Bypassing the Incumbent: Leadership Tenure and Foreign Aid Channels

The traditional perception is that foreign aid provides leaders in recipient states access to nontax revenue, which incumbents may be able to manipulate in order to provide public or private goods as is required to maintain ofﬁce. Recognizing this possibility, donors have been shifting away from direct government-to-government aid and toward bypass aid, administered by NGOs and civil society groups rather than the regime directly. Receiving aid through alternative channels is designed to increase aid’s effectiveness, yet it has potential implications for incumbent leaders who may have less ability to manipulate aid receipts to their beneﬁt. In this paper, we argue that bypass aid should negatively affect leaders’ tenure, conditional on the political institutions in place. Using data on bypass aid from 2004 to 2017, we ﬁnd evidence of a punishing effect and conclude that bypass aid is a particularly acute problem for leaders in authoritarian regimes. The ﬁndings imply that efﬁciency and stability may present a tradeoff for potential donors and that bypass aid may not be the panacea of the aid effectiveness movement.


Introduction
States give billions of dollars in foreign aid each year with the goals of fueling economic growth and eradicating poverty in developing countries.To maximize aid's effectiveness, the last two decades have witnessed a fundamental shift in the aid allocation process.During the Cold War, most foreign aid was given directly to recipient governments.As the political climate shifted and donors' concerns about aid leakage and corruption grew, they allocated a greater share of aid monies to particular projects in order to increase oversight.More recently, donor governments have funneled an increasing amount of these resources through nongovernmental and civil society organizations (NGOs and CSOs), effectively bypassing the role of corrupt or inefficient recipient governments altogether.Research suggests that the decision to allocate aid through bypass channels is influenced by both recipient (Dietrich 2013;Acht, Mahmoud, and Thiele 2015;Dietrich and Murdie 2017) and donor characteristics (Dietrich 2016;Allen and Flynn 2017).Aid that is implemented by nongovernment actors may have important implications for the relationship between a recipient government and its citizens (Baldwin and Winters 2020).Yet, little is known about the longer-term political effects of different aid delivery strategies.
One potential ramification of alternative aid modalities is that foreign aid can help incumbent politicians in recipient countries maintain their hold on power.Previous work has demonstrated that aid has an impact on the survival of leaders, but not all leaders are affected in the same way (Wright 2008;Kono and Montinola 2009;Licht 2010;Ahmed 2012).Despite anecdotal evidence, foreign aid does not appear to prop up dictatorships in every instance; aid enhances survival only under specific circumstances.Building on this prior work, we explore how, and under what circumstances, aid delivery strategies that bypass the recipient government might influence a leader's ability to stay in power.
We argue that the impact of foreign aid on leader survival is conditioned on the way aid is allocated.While the fungibility of foreign aid varies (Feyzioglu, Swaroop, and Zhu 1998;van de Walle and Mu 2007;Bermeo 2016Bermeo , 2017)), the expectation is that direct government-to-government transfers and government dispersed project aid are easier for leaders to manipulate than aid allocated through nonstate actors.When aid is more fungible, leaders can use foreign assistance to provide a combination of private and public goods to co-opt support or pay for repression to maintain power.As more aid is awarded through NGOs versus traditional bilateral channels, such assistance is harder to manipulate in order to abet leaders' ulterior motives-it cannot be easily shifted toward the provision of goods to key supporters.Given the same amount of total aid but with a diminished ability to share the fruits of that aid with supporters, where the proportion of foreign aid allocated through bypass channels is higher, leaders will be at a greater risk of turnover.Using data on aid allocations from 2004 to 2017, we find evidence of a punishing effect and conclude that increasing bypass aid is a particularly acute problem for leaders in autocratic regimes, where private goods provision is more important.
While bypass aid projects are not designed to have a political impact, when aid is designated through this alternate channel, there are likely to be political consequences.For the most part, these impacts have gone unexplored, and we contribute to the literature on bypass aid in several ways.First, we offer one of the first cross-national empirical studies with bypass aid as the explanatory variable. 1Traditionally, bypass aid has been the outcome of interest (Dietrich 2013(Dietrich , 2016;;Acht, Mahmoud, and Thiele 2015;Allen and Flynn 2017) or the effects of bypass aid have been studied within specific countries (Dietrich and Wright 2015;Winters, Dietrich, and Mahmud 2018;Baldwin and Winters 2020).Secondly, our findings imply that the relationship between foreign aid and leadership is more nuanced than has previously been assumed; it is conditional on both aid modality and recipient political institutions.Moreover, we show that the effects of aid modality on tenure are different and independent of the effects of government-to-government aid.Finally, we point to a potential unintended consequence of bypass aid.For donors, allocating aid through NGOs presents a tradeoff.Bypassing recipient governments by using NGOs can increase aid effectiveness, but it might also exacerbate political instability in developing countries.

The Logic of Bypass Aid
Foreign aid is an inducement for recipient countries-one that offers the possibility of additional, unearned resources, which may help leaders maintain power.With aid, leaders may be able to satisfy core constituents, but leaders' ability to direct those funds varies.While much of the state-tostate aid that is awarded to recipient nations comes in the form of project aid rather than budgetary support (Jelovac and Vandeninden 2014), when funds are earmarked for specific development projects and the recipient government is the implementing agency, there is the potential for leakage, either through aid capture or bureaucratic inefficiency (Brautigam and Knack 2004;Reinikka and Svensson 2004;Djankov, Montalvo, and Reynal-Querol 2008b).Misuses of aid funds and heightened attention to the importance of recipient good governance have led to a growing belief among donors that public services can be supplied more effectively by nonstate actors without negatively affecting the foreign population in need of assistance (Reinikka and Svensson 2010). 2mid concerns that foreign aid stalls development and that it abets authoritarian leaders in aid-receiving countries, scholarship on donor use of alternative aid channels illustrates a shift in donor practices to contain mismanagement.Many see it as an article of faith that NGOs are closer to the poor than official aid agencies (Tendler 1982;Edwards and Hulme 1998), and allocation of aid through NGO channels is perceived to be less distorted by commercial and political mandates.Donor governments appear to share the view that NGOs have an important role to play, and an increasing percentage of bilateral official aid is administered through NGOs, bypassing the recipient government.Between 2004 and 2017, the average share of bilateral ODA allocated through NGOs and CSOs rose from about 5 percent of the total aid given to almost 15 percent. 3Findings from studies that explore the effects of variation in donor practice across aid sectors and the use of aid delivery mechanisms on development outcomes in recipient countries indicate enhanced donor selectivity with respect to recipient country governance (Winters and Martinez 2015;Ferry, Hafner-Burton, and Schneider 2020).
Research on aid bypass is largely motivated by the understanding that foreign aid allocation is not predominantly driven by strategic concerns (Alesina and Dollar 2000) or policy concessions (Bueno de Mesquita and Smith 2007).Donors are presumed to be committed to poverty reduction either as a goal in itself or for producing global public goods such as peace and security (Dietrich 2013;Acht, Mahmoud, and Thiele 2015;Bermeo 2017).NGOs' aid allocations are expected to be less distorted by the political interests of donor governments (Nancy and Yontcheva 2006) and more directly focused on poverty alleviation, as NGOs are often better positioned to deal directly with local populations (Riddell et al. 1995).Subsequently, donor countries find it more efficient to outsource higher proportions of foreign aid to nonstate actors in aid-receiving countries with lower quality of governance (Dietrich 2013;Knack 2014;Acht, Mahmoud, and Thiele 2015;Bermeo 2017).In recent years, the number of service delivery NGOs and donor support to these organizations has expanded significantly, especially in comparison to advocacy NGOs (Kabeer, Mahmud, and Castro 2012;Lewis 2017).
Despite benevolent intentions, the outsourcing of aid still involves political and strategic considerations.Dietrich (2016) analyzes the role of donor political economies in determining the choice of aid delivery channels.She finds that donors that prefer a wider role for the market in service delivery are more likely to opt for aid bypass in countries with weak state institutions.Relatedly, Allen and Flynn (2017) find evidence that right-wing governments prefer state-tostate aid delivery in recipient countries while more liberal governments choose to bypass recipient governments.
Donors are also attuned to public opinion on aid effectiveness, and this has led donors to condition aid decisions in favor of disbursement through nonstate actors (Milner 2006;Knack 2013Knack , 2014)).Dietrich and Murdie (2017) find that international NGOs' shaming of recipient governments as human rights violators biases public opinion against stateto-state aid delivery, thus increasing the proportion of bypass aid given to those recipient countries.Alternatively, Adhikari (2019) finds evidence of heterogeneity in donor behavior on when to bypass; major power donors are less swayed by recipient needs and are more likely to use government channels for strategic considerations.
Bypass aid is likely to have political effects in the recipient country.Previous work has focused primarily on how such aid might influence perceptions of state legitimacy.Experimental studies find evidence that upon learning about the source of funding, citizens have a stronger preference for foreign aid-funded projects relative to government programs in Uganda (Milner, Nielson, and Findley 2016;Findley et al. 2017;Springman 2022), and better perceptions of service quality in Bangladesh (Winters, Dietrich, and Mahmud 2017).Likewise, information on donor funding of projects is found to be positively associated with perceptions of state legitimacy, local government performance, and quality of governance among Indian and Bangladeshi respondents (Dietrich and Wright 2015;Dietrich, Mahmud, and Winters 2018;Winters, Dietrich, and Mahmud 2018).Findings from Sacks (2012) also reflect that citizens' perceptions of a greater role for foreign donors in assisting recipient countries strengthen the fiscal contract in sub-Saharan African countries.Using nonstate channels for aid delivery does not appear to undermine citizens' legitimating beliefs in recipient countries with limited state capacity, but does it affect the leader's ability to remain in power?

The Political Impact of Bypass Aid
When foreign aid is delivered through NGOs rather than through bilateral transfers, there are likely to be political consequences.NGOs primarily engage in service provision, which can result in free or subsidized services for needy individuals and households in society (Lewis 2017).In this way, NGOs can deliver services in a way much more similar to governments than private service providers.For example, Brass (2016) finds that service provision by NGOs in Kenya improved perceptions of government legitimacy and expanded the reach of the state.Here, we develop a theory about how aid that bypasses the recipient government and is delivered by NGOs could affect the tenure of incumbent leaders.How does allocating a higher proportion of aid through bypass channels affect leaders?
Our argument about the effect of aid modalities on leadership tenure borrows from the foundations of Bueno de Mesquita et al. (2005)'s selectorate theory.First, we assume that the primary goal of political leaders is to gain and retain power.Second, we assume that in order to achieve this objective, leaders require support from a segment of the population, referred to as the winning coalition (W).The size of the winning coalition varies across domestic political institutions, containing elite generals in military juntas and half the electorate in democracies.Leaders provide goods to their winning coalitions to retain loyalty and limit defections.
In addition to desiring to stay in power, leaders are utility maximizers; they optimize the efficiency of goods provision based on their available revenue, which is scarce.The original conception of the selectorate model is based on tax revenues; however, tax revenues cannot be raised without negatively impacting the population's support.The model has since been expanded to incorporate other forms of unearned income (Bueno de Mesquita and Smith 2007).One source of revenue is foreign aid.Many developing countries are highly dependent on foreign aid, and many low-income countries receive foreign aid upwards of 10 percent of their gross national income.Crucial to our argument, foreign aid benefits leaders by relaxing their resource constraints and allowing them to substitute foreign aid for domestic expenditures.Incumbents are office seeking, and this implies that leaders should use aid to augment the support of their winning coalition, either by purchasing loyalty or by repressing the opposition (McDonald 2011).
Recent scholarship highlights that fungibility varies temporally (Bermeo 2016) and sectorally (Feyzioglu, Swaroop, and Zhu 1998;van de Walle and Mu 2007).While fungibility may also be driven by the types of donors providing aid (Ferry, Hafner-Burton, and Schneider 2020), the potential for aid capture is not entirely abated in the "good governance era."In the post-Cold War era, foreign aid has largely progressed from direct budget support to projectspecific aid.It is also more closely monitored than in previous eras and governed by anticorruption protocols; yet, governments are still granted leeway by virtue of being the implementing agent. 4Project-based aid money still flows through the government's hands.This may allow for some aid-capture of project-based aid (Andersen, Johannesen, and Rijkers 2022).As a less blatant violation, this also allows the recipient government to award contracts, locate projects in key electoral districts, etc. in order to maximize key electoral benefits (Briggs 2012(Briggs , 2014;;Jablonski 2014;Öhler and Nunnenkamp 2014;Dreher et al. 2019).In these cases, the flexibility of government-implemented foreign aid abets the government's office-seeking motivations without necessarily being flagged as divergent from the aid's intended purposes.
The foreign aid community has increasingly allocated aid to developing countries through NGOs, effectively bypassing the role of the recipient government.Holding constant the total amount of aid provided, when aid is awarded through these alternative actors, the administration of aid projects is outside of the government's control.Importantly, elite surveys in Ghana find that government officials themselves believe they have less influence over and knowledge of foreign aid when it is allocated through NGOs (de la Cuesta et al. 2019).Therefore, for a given amount of foreign assistance, funds distributed through bypass channels are not flexible enough to abet a recipient government's ulterior motives, and the incumbent is less able to use foreign aid to augment the provision of goods, of any type, to satisfy the winning coalition.The governments' cooptation strategies become less effective.
The first implication of our argument is that bypass aid decreases government control over funds.All leaders-even the most accountable-would prefer greater control over how the aid is spent.Ideally, every leader would like to determine how benefits are shared, deciding which goods to provide and to whom.When more aid is awarded through bypass channels, leaders become constrained.Where more aid is allocated through NGOs, incumbents should be less successful in satisfying their winning coalition and maintaining office, compared to incumbents who receive more aid through government-to-government channels.The greater the amount of foreign aid that is bypassed, the greater the threat to the leader's incumbency.
H 1 : An increase in foreign aid allocated through bypass channels (relative to GDP) will increase the risk of leader turnover.This is not to say that recipient governments do not still receive benefits from bypass aid.While aid delivery through nonstate development actors may limit government control, recipient incumbents can still use nonstate service delivery for claiming undue credit.Baldwin and Winters (2020) find that the majority of Ugandans believe that the government plays a central role in NGO-provided projects.This accords with Cruz and Schneider (2017) who find that incumbent mayors in the Philippines use projects administered by the World Bank to claim credit for ensuring reelection.Bypass aid also increases support for leaders in Bangladesh (Shammama 2022).The political effects of bypassing recipient governments may even manifest in the form of lower protest activities in authoritarian regimes (DiLorenzo 2018).Our argument does not dispute that strategic politi-cians can use NGO-provided aid, particularly NGO-provided service delivery, to their benefit.Instead, it makes the more specific claim that, given the choice, governments would prefer to receive more flexible forms of aid so that they can direct the benefits to the groups or locations that are of the most electoral importance.The more governments are able to generate office-seeking benefits from bypass aid, the more this biases against finding support for our argument.
The effect of bypass aid on political support should also depend on the political institutions in place.Access to foreign assistance matters for leadership tenure, especially for authoritarian leaders (Smith 2008;Djankov, Montalvo, and Reynal-Querol 2008a;Morrison 2009;Bueno de Mesquita andSmith 2009, 2010;Licht 2010;Ahmed 2012).When receiving aid is conditional on political reforms, however, even leaders in these regimes may democratize (Wright 2009), especially when the donors themselves are democracies (Bermeo 2011).While there remains scholarly debate about the direction of institutional effects, previous work suggests that the effect of foreign aid on leader survival might vary by the domestic regime type of aid-receiving countries (Kono and Montinola 2009).Variation in the choice of aid modality might be similarly differentiated.
Most broadly, access to funding is a critical issue for autocratic incumbency.Empirically, reduced access to credit and credit-rating downgrades hurt autocratic leaders' holds on power more than their democratic counterparts (DiGiuseppe andShea 2015, 2016).On the positive side, the existence of foreign aid and remittances help autocrats more as they can utilize the unearned income-both to the government and to individual households-more successfully (Morrison 2009;Ahmed 2012).The same is true for oil rents (Omgba 2009) and other nonlootable resources (Andersen and Aslaksen 2013), although Bermeo (2016) acknowledges that these analogies are no longer as applicable to foreign aid.
More specifically, one of the canonical conclusions derived from the selectorate model is that the size of the winning coalition determines what type of goods leaders provide-public versus private.Private goods are enjoyed only by members of the winning coalition, whereas public goods are nonexcludable and are enjoyed by all.Therefore, where the winning coalition is small, as in an autocracy, a leader only needs the support of a few individuals.If the leader allocates the majority of their resources to private goods, then each member of the winning coalition receives a large proportion of the available transfers, and their support for the leader remains strong.Where the winning coalition is large, as in democracies, transferring resources to each individual member of the winning coalition becomes prohibitively costly.Democratic leaders receive a greater benefit from providing public goods, which enrich all citizens collectively.Consequently, autocracies allocate more resources to private goods, while democracies allocate more resources to public goods.Increased bypass aid can affect nondemocratic leaders in two different ways.First, we consider negative effects for the most autocratic leaders.
From the above logic, foreign aid is linked to leader tenure because it is a particularly useful tool for leaders wishing to provide individuals with private benefits.In small W systems, foreign aid enables the government to have more discretionary income to distribute.Leaders can provide each member of the small winning coalition with a greater individual payoff, increasing their loyalty to the incumbent regime.When states receive aid, it serves to exacerbate the difference in payoffs between the winning coalition and the selectorate (S) in such states.This dynamic is the basis for the aid-curse argument that foreign assistance provides funds that prop up autocratic regimes.Autocrats who receive foreign aid, under certain conditions, experience an improvement in their leadership tenure (Kono and Montinola 2009;Licht 2010). 5imilar to when unearned income declines due to commodity price drops or negative aid shocks, autocrats might be unable to deliver expected private goods as aid modality shifts (Nielson et al. 2011;Arezki et al. 2021).Supporters make retrospective judgments about how well-off they are under the regime in power (Kono and Montinola 2009).When negative aid shocks occur, governments find themselves with much less revenue than anticipated.Unable to provide goods to supporters, autocratic leaders run a greater risk of losing power.Aid volatility and negative aid shocks, in particular, can have a range of severe economic (Bulir andHamann 2003, 2008;Eifert and Gelb 2008) and political consequences (Nielson et al. 2011;Asongu and Nnanna 2019).
A shift in aid modalities toward more bypass aid should have the same impact on autocratic leaders as a negative aid shock.In order to guarantee the loyalty of their supporters, autocratic leaders need money to provide private goods.If the mechanism of aid allocation switches to NGOs, where it used to be administered directly through the recipient government, autocratic leaders may struggle to manipulate unearned revenue to supporters via private goods provision.When members of the winning coalition feel that their material well-being is diminished, the likelihood of shifting their support to the challenger increases.While the rational response of leaders may be to limit this effect by restricting NGO activity, there is no evidence that autocrats receive less bypass aid than democracies.6DiLorenzo (2018) and Springman (2022)'s work also suggests that autocrats' incentives to curtail bypass are not clear-cut.Bypass aid may be less flexible in appeasing autocratic elites via private goods provision, but it does provide a tool to appease the masses.In other words, autocrats may still claim credit, but their ability to buy the support of elites-their selectorate-is weakened.
On the other hand, democratic leaders are less likely to be able to manipulate aid to their advantage.First, the large size of their winning coalition makes private goods provision inefficient.Instead, it implies that democratic governments must devote all of their resources from foreign assistance to public goods, which benefit both members and nonmembers of the winning coalition alike and narrow the payoff differential between the winning coalition and the selectorate.Second, even if foreign aid gave leaders in large W systems the resources to provide sufficient individual transfers, the heightened level of checks and balances in political institutions would make it more difficult to target supporters over nonsupporters.This rationale provides the basis for the "good governance" initiative in foreign aid (Burnside and Dollar 2000).
As NGO provision is intended to increase aid's effectiveness, aid that bypasses the recipient government in democracies will provide the same, if not greater, benefits to incumbent politicians.It honors the objectives that democratic leaders with large winning coalitions are trying to achieve with their fiscal expenditures.Increasing the percentage of bypass aid should have little impact on leaders who already rely primarily on public goods to stay in power.
Therefore, the effect of bypass aid on leader tenure should be conditioned on the political institutions in place.
If it is about the size of the winning coalition, greater bypass aid should be particularly problematic for autocratic regimes, increasing the probability of leader turnover.
H 2 : An increase in foreign aid allocated through bypass channels (relative to GDP) will increase the risk of leader turnover more in autocratic regimes compared to other types of leaders.
Another potential way to interpret the effect of aid on domestic political institutions relates to the importance of the loyalty norm in the original selectorate model.The model argues that the payment necessary to keep members of the winning coalition loyal may vary with the stability of political institutions.For example, the payment required to maintain loyalty in a pure autocracy, where dictators enjoy a long time horizon, is lower than in mixed regimes because the relative likelihood of being excluded from the winning coalition is higher (Bueno de Mesquita et al. 2005).As a result, a shift in aid money toward bypass channels might hurt leaders less in consolidated authoritarian regimes than in mixed regimes.
Bueno de Mesquita et al. (2005) define autocratic states by their small winning coalition.However, in some autocratic states like China and the former Soviet Union, the selectorate is relatively large as well.The risk of being replaced in W in these countries is high because there is a plethora of potential members in the selectorate.In these systems, the leader could choose to replace any single member of the winning coalition, and a large number of alternative members always exist.At the risk of exclusion, members of the winning coalition are likely to develop a strong loyalty norm.Strong loyalty, in turn, provides a large benefit to leaders who can afford to spend less in buying off their winning coalition and more on their own consumption or future investment.A high loyalty norm also has the additional benefit of diminishing the likelihood of a successful challenger-requiring fewer resources to fight off threats to the regime.
Where does the size of the winning coalition and selectorate make loyalty most likely?The size of the winning coalition relative to the size of the selectorate is typically larger in unconsolidated regimes in comparison to a stable autocracy.This implies that the loyalty norm is stronger in stable autocracies and weaker in mixed or transitioning regimes.What does this mean for bypass aid?Stability, backed by high loyalty, provides leaders with a longer time horizon and incentivizes them to act like stationary bandits (Olson 1993).Stable autocrats are more likely to invest aid receipts in long-term public goods like property rights and education, similar to democratic leaders (Wright 2008).
On the other hand, weak loyalty norms in mixed or transitional regimes make leaders more vulnerable, shortening their time horizon.New competition between opposing parties in transitioning regimes can also make domestic competition particularly volatile as newfound mass politics competes with traditional elite power.Competition increases the chances of defection from the leader's winning coalitionbecause defection is less costly with a viable challenger (Mansfield and Snyder 1995;Hegre et al. 2001).Given this threat, leaders in mixed regimes are incentivized to allocate aid receipts differently.They are more likely to raid aid revenues in order to pay for repression or buy off emerging threats to the regime (Wintrobe 1998;Wright 2008).Providing private goods to maintain potentially disloyal supporters is paramount for incumbents in Downloaded from https://academic.oup.com/isagsq/article/3/1/ksad007/7056569 by University of Mississippi user on 04 April 2023 unconsolidated regimes to stay in power, yet it requires substantial resources.When foreign aid is provided through NGOs in higher proportions, leaders who have not cultivated strong loyalty are less likely to be able to satisfy their winning coalition.
While our argument has focused on the provision of private goods as the predominant means of maintaining support without a strong loyalty norm, we acknowledge that leaders can also use repression as an alternative instrument to safeguard their incumbency (Wintrobe 1998;Escribà-Folch 2013).While empirically indistinguishable from goods provision, the effects of aid bypass on the ability to repress are the same because repression also requires significant financial resources.In order to successfully repress, a state security apparatus is needed to restrict the public's civil and political liberties (Davenport 2007).Theoretically, we understand that less stable autocrats have an incentive to invest in repression, while consolidated regimes no longer have to (Wright 2008).Empirically, we know that more human rights violations occur in transitional or mixed regimes (Regan and Henderson 2002).
When more of the fungible resources are bypassing the government, initiating or maintaining existing repression will be more difficult for transitional regimes.On the other hand, democracies have institutionalized mechanisms for the expression of discontent, and consolidated autocracies have already successfully co-opted opponents or developed a reputation for punishing challengers.Thus, the need for investing in repression is less vital for incumbents in these regimes.Bypass aid hinders the ability of incumbents in transitional regimes to invest in repression as it does their ability to provide private goods.
While we cannot clearly delineate between whether transitional regimes are hurt more by a decreased ability to offer private rewards to supporters or by a diminished capacity to repress challengers, a final implication of our argument hinges on the size of W relative to the selectorate.If the loyalty norm substitutes for private goods provision, then as the proportion of aid allocated through bypass channels increases, leader turnover should be more likely in unconsolidated regimes.
H 3 : An increase in foreign aid allocated through bypass channels (relative to GDP) will increase the risk of leader turnover more in mixed regimes compared to other types of regimes.
We acknowledge that the link between bypass aid and tenure may give rise to potential concerns about endogeneity if donor governments, as a group, factor the potential for turnover into their choice to allocate aid through bypass.While the data does not suggest that donors are using bypass aid in greater proportions in autocratic or unconsolidated regimes 7 -donors might still adjust the level of aid strategically by regime type-a more benign possibility is that the relationship is driven by additional variables that jointly impact bypass aid and tenure.For example, if states with high capacity and entrenched leaders can block NGOs from operating within their borders, thereby decreasing bypass aid, then the causal chain we propose is interrupted.We adopt several strategies to deal with these concerns in our empirical models.For now, we emphasize that our focus on bypass aid as an explanatory variable is novel, and we pay heightened attention to controlling for the known determinants of bypass aid in our research design described below. 7See online appendix B.

Research Design and Data
To test our theoretical arguments, we conduct a quantitative analysis using a dataset on bypass aid and leader turnover from 2004 to 2017. 8Data on political leadership comes from the Rulers, Elections, and Irregular Governance (REIGN) dataset, which yields more than 350 unique leaders across 133 recipient countries with 279 turnovers.The dependent variable Leader Exit takes on the value of 1 in country-years that experienced a change in leadership and 0 otherwise. 9In our sample, approximately 16 percent of country-years experience a leader exit.
To construct our main explanatory variable, we rely on data from the OECD's Credit Reporting System (CRS) aid activity database (Organization for Economic Cooperation and Development 2017).Data on foreign aid commitments is available in donor-recipient dyads of aid allocations to recipient country i from donor country j in year t.As we are interested in the aggregate effect of bypass aid on leader tenure, we aggregate the data from DAC donors for which we have data across the full sample period to the recipientyear level. 10We then rely on the CRS's reporting by aid channel, which includes the following categories: the public sector (i.e., traditional government-to-government aid), NGOs and civil society organizations, multilateral organizations, public-private partnerships, teaching and research institutions, private sector institutions, and an "other" category of official development assistance.Building on direct parallels in the aid-curse literature (Kono and Montinola 2009;Licht 2010), we operationalize our independent variable, Bypass Ratio, as the total assistance committed in a recipientyear through NGO and civil society organizations divided by GDP. 11Standardizing our measure by GDP has several advantages.Following Kono and Montinola (2009), the impact of aid on leader tenure is conditional on how much support needs to be bought and the income of potential supporters.Unearned income goes further in buying support when there are fewer citizens and citizens are poorer.GDP correlates with both production and population, allowing us to account for both.It also allows us to capture general aid dependence, whereby leaders who are more reliant on aid should be more likely to lose office when aid changes to less fungible modalities.To reduce the influence of outlier observations and ensure temporal ordering, we use the natural log of Bypass Ratio and lag the variable by one year.Bypass 8 The temporal sample is based on data availability.Data on bypass aid only becomes available after 2004. 9There exist alternative turnover indicators, notably Goemans, Gleditsch, and Chiozza (2009)'s ARCHIGOS dataset and Licht (2010)'s supplement.These alternative indicators are not available for the full temporal sample.Bypass aid is a relatively recent phenomenon, and REIGN provides the most recent update, albeit at the expense of analyzing how turnover happens.We explore the method of removal in online appendix H. 10 This also has implications related to causality.While donors could factor the potential for turnover into their choice of aid modality, donors would have to systematically act according to the same allocation criteria to bias our results.While certain donors are more likely to choose bypass for strategic reasons (Dietrich 2016;Allen and Flynn 2017), there also exists significant heterogeneity (Alesina  and Dollar 2000; Adhikari 2019; Dietrich 2021).  1We focus specifically on NGOs as the implementing agent for two reasons.
First, donors' decision to allocate aid through multilateral organizations is studied elsewhere (Milner 2006).Second, in line with our theoretical argument, aid implemented through NGOs is perceived as less distorted by commercial or political mandates than aid implemented through other bypass channels like IGOs.Therefore, we have the clearest expectations when we focus on NGO agents exclusively.We show in online appendix G that the results are robust for measuring bypass aid as the sum of aid administered through all nongovernment channels.
Ratio takes the following form: To separate our findings from previous work, we are interested in the effect of bypass aid on leadership tenure, holding the level of government-to-government aid constant.In other words, the relevant comparison is between leaders who have similar levels of aid dependence, but aid is given through different channels (government-togovernment versus bypass).We therefore include a lagged Government Ratio, as the natural log of public sector (i.e., government-to-government) aid received in a given year divided by GDP, in all model specifications.
Our theoretical argument implies that bypass aid's effect on leader survival should vary with the quality of political institutions.To account for the recipient country's level of Democracy, we include the Varieties of Democracy Project's (V-Dem) electoral democracy index (polyarchy), normalized on a 0 to 1 scale (Coppedge et al. 2020).In V-Dem's coding, the electoral democracy index is an essential component of any other form of representative democracy (liberal, deliberative, participatory, or egalitarian).Higher values represent a higher achievement of democratic principles.While our primary measure is continuous, we also include additional models using V-Dem's categorical measure of Regime type.Hypothesis 2 implies a linear relationship, where the interactive effect should be negative because autocracies are more disadvantaged by a decline in aid fungibility.Hypothesis 3 is about unconsolidated regimes and implies that the interactive effect should be significant at intermediate values of democracy, where loyalty cannot substitute for fungibility.Our categorical measures of regime type will be particularly helpful in differentiating the effects of Hypotheses 2 and 3. 12  We also control for a number of economic and political factors in line with previous work on leadership tenure.First, to control for the effect of economic conditions on incumbency, we include the log of Per Capita Income, GDP Growth, and the log of Population.Research suggests that leaders in populous countries with low per capita income and low economic growth are more likely to lose office (Chiozza and Goemans 2004).Data on these variables comes from the World Bank's World Development Indicators.Second, we rely on data from the REIGN dataset to control for Leader Age, as older leaders are more likely to lose power (Bienen and de Walle 1991).Third, because domestic conflicts increase the likelihood of turnover, we also include a dummy for Civil War from the Correlates of War Project (Sarkees and Wayman 2010).Finally, we know that leaders face different a priori risks of turnover and that donors may provide bypass aid with this in mind.Therefore, we include a measure of political risk in each of our main models.We rely on the International Country Risk Guide's (ICRG) measure of Government Instability.We also include the ICRG's measure of recipient Government Corruption, as a determinant of donors' decisions to bypass.We invert ICRG scores so that higher values of both variables imply greater instability and corruption, respectively.
Because the dependent variable, Leader Exit, takes on the value of 0 or 1 in a given year, we rely on a logistic regression. 13We include unit-means for all explanatory and 12 For alternative tests of Hypothesis 3, see online appendix E. 13 While an alternative approach would be to use a duration model, such as the Cox proportional hazard model (Box-Steffensmeier and Jones 2004), logistic and probabilistic models of time series cross-sectional data are treated as grouped control variables as proxies for country fixed effects, as well as time polynomials to capture temporal dependence in leader tenure (Carter and Signorino 2010).This specification allows us to capture "within" country variation without dropping countries without a leadership change in our sample period (i.e., Russia).This is important to both our theoretical argument and to preserving our sample size, which is already limited by the availability of bypass data.We lag all time-varying independent variables by one year and utilize recipient-level clustered standard errors.

Results
Table 1 presents our main results.Model 1 provides the bivariate relationship between leader exit and our main explanatory variable, Bypass Ratio.Model 2 adds Democracy and Model 3 includes their interaction.Model 4 includes control variables and unit means.Model 5 replicates our main model using V-Dem's Regime classifications in lieu of our continuous measure. 14 The positive and significant coefficient for Bypass Ratio indicates that where more foreign aid (as a percentage of GDP) is allocated through NGO channels, leaders are more likely to lose office, holding government-implemented aid constant.Since the size of the effect cannot be interpreted straightforwardly, the top panel of figure 2 graphs the predicted probability of Leader Exit across different values of Bypass Ratio, based on Model 4 with 95 percent confidence intervals. 15In substantive terms, the probability of turnover for governments that receive mean levels of bypass aid relative to GDP (e.g., Horacio Cartes in Paraguay in 2015) is approximately 10 percent.As countries increase their reliance on bypass aid by one standard deviation (e.g., Jakaya Kikwete in Tanzania in 2015), the risk increases to 13 percent.The tabular and graphical results both provide strong support for Hypothesis 1. Aid modalities that result in less fungibility are harmful to political incumbents, increasing the probability of leadership turnover.
The results from our interactive models provide additional nuance to understand the relationship between bypass aid and leadership tenure.In our main empirical specification (Model 4), the interactive effect between Bypass Ratio and Democracy is negative and highly significant.The effect of bypass aid differs across domestic political institutions.Autocratic regimes are the most likely to witness turnover as bypass aid increases.Presented graphically, the middle panel of figure 2 plots the conditional marginal effects of Bypass Ratio on Leader Exit, this time for different values of Democracy. 16The distribution of Democracy is plotted below.The conditional effect of bypass aid on leader tenure is only significant for the most autocratic regimes.While bypass aid is costly for autocratic leaders, it does not have a similar effect on democratic leaders, most likely because bypass aid continues to provide the public goods on which duration data, where we control for temporal dependency on the right hand side of the model (Beck, Katz, and Tucker 1998).Our empirical modeling choices are driven by the discrete observations of our key variables over a limited time frame.Given the binary nature of our data, cubic polynomials should produce a hazard shape similar to typical proportional hazards models (Carter and Signorino 2010).See online appendix G for alternative specifications.
14 Closed autocracies serve as the omitted category.Negative coefficients imply a lower probability of leader exit relative to closed autocracies. 15All other variables held at their mean value. 16Based on Model 4. All marginal effect plots are created using margins and   2005)'s model is predicated not just on the size of the winning coalition but also on the government's ability to generate loyalty.The payment required to maintain loyal (versus potentially disloyal) members of the winning coalition is lower in pure autocracies, so loyalty may substitute for a lack of aid fungibility.If this logic is correct, then bypass aid should have a stronger effect on leader tenure at intermediate values of democracy.While the middle panel of figure 2 provides evidence against this interpretation of the loyalty norm, the bottom panel of figure 2 plots the same relationship using the Varieties of Democracy's Regime classifications. 17The categorical classification makes the distinction between H 2 and H 3 more clear.Table 1 indicates that as bypass aid increases, electoral democratic and liberal democratic regimes are significantly less likely to experience turnover, as compared to consolidated autocratic regimes (the omitted category). 17Based on Model 5.The bottom panel of figure 2 presents the conditional marginal effects (first differences) of a one-unit change in Bypass Ratio on the probability of Leader Exit, across V-Dem's four regime categories, holding all other variables constant at their mean with 95 percent confidence intervals.
Similarly, in the marginal effects plot, the interactive effect is only significant for closed autocracies.We use alternative proxies for the loyalty norm in online appendix E and continue to find that loyalty is not a substitute for fungibility when it comes to maintaining political office.Because the loyalty norm would imply that pure autocracies are less vulnerable to nonfungible aid, we interpret this as evidence against our third hypothesis.
Finally, we find that recipient characteristics impact leaders' tenure in line with the previous literature.Poor and unstable countries are more likely to experience turnover.Older leaders are also more likely to lose office.Most of the variation in other recipient characteristics is likely accounted for in the unit means that we include as proxies for fixed effects in our estimating equations.
To ensure that our results are not dependent on empirical modeling choices, we conduct additional robustness checks, which we note here and report in full in the online appendix.First, in online appendices C and D, we probe our conditional findings on political institutions by using alternative measures of regime type.Online appendix C uses data from the Polity IV Project, and online appendix D uses Freedom House's political and civil liberties.1 .15 .2 .25 .3 .35 .4 .45 .5 .55 .6 .65 .7 .75 .8 .85 .9 .95Second, online appendix E provides alternative tests for Hypothesis 3. Above, we assume the strength of the loyalty norm based on the recipient's political institutions.The size of the winning coalition relative to the selectorate is typically larger in unconsolidated regimes in comparison to a stable autocracy.This implies that the loyalty norm is stronger in stable autocracies and weaker in mixed or transitioning regimes.Online appendix E uses alternative measures of loyalty that are not dependent on political institutions (years in office and years since regime failure).While this more clearly distinguishes Hypotheses 2 and 3, the results do not change.There is little evidence that loyalty substitutes for less fungible aid modalities.
Third, online appendix F addresses the determinants of bypass aid.Absent a viable instrument that meets the exclusion restrictions, we choose to control for leaders' a priori risk of turnover by including several measures of political risk and governance quality. 18Results are robust to additional measures of coup risk and the World Bank's Worldwide Governance Index.
An additional selection concern may arise from recipient governments' control over NGO operations (Dupuy, Ron, and Prakash 2016;Dupuy and Prakash 2020).If consolidated autocratic regimes impose tighter restrictions on donor and NGO activities, then this could limit the opportunities for bypass and provide an alternate explanation.In online appendix F, we show support for this intuition; autocratic regimes are more likely to have NGO financing laws in place.Yet, this does not appear to condition donors' preferences for bypass aid in autocratic regimes; there is no significant difference in the proportion of bypass aid allocated across regime types.To further ensure that this does not drive our main results, we include Dupuy, Ron, and Prakash (2016)'s measure of NGO finance laws (0,1) and V-Dem's measure of CSO repression in our estimations.We continue to find support for the punishing effect of bypass aid, especially in autocratic regimes.
Fourth, online appendix G turns to specification.It adds additional control variables that may jointly effect bypass aid and leader tenure.Following Kono and Montinola (2009) and Licht (2010), it also interacts Government Ratio with Democracy, and the results do not change.Neither does operationalization change the results.Our findings are robust for operationalizing bypass aid as a percentage of total aid, including other nongovernment channels, and for excluding aid for democratic purposes from our bypass measure.Online appendix G also focuses on empirical modeling choices.Our results do not change with the inclusion of temporal splines, a quadratic specification, or a Cox proportional hazards model.Because our variable of interest is an interaction, we also employ diagnostic tests from Haimnueller, Mummolo, and Xi (2019).Our results are further robust to using robust standard errors and recipient country fixed effects (in lieu of unit means).
Finally, turnover looks different under different political institutions.For example, in consolidated autocracies, leadership change may take place as a consensual prearranged affair.Leaders who die in office may be replaced by their biological heirs or term limits may force autocrats to step down, but be replaced by another elite member of the same political party.In these cases, leader change is less meaningful.Online appendix G presents results using winning coalition failure as an alternative measure of leadership change (Licht 2010).Using a dependent variable that more accurately captures "irregular" change in both democracies and autocracies, our results remain unchanged.

Conclusion
Foreign aid remains a source of debate in academic and policy circles, with some arguing that substantial aid is required to eliminate poverty and others arguing that aid can do more harm than good.A recent caveat in this debate is that aid administered and managed by nonstate actors may be more effective and efficient, especially where recipients fail to meet good governance standards (Dietrich 2013).However, bypass aid also has political consequences, and to the best of our knowledge, this paper is the first to stake the claim that by minimizing incumbents' perceptions of fungibility, aid allocated through bypass channels impacts political survival.Foreign aid constitutes a major unearned part of government finance for developing countries, which politicians would prefer to allocate as is necessary to stay in power.Bypass aid weakens their ability to do so, and we find that aid modalities that diminish government control of aid expenditures also increase the risk of leader turnover, but only in autocratic regimes.
Our results have several implications for public policy and academic research.First, aid fungibility varies across many dimensions (Feyzioglu, Swaroop, and Zhu 1998;van de Walle and Mu 2007;Bermeo 2016Bermeo , 2017)).While we argue that NGO-administered aid is less fungible than government-administered aid, this does not preclude incumbents from benefiting from NGO projects (Shammama 2022;Springman 2022).Nor does it probe the idea that different types of NGO projects may be more or less fungible for political purposes.Our challenge, the same as all researchers, is that because foreign aid donors are concerned about leakage and adopt policies to prevent mismanagement, fungibility is a clandestine phenomenon.It is difficult to measure precisely.Our results are not causal, yet the less fungible government-to-government aid flows or the more fungible NGO aid flows, the more this biases against our findings.This should increase confidence in our results.It also opens new frontiers to understand how the potential for fungibility varies across NGO-administered aid projects.
Second, we emphasize the importance of downstream effects that can stem from a heightened use of bypass channels.While we offer an explanation for one potential consequence, it is likely that bypass aid affects additional political phenomena as diverse as government legitimacy and partisan alignment.Additionally, while we demonstrate that bypass aid impacts leaders' tenure, data limitations only offer suggestive evidence about the method of leaders' removal.We encourage future scholars to probe deeper into the effects of bypass aid to ask questions about the method of removal and the reactions of aid donors to bypass aid-induced turnover.If bypass aid is causing more irregular leadership turnovers via coups or leading to the rise of extreme leaders who are more antagonistic toward donor countries, then donors may be trading the problem of aid capture for a more serious political problem in the long term.
Third, while we do not explore the link directly, previous findings suggest that government turnover is generally not conducive to economic development (Aisen and Veiga 2013).Political instability decreases incentives for capital accumulation.It can also inhibit growth by undermining foreign channels of investment.If bypass aid-induced turnover undermines growth, then the impact of aid on development might be cancelled out.More optimistically, it might also suggest that studies exploring the impact of NGO service provision on economic growth in recipient countries might be capturing downward bias from political turnover.
Finally, a significant portion of contemporary foreign aid is directed toward democracy promotion.Even where democratization is not an explicit criteria for aid, democratic states receive larger aid receipts (Alesina and Dollar 2000).Our findings about the role of domestic political institutions imply that the policy debate around foreign aid and democratic transition is more nuanced when we account for aid delivery mechanisms.While the most entrenched autocratic regimes might represent the cases where donors would prefer to use bypass channels, they are also the states where bypass aid is likely to have the largest impact on incumbency.Who replaces ousted leaders is outside the scope of our study, but we conjecture that bypass aid poses the greatest threat to leaders in precisely those states where donors prefer to use bypass aid.Therefore, our results speak to the literature on foreign aid and democratization, but with a different approach.Donors have often tied aid to democratization measures like electoral reform, yet we suggest a more direct means through which the process may be, inadvertently, influenced.

Figure 1 .
Figure 1.Average aid delivered through NGOs and CSOs (as a percent of total aid) to recipient countries, 2004-2017.
marginsplot, Stata v.16.The figure presents the conditional marginal effects (first differences) of a one-unit change in Bypass Ratio on the probability of Leader Exit, across different values of Liberal Democracy holding all other variables constant at their mean with 95 percent confidence intervals.