Master Thesis

(1)

33

University of Groningen

Faculty of Economics & Business

24 January 2013

The effectiveness of Aid:

Improving human welfare?

Master

Thesis

Jessica Meijer

S2176742

j.meijer.28@student.rug.nl

Supervisor:

Dr. R.C. Inklaar

Co-assessor:

(2)

2

Abstract

This research explores the effect of development aid on human welfare. A new aggregate human welfare measure is used. For comparison the effect of aid on economic growth is tested too. Furthermore the impact of aid on individual human welfare components health, inequality, consumption and leisure is investigated. The results point out that aid positively affects economic growth and human welfare. However, this effect could not be compared due to puzzling results in terms of economic level and human welfare growth. The individual components show that aid increases inequality and leisure, but there is no evidence regarding consumption and health.

(3)

3

1. Introduction

Poverty is still a major problem in the world: 21 percent of the people in the developing world is living below $1.25 a day in 2010. This percentage reflects a large group of 1.22 billion people (World Bank, 2013). Even if development aid has been provided since the 1960s, it is very hard to alleviate poverty. According to Sachs (2005) there is a danger of 'poverty traps', meaning that countries are stuck in a cyclical chain of events that keeps them in poverty. Another vicious circle mentioned is that poverty causes disease and disease causes poverty. Hence poverty is a serious problem and it is difficult to breakthrough.

An important instrument in reducing poverty could be development aid. During the past 50 years, richer nations donated more than 1 trillion dollars to poorer nations with the intention to support economic development and decrease poverty (Moyo, 2009). However, the effectiveness of aid is a widely discussed subject in literature. As stated by Doucouliagos & Paldam (2009): in theory aid must be effective, but data shows otherwise. Scholars found that aid is ineffective and does not lead to economic growth (Easterly, 2006; Moyo, 2009; Rajan & Subramanian, 2008). Reasons mentioned are that aid is counterproductive and leads to corruption and economic failure. Furthermore it also makes recipient countries dependent on aid (Benmanoun & Lehnert, 2013; Hammarstrand & Sundsmyr, 2013; Wolf, 2007) and donors' self-interest plays an important role (Boone, 1996; Burnside & Dollar, 2004; Uneze, 2012). On the other hand, different scholars found that, there are indications of a positive effect of aid on growth, even if minor sometimes (Benmamoun & Lehnert, 2013; Minoiu & Reddy, 2007; Levy, 1988; and others). Despite the strand of literature giving clear positive or insignificant effects of development aid on growth, conditional effects are found as well. Aid can have a positive influence conditional on the institutional environment, degree of democracy or type of aid (Burnside & Dollar, 2000; Collier & Dollar, 2001; Kosack, 2003; Agénor & Yilmaz, 2013). Overall, there is no consensus if development aid has a positive effect on economic growth, so if aid is an effective instrument in alleviating poverty.

In the early 2000s, the United Nations set up the Millennium Development Goals: a list of goals focusing on poverty reduction, higher health standards and better education with a target year of 2015. These goals led to increased attention to the aid debate (UNDP, 2005). There is a general consensus that the first objective of aid is to increase human development levels. The focus has been on economic growth, but there are other ways in which aid can contribute to human development. Even if research does not always show a positive effect on growth, aid can still enhance development in terms of health or education for instance. The quality of life is important, which does not necessarily depend on income (Kosack, 2003). On top of that, GDP per capita seems not to be a reliable indicator of wellbeing, whereas rising life expectancy can occur independent of changes in income (Quayyum, 2012). Moreover income does not illustrate living standards per se, because it depends on the distribution of income, thus income inequality is important to be considered.

(5)

5 Sundsmyr, 2013; Herzer & Nunnenkamp, 2012; and others) but a general conclusion is lacking in this case. Regarding human development the Human Development Index of the UN is a popular measure, but a debate is going on about the meaning, interpretation and robustness of this index (Ravallion, 2010). The HDI is composed out of data on life expectancy, literacy, school enrollment and GDP per capita. These rates are re-scaled and add up in an aggregate indicator. The main reason why the HDI was constructed is that GDP per capita as a measure of human welfare seemed to be misleading, because it did not include any social elements and ignored the distribution of income. Therefore welfare should not be measured in terms of money, but still GDP is included in the HDI, which is not in line with its intention. Furthermore Ravallion states that it is unclear what is actually being measured with the HDI. It was intended to reflect human capabilities, but there are no clear theoretical roots concerning the actual meaning and how countries can be compared. Therefore, instead of the HDI measure, a human welfare measure developed by Jones & Klenow (2011a) will be used. It is an indicator set up by combining data on consumption, leisure, inequality and life expectancy. It gives a better picture of differences in welfare between countries and has an economic meaning, because it is based upon a utility model.

This Master thesis will add to the literature, because another more reliable measure of human welfare will be used in investigating the effect of aid, which has not been tested in this way before. The intention is to test the effect of aid on the distinct elements of human welfare individually, which can illustrate if there are differences between the individual indicators that cannot be observed in an aggregate measure. Hence it might illustrate in which area of human development, development aid is most effective. To have a complete overview economic growth will be added to the model as well to find out what the differences are between the effect of development aid on economic growth and human welfare.

As a result of this, the research questions are:

- What is the effect of development aid on economic growth?

- What is the effect of development aid on the elements of human welfare individually? - What is the effect of development aid on human welfare in total?

- What is the difference between the effect of aid on economic growth and on human welfare?

(6)

6

2.1. Literature review

Development aid

Development aid is a broad concept and different types of development aid exist. First of all one can distinguish between bilateral, multilateral and non-government aid. Bilateral aid is assistance given from one government directly to the government of another nation. Multilateral aid is assistance provided to international organizations such as the World Bank, UNDP and IMF which is used to reduce poverty in the developing world. Moreover, there is non-government aid provided by non-government organizations (NGO's) or companies. Examples are Oxfam Novib or Artsen zonder grenzen (OECD, 2013). Secondly, Hammarstrand & Sundsmyr (2013) make another distinction with seven types of aid which is used by the OECD as well. The first type is budget support, which is monetary aid provided to a country's governmental budget. In that case the donor is not in control of the way the money is spent exactly. When aid is provided to non-government parties one speaks of core contributions or pooled programs and funds. This includes aid to NGO's or multilateral organizations. Furthermore aid could be given in terms of project type interventions, which are funds for specified projects with a specific duration, objective and budget. For example aid for building a school or a hospital. Moreover a donor could grant aid in the form of technical assistance. This means that experts will give advice and monitor reforms in a developing country. Debt relief is another type of aid, meaning that developing countries do not have to pay back a loan to the donor country. In addition, scholarships can be seen as a type of aid too, which refers to financial help for students. Finally, when splitting up total aid in the different parts discussed before, some administrative costs remain. These are costs for delivering aid such as salaries for agency personnel.

In literature the most applied measure of aid is Official Development Aid (ODA). This is an aggregate measure of governmental aid and its main goal is to increase the level of development of developing countries. ODA includes bilateral aid between countries as well as multilateral aid to international agencies. The bilateral aid share is aid provided by official agencies to countries that are on the DAC list of ODA recipients (OECD, 2013). The DAC is a Development Assistance Committee in which several OECD members focus on aid and poverty alleviation (Hammarstrand & Sundsmyr, 2013).

Economic growth

(7)

7 growth. This means that aid gives rise to a real revaluation of the currency which undermines the competitive position of a country. As a matter of fact, least-developed countries experience more exchange rate changes and inflation compared to advanced nations. Thus it is assumed that the Dutch disease effect has a similar influence as the aid obtained, resulting in no effect of aid on growth (Doucouliagos & Paldam, 2009).

However, there are studies that report positive effects as well. Levy (1988) concludes that aid has a positive influence on investment and growth. Aid complements domestic investment which leads to growth when capital accumulation takes place. Levy also warns that the effect can be smaller due to the Dutch disease problem. Juselius, Møller & Tarp (2012) came to a similar result. They state that the difference in conclusions is caused by the use of distinct econometric methods and endogeneity assumptions as well as different types of data transformations. Another important reason is that some authors misinterpret the results: an insignificant coefficient indicates absence of evidence and not evidence of absence. Moreover, Minoiu & Reddy (2007) found that aid has a positive effect on economic growth on the long term. This means that the effect cannot be remarked on the short term and donors and recipients need to be patient. Their policy implication is that total ODA should be increased and in particular the share of developmental aid, because this type of aid is likely to spur growth. The developmental type of aid is mostly provided by multilateral organizations. Bilateral aid is found to be ineffective, because donors' political self-interest plays a large role and the actual main goal is not economic growth of the recipient. In addition, Benmamoun & Lehnert (2013) investigate the effect of FDI, remittances and ODA on growth. They find that all three factors have a positive influence, but remittances are most beneficial for growth. For low income countries FDI is important for upgrading infrastructure and ODA has more impact on social effects such as enhancing health and education.

On the other hand, authors found that aid is ineffective in influencing economic growth. Easterly (2006) explains that even though the Western world provided aid since the 1960s, it could not realize that the poorest people fulfill their most basic needs, so aid seems to be ineffective. A reason is that many aid providers do not feel actually responsible and operate from a distance and there is too much bureaucracy. Moyo (2009) agrees and says that aid resulting in long run growth is a myth. In addition, Rajan & Subramanian (2008) find that there is no evidence of an effect of aid on growth. These authors argue that the device of development aid should be rethought to increase its effectiveness.

(8)

8 important for poverty reduction. Furthermore another conditional research indicates that aid is only effective when tied to productive government spending (Agénor & Yilmaz, 2013). This type of spending refers to investments in infrastructure or health for example. These scholars found that the Dutch disease effect does not play a role.

Overall the most recent meta-analysis on this subject is carried out by Mekasha & Tarp (2013). They tried to build upon the work done by Doucouliagos & Paldam (2009) and found an overall positive effect of aid. The difference in conclusion is due to making four distinct analytical choices. Mekasha & Tarp assumed that the impact of aid is heterogeneous, because the type of aid, way of delivery and the donor-recipient relationships vary and change over time. Therefore they used random-effects meta-analysis instead of fixed-effects. Even if Doucouliagos & Paldam reported that almost three-quarters of the studies found a positive effect, they concluded that no effect is present. Mekasha & Tarp devote this conclusion to the differences in analytical methods and state that overall aid has a positive influence on growth.

Human welfare

Human welfare is a concept that represents human wellbeing. Despite only considering economic levels it also captures non-income dimensions. Human welfare is a broad notion and scholars use varying indicators including different aspects to describe human welfare. For instance Blaikie & Jeanrenaud (1996) state that human welfare is about the extent to which people can fulfill their needs. Important elements are participation and authorization of human beings to shape their own lives. Furthermore Gomanee, Morissey, Mosley & Verschoor (2005b) mention that the quality of life is the central point of human welfare instead of possessing money or goods. According to them main components are longevity, education and access to resources. In addition Llavador, Roemer & Silvestre (2011) indicate that education, physical capital, knowledge and the environment are essential dimensions of human welfare.

(9)

9 but it is unsure if he will be living in wealth or poverty; if he will be working all the time or having a lot of leisure time; or if he will be killed by a disease before the year is over. Overall it is about the share of Rawls consumption that makes him indifferent to living in a particular other country. The human welfare model is based upon a utility model in which the mortality risk, consumption inequality and utility from leisure and home production are included. Jones & Klenow apply several assumptions to their human welfare model: preferences over consumption and leisure are separable; consumption does not vary by age; consumption and mortality are uncorrelated; and consumption in a country is log-normally distributed with arithmetic mean c and standard deviation σ. The main formula to construct the human welfare measure will be presented in the methodology and data section.

The Jones & Klenow measure might be illustrative for total human welfare, but it should be considered that only several elements of human welfare are included. For example the authors do not look at participation and authorization of someone's own life as Blaikie & Jeanrenaud (1996) do. This could be partly included in the leisure component, but institutions and political systems are not enclosed. Moreover education is not considered in the Jones & Klenow measure, which is an important aspect of future wellbeing (Gomanee et al, 2005b; Llavador et al, 2011). Jones & Klenow mainly focus on the present quality of life while the level of education is an important determinant of wellbeing in the future. In addition monetary aspects are excluded from the Jones & Klenow measure, while access to resources and physical capital for instance can still be important elements of human wellbeing (Gomanee et al, 2005b; Llavador et al, 2011). Finally Jones & Klenow themselves already explain that their measure does not include aspects such as morbidity, crime or a pristine environment, which affect the quality of life as well.

With regard to development aid, several scholars tried to investigate the effect of aid on human welfare. As mentioned, the effect of aid on human welfare is important to be considered, because non-monetary indicators may be better illustrators of poverty than income measures (Gomanee, Girma & Morrisey, 2005a). The most applied measure in this case is the HDI which is composed out of life expectancy, income and educational data. There is no research done yet concerning aid and the human welfare measure of Jones & Klenow (2011a), so the main literature outcomes consulting the HDI will be discussed.

(10)

10 hypothesis and concludes that aid has a large positive effect on HDI. However, this scholar finds that the effect is smaller inside SSA compared to other poor countries. He dedicates this difference to HIV epidemics, civil wars and conflicts in Sub-Saharan Africa leading to a lower life expectancy. Quayyum tries to explain why no positive effect of aid on economic growth is found and he gives several reasons. First of all when aid increases life expectancy (as he discovered) this leads to a relative growth in population which could result in a fall in output per capita. Secondly, aid decreases infant mortality, so there will be more younger people relatively who are not participating in the workforce yet and thus not contributing to the country's income. In the third place the author thinks that the Dutch disease effect of exchange rate appreciation is an important cause for lower economic growth as well, which has been mentioned before by different scholars.

On the other hand an indirect effect of aid on human welfare might be present as studied by Kosack (2003). He concludes that development aid positively influences the HDI in democratic countries, but it is ineffective in autocracies. Kosack discovers after his investigation that in democracies people are treated better in general. In democracies citizens and especially minority groups can join the political process and advocate their needs. There is also a free press to voice the population's needs and an opposition to criticize the government when necessary. As a result of that the policy implication is to encourage democratization to enhance aid effectiveness.

In addition Boone (1996) concludes that aid raises consumption, but it does not alleviate poverty. The HDI is not influenced significantly by aid. Also he did not find a difference between democracies or autocracies. In general people in democracies experience lower infant mortality, because more basic health services are provided to the citizens. However this is not related to development aid, so contradicting to the result of Kosack (2003). Boone devotes the ineffectiveness of aid to donors' interests. In the case of bilateral aid countries often act out of political, strategic or welfare interests. For example the United States provides aid especially for military or strategic reasons; French and British aid is mainly given to former colonies; and the OPEC mostly supports adjacent countries and Arab League members.

Health

(11)

11 Conversely, Wolf (2007) states that there is no influence of ODA on health while aid provided for education or health specifically has a positive effect. The author argues that aid targeted to specific sectors may have a desired impact, but an increase in total ODA will have additional negative effects. For instance Dutch disease, more opportunities for corruption and less accountability of governments in the direction of local populations. Mishra & Newhouse (2007) report a similar result. They look at health aid specifically as well and find a positive effect on infant mortality. If health aid per capita will be doubled, this leads to a 2% decrease in infant mortality. The effect is significant, but compared to the Millennium Development goals it is quite small. The overall effect of total aid is half that of health aid and not significant. Therefore the authors advice to increase the share of aid devoted to the health sector. Gupta, Verhoeven & Tiongson (2002) agree with this policy implication and say that aid specific for health and education increases public expenditure in these sectors. This, in turn, is found to result in lower infant mortality and better access to and attainment in schools.

Income inequality

(12)

12 Conclusion

Overall, there is no consensus in literature about the actual effect of development aid on economic and social levels. Some authors publish positive results, but other authors find that aid is ineffective. These differences are likely to be due to varying econometric methods and assumptions. Furthermore data availability may be an important problem, because in the aid discussion the poorest countries are the focal point and for these countries reliable data might be missing in certain cases. Even if the focus in literature moved from the effect of aid on economic growth to more social indicators, the actual effect is still not clear. Not only the conclusions but also the source of the papers are important to consider, because there could be a difference in reliability. Most sources used in this literature review are scientific papers, but for instance the paper by Hammarstrand & Sundsmyr (2013) is a bachelor thesis and Quayyum (2012) is a PhD candidate. This does not indicate that their results are not reliable, but it is important to keep the type of source in mind. With regard to aid effectiveness, hopefully this thesis could give more consensus and can contribute to the existing literature.

2.2. Hypotheses

Scholars do not agree on the effect of development aid on economic growth, but Mekasha & Tarp (2013) try to reach consensus is their meta-analysis. As explained they found a positive effect overall and the differences in results are devoted to distinct econometric models, statistical choices and data coding. Therefore it will be assumed that development aid positively influences economic growth. Aid can be effective through the channel of investment. Higher investment will lead to an increase in output and hence economic growth (Burnside & Dollar, 2000; Uneze, 2012). Thus hypothesis 1 is:

- H1: Development aid has a positive effect on economic growth

The individual elements of human welfare will be tested which are health (measured by life expectancy), inequality, consumption and leisure. With regard to the influence of aid on health, the literature shows that different effects are found ranging from a positive effect till aid ineffectivess. Scholars also argue that it might depend on the type of aid given and that aid specifically provided to the health sector has a positive outcome (Mishra & Newhouse, 2007; Wolf, 2007). However, as Williamson (2008) explains, health aid accounts for a fixed part of total aid approximately, namely around 7%. While this is a fixed percentage, it will be expected that the effect of total aid on life expectancy is positive, in line with Mishra & Newhouse (2007) and Wolf (2007). Hence hypothesis 2 is:

- H2: Development aid has a positive effect on health in terms of life expectancy

(13)

13 provided does not have a main goal of benefiting the poor, thus income inequality remains. Consequently the following hypothesis is formulated:

- H3: Development aid has a positive effect on income inequality

Thirdly, with respect to consumption, Jones & Klenow consider an aggregate measure of public and private consumption. Whereas total income depends partly on consumption: Y = C + I + G + (X-M). Y = total income; C=private consumption; I = investment; G = government spending (investment + consumption); X = export; M = import. Thus total consumption depends on C plus a part of G. While it is assumed that aid has a positive influence on economic growth, leading to a higher total income Y, this is likely to be partly determined by an increase in consumption, as the equation shows. Therefore the effect on consumption will be assumed to be positive too. Hence hypothesis 4 is:

- H4: Development aid has a positive effect on consumption

The fourth element of human welfare is leisure. This is measured as the difference between a time endowment and time spent in employment. Jones & Klenow's (2011b) data shows that the amount of leisure time is pretty high everywhere and does not vary that much as expected. The correlation between leisure and GDP per capita is only 0.1, so these two measures are not that related. Hence there is no clear indication that a higher level of development implies that there is more leisure time. Whereas there is no literature on the effect of development aid on leisure, there is no theoretical background to assume a certain relationship. Development aid is expected to increase economic growth, leading to a higher level of economic development and the question is if this impacts leisure time as well. However, by logical consideration it seems reasonable that aid can create more jobs and decrease unemployment. This results in less leisure time, because more time is spend on the work floor. To conclude, the following hypothesis is formulated:

- H5: Development aid has a negative effect on leisure

Overall the effect of human welfare depends on H2 till H5. The magnitude of the effect on the individual elements cannot be analyzed yet, because it is an important question of the thesis which should become clearer after the empirical analysis. Without considering a magnitude and difference in weight between the elements, the overall effect on human welfare is expected to be positive. A reason for this is that there are assumed to be three positive effects and one negative effect. In fact a positive effect on income inequality is not beneficial for overall human welfare, because it means that inequality increases. The negative effect on leisure is not necessarily detrimental for human welfare, because it could mean less leisure time but a better income due to employment. With a higher income the remaining leisure time can be spend more desirably, which might be a positive outcome for human welfare. In short, hypothesis 6 is:

- H6: Development aid has a positive effect on human welfare

(14)

14 economic level and on human welfare growth will be estimated too, to have more coherent and comparable results. The effect of aid on the economic level will be assumed to be in the same direction as hypothesis 1, because when aid positively influences growth, the economic level is expected to rise as well. With regard to human welfare growth, it will be assumed that aid has a positive influence in line with hypothesis 6. The main reason is that when aid is expected to lead to higher human welfare levels, it will be likely that aid also leads to human welfare growth. As a result of that hypothesis 7 and 8 are:

(15)

15

3.1. Methodology

The model

As the hypotheses already may indicate, there are eight distinct models with different dependent variables to be tested. These variables are: economic growth, human welfare, health, inequality, consumption, leisure, economic level and human welfare growth. The independent variable in all models is development aid. Each model also needs a number of control variables which are held constant to test the relative impact of development aid. These controls will be highlighted in the data section. A panel data approach will be used, because in that case the behavior of different countries over time can be investigated. This has been chosen over cross-sectional analysis, because when more points in time are observed the sample is larger which can lead to more reliable results. With panel data, changes over time can be analyzed while differences between countries can be controlled for. As a consequence, the eight models are:

Economic growthit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (1)

Human Welfare levelit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (2)

Healthit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (3)

Income inequalityit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (4)

Consumptionit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (5)

Leisureit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (6)

Economic levelit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (7)

Human Welfare growthit = βi + βt + β1 (Aid/GDP)it + δ(Controls)it + εit (8)

In these models ε is the error term, i is an index for the recipient countries and t is the time index. βi is added to control for individual heterogeneity between countries and βt is added to control for heterogeneity over time. It is not sure if a fixed or random effects model is appropriate, so a Breusch-Pagan test will be performed to check if random effects are present. To confirm if random or fixed effects should be used a Hausman test has to be applied.

Assumptions

(16)

16

3.2 Data

Dataset

Regarding the data on human welfare and its components the dataset developed by Jones & Klenow (2011b) will be used as example. Unfortunately their dataset only includes data for 1980 and 2000, so the formulas from the Jones & Klenow (2011a) paper are used to construct the human welfare measure and create data for a larger time span. A time period from 1990 till 2010 has been chosen, because it is the most recent data available and there is a data availability problem for the control variables before 1990. Furthermore the same points in time will be used for the economic growth model as well, to make the overall model coherent. In that case the effect of aid on different dependent variables will be better comparable as well.

Sample

The total sample includes 155 countries which are listed in appendix A. The sample per model differs, because not all data for every country and year is available for all variables. The samples per model are:

- Economic growth model: 94 countries and 1582 observations - Human welfare level model: 80 countries and 849 observations - Health model: 90 countries and 653 observations

- Income inequality model: 81 countries and 881 observations - Consumption model: 114 countries and 1866 observations - Leisure model: 105 countries and 985 observations

- Economic level model: 94 countries and 1582 observations - Human welfare growth model: 80 countries and 849 observations

The dataset is unbalanced, because some observations are missing for some years.

Dependent variables

Economic growth is the first dependent variable that will be used. It will be measured as GDP

annual average growth rates. These growth rates are based on the gross domestic product at constant 2005 U.S. dollars.

Human welfare is the next dependent variable. This indicator is created by means of the Jones

& Klenow (2011a) method and is a composite measure of life expectancy, health, consumption and leisure. The authors compare welfare across countries by using a fictitious person called Rawls. It is about Rawls as a random citizen of the United States and the share of his consumption that makes him indifferent to living in a specific other country. This proposition is answered by a consumption equivalent measure of welfare. Consequently the welfare measure is constructed as an index number with the United States as reference country with a human welfare of 100. The following formula out of the Jones & Klenow paper is applied to construct the human welfare measure:

(17)

17

λ represents total human welfare; e stands for life expectancy; c is a measure of consumption; v(l) is a function of leisure; σ indicates income inequality; ū is a constant; i is a subscript for

the specific country; and the subscript us refers to the value for the United States, which is the reference country. For more details of the individual parts of the formula the Jones & Klenow (2011a) paper can be considered.

Health is the first component of human welfare and is measured by the average life

expectancy of people. With respect to Rawls, when he wants to consume in another country he is facing a certain risk that he will die and not be able to consume. This is called the mortality risk and is determined by the life expectancy at birth.

Income inequality is the second element of welfare and is usually measured by means of the

GINI index. The GINI index displays a country's income distribution and ranges from zero to hundred: zero reflects complete equality and hundred reflects complete inequality between citizens. Jones & Klenow (2011a) use an indicator called sigma squared to represent income inequality which is derived from the GINI index. Sigma squared is the standard deviation of log consumption, which shows the inequality in consumption in a specific country. An inverted version of formula 10 is used to derive sigma squared:

(10)

Where G stands for the GINI index; Ф is the cdf of the standard normal distribution; and σ is the standard deviation of log consumption. For the GINI index averaged data is used, because data was unavailable for a large number of years. Averaging GINI's over several years will not be a problem, because income inequality is not expected to change rapidly over time. Data is averaged over: 1990 till 1995; 1996 till 2000; 2001 till 2005; and 2006 till 2010. Sigma squared will be used to indicate income inequality instead of the GINI index, because it is an important part of the overall human welfare model. This makes the result of the inequality model better comparable to the human welfare model's result.

Consumption is the third component of the human welfare indicator. It is measured as the

share of consumption in GDP reflecting private and public consumption. Rawls will be more willing to live in a country where he has a higher consumption share, so this reflects a higher level of human welfare relative to income.

Leisure is the last aspect included in human welfare and determined by the gap between a

time endowment and time spent in employment. In employment time the intensive margin (annual hours worked per employee) and extensive margin (a measure of employment divided by total adult population) are considered. The time endowment is based on the number of hours per year excluding sleep, because this is not seen as work nor leisure time. The formula is:

(18)

18

Economic level is an additional dependent variable. Jones & Klenow (2011b) also include an

economic level variable in their dataset to compare the human welfare measure to. Their tactic is to convert the economic level data to index numbers with the United States as the base of 100. Whereas this approach is used for the human welfare measure as well, it will be applied to the economic data too for coherency. The economic level will be measured as GDP per capita at Purchasing Power Parity. GDP per capita is favored over real GDP levels in this case, because it makes countries better comparable on the level of economic development. When a country's population is not taken into account this could lead to numbers that could give a wrong picture. For instance the richest countries could be the ones with most citizens, while there are some rich countries that are lower populated as well. Hence it does not give an actual picture of the level of economic development.

Human welfare growth will be used as the last dependent variable. To calculate this growth

rate formula 12 will be used from the Jones & Klenow (2011a) paper. g represents the human welfare growth rate; T stands for the number of periods; λ is the human welfare indicator; and

i is a country subscript. T has the value of one in this case, because the growth rate per year is

used.

(12)

Independent variable

Aid is a broad concept and as explained before different types of development aid exist. In

literature the most applied measure of aid is Official Development Aid (ODA). This measure includes bilateral as well as multilateral aid. ODA will be used as variable, because it is an aggregate measure of governmental aid which is intended to increase the level of development of developing countries. This total measure is more applicable compared to specific aid flows for health or education for instance, because the thesis is about the effect of aid on different elements of development. Namely ODA can be used for all hypotheses which makes the model more integrated and gives the possibility to compare individual welfare elements. An ODA divided by GDP measure will be used, because this is a clear relative measure of how much aid is received by a developing country. Just the amount of ODA received is not that illustrative, because it is likely that poor countries receive more or countries with more citizens. It also shows to what extent a country is dependent on aid.

Control variables

A set of control variables is needed to include in the model to control for possible other influences on the dependent variable despite aid. To have a clear overview, the control variables are numbered and listed in table 1.

(19)

19 development can determine the chances for growth. It will be measured as the log of GDP per capita at purchasing power parity. Furthermore, FDI can impact growth. When a foreign firm invests in your country, this can create opportunities for growth for instance in terms of employment or exports. It will be measured as the amount of inward direct investment in the reporting country per year in logs. Moreover openness is important to be considered. Trade theory suggests that openness to trade is growth enhancing. It stimulates firms to become competitive and to focus on their qualities while the country can take advantage of foreign qualities. Openness will be determined as export plus import divided by GDP. In addition inflation can cause uncertainty in the business environment and macroeconomic instability which can affect economic growth negatively. The inflation rate will be measured as the consumer price index. Finally human capital accumulation can impact economic growth. When more people join secondary schooling this improves the average educational level and that can increase citizens chances for a better paid job. Hence this can mean that more income is generated and could result in economic growth. Human capital accumulation will be measured as the average years of secondary schooling of citizens above 15 years old.

The second model is the human welfare model for which control variables are needed as well. Human welfare is constructed by health, inequality, consumption and leisure measures and correlated with the economic level as explored by Jones & Klenow (2011a). Therefore an almost similar set of control variables will be used as in the economic level model below, which are variable 1, 3, 5, 6 and 11. Variable 1 will be added to the model, because the level of GDP is likely to explain a large part of the variation in human welfare despite aid. As Jones & Klenow found, these variables move closely together, so the relationship is expected to be positive. Similar to the economic level model, FDI could also impact human welfare. It could lead to job creation leading to a higher income and higher consumption for instance. On the other hand, high inflation can deteriorate human welfare, because a higher relative price level is likely to reduce consumption. Furthermore population growth can have a positive or negative effect on human welfare. For instance when there are more citizens and the employment and health facilities are the same this could possibly reduce overall health and at the same time raise leisure time. In addition human capital accumulation can influence the human welfare level. Education itself is not included in the human welfare indicator, but a higher level of education can lead to better future prospects in terms of income. This can increase the human welfare level via higher consumption. Another channel is income inequality, this can fall when secondary schooling is more accessible for everyone.

(20)

20 population of a country. Fourthly better access to sanitation is likely to increase health and lead to a higher life expectancy. It will be measured as the percentage of the population with access to improved sanitation facilities. Fifthly a high prevalence of HIV could lead to a low life expectancy, because more people die from disease instead of from a natural old age. It is measured as the prevalence of HIV as a percentage of the total population between 15 and 49 years old.

The control variables for the income inequality model are variable 1, 2, 3, 6 and 11. First of all the level of GDP per capita can determine income inequality as implied by the inverted U-hypothesis developed by Kuznets (Deininger & Squire, 1996). It shows that inequality rises when GDP increases, then flattens out and finally declines as income keeps rising. Therefore the level of GDP per capita and a squared GDP term will be included in the model. The squared GDP term is useful, because it is related to the inverted U-curve relationship. FDI could influence inequality when the investments mainly benefit skilled and educated people, which makes the gap between unskilled and skilled people larger. Population growth can worsen income inequality as well, because a higher natural population growth is generally experienced in lower income groups (Ahluwalia, 1976). The last control variable is capital accumulation. If education is improved and more people have access to secondary education, there are better chances for everyone on the labor market. Hence the gap between the rich and the poor is expected to fall. Moreover if poorer inhabitants are better educated, they could perhaps exert more influence on politics by advocating their needs, which could decrease the importance of the rich minority's needs. In turn a more equal distribution of income could arise (Ahluwalia, 1976).

Consumption is a part of total income as shown in the basic formula below. So factors influencing Y can affect the level of public and private consumption as well. Therefore an almost similar set of control variables will be used as the one in the economic level model below which are variable 1, 3, 4, 5 and 6. Firstly when the level of GDP per capita rises, consumption can rise as well, because there is a higher income that can be spent. Secondly FDI is expected to increase employment hence income, which will have a positive effect on consumption as well. Thirdly, more openness indicating more trade could mean that there is more competition from foreign firms on the national market. This forces down prices which could raise consumption. At the same time more international firms on the market will mean that the product range available for consumers increases, which can also lead to higher consumption. Fourthly as mentioned above high inflation is expected to reduce consumption, because of a relatively higher price level. Lastly an increasing population could mean that more people are contributing to the economy, so there might be a larger working population, resulting in a higher consuming population.

Y = C + I + G + (X-M) (13)

(21)

21 (nr. 12 and 14 in the table). Life expectancy and population growth can influence the size of the adult population which is one of the factors in the leisure equation. There can be differences in age distribution caused by these two elements which changes the working population. The number of jobs is expected to increase when the economy grows, so more people will be employed and leisure time falls. Furthermore FDI can also lead to job creation, having the same effect as economic growth. In addition, unemployment can influence leisure time, because having no job means having more leisure time. This will be measured by the percentage of the total labor force that is unemployed.

For the economic level model several control variables have been selected which are variable 3, 4, 5, 6 and 11. According to Doucouliagos & Paldam these variables are recommended to use in economic growth models, because they influence growth besides aid. However, the variables will be appropriate in the economic level model as well, only the interpretation of the possible effects changes slightly. First of all foreign direct investment can impact the economic level, because it is money invested in a country directly which could result in a higher GDP level. For example when a foreign firm sets up a subsidiary in another country, in this subsidiary new jobs can be created for local people which means that their income rises resulting in a higher total GDP level. Secondly, more openness indicating more trade could increase the economic level. For instance when a country can export more it will generate more income or when cheap intermediate products can be imported this decreases costs for the local firms. In the third place the inflation rate can affect GDP per capita. High inflation could lower consumption due to relatively higher prices. It could also deteriorate a country's competitive position resulting in less trade which leads to a fall in GDP per capita. Fourthly population growth can impact the economic level, because the level of GDP needs to be divided over more citizens which leads to a lower amount per capita. On the other hand a growing population means that more citizens can work and contribute to the economy, which can result in a higher GDP level. This control variable will be measured by the population growth rate. Finally human capital accumulation can impact the GDP level. If more people have access to secondary education, this is expected to lead to more employment after graduation. Furthermore when the overall population is better educated this could result in more inventiveness in terms of making money which can affect the GDP level positively.

(22)

22 could lead to lower inequality, higher employment and more consumption as remarked. Therefore it is expected to have a positive influence on human welfare growth.

Number Variable Measurement Source

1 Initial level of GDP Log GDP per capita PPP World Bank, 2013a

2 GDP level squared Log GDP per capita PPP squared

World Bank, 2013a

3 FDI Log inward FDI annual,

current US$, in millions

UNCTAD, 2013b

4 Openness (Export + import) / GDP World Bank, 2013i

5 Inflation Consumer price index, annual

average growth rate

UNCTAD, 2013a

6 Population growth Population growth rate World Bank, 2013g

7 Number of physicians Nr of physicians per 1000 people

World Bank, 2013e

8 Urban population % urban population World Bank, 2013k

9 Access to sanitation % of population with access to improved sanitation

World Bank, 2013b

10 Prevalence of HIV Prevalence of HIV as % of total population of age 15-49

World Bank, 2013h

11 Human capital accumulation

Average years of secondary schooling, age 15+, total

Encyclopedia of the Nations, 2011

12 Life expectancy Life expectancy at birth World Bank, 2013c

13 Unemployment % of total labor force being unemployed

World Bank, 2013j

14 Economic growth GDP growth rate at constant 2005 US dollars

UNCTAD, 2013c

Data sources

The data on the dependent variable economic level is extracted from the World Bank Development Indicators (2013a). The Worldbank is used for data on life expectancy as well (2013c). Income inequality data is found in UNU-WIDER World Income Inequality database version 2.0c (2008). For the intensive margin of the leisure indicator The Conference Board's Total Economy Database is applied (2013). With regard to the extensive margin data is found at the World Bank (2013f) and the Penn World tables version 8.0 (Feenstra, Inklaar & Timmer, 2013). The latter database is approached for the consumption data as well (Feenstra et al, 2013). For data regarding economic growth the UNCTAD database is considered (UNCTAD, 2013c). Furthermore data on the independent variable aid will be extracted from the World Bank dataset (2013d). Concerning the control variables, the level of GDP and its squared term will be determined by using data from the same source (World Bank, 2013a). Data regarding FDI will be found in the database of UNCTAD (2013b). The openness of a country will be checked in the database of the World Bank again (2013i). With regard to inflation rates, the UNCTAD dataset will be applied (2013a). Moreover population growth data will be retrieved from the World Development Indicators of the World Bank (2013g) as

(23)

(24)

24

4.1. Empirical results

Assumptions

First of all there should be no multi-collinearity between the independent variables. It is tested via correlation matrices including the independent and control variables. A threshold of 0.7 will be used to determine if multi-collinearity is present or not. The matrices are depicted in appendix B. As the tables points out, almost all correlations are below the threshold of 0.7. The only exception is the correlation between LogGDP and GDP squared in the inequality model, which is 0.998. However, this high correlation is logical, because the latter variable is the squared version of the former. Hence there is no problem of multi-collinearity.

Secondly there may be no autocorrelation, indicating that there should be no correlation between values of the same variable at different points in time. The corresponding equation is:

et = ρet-1 + vt. The Wooldridge testcan be used to test for autocorrelation and the hypotheses are: H0: ρ=0 and H1: ρ≠0. If the null hypothesis can be rejected, this means that there is autocorrelation. The outcomes of the test show that there is autocorrelation in all eight models: economic growth model (F=24.632, p=0.000), human welfare level model (F=54.793, p=0.000), health model (F=43.320, p=0.000), inequality model (F=10101.748, p=0.000), consumption model (F=28.956, p=0.000), leisure model (F=44.418, p=0.000), economic level model (F=386.167, p=0.000) and human welfare growth model (F=54.793, p=0.000). Consequently autocorrelation consistent standard errors will be applied to all models.

In the third place heteroskedasticity may not be present as explained before. The Likelihood Ratio test will be used to verify this. The corresponding hypotheses are: H0:σ2t=σ2 against the alternative H1:σ2t=σ2h(α2zt2+..+αszts). If the null hypothesis can be rejected, this means that there is a problem of heteroskedasticity. The LR test statistics imply that the null hypothesis must be rejected for all models: economic growth (LR chi2=1747.96, p=0.000), human welfare level (LR chi2=3084.67, p=0.000),health (LR chi2=915.74, p=0.000), inequality (LR chi2=569.00, p=0.000), consumption (LR chi2=1386.33, p=0.000), leisure (LR chi2=556.67 p=0.000), economic level (LR chi2=1206.31, p=0.000) and human welfare growth (LR chi2=407.54, p=0.000). Hence there is a heteroskedasticity problem for all eight models and robust standard errors should be used.

(25)

25

Variable Obs. Mean Stand. Dev. Min. Max.

Economic growth 3167 3.944 6.664 -66.120 106.280 Human welfare 1847 27.844 40.936 0.1 202.050 Life expectancy 3255 66.412 10.511 26.0 85.0 Inequality 1889 0.296 0.179 0.070 1.217 Consumption 3255 0.820 0.145 0.218 1.2 Leisure 3255 0.804 0.045 0.626 0.924 Economic level 3159 29.378 32.750 0.298 182.297

Human welfare growth 1826 1.635 1.439 -0.305 5.940

The histograms are depicted in appendix C. With regard to the variable economic growth, appendix figure 2a shows that it is quite normally distributed. Using a log term in figure 2b only increases the standard deviation, but the distribution remains similar. As figure 3a points out, the variables human welfare is skewed to the right. To correct for this skewness the log of these values is taken which results in a more normal distributed variable as figure 3b shows. Furthermore the variable life expectancy is quite normally distributed, but the data is a little skewed to the left as indicated by figure 4a. However, using the log of life expectancy in figure 4b does not seem to solve the skewness, so the original values will be used. Moreover the inequality values are skewed to the right as depicted in figure 5a. The log transformation of inequality in figure 5b leads to a normal distribution of the variable. With regard to consumption, the data is somewhat skewed to the left as figure 6a shows. Though applying the log of consumption in figure 6b only makes the skewness worse, so the original values will be used. Moreover figure 7a indicates that the leisure variable is normally distributed and there is no skewness. Figure 7b shows that using a logarithmic function does not change the picture at all, so the original values for leisure will be used. In addition figure 8a shows that the economic level variable is skewed to the right and applying the logarithm leads to a normal distributed variable in figure 8b. Finally, the human welfare growth variable is quite normally distributed, but somewhat skewed to the right in figure 9a. However, using the log term only makes the skewness worse in figure 9b, so the original values will be used. Concerning the first dependent variable economic growth, the original values will be used as well, because this will make the results of this model better comparable to the human welfare growth model. In short, the logarithmic term will be used for the variables human welfare level, inequality and economic level.

Summary statistics

The summary statistics of the eight different models can be found below in table 3 till 10. As the tables show, the number of observations differs per variable due to missing data.

Eco. growth 3167 3.944 6.664 -66.120 106.280 ODA 2284 7.020 10.749 -0.690 181.187 LogGDP 3159 8.500 1.321 4.415 11.348 LogFDI 2997 5.976 2.778 -11.513 12.657 Openness 3167 84.665 50.151 10.831 444.1 Inflation 2949 38.531 509.769 -16.117 24411.0 Capital accumulation 2730 2.480 1.391 0.070 7.470

(26)

26

Log Human welfare 1847 0.884 3.319 -9.072 5.309

ODA 2284 7.020 10.749 -0.690 181.187 LogGDP 3159 8.500 1.321 4.415 11.348 LogFDI 2997 5.976 2.778 -11.513 12.657 Inflation 2949 38.531 509.769 -16.117 24411.0 Population growth 3250 1.538 1.498 -7.597 17.483 Capital accumulation 2730 2.480 1.391 0.070 7.470

Life expectancy 3255 66.412 10.511 26.0 85.0 ODA 2284 7.020 10.749 -0.690 181.187 LogGDP 3159 8.500 1.321 4.415 11.348 Physicians 1587 1.940 1.372 0.007 6.167 Urban Population 3255 55.107 23.434 5.416 100.0 Sanitation 3027 68.675 31.527 2.3 100.0 HIV 1932 2.653 5.025 0.1 28.2

Log Inequality 1889 -1.388 0.582 -2.666 0.196 ODA 2284 7.020 10.749 -0.690 181.187 LogGDP 3159 8.500 1.321 4.415 11.348 LogGDP Squared 3159 73.992 22.380 19.490 128.769 LogFDI 2997 5.976 2.778 -11.513 12.657 Population growth 3250 1.538 1.498 -7.597 17.483 Capital accumulation 2730 2.480 1.391 0.07 7.47

Consumption 3255 0.820 0.145 0.218 1.2 ODA 2284 7.020 10.749 -0.690 181.187 LogGDP 3159 8.500 1.321 4.415 11.348 LogFDI 2997 5.976 2.778 -11.513 12.657 Openness 3167 84.665 50.151 10.831 444.1 Inflation 2949 38.531 509.769 -16.117 24411.0 Population growth 3250 1.538 1.498 -7.597 17.483

Leisure 3255 0.804 0.045 0.626 0.924 ODA 2284 7.020 10.749 -0.690 181.187 LogFDI 2997 5.976 2.778 -11.513 12.657 Population growth 3250 1.538 1.498 -7.597 17.483 Unemployment 1838 8.724 5.841 0.1 59.5 Life expectancy 3255 66.412 10.511 26.0 85.0 Economic growth 3167 3.944 6.664 -66.120 106.280

Table 4: Summary statistics human welfare level model

Table 5: Summary statistics health model

Table 6: Summary statistics inequality model

Table 7: Summary statistics consumption model

(27)

27 Fixed or random effects

The next question is if fixed or random effects should be used. By means of a Breusch-Pagan test it is investigated if random effects are present. The corresponding hypotheses are: H0: σ2

u=0 and H1: σ2u>0. If the null hypothesis can be rejected, this indicates that random

individual differences are present among the countries. In order to be sure which type of effects is more appropriate to use, a Hausman test is carried out. The hypotheses are: H0:

cov(ui, xkit)=0 which states that the difference in effects is not systematic against the

alternative hypothesis: H1: cov(ui, xkit)≠0. If the null hypothesis can be rejected, the fixed effects model is the most appropriate one to use. The Breusch-Pagan test shows that random effects are present all eight models: economic growth (chibar2=83.6, p=0.000), human welfare (chibar2=2032.28, p=0.000), health (chibar2=753.91, p=0.000), inequality (chibar2=2672.56, p=0.000), consumption (chibar2=5158.09, p=0.000) leisure (chibar2=2920.22, p=0.000) and economic level and human welfare growth model (chibar2=6951.26, p=0.000 and chibar2=1992.62, p=0.000). Next, the Hausman test indicates that fixed effects are most appropriate to use in the economic growth (chi2=17.31, p=0.008), human welfare level (chi2=174.16, p=0.000), health (chi2=19.99, p=0.003), inequality (chi2=21.37, p=0.002), consumption (chi2=17.19, p=0.009), economic level (chi2=107.58, p=0.000) and human welfare growth model (chi2=170.79, p=0.000). This means that for the leisure model random effects are most appropriate and will be used (chi2=8.16, p=0.227).

Estimation results

The models have been estimated with the appropriate approach (fixed or random effects) and the right corrections regarding heteroskedasticity and autocorrelation. The estimation results are outlined in table 11. The r-squared of the health model is quite high (0.593) meaning that 59.3% of the variation in health is explained by the model. Furthermore the r-squared for the

Log Economic level 3159 2.671 1.292 -1.212 5.206

ODA 2284 7.020 10.749 -0.690 181.187 LogFDI 2997 5.976 2.778 -11.513 12.657 Openness 3167 84.665 50.151 10.831 444.1 Inflation 2949 38.531 509.769 -16.117 24411.0 Population growth 3250 1.538 1.498 -7.597 17.483 Capital accumulation 2730 2.480 1.391 0.07 7.47

Human welfare growth 1826 1.635 1.439 -0.305 5.940

ODA 2284 7.020 10.749 -0.690 181.187 LogGDP 3159 8.500 1.321 4.415 11.348 LogFDI 2997 5.976 2.778 -11.513 12.657 Openness 3167 84.665 50.151 10.831 444.1 Inflation 2949 38.531 509.769 -16.117 24411.0 Capital accumulation 2730 2.480 1.391 0.070 7.470

Table 9: Summary statistics economic level model

(28)

28 leisure, human welfare level and human welfare growth model are somewhat lower (0.172, 0.150 and 0.147 respectively). In addition the r-squared of the economic level, consumption, economic growth and inequality model are quite low (0.095, 0.056, 0.036 and 0.002 respectively).

When looking at the main variable ODA, it becomes clear that ODA has a positive significant influence on economic growth and the human welfare level. The interpretation of these results is that when ODA as a share of GDP increases by 1%, economic growth will increase by 0.036 points, ceteris paribus. When ODA as a share of GDP rises by 1%, human welfare will rise by 2.1%, ceteris paribus. The individual elements of human welfare in model 3 till 6 give a puzzling result. ODA does not have a significant impact on health nor consumption, so the total effect of aid on human welfare is not expected to come from these two aspects. However, a positive effect of ODA on leisure arises, but this effect is quite small. An increase in ODA as a share of GDP by 1% will lead to a rise in leisure of 0.0005 points, ceteris paribus. Moreover the effect of ODA on income inequality is positive and significant, meaning that it is welfare deteriorating. Namely a rise of ODA as a share of GDP of 1% leads to an increase of 0.8% in inequality, ceteris paribus. This means that the GINI-coefficient increases, so more aid results in more income inequality. Moreover ODA appears to have a negative significant influence on the economic level in model 7. This can be interpreted as: when ODA as a share of GDP increases by 1%, the economic level will fall by 0.14%, ceteris paribus. In addition, model 8 shows that there is a negative and significant relationship between ODA and human welfare growth. This indicates that when ODA as a share of GDP rises by 1%, human welfare growth falls by 1% too, ceteris paribus.

As table 11 shows, a number of control variables are insignificant, so there is no evidence of an effect on the dependent variable. For the economic growth model, FDI and openness are positive and significant, which is in line with the expectations. In this model inflation has a negative significant influence on economic growth, which is also as expected. However, capital accumulation is negative and significant, which indicates that when more people join secondary education economic growth decreases. With respect to the human welfare level model LogGDP is significant and positive, which confirms the assumptions. On the other hand the FDI variable reports a negative coefficient, so the reasoning that FDI increases consumption seems not to be right. Inflation has a small positive effect on human welfare which is also unexpected. In addition there were reasons for a positive and negative coefficient of population growth and it turns out to be positive. This means that population growth could lead to more leisure time when no new jobs are created at the same time.

(29)

29

* Significant at 10%, ** Significant at 5%, *** Significant at 1% level. Robust standard errors in parentheses (1) Eco. growth (2) Log Human welfare (3) Health (4) Log Inequality (5) Consumption (6) Leisure (7) Log Economic level (8) Human welfare growth ODA 0.036** (0.019) 0.021*** (0.004) -0.006 (0.011) 0.008*** (0.002) 0.0005 (0.0003) 0.0005** (0.0002) -0.0014** (0.0006) -0.010*** (0.002) LogGDP 0.603 (0.498) 0.760*** (0.080) 2.962*** (0.331) -0.147 (0.304) -0.045*** (0.008) -0.339*** (0.036) GDP squared 0.007 (0.019) LogFDI 0.265*** (0.097) -0.050*** (0.014) 0.009 (0.007) 0.0003 (0.0016) -0.0019*** (0.0006) 0.018*** (0.003) 0.022*** (0.006) Openness 0.023*** (0.008) -0.0005*** (0.0001) 0.0001 (0.0003) 0.0011** (0.0005) Inflation -0.001*** (0.0004) 0.0001*** (0.00004) 4.06e-06 (7.31e-06) -0.00002 (0.00001) 0.00006*** (0.00002) Pop. growth 0.076*** (0.029) 0.002 (0.014) -0.002 (0.003) 0.0006 (0.0013) -0.013*** (0.005) Physicians -0.498 (0.335) Urban pop. 0.171*** (0.034) Sanitation 0.117*** (0.023) HIV -0.778*** (0.049) Capital acc. -0.744*** (0.285) 0.029 (0.040) 0.009 (0.015) 0.051*** (0.009) -0.014 (0.017) Unemployment 0.0009*** (0.0002) Life expect. 0.0018*** (0.0003) Eco. growth 0.0002 (0.0002) Constant -2.413 (3.463) -6.438*** (0.605) 27.899*** (1.881) -0.491 (1.217) 1.242*** (0.056) 0.690*** (0.020) 2.011*** (0.027) 4.738*** (0.261) Observations 1582 849 653 881 1866 985 1582 849 R2 0.036 0.150 0.593 0.002 0.056 0.172 0.095 0.147

(30)

30 consumption model, two significant control variables are presented: LogGDP and openness. The signs are negative, which was not expected. It seems that consumption is higher at a lower level of GDP and this can be logical, because consumption is measured as a share of GDP. So the share of consumption to GDP is higher when the level of GDP is lower. The effect of openness is small and near zero. With respect to the leisure model the signs of the significant control variables are as expected. For example a higher unemployment rate appears to result in more leisure time, which seems logical, because less people have a job and hence more spare time.

Model 7 shows that FDI positively affects the economic level. So the assumptions made in section 3.2 may be right. There appears to be a negative and significant effect of population growth on the economic level, so it seems to be true that the level of GDP per capita falls, due to a division over more citizens. In addition, capital accumulation positively influences the economic level which is in line with the expectations. In the matter of the human welfare growth model, the GDP level is displayed with a negative and significant sign. This was not assumed, so it does not mean that a higher level of GDP includes that public and private investment is more successful resulting in human welfare growth. FDI and openness are positively related to human welfare growth, as expected. Finally, the inflation variable reports a positive sign, which is not in line with the expectations, but the coefficient is very small.

Additionally the correlation between the economic level and human welfare level will be checked. Jones & Klenow (2011a) found in their dataset that the human welfare indicator was highly correlated with the economic level. Therefore it might be illustrative to test the correlation in this extended dataset as well, to see if the data is consistent. In addition, the correlation between the two growth variables will be checked, which might be helpful for interpretation of the results later on. The results point out that the economic and human welfare level are indeed highly correlated (0.920), so this is in line with the correlation found by Jones & Klenow. However, the correlation between economic growth and human welfare growth is quite low (0.099).

Indirect effect

Whereas human welfare is an aggregate measure of health, inequality, consumption and leisure, there could be an indirect effect of aid on human welfare via one of these four channels. This could be clarified by means of mediation testing as explained by Newsom (2012). Figure 1 shows the basic diagram that Newsom uses. In this diagram X stands for ODA and Y for human welfare. M refers to the four elements of human welfare: health, inequality, consumption and leisure. a, b, and c relate to the direct and indirect effects between the main variables.

The first step is to test for the direct effect c which is the effect of ODA on human welfare. As table 11 indicates this effect is significant. Step two is to test for effect a, so the influence of

Master Thesis

University of Groningen

Faculty of Economics & Business

24 January 2013

The effectiveness of Aid:

Improving human welfare?

Master

Thesis

Jessica Meijer

S2176742

j.meijer.28@student.rug.nl

Supervisor:

Dr. R.C. Inklaar

Co-assessor:

Abstract

Table of contents

1. Introduction

2.1. Literature review

2.2. Hypotheses

3.1. Methodology

3.2 Data

4.1. Empirical results