University of Groningen and University of G¨
ottingen
Foreign Aid and Multidimensional Poverty
A Cross-Country Analysis of 67 Developing Countries
Abstract
This thesis tries to approach the subject aid effectiveness from a new perspective by combin-ing concepts of poverty research and existcombin-ing aid effectiveness literature. It tries to integrate insights from the capability approach of multidimensional poverty with findings related to aid effectiveness contingent on policy frameworks and different types of foreign aid.
The empirical results suggest the absence of a relationship between higher levels of foreign aid and changes in multidimensional poverty The results also show that foreign aid is not more effective in more democratic countries and that the changes in the composition of aid flows do not lead to larger improvements in multidimensional poverty.
Additionally, the thesis provides some methodological insights into the use of the MPI in re-gression analysis.
Contents
1 Introduction 1
2 Theory and Literature 2
2.1 Concepts and Measurements of Poverty . . . 2
2.2 Aid-Growth and Aid-Poverty Frameworks . . . 6
2.3 Empirical Evidence on the Relationships between Aid, Growth and Poverty . . . 9
2.4 Summary and Hypotheses . . . 14
3 Methodology and Data 16
3.1 Model Specification . . . 17 3.2 Data and Descriptive Statistics . . . 21
4 Empirical Results 28
5 Robustness Analysis 35
6 Conclusion 36
Appendices 39
List of Figures
1 Density Plots for Average Annual Changes in Poverty . . . 26
2 Density Plots for Flows of Aid . . . 27
3 Scatter Plots Changes in Poverty and Total Aid Flows . . . 28
4 Average Marginal Effects of Multilateral and Bilateral Aid on Health Poverty . . 36
List of Tables
1 Summary Statistics . . . 252 Effects of Total ODA on Changes in MPI and Its Sub-dimensions . . . 30
3 Effects of Types of Aid on Changes in MPI and Its Sub-dimensions . . . 32
1
Introduction
In September 2000 world leaders gathered in New York to sign the United Nations Millennium Declaration. They promised to collectively increase their efforts towards supporting developing nations across the world and to eradicate extreme poverty by the year 2015. Aid flows from bilateral and multilateral sources nearly doubled from about 96 billion US dollars in the year 2000 to 187 billion US dollars in 2013. The population living from less than one dollar a day halved during the same time period (World Bank, 2015).
Yet, although the evidence of the promotive impact of foreign aid seems to be evident, aca-demic literature has been debating its effectiveness for the last 60 years. As early as 1958 Milton Friedman doubted its benefits, stating that it would perpetuate the government’s role in the economic sphere and, thus, prevent private economic activity that was needed for growth (Friedman, 1958). Empirical literature on the subject has also not been able to provide clear evidence or to even reach consensus on aid’s role in economic growth and poverty reduction. This thesis tries to approach the subject from a new perspective by combining concepts of poverty research and existing aid effectiveness literature. It tries to integrate insights from Amartya Sen’s capability approach, namely that poverty is a concept too wide to be cap-tured by income measures alone, with findings related to aid effectiveness contingent on policy frameworks and different types of foreign aid. At the core of the analysis will be the Oxford Poverty and Human Development Iniative’s (OPHI) Multidimensional Poverty Index (MPI), which combines the methodology of established poverty measures with the conceptual frame-work of the capability approach. Cross-country regression analysis for a sample consisting of 67 developing countries will be used to assess the impact of changes in various types of foreign aid on multidimensional poverty and poverty in the dimensions of education, health, and living standards.
The empirical findings provide evidence that foreign aid is not the main determinant of poverty reduction in developing countries. Changes in multidimensional poverty as well as its sub-dimensions are not driven by varying levels of foreign aid or its composition into grants and loans as well as multilateral and bilateral aid. Additionally, differences in levels of democrati-zation or policy environments in recipient countries do not significantly influence the change in poverty.
2
Theory and Literature
The following section will describe the definitions and concepts that have been used when analyzing poverty, both in terms of incomes as well as capabilities. Furthermore, it will explain the theoretical frameworks that can be applied when assessing the relationship between foreign aid and economic growth as well as foreign aid and poverty. The last part will provide an overview of the empirical literature on the subject.
2.1
Concepts and Measurements of Poverty
This subsection will give an overview of the different concepts and definitions that have been
used when analyzing poverty in developing as well as developed countries. It will further
introduce a number of indicators that were used by researchers along with their respective advantages and disadvantages.
Poverty as Income Poverty
Historically, poverty has been connected to people’s income, expenditure or consumption. A person was considered poor if she lacked a certain amount of resources needed for physical subsistence. Poverty was defined as having an income that was below a pre-specified threshold needed to achieve a minimum level of socially accepted welfare, the so-called poverty line (Atkinson, 2008).
The first studies to examine poverty from this point of view and in a systematic way were those of Booth between the late 1880s and 1902-03, Rowntree in 1901, and Bowley in 1915. Booth examined the extent of poverty in the English capital recording occupation and wealth, and taking individual streets as the unit of analysis. Rowntree’s study of 1901 compared Booth’s insights with poverty patterns in provincial towns. His methodology was different from Booth’s in two important aspects. First, his unit of analysis were households and their family incomes instead of streets and second, he constructed a poverty line based on nutritional requirements and other basic goods instead of setting a more or less arbitrary threshold of income. Bowley’s article on ‘Working class households in Reading’ extended the methodology of Rowntree by introducing sampling methods. These seminal studies were later replicated for other cities in Britain and in other countries in general (Atkinson, 2008).
By the 1950s governments in developed countries were proclaiming the end of poverty with the introduction of the welfare state and ongoing economic growth. However, the reality was different as there were still a large number of people living in poverty, a result confirmed later by Townsend and Abel-Smith for Great Britain (Atkinson, 2008). Their findings led Townsend to remark that absolute poverty was not a sufficient yardstick to assess poverty in a society but that
encouraged and approved, in the societies in which they belong” (Townsend, 1979: p. 39).
Townsend realized that simply creating a more or less arbitrary absolute threshold of poverty and counting the people not passing this threshold was not a sensible approach. It had to be accompanied by some measure of relative poverty. This meant that people were categorized as poor not if they lacked a certain level of absolute income or consumption but in case their income was lower than a certain fraction of the median income. This level was subject to constant change depending on the overall development of the economy. Furthermore, it was different from society to society not only because prices for the bundles of goods were different between two countries but also because these bundles could contain different goods in the first place (Ravallion, 2008).
Absolute and relative poverty measures do not allow for differentiation among the poor, how-ever. People just above the poverty line are treated fundamentally different than those just below. At the same time people falling barely below the poverty line are regarded exactly the same as the poorest members in society. In other words, simply counting the number of poor people in a society does not allow for an assessment of depth and severity of poverty in a country.
This insight led to the creation of a number of additional measures that came to be known as the Foster-Greer-Thorbecke (FGT) measures introduced in 1984 (Foster et al. 1984). They comprise a general class of indicators related to poverty of which three particular specifications have become widely used. The first is the so-called headcount ratio representing the fraction of people in a country with incomes below the poverty line. It is, thus, the simplest absolute measure and represents the breadth of poverty. The second indicator is the poverty gap prox-ying for the depth or intensity of poverty. It shows the average distance from the poverty line of the people defined as poor, i.e. the average income shortfall to reach the poverty line of the poor population. The last indicator, the so-called squared poverty gap, is a measure of severity of poverty. It is the average squared distance from the poverty line, thus, giving more weight to people farther away from the poverty line (Haughton and Khandker, 2009). Together these indicators allow for a much more thorough analysis of poverty within or across societies as well as over time. Specifically, they allow conclusions about developments among the poor, e.g. have the poor become poorer or has their situation been improved on average.
Poverty as Lack of Capabilities
However, restricting focus on pecuniary elements of poverty necessarily left out a wide array of other issues going hand in hand with income poverty. Other aspects of being poor such as being in bad health or having no access to basic facilities had to be subsumed under the income parameter or were assumed to simply be correlated with income.
One of the first to voice this problem was Amartya Sen who had studied inequality beginning in the 1970s but later broadened his focus to include poverty and other social problems as well. Sen argued that income and resources are an important aspect when assessing poverty but that they are only means rather than ends in itself (Sen, 1995). Their possession is merely instru-mental for escaping poverty and does not constitute poverty or rather non-poverty. Poverty in his framework is defined as the lack of capabilities, i.e. the lack of freedom to choose between vectors of functionings or to achieve certain functionings at all (Sen, 1995). Functionings are all “beings and doings” (Sen, 1995: p. 39) of an individual ranging from basic states such as being healthy or being well-nourished to more complex concepts such as being happy, being loved or to “live without shame” (Sen, 1983: p. 163).
Sen is not precise about which capabilities are relevant and necessary for a person to be con-sidered non-poor. He states that there are basic capabilities that are equally important for all individuals in all societies such as “the ability to be well-nourished and well-sheltered, the capability of escaping avoidable morbidity and premature mortality” (Sen, 1995: pp. 45-46). However, he adds that different societies may have different definitions of poverty going beyond these, and that a consensus about which these are and to what extend they need to be satisfied has to be reached within each society individually (Sen, 1995).1
The latent character of capabilities makes their measurement a hard undertaking. Since ca-pabilities cannot be observed directly, their measurement has to rely on indirect procedures connected to associated functionings and inputs. A first attempt in this direction was made by the United Nations Development Programme (UNDP) with the Human Development Index (HDI). Human development is defined by the UNDP as “enlarging people’s choices” (United Nations Development Program, 1997: p. 15), in Sen’s words to increase people’s capabilities. The basic capabilities identified by the UNDP are the ability to live a long and healthy life, the ability to enjoy an education, and the ability to have a certain standard of living. Con-versely, poverty is defined as the lack of all or at least some of these abilities (United Nations Development Program, 1997). The index consists of functionings associated with each of the three basic abilities and is meant to give a comprehensive overview of human welfare that goes beyond income alone. It has been an element of UNDP’s annual Human Development Report (HDR) since 1997 and, over time, has become one of the most watched indicators assessing human development.
The capability approach expands the income approach by an element of freedom of choice and includes important aspects of poverty independent from income. It is highly individualistic
1Other authors have been more precise and formulated a wide range of capabilities connected to poverty. An
taking into account people’s characteristics, needs, and preferences. The result is a compre-hensive framework for poverty analysis that is flexible with respect to the dynamics of poverty and differences between societies.
However, the individualistic character of the capability approach makes it difficult to establish a specific poverty line for an entire country. Because persons differ with respect to their needs and values, it is almost impossible to define poverty except by the lack of the most basic ca-pabilities. The comprehensive capability approach then collapses back to a simpler framework taking away much of its analytical strength. The latent character of the capabilities themselves makes it very difficult to set thresholds based on them. Measures of poverty must always rely on the presumed link between functionings and their underlying capabilities.2
Indicators, such as the HDI, do not constitute measures of poverty comparable to the head-count ratio or the (squared) poverty gap index. They provide an overview of the general state of development of a country without defining poverty lines. They are also confined to a limited number of indicators presumed to be associated with capabilities.
The Oxford Poverty and Human Development Initiative’s (OPHI) Multidimensional Poverty Index (MPI) seeks to improve on these problems. The MPI is constructed in the tradition of the FGT-measures using the method of identification and aggregation (Alkire and Foster, 2007). It provides a clear definition of poverty as deprivation in certain indicators and dimensions and a poverty line linked to these deprivations. It is decomposable by regions or sub-groups of populations as well as into its different sub-dimensions. Its dimensions are similar to those of the HDI incorporating education, health, and standards of living but unlike the former it is based entirely on household survey data (Alkire and Foster, 2007).
However, it is not a remedy for all of the problems facing older measures, as it also is reliant on the connection between observed functionings and underlying capabilities. Data availability limits its dimensions and it is unable to differentiate between societies, postulating identical poverty lines for each country. Lastly, these poverty lines are themselves value judgments and to a certain extent arbitrary. However, since it is the only multidimensional poverty measure available for more than one hundred countries at this time, it will be used as the main measure of poverty in the remainder of this thesis.3
The section above shows that poverty is a complicated and, at times, diffuse concept that cannot be captured by income alone. A substantial effort has been put into its elimination by national governments in developed countries and multinational development agencies. Foreign aid has been identified as one of the principal tools for this complex task and donors have shifted their focus away from trying to increase growth and income towards targeting poverty directly. The next section will describe the theoretical foundations underlying the relationship between foreign aid, growth and poverty.
2Another approach used in the literature is that of latent variable estimation. However, it is also not a measure
of poverty, providing no indicator of individual capabilities. It will not be elaborated further here but for a
more detailed description see, for example, Krishnakumar (2007). Wagl´e (2005) uses a similar methodology in
his empirical study on poverty in Kathmandu, Nepal.
2.2
Aid-Growth and Aid-Poverty Frameworks
Mosley et al. (2004) provide a helpful formalization of the total impact of aid (A) on poverty (P ) that can be broken into a direct effect, an effect working through income (y), and an effect operating via policies (Ω). They write
dP dA = ∂P ∂A + ∂P ∂y ∂y ∂A + ∂y ∂Ω ∂Ω ∂A + ∂P ∂Ω ∂Ω ∂A. (1)
However, even though the equation above gives a clear picture of the potential interrelations between the variables in question, the literature does not follow the proposed relationships as neatly. Rather, authors usually examine aid’s impact on growth or poverty directly, taking into account different policy environments. Therefore, this sub-section will first explain the theoretical relationship between aid and growth, and how different authors have included the policy dimension. In a second step, it will draw a line from growth in income to reduction of poverty, again considering different policy scenarios.
The Relationship between Aid, Growth, and Poverty
The aid effectiveness literature can be divided into three different waves. The first wave ana-lyzed the relationship between aid and savings and investments while the second wave assessed aid’s impact on growth directly. The focus of the third wave remained on the relationship between aid and growth taking into account possible non-linearities (Doucouliagos and Pal-dam, 2009; Hansen and Tarp, 2001). The first two waves will be discussed jointly as they are conceptually similar. The third wave follows below.
The so-called two-gap model by Chenery and Strout (1966) was the first attempt to ground the effect of foreign aid on saving, investment and growth in economic theory and was either explicitly or implicitly used to analyze the impact of foreign aid in the first two waves of the aid effectiveness literature. Even though the model was initially conceived for a Harrod-Domar framework, its logic also applies in the growth model of Solow, as both regard the per capita
capital stock as the main determinant for per capita output.4 The underlying assumption is
that higher levels of savings, and in turn investments, lead to higher incomes as the per capita capital stock is increased. The major problem for developing countries in these frameworks are low saving rates, which lead to slow or even negative growth, a problem that can be mitigated by foreign aid.
In the model of Chenery and Strout (1966) developing countries might not only be hindered in their development through low or negative saving rates but also by a lack of foreign exchange income. This will be the case if a country is dependent on foreign capital goods or foreign intermediate inputs. It will become foreign exchange constrained if the value of its exports is not sufficiently large to cover the necessary imports for investment and production. In this
4While output per capita is a function of capital per capita in both models, they differ in their underlying
case, the country is unable to increase the domestic capital stock or produce at the optimal level because it lacks the foreign exchange reserves (Hansen and Tarp, 2000).
In both cases foreign aid can be used to overcome the country’s constraints, albeit with different implications for savings and investment. Increases in foreign aid will increase domestic saving directly if a country is saving constrained, as it will be substituting for the lack of domestic savings. This will lead to investments in the per capita capital stock and, in turn, to higher equilibrium per capita income. Growth in income will be different depending on the framework applied: in the Harrod-Domar model the economy will experience constant increases in per capita income dependent on the capital-output ratio. In the Solow model the growth in per capita income will decrease as the new steady-state comes closer. However, both models predict decreases in poverty until the new equilibrium is reached, as higher per capita income means less people falling below the poverty line.5
If, on the other hand, the country is foreign exchange constrained, aid will have no immedi-ate impact on domestic savings. Higher flows of foreign aid will lead to higher investment if they are used to purchase foreign capital goods. It will lead to higher output directly if they are used to buy intermediate foreign goods necessary for final production (Hansen and Tarp, 2000). Independent from the exact channel, increases in aid will lead to increases in per capita output either directly or indirectly in a foreign exchanged constrained economy. As output and incomes rise, the absolute level poverty will fall in return.
Chenery and Strout (1966) also propose a third scenario, namely a lack of absorptive capacity. Developing countries might be unable to channel foreign aid into productive projects increasing savings, investment and income because they do not have the necessary government structures or entrepreneurial abilities do so (Hansen and Tarp, 2000). Increases in aid would then have no effect on saving, investment and income.
The crucial point is that poverty reductions are always only temporary in a two-gap model. This effect is independent from the underlying framework and holds true in both the Harrod-Domar as well as the Solow growth model. Increasing the level of output and income always requires an increase in the level of external funding if the economy is either savings or foreign exchange constrained. If the level of foreign aid stays constant or even declines, the level of absolute poverty will also stay constant or even increase.
The last wave of studies worked under the assumptions that aid’s effect on poverty might be non-linear. Articles published in this framework either tied aid’s effects on favorable policy environments (e.g. Burnside and Dollar, 2000) or assumed diminishing marginal returns to increases in aid (e.g. Lensink and White, 2001).
The economic reasoning behind differences of aid effectiveness can be shown in a neo-classical growth model. Aid will only be effective if it is used for productive investments in the recipient
5One remark must be made for the analysis in all neo-classical models. All of them assume representative
country. Differences in policy environments are directly related to the amount of foreign aid money that is invested – countries with sound policies will invest a larger fraction of foreign aid. Countries lacking these policies will consume the additional funds instead of investing them (Burnside and Dollar, 2000).
The return of these investments will also depend on policy levels. It will be higher in countries with less distortionary policy environments than in those with bad policies. High inflation rates or high levels of corruption will therefore not only be critical for the investment decision itself but also for the return of that investment (Burnside and Dollar, 2000).
Lensink and White (2001) argue that not only is aid subject to decreasing marginal returns but that its effect can actually turn negative. The first claim follows directly from the neo-classical growth model: because the production function is convex, increases in the per capita capital stock will lead to decreasing marginal effects on per capita income. Since it is assumed that investments in the per capita capital stock are financed via foreign aid in developing countries, increasing foreign aid will lead to diminishing marginal returns in a Solow framework (Lensink and White, 2001). The second effect is dependent on aid’s impact on the rate of technical progress and productivity. The authors suggest that foreign aid might decrease the productiv-ity of investments. This can be the case if the recipient economy is lacking absorptive capacproductiv-ity to invest additional funds productively, or if increased aid flows are used to advance donor in-terests instead of economic growth in the recipient economy. If the effect on technical progress and labor productivity becomes large enough, the marginal effect of aid on per capita income might not only be decreasing, it might even turn negative. Lensink and White (2001) call this the “Aid Laffer Curve Model” (Lensink and White, 2001: p. 49).
Therefore, the aid-growth nexus is also dependent on the policy environment. Aid will only be poverty reducing if it is invested and if its returns stay positive. If either one of these prereq-uisites is violated, increases in foreign aid might actually increase income poverty. However, as long as at least a fraction of aid is invested with positive marginal returns on income, absolute poverty will decrease – again until the new steady-state is reached.
However, the effects described above only apply to poverty defined as lack of income or con-sumption. When analyzing poverty in the framework of the capability approach, increases in per capita income might not always be an effective method for reducing the level of absolute poverty. The ability to receive a formal education or the ability to live a long and healthy life is largely dependent on a person’s environment and not on her income level alone. Education and health care are public goods that cannot be bought in private markets and need to be
pro-vided by the central government.6 Achieving the functionings of education and health might
be independent from a person’s income, at least to some extent.
Productive government expenditures increase the per capita capital stock, leading to higher levels of steady-state per capita income. Unproductive government expenditure will not be
6There are certainly both private schools and private health care in developed countries and to some extend
invested but consumed. Increases in foreign aid can have a poverty reducing effect if they allow for higher levels of saving, investment and growth. This will lead to higher overall income and therefore to higher tax revenues to finance unproductive government expenditure. If at least a fraction of these increased tax revenues are used for government consumption that provides public services or improves social infrastructure, capability poverty will be lowered.
2.3
Empirical Evidence on the Relationships between Aid, Growth
and Poverty
Economic theory provides a number of channels through which aid might be able to increase growth and decrease poverty. The following sub-section will give an overview about the em-pirical evidence based on this theory. It will summarize the evidence concerning growth and poverty as well as aid and poverty for different poverty concepts. Additionally, the non-linear relationship between aid and growth as well as aid and poverty, as examined in the literature, will be taken into account.
Aid and Growth
Hansen and Tarp (2000) survey a number of studies analyzing the relationship between aid and savings, aid and investments as well as aid and economic growth. The article summarizes the findings of 29 different papers published between 1969 and 1998, and 131 regression specifica-tions within these.
According to the study of Hansen and Tarp (2000) the majority of papers examining the effect of aid on savings report significant positive results. In aggregate, increases in aid are correlated with higher saving in their sample of academic articles.7 A similar picture arises when focus is shifted towards the relationship between aid and investments. Of the regressions covered in the survey article, a majority finds a positive relationship between higher aid flows and higher investment. They conclude that in the Harrod-Domar model this is convincing evidence that increasing aid will lead to higher rates of growth.
This result is not supported by Boone (1996) who finds no significant effect of aid on invest-ment. Two qualifications must be made, however. First, Clemens et al. (2012) find that this result is contingent on the sample. If all countries are included in the regression, the effect of aid on investment is positive and significant. Second, even if the estimated coefficient were insignificant, aid could still increase investment via increases in the growth rate in a foreign exchange constrained economy. In the case of Boone (1996) the coefficient for growth in per capita GNP enters significantly into the regression, leaving this possibility open.
Hansen and Tarp’s (2000) article also presents evidence for a growth-enhancing effect of aid.
7Hansen and Tarp (2000) re-formulate the null-hypothesis to take account for the fact that positive effects on
saving do not require estimated coefficients to be larger than zero. They show this in equations (4) and (5) of
their paper (Hansen and Tarp, 2000: p. 379). If investment is described by it≡ st+ at+ fpt+ fot and the
partial derivatives of private foreign resources, fpt, on aid, at, and other foreign resources, fot, on aid are zero,
then ∂it
∂at =
∂st
∂at + 1. The overall impact of a marginal increase of aid on investments is only smaller than zero
They argue that only if estimated coefficients for both aid and savings are insignificant, does that mean that aid has no effect on growth. The channel through which aid ultimately affects growth is dependent on the constraint the economy is facing. In case of a foreign exchange constraint this would lead to a significant estimator of aid. If the economy were saving con-strained, the estimator of saving would be significant, capturing an indirect effect of increases in aid.
The last part of the paper by Hansen and Tarp (2000) explores the possibility that the rela-tionship between aid and growth might be non-linear, an alternative that will be discussed in more detail below. The results of two of the most influential studies on the non-linear relation-ship between aid and growth will be presented, namely those of Burnside and Dollar (2000) and Rajan and Subramanian (2008). Unlike older studies, these authors use panel instead of cross-country regression techniques to analyze growth-enhancing effects of aid. They further employ IV-regression to account for possible endogeneity issues of the aid variable – donors might either give more aid to countries with bad growth experiences or they might give aid to countries that have experienced high rates of growth. Another common feature is the explicit modeling of possible non-linearities of aid. Aid’s effectiveness might either be contingent on favorable policy environments or it might be subject to diminishing marginal returns.
Burnside and Dollar (2000) find no significant effect of aid alone on countries’ growth perfor-mance. However, they find that in countries with good policies aid has a growth-increasing effect, albeit with diminishing marginal returns. Even so, their results are dependent on the exclusion of influential outliers, a result that was later stressed by Easterly et al. (2004) who extended the original dataset of Burnside and Dollar (2000). This led them to conclude that there was, in fact, no effect of aid on growth, even in favorable policy environments, and that the results of Burnside and Dollar (2000) were due to their preference for specific samples and regression techniques.
Rajan and Subramanian (2008) article on the subject arrives at similar conclusions. The au-thors find no significant effect of aid on growth in any of their specifications. These results are robust for different time horizons or time periods, different regression settings (cross-section versus panel data), policy or geographical environments, and different types of aid (donors type and purpose). They further conclude that, even if recipients invested all foreign aid instead of consuming part of it, growth effects would still be small, and increasing growth by only about 0.16% or less in the long-run.
Clemens et al. (2012) argue that these studies exhibit a number of problems. They start by claiming that aid’s impact on growth will mostly likely be delayed so that including contempo-raneous levels of aid into the growth regression is unwarranted. Furthermore, they argue that most of the previous aid-growth literature suffers from weak instruments for aid.8 Lastly, they contend that not all aid can be expected to increase growth so that the overall level of aid is
8Specifically, they argue that most of the instruments used in previous articles are a combination of donor ties to
not the right variable to include in the regression analysis.9 They go on to suggest possible solutions for each of the perceived problems: they take first differences of aid instead of using more complicated first-stage regressions, they lag these first differences to account for the de-layed effect of aid, and they distinguish between aid that is expected to be growth-enhancing and that, which is not. They call the former ‘early impact’ aid.
Clemens et al. (2012) re-estimate the studies cited above using the modifications suggested by them and arrive at different results. They find that aid has growth-enhancing effects subject to diminishing marginal returns in the dataset of Burnside and Dollar (2000) and also for similar data used in a paper by Hansen and Tarp (2001). The authors do not include, or at least do not report, interaction effects with different policy regimes, however.
Their results for the dataset of Rajan and Subramanian (2008) are less striking, as they do not find any significant effect of aid on growth. The result holds as well after they include a term of squared aid into the regression. Only after including additional observations preceding and succeeding the initial dataset do they find evidence for a positive effect of aid on growth. They conclude by stating that these results are not a confirmation of the original findings by Rajan and Subramanian (2008) but that they rather show that the original study suffers from limited sample years, imposed linearity between aid and growth, and weak instruments.
The empirical evidence concerning aid’s ability to increase savings, investments, and growth is mixed. Early studies have found strong reasons to believe that aid is indeed able to positively influence all three of them (Hansen and Tarp, 2000). Yet, more recent studies cast doubt on these findings in at least two ways: first, not all of them find a statistically significant relation-ship between aid and the three dependent variables (Boone, 1996; Rajan and Subramanian, 2008). Second, if studies do find significant evidence, it is usually dependent on policy environ-ments (Burnside and Dollar, 2000) or it is effective but only with diminishing marginal returns (Dalgaard and Hansen, 2001). The results are also not very robust and highly dependent on the sample and methodology used. Doucouliagos and Paldam’s (2009) survey of the aid ef-fectiveness literature (AEL) comes to the conclusion that the failure of converging empirical results is driven by a ‘reluctance bias’ of researchers: results are only published if they confirm the hypothesis that aid is beneficial for growth. Their conclusion is rather bleak, stating that “aid has failed in its primary mission” and that “meta-analyses of the AEL have failed to find evidence of a significantly positive effect of aid. Consequently, if there is an effect, it must be small” (Doucouliagos and Paldam, 2009: p. 457).
Aid, Growth, and Poverty
Dollar and Kraay (2002) provide evidence that economic growth does indeed lead to lower levels of poverty. In their sample consisting of over 130 countries and for the period between 1950-1999 they find convincing support that growth in mean per capita income leads to one-for-one
9In an earlier version of their paper the authors concentrated on the lagged impact of aid only (Clemens et al.
growth in the income of the bottom wealth quintile.10
Collier and Dollar (2002) use a specification similar to Burnside and Dollar (2000) to analyze the relationship between aid and growth. Their results support the hypotheses that aid’s effec-tiveness is contingent on the policy environment and subject to decreasing marginal returns. They conclude giving to the poorest countries with the best policies would almost double its effectiveness leading them to advocate for strict conditionality. However, their approach was criticized for a number of reasons. Collier and Dollar (2002) assume exogeneity of both aid and policies unlike many other studies on the subject, for example Boone (1996), Dalgaard and Hansen (2001) or Hansen and Tarp (2001). They argue that donors have no control over the way that aid is used in recipient countries and that the growth elasticity of poverty is set equal for all countries at a level of 2. An increase in income of 1 percent would lead to an average decrease in poverty of 2 percent.
Mosley et al. (2004) agree that the effectiveness of aid is dependent on the recipient country’s policies but criticize the approach of Collier and Dollar (2002) for the reasons above. They argue for a different approach, concentrating on different policy variables for the construction of their policy index following Gomanee et al. (2005). Instead of proxying policy environ-ments with different macroeconomic conditions, they use government expenditure targeted to the social sectors, so-called pro-poor expenditure.11 This choice reflects the assumption that donors can use their leverage to change recipients’ behavior and that aid can change the policy environment in the recipient country. Furthermore, Mosley et al. (2004) construct a system of equations using GMM three-stage least squares (3SLS) instead of simple OLS to control for po-tential endogeneity of independent variables. They find that pro-poor expenditure significantly reduces income poverty, however only by an amount about half as much as that assumed by Collier and Dollar (2002). The driving effect behind increases in these expenditures is increases in income. They also find that higher amounts of income inequality and corruption are associ-ated with higher levels of poverty and that aid is able to alter recipients’ policies in low-income countries only (Mosley et al. 2004).
Alvi and Senbeta (2012) arrive at slightly different conclusions and find that aid has a direct poverty reducing effect even when controlling for per capita GDP and inequality. Their results hold consistently for the three standard FGT-measures headcount index and (squared) poverty gap. Their conclusion is that foreign aid does have a mitigating effect on poverty even when controlling for income and inequality and that questionable aid-growth-poverty nexus is there-fore not of utmost importance when trying to lift people out of poverty.
However, the evidence cited above focuses exclusively on poverty defined as income. The ques-tion remains if aid and growth are effective tools to reduce poverty independent from income and if this effect is dependent on different policies, too. In other words, what is the effect of aid on the different dimensions of multidimensional poverty?
10However, in contrast to, for example Ravallion (1995), their definition of poverty is relative instead of absolute
– people are considered poor if they are among the bottom wealth quintile in their country. Therefore, their results cannot provide any information concerning the development of people in absolute poverty.
The study of Mosley et al. (2004) provides a starting point for this question, as it also evalu-ates the effect of health spending on infant mortality and the effect of aid on health spending. They find that increases in health spending lead to significantly lower levels of infant mortality. However, unlike in the case of income poverty, increases in aid do not lead to increases in health expenditures. Increases in income are the main determinant also.
Gomanee et al. (2005) arrive at very similar results. They find only limited evidence that increases in pro-poor expenditure have a significant impact on measures of well-being. Only in the case of middle-income countries does higher pro-poor government expenditure lead to higher levels of the HDI and in none of the specifications is there an effect of higher pro-poor expenditure on the level of infant mortality. Aid, on the other hand, is found to have a positive and significant effect on poverty. It is found to be an effective tool to increase HDI and decrease infant mortality in the majority of the authors’ specifications.
Williamson (2008) analyzes the impact of health aid on a number of different health outcomes. She finds that increases of aid targeted to the health sector does not have a significant effect on health outcomes in almost all of the specifications while the number of physicians does enter significantly. However, she does not account for the fact that increases in health aid might actually be responsible for increases in the number of physicians, therefore having an indirect effect on the observed health outcomes.
Kosack (2003) uses the growth rate of the human development index (HDI) to examine aid’s impact on general well-being, contingent on the level of democratization in the recipient coun-try. His results largely confirm previous findings that increases in the ratio of aid to GDP do not lead to increases in the growth rate of well-being, specifically the HDI. The level of democracy is also not significantly correlated with higher growth rates in the HDI. When including the interaction effect of aid and democratization, all of the relevant estimated coefficients become significant. There exists a positive interaction effect between aid and democratization, i.e. aid is more effective in countries with higher levels of democratization than in less democratic ones. Both of the coefficients for aid and democratization, on the other hand, are negative, suggesting that aid will have a negative impact on the growth rate of the HDI in undemocratic regimes and that undemocratic regimes will experience negative rates of HDI growth without aid (Kosack, 2003).
Batana (2010) uses stochastic dominance procedures to evaluate countries’ developments in assets, health, and a combined measure of the two. He relates these two dimensions to different levels of aid and government effectiveness in his sample of ten developing countries located in Africa. His results suggest that there is no relationship between higher levels of aid or gov-ernment effectiveness and developments in health, assets or the combined measure of the two. However, care must be taken, as his study examines only a very narrow sample both in terms of observations as well as geographically.
results are ambiguous. Countries receiving aid in the highest tertile of the aid distribution are more likely to improve access to drinking water, yet not to better sanitational facilities. On the other hand, higher access to clean drinking water does not lead to lower mortality rates, while better sanitational facilities do, at least for the highest tertile of the distribution. They offer possible explanations for their results, the first being a higher availability of improved sanitation compared to clean drinking water, and second population growth leading to a constant ratio of people with access to these facilities even though the absolute number might have increased. Dreher et al. (2008) examine the effect of aid targeted to the education sector on primary en-rollment rates in developing countries. Their results suggest that higher per capita aid devoted to education significantly increases enrollment rates while higher government expenditure in this area has no significant effect. They also find that other variables such as the level parental education or GDP per capita have no systematic effect on enrollment outcomes, and that in-creases in aid do not lead to inin-creases in government expenditure.
The study by Alesina and Dollar (2000) finds that there are large differences in donor motives between bilateral and multilateral aid. Bilateral donations seem to be driven by strategic in-terests or due to bilateral ties between donor and recipient while multinational ones are not. This suggests that poverty reduction is not the main objective for bilateral donors whereas multinational organizations are more focused on poverty alleviation. Kosack (2003), Masud and Yontcheva (2005), and Alvi and Senbeta (2012) confirm these results. They also find dif-ferent effects between grants and loans. Grants differ from loans because they do not have to be repaid. Therefore, they do not have to be invested into projects that lead to immedi-ate income-increasing effects. They might therefore be more effective in reducing non-income poverty as they can be used to finance government expenditure that is connected to public goods and social services.
2.4
Summary and Hypotheses
An extensive effort has been put into the analysis of the aid-growth as well as the aid-poverty nexus. Yet, even after almost 60 years of research, empirical findings have not converged. Some authors find no relationship at all (Rajan and Subramanian, 2008), others suggest that foreign aid only work in the right policy environment (Burnside and Dollar, 2000; Mosley et al. 2004), and again others that only particular types of aid work, yet with delayed effects (Clemens et al. 2012).
The positive effect of growth on poverty, if measured in terms of income, is less ambiguous (Dollar and Kraay, 2002; Ravallion, 1995). This is in line with neo-classical growth mod-els predicting that economic growth will lead to higher incomes and, in turn, lower absolute poverty. The same positive relationship holds true between aid and income poverty (Alvi and Senbeta, 2012). Aid can effectively decrease poverty by either increasing growth in constrained economies or by increasing income directly via income transfers to the poor.
anal-ysis focuses on measures of well-being assumed to be correlated with multidimensional poverty, such as child mortality or education outcomes, but this does not allow for direct conclusions about the level poverty itself. The aim of this thesis is to try and fill this gap in the existing literature. The MPI constitutes a natural starting point for this undertaking, as it extends the concept of income poverty to reflect capabilities also. It provides a clear poverty indicator that is in line with Sen’s ‘basic capabilities’ of being “well-nourished and well-sheltered, the capability of escaping avoidable morbidity and premature mortality” (Sen, 1995: pp. 45-46) and is comparable to the well-established HDI when it comes to the underlying indicators. This thesis will use the insights from existing aid effectiveness literature as well as the capability approach to analyze aid’s effects on changes in multidimensional poverty. In one way, it will stay within the scope of the existing literature in that it primarily focuses on short-term im-pacts of aid. This choice is made because long-term effects of foreign aid are almost impossible to capture in cross-sectional regression analysis. As the gap between aid flows and observed poverty gets larger, the direct effect of aid is ever harder to detect since it gets confounded with an increasing number of other factors.
However, it will also extend the existing literature by explicitly focusing on multidimensional poverty with its three sub-dimensions of education poverty, health poverty, and standards of living poverty. Insights can be gained from this approach, as many factors constituting poverty are not reflected in a person’s income alone. Empirical analysis incorporating the MPI instead of conventional measures of income headcounts will offer additional insights into the relation-ship between foreign aid and poverty. It will shed light on changes in poverty that cannot be explained with developments of income alone.
In the case of multidimensional poverty aid can work through a number of channels. If aid bridges constraints in the recipient economy, higher incomes can directly affect poverty even in non-income dimensions, as households are able to afford sending children to school and to buy medicine in case of sickness. Aid might also be effective by increasing spending in public goods and social infrastructure independent from economic growth and personal income. Increased access to clean drinking water, schools, or hospitals will have beneficial effects on multidimen-sional poverty.
Hypothesis 1: Higher levels of foreign aid will lead to larger improvements in multidimensional poverty.
more effective at reducing multidimensional poverty in more democratic regimes, as govern-ments have to take into account the well-being of their citizens and not only a narrow political elite. Furthermore, they are subject to penalties of the democratic system if they fail to explain themselves and their use of public funds.
Hypothesis 2: Higher levels of democratization will lead to larger improvements in multidimen-sional poverty at every given level of foreign aid.
Even though higher levels of foreign aid are expected to lead to improvements in multidimen-sional poverty, their effects are most likely not equally strong on each of the three different sub-dimensions. This is the case for at least two reasons: first, not all underlying indicators can be expected to be directly influenced by foreign aid in the first place. Second, not every indicator will be affected in the short-term – changes might take place only gradually and over a longer amount of time.
Hypothesis 3: Higher levels of foreign aid will affect each of the indicators differently under the same policy environments.
The effectiveness of foreign aid is not only contingent on the prevalent democratic environment, it is also dependent on the type of aid itself. Loans need to be repaid and must, thus, be used for projects that yield immediate monetary returns. This does not hold true for grants that can be utilized to improve poverty directly, also in non-income dimensions. Poverty reduction will be increased if countries receive higher amounts of grants. This will not be true for higher amounts of loans.
Hypothesis 4: Higher levels of grants will result in larger improvements in multidimensional poverty than will higher levels of loans.
Donor strategies differ. The literature has found that bilateral donors’ main objective might not be poverty alleviation but rather the furthering of strategic interests (Alesina and Dollar, 2000). Multilateral donors’ interests, on the other hand, are assumed to be more in line with the poor population in the recipient countries. The main objective for multilateral donations is poverty alleviation. Higher amounts of multilateral aid will lead to increased reductions in multidimensional poverty, whereas increases in bilateral aid money will not.
Hypothesis 5: Higher levels of multilateral aid will result in larger improvements in multidimen-sional poverty than will higher levels of bilateral aid originating from DAC-countries.
3
Methodology and Data
3.1
Model Specification
The dataset used for empirical analysis is a cross-sectional dataset including 148 observations of the MPI for 67 developing countries.12 After calculating first differences of the multidimensional
poverty index to assess changes in poverty over time, the remaining sample consists of 81 observations.
The basic model will use ordinary least squares regression and will take the form
∆Pi = β0+ β1 aidi+ β2 aidi× democi+ β3 democi+ β4 policyi+ x0β + εi (2)
where ∆P represents the average annual change in either MPI or one of its three sub-indices. The term aid represents the amount of aid flows to the recipient country. The variable democ is a measure of a country’s democratization and will enter the regression independently as well as an interaction term with aid flows to assess the effect of democracy on changes in multidimensional poverty and aid effectiveness under different levels of democracy. The variable policy proxies for macroeconomic and government expenditure policies that are expected to influence changes in multidimensional poverty independent from foreign aid and democratization. Vector x contains all other control variables affecting the change in poverty unrelated to aid, democratization or government policies. The last term, ε, is the error-term of the regression model. The subscript i indicates the observations and β the estimated coefficients for the respective variable.
The model’s aim is to make predictions about variations in the average annual change in multidimensional poverty. Estimated coefficients must therefore be interpreted as either leading to positive or negative changes in poverty. Aid’s impact on the change in poverty is further assumed to be non-linear, depending on the level of democratization in the recipient country. The effect of a change in foreign aid on the rate of change in multidimensional poverty will then be
∂∆P
∂aid = β1+ β2 democ. (3)
The overall effect of aid on the change in poverty will depend on the sign and size of the estimated coefficients. If both estimators are negative, marginal increases in foreign aid will unambiguously lead to faster reductions in multidimensional poverty and will do so more ef-fectively in better policy environments. If one of the two coefficients is positive, the overall effect is not clear a priori. In the unlikely case that β1 < 0 and β2 > 0, higher levels of aid
will lead to larger negative changes in poverty measure even at low levels of democratization. However, the impact will decrease with higher levels of democratization. The absolute effect on the change of poverty will depend on the size of the coefficients, however. In other words, the overall effect of higher levels of aid on changes in poverty could still be negative if the equation β1+ β2 democ < 0 at the maximum value of democratization. In the case that β1 > 0
and β2 < 0, higher levels of aid will lead to positive changes in the poverty measure in less
democratic countries. This effect will decrease as the level of democratization increases. A
12An overview of the countries included as well as the individual time periods used to calculate the average
similar reasoning applies concerning the absolute effect on the change of poverty at higher lev-els of aid. The overall effect might be positive if β1 + β2 democ > 0 for all possible levels of
democratization.
Higher levels of aid are assumed to accelerate the reduction of poverty as stated in hypothesis 1 since they can be used for larger unproductive government expenditure financing public goods and social infrastructure. Therefore, the estimated coefficient for β1 is expected to be negative.
The same holds true for the estimated coefficient of the interaction term, β2, connected to
hypothesis 2. Aid is expected to be more effective in more democratic countries leading to a negative coefficient. As stated in hypothesis 3, β1will vary for the three different sub-dimensions
of the MPI because not all of the underlying indicators can and will be affected equally. Finally, the coefficient of β3 is not clear a priori. More democratic countries could be characterized by
higher levels of re-distribution, which would lead to faster decreases of poverty. On the other hand, more democratic countries could see positive changes in poverty indices if the democratic majority took actions that be beneficial to them but would harm the less influential interest group of the poor population.
In the case of hypotheses 4 and 5, the model will be modified slightly to include two aid terms and two corresponding interactions:
∆Pi = β0+ β1 aid1i+ β2 aid1i× democi+ β3 aid2i+ β4 aid2i× democi+ β5 policyi+ x0β + εi. (4)
It is expected that β1 < β3 if aid1icorresponds to grants or multilateral aid and aid2imeasures
flows of loans or bilateral aid, respectively. Coefficients connected to grants are expected to be smaller than those for loans as stated in hypothesis 4. The same holds true for the coefficients of multilateral aid compared to bilateral aid as formulated in hypothesis 5.
The control variables included in the model are taken from various sources in the literature. Macroeconomic policies are in line with those used by Burnside and Dollar (2000) while gov-ernment expenditure variables are taken from Mosley et al. (2004), respective Gomanee et al. (2005). The remaining controls are growth in per capita income, growth in urbanization, and a dummy for Sub-Saharan Africa. These are chosen in line with the existing aid effectiveness literature on growth and poverty, such as Boone (1996), Dollar and Kraay (2002) or Rajan and Subramanian (2008).
Government expenditure policies are expected to have a positive effect on poverty reduction. In other words, higher levels of government expenditure on either health or education are as-sumed to lead to faster decreases in multidimensional poverty. A negative estimated coefficient is expected for these two variables. However, including government expenditure and aid flows into the same regression might cause insignificance of the aid term. This will be the case if higher flows of foreign aid will lead to higher government expenditures. Aid’s effect on poverty would then work indirectly via the recipient’s expenditure and would not result in a significant estimator. The argument is similar to that applied by Hansen and Tarp (2000) concerning aid and saving coefficients in growth regressions: only if both estimators are insignificant, can one conclude that aid has no effect on poverty reduction. Nevertheless, both terms should be included into the regression since the evidence that aid significantly increases government expenditure is weak (Gomanee et al. 2005; Mosley et al. 2004). Additionally, aid might still have a direct effect on poverty even after controlling for pro-poor expenditure, for example via NGO-projects affecting the population directly and without the help of the recipient’s govern-ment.
Increases in per capita income are expected to result in larger negative changes of multidimen-sional poverty. Notwithstanding the fact that per capita GDP largely accounts for the monetary situation of the population, it is also correlated with the general economic environment of the country. Countries with larger increases in per capita income are expected to reduce poverty more quickly.
The effect of growth in urbanization is ambiguous. On the one hand, the task of providing basic social services for a larger share of the population will be easier in more urban countries, as more people can take advantage of the existing infrastructure at a time. This should result in higher levels of people with access to education, health care as well as access to sanitation and clean drinking water – the estimated coefficient would then be negative. On the other hand, higher population density facilitates the spread of disease. Also, urban slums seldom receive widespread access to electricity or other basic infrastructure. If this effect is predominant, the estimated coefficient will be positive.
Lastly, changes in multidimensional poverty might be contingent on geography. If there are systematic differences in poverty development in Sub-Saharan Africa, this would bias the re-sults of the estimation, as almost half of the countries in the dataset are located in that part of the world.
The second exception will be growth in per capita income. Income increases have been found to be strongly correlated with poverty reduction (Dollar and Kraay, 2002) but are also volatile and dependent on overall economic performance of the country and the world economy itself. Income growth at the beginning of the observation period does, therefore, not proxy well enough for overall increases in income so that growth in income will enter as the average growth in per capita GDP between the two observations.
The remaining variables do not show large amounts of volatility over time. The level of democ-ratization, for example, changes only gradually. The same holds true for growth in urbanization or trade openness.
The following example clarifies the setup of each individual observation: MPI scores for Kenya come from the years 2003 and 2009. The change in multidimensional poverty is regressed on aid flows averaged over the five years preceding and including the first observations, i.e. aid flows in the years between 1999 and 2003. Income growth will enter the regression as the average growth of income between 2003 and 2009. All other variables will enter the regression with their respective values in 2003.
Theroretical Assumptions
The validity of the ordinary least squares model, its estimated coefficients as well as the stan-dard errors of these coefficients is reliant on a number of assumptions in cross-section analysis, the so-called Gauss-Markov assumptions. If any of these assumptions are violated, estimators could be biased, inconsistent and inefficient. Standard errors could be invalid, leading to mean-ingless t-statistics and significance levels. However, even if all of the Gauss-Markov assumptions are fulfilled, influential outliers in the dataset could still lead to biases in the results.
growth or poverty leading to higher aid flows to these countries. In the context of this thesis possible endogeneity of aid will be mitigated by including the aid term as the average of the five years preceding and including the first observation of the MPI. Aid flows in the past should not be caused by potential poverty reduction performances in the future.
Another prerequisite of ordinary least squares estimation is the assumption of no perfect collinearity between the regressors, i.e. there are no exact linear relationships between in-dependent variables and none of them is a constant. Failure to satisfy this assumption would result in singularity of the matrix (X0X) in the equation for OLS-estimation so that OLS can-not be applied (Wooldridge, 2002). However, in the multivariate regression model problems can arise even if independent variables are not perfect linear combinations but are instead highly correlated, i.e. if there exists multicollinearity. This results in inflated standard errors of the estimated coefficients and might lead to falsely rejecting null-hypotheses of coefficient significance (Wooldridge, 2002). Possible issues concerning multicollinearity will be assessed by analyzing cross-correlations between the independent variables. If high correlations are found, this will be taken into account when specifying the individual models. Additionally, variance inflation factors (VIF) will be calculated after the regressions to assess if multicollinearity is still a problem.
The final assumption of multivariate least squares estimation concerns homoscedasticity and normality of the estimated residuals. Residuals obtained after estimating the empirical mod-els should have mean zero and a constant variance around that mean. They should also be distributed normally around the mean (Wooldridge, 2002). Homoscedasticity will be checked after the initial regressions and will be accounted for using robust standard errors if necessary. The problem of influential outliers is independent from the theoretical assumptions of the em-pirical model. Even if all assumptions are fulfilled, estimation results might not approximate the true linear relationship between the independent and dependent variables in case there are observations that differ substantially from the rest of the dataset. Since OLS minimizes the sum of squared residuals, outlying observations will receive an over proportionately high weight when estimating the slope coefficients. This might influence both the estimated coefficients as well as their standard errors. The problem is especially relevant in small datasets such as in the case of this thesis (Wooldridge, 2002). Potential outliers will be addressed by calculat-ing Cook’s Distances (Cook, 1977) for each observation and removcalculat-ing influential observations before re-running regressions.
3.2
Data and Descriptive Statistics
The following section describes the data that will be used for the empirical analysis below. It will further provide descriptive statistics of these data.
Data
Oxford Poverty and Human Development Initiative (OPHI) and uses household survey data to calculate the multidimensional index of poverty and the three sub-indices. These dimensions correspond closely to those used in the HDI, namely the ability to be educated, the ability to live a long and healthy life and the ability to enjoy a basic standard of living.
The MPI is computed by applying the so-called “dual cutoff method of identification” (Alkire and Foster, 2007: p.3.). After identifying relevant indicators of multidimensional poverty, in a first step a poverty line for each indicator is established. These are the deprivation cutoffs. A person is considered deprived in that indicator if she falls below the pre-specified cutoff. How-ever, limited data availability prevents the authors from basing the MPI on individual persons so that their unit of analysis is the individual household.13 Individual weights are assigned to the indicators to calculate a person’s deprivation score, i.e. the weighted sum of her depriva-tions. People are defined as poor if their deprivation score is above a pre-defined value, the poverty cutoff. Aggregating all households defined as poor yields the unadjusted headcount ratio of multidimensional poverty or incidence of poverty. To receive the MPI, this unadjusted headcount is multiplied with the average weighted ratio of deprivations experienced by the poor population (Alkire and Foster, 2011). An overview of the different indicators, their deprivation cutoffs and their respective weights can be found in Table A2 the Appendix.
The MPI is aimed at providing an overview of current multidimensional poverty across the world. Therefore, each individual release only carries one observation for each country at a time based on the most recent household survey carried out in the respective country. How-ever, since its initial publication in 2010 the index has been updated four times replacing a number of poverty scores with more recent data. To construct the dataset for this thesis all five releases of the MPI were retrieved from OPHI. Starting with the first report of 2010 and with the help of the Methodological Notes (Alkire et al. 2011, 2013, 2014, 2015) accompanying each individual release, the updated values for the MPI were picked from the subsequent reports and included in the dataset. Sub-indices for the three observations were computed using the methodology applied by OPHI to calculate the overall MPI: for each dimension the weighted average of the respective indicators was calculated and also included in the final dataset (Alkire and Santos, 2010).
This resulted in a total 67 countries with at least two observations and a total number of 148 observations. For the construction of the dependent variable used in the regression analysis, the value of the initial observation was first subtracted from that of the subsequent one and then divided by the number of years between observations. The dependent variable is therefore the average annual change in MPI or its three sub-dimensions between two points in time for each individual country. For example, the observations for Rwanda come from OPHI’s releases
13The indicators are chosen in a way, however, to account for the fact that deprivations of one member of the
in 2010 and 2013 and cover the years 2005 and 2010 respectively. Multidimensional poverty as measured by the MPI declined from 0.44 to 0.35 during that time frame. Thus, the average annual change in poverty is the difference between these two scores divided by the five years in between observation and is equal to a reduction of 0.02 per year. The average gap between observations in the dataset is slightly over five years, however, the data shows large variation for the time period between observations. In some cases household surveys were only one year apart as, for example, in South Africa with observations in 2007 and 2008. On the other end of the distribution are the Central African Republic with ten years difference between obser-vations or the Comoros and Gabon with twelf years in between household surveys. It can be expected that countries with larger time gaps between observations experienced larger changes in poverty. For this reason the overall change was divided by the number of years in between observation.
The overall number of observations that was obtained using this methodology amounts to 81. Most of the countries enter with only one observation into the dataset, i.e. one average change in poverty. Ten countries have two observations each. These are the Democratic Republic of Congo, the Republic of Congo, Jordan, Mozambique, Nicaragua, Peru, the Philippines, Senegal, South Africa, and Suriname. Two countries, Nigeria and Sierra Leone, have three observations. Including several observations for some countries leads to a higher weight of the developments within these countries compared to the remaining sample. If countries with more than one observation are systematically different from the others, this could bias the results. Therefore, robustness analysis will check for possible distortions of the overall results.
Hypothesis 3 stated that higher levels of foreign aid will affect each dimension differently. The reason for this is the exact definition of the indicators within the MPI. Particularly, this is the case of the indicators of the level of schooling and child mortality. They are defined in a way that make short-term changes caused by higher levels of aid unlikely. A household is deprived in the indicator of formal schooling if none of its members has a formal education of five years or more. Even if foreign aid is able to increase enrollment rates, the effect on households will materialize only with delayed effects since the majority of the population will not benefit from it. The same holds true for the indicator of child mortality. A household is considered de-prived if at least one child has died in that household at some point in the past. Even if aid were able to drastically reduce child mortality, the effect would still not be reflected in those households, which have experienced child death even though the actual risk of child mortality would be lower. Additionally, the choice of flooring or cooking fuel might take longer time to be responsive to higher aid flows since it is unlikely that households will immediately spend higher incomes on a new house but might instead increase consumption.
access to electricity, clean drinking water or improved sanitation can be directly influenced by foreign aid. Therefore, each dimension of multidimensional poverty is at least in principle responsive to changes in the level of aid making the MPI and its sub-dimensions a valid part of regression analysis.
The data for aid flows is retrieved from the Organisation for Economic Co-operation and De-velopment’s (OECD) Development Assistance Committee’s (DAC) database. Aid is defined as official development assistance (ODA). Five different flows of aid are included in the empirical analysis, which differ in terms of type and source. Analysis will be done with grants and total net loans as well as the sum of these two terms, total net ODA. Furthermore, it will be distin-guished between multilateral donors and DAC-countries. Moreover, aid flows will be divided by gross national income (GNI) of the recipient country in the respective year to control for differences in the different economies. The resulting ratio of aid flow to GNI provides a measure of aid dependency of the recipient country.
The level of democratization and accountability will be proxied by the Center for Systemic Peace’s Polity IV index, which combines scores for a country’s level of autocratic and demo-cratic characteristics. The Polity IV index is calculated by subtracting the score of autocracy from that of democracy leading to index values between -10 and 10. The combined Polity IV index will be used for analysis instead of of the dimension of institutionalized democracy only. The reason for this is that the overall level of democratization not only depends on the presence of democratic institutions but also on the absence of autocratic institutions. For example, sup-pression of political competition is a sign of lower levels of democratization, yet it is included in the measure of autocracy in the Polity IV index. For the purpose of this thesis, the index is recoded so that it only takes positive values. A constant of 10 is added to each score so that the variable now ranges from 0 to 20 instead of -10 to 10.
Data for the remaining control variables come from the World Bank. Data for macroeconomic policies includes inflation in consumer prices, trade openness measured by the ratio of exports plus imports over GDP, and budget surplus defined as the cash surplus or deficit as a per-centage of GDP. Health expenditure is measured by public health expenditure as a perper-centage of GDP. Education expenditure is defined as government expenditure on education, also as a percentage of GDP. Finally, per capita income growth and growth in urbanization also enter the regressions.
Descriptive Statistics
Table 1 provides descriptive statistics of the variables that will be used in the empirical anal-ysis below. The first four rows show data for the average annual change of the MPI and the three sub-indices for education, health and living standards. The following five rows present the results for overall ODA, different donors and different types of aid. The remaining rows contain data for the policy and control variables.
