Grants versus loans: the
impact on democracy
Master’s Thesis Economics
Specialisation: International Economics and Globalisation Marijn Knieriem
10018735
marijnknieriem@hotmail.com 30 June 2015
Supervisor: Dr. Maja Micevska Scharf Second reader: Dr. Dirk Veestraeten
16221 words
Abstract
This paper investigates whether levels of democracy in recipient countries show different responses to loans than they exhibit to grants. I discuss three channels that predict that loans lead to higher levels of democracy than grants; I also discuss three channels that predict the opposite effect. In a cross-country analysis I find the quite robust result that there is a statistically significant positive relationship between grants and democratization when aid is expressed as a percentage of the recipient’s GDP, and democracy is measured by the Freedom House index. Using these measures, no statistically significant relationship is found between loans and changes in levels of democracy. These results, however, do not hold when aid is expressed in per capita terms, or when democracy is measured by the Polity IV index.
Statement of Originality
This document is written by Marijn Knieriem who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
1. Introduction p. 4
2. Loans versus grants and their impacts on the levels of democracy in recipient p. 6 countries
2.1 What is the difference between grants and loans? p. 6
2.2 The effects of grants and loans on democracy levels p. 8
2.3 The empirical evidence regarding the effect of foreign aid on democracy p. 10
3. Empirics p. 14 3.1 Expectations p. 14 3.2 Data p. 14 3.3 Econometric methodology p. 21 3.4 Results p. 23 3.5 Robustness checks p. 27 3.6 Limitations p. 34
4. Conclusion and discussion p. 36
1. Introduction
How could the funds that are available for development aid be used most efficiently? Several factors that influence the effectiveness of foreign aid are known nowadays. One of those factors is the form of government: Svensson (1999) found that aid is more favourable to economic growth in countries with a democracy. This might be explained by Olson’s (1993) argument that democracy and economic growth require the same set of institutions. Democracy needs the security of individual rights, such as the right to free speech and association, in order to be viable. The same respect for individual rights (in this case contract and property rights) is necessary for economic growth, since it prevents that the political leader could extract the economic surplus from his people. Moreover, these institutions increase the stability of a country, which could stimulate people to invest in long-term projects. Santiso (2001) furthermore argues that democracy is the political system that is best suited to implement sound economic policies that could increase the effectiveness of development aid. Now, if aid is more effective in democratic countries, and if one wants to know how aid could be disbursed most efficiently, then it would be interesting to look at the issue whether foreign aid affects the level of democracy in the recipient country. This would tell something about whether aid influences the conditions for its own effectiveness.
The effect of aid in general on the levels of democracy in recipient countries has already been investigated. The empirical results are mixed; e.g. Knack (2004) finds no effect of foreign aid on democracy levels, Kersting & Kilby (2014) find that development aid leads to higher levels of democracy, and Djankov et al. (2008) get the result that aid leads to lower levels of democracy in recipient countries. These and other findings will be discussed in more detail later on in the second section. In this paper the consequence of foreign aid for the recipients’
democracy levels will be discussed in relation to another debate in the literature concerning the effectiveness of foreign aid. This is the discussion on loans versus grants. While the literature regarding this topic is quite extensive, there has not yet been a study that investigates whether loans and grants have differing effects on the levels of democracy in recipient countries.
This paper tries to fill this gap in the literature. It will try to answer the following question: do levels of democracy in recipient countries show different responses to loans than they exhibit to grants?
This study will be a cross-country analysis. The idea of Knack (2004) that
democratization is a long-term process will be followed, and hence the goal of this paper is not to analyze changes in levels of democracy that occur from year to year. The period under investigation is from 1991 to 2013. The rationale to start in 1991 is given by the fact that the Cold War ended in 1990, and the effects of aid might be different after the Cold War than they
were during it. This study is concerned with the effects of current aid, and hence it seems reasonable to start the analysis in 1991.
Total Official Development Aid (ODA) is the flow of aid that will be analyzed. The data for this variable comes from the Organisation for Economic Co-operation and Development (OECD). By focussing on total ODA I will follow the bulk of the empirical literature on loans versus grants (e.g. Iimi & Ojima, 2008; Gupta et al., 2003; Djankov et al., 2006; Odedokun, 2003). I will express the total ODA flow as a percentage of the recipient’s Gross Domestic Product (GDP), in order to normalize aid for the size of the country. Moreover, I will divide the flow of ODA by the
population of the recipient country, in order to check the robustness of the results when ODA is normalized for the size of the recipient country in a different way.
In the literature, two proxies for the level of democracy are most often used: the Freedom House index and the Polity IV index. As Knack (2004) describes, the former is a measure of the rights of the country’s population, while the latter measures the institutional environment of a country, focusing on the openness and competitiveness of the election process. Democracy, understood as the ability of the population to influence the political process of a country, seems to be captured quite well by these two measures. In this paper, there will be mainly focused on the Freedom House index, and the Polity IV index will be used as an alternative to check the robustness of the empirical results.
This paper is structured as follows. First, I will discuss loans versus grants in the context of their respective effects on the levels of democracy in recipient countries. Then, I will turn to the empirical part of this essay: as a start, the data used and the method that is employed will be discussed in detail. After this I will present the results of the empirical analysis, before I will turn to some robustness checks. There will be ended with a conclusion and a discussion.
2. Loans versus grants and their impacts on the levels of democracy in recipient countries
2.1 What is the difference between grants and loans?
The distinction between grants and loans is a simple one: grants do not need to be repaid, whereas loans do. When loans are a part of ODA, these repayments are concessional: the present value of the future repayments is smaller than the value of the future repayments of a loan at market terms. The grant element in concessional loans is made up of three elements. These are the grace period, the longer maturity and lower interest rates.
Nevertheless, several scholars argue that this difference between grants and loans is not of much relevance for the recipient (Schmidt, 1964; Lerrick & Meltzer, 2002). This argument could be illustrated by a numerical example (see e.g. Klein & Harford, 2005). Let us assume perfect capital markets, in which everyone could borrow and lend at a real interest rate of 100%. The repayments are due in the next year. We will furthermore assume that concessional loans, on the other hand, require real interest rates of only 50%. Now, a concessional loan of 10 million dollars would in this case imply a cash outflow of 15 million dollars in the next year. This loan could be turned into a grant of 2.5 million dollars: the recipient could lend the remaining 7.5 million on the capital market, which yields 15 million dollar in the next year. This would equal the repayment that the recipient owes to the donor country. Similarly, a recipient could transform a 2.5 million dollar grant into a 10 million dollar concessional loan: the recipient could borrow 7.5 million dollar on top of the 2.5 million dollar grant on market terms of 100%, which would also yield a repayment obligation of 15 million in the next year. Thus, if the recipient has good access to commercial capital, a concessional loan would be equivalent to a combination of a market loan and a grant. Hence, if the recipient could borrow extra money on reasonable terms on the capital market, ODA grants and ODA loans are equivalent.
Be that as it may, the assumption of perfect capital markets is a problematic one.
Developing countries often have little access to capital, other than to the capital that is provided by development organizations. The regular market agents often are reluctant to lend money to these countries, since poor countries are perceived as too risky: Moody’s rates the majority of the bonds issued by low and middle-income countries at ‘Ba’ or lower (Iimi & Ojima, 2008).
The phenomenon that capital does not flow to poor countries, even though the expected return on capital in these countries is high, is called the ‘Lucas paradox’ (Cohen, Jacquet & Reisen 2005). The expectation of a high return on capital follows from the neoclassical assumption of decreasing returns to capital, which predicts that capital scarce countries have profitable investment opportunities that have not been reaped yet. Lucas (1990) provides four
that all come down to the same intuition: the returns on capital do not solely depend on the project in which the money is invested. They also depend on the conditions surrounding this project, like the human capital in a country and the presence of a good infrastructure. In developing countries, these conditions are often inferior. Moreover, since the investments in these conditions are not directly profitable, but only indirectly and after a very long time through higher taxes, it is hard to collect private capital for these investments. Furthermore, these conditions often have the nature of public goods, such that economic theory predicts that they will be subject to underinvestment in the absence of a well-functioning government. Hence, according to Cohen, Jacquet and Reisen (2005), from a social welfare perspective there is a rationale to invest in these conditions, even though the capital for these investments is not available on the private capital markets. And, since these investments will eventually yield returns, loans are better suited to finance these investments than grants are. Concessional loans, with their grace period, longer maturity and lower interest rates, are especially well fitting for these investments. As a consequence, there is a justification to provide ODA in the form of loans, a justification that would be lacking if grants and loans were equivalent. Because of the failure of the market to provide funds to invest in these public goods, i.e. because developing countries do not have perfect access to capital markets, ODA loans remain an important instrument to finance certain kinds of projects.
Consequently, we can conclude that grants and loans are different from the perspective of the recipient. Grants are a one-way flow of money from the donor to the recipient. Loans are a two-way flow: first the aid flows from the donor to the recipient, and later on the repayments flow from the recipient to the donor. Hence, for an equal grant element, the amount of aid that the recipient gets at the moment of disbursement is larger when the aid is given in the form of a loan. This difference will be offset later on by the repayment obligation that the loan incurs. So, the choice between grants and loans is, from the perspective of the recipient, basically a trade-off between current and future consumption: loans provide a higher current consumption than grants; however, this higher current consumption of loans comes at the cost of future
consumption, whereas grants are free in terms of future consumption.
Because of this difference, grants and loans could lead to different incentives for the people in the recipient countries. As a consequence, it is possible to identify several channels through which the forms of aid may have different consequences for the levels of democracy in the recipient countries. I will discuss three channels that predict that loans lead to higher levels of democracy than grants; I will also discuss three channels that forecast the opposite effect.
2.2 The effects of grants and loans on democracy levels
2.2.1 How loans could lead to higher levels of democracy in the recipient country than grants Because ODA loans have to be repaid, they impose a burden on future generations, or on the future consumption of the current generation. That is to say that loans are not free. Because of the repayment obligation, loans from the International Development Association (IDA) are only disbursed after the recipient country has guaranteed the transparency of its borrowing activities by informing the public, e.g. via parliamentary processes (International Development
Association, 2001). If there are hard budget constraints, such that the government is not
authorized to agree on new debt without popular approval, i.e. without a referendum, this public transparency obligation holds for all kinds of ODA loans (Bräutigam, 2000). Hence, the
repayment obligation of loans assures that the government has to be transparent regarding the financing of its expenditures. In the case of grants, the resources are free, and hence there is less of a necessity to inform the public about the receipts of this kind of aid, since it has no downside at later points in time. Thus, via the transparency that is required for the acceptance of loans by the recipient country, it is possible that loans lead to higher levels of democracy than grants do.
Moreover, several scholars (e.g. Gupta et al., 2003; Odedokun, 2003) argue that grants replace tax revenues, while loans increase the receipts of taxes. This argument is based on the idea that the future repayments, that are the consequence of loans, provide an incentive for the government to increase tax collection efforts. When aid is given in the form of grants, such an incentive is lacking, and the windfall in free revenues might even reduce the efforts by the government to collect taxes. Thus, when the aid is disbursed as a grant, this implies that the recipient becomes more dependent on the inflow of aid. As a consequence, governments become more accountable to their donors, but less to the taxpayers (Heller & Gupta, 2002). Those in power may feel less necessity to keep their people informed about how they spend the
government revenues, since the tax revenues are a less important part of government revenues. The political process may lose its transparency, and hence it becomes harder for the people to influence the political outcome. Therefore, grants may cause the level of democracy to decrease.
A third channel that would predict that loans lead to higher levels of democracy than grants is based on the idea of the fungibility of aid. This is especially problematic when government expenditures are not very transparent; as we have seen, this is in particular a problem associated with grants, since these do not have to go through the same process that ensures transparency as loans. So in the case of grants, the leaders face fewer obstacles to spend money on goals for which it was not intended. When this happens, problems concerning
patronage might increase (Bräutigam, 2000). When there is less need for the government to justify its expenditures to the public or to the parliament, it becomes easier for the leaders to
channel money to politically important people or regions in order to get political support. As a consequence, some voices that could object to the policy of the leader are silenced, and the leader has more room to continue his policy without the checks and balances that are necessary for a well-functioning democracy. Therefore, the increased patronage as a consequence of grants may lead to lower levels of democracy.
2.2.2 How grants could lead to higher levels of democracy in the recipient country than loans Grants do not require repayment. This means that they can be used for a larger variety of projects than loans (Mourmouras & Mayer, 2004). Indeed, if there is no repayment obligation, the amount of aid received could be used for projects that do not generate monetary returns, e.g. investments in sanitation. Loans are not suitable for these kinds of projects, because the lack of monetary returns makes that the credit cannot be repaid. Now, let us assume that there is a part of the population that is only interested in investing in sanitation. In the case of loans, they would not be interested in having a say about the destination of the money, since this form of aid is not suitable for the only investment that they care about. However, in the case of grants, they do want to influence where the money goes to, since it could be used to finance the sanitation. So in the case of grants, there is a potentially larger share of the population that wants to influence the destination of the funds. We could interpret this as a larger demand for having influence in the political outcome, which could increase the efforts to achieve this goal. Through this channel, grants may increase the recipient’s level of democracy.
As a second channel, let us consider an autocratic regime that receives loans. Furthermore, we make the assumption that the ruler has a high discount factor, i.e. he has a short time-horizon. This could happen for example when the one in power expects that he will not be able to hold his position for a long time. Olson (1993) presents a theory that assumes that the incumbent wants to maximize his income during the time that he holds office. This income consists of two parts: the tax revenues, and the confiscation of those assets of which the total value is higher than the tax they generate during the period that the ruler is in office. Let us consider an asset that requires large initial investments, which are only compensated by the returns after a long time. Furthermore, we assume that the ruler cannot confiscate this asset (e.g. in the case of a road). Now, since loans could be disbursed at higher levels, they are better suited than grants to finance projects that require large initial investments (Iimi & Ojima, 2008). If these projects are the best choice from a social welfare perspective, the people in the recipient country would want that the loans are used for financing these projects. The ruler, however, would want to prevent this, because his high discount factor implies that he expects to be unable to reap the returns of these investments. As a consequence, the ruler has an incentive to make
sure that his people cannot influence the destination of the money, i.e. that they cannot influence the political process. Via this channel, loans could prevent the development of democracy.
Because loans require repayments, the disbursement of loans implies by necessity the emergence of a long-term donor-recipient relationship (Iimi & Ojima, 2008). However, it is argued that donors do not have any enforcement mechanisms to make sure that the repayment obligation is actually met by the recipient (Bulow & Rogoff, 2005). This could lead to moral hazard on the side of the recipient. Following this line of reasoning, it could be argued that, from the perspective of the recipient, loans are perceived as simply higher levels of aid than grants. Svensson (2000) presents a game-theoretical model in which higher levels of aid lead to an increase in rent-seeking behaviour. As a consequence, less money remains for the provision of public goods, such that a higher level of aid implies a lower expenditure on public goods. Even though Svensson does not define it as follows, one could see that examples of public goods include property and contract rights, and protection to ensure that free speech is possible. Indeed, these rights are non-excludable and non-rivalrous. Moreover, they are necessary for a lasting democracy: without them, an autocrat could threaten his adversaries with confiscation of their property, such that they are deterred from speaking freely (Olson, 1993). Thus, if the recipient regards loans as simply higher levels of aid than grants, then loans might have the consequence that less public goods become available: this is also true for public goods that are fundamental for a lasting democracy. Hence, through this channel loans could decrease the levels of democracy in recipient countries.
2.2.3 Recapitulation
Now, let us summarize the above. I have discussed three channels through which loans could lead to higher levels of democracy in the recipient country than grants do. I also discussed three channels through which grants are expected to lead to higher levels of democracy. Therefore, it is not a priori clear which form of aid leads to higher levels of democracy. Hence, it is necessary to look at the empirics. However, before we turn to the empirical part of this paper, I will discuss some empirical research concerning the effect of foreign aid on democracy levels in recipient countries. This would provide the necessary background information on the issues related to the empirical studies regarding democracy changes.
2.3 The empirical evidence regarding the effect of foreign aid on democracy
Basically, one could divide the research concerning the relationship between foreign aid and democracy into two groups. The first type of studies performs a cross-country analysis. One of the most influential studies concerned with the relationship between foreign aid and democracy
is such a cross-section study. This research is performed by Knack (2004). Knack studies the period 1975-2000, and investigates whether the levels of democracy in recipient countries have changed as a consequence of the inflow of foreign aid. He controlled for reversed causality, because levels of democracy might influence how much aid donors are willing to give to development countries (this problem of reversed causality will be discussed in more detail in section 3.3). Knack did this by performing an Instrumental Variable (IV) regression, using the infant mortality rate of the recipient country, its number of inhabitants and a set of colonial heritage dummies to predict levels of aid. However, Knack does not report first stage F-statistics, and therefore it is hard to evaluate whether he succeeds in solving the problem of endogeneity. Nevertheless, using several different model specifications, Knack finds no evidence that aid has a significant effect on democracy levels. Goldsmith’s (2001a) cross-sectional analysis however, using a sample of sub-Saharan African countries over the 1980s and the 1990s, found a positive link between development aid and democracy. Goldsmith dealt with the endogeneity problem by employing life expectancy, literacy rate and a dummy for former French colonies as
instruments. Goldsmith also does not report first stage F-statistics. Heckelman (2010)
investigated whether aid affected democracy in the transition economies of Eastern Europe and the former Soviet-Union over the period 1997-2007. Heckelman found that aid per capita is a significant determinant of democratization in the recipient country, while aid as a percentage of Gross National Income (GNI) is not. Heckelman controlled for endogeneity by splitting up the period under investigation; he then regressed the amount of aid received in the early period (1997-2001) on the changes in democracy levels over the whole period (1997-2007). Because this early aid was not given as a reaction to changes in democracy levels over the whole period (since these changes, or at least a part of them, occur later), Heckelman argues that this
specification controls for endogeneity. Finally, in the most recent cross-country study, Kersting and Kilby (2014) performed an analysis quite similar to the one completed by Knack (2004), and found a statistically significant positive effect of foreign aid on democracy levels in recipient countries. An explanation for this different result might be found in the fact that Kersting and Kilby used data up to 2011, such that they have eleven more years of post-Cold War data; the end of the Cold War might have had an effect on the credibility of the donor to enforce its conditionality. Kersting and Kilby controlled for endogeneity in two ways. First, they followed Knack’s (2004) approach by performing an IV regression and, using the same set of instruments, found a first stage F-statistic of 9.69. Second, they used Heckelman’s (2010) strategy by splitting up the period under investigation. These strategies led to similar results.
The second type of research performed to investigate the relationship between foreign aid and democracy employs panel data. The first study performed in this respect is from Goldsmith (2001b). This study investigated whether the levels of democracy in sub-Saharan
African countries were affected by development aid. A small positive relationship between aid and democracy was found. This is in line with the cross-country analysis of this scholar that was discussed earlier. He used a generalized instrumental variable method with per capita income, a dummy for former French colonies and population size as instruments; again no first stage F-statistics were presented. Using the data from Goldsmith (2001b), Dunning (2004) investigated whether the effect of foreign aid on democracy levels depends on the strategic interests of donors. According to Dunning, the small positive effect that Goldsmith found should be entirely attributed to the effect of aid on democracy in the post-Cold War period, thus when the
geopolitical interests of donors were small. Before the end of the Cold War, Dunning finds no significant effect of aid on democracy. Hence, the global political constellation seems to influence the effect of foreign aid. Dunning controlled for endogeneity in the same way as Goldsmith (2001b). In contrast to these studies, that both found a positive relationship between aid and democracy, Djankov et al. (2008) found a statistically significant negative relationship between foreign aid and democratization. Employing panel data of 108 recipient countries during the period 1960-1999, they found that foreign aid reduces democracy levels. This result also holds when they investigate the period from 1990 onwards. Thus, this contrasts with the findings of both of the panel analyses reported above. This discrepancy might be explained by the fact that the study of Djankov et al. has a different regional scope than the studies of Goldsmith (2001b) and Dunning (2004). Djankov et al. (2008) controlled for reversed causality by employing different combinations of instruments; all these combinations yield first stage F-statistics that exceed the value of 10.
Besides the discussion on what the effect of foreign aid on democracy is, several authors argue that one should not speak of the effect of foreign aid on democracy. Dutta et al. (2013) advocate that the influence of aid on democratization depends on the initial governmental orientation of the recipient. They contend that aid only amplifies the existing political
orientation of the recipient nation, leaving autocracies more autocratic, and democracies more democratic. The reason for this effect is that the leaders in autocratic countries could use aid to support their own position; on the other hand, the leaders in democratic countries face more checks and balances and hence should spend the money more in line with what the people demand. This should give the people more resources to control the government in the future. Dutta et al. find empirical support for their hypothesis of the amplification effect. Related to the amplification effect is the idea of Morrison (2009) that foreign aid stabilizes both democracies and autocracies. His argument is as follows: the general population is a threat to the stability of an autocracy, while the elite is a threat to the stability of a democracy. The inflow of foreign aid makes that the autocrat could buy off (a part of) the population in order to keep them calm; in a democracy, the windfall in revenues makes that there is less of a necessity to increase taxes,
such that the elite does not have much of an incentive to seize power. Hence, the inflow of foreign aid should stabilize both democracies and autocracies. Morrison presents empirical support for this hypothesis. Wright (2009) also argues that the effect of aid on democracy levels depends on a third factor. According to this author, whether an autocratic leader will use the foreign aid for democratic reforms depends on his expectations about whether he will be elected in free elections. If an autocrat expects to be elected, he has an incentive to democratize, since this will yield higher future levels of aid in Wright’s theoretical setup. If an autocrat expects that the people will choose another leader, he does not want to reform, since he then cannot control the higher level of foreign aid that flows into the country. Wright also presents empirical support for this hypothesis.
As the above discussion concerning the relationship between foreign aid and democracy should clarify, there is little consensus in the literature on the results in this field. The same holds for the views regarding the methodology that is best suited for this task. Some authors prefer an analysis employing panel data, while others choose a cross-sectional analysis. Wright (2009) is the most severe advocate for a panel analysis, and at the same time the harshest critic of the cross-sectional method that was employed by Knack (2004), Goldsmith (2001a),
Heckelman (2010) and Kersting and Kilby (2014). According to Wright, the cross-country methodology that Knack and the others employ suffers from several shortcomings. Knack only considers the changes in the democracy indices over the whole period of 1975-2000. Wright argues that this approach abstracts from changes within this period. Hence, countries that had the same democracy score in 1975 as in 2000 are treated by Knack like they have not
experienced any changes in their democracy levels, even though these levels might have fluctuated heavily within this period. Despite this criticism, a cross-sectional study seems best suited to capture the persistent effects of foreign aid on the emergence of democracies, i.e. it is the most appropriate setup to seize the long run relationship between aid and democracy (Kersting & Kilby, 2014).
A second argument of Wright (2009) against Knack’s (2004) cross-section analysis, is that the results are very sensitive to the choice of begin and end years of the period under investigation. He reminds the reader of the example of Thailand: this country enjoyed a year of democracy in 1975, in between two years (1974 and 1976) of autocracy. Now, if the year 1974 would have been chosen as the begin year of the period under investigation, the results that Knack found might have been very different. This is a convincing argument, even though the effects of the choices of begin and end years might be very small in a large sample, such that the results are not significantly altered as a consequence of these decisions. Nevertheless, there should be dealt with this problem. The strategy to handle this issue will be discussed in the methodological section of this paper.
3. Empirics
3.1 Expectations
As has already been discussed in section 2.2.2, there are several ways through which loans are expected to lead to higher levels of democracy than grants. On the other hand, there are also several channels that predict that grants lead to higher levels of democracy than loans. Three channels were discussed to support both of these claims. Precisely because of this reason, the research question has to be answered empirically. Thus, there is no clear-cut prediction about what the results of the empirical analysis will be.
However, as a consequence of the causal channels that have been discussed, it is plausible that the different types of aid have different effects on democracy levels in recipient countries. The only two situations in which it is not expected to find a difference are if the effects of these causal mechanisms are too small, or if they are approximately of an equal size. Hence, if the effects of the discussed causal channels are large enough and not of an equal size as each other, then it is expected that it will be found that loans and grants have different consequences for the levels of democracy in recipient countries.
The next section starts with a description of the data that is employed. I also present some descriptive statistics.
3.2 Data
The period under investigation is from 1991 until 2013. The rationale to start in 1991 is given by the fact that the Cold War ended in 1990. During the Cold War, donors used foreign aid to strengthen the position of their allies. After the Cold War ended, the donor countries had more freedom to stop the flows of foreign aid, since this would not imply anymore that their allies were vulnerable to influence by the enemy. This gave the donor countries some leverage, and made it possible for them to enforce economic reforms that could induce economic growth (Bearce & Tirone, 2010). Moreover, Dunning (2004) used a sample of sub-Saharan African countries to investigate whether the effect of foreign aid on democracy was different during the Cold War than after this period. He found that in the post-Cold War era, foreign aid had a positive significant effect on democracy levels in the sub-Saharan African countries, while there was no effect during the Cold War. Therefore, after the end of the Cold War, the effect of foreign aid might have been different than the effect during this period. This consideration led Wright and Winters (2010, p. 65) to pose the rhetorical question: “is it fair to judge the current state of foreign aid by looking at historical data, when we know that aid was distributed and used under
Table 1: Summary statistics
Variable N Mean Standard
deviation Min Max
Change in FH index 1991-2013 110 -.41 1.30 -4.5 4
FH index 1991 110 3.97 1.76 1 7
Change in Polity IV index 1991-2013 72 3.22 5.57 -13 15
Polity IV index 1991 72 1.25 6.65 -10 10
Net total ODA (% of GDP, 1991-2013) 110 8.89 10.57 .03 59.28
Net loans (% of GDP, 1991-2013) 110 .86 1.12 -1.01 4.78
Grants (% of GDP, 1991-2013) 110 8.03 10.18 .03 57.24
Ln GDP per capita 1991 110 6.71 .93 4.96 8.79
GDP per capita growth 1991-2013 110 2.24 2.39 -2.20 19.27
conditions very different from those we face today?” For this reason, it seems reasonable to use the data from 1991 onwards.
The most often employed measure in empirical research concerned with the
determinants of democracy is the Freedom House index. This index is published since the 1950s. The Freedom House index consists of two rating systems: the political rights rating and the civil liberties rating. The political rights rating is determined based on a set of ten questions: three related to the electoral process, four questions that measure political pluralism and participation and three questions are concerned with the functioning of the government. The civil liberties rating is based on a set of fifteen questions. These relate to freedom of expression and belief, associational and organizational rights, the rule of law, and personal autonomy and individual rights (Freedom House, 2015).
For each of these questions, experts assign points to the individual countries (Freedom House, 2015). While there is a subjective element in this process, this subjectivity is kept to a minimum because the assigned ratings have to be defended in front of a panel of experts in the field. This should ensure the reliability of them. These scores eventually determine the political rights and civil liberties ratings. Both range from 1 to 7, with 1 representing the greatest degree of freedom, while 7 stands for the smallest degree of freedom. The average of a country’s
political rights rating and civil liberties rating is the Freedom Rating. This is the rating of interest in this essay. We are interested in the change in the Freedom Rating as a consequence of foreign aid. Since lower numbers denote higher levels of freedom, a negative change in the Freedom Rating means that a country has become more democratic, whereas a positive change denotes that a country has become more autocratic. Some summary statistics for the Freedom House index are reported in table 1. Figure 1 shows a histogram with the changes in the Freedom House index.
Figure 1: Changes in the Freedom House index
The other variable measuring the change in democracy is the Polity IV index. This variable is used to check the robustness of the results that are found using the Freedom House index. This is common practice in the literature concerned with the relationship between foreign aid and democracy (e.g. Knack, 2004; Kersting & Kilby, 2014). I will look at the measure ‘Polity’ in this dataset. This measure could vary from +10 (denoting a full democracy) to -10 (denoting a full autocracy). Hence, higher scores in the change in the Polity variable denote countries that have become more democratic, while lower scores denote countries that have become more autocratic. Polity is constructed as a combination of two other measures. The first is the measure ‘institutional democracy’. This measure is concerned with the role of institutions in achieving three goals: the effective expression of citizens’ preferences regarding alternative policies and leaders, constraints on the government’s exercise of power, and the guarantee to civil liberties for all citizens (Marshall et al., 2013). The second measure to influence the Polity score is the measure for ‘institutional autocracy’. The autocracy scale is constructed by looking at the competitiveness and regulation of political participation, constraints on the political leader and the openness and competitiveness of executive contracting (Marshall et al., 2013). The eventual Polity score is constructed by subtracting the autocracy score from the democracy score. Some summary statistics for the Polity IV index are reported in table 1. Figure 2 shows a histogram with changes in the Polity IV index.
0 5 10 15 20 25 F re q u e n cy -4 -2 0 2 4 Change in FH index 1991-2013
Figure 2: Changes in the Polity IV index
The final thing to say about the different measures for democracy is that they do not contain the same information. This is what one would expect, since they are constructed based on related, but not identical criteria. This comes clear from their correlation coefficient, which has the value of -0.5852. Figure 3 shows a plot with the changes in the Freedom House index and the changes in the Polity IV index.
Figure 3: Plot of the changes in the Freedom House index on the changes in the Polity IV index
The data for development aid comes from the database of the OECD. This data consists of each transaction that has a concessional nature and a grant element of at least 25 percent
(calculated at a discount rate of 10 percent). This ODA data includes both bilateral and
0 5 10 15 F re q u e n cy -20 -10 0 10 20
Change in Polity IV index 1991-2013
-4 -2 0 2 4 C h a n g e i n F H i n d e x 1 9 9 1 -2 0 1 3 -20 -10 0 10 20
Change in Polity IV index 1991-2013
multilateral aid (OECD, 2015). Net total ODA (the total inflow of ODA minus the repayments of the principal on loans of prior years), total grants and net loans (the inflow of loans minus the repayments of the principal on loans of prior years) are the flows of aid that will be looked at. Since the data from the OECD database is expressed in dollars, it does not tell us directly how large the influence of the development aid is. Therefore, two measures that normalize aid for the size of the country had to be constructed. The first is aid as a percentage of the recipient’s GDP. The data on GDP comes from the World Bank’s World Development Indicators. The second measure for aid is constructed by dividing the amount of aid received by the population of the recipient country. This measures the aid per capita for each recipient country. Population data also comes from the World Bank’s World Development Indicators. Two different measures for aid are used in order to check the robustness of the results. E.g. Heckelman (2010) did find a
statistically significant effect of aid on democracy changes when aid is measured in per capita terms, while he found statistically insignificant results when aid as a percentage of GNI was employed. Some summary statistics for aid as a percentage of the recipient’s GDP averaged over the period 1991-2013 are reported in table 1.
As control variables I use, following Knack (2004), the log of income per capita in 1991, and the average per capita income growth rate over the period 1991-2013. The data for both variables comes from the World Bank’s World Development Indicators. Some summary statistics for logged initial GDP per capita and average GDP per capita growth are reported in table 1. Knack (2004) moreover uses illiteracy as a control variable. However, nowadays the data on illiteracy is scarce; Kersting and Kilby (2014) suggest that this is a consequence of improved quality control standards. Kersting and Kilby therefore simply leave out illiteracy as a control variable. However, the downside of this is that they only control for economic development, but not for human capital. Nevertheless, the latter is probably also a determinant of democracy (Lipset, 1959). Therefore, an alternative measure for human capital will be used. This variable is the net enrolment rate in primary education. I will look at the 1991 levels of this variable. The data again comes from the World Bank. Among other things, primary education provides children with reading and writing skills. Hence, the conceptual link between the illiteracy measure employed by Knack (2004) and the education measure that is used in this essay should be clear. Moreover, it provides a more comprehensive set of control variables than Kersting and Kilby (2014) have. Nonetheless, just like for the illiteracy rates, there is a lot of data missing for this variable, such that the sample size is a lot smaller when education levels are included in the regression. Therefore, they will only be used as a robustness check. The regional dummies form a third collection of control variables. These dummies were created based on the grouping that is used in the OECD data that was employed for the numbers on ODA.
The robustness of the results is checked with additional control variables. I will follow Knack (2004) by including the share of the population that lives in urban areas. The data for this variable comes from the World Bank’s World Development Indicators. Moreover, the change of this variable in the period 1991-2013 divided by its 1991 level will be included as an
independent variable. Another control variable that will be used is the percentage of the population that is Muslim, since religious orientations of countries are also linked to democratization. I will follow Kersting and Kilby (2014) by including only one variable concerned with religion. Kersting and Kilby controlled for the share of the population that is Christian. They justify this choice by pointing towards the high negative correlation that exists between the two major religious traditions, namely Christianity and Islam. However, because the only statistically significant religion variable in Knack (2004) is the percentage of Muslims in a country, I chose to use the percentage of Muslims as a control variable. The data on the
percentage of Muslims is taken from Maoz and Henderson (2013). Since this dataset only contains data for every five years, the percentage of Muslims in 1990 was used.
Moreover, I will use additional control variables that were not employed by Knack (2004), in order to add more robustness to the results. The first additional control variable will be the average oil rents over the period 1991-2013 as a percentage of GNI. The data for this variable comes from the World Bank. By including oil rents in the regression, I follow Djankov et al. (2008). They argue that natural resources may have the same adverse consequences on institutions as foreign aid may have. The reason for this similarity is that both are a windfall in revenues, and may lead to the same kind of rent-seeking behaviour. Hence, both kinds of revenues may provide an incentive for corruption, and a disincentive for democratic legitimacy. The inclusion of the variable for oil rents controls for the effect of natural resources on changes in the levels of democracy. The second additional control variable is ethnic fractionalization. Kersting and Kilby (2014), Heckelman (2010) and Goldsmith (2001a) all use some measure for ethnic heterogeneity. As Goldsmith writes, ethnic heterogeneity is often thought to impair processes of democratization. In countries with a relatively homogeneous population, the public choices are closer to the choices of the average individual than in more heterogeneous countries (Alesina & Spolaore, 1997). So, if the preferences of different people are closer to each other, the result of the majority vote will be more likely to satisfy every individual. Hence, there is more to gain from democracy for the people in a homogeneous country than for the people in a more heterogeneous country. As a consequence, ethnic heterogeneity might have adverse effects on democratization. Data for this ethnic fractionalization variable is drawn from Alesina et al. (2003). The measure from Alesina et al. expresses the chance that two randomly drawn individuals from a population belong to different groups. Hence, a higher score in the fractionalization data denotes a higher level of ethnic heterogeneity.
Finally, the data employed for the IV regressions should be discussed (the problem of endogeneity will be discussed in more detail in the next section). I will use five measures to predict the exogenous part of foreign aid that a country receives. The first three instruments are identical to those used by Knack (2004) and Kersting and Kilby (2014). These are infant
mortality, population of the recipient country and dummies that denote colonial heritages. Infant mortality is a measure for the recipient’s need: countries with higher levels of infant mortality are in more urgent need for aid than countries with lower levels. The data for this variable, the number of infants who died before they have reached the age of one year per 1000 live births, is taken from the World Bank’s World Development Indicators. The population of the recipient country and the set of colonial heritage dummies are both measures for the interest of the donor. The data concerning the population is taken from the World Bank. The colonial heritage dummies were constructed, based on the data from Hensel (2014). The three measures discussed thus far all relate to development aid in general, and are at first sight not particularly well equipped to distinguish between grants and loans. Hence, additional instruments have to be employed for this task. The first one is debt. More specifically, the external debt stocks as a percentage of GNI is used as an instrument. The data for this variable comes from the World Bank. In the 1980s, several scholars wrote about the problem of debt overhang (e.g. Sachs, 1984; Krugman, 1988). If countries have very high levels of debt, the possible returns on investments have to be used for their repayment obligations. Hence, the debtor cannot profit from these returns, and may face the incentive to consume rather than invest. This could impede economic growth in the debtor country. Moreover, if the levels of debt are too high, the repayments could make it happen that the debtor countries do not have enough money left to invest in order to induce growth: the repayments could deplete export revenues and prevent the import of investment goods (Roubini & Wachtel, 1998). The recognition of this potential problem caused by loans led donors to switch from giving aid in the form of loans, to disbursing it in the form of grants. Hence, it is expected that countries with high levels of debt receive relatively more grants, whereas countries with lower levels of debt receive relatively higher levels of loans. This hypothesis was tested by Marchesi and Missale (2013). They indeed found that higher debt ratios of recipient countries led to a higher inflow of grants and to a lower inflow of loans. Hence, the debt level of a recipient country seems to be a relevant instrument to distinguish between grants and loans. The exogeneity of debt as an instrument is more of an issue: it has been argued that democratic leaders are punished when they do not repay their debt. Those who lend their money do not give their vote to the political leader if he does not repay his debt. This possibility of punishment leads to better credit ratings for democratic countries, which makes that they could borrow at lower interest rates and hence accumulate less debt. Beaulieu et al. (2012) present empirical evidence to support this view. On the other hand, it has been argued that this
‘democratic advantage’ does not hold for developing countries, since these countries hardly borrow any money domestically. Hence, they are not punished by their lenders during the elections. Empirical support to this latter claim is given by Saiegh (2005) and by Archer et al. (2007). Based on all this, it cannot simply be assumed that debt is an exogenous instrument, and so this should be determined by an overidentification test. Finally, I will follow Selaya and Thiele (2012) by employing the aid commitments to developing countries as instruments. Since three different instruments with commitments are used (for grants; for loans and other long term capital; and total commitments) this should also help to distinguish the different types of aid. The instruments that were taken from the literature are summarized in table 2. This table also displays information about the strength of the instruments.
Table 2: Summary of the instrumental variables used in the literature
Article Instruments used Strong or weak?
Knack (2004) - Initial infant mortality
- Initial population
- Colonial heritage dummies
Knack does not report first stage F-statistics, but concludes that his instruments are effective merely by looking at the R2 in
the first stage Kersting & Kilby (2014) - Use the same set of
instruments as Knack
The first stage F-statistic is below 10 (9.69)
Selaya & Thiele (2012) - Initial infant mortality - Initial population
- Initial aid commitments to recipients
The first stage F-statistics are above 10
3.3 Econometric methodology
As already has been discussed, in this paper a cross-country analysis will be performed. I will look at changes in the levels of democracy over the whole period 1991-2013. It is therefore not the concern of this paper to analyse year-to-year changes in levels of democracy. The long run effect of foreign aid on democracy in the recipient country is the subject of interest here. This leads to the following base specification, based on the one used by Selaya and Thiele (2012):
where is the change in the level of democracy in country i over the period 1991-2013, is the level of democracy in country i in 1991, is the average level of type t aid received, where t denotes net total ODA, net loans or grants, is a vector containing several control variables and is the error term.
However, in the discussion of the data it was already indicated that there may be a problem with this base specification: the problem of endogeneity. More specific, there might be reversed causality involved in this specification. That is to say that foreign aid may not merely influence the level of democracy in the recipient country, but the recipient’s level of democracy may also cause the inflow of foreign aid to change. There are several channels that could explain how levels of democracy of recipient countries affect the flows of development aid. Two of them will be presented here, based on the discussion of Bräutigam(2000). The first channel predicts that higher levels of democracy lead to a higher level of foreign aid. Since donors are concerned with the effectiveness of foreign aid, they tend to send more money to countries that are most likely to use it in an efficient way. Countries with good institutions are expected to use the development aid more effectively. Hence, donors may send higher levels of aid to more
democratic countries, since the expected productivity of the aid sent to these countries is higher. The second channel predicts the opposite effect. Recipient need is an important factor in the decision to send development aid to a foreign country. Countries with deteriorating institutions signal that their development does not progress well. As a reaction, donors may choose to send more money to these countries, since they most urgently need it. Hence, donors can choose to disburse more aid to countries with lower levels of democracy. So even though it is not clear at first sight how the level of democracy affects the inflow of aid, it is quite plausible that it does.
IV regressions are used to correct for this problem of endogeneity. As discussed in the above section, one instrument that should capture recipient need, two instruments to capture the interest of the donor and two instruments to distinguish between grants and loans are used. The Two Stage Least Squares (2SLS) regressions will mainly be performed to add robustness to the results. Knack (2004), Goldsmith (2001a) and Kersting and Kilby (2014) all have used 2SLS regressions to correct for endogeneity; in all of these studies the results obtained using Ordinary Least Squares (OLS) did not differ significantly from those achieved using 2SLS. Hence, the expectation is that the effect of reversed causality is only small, or that the channels discussed above tend to offset each other. However, in order to provide more certainty that the obtained results are not a consequence of reversed causality, a 2SLS regression will be performed.
Besides the problem of endogeneity, there is also a problem with using OLS in the
analysis of an ordinal variable. Both the Freedom House index and the Polity IV index are ordinal variables. Even though they are often treated like interval variables in empirical studies, their
ordinal character may bias the results when OLS is used. I will follow Kersting and Kilby (2014) in employing an ordered probit regression model to check whether the results are biased in the OLS regression.
Furthermore, a problem that was already discussed in section 2.3 should be dealt with. This is the problem to which Wright (2009) directed attention. Wright argued that the results of a cross-country analysis are very sensitive to the choice of begin and end years of the period under investigation. It will be checked whether this is the case. This will be done by performing a regression that analyzes the period from 1992 to 2012 instead of the period from 1991 to 2013.
Moreover, the period under investigation will be split up in two separate periods: the first period covers the years from 1991 to 2001, and the second period covers the years 2002-2013. The reason to split up the period under investigation, and for choosing these specific periods, is given by the fact that the World Bank’s International Development Association (IDA) decided in 2002 that 18-21% of IDA funds would be disbursed in the form of grants (Radelet, 2005; see Sanford, 2001, for an overview of the arguments preluding this decision). Therefore, it would be interesting to see whether the effects of foreign aid on democracy levels were different in these subsequent periods.
Finally, some additional variables will be included in the regression analysis. This allows us to control for other factors that might have an effect on levels of democracy in recipient countries. From these results it should be possible to infer whether not including them in the earlier analyses could have biased the results.
3.4 Results
Table 3 shows the results of the OLS regressions with the different forms of aid as a percentage of the recipient’s GDP as independent variables. Columns 1-3 present the effects on the changes in the Freedom House index over the period 1991-2013, whereas columns 4-6 present the effects on the changes in the Polity IV index. The results show that net total ODA as a percentage of the recipient’s GDP is a statistically significant determinant of the changes in the Freedom House index. The same holds for grants. They are both significant at the 1% level and positively related to democratization. Loans, however, do not exhibit a statistically significant effect on the Freedom House index. These results change when we look at the changes in the Polity IV index. In that case, net total ODA, net loans and grants all are not statistically significant. So the statistical significance of the different forms of aid expressed as a percentage of the recipient’s GDP, does not seem to be robust to the employment of another index for democracy. This could be explained, either by the fact that the Freedom House index and the Polity IV index measure
Table 3: OLS regressions with aid as a percentage of the recipient’s GDP as independent variable 1 Change in FH index 1991-2013 2 Change in FH index 1991-2013 3 Change in FH index 1991-2013 4 Change in Polity IV index 1991-2013 5 Change in Polity IV index 1991-2013 6 Change in Polity IV index 1991-2013 FH index 1991 -0.479 (5.16)*** -0.467 (4.80)*** -0.473 (5.05)*** Polity IV index 1991 -0.593 (4.36)*** -0.612 (4.47)*** -0.592 (4.29)***
Net total ODA (% of
GDP, 1991-2013) -0.039 (4.29)*** (1.36)0.159 Net loans (% of GDP, 1991-2013) -0.194 (1.64) 0.721 (1.44) Grants (% of GDP, 1991-2013) -0.039 (4.30)*** 0.167 (1.17) Ln GDP per capita 1991 -0.332 (1.95)* -0.198 (1.13) -0.314 (1.84)* 0.590 (0.71) 0.282 (0.38) 0.535 (0.62) GDP per capita growth 1991-2013 -0.051 (0.64) -0.031 (0.39) -0.054 (0.68) -0.295 (2.34)** -0.357 (2.94)*** -0.284 (2.14)**
North Africa &
Middle East 0.579 (1.08) 1.165 (2.23)** 0.600 (1.12) -2.567 (0.70) -2.969 (0.84) -2.537 (0.69)
Sub-Saharan Africa 0.032 0.460 0.038 0.643 0.838 0.684
(0.08) (1.18) (0.10) (0.19) (0.26) (0.20)
South America -0.612 -0.125 -0.576 1.916 1.995 1.895
(1.39) (0.28) (1.31) (0.60) (0.64) (0.59)
Far East Asia -0.236 0.503 -0.234 1.251 0.323 1.351
(0.42) (0.94) (0.41) (0.34) (0.09) (0.36)
South & Central Asia -0.593
(1.24) 0.155 (0.34) -0.588 (1.21) 0.869 (0.25) 0.089 (0.03) 0.952 (0.27)
North & Central
America -0.776 (2.38)** -0.291 (0.92) -0.748 (2.30)** 3.145 (1.01) 3.237 (1.07) 3.095 (0.98)
Constant 4.394 2.758 4.204 -1.326 1.441 -0.916
(3.07)*** (1.94)* (2.93)*** (0.19) (0.26) (0.13)
R2 0.30 0.27 0.30 0.50 0.50 0.50
N 110 110 110 72 72 72
Absolute values of t-statistics in parentheses based on robust standard errors Asterisks denote the following: * p<0.10; ** p<0.05; *** p<0.01
democracy in a different way, or by the fact that the Freedom House index contains 38 more observations in our sample than the Polity IV index.
However, as already indicated, the results may be biased as a result of reversed causality. Therefore, it has to be checked whether the results in table 3 are the consequence of
endogeneity. This is investigated with 2SLS regressions. The results of these regressions are presented in table 4. The following sets of instruments were used. To predict the level of
Table 4: 2SLS regressions with aid as a percentage of the recipient’s GDP as instrumented variable 1 Change in FH index 1991-2013 2 Change in FH index 1991-2013 3 Change in FH index 1991-2013 4 Change in Polity IV index 1991-2013 5 Change in Polity IV index 1991-2013 6 Change in Polity IV index 1991-2013 FH index 1991 -0.520 -0.497 -0.515 (4.96)*** (4.96)*** (4.97)*** Polity IV index 1991 -0.632 (5.23)*** -0.665 (5.65)*** -0.627 (5.02)***
Net total ODA (% of
GDP, 1991-2013) -0.061 (2.04)** 0.288 (1.96)** Net loans (% of GDP, 1991-2013) -0.218 (1.20) 1.276 (2.11)** Grants (% of GDP, 1991-2013) -0.062 (2.09)** 0.342 (2.13)** Ln GDP per capita 1991 -0.365 (1.82)* -0.192 (1.06) -0.339 (1.81)* 1.477 (1.83)* 1.022 (1.41) 1.519 (1.94)* GDP per capita growth 1991-2013 -0.249 (3.28)*** -0.036 (0.43) -0.263 (3.45)*** -0.184 (0.59) -0.370 (3.26)*** -0.140 (0.44)
North Africa &
Middle East 0.596 (1.12) 0.903 (1.85)* 0.605 (1.12) -2.800 (0.82) -3.589 (1.10) -2.658 (0.76)
Sub-Saharan Africa 0.021 0.218 0.009 0.229 0.574 0.200
(0.05) (0.58) (0.02) (0.07) (0.19) (0.06)
South America -0.316 -0.457 -0.294 1.254 1.962 1.237
(0.68) (1.10) (0.63) (0.42) (0.69) (0.41)
Far East Asia 0.406 0.286 0.434 0.960 -0.046 1.173
(0.56) (0.55) (0.61) (0.25) (0.01) (0.31)
South & Central Asia -0.000
(0.00) 0.149 (0.35) 0.019 (0.04) 3.199 (1.02) 2.177 (0.73) 3.387 (1.07)
North & Central
America -0.827 (2.02)** -0.628 (1.94)* -0.812 (2.01)** 3.342 (1.16) 3.452 (1.25) 3.288 (1.13) Constant 5.296 3.145 5.090 -8.257 -3.769 -8.751 (2.97)*** (2.05)** (3.14)*** (1.20) (0.69) (1.30) First stage F-statistic 16.2683 14.9557 20.555 18.3029 14.9634 47.5483 Overidentification test (p-value) 0.3976 0.9336 0.4399 0.9963 0.7802 0.9985 R2 0.36 0.28 0.36 0.52 0.52 0.51 N 92 103 92 62 69 62
Absolute values of t-statistics in parentheses based on robust standard errors The overidentification test is Woodridge’s robust test statistic
Asterisks denote the following: * p<0.10; ** p<0.05; *** p<0.01
net total ODA, it was instrumented on the initial infant mortality rate, the initial population, on the initial external debt levels as a percentage of GNI, on a dummy denoting whether the
recipient country is a former British colony and on the total aid commitments in 1991 divided by the recipient’s GDP. The level of net loans was predicted by the employment of the initial
population, two dummies denoting whether the recipient is a former Belgian or Portuguese colony and the commitments to loans and other long term capital in 1991 divided by the recipient’s GDP as instruments. Finally, grants were instrumented on the initial infant mortality rate, the initial population, the initial external debt levels as a percentage of GNI, on a dummy that denotes whether a recipient is a former British colony and on the commitments to grants in 1991 divided by the recipient’s GDP. Again, columns 1-3 show the results for the regressions with the change in the Freedom House index as dependent variable, and columns 4-6 show the results with the change in the Polity IV index as dependent variable. First of all, the first stage F-statistics exceed the ‘rule of thumb’ value of 10 (Stock & Watson, 2012). Hence, it can be concluded that the instruments are strong enough to predict the levels of aid in the first stage of the regressions. Moreover, the p-values of Woodridge’s robust test statistic for
overidentification are high in all of the regressions, such that the null hypothesis ofexogeneity of the instruments cannot be rejected.
When the change in the Freedom House index is analysed, table 4 shows results quite similar to table 3: the effects of net total ODA and grants are statistically significant, whereas this is not the case for loans. The difference is that in the OLS regressions, net total ODA and grants were significant at the one percent level, while they are only significant at the five percent level in the 2SLS regressions. Hence, reversed causality does not affect the results very much.
Comparing the results of table 3 and table 4 to each other when the change in the Polity IV index is the dependent variable, the difference is more striking. Using OLS, none of the coefficients of the aid variables were statistically significant; in the 2SLS regressions, they all turn out to be significant at the five percent level. Hence, the problem of reversed causality was more severe in the analysis of the Polity IV index.
So, our preliminary results show that, when democracy is measured using the Freedom House index, grants are associated with higher levels of democracy in the recipient country than loans. This suggests that the channels that predicted that grants lead to higher levels of
democracy than loans have had more effect than those channels that forecasted the opposite effect. Moreover, the net total ODA has a statistically significant positive effect on levels of democracy in recipient countries. This result is in line with the finding of Kersting & Kilby (2014), who also found a statistically significant positive relationship between aid and democratization. And ipso facto it is contrary to Knack’s (2004) result, who did not find a statistically significant relationship between aid and democracy. However, the difference between grants and loans is no longer found when democracy is measured with the Polity IV index. In that case, loans and grants exhibit no different effects; they are both statistically
significant. The same holds for net total ODA. This latter finding is in line with the results of both Knack (2004) and Kersting and Kilby (2014), who all found that their results were robust to the
employment of an alternative democracy index. In the next section, some robustness checks will be performed.
3.5 Robustness checks
A factor that might have affected the results shown above is the way in which aid is normalized for the size of the recipient country. In table 3 it was expressed as a percentage of the recipient’s GDP. To check whether this choice has an influence on the results, aid is expressed in per capita terms in table 5. Just as in the previous table, columns 1-3 show the results for the change in the Freedom House index, and columns 4-6 show the results for the change in the Polity IV index. When we look at the change in the Freedom House index, there is a clear difference with the results obtained in the regressions with aid as a percentage of the recipient’s GDP. When aid is measured in per capita terms, net loans turn out to be significant as well. The coefficients for net total ODA and grants remain significant also. Hence, the difference between grants and loans that was found when aid is expressed as a percentage of the recipient’s GDP, vanishes when it is expressed in per capita terms. When the change in the Polity IV index is analysed, we even find a result opposite to the one found when the change in the Freedom House index was regressed on aid as a percentage of GDP: now loans turn out to have a statistically significant effect on
changes of democracy in the recipient country, while grants do not. However, as has been found in the 2SLS regressions, the results relating to the change in the Polity IV index might suffer from a severe problem of endogeneity. Unfortunately, there has not been found a set of instruments for aid in per capita terms for which the first stage F-statistic exceeds the value of ten; hence, we cannot check whether reversed causality is responsible for this result, and so we should be cautious with drawing too strong conclusions from it.
Because the Freedom House index and the Polity IV index are ordinal variables, OLS might lead to biased results. Therefore, Knack (2004) and Kersting and Kilby (2014) use ordered logit and ordered probit regressions respectively, to check whether the ordinal nature of these variables has consequences for the results. Table 6 shows the results of ordered probit
regressions.1 Employing ordered probit regressions does not alter the results drastically; again,
using the change in the Freedom House index as dependent variable, net total ODA and grants turn out to have an effect on the changes in the levels of democracy that is statistically significant at the one percent level. Loans do exhibit a statistically significant effect only at the 10 percent level. When the Polity IV index is used as outcome variable, both loans and grants do not show a
1When ordered logit regressions are employed, the results are quite similar. The main difference is that in the
ordered logit regressions loans are no longer statistically significant at the ten percent level when they are regressed on the change in the Freedom House index.
Table 5: OLS regressions with aid per capita as independent variable 1 Change in FH index 1991-2013 2 Change in FH index 1991-2013 3 Change in FH index 1991-2013 4 Change in Polity IV index 1991-2013 5 Change in Polity IV index 1991-2013 6 Change in Polity IV index 1991-2013 FH index 1991 -0.485 -0.479 -0.480 (4.97)*** (5.16)*** (4.85)*** Polity IV index 1991 -0.615 (4.56)*** -0.631 (4.59)*** -0.611 (4.49)***
Net total ODA (per
capita, 1991-2013) -0.002 (3.38)*** 0.012 (1.77)*
Net loans (per
capita, 1991-2013) -0.019 (4.21)*** 0.058 (2.71)***
Grants (per capita,
1991-2013) -0.002 (3.28)*** 0.012 (1.19)
Ln GDP per capita
1991 -0.118 (0.74) -0.074 (0.49) -0.122 (0.75) 0.009 (0.01) 0.010 (0.01) -0.013 (0.02)
GDP per capita
growth 1991-2013 -0.045 (0.54) -0.024 (0.34) -0.047 (0.56) -0.335 (2.70)*** -0.366 (3.10)*** -0.324 (2.51)**
North Africa &
Middle East 0.706 (1.36) 1.119 (2.26)** 0.739 (1.42) -2.940 (0.79) -3.112 (0.89) -2.898 (0.77)
Sub-Saharan Africa -0.023
(0.06) 0.359 (0.98) 0.008 (0.02) 1.025 (0.31) 1.022 (0.33) 1.043 (0.31)
South America -0.620 -0.336 -0.561 2.125 2.269 2.047
(1.39) (0.78) (1.26) (0.66) (0.75) (0.63)
Far East Asia -0.097 0.323 -0.059 1.142 0.668 1.125
(0.17) (0.62) (0.10) (0.31) (0.19) (0.30)
South & Central Asia -0.420
(0.88) 0.145 (0.32) -0.395 (0.83) 0.656 (0.19) 0.162 (0.05) 0.687 (0.20)
North & Central
America -0.735 (2.16)** -0.312 (1.03) -0.699 (2.03)** 3.256 (1.04) 3.573 (1.21) 3.135 (0.99)
Constant 2.775 2.047 2.732 2.976 3.408 3.184
(2.07)** (1.73)* (2.00)** (0.53) (0.62) (0.56)
R2 0.28 0.30 0.28 0.49 0.50 0.49
N 110 110 110 72 72 72
Absolute values of t-statistics in parentheses based on robust standard errors Asterisks denote the following: * p<0.10; ** p<0.05; *** p<0.01
statistically significant effect; net total ODA is only significant at the ten percent level. So the results do not seem to be biased very much as a result of treating the Freedom House index and the Polity IV index as interval variables.
Now, as already mentioned, there should be dealt with the critique of Wright (2009) on the cross-sectional approach that is also used in this paper. Does the choice for begin and end years have an effect on the results that are obtained? To check for this, different begin and end
