniques in the Social Sciences.
Rodrik, Dani, Arvind Subramanian, and Francesco Trebbi. 2004, ‘Institutions Rule: The Primacy of Institutions Over Geography and Integration in Economic Development’, Journal of Economic Growth, 9(2):
131–165.
Rose-Ackerman, Susan. 1999, Corruption and Government: Causes, Consequencse, and Reform, Cambrdige University Press.
Sala-i-Martin, X., Doppelhoffer, G., and R. Miller. 2004, ‘Determinants
of Long-Term Growth: A Bayesian Averaging of Classical Estimates
(BACE) Approach’, American Economic Review, 94(4): 813–835.
Sala-i-Martin, Xavier X. 1997, ‘I Just Ran Two Million Regressions’, American Economic Review Papers and Proceedings of the Hundred and
Fourth Annual Meeting of the American Economic Association, 87(2):
178–183.
Bibliography
135
Shapiro, S. S. and M. B. Wilk. 1965, ‘An Analysis of Variance Test for
Normality (Complete Samples)’, Biometrika, 52(3-4): 591–611.
Shapiro, S. S. and R. S. Francia. 1972, ‘An Approximate Analysis of
Variance Test for Normality’, Journal of the American Statistical Association, 67(337): 215–216.
Shleifer, Andrei and Robert W. Vishny. 1993. ‘Corruption’, Quarterly
Journal of Economics, 108(3): 599–617.
Shorrocks, A. F. 1978, ‘The Measurement of Mobility’, Econometrica,
46(5): 1013–1024.
Silverman, B. W. 1986, Density Estimation for Statistical and Data Analysis., Chapman and Hall.
Simmons, Beth A. and Zachary Elkins. 2004, ‘The Globalization of Liberalization: Policy Diffusion in the International Political Economy’,
American Political Science Review, 98(1): 171–189.
Søreide, Tina. 2003, ‘Estimating Corruption: Comments on Available
Data’, Mimeo, Chr. Michelsen Institute.
Starr, H. 1991, ‘Democratic Dominoes: Diffusion Approaches to the Spread
of Democracy’, Journal of Conflict Resolution, 35(2): 356–381.
Sturm, Jan-Egbert and Jakob de Haan. 2005, ‘Determinants of longterm growth: New Results Applying Robust Estimation and Extreme
Bounds Analysis’, Empirical Economics, 30(3): 597–617.
Swamy, Anand, Stephen Knack, Young Lee, and Omar Azfar. 2001, ‘Gender and Corruption’, Journal of Development Economics 64(1): 25–55.
Tanzi, Vito and Hamid R. Davoodi. 1997, ‘Corruption, Public Investment,
and Growth’, IMF Working Paper 139.
Tanzi, Vito. 1997. ‘Comments’ in Corruption and the Global Economy,
Kimberly Ann Elliot (ed.), Institute for International Economics: 163–
168.
Tavares, Jose. 2003, ‘Does Foreign Aid Corrupt¿, Economic Letters 79(1):
99–106.
Temple, Jonathan. 1998, ‘Robustness Tests of the Augmented Solow
Model’, Journal of Applied Econometrics, 13(4): 361–375.
136
The Guardian, 26 March 2004.
Time Asia, 24 May 1999.
Tirole, Jean. 1996, ‘A Theory of Collective Reputations (with Applications
to the Persistence of Corruption and to Firm Quality)’, Review of
Economic Studies, 63(1): 1–22.
Tobler, W. R. 1970, ‘A Computer Movie Simulating Urban Growth in the
Detroit Region’, Economic Geography (Supplement: Proceedings. International Geographical Union. Commission on Quantitative Methods.), 46(2): 234–240.
Tornell, Aaron and Philip Lane. 1998, ‘Voracity and Growth’, NBER
Working Paper W6498.
Transparency International. 2004, The Global Corruption Report, Pluto
Press.
Treisman, Daniel. 2000, ‘The Causes of Corruption: A Cross-National
Study’, Journal of Public Economics, 76(3): 399–457.
van Rijckeghem, Caroline and Beatrice Weder. 1997, ‘Corruption and
the Rate of Temptation: Do Low Wages in the Civil Service Cause
Corruption?’, IMF Working Paper 73.
Ventelou, Bruno. 2002, ‘Corruption in a Model of Growth: Political Reputation, Competition and Shocks’, Public Choice, 110(1-2): 23–40.
Wansbeek, Tom and Erik Meijer. 2000, Measurement Error and Latent
Variables in Econometrics. Elsevier Science.
Ward, Michael D. and Kristian Skrede Gleditsch. 2002, ‘Location, Location, Location: An MCMC Approach to Modeling the Spatial Context
of War and Peace’, Political Analysis, 10(3): 244–260.
Waterbury, John. 1973, ‘Endemic and Planned Corruption in a Monarchical Regime’, World Politics, 25(4): 533–555.
Wei, Shang-Jin. 2000a, ‘How Taxing is Corruption on International Investors?’, Review of Economics and Statistics, 82(1): 1–11.
Wei, Shang-Jin. 2000b, ‘Natural Openness and Good Government’, NBER
Working Paper 7765.
Bibliography
137
World Bank and United Nations. 2007, ‘Stolen Asset Recovery (StAR)
Initiative: Challenges, Opportunities, and Action Plan’, The World
Bank.
Wyatt, Geoffrey. 2003, ‘Corruption, Productivity, and Socialism’, Kyklos,
56(2): 223–244.
Appendix
2003
2004
2002
2006
1999
2005
2003
2001
2003
Damania et al.
Fisman & Gatti
Fréchette
Goldsmith
Gurgur & Shah
Herzfeld & Weiss
Knack
Knack & Azfar
2006
Brown et al.
2007
2002
Broadman & Recanatini
Chang & Golden
2001
Broadman & Recanatini
Brunetti & Weder
2004
Braun & Di Tella
2003
Alt & Lassen
2007
2003
Ali & Isse
2001
1999
Ades & Di Tella
Bonaglia et al.
2000
Abed & Davoodi
Aslaksen
Year
Authors
Appendix 1
1, 6, 7, 8
4
4, 5, 6
6
6
4, 5, 6
1, 3, 4, 5, 6, 9
6, 7
6, 7
4, 6, 7, 5
7
7
7
4, 6, 7, 5
4, 6
4
10
4, 6
1, 9
6
Data
a
Corruption
5
3
2
7
2
3
4
5
2
4
2
5
5
6
6
4
8
6
4
5
ED
1
3
6
2
2
3
4
7
4
10
1
1
1
2
4
7
2
1
2
PPI
1
4
1
2
1
1
2
1
1
1
JB
Determinantsb
5
3
1
3
2
3
2
8
7
1
4
2
GC
40-165
68-80
40-1061
30
34
947-1168
28-55
50-79
24-39
46-497
162
19
19
50-1125
75-305
49-1599
45
57-78
31-186
25-84
Obsv.
No. of
c
1
1
1, 2
1
1
2
1
1
1
1, 2
2
1
1
1, 2
1, 2
1, 2
1
1
1, 2
1, 2
Type
Data
80-98
82-95
82-97
90-99
95-96
82-97
80-95
97-98
96-98
82-98
96-00
96-99
96-99
80-94
80-00
82-97
02
80-90
80-90
94-98
25
2
5
3
6
24
19
3
10
23
1
1
2
22
22
42
21
2
12
8
Model
No. of
1
2
1, 3, 2, 5
1
1
1, 2, 4
1, 2
1
1
1, 4
4
1
1
1, 2
1, 2
1, 4, 5
1
1
1; 4
1
Estimatord
Continued on next page. . .
Coverage
Year of
140
4
3
3
4
1
3
1
3
4
12
39-66
22-33
31-81
181-200
32-98
28-35
60-640
30-37
41-100
85
72
364-1158
27-30
114
54-94
Obsv.
1
1
1
2
1
1
1, 2
1
1
1
1
2
1
1
1
Typec
b:
10
6
50
19
10
5
19
2
30
55
5
10
3
3
15
Model
No. of
1
1
3, 1
1, 2
3
1
1, 3, 6, 7
1
1, 2
1
2
1
2
1
3
Estimatord
1=Cross-section, 2=Panel
80-83
82-95
80-98
NA
82-97
81-90
95-00
NA
94-99
98
70-90
84-97
90-95
82-95
93-01
Coverage
1=OLS; 2=IV, 2SLS; 3=WLS; 4=Fixed Effect; 5=Random Effect; 6=Country Effect; 7=Time Effect
1
2
6
9
1
12
1
6
11
6
2
9
6
GC
Year of
ED=Economic-Demography, PI=Political Institutions, JB=Judicial-Bureaucracy, GC=Geography-Culture
Wei
2
1
1
3
1
1
1
JB
Data
d:
1997
2000a
van Rijckeghem & Wede
1, 2, 4, 6
2
4
17
4
1
1
8
13
PPI
No. of
1=BI, 2=Gallup, 3=GEI, 4=ICRG, 5=IMD, 6=TI, 7=WB-Kaufmann, 8=WB-CPIA, 9=WEF, 10=others;
2000
Treisman
4
6, 7
3
4
3
5
2
3
5
4
2
ED
Determinantsb
c:
2003
1, 4
4, 6, 7
6
6
6
4
4
1
4
7
Dataa
Corruption
a:
2001
Tavares
2002
Paldam
Swamy et al.
2001
Paldam
2000
1999
Leite & Weidmann
Rauch & Evan
2005
Lederman et al.
2003
1999
Laffont & N’Guessan
2003a
1999
La Porta et al.
Persson et al.
2005
Kunicová & Rose-Ackerman
Park
Year
Authors
Appendix
141
Human Development Report (HDR) 2005
http://devdata.worldbank.org/data-query/
http://www.nationmaster.com/graph-T/eco deb ext gdp
Human Development Report (HDR) 2005
Human Development Report (HDR) 2005
11d. FDI Inflow
PWT 6.2
PWT 6.2
PWT 6.2
http://genderstats.worldbank.org/dataonline/
PPP
PPP
PPP
Principal Component Analysis (PCA)
2000
11c. Trade Volume
12. Human Capital
2000
2000
11b. Export Share
2000
Principal Component Analysis (PCA)
11a. Import Share
11. Economic Integration
Continued on next page. . .
(the sub-indicators not included; the parallel variable—Fraser Institute and its sub-indicators—not included)
http://www.heritage.org/research/features/index/downloads/2007PastScores.xls
2000
10. Heritage Foundation
2000
7c. The Richest 20 %
9. Aid per capita
93-00
7b. The Richest 10 %
8. Debt per GDP
93-00
93-00
7a. Gini Index
Principal Component Analysis (PCA)
7. Income Distribution
PWT 6.2 and CIA Factbook
PWT 6.2
PWT 6.2
2000
PPP
PWT 6.2
PWT 6.2
2000
4. Investment per GDP
PPP
PPP
PWT 6.2
Source
6. Population Density
2000
3. Govt. Expenditure per GDP
PPP
Description
5. Ln Population
90-00
2000
2. Growth GDP per capita
2000
Year
1. GDP per capita
Economic Variables:
Variable
Appendix 2
142
2000
2000
12c. Gross Enrol. (Tertiary School)
12d. Illiteracy Rate (15 year+)
Source
Aggregated Index
Kaufmann (2007)
http://www.uis.unesco.org/TEMPLATE/html/Exceltables/education/
http://devdata.worldbank.org/edstats/query/default.htm
http://devdata.worldbank.org/edstats/query/default.htm
http://devdata.worldbank.org/edstats/query/default.htm
Description
2000
2000
2000
15c. Sub-Nat. Exp. per GDP
15d. Sub-Nat. Exp. per Total Exp.
15e. Sub-Nat. Rev. per Total Rev.
2000
18. Political Polarization
19. Political Pluralty
96-00
2000
2000
20. Wage Bill per GDP
21. Government Effectiveness
22. Regulatory Quality
Bureaucracy and Judicial Environment:
2000
2000
17. Presidentialism
2000
2000
15b. Sub-Nat. Rev. per GDP
16. District Magnitue
2000
15a. Sub-Nat. Exp. per GDP
Aggregated Index
Aggregated Index
Kaufmann (2007)
Kaufmann (2007)
Continued on next page. . .
http://www1.worldbank.org/publicsector/civilservice/development.htm
http://siteresources.worldbank.org/INTRES/Resources/
http://siteresources.worldbank.org/INTRES/Resources/
http://siteresources.worldbank.org/INTRES/Resources/
http://siteresources.worldbank.org/INTRES/Resources/
Government Financial Statistics
Government Financial Statistics
Government Financial Statistics
Government Financial Statistics
Government Financial Statistics
Kaufmann (2007)
15. Decentralization
Aggregated Index
2000
Principal Component Analysis (PCA)
14. Political Stability
Freedom of Speech, Freedom of Political Participation, Religious Freedom (http://ciri.binghamton.edu/)
Freedom of the Press (Freedom House), Polity Index (Polity IV Project), Democratic Accountability (ICRG),
Related variables (not included in the regressions): Political Rights, Civil Liberty,
13. Voice and Accountability
2000
2000
Political Institutions:
2000
12b. Gross Enrol. (Secondary School)
Year
12a. Gross Enrol. (Primary School)
Variable
Appendix
143
2000
23. Rule of Law
http://ciri.binghamton.edu/
http://ciri.binghamton.edu/
http://ciri.binghamton.edu/
La Porta et.al. (1999)
La Porta et.al. (1999)
La Porta et.al. (1999)
La Porta et.al. (1999)
39. French law
40. German Law
41. Scandinavian Law
Continued on next page. . .
http://devdata.worldbank.org/edstats/query/default.htm
38. Socialist Law
2000
La Porta et.al. (1999)
36d. Female Labor Force
0 (no right) to 3 (guaranteed)
0 (no right) to 3 (guaranteed)
0 (no right) to 3 (guaranteed)
http://www.carleton.ca/cifp/rank.htm
http://www.carleton.ca/cifp/rank.htm
http://www.worldchristiandatabase.org/
http://www.worldchristiandatabase.org/
http://www.worldchristiandatabase.org/
37. British Law
2000
36. Women’s Roles
2000
0 (low) to 3 (highly diversed)
Principal Component Analysis (PCA)
35. Religious Diversity
36c. Women’s Social Rights
0 (low) to 3 (highly diversed)
34. Ethnic Diversity
36b. Women’s Political Right
2005
33. Protestant
2000
2005
32. Orthodox
36a. Women’s Economic Right
2005
31. Marginal
http://www.worldchristiandatabase.org/
http://www.worldchristiandatabase.org/
2005
2005
http://www.worldchristiandatabase.org/
http://www.worldchristiandatabase.org/
http://www.worldchristiandatabase.org/
http://www.worldchristiandatabase.org/
30. Independent
2005
28. Anglican
Kaufmann (2007)
Source
http://www.worldchristiandatabase.org/
Aggregated Index
Description
29. Catholic
2005
2005
27. Nonreligion
2005
25. Hindu
26. Muslim
2005
24. Budha
Geography and Culture:
Year
Variable
144
https://www.cia.gov/library/publications/the-world-factbook/index.html
2000
2000
44. Export of Fuel
45. Export of Ores and Metal
http://devdata.worldbank.org/data-query/
http://devdata.worldbank.org/data-query/
https://www.cia.gov/library/publications/the-world-factbook/index.html
43. Log of Area
Source
42. Absolute Latitude
Description
Year
Variable
Appendix
145
146
WDI CD-ROM
1985
1984
1984
7. Import Tariff Rate
8. Industrial Share
1985
17. Term of Trade Schock
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
19. East Asia Dummy
Continued on next page. . .
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
WDI CD-ROM
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
PWT (http://pwt.econ.upenn.edu/php site/pwt index.php)
PWT (http://pwt.econ.upenn.edu/php site/pwt index.php)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
18. Distance to Economic Centers
Geography:
1985
1985
16. Tariff Restriction
1985
14. Liquid Liabilities
15. Openess
1985
1984
13. Ln GDP per Capita
1984
11. Investment Price
12. Log Black Market Premium
1984
10. Investment Share
9. Inflation
WDI CD-ROM
1985
6. Import Non-Tariff Rate
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
WDI CD-ROM
1984-2004
WDI CD-ROM
1985
1984
3. FDI
Heritage Foundation (http://www.heritage.org/)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
5. Import Share
1985
Source
4. GDP per Capita Growth
1980
2. Export Share
Year
1. Economic Freedom
Economic Variables:
Variable
Appendix 3
Appendix
147
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
24. OECD Dummy
25. Sahara Dummy
26. Size of Land
27. South Asia Dummy
28. Transition Economies Dummy
29. Western Europe Dummy
1985
34. Recurring Expenditure on Education
1985
1985
1985
1985
1985
1985
1985
35. Higher School Enrolment
36. Primary School Enrolment
37. Secondary School Enrolment
38. Years of Higher Education
39. Years of Primary Education
40. Years of Secondary Education
41. Total Years of Schooling
Human Capital :
1980
1985
33. Public Investment
1980
31. Expenditure on Education
32. Government Expenditure
1975
30. Expenditure on Defense
Continued on next page. . .
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
23. Oceania Dummy
Government Expenditure:
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
22. North Africa Dummy
1970
20. East Asia Religion
Source
21. Middle East Dummy
Year
Variable
148
1985
1984
1985
1985
43. Life Expectancy
44. Log Labor Force
44. Population Growth
45. Worker Ratio
1984
1980-85
1960-85
53. Polity 2
54. Revolution
55. War Dummy
1970
1970
1985
1970
1970
1970
56. Buddha Fraction
57. Catholic Fraction
58. Ethnolinguistic Fractionalization
59. Herfindal Index of Religion
60. Hindu Fraction
61. Jews Fraction
Religion and Culture:
1980-85
1984
52. Political Right
1980-85
50. Political Instability
51. Political Assassination
1960-85
1984
49. Our Governance Index
1980-85
47. Coup
48. External War Involvement
1984
46. Civil Liberty
Political Variables:
1985
Year
42. Infant Mortality
Population:
Variable
Continued on next page. . .
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Roeder (http://weber.ucsd.edu/ proeder/elf.xls)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Polity (http://www.cidcm.umd.edu/polity/)
Freedom House (http://www.freedomhouse.org/uploads/fiw/FIWAllScores.xls)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
ICRG-based
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
(http://www.freedomhouse.org/uploads/fiw/FIWAllScores.xls)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
WDI CD-ROM
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Source
Appendix
149
1970
1970
1970
1970
1970
64. Non-Religion Fraction
65. Orthodox Fraction
66. Orthodox Christian Fraction
67. Other Religions
68. Protestant Fraction
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro (http://www.economics.harvard.edu/faculty/barro/data.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
Barro-Lee (http://www.cid.harvard.edu/ciddata/ciddata.html)
1970
63. Moslem
Source
Shleifer (http://post.economics.harvard.edu/faculty/shleifer/data.html)
Year
62. Legal Origin
Variable
150
Appendix
151
Appendix 4
Appendix 4a. The Dimensions of Governance: Definitions (Chapter 6)
• Voice and Accountability: The extent to which a country’s citizens
are able to participate in selecting their government, and the degree
of freedom of expression and of association, and the presence of free
media.
• Political Stability and Absence of Violence: Perceptions of the
likelihood that the government will be destabilized or overthrown by
unconstitutional or violent means, including political violence and terrorism.
• Government Effectiveness: The quality of public services, the
quality of the bureaucracy and its independence from political pressures, the quality of policy formulation and implementation, and the
credibility of the government’s commitment to such policies.
• Regulatory Quality: The ability of the government to formulate
and implement sound policies and regulations that permit and promote private sector development.
• Rule of Law: The extent to which agents have confidence in and
abide by the rules of society, and in particular the quality of contract
enforcement, the police, and the courts, as well as the likelihood of
crime and violence.
• Control of Corruption: The extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ”capture” of the state by elites and private interests.
Source: Kaufmann et al. (2006).
152
Appendix 4b. Countries in Our Sample
• (1) Afghanistan, (2) Albania, (3) Algeria, (4) Angola, (5) AntiguaBarbuda, (6) Argentina, (7) Armenia, (8) Australia, (9) Austria, (10)
Azerbaijan,
• (11) Bahamas, (12) Bahrain, (13) Bangladesh, (14) Barbados, (15)
Belarus, (16) Belgium, (17) Belize, (18) Benin, (19) Bermuda, (20)
Bhutan, (21) Bolivia, (22) Bosnia-Herzegovina, (23) Botswana, (24)
Brazil, (25) Brunei, (26) Bulgaria, (27) Burkina Faso, (28) Burundi,
• (29) Cambodia, (30) Cameroon, (31) Canada, (32) Cape Verde, (33)
CAR, (34) Chad, (35) Chile, (36) China, (37) Colombia, (38) Comoros, (39) Congo, Rep., (40) Congo, DR, (41) Costa Rica, (42)
Croatia, (43) Cuba, (44) Cyprus, (45) Czech Republic,
• (46) Denmark, (47) Djibouti, (48) Dominica, (49) Dominican Republic,
• (50) Ecuador, (51) Egypt, (52) El Salvador, (53) Equatorial Guinea,
(54) Eritrea, (55) Estonia, (56) Ethiopia,
• (57) Fiji, (58) Finland, (59) France,
• (60) Gabon, (61) Gambia, (62) Georgia, (63) Germany, (64) Ghana,
(65) Greece, (66) Grenada, (67) Guatemala, (68) Guinea, (69) GuineaBissau, (70) Guyana,
• (71) Haiti, (72) Honduras, (73) Hong Kong, (74) Hungary,
• (75) Iceland, (76) India, (77) Indonesia, (78) Iran, (79) Iraq, (80)
Ireland, (81) Israel, (82) Italy, (83) Ivory Coast,
• (84) Jamaica, (85) Japan, (86) Jordan,
• (87) Kazakhstan, (88) Kenya, (89) Kiribati, (90) Korea, DR., (91)
Korea, Rep., (92) Kuwait, (93) Kyrgyzstan,
Appendix
153
• (94) Laos, (95) Latvia, (96) Lebanon, (97) Lesotho, (98) Liberia, (99)
Libya, (100) Lithuania, (101) Luxembourg,
• (102) Macao, (103) Macedonia, (104) Madagascar, (105) Malawi, (106)
Malaysia, (107) Maldives, (108) Mali, (109) Malta, (110) Mauritania,
(111) Mauritius, (112) Mexico, (113) Micronesia, (114) Moldova, (115)
Mongolia, (116) Morocco, (117) Mozambique,
• (118) Namibia, (119) Nepal, (120) Netherlands, (121) Netherlands Antilles, (122) New Zealand, (123) Nicaragua, (124) Niger, (125) Nigeria,
(126) Norway,
• (127) Oman,
• (128) Pakistan, (129) Panama, (130) Papua New Guinea, (131) Paraguay,
(132) Peru, (133) Philippines, (134) Poland, (135) Portugal, (136)
Puerto Rico,
• (137) Qatar,
• (138) Romania, (139) Russia, (140) Rwanda,
• (141) Samoa, (142) Sao Tome-Principe, (143) Saudi Arabia, (144)
Senegal, (145) Serbia-Montenegro, (146) Seychelles, (147) Sierra Leone,
(148) Singapore, (149) Slovak Republic, (150) Slovenia, (151) Solomon
Islands, (152) Somalia, (153) South Africa, (154) Spain, (155) Sri
Lanka, (156) St. Kitts-Nevis, (157) St. Lucia, (158) St. VincentThe Grenadines, (159) Sudan, (160) Suriname, (161) Swaziland, (162)
Sweden, (163) Switzerland, (164) Syria,
• (165) Taiwan, (166) Tajikistan, (167) Tanzania, (168) Thailand, (169)
Togo, (170) Tonga, (171) Trinidad-Tobago, (172) Tunisia, (173) Turkey,
(174) Turkmenistan,
• (175) Uganda, (176) Ukraine, (177) UAE, (178) UK, (179) USA, (180)
Uruguay, (181) Uzbekistan,
154
• (182) Vanuatu, (183) Venezuela, (184) Vietnam,
• (185) Yemen,
• (186) Zambia, (187) Zimbabwe.
Samenvatting
Corruptie leidt ertoe dat mensen arm en analfabeet blijven, lijden onder
een hoge kindersterfte, een laag geboortegewicht van baby’s, alsmede een
hoge uitval op de basisschool (Kaufmann en Al., 1999; Gupta ea., 2001).
Tevens verslechtert de inkomensverdeling binnen landen (Li et al., 2000), de
kwaliteit van de openbare infrastructuur (Tanzi en Davoodi, 1997), en de
productiviteitsgroei (Lambsdorff, 2003a) door corruptie.
De International Country Risk Guide (ICRG) publiceert een corruptieindex waaruit blijkt dat 79 procent van de ongeveer 140 landen in de wereld
in 2005/06 worden geregeerd door corrupte politici en bureaucraten. Een
alternatieve maatstaf voor corruptie is de index van Kaufmann et al. (2007)
van de Wereld Bank (WB). Volgens deze indicator is 59 procent van de 207
landen corrupt. De groep van corrupte landen bestaat voornamelijk uit
landen in Afrika, Azie en Latijns-Amerika, terwijl de groep van landen met
weinig corruptie hoofdzakelijk westerse landen bevat.
Een algemeen aanvaarde definitie van corruptie is van Waterbury (1973),
namelijk: het misbruik van publieke macht en invloed voor priv doeleinden.
Een alternatieve definitie wordt gegeven door Jain (2001). Deze geeft de
sociaal-politieke context aan van corruptie; corruptie is een daad waarin de
macht van een openbaar ambt wordt gebruikt voor persoonlijk gewin op een
wijze die in strijd is met de regels. Corruptie weerspiegelt poor governance:
ambtenaren die niet gekwalificeerd zijn, wetten die niet worden nageleefd en
gebrek aan transparantie en publieke verantwoording. Hoewel de definities
van governance verschillen, lijkt er brede consensus te bestaan dat good
156
governance betekent dat de regering accountable is, en ook transparant,
flexibel, effectief en efficient opereert en hierbij de regels van de rechtsstaat
volgt, zodat corruptie wordt geminimaliseerd.
Dit proefschrift behandelt voornamelijk de oorzaken en gevolgen van
corruptie en good governance. De eerste onderzoeksvraag waar we ons mee
bezighouden is wat bepalend is voor de verschillen in corruptie tussen landen.
Corruptie komt niet alleen veel voor, veel auteurs menen ook dat corruptie een blijvend karakter heeft. Net als andere (politieke) instituties is
corruptie echter mogelijk niet een statisch fenomeen. Daarom is onze tweede
onderzoeksvraag in hoeverre corruptie persistent is.
Omdat corruptie nauw verbonden is met governance, is onze derde onderzoeksvraag wat de relatie is tussen governance en economische groei.
Knack en Keefer (1995), Barro (1997), Keefer en Knack (1997), en Chong
en Calderon (2000) stellen dat good governance een positieve invloed heeft
op economische groei. Anderen betogen het tegendeel (Quibria, 2006). Zo
zijn bijvoorbeeld China and Vietnam nog ver verwijderd van good governance, maar hun economische groeicijfers zijn opmerkelijk. Indonesie is een
‘Aziatische tijger’ terwijl het land lange tijd leed onder het kleptocratische
regime van Soeharto. Om een bijdrage aan dit debat te leveren, hebben we
opnieuw gekeken naar de relatie tussen economische groei en good governance.
Landen staan onder invloed van internationale economische en politieke ontwikkelingen. Met name interacties met buurlanden kunnen invloed
hebben op governance. Dit betekent dat er een ruimtelijke afhankelijkheid
kan zijn, zoals die ook is aangetoond voor democratie, oorlog en vrede, en
economische vrijheid (O’Loughin et al., 1998; Ward en Gleditsch, 2002; Simmons en Elkins, 2004). Deze ruimtelijke afhankelijkheid kan het gevolg zijn
van spillovers van adoptie- en diffusieprocessen, convergentie in economisch
beleid (Mukand en Rodrik, 2005), onderlinge afhankelijkheid van beleidsbeslissingen (Brueckner, 2003) of de transmissie van overheidsvormen (Starr,
1991). Aanwezigheid van ruimtelijke afhankelijkheid betekent dat landen
Samenvatting (Summary in Dutch)
157
geografisch kunnen worden geclusterd op basis van hun governance. Tegelijkertijd is het effect van de determinanten van governance mogelijk niet homogeen, dat wil zeggen dat de determinanten verschillende gevolgen hebben
in verschillende landen. Deze kwestie zal als laatste aan de orde komen in
dit proefschrift.
De theoretische literatuur biedt vele verklaringen voor het bestaan van
verschillen in corruptie tussen landen. Omdat er geen eenduidige theorie
bestaat over de oorzaken van corruptie, bepalen wij met behulp van een
Sensitivity Analysis (SA) wat de determinanten van corruptie zijn. Het
idee achter deze aanpak is om door een reeks van schattingen, waarbij allerlei combinaties van variabelen worden gehanteerd die mogelijk van invloed
zijn op corruptie, te bepalen welke determinanten robuust zijn. In de analyse worden 45 variabelen onderzocht die allemaal in eerdere studies zijn
gebruikt. Wij vinden slechts twee variabelen die robuust zijn, namelijk de
effectiviteit van de regering en de mate waarin sprake is van een rechtsstaat.
Over het algemeen komt theoretisch en empirisch onderzoek naar corruptie tot de conclusie dat corruptie persistent is. Met behulp van ICRG
gegevens voor de periode 1984-2003 vinden we echter sterke aanwijzingen
dat corruptie verandert in de tijd. Veel corrupte landen zagen hun niveau
van corruptie dalen, terwijl veel ‘schone’ landen corrupter werden in dezelfde
periode. Wij vinden ook dat dit convergentieproces niet continu is, maar
dat er een wereldwijde verbetering ten aanzien van corruptie plaats vond in
het eerste deel van onze schattingsperiode, maar een verslechtering in het
tweede deel.
Een eenvoudige analyse van de correlatie tussen het huidige niveau van
corruptie en dat in eerdere jaren, toont aan dat de correlatie afneemt naarmate de afstand tussen de beide jaren toeneemt. Ook vinden we aanwijzingen voor zogenoemde beta-convergentie wanneer we de verandering in
corruptie regresseren op zijn initile waarde. Ook vinden we bewijs voor
sigma-convergentie: de standaarddeviatie en de variatiecofficint van corruptie nemen af in de loop van de tijd.
Deze bevindingen worden bevestigd in onze analyse van de dynamiek
158
van de verdeling van corruptie. Er is een significante verschuiving in de
globale distributie van corruptie in de loop van de tijd. Met behulp van
een Gaussian-kernel functie kunnen we de verandering in de verdeling van
de corruptie van een bimodale naar een monomodale distributie aantonen.
Ten slotte vinden we op basis van een Markov chain analyse, opwaartse en
neerwaartse ’interclass’ verschuivingen.
De literatuur over de relatie tussen good governance en economische
groei is niet eenduidig. Er dienen twee belangrijke stappen te worden gezet
om deze relatie te onderzoeken: analyse van de robuustheid van de relatie enerzijds en de manier waarop governance wordt gemeten anderzijds.
Onze bijdrage aan dit debat is dat we een governance index construeren
met behulp van Confirmatory Factor Analysis (CFA) gebruikmakend van
verschillende ICRG indicatoren, namelijk democratic accountability, overheidsstabiliteit, kwaliteit van de bureaucratie, corruptie, en rechtsorde. We
concluderen dat deze indicatoren kunnen worden gecombineerd tot n index.
Met onze CFA-index testen we vervolgens de relatie tussen governance en
economische groei gebruik makend van de eerder vermelde Sensitivity Analysis. We concluderen dat onze index behoorlijk robuust is in een reeks van
regressies met verschillende controlevariabelen, ook als we het aantal observaties laten variren. In bijna alle gevallen vinden wij dat good governance
de economische groei aanzienlijk bevordert.
Wetenschappers erkennen over het algemeen de rol van geografie als
het gaat om de verklaring van verschillen in de kwaliteit van governance
tussen landen. Verschillende indicatoren zijn gebruikt voor de geografische
ligging zoals de breedtegraad, klimaat, temperatuur, de grootte van het
land, klimaatgerelateerde ziekten, of dummyvariabelen die aangeven dat
een land aan zee grenst of behoort tot een bepaalde regio (Acemoglu et
al., 2001, 2002; Easterly en Levine, 2003; Rodrik et al., 2004; Olsson en
Hibbs Jr, 2005). Deze indicatoren geven echter niet adequaat de ruimtelijke
dimensie van governance weer.
We beschouwen twee ruimtelijke dimensies van governance, namelijk
ruimtelijke afhankelijkheid en ruimtelijke heterogeniteit. De eerste dimen-
Samenvatting (Summary in Dutch)
159
sie verwijst naar de afhankelijkheid van de kwaliteit van governance in een
land en de kwaliteit van de governance in de buurlanden. De tweede dimensie verwijst naar de wisselende invloed van de determinanten van governance. Uit onze analyse blijkt dat hoe dichter het land is gelegen bij
het land met de beste (of slechtste) governance, hoe hoger (of lager) de
kwaliteit van zijn eigen governance. We ontdekken ook een positieve globale ruimtelijke afhankelijkheid, waarbij poor (good) governance landen geografisch zijn geclusterd met poor (good) governance landen.
We passen ook een ruimtelijke econometrische methode toe die voorgesteld
is door Anselin (1988). Hierbij nemen we niet alleen de geografische afstand,
maar ook de ‘politieke afstand’ in beschouwing. Wij vinden dat governance
in een land een positieve relatie vertoont met governance in naburige landen.
Tot slot hebben wij ook de Geografisch Gewogen Regressie methode
van Brunsdon et al. (1999) toegepast. Deze methode stelt ons in staat het
probleem te analyseren van de ruimtelijke niet-stationariteit, gedefinieerd als
de variatie in de relaties en processen in de ruimte (Brunsdon et al., 1999).
We concluderen dat de cofficinten van de determinanten van governance
niet constant zijn, maar variren tussen landen.