Can foreign aid donors credibly threaten to suspend aid? Evidence from a cross-national survey of donor officials

ABSTRACT Under what conditions are foreign aid donors willing to suspend foreign aid to punish political transgressions, such as election fraud, corruption scandals or political repression? Prior scholarship has emphasized that political sanctions, including foreign aid suspensions, are constrained by the geostrategic considerations of donor countries. However, foreign aid suspensions often occur in strategically important countries, and donors respond differently to different types of political transgressions within the same county. To shed light on this puzzle, in this article, I present evidence from an original survey of top-level donor representatives in 20 African countries, including a list experiment designed to elicit truthful responses about the conditions under which donors are willing to suspend foreign aid. I argue that the likelihood of a foreign aid suspension depends not only on the strategic considerations of the donor government, but also on the institutional incentives of the donor agency. A donor agency's institutional incentives are shaped by the agency's organizational design, as well as by its foreign aid portfolio in the recipient country.


INTRODUCTION
Citizens in donor countries have a strong preference for conditioning foreign aid on good governance. Over 90% of respondents in the 2011 Eurobarameter, which sampled more than 25,000 individuals from 27 countries, expressed that foreign aid should be conditioned on democracy, human rights and governance (Bodenstein and Faust 2014). But, how willing are aid donors to condition foreign assistance on good governance? Given the strategic nature of foreign aid, can donor agencies credibly threaten to suspend foreign aid, if a recipient government commits a political transgression such as election fraud, corruption or political repression?
In this article, I argue that political conditionality, or the willingness of donors to condition foreign aid on governance (Baylies 1995), only works when the institutional incentives of donor agencies are aligned to allow the threats of donor agencies' to be credible. By institutional incentives, I mean how costly an aid suspension is to a donor agency in terms of organizational performance and reputation. Such incentives are shaped by both the donor's foreign aid portfolio in the recipient country, and organizational design features of the agency that make it easier or harder for aid to be suspended.
Prior research largely assumes that the main constraint facing political conditionality is the geopolitical motivations of donor countries (e.g. Dunning 2004;Lebovic 2005;Lebovic and Voeten 2009;Nielsen 2013). That is, donor agencies are unwilling to enforce good governance criteria, because it is diplomatically and commercially costly to donor governments. However, donor agencies sometimes sanction recipient governments who are strategically important to donor countries, and donors' strategic interests cannot explain why donor agencies respond differently to different types of political transgressions within the same country.
The possibility that institutional incentives affect the likelihood of an aid suspension occurs, because donor governments delegate the disbursement of foreign aid to donor agencies (Gibson et al. 2005;Hawkins et al. 2006;Martens 2002;Nielson and Tierney 2009;Ostrom et al. 2002;Svensson 2006). Delegation to aid agencies means that institutional incentives to suspend aid could -at least under certain conditions -supersede strategic incentives on the part of donor governments not to suspend aid, influencing how foreign aid diplomacy is practiced abroad.
To test the argument that institutional incentives have important consequences on the willingness of donors to sanction political transgressions, I collected original data on the likelihood of an aid suspension by directly querying heads of development cooperation, commonly referred to as HoCs, from 23 different donor agencies across 20 countries in sub-Saharan Africa. HoCs are the key interlocutors between donor agencies and recipient governments and are responsible for negotiating and implementing programming at the recipient-country level. In addition to vignette questions that asked the HoC to indicate how likely it is that their agency will suspend aid in a number of different scenarios, the survey included a list experiment designed to overcome social desirability 455 bias and elicit truthful information regarding the willingness of their donor agency to suspend aid following severe political repression.
Consistent with my argument, results from both the vignette questions and the list experiment vary in important ways related to the institutional incentives faced by donor agencies. First, respondents are more likely to report that their agency would suspend aid, if the political transgression directly affects the donor agency's aid portfolio 1 in the recipient country; for example, if there is corruption in a project their agency is supporting. Second, respondents report a greater willingness to sanction political transgressions, if their agency provides direct budget support to the recipient government. Third, respondents report a higher willingness to suspend aid, if there are fewer domestic veto players that can block an aid suspension decision.
My findings have important practical and theoretical implications for scholars of foreign aid and international relations more broadly. Results from the survey suggest that institutional incentives can affect how aid diplomacy is actually exercised in recipient countries. Contributing to a 'bureaucratic turn' in studies on foreign aid, 2 my results provide us with a richer understanding of when and where we might expect donor agencies to actually sanction political transgressions by suspending aid. Most bilateral and multilateral donor agencies formally condition their aid on governance. However, our ability to measure the willingness of donor agencies to suspend aid in response to a political transgression has been limited with the data hitherto available. I draw on novel data that allows us, for the first time, to directly measure the willingness of donor agencies to suspend foreign aid following a political transgression. In addition to highlighting the important role of institutional incentives in aid suspensions, the study is an important step in measuring whether political conditionality is actually effective. Aid suspensions punish recipients for political transgressions by either partially or completely stopping the disbursement of foreign aid. If donor agencies cannot credibly commit to suspending aid in response to a political transgression, then we can hardly expect recipient governments to be incentivized to improve governance based on donors' threats to suspend aid.
For scholars of international relations more broadly, my findings emphasize the importance of understanding institutional incentives in foreign policy decision-making. My results suggest that political conditionality only works when institutional incentives are aligned in such a way that donor agencies' threats are credible. This finding is relevant beyond the aid literature. In particular, the results have potential application in the extensive debate on the effectiveness of economic sanctions (e.g. Baldwin 2000; Bapat and Kwon 2015;Drezner 2003;Hufbauer, Schott, and Elliott 1990;Pape 1998;Peterson 2013). My findings suggest that scholars interested in the effectiveness of sanctions, of which aid 456 suspensions are one type, need to not only pay attention to the political motives of states, but also to the institutional incentives of implementing agencies.
Finally, methodologically, the article demonstrates the potential usefulness of elite survey experiments in foreign policy and international relations. As Susan Hyde argues, 'conducting experimental research directly with elites is one way to make experiments in IR more realistic and potentially more relevant ' (2015, 409). By directly surveying highlevel development practitioners, I am able to answer to offer a more precise answer to an important question long-debated by aid scholars, and the use of a list experiment design allows me to address a major threat to inference in the study of aid suspensions -social desirability bias.
The paper is organized as follows. First, I provide an overview of what we know about politically conditionality and the willingness of donors to suspend aid in response to political transgressions. Second, I make the case for taking the institutional incentives of donor agencies seriously when it comes to aid suspensions and present three testable predictions. Third, I provide a brief overview of the survey and the type of data it can provide. Fourth, I evaluate my argument using original survey data. Fifth, I provide an external validity check of the data by comparing my survey results to actual suspension data. Finally, I conclude by outlining the implications of the study and providing suggestions for future research.

THE POLITICS OF AID SUSPENSIONS
Following the end of the Cold War, almost all donor agencies began formally conditioning their aid on governance (Carothers and de Gramont 2013). Even aid agencies that deliberately avoid formal democratization criteria, such as the World Bank, condition aid on good governance, emphasizing sound public financial management and the protection of basic human rights (Lebovic and Voeten 2009). But, can aid agencies credibly commit to suspending aid following political transgressions by recipient governments? Does aid conditionality actually work?

POLITICAL CONDITIONALITY AND AID SUSPENSIONS
The effectiveness of political conditionally has drawn a great deal of attention from scholars. In the 1990s, many scholars took a critical perspective, emphasizing that donors often appeared unwilling to enforce conditionality. 3 At the turn of the century, the focus shifted to aid selectively. Evidence from this body of literature is mixed. 4 There is some evidence that select donors reward democratic and democratizing states with higher volumes of aid at different periods of time (Bermeo 2011;457 Claessens, Cassimon, and Van Campenhout 2009;Dunning 2004;Freytag and Pehnelt 2009;Hout 2007;Reinsberg 2015), and several scholars find that donors are more willing to condition certain types of aid -namely aid given directly to the recipient government -on governance performance (Clist, Isopi, and Morrissey 2012;Dietrich 2013;Nielsen 2013;Winters and Martinez 2015). However, others conclude that the salience of governance in the recipient country frequently takes a backseat to the strategic interests of donor countries (e.g. Alesina and Dollar 2000;Carey 2007;Hook 1998;Schraeder, Hook, and Taylor 1998;Younas 2008). Some scholars even argue that corrupt governments might get more aid (de la Croix and Delavallade 2013; Easterly 2008; Easterly and Williamson 2011). 5 Despite an extensive bibliography, the literature on political conditionality is presently limited in two key ways. First, current methodological tools make it difficult to capture whether a donor is willing to suspend aid -that is to stop the release of aid already promised to a recipient (either in part or in full) -in response to a particular political transgression. Cross-national studies of political conditionality have almost exclusively relied on regression analysis to provide an estimated effect of yearly changes in a country's score on governance indicators -e.g. the Freedom House Indicators or the Polity Index -on annual aid volumes. 6 This approach highlights important trends in aid allocation, telling us whether donors are willing to reward or punish recipients for aggregate gains in governance by increasing or decreasing the amount of aid a recipient is allocated in subsequent years. However, at least with the data currently available, it cannot speak to the micro-foundations of aid suspensions, because it is not able to capture in real time how donors respond to particular political transgressions. This is a big gap in our knowledge about political conditionality. Although aid suspensions are only one of the tools at policy-makers' disposal (Hackenesch 2015;Koch 2015;, they are vital to the success of political conditionally. Without the ability to sanction recipient governments by suspending aid, attempts to incentivize recipient governments to uphold good governance principles based on the promise to suspend aid if certain conditions are not met will most certainly be less effective. Recipient governments work closely with donor agencies over many years. If donor agencies do not follow through with their threats to suspend aid in response to specific political transgressions, recipient-country officials are unlikely to be compelled to change their governance practices on account of such threats. The second limitation is theoretical. Thus far, scholars have largely assumed that enforcement of political conditionality is mainly limited by the strategic interests of donor governments, emphasizing the political and commercial uses of aid (Lancaster 2007;Milner and Tingley 2013).

458
However, donors suspend aid even in countries that have a high level of strategic importance. For example, in November 2007, the United Kingdom's Department for International Development (DfID) suspended £20 million of foreign aid to Pakistan in response to the government's decision to declare a State of Emergency and issue a provisional constitutional order (DfID 2008). Similarly, in November 2010, DfID delayed a payment of £85 million to Afghanistan because of concerns about financial management and accountability (DfID 2011). In both cases, the UK had important strategic interests in the recipient country. Nonetheless, aid was suspended.
Strategic motivations are also unable to explain variations in how donors respond to different types of political transgressions within the same country. Take, for example, Uganda: a key ally for Western countries in the fight against terrorism in East Africa and the Horn of Africa. Despite Uganda's strategic importance, political transgressions by the Ugandan government have resulted in multiple aid suspensions over the past decade. In 2013, donors suspended over $100 million dollars in foreign aid in response to the passage of a bill that punished some homosexual acts (Reuters 2014). The bill was widely reported on and condemned in the international media. This suspension, however, was relatively minor compared a suspension that took place the year before when it was discovered that $11 million in donor funds had disappeared from an account held by the Office of the Prime Minister. Even though the theft of funds was discovered and reported to donors by the Ugandan government, donors suspended over $300 million in direct budget support to the government (Government of Ireland 2012). This caused a huge hole in the national budget, infuriating the Ugandan government (Swedlund 2017). Why were international donors willing to suspend over three times as much aid in response to the corruption scandal?

DELEGATION AND INSTITUTIONAL INCENTIVES
The delegation of aid disbursement to aid agencies opens up the possibility that institutional incentives influence the willingness of donors to sanction political transgressions by suspending foreign aid. In Uganda, donors reacted so strongly to the corruption scandal, because the disappearance of the aid money reflected poorly on donor agencies that were directly supporting the Uganda government via cash transfers. 7 The scandal was particularly embarrassing for Irish Aid, as Ireland was at the time facing a domestic financial crisis. For several days after the scandal broke, the Irish papers were filled with news articles and editorials about the missing aid funds. Because donor programming had been directly affected, it would have looked irresponsible not to suspend. 8 In contrast, although reprehensible to many international donors, suspending aid because of the 459 anti-homosexuality bill would have disrupted projects and programs that were in many cases substantively unrelated to the transgression.
While agency problems are rarely mentioned in the literature on political conditionality, it is well documented that institutional incentives can affect the quality and sustainability of foreign aid (Gibson et al. 2005;Martens 2002;Ostrom et al. 2002;Svensson 2006). Agency problems arise during aid distribution, because responsibility for disbursement is delegated to agencies whose preferences do not directly align with donor governments (e.g. Bush 2015; Gibson et al. 2005;Hawkins et al. 2006;Martens 2002). We know, for example, that policy conditionality is often not enforced because of a 'pressure to disburse' (Ostrom et al. 2002;Svensson 2006). Donor agencies seek to maximize their aid budgets and are therefore incentivized to keep disbursing, even if the recipient fails to meet the conditions of the loan or grant. 9 This creates a moral hazard problem, allowing recipient countries to put off promised reforms. 10 In the event of a political transgression, institutional incentives on the part of donor agencies often overlap with the strategic interests of donor governments not to suspend aid. Suspending aid means that development programs have to be stopped mid-stream. Not only is this often undesirable from a developmental perspective, it is practically difficult to do in cumbersome, bureaucratic organizations like aid agencies (Brown 2005). Knowledge has to be filtered up and then acted upon. 11 In addition, once suspended, it can be difficult to restart aid programs, which can adversely affect the amount of aid allocated to the agency in the future. As Ostrom et al. explain, 'A nearly universal pressure exists within almost all development agencies…to spend the money that is allocated in one budgetary cycle, as parliamentarians are likely to interpret unallocated funds as evidence that their funds are not needed ' (2002, 70).
Importantly, however, there may also be cases where the strategic interests of the donor country and the institutional incentives of the donor agency clash. For example, because of bad press, a donor agency may find it costly not to suspend aid, even if the donor country considers the recipient government to be an important political ally. Similarly, a donor agency may find its work in a recipient country compromised by a political transgression, such as corruption, and thus find it costly to continue to give aid, even if, all other things being equal, the donor government would prefer not to suspend aid. My interest is in these cases, as it is here that we should see important variations in the willingness of donor agencies to sanction political transgressions.

HYPOTHESES AND OBSERVABLE IMPLICATIONS
If institutional incentives matter for the willingness of donor agencies to uphold political conditionality, I predict that donor agencies will be 460 more likely to suspend foreign aid in three circumstances: (1) if the political transgression committed directly affects the agency's aid portfolio in the recipient country, (2) if the donor agency provides direct budget support to the recipient government, and (3) if there are fewer veto players that can block an aid suspension.
A donor agency's combined set of projects and programs in a given recipient country are collectively referred to as their aid portfolio. As we saw in the case of Uganda, there is variation in how directly different types of political transgressions affect a donor agency's aid portfolio. In some cases, the political transgression may be normative upsetting to donor officials but does not directly affect their agency's work in the recipient country. In other cases, the political transgression may affect the donor agency's ability to pursue its main objective: the disbursement of aid. I expect donors to be more sensitive to the latter types of political transgressions.
H1 -A donor agency is more likely to suspend foreign aid if the political transgression directly affects its aid portfolio in the recipient country.
Imagine that corruption is discovered in a project a donor agency is supporting in a given recipient country. Not only has the project, which the donor agency would presumably like to be successful, been compromised, it would look irresponsible for the donor agency not to suspend aid after corruption has been reported. Not suspending aid could not only result in bad press but may even lead to a decrease in the amount of money that the donor agency is allocated in the future. Alternatively, imagine that the same recipient government is accused of preventing opposition parties from registering for parliamentary elections. While donor officials may normatively object to the government's actions, their aid portfolio has not been directly compromised. Thus, we can expect that there are fewer institutional incentives to suspend aid.
In previous studies, the distinction between political transgressions that directly affect a donor agency's aid portfolio and those that do not has been systematically overlooked. The hypothesis is, however, consistent with anecdotal evidence by Fisher (2015), who observes that, in the case of the UK, corruption is more likely to lead to an aid suspension than other transgressions. It is also consistent with actual donor policies regarding aid suspensions. While many donor agencies have explicit policies regarding aid sanctions in the case of poor fiscal management, they typically lack clear policies when it comes to political governance problems. Norway, for example, has a zero-tolerance policy on financial irregularities. All recipients of Norwegian funding are required to incorporate this policy into their financial and administrative procedures, and there is a special unit within the Norwegian aid agency (NORAD) that manages all cases related to financial irregularities and operates a 461 whistleblower hotline. 12 In comparison, its policies for suspending aid on account of political governance or human rights violations are much less specified and open to interpretation. 13 There may be, however, certain circumstances under which donor agencies are more likely to suspend aid more broadly. Agencies that provide budget support disburse resources directly into the treasury of the receiving country, allowing the recipient to use their own allocation, procurement and accounting structures (Koeberle, Stavreski, and Walliser 2006). Instead of walling off funds in self-standing project accounts, when they provide budget support, donors directly fund the recipient government (Swedlund 2013(Swedlund , 2014. As a result, all other things being equal, any type of political transgression by the recipient country should implicate the donor's programming in the country more than if the aid was being provided through stand-alone projects. As a donor official working in Ghana explained, when a donor agency provides budget support and things go 'off-track', their agency is implicated. 14 In contrast, donor agencies that do not provide budget support are less likely to be held responsible for the political transgressions committed by the recipient government.
H2 -Donor agencies are more likely to suspend foreign aid if they provide budget support to the recipient government.
This prediction is consistent with recent cross-national findings that donor agencies are more willing to condition economic program aid on governance (Clist, Isopi, and Morrissey 2012;Dietrich 2013;Nielsen 2013;Winters and Martinez 2015). It is also consistent with empirical observations that donor agencies are more frequently held accountable for directly supporting governments that commit political transgressions, and that design features of budget support often give donors more leeway to suspend for political reasons (Faust and Koch 2014;Molenaers 2012). 15 Finally, anecdotally, it helps explain the strong reaction by budget support donors to the aforementioned corruption scandal in Uganda. It would have looked highly irresponsible for budget support donors not to suspend aid, because the scandal highlighted the vulnerability of the budget, which was being partially funded by aid, to fiscal misuse (Swedlund 2017).
We should not expect, however, that all donor agencies should suspend aid in equal measure, even if the political transgression is similar. In comparative studies of foreign aid, 'donors' are often discussed as if they are a monolithic group. However, donor agencies are organized in many different ways, and these variations in design can moderate the incentives of donor agencies to comply with the strategic preferences of the donor government (Arel-Bundock, Atkinson, and Potter 2015). In some cases, it may be organizationally more difficult for a donor agency 462 to suspend aid even if, all other things being equal, the donor agency would prefer to suspend aid. In particular, I predict that donor agencies with fewer veto players should suspend aid more often.
H3 -Donor agencies are more likely to sanction political transgressions if there are fewer veto players that can block an aid suspension.
According to Tsebelis (1995Tsebelis ( , 2002, the more individual or collective actors that can block changes to the status quo -i.e. the number of veto players -the less likely there is to be a change in policy. If we apply Tsebelis' logic to aid suspensions, the status quo is a planned disbursement of aid, i.e. the failure to suspend aid. Donor agencies with fewer veto players should be more likely to suspend aid following a political transgression, because there are less opportunities for either institutional or strategic interests to get in the way of a suspension. The precise number and type of veto players are likely to vary considerably across donor agencies. However, as a first test of this argument, we can distinguish between three types of donor agencies: (1) multilateral donors agencies, (2) bilateral donor agencies with a minister for international development (ministerial bilaterals), and (3) bilateral donor agencies without a minister for international development (non-ministerial bilaterals). Suspending aid following a political transgression should be the most difficult for a multilateral agency, because there are many potential veto players, all of whom have multiple incentives -both strategic and institutional -not to suspend. Alternatively, ministerial bilaterals should be the most likely to suspend, because responsibility for development cooperation rests with a singular high-ranking, political appointeethe minister of international development. Therefore, if they opt to deviate from the status quo and suspend foreign aid, it should in principle be easier for them to do so. Non-ministerial bilaterals should fall somewhere in between.
To clarify this argument, it is useful to look at how aid suspension decisions are made in these three different types of agencies. At the European Commission, a multilateral aid agency, it is the European Councilwhich is composed of ministers from each member state -that undertakes the decision to suspend aid. 16 This means that any suspension decision has to overcome a number of potential veto players (all of which are collective veto players, according to Tsebelis' logic). Similarly, at the World Bank, it is the Board of Executive Directors (appointed by the Board of Governors) that must green light suspensions. In terms of dayto-day management, the board has delegated a great deal of authority to the World Bank's management. However, the board is still ultimately responsible for approving financial packages and therefore must sign off on any changes in approved packages, including aid suspensions. 17 This opens up the possibility that one or more member states could block a 463 suspension, decreasing the likelihood that the agency will actually sanction a recipient government following a political transgression.
In contrast to multilateral agencies, bilateral agencies act on behalf of a single donor country, effectively reducing the number of potential veto players. However, donor countries organize their bilateral aid programs differently. In some donor countries, a dedicated minister for international development oversees foreign aid. The Danish Minister for Development Cooperation, for example, is charged with 'coordinat[ing] Denmark's participation in international negotiations relating to development policy issues ' and 'administer[ing] the Danish State's bilateral and multilateral development cooperation' (Government of Denmark 2012, 1). In this capacity, the minister is the ultimate arbitrator of decisions regarding aid disbursements. Similarly, in the UK, decisions about aid disbursement -including aid suspensions -are ultimately the prerogative of the minister responsible for development cooperation. 18 In contrast, in non-ministerial bilaterals, foreign aid is managed by subordinate agencies that are subject to oversight by many different domestic actors, all of which could potentially veto an aid suspension. In the United States, for example, the majority of aid is disbursed through the United States Agency for International Development (USAID); 19 a sub-cabinet agency that takes foreign policy guidance from the executive branch via the Secretary of State and is closely monitored by the United States Congress (Lancaster 2007;Lancaster and Van Dusen 2005). Similarly, in France, aid is jointly managed by the Ministry of Foreign Affairs, the Ministry of Economic Affairs and the Agence Française de D eveloppement (AFD), with the executive playing a significant role (de Felice 2015). The same goes for Austria, where suspension decisions require a concerted decision by the managing director of the Austrian Development Agency (ADA), the Austrian ministry of foreign affairs, executive staff members of ADA and even other donors. 20 In each of these cases, the number of veto players who can challenge aid suspension decisions is greater than in ministerial bilaterals, where ultimately it is the minister who decides whether or not to suspend aid.
When it comes to aid allocation, independence from the executive should increase a donor agency's ability to pursue developmental objectives (Arel-Bundock, Atkinson, and Potter 2015). Alternatively, I am arguing that bilateral agencies that are overseen by a minister for international developmentpolitical appointees and members of the ruling party's cabinet -should find it easier to sanction recipient governments for political transgressions. In the case of an aid suspension, what is required is a change in the status quo. Under these circumstances, it is the number of players who can potentially block an aid suspension that should be important.
One important question raised by this hypothesis is, to what degree is it the strategic orientation of the donor that influences the bureaucratic setting or vice versa? Certain aid agencies are arguable designed to allow 464 them to have more freedom, while others are more dependent on consent from political actors. My argument is that political conditionality only works when the institutional incentives of donor agencies are aligned to allow donor agencies' threats to be credible. That is, I am not arguing that institutional incentives are the only factor driving aid suspensions, but rather that, at times, institutional incentives can overcome strategic motivations not to suspend aid because delegation, by design or not, allows agencies to bypass strategic motivations that decrease the likelihood of a suspension. The inclusion of three distinct hypotheses allows me to test this argument in diverse ways.

DATA AND METHODS
Between March 2013 and July 2014, I carried out an original survey of high-level donor representatives working at the recipient-country level. My survey targeted HoCs, or the senior civil servant working at the recipient-country level for a particular donor agency. 21 The fact that donor officials are not widely studied is a missed opportunity (Brown 2011); they are the human face of donor agencies and thus have important insights into the practices and behaviors of donors. HoCs are a particularly interesting group, because they are the key interlocutors between the donor agency and the recipient government and are responsible for negotiating and implementing programming at the recipient-country level.
In regards to aid suspensions, HoCs working at the mission level are in an excellent position to tell us under what conditions their employer -the donor agency -is likely to suspend aid. A HoC is typically not the one who makes the final decision to suspend aid. However, as the core representative of a donor agency in a recipient country, the HoC is the donor official who has the most information about the actual likelihood of an aid suspension in a particular recipient country. Take, for example, the UK. While suspension decisions are ultimately ministerial, decisions are undertaken based on the advice of the head of office (DfID's HoC equivalent). As the head of development cooperation for a given recipient country, HoCs have considerable discretion in reporting and evaluating changes in the political environment and in recommending a course of action to the agency's headquarters. Surveying donor officials working at headquarters would likely over-represent an agency's willingness to suspend aid, because headquarter staff would only be able to answer questions about aid suspensions based on their knowledge about formal rules and guidelines rather than their personal experience working in the recipient country. At the same time, staff members working on specific initiatives at the mission level may not have the broad knowledge about the agency's development program as a whole and/or the knowledge 465 about how disbursement decisions are made to be able to sufficiently answer questions about aid suspensions.
In total, 114 HoCs from 23 different donor agencies in 20 recipient countries participated in the survey for a response rate of 53%. The survey took respondents approximately 20 minutes to complete. 22 Respondents in Francophone countries were given the opportunity to complete the survey in either French or English. Potential respondents were told upfront that all data collected would be anonymized and not associated with either their name or their agency's name. 23 Tables A1 and A2 provide a more extended description of the sample, including the number of respondents per recipient and donor country. 24 Given that the survey was carried out in sub-Saharan Africa, one concern might be that the continent's heavy reliance on aid would mean that aid agencies have a stronger influence over government policies in the region. To account for this, in my sampling strategy, I selected countries with a wide range of aid dependence (see Table A1).
It is worth underscoring that data from survey do not tell us what the HoC actually thinks his or her agency should do in a given situation. Rather the questions are designed to gauge what the respondent believes their agency is likely to do in a given scenario. Thus, the results are not indicative of the preferences of the respondent, who may believe that aid should be suspended but is not likely to be, or vice-versa. Instead, the data should be interpreted as depicting what respondents believe will happen based on their (often extensive) experience working in the field of international development. 25

UNDER WHAT CONDITIONS CAN DONOR AGENCIES CREDIBLY COMMIT TO SUSPENDING AID?
The survey included two sets of questions on the willingness of donor agencies to suspend aid in response to a political transgression: a set of vignette questions designed to measure variation between how donors respond to different types of political transgressions and a list experiment designed to account for potential social desirability biases. If the institutional incentives of donor agencies have important consequences for aid conditionality, then I expect the results to vary based on whether the political transgression directly affects the aid portfolio of the donor agency (H1), whether the donor agency provides budget support to the recipient government (H2), and whether there are fewer veto players that can block an aid suspension decision (H3). In this section, I first present the logic and results from the vignette questions, before turning to the design and results of the list experiment.

DONORS' WILLINGNESS TO SUSPEND AID
I first asked survey respondents about the probability their agency would suspend aid in a number of different scenarios. The logic behind this approach is very straightforward. I wanted to know how HoCs believed their agency would respond to different types of political transgressions. Therefore, I asked them to indicate how likely their agency was to suspend aid conditional on different types of political transgressions. The question was phrased as follows: For each of the following events, how likely is it that your agency would suspend aid to [Country X] if the event occurred in [Country X] at some point in the future? A: moderately fraudulent elections B: corruption scandal in a project your agency is supporting C: deterioration in the investment climate D: changes in headquarter priorities E: highly fraudulent election F: deterioration in respect for civil liberties G: corruption scandal in the government at large In the individual surveys, [Country X] was replaced with the name of the country where the respondent was working (i.e. Uganda), and the different scenarios were randomized. The response options were unlikely, moderately likely or very likely. In the analysis, I rely on the results for 'very likely', as any potential respondent bias should be lower in this category. While respondents may feel pressured to avoid admitting that their agency would not suspend aid in certain scenarios, they should feel less pressure to select that their agency is very likely to suspend aid.
In addition to the governance scenarios, I also included the options 'changes in headquarter priorities' and 'deterioration in the investment climate'. I included the option regarding headquarter priorities, because I was interested to see which scenarios would be more or less likely to lead to a suspension when compared to this relatively mundane (but frequent) bureaucratic event. I included the option on investment climate as a measure of the commercial incentives of the donor country, which many suggest influence aid allocation decisions.
Consistent with H1, HoCs report that donor agencies are more sensitive to transgressions that directly affect their aid portfolio (see Figure 1). Only 17% of those surveyed responded that deterioration in the respect for civil liberties was very likely to lead to an aid suspension. This is roughly equivalent to the proportion of HoCs indicating that a change in headquarter priorities is very likely to result in their agency suspending 467 aid. In comparison, 53% of HoCs reported that corruption in a project their agency is supporting is very likely to lead to an aid suspension. If the corruption takes place in the government-at-large, only 36% HoCs reported that their agency is very likely to suspend aid. These results are consistent with my argument about institutional incentives. Corruption in the government-at-large may or may not affect the agency's portfolio in the country, while corruption in a project the donor agency supports clearly affects the agency's activities in the recipient country.
Interestingly, 38% of HoCs report that a highly fraudulent election is very likely to lead to an aid suspension. While somewhat surprising, this finding is not inconsistent with my argument that agencies are more willing to suspend in scenarios where day-to-day operations are affected. Highly fraudulent elections can have enormous reputational costs for the aid agency, particularly if the donor agency is funding parts of the electoral process. The results are particularly interesting when one considers that only 2% of respondents (only 2 out of 86) report that moderately fraudulent elections are very likely to lead to a suspension. In the real world, how election fraud is described and understood by donors and the public may vary considerably from case to case, even if the facts are similar. If aid agencies have a preference for continuing to provide aid, they may be incentivized to downplay the degree of election fraud.
Only 1% percent of HoCs report that a suspension is very likely in the case of deterioration in the investment climate. While this finding does not discredit the idea that commercial interests can drive decisions about aid allocation, 26   If we break the scenario question down by whether or not the agency provides budget support, as predicted by H2, we see clearly differences between how the two groups of donors respond to political transgressions (see Figure 2). 27 HoCs working for agencies that provide budget support to the recipient country in question are twice as likely to report that their agency is very likely to suspend aid in the case of a highly fraudulent election or a deterioration in respect for civil liberties. They are also 15% more likely to report that their agency is very likely to suspend aid in the case of government corruption. The pattern is slightly different for corruption discovered in an agency's project. Here HoCs working for agencies that do not provide budget support report being more sensitive. Given that non-budget support agencies provide more project aid, it makes sense that they would be more sensitive to this type of corruption than budget support donors.
Finally, as predicted by H3, responses also vary based on the type of donor agency the HoC works for (see Figure 3). Only 10% of HoCs working for multilateral agencies indicated that their agency is very likely to

469
suspend if there is deterioration in respect for civil liberties. Alternatively, 18% of HoCs working for non-ministerial bilateral agencies and 25% of HoCs working for ministerial bilateral agencies selected that their agency is very likely to suspend in this scenario. There is a similar pattern for highly fraudulent elections. The patterns are slightly different for corruption with non-ministerial bilaterals reporting the most sensitivity to project-level corruption. In practice, many non-ministerial bilaterals do not give budget support. This is not likely to be a coincidence. Controversial aid modalities, such as budget support, may also be less likely to be adopted if there are more veto players. However, in the sample, the two measurements vary in important ways that help to mitigate endogeneity concerns. First, whether or not a donor agency provides budget support varies according to whether or not a particular donor agency gives budget support to the recipient country where the respondent is working, whereas the type of agency (multilateral, ministerial bilateral or non-ministerial bilateral) is a sample-wide measurement. Second, there are donor agencies that give budget support but do not have a dedicated minister for international development. For example, Japan gives budget support to three recipient countries in the sample (Ghana, Tanzania and Zambia), but does not have a minister for international development. 28 Third, whether or not an agency provides budget support is not limited to bilateral agencies, as Proportion of HoCs reporting that their agency is 'very likely' to suspend aid in several fictional scerarios, by agency type (with 95% confidence intervals for the population proportion).

TOLERANCE FOR POLITICAL REPRESSION
A significant challenge when directly asking donor officials about their agency's willingness to suspend foreign aid is social desirability bias. In the vignette question, I try to limit potential social desirability bias by ensuring the anonymity of the respondent and by focusing on the 'very likely' responses. However, nonetheless, respondents might still be reluctant to directly admit that their agency is unlikely to suspend aid, because freely admitting this could possibly undermine the credibility of their bargaining position vis-a-vis the recipient government. Given this concern, I also carried out a list experiment designed to estimate the percentage of donor-recipient dyads where the donor agency would be unwilling to suspend aid even after an incident of severe political repression. List experiments use an experimental design to elicit truthful responses to a sensitive topic by circumventing respondent-related biases that could arise from social desirability or privacy concerns (Blair and Imai 2012;Lavrakas 2014). In a list experiment, the respondent is not asked to openly admit to holding a controversial belief (or in this case to admit that their agency might act against its publicly stated principles). Rather they are asked how many items are true from a list of statements that includes a sensitive item. The mean response of the treatment group is then compared to the mean response of a baseline group that received exactly the same statements, minus the sensitive item. From a comparison of these two means, an estimate of how many respondents believe the sensitive item to be true can be derived. 29 Respondents in the baseline group were asked to indicate how many statements were true out of a list of four baseline statements (designed to avoid ceiling and floor effects). 30 In addition to the baseline statements, the treatment group was presented with the following sensitive item: As long as the [Government of X] remains cooperative on our agency's main priorities, political repression (including suspected assassinations of opposition leaders) would not lead to the suspension of aid.
In the individual surveys, [Government of X] was replaced with the government of the recipient country in which the respondent was working (i.e. Government of Uganda). Whether or not the HoC was in the baseline or treatment group was randomized, as were the statements.
I designed this question to be a strong test of the hypothesis that development agencies have an incentive to continue providing aid despite a 471 political transgression. The decision to focus on political repression and to mention 'suspected assassinations of opposition leaders' was made in order to present a vivid scenario to respondents that was clearly beyond acceptable norms of behavior. I wanted the question to be both realistic (hence the use of the word 'suspected'), but also to clearly imply that, while the recipient government had crossed a line, it remained cooperative with the agency in other ways.
Using a difference-in-means estimator, I estimate from HoCs' responses to the list experiment that 45% of HoCs believe that their agency is unlikely to suspend in the treatment scenario (see Table A3 in Appendix). 31 Using a one-sided permutation test, the hypothesis that no donor representative affirmed the sensitive item (p D .017) can be rejected. If close to half of HoCs report that a suspension is unlikely in the case of severe repression, it seems reasonable to expect that suspension rates will be even lower in more moderate cases of political repression.
An important consideration in the analysis of list experiments is the assumption of no design effect; i.e. the assumption that inclusion of the sensitive item had no systematic effect on respondents' answer to the central item. Using Blair and Imai's (2012) test, I find no evidence of design effects (see Table A4 in Appendix). There is also no evidence of any systematic variation across key characteristics -i.e. the type of agency, language of the survey, colonial history and mean length in office -between the baseline and treatments groups (see Table A5 in Appendix).
Just as with the vignette questions, results from the list experiment are consistent with my argument about institutional incentives. Consistent with H2, I estimate that 94.4% of HoCs representing agencies that do not provide budget support to the given recipient country believe that their agency would be unlikely to suspend in response to severe political repression. Alternatively, the proportion of HoCs representing agencies that have provided budget support to the recipient country is just 20% (see Figure 4). Consistent with H3, donor agencies with fewer veto players also report that their agency is more likely to suspend aid following severe political repression (see Figure 5). The estimated proportion of HoCs reporting that their agency is unlikely to suspend aid in the case of political repression is over 71% for multilateral agencies, compared to 46% for HoCs representing non-ministerial bilaterals and 8% for HoCs representing ministerial bilaterals.
Results from the list experiment are thus consistent with the results from the vignette questions, and my broader argument about the role of institutional incentives in aid suspension decisions.

DO THE RESULTS MIRROR REAL-WORLD DECISION-MAKING?
In both the vignette questions and the list experiment, respondents are presented with several imaginary examples and asked how their agency is likely to respond. Presenting respondents with a fictional scenario is advantageous in that it helps us rule out exogenous factors that may influence a donor's decision to suspend or not suspend aid. Because respondents are given only limited information about the hypothetical situation, they have to make their decision based on the information presented to them. This allows us to know with greater confidence that the decision to suspend is a result of the political transgression. However, it also means that the results do not take into account a variety of other potential factors not captured in the simple scenarios. How then do we know that these results mirror real-life decision-making regarding aid suspensions? One way to verify the accuracy of the results is to look at actual cases of aid suspensions. Testing our predictions using such data would require selecting on the dependent variable. However, aid suspension data are useful for verifying that the patterns observed in the survey mirror actual suspension decisions.
Most aid agencies do not publicly release data on aid suspensions. However, as an initial test, I obtained data from DfID, a ministerial bilateral agency that provides budget support to many countries around the world. In 2006, DfID began publicly reporting aid suspensions in accordance with a new transparency law. 32 Between 2006 and 2014, DfID reported 55 changes to their programming, of which 39 can be classified as an aid suspension. 33 Of these aid suspensions, only 10 were related to problems with human rights or political governance. The remaining suspensions came about because of poor fiscal management and/or corruption. 34 In total, this means that in 75% of the cases where DfID suspended aid between 2006 and 2014, the aid suspension occurred because of issues related to the integrity of the program rather than problems with human rights or political governance. These results are consistent with results from the vignette questions suggesting that -at least on this measure -the results from the survey are consistent with real-world decision-making.

CONCLUSIONS
How willing are donors to condition foreign assistance on good governance? Given the strategic nature of foreign aid, can donor agencies credibly threaten to suspend foreign aid, if a recipient government commits a political transgression such as election fraud, corruption or political repression? In this paper, I argue that political conditionality only works when the institutional incentives of donor agencies are aligned to allow the threats of donor agencies' to be credible. Using data from an original survey of 474 high-ranking donor officials at the recipient-country level, I substantiate this argument by providing evidence that donors are more willing to sanction political transgressions by the recipient government: (1) if the political transgression directly affects the donor agency's aid portfolio, (2) if the donor agency provides budget support to the recipient government, and (3) if there are fewer veto players that can prevent an aid suspension.
In future scholarship, it would be useful to test more detailed predictions about donors' willingness to suspend aid for specific donor agencyrecipient country dyads. A larger sample size that includes HoCs working in missions beyond Africa would make this possible and, at the same time, add broader validity to the findings of the study. Nonetheless, the study is an important first step in understanding when and where donor agencies can credibly commit to sanctioning political transgressions by suspending foreign aid.
Theoretically, my findings suggest that institutional incentives can change how aid diplomacy is actually exercised in recipient countries. Once aid is committed, disbursement is left to donor agencies, whose preferences do not perfectly align with donor governments. Even if aid is given largely for strategic reasons, the institutional incentives of donor agencies may at times override the strategic preferences of donor governments. These findings have potential application beyond the aid literature, as they suggest that in contexts where decision-making is delegated, institutional incentives can at times override the geopolitical and commercial interests of states. They also speak to the broader question of when and where political conditionality is likely to be effective.
Methodologically, the article demonstrates the potential usefulness of elite survey experiments in foreign policy and international relations. By directly surveying high-level development practitioners, I am able to provide a more precise answer to an important question long-debated in the aid literature: how credible are the threats of donors to suspend aid? The use of a list experiment design allows me to overcome possible bias from social desirability pressures.
Practically, the findings not only help us to understand variation in aid suspensions decisions, but also suggest that donor agencies can be designed in a way to give them more independence over aid suspension decisions. If the goal is to condition aid on good governance, the number of potential veto players that can block an aid suspensions decision should be limited. This means that donor countries, if they so choose, can take steps to shield their foreign aid from political influences and increase the credibility of threats to condition aid on governance. From the recipient-country perspective, my findings suggest that to maintain aid flows, recipient governments should focus their attention on fiscal management. Corruption, more than violations related to human rights or political governance, is like to result in a suspension of foreign aid.

ACKNOWLEDGMENTS
Samuel Andreas Admasie, Rik van Hulst, Marije Oude Hengel, Elisabeth Kr€ amer and Maya Turolla all provided valuable research assistance with the survey. The author is grateful for comments and advice received at the 2014 EADI General Conference, 2014 APSA Annual Meeting, presentations at the University of Albany, Georgetown University and Yale University, and during an IPES virtual session. She is particularly indebted to Vincent Arel-Bundock, Sarah Bermeo, Sarah Bush, Stephen Brown, Simone Dietrich, Raymond Hicks, Naomi Levy, Malte Lierl, Martin Steinwand, Micheal Tierney, Joseph Wright and three anonymous reviewers for their comments and suggestions.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.

Research Executive Agency, European Commission (CIG #322228).
NOTES 9. At the World Bank, for example, not only are country loan officers under pressure to meet country disbursement targets, there is a coordination problem in that staff are aware that it would not be financially productive to make an example of a particular country by refusing to disburse funds. Policy conditionality can also conflict with other objectives, namely the goal of providing quick-disbursing finance to limit potential loan defaults . 10. See, for example, Azam and Laffont (2003), Burnside and Dollar (2000), Collier (1997), Dollar and Svensson (2000), Hudson and Mosley (2001) and Killick (1998). 11. For example, in the 1970s, the head of Norway's development program in Tanzania reportedly discouraged aid personnel working on the ground from disclosing problems with the country's forced villagization program (ujamaa), because Tanzania had a special place in Nordic aid policy. Even once the issue was taken up in Oslo, it took years before the full extent of the forced removals was widely known, because donor officials were afraid that the situation would be damaging to a still new aid system (Simensen 2007, 171). 12. Norway is not alone in having a whistleblower hotline. USAID, for example, also manages a hotline for the reporting of fraud, waste and abuse in projects sponsored by the agency. 13. E-mail exchange with Norad representatives, October 2014. 14. Interview with bilateral donor representative: May 2013; Accra, Ghana. 15. Most budget support programs include a set of underlying principles that are almost always related to governance. 16. E-mail correspondence with the European Commission, January 2015. As an example, see council decision 2002/148/EC, which suspended aid to Zimbabwe in 2002. For more on how foreign aid works at the EC, see Schneider and Tobin (2013). 17. In practice, the board does not typically 'suspend' aid but rather pulls financing or fails to renew financing. Following the anti-homosexuality bill, for example, the Ugandan government could continue to borrow from the World Bank but financing for health was pulled. 18. E-mail correspondence with DFID, September 2014. 19. In any given year, up to 15 different government agencies are involved in the disbursement of aid in the United States (Tierney et al. 2011). The largest amount of aid, however, flows through USAID. 20. E-mail correspondence with ADA, November 2014. 21. Official names for HoCs vary by donor. Within UNDP the equivalent would be the resident representative, while at the World Bank it would be the country director. Most bilateral agencies have a Head of Cooperation or Head of Development Cooperation. This position is different from the ambassador, who is responsible for the political relationship between the two countries. Instead, the HoC is responsible for the country's development portfolio in the recipient country. 22. A full draft of the survey protocol is available at: http://haleyswedlund. com/files/SurveyProtocol.pdf. 23. The most difficult part of the survey process was identifying potential respondents, as many aid agencies are less than transparent. After the names and e-mails of the HoCs in a particular country were collected, an e-mail requesting participation was sent. Potential respondents were sent up to five e-mails requesting their participation; one approximately every week for four weeks and a final reminder before the survey closed. In addition, in some of the Francophone countries, respondents were also sent an additional e-mail in French to see if this would increase response rates. (It did not.) 477 24. Only traditional, OECD-DAC donors and affiliated multilateral institutions were sampled. Accordingly, responses cannot be considered representative of how donor officials from non-traditional donor countries, such as China or Brazil, might respond. 25. On average, respondents reported holding in their current position for 24.7 months (s.e. 1.5 months). Thus, respondents were not, on average, new to the position or the recipient country. (Country postings for donor staff are generally 3 to 4 years with postings for 'hardship' countries being considerably shorter at 1-2 years.) 26. A donor country may still choose, at least in part, to provide aid to a particular recipient country for economic reasons. 27. In the 'budget support' category, I include HoCs that selected that their agency is currently providing budget support, as well as HoCs that selected that their agency provided budget support in the recent past. The latter category is quite small and theoretically, it makes more sense to include them with current budget support donors; the overwhelming majority reported that they ended budget support because of a political transgression by the recipient government. The results do not change when this group is dropped from the sample. 28. Japanese development cooperation falls under the Ministry of Foreign Affairs, which has an International Cooperation Bureau overseen by a director-general. 29. Political scientists have used list experiments to address several sensitive topics, such as race relations, support for militant groups, and support for female political candidates (Bullock, Imai, and Shapiro 2011;Kuklinski et al. 1997;Streb et al. 2008). 30. Respondents were instructed to 'Please read the following statements and indicate how many are true.' The baseline statements were as follows: (1) (37) Belgium (3) Australia (2) African Development Bank (10) Canada (1) Austria (3) European Union (12) Denmark (5) France (10) UNDP (9) Finland (2) Italy (4) World Bank (6) Germany (7) Japan (7) Ireland (6) Korea (3) Netherlands (4) Norway (3) Sweden (4) Spain (2) United Kingdom (3) Switzerland (4) United States (4) ¡0.0208 0.0206 H0: No design effect. Bonferroni-corrected p-value 0.5418 Note: This statistical test is designed to detect potential violations of the assumption that the inclusion of a sensitive item has no effect on respondents' answers to control items, which may arise if respondents evaluate the control items relative to the sensitive item. For a full explanation of the method, see Blair and Imai (2012, Section 3.1, pp. 63-5).