Access to Justice and Economic Development: Evidence from an International Panel Dataset

Abstract

We empirically investigate the impact of access to justice (ATJ) on GDP per capita growth in a panel of 83 countries from 1970 to 2014. Our analysis relies on a new database documenting the number of judges per capita as a proxy for capturing the cross-country evolution of ATJ. The proxy measures the extent to which disputes between economic actors can be resolved at a relatively low cost, without dysfunctional delay and discrimination. In a dynamic panel setting using internal instruments, we find that increasing ATJ by 1% increases the five-year growth rate of GDP per capita by 0.86 p.p. (0.17 p.p. annually) with diminishing marginal returns. In line with the diminishing marginal returns argument, we find that the effect of ATJ is two times smaller in Europe compared to other regions due to higher levels of ATJ. We find no evidence of a differential effect of ATJ across other regions, income levels, legal origins, democracy, corruption of the judicial system or human capital levels.

Full text

Access to Justice and Economic Development:
Evidence from an International Panel Dataset
A. Deseau, A. Levai, and M. Schmiegelow
Discussion Paper 2019-9

Access to Justice and Economic Development:
Evidence from an International Panel Dataset ∗
Arnaud Deseau+, Adam Levai++, and Mich`ele Schmiegelow+++
+Universit´e Saint-Louis – Bruxelles, CEREC
++UCLouvain, CRIDES, IRES
+++UCLouvain, CRIDES
April 2019
Abstract
We empirically investigate the impact of access to justice (ATJ) on GDP per capita
growth in a panel of 83 countries from 1970 to 2014. Our analysis relies on a new
database documenting the number of judges per capita as a proxy for capturing the
cross-country evolution of ATJ. The proxy measures the extent to which disputes
between economic actors can be resolved at a relatively low cost, without dysfunctional
delay and discrimination. In a dynamic panel setting using internal instruments, we
find that increasing ATJ by 1% increases the five-year growth rate of GDP per capita
by 0.86 p.p. (0.17 p.p. annually) with diminishing marginal returns. In line with the
diminishing marginal returns argument, we find that the effect of ATJ is two times
smaller in Europe compared to other regions due to higher levels of ATJ. We find no
evidence of a differential effect of ATJ across other regions, income levels, legal origins,
democracy, corruption of the judicial system or human capital levels.
Keywords: Access to Justice, Legal development, Economic Development, Growth, Insti-
tutions
JEL Codes: E02, K00, O11, O43, O47
∗We would like to thank Alexia Autenne, Fr´ed´eric Docquier, Filip Dorssemont, Alexandre Girard, William
Parient´e and Henrik Schmiegelow for their invaluable comments. We also thank the participants at seminars
and workshops at University of Bonn, UCLouvain, University Saint-Louis Bruxelles, Erasmus University
Rotterdam and Spring Meeting of Young Economists for helpful comments and suggestions. We are grateful
to: the Fonds Quetelet (Brussels) for granting us access to its cross-country collection of statistical yearbooks;
the Ministries of Justice in Botswana, Belgium, Croatia, Estonia, Finland, Ireland, Latvia, Mali, Malta, New
Zealand, Niger, Poland, Romania, Slovenia, South Korea, Sweden, Uruguay, Venezuela; and to National
Statistical Service of Australia, Armenia, Qatar and United States for sharing their official data with us.
This work was supported by the Fonds de la Recherche Scientifique - Projet de Recherches (FNRS-PDR),
under the grant number “26036434”. The usual disclaimers apply.

“Promote peaceful and inclusive societies for sustainable development, provide access to
justice for all and build effective, accountable and inclusive institutions at all levels ”
Sustainable Development Goal 16 - United Nations (2015)
1 Introduction
A country can have a very well-defined legal system and legal rules but if they are not
accessible for the majority of people, due for example to high costs and long (dysfunctional)
delays, then the legal system and rules, both substantive and procedural, will only have a
limited impact on economic activity. For example, even though France and Senegal (a former
French colony) have very similar law codes, the effective access to justice (ATJ) of citizens
in both states is significantly different, leading thus to different economic outcomes.
The legal origin theory shows how differences in legal codes may affect financial develop-
ment (La Porta et al.; 1997, 1998, 1999; Djankov et al.; 2007), labor markets (Botero et al.;
2004), competition (Djankov et al.; 2002, 2003) and finally economic growth (Beck et al.;
2000; Levine et al.; 2000; Mahoney; 2001).1More recent research emphasizes the role of
justice quality and efficiency in fostering entrepreneurship, credit, agricultural and indus-
trial activities at the country level (Chemin; 2009a,b; Jappelli et al.; 2005; Visaria; 2009;
Amirapu; 2017). Yet to date, surprisingly little is known about the importance of ATJ in
explaining cross-country income per capita differences.
As a first step in bridging this gap in the literature, we build a new database documenting
the number of judges from 1970 to 2014 for 107 countries worldwide. This measure serves
as a macro level proxy for ATJ, capturing the extent to which disputes can be resolved
at a relatively low cost, without dysfunctional delays and discrimination. Using the panel
structure of our data, we are able to address endogeneity issues and look at the impact of
1See La Porta et al. (2008) for a complete review of the contribution of the legal origin theory to the
literature.

ATJ on economic growth using a difference GMM estimation.2Our benchmark specification
suggests that a 1% increase in ATJ leads to, on average, a 0.86 p.p. increase in the five-year
GDP per capita growth rate with diminishing marginal returns. This finding is robust to
different subsamples of countries.
We then assess whether ATJ can foster economic growth differently across continents/regions,
income groups, legal origins, political regime, corruption of the judicial system and human
capital levels. Our results show no statistical difference in the effect along those dimensions.
Only in the case of Europe, we find a significantly weaker effect: the impact of ATJ is al-
most two times lower compared to the rest of the world, which is in line with the diminishing
marginal returns argument.
This paper makes three main contributions. First, we add to the empirical law and
economics literature by providing a new database tracking the number of judges per capita
in 107 countries between 1970 and 2014. This is notable as there is very few panel data
allowing for cross-country comparisons of the effective judicial supply both for developed and
developing countries. Moreover, it is the first database offering a comprehensive proxy for
ATJ across country and through time. This is crucial as the United Nations has currently no
historical data which would allow them to track the progress of the Sustainable Development
Goal 16 from the ATJ aspect.
Second, we add to the literature analyzing empirically the determinants of economic
growth as, to the best of our knowledge, we provide the first panel cross-country analysis
looking at the effect of ATJ on economic growth. On that aspect, we complement the scant
literature trying to explain cross-country income and growth differences by difference in the
judicial characteristics (Berkowitz et al.; 2003; Feld and Voigt; 2003; Voigt et al.; 2015).
Third, we advance the existing literature on the macroeconomic effect of justice by ad-
dressing endogeneity issues in a more satisfactory way than existing contributions. While
2Our specifications focus on the effect of access to justice in log (as proxied by the number of judges
per 100.000 inhabitants) on growth in GDP per capita. For brevity we will sometimes describe this as, with
some abuse of terminology, ”the impact of ATJ on growth” or ”the impact of access to justice on economic
development” (rather than the impact of access to justice on GDP per capita growth).
2

endogeneity is a long recognized issue in cross-country analysis, recent literature has ad-
vanced in better panel cross-country identification strategies. Benefiting from the panel
structure of our data, we are able to use a difference GMM estimation which relies on first
difference, to mitigate the confounding effect of country-level unobserved heterogeneity, and
internal instrumentation to address reverse causality between ATJ and economic growth.
The rest of the paper is organized as follows. Section 2 discusses the relevance of our proxy
and presents our new database with some stylized facts. Section 3 explains the econometric
specification and estimation methods. Section 4 presents the results of the aggregate effect of
access to justice on economic growth, evaluates its robustness, and explores the heterogeneity
of the impact regarding income level, region, legal origin, political regime, judicial corruption
and human capital. Section 5 concludes.
2 Access to Justice
2.1 Theory and Measurement
ATJ is a difficult concept to define, as it comprises many dimensions which are not solely
economic. We can decompose ATJ into four main dimensions: 1) information, 2) social, 3)
geography and 4) cost dimension (Woolf; 1996; Hammergren; 2014). A society with a high
level of ATJ should enable the majority of its citizens to: know their rights, provide them
with information on which steps to undertake, and which competent people to meet in order
to introduce a plaint or defend themselves against it (dimension 1); they should not face
discrimination of any kind along the judicial process (dimension 2); they should be capable
to go effectively to court, meet the legal staff and exchange with them (dimension 3); finally,
they should be facing a procedural cost which is not prohibitive both in terms of monetary
and time cost (dimension 4).
The complexity of ATJ makes it difficult to measure accurately across countries and
through time. For this reason, there are very few publicly available datasets that one can
3

exploit for a cross-country comparison of ATJ. In addition, for existing datasets, the time
dimension is small which is not suitable for a meaningful panel analysis. One example is The
World Justice Project: this non-profit organization has collected an impressive amount of
data via general population polls and questionnaires sent to in-country law professionals to
build an index describing the Rule of Law at the national level. One of the 44 sub-factors of
the Rule of Law Index concerns directly ATJ as it measures if “people can access and afford
civil justice” (sub-factor 7.1) but is available only from 2012. Another initiative launched
by The World Justice Project called Global Insights on Access to Justice have gathered data
on legal needs and public ATJ in 45 countries but again, the time dimension is small as the
first available year is 2017.
We argue that the total number of all types of professional judges per capita is a relevant
proxy for ATJ at the macro level. We can identify two main channels through which the
density of judges directly affects ATJ. First, the number of judges per capita directly improves
ATJ through a quantity channel: more judges means more resolved cases, reducing both the
monetary and time cost of accessing to justice (dimension 4). Specifically, the literature on
the determinants of court output has established that this channel is more important for
developing, middle-income economies and small-size courts (Dimitrova-Grajzl et al.; 2016;
Grajzl and Silwal; 2017) rather than for developed countries or urban courts (Beenstock and
Haitovsky; 2004; Dimitrova-Grajzl et al.; 2012).
Second, the literature provides evidence that the number of judges per capita can affect
ATJ through the quality of justice channel. At first glance, one can argue that the positive
effect of the quantity channel can be cancelled out by a decrease in the quality of justice.
However, most of the empirical studies reject the existence of a quantity-quality trade-off
within courts (Rosales-L´opez; 2008; Coviello et al.; 2015; Dimitrova-Grajzl et al.; 2016; Grajzl
and Silwal; 2017). Moreover, in a cross-section of 36 European countries, Voigt and El-Bialy
(2016) shows that more judges per court is even positively associated with a measure of
judicial independence (quality of justice) while it is not significantly correlated with the
4

resolution rate (quantity of justice). This last result provides indications that adding more
judges in an already developed judicial system can improve the quality of justice. A greater
quality of justice can be measured by reducing the number of appealed cases or verdict
reversals.
Overall, we view our proxy as capturing shocks to the aggregate supply of justice: an
increase in the number of judges per capita decreases the trial length and increases the total
number of treated cases, ceteris paribus. Consequently, ATJ is enhanced primarily via a
reduction in both monetary and time cost of justice.
Our measure is coherent and transparent as the data is collected using official sources.
Data on the number of judges are gathered from public institutions, international organi-
zations and academic publications (see table A3 for more details) and are, for the most,
publicly available. Moreover, the judge has the universal role of supplying justice by resolv-
ing disputes in a court of law across all national legal systems, allowing for cross-country
comparisons.
Figure 1 provides an empirical validation of our proxy by showing a strong and positive
correlation between our measure of ATJ (log density of judges) and the access to civil justice
score as measured by the 2019 edition of The World Justice Project - Rule of Law Index,
conditional on log of GDP per capita. The log density of judges is averaged across the
whole 2000-2014 period to maximize sample size. In particular, the score constructed by the
World Justice Project measures: “the accessibility and affordability of civil courts, including
whether people are aware of available remedies [dimension 1]; can access and afford legal
advice and representation [dimensions 3 and 4]; and can access the court system without
incurring unreasonable fees [dimension 4], encountering unreasonable procedural hurdles, or
experiencing physical or linguistic barriers [dimensions 2 and 3]”. Figure 1 therefore gives
direct evidence that the density of judges is a good proxy for ATJ as it captures all four
dimensions described above.3
3In appendix figure A1, we test the quality of our proxy to alternative measures depicting more the
dimension 2, 3 or 4 of ATJ with similar results.
5

Figure 1: The Partial Effect of the Density of Judges on Access to Civil Justice
GTM
SGP
ETH
ECU
MYS
JPN
MEX
GBR
CAN
NZL
KOR
AUS
IND
NPL
KHM
NOR
PHL
DNK
BWA
ZAF
CHL
NLD
USA
TTO
VEN
THA
MUS
ESP
FRA
PER
SWE
ITA
BLR
BRA
PAN
DZA
PRT
GEO
MMR
UKR
TUR
FIN
EST
RUS
GRC
AUT
BGR
BEL
KGZ
COL
ROU
DEU
URY
KAZ
CHN
MDA
BOL
POL
CRI
ALB
HUN
CZE
MNG
SVN
HRV
MKD
BIH
-.2 -.1 0 .1 .2 .3
People can Access &
Afford Civil Justice
-2 -1 0 1 2
Log Density of Judges
coef = .03244023, (robust) se = .01110696, t = 2.92
Note: This figure plots the relationship between the log density of judges (averaged between 2000-2014) and
the access to civil justice score of The World Justice Project (year 2019). This adjusted partial residual plot
is based on an OLS regression where the log GDP per capita is used as a control variable and with robust
standard errors.
Obviously, our proxy comes with some disadvantages. First, we measure access to a
formal court of justice, neglecting the effect of informal justice and out-of-court settlements
despite their importance in some developed and developing countries (Galanter; 1981; Plat-
teau; 2015). This is a concern as we tend to underestimate the ATJ in some countries.
However, many alternative dispute resolution mechanisms such as mediation require court-
backed enforcement and thus the intervention of formal justice. In addition, in a context of
competition between informal and formal legal institutions (known as legal pluralism), Al-
dashev et al. (2012) shows that formal law acts as an outside anchor on the custom. In that
perspective, increasing access to the formal justice improves outside options of the plaintiff,
letting him the choice to exit the informal system or forcing custom to adjust.
Second, due to data availability, we collect data on professional judges in all types of
6

courts: first instance, second instance (appellate courts), supreme courts etc.; dealing with
all types of cases: civil, criminal or other types of cases. A priori, we cannot say that some
type of judges dealing with some type of cases are not important for economic activities.
Judges often act as generalists: they adjudicate all relevant cases both civil and criminal.
Only in large courts and developed economies judges tend to specialize in one specific domain
(e.g., marital matters for civil cases). The civil cases are potentially the most important type
of cases for economic development since they deal directly with economic activity and they
have the highest monetary value for an average case. Moreover, the majority of legal issues
are civil rather than criminal (Pleasence et al.; 2013). Therefore, by considering professional
judges dealing with all types of cases can only bias our results downwards since an additional
professional judge dealing with civil cases (compared to a judges dealing with criminal cases)
contributes more to economic development. Figure 1 indicates that we are well capturing
access to civil justice.
Finally, in our empirical exercise we focus on the quantity rather than quality of judges,
while the quality of judges can influence ATJ via the same channels of quantity and quality
of judgments (Ramseyer; 2012; Bielen et al.; 2018). That omission is a concern only in the
case of a specific trend in the quality of judges on the period 1970-2014. Although the quality
of judges might have changed over time, we think that this evolution stays minor compare
to the development of the number of judges on the same period. Moreover, we control for
time and country fixed-effects in our specifications, capturing the general trend and country
specificities in terms of education which are good approximation of the quality of judges.
2.2 Data and Stylized Facts
Figures on the number of judges used to proxy ATJ have been gathered firstly from interna-
tional organizations such as the Inter-university Consortium for Political and Social Research
(ICPSR), the Commission Europ´eenne Pour l’Efficacit´e de la Justice (CEPEJ), the United
Nations Office on Drugs and Crime (UNODC) and the Organization of American States
7

(OAS). The data provided by those four international organizations is long recognized for
their transparency and expertise on various judicial indicators. Our contribution is to have
created a unified database by: 1) merging the existing data and 2) verifying and supple-
menting the available data with the help of ministries of justice, supreme courts, national
statistical offices and academic publications. We end up with an unbalanced panel of judges
per capita for 107 countries from year 1970 to 2014. See section A.2 of the appendix for a
complete description of the merging process, sources, definitions (table A3), and descriptive
statistics (table A4).
Figure 2 gives an overview of our data by plotting the average number of judges per
100.000 inhabitants over the whole 1970-2014 period categorized by deciles. By this mean,
we can see two things: first, the 107 countries for which we managed to collect at least
one data point; second, the high cross-country variation of the number of judges per capita.
The highest average density of judges for the 1970-2014 period is reached in Montenegro
with 40.14 judges per 100.000 inhabitants; the lowest is found in Ethiopia with 0.24 judges
per 100.000 inhabitants. Even within-continents the cross-country variation stays sizeable.
For example, in Europe we find both countries belonging to the top decile (like Germany
or Serbia) and countries belonging to the third lowest decile (like the United Kingdom or
Ireland). The cross-country variations in terms of judges per capita are driven by differences
in factors such as GDP per capita, political regime, legal origin, culture or ethnic composition
of the population.4For instance, the legal origin is one variable that can help us understand
the difference in the density of judges between two countries that have similar income per
capita and similar political regime like Germany and the United Kingdom.
Figure 3 looks at the evolution of five-year averages of the density of judges across different
regions. Even though during the last 45 years the world population has increased, the number
of judges has increased even more. The world average of the density of judges has more than
doubled, going up from 4.26 to 10.63 judges per 100.000 inhabitants between 1970 and 2014.
4See Appendix A.4. for more details on the determinants and correlates of the density of judges in our
sample.
8

Figure 2: Average density of judges around the world between 1970-2014 (deciles)
(21.8,40.1]
(14.9,21.8]
(11.4,14.9]
(9.5,11.4]
(7.5,9.5]
(6.4,7.5]
(4.2,6.4]
(2.6,4.2]
(1.1,2.6]
[0.2,1.1]
No data
Note: Country-level distribution of the number of judges per 100.000 inhabitants (averaged between 1970-
2014). Each color represents a decile from the first (blue) to the tenth (red).
This doubling of the density of judges was achieved at a quite stable growth rate over the
analyzed period.
Europe consistently displays the highest average density of judges among all continents,
starting from 7.78 judges in 1970, to 21.63 judges per 100.000 inhabitants in 2010. Testing
the difference of means between Europe and the rest of the world across the whole period
we find that, on average, European countries have 10 more judges per 100.000 inhabitants
than the rest of the world.
We document a remarkable increase in the number of judges per capita in the Post-Soviet
countries since the 1990’s.5Since 1980 the average number of judges in Post-Soviet countries
has more than tripled going from 4.53 to 14.62 judges per 100.000 inhabitants in 2010. Post-
Soviet countries have now the second highest density of judges in the world after Europe.
Since the 1990’s, an average Post-Soviet country has 2.9 more judges per 100.000 inhabitants
than an average American country. One of the main explanations for the judicial staff
increase in Post-Soviet countries after 1990, is the replacement of the soviet administration
after the fall of USSR which had to meet legal needs following the transformation from
5In our sample, Post-Soviet countries are comprised of Armenia, Azerbaijan, Belarus, Estonia, Georgia,
Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Moldova, Russia and Ukraine.
9

planned to market economies (Dietrich; 2000; Murrell; 2001).
Finally, figure 3 highlights the sluggish evolution of the average density of judges in the
other considered regions: Africa, Asia, the Middle-East, North America and Latin America.
Even though North America (United States and Canada) always have a positive growth rate
on the number of judges, the population growth rate is even higher in some periods (e.g.
between 1985 and 1995), leading to an overall decrease in the average density of judges.
Figure 3: Average density of judges across regions between 1970-2014
5
10
15
20
25
Density of Judges
1970 1975 1980 1985 1990 1995 2000 2005 2010
Period
Sub-Saharan
Africa
Asia and
Pacific Europe World
Middle East and
North Africa
Post
Soviet
North
America
Latin
America
Note: This figure plots trends in the density of judges (number of judges per 100.000 inhabitants) between
1970-2014. The graph covers 107 countries categorized into seven regions plus the World. The seven regions
used are identical to the World Bank classification with the exception of the post-Soviet region: Armenia,
Azerbaijan, Belarus, Estonia, Georgia, Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Moldova, Russia,
Tajikistan and Ukraine.
3 Empirical Strategy
Our goal is to empirically investigate the causal effect of ATJ (as proxied by the number
of judges per capita) on economic development (as proxied by the growth rate of GDP per
capita). In this section, we present our identification strategy and we discuss the diagnostic
10

tests that we use throughout the paper to deal with endogeneity issues. We use five-year
averages of all variables to smooth the short-run fluctuations and handle the annual gaps in
the data. Moreover, changes in the judicial system are expected to have their full effect on
the economy after a few years rather than immediately in one year of time.
To explain our estimation strategy, we take the following dynamic model as a starting
point:
ln yi,t
yi,t−1=β1ln(yi,t−1) + β2ln(AT Ji,t−1) + αi+δt+εi,t (1)
Where yi,t is GDP per capita in country iat time t,AT Ji,t is the number of judges per
100.000 inhabitants in country iat time t.αiand δtare country and time fixed-effects,
and εi,t is the error term. Equation (1) corresponds to a beta-convergence model, which is
a standard specification in the empirical growth literature since Islam (1995). If negative
and above -1, the β1coefficient captures the speed at which an economy is converging to
the common long-run GDP per capita level. By controlling for the dynamic component of
GDP per capita, this model can isolate the effect of the other right hand-side variables on
the steady-state of the economy. Our coefficient of interest is β2as it captures the short-run
effect of ATJ on economic growth.
Estimating equation (1) by pooled OLS will generate inconsistent estimates. The key
issue is the endogeneity of our variable of interest, ATJ. First, ATJ can affect economic
growth (for example through contract enforcement) and be affected by economic growth (as
citizens with higher income can more easily afford to go on court). This means that we
are facing reverse causality. Second, access to justice and economic prosperity can be jointly
affected by a third omitted variable (e.g. political regime), leading to a omitted variable bias.
Thirdly, the dynamic panel specification leads to an asymptotic bias of order 1/T known as
the Nickel bias (Nickell; 1981). In our case, this is a concern as we have a relatively short
time dimension (in our estimations T= 9).
To address the aforementioned biases, we apply an instrumental variable approach. More
specifically, we estimate equation (1) using a two-step difference Generalized Method of
11

Moments (GMM):
∆ ln yi,t
yi,t−1=β1∆ ln(yi,t−1) + β2∆ ln(AT Ji,t−1)+∆δt+ ∆εi,t (2)
In that form, equation (2) is estimated in the spirit of the seminal paper of Arellano and
Bond (1991). By taking equation (1) in first difference and including time fixed-effects, we
are able to control for all the time invariant country characteristics affecting both economic
growth and ATJ such as: legal origin, culture, ethnic composition of the population, struc-
tural criminality, geography, etc. The inclusion of time fixed-effects allows to capture the
world economy trends and the business cycle effects. The identification strategy relies on
internal instrumentation: lagged levels of the right-hand side variables are used to instru-
ment current differences. If the residuals are not serially correlated, difference GMM yields
consistent estimates in a dynamic panel model with small T, helping to mitigate both the
Nickel bias and the reverse causality issues. Treating ATJ as endogenous means that the
first available instrument for ∆AT Ji,t−1is AT Ji,t−2with the assumption that the level of
ATJ observed at t−2 (i.e. 10 years before) is not correlated with current shocks of economic
growth. We treat the log of GDP per capita as predetermined and time-period dummies as
exogenous.6Specifically, we focus on the following moment conditions:
E[(εi,t −εi,t−1) (ln(yi,t−j),ln(AT Ji,t−k))] = 0 for all 1 ≤j≤4 and k= 2 (3)
The exact choice of instruments follows a trade-off between the moment condition used
(which is most likely to hold using deeper lags) and the relevance condition (the fact that
instruments should be sufficiently correlated with the instrumented variables), which is most
likely to be satisfied with shorter lags.7
6In the growth empirics literature using GMM, there is no consensus on how to treat the initial level of
GDP per capita: while some studies treat it as predetermined (DeJong and Ripoll; 2006); some treat it as
endogenous (Hauk and Wacziarg; 2009; Voitchovsky; 2005). In the appendix table A7, we provide evidence
that our results are qualitatively the same if we treat GDP per capita as endogenous.
7In appendix table A7 we provide evidence of the robustness of our benchmark specification to different
moment conditions (i.e. different choice of internal instruments).
12

Difference GMM estimation of equation (2) constitutes our parsimonious benchmark
specification and is used to explore further heterogeneous effects of ATJ on economic growth
via the following specification:
∆ ln yi,t
yi,t−1=β1∆ ln(yi,t−1) + β2∆ ln(AT Ji,t−1) + γ0∆(ln(AT Ji,t−1)∗Θi,t−1)+∆δt+ ∆εi,t
(4)
Where Θi,t−1is a vector of time-variant (e.g. level of democracy or education) or time-
invariant variables (e.g. the legal origin or geographical area) interacted with ATJ in country
iat time t. When estimating equation (4), we keep the same moment conditions for the
interaction terms as those of ATJ.
To evaluate the quality of our internal instrumentation, we systematically report p-values
of the Hansen (1982) Jtest and the Arellano-Bond test for AR2. These are the two stan-
dard GMM diagnostic tests for the quality of the instrumentations. The Hansen test is
heteroskedasticity robust and it evaluates the joint validity of all the instruments. The
Arellano-Bond test for AR2 evaluates the second order autocorrelation of residuals; this is
required for a good instrumentation as the presence of second order autocorrelation would
yield AT Ji,t−2to be an invalid instrument. Since Roodman (2009) and Windmeijer (2005),
economists are aware of the problem of too many instruments when using GMM.8A common
solution is to reduce the number of instruments below the number of individuals. In our
estimations, we always keep the instrument count well below the number of countries in our
sample. In particular, our benchmark specification uses a collapsed matrix of instruments:
a method allowing to reduce significantly the instrument count without losing information.9
To evaluate further the quality of our internal instrumentation, we follow Bazzi and
Clemens (2013) testing for underidentification and weakness of the GMM instrument ma-
trix in a Two Stage Least Squares (2SLS) context. We first provide p-values for a test of
8In a difference GMM estimation without any restriction on the number of lags, the number of instru-
ments is quadratic in T.
9In the collapsed form the matrix of instruments contains one column per lag, instead of one column per
lag and time-period in the non-collapsed form.
13

underidentification based on the Lagrange Multiplier version of Kleibergen and Paap (2006)
rk statistic (hereafter KP-LM). The KP-LM test evaluates if one or more of the canonical
correlations is zero, providing a lower hurdle test for the weakness of instruments. Sec-
ond, we provide p-values to test the weakness of instruments as implemented by Bazzi and
Clemens (2013). In particular, this test uses a Wald Fstatistic based on the Kleibergen
and Paap (2006) rk statistic and the critical values classified by Stock et al. (2005). The
Kleibergen-Paap Wald test (hereafter KP-W) evaluates if the bias in the point estimate of
our endogenous variable is greater than 30 percent of the OLS bias.
Another GMM estimator, the Blundell and Bond (1998) system GMM is often used in
the economic literature. This estimator jointly estimates the difference equation with a level
equation in which the levels of right hand side variables are instrumented with lagged dif-
ferences. However, this estimator is not suitable in our case for two main reasons. Firstly,
system GMM requires additional moment conditions to hold between the current level of
the error term and the lagged differences. The crucial point here is that the error term
in the level equation still contains the fixed-effect. Roodman (2009) shows that the addi-
tional moment condition is not satisfied if the individuals in the sample have different initial
deviations from their long-run mean. As our sample is comprised of many heterogeneous
economies (transition economies, newly industrialized, developed) this condition is not likely
to hold. Secondly, the system GMM estimation can produce weak instruments too. Bazzi
and Clemens (2013) show that in several published growth papers, the superiority of system
GMM can be questioned and that the weakness of instruments should be directly tested:“In
practice, most applications of system GMM simply assume that instruments are strong. We
argue that instrument strength is an empirical question that can and should be directly
tested”. Other researchers advance a similar argument in the literature such as Hayakawa
(2009) and Bun and Windmeijer (2010).10
10Bun and Windmeijer (2010) show that when the ratio of the variance of the country fixed-effect to
the variance of the idiosyncratic shocks is higher than unity, the problem of weak instruments in the level
equation of system GMM is amplified. Considering equation (2), we find a ratio of 2.6, indicating that
system GMM is likely to produce weak instruments.
14

Finally, the potential non-stationarity of both ATJ (looking at figure 3) and of GDP
per capita is not an identification issue in our setting for several reasons. First, contrary
to panels with large T, unit-root is not a major concern in panels with relatively small T
(T= 9 in our case). Second, we detrend our potentially non-stationary panel by estimating
our benchmark specification in first differences using equation (2). Furthermore, the lagged
dependent variable coefficient in equation (2) is significantly negative which is consistent
with the stationarity assumption.
4 Results
The results are organized in two subsections. First, we present the main results obtained
from the two-step difference GMM estimation and discuss their robustness. In the second
subsection, we check for heterogeneous effects across geographical areas, income levels, legal
origin, political regime and education.
4.1 Baseline Results and Robustness Checks
Table 1 presents our results based on equation (2) corresponding to a two-step difference
GMM estimation on an unbalanced panel of 83 countries for the 1970-2014 period. Column 1
is our benchmark specification. Columns 2 to 4 show alternative difference GMM estimations.
Columns 5 to 11 show the robustness of the benchmark specification to different subsamples.
For each estimation, we report four diagnostic tests to keep track of the quality of our GMM
estimations (AR2, Hansen, KP-LM and KP-W).
In our benchmark specification (column 1), we find that increasing ATJ by 1% causes an
increase in the five-year growth rate of GDP per capita by 0.86 p.p. (0.17 p.p. annually),
which is a sizeable effect. Our benchmark results are in line with other cross-country studies
quantifying the effect of effective judicial institutions on economic growth. In particular,
Voigt et al. (2015) in a cross-section of 100 countries find that increasing de facto judicial
15

independence by one standard deviation leads to a 0.3 p.p. increase in annual GDP per
capita growth. Melcarne and Ramello (2016) in a panel of 175 countries over 12 years using
fixed-effect estimation, show that an extra year in judicial delays of private litigation lowers
annual growth rate by over 1 p.p.. Our estimates are even more plausible as we tackle
explicitly endogeneity issues.
In our benchmark specification, we deal with the issue of too many instruments inherent
to GMM estimations by collapsing the instrument matrix: this allows us to keep the number
of instruments well below the number of country in our sample as suggested by Roodman
(2009). The internal instrumentation seems valid as both the AR2 and the Hansen tests are
not rejecting their null hypothesis. In addition, following Bazzi and Clemens (2013) we give
evidence that the set of instruments used does not suffer from underidentification as we can
reject the null hypothesis of the KP-LM test at the 1% level, and the KP-W suggests strong
enough instruments to withdraw a sizeable portion of the OLS bias as we can reject the null
hypothesis at the 5% level.
Table 1: Main results and robustness checks
Alternative GMM estimations Robustness to subsamples
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Bench NoColl Ortho OrthoNoColl <2Mln Inh Low Judge High Judge <4 Obs <6 Obs
L.ln(GDPpc) -0.705*** -0.692*** -0.765*** -0.659*** -0.699*** -0.684*** -0.713*** -0.688*** -0.636***
(0.086) (0.072) (0.092) (0.085) (0.087) (0.082) (0.096) (0.081) (0.049)
L.ln(ATJ) 0.860*** 0.458*** 0.733*** 0.370*** 0.845*** 0.865*** 0.966*** 0.804*** 0.523***
(0.216) (0.141) (0.189) (0.099) (0.218) (0.204) (0.255) (0.202) (0.104)
AR2 0.498 0.851 0.756 0.474 0.598 0.455 0.384 0.546 0.768
Hansen 0.344 0.125 0.197 0.122 0.434 0.453 0.376 0.660 0.908
KP-LM 0.001 0.041 0.019 0.178 0.002 0.000 0.005 0.001 0.011
KP F-stat 6.18 3.67 3.61 2.47 6.09 7.02 5.06 5.86 11.87
KP-W 0.022 0.225 0.224 0.850 0.024 0.009 0.064 0.030 0.000
Instruments 12 39 12 39 12 12 12 12 12
Countries (N) 83 83 83 83 71 75 75 49 22
Observations (NT) 241 241 273 273 209 231 223 202 121
Note: Table reports two-step difference GMM estimations with Windmeijer-corrected standard errors in parentheses. All regressions include period fixed
effects. Our dependent variable is five-year average growth rate of GDP per capita. All the estimations keep the same collapsed instrument lag structure,
treating the lag dependent as predetermined (instrumenting it from lag 1 to lag 4) and the variable of interest as endogenous (instrumenting it only with lag
2). The KP LM underidentification test uses a rank test procedure from Kleibergen and Paap (2006). KP F-stat is heteroskedasticity robust multivariate
analogues to the first-stage F statistics. For the KP-W test, since critical values do not exist for the KP statistic, we follow the approach suggested by Baum
et al. (2007) and use the Stock et al. (2005) 30 percent of the OLS bias critical values for the multivariate statistic. Standard errors are clustered at country
level. * p<0.1, ** p<0.05, *** p<0.01.
In column 2 we do not collapse the matrix of instruments, which raises the instrument
16

count from 12 in our benchmark to 39. The instrumentation appears to be of lower quality
with respect to the benchmark: we are able to reject the lower hurdle of the KP-LM test only
at the 5% level. The set of instruments used is clearly weaker as the KP-W test indicates
that our estimates contain more than 30% of the OLS bias. In particular, the ATJ coefficient
is almost two times smaller, suggesting a downward bias.11
Columns 3 and 4 show results using forward orthogonal deviation instead of first difference
transformation. By subtracting the average of all future observations for each variable, this
technique allows to get rid of the time-invariant country characteristics without losing too
much observations in an unbalanced panel.12 Column 3 shows the estimation using both
forward orthogonal deviation and a collapsed matrix of instruments. The ATJ coefficient is
in line with the magnitude and the significance of our benchmark specification. Contrary
to the benchmark, the KP-W test indicates a problem of weak instruments. Using forward
orthogonal deviation without collapsing the matrix of instruments (column 4), leads to a
weaker instrumentation as we are not able to reject the null hypothesis of both the KP-LM
and the KP-W tests. The ATJ coefficient is now two times smaller with respect to the
benchmark, revealing a clear bias toward OLS.
To reassure on the fact that our results are not driven by a specific group of countries,
we report in columns 5 to 11 the robustness of our benchmark specification across different
subsamples. In column 5, we remove countries with less than two million inhabitants to test
if our results are not driven by low populated countries. Indeed, for a given minimum size of
the judicial system, low populated countries have a higher density of judges and consequently
higher ATJ. After dropping the twelve least populated countries from our sample, the results
remain very robust both in terms of magnitude, significance and quality of instrumentation.13
11Table A6 shows the OLS estimation of equation (2). In particular, column 2 depicts an ATJ coefficient
almost 10 times smaller with respect to our benchmark.
12The first difference transformation can be demanding in terms of losing observations in an unbalanced
panel. However, in our case the use of forward orthogonal deviation leads to a modest gain of 32 observations
in columns 3 and 4 with respect to the benchmark.
13Countries bellow 2 million inhabitants in 1990 in our sample: Botswana, Cyprus, Estonia, Iceland,
Luxembourg, Macedonia, Malta, Mauritius, Montenegro, Qatar, Slovenia, Trinidad and Tobago.
17

In columns 6 and 7, we check the validity of our results to the removal of top and the
bottom decile in terms of the density of judges (8 countries). Removing the bottom decile
(column 6) does not change the significance or the magnitude of the ATJ coefficient with
respect to the benchmark.14 The quality of the instrumentation is now even better as we are
able to reject the null hypothesis of the KP-W test at the 1% level. Removing the top decile
(column 7) increases the ATJ coefficient with respect to the benchmark while the quality of
the instrumentation stays close to the benchmark.15
In columns 8 and 9, we remove countries with less than 4 and 6 points of data for the
density of judges. By doing so, we move gradually to a more balanced panel and we can
verify if our results are driven by the unbalanced nature of the data. In both estimations,
the ATJ coefficient stays positive and highly significant. Even when we are dropping almost
75% of the countries and 50% of the observations in the more demanding estimation (column
9), the results stay in line with the benchmark specification both in terms of significance and
the quality of the instrumentation.
In the appendix table A7, we run additional robustness checks by considering different
moment conditions (i.e. different sets of internal instruments) compared to the benchmark
specification reported in column 1. Overall, our results remain robust across different choices
of instruments unless we use too deep lags (see discussion in section 3).
4.2 Heterogeneity of the Results
Table 2 reports the results of the two-step difference GMM estimation of equation (4) for
the 1970-2014 period. Columns 1 through 6 present the heterogeneity of the impact across
different regions, and columns 7 to 9 across different income levels.
In column 1 we find a significantly smaller effect of ATJ on economic growth in Europe
compared to the rest of the world. The quality of the instrumentation is even higher than
14The eight dropped countries happen to be mostly low income countries: Cambodia, Ecuador, Ethiopia,
Guatemala, India, Malaysia, Mexico and Nepal.
15The eight dropped countries are all Central European and Balkan countries: Croatia, Czech Republic,
Hungary, Luxembourg, Macedonia, Montenegro, Serbia and Slovenia.
18

in the benchmark as we can reject the null hypothesis of weak instruments at the 1% level.
Increasing ATJ by 1% rises the five-year GDP per capita growth rate by 0.45 p.p. on average
in Europe, compared to 0.84 p.p. outside Europe. We find this result in line with the
literature on court productivity determinants (Beenstock and Haitovsky; 2004; Dimitrova-
Grajzl et al.; 2012, 2016; Grajzl and Silwal; 2017). As discussed in section 2.1, there are
higher returns to GDP for countries with understaffed judiciaries or low ATJ. This is typically
the case outside Europe since the other continents/regions have significantly lower density
of judges (figure 3). This is in line with our general finding on diminishing marginal returns:
the expected economic gain of increasing ATJ is higher for lower initial levels of ATJ.
When looking at other regions, we do not find significant heterogeneous effects of ATJ
in North America, Latin America, Sub-Saharan Africa, Middle East and North Africa, and
Asia and Pacific (column 2 through 6). This means that ATJ is equally important for
development in those geographical areas compared to the rest of the world. In other words,
continental heterogeneity does not seems to affect the impact of ATJ in this regions. This is
in line with figure 3 showing similar levels of density of judges across the globe, except for
Europe. The quality of the instrumentation stays stable relatively to the benchmark, except
in column 2 and 6 where we are not able to reject the null of weak instruments.
Exploring the heterogeneity of the effect by income levels (columns 7 to 9), we do not
find any significant difference in the effect of ATJ across poor or rich economies. On the
other hand, the ATJ coefficient stays in line with the significance and the magnitude of the
benchmark. The quality of the instrumentation is in line with the benchmark, except in
column 9 where we cannot reject the null of weak instruments.
Table 3 presents the heterogeneity of the results by legal origins, political regime and
human capital levels. Column 1 suggests that the positive impact of ATJ on economic
growth is equally important in common law as in civil law countries. This is an interesting
finding as we would have expected a significantly different effect due to their fundamental
difference. Indeed, the comparative law literature describes common law as an adversarial
19

Table 2: Heterogeneity across regions and income levels
Region Income level
(1) (2) (3) (4) (5) (6) (7) (8) (9)
EU N AMER LAC SSA MENA APAC low midinc up mid inc high inc
L.ln(GDPpc) -0.738*** -0.699*** -0.702*** -0.706*** -0.711*** -0.767*** -0.618*** -0.709*** -0.695***
(0.089) (0.103) (0.085) (0.086) (0.085) (0.105) (0.097) (0.085) (0.092)
L.ln(ATJ) 0.843*** 0.856*** 0.860*** 0.860*** 0.863*** 0.778*** 0.575*** 0.790*** 0.933***
(0.199) (0.224) (0.212) (0.216) (0.221) (0.220) (0.209) (0.235) (0.233)
L.ln(ATJ) X Θ -0.395** -1.041 0.303 -0.509 0.121 0.296 0.817 0.069 -0.139
(0.191) (9.800) (0.443) (0.327) (0.530) (0.291) (0.683) (0.244) (0.246)
AR2 0.883 0.508 0.447 0.496 0.486 0.929 0.398 0.545 0.381
Hansen 0.156 0.281 0.429 0.346 0.318 0.458 0.280 0.295 0.341
KP-LM 0.000 0.424 0.001 0.001 0.001 0.029 0.001 0.001 0.005
KP F-stat 6.69 0.89 4.87 4.98 4.65 2.40 5.99 5.04 3.86
KP-W 0.005 0.930 0.048 0.043 0.061 0.469 0.013 0.040 0.139
Instruments 13 13 13 13 13 13 13 13 13
Countries (N) 83 83 83 83 83 83 83 83 83
Observations (NT) 241 241 241 241 241 241 241 241 241
Note: Table reports two-step difference GMM estimations with Windmeijer-corrected standard errors in parentheses. All regressions include period
fixed effects. Our dependent variable is five-year average growth rate of GDP per capita. Θ is our interaction term component which takes the
value of the corresponding title of the column. All the estimations keep the same collapsed instrument lag structure by treating the lag dependent
as predetermined (instrumenting it from lag 1 to lag 4) and the variable of interest together with its interaction term components are treated as
endogenous (instrumenting it only with lag 2). Standard errors are clustered at country level. * p<0.1, ** p<0.05, *** p<0.01.
dispute resolution system, where the judge has less investigative powers than lawyers, while
civil law countries dispute resolution system is inquisitorial, giving more abilities to the judge
to manage the procedure (Zweigert et al.; 1998). The ATJ coefficient stays in line with the
benchmark both in terms of magnitude and significance. However, looking at the KP-W test
we conclude at the presence of weak instruments. Therefore, the size of the effect remains
uncertain.
In columns 2 to 4, we explore heterogeneity within the civil law legal family, which
can be divided into three main categories: French, German and Scandinavian legal origin.
As the corresponding interaction terms are statistically insignificant, we cannot conclude a
differential effect of French or German legal origin compared to other legal families. We do
find a significant difference for Scandinavian legal origin at 10% level but as the number of
observations for this legal family is very limited in our sample, we cannot infer a general
tendency for this result.16 Moreover, the KP-W test indicates a bias in the coefficients
16Countries of Scandinavian legal origin in our sample are: Denmark, Finland, Iceland, Norway and
Sweden. Data coverage for that group is limited as the average number of observations is 2. Moreover, half
20

towards OLS in columns 3 and 4.
In columns 5 and 6, we test for differences in the effect across the level of democracy and
judicial corruption. In column 5, we use the Polity2 score for measuring democracy, which
is interacted with ATJ and used as a separate control. Although the effect of ATJ stays
close to the benchmark in terms of magnitude and significance, we do not find a significant
interaction between Polity2 and ATJ. This result is surprising as we would have expected a
higher effect of ATJ in democracies. Indeed, democracy is a sign of inclusive institutions and
a higher trust in judiciary institutions. In column 6, we test if increasing ATJ in a corrupted
judiciary leads a heterogeneous effect. Using a score on corrupted judicial decision, we do
not find a significant heterogeneous effect of ATJ. The ATJ coefficient stays in line with the
benchmark both in terms of significance and magnitude. Althought we can reject the null
of underidentification, we cannot reject the null of the KP-W test indicating a bias towards
OLS.
In the last three columns we check if there are differential effects of ATJ across different
human capital levels, as proxied by the share of people having completed primary (column
7), secondary (column 8) and secondary or tertiary education (column 9). Along the three
education dimensions we find similar results compared to the benchmark in both the mag-
nitude and the significance of the ATJ coefficient, while we do not a find coefficient of the
interactions with education levels. There is a strong concern for the quality of the instru-
mentation as we fail to lower hurdle of underidentification in columns 8 and 9 and we fail to
reject the null of weak instruments in columns 7 through 9.
In the appendix table A8, we check if the effect of ATJ was significantly larger in some
specific periods. Overall, we conclude that ATJ was equally important for GDP growth
throughout the whole 45-year period between 1970 and 2014. In addition, we do not find
significant differential effect comparing pre-1990 to post-1990 period, indicating that ATJ
remains of equal importance for economic development in current times.
of the available observations are for the most recent period (2010-2014)
21

Table 3: Heterogeneity across legal origins, political regimes and education
Common law Civil law Political regime Education
(1) (2) (3) (4) (5) (6) (7) (8) (9)
legor uk legor fr legor ge legor sc polity2 ju corrupt lpc lsc lhc lsc
L.ln(GDPpc) -0.718*** -0.722*** -0.684*** -0.711*** -0.717*** -0.692*** -0.717*** -0.669*** -0.660***
(0.080) (0.075) (0.086) (0.091) (0.091) (0.106) (0.078) (0.094) (0.112)
L.ln(ATJ) 0.786*** 0.839*** 1.008*** 0.867*** 0.828*** 0.986*** 0.817*** 0.772*** 0.836***
(0.209) (0.219) (0.322) (0.234) (0.194) (0.305) (0.193) (0.268) (0.274)
L.ln(ATJ) X Θ 0.062 -0.058 -0.205 -1.965* -0.006 -0.150 -0.003 0.002 0.002
(0.451) (0.220) (0.244) (1.037) (0.008) (0.406) (0.005) (0.004) (0.004)
L.Θ 0.012 0.247 0.016 -0.019 -0.025
(0.023) (0.680) (0.013) (0.022) (0.026)
AR2 0.565 0.573 0.383 0.522 0.709 0.837 0.747 0.483 0.564
Hansen 0.163 0.321 0.321 0.223 0.223 0.455 0.933 0.650 0.950
KP-LM 0.038 0.001 0.011 0.007 0.004 0.057 0.002 0.231 0.247
KP F-stat 2.10 5.57 3.69 3.31 3.36 1.30 3.32 0.89 0.83
KP-W 0.567 0.021 0.163 0.230 0.190 0.846 0.197 0.943 0.954
Instruments 13 13 13 13 14 14 14 14 14
Countries (N) 83 83 83 83 78 79 76 76 76
Observations (NT) 241 241 241 241 228 230 227 227 227
Note: Table reports two-step difference GMM estimations with Windmeijer-corrected standard errors in parentheses. All regressions include period
fixed effects. Our dependent variable is five-year average growth rate of GDP per capita. Θ is our interaction term component which takes the
value of the corresponding title of the column. All the estimations keep the same collapsed instrument lag structure by treating the lag dependent
as predetermined (instrumenting it from lag 1 to lag 4) and the variable of interest together with its interaction term components are treated as
endogenous (instrumenting it only with lag 2). Standard errors are clustered at country level. * p<0.1, ** p<0.05, *** p<0.01.
5 Conclusion
The importance of ATJ today is revealed by the 2030 Agenda of Sustainable Development
Goal of the United Nations (2015). Goal 16 states: “Promote peaceful and inclusive societies
for sustainable development, provide access to justice for all and build effective, accountable
and inclusive institutions at all levels”. Despite this renewed interest, there is no systematic
panel cross-country analysis trying to establish and quantify the economic impact of ATJ.
This is important not only in achieving this goal but also other Sustainable Development
Goals such as the goal 1: no poverty.
In this paper, we build a new database on the number of judges per capita in 107 countries
from 1970 to 2014. This measure allows us to proxy ATJ across countries and through time,
bridging the actual data limitation in the literature. Secondly, we tackle endogeneity issues
using difference GMM estimation with internal instruments and we uncover diminishing
marginal returns to GDP per capita growth by improving ATJ. This result is robust to
22

various subsamples of countries. In a dynamic panel setting we find that increasing ATJ
by 1% increases GDP per capita growth rate by 0.86 p.p., with dimishing marginal returns.
Third, checking for heterogeneous effects across continents, we find that increasing ATJ
in Europe (the region endowed with the highest density of judges worldwide), leads to an
two times smaller effect which is in line with the diminishing marginal returns argument. In
contrast to that finding, we find no effect heterogeneity by income level, legal origin, political
regime, corruption of the judiciary or education levels.
Overall, our results suggest that ATJ is an important factor affecting economic prosperity.
Identifying the exact channels through which ATJ affects economic growth is an important
question for future research. Moreover, by further expanding the time-dimension we can
learn more about the long-run effects of access to justice which is another interesting area
for future research.
23

References
Acemoglu, D., Johnson, S. and Robinson, J. A. (2005). Institutions as a fundamental cause
of long-run growth, Handbook of economic growth 1: 385–472.
Aldashev, G., Chaara, I., Platteau, J.-P. and Wahhaj, Z. (2012). Using the law to change
the custom, Journal of Development Economics 97(2): 182–200.
Amirapu, A. (2017). Justice delayed is growth denied: The effect of slow courts on
relationship-specific industries in india, Technical report, School of Economics Discussion
Papers 1706.
Arellano, M. and Bond, S. (1991). Some tests of specification for panel data: Monte carlo
evidence and an application to employment equations, The review of economic studies
58(2): 277–297.
Baum, C. F., Schaffer, M. E. and Stillman, S. (2007). Enhanced routines for instrumen-
tal variables/generalized method of moments estimation and testing, The Stata Journal
7(4): 465–506.
Bazzi, S. and Clemens, M. A. (2013). Blunt instruments: Avoiding common pitfalls in
identifying the causes of economic growth, American Economic Journal: Macroeconomics
5(2): 152–86.
Beck, T., Levine, R. and Loayza, N. (2000). Finance and the sources of growth, Journal of
financial economics 58(1-2): 261–300.
Beenstock, M. and Haitovsky, Y. (2004). Does the appointment of judges increase the output
of the judiciary?, International Review of Law and Economics 24(3): 351–369.
Berkowitz, D., Pistor, K. and Richard, J.-F. (2003). Economic development, legality, and
the transplant effect, European Economic Review 47: 165 – 195.
Bielen, S., Marneffe, W., Grajzl, P. and Dimitrova-Grajzl, V. (2018). The Duration of
Judicial Deliberation: Evidence from Belgium, Journal of Institutional and Theoretical
Economics (JITE) 174(2): 303–333.
Blundell, R. and Bond, S. (1998). Initial conditions and moment restrictions in dynamic
panel data models, Journal of econometrics 87(1): 115–143.
Botero, J. C., Djankov, S., Porta, R. L., Lopez-de Silanes, F. and Shleifer, A. (2004). The
regulation of labor, The Quarterly Journal of Economics 119(4): 1339–1382.
24

Bun, M. J. and Windmeijer, F. (2010). The weak instrument problem of the system gmm
estimator in dynamic panel data models, The Econometrics Journal 13(1): 95–126.
Chemin, M. (2009a). Do judiciaries matter for development? evidence from india, Journal
of Comparative Economics 37(2): 230–250.
Chemin, M. (2009b). The impact of the judiciary on entrepreneurship: Evaluation of pak-
istan’s access to justice programme, Journal of Public Economics 93(1-2): 114–125.
Coviello, D., Ichino, A. and Persico, N. (2015). The inefficiency of worker time use, Journal
of the European Economic Association 13(5): 906–947.
DeJong, D. N. and Ripoll, M. (2006). Tariffs and growth: an empirical exploration of
contingent relationships, The Review of Economics and Statistics 88(4): 625–640.
Dietrich, M. K. (2000). Legal and judicial reform in Central Europe and the former Soviet
Union-voices from five countries, The World Bank.
Dimitrova-Grajzl, V., Grajzl, P., Slavov, A. and Zajc, K. (2016). Courts in a transition econ-
omy: Case disposition and the quantity–quality tradeoff in bulgaria, Economic Systems
40(1): 18–38.
Dimitrova-Grajzl, V., Grajzl, P., Sustersic, J. and Zajc, K. (2012). Court output, judi-
cial staffing, and the demand for court services: Evidence from slovenian courts of first
instance, International Review of Law and Economics 32(1): 19–29.
Djankov, S., La Porta, R., Lopez-de Silanes, F. and Shleifer, A. (2002). The regulation of
entry, The quarterly Journal of economics 117(1): 1–37.
Djankov, S., McLiesh, C., Nenova, T. and Shleifer, A. (2003). Who owns the media?, The
Journal of Law and Economics 46(2): 341–382.
Djankov, S., McLiesh, C. and Shleifer, A. (2007). Private credit in 129 countries, Journal of
financial Economics 84(2): 299–329.
Feld, L. P. and Voigt, S. (2003). Economic growth and judicial independence: cross-country
evidence using a new set of indicators, European Journal of Political Economy 19(3): 497–
527.
Galanter, M. (1981). Justice in many rooms: Courts, private ordering, and indigenous law,
The Journal of Legal Pluralism and Unofficial Law 13(19): 1–47.
25

Grajzl, P. and Silwal, S. (2017). The functioning of courts in a developing economy: evidence
from nepal, European Journal of Law and Economics pp. 1–29.
Hammergren, L. (2014). Reforming the reformers: A different approach to access issues,
Institutional Competition between Common Law and Civil Law, Springer, pp. 349–379.
Hansen, L. P. (1982). Large sample properties of generalized method of moments estimators,
Econometrica: Journal of the Econometric Society pp. 1029–1054.
Hauk, W. R. and Wacziarg, R. (2009). A monte carlo study of growth regressions, Journal
of Economic Growth 14(2): 103–147.
Hayakawa, K. (2009). On the effect of mean-nonstationarity in dynamic panel data models,
Journal of Econometrics 153(2): 133–135.
Islam, N. (1995). Growth empirics: a panel data approach, The Quarterly Journal of Eco-
nomics 110(4): 1127–1170.
Jappelli, T., Pagano, M. and Bianco, M. (2005). Courts and banks: Effects of judicial
enforcement on credit markets, Journal of Money, Credit and Banking pp. 223–244.
Kleibergen, F. and Paap, R. (2006). Generalized reduced rank tests using the singular value
decomposition, Journal of econometrics 133(1): 97–126.
La Porta, R., Lopez-de Silanes, F. and Shleifer, A. (2008). The economic consequences of
legal origins, Journal of economic literature 46(2): 285–332.
La Porta, R., Lopez-de Silanes, F., Shleifer, A. and Vishny, R. (1999). The quality of
government, The Journal of Law, Economics, and Organization 15(1): 222–279.
La Porta, R., Lopez-de Silanes, F., Shleifer, A. and Vishny, R. W. (1997). Legal determinants
of external finance, Journal of finance pp. 1131–1150.
La Porta, R., Lopez-de Silanes, F., Shleifer, A. and Vishny, R. W. (1998). Law and finance,
Journal of political economy 106(6): 1113–1155.
Levine, R., Loayza, N. and Beck, T. (2000). Financial intermediation and growth: Causality
and causes, Journal of monetary Economics 46(1): 31–77.
Mahoney, P. G. (2001). The common law and economic growth: Hayek might be right, The
Journal of Legal Studies 30(2): 503–525.
26

Melcarne, A. and Ramello, G. (2016). Justice delayed, growth denied: Evidence from a
comparative perspective, Unpublished manuscript .
Murrell, P. (2001). Demand and supply in romanian commercial courts: Generating infor-
mation for institutional reform, Available at SSRN 280428 .
Nickell, S. (1981). Biases in dynamic models with fixed effects, Econometrica: Journal of
the Econometric Society pp. 1417–1426.
Platteau, J.-P. (2015). Institutions, social norms and economic development, Routledge.
Pleasence, P., Balmer, N. J. and Sandefur, R. L. (2013). Paths to justice: a past, present
and future roadmap.
Ramseyer, J. M. (2012). Talent matters: Judicial productivity and speed in Japan, Interna-
tional Review of Law and Economics 32(1): 38–48.
Roodman, D. (2009). A note on the theme of too many instruments, Oxford Bulletin of
Economics and statistics 71(1): 135–158.
Rosales-L´opez, V. (2008). Economics of court performance: an empirical analysis, European
Journal of Law and Economics 25(3): 231–251.
Stock, J. H., Yogo, M. and Andrews, D. W. (2005). Identification and inference for econo-
metric models: Essays in honor of thomas rothenberg, Cambridge Univ Press .
United Nations (2015). Transforming our world: The 2030 agenda for sustainable develop-
ment, Resolution adopted by the General Assembly .
Visaria, S. (2009). Legal reform and loan repayment: The microeconomic impact of debt
recovery tribunals in india, American Economic Journal: Applied Economics 1(3): 59–81.
Voigt, S. and El-Bialy, N. (2016). Identifying the determinants of aggregate judicial perfor-
mance: taxpayers money well spent?, European Journal of Law and Economics 41(2): 283–
319.
Voigt, S., Gutmann, J. and Feld, L. P. (2015). Economic growth and judicial independence,
a dozen years on: Cross-country evidence using an updated set of indicators, European
Journal of Political Economy 38: 197 – 211.
Voitchovsky, S. (2005). Does the profile of income inequality matter for economic growth?,
Journal of Economic growth 10(3): 273–296.
27

Windmeijer, F. (2005). A finite sample correction for the variance of linear efficient two-step
gmm estimators, Journal of econometrics 126(1): 25–51.
Woolf, H. (1996). Access to Justice: final report to the Lord Chancellor on the Civil Justice
System in England and Wales, HM Stationery Office.
Zweigert, K., K¨otz, H. and Weir, T. (1998). Introduction to Comparative Law, Clarendon
Press.
URL: https://books.google.be/books?id=OBuQQgAACAAJ
28

Appendix
A.1. List of Variables
Table A1: Variable definitions and sources - common variables
Variable Description and Definition Source
PANEL A - Time variant
ln(GDPpc) GDP per capita in logs. Real GDP/capita at constant 2011
national prices (in mil. 2011US$).
Feenstra et al. (2015)
ln(ATJ) Log of access to justice proxied by the density of judges. Den-
sity of judges is calculated as number of judges per 100.000
inhabitants, where the population data is from World Bank
(2018).
Authors’ calculation
polity2 The Polity2 score is based on the constraints placed on the
chief executive, the competitiveness of political participation,
and the openness and competitiveness of executive recruit-
ment. The score ranges from -10 to +10, with higher values
indicating stronger democratic institutions.
Marshall et al. (2002)
ju corrupt Judicial corruption decision measuring how often do individ-
uals or businesses make undocumented extra payments or
bribes in order to speed up or delay the process or to ob-
tain a favorable judicial decision. Time variant dummy = 1
if value is higher than the median level in a given year.
Coppedge et al. (2018)
lpc Percentage of population with primary complete education. Barro and Lee (2013)
lsc Percentage of population with secondary complete education. Barro and Lee (2013)
lhc lsc Percentage of population with secondary and tertiary com-
plete education.
Barro and Lee (2013)
PANEL B - Time invariant
EU Dummy = 1 for Europe countries. World Bank
N AMER Dummy = 1 for North America countries. World Bank
LAC Dummy = 1 for Latin America countries. World Bank
SSA Dummy = 1 for Sub-Saharan Africa countries. World Bank
MENA Dummy = 1 for Middle East and North Africa countries. World Bank
APAC Dummy = 1 for Asia and Pacific countries. World Bank
low midinc Dummy = 1 for Lower-middle-income economies. World Bank
up midinc Dummy = 1 for Upper-middle-income economies. World Bank
high inc Dummy = 1 for High-income economies. World Bank
legor uk Dummy = 1 for Common law and = 0 Civil law otherwise. La Porta et al. (2008)
legor fr Dummy = 1 for French civil law. La Porta et al. (2008)
legor ge Dummy = 1 for German civil law. La Porta et al. (2008)
legor sc Dummy = 1 for Scandinavian civil law. La Porta et al. (2008)
access and afford just People can Access and Afford Civil Justice. Measures the ac-
cessibility and affordability of civil courts, including whether
people are aware of available remedies; can access and af-
ford legal advice and representation; and can access the court
system without incurring unreasonable fees, encountering un-
reasonable procedural hurdles, or experiencing physical or lin-
guistic barriers.
The World Justice
Project (2019)
29

Table A2: Summary Statistics
Variable Mean S.D. Min. Max. Obs.
ln(GDPpc) 9.60 0.85 6.36 11.82 241
ln(ATJ) 2.07 0.92 -1.26 3.94 241
polity2 6.94 5.05 -10.00 10.00 230
ju corrupt 0.32 0.47 0.00 1.00 234
lpc 19.34 12.06 0.04 51.01 227
lsc 29.47 14.84 2.92 71.91 227
lhc lsc 41.50 18.15 3.54 84.69 227
EU 0.39 0.49 0.00 1.00 241
N AMER 0.02 0.16 0.00 1.00 241
LAC 0.13 0.34 0.00 1.00 241
SSA 0.02 0.13 0.00 1.00 241
MENA 0.08 0.28 0.00 1.00 241
APAC 0.24 0.43 0.00 1.00 241
low midinc 0.10 0.29 0.00 1.00 241
up midinc 0.28 0.45 0.00 1.00 241
high inc 0.61 0.49 0.00 1.00 241
legor uk 0.21 0.41 0.00 1.00 241
legor fr 0.46 0.50 0.00 1.00 241
legor ge 0.28 0.45 0.00 1.00 241
legor sc 0.04 0.20 0.00 1.00 241
A.2. A New Database on Judges: Method and Sources
We construct a new database on number of judges by collecting and pooling different official
sources: international organizations, ministries, statistical offices and academic publications.
The coverage of each database can be complementary or overlapping as they focus on different
countries and periods.17 Therefore, our contribution is to have collected and unified different
data sources in one database.
During the merging process, we have followed 3 main criteria:
1. Maximize the coverage for a given country with one dataset. This criterion enables
us to minimize the measurement error inherent to different definitions or counting
17To illustrate the complementary case between multiple datasets, lets consider the example of France:
CEPEJ provides data for France and for European countries from 2003 to 2014. The ICPSR provides data
for years 1982 and 1984; whereas Pistor et al. (1999) delivers the first available data point for year 1977. As
a result, we use all three sources as they are complementary to each other. To illustrate the overlapping case
between multiple datasets, lets consider the example of Belgium: CEPEJ provides data for Belgium and for
European countries from 2003 to 2014. However, the Ministry of Justice of Belgium provided us with their
official statistics from 1970 to 2014. As a result of overlapping between the two datasets, we decided to use
only the data from the Ministry of Justice.
30

methods while merging different sources. This rule is important as it minimizes fake
evolutions in a country series.
2. Control for unrealistic change between the two closest merged country-year data points.
If we observe jumps in the number of judges (annual change larger than 50%) between
the two datasets, we do not merge these datasets. There could be a judicial reform at
play that radically changes the number of judges for the exact country-year observation
when we merge different datasets, however this can only be the case for very few
countries. There is a significantly higher probability of large changes in values when
merging different dataset than it is to observe in the same period a drastic judicial
reform.
3. Rank the importance of datasets when merging. In the case of overlapping data with
similar country-year coverage, we chose the more preferred database according to the
specified list: Ministries of Justice and Supreme Courts, National Statistical Offices,
international organizations and academic publications. We prefer national sources
given their responsibility to issue first such statistics. Moreover, we favor more recent
datasets as they are more likely to be precise since they can contain corrected estimates
on the number of judges for previous years.
Overall, our measure of the number of judges is available for 107 countries and covers
the period 1970-2014. This is an unbalanced panel, meaning that we cannot follow each
country-year observation in our cover period. To obtain the density of judges, we divide
the number of judges by population as reported by World Bank (2018). We then calculate
five-year averages to smooth the short-run fluctuations and handle the annual gaps in the
data. Table A3 details the sources, definitions, countries and periods cover by each used
source of data. Table A4 provides descriptive statistics.
31

Table A3: Number of Judges: Sources, Definitions and Coverage
Source Definition Sample used
PANEL A - Interna-
tional Organizations
ICPSR (2010) (United Na-
tions Surveys of Crime
Trends and Operations of
Criminal Justice Systems)
Professional judges or magistrates may be understood to
mean both full-time and part-time officials authorized to
hear civil, criminal and other cases, including in appeal
courts, and make dispositions in a court of law.
67 countries, 1973-
2014 (Online database
and rapports)
CEPEJ (2016) (Council of
Europe European Commis-
sion for the efficiency of jus-
tice)
Total number of professional judges in all types of
courts. Professional judges are those who have been
trained and who are paid as such and whose main func-
tion is to work as a judge and not as a prosecutor; the
fact of working full-time or part-time has no consequence
for their status. It does not include the court clerks that
exist in some member states.
34 European countries
2002-2014 (2002-2009 rap-
ports and 2010-2014 online
database)
UNODC (2016) (United
Nations Office on Drugs
and Crime)
Professional Judges or Magistrates means both full-time
and part-time officials authorized to hear civil, criminal
and other cases, including in appeal courts, and to make
dispositions in a court of law. Includes also authorized
associate judges and magistrates.
24 countries, 2003-
2015 (Online database
and rapports)
OAS (2016) (Organization
of American States)
Professional judges or magistrates understood as both
full-time and part-time officials authorized to hear civil,
criminal and other cases, including in appeal courts, and
to make dispositions in a court of law. Including autho-
rized associate judges and magistrates.
17 countries, 2003-2014
(Online database and
rapports)
EUROSTAT (2016) (Euro-
pean Statistical Office)
Both full-time and part-time officials authorized to hear
criminal and civil cases, including in appeal courts, and
to make dispositions in a court of law. Authorized as-
sociate judges and magistrates are included.
7 European countries, 2002-
2013 (Online database and
rapports)
PANEL B - Public Insti-
tutions
Ministries of Justice and
Supreme Courts
Data for: Botswana, Belgium, Croatia, Estonia, Fin-
land, Ireland, Latvia, Mali, Malta, New Zealand, Niger,
Poland, Romania, Slovenia, South Korea, Sweden,
Uruguay, Venezuela
18 countries, 1970-2014
National Statistical Offices Data for: Australia, Armenia, Qatar and United States 4 countries, 1980-2014
PANEL C - Publications
Albers (2003) Data for: Armenia, Botswana, Cambodia, Ghana, Mex-
ico, Mozambique, Pakistan, Papua New Guinea, Peru,
Trinidad and Tobago
10 countries, 2000
Pistor et al. (1999) Data for: China, France, Germany, Japan, India, South
Korea and Malaysia
7 countries, 1970-1995
Calleros-Alarc´on (2008) Data for: Bolivia, Brazil, Paraguay and Venezuela 4 countries, 1993-2001
Schmiegelow and
Schmiegelow (2014)
Data for: Cambodia, Laos and Tajikistan 3 countries, 1995-2011
Contini (2000) Data for: Austria and Spain 2 countries, 1980-1990
Dakolias (1999) Data for Ecuador 1 country, 1998
K¨uhn (2011) Data for Poland 1 country, 1981
Turner (2009) Data for Serbia 1 country, 2002
IMF (2012) Data for Burundi 1 country, 2005-2010
Note: Overlapping coverage implies prioritization of data sources by its level of reliability and allows for cross-validation of data.
32

Table A4: Number of judges per 100.000 inhabitants - 107 countries, 1970-2014
Country Mean S.D. Min. Max. Country Mean S.D. Min. Max.
Afghanistan 8.95 . 8.95 8.95 Kyrgyz Republic 6.18 0.64 5.32 6.73
Albania 11.42 2.12 8.53 13.27 Laos 5.92 . 5.92 5.92
Algeria 8.72 3.53 3.89 12.32 Latvia 14.91 8.42 6.35 26.05
Argentina 3.6 1.87 1.44 4.75 Lithuania 15.6 8.42 5.39 25.49
Armenia 6.7 0.8 5.95 7.54 Luxembourg 34.43 5.94 25.85 39.27
Australia 4.82 0.18 4.64 5 Macedonia 28.63 4.57 20.79 31.52
Austria 19.71 0.95 18.38 21.04 Malaysia 1.31 0.37 0.83 1.67
Azerbaijan 4.66 1.78 2.52 6.46 Mali 2.58 . 2.58 2.58
Bahamas 9.16 1.49 8.11 10.21 Malta 6.08 2.71 2.85 10.09
Bahrain 11.44 2.27 9.83 13.05 Mauritius 4.66 1.15 3.45 5.73
Barbados 8.2 0.42 7.9 8.5 Mexico 2.15 1.88 0.83 4.31
Belarus 7.43 2.68 4.64 10.27 Moldova 8.9 3.07 4.69 12.45
Belgium 19.72 2.68 16.28 22.9 Mongolia 15.77 0.7 14.96 16.24
Bolivia 8.85 1.35 7.4 10.05 Montenegro 40.14 1.15 39.46 41.47
Bosnia and Herzegovina 21.75 4.25 18.25 26.47 Morocco 10.22 . 10.22 10.22
Botswana 2.82 1.22 2.1 4.22 Mozambique 0.97 . 0.97 0.97
Brazil 7.46 0.99 6.13 8.35 Myanmar 2.72 0.23 2.46 2.86
Bulgaria 14.22 10.48 2.82 30.43 Nepal 1 0.19 0.81 1.2
Burkina Faso 1.81 0.12 1.73 1.9 Netherlands 9.98 3.76 5.77 14.48
Burundi 14.41 6.77 6.62 18.88 New Zealand 3.77 0.87 2.18 4.86
Cambodia 1.11 0.71 0.47 1.86 Nicaragua 0.94 0.34 0.7 1.19
Canada 5.29 3.1 0.68 7.4 Niger 1.84 0.14 1.75 1.94
Chile 5.12 3.19 2.51 10.54 Norway 9.07 1.96 6.67 11.12
China 8.75 6.05 2.58 16.62 Pakistan 1.05 . 1.05 1.05
Colombia 9.93 0.37 9.4 10.34 Panama 7.51 0.83 6.04 8.42
Costa Rica 14.28 7.04 4.98 25.48 Papua New Guinea 2.48 . 2.48 2.48
Croatia 36.7 9.43 22.24 44.55 Paraguay 10.39 0.16 10.27 10.51
Cyprus 6.9 1.93 4.73 9.31 Peru 4.73 2.13 2.99 7.58
Czech Republic 28.55 0.97 27.43 29.19 Philippines 2.17 0.47 1.73 2.85
Denmark 6.68 0.19 6.47 6.83 Poland 17.29 6.7 8.62 26.04
Dominican Republic 6.92 0.09 6.86 6.99 Portugal 12.13 4.93 5.34 18.8
Ecuador 0.97 0 0.97 0.97 Qatar 7.39 1.9 5.37 9.62
Egypt 7.11 3.7 4.5 9.73 Romania 13.45 6.1 5.22 20.56
El Salvador 10.75 0.08 10.69 10.8 Russia 15.98 8.41 6.16 22.91
Estonia 13.8 4.62 5.13 16.86 Saudi Arabia 3.17 0.17 3.05 3.3
Ethiopia 0.25 0.08 0.16 0.3 Serbia 34.54 2.56 32.32 37.35
Finland 16.44 1.51 13.73 18.07 Singapore 1.82 0.56 1.08 2.5
France 9.51 1.68 6.56 10.62 Slovak Republic 22.59 2.6 18.76 25.34
Georgia 6.44 0.99 4.76 7.19 Slovenia 38.97 10.83 25.82 51.51
Germany 22.04 3.13 17.89 26.08 South Africa 3.74 0.55 3.03 4.22
Ghana 0.93 . 0.93 0.93 South Korea 2.76 1.45 1.28 5.37
Greece 17.51 7.77 9.7 31.23 Spain 8.6 2.04 5.07 10.85
Guatemala 0.67 0.02 0.66 0.69 Sweden 11.49 0.73 10.78 12.37
Hungary 22.34 5.97 12.66 28.55 Switzerland 13.15 2.58 10.57 15.41
Iceland 16.23 1.27 15.14 17.79 Tajikistan 3.19 2.36 0.47 4.63
India 1.06 0.08 1 1.15 Thailand 4.23 1.71 2.24 6.35
Indonesia 1.63 0.03 1.6 1.65 Trinidad and Tobago 6.68 0.94 5.6 7.34
Ireland 2.61 0.49 2.07 3.22 Turkey 9.89 0.7 9.19 11.02
Israel 6.81 1 5.55 8.1 Ukraine 9.68 5.35 4.37 17.24
Italy 12 1.33 10.74 14.67 United Kingdom 2.74 0.94 1.57 3.57
Jamaica 2.57 1.01 1.07 3.25 United States 10.12 0.75 9.27 11.32
Japan 2.44 0.2 2.28 2.88 Uruguay 13.19 0.97 11.45 14.4
Kazakhstan 14.99 1.41 13.38 15.97 Venezuela 6.5 0.36 6.1 6.8
Kenya 1 0.24 0.83 1.17
Note: List of all 107 countries with at least one 5-year period observation on the number of judges per 100.000 inhabitants. Countries
with standard deviation = (.) have one 5-year period observation.
33

A.3. Density of Judges and Alternative Measures of ATJ
In this section, we present additional evidence for the validity of our macro-level proxy of
ATJ: the number of judges per capita. Figure A1 shows the partial effects of the log density
of judges across four different measures of ATJ, conditionnal on log GDP per capita. The
log density of judges is averaged across the whole 2000-2014 period to maximize sample size.
All the conditional correlation results presented in figure A1 are robust to additional controls
such as the democracy/autocracy indicators from Polity2.18
Panel a. shows that the positive partial correlation between ATJ and the log density of
judges shown in figure 1 is robust to using alternative years. In particular, we test the other
extreme by taking the earliest available data in The World Justice Project, 2012 instead of
2019.
Panel b. tests the link between the log density of judges and the presence of discrimination
along the judicial process (dimension 2 of ATJ, see section 2.1.) as measured by the 2019
edition of The World Justice Project. In particular, we use the sub-factor 7.2 measuring
“whether the civil justice system discriminates in practice based on socio-economic status,
gender, ethnicity, religion, national origin, sexual orientation, or gender identity.” We find
a positive and significant correlation confirming that our proxy is able to capture dimension
2 of ATJ. It should be noted that the relationship is hampered by Hungary which is clearly
an outlying country in that case.
Panel c. tests the link between our proxy and the overall efficiency of the judicial system
as measured by the share of unsentenced detainees in overall prison population. This is the
official indicator chosen by the United Nations to track progress on the Rule of Law side
of the Sustainable Development Goal 16.3. We find a robust negative correlation, meaning
that the density of judges is positively associated with overall judicial efficiency.
Panel d. shows the partial correlation between our proxy and geographical ATJ (dimen-
sion 3 of ATJ, see section 2.1). In particular, we use a specific variable in the Global Integrity
Index of 2007 and 2008 constructed by the non-profit organization Global Integrity asking
whether “in practice, all citizens have access to a court of law, regardless of geographic lo-
cation.” Even with that noisy measure and a sample reduce to 38 countries, we are able to
find a significant positive relationship between that measure of geographical accessibility to
justice and the density of judges.
18Results available upon request
34

Figure A1: The Partial Effect of the Density of Judges on Alternative Access to Justice
Measures
ETH
MYS
GTM
SGP
ECU
JPN
MEX
GBR
NPL
AUS
NZL
KOR
PHL
DNK
CAN
KHM
NOR
BWA
ZAF
USA
VEN
THA
FRA
ESP
SWE
ITA
CHL
NLD
PAN
BRA
PER
TUR
GEO
FIN
BLR
AUT
COL
DEU
PRT
URY
BEL
KAZ
EST
KGZ
BOL
ALB
ROU
GRC
RUS
UKR
POL
CZE
CHN
MDA
MNG
HUN
BGR
BIH
SVN
HRV
MKD
-.2 -.1 0 .1 .2
People can Access &
Afford Civil Justice
-2 -1 0 1 2
Log Density of Judges
coef = .02311228, (robust) se = .0085035, t = 2.72
Panel a.
ETH
MYS
GTM
SGP
ECU
MEX
JPN
GBR
NPL
AUS
NZL
KOR
PHL
DNK
KHM
CAN
NOR
MUS
BWA
TTO
ZAF
USA
VEN
MMR
THA
FRA
ESP
SWE
ITA
CHL
NLD
PAN
BRA
PER
TUR
GEO
FIN
BLR
DZA
AUT
COL
DEU
PRT
URY
BEL
KAZ
KGZ
EST
BOL
ALB
ROU
GRC
RUS
UKR
POL
CZE
CHN
MDA
MNG
HUN
CRI
BGR
BIH
SVN
HRV
MKD
-.4 -.2 0 .2 .4
Civil Justice is
Free of Discrimination
-2 -1 0 1 2
Log Density of Judges
coef = .02977716, (robust) se = .01605587, t = 1.85
Panel b.
GTM
MYS
ECU
SGP
MEX
JPN
IRL
KHM
PHL
GBR
AUS
KOR
NZL
BWA
ZAF
MUS
DNK
CAN
TTO
AZE
THA
VEN
ISR
USA
NOR
CHL
CYP
MLT
PAN
BRA
PER
FRA
ARM
ESP
GEO
ITA
SWE
CHE
KGZ
TUR
NLD
BLR
DZA
COL
ISL
BOL
URY
FIN
KAZ
PRT
ALB
AUT
EST
DEU
MDA
UKR
BEL
ROU
MNG
RUS
LVA
LTU
GRC
CRI
POL
SVK
CZE
BIH
HUN
BGR
LUX
MKD
SVN
HRV
-20 0 20 40
Share of Unsentenced Detainees
in Overall Prison Pop.
-2 -1 0 1 2
Log Density of Judges
coef = -5.1363614, (robust) se = 1.6132017, t = -3.18
Panel c.
ETH
JPN
ECU
MEX
CAN
USA
NPL
PHL
ZAF
FRA
ESP
ITA
KHM
THA
CHL
AZE
TUR
PER
BLR
DZA
ARM
GEO
COL
KAZ RUS
ROU
KGZ
LTU
POL
LVA
UKR
HUN
CRI
CHN
BGR
MDA
BIH
MKD
-40 -20 0 20 40
Geographic Access to Justice
-2 -1 0 1 2
Log Density of Judges
coef = 8.1671518, (robust) se = 3.5341762, t = 2.31
Panel d.
Note: This figure plots the relationship between the log density of judges (averaged between 2000-2014) and
four measures of the accessibility and efficiency of the judicial system: the access to civil justice score of
The World Justice Project in year 2012 (panel a.), the civil justice free of discrimination score of The World
Justice Project in year 2019 (panel b.), the share of unsentenced detainees in overall prison population from
UNODC in year 2016 (panel c.) and a score describing geographical access to justice from Global Integrity
in year 2007-2008. This adjusted partial residual plots are based on OLS regressions where the log GDP per
capita is used as a control variable and with robust standard errors.
35

A.4. Determinants and correlates of the Density of Judges
In this section, we look at the cross-country determinants and correlates of the density of
judges. We estimate the following equation by OLS in an unbalanced panel setting:
ln(Judgespc)i,t =α+β1ln(yi,t) + β2LOi+γ0Xi,t +δt+εi,t (5)
Where the dependent variable Judgespc is the number of judges per 100.000 inhabitants
in country iat time t,yi,t is GDP per capita in country iat time t,LO is the legal origin
of country i,Xit is a vector of other explanatory variables in country iat time t,δtis a
time fixed-effect and εi,t is an error term. Each period tcorresponds to a five-year average.
Depending on the specification, Xi,t will contain the following variables: Polity2 score of
democracy, the share of population having completed tertiary education, absolute latitude,
percentage of population at risk of contracting malaria, ethnic fractionalization, percentage of
population of European descent and share of population of a given religion. The mentioned
variables are good candidates for explaining the cross-country variance of the density of
judges and are widely used in the macroeconomic literature. Their sources and definitions
are available in table A2.
The main finding of table A5 is that the legal origin is a robust predictor of the density
of judges. We find that civil law countries of either French, German or Scandinavian legal
origin, have significantly higher density of judges than comparable common law countries.
Based on our preferred specification (column 3), being a French legal origin country increases
the density of judges by 62% on average and being a German legal origin country more than
double the density of judges, ceteris paribus. Moreover, inside the civil law legal family,
we find that German legal origin countries have significantly higher density of judges than
comparable French legal origin countries. These findings are consistent with the comparative
law literature describing common law as an adversarial dispute resolution system, where the
judge has less investigative powers than lawyers, while civil law countries dispute resolution
system is inquisitorial, giving more power to the judge (Zweigert et al.; 1998). Common law
legal system requires therefore less judges and more lawyers than civil law.19 Differences
inside the civil law family can be explained by specificities of the German legal origin in
terms of judicial organisation (Schmiegelow and Schmiegelow; 2014). Finally, we confirm
figure 3 showing that the share of population of European descent is a strong predictor of
the density of judges.
Column 1 shows a positive and significant relationship between each of the three civil
19Explaining the cross-country differences in the number of lawyers, Massenot (2012) finds that common
law and French civil law countries have more lawyers than German and Scandinavian legal origins countries.
36

law legal origin dummies (French, German and Scandinavian) and the log of the density of
judges, controlling for the log of GDP per capita and year fixed-effects. Adding a measure
of democracy and tertiary completed education (column 2) or a full set of macro controls
proxying temperature, disease environment, colonisation patterns, ethnic composition of the
population or cultural traits (column 3) does not alter the significance of the relationship
found, except for Scandinavian legal origin coefficient which is still significant at the 10%
level. Concerned by the possibility that post-soviet countries have changed their legal origin
after the fall of the Berlin Wall (La Porta et al.; 2008), columns 4 to 6 replicate the first three
columns taking the 1990-2014 period. Consistent with our previous findings, we confirm the
highly significant and sizeable effect found for the three civil law legal origins. Performing
a Wald test in each of the 6 mentioned columns, we are able to reject the hypothesis that
the three civil law legal origin dummies are jointly insignificant at the 1% level, confirming
the significance of the effect found. In columns 3 and 6, testing the equality of coefficients
between the three civil legal origin families, we can reject the null hypothesis only when
comparing German and French legal origins. It means that, among the civil law legal family,
German legal origin countries enjoy on average more judges than their French counterpart,
while Scandinavian legal origin countries are the in between category.
Among the control variables, one other important factor explaining the cross-country
variance in the density of judges is the share of population of European descent (columns 3
and 6). This reflects our previous finding figure 3 as Europe is the host of nations with the
highest density of judges worldwide. Furthermore, outside Europe this variable reflects the
link between the colonization pattern and current institutions (Acemoglu et al.; 2005).
37

Table A5: Determinants and Correlates of the Density of Judges
(1) (2) (3) (4) (5) (6)
1970-2014 1970-2014 1970-2014 1990-2014 1990-2014 1990-2014
ln(GDPpc) 0.455*** 0.346*** 0.122 0.455*** 0.364*** 0.123
(0.095) (0.085) (0.100) (0.096) (0.089) (0.119)
French Legal Origin 0.919*** 0.950*** 0.623*** 0.990*** 0.995*** 0.604***
(0.162) (0.154) (0.211) (0.160) (0.154) (0.217)
German Legal Origin 1.243*** 1.276*** 1.037*** 1.425*** 1.464*** 1.117***
(0.256) (0.289) (0.260) (0.229) (0.261) (0.264)
Scandinavian Legal Origin 0.742*** 0.729*** 0.728* 0.734*** 0.686*** 0.678*
(0.264) (0.274) (0.370) (0.243) (0.253) (0.346)
Polity2 0.024** 0.011 0.018 -0.006
(0.012) (0.015) (0.014) (0.019)
% of Tertiary Completed 0.007 0.001 0.008 0.000
(0.012) (0.011) (0.012) (0.010)
Absolute latitude -0.000 0.000
(0.007) (0.007)
% of pop. at risk of contracting malaria -0.349 -0.490
(0.440) (0.409)
Ethnic fractionalization 0.336 0.253
(0.350) (0.369)
% of population of European descent 1.155*** 1.180***
(0.326) (0.324)
Share of Protestants in the population -0.004 -0.005
(0.004) (0.004)
Share of Roman Catholics in the population -0.001 -0.001
(0.002) (0.003)
Share of Muslims in the population 0.004 0.003
(0.003) (0.003)
Year FE Yes Yes Yes Yes Yes Yes
adj. R-sq 0.43 0.45 0.60 0.41 0.43 0.60
Countries (N) 104 92 89 104 92 89
Observations (NT) 465 416 407 365 323 314
Note: the dependent variable is the number of judges per 100k inh. Standard errors are clustered at country level. * p<0.1, ** p<0.05,
*** p<0.01.
A.5. Supplementary Results
38

Table A6: Pooled OLS and Fixed-effects Results
OLS FE OLS FE
(1) (2) (3) (4)
1970-2014 1970-2014 1990-2014 1990-2014
L.ln(GDPpc) -0.045*** -0.374*** -0.054*** -0.623***
(0.009) (0.071) (0.012) (0.106)
L.ln(ATJ) 0.005 0.089* 0.012 0.180***
(0.010) (0.045) (0.008) (0.051)
cons 0.585*** 3.382*** 0.543*** 5.572***
(0.084) (0.626) (0.113) (0.969)
Time FE Yes Yes Yes Yes
Country FE No Yes No Yes
Adj.-R2 0.18 0.36 0.23 0.51
Countries (N) 83 83 83 83
Observations (NT) 356 356 260 260
Note: Our dependent variable is five-year average growth rate of GDP per
capita. Standard errors are clustered at country level. * p<0.1, ** p<0.05, ***
p<0.01.
Table A7: Robustness of benchmark to different moment conditions
(1) (2) (3) (4) (5) (6) (7)
L.ln(GDPpc) -0.705*** -0.709*** -0.708*** -0.703*** -0.702*** -0.560*** -0.702***
(0.086) (0.086) (0.085) (0.087) (0.086) (0.151) (0.086)
L.ln(ATJ) 0.860*** 0.849*** 0.805*** 0.877*** 0.872*** 0.907*** 0.832***
(0.216) (0.217) (0.217) (0.217) (0.214) (0.253) (0.192)
First ln(GDPpc) 1 1 1 1 1 6 1
Last ln(GDPpc) 4 4 3 4 4 7 5
First ln(ATJ) 2 2 2 3 4 2 6
Last ln(ATJ) 2 3 3 4 5 3 7
AR2 0.498 0.515 0.553 0.493 0.490 0.533 0.524
Hansen 0.344 0.435 0.391 0.411 0.504 0.820 0.532
KP-LM 0.001 0.002 0.006 0.002 0.002 0.001 0.003
KP F-stat 6.18 5.09 4.80 4.83 4.80 4.68 4.04
KP-W 0.022 0.055 0.082 0.072 0.074 0.091 0.153
Instruments 12 13 12 13 13 11 14
Countries (N) 83 83 83 83 83 83 83
Observations (NT) 241 241 241 241 241 241 241
Note: Table reports two-step difference GMM estimations with Windmeijer-corrected standard errors in parentheses.
All regressions include period fixed effects. Our dependent variable is five-year average growth rate of GDP per capita.
All the estimations report different instrument structure for each of the two right hand side variables. The matrix of
instruments is collapsed across all estimations. Standard errors are clustered at country level. * p<0.1, ** p<0.05, ***
p<0.01.
39

Table A8: Heterogeneity by Different Time-periods
(1) (2) (3) (4) (5) (6) (7) (8) (9)
period1 period2 period3 period4 period5 period6 period7 period8 post1990
L.ln(GDPpc) -0.707*** -0.712*** -0.708*** -0.708*** -0.700*** -0.707*** -0.705*** -0.705*** -0.712***
(0.087) (0.087) (0.089) (0.085) (0.088) (0.081) (0.084) (0.087) (0.083)
L.ln(ATJ) 0.857*** 0.846*** 0.867*** 0.865*** 0.889*** 0.901*** 0.843*** 0.859*** 0.872***
(0.217) (0.217) (0.222) (0.213) (0.236) (0.216) (0.212) (0.218) (0.210)
L.ln(ATJ) X Θ 0.020 0.074 -0.078 0.026 0.027 -0.043 0.016 -0.002 -0.032
(0.045) (0.048) (0.064) (0.027) (0.032) (0.028) (0.026) (0.051) (0.049)
AR2 0.494 0.464 0.412 0.522 0.563 0.651 0.525 0.507 0.572
Hansen 0.348 0.309 0.315 0.373 0.364 0.498 0.367 0.194 0.365
KP-LM 0.001 0.001 0.001 0.001 0.006 0.001 0.002 0.000 0.001
KP F-stat 6.12 4.95 4.76 5.28 4.30 4.90 4.61 6.11 4.88
KP-W 0.014 0.044 0.054 0.030 0.089 0.046 0.064 0.014 0.048
Instruments 13 13 13 13 13 13 13 12 13
Countries (N) 83 83 83 83 83 83 83 83 83
Observations (NT) 241 241 241 241 241 241 241 241 241
Note: Table reports two-step difference GMM estimations with Windmeijer-corrected standard errors in parentheses. All regressions include period
fixed effects. Our dependent variable is five-year average growth rate of GDP per capita. Θ is our interaction term component which takes the
value of the corresponding title of the column. All the estimations keep the same collapsed instrument lag structure by treating the lag dependent
as predetermined (instrumenting it from lag 1 to lag 4) and the variable of interest together with its interaction term components are treated as
endogenous (instrumenting it only with lag 2). Standard errors are clustered at country level. * p<0.1, ** p<0.05, *** p<0.01.
40

Appendix references (not cited in the main text)
Albers, P. (2003). Evaluating judicial systems a balance between variety and generalisation.
European commission for the efficiency of justice (CEPEJ).
Barro, R. J. and Lee, J. W. (2013). A new data set of educational attainment in the world,
1950–2010. Journal of development economics, 104:184–198.
Calleros-Alarc´on, J. C. (2008). The unfinished transition to democracy in Latin America.
Routledge.
CEPEJ (2016). Council of europe european commission for the efficiency of justice. Retrieved
from https://www.coe.int/en/web/cepej.
Contini, F. (2000). European database on judicial systems. European Research Network on
Judicial Systems.
Coppedge, M., Gerring, J., Knutsen, C. H., Lindberg, S. I., Skaaning, S.-E., Teorell, J.,
Altman, D., Bernhard, M., Cornell, A., Fish, M. S., et al. (2018). Varieties of democracy
(v-dem) database v8.
Dakolias, M. (1999). Court performance around the world: a comparative perspective. The
World Bank.
EUROSTAT (2016). European statistical office. Retrieved from https://ec.europa.eu/
eurostat.
Feenstra, R. C., Inklaar, R., and Timmer, M. P. (2015). The next generation of the penn
world table. American economic review, 105(10):3150–82.
ICPSR (2010). United nations surveys of crime trends and operations of criminal justice
systems series, waves 1-10, 1970-2006. Retrieved from https://www.icpsr.umich.edu/
icpsrweb/ICPSR/studies/26462.
IMF (2012). Burundi: Poverty reduction strategy paper 2. IMF Country Report, 12(224).
K¨uhn, Z. (2011). The judiciary in Central and Eastern Europe: mechanical jurisprudence
in transformation? Brill.
Marshall, M. G., Jaggers, K., and Gurr, T. R. (2002). Polity iv dataset. Computer file.
Massenot, B. (2012). Why do some countries have so many more lawyers than others?
Working Paper.
41

OAS (2016). Organization of american states. Retrieved from http://www.oas.org/dsp/
Observatorio/database/indicatorsdetails.aspx?lang=en&indicator=72.
Pistor, K., Wellons, P. A., and Sachs, J. (1999). The Role of Law and Legal Institutions in
Asian Economic Development, 1960-1995. Oxford University Press, USA.
Schmiegelow, M. and Schmiegelow, H. (2014). Institutional Competition between Common
Law and Civil Law. Springer.
The World Justice Project (2019). World justice project rule of law index. Retrieved from
The World Justice Project, http://worldjusticeproject.org.
Turner, B. (2009). The Statesman’s Yearbook 2010: The Politics, Cultures and Economies
of the World. Springer.
UNODC (2016). United nations office on drugs and crime. Retrieved from https://www.
unodc.org/unodc/en/data-and-analysis/statistics.html.
World Bank (2018). World development indicators database. The World Bank.
42

ISSN 1379-244X D/2019/3082/9