Abstract Several studies in the past investigate the impact of corruption on various aspects of economic life using the best available measures as viable proxies of real corruption levels. The dominant type of index employed in such research has been the perception-based measure of corruption. Despite its popularity, researchers use caution when interpreting results due to the subjectivity of this type of indicator. The focus of this paper is understanding how information presented in the media affects perceptions of the public regarding the spread of corruption, to emphasise how this link impacts sustainable development of the economy. Since the attention of this thesis is on corruption in public office and how censorship might affect its image, it is expected that the preferred perception-based index understates the level of actual corruption in countries that have reduced media freedom, which in turn underestimates the negative effect of corruption on sustainability. The adopted estimation approach in the hypothesis test allows for an analysis of temporal and cross-country effects on sustainability separately for most factors of development including for the interaction term between corruption and media freedom. The results find no significant evidence in favour of the hypothesis. The estimates of the link between censorship and corruption do have the expected negative sign for temporal effects, but they show positive correlation with sustainable development on a time-fixed cross-country level.
MSc International Economics and Business
The Impact of Corruption on
Sustainable Development: The Role of
Media Freedom
Ivan Romanski S3071219
1/26
1. Introduction
Corruption is a complex phenomenon that has been part of human development since at least as early as ancient Mesopotamia (Delaporte, 2013). Its effects on society and development have been notoriously hard to examine due to the term’s conceptual vagueness and covert nature (UNDP, 2008). Because of the difficulty to conclusively define the concept and observe the precise magnitude of corruption (of various forms), a dominant way of looking at its impact on economic activity is to rely on indices that measure levels of perceived corruption. Such measures base their values on “expert” surveys that aim to proxy actual levels of corruption by accounting for a variety of biases that might affect opinion (Lambsdorff, 2006).Yet, such indexes fail to capture the wider spectrum of subjective influences such as the character of the surveyed, their current mood, or the news presented in media (Galtung, 2006). Several studies in the past have used perception-based measures to analyse the effect of corruption on economic growth, productivity and sustainable development (Mauro, 1995; Mo, 2001; Svensson, 2005; Aidt 2009), but have taken caution in interpreting the results – and some even question their validity (Svensson, 2005; UNDP, 2008) – as “close-enough” depiction of actual effects.
In this thesis I explicitly acknowledge that I use ‘perception of corruption’ as a control for sustainable development – the ability of present generations to produce wealth without compromising the needs of future generations. As alluded above, the media has a profound influence on the popular image of the level of corruption within society (Miller, 2006). This influence is due to the significant role of the media as mediators of information and sources of knowledge in modern societies (Alam and Ali Shah, 2013). As a consequence, in societies where media freedom is limited perceptions of corruption might be increasingly biased towards the image presented in the media. It is likely that the perceived corruption would be lower than actual corruption in places where the corrupt control the media. Previous research has observed a negative impact of corruption on economic progress (de Soto,1990; Mauro, 1995; Reinikka and Svensson, 2005; Aidt 2009) and also that the media plays a significant role in shaping the actions of economic agents via their subjective perceptions (Deaton, 2012). Hence, the negative effect of corruption on economic development (found in the aforementioned studies) might be understated in societies where the corrupt are also the censors. Therefore, while controlling for direct effects of both media freedom and corruption perception, this thesis’ main objective is to examine how the relationship between freedom of the media and perceived corruption affects sustainability.
2/26 for example. It can include virtues such as health, knowledge, and social connections, among other constituents wellbeing. Still, consumption itself should not be disregarded – it is a highly complex activity which provides people not only with the means to survive, but as anthropologists Douglas and Isherwood (1979) assert it is “an instrument of social interaction allowing people to attain their humanity”. Thus, it plays a central role as a determinant of wellbeing.
The maintained hypothesis in this thesis is that sustainable development is negatively affected by the influence of a controlled media on subjective measures of corruption perception. This hypothesis is tested using an interaction term between corruption perception and freedom of the media after controlling for a wide array of other factors that influence sustainability including the direct effects of the two variables of the interaction term. The tests are performed using a cross-country panel that includes data for 104 counties for the period of twelve years between 2002 and 2013. The estimation approach employed for the more inclusive models (those factoring a greater number of the determinants of development) enabling a distinction between the cross-country and the time effects of various controls to take full advantage of the panel. And while the most of the coefficients of the regressions have the expected signs (for instance, negative for the interaction term), the analysis fails to produce significant results for the majority of factors, which is conflicts theory implications and the current hypothesis. However, since regression analysis produces the appropriate signs for the results, their statistical insignificance might be due to the examination of confined period and imperfect factor. Another point of caution in the regression analysis is the use of an interaction term itself. While it is a useful analytical tool for determining whether links between factors influence the dependent variable, its effect is highly dependent on the magnitude of any of the variables it captures. High values of composite variables impact the estimated results of the term depending on their individual significance in the regression, so its effect too should be interpreted with caution. A further disadvantage of the analysis in this paper is that it is constricted to the realm of political corruption. This limitation is in line with what Acemoglu calls “the general convention” (Acemoglu and Verdier, 2000), but is based on how I define this contested concept for the purposes of the analysis and does not defy the important role of private agents in promoting corruption.
3/26
2.
Corruption as Part of the Institutional Setting of a Country: Concept
and Impact – A Review of Literature
This thesis examines the cross-country and intertemporal impact of the relationship between corruption perception and media freedom on sustainability as a determinant of wellbeing. Is sounds like a straightforward analysis of properly defined regression, but one has to consider that the interactions between these socially determined concepts are heavily dependent on local social specificities since all three concepts are culture-contingent (Philip, 2006, for corruption; Tran, 2011, for civil liberties; Dasgupta, 2001, for wellbeing). How they interact and what is the result is dependent on the local environment as it determines the magnitude of said concepts and their influence on one-another, but also constitutes the very notions of corruption and wellbeing. Thus, this section defines the term ‘corruption’ and how it is going to be used in the analysis in order to put into perspective its influence on welfare. It also specifies the determinants of corruption from the local environment and how they are related to economic growth and social wellbeing in general. Lastly, the section provides empirical evidence on the effect of corruption on economic activity both at a firm level and across countries.
2.1. Defining, Specifying and Determining Corruption
4/26 on economic performance (Svensson 2005; Langseth, 2006). Thus, some of the main forms of corruption in terms on their impact on development are presented below.
The most popular specification of corruption focuses on “abuse of public office for private gain” (UNDP, 2008). However, in line with the ‘comprehensive definition’ of the concept, corruption can occur in any organisation where an abuse entrusted power occurs. So, I begin the review of the forms of corruption with transaction that are confined entirely to the private sector, or what is referred to as ‘private corruption’. For example, when a bank employee in view of her personal agenda lends large amounts of money at harmfully low interest and these transactions lead to instability or even default of the bank, the act can be viewed as indirect embezzlement or fraud, since banks lend other people’s capital. What is more important, instability in large private organisations causes economic damage to individuals and even entire economies (Langseth, 2006) through losses of income and reduction in trust required for a stable business environment (Tirole, 1995). These welfare reductions are the reason the Council of Europe has criminalised private section corruption acts such as bribery, trading in influence and account offences (EU, 1998).
Touching on the subject of lawfulness, legal definitions of corruption fail to capture some of corrupt activities where transactions become institutionalised in the state and the economy. This is recognised in recent work on ‘legal corruption’ (Kaufmann and Vicente, 2011) that examines acts which violate the norms or public expectations for the person occupying a position of entrusted power, yet the legal framework of the state fails to criminalise and pursue them. The reason is that some group (‘the elite’) has significant influence over the policy-making within the state and prevents certain forms of corruption to be made illegal for their own benefit. In some countries, even if corrupt acts are de jure criminalised, the law is not enforced for ‘the elite’ which makes corruption for them de facto legal. In their paper, Kaufmann and Vicente (2011) endogenises corruption and the related legal framework in order to identify the possible equilibrium outcomes. The authors are able capture influences induced by the private sector, as ‘the elite’ is not limited to persons of public office. The model produces highest equilibrium for the case of ‘no corruption’ since both legal and illegal forms of moral misconduct are associated with some type of costs (be it opportunity cost of unrest for the public, or cost of setting a legal framework and the associated misallocation of productive talent for the elite). In other words, corruption (illegal and legal) is related to suboptimal economic performance due to increase of costs. While this finding speaks loudly for the importance of not restricting corruption analyses to practices that are outlawed by legislators it is also a sign that examination of overall effects of corruption should not be based on the legality of practices at all. The specification of choice should be based on its influence on the dependent concept.
5/26 the state in social order. One does not need to be ‘egalitarian’, holding a strong belief that that the state has a central role in assuring fair distribution of goods (Rawls, 1971), to find the benefit of government institutions in structuring society. Even in a minimalistic sense Robert Nozick (1974) asserts the need of a “night-watchman state” to constrain what can be done and to seek compensation for wrong-doings. From this ‘utilitarian’ perspective, the state is the “referee” that enforces contracts between parties, when information is deficient. Thus, government institutions are socially desirable as they economise on losses arising from hidden information and unobservable actions. When there is lack of trust between parities due to information deficiencies, the state – and by extension the public servants – are entrusted with power to enable mutually beneficial transactions between people. Furthermore, in practice most governments are entrusted with regulatory power way beyond mere guardianship of rights. Their role is often equated to promotion of social optima (Acemoglu and Verdier, 1998, 2000), which entails sufficient impersonalism and universalism in the conduct public officials, which in turn requires separation between private and official interest of acting public servants (Brown, 2006). So, if there are no government failures (regulations and regulatory practices reducing wellbeing) private corruption is controlled for by the regulator in a way in which the market is socially optimal (maximises social welfare). However, when government officials misuse their position they can influence the very environment through law-making and regulation-applying, undermining the very integrity of institutions that govern economic life. In fact, research has found causal relationship between ‘public corruption’ and market distortions (Aidt and Dutta, 2008).
The role of government institutions in optimisation of social conduct is central for an analysis of the effects of corruption on welfare1. In view of previous research (Acemoglu and Verdier, 1998, 2000; Svensson, 2005) public corruption has a complex relationship with institutions as they interact on multiple levels. First, government institutions act as the medium in which public officials carry out corrupt transaction. Acemoglu and Verdier (1998) argue that, since institutions are socially desirable but some government agents are corruptible, public office corruption is unavoidable to occur and due to its secretive nature corruption it is just too costly to eradicate. Also, in their follow-up paper Acemoglu and Verdier (2000) assert that since there is а trade-off between market failures (inefficient allocation of scarce resources) and corruption-induced government failures, a welfare maximising market does entail some degree of public corruption. However, what is more intricate in the relationship between government institutions and public corruption is that both are subject to similar determinants. This is due to the fact that both concepts
1 My focus on ‘public corruption’ does not disregard the significant impact of other forms of corruption on society, progress,
6/26 are founded in the culture of the local society and are determined by the political, economic and information environments that are endogenous to it. (Acemoglu, et. al., 2004; La Porta, 1998, 1999). The ‘depth’ and ‘breadth’ of corruption within government structures and its effects on welfare are dependent on these determinants and the controls constituted in response to environment. Still, in order to understand how public corruption as an institutional feature affects wellbeing, one first has to understand how the latter is quantified.
The chosen measure for sustainable development in this thesis – ‘growth in genuine wealth’ adopted from the work of Arrow et. al. (2004) – manages to quantify the ability of a country to sustain its standard of living (the determinant of wellbeing). This is done by weighing the ‘productive base’ of an economy by the attainable inter-temporal social costs – the ‘shadow prices’. To elaborate, the ‘productive base’ includes all capital assets of the economy (physical, human, and natural, jointly referred to here as ‘productive capital’), as well as the knowledge base and institutions of a country which determine the efficiency and distribution of production (Dasgupta, 2001). By weighing the productive base on shadow prices, the measure includes a diversity of aspects of well-being of current and future generations since: (1) future forfeitures by the society are captured by current shadow prices, while (2) net production (or “the consumption stream”) encompasses the determinants of welfare which in their total should equal the aggregate of the constituents of well-being (Dasgupta, 2001). Thus, the ability to sustain the productive base measured by the indicator is equated with the ability to sustain intertemporal social welfare. In comparison to alternative popular measures that are often used as indexes of well-being, growth in genuine wealth has the advantage of controlling for time-effects of social development captured by shadow prices. For example, neither GDP, nor net domestic product (NDP), nor even the Human Development Index (HDI) are sensitive to intertemporal changes constituents of well-being as they ignore a diversity of social costs over time. Yet, it should be noted that despite well-being a much more comprehensive measure of sustainability due to inclusion of shadow prices, the current list of costs to society is too short and full of subjectivity-induced inaccuracies for the measure to considered “prefect” (Dasgupta, 2001).
7/26 colleagues (1998, 1999) stress the importance of society’s long-held beliefs formed by cultural influences and regional understanding of morality for the formation of institutions. The authors observe that adoption of a particular religion or transactions with European colonisers have a profound influence on the institutional structure within a county. As a result, institutional characteristics that determine national prosperity are strongly dependent on historical cultural events. In view of these theories, culture and natural endowments can be considered the initial conditions that determine the political power within the country that constitutes the regulatory apparatus of the state. The ‘depth’ of corruption is also determined by these initial conditions since the concept is culturally dependent and has a complex relationship with constituted institutions. Further, the initial conditions construct the culturally-established notions of society and social wellbeing, and the initial distribution of productive capital.
8/26
2.2. Economic Effects of Corruption
There is a body of works that analyses how determinants of corruption such as the economic or the legislative environment, or the political system affect the extent of perceived corruption in a country. For instance, Ades and Di Tella (1999) account for market restrictions with variables for openness to external competition from imports. De Soto (1990) examines the impact of the legislative environment on the spread of corruption to find that the more links there are in licensing procedures, the bigger the opportunity for corruption, yet at low levels of regulation corruption also thrives. As for the effect of political accountability on the extent of corruption Persson and Tabellini (2004) examine parliamentary versus presidential and majoritarian versus proportional political systems in democratic countries. Their finding are mixed as no particular pattern is found. Moving to productive capital, the relationship between income and corruption has been widely agreed upon to be negative with some measure of income levels used as control for corruption in most studies. Concentrating on the specifics of this relationship Saha and Gounder (2013) analyse the income-corruption nexus finding patterns of non-linearity the scale of corruption based on income levels. Education is the other factor that is considered essential for the extent of corruption by all researchers – as shown in Svensson (2005), years of schooling is highly significant for corruption magnitudes controlling for whichever other factors.
Thus, cross-country evidence presented in the aforementioned papers suggest that corruption is significantly negatively correlated to accumulation of productive capital (represented by GDP per capita and educational attainment). Yet the economic, legal, and political environments have a very diverse effect on the extent of perceived corruption for different countries. This finding suggests that magnitude of perceived corruption is environment-specific – increased regulation does not necessarily translate to increased corruption, same as market restrictions or any particular political system. This is an important reminder that the image of what socially unacceptable varies across countries based on cultural specificities and the balance of political power. This results in a various levels of control of corruption and corruption perception. It can be inferred that the extent of corruption is determined by the level complementarity of all factors, but this does not imply that culture-specific views change the effect of corruption on production capital.
9/26 markets of the trusted clients, which in many cases reduces innovation and entrepreneurship. Even further, bribes increase the cost of doing business – costs which are not tax deductible. This in effect also reduces entrepreneurships incentives relative to rent-seeking activities, since new products and businesses are the ones that require more interaction with bureaucracy and incur the higher costs. Through corruption increased opportunity costs for entrepreneurs result in slowed-down economic development (Bardhan, 1997).
There exists an alternative view on the impact of corruption, which tries to explore any benefits of corruption. Leff’s (1964) “greasing the wheels” hypothesis has maintained that corruption helps efficient economic agents circumvent market deficiencies. As the theory goes, imperfect markets that bring about welfare losses because of some exogenous negative influences can be corrected by corrupt practices as they allow market players to “bend the rules” in favour of a preferable outcome. Thus, the benefit of corruption is that it allows for welfare-improving events that are unobtainable within the confines of the market rules. For example, bribe payments may facilitate the fast undergoing of otherwise cumbersome administrative procedures. When time is of the essence because the entrepreneur needs to retain competitive advantage, the bribe option incentivises investment in the entrepreneurial endeavour. The hypothesis stipulates that it is the lowest-cost firm – and hence the most efficient – that would be able to pay the biggest bribe to the corrupt public official as it has the biggest competitive advantage. So the act itself is efficiency improving. Since Leff (1964), the theory has been further developed with prominent works including Huntington (1986) who coined the “greased wheels” phrase and analysed how deficiencies emerge, Lui (1985) and Beck and Maher (1986) who set the economic foundations, and more recently some research has been done in finding empirical evidence for the hypothesis (Levy, 2007).
10/26 fallacy of efficient corruption”. Thus, even theoretically the idea of growth-enhancing corruption is flawed since in the long run societies aim at a first-best scenario of economic environment. Still, empiric evidence are needed to prove the legitimacy of any of the two positions, which is what follows until the end of the section.
Svensson (2005) and Aidt (2009) offer an account of many of the effects of corruption both on a firm level and case studies, and as cross-country comparison based on works by de Soto (1990), Murphy et. al (1991, 1993), Mauro (1995), Choi and Thum (1998), Mo (2001), Svensson (2003), and Reinikka and Svensson (2005), among others. In general, the authors show that on a micro level there is sufficient evidence against efficient corruption due to its detrimental effects to the pillars of economic growth. On its own corruption does not have the most adverse effect on aspects of society. It is related to misallocation of resources and capacities due to survival of inefficient firms, sub-optimal investment, misallocation of entrepreneurial talent by creation of a market for corrupt positions in the administration, and preference of investors toward smaller scale enterprises and the informal sector in order to reduce costs of corrupt bureaucracy. Also, empirical studies show that, ceteris paribus, in countries with similar levels of corruption ‘speed money’ (bribes for facilitating the administration of firm practices) does not benefit, but actually reduces firm growth (Kaufmann and Wei, 1999). The aforementioned studies also suggest that corruption has adverse impact on education attainment, food and health procurement, and poverty elevation which is harmful not only on efficiency considerations but also from a welfare point of view.
11/26 of corruption on particular types of economies. Based on the level of political freedom it can be seen that countries that are more liberal have a growth-maximising level of corruption at the low levels of corruption. For more politically restrictive countries, no such level exists as they are worse off with corruption at any level. Hence, when corruption is occasional it may be efficiency-inducing, whereas in countries with systematic corruption there is no evidence of an effect on growth. From the results it can be concluded that the effect of corruption is dependent on the political regime, but countries sort themselves in particular regimes based on their institutional setting, which on its own affects their growth opportunities. In fact, Saha and Gounder (2013) examined the non-linear effect of income on corruption and found evidence that in countries with low levels of income corruption rises with income whereas mid- and high-income countries have a negative correlation between income and corruption. Thus, over the course of development corruption levels are non-homogeneous, initially increasing but falling naturally after a certain level of earnings. This confirms the idea that corruption may be beneficial at low levels of economic development but becomes an obstacle of progress.
As it follows, the empirical research on the effect of corruption on growth concludes that on a cross-country level it averages to zero as it is highly dependent on local circumstances. It has benefits at small doses for developed countries but a negative marginal effect. Meanwhile, it indeed greases the wheels of cumbersome transition economies that have to focus resources on more immediate threats to growth before they invest in costly anti-corruption structures. However, in developing economies the institutional setting is a second-best scenario which needs to be corrected if the economy is to develop further. Turning the focus to specific factors of growth inside particular economies it has been shown that corruption is efficiency reductive for firms and damaging to investments.
12/26
3. The Influence of the Media on Corruption Perceptions and Economic
Development
The discussion so far has presented the idea that political power and the way it is used is a significant determinant of the spread of corruption within government institutions. In their analysis Brunetti and Weder (2003) maintain that rule of law (proxied by an index of political freedom) is a powerful control for corruption and allows for a more refined estimate of level of misconduct. However, political freedom is a term that encompasses a variety of rights and liberties. It is best summarised by Article 19 of the Universal Declaration of Human Rights adopted by all members of the United Nations: “Everyone should have a right to freedom of opinion and expression” (UN, 1948). Thus, an essential part of political freedom is freedom of speech and its embodiment in journalists representing the free media. The extent of media independence from outside pressures as a component of political freedom is correlated to the degree in which perceived and actual corruption differ. Miller (2006) shows that images of corrupt behaviour of public officials in people who rarely deal with bureaucracy are highly dependent on information from the media and not so much on experience. Thus, how close perceptions of corruption are to actual levels of corruption is highly dependent on the quality of information provided in the media.
13/26 Another benefit of an unrestrained information environment is access to additional information and opinion. To elaborate, consider the measure of sustainability of ‘genuine wealth’. An important aspect of the indicator is input prices derived by information regarding social costs – the ‘shadow prices’. Subjectivity of shadow prices and knowledge of externalities, in combination with attitude toward risk, makes the information environment essential to the computation of the measure. In countries where journalists are economically dependent on companies involved in unsustainable production for example, their reduced ability to inform the public on may have direct negative effects on sustainability due to unawareness of the consequences by society. In fact Alam and Ali Shah (2013) show empirical evidence on the positive correlation between press freedom and economic growth (in terms of GDP) across countries. Their conclusion is that, although there is high degree of simultaneous causality, a free media provides independent feedback on government policies and firm practices. In fact, independent in-depth press coverage is able to provide comprehensive analyses from a wide spectrum of opinions free of charge for a very short period of time. This makes the independent media not only informative for the public but also constructively critical for key stakeholders. Thus, impeding this information channel by limiting media freedom would translate not only to unsustainability form a welfare point of view, but also to inefficiency from purely productive perspective.
14/26 philosophical perspective this would mean that freer press contributes to welfare on the grounds of palpable wellbeing. Of course this would not mean that society feels better off, but it provides a ‘solid foundation’ for further development.
4. The Impact of Media Freedom on Corruption Perception and thereby
Sustainable Development: The Methodology
Returning to the tangible aspects of economic development such as increases in the determinants of life quality, the objective of this thesis is to examine the effect of the link between corruption perception and media freedom on sustainable development. Following the review of literature in the previous sections, I hold that the impact of corruption will be understated in countries with controlled media since perception-based measures are influenced by information presented in the media. Thus, I postulate the following testable hypotheses:
The interaction between corruption perception and media freedom will have a negative impact on sustainability.
This section discusses the technicalities in deriving the measure of sustainability, presents the preferred indicator of corruption perception, outlines specifics in the measure of media freedom and describes the methodology used in this thesis to address the main research question. While considering the impact of corruption perception, channelled through media freedom, I also control for other potential factors that could influence sustainable development. These control variable are also discussed in this section. The variables employed in the hypothesis tests form panel data of twelve years for the period between 2002 and 2013 for 104 countries. The period has been chosen as it is the most recent time-frame for which an abundancy of information is available for most countries for all examined measures.
Based on the works of Hamilton and Clemens (1999) sustainability is defined as the ability of a society to maintain its living standards through time. Arrow et.al. (2004) refine the notion by asserting that an economy is sustainable when intertemporal social utility is non-decreasing. Thus, economic progress is desirable when it does not compromise social utility over time. This idea relates more comprehensively intertemporal consumption and investment to a general view of welfare since it accounts for their effect on utility within the community over time. In other words, the preferred measure of sustainability – “growth in genuine wealth” (GGW) – is more closely related to wellbeing than traditional measures of production capabilities (such as GDP) on the grounds of intertemporal social utility captured by the attainable shadow prices. Still, it should be noted that the current list of costs to society is far from exhaustive and full of subjectivity-induced inaccuracies which makes GGW a good, but imperfect proxy of sustainability.
15/26 used in this paper is derived starting from an estimate of Adjusted Net Savings (ANS). ANS is part of the World Bank’s World Development Indicators (WDIs) and is computed as Gross National Savings adjusted for capital depreciation and accounting for investment in education and costs of natural resources used and pollutants emitted in production (World Bank, 2016). ANS is a rough estimate of genuine wealth. Using the approach of Arrow et. al. (2004), ANS as percentage of gross national income (GNI) can be converted into growth in genuine wealth per capita by multiplying with a presumed ratio of GNI-to-wealth. This ratio is related to the availability of (physical, human and natural) productive capital in the country since annual national output and income are dependent on the productive base of the economy. The ratios used in this thesis are also the ones adopted in Arrow et. al. (2004) – 0.15 for developing and oil-rich economies and 0.20 for advanced economies. These indexes depict the assumption that advanced economies have generally more expensive productive capital. Continuing with the measure of sustainability, I need to account for changes in production technology and population. Thus the result of the conversion is adjusted for the growth of population within the country and any technological advancements in the economy using specific population growth and Total Factor Productivity (TFP) growth rates. Both rates are provided by the Total Economy Database of The Conference Board (CB) organisation (CB, 2016). The CB estimates TFP growth as a residual from a general growth accounting regression examining both quantitative and qualitative contributions of labour and capital to output growth. Thus, TFP growth encompasses all unaccounted contributions such as changes in production processes, R&D investments, intangible assets, etc. A refined measure of growth in genuine wealth per capita (GGW) is the result of all conversions and adjustments to be used in the hypothesis tests.
16/26 amount of perceived corruption (e.g. for 2015 Denmark is perceived as the least corrupt country with a score of 91). Yet, earlier versions of the CPI were on a decigrade scale (from 0 to 10) with indexes rounded after the first decimal point (e.g. the score of Denmark for 2003 is 9.5). For comparativeness in the following analysis earlier versions of the CPI are increased by a factor of 10.
Turning to the effect of media freedom on sustainable development, the analysis employs Freedom House’s Freedom of Press measure (FH, 2016a). Freedom House is an independent think tank that publishes indexes of political freedoms across countries based on expert opinions. The think tank makes the composite index of press freedom based on findings of international human rights organisations and opinions of journalists. It accounts for economic and political pressures in addition to legal regulations that may affect media freedom. The index uses a centigrade scale similar to new versions of the CPI, yet the lowest result shows the country with most independent press (e.g. for 2015 Norway has the most independent media outlets with a score of 9). Thus, for the hypothesis test a rescaling of the measure is made by subtracting the index from 100 so the indicator of media freedom used is MF = 100 – freedom of press index (MF for Norway in 2015 is 91). This is done so that the correlation of media freedom with sustainability is with the same as the sign of the effect of PF. In addition, the rescaling of MF allows for direct inference of the effect on GGW of the interaction variable of MF and CPI.
17/26 As for the index of political freedom (IPF), it is used to represent the institutional environment within an economy that sets the framework in which the economy has to develop. A remark to the index of political freedom (IPF) is that it is derived from Freedom House’s Freedom in the World index (FH, 2016b). This measure aims to encompass all relevant civil rights and social liberties so it also accounts for media freedom. To avoid collinearity IPR is an adjusted version of the more general index by Freedom House. As 4% of the Freedom in the World index is determined by media freedom, the index is corrected to IPF by subtracting from it the value of MF reduced by a factor of 25 (IPF = Freedom in the World – 0.4*MF). Lastly, regional dummies (Region) are included in the test to control for a variety of relatively fixed country-specific factors such as common culture, beliefs and natural endowments. As mentioned in section 2 of this thesis, such factors set the pace and path of economic development within a region (Acemoglu, et. al., 2004; La Porta, 1998, 1999). Still, the regions specified in the dummy variables divide countries more or less according to cultural similarities with religion or significant ideological movements such as the creation of the Soviet bloc being major factors, and geography being secondary consideration. The reasoning behind the specified regions is thoroughly described in Appendix 4 to this thesis. It can be noted that all of the factors used in the test are also determinants of corruption discussed in the sections above. This allows for an estimation of a more pure effect of such practices on sustainability.
The question of interest in the following empirical tests is whether the theoretical link between press freedom and corruption perception argued in the previous sections significantly affects sustainable development. To answer this question the employed measure of sustainability (GGW) is regressed on an interaction variable between indexes of media freedom (MF) and corruption perception (CPI). I also factor in the direct effects of corruption perception and media freedom in the ‘base model’ of my hypothesis, while the other controls are included in the analysis in more inclusive regressions that follow. The preferred specification of the base regression of my hypothesis test is:
𝐺𝐺𝑊𝑖𝑡 = 𝛽0 + 𝛽1(𝑀𝐹𝑖𝑡𝑥𝐶𝑃𝐼𝑖𝑡) + 𝛽2𝑀𝐹𝑖𝑡+ 𝛽3𝐶𝑃𝐼𝑖𝑡+ 𝑎𝑖+ 𝜀𝑖𝑡
Examining the specified base model I employ a fixed effects estimation approach to the panel data available. The estimation method is chosen as it provides the best fit for the data according to the Akaike Information Criterion (AIC) (for further discussion and computations, see Appendix 2).
Moving to more inclusive regressions I test the hypothesis specifying two more models. By adding GDP and ALR to the ‘base model’ I control for simultaneously for income and education of the population just as I control for changes in physical and human capital in the productive base. As these factors affect directly sustainability and productivity growth, the specification is referred to as ‘included direct factors’ (IDF) model and is specified as:
𝐺𝐺𝑊𝑖𝑡 = 𝛽0+ 𝛽1(𝑀𝐹𝑖𝑡𝑥𝐶𝑃𝐼𝑖𝑡) + 𝛽2𝑀𝐹𝑖𝑡+ 𝛽3𝐶𝑃𝐼𝑖𝑡+ 𝛽4𝐺𝐷𝑃𝑖𝑡
18/26 Further inclusion of controls adds IPF and region dummies to the analysis. The aim of these variables is to control for the institutional environment and the historical cultural and natural conditions within the country. These factors are rather indirect to sustainability but are important as they determine the environment in which development takes place. The inclusion of indirect factors of sustainable development sets the ‘full model’ specified as:
𝐺𝐺𝑊𝑖𝑡 = 𝛽0+ 𝛽1(𝑀𝐹𝑖𝑡𝑥𝐶𝑃𝐼𝑖𝑡) + 𝛽2𝑀𝐹𝑖𝑡+ 𝛽3𝐶𝑃𝐼𝑖𝑡+ 𝛽4𝐺𝐷𝑃𝑖𝑡
+ 𝛽5𝐴𝐿𝑅𝑖 + 𝛽6𝐼𝑃𝐹𝑖𝑡+ 𝛽𝑘𝑅𝑒𝑔𝑖𝑜𝑛𝑘+ 𝑎𝑖 + 𝜀𝑖𝑡
The inclusion of the time-invariant factors ALR and region dummies in the analysis requires a different estimation approach to panel data. AIC shows that fixed effects estimation still provides the best fit of the data (see Appendix 2), yet this methodology eliminates the theorised important time-invariant factors of sustainability. On the other hand, random effects estimation allows for examination of the effect fixed characteristics but assumes that the differences in unobserved characteristics are unrelated to the factors included in the model which is a caveat. Thus, for the test of the ‘IDF model’ and the ‘full model’ I also adopt a mixed fixed-effects/random-effects model (Schunk, 2013). The specific approach is referred to as correlated random effects (CRE) (Wooldridge, 2013) and aims to control for country-specific unobserved effects while still being able to determine the coefficient and significance of time-invariant factors (for further discussion see Appendix 1). The CRE estimation approach allows for the observation of both country-specific effects of time-variate factors and the cross-country effects of all factors including fixed ones. Comparing the fit of the data in the different estimation methodologies the AIC shows that CRE is most appropriate approach for the specific models examined.
5. The Impact of Media Freedom on Corruption Perception and thereby
Sustainable Development: Results and Implications
This section provides a summary of the statistics on the data used in the hypothesis tests (Table 1), as well as the estimation results of the three models (Table 2). Robustness is also discussed, but checks and adjustments are presented in Appendix 3. In summary, estimation of all three models finds the expected negative sign for the time-effect of the media-corruption link on sustainable development. However, as the results are statistically insignificant there is no sufficient evidence in support of the main hypothesis of this thesis. Robustness tests show that the control variables are highly heteroskedastic across countries, and correction with cluster-robust standard errors across countries reduces the significance of the estimates even further. The discussion attributes the insignificance of the results to the rather short time-period of the analysis.
19/26 Table 1: Summary Statistics of the Data
Variable Mean St. Dev. Median Min. Value Max. Value
GGW 0.9954 4.1002 0.9185 -21.8156 22.98213 CPI 45.2981 22.3066 37 12 97 MF 56.4599 22.6113 56 7 92 GDP 16,250.11 20,211.49 7,367.49 257.11 110,001.10 ALR 89.4644 16.2334 96.7011 19.1026 100 IPF 66.5641 26.3918 71 7 100 Region ‘Protestant’ 0.1250 0.3308 0 0 1 Region ‘Catholic Europe’ 0.1731 0.3785 0 0 1 Region ‘Former Eastern Bloc’ 0.1346 0.3414 0 0 1 Region
‘North Africa and Middle East” 0.1538 0.3609 0 0 1 Region ‘Sub-Saharan Africa’ 0.1250 0.3308 0 0 1 Region ‘East Asia’ 0.0385 0.1924 0 0 1 Region ‘South Asia’ 0.0865 0.2813 0 0 1 Region ‘Latin America’ 0.1538 0.3609 0 0 1
20/26 and medians shows that all of the variables have skewed distributions – most evident in the GDP and the ALR indicators. For GDP the data shows positive skewness meaning that while the most occurring value of real GDP per capita is USD 7,367.48 per year (at 2010 levels) there are countries with much much higher values for the index offsetting the average for the dataset – the maximum value can be observed to be USD 110,001.10. Inversely, ALR is negatively skewed being evident of outlier nations with very low levels of education compared to the average country – the minimum value of ALR contained in the dataset is 19.10%.
Turning to estimation results for the ‘base model’, the fixed effects estimate of the effect of the media-corruption interaction has the expected negative sign (Table 2). Meanwhile, estimation shows that both CPI and MF indexes have positive stand-alone effect on GGW. The estimated result for the interaction term is rather small at about (-0.0004) and since both variables constituting the interaction are continuous stand-alone interpretation of the value is rather difficult (Williams, 2015). However, the negative result signals that at higher levels of media freedom the effect of improvements in perceived corruption is moderated since the upward bias arising from the censorship of the media by the corrupt is controlled for. Also, for lower levels of media freedom the stand-alone effect of corruption perception understates the negative effect negative of actual corruption on sustainability. Controlling for other factors in the model, the effect of an improvement in corruption perception alone by one index unit will cause growth in genuine wealth by 0.0151 percentage points (pp). Or in other words if the CPI of the USA in 2013 (73) was the same as the one in Canada (81) for the same year, the results estimate that GGW for the US would have been nearly a third higher (28.7%) rising form (-0.4206) to (-0.2998). However, bearing in mind the negative effect of the interaction term as a control for media-induced bias, the actual effect is more modest. As for media freedom, if the level of MF of the USA in 2013 (79) was the same as that of Canada in 2013 (81), estimation shows that GGW for the US would have been more than a third (37.2%) higher rising from (-0.4206) to (-0.2640). Still, these estimates ignore a wide array of important determinants to development, which can be observed in the estimation results of the wider models.
21/26 Table 2: Robust Estimators of Models
Base Model IDF Model Full Model
MFxCPI -0.0004 -0.0004 -0.0003 (0.0010) (0.0010) (0.0010) MF 0.0783 0.0724 0.0549 (0.0560) (0.0550) (0.0548) CPI 0.0151 0.0240 0.0148 (0.0753) (0.0725) (0.0732)
MFxCPI Period Average 0.0016 0.0010
(0.0011) (0.0010)
MF Period Average -0.1129* -0.0675
(0.0582) (0.0570)
CPI Period Average -0.1319 -0.0724
(0.0863) (0.0835) GDP -0.0001* -0.0002* (0.0001) (0.0001) GDP Period Average 0.0001* 0.0002** (0.0001) (0.0001) ALR 0.0795*** 0.0193 (0.0155) (0.0169) IPF 0.0287 (0.0376)
IPF Period Average -0.0232
(0.0443) Region 0.2815 ‘Protestant’ (0.7096) Region 0.3017 ‘Catholic Europe’ (0.5977) Region ‘Former 3.5405*** Eastern Bloc’ (0.8307)
Region ‘North Africa -0.9872
and Middle East’ (0.9652)
Region -1.1751*
22/26 Region 3.7390*** ‘East Asia’ (1.1958) Region 2.2487*** ‘South Asia’ (0.7088) Constant -2.9927 -2.3430 -0.3875 (3.7852) (2.1949) (2.3505)
Estimation Method FE CRE CRE
Observations 1248 1248 1248
F-statistic 1.49
Wald Chi2-statistic 44.28*** 86.85***
Robust standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
F Test: Null hypothesis is that the coefficients of all controls are zero Wald Test: Null hypothesis is that the coefficients of all controls are zero
countries since the media can influence subjective wellbeing (Deaton, 2012) and boost morale and productivity.
23/26 compared to other countries (to the level of Finland in 2013, for instance) on average its growth in genuine wealth would have been 0.0232 pp higher. Lastly, region dummies show diverse effects of the geography/culture controls (see Appendix 4). While RegSSA, RegNAME, and RegLatAm (contained in the constant) have a negative effect on growth of genuine wealth, the effect of RegProt and RegCath are positive and this effect is even more pronounced in RegFormSov, RegSAsia and RegEAsia.
It should be noted that the robust results for most factors are statistically insignificant. For the ‘base model’ the media-corruption interaction is with a very high p-value of 0.70 meaning that it cannot be used as irrefutable evidence for the hypothesis. Similarly, the ‘IDF model’ and the ‘full model’ also produce highly insignificant results for the within-country temporal effect of the interaction variable (with p-values of 0.66 and 0.75, respectively). The p-values for the country-specific effect are relatively higher but still too low to produce econometrically significant results (0.14 and 0.34, respectively). It is also interesting to observe that in the ‘base model’ both the measure of corruption perception and of media freedom are also statistically insignificant at any appropriate level. This finding is in conflict with results in previous research estimating cross-country effects of corruption perception on sustainability (Aidt, 2009) or the influence of freedom of the media on economic development (Alam and Ali Shah, 2013). In fact, including control variables to constitute the ‘IDF model’ or even the ‘full model’ produces much similar results. In both models estimated via the appropriate CRE methodology most robust estimators are statistically insignificant. At 99% significance, exceptions make in ALR in the ‘IDF model’ and three of the eight regions in the ‘full model’. Relaxing the significance to 90%, includes both GDP effects for both models. Notably, while the cross-county effect of media freedom is significant at 90% in the ‘IDF model’, further expansion to the ‘full model’ makes both MF and ALR insignificant.
24/26 the time effects of other controls cannot be appropriately estimated in the period of 12 years for which data is available. In fact, since the CRE estimation approach allows for disaggregation of the effects into time-variant and time-fixed for time-changing controls it can be observed that the statistical significance of time-fixed country-specific effects of such controls is higher in all instances in both the ‘IDF model’ and the ‘full model’. Thus, further research on the topic may need to focus on a larger time-frame when data is available if it is to produce significant results.
Despite the negative sign for the interaction terms being in line with previously discussed literature, the results of the test cannot be used as evidence in favour of the hypothesis. This is due to the statistical insignificance of the estimated effects. It should be noted that the in the more inclusive specifications of the ‘IDF model’ and the ‘full model’ CRE estimation shows negative correlation only for the temporal effect of the interaction. This would mean that while within the average country over time elevation of corruption awareness is positively related with sustainability, between countries at particular point in time the average effect is reversed. This result can me related to Deaton’s finding (2012) that the media can affect subjective measures of well-being, which in turn can affect rational decision-making. When the population is more aware of the true level of corruption, for the time being its trust in government policies may be reduced, decreasing the compliance with polices, some of which may even be aimed at improving sustainability. Inversely, improvements of awareness over time can be perceived as a sign of improvements of political freedoms (captured by the temporal interaction variable effects) and affect positively sustainability through belief of progress in the political system.
25/26 The results for the effects of regional dummies are somewhat surprising. Indeed, the regions that contain the most developed economies (the Western world) do in fact show positive correlation with sustainability and the regions with the poorest nations and military turmoil (Africa, the Middle East, and Latin America) are with negative effect on GGW in accordance to the findings of Arrow, et. al. (2004). However, the results are highly positive for the Far East and the countries of the former Soviet Bloc. These results can be attributed to convergence in productivity of fast industrialisers such as ‘Asian Tigers’ countries (South Korea, Singapore), or the ‘Global Factory’ of China, India and more recently Indonesia, the Philipines, etc. Similarly, Eastern European countries have managed to raise their productivity through ‘regionalisation’ as they captured production activities from Western Europe (Timmer, et. al., 2013), while other former Soviet countries invested in productivity enhancements financed by extraction industries (WEF, 2016).
6. Conclusion: Policy Suggestions and Further Research
The analysis in the previous section fails to provide significant evidence in favour of the main hypothesis in this paper. Yet, results suggest a positive correlation between the sustainable development of an economy and improvement in media freedom and corruption prevention measures over time. What is more, regression estimates show insignificant but negative relationship between sustainability and the link between corruption and influenced media over the course of development within a country which is in line with the theory. On the other hand, the adopted estimation approach allows for separate observation of but cross-country effects time-average, in which the results have the opposite signs (albeit again insignificant). Thus, the observed signs of the effects suggest that sustainable progress of an economy is related to media independence and low levels of corruption due to their effect on the subjective motives behind entrepreneurial decision-making. Since measures of wellbeing that focus in self-evaluation are greatly impacted by news (Deaton, 2012), policy makers need to conduct the legislation process as transparent as possible in order to build trust in citizens based on public awareness. They should also enable the independent media to work as a control for corruption (Brunetti and Weder, 2003) and provide two-way feedback between the government and the public (Alam and Ali Shah, 2013) to ensure sustainability of economic growth.
27/26
References
Acemoglu, & Verdier. (1998). Property Rights, Corruption and the Allocation of Talent: A General Equilibrium Approach. Economic Journal, 108, 1381-1403.
Acemoglu, & Verdier. (2000). The Choice between Market Failures and Corruption. American Economic
Review, 90(1), 194-211.
Acemoglu, Johnson, & Robinson. (2004). Institutions as the Fundamental Cause of Long-Run Growth.
Handbook of Economic Growth, 386-414.
Aidt, T. (2009). Corruption, Institutions, and Economic Development. Oxford Review of Economic
Policy, 25(2), 271-291.
Alam, A., & Ali Shah, S. Z. (2013). The Role of Press Freedom in Economic Development: A Global Perspective. Journal of Media Economics, 26 (1), 4-20.
Alvaredo, F., Atkinson, A. B., Piketty, T., & Saez, E. (2013). The top 1 percent in international and historical perspective. The Journal of Economic Perspectives 27 (3), 3-20.
Arrow, K., Dasgupta, P., Goulder, L., Daily, G., Ehrlich, P., Heal, G., . . . Walker, B. (2004). Are We Consuming Too Much? Journal of Economic Perspectives, 18 (3), 147–172.
Assembly, U. G. (1948). Universal Declaration of Human Rights. Paris: United Nations, 217 (III) A. Bardhan, P. (1997). Corruption and Development: A Review of Issues. Journal of Economic Literature,
35(3), 1320-1346.
Baten, J. (2016). A History of the Global Economy. From 1500 to the Present. Cambridge University Press.
Beck, P., & Maher, M. W. (1986). A Comparison of Bribery and Bidding in Thin Markets . Economic
Letters, 20, 1-5.
Bhattacharya, U., & Daouk, H. (2002). The World Price of Insider Trading. Journal of Finance 57 (1), 1-24.
Black, J. (1998). Maps and Politics. Chicago: University of Chicago Press.
Braguinsky, S. (1996). Corruption and Schumpterian Growth in Different Economic Environments.
Contemporary Economic Policy 14 (3), 14-25.
Brown, A. (2006). What are We Trying to Measure? Reviewing the Basics of Corruption Definition (Chapter 4). In C. Sampford, A. Shacklock, & C. Connors, Measuring Corruption (pp. 57-77). Burlington: Ashgate Publishing Limited.
Brunetti, & Weder. (2003). A Free Press is Bad News for Corruption. Journal of Public Economics, 87
(7-8), 1801-1824.
Choi, J. P., & Thum, M. P. (1998). The Economics of Repeated Extortion. Columbia University Working
Paper No. 9899-03.
Conference Board, The. (2016). Total Economy Database 2016. Retrieved from The Conference Board: https://www.conference-board.org/data/economydatabase/index.cfm?id=27762
28/26 De Soto, H. (1989). The Other Path. New York: Harper and Row.
Deaton, A. (2012). The financial crisis and the well-being of Americans. Oxford Economic Papers 64 (1), 1-26.
Delaporte, L. (2013). Mesopotamia: The Babylonian and Assyrian Civilization. New York: Routledge. Douglas, M., & Isherwood, B. C. (1979). The World of Goods: Towards an Anthropology of
Consumption. Oxford: Allen Lane.
Duncan, N. (2006). The Non-Perception Based Measurement of Corruption: A Review of Issues and Methods from a Policy Perspective (Chapter 7). In C. Sampford, A. Shacklock, & C. Connors,
Measuring Corruption (pp. 132-160). Burlington: Ashgate Publishing Limited.
Dynan, Skinner, & Zeldes. (2004). Do the Rich Save More? Journal of Political Economy, 112 (2), 379-444.
European Union. (1998). ETS No. 173. The Council of Europe’s Criminal Law Convention on
Corruption.
Freedom House. (2016a). Freedom of the Press 2015. Retrieved from Freedom House: https://freedomhouse.org/report-types/freedom-press
Freedom House. (2016b). Freedom in the World 2015. Retrieved from Freedom House: https://freedomhouse.org/report-types/freedom-world
Galtung, F. (2006). Measuring the Immeasurable: Boundaries and Functions of (Macro) Corruption Indices (Chapter 6). In C. Sampford, A. Shacklock, & C. Connors, Measuring Corruption (pp. 101-130). Burlington: Ashgate Publishing Limited.
Hamilton, K., & Clemens, M. (1999). Genuine Savings Rates in Developing Countries. World Bank
Economic Review, 13 (2), 333-356.
International Monetary Fund. (2000). Transition Economies: An IMF Perspective on Progress and
Prospects. Retrieved from IMF: http://www.imf.org/external/np/exr/ib/2000/110300.htm
Kaufmann, D. A., & Wei, S.-J. (1999). Does “Grease Money” Speed Up the Wheels of Commerce?
NBER Working Paper 7093.
Kaufmann, D., & Vicente, P. V. (2011). Legal Corruption. Economics & Politics 23 (2), 195–219. Kaufmann, D., Kraay, A., & Mastruzzi, M. (2003). Governance Matters III: Governance Indicators for
1996-2002. Washington, D.C.: World Bank Policy Research Working Paper No. 3106.
Kennedy, R. F. (1968, March 18). Address. (U. o. Kansas, Interviewer)
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R. (1998). Law and Finance. Journal of
Political Economy, 1113-1155.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R. (1999). The Quality of Government.
Journal of Law, Economics and Organization, 15: 1, 222-279.
29/26 Lancaster, T., & Montinola, G. (2001). Comparative Political Corruption: Issues of Operationalization
and Measurement. Studies in Comparative International Development 36 (3), 3-28.
Langseth. (2006). Measuring Corruption. In C. Sampford, A. Shacklock, & C. Connors, Measuring
Corruption (pp. 7-44). Burlington: Ashgate Publishing Company.
Leeson, P. T. (2008). Media Freedom, Political Knowledge, and Participation. of Economic Perspectives,
22 (2), 155-169.
Leff, N. (1964). Economic Development through Bureaucratic Corruption. American Behavioral
Scientist, 82:2, 337-341.
Levy, D. (2007). Price Adjustment under the Table: Evidence on Efficiency- enhancingCorruption.
European Journal of Political Economy, 23, 423–47.
Lui, F. (1985). An Equilibrium Queuing Model of Bribery. Journal of Political Economy. 93:4, 760-781. Mauro, P. (1995). Corruption and Growth. Quarterly Journal of Economics, 110, 681-712.
McCoy, J., & Heckel, H. (2001). The Emergence of a Global Anti-Corruption Norm. International
Politics 38 , 65-90.
Miller, W. L. (2006). Perceptions, Experience and Lies: What Measures Corruption and What do Corruption Measures Measure? (Chapter 8). In C. Sampford, A. Shacklock, & C. Connors,
Measuring Corruption (pp. 163-185). Burlington: Ashgate Publishing Limited.
Mo, P. H. (2001). Corruption and Economic Growth. Journal of Comparative Economics, 29, 66–79. Nelson, R. R., & Pack, H. (1999). The Asian Miracle and Modern Growth Theory. The Economic
Journal, Vol. 109, No. 457, 416-36.
Nozick, R. (1974). Anarchy, State, and Utopia. New York : Basic Books.
Organisation of Economic Co-operation and Development. (2016). List of OECD Member countries. Retrieved from OECD: https://www.oecd.org/about/membersandpartners/list-oecd-member-countries.htm
Persson, T., & Tabellini, G. (2004). Constitutions and Economic Policy. Journal of Economic
Perspectives, 18 (1), 75-98.
Philip, M. (2006). Corruption Definition and Measurement (Chapter 3). In C. Sampford, A. Shacklock, & C. Connors, Measuring Corruption (pp. 45-56). Burlington: Ashgate Publishing Limited.
Rawls, J. (1971). A Theory of Justice. Cambridge: Belknap Press.
Reinikka, R., & Svensson, J. (2005). Fighting Corruption to Improve Schooling: Evidence from a Newspaper Campaign in Uganda. Journal ofEuropean Economic Association.
Rents, C. a. (1999). Ades, Alberto; Di Tella, Rafael. American Economic Review, 89 (4), 982-993. Saha, S., & Gounder, R. (2013). Corruption and economic development nexus: Variations across income
levels in a non-linear framework. Economic Modelling 31, 70-79.
30/26 Schunck, R. (2013). Within and between estimates in random-effects models: Advantages and drawbacks
of correlated random effects and hybrid models. The Stata Journal, 13(1), 65-76.
Seligson. (2006). The Measurement and Impact of Corruption Victimization: Survey Evidence from Latin America. World Development 34 (2), 381–404.
Show, T. D. (2016). State Capture in South Africa. (T. Noah, Performer) Comedy Central, USA. Solow, R. M. (1956). A Contribution to the Theory of Economic Growth. he Quarterly Journal of
Economics, 70 (1), 65-94.
Svensson, J. (2005). Eight Questions about Corruption. Journal of Economic Perspectives, 19(3). Timmer, M. (2016). New measures of welfare (Part II). Lecture Course: Growth and Development
Policies. University of Groningen.
Timmer, M., Los, B., Stehrer, R., & de Vries, G. (2013). Fragmentation, incomes and jobs: an analysis of European competiveness. Economic Policy 28, 613-666.
Tirole, J. (1996). A Theory of Collective Reputations. Review of Economic Studies, 63 (1), 1-22.
Tran, H., Mahmood, R., Du, Y., & Khrapavitski, A. (2011). Linking measures of global press freedom to development and culture: Implications from a comparative analysis. International Journal of
Communications, 5, 170–191.
Transparency International. (2016). Corruption Perception Index 2015. Retrieved from Transparency International: https://www.transparency.org/cpi2015/
United Nations General Assembly. (1948). Universal declaration of human rights. Paris: UN. van den Bergh, J. C. (2009). The GDP Paradox . Journal Of Economic Psychology, 30(2), 117-135. Wikipedia. (2016a). East Asian Cultural Sphere. Retrieved from Wikipedia:
https://en.wikipedia.org/wiki/East_Asian_cultural_sphere
Wikipedia. (2016b). South Asia. Retrieved from Wikipedia: https://en.wikipedia.org/wiki/South_Asia Wikipedia. (2016c). Southeast Asia. Retrieved from Wikipedia:
https://en.wikipedia.org/wiki/Southeast_Asia
Wikipedia. (2016d). Latin America. Retrieved from Wikipedia: https://en.wikipedia.org/wiki/Latin_America
Williams, R. (2015). Interaction effects between continuous variables (Optional). Retrieved from University of Notre Dame: https://www3.nd.edu/~rwilliam/stats2/l55.pdf
Wooldridge, J. (2013). Correlated Random Effects Panel Data Models. Retrieved from IZA Institute of Labour Economics:
http://conference.iza.org/conference_files/SUMS_2013/slides_1_linear_iza.pdf World Bank, The. (2016). World Development Indicators. Retrieved from The World Bank:
http://data.worldbank.org/products/wdi
31/26
Appendix 1
Correlated random effects (CRE) estimation is an approach that combines random effects (RE) and fixed effects (FE) estimation methods for panel data as it controls for unobserved individual-specific characteristics of the analysed panel (similarly to FE), while allowing for particular time-invariant controls to be included into an estimation model (same as RE) (Wooldridge, 2013). Estimation via CRE allows for observation of cross-section characteristics between individuals (changes caused by differences between analysed individuals) along with temporal effects of the time-variant factors included in the model (changes caused by developments within factors over time). CRE estimation disaggregates individual-specific from temporal effects of the different time-variant regressors in a given panel data model by including their time-averaged values in the specification (Schunk, 2013). The estimation delivers FE estimates to time-varying controls when the effect of time-averages is above zero. The benefit of such estimates is that they are consistent throughout the sample with a probability converging to the true value of the estimated parameters (Wooldridge, 2013).
In order to determine whether the estimation method is appropriate for the data a Hausman specification test can be performed to determine the consistency of RE estimators when compared to FE estimators of the factor variables. The test analyses the consistency of the feasible generalised least squares (FGLS) estimates of the RE method to estimates of the FE approach since the latter are produced through generalised least squares (GLS) estimation and are always consistent (Wooldridge, 2013). The null hypothesis of the test states that comparing the estimators of the two methods would produce similar estimates, or in other words there are no systematic differences between FGLS estimates and GLS estimates when the data is robust. Thus, if the null hypothesis holds RE also produces consistent estimates which might be relatively more efficient than FE estimates. If the null is rejected, RE estimates are considered inconsistent so FE estimation should be preferred (Wooldridge, 2013). Thus, if the Hausman test rejects the null, we should have a preference towards FE estimates of control variables. Furthermore, even if the Hausman test fails to reject the null, a comparison between the fit of the data for the estimation methods of the specified models should be carried out (see Appendix 2).
According to the Hausman specification tests for the three models of the main hypothesis of this thesis the null is rejected at 95% confidence for the ‘IDF model’. At 90% confidence the null hypothesis can be rejected for all three specifications. The specific results of the Hausman tests can be observed in Table 1 below. These results lead to further analysis of the fit of the data to the models according to the Akaike Information Criterion to determine the best estimation approach for the data (Appendix 2).
Table A1: Results of Hausman specification test of FE vs RE for the main controls in the models
‘Base Model’ ‘IDF Model’ ‘Full Model’
χ 2 (df) 6.67 (3) 11.54 (4) 9.73 (5)
p-values 0.0831 0.0211 0.0832
32/26
Appendix 2
The Akaike information criterion (AIC) is a measure of the relative quality of several models/estimation methods for a given data set, quantifying the quality of each model or estimation approach to be used for selection of the one best fitting the data among them. The criterion provides a comparison based on the trade-off between the goodness-of-fit of a specification/estimation and its complexity (Wooldridge, 2013). Thus, the AIC does not provide any information on absolute quality – in econometric sense, there is no information in AIC whether the values of the estimators do in fact provide estimates that are significantly close to the real values of parameters (Wooldridge, 2013). So AIC is used only as a tool for comparing the quality of models/estimation approaches for the data provided and cannot be used as a basis for making factual statements on the truthfulness of the information provided by a model.
Since the Hausman test provides information only regarding the consistency of RE estimates based on a comparison with FE estimates and the results of the test are inconclusive (see Appendix 1), I use the AIC to determine which type of estimation method of a model specification best fits the data analysed in this thesis. However, the AIC is based on an estimation of maximum likelihood (ML) function on the parameters in the model while RE estimation uses iterated feasible generalised least squares (FGLS) and is not derived from likelihood. To account for that maximum likelihood estimation (MLE) can be used to produce ML estimates for the RE model to be used as substitutes for RE estimates in the AIC. This substitution is possible as FGLE estimates converge to ML estimates with big enough samples (Sanchez-Penalver, 2014). Thus, when the coefficients of RE estimates and ML estimates do not differ sufficiently (Sanchez-Penalver, 2014) the latter can be used to represent RE estimation in cases where likelihood is required such as for AIC. In making the AIC comparison for the estimation methodologies across the three models I have acknowledged that the values of the some ML estimates differ from their respective RE counterparts (see the respective tables), but these differences are in most cases less than a thousandth which is sufficiently small to consider ML and RE estimates compatible.
