[COVID-19] … calls for a global coordination mechanism to accelerate matching demand with supply of what is necessary for the health system to respond. And I can tell you that we, the World Bank and the IMF are stepping in, are taking our responsibility for this global coordinating mechanism. —Kristalina Georgieva, IMF Managing DirectorFootnote 1
Policymakers in international organizations (IOs) often pursue “coordination.” Indeed, existing literature frames coordination as the primary way for IOs to eliminate redundancies among their activities and thereby mitigate the pathologies of competition under regime complexity, such as forum shopping.Footnote 2 To achieve this, IOs can develop a division of labor,Footnote 3 defer to one another,Footnote 4 or integrate their activities.Footnote 5
While effective at mitigating overlap, coordination can be quite costly for IOs because they are asked to limit the scope of their activities until they occupy a niche over which they possess a monopoly. Existing work shows that IOs prefer to creep their missions, increase their independence from member states, and expand the breadth of their activities. For instance, Johnson shows that bureaucrats working in existing IOs may help insulate new IOs from member state influence during the design phase.Footnote 6 Similarly, Abbott and Snidal argue that independent IOs can more ably pursue agenda setting, appear neutral in international affairs, and resolve members’ disputes.Footnote 7 Others show that IOs often creep their missions, either to fill governance voids when new issues arise, or by leveraging rational-legal authority to pursue their own agendas.Footnote 8 As a result, IOs often select alternative strategies to coordination when responding to competition under regime complexity.
I focus on one such alternative—pooling—which does not require any restriction of IO activities. IOs instead directly combine resources and expertise to pursue their mandates in tandem.Footnote 9 While pooling carries its own costs as IOs encounter friction when they work together, these are small relative to a permanent restriction of the scope of organizational activities. IOs might then pool resources both because it is relatively less costly than coordination and because it offers several benefits to them, since it prevents member states from forum shopping and allows IOs to leverage their comparative advantages.Footnote 10 Pooling can involve financial resources (co-financing) or information and expertise (information sharing). While these activities are important and often pursued by IOs, existing work has largely overlooked them.Footnote 11 Formal definitions of pooling, coordination, and competition can be found in Table 1.
Key terms and definitions
Given the prevalence of IO pooling and asymmetric focus on coordination in the literature, I address the following question: when are IOs willing and able to pool resources and expertise? Drawing on interviews with officials from prominent international financial institutions, I argue that the interests of the most powerful member state in each IO are pervasive regardless of whether or not member states have formal control over the approval of pooling arrangements. Therefore, the relations between the most powerful stakeholders in each IO should dictate whether IOs pool or duel.
This argument extends and revises the large body of work illustrating how powerful member states shape policymaking in IOs broadly and international financial institutions in particular.Footnote 12 In particular, I build on recent work that shows how these approaches are compatible with those portraying IOs as autonomous actorsFootnote 13 because IO staff often design policies that align with the interests of leading member states even in the absence of member state intervention or horse trading.
With respect to both co-financing, over which member states retain oversight, and information sharing, over which they lack such oversight, IO staff design policies favorable to powerful stakeholders. They may do so in anticipation of a principal's preferences, because they are selected with input from and socialized by powerful member states, or because they share training and policy views supported by such states. IOs are typically headquartered in their most powerful member state, which permits staff interaction with government officials and socialization to their perspectives. Moreover, staff are disproportionately supplied by powerful member states, and these states may influence staff hiring and promotions. Staff then choose to cooperate with IOs dominated by states that are similar to their own most powerful principals, either to satisfy their leading shareholders or because they share the world views that staff at the other IOs possess.
I corroborate this theory in several ways. First, I introduce an original data set of patterns of cooperation in the development issue area over the course of its entire operational history (1944 to the present)—the first such data set of its kind. Through observational regression analysis, I show that IOs dominated by geopolitical friends are much more likely to pursue cooperation than those controlled by foes. Then, to generate causal leverage, I exploit an elite survey experiment performed on staff from development IOs recruited through LinkedIn. Across a battery of tests, support for geopolitical explanations dwarfs that for alternative explanations derived from existing research, including geographic and resource considerations.Footnote 14
When Do IOs Pool Resources?
While pooling is often pursued by overlapping IOs, scholars have focused their attention and data collection efforts primarily on coordination. Coordination is distinct from pooling because it requires IOs to minimize overlap in rules and activities. Such restrictions can be undesirable for IO staff, who prefer to creep their missions and increase their influence relative to member state principals.Footnote 15 Similarly, key member states might resist any narrowing of organizational mandates because it threatens their influence.Footnote 16 Therefore, IOs and their principals might prefer to pool resources in lieu of developing a division of labor, deferring to other IOs, or pursuing specialization.Footnote 17
Because the literature tends to conflate coordination and related activities like pooling,Footnote 18 I am among the first to draw a clear theoretical and empirical distinction between the two. However, this is not to say that the two are independent of one another. In fact, because pooling is most attractive when IOs’ activities substantively and geographically overlap,Footnote 19 it may occur most often when IOs are uncoordinated. I therefore focus on IOs whose activities substantively and geographically overlap in this study.
Importantly, my theory should apply to only operational as opposed to regulatory IOs. Where regulatory organizations create and enforce rules and adjudicate disputes, operational organizations undertake programs and carry them out on the ground.Footnote 20 My theory applies to only the latter group of organizations because states can typically pursue regulatory activities, such as an investment dispute, at a single IO at a time. This makes coordination attractive to regulatory IOs because they can recover monopoly power over a regulatory niche. In contrast, states can solicit assistance from a number of operational IOs, such as development banks, simultaneously. Therefore, the potential for pooling among IOs is much higher among operational organizations, and my theory should apply primarily to them. Surprisingly, few scholars have examined pooling among these IOs.Footnote 21
When do overlapping IOs pool resources and expertise? I argue that pooling is driven by the preferences of the most powerful stakeholders in each institution. I therefore extend and revise the large body of work showing how powerful member state preferences are felt in multilateral policymaking. Older work in this tradition contends that IOs are so controlled by powerful states that they are epiphenomenal.Footnote 22 More recent work provides evidence that powerful states intervene in IO policymaking directly,Footnote 23 or else engage in horse trading to realize their preferred outcomes.Footnote 24 For instance, US influence has been shown to affect outcomes like the size of loans, speed of disbursements, and stringency of conditionality at IOs like the International Monetary Fund (IMF) and World Bank.Footnote 25 I build on these literatures by showing that powerful stakeholder preferences also pervade decision making about pooling. I also inject needed nuance into this body of work, as I show that even autonomous IO staff pursue policies that align with powerful shareholder preferences.Footnote 26
I suggest that powerful member state interests are pervasive regardless of whether member states have the opportunity to actively intervene in the negotiation process. I therefore focus on two forms of pooling that vary on the dimension of member state control: co-financing and information sharing. Co-financing occurs when two IOs combine financial resources and personnel to tackle a fixed-length operation. For example, in 2014, the IMF and Eurasian Fund for Stabilization and Development (EFSD) co-financed a loan program in Armenia, with the EFSD shouldering 70 percent of the cost. Because these operations are subject to direct vote by IO member states, stakeholders have an opportunity to weigh in on co-financing deals. Additionally, co-financing carries costs for IOs since each organization must compromise to reach consensus on a policy framework. However, I suggest these temporary concessions are less costly than permanent changes to IOs’ mandates, as are often required under coordination. Additionally, co-financing carries clear benefits by minimizing each IO's financial burden and diversifying loan risk across IOs. Similarly, when IOs co-finance, they can leverage comparative advantages, as a former IMF senior economist indicated when describing co-financing between the IMF and Asian Development Bank (ADB) in Pakistan: “The IMF will sometimes allow those with more expertise to design [part of] the program … in Pakistan, the ADB directed energy sector reform.”Footnote 27
Information sharing, meanwhile, is characterized by the exchange of otherwise private knowledge.Footnote 28 Shared information is often statistical, as when IOs pool economic data, but it can also be more technical, as when IOs share management practices. These arrangements tend to persist indefinitely, and they are negotiated by IO staff in the absence of direct oversight by member states. For instance, leaders from the African Development Bank (AfDB) and Islamic Development Bank (IsDB) signed a memorandum of understanding establishing information sharing channels in 1987.Footnote 29 Like co-financing, information sharing can be costly for IOs because private information allows IOs to offer members unique expertise.Footnote 30 Additionally, members might be less willing to transmit their sensitive data to IOs when the organizations must share it with their peer institutions.Footnote 31 However, information sharing also carries benefits for IOs because it reduces each organization's reliance on member states.Footnote 32 Moreover, because IOs may draw on different sources of information and possess divergent expertise, information sharing allows IOs to leverage comparative advantages.
Although powerful member states can interject in the design of co-financing but not information-sharing agreements, similar mechanisms apply to each. In the former case, IO staff know that co-financing is subject to powerful stakeholders’ approval. However, these states generally lack formal veto power over lending arrangements—for example, the United States does not have veto power over lending at the World Bank despite being the largest stakeholder.Footnote 33 A former senior economist from the IMF noted that co-financing frameworks are designed by staff possessing high levels of autonomy.Footnote 34 Information-sharing agreements are similarly negotiated by staff without member state intervention. These agreements take the form of memorandums of understanding, typically signed by leaders from each organization and accompanied by signing ceremonies.Footnote 35
Despite the fact that IO staff negotiate pooling frameworks autonomously, they are still likely to design policies that accord with powerful member state preferences. First, organizations are typically headquartered in the territory controlled by the most powerful institutional stakeholder.Footnote 36 This means that IO staff are often primarily natives of that country, and staff have a unique opportunity to interact with officials from that government.Footnote 37 For example, at the World Bank, 25 percent of staff over the period from 2006 to 2015 were US citizens.Footnote 38 These staff members, as US natives, may pursue policies that align with American preferences as diligent agents of their home nation. They may also be selected or promoted with input from the US government, which retains control over key appointments.Footnote 39 American staff may then be selected or promoted because they share policy preferences and training favorable to the United States.Footnote 40
Other, non-native staff members, meanwhile, may become socialized as they interact with American staff members and US government officials.Footnote 41 In other words, we might expect IO staff to express preferences similar to those of government officials in the most powerful member state, either because they are natives, they anticipate and match the preferences of the leading stakeholder, they are selected with input from this state, or because they are socialized as a result of frequent interaction with natives and the government. Bureaucrats then, intentionally or not, opt to cooperate with IOs that are dominated by states that have friendly relations with their organization's most powerful principal. This theory is highly generalizable, and I discuss the situations to which it applies in the conclusion.
Evidence from an interview with a former senior economist at the IMF offers preliminary support for this theory. In her experience, “co-financing is most seamless when it is between similar organizations like the IMF and World Bank due to locational and US advantages. You have the same economists and country coverage.”Footnote 42 The same interviewee stressed that “there is friction behind the scenes” between staff even when geopolitically and institutionally similar organizations like the World Bank and IMF cooperate.Footnote 43 However, she reported that this friction is most obstructive when IOs are controlled by rivals.
I therefore argue that two IOs should be more likely to pool resources when they are controlled by countries with similar geopolitical preferences. In contrast, IOs that are dominated by geopolitically dissimilar countries should not pursue cooperation as often. This discussion yields the following hypothesis:
H1 IOs that are controlled by geopolitically aligned member states should pool resources more often than IOs led by geopolitically unaligned member states.
A Data Set of IO Pooling
To test my theory, I focus on the development space—one of the most prominent contexts governed by operational IOs. Theoretically, it permits comparison with existing studies of powerful state influence.Footnote 44 Moreover, this issue space is advantageous for empirical study because it is characterized by high degrees of transparency. This allows for rigorous and comprehensive coding of pooling among overlapping IOs.
As I discussed, I study two common forms of pooling: co-financing and information sharing. I focus on these activities for two reasons. First, both are systematically documented by IOs, making them empirically tractable. Second, co-financing and information sharing are prevalent across many issue areas, including but not limited to development and emergency lending.Footnote 45
My original data therefore track co-financing and information sharing among all overlapping IOs in the development issue space. To my knowledge, I am the first to document pooling among IOs over time. The unit of analysis is the IO dyad-year. Importantly, the data consider only dyads whose activities substantively overlap. This means that the IOs have to perform activities that are at least partial substitutes and their geographic coverage must overlap. Otherwise, there is no concern about forum shopping by member states, and pooling is unlikely to manifest. For example, while the ADB and World Bank may pool resources, the West African Development Bank (BOAD) and Development Bank of Latin America (CAF) have little reason to do so.
The co-financing variable is a count measure that captures the number of co-financed programs initiated between two IOs in a given year. The information-sharing variable, meanwhile, measures the number of information-sharing agreements that are active between two IOs in a given year. These measures were hand-coded based on my reading of thousands of program documents, press releases, and other pieces of IO legislation covering the period between 1944 and 2018. Additional details about coding procedures can be found in Appendix 2.
Regression Results
To start, I run a series of negative binomial regressions, which are appropriate given overdispersion in my pooling measures. In addition to the co-financing and information sharing variables discussed earlier, which serve as dependent variables in these analyses, I deploy a series of dyadic independent variables designed to capture geopolitical closeness between key member states.
More specifically, I look at the relationships between the leading shareholder in each organization as measured by vote share in and contributions to the organization.Footnote 46 I consider three measures of geopolitical relations among leading IO stakeholders: UN voting (ideal point distance), alliance ties, and the rivalry-peace variable from Goertz, Diehl, and Balas.Footnote 47 UN voting data come from Bailey, Strezhnev, and Voeten and are often used to measure geopolitical closeness.Footnote 48 Similarly, countries that have signed formal defense agreements are likely to share geopolitical goals.Footnote 49 Last, the rivalry-peace measure captures both geopolitical closeness and hostility. It is measured on a 0 to 1 scale, with 0 and 1 representing severe rivalry and security community respectively. I anticipate that IOs led by countries that vote more similarly in the UNGA, have formal defense ties, and are more peaceful toward one another will pool resources more often.Footnote 50
I also include several covariates derived from existing literatures. First, I control for the difference in the number of member states belonging to each IO because existing work suggests that IOs with relatively few resources may prefer to cooperate with richer IOs.Footnote 51 While I would prefer to control directly for the amount of capital available to each IO, the data on resource holdings at many IOs are incomplete. However, in general, IOs with more member states possess greater resources, and data on membership are widely available. I more directly test for the importance of resource disparities in my elite survey, and I offer an alternate measure of each IO's “aid budget” in a robustness check.Footnote 52
Beyond resource dependencies, I control for geographic proximity between IOs.Footnote 53 IOs with headquarters located in close proximity likely have similar staff compositions, and their staff members may interact frequently.Footnote 54 These IOs might therefore be more likely to pool resources. For robustness, I also include a test that incorporates a binary indicator for IOs co-located in the same city.Footnote 55
Next, I control for the extent to which two IOs’ memberships overlap. To construct this measure, I simply calculate the share of all members of the IOs in each dyad that belong to both organizations. IOs with greater member overlap should pool resources more often because doing so reduces the likelihood of forum shopping by members.Footnote 56 Last, because pooling has become more common over time,Footnote 57 I include a linear time trend in subsequent tests. Complete variable descriptions and sources are located in Appendix Table A3, and descriptive statistics appear in Appendix Table A4.
Results with the information sharing dependent variable can be found in Table 2, and they align with theoretical expectations. Two of the three measures of geopolitical closeness—un voting ideal pt. distance and rivalry-peace—attain statistical significance in the expected direction. Geopolitics then clearly shapes pooling outcomes because IOs whose leading shareholders vote more similarly in the UNGA and have more peaceful foreign relations are more likely to sign information sharing agreements. Also in line with expectations, IOs whose memberships overlap more are more likely to pool resources, perhaps because they wish to minimize forum shopping. In contrast, results for resource dependencies are mixed, and HQ proximity appears to matter little for information sharing.
Development regression results (information sharing)
Notes: Robust standard errors are clustered at the dyad level. * p < .10; ** p < .05; *** p < .01.
The geopolitical effects that I detect in these models are substantively meaningful. A one-point increase in un voting ideal pt. distance (on a six-point scale) between leading IO stakeholders is associated with a 20 percent reduction in the number of information-sharing agreements signed between IOs. Similarly, moving from 0 to 1 on the rivalry-peace scale leads the number of information-sharing agreements signed between organizations to increase by over 30 percent. These effect sizes and their associated confidence intervals are plotted in panel (a) of Figure 1.
Substantive effect sizes
Results are even more consistent with my hypothesis when I examine co-financing patterns. As Table 3 shows, all three measures of geopolitical relations attain statistical significance at conventional levels. The effects are again substantively quite large—a one-point increase in un voting ideal pt. distance is associated with a reduction in the number of information-sharing agreements signed between IOs by nearly 20 percent. Similarly, moving from 0 to 1 on the rivalry-peace scale is associated with an increase in the number of co-financed agreements by nearly 250 percent. Last, IOs controlled by allies sign over 35 percent more agreements than those controlled by non-allies. These effects and their associated confidence intervals can be found in panel (b) of Figure 1.
Development regression results (co-financing)
Notes: Robust standard errors are clustered at the dyad-level. * p < .10; ** p < .05; *** p < .01.
Looking to the covariates, we see that member overlap again attains statistical significance because IOs are much more likely to co-finance projects if they share more member states. However, in contrast to the information sharing results, resource dependency considerations appear important for co-financing. This finding is perhaps unsurprising, since co-financing decisions might be heavily influenced by organizational financial constraints, given the substantial expenses associated with many development programs.
I perform a number of robustness checks to increase confidence in my findings. First, I swap the time trend for year fixed-effects.Footnote 58 Second, I examine the bivariate relationship between each of the geopolitical and pooling measures.Footnote 59 Results are robust.
Next, I incorporate proxies for portfolio similarity between IOs and the riskiness of IO lending, which represent important alternative explanations. IOs that have more similar lending portfolios might cooperate more, generating basic efficiency gains.Footnote 60 Similarly, IOs may cooperate more when they lend to particularly risky countries, since cooperation spreads risk across IOs.Footnote 61 I opt not to include these measures in my primary models because comprehensive lending data are available for only fourteen of the twenty-eight development IOs. Incorporating these variables then substantially reduces my sample size. However, results remain similar.Footnote 62 Last, I iteratively drop each IO from the data in order to ensure that the models are not overly sensitive to influential points.Footnote 63
Experimental Results
While the observational regression results presented thus far provide support for my hypothesis, they offer no causal leverage. To provide causal support for my mechanism, I deployed a survey experiment to a sample of bureaucrats employed at various development IOs through LinkedIn. I believe I am the first to use this method for elite sampling. The advertisement linked to a Qualtrics survey.Footnote 64 I targeted employees of the twenty-eight development IOs that I list in Appendix Table A2.Footnote 65 The resulting potential pool of respondents included around 60,000 LinkedIn members who were either currently or recently employed at one of the IOs. From this group, I was able to recruit around 200 respondents.Footnote 66 Of these, 147 responses were usable.Footnote 67 The breakdown of these respondents by IO of employment can be found in Appendix Table A17. Respondents were entered into a lottery to win a USD 250 Amazon gift card for their participation.Footnote 68
In the survey, each respondent answered a series of demographic and employment-related questions before they were presented with a profile for a hypothetical development organization. The respondent was informed that the organization's activities substantively and geographically overlap with those of their own organization. Each respondent was asked to assess five randomly selected organizational profiles before the end of the survey. Each organizational profile involved three key treatment arms.
First, to get at geopolitical relations between leading IO shareholders, I provided information about the location of the organization's headquarters and largest contributing shareholder; the host country was always the same as the primary shareholder. The country and headquarters' city of interest were one of Russia (Moscow), China (Beijing), or the US (Washington, DC). These countries serve as the leading stakeholders of several prominent international financial institutions, and they vary in terms of their geopolitical stances.Footnote 69 Second, I provided the number of member states that belong to the organization (5, 50, or 100). These three options respectively correspond to the membership profiles of the smallest development organizations (e.g., the Development Bank of the Great Lakes States), regional development banks (e.g., the Asian Development Bank), and more global organizations (e.g., the World Bank or International Fund for Agricultural Development). Third, to account for resource considerations, I offered details about the volume of lending administered by the organization in 2018 (USD 500 million, USD 10 billion, or USD 20 billion). The amounts are roughly representative of the tertiles of disbursements by development organizations in 2018. A sample organizational profile for an IO led by China with high disbursement levels and inclusive membership appears in Figure 2. The number of responses falling into each treatment condition can be found in Appendix Table A18.
Sample organizational profile
I utilize characteristics from the organization of employment for each respondent as well as the information in the organizational profiles to construct comparable variables to those used in the observational regression analyses. I then similarly model the survey data in a regression framework. The dependent variable is support for pooling with the hypothetical organization described in the profile. Each respondent is asked to report their level of support for both co-financing and information sharing with the organization in question on a scale of 1 to 10. The geopolitical covariates remain the same as in the previous models, and they measure alignment between the leading shareholders in the hypothetical organization and the organization that employs the respondent: un voting (ideal pt. distance), rivalry-peace, and alliance.Footnote 70
To ensure that all variation in the geopolitical measures is driven by the experimental manipulation of the leading shareholder, I include respondent organization fixed-effects in these models. With organization fixed-effects, the treatment alters the geopolitical covariates in the same way for each respondent from that organization, so variation is “as if” randomly assigned. In other words, because each respondent from a given organization shares the same baseline values on the observational variables, all variation in these measures is driven by the randomly assigned change in the most powerful stakeholder. The experiment therefore allows me to generate exogenous variation in geopolitical relations among leading IO shareholders.
I also include relevant control variables. To capture resource dependencies, I utilize the difference in disbursement amounts.Footnote 71 I also control for the number of members in the IO described in the organizational profile, since staff may prefer pooling with more inclusive IOs. I then include several individual-level covariates, namely age, tenure, and gender.Footnote 72 The full survey questionnaire including all organizational profiles, appears in Appendix 5.2, and descriptive statistics for these data can be found in Appendix Table A19.
The unit of analysis for these models is the respondent-organization. In other words, I treat each of the five organizations that a respondent evaluated over the course of the survey as a separate observation. This approach has several advantages. First, it allows me to maximize the number of observations despite having a relatively small pool of respondents; this is a common problem with elite samples. Second, it allows me to utilize incomplete responses—several respondents failed to complete the entire survey, though they did evaluate at least one organizational profile. I therefore also control for the response iteration. Model type is ordinary least squares.
The results with the information-sharing dependent variable can be found in Table 4. Geopolitical relations between leading member states are clearly an important factor for respondents because rivalry-peace is highly statistically significant. The result is also substantively meaningful—a change from rivalry to security community on this measure is associated with a fourteen-percentage-point increase in support for information sharing. In contrast to the observational results, un voting ideal pt. distance is statistically insignificant in this model, which perhaps suggests that bureaucrats are primarily averse to information sharing with adversaries. Meanwhile, resource dependencies again appear unimportant for information sharing, and bureaucrats do not seem concerned with the number of members in the hypothetical IO. In contrast, geographic considerations seem to matter because support for information sharing is higher when the hypothetical IO is geographically proximate. Last, none of the individual-level covariates achieve statistical significance.
Experimental results (nformation sharing)
Notes: Robust standard errors are clustered at the respondent organization-level. Respondent organization fixed-effects are included. * p < .10; ** p < .05; *** p < .01.
We see similar results with the co-financing DV, as the rivalry-peace and alliance variables both attain statistical significance. Moving from 0 to 1 on rivalry-peace and alliance generates increases of eighteen and eleven percentage points in support for co-financing respectively. Moreover, as is the case for information sharing, resource dependency concerns do not appear particularly salient to bureaucrats, while geographic considerations remain important. On the whole, then, geopolitical factors are clearly top of mind for IO bureaucrats when they consider co-financing and information-sharing opportunities.Footnote 73
I again run several robustness checks. First, I restrict the experimental data set to only those respondents who completed the survey.Footnote 74 Rivalry-peace and alliance attain statistical significance in the expected direction.Footnote 75 Next, I remove respondents whose job titles suggest that they are not involved in the negotiation and implementation of pooling arrangements. These include interns, receptionists, human resources staff, and research assistants. The remaining staff are likely directly involved in program negotiation and implementation—their roles include “local expert,” “project executive,” “lead economist,” and “regional director.” Interviews confirm that these are the types of staff involved in executing pooling arrangements.Footnote 76 Results are even stronger than in the initial models, as the rivalry-peace and alliance measures attain statistical significance for both dependent variables under this specification.Footnote 77
In another model, because my sample contains more staff from Western than non-Western IOs and is therefore not representative of staff across all IOs in the issue space, I weight each response by the share of total development IO staff employed by the relevant IO.Footnote 78 Additionally, I swap out organization fixed-effects for respondent fixed-effects. I do not use this as my primary model because it raises concerns about statistical power, but within-respondent variation in my observational measures is again randomly assigned in this specification. Results are robust.Footnote 79
Finally, to better assess the mechanism behind my results, I asked respondents in an open-ended question to describe the factors that were most important for their cooperation assessments. A word cloud derived from these responses can be found in Appendix Figure A4. The responses point predominantly to geopolitical concerns. For instance, a World Bank official said that “Most of the state institutions in both Russia and China, development institutions not being an exception … many of the financing decisions are made based on political interests.” Some staff even admitted to ideological biases that favor their own organization's primary shareholder, as was the case for a former country officer from the World Bank: “I confess to a Washington consensus bias.” Similarly, another respondent stated “The information should not be shared with any country who are not a friend. The information sharing must be on the base of mutual agreement and friendship.” A number of other respondents pointed to perceptions of trust or the reputation of the host country, both of which are shaped by individuals’ ideological biases. For example, an economist from the ADB responded that a major consideration is “mutual trust at the country level based on the international reputation.” In sum, autonomous IO staff commonly cite geopolitical factors as salient when considering cooperation with other organizations, and their concerns align with the preferences held by the organization's most powerful stakeholder.
Experimental results (co-financing)
Notes: Robust standard errors are clustered at the respondent organization-level. Respondent organization fixed-effects are included. * p < .10; ** p < .05; *** p < .01.
Conclusion
IOs increasingly pursue pooling as a strategy to combat institutional crowding, though it is understudied. Drawing on elite interviews, I argue that IOs that are dominated by friendly states are more likely to pool resources than those controlled by foes. I test my argument with original, comprehensive data on patterns of co-financing and information sharing in the development issue area. I also supplement this analysis with a survey experiment performed on a sample of development bureaucrats through LinkedIn. I find that for both forms of pooling and across an array of tests, geopolitics dominates competing explanations.
This project builds on the large body of literature identifying powerful member state influence in IOs broadly and international financial institutions in particular.Footnote 80 These IOs are an important channel through which powerful states realize their foreign policy objectives, and this is true even for policies over which IO staff have a good deal of autonomy, such as information sharing and co-financing. This is because IO staff sometimes act in the interest of important principals even in the absence of any formal intervention by these powerful member states.Footnote 81 IO staff can then be both autonomous and exhibit bias toward major stakeholders.
Future work should examine the applicability of this theoretical framework in other issue contexts. My argument specifically applies to any situation where operational IOs perform overlapping activities and IOs are asymmetrically controlled by a single powerful state or set of homogeneous states. Scholars might also probe more puzzling instances of cooperation between IOs. A particularly salient case is World Bank cooperation with the Chinese-led Asian Infrastructure Investment Bank (AIIB)—they signed a cooperation framework in 2017.Footnote 82 My theoretical framework suggests that such cooperation may be short-lived since staff should prefer to cooperate with more like-minded IOs. One possible explanation for this puzzling trend is that the AIIB sought cooperation with the World Bank to boost its legitimacy and attract business as a relatively new development IO. Cooperation may then cease in future years as the AIIB establishes itself as a premier development institution and geopolitical tensions begin to take hold.
Acknowledgments
I thank Ryan Brutger, Allison Carnegie, Don Casler, Lindsay Dolan, Benjamin Faude, Nikhar Gaikwad, Jessica Green, Cleo O'Brien-Udry, Jack Snyder, Keren Yarhi-Milo, Noah Zucker, and the editor and two anonymous reviewers at International Organization for helpful comments on previous drafts. Participants at several workshops at Columbia University also provided valuable feedback.
Data Availability Statement
Replication files for this research note may be found at <https://doi.org/10.7910/DVN/0JFJ5V>.
Supplementary Material
Supplementary material for this research note is available at <https://doi.org/10.1017/S0020818321000229>.
