E CONOMIC GROWTH AND GOVERNANCE: A ROBUSTNESS ANALYSIS MSc International Economics and Business Thesis

MSc International Economics and Business Thesis

E

CONOMIC GROWTH AND GOVERNANCE:

A ROBUSTNESS ANALYSIS

Emmanuel Pandu Nugroho (S 1344641)

Msc International Economics and Business

University of Groningen Groningen, the Netherlands e-mail: e.p.nugroho@student.rug.nl

Supervisor

: Prof. Dr. H.H. van Ark

A

BSTRACT

Finding the most significant determinant of growth is the major attention of economists recently. Previous studies have shown that governance matters for economic development outcome. This thesis conducts an extensive robustness analysis regarding the relationship between economic growth and governance. Using two robustness tests (Extreme Bound Analysis and Cumulative Density Functions), we show that governance is robust in explaining variations of income growth. The thesis examines two channels – productivity changes and investment – on how governance affects growth. Other findings are that several economic, social, and geographic conditioning variables are also robust in determining the rate of growth. Some of these conditioning variables, including some political variables, are proven to be important in giving governance its biggest impact on income growth.

T

ABLE OF CONTENTS

A

BSTRACT ii

A

CKNOWLEDGEMENTS iv

Chapter 1:

I

NTRODUCTION 1

Chapter 2:

T

HEORETICAL REVIEW 3

Chapter 3:

M

ODELLING AND METHODOLOGY 11

Chapter 4:

R

ESULTS AND DISCUSSION 23

Chapter 5:

G

OVERNANCE INDEX ANALYSIS: AN EXTENSION 37

A

CKNOWLEDGEMENTS

This thesis is written as a final prerequisite for obtaining Master of Science degree in International Economics and Business, in the University of Groningen, The Netherlands. Therefore, I would like to thank several people who have lent their hand during the completion of this work.

First of all, I kindly gratitude Prof. Dr. Bart van Ark and Drs. Ewout Frankema for their extensive supervision and assessment; also Prof. Dr. Jakob de Haan as the second assessor for this thesis. I also thank Groningen Growth and Development Center, particularly Dr. Jan-Pieter Smits and Dr. Herman de Jong, for the financial assistance so that I could reach this point of study. A special thanks to Meneer Harry Seldadyo for all fruitful discussions during the process. Moreover, I would never forget my parents, mom and dad, and also my dearest sister in Indonesia for all love, prayer and support they have given. They are more than anything the world could grant to me. I also would like to show my deep gratitude to all my friends here in Groningen, whom I could not mention, each and every one of them. They are just too many, and they all are too precious to be forgotten. Last but not least, to Ikarina Ratna Kusuma, for brighten up my days and nights, and for being my inspiration, during the writing of this thesis. I would never be able to finish the work without help from the people mentioned above.

This thesis is still far from perfectly done, and I appreciate all constructive comments and suggestions to improve it. Hopefully, I could still give my further contribution to this field of study, to the empirics of economic growth and governance.

Groningen, 4th _{of July 2006}

Chapter 1

I

NTRODUCTION

“The only justifiable purpose of political institutions is to ensure the unhindered development of the individual”

Albert Einstein, physicist (1879-1955)

For decades now, economists have sought for the most important determinants of growth. Ranging from the basic concept of factor accumulation (Solow, 1956; Mankiw et al., 1992) to the importance of total factor productivity (Easterly and Levine, 2001), and including an abundant potential of other explanatory variables (eg.: Levine and Renelt, 1992; Barro and Sala-i-Martin, 1995; Sala-i-Martin, 1997), all of these are prospective playground for growth economists. Rodrik (2003) summarized these determinants into only two categories: proximate and deep determinants. Factor accumulation and productivity are included as proximate determinants, while deep determinants consist of geography, trade, and institutions. This thesis emphasizes the importance of institutions – through governance – in explaining the variation in economic growth between countries.

Early studies have shown that governance matters in explaining economic development outcome. Kauffmann et al. (1999a) found a strong causal relationship from better governance to better development outcomes. His six governance indicators are all associated with higher income per capita, lower infant mortality, and higher adult literacy. Another appealing research result originates from Paolo Mauro (1995, 1997), which empirically showed that corruption worsens countries’ income growth through lower investment. Furthermore, Keefer and Knack (1997) explained the importance of institutions by using a new concept of “social infrastructure”, which was derived from indexes of law and order, bureaucratic quality, corruption, risk of expropriation, and government repudiation of contracts. Finally, Hall and Jones (1999) discovered a strong relationship between these indexes and levels of income per capita.

model specifications, yet with results which were still dependent on conditioning variables. In other words, changing the model specification or variables leads to different conclusions. Hence the main idea of this thesis is to check the robustness of governance variables in explaining economic growth more extensively.

Following the general acceptance of two concepts of robustness criteria by Leamer (1985), Levine and Renelt (1992), the so-called Extreme Bound Analysis (EBA) and Sala-i-Martin’s (1997) concept of Cumulative Density Function (CDF), the research objective in this thesis is to test whether governance is robust in explaining economic growth. Furthermore, we also question specific conditions that give greater importance for governance, in its relation to income growth.

The analysis here utilizes approximately 85 variables taken from various sources, in order to run thousands of iterated regressions, consequently alternating specifications, with the aim to finally reach conclusions regarding robustness. We use quality of bureaucracy, sound rule of law, and cleansing of corruption, as indicators of good governance.

We conclude that these governance variables are truly robust in explaining economic growth. In our extensive analysis, we also construct a new index – the so-called GOVERNANCE index – which is composed from the three governance variables. Again, we find a robust and positive relationship in all model specifications. Other important findings from this thesis, concern the role of conditioning variables that could potentially explain economic growth also. Our evidence partly confirms and partly contradicts previous studies. Further, we analyze the causal link between governance and growth, and show that governance influences growth through two channels – a direct channel through productivity and an indirect channel through investment. Lastly, we find several conditions that give greater importance to governance in determining growth rate.

Chapter 2

T

HEORETICAL REVIEW

Before we begin with our main analysis, we need a good understanding of the matters discussed in this thesis. This chapter sets out theoretical foundation concerning economic growth and governance. We shall discuss the state of the art concerning the theory of economic growth, its determinants, and the concept of governance itself. Finally, this chapter provides a background discussion on robustness test.

2.1. Economic growth and governance

Economic growth is central to the study of economics. Raising questions such as why one country is richer than the other, or why offspring’s income increases multifold compared to their parents, is of economists’ interest in this field of study. Barro and Sala-i-Martin (1995) stated that in order to understand why countries differ in standards of living, one needs to grasp the reasons for such sharp divergences in long-term growth rates. To understand large differences in long-long-term growth rates between countries, one needs to comprehend the basics of the theory of economic growth.

According to Rodrik (2003), these accumulation factors are called “proximate” determinants of growth. Following the work of economic historians and scholars on growth, Rodrik stressed that there is “deeper” source of growth. These include geography, trade (market integration), and institutions.

Figure 2.1

Determinants of income

Source: Rodrik (2003)

Figure 2.1 shows the association between the determinants of growth. Geography concerns the specific condition given by country’s particular physical location (e.g., latitude, weather and climate, etc). Trade or market integration implies the market size and welfare gain from international trade (e.g., import and export share, terms of trade, openness, etc). Institutions include quality of formal and informal sociopolitical arrangements (e.g., legal system, bureaucracy, other political institutions, etc). Geography, trade and institutions are all influencing income through endowments and productivity (the “proximate” causes of growth).

Income

Factor endowments Productivity

Trade

Geography

However we see that the arrows reflect the endogeneity or causality problem. All determinants show reverse causality direction, except for geography (for it is widely accepted as definitely exogenous determinant of income). Rodrik et al. (2002) discussed how an integrated market, hence more trading activity improves the utility of endowments and productivity; meanwhile, expanded markets could also be the result of increased income or productivity. Another case is that a good institutional setting enhances efficient allocation of resources and increase investment rate, thus income; while actually property rights, law and order and conducive institutional environment could also have been determined by level of income or productivity. Even trade and institution themselves could explain each other. Nonetheless, as has been proven empirically, all of these determinants are significantly correlated with income per capita.

and development. Even La Porta et al. (1999) concluded that good institutions, through established property rights, are instrumental to economic growth. Keefer and Knack (1997) also emphasized the importance of securing property rights.

Yet property rights is just one channel, through which institutions could influence growth and development. Dawson (1998) points out two channels:

1. A direct channel – via productivity changes (i.e., greater efficiency through better resource allocation).

2. Indirect channel – via a higher investment rate because of an improved business environment.

This thesis puts emphasis on the direct channel between institutions and growth1_{. It} changes the aggregate production function, causing variation in productive efficiency across countries. Institutional infrastructure is needed to support efficient allocation of resources. According to Keefer and Knack (1997), its absence is the cause of countries’ well-being divergence. Easterly and Levine (2001) concluded that national policies (as an outcome of sound institutions) are strongly linked with long-run economic growth rates. Yet, the investment link between institutions and growth is also important. The key point here is that institutions that provide dependable property rights, maintain law and order, and align economic incentives with social costs and benefits – hence stimulating production – which are the foundation of long-term growth (Rodrik, 2003).

Yet, how can institutions be best-approximated? Dawson (1998) raised the notion of “institutional infrastructure” to govern these social constraints. Meanwhile, Hall and Jones (1998) defined “social infrastructure” as institutions and government policies that cause differences in capital accumulation, productivity and output per worker. Furthermore, all economic, political, and cultural institutions (La Porta et al., 1999) influence the quality of government. Hence we obtain a concept that could indicate best approximation of institutions that influences development, and that is the one so-called as “governance”.

1 _{In chapter 5, when we deal extensively with our composite index, we shall see that the indirect channel is}

The Oxford Advanced Learner’s Dictionary2 _{describes governance as: [1] the activity}

of governing a country or controlling a company or an organization; the way in which a country is governed or a company (or institutions) is (are) controlled. Kaufmann et al. (1999) have indicated the link between institutions and governance, as they have defined governance as the set of traditions and formal-informal institutions by which authority in a country is exercised for common good. It includes [1] the process by which government is selected, monitored and replaced, [2] the capacity of the government to effectively formulate and implement sound policies, and [3] the respect of citizens for the institutions that govern economic and social interaction among them. The World Bank uses the concept of “governance” which includes creation, protection and enforcement of property rights, provision of sound macroeconomic policies, and an environment in which bureaucrats and administrators are not affected by graft practices. The next concerns measurement of governance. Again, Kauffman et al. (1999) measured governance using their six indicators of governance, i.e., voice and accountability, political instability and violence, government effectiveness, regulatory burden, rule of law, and graft. According to the authors, these measures are the best indicators of the quality of governance in different countries3_{. La Porta et al. (1999)} assessed government performance using measures of government intervention, public sector efficiency, public good provision, size of government, and political freedom. Hall and Jones (1998) even quantified the concept of their “social infrastructure” using measures of wedges between private and social returns. Positive returns are [1] law and order, and [2] bureaucratic quality. Negative returns arise from [1] corruption, [2] risk of expropriation, and [3] government repudiation of contracts. Paulo Mauro (1995) emphasized negative impact from corruption in degrading quality of governance, hence economic development. Summarizing these proxies, Kaufmann and Kraay (2003) plotted three most important measures of governance as: [1] control of corruption, [2] protection of property rights and rule of law, and [3] voice and accountability. These three were

2 _{http://www.oup.com/oald-bin/web_getald7index1a.pl}

3 _{It is important to note that we explore the characteristics of quality of institutions that influence}

used by Knack (2001) as the best indicators (or approximation) of the quality of governance.

Given all of these proxies, we could have taken the six indicators of governance from Kaufmann et al. (1999) and used it as our measures for governance variables. Nonetheless, due to substantial lack of date for earlier years (i.e.: prior to 1984) in the Kaufmann et al. (1999) dataset, we were not able to utilize it. The only possible dataset for our study is from PRS-ICRG (Political Risk Services – International Country Risk Guide), which provides us with complete data series required; i.e., series for Bureaucratic quality, Rule of Law, and Corruption.

In conclusion, there are three major characteristics of governance quality itself; which are the presence of qualified bureaucracy, existence of rule and law, and absence of corruption. Hence these are three governance variables that become our main interest in this study, and will be tested whether these variables robustly determine economic growth.

2.2. Robustness analysis

On the issue of robustness, Levine and Renelt (1992) used cross-country regressions to find an empirical relationship between long-run growth rates and a variety of variables, including economic policy, political, and institutional indicators. Their paper utilized the Extreme Bound Analysis (EBA) method introduced by Edward Leamer (1985), in order to provide robustness of “a-priori” suspected variables of determinants of growth. Leamer put emphasis on the importance of an organized sensitivity analysis, in order to obtain a convincing statistical inference. This method determines if features of post-regression result depend importantly on the way the specification is defined prior-regression. If the regression result (statistical inferences) is consistent, and the same for all choices of linear combination of various variables, then that particular model (or variables) is (are) robust. The estimation is “fragile”, or not robust, if the model specification is specific to such a result, or in other words, when changing the model specifications or conditioning variables alter the result. Using this EBA method, Levine and Renelt proved that many variables that determine growth are actually fragile. Only the share of investment in GDP, initial income, the secondary school enrollment rate and a few other explanatory variables are robustly correlated with cross-country growth rates.

assigning some level of confidence to each of the variables. Obviously, by using these criteria, number of robust variables that potentially explain income growth increases.

There are actually many other criteria regarding robustness checks. Similar to the fate of governance definition, there is no consensus regarding the concept of robustness. Bhattacharya (2004) proved the robustness of deep determinants of growth by changing the measurement of an indicator. Acemoglu et al. (2001) also used several variables that could best approximate the quality of institutions. Rodrik et al. (2002) used alternating number of samples, model specification and methods of regression, in order to find robust results. However, Beugelsdijk et al. (2004), Bengtsson et al. (2005), and Freille at al. (2006) used EBA and CDF for their robustness checks. Following these three latest papers, and arguing that we need a systematic parameter for robustness checks, this thesis also uses these two tests.

Chapter 3

M

ODELLING AND METHODOLOGY

In this chapter, we begin our analysis by deriving our model specification from Robert Solow’s neo-classical growth model (Solow, 1956). The purpose of this is to have a theoretical justification in setting up our empirical model. The next part of the chapter is aimed at explaining the dependent and independent variables included in this thesis, and building preliminary hypotheses regarding the relationship between these variables. Furthermore, the method of conducting step-by-step robustness analysis will be introduced in this chapter.

3.1. Growth theory and the production function

This thesis uses the basic theoretical framework of neo-classical growth model of Solow (Solow, 1956), which then extended further in the influential paper of Mankiw, Romer, and Weil (Mankiw et al., 1992). This section shows only relevant equation, which will be used in the analysis here4_.

The model, as in Solow (1956) and Mankiw et al (1992), utilized a Cobb-Douglas production function with constant returns to scale but diminishing returns to factors. We consider directly the human capital augmented version of Solow model, which then gives the following production function

(1) Y(t)=K(t)αH(t)β(A(t)L(t))1−α−β

α

,

β

>0,

α

+

β

<1 where Y is output, K is capital, H is human capital, L is labor, and A is the level of (labor-augmenting) technology. As in other standard neo-classical growth literature, labor (L) and technology (A) are assumed to grow at exogenously rate of n and g, respectively. By

substitution and taking logarithms, Mankiw et al. (1992) derived the equation for income per capita as:

(2) ln( ) 1 ) ln( 1 ) ln( 1 ) 0 ( ln ) ( ) ( ln

δ

β

α

β

α

β

α

β

β

α

α

_{+ +} − − + − + − − + − − + = A s s gt n g t L t Y h k

Furthermore, in order to capture the transitional dynamics of economy’s path in reaching steady-state, Mankiw et al (1992) provided quantitative predictions regarding the speed of convergence to steady-state, in linear approximation. Hence we have our income growth equation as defined by

(3) ln( ) 1 ) 1 ( ) ln( 1 ) 1 ( )) 0 ( ln( )) ( ln(y t − y = −e−ϕt ₋

_α

α

₋

_β

sk + −e−ϕt ₋

_α

β

₋

_β

sh ln( ) (1 )ln( (0)) 1 ) 1 ( e ϕt n g

_δ

e ϕt y

β

α

β

α

− − _{+ + − −} − − + − −

This neo-classical growth model (equation 3) implies that growth of income depends on accumulation of physical (sk) and human (sh) capital, population growth, and initial level

of income.

Equation (3) can be estimated simply by using average of income growth within a certain period of time as dependent variable, and price of investment goods, secondary enrollment rate in education, population growth and initial level of income at certain point of time as independent variables. This thesis also tries to gives a new insight of how this Total Factor Productivity plays its role in explaining income growth.

(4)lny(t)−lny(0)=b₀ +b₁ln(y(0))+b₂ln(s_k)+b₃ln(s_h)+b₄ln(n+g+

δ

)+b₅ln(A)+

ε

where is the error term. Equation (4) is now more applicable to be tested using any regression method. While empirical studies on growth have proven the significance of initial level of income, physical and human capital, and population growth in explaining income growth (Mankiw et al., 1992; Barro, 1989, 1993; Islam, 1995, etc), we are interested in finding the significance of the A term. Many previous empirical growth studies have also shown the significance of this A term, by using various means of approximation of A term5_{. This thesis’ main idea is again to find the significance of term} A, as represented by institutions, whether it is hold in general. Thus, this term shows the direct link between indicators of institutions and economic growth. Equation (4) is our basic specification (by giving justification for our fixed explanatory variables) in developing the robustness analysis, which will be presented in the next sub-chapter.

3.2. Model specification and data

In order to answer our main research question on the robustness of the relationship between governance variables and economic growth, we first need to understand the structure of our statistical tools. We begin by describing our main model specification, derived from equation (4) above.

Following the work of Leamer (1985) and Levine and Renelt (1992), we can rewrite equation (4) as

(5) y=

α

_jF+

β

_ijx_i +

γ

_jC_j +

ε

_j ∀i,j

5 _{For instance, Rodrik et al. (2002) used measures of institution and geography when estimating the A term,}

where subscript i indexes the variable of interest, and subscript j indexes different combinations of regression (or in other words, different combinations of conditioning variables captured by C ). Equation (5) is our main model specification in robustness analysis, which links to equation (4) above. In essence, equation (5) only rewrites the Solow growth accounting equation in a slightly different manner. Initial income, physical and human capital accumulation are all captured in F term in equation (5), while both xi and Cj terms actually represent the A term (residual or TFP) of Solow’s

growth model. Thus we ask here what factors are included in the residual as growth determinants. This section is dedicated in explaining this model thoroughly, including giving description and sources of data6_.

3.2.1. Dependent variable (y)

We take average of per capita Gross Domestic Product (GDP) growth over the period 1984-2003 as our dependent variable. This series of data is constructed from World Development Indicator 2005 (World Bank). We collected date for 145 countries of observation (N=145), based on the data availability for the year 1984 from Political Risk Services – International Country Risk Guide (PRS-ICRG) dataset. This dataset allows us to collect all required value of data for our variables of governance, which will be explained below.

3.2.2. Fixed variables (F)

According to Levine and Renelt (1992), the F term in equation (5) is a set of variables which are always included in the regression. These are explanatory variables other than the A variable, which are derived directly from Solow’s growth theory and already proven as always significant in empirical growth literatures (mainly by Mankiw et al., 1992). These variables include initial level of per capita income, physical and human capital, and population growth rate. Hence equation (4) provides the justification of using these fixed variables in our model.

6 _{Descriptions of the data and their respective sources can be seen in appendix 3, while the descriptive}

As a proxy for initial level of per capita income, we use log of income per capita (PPP adjusted, in constant dollar) for year 1984, from World Development Indicators 2005 (World Bank). Following Beugelsdijk et al. (2004) we also use price of investment goods ((PPP/exchange rate)*100 in current price) for 1984, from Penn World Table 6.17_. Bengtsson et al. (2005) justified the choice of investment price as proxy for physical capital accumulation, due to its nature of exogeneity (using the share of investment in GDP instead would create endogeneity with respect to growth). For the human capital accumulation variable, we include the average of total gross enrollment ratio for secondary education over the period 1980-1985 from the Barro-Lee (1993) dataset.

Previous studies (Sala-i-Martin, 1997; Beugelsdijk et al., 2004; Bengtsson et al., 2005) have given similar result regarding the relationship between these fixed variables and economic growth. The initial level of income is usually found to be significantly negative (conditional convergence effect), the price of investment goods is also to be found as significantly negative related to growth (higher price discourages investment), and the secondary level of enrollment rate in education tends to be positively significant to growth.

3.2.3. Variables of interest (xi)

Next to the set of fixed variables, our variables of interest are also always included in all regressions. As already explained in previous chapter, we use three different indexes as a proxy for these variables of interest, which are: Bureaucratic quality, Rule of Law, and Corruption, for year 1984, constructed from PRS-ICRG dataset8_{. These variables are underlying the A term in equation (4) above. As these are} our main focus, we establish our countries of observations based on the data availability in this dataset, which then result in 145 observations available (the list of countries of observations can be found in the appendix 2). An important remark regarding these three indexes concerns the interpretation of the index score. Bureaucratic quality is measured on zero-to-four scale, while Rule of Law, and Corruption are measured on zero-to-six

7 _{Alan Heston, Robert Summers and Bettina Aten, Penn World Table Version 6.1, Center for International}

Comparisons at the University of Pennsylvania (CICUP), October 2002.

8 _{Kaufmann et al (1999) mentioned arguably the most complete indexes to measure governance variables.}

scale, where six (or four) is the most favorable condition. Hence the higher (lower) the index score, the better (worse) governance of such country, measured by those three indexes. One should note here that higher value of Corruption indexes means a less corrupt governance. In chapter 5, we shall extend our analysis by constructing our own index of governance, by using these three indexes. This governance index is established by aggregating the weighted indexes of Bureaucratic quality, Rule of Law, and Corruption using the standard deviation of each index series as weighting. We shall see that this new Governance index even gives a more robust result, compared to the first three indexes.

3.2.4. Switch (conditioning) variables (Cj)

Another part of the A term in equation (4) concern switch or conditioning variables. Levine and Renelt (1992) indicate that these variables should be significantly correlated with growth in at least one regression. Beugelsdijk et al. (2004) claimed that these are variables that have previously been proven as potentially relevant explanatory factors for economic growth. We have found and assembled 77 series of data regarding these switch variables.

The whole list of these variables, including its definition and sources is presented in appendix 3. We classify the whole dataset into four different groups; which are: Economics, Social, Politics, and Geography. Table 3.1 below shows this classification. We shall explore these switch variables briefly.

Economic

Table 3.1

Groups of switch variables

Economics Social Politics Geography

Urbanization rate1984 Years of primary schooling Assassinations in politics East Asian Gini index over period

1975-1984 Years of secondary schooling Coups Latin American

Share of investment to

GDP Years of higher schooling Revolution OECD countries

Net Gov. Expenditure Total schooling years Revolution and Coups Sub Sahara Africa Gov. Expenditure in

Education Gross enrollment ratio in primary education War Dummy Transition Economies Gov. recurring education

expenditure Gross enrollment ratio in higher education War Time Area Gov. defense expenditure Worker ratio Political instability Distance Public investment Population growth Political rights Latitude

Export Infant mortality Civil liberties Land-lock countries

Import Life expectancy LO-English Tropical countries

Terms of Trade growth Labor force LO-Socialist Black market premium Ethno-lingualistic fraction LO-French

Log (BMP) Buddha LO-German

Import Tariff Hindu LO-Scandinavian

Non-Tariff barriers Muslim Democracy

Trade Openness Nonreligious Regime

Tariff Restriction Catholic Size of Military

Liquid liabilities Protestant Democratic

Accountability Average of Inflation, over

Social

We include measures of religion, education, and demographics indicators, as proxies for social factors that could potentially explain economic growth. Again, similar as in the economics group, the nature of endogeneity in this set of variables is mixed. Nevertheless, we shall see later in the result part of this thesis that most of these variables give new evidence on its relationship with economic growth.

Politics

Among this whole set of switch variables, political indicators are those which are mostly interesting, as they are also proxies of institutional setting. These variables include legal origin (i.e., common law, civil law, or socialist), political stability (e.g., revolutions, coups, war dummy), and government regime (e.g., democracy, polity, military regime, etc). We could initially suspect that these variables should be highly correlated with our governance indicators, as these political indicators should influence the building and therefore the quality of governance itself (La Porta et al., 1999). It is commonly expected that less political instability, a more democratic (thus higher civil liberties also) and common-law countries should encourage a higher economic rate of growth. Nevertheless, we will also find mixed results regarding the relationship between these political variables and economic growth, as not all of these premises should correctly hold.

Geography

These geographical variables are mostly exogenous, in the sense that none of these are resulted from economic developments. Rodrik (2003) argued geography determines income due to its natural resources endowed. Furthermore, locations also shape the pattern and behavior of specific countries, hence their incomes. We expect that those countries which lie further from equator, close to economic center, and not being in sub-Saharan Africa are growing more rapidly than those which are not.

3.2.5. Tackling causality problem in the model

our analysis. Particularly, we set our variables of interest (Bureaucratic quality, Rule of Law, and Corruption; and later with Governance index) at the initial year (1984) of the period of long-term growth (1984-2003), in order to see how an initial institutional setting could influence the variation in income growth between countries. We also take all other explanatory variables as close as to the initial year (1984) to minimize endogeneity. Of course, reverse causality could also be justified, as Rodrik (2003) showed that institutions can be determined by economic developments, hence become endogenous. Kaufmann and Kraay (2003) explored this causality question in depth. They first considered the effect of governance on per capita income, and concluded that the widening gaps in per capita income between countries are contributed by deep historical differences in institutional quality (there is powerful impact of initial institutional quality on growth in long run). In contrast, Kaufmann and Kraay argued that richer countries are better able to provide competent government bureaucracy, sound rule of law, and less corrupt environment, hence better institutions. Nevertheless whatever the direction of causality is, the conventional wisdom in these studies is that the correlation between governance and income growth is always positive9_{. Hence our} model specification endorses the argument that it is institutional setting (governance) that influences per capita income growth.

3.3. Methodology

In this section, we shall explain our step-by-step analysis in conducting robustness checks. It begins by plotting a scatter diagram in order to obtain a quick glance of what should be expected from the relationship between governance variables and income growth. We then employ an Ordinary Least Squares (OLS) regression which involve fixed and our variables of interest only. Again, this only provides a preliminary sense of the direction of the relationship. Starting our main analysis, we then proceed by screening switch variables to avoid the multicollinearity problem. Hence not all above

9 _{Although this is the accepted norm of thoughts in this area, Kaufmann and Kraay (2003) also provided an}

mentioned conditioning variables will be used in the regressions. Next, we run all regressions for our model specification using software MetaGrowth 1.0 10_{. Analyzing the} results of these regressions, we set our robustness criteria, and check the robustness results for each variable (particularly for our governance variables). The results and analysis for this study is thoroughly explained in the next chapter. Extending our analysis, we then develop our own GOVERNANCE index and repeat the same procedure of analysis as before. Chapter 5 will then show the result and analyze further in what specific conditions governance should robustly explain economic growth.

3.3.1. Screening the switch variables

The most common problem in empirical growth studies (other than endogeneity problem, of course) is multicollinearity. Hill et al. (2001) mentioned several consequences regarding this collinearity symptom. Nevertheless, for our case, having high correlation between explanatory variables would give redundant information, which unable us to extract a meaningful economic relationship or coefficient of estimation.

Thus in order to avoid this problem, we select only few switch variables (out of 77 variables) which have the lowest correlation coefficient against our variables of interest (Bureaucratic quality, Rule of Law, and Corruption). For each variables of interest, we establish a (pair wise) correlation matrix, and select only switch variables which have a correlation coefficient of less than 0.25 (in absolute value). This procedure leaves us with (names of the sets are in brackets):

-. 25 switch variables for Bureaucratic quality (BurQua25 big sample) -. 21 switch variables for Rule of Law (RuleOfLaw25 big sample) -. 28 switch variables for Corruption (Corruption25 big sample)

Multicollinearity symptoms might not appear between variables of interest and switch variables only, but could also appear between switch variables themselves. Thus we have to screen the correlation within switch variables as well. For this we set a threshold of 0.50 correlation coefficient (in absolute value), and establish another correlation

10 _{Source: http://henridegroot.net. See Heijungs et al. (2001) for further description and application of this}

matrix for variables which have already survived the previous 0.25 criteria. From this, we select only switch variables which score less than 0.50. Due to this procedure, our switch variables are condensed even more, which leaves us with11 _{(names of the sets are} in brackets):

-. 19 switch variables for Bureaucratic quality (BurQua25 small sample) -. 17 switch variables for Rule of Law (RuleOfLaw25 small sample) -. 23 switch variables for Corruption (Corruption25 small sample)

In addition to this, we also would like to have a wider set of variables that could potentially explain the relationship between governance and income growth. We weaken our correlation coefficient criteria by establishing (pair-wise) correlation matrix again, yet now we only select switch variables which have a correlation coefficient of less than 0.50 (in absolute value). This procedure leaves us with (names of the sets are in brackets):

-. 50 switch variables for Bureaucratic quality (BurQua50 big sample) -. 49 switch variables for Rule of Law (RuleOfLaw50 big sample) -. 52 switch variables for Corruption (Corruption50 big sample)

Following the same procedure as before, we establish again (pair-wise) correlation matrix and set a threshold of 0.50 correlation coefficient to check multicollinearity between switch variables themselves. From this, we are left with (names of the sets are in brackets):

-. 40 switch variables for Bureaucratic quality (BurQua50 small sample) -. 38 switch variables for Rule of Law (RuleOfLaw50 small sample) -. 39 switch variables for Corruption (Corruption50 small sample)

3.3.2. Regression and robustness checks

Proceeding from these 12 different datasets (of switch variables), we run our model specification as in equation (5) in iterated regressions. For each regression, we always include initial income level, price of investment goods, schooling, and indexes of governance measures. For each regression, we also include only 3 switch variables12_,

hence we always have seven explanatory variables (three fixed variables, one variable of interest, and three switch variables) in each regression.

Regarding robustness checks, first we implement the Extreme Bound Analysis (EBA) method (Leamer, 1985; Levine and Renelt, 1992). For each model j, we obtain an estimate

β

ˆij and standard deviation

σ

ˆij. Upper extreme bound is the maximum value of (6)

_β

_ˆ

₂

_σ

_ˆ

ij ij+

and lower extreme bound is (7)

_β

_ˆ

₂

_σ

_ˆ

ij ij−

Variable xi is robust if both the upper and lower extreme bounds are both of the same

sign (strong or weak sign test). Stricter criteria in Extreme Bound Analysis is the so-called strong EBA (where all coefficients of estimation signs are equal and significant for all regressions), and weak EBA (where 95% of all coefficient of estimation signs are equal and significant for all regressions). Another EBA criteria is the so-called weighted EBA (where 95% of all coefficient of estimation signs are equal and significant for all regressions after being weighted with the log-likelihood). Failure to pass these tests makes a variable to be labeled as not robust or fragile, according to the EBA criteria.

Levine and Renelt (1992) concluded that none of the explanatory variables (except for gross enrollment in secondary level of education) are passing the strong and weak EBA test. Hence nothing can be learned from the theory of growth, using EBA criteria only. Because EBA method is considered to be too strict, we balance the analysis by using Cumulative Density Function (CDF) method of Sala-i-Martin (1997). This method simply takes the confidence interval on the fraction of density function of the estimated coefficient that is positive (lying to the right of zero). In this thesis, we shall apply a common ‘critical fraction’ of 0.95 (or 95%) as in Sala-i-Martin (1997) and Beugelsdijk et al. (2004). In other words, a variable is labeled as robust if the averaged 95% confidence interval of a regression coefficient does not include zero. We label a variable robust if this fraction exceeds 95% or is less than 5%13_.

13 _{Less than 5% criteria means that 95% or more of the distribution is lying to the left of zero, while only 5%}

Chapter 4

R

ESULTS AND DISCUSSION

A hypothesis test of the relationship between governance (Bureaucratic quality, Rule of Law, Corruption) and economic growth is presented here by using scatter diagram and simple OLS regression. We then proceed by screening switch (conditioning) variables which are suitable for performing our main iterated regressions. Using the results from these thousands of regressions, we try to grasp the main inference of the analysis. Some interesting remarks and points will bottom-line our discussion, and lead into our main conclusion.

4.1. What should we expect from growth and governance?

Before we begin our main robustness test, it is useful to first have a sense of what should be expected from the relationship between governance and economic growth. Figure 4.1 below shows a simple scatter diagram for each governance variable against the average of per capita income growth.

Figure 4.1

Scatter plot for three governance variables against income growth14

-1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0 1 2 3 4 5 BURQUA_1984 A V E G D P G 84 _0 3 AVEGDPG84_03 vs. BURQUA_1984 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0 1 2 3 4 5 6 7 RULEOFLAW_1984 A V E G D P G 84 _0 3 AVEGDPG84_03 vs. RULEOFLAW_1984 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 0 1 2 3 4 5 6 7 CORRUPTION_1984 A V E G D P G 84 _0 3 AVEGDPG84_03 vs. CORRUPTION_1984

14 _{Each diagram in figure 4.1 shows the average of GDP per capita growth between 1984 and 2003 on the}

As expected from previous studies, all of these governance variables (Bureaucratic quality, Rule of Law, and Corruption) have a positive linear relationship with economic growth. Loosely speaking, we should expect a positive relationship between growth and governance15_.

Next to these simple scatter diagrams, we also check for preliminary result using Ordinary Least Squares (OLS) regression to find the expected hypotheses for growth-governance relationship. In these regressions, we include only our fixed variables (F) and each time one of our variables of interest (xi). Hence we have the following four

different model specifications. Table 4.1 presents the results for these simple OLS regressions.

Table 4.1

Cross-Country Growth Regressions, 1984-2003

Estimation by OLSa

Variables of interest included

Variables (none) BurQua RuleOfLaw Corruption

Constant 0.794

(1.153) 3.105*** (0.294) 3.133*** (0.295) 3.163*** (0.291)

Log Initial GDP per capita 1984 0.218

(0.389) -0.415*** (0.101) -0.428*** (0.102) -0.449*** (0.102)

Investment price 1984 -0.001

(0.003) -0.002*** (0.001) -0.002*** (0.001) -0.002*** (0.001)

Secondary Enrollment Ratio 1984 0.719

(0.621) 0.651*** (0.144) 0.628*** (0.145) 0.637*** (0.142) Bureaucratic Quality 1984 0.049** (0.023) Rule Of Law 1984 0.046** (0.019) Corruption 1984 0.053*** (0.019) N 89 79 79 79 R2 _{0.114 0.347 0.353 0.367} F-Statistic 3.631 9.843 10.114 10.741 Notes:

a _{The dependent variable is average GDP per capita Growth, 1984-2003. Standard errors are in parentheses.}

** denotes statistical significance at 5% level, while *** denotes statistical significance at 1% level.

15 _{A positive linear relationship between Corruption and average of income growth does not entail a}

One striking outcome in this simple regression is that we found none of any fixed variables – either it is the initial income, physical or human capital – are significant, although their signs are as expected as in previous studies (except for the initial income16_{). The model also explains only for about 11% of the whole relationship when} we only include our fixed variables (F). Nevertheless, things change when we include our variables of interest – either Bureaucratic quality, Rule of Law, or Corruption. All explanatory variables (including constant) are now becoming significant, and the sign for the initial income now turns to verify the “conditional convergence” (Mankiw et al., 1992; Keefer and Knack, 1997) hypothesis. A negative relationship between investment goods’ price and income growth is obvious as an increase in price discourages accumulation of physical capital, while a positive coefficient of estimation of gross enrollment rate in secondary education confirms the importance of accumulation in human capital for income growth. Another striking point is that the coefficient of determination (R2_{) increases by almost four-fold when we include governance variables} in our model. We may conclude that including one of the governance variables increases the explanatory power of the model. As expected and shown by scatter diagrams above, coefficients of estimation for each governance variables also give a significantly positive relationship with average income growth. Hence we predict a very much strong impact of governance in explaining economic growth.

4.2. Results of the robustness test

As mentioned before, we shall initiate this analysis by selecting switch variables which are not highly correlated to Bureaucratic quality, Rule of Law, and Corruption. This is aimed to reduce the problem of multicollinearity between explanatory factors. We proceed in four steps of screening categories, as explained in previous chapter.

Hence we are left with considerably reduced amount of conditioning variables. Table A.5.1 in appendix 5 lists the switch variables that survive the first criteria (big sample of 0.25 correlation coefficient). Continuing with our second screening criteria (small sample of 0.25 correlation coefficient), we eliminate some variables listed in table

A.5.1 of appendix 5. Table A.5.2 in appendix 5 presents the new list of switch variables that survive this second category.

Next to these two initial steps, we would like to see broader specifications which can be applied in general. This appeals for more conditioning variables included in the regressions. Hence we consider less stringent criteria when se select these conditioners. We apply our third and fourth criteria above, in order to look and take wider conclusion in this study. Table A.5.3 and A.5.4 in appendix 5 present the lists of variables when we apply 0.50 correlation coefficient criteria for big and small sample, respectively.

Moving further on, this section presents all results of our robustness analysis. Main iterated regressions from our model specification (see equation (5) in chapter 3) will be run here, and the results will be shown in tables. Nevertheless we will put the tables in the appendix17_{, while summarizing the most important findings here. We} discuss these findings according to groups of variables (fixed variables, variables of interest, and conditioning) which are robust in explaining average income per capita growth.

4.2.1. Fixed variables

One of the most important results in our analysis is that we confirm the validity of theory of economic growth (Solow, 1956; Barro and Sala-i-Martin, 1995). Capital – both physical and human – accumulation and initial income are indeed basic explanation for growth. This again confirms the results from Mankiw et al. (1992), where we found a very significant effect of initial income and capital accumulation on the rate of income growth. What makes this result even more fascinating is that these fixed variables are robust in all model specifications18_{. Furthermore, all signs and directions of} all of these variables are as expected and again verify previous studies. Tables A.6.1, A.6.2, and A.6.3 in appendix 6 give our complete findings. We only present tables which include most conditioning variables here (big sample of 0.50 correlation coefficient criteria). We shall discuss these results for each member of our fixed variables.

17 _{Complete results of regression (Metagrowth 1.0 output form) is available upon request.}

18 _{We take CDF criteria as a default to take conclusion on robustness. In the following tables, CDF column reports the}

Log initial GDP per capita 1984

In all model specifications, we always obtain a robustly negative coefficient estimates for the initial income per capita (in logarithmic term). Initial income is robust not only according to CDF criteria, but also to weak EBA test in few specifications: BurQua25 small sample, RuleofLaw50 – both big and small sample, and in Corruption25 small sample. Hence initial income is always important in explaining growth, regardless the measures of governance that we take.

Another important result concerns the sign (direction) of this fixed variable. All coefficient estimates of log initial income per capita are negative, which implies that lower initial income increases the rate of growth. This result validates the “convergence” hypothesis (Mankiw et al., 1992; Keefer and Knack, 1997). Poorer countries should grow faster than richer countries. Barro and Sala-i-Martin (1995) mentioned several reasons behind this phenomenon, as diminishing returns to physical capital play a big role here. This diminishing returns to capital is more experienced by advanced countries, hence they grow slower than “relatively behind”-countries. Undeveloped countries could gain a technological advantage from developed countries, and then utilize it to lower their cost of industrialization, thus foster growth. Nevertheless we set this convergence trend as conditional when we control for other variables, such as: investment, trade, or even institutions. Yet, we have seen that initial income is robust and insensitive to any model disruptions.

Table 4.6

Robustness of fixed variables for Bureaucratic quality specification Bureaucratic quality (BurQua50 big sample) 50 switch – 19,600 regressions

Variable Mean value of (S.E.) CDF (in %)

Log initial GDP per capita 1984 -2.620 (0.052) 0.1

Investment price 1984 -0.011 (0.000) 1.0

Secondary enrollment ratio 1984 4.120 (0.060) 100.0*

Table 4.7

Robustness of fixed variables for Rule of Law specification Rule of Law (RuleOfLaw50 big sample) 49 switch – 18,424 regressions

Variable Mean value of (S.E.) CDF (in %)

Log initial GDP per capita 1984 -2.860 (0.055) 0.0*

Investment price 1984 -0.012 (0.000) 0.4

Secondary enrollment ratio 1984 4.080 (0.059) 100.0*

Notes: * = also passing weak EBA test. ** = also passing strong and weak EBA test. Standard errors are in parentheses.

Table 4.8

Robust fixed variables for Corruption specification Corruption (Corruption50 big sample) 52 switch – 22,100 regressions

Variable Mean value of (S.E.) CDF (in %)

Log initial GDP per capita 1984 -2.750 (0.054) 0.1

Investment price 1984 -0.012 (0.000) 0.7

Secondary enrollment ratio 1984 4.170 (0.058) 100.0*

Notes: * = also passing weak EBA test. Standard errors are in parentheses.

Investment price 1984

Gross enrollment ratio in secondary level of education in 1984

Sala-i-Martin (1997) used primary school enrollment rate to measure this human capital accumulation, however he even mentioned that results for primary enrollment rate were mixed. Yet, an earlier study by Mankiw et al (1992) has already used secondary enrollment rates. This was then followed by other researches in empirical growth (Levine and Renelt, 1992; Keefer and Knack, 1997; Beugelsdijk et al., 2004; Bengtsson et al., 2005), to use enrollment rate in secondary level of education. By using this measurement, we confirm the findings in these previous studies. We find a robust result regarding this human-capital accumulation variable. According to CDF criteria, secondary enrollment rate is statistically significant in 100.0% of the cases. Furthermore, this fixed variable also passes strong EBA test in RuleOfLaw25 small sample specification, and passes weak EBA in all specifications. This is the only explanatory variable that we find consistently significant in all model specifications. Regarding the sign (direction) of its coefficient estimates, again we validate the findings in all previous literatures. A positive magnitude of this estimate signifies the importance of investment in human capital in order to foster income growth. It means that secondary enrollment rate is extremely important in explaining average income per capita growth.

4.2.2. Governance variables

This section gives the answer to our main research question, as we will obtain results regarding the robustness of our three governance variables; i.e., Bureaucratic quality, Rule of Law, and Corruption. We give the complete results in the appendix, while underpin the core picture here.

Table 4.9, 4.10, and 4.11 present these results. In general, we find a very confirming result, as what we have hypothesized before. We could now certainly believe that governance matters (Kaufmann et al., 1999) in explaining economic growth, in a form as institutions (Dawson, 1998; Rodrik, et al., 2002; Rodrik, 2003; and many other studies). We discuss the results in-depth for each variable.

Table 4.9

Robustness of Bureaucratic quality

Specification categories Mean value of (S.E.) CDF (in %)

BurQua25 big sample 0.305 (0.016) 93.0

BurQua25 small sample 0.307 (0.016) 94.3

BurQua50 big sample 0.342 (0.011) 95.6

BurQua50 small sample 0.340 (0.013) 95.4

Notes: Standard errors are in parentheses.

Bureaucratic quality

We see in table 4.9 that all of the coefficient estimates are in positive sign, and robust. The positive sign verifies our hypothesis that better quality of bureaucracy, leads into a better governance, hence promote economic growth. Bureaucratic quality is robust according to 5% CDF criteria in larger set of switch variables specification (i.e., BurQua50 big sample, BurQua50 small sample). Nevertheless in the other two categories, bureaucratic quality is also robust if we apply 10% CDF criteria (i.e., BurQua25 big sample, BurQua25 small sample). The coefficient estimates’ significance is also consistent throughout all specifications. Consequently, we may conclude that bureaucratic quality is robust as one of the governance elements in explaining income growth.

Rule of Law

Table 4.10

Robustness of Rule of Law

Specification categories Mean value of (S.E.) CDF (in %)

RuleOfLaw25 big sample 0.277 (0.012) 95.1

RuleOfLaw25 small sample 0.293 (0.015) 96.0

RuleOfLaw50 big sample 0.335 (0.008) 97.6

RuleOfLaw50 small sample 0.341 (0.010) 97.9

Notes: Standard errors are in parentheses.

Table 4.11

Robustness of Corruption

Specification categories Mean value of (S.E.) CDF (in %)

Corruption25 big sample 0.329 (0.014) 98.1

Corruption25 small sample 0.335 (0.017) 98.5

Corruption50 big sample 0.327 (0.008) 97.9

Corruption50 small sample 0.328 (0.010) 98.0

Notes: Standard errors are in parentheses.

Corruption

We obtain a very strong similar result, as we see in table 4.11, that corruption is negatively19 _{and robustly related to average income per capita growth, throughout all} specifications. Akin to Bureaucratic quality and Rule of Law, the coefficient estimates of Corruption also consistently passes weighted EBA and is statistically significant at above 98% in all cases. This result does not agree with the argument of Leff (1964) where he suggested that corruption may promote growth. Using this reassuring result, we could conclude that corruption acts as “sand in the machine”, that hampers the process of economic development.

4.2.3. Switch variables

This section gives results regarding the robustness of conditioning variables. In other words, we solve the question of which conditioning variables that robustly explain income growth. Interestingly, there are variables that constantly robust associated with average income per capita growth, e.g., East Asian countries dummy, Buddhist countries, and average per standard deviation of inflation rate. This means that these

19 _{Remember that ICRG data index has reversed the order of Corruption index score, with higher score}

variables are always significant throughout whole model specifications. Table 4.12, 4.13, and 4.14 report switch (conditioning) variables which are robust (according to 95% CDF criteria) in bureaucratic quality, rule of law, and corruption specifications, respectively. Similar to the fixed variables sub-section above, the complete list of results are given in appendix 6 (table A.6.4, A.6.5, and A.6.6). Table 4.15 summarizes these variables according to their groups. We shall discuss these remarkable findings for each group of variables.

Table 4.12

Robust conditioning variables for Bureaucratic quality specifications Bureaucratic quality (BurQua50 big sample) 50 switch

Variable Mean value of (S.E.) CDF (in %)

Labor force 0.000 (0.000) 99.7

East Asia 1.500 (0.042) 99.6

Exporters of Manufacturer 1.240 (0.050) 99.1

Buddha 0.024 (0.001) 98.5

Average/std. dev of inflation 0.382 (0.020) 95.9

Ethnolinguistic fractionalization -1.210 (0.060) 4.4

Protestant -0.016 (0.001) 2.5

Gini index -0.057 (0.002)* 0.5

Sub Sahara Africa -1.630 (0.043)* 0.2

Notes: * = also passing weak EBA test. Standard errors are in parentheses.

Table 4.13

Robust conditioning variables for Rule of Law specifications Rule of Law (RuleOfLaw50 big sample) 49 switch

Variable Mean value of (S.E.) CDF (in %)

East Asia 1.520 (0.042) 99.7

Labor Force 0.000 (0.000) 99.6

Buddha 0.027 (0.001) 99.5

Average/std. dev of inflation 0.390 (0.018) 96.5

Exporters of non-fuel -0.752 (0.031) 3.6

Protestant -0.017 (0.001) 1.7

Gini index -0.051 (0.001) 1.2

Sub Sahara Africa -1.330 (0.048) 0.8

Table 4.14

Robust conditioning variables for Corruption specifications Corruption (Corruption50 big sample) 52 switch

Variable Mean value of (S.E.) CDF (in %)

East Asia 1.670 (0.036) 99.9

Labor Force 0.000 (0.000) 99.6

Buddha 0.024 (0.001) 99.0

Average/std. dev of inflation 0.410 (0.019) 97.3

Exporters of non-fuel -0.831 (0.028) 2.2

Protestant -0.016 (0.001) 2.1

Gini index -0.056 (0.001) 0.6

Sub Sahara Africa -1.690 (0.051)* 0.2

Notes: * = also passing weak EBA test. Standard errors are in parentheses.

Table 4.15

Groups of robust conditioning variables

Economics Social Politics Geography

Average/st. dev of inflation (+) Average inflation (-) Buddha (+) Ethnolinguistic fractionalization (-) --- East Asia (+) Sub Sahara Africa (-) Exporters of

manufacturer (+) Labor force (+) Protestant (-) Exporters of non-fuel (-)

Gini index (-)

Notes: (+) denotes a positive association with income growth, while (-) denotes the opposite.

Economic

fluctuation of inflation level across countries20_{. The results demonstrate a positive} coefficient for this stability of inflation variable. Average per standard deviation of inflation level result indicates that countries with stable inflation generate more investment rate, hence foster income growth. Roughly speaking, this validates the argument that a minimum level of affordable inflation is favorable for growth. Second, we have dummies for countries, whose major export commodities are manufacturing products and non-fuel products, as robust in explaining income growth. However, these two variables are opposite to each other regarding its sign (direction). Exporters of manufacturing products are positively related to income growth, while exporters of non-fuel products21 _{are negatively correlated.}

A positive relationship between exporters of manufacturing products and growth is obvious as it shows the tendency of industrialization progress of such an economy. Nevertheless the interesting finding here is on the negative relationship between non-fuel exporters and growth. These results are in line with Sachs and Warner’s (1997) conclusion that economies with a high ratio of natural resources export to GDP tended to grow slower. Resource-poor economies’ growth rate often actually outperform resource-rich economies’. Sachs and Warner hypothesized that this phenomenon could be explained due to unproductive rent-seeking activities by authorities. Concerning there could be an abuse of power, high natural resources abundance leads to increased rent-seeking, corruption, and poorer overall bureaucratic efficiency. Loosely speaking, this trade issue proves the significance of governance channel in influencing income growth.

The last economic variable that is robust in explaining growth is the inequality of income across countries itself. We found that the coefficient for Gini index is negatively correlated with income growth, which implies that unequal income distribution (higher value of Gini index) reduces the rate of GDP per capita growth. Barro (1999) showed that higher income inequality reduces rate of growth in poor countries, while encourages rate of growth in richer countries. Hence there is a sense of poverty trap regarding this

20 _{Hence we have a new so-called “Corrected Inflation” variable.}

21 _{Standard International Trade Classification (SITC) classifies non-fuel products as: food and live animals}

hypothesis. Poor countries are getting poorer, while richer countries are getting even richer. Our findings definitely support this argument, as we shall see later that Sub Saharan countries are perfect examples for this idea.

Social

Table 4.15 mentions Buddhist and Protestant countries, ethno-linguistic fractionalization, and labor force as robust conditioning variables in explaining growth. First, similar to Sala-i-Martin (1997), we found a positive relationship for Buddhist countries, and a negative relationship for Protestant countries. He reasoned that the explanation behind these interesting results is related to regional dummies variables. Buddhist countries tend to be East Asian countries, while Protestant countries tend to be developed or already advanced countries (e.g., European countries).

Ethno-linguistic fractionalization measures the heterogeneity of a society. La Porta et al. (1999) hypothesized that in ethnically heterogeneous countries (higher value of this index), the tendency for social conflict and ethnic wars is increasing. Furthermore, political theories concluded that in this chaotic situation, governments become more interventionist and less efficient, hence reducing the provision of public goods and lowering the quality of it. This mechanism once again endorses our argument of the importance of governance in fostering economic growth.

A very much robust labor force is explained by the theory of growth itself22_. Labor force is always robust in all model specifications. It is obvious as labor force itself is actually one of the fixed variables included in Mankiw et al.’s study (1992). Our positive relationship between labor force and income growth signifies the importance working-age population growth needed to smoothen production process.

Politics

We find no robust conditioning political variables. Nevertheless, we shall see later in the subsequent chapter, when we discuss our own new index, politics is one of

22 _{The mean value of 0.000 coefficient estimates for labor force does not imply its non-influencing effect to}

the most important conditioning variables group that describes conditions where governance explains income growth at best.

Geography