Donor Fragmentation,Aid Shocks,and Violent Political Conflict

Abstract

Recent debates have focused on the negative role of the proliferation of foreign aid facilities and donor fragmentation for development outcomes and recipient country institutions. This article investigates an overlooked positive side effect of donor proliferation. With an increasing number of donors, exposure to negative aid shocks decreases, as well as the impact of such shocks on violent political conflict. Using data on 106 recipient countries for the years 1970 to 2008 and employing event history and mediation analysis, we find strong evidence that fragmentation significantly reduces the risk for political destabilization associated with aid shocks.

Keywords

foreign aid conflict bargaining internal armed conflict political economy

The proliferation of governmental and nongovernmental organizations (NGOs) that provide aid to developing countries has been at the center of recent academic and policy debates (see, e.g., Aldasoro, Nunnenkamp, and Thiele 2010; Easterly 2007). There is evidence that donor fragmentation has negative consequences, both for the efficacy of aid (Djankov, Montalvo, and Reynal-Querol (2009) and for domestic institutions in recipient countries (Morss 1984; Knack and Rahman 2007).

While these negative effects are real and have potentially serious consequences, donor fragmentation may have a largely overlooked positive side effect. Highly concentrated donor structures mean that unexpected aid shortfalls by one main donor can do serious harm to overall aid flows to a recipient country. For example, countries that receive more than half of aid from just one donor experience aid shocks that are on average 21 percent larger than those with a broader donor base (5.90 cents per dollar of gross domestic product [GDP] vs. 4.86 cents).¹ By making a recipient country less dependent on a single donor, or a small group of donors, donor proliferation can reduce both the frequency and the impact of shocks to aid flows.

This connection between donor fragmentation and aid volatility has important consequences. In this article, we focus on the relationship between donor fragmentation and violent political conflict. When aid receipts make up a substantive share of government revenue, unpredictable shortfalls have direct consequences for domestic distributional outcomes. This can affect political stability in direct and indirect ways.² Aid volatility affects the predictability of government revenue. This translates into uncertainty about the value of holding office, making it harder to strike binding bargains with political stake holders (Arcand and Chauvet 2001). Shocks to revenue also affect the government’s ability to suppress unrest, shifting the balance of power between the government and would-be rebel groups (Nielsen et al. 2011). Aid volatility affects political stability also indirectly, because it disrupts government spending on welfare and infrastructure programs (Agénor and Aizenman 2010; Arellano et al. 2009) and reduces short-term economic growth (Lensink and Morrissey 2000; Kharas 2008). Economic performance in turn is one of the most robust cross-sectional correlates of violent political conflict (Blattman and Miguel 2008).

A clear, if perhaps extreme, example for the importance of aid shocks comes from the Cold War period. On March 8, 1977, the rebel group Front de la Libération Nationale Congolese (FLNC) started an attack on the mineral rich Shaba (Katanga) region in southeastern Zaire from the neighboring Angola. The FLNC fighters made fast inroads and quickly gained control of sizable portions of the territory. The Zairian forces crumbled in the face of the onslaught, and only military supplies from the United States, France, and Belgium, as well as pilots and 1,500 combat troops from Egypt and Morocco saved the day for the regime of Mobutu (Kelly 1993).

What is puzzling about this episode of violent political conflict is why the FLNC was willing to take on Zaire’s government at all. Mobutu’s Zaire was one of the main beneficiaries of US aid during the Cold War. In the five years preceding the raid, the US and other bilateral donors pumped an average of $420 million per year in official development aid (ODA) into the country. The numbers are less clear for military aid, but the amounts were substantive. The economically important Shaba region had been a focal point of violence before, and ensuring territorial integrity must have been a high priority. Why then did the Mobutu regime not have the military capability to deter rebellion?

The answer lies in a sudden restriction by US policy makers to support Zaire financially. When the Ford administration’s involvement in the civil war in neighboring Angola became public in 1976, Congressional action (the Tunney/Clark amendment) ensured that a steady flow of American weapons and money through Zaire came to an abrupt halt. This disruption in turn hurts the relative strength of Mobutu, who had served as a conduit for US aid and weapons destined for Angola. Continued US support for Zaire suddenly was cast into doubt. While this example of an aid shock extends beyond the realm of ODA, Zaire’s dependence primarily on American support had dramatic consequences for its political stability (though in this instance Western military intervention sought to correct the political results). Arguably, had Zaire relied less on the United States as its main donor, a sudden change in American policy would have had less disruptive effects.

While this anecdote is informative, we need to establish a more systematic understanding of the relationship between the concentration/fragmentation of the donor system and the risk of violent political conflict. In the following, we therefore first develop the theoretical framework that ties donor fragmentation to both the frequency and makeup of aid shocks and the impact of these shocks on the onset of violent political conflict. We identify two potential pathways through which fragmentation can attenuate the negative consequences of aid shocks, distinguishing between a moderated and a mediated relationship. We then present results from a statistical analysis of 106 recipient countries of ODA from 1970 to 2008. We show that donor fragmentation substantively moderates downward the risk of violent political conflict associated with aid shocks, producing a meaningful insurance effect against political instability. Fragmentation also reduces the incidence of aid shocks but not enough for a substantively significant mediated relationship between fragmentation and conflict onset.

Our article speaks to the growing literature that evaluates the negative effects of donor fragmentation for economic development. To the extent that donor fragmentation reduces the detrimental effects of aid shocks on political stability and short-term growth, this directly works against its documented tendency to weaken state capacity and undermine growth in the long run. Our findings also have important implications for the policy debate on how to reduce donor fragmentation. In particular, calls for greater donor coordination run the danger of undermining the insurance function of donor fragmentation against negative aid shocks and political destabilization.

From a larger theoretical perspective, the issue of donor fragmentation belongs to a subset of questions about collective action problems in the provision of global public goods. Donor fragmentation is to some extent driven by bureaucratic self-interest but also is indicative of coordination problems in the provision of developmental goals that have public good characteristics (Steinwand 2015). Dependent on the main source of behavior, scholars tend to either emphasize welfare losses (fragmentation overwhelms recipient capacity) or welfare gains (fragmentation and insurance against shocks). Such trade-offs exist for other policy areas as well, including, for example, preferential trade agreements and environmental regimes. Our work helps sharpen awareness of such trade-offs.

Donor Fragmentation as Insurance against Armed Political Conflict

In order to study the effects of donor fragmentation on violent political conflict, we need to bring together the relationship between donor fragmentation and aid volatility, the effects of volatility on political stability, and how donor fragmentation helps to lessen the incidence and consequences of aid shocks. We argue that donor fragmentation can serve as insurance against the negative consequences of aid volatility by reducing the risk of negative aid shocks as well as moderating how shocks affect political stability. As a result, donor fragmentation helps to decrease the number of episodes of violent political conflict associated with aid shocks.

Figure 1 illustrates the two causal pathways that lead from donor fragmentation to the onset of violent conflict. On the first pathway, donor fragmentation mediates the relationship between aid shocks and conflict onset. That is, donor fragmentation reduces the risk of conflict indirectly by reducing the likelihood that a recipient country experiences an aid shock in the first place. Symbolically, in Figure 1, the causal arrows run from donor proliferation to reduced frequency of aid shocks to conflict, with no short-term direct effect of donor proliferation on conflict onset.

Figure 1.

Causal pathway.

The second pathway describes a moderation relationship between proliferation, aid shocks, and conflict. That is, fragmentation reduces the negative effects of those shocks that actually occur. In Figure 1, this is symbolized by the arrow pointing from donor proliferation to the link between aid shocks and conflict, showing that proliferation reduces the risk of conflict onset from aid shocks that do occur. We discuss the elements of these causal paths in turn, beginning with the relationship between aid shocks and violent political conflict.

A major venue through which aid affects political stability is through government finances. The basic notion is that the government’s financial capacity shapes societal agreements on how to redistribute economic and political rents. Aid enters into this in at least three ways. First, aid provides direct rent extraction opportunities and thus increases the value of holding office (Grossman 1992). Second, aid can go toward transfers to the population, helping to quell dissatisfaction and paying off potential rebels (Morrison 2007). And third, aid increases the government’s motivation and ability to suppress political dissent (Bueno de Mesquita and Smith 2009; Nielsen et al. 2011; Steinwand 2014).^3,4

In each of the three instances, foreign aid alters distributional arrangements, typically in favor of the government. Unequal rent extraction and a skewed wealth distribution may increase the potential for conflict (Acemoglu and Robinson 2000) but to explain the actual breakdown of domestic redistributional arrangements we need to look further.⁵

Aid shocks disrupt bargaining processes and therefore can trigger political instability. Drawing on the rationalist literature on interstate and civil war (Fearon 1995, 2004), a variety of authors have tied negative aid shocks to a breakdown of bargaining between the government and societal stake holders or rebel groups. For example, Arcand is among the first to point out that uncertainty about current aid flows can lead to bargaining breakdown (Arcand and Chauvet 2001). Further, Nielsen and colleagues (Nielsen et al. 2011) argue that variation in aid flows affects the government’s future ability to fight rebellion and therefore creates a commitment problem that makes it impossible to find peaceful agreements in the present. The authors find statistical evidence that severe negative aid shocks are associated with an increased risk of civil war.

We follow the approach taken by Nielsen et al. and concentrate on negative aid shocks, that is, sudden large decreases in aid allocations. Negative aid shocks degrade the government’s ability to either buy support of opposition groups or threaten the use of force, resulting in a greater likelihood of conflict. In contrast, aid windfalls, that is, unexpected large inflows of money, have no such pernicious effect. To see why, consider the structure of the government’s commitment problem. The commitment problem arises if the opposition is in a weaker bargaining position in the future than in the present. In this scenario, sudden aid shortfalls mean that the government cannot credibly promise to uphold a deal that reflect’s the opposition’s current strength in the future. Therefore, the opposition exploits its current position of strength to start a conflict (see, e.g., Acemoglu and Robinson 2000). Positive aid shocks do not follow this logic. Aid windfalls put the government in an unexpected position of strength, as it has more money to either increase suppression or payoff opposition groups. The government can offer the opposition a relatively bad deal that reflects its current strength, without limiting its ability to offer a better deal in the future once it reverts to a relative position of weakness. The opposition has nothing to gain from starting a conflict at this point in time, as it is receiving an optimal deal today and in the future.⁶

Aid volatility affects political stability also indirectly. A number of studies find that volatility reduces economic growth (Lensink and Morrissey 2000; Kharas 2008).⁷ Aid volatility tends to be pro-cyclical (Pallage and Robe 2001; Bulíř and Hamann 2003, 2007), exacerbating economic contractions. It also has the potential to disrupt financing of government programs and infrastructure spending (Agénor and Aizenman 2010; Arellano et al. 2009), which in turn also has detrimental effects for development. These negative economic effects are undesirable in their own right, but they also are problematic for political stability. Overall economic prosperity is perhaps the best available predictor of political stability across time and space (Fearon and Laitin 2003; Blattman and Miguel 2008). More importantly, economic knock-on effects exacerbate the direct effects that sudden aid shocks have for the relationship between the government and societal stake holders. The destabilizing role of economic shocks is well understood (Miguel, Satyanath, and Sergenti 2004; Savun and Tirone 2012). Aid shocks tend to translate into economic shocks, compounding direct with indirect effects. For example, pro-cyclical drops in aid reduce the government’s fiscal position directly but also depress tax receipts.

The next element in the connection between donor fragmentation and political stability is the relationship between donor fragmentation and aid shocks. The central focus of the policy discussion and academic work has been on negative effects of fragmentation on program implementation (Easterly 2007), local government performance (Knack and Rahman 2007; Frot and Santiso 2010), and corruption (Djankov, Montalvo, and Reynal-Querol 2009). These problems can adversely affect development, and Djankov, Montalvo, and Reynal-Querol (2009) present evidence that fragmentation is associated with slower economic growth. Taken together, these findings suggest that donor fragmentation could have a negative effect on political stability through its negative impact on institutions and growth.

We do not dispute that donor fragmentation can put pressure on long-term growth rates and institutional performance. However, it is not clear that donor fragmentation has similar effects in the short term. Instead, we have good evidence that aid shocks and economic knock-on effects contribute to violent political conflict. While this is a short-term risk, its potential consequences for economic development can be catastrophic if violence escalates into full-fledged civil war (Collier et al. 2003). We believe that existing works have ignored how donor fragmentation can help mitigate this risk.

Donor fragmentation can affect political stability through both the frequency of aid shocks (mediation) and the makeup of aid flows (moderation). For the frequency of aid shocks, we rely on work by Hudson and Mosley (2008). The authors note that donor fragmentation should serve to reduce aid volatility and they provide descriptive statistics to underscore this point. We follow their demonstration of the relationship between donor fragmentation and aid volatility.

Let X be the total aid provided by just one donor, with expected (average) payout $E (X) = μ$ . Volatility is measured as the variance, $Var (X)$ . Now assume the donor gives X but for accounting purposes divides this amount into two equal parts $X_{1 / 2}$ , such that $2 X_{1 / 2} = X$ . Since this is only an accounting change, the two parts $X_{1 / 2}$ are perfectly correlated with each other, and we can decompose $Var (X)$ into

Var (X) = Var (2 X_{1 / 2}) = 2 Var (X_{1 / 2}) + 2 Cov (X_{1 / 2}, X_{1 / 2}) = 4 Var (X_{1 / 2}) .

This result only obtains because the two quantities $X_{1 / 2}$ perfectly covary with each other and therefore $2 Var (X_{1 / 2}) = 2 Cov (X_{1 / 2}, X_{1 / 2})$ .

Consider now a situation where the same total amount of aid X is provided by two donors, $i = 1, 2$ , each giving equal aid shares $X_{1 / 2}^{i}$ . Thus, average total aid provision is unchanged $E (X_{1 / 2}^{1} + X_{1 / 2}^{2}) = μ$ . However, now total volatility is

Var (X_{1 / 2}^{1} + X_{1 / 2}^{2}) = Var (X_{1 / 2}^{1}) + Var (X_{1 / 2}^{2}) + 2 Cov (X_{1 / 2}^{1}, X_{1 / 2}^{2}) .

This is always less than $4 Var (X_{1 / 2})$ because $2 Cov (X_{1 / 2}^{1}, X_{1 / 2}^{2}) < Var (X_{1 / 2}^{1}) + Var (X_{1 / 2}^{2})$ as long as the correlation between $X_{1 / 2}^{1}$ and $X_{1 / 2}^{2}$ is less than 1. In reality, aid of several donors never should be perfectly correlated. This condition will always be met even if donors exhibit ‘herding’ behavior, that is, imitate each other. While we develop our example with only two donors, dividing up a fixed amount of aid between more than two donors always leads to additional decreases in volatility.⁸ Intuitively, aid flows to a given country depend less on a particular donor’s actions if this donor provides a smaller share of aid, and increasing the number of donors decreases each donor’s share.

To relate negative aid shocks to aid volatility, we can think of aid provision as a stochastic process. Aid shocks are extreme realizations of this process. As the variance of the process increases, more probability mass is located in the tails of its distribution, and aid shocks become more likely. Likewise, a reduction in variance reduces the probability of aid shocks.⁹ In line with this reasoning, we conceptualize aid shocks as deviations from an expected value that pass a threshold. For purposes of the empirical analysis, we think of the expected value as a long-term observed trend and the threshold is a function of the overall distribution of aid. Since donor fragmentation reduces the variance of aid, it also decreases the probability of negative aid shocks.

Finally, we turn to how donor fragmentation helps to reduce the negative effects of aid shocks for political stability. It is useful to emphasize the difference between this and the previous causal pathway. To the extent that donor fragmentation prevents aid shocks from happening in the first place, the makeup of the donor system mediates the relationship between aid shocks and the onset of political conflict. However, there are good reasons to believe that donor fragmentation also reduces the negative effects of those shocks that actually occur, that is, donor fragmentation moderates such effects.

We argue that fragmentation of aid delivery goes hand in hand with changes to the composition of aid flows. As aid becomes more oriented toward developmental goals, the less harmful the effects of aid shocks will be. Beginning with the first part of this argument, increasing donor fragmentation results in part from fundamental changes in aid practices after the end of the Cold War. Symbolized by the Millennium Development Goals, aid policies have generally become more oriented toward achieving developmental goals and less self-interested on the part of donors (Lancaster 2007). For example, Steinwand (2015) shows that instances of highly concentrated aid delivery in the hand of a single “lead donor” have declined as the overall aid system has become more fragmented. Lead donorship in turn is associated with self-interested donor goals such as export promotion and access to raw materials.

In a similar vein, Girod (2012) shows that after the end of civil conflict, aid provided for strategic reasons is less effective in fostering development, as recipient governments are relatively free to ignore conditionality that seeks to foster development.¹⁰ Important measures of donor strategic interests are associated with higher donor concentration, such as colonial history and shared military ties, and defensive treaties and troop deployments by the donor to the recipient country. In a narrower setting, Steinwand (2014) demonstrates that US stability interests lead to decreased aid volatility, but aid shocks are associated with a greater risk of conflict when stability interests are high.¹¹ Again, US stability interests are associated with greater donor concentration.

Donors use ODA to pursue a variety of goals, including genuine developmental aims, but also strategic and self-interested political goals (e.g., Lancaster 2007; Dreher, Nunnenkamp, and Thiele 2008). ODA flows always have some developmental and some nondevelopmental functions. The relative mixture of these characteristics can be tied to how donors deliver aid and in turn to donor concentration. Looking at sectoral aid allocations, Dreher, Nunnenkamp, and Thiele (2008) show that program aid, such as general budget support and debt relief, is closely tied to support for US voting positions in the UN General Assembly. In contrast, project specific aid does not exhibit such a political dimension. In a similar vein, Dietrich (2013) shows that donors who care about the effective implementation of specific projects “bypass” recipient governments with bad track records and administer ODA through NGOs.

Our own data show a pattern that is consistent with systematic differences in the developmental content of ODA in fragmented versus concentrated donor systems. There is a clear relationship between donor concentration and the choice of delivery channel. As the left panel of Figure 2 shows, the share of total aid given through government channels increases as the donor system becomes more concentrated (r = .176). At the same time, for aid to NGOs, there is a strong negative relationship with donor concentration (right panel, r = −.366). NGO aid is more developmentally oriented in nature, as governmental aid is subject to rent extraction.¹²

Figure 2.

Relationship donor concentration and delivery channel.

When breaking ODA allocations down by sector, a similar picture emerges. Higher donor concentration is associated with more politically motivated program aid such as general budget support and less developmentally oriented project aid.¹³ Regressing the program aid share of ODA on donor concentration, we find a positive relationship. Moving from complete donor fragmentation to full concentration increases the share of program aid on average by about 10.5 percentage points, while at the same time decreasing the more developmentally oriented project aid content by the same amount.¹⁴

Shocks to ODA with a greater developmental content should be less politically destabilizing for a host of reasons. Of the direct links between aid and political stability that we discussed above, a majority focuses on nondevelopmental uses of aid, whether aid is thought to increase the value of office holding (Grossman 1992), is used to suppress rebel activity (Steinwand 2014), or alters the terms of agreement in domestic bargaining between the government and important constituent groups (Bueno de Mesquita and Smith 2009; Nielsen et al. 2011). In contrast, developmental aid is geared toward providing public goods and is not as easily ransomed by domestic constituencies, especially if administered through NGO channels. Shocks to ODA that is developmental in character therefore have less potential to upset existing political arrangements and trigger conflict. On the other hand, because of its public goods nature, such shocks will affect economic growth, which in turn links aid shocks and conflict in an indirect manner. On balance however, greater developmental content of ODA should result in fewer problems for political stability. In summary, we expect donor fragmentation is associated with a more developmentally oriented mix of aid allocations, which in turn should reduce the risk of conflict onset associated with shocks to ODA.

We now develop our argument about the relationship between negative aid shocks and violent political conflict in a more formal fashion. Typical models of civil conflict treat a stake holder’s decision to fight as related to the reservation value for a peaceful settlement. If the expected payoffs from fighting for the would-be rebels seem more promising than the alternatives, fighting is a rational decision, despite the expected costs for life and property. As we discussed above, aid shocks disrupt the bargaining between societal stake holders because they introduce uncertainty about aid allocations.¹⁵ Once an aid shock is observed, uncertainty about the current state of the world is resolved. Formally, a rebel group fights after observing an aid shock if

D (Aid) = F (Aid) - SQ (Aid) > 0,

where F(Aid) is the expected utility from fighting given observed aid levels, SQ(Aid) is the expected utility of remaining at peace, and the difference between the two is D(Aid).¹⁶ Both utilities are a function of realized aid provisions, as aid either influences relative government capabilities or is used as subsidy to the population. Both causal pathways affect the outcome in the same fashion, since less aid makes fighting more attractive and staying at peace less attractive, that is, $\partial SQ (Aid) / \partial Aid > 0$ and $\partial F (Aid) / \partial Aid < 0$ . The severity of the aid shock therefore determines the ex-post probability of conflict, conditional on observing an aid shock. In the statistical analysis, this is a function of the systematic differences between countries that affect the reservation price for fighting and a random disturbance. In contrast, the ex-ante probability of conflict is a function of the likelihood of experiencing an aid shock that is sufficiently large to meet the conflict threshold,

Pr (Conflict) = Pr (D^{*} (Aid) > 0),

where $D^{*} (Aid)$ varies with the realization of Aid. $D^{*} (Aid) = 0$ defines the reservation value for peace. If we rewrite the probability of fighting in terms of the cumulative distribution function of Aid, and the derived random variable $D^{*} (Aid)$ , we have $Pr (D^{*} (Aid) > 0) = F_{D^{*} (Aid)} (0)$ . Note that the distribution described by $F_{D^{*} (Aid)}$ is not centered on zero, because on average the peace condition is met. Higher aid volatility leads to increases in $Var (D^{*} (Aid))$ . And since the reservation value for peace is fixed at 0, this in turn causes increases in $Pr (D^{*} (Aid) > 0)$ . Higher aid volatility thus increases the risk of an aid shock and in turn the probability of conflict. Similarly, as aid shocks become more severe, the mean of the distribution shifts toward 0, making conflict more likely in expectation. Since only negative deviations from expected aid flows will trigger violent conflict, we concentrate on this aspect.

Together, the relationships between donor fragmentation, negative aid shocks, and violent conflict cover the causal pathways from Figure 1. Since donor fragmentation mediates the risk of conflict onset by reducing the likelihood that aid shocks occur and mitigates the negative effects on political stability of those aid shocks that actually occur, donor fragmentation serves to reduce the overall risk of conflict onset. The two causal pathways define the central hypotheses of our article. In the next section, we discuss the data and the statistical setup with which we test these hypotheses. We then present results from the empirical analysis.

Empirical Analysis and Data

We assemble a data set covering 106 countries that have received ODA between 1970 and 2008.¹⁷ Calculating moving averages and taking lags effectively reduces the years in the analysis to 1972 to 2007.

Aid Shocks

We begin with coding aid volatility. The first choice concerns the measure of aid flows. We use bilateral net ODA flows, taken from the Organisation for Economic Co-operation and Development’s (OECD) International Development Statistics (2009). Net flows measure total aid disbursements minus repayments of non-concessional loans. Unlike aid commitments or gross disbursements, looking at net flows gives an accurate picture of the total free resources from aid available in a given year. We measure aid by aggregating total aid flows for each recipient country and year. Using cross-sectional aid data also requires that we standardize aid flows. We divide aid per GDP to make the economic impact of aid shortfalls comparable across countries. There are twenty-two OECD donors in the analysis and we concentrate on bilateral aid. While growth in multilateral institutions and NGOs is an important aspect of donor fragmentation, bilateral aid tends to be more discretionary in character. It therefore plays a bigger role in sustaining recipient country governments and is more consequential for political stability.¹⁸

Next, we calculate a five-year sliding mean of the aid variable for each country. We measure volatility as deviation from this trend, adding a one-year lag. The resulting variable takes the form

Δ y_{t_{0}} = y_{t_{0}} - \frac{1}{5} \sum_{j = 1}^{5} y_{t_{- j}} .

This conceptualization follows the standard approach in the literature on aid volatility.¹⁹ From a theoretical perspective, aid shocks are deviations from expected aid flows. Expectations of future aid flows are a function of policy decisions made in various donor countries, as well as economic circumstances. It is difficult for officials to form these expectations with precision and even harder for researchers to measure them. It therefore makes sense to use a backward-looking element to establish a baseline for what constitutes a shock. To capture negative shocks, we code an indicator variable to 1 for observations in the lower fifteenth percentile of $Δ y_{t_{0}}$ and 0 otherwise.²⁰ This approach has the advantage that it identifies shocks only if they are substantively important relative to a country’s GDP. Large absolute shifts in aid allocations will not affect political stability if they are not of economic importance and thus should not be coded as shocks.

Relatively small deviations from long-term aid trends are the norm. In our sample, the mean deviation from the five-year moving average of aid is −0.00289 percent of GDP, with a standard deviation of 0.418 percent. Guinea-Bissau experienced the most extreme negative shock to aid flows in 1997, where the deviation in aid per GDP from the five-year moving average was −22.9 percent of GDP. On the positive side, the Democratic Republic of the Congo in 2003 had a positive deviation of 120 percent of GDP. In terms of our dichotomous aid shock variable using the lower fifteenth percentile cutoff, the proportion of aid shocks in our analysis is 0.167 with a standard deviation of 0.373.

Conflict

We take conflict data from the Uppsala Conflict Data Program/Peace Research Institute Oslo Armed Conflict Dataset v.4-2009 (Gleditsch et al. 2002). This variable measures violent political conflict and is coded 1 if twenty-five or more battle deaths occurred in a given year between the government and at least one organized group and 0 otherwise. The advantages of the Armed Conflict Data for our purpose are that it taps into relatively low-intensity episodes of fighting and it focuses only on conflicts that involve the government. Other measures of violent political conflict, such as the Correlates of War project, focus on full-blown civil war and are not sensitive enough to register the type of violent episodes triggered by aid shocks. At the same time, we do not want to pick up, for example, interethnic violence or the activities of organized crime, as these follow a different logic than violent conflict with the government.

Donor Concentration

Our key independent variable is donor concentration. We follow a widespread practice and calculate the Herfindahl–Hirschman index (HHI) of aid shares. The HHI is calculated by taking the sum of squared aid shares of all donors, that is, $HHI = \sum_{i} {({aid}_{i} / total aid)}^{2}$ , where i indexes all donors who give aid. The HHI provides a measure of donor concentration, with 0 corresponding to perfect fragmentation and 1 to total concentration. The HHI has the advantage of accounting for increases in the numbers of donors, while correcting for relative size of aid share. For example, when moving from one to two donors with equal aid shares, donor concentration decreases from 1 to 0.5. However, if aid allocated by the two donors splits 3/4 to 1/4, the resulting HHI is 0.625, reflecting the larger influence of the main donor.

For the thirty-five years in our analysis, the mean of the HHI variable is 0.312 with a standard deviation of 0.176. As illustrated in Figure 3, donor concentration is positively skewed, with most countries clustered somewhat below the mean. Among the countries in our analysis, Mongolia in 1984 had the most concentrated aid, with an HHI of 1. Turkmenistan also had high donor concentration, with an HHI between 0.933 and 0.993 in 1993 and 1994, respectively. At the other end of the spectrum, Mozambique had the most fragmented aid with an HHI between 0.0756 and 0.0889 from 1994 to 2007.

Figure 3.

Distribution of donor concentration.

Regionally, the most fragmented aid went to Central & South Asia (mean HHI = 0.276). Aid to Sub-Saharan Africa and South America is similar in structure (mean HHI Africa = 0.286, mean HHI S. America = 0.298). A second group of regions has more concentrated aid, with HHI values clustering around 0.40. This group includes Central America and the Caribbean (mean HHI = 0.381), Europe (mean HHI = 0.413), and the Middle East and North Africa (mean HHI = 0.414). Countries in Oceania receive the most concentrated aid (mean HHI = 0.471). The data bear out the much discussed increase in fragmentation over time. The average HHI was 0.411 in the 1970s, declined to 0.352 in the 1980s, to 0.324 in the 1990s, and finally to 0.307 for the years after 2000.

Empirical Models

To test our hypotheses that donor fragmentation reduces both the frequency of aid shocks and lessens their negative impact on civil conflict, we estimate two separate models. We begin with the latter.

Moderation Model

To capture how fragmented aid delivery conditions the effect of aid shocks (i.e., the moderation effect), we use an interaction design between aid shocks and donor concentration:

Y_{i, t}^{*} = β_{1} Aid {Shock}_{i, t_{- 1}} + β_{2} Donor {Concentration}_{i, t_{- 1}} + β_{12} Aid {Shock}_{i, t_{- 1}} \times Donor {Concentration}_{i, t_{- 1}} + β X_{i, t_{- 1}} + g (t) + u_{i, t}

The dependent variable $Y^{*}$ is the unobserved latent tendency for conflict. We only observe a binary censored version of this, which we assume takes on values of Y = 1 if $Y^{*} \geq 0$ and Y = 1 otherwise. All independent variables enter the analysis with a one-year lag.²¹ The main independent variables are Aid Shocks and Donor Concentration and their interaction term. We also include a number of control variables $X_{i, t_{-}_{1}}$ that we discuss in detail below. The term g(t) represents the baseline hazard of conflict initiation (Beck, Katz, and Tucker 1998). We follow Carter and Signorino and include a third degree polynomial of the years since the last conflict episode (Carter and Signorino 2010).

In principle, this model can be estimated using standard logit or probit, but the most appropriate setup for event history data is the complimentary log–log model (Alt, King, and Signorino 2001).²² However, the onset of violent political conflict is a rare event. Of the 2,648 observations in our main specification, there are only 128 instances of conflict onset (4.83 percent). It is well known that for low likelihood events like this, event history models suffer from rare events bias (King and Zeng 2001). We correct for this by using the rare events logistic regression proposed by King and Zeng. We also replicate the analysis fitting a cloglog-model without rare events correction, with only minor loss in statistical certainty.²³

Mediation Model

To test how donor fragmentation reduces the incidence of aid shocks and as a result the risk of civil conflict (i.e., mediation), we need a two-equation setup. The first equation captures aid shocks:

Aid {Shock}_{i, t_{- 1}}^{*} = γ_{1} Donor {Concentration}_{i, t_{- 2}} + γ X_{i, t_{- 2}} + h (t) + ∊_{i, t_{- 1}} .

Here, the dependent variable ${Aid Shock}^{*}$ is the unobserved latent tendency for an aid shock, with observed values of $Aid Shock = 1$ if $Y^{*} \geq 0$ and Y = 0 otherwise. Note that in this equation the dependent variable is lagged by one time period and all independent variables by two time periods, to ensure the proper sequence of events leading from donor concentration to conflict onset. The main independent variable is Donor Concentration, and as in the conflict equation, we also include a number of control variables $X_{i, t_{- 2}}$ . The term h(t) represents the baseline hazard of aid shocks, which we model with a third degree polynomial of the years since the last aid shock.

The second equation in the mediation setup again models the occurrence of conflicts. It is identical to equation (6), but we drop the interaction between aid shocks and donor concentration. In order to estimate how much fragmentation reduces the incidence of conflict through lowering the frequency of aid shocks (i.e., the mediated, or indirect, effect), we first separately estimate the aid shock and conflict equations. This provides the effect of donor concentration on aid shocks (i.e., the direct effect) and the effect of aid shocks on conflict. To obtain the mediated effect of concentration on conflict, we use the mediation package in R (Tingley et al. 2014) with Donor Concentration as the treatment and Aid Shocks as the mediator.²⁴ Evidence of a mediated effect indicates that the causal arrow goes from concentration to aid shocks to conflict.

Mediation analysis involves stronger assumptions than typical regression. In particular, sequential ignorability must be met in order to draw causal inferences from mediation models. Sequential ignorability consists of two assumptions: (1) conditional on the observed pretreatment covariates, the treatment is independent of all potential values of the outcome and mediating variables; and (2) the observed mediator is independent of all potential outcomes, given the observed treatment and pretreatment covariates (Imai, Keele, and Tingley 2010). Simply, the first assumption means that conditional on a profile of pretreatment covariates, the treatment can be treated as if randomly assigned. Under assumption 2 of sequential ignorability, the model must not omit any potential confounders, or unobserved pretreatment covariates, that affect both the mediator and outcome. Violations of no omitted variables lead to biased estimates, and the bias caused by these unmeasured confounders leads to overstating mediation effects (Bullock, Green, and Ha 2010).

Both of these assumptions are difficult to meet, even in controlled experiments. For example, it would be unrealistic to assume that donor fragmentation should be considered randomly distributed across countries. As discussed above, strategically important countries tend to have higher levels of donor concentration. An approach taken by many researchers to overcome this limitation in nonrandom treatment assignment is to collect as many pretreatment covariates as possible to adjust for selection bias (Imai, Keele, and Tingley 2010; Rosenbaum 2002). Other methods apply Bayesian techniques to calculate identification bounds on the average treatment effect, such as Molinari bounds (Mebane and Poast 2013; Molinari 2010). For our purposes, we begin by including a host of covariates (discussed below under controls) to help mitigate potential selection bias.

A more fundamental difficulty with causal mediation analysis is that there may exist unobserved confounders that causally affect both the mediator and the outcome, even after conditioning on the observed treatment and pretreatment covariates. This second assumption of sequential ignorability is non-refutable, meaning that it cannot be directly tested with the data. Various sensitivity analysis methods allow researchers to quantify the degree to which violations of ignorability undermine causal inference (e.g., Imai, Keele, and Tingley 2010; Mebane and Poast 2013; Quinn 2008; Rosenbaum 2002). While the specific methods vary in their implementation, they allow researchers to answer the same basic question: what happens to my estimated parameters if I simulate the effect of unmeasured confounders? For example, the mediation package provides a simple to implement sensitivity analysis following estimation of direct and indirect effects, but the package is not yet able to perform sensitivity analysis when both the mediator and outcome variables are binary, such as is the case with our setup. However, other methods for conducting sensitivity analysis also exist (Rosenbaum 2002; Quinn 2008).

Thus, while the mediation approach has its limitations, it provides the most direct test of pathway 2. In addition, an initial point estimate of the mediation effect can tell us whether follow-up methods are warranted. Since bias resulting from omitted variables leads to overly optimistic results (Bullock, Green, and Ha 2010), a null finding in the initial estimation alleviates the need for further sensitivity analysis. To preview our findings, we find no evidence of a substantively important mediation effect and accordingly do not further investigate violations of sequential ignorability.

Controls

We include a set of standard control variables into our analysis of conflict onset. Following Fearon and Laitin (2003), we use GDP per capita, population, and a democracy indicator.²⁵ As a measure of the government’s ability to control its territory, we include the country’s land area (this variable is available for more countries than mountain coverage, the measure proposed by Fearon and Laitin (2003)). In addition, we include two indicator variables, which take on the value of 1 if a country was a former British or French colony and 0 otherwise. Because conflict also can arise as a result of exogenous shocks to largely agricultural economies (Miguel, Satyanath, and Sergenti 2004), we also include the number of natural disasters that strike a country per year. Recorded events include floods, famines, droughts, earthquakes, epidemics, and windstorms.²⁶ For our aid shock equation, we reviewed the existing literature. There are surprisingly few works that consider the sources of aid volatility.²⁷ However, a number of variables that have an effect on conflict are also relevant predictors of aid volatility. These include population size, regime type, GDP per capita, and natural disasters (Hudson and Mosley 2008; Desai and Kharas 2010). In addition, we include the real effective exchange rate of the US dollar against a basket of international currencies (REER; The World Bank Group 2011). Since aid data are denominated in US dollars, movements in the REER will have an immediate effect on recorded aid flows (from non-US sources).

Results

In our analysis, we test separate models. The first addresses whether donor fragmentation reduces the negative effects of aid shocks on civil conflict onset (the moderation effect). We begin by estimating the conflict equation (6) as a rare events logit model without the interaction of donor concentration and aid shocks (model 1). Table 1 reports results. They mostly confirm existing findings. As a key variable, aid shocks increase the risk of violent conflict. Control variables behave largely as expected, with wealthier countries and former British colonies on average being more peaceful. Land area and disasters are not statistically significant at the usual levels but point in the right direction. Importantly, donor concentration does not have an independent effect on the likelihood of conflict.

Table 1.

Predicting Violent Conflict in Recipient Countries.

Conflict	1	2	3	4
		Rare events logit		Cloglog
Aid shock	0.478	−0.145	0.0372	0.00814
	(0.204)	(0.255)	(0.308)	(0.313)
Donor concentration	−0.491	−2.14	−0.899	−1.0187
	(0.475)	(0.348)	(0.548)	(0.522)
Aid shock × concentration		2.49	1.689	1.594
		(0.738)	(0.911)	(0.926)
Aid per GDP pc	−1.20 × 10⁶		−1.13 × 10⁶	−1.27 × 10⁶
	(1.24 × 10⁶)		(1.23 × 10⁶)	(1.02 × 10⁶)
Per capita GDP	−0.0291		−0.0273	−0.0290
	(0.0103)		(0.0103)	(0.00820)
Land area	8.38 × 10⁻⁷		7.98 × 10⁻⁷	7.10 × 10⁻⁷
	(6.24 × 10⁻⁷)		(6.20 × 10⁻⁷)	(7.57 × 10⁻⁷)
Democracy	−0.00901		0.00481	−0.00904
	(0.0370)		(0.0372)	(0.0331)
French colony	−0.193		−0.211	−0.218
	(0.201)		(0.202)	(0.175)
British colony	−0.307		−0.335	−0.354
	(0.185)		(0.186)	(0.170)
Population	−1.35 × 10⁻⁹		−1.33 × 10⁻⁹	−1.40 × 10⁻⁹
	(1.21 × 10⁻⁹)		(1.20 × 10⁻⁹)	(1.14 × 10⁻⁹)
Disasters	0.0248		0.0250	0.0190
	(0.0170)		(0.0170)	(0.0156)
Time	−0.428	−0.481	−0.416	−0.398
	(0.0598)	(0.0500)	(0.0600)	(0.0583)
Time²	0.0248	0.0284	0.0241	0.0233
	(0.00462)	(0.00422)	(0.00461)	(0.00466)
Time³	−0.000447	−0.000514	−0.000436	−0.000427
	(0.0000958)	(0.0000917)	(0.000095)	(0.0000999)
N	2,648	3,101	2,648	2,648

Note: Entries are model coefficients with standard errors in parentheses. Bolded coefficients are twice their standard error and bold italics indicate p ≤ .10. GDP = gross domestic product.

In model 2, we add the interaction term between donor concentration and aid shocks, leaving out the control variables. The interaction term is highly statistically significant (p = .001) and has the expected sign. Adding the control variables (model 3) leads to slight decrease in the coefficient and statistical significance (p = .064). Importantly, in both models 2 and 3, the main effect of aid shocks loses statistical significance once we control for the shared variation between donor concentration and shocks. Because of the presence of the interaction term, this means that with perfect fragmentation (HHI = 0), aid shocks do not increase the risk of conflict anymore. To correctly evaluate the effect of aid shocks across different values of donor concentration, we need to vary the HHI variable while keeping other variables constant (Ai and Norton 2003). To this end, we create a representative profile of other regressors (holding them at their median) and calculate the difference in predicted probabilities of conflict with and without aid shocks (i.e., the marginal effect of aid shocks).The baseline risk of conflict in this scenario is quite low, remaining below 5 percent without aid shocks. This is not surprising, since violent political conflict overall is a rare event. To gauge the full potential effect of donor fragmentation, we also create a high-risk profile, including only cases for which the predicted risk of conflict without aid shocks is at least 30 percent.

Figure 4 shows the marginal effect of aid shocks across the range of the donor concentration variable for both the average and high-risk profiles, including 95 percent confidence bands.²⁸ In both scenarios, aid shocks have a strong effect on the risk of conflict when donors are highly concentrated, but this effect becomes indistinguishable from zero for low donor concentrations.²⁹ For the highest observed concentration in the data (HHI = 1, only one donor gives aid), an aid shock strongly increases the probability of conflict onset. A shock adds on average ten percentage points (thirty-five percentage points for high-risk countries) to the probability of conflict. Donor fragmentation strongly reduces this destabilizing effect. For the average donor concentration in the sample (HHI = 0.31), aid shocks are associated with an increase in the risk of conflict of only 2.7 percentage points (10 percent points).

Figure 4.

Effect of aid shock, conditional on donor concentration.

Thus, by moving from a single-donor system to a more typical degree of donor fragmentation, the probability of conflict associated with aid shocks decreases by more than two-thirds (7.3 and 25 percentage points). This is a profound reduction in the risk of political destabilization. The results do not change substantively without the rare events correction. The cloglog model (4) has a slightly different functional form than a logit link and no rare events correction. Yet, the interaction term and effect sizes remain largely unchanged. In addition, our results are robust against varying the time window and percentile cutoffs in the calculation of aid shocks (reported in the online supplement).

We therefore have strong evidence of a large dampening effect of donor fragmentation on how aid shocks affect civil conflict onset (i.e., donor fragmentation moderates the effect of aid shocks downwards). This effect is statistically and substantively strong. To illustrate its relevance, we return to our example from the introduction. In 1976, developmental aid to Zaire was fairly concentrated (HHI = 0.44). In reaction to the internal conflicts of 1977 and 1978, the Western donor community started to engage more broadly, resulting in more fragmented aid delivery (HHI = 0.30 in 1979 and 0.31 in 1980). According to our model, in 1976, Zaire had a baseline risk of conflict of around 8.9 percent, which an aid shock would roughly double to 16 percent. Had aid been fragmented to the extent that it was after the crisis, the increase in risk from an aid shock would have been limited to 4.6 percentage points, about 50 percent lower than in 1976.³⁰

We now turn to our second model, which relates donor fragmentation to a decrease in the frequency of aid shocks, and a resulting reduction in the risk of violent conflict onset (the mediation effect). We find evidence for a statistically significant but substantively very modest indirect effect of donor concentration through aid shocks. The results are reported in Table 2. First, we do find that donor concentration is significantly and positively related to an increased likelihood of an aid shock, and this effect is also substantial.³¹ When holding all covariates at their median values, moving from the most fragmented to the most concentrated donor system increases the probability of experiencing an aid shock by six percentage points on average (over a 13 percent baseline probability of an aid shock). The results for the conflict equation are substantively the same as in model (1), though effect size and statistical significance of aid shocks decrease.³² Turning to mediation, as shown in Table 2, the indirect effect of concentration through the increased probability of aid shocks is statistically significant. However, this effect is substantively very small. Moving from low to high donor concentration, the probability of conflict corresponding to an aid shock increases by a mere 0.368 percentage points (average causal mediation effect [ACME]). In other words, even though donor concentration increases the probability of aid shocks, and aid shocks in turn increase the likelihood of conflict, the indirect effect of concentration through aid shocks is negligible.

Table 2.

Mediated Effects of Donor Fragmentation on Violent Conflict.

	Aid shock	Conflict
Donor concentration	0.468	−0.277
	(0.226)	(0.234)
Aid shock		0.260
		(0.109)
Real exchange rate, USD	−0.00890
	(0.00169)
Aid per GDP pc	8.057 × 10⁶	−1.103 × 10⁶
	(6.081 × 10⁵)	(5.876 × 10⁵)
Per capita GDP	−0.00152	−0.0150
	(0.00536)	(0.00382)
Land area	1.65e-07	3.041e-07
	(5.55e-07)	(3.870e-07)
Democracy	−0.0440	0.0132
	(0.0217)	(0.0177)
French colony	0.116	−0.161
	(0.0942)	(0.0952)
British colony	0.166	−0.175
	(0.0981)	(0.0892)
Population	−1.609 × 10⁻⁹	−7.487 × 10⁻¹⁰
	(1.316 × 10⁻⁹)	(5.497 × 10⁻¹⁰)
Disasters	−0.0127	0.0156
	(0.0139)	(0.00872)
Time	−0.157	−0.244
	(0.0317)	(0.0305)
Time²	0.00777	0.0148
	(0.00421)	(0.00242)
Time³	−0.000114	−0.000142
	(0.0000490)	(0.0000517)
Mediated effects of donor concentration		90 percent confidence interval
Total effect	−0.0365	[−0.0373, 0.0254]
Indirect effect (ACME)	0.00368	[0.000151, 0.00836]
N	2,541	2,541

Note: Coefficients are estimated using probit and standard errors are in parentheses. Bolded coefficients are twice their standard error and bold italics indicate p ≤ .10, two-tailed tests. The indirect effects are the change in the probability of conflict corresponding to a change in the value of an aid shock produced by moving from presence to absence of a shock. GDP = gross domestic product; ACME = average causal mediation effect; USD = US dollar.

Given that bias from omitted variables favors finding mediated effects, as discussed above, together with the substantively meager mediation results we estimate, we find it unnecessary to move forward with more sophisticated (e.g., Molinari bounds) or follow-up methods (e.g., sensitivity analysis). Rather, we conclude that while donor concentration increases aid volatility, this does not substantively affect conflict through this causal pathway.

Summing up our results, the evidence provides solid support for our argument that donor fragmentation provides insurance against short-term political fallout from aid shocks. The causal pathway through which this occurs is moderation. That is, if aid is delivered in a more fragmented fashion, aid shocks lose much of their politically destabilizing force. We have presented some preliminary evidence that fragmentation changes the composition of aid flows, pointing to a more developmental and public goods oriented outlook of aid delivery. A better understanding of the mechanisms that makes fragmented aid less politically destabilizing awaits further research. Overall, our findings suggest that students of donor fragmentation and aid effects should pay closer attention to the composition of aid flows and the modes of aid delivery. Considering the second causal pathway, we found some narrow evidence that diversified donor portfolios reduce the incidence of aid shocks and the resulting risk in conflict onset (mediation effect). However, in our analysis, this effect appears too small to be of substantive importance to researchers and policy makers.

Conclusion

The analysis shows that donor fragmentation can effectively insulate a recipient country from the risk of political destabilization by moderating the negative effects of aid shocks. Our findings therefore highlight that donor fragmentation can have positive side effects. We believe that our work is the first to identify such a positive side effect.

The consequences of our findings for aid policy are important and call for more research on the link between donor fragmentation and the composition of aid flows and aid volatility. There are at least two areas that should receive attention. First, donor fragmentation protects against the immediate political downside that arises from short-term changes in aid flows. Critics of donor fragmentation have focused on longer-term processes, such as the deterioration of institutional capacity in aid recipient countries. This suggests that there exists a trade-off between short-term gains and long-term harm from donor fragmentation. It is an open empirical question which of the two sides is more prevalent. More research is needed to understand the trade-off and under which circumstances short-term or long-term effects prevail. Such work should shed light on the possibility of mitigating the negative long-term consequences, while preserving short-term benefits.

Second, political destabilization is not the only negative effect of aid volatility. In addition to reducing the political risk associated with aid shocks, donor fragmentation might help to alleviate problems arising from aid shocks for economic growth. Similar to political stability and institutional performance, a trade-off likely exists for the effects of fragmentation on short-term versus long-term economic growth. A full assessment of the benefits and costs of donor fragmentation should take account of this issue area as well.

From a larger theoretical perspective, the dual nature of fragmentation as overburdening recipient bureaucracies (an externality) and providing insurance against aid shocks (a public good) is not unique to donor fragmentation. Many forms of international cooperation are welfare improving in some aspects but also generate externalities. Regional trade agreements, for example, benefit participants but impose costs on neighboring countries not covered by the agreement. In this situation, cooperation drives a wedge between the benefactors of cooperation and those on the outside. For donor fragmentation, the situation should be in principle more beneficial since the key actors have a stake in avoiding externalities. A more actor-centric theoretical framework that pays attention to the incentives of individual aid providers could help cast more light on this specific international cooperation problem.

Footnotes

Acknowledgment

The authors would like to thank Desha M. Girod, Matt Lebo, Gabriella Montinola, Richard A. Nielsen, John Tuman, Matthew S. Winters, and two anonymous reviewers. Taylor Grant provided excellent research assistance.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material

The supplemental materials are available at .

Notes

References

Acemoglu

Daron

Robinson

James A.

. 2000. “Why Did the West Extend the Franchise? Democracy, Inequality, and Growth in Historical Perspective.” Quarterly Journal of Economics 115 (4): 1167–99.

Agénor

Pierre-Richard

Aizenman

Joshua

. 2010. “Aid Volatility and Poverty Traps.” Journal of Development Economics 91:1–7.

Chunrong

Norton

Edward C.

. 2003. “Interaction Terms in Logit and Probit Models.” Economics letters 80:123–29.

Aldasoro

Inaki

Nunnenkamp

Peter

Thiele

Rainer

. 2010. “Less Aid Proliferation and More Donor Coordination? The Wide Gap between Words and Deeds.” Journal of International Development 22:920–40.

Alt

James E.

King

Gary

Signorino

Curtis S.

. 2001. “Aggregation among Binary, Count, and Duration Models: Estimating the Same Quantities from Different Levels of Data.” Political Analysis 1:21–44.

Arcand

Jean-Louis

Chauvet

Lisa

. 2001. “Foreign Aid, Rent-seeking Behavior and Civil War.” Working Paper, CERDI-CNRS, Université d’Auvergne, Clermont-Ferrand, France.

Arellano

Cristina

Bulir

Ales

Lane

Timothy

Lipschitz

Leslie

. 2009. “The Dynamic Implications of Foreign Aid and Its Variability.” Journal of Development Economics 88:87–102.

Beck

Nathaniel

Katz

Jonathan N.

Tucker

Richard

. 1998. “Taking Time Seriously: Time-series-cross-section Analysis with Binary Dependent Variable.” American Journal of Political Science 42:1260–88.

Blattman

Christopher

Miguel

Edward

. 2008. “Civil War.” Journal of Economic Literature 48:3–57.

10.

Bueno de Mesquita

Bruce

Smith

Alastair

. 2009. “Political Survival and Endogenous Institutional Change.” Comparative Political Studies 42:167–97.

11.

Bulíř

Aleš

Hamann

Javier

. 2003. “Aid Volatility: An Empirical Assessment.” IMF Staff Papers 50:64–89.

12.

Bulíř

Aleš

Hamann

Javier

. 2007. “Volatility of Development Aid: An Update.” IMF Staff Papers 54:727–39.

13.

Bullock

John G.

Green

Donald P.

Shang E.

. 2010. “Yes, but What’s the Mechanism? (Dont Expect an Easy Answer).” Journal of Personality and Social Psychology 98 (4): 550.

14.

Carter

David B.

Signorino

Curtis S.

. 2010. “Back to the Future: Modeling Time Dependence in Binary Data.” Political Analysis 18 (3): 271–92.

15.

Collier

Paul

Elliott

V. L.

Hegre

Håvard

Hoeffler

Anke

Reynal-Querol

Marta

Sambanis

Nicholas

. 2003. Breaking the Conflict Trap: Civil War and Development Policy. Washington, DC: The World Bank.

16.

Desai

Raj M.

Kharas

Homi

. 2010. “The Determinants of Aid Volatility.” Brookings Global Economy and Development Working Paper 42, Global Economy and Development Program, the Brookings Institution, Washington, DC.

17.

Dietrich

Simone

. 2013. “Bypass or Engage? Explaining Donor Delivery Tactics in Foreign Aid Allocation.” International Studies Quarterly 57:698–712.

18.

Djankov

Simeon

Montalvo

Jose G.

Reynal-Querol

Marta

. 2009. “Aid with Multiple Personalities.” Journal of Comparative Economics 37:217–29.

19.

Dreher

Axel

Nunnenkamp

Peter

Thiele

Rainer

. 2008. “Does US Aid Buy UN General Assembly Votes? A Disaggregated Analysis.” Public Choice 136:139–64.

20.

Easterly

William

. 2007. “Are Aid Agencies Improving?” Economic Policy 22:633–78.

21.

Fearon

James D.

1995. “Rationalist Explanations for War.” International Organization 49 (3): 379–414.

22.

Fearon

James D.

2004. “Why Do Some Civil Wars Last So Much Longer Than Others?” Journal of Peace Research 41:275–301.

23.

Fearon

James D.

Laitin

David D.

. 2003. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97:75–90.

24.

Findley

Michael G.

Powell

Josh

Strandow

Daniel

Tanner

Jeff

. 2011. “The Localized Geography of Foreign Aid: A New Dataset and Application to Violent Armed Conflict.” World Development 39:1995–2009.

25.

Freedom House. 2010. “Freedom in the World: Country Ratings by Region.” Accessed May 3, 2010. http://www.freedomhouse.org/template.cfm?page=439.

26.

Frot

Emmanuel

Santiso

Javier

. 2010. “Crushed Aid: Fragmentation in Sectoral Aid.” Working Paper 248, Development Centre, OECD, Paris, France.

27.

Girod

Desha M.

2012. “Effective Foreign Aid Following Civil War: The Nonstrategic-desperation Hypothesis.” American Journal of Political Science 56:188–201.

28.

Gleditsch

N. P.

Wallensteen

Eriksson

Sollenberg

Strand

. 2002. “Armed Conflict 1946-2001: A New Dataset.” Journal of Peace Research 39:615–37.

29.

Grossman

H. I.

1992. “Foreign Aid and Insurrection.” Defence Economics 3:275–88.

30.

Heinrich

Tobias

Kobayashi

Yoshiharu

. 2009. “Foreign Aid and the Tenure of Leaders.” Paper presented at Midwest Political Science Association National Conference, Chicago, IL.

31.

Hodrick

Robert J.

Prescott

Edward C.

. 1997. “Postwar US Business Cycles: An Empirical Investigation.” Journal of Money, Credit & Banking 29:1–16.

32.

Hudson

John

Mosley

Paul

. 2008. “Aid Volatility, Policy and Development.” World Development 36:2082–102.

33.

Imai

Kosuke

Keele

Luke

Tingley

Dustin

. 2010. “A General Approach to Causal Mediation Analysis.” Psychological Methods 15:309.

34.

Kelly

Sean

. 1993. America’s Tyrant: The CIA and Mobutu of Zaire. Washington, DC: American University Press.

35.

Kharas

Homi

. 2008. “Measuring the Cost of Aid Volatility.” Wolfensohn Center for Development Working Papers 3, Washington, DC.

36.

King

Gary

Zeng

Langche

. 2001. “Logistic Regression in Rare Events Data.” Political Analysis 9:137.

37.

Knack

Stephen

Rahman

Aminur

. 2007. “Donor Fragmentation and Bureaucratic Quality in Aid Recipients.” Journal of Development Economics 83:176–97.

38.

Lancaster

Carol

. 2007. Foreign Aid: Diplomacy, Development, Domestic Politics. Chicago: University of Chicago Press.

39.

Lensink

Robert

Morrissey

Oliver

. 2000. “Aid Instability as a Measure of Uncertainty and the Positive Impact of Aid on Growth.” Journal of Development Studies 36:31–49.

40.

Mebane

Walter R.

Poast

Paul

. 2013. “Causal Inference without Ignorability: Identification with Nonrandom Assignment and Missing Treatment Data.” Political Analysis 21 (2): 233–51.

41.

Miguel

Edward

Satyanath

Shanker

Sergenti

Ernest

. 2004. “Economic Shocks and Civil Conflict: An Instrumental Variables Approach.” Journal of Political Economy 112:725–53.

42.

Molinari

Francesca

. 2010. “Missing Treatments.” Journal of Business & Economic Statistics 28 (1): 82–95.

43.

Morrison

Kevin M.

2007. “Natural Resources, Aid, and Democratization: A Best-case Scenario.” Public Choice 131:365–86.

44.

Morss

Elliott R.

1984. “Institutional Destruction Resulting from Donor and Project Proliferation in Sub-Saharan African Countries.” World Development 12:465–70.

45.

Nielsen

Richard A.

Findley

Michael G.

Davis

Zachary S.

Candland

Tara

Nielson

Daniel L.

. 2011. “Foreign Aid Shocks as a Cause of Violent Armed Conflict.” American Journal of Political Science 55:219–32.

46.

OECD (Organization for Economic Cooperation and Development). 2009. International Development Statistics on CD ROM. Paris, France: OECD.

47.

Paine

Jack

. 2014. “A Blessing and a Curse: How Oil Impacts Center-seeking and Separatist Civil Wars.” Paper presented at Annual Conference of the International Society for New Institutional Economics. Accessed December 2, 2014. http://extranet.isnie.org/uploads/isnie2014/paine.pdf.

48.

Pallage

Stéphane

Robe

Michael A.

. 2001. “Foreign Aid and the Business Cycle.” Review of International Economics 9:641–72.

49.

Quinn

Kevin M.

2008. “What Can Be Learned from a Simple Table? Bayesian Inference and Sensitivity Analysis for Causal Effects from 2× 2 and 2 × 2 × K Tables in the Presence of Unmeasured Confounding.” Working Paper, Harvard University, Cambridge, MA. Accessed July 15, 2015. http://polmeth.wustl.edu/media/Paper/Causal2by2.pdf.

50.

Rosenbaum

Paul R.

2002. “Covariance Adjustment in Randomized Experiments and Observational Studies.” Statistical Science 17 (3): 286–327.

51.

Savun

Burcu

Tirone

Daniel C.

. 2012. “Exogenous Shocks, Foreign Aid, and Civil War.” International Organization 66:363–93.

52.

Steinwand

Martin C.

2014. “Foreign Aid and Political Stability.” Conflict Management and Peace Science. Accessed July 15, 2015. http://cmp.sagepub.com/content/early/2014/07/17/0738894214541227. doi:10.1177/0738894214541227.

53.

Steinwand

Martin C.

2015. “Compete or Coordinate? Aid Fragmentation and Lead Donorship.” International Organization 69 (2): 443–72. doi:10.1017/S0020818314000381.

54.

The World Bank Group. 2011. “World Development Indicators Online.” Accessed February 24, 2011. http://databank.worldbank.org/ddp/home.do?Step=12&id=4&CNO=2.

55.

Tingley

Dustin

Yamamoto

Teppei

Hirose

Kentaro

Keele

Luke

Imai

Kosuke

. 2014. “Mediation: R Package for Causal Mediation Analysis.” Journal of Statistical Software 59 (5): 1–38.

56.

Université catholique de Louvain. 2012. EM-DAT: The OFDA/CRED International Disaster Database. Brussels, Belgium. Accessed March 12, 2012. www.emdat.be.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.55 MB

0.00 MB