Spring Semester 2020
Are we Making Promises without Proof?
An empirical analysis of the impacts that democracy support and aid targeting education have on democratization
Abstract
As democracy promotion has become an increasingly important aspect on the agenda of foreign aid donors, and since such prioritization of funding comes at the expense of other development areas, it is arguably of interest for donors as well as researchers to investigate its actual impact on democratization. This study endeavors to examine the influence of two types of foreign assistance that could potentially contribute to a democratic development, directly through democracy support and indirectly through aid focused on education. Four models of regression analysis are applied on a data set of 65 developing countries receiving Official Development Assistance (ODA), during the period of 2006–2018. The findings of this study are inconclusive in determining the influence of these aid types, as the main results show no significant effects on the Freedom House grading of the recipient countries. Yet, when using an alternative measurement, the Democracy Index, directly focused democracy support appears to have a slight positive and significant impact on democratization. These results should however be interpreted with caution due to the risk of reversed causality.
Keywords: foreign aid; aid effectiveness; democratization; democracy aid; education aid
Table of Contents
1. Introduction ... 3
2. Theoretical framework and previous empirical research ...6
2.1 Theoretical framework ... 6
2.1.1 Foreign aid in general and its expected impact on democratization ... 6
2.1.2 Democracy aid and its expected impact on democratization ...7
2.1.3 Aid targeting education and its expected impact on democratization ...8
2.2 Previous empirical research in the field ...10
2.2.1 Previous research on the correlation between foreign aid and democracy…………...10
2.2.2 Previous research on the correlation between foreign aid, education and democracy...11
3. Data …...13
3.1.1 Data on democracy-levels ...…...13
3.1.2 Aid Data ………...…...15
3.1.3 Democracy aid ………...…...16
3.1.4 Education aid ...…...16
3.1.5 Decisions to make when collecting data on aid ...…...17
3.1.6 Data on additional variables used in the analysis ……….…………...17
3.2 Descriptive statistics ………...………...………...….……...19
3.3 Missing data and sample selection bias...……...…………...………...….……...20
4. Methodology ...…...20
4.1 Regression models …....…………..…………...…...20
4.1.1 Model 1- model 3…....………...…...21
4.1.2 Model 4 ...…...23
4.1.3 Model 5 ...…...25
5. Results ...…...26
6. Discussion…………...…...29
6.1 Analysis of the results ...…...29
6.2 Limitations of this study ...…...32
6.3 Suggestions for future research ……...33
7. Conclusion …………...…...34
8. References…….…...35
9. Appendix …………...…...39
1. Introduction
The challenge of measuring the effectiveness of foreign aid has been on the agenda for researchers in economics as well as social science for decades. The actual impact of foreign assistance is a highly controversial subject. For the greatest part of the history of aid its main focus has been on poverty reduction and promotion of economic growth. Consequently, previous research on aid effectiveness has mainly studied the correlation between aid flows and such progress, or absence of progress (Easterly 2003; Wright and Winters 2010).
However, for the past twenty years, democracy and human rights have become increasingly important aspects of international development aid. In Sweden, the current Minister for Development Assistance, Peter Eriksson emphasizes the importance of international development policies that are promoting democratization1. Such focus is apparent in the Swedish Statement of Foreign Policy in which it is stated that Sweden will intensify the work to promote and defend democracy and thereby increase the amount of aid allocated to
democracy assistance (Swedish Government, 2020). In the 2020 budget proposal by The Swedish International Development Cooperation Agency (Sida), twenty five percent of the total aid amounts has been allocated towards promoting democracy (Sida, 2020a).
Yet, the empirical findings of earlier quantitative studies on the impacts of foreign aid on democracy are rather inconclusive. An extensive and well cited study conducted by Stephen Knack (2004) analyzed a data set of 104 developing countries receiving foreign aid over the period 1975–2000, suggests that there seems to be no signs of a positive correlation between foreign aid and democratization. There are even scholars arguing that foreign aid has a negative influence on democratization (Kalyvitis and Vlachaki 2012). At the same time, two
1 There are numerous ways to define democratization, some definitions imply that the direction of democratization could be both positive and negative in reaching democracy. Further, scholars often divide democratization into different phases, yet, this research will refer to democratization simply as the process of making countries more democratic in line with Robert Dahl's definition of a polyarchal democracy that will be further discussed on page 13.
more recent studies (Scott and Steele 2011; Altunbaş and Thornton 2014), who both included data on the first decade of the 21st century, suggest that foreign aid appears to have a small positive impact on democracy over time. To conclude, the findings of earlier scholars are inconclusive in determining the influence of foreign aid on demoralization.
The emerged focus on democracy as an aid target, comes at the expense of development assistance allocated to other areas. Thus, it is arguably of interest for aid donors as well as researchers, to reach further knowledge about what aid forms that are most likely to have a positive impact on the democratization. In the light of the currently favored approach of result-based management (RBM) among aid donors, measurable and verifiable results of the impacts of foreign assistance are of often regarded as important aspects when improving aid effectiveness (Brolin, 2016).
Further, it is likely that recent data from the latest decade differentiate from previous
research. In addition, at the contrary to the previous mentioned scholars, this study builds on data from the Organization for Economic Cooperation and Development (OECD), after 2002 when the database made great improvements in terms of credibility and accuracy (Cornell, 2008).
Moreover, a great share of the total empirical research on aid and democracy studies the aggregated amount of foreign aid and thus does not consider that aid projects have different purposes and are not all expected to have an impact on democratization. This study will focus on the effects of two aid forms that are likely to improve democratic development in two different ways, directly through democracy support and indirectly through aid focused on education. Foreign assistance that generate improvements in education of the recipient country can thereby enhance structural social changes and increase the demand for democracy. The theoretical arguments behind including education aid as a potential
explanatory variable for democratization are further explained in the next chapter. The total amount of aid flows allocated to projects promoting education is comparable to the amount of aid means targeting democratization. The average amount of democracy support per capita between 2006–2018 in the 65 observed countries was approximately 3.1 US dollars, while that amount was roughly 4.2 US for education aid (see table 1).
Hence, this study intends to provide a more updated and recent contribution to the existing research in this field, and also to introduce a comparative dimension by analyzing the impacts of two different aid forms. The research question for this study is:
- What impact does democracy aid and education aid have on democratization?
In an attempt to investigate the subject of interest, this study will analyze a panel data set of 65 countries that have been receiving both democracy support and aid promoting education that are included in the definition Official development assistance (ODA)2 between the years of 2006–2018. As indicator for democratization, two different measurements of democracy are tested, The Freedom House and The Democracy Index. Four different panel data
regression analysis models are applied, one of them including instrument variables with the purpose to mitigate the risk of reversed causality. An additional regression is added to further investigate the influence of education aid on the education level of the recipient countries.
Based on the findings of this study, there are no implications that neither of these aid forms have any significant impact on the of The Freedom House grading of the recipient countries.
Yet, when instead using the Democracy Index as outcome variable, the estimated impact of democracy support is positive and statistically significant.
This research begins by providing a brief introduction to the theoretical framework in chapter two, in which the expected impacts on democratization of foreign aid in general, democracy support and education aid in particular. The other part of chapter two review previous empirical research that is of relevance to this thesis. Earlier quantitative studies on the correlation between foreign aid and democracy are brought up, as well as previous research on the relationship between foreign aid, education and democracy.
2 Official development assistance (ODA) is foreign aid provided by the Development Assistance Committee (DAC) of the Organization for Economic Co-operation and Development (OECD). Although foreign aid and ODA are often used interchangeably, what can technically be counted as ODA is more restrictive than what is regarded as general foreign aid. Firstly, for means to be qualified as ODA, funding must be provided by governments. Hence, donations from private actors or non-governmental organizations (NGOs), are not included in what counts as ODA. Secondly, military assistance and export credits meant primarily to promote sale of goods from the donor country, are not included in ODA (Brown, 2015).
Chapter three will describe the variables and the data of this study. First, the two chosen measurements of democracy are reviewed. Secondly, data and measurements on aid are described together with specific information regarding the two aid forms and decisions to make when examining aid effectiveness. Third, data on the control variables and the instrument variable are presented. Fourth, descriptive statistics for all variables included in the main analysis are presented. Lastly, potential challenges and sources to bias related to missing data are briefly discussed. In chapter four, the regression models that are applied in this analysis are described and motivated. Followed by the fifth chapter in which the main results are presented and briefly discussed.
Chapter six starts with a more thorough analysis of the main findings from the results. The discussion continues with a comparison of these findings in relation to previous mentioned research and the theoretical framework of this study. Suggestions for further research is brought up, and the limitations of this study are discussed. Lastly, in chapter seven, conclusions are drawn based on the findings of the empirical analysis.
2. Theoretical framework and previous empirical research
2.1 Theoretical framework
2.1.1 Foreign aid in general and its expected impact on democratization
Previous scholars are not in agreement regarding the relationship between foreign assistance and democratic development. Grossman (1992) highlights the risk that foreign aid can encourage coup attempts and create political instability due to increased incitements to be in control of the government by making it a more valuable prize. Another potential risk that appears when aid is allocated directly to the recipient government is brought by Friedman (1958) who argues that the role of the state then tends to strengthen relative to the private sector and such development tends to have a negative impact on a democratization process.
Moreover, Knack (2004) discusses the fact that foreign aid will reduce the government´s dependence on tax incomes. If a great share of the government revenue is based on foreign aid, the recipient government is accountable to foreign donors rather than to the citizens
paying taxes. Such issues with non-tax revenue is similar to when the economy of a
government is highly dependent on natural resource wealth. It creates a situation in which the leaders do not need to consider the public opinion when making decisions, and consequently reduce the incentives for citizens to participate in the political process.
Notwithstanding, Knack also emphasizes that such dependency on foreign donors can instead work in favor of democracy promotion since it makes it possible for donors to put certain demands or conditions on the recipient country to continue to receive aid. Collins (2009) summarizes the positive ways in which general development assistance may promote
democracy, either as a byproduct from structural changes within the recipient country or as a consequence of top-down aid conditionally.
Conditional aid is foreign aid that comes with certain conditions. The donor of such
assistance induces the recipient country to pursue certain goals and to adopt certain policies, to which the recipient would otherwise not be given the same amount of aid, or even not be given assistance at all (Selbervik, 1999).
2.1.2 Democracy aid and its expected impact on democratization
In terms of aid that is directly targeting democracy promotion, Scott and Steele (2011) among other scholars (Collins 2009; Finkel et al. 2007) in this field use the terms Anticipated
Reactions and Agent Empowerment, when describing the mechanisms that are linking democracy support and democratization. The first term describes how means of democracy aid is allocated, while the secondly mentioned term explains how such means are distributed within the recipient country.
Scott and Steele (2011) explain Anticipated Reactions as the way that both donors and
recipients are strategic actors. Donors tend to calculate and predict the outcome of democracy promoting aid in a given situation. Such predictions do not necessarily result in allocation of democracy support to countries that are already democratizing, it rather implies that donors aim to identify recipient countries where a democratization process is possible to proceed.
USAID (United States Agency for International Development) is brought up as an example of a donor that focuses on targeting the developing countries that are most likely to achieve democracy when allocating democracy aid.
Moreover, Scott and Steele address that the behaviors of the recipients are also likely to be driven by anticipated reactions. They argue that democracy support is considered to be a rather prestigious category within foreign assistance, and receiving such aid is regarded to be a signal of preference by the donor country. Recipient countries that are eligible for
democracy support are then seen as potential recipients for other aid types, such as general economic aid. For the recipient to continue to receive democracy support as well as other forms of aid, progress must be done, which of course the recipients are aware of. The donors often expect evidence of such progress before renewing aid commitments. There are
therefore incitements for the recipients to effectively spend the aid means. If the democracy support is not generating the expected outcome, the donors can strategically shift means to where it is more likely to be effective, either within the recipient country or to another country.
The second mentioned link between democracy aid and democratization, Agent
Empowerment, can be viewed as more direct than previously mentioned potential outcomes of how foreign aid might affect democracy development. As noted, when describing the relationship between aid in general and democratization, foreign assistance may indirectly produce certain conditions that are in favor of a democratic development. However, Finkel et al. (2007), describe how democracy promoting assistance often focuses on empowering certain actors such as individuals, NGOs (non-governmental organizations) and political parties. Collins (2009) explains such Agent Empowerment as both the process in which donors attempt to cultivate independent and qualified legislatures and judiciaries, as well as aid packages assisting civic NGOs, democracy activists and the independent media. Also, actors within the private economic sector are included in these agent groups, in line with the previously mentioned arguments by Friedman (1958) who highlights the importance of a strong private sector as a counterweight to the power of the state.
2.1.3 Aid targeting education and its expected impact on democratization
When determining the possible impact that aid targeting education might have on the democratization of a country, it is of relevance to discuss Seymour Lipset's modernization theory. Lipset claims that democracies were created and preserved through a process of modernization, in which there is a strong link between economic development and
democratic development (Lipset, 1959: 75). Lipset argues that “all the various aspects of economic development – industrialization, urbanization, wealth, and education – are so closely interrelated as to form one major factor which has the political correlate of
democracy” (Lipset, 1960: 41). Simplified, Lipset argues that factors that are closely related to economic growth such as industrialization and increased levels of education are well correlated with each other, and together these factors increase the likelihood for a country to reach a democratization. In particular, the modernization process is expected to strengthen the middle class, who no longer will accept oppressive regimes. As a consequence, a democracy will evolve (Lipset, 1959: 80).
Accordingly, if aid successfully manages to improve the education of a country and such development increases the general material standard, it could create increased demand for democracy among the citizens. Although Lipset's Modernization theory might be useful to some extent when explaining the emergence of the democratic culture of the western world, there are more recent historical examples such as China in which the theory does not seem to be as plausible.
Another concept of interest when discussing the relationship between education and democracy is the existence (or non-existence) of democratic norms. Almond and Verba (1963) emphasized the importance of certain values and beliefs of the citizens for democratic advancement to be reached within a country. Democratic norms are crucial for creating the desire and ability of citizens to participate individually and together in the public affairs that are affecting them. Almond and Verba (1963) claimed that a stable and effective democracy depends upon more than the structure of government and politics, the orientations of the citizens are essential elements in democratization. Arguably, political norms are partly related to education and the way in which knowledge and beliefs about democracy is given to
children in school as well as students in higher degrees of education. By allocating aid means to projects that successfully increases the levels of education, such aid can contribute to create a democratic culture in which the citizens attitudes towards a democratic political system are positive. This is however based on the assumption that the provided education is grounded in values that are positive towards democracy as a political system, and such condition might not always be satisfied.
2.2 Previous empirical research in the field
2.2.1 Previous research on the correlation between foreign aid and democracy
The increased ambition to target democracy when allocating means of aid is something that has grown recently, even though the body of empirical literature is growing, the subject is far less studied than for instance aid´s impact on economic growth. The effects that aid has on democracy development have only been studied less. It is also notable that a great share of earlier studies do not distinguish democracy assistance from other aid but look at the overall effect of the aid (Knack 2004; Altunbaş and John Thornton 2014), or solely examined the effects of US democracy promotion (Finkel et al. 2007; Carothers 2009; Scott and Steele 2011). Due to the risk of simultaneous causality3 when investigating the influence on aid on democratization, the majority of previous scholars on the subject include empirical models that are controlling for endogeneity.
One of the first and most influential studies on this subject is the one done by Lead
Economist at the World Bank Stephen Knack (2004), who provides a multivariate analysis based on a large sample of developing countries receiving aid from the DAC-donors over the period 1975–2000. Knack uses two different measures of democratic development (Polity IV and Freedom House) and he also includes two different measures of aid intensity as the explanatory variables, he uses both Aid/ GNP and Aid/ Government expenditure, which adds robustness to his results. However, an issue with the first mentioned measure of aid intensity, Aid/ GNP, is brought up by Cornell (2008) who highlights the risk of using GNP as a
nominator since fluctuations in the GNP are likely to have a large effect on the measure of aid intensity accordingly also the results.
Another potential weakness of his model is that he uses the change in democracy level between the period 1975–2000 as the dependent variable, and the mean of the percentage of aid during this period as the independent variable. Such ways of measuring the studied
3 Simultaneous causality appears when in addition to the causal link between the independent variable (x) and dependent variable (y), there is also a causal link between the dependent variable (y) and independent variable (x) (Stock and Watson, 2020). This is a potential source to bias when examining the impacts of foreign
assistance as donors are likely to “reward” and allocate aid funding to a greater extent to recipients that are more likely to achieve democracy. This will be discussed further in this paper.
variables simplify the calculations of the model by giving each observed country one value for each variable. At the same time, it does not consider that democracy levels might not have developed in a linear way and the fact that the amount of aid could have changed between one year to another in a way that might have had an impact on the level of democracy.
In a more recent study, Altunbaş and Thornton (2014) estimated the impact of foreign aid on democracy by using a panel of 93 developing DAC aid-recipient countries during the time period of 1971–2010. Likewise, to Knack, they examine the impacts of total amount of foreign aid. However, on the contrary to the results of Knack, the findings of their study imply that foreign aid does have a small positive impact on democracy over time.
Finkel et al. (2007) as well as Scott and Steele (2011) focuses on the influence of democracy assistance and not general development assistance, yet they examine only assistance from USAID. The first mentioned builds on data on 165 countries over the period 1990–2003 and the latter one observes 108 countries between 1988–2011. Both of these studies suggest that democracy support has a positive impact on democracy, even when controlling for
endogeneity.
Narrowing the data to only one donor makes it possible to use models that better control for donor behavior and aid allocations patterns which makes it easier to mitigate the risk of simultaneous causality. However, Scott and Steele (2011) stresses that USAID to a great extent tends to focus their allocation of democracy support to the countries which are most likely to successfully become democracies. When only studying the impacts of USAID in their recipients’ countries, there might be sample bias that threatens the external validity of the study, since these countries are not representative for developing countries in general.
Lastly, according to the findings Kalyvitis and Vlachakis (2012), foreign could in fact have a negative impact on democratization. According to their study on 64 ODA-recipients during the 1967–2002 period, aid flows decreases the likelihood for the recipient country to achieve democratization. However, they treat the outcome variable (democracy) as binary, taking on the values 0 or 1, since democratization arguably is a gradually process, such way of
measuring to not capture nuances that are of relevance when analyzing change in the democracy level of the recipient country.
2.2.2 Previous research on the correlation between foreign aid, education and democracy The finding of a study by Asongu and Tchamyou (2019), focusing on the effects of foreign aid on education levels in Africa between 1996–2010, implies that there is a small positive effect. The impact is mostly referring to an increased level of primary school completion rates. They conclude that a 1 percent increase in education aid raises the primary completion rate by 0.20 percentage. Their results are in line with a study on the correlation between aid and education between 1995–2010 done by Riddell and Niño -Zarazúa (2016) who states that aid has made a positive contribution to education in the recipient contribution in terms of expanding enrolments. Yet, they also emphasize that such improvements are small and are found mainly on basic education.
The correlation between education and democracy, which is relevant to investigate further when examining the impact of education support and democracy, has been the subject for numerous empirical studies. L.Glaeser et al. (2007) imply that there is a correlation between education and democracy, although the reason for this correlation is not as clear. Meaning that is not possible to conclude that this positive relationship is due to causality between the variables. Additionally, the findings from a research done by Daron et al. (2005) suggests that such positive correlation is not robust when including fixed effects and hence controlling for the within-country variation, which implies that the findings of a positive correlation in cross- sectional studies are positively biased due to omitted factors influencing both education and democracy. Based on these ambiguous results, the correlation between education aid and democracy estimated on the data of this study will be of interest.
3. Data
In order to examine how the two independent variables, democracy support and aid targeting education, have been affecting the outcome variable, the democracy level, data on 65
countries receiving ODA during the period of 2006–2018 are examined, thus the total number of observations are 845. A complete list on the observed countries is found in the Appendix.
Since there are missing values for certain country-year observations the constructed data set is unbalanced. There are more than 65 countries receiving aid from the DAC-donors, the decision not to include all of these countries is motivated by lack of complete data on the variables on the other ODA receiving countries.
3.1. Data on democracy levels
To capture changes in the outcome variable, democracy, two different measurements of democracy are used; Freedom House and The Democracy Index. Focus is mainly on the Freedom House estimations, while the measures of The Democracy Index are reported as control for robustness. Both of these indexes are in line with the inclusive definition of democracy provided by Robert Dahl (1989) who uses the term polyarchy to explain its meaning. According to Dahl´s definition, various democratic conditions4 can be quantified and fulfilled at different levels. For a country to be considered to be a polyarchy, the score of all of these conditions needs to be maximized, however such a state does not exist in practice, it should rather be seen as a democratic utopia for states aiming to be democratic to strive for.
Freedom House does not only include political procedures in its grading, it also considers institutions and political freedoms, which reflects the extensive Dahlian definition of a polyarchal democracy (Cornell, 2008). Moreover, both Freedom House and The Democracy Index are graded measures of democracy and can better capture nuances of democracy as
4 A polyarchy has to fulfill following conditions (1989: 333): 1) The control of governmental decisions about policy is constitutionally vested in officials that have been elected. 2) The elections of such officials are done relatively frequently and in a complete free and fair way. 3) Practically all adult citizens have the right to vote in the elections. 4) Most adult citizens have the right to run as candidates in these elections. 5) The citizens have an effectively enforced right to freedom of expression. Such freedom includes the right to criticize the government, the prevailing system, the economic, political and social system. 6) The citizens have access to alternative sources of information that are not monopolized by the government. 7) The citizens have an enforced right to form and join autonomous associations, including political associations such political parties that can attempt to influence the government and are allowed to compete in elections.
compared to dichotomous measurements, which makes it possible to study gradual democratic developments (Collier and Adcock, 1999).
The measurement conducted by the non-government organization Freedom House is often used by scholars as a variable indicating the democratic status of a country. Although the index has received critique for including freedoms that are not completely related to
democracy, such as economic freedoms, it is still regarded to be one of the most accurate and measurements of democratic development (Cornell, 2008). The reports done by The Freedom House covers every state of the world and the index has been published annually since 1973.
It is based on a two-tier system consisting of scores and status. Due to the primarily focus on gradually changes in democracy, only the score system is applied in this research. Based on 25 indicators, divided into 7 main topics 5, each country or territory is awarded a score between negative 4 and 100, in which the optimal democracy receives 100 (Freedom House, 2020).
As a robust check, the results from the Democracy Index will also be reported. It is an index compiled by the UK based company the Economist Intelligence Unit (EIU). It measures the level of democracy based on 60 indicators within five categories6. Each observed country receives a number between 0–10, rounded with two decimals, in which a country that is regarded as full democracy receives the score 8.01–10 (EIU, 2020).
Since the Democracy Index is relatively new (it was first published in 2006) it has not been used to a great extent by previous scholars in the field. The results of this measurement are hence of interest to compare to the more commonly used Freedom House index. It has however received critique for lacking transparency in its process of grading (Tasker, 2016).
Due to such critique along with the fact that it is not as frequently used as Freedom House by
51) Electoral Process: executive and legislative elections, and electoral framework. 2) Political Pluralism and Participation: party system, competition, freedom to exercise political choices, and minority voting rights. 3) Functioning of Government: corruption, transparency, and ability of elected officials to govern in practice. 4) Freedom of Expression and Belief: media, religious freedom, academic freedom, and free private discussion. 5.
Associational and Organizational Rights: free assembly, civic groups, and labor unions. 6. Rule of Law:
independent judges and prosecutors, due process, crime and disorder, and legal equality 7. Personal Autonomy and Individual Rights: freedom of movement, property rights, personal and family rights, and freedom from economic exploitation.
6 1) Electoral process and pluralism. 2) Civil liberties 3) The functioning of government 4) Political participation 5) Political culture.
previous scholars, it is used as a control measurement rather than the main outcome variable in this study. When estimating the relationship between these two measurements of
democracy, it is notable that they are similar in their grading since the correlation between them is 0.891. There are however significant differences in their scoring for some
observations, for instance Belarus in average received 15.46 by Freedom House, as compared to 33.12 by the Democracy Index. The reason for not using the measurement Polity IV, which is most commonly used in similar research, together with Freedom House, is due to the fact that the annual data on this index was not generally available at the time when the data was collected.
3.2 Aid data
Similar to previous researchers in this field, the Official development assistance (ODA) definition of foreign aid will be used. This definition is used by OECD Development Assistance Committee (DAC) and it refers to foreign assistance as government aid that promotes and specifically targets the economic development and welfare of developing countries (OECD, 2020a).
The chosen data on aid that is analyzed in this study is retrieved from the OECD database, which is regarded as the most comprehensive source on developmental aid that there is today, and its data is widely considered to be accurate and reliable (Holden, 2016). OECD provides a specific database named Creditor Reporting System (CRS) in which aid flows are
categorized and divided in a way that is suitable for the purpose of this study. CRS provides information on the aid flows given by the member donors as well as the specific purpose of every aid project. There are standardized OECD/DAC classifications of purposes on each reported aid activity in the database is assigned a so-called purpose code (OECD. Stat, 2020).
By using these standardized purpose codes, it becomes easier to analyze aid flows than if one were to compare and gather separate data from each donor. Such a method would also require that the researcher on its own interpreted the specific purpose of each project from the
different donors and then decided if these projects are comparable or not. It would also involve risks in terms of measurement errors when calculating exchange rates and deflators
based on different currencies used by the donors. At the same time, it should be stressed that there is still a risk that the interpretation of the CRS purpose codes might differentiate
between donors who for instance might report similar projects under different codes (Cornell, 2008).
3.3 Democracy aid
Democracy support does not have its own purpose code, but there are existing purpose codes that together capture projects that can be included as aid focused on the chosen definition of democracy. Democracy promoting assistance is in this research, in line with the study of Kalyvitis and Vlachaki (2012), defined as aid flows included under the purpose code 151:
Government and Civil Society. This purpose code includes aid flows addressed to various sorts of projects that can be seen as promoting democracy as the chosen way to define its meaning. Included sub-categories in this sector are for instance media and free flow of information, election support and anti-corruption (OECD. Stat, 2020). Examples of projects of democracy support programs focusing on creating well-functioning electoral authorities, training of journalists, support to non-governmental organizations, training in IT security and how to circumvent censorship.
3.4 Education aid
Aid targeting education has its own purpose code, 110: Education. There are sub-categories under this sector such as teacher training, school feeding and building of education facilities (OECD.stat, 2020). In this study all projects coded within this sector are used, including everything from aid projects targeting primary education to assistance intending to improve a country's higher education (Sida, 2020b).
3.5 Decisions to make when collecting data on aid
An issue that is faced here by scholars measuring development aid and its effect, is to decide whether or not to use a denominator to put the aid flow in relation to the size of the
population or the economy of the recipient country. It can be argued that it is no need to use a denominator, based on the assumption that aid flows are allocated by the donors in relation to the population and economy of the country receiving aid. However, when dividing each aid flow to each country every year by its population, it is notable that there is quite a difference in the received amount of aid per capita. This study will use population as a denominator based on the impact of aid relative to the size of the population. It is reasonable that a country with a large population would be affected less if that country would receive the same amount of aid as a much smaller country.
Another decision to make when studying aid flows is whether to use data on the
disbursements or the commitments. Disbursements are the actual payments to the recipient country, while commitments are donors’ intentions and can be seen as written obligations from the donor countries (OECD, 2020b). Since the commitments can be seen more as an indication about the future aid flows, the disbursements are more of relevance when analyzing the earlier effects of aid flows over time and will hence be used in this study.
3.6 Data on additional variables used in the analysis
The following three control variables are included in the models (unless otherwise stated):
GDP (Gross domestic product) per capita, trade as percentage of GDP and natural resources rents as percentage of GDP7. These indicators are included as regressors since they are assumed to have an impact on the outcome variable and they could potentially be correlated with the independent variables, they could cause omitted variable bias on the estimated impacts of aid on democracy if they are neglected.
The variable GDP per capita is included since it is likely that economic development will have a positive impact on the democratic development, for instance through the previously mentioned modernization theory by Lipset (1959). To better represent the growth rate of
7 Data on the additional variables are retrieved at World Bank's World Development Indicators database.
https://data.worldbank.org/indicator (2020-05-07).
GDP, the log of the variable is used. Trade as a percentage of GDP is included since greater integration in the world economy has been found to have a positive impact on
democratization (Lopez-Cordova and Meissner, 2005). Natural resources rent as a share of GDP is included as a regressor based on the hypothesis that natural resources could breed corruption and decrease the accountability of the political leaders due to a decreased dependence on tax income (Leite and Weidmann 2002; Sala-i-Martin and Subramanian 2003).
There are other variables that have been used by scholars as explanatory factors on
democratization, such as colonial heritage, main religion and the level of heterogeneity of the population. However, based on the assumption that these variables are more or less constant over time within the country, they are controlled for when using fixed effects.
Furthermore, in an attempt to control for the potential risk of simultaneous causality between the dependent and independent variables, meaning in this context that the amounts of aid flows could potentially be influenced by the democracy level of the recipient country, an instrumental variable approach will be applied. The population size of the recipient country will be used as an instrument for aid. This is based on the theory that countries with small populations tend to receive greater amounts of aid than larger countries. This is due to the fact that donor countries want to “put a flag” on their success in development projects
(Knack, 2004). Therefore, the size of the population is expected to have a negative impact on the aid per capita variable, which is also the case in the data set of this study. More details about the models are provided in the next chapter.
The variable gross enrollment ratio for tertiary school8 is also included in the data set but not as a control variable since that could cause post treatment bias due to its potential relationship with education aid. Instead it will be used to further investigate the impact of such assistance on the education level of the recipient country, with the purpose to analyze the influence of education aid on democracy more thoroughly.
8 The usage of gross enrollment ratio for tertiary school as an indicator is mainly due to lack of available data on other education measurements such as literacy.
3.7 Descriptive statistics
Table 1. Descriptive statistic – original data set
Variable Description Mean Std. Dev. Min. Max. N
Freedom
House Democracy
measurement (value between -4 and 100)
48.071 23.774 3 92 843
Democracy index
Democracy measurement (value
between 0-10, here multiplied by 10)
48.277 17.51784 14.9 81 845
Democracy aid
Amount of democracy aid/ population in US
dollar
3.061 3.995 0.0229 38.169 845
Education
aid Amount of education
aid/ population in US dollar
4.262 7.15 0.096 76.618 845
School
enrollment School enrollment rate for Tertiary education
in %
27.527 21.769 0.993 117.1 789
GDP/ capita
(log) Gross domestic
product divided by midyear population
in US dollar
3.346 0.432 2.224 4.192 794
Trade dependence
The sum of exports and imports of goods and services measured
as % of GDP
76.883 33.333 19.101 208.307 840
Resource
dependence Sum of oil rents, natural gas rents, coal
rents (hard and soft), mineral rents, and forest rents as % of
GDP
9.139 10.302 0.108 56.609 813
Population All residents included, regardless of legal status or citizenship.
7.28e+07 2.26e+08 469170 1.39e+09 845
3.8 Missing data and sample selection bias
A challenge that arose when conducting the empirical research of this study, was the issue of missing data. It is notable when collecting data on some of the chosen variables, particularly school enrollment, that there is a great lack of data on many developing countries. Some of the countries that were initially included were omitted since there was no data on this variable. However, countries for which there was a lack of data only for one or a few years remained included in data set. For the missing values mean values have been generated based on the values on the variables that are closest in time for that particular country. It should be stressed that there was no missing data on the aid variables, and only two missing values in the total data on the Freedom House.
The fact that there are lack of data affecting the sample selection is however not only problematic when applying the econometric models since Stata automatically will omit the observed country for the year as soon as only one variable value is missing, it might also be a source of selection bias. That could be the case for this study since there is a risk that
countries that were excluded due to missing data (for example Syria, Djibouti and Lebanon), are lacking data for reasons that are related to the political situation in these countries and thus also the democracy level. It is reasonable to assume that countries with well-functioning government institutions are more likely to collect and report data, as compared to a country such as Syria in which there has been political instability and civil war for the past decade and the scars resources are put elsewhere. This potential systematic trend in missing values could introduce sample selection bias to the estimated results.
4. Methodology
4.1 Regression models
The methods applied to examine the impacts of democracy support and education aid on democracy are different models of panel data regression analysis, in which the two aid forms are the independent variables and Freedom House is the outcome variable.
All the models are combined entity and time fixed effects regression models, further referred to as TFE-models. The first three models are similar in terms of variables except from the
inclusion of lagged time effects, with a one-year lag (t–1) in the second equation, and then a two-year lag (t–2) in the third equation.
Equation four is also a TFE-model, although this model is based on an instrumental variable approach, further referred to as an IV-model. The intention behind including instrumental variables is to filter out the endogenous variation within aid and thereby measure the exogenous variation only.
The independent variables democracy aid and education aid are studied separately and not in the same regressions, mainly due to that one of the aid variables would be automatically omitted in the IV-model since these variables are instrumented with the same instrumental variable. Hence, each regression model is applied two times, firstly on democracy aid, the results of these estimations are named with an additional “a”, and secondly on education aid, when reporting these results an additional “b” is included.
To further investigate the impacts of education aid on democracy, one additional regression is applied. Model five estimates the correlation between education aid and education in terms of school enrollment.
For robustness, the four first models will also be estimated with the alternative measurement of democracy, The Democracy Index, as the outcome variable. To test for heterogeneity, the first model will also be applied within each country group9. The results of these additional estimations are presented in the Appendix which can be found at the end of the study.
4.1.1 Model 1 - model 3
Fixed effects regression models are plausible to use when heterogeneity between the entities is expected to bias the results. In this context that is unobserved country-specific
characteristics that could be correlating with both the dependent and independent variables, and thus cause omitted variable bias on the estimations if they are not excluded. These are for instance political, cultural and historical circumstances. To be able to control for this
9 Country groups: Europe, Eurasia, America, Sub-Saharan Africa, Asia, America and MENA (Middle East and North Africa). This categorizing of countries is according to the regions provided by Freedom House.
potential bias, each country is assigned a dichotomous variable that captures all factors that vary between the countries but do not vary over time.
Yet, there might be omitted variables causing bias on the estimations that vary over time but not across countries. To control for such bias, a time fixed effect variable is included for each observed year. Such time-inconstant variation across countries could for instance be global political trends or change in regulations by the United Nations. The usage of entity and time fixed effects does not consider the possibility that there might be variables affecting the variables of interest that are varying over time and within the countries, therefore the previous mentioned control variables, within such “variation-category”, are included.
Due to the possibility that unobserved errors within each country group are not
i.i.d.(independently and identically distributed) since they might be correlated over time, all the panel data regressions have country cluster adjusted standard errors. The first model can be formally written as:
Democracy-level(𝑖,𝑡) = 𝛽1Democracy aid(𝑖,𝑡) + 𝛽2Education aid(𝑖,𝑡) + 𝛽3GDP/capita(𝑖,𝑡) + 𝛽4Trade dependence(𝑖,𝑡) +𝛽5Resource dependence(𝑖,𝑡) + αi + λt + 𝜀(𝑖,𝑡) (1)
In which i = country, t = year, 𝛽1-𝛽5 = the coefficients for the independent and control variables, αi = unobserved fixed effect that captures factors that are constant over time within the country, λt = time-fixed effect that captures factors that are constant between the countries but vary over time and 𝜀(𝑖,𝑡) = unobserved time variant and individual error term.
The same variables as equation 1 are included in model two and three, except in these regressions the independent variables are lagged, meaning that the estimations in the second model capture the influence of the two aid forms from the previous year (t–1), and in the third model, the impacts of aid from two years earlier (t–2) .The theoretical reason behind
including time lags is based on the hypothesis that democratic development is believed to be a rather slow process in which the impacts of variables that might contribute democratization might not be visible within the same year.
Further, the usage of lagged effects could also mitigate the potential issue of reversed causality, since the allocated amount of aid in year t–1 or t–2 could not be based on the outcome variable in year t. Notwithstanding, although the practice of replacing a potential simultaneously determined explanatory variable with its lagged value is commonly applied in political science and economic, there is a lack of research analyzing the way in which lagged independent variables are appropriate to use as an response to endogeneity concerns (Reed 2015; Bellmare, Masaiki and Pepinsky 2017).
4.1.2 Model 4
Entity and time fixed effects are still included in the fourth model, although this model is an instrumental variables (IV) regression, using a two stage least square method (2SLS). The purpose of this model is to mitigate the risk of simultaneous causality. As stated, it is likely that the recipient countries that reach democratic development, will receive more foreign aid in the future as compared to countries in which democratic improvements are not made. If this potential endogeneity is not considered, the estimated impact of aid could be positively biased. Thus, the aim of using an IV regression, is to isolate the variation within democracy support and education aid that are uncorrelated with the error term 𝜀, and thereby only measure the exogenous variation.
In the fourth model, an attempt to capture such exogenous variation within the independent variables, is made by filtering out the initial allocation motives from the part of aid that is related to democracy development. As an indicator for allocation motives, population size as an instrumental variable, Z. The intuition behind using population size as an instrumental variable is the fact that donors tend to allocate aid to countries with small populations and thereby be able to “claim” improvements as results of such aid. Accordingly, population size is expected to have a negative impact on aid flows.
First stage-regression:
Aid(𝑖,𝑡) = π0 + π1PopulationSize(𝑖,𝑡) + π2GDP/capita(𝑖,𝑡) + π3Trade dependence(𝑖,𝑡) +
π5Resource dependence(𝑖,𝑡) + αi + λt + ν(i) (2)
In which π0-π5 are coefficients and ν(i) is an error term.
The first stage regression (equation 2), estimates the correlation between population and the two forms of foreign aid. For population size to be considered to be a valid instrument it needs to be both relevant10 in terms of explaining variance in democracy as well as education aid, and exogenous11, meaning that population should not itself explain variation in the outcome variable.
The results of the first stage regression equation, see table 3, illustrates that the estimated coefficient for population for both aid types are statistically insignificant, in addition, both coefficients are positively signed and not negative as expected. Moreover, the F-values when predicting the correlation between population size and democracy aid as well as education, are 0.51 and 0.44, hence the instrument relevance condition is not fulfilled for neither of the aid types12. Due to this lack of correlation between the independent variables and chosen instrument, population size is considered to be a weak instrument for this data set.
Consequently, the further presented results of this model might suffer from bias and should be interpreted with caution. Other instruments for aid were also tested, such as infant mortality and colonial heritage, however none of these fulfilled the relevance condition.
There is no direct way to test whether or not an instrument is exogenous, which make the second condition a question of intuition. Although population size has been used in a similar study by Knack (2004) as an instrument for aid, one still needs to consider the risk that there might be a relationship on its own between the democracy-level and the size of the
population. However, it is reasonable to assume, based on the variation of population size of existing democracies as well as authoritarian states, that neither a large nor a small population should be favorable for democracy development. At the same time, population size may be affected by the process of demographic transition that is in turn affected by the development process,
10 Cov(z, x) ≠ 0 is the formal condition for instrument relevance, meaning that the correlation between the instrument and the independent variable should not be equal to zero.
11 Cov(z, ε) = 0 is the formal condition for the exogeneity condition, implying that the only impact that the instrument has on the dependent variable should go through the independent variable.
12 A F-value > 12 is the rule of thumb according to Stock and Watson (2020) for an instrument to be considered relevant. It should be addressed that there are other rules and indicators applied to test for instrument relevance.
and thus potentially democratization. Yet, the exogenous condition is arguably even more uncertain if using infant mortality or colonial heritage as instrument variable. Model 4 can formally be written:
Democracy-level(𝑖,𝑡) = β1𝐷𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦 𝑎𝚤𝑑0 (𝑖,𝑡)+ β2𝐸𝑑𝑢𝑐𝑎𝑡𝚤𝑜𝑛 𝑎𝚤𝑑0 (𝑖,𝑡) + 𝛽3GDP/capita(𝑖,𝑡) +
𝛽4Trade dependence(𝑖,𝑡) +𝛽5Resource dependence(𝑖,𝑡) + αi + λt + 𝜀(𝑖,𝑡) (3)
In which β1 = the coefficient for the model-estimated values of democracy aid, and β2 = the coefficient for the model estimated values of education aid.
4.1.3 Model 5
To be able to address to what extent education aid has been effective, model 5 estimates the impacts of education aid on the education level. Since the purpose of this model is not the same as in previous models, control variables are not included in these regressions. Model 5 can be formally written as:
School enrollment(𝑖,𝑡) = 𝛽1Education aid(𝑖,𝑡) + αi + λt + 𝜀(𝑖,𝑡) (4)
5. Results
Table 2. Regression results - Aid dependence and the Freedom House index
Equation 1a 1b 2a 2b 3a 3b 4a 4b
Variable
(TFE) The Freedom
House
(TFE) The Freedom
House
(TFE) The Freedom
House
(TFE) The Freedom
House
(TFE) The Freedom
House
(TFE) The Freedom
House
(TFE + IV) The Freedom
House
(TFE + IV) The Freedom
House
Democracy Aid
0.143 (0.1896)
0.2449 (0.2482)
0.334 (0.2254)
0.1205 (4.62)
Education Aid -0.0643
(0.1427)
-0.0341
(0.1149) -0.0516
(0.1064) 0.0497
(1.911) Control
variables GDP per capita
-0.0001 (0.002)
-0.0001 (0.002)
-0.0001 (0.0005)
-0.0001 (0.0005)
-0.0002 (0.0005)
-0.0002 (0.0005)
-0.0001 (0.0006)
-0.0001 (0.0007)
Trade
Dependence -0.0257
(0.0386) -0.026
(0.039) -0.0333
(0.0407) -0.0368
(0.0412) -0.0366
(0.0401) -0.0429
(0.0414) -0.0216
(0.0205) -0.0215 (0.0208) Resource
Dependence
0.189 * (0.1)
0 .1944*
(0.0988)
0.2303**
(0.1028)
0.2344**
0.102
0.2319**
(0.0951)
0.2333**
(0.0948)
0.234***
(0.0612)
0.2359***
(0.0814)
Country Fixed yes yes yes yes yes yes yes yes
Time Fixed yes yes yes yes yes yes yes yes
Time Lag (one
year) no no yes yes no no no no
Time Lag (two years)
no no no no yes yes no no
Total
observations 854 845 780 780 715 715 845 845
Number of
countries 65 65 65 65 65 65 65 65
Adjusted R-
squared 0.014 0.0122 0.0158 0.0089 0.0211 0.0077 Missing Missing Notes: cluster adjusted standard errors on country level in parentheses.
Statistical significance is denoted: *** p<0.01, ** p<0.05, * p<0.1.
The Freedom House as dependent variable.
In this section, the essential results from the regressions are described and briefly discussed.
The results from the main equations 1a–4b are presented in table 2 above, the results from all other equations are presented in the appendix in the end of this study. All the statistical estimations are produced in Stata v.15.1.
As illustrated in the first column in table 2, the estimated TFE-coefficient for democracy aid is 0.143, meaning that an increase in US$1 democracy aid per inhabitant results in an
increase of about 0.14 units of the Freedom House grading. When including a time-lag of first one year (equation 2a), and then two years (equation 3a), such predicted positive influence of democracy support increases to approximately 0.25 and 0.34 units. Yet, since none of these estimated coefficients are statistically significant13, and the reported standard errors are in average the same size as the coefficients, such predicted positive impact of democracy support should be interpreted with caution. The results from equation 1b, 2b and 3b indicates that the predicted impact of education aid is slightly negative, almost non- existing. Yet, these estimations are also statistically insignificant.
When introducing the instrumental variable (equation 4a and 4b), the estimated impact of democracy support on the Freedom House grading decreases slightly and the education aid coefficient becomes positive. If population size would have been a valid instrument for democracy aid, and equation 4a corrected for potential positive bias from the first three equations, it would be an indication of that the previous mentioned risk of simultaneous causality is an issue in this dataset. Yet, due to the statistical insignificance of these
estimations, along with the concerns of validity of the instrumental variable, also these results should be interpreted carefully.
The only estimated coefficient of the control variables that reaches statistically significance through these four equations are the coefficients for resource dependence. The predicted value of the coefficient for natural resource dependence is between 0.19 and 0.26, suggesting
13As illustrated in the tables of regression results, the coefficients are presented with a *(***) depending on its significance level. When further referring to statistical significance/ insignificance, it will be based on whether or not the estimates are significant at the 5% level (**).
that one more percent of natural resources as a share of GDP, increases the Freedom House grade by approximately 0.19–0.26 units, which is rather interesting since when including resource dependence as a control variable, its expected impact on democratization was negative.
The results of equation 5 (see table 4 in the Appendix), imply that education has small negative influence on the school enrollment rate, the predicted coefficient is - 0.098, which suggests that when increasing the amount of education aid by US$1 per inhabitant, the rate of school enrollment decreases by approximately 0.1 percent. Yet, the standard error is
approximately the same as the coefficient, and these calculations are not statistically significant.
There are a mainly two interesting results from the equations that were included for
robustness. Firstly, when using the Democracy Index as the outcome variable (see table 4 in the Appendix), the coefficients for democracy aid in equation 6a, 7a and 8a are greater as compared to when using the Freedom House grading. In addition, the estimated coefficients for democracy aid in equation 7a and 8a are statistically significant.
Secondly, when testing for heterogeneity and applying the first TFE-model separately for each geographical region (illustrated in table 5 in the Appendix), some of the coefficients for democracy aid and education aid reaches statistical significance. The estimated influence appears to be both positive and negative, varying between the regions. Although these results imply that impact of these two aid types is correlated with geographical position of the recipient country, it needs to stressed that the number of countries in these regionals groups included in this data set it small and hence not representative for the total amount of countries in these regions. Further, due to the small number of countries within each region, the sample population for the equations when tested separately are insufficient for providing reliable estimations.
