Trusting relationships,learning bureaucrats: International organizations and early-stage policy diffusion

Abstract

This article examines the diffusion of the conditional cash transfer (CCT) model to Turkey and Indonesia, and the role of World Bank bureaucrats in these cases of early-stage diffusion. The article finds that learning, and not coercion or emulation, is the primary mechanism of policy diffusion in both cases. This learning was mediated by the World Bank bureaucrats even before the CCT model gained mainstream acceptance inside the World Bank. The findings from these two cases suggest World Bank bureaucrats to be engaging in domestic policy processes not by ‘powering’ but by participating in the national bureaucrats’ ‘puzzling’. The findings also underline the importance of trusting relationships between international and national bureaucrats in these policy processes. More broadly, the article makes the case for conceptualizing international organizations (IOs) as organizations with heterogeneous staff who play more nuanced and contingent roles in policy diffusion processes than is commonly conceived.

Introduction

In July 2001, International Food Policy Research Institute (IFPRI) published the first report on the evaluation results of Mexico’s pioneering cash transfer programme, Progresa (Skoufias and McClafferty, 2001). Just 2 months later, the Turkish government announced the launch of a conditional cash transfer (CCT) programme with striking similarities to Progresa. Four years later, on the other side of the Asian continent, Indonesia followed suit. Turkey and Indonesia were the first countries in their respective regions to adopt the CCT model à la Mexico.

How was this possible? It seems unlikely that policymakers in Mexico, Turkey and Indonesia independently came up with such similar programmes, nor does it seem likely that an evaluation report from a Mexican programme would be read and adopted by Turkish and Indonesian bureaucrats without intermediation. In both cases, World Bank bureaucrats were actively involved. But this only partly answers the question and raises an even more puzzling one given that the CCT model was barely known inside the Bank in 2001 and acquired only limited internal support by 2005.

How is it possible for an international organization (IO) to play a role in a policy model’s diffusion before that model is widely accepted internally? Without such acceptance, an IO is unlikely to push for a model’s adoption, to promote its normative value, or to market it as best practice. Yet CCT model’s diffusion to Turkey and Indonesia shows that even before a policy’s widespread internal acceptance, an IO could still play a role, in fact a critical role, in transnational policy diffusion.

Empirical studies on IOs’ engagement in policy diffusion almost exclusively focus on later stage cases. This article examines IOs’ role in early-stage diffusion by analysing diffusion processes, identifying diffusion mechanisms and examining IO bureaucrats’ roles. The findings include several insights about the micro-dynamics of IO bureaucrats’ influence on domestic policymaking and learning as a diffusion mechanism mediated by IOs.

These findings suggest that IO bureaucrats’ engagement in domestic policymaking is in some cases less about ‘powering’ and more about ‘collective puzzlement’, two concepts employed by Heclo (1974) in examining social policymaking in Britain and Sweden. In the cases studied, IO bureaucrats and national bureaucrats were not strangers-in-suits sitting across from each other exchanging bargaining chips and making subtle threats. They were individuals who trusted and respected each other, hovering around the same analyses on a computer screen and discussing ideas. The article contends these trusting interpersonal relationships to be critical in early-stage policy diffusion via learning. When juxtaposed with earlier studies of CCT’s diffusion (Borges, 2018; Brooks, 2015; Hall, 2015; Simpson, 2017; Sugiyama, 2011), these findings demonstrate how diffusion mechanisms and IOs’ roles vary across time and space even for the same policy.

Literature review

Empirical research on IOs’ influence on domestic policymaking reveals a wealth of tools utilized by IOs, including hard power tools such as conditionalities, inducement, paternalism and signalling (Stone, 2004); soft power normative tools such as contestation and integration (Bauhr and Nasiritousi, 2012); and soft power knowledge tools such as framing and marketing (Brooks, 2005), meetings and study tours (Bennett et al., 2015), and data and advice (Knill and Bauer, 2016). This article builds on a growing strand of this literature focussing on a particular form of IO influence, that is, transnational policy diffusion whereby ‘policy decisions in a given country are systematically conditioned by prior policy choices made in other countries’ (Simmons et al., 2006). In this literature, IOs are predominantly conceptualized as monolithic actors in pursuit of predefined policy goals and portrayed as policy diffusion agents mediating the diffusion process using their soft normative or knowledge powers (Brooks, 2005; Ogden et al., 2003; Walt et al., 2004). This article focusses on an understudied topic in this literature: IOs’ influence in early-stage policy diffusion. By doing so, the study findings put a spotlight on individual IO bureaucrats and their engagement in domestic policymaking before a policy agenda emerges inside their IO.

Methodology

The study uses within-case process tracing with a comparative angle (Starke, 2013). It employs a ‘process-tracing strategy’ as opposed to a ‘pattern finding strategy’ (Lee and Strang, 2006), and it complements within-case analyses with cross-case analyses (Obinger et al., 2013) to draw causal inferences (Fairfield, 2013). Relatedly, the study employs a micro-level perspective. Macro-level studies of policy diffusion employing a pattern-finding strategy and using event history analysis are approaching a point of saturation; yet, the need remains for studies investigating diffusion mechanisms (Dobbin et al., 2007; Lee and Strang, 2006; Maggetti and Gilardi, 2013; Marsh and Sharman, 2009; Obinger et al., 2013). Hence, analysing micro-processes is a promising approach to investigating diffusion mechanisms and IOs’ roles in these mechanisms.

Case selection

The cases were selected using a two-staged approach: selection of the policy followed by selection of countries. Two reasons underlie CCT model’s selection. First, the World Bank was already identified as a player in CCT’s transnational diffusion (Hall, 2007; Sugiyama, 2011), making it a positive case for investigating IO’s role. Second, the model built on Mexico’s Progresa programme is a ‘hard’ policy as opposed to a ‘soft’ policy, a distinction put forward by Evans and Davies (1999), which, according to Weyland (2006), minimizes the possibility of ‘spurious diffusion’ and, according to Braun and Gilardi (2006), facilitates tracking the diffusion process.

For country selection, a subset of positive cases of early adopters outside of Latin America, the primary region of diffusion, were identified – a case selection approach recommended for investigating diffusion processes (Weyland, 2006). Turkey and Indonesia were selected from this subset as ‘diverse cases’ to maximize variation on World Bank’s role, a selection criterion considered appropriate for policy diffusion studies (Obinger et al., 2013; Starke, 2013). World Bank’s tools of engagement (financial products and technical assistance) guided this selection. Specifically, the Bank gave a loan, a financial product, to Turkey and provided technical assistance to Indonesia. By examining diverse cases, the study investigated if and how the Bank’s role in CCT’s early diffusion varied across engagement tools.

Data collection

Elite interviews were the primary data collection method. Chain-referral sampling was used with multiple entry points identified through document reviews and personal contacts; 39 interviews were conducted in Ankara, Jakarta and Washington, DC and via Skype. ¹ Almost all individuals identified in document reviews and mentioned by interviewees as pertinent actors in CCT’s diffusion processes were interviewed. ² Individuals in national agencies and the Bank who were unsupportive of CCT’s adoption were also interviewed to minimize confirmation bias.

Using narrative interviewing, interviewees were asked to tell their story of CCT’s adoption. A preliminary document review was done before the fieldwork to inform the interviews. Additional document review was conducted after coding transcriptions to strengthen the evidentiary basis. Documents reviewed included publicly available reports, programme documents provided by the World Bank Archives and documents shared by interviewees.

Findings on the diffusion processes

Although the Bank’s engagement tools were different, with Turkey receiving a loan and Indonesia receiving technical assistance, the similarities between the diffusion processes and Bank bureaucrats’ roles are striking. In both cases, an economic shock triggered the search for a new policy (2001 economic crisis in Turkey and 2005 fuel subsidy adjustment in Indonesia); the decision process was driven predominantly by bureaucrats with a single minister giving the ultimate approval; and Bank bureaucrats played a critical role in CCT’s introduction. Also in both cases, trusting relationships between national bureaucrats and Bank bureaucrats played a mediating role. The rest of this section presents the findings about these diffusion processes focussing on framing events, policymaking processes and enabling dynamics.

CCT’s diffusion to Turkey

Framing events

The CCT model was adopted in 2001 as part of a social assistance project designed in response to an economic crisis and funded partially by a Bank loan. Two events after the crisis shaped the course of the subsequent policymaking process. The first was a cabinet-level meeting about the post-crisis World Bank loan package. Following a suggestion by the state minister responsible for the social assistance portfolio, the decision was made to allocate US$500 million from the loan package for assistance to poor households (interviews 27, 33, 36). This cabinet-level decision set the boundaries of the fiscal space of the social assistance package. The subsequent policymaking process was, in part, an exercise in pulling together a ‘just-large-enough’ package of interventions.

The second was the decision of the Bank’s Ankara-based lead economist to invite a particular social protection expert from the headquarters. This expert had worked in Turkey and just helped prepare the loan agreements with Colombia and Jamaica that included the CCT model (interviews 2, 4). The lead economist had been a peer reviewer of this loan agreement with Jamaica and thought the CCT model would be appropriate for post-crisis Turkey (interview 4). By inviting this particular expert, the lead economist indirectly shaped the plausible set of policy proposals. Predictably, this expert’s initial dialogues with Turkish bureaucrats and preparatory document presenting plausible post-crisis social assistance interventions included the CCT model (interviews 2, 4, 6, 27, 35). More importantly, this document expressed a preference for the CCT model. ³ Despite doubts raised by some Bank bureaucrats about its appropriateness in internal discussions, ⁴ CCT was pushed onwards as a recommended policy in conversations with national bureaucrats. In September 2001, less than 6 months after a Bank expert introduced the CCT model to Turkish bureaucrats, the agreement for a US$500 million loan package with CCT as its largest component was signed.

Policymaking process

The above-mentioned Bank expert engaged with a small circle of bureaucrats from the Treasury, the State Planning Agency, and the Social Assistance and Solidarity Fund (Sosyal Yardımlaşma ve Dayanışmayı Teşvik Fonu (SYDTF)) concerning the CCT component of what eventually became the Social Risk Mitigation Project (SRMP). When the Bank expert described the CCT model to these national bureaucrats, it was their first time hearing about it (interviews 27, 33, 35). An SYDTF bureaucrat assigned to design the social assistance project emerged as the intermediary between Bank bureaucrats and the pertinent state minister who was the ultimate decision maker.

Data from interviews and document reviews are insufficient to draw inferences with certainty about the decision point for CCT’s inclusion in SRMP. The most likely scenario involves the Bank expert convincing the SYDTF bureaucrat about CCT’s desirability followed by the SYDTF bureaucrat persuading the state minister. During the same period, Bank bureaucrats also had meetings with the minister and provided information about other countries’ experiences with SRMP’s different components, including the CCT (interviews 1, 2).

Details of the SRMP were determined during working meetings attended by national and Bank bureaucrats. This was followed by the project information document’s (PID) preparation. During the same period, a crisis social assessment and a welfare distribution impact simulation were conducted by Bank bureaucrats. Subsequently, Bank bureaucrats prepared the project appraisal document (PAD) detailing out the components of the SRMP, including the CCT.

After the loan agreement was signed in September 2001, the CCT model was piloted by SYDTF and subsequently implemented nationwide. The same Bank expert was involved in the technical design of the CCT’s piloting and implementation. In addition to providing direct technical support, Bank bureaucrats organized study visits to Latin America for national bureaucrats and made relevant documents available during this period (interviews 1, 2, 6, 27, 33).

Enabling dynamics

Two dynamics acted as the grease oiling the wheels of CCT’s diffusion to Turkey. The first concerns the motivation of relevant national and Bank bureaucrats. They advocated for the CCT model’s adoption not only to mitigate the effects of the economic crisis on poor households but also to improve social assistance delivery mechanisms (interviews 2, 27). This latter motivation contradicts the dominant view in the diffusion literature about the satisficing bureaucrat who draws lessons from other places to find quick solutions in the face of a crisis (Rose, 1991). In contrast, these bureaucrats used fiscal and political enablers created by a crisis to change an inadequate, yet unbroken, social assistance system.

The second concerns the relationships between some national bureaucrats and Bank bureaucrats. Characterized by positive affect and trust, some were social in nature and involved informal interactions in social settings. Many were built progressively through repeated work interactions, particularly in emotionally charged situations, like the relief efforts after the 1999 Marmara earthquake (interviews 1, 2, 27). Some persisted personally or professionally long after 2001 (interviews 2, 5, 6, 27, 35). These relationships characterized by positive affect and trust played an important role in CCT’s diffusion. Not only national bureaucrats were willing to meet and listen to the Bank bureaucrats, they were also willing to trust the information and advice presented.

CCT’s diffusion to Indonesia

Framing events

The CCT model came onto the policy agenda in 2005 after a major fuel subsidy adjustment (Widianto, 2013). This subsidy adjustment was critical in CCT’s diffusion: it triggered the search for assistance programmes to accompany the adjustment and created the fiscal space for them. During this period, Bank bureaucrats in Jakarta provided technical analyses on the welfare effects of subsidy adjustment options and compensation packages to senior bureaucrats (interview 20). At this point, Bank bureaucrats in Jakarta had limited knowledge about targeted cash transfer models including the CCT, so they hired a consultant with expertise in compensation schemes to build their own knowledge base (interviews 10, 21, 28).

Bank bureaucrats and this consultant introduced the idea of cash transfers to a senior bureaucrat and an advisor during informal meetings in July 2005 (interviews 10, 21). Soon after, at a meeting attended by cabinet ministers and following a presentation by Bank bureaucrats and consultant on compensation schemes and cash transfers, the Vice President opted for a short-term unconditional cash transfer (UCT) programme to expand the compensation package and rejected the CCT option (interviews 9, 10, 19, 20, 21). Cash transfers were included in the compensation package partly because existing assistance programmes’ expansion was too slow and small to fill up the fiscal space created by the adjustment (interview 12). UCT was chosen over CCT partly because it could dispense large amounts of money quicker (interviews 12, 18) (World Bank, 2005).

Policymaking process

Despite Vice President’s rejection, the CCT idea gained considerable traction among key national bureaucrats and Bank bureaucrats, opening the path to its eventual piloting. Serendipity’s role is noteworthy in this regard. During their consultant search, Bank bureaucrats did not look for a CCT expert, suggesting that CCT was not their priority. The consultant hired happened to have extensive knowledge about CCTs and he influenced both Bank and national bureaucrats. If this consultant instead knew only about UCTs, the CCT model would probably not diffuse to Indonesia then.

The piloting decision took place in a small circle of Bank bureaucrats and national bureaucrats. One month after UCT’s first tranche in October 2005, a World Bank document already referred to the Government’s agreement to pilot the CCT programme upon UCT’s completion in 2006 (World Bank, 2005). Contrary to findings of another study on CCT’s adoption in Indonesia (Kwon and Kim, 2015), it is unlikely that CCT was introduced to replace UCT because of negative public reactions to UCT, given this reference to Government’s agreement to piloting the CCT in a Bank document in October. This is supported by the fact that only four coupons were distributed for the quarterly payments under the UCT, suggesting that it was temporary by design.

Data from interviews and document reviews are insufficient to draw inferences with certainty about the precise decision point for piloting the CCT model. The most likely scenario involves a small group of Bank bureaucrats and national bureaucrats making use of the fiscal space and policy attention to introduce what they saw as the next generation of poverty alleviation programmes. Informal meetings between Bank bureaucrats and national bureaucrats after the above-mentioned meeting with the Vice President influenced the solidification of the bureaucratic support for CCTs (interviews 15, 18, 19, 20).

An informal fiscal green light for CCT’s piloting was given in early 2006. The subsequent programme design process received technical support from the Bank in the form of expert secondments and provision of technical documents (interviews 12, 13, 14, 19, 20). In June 2006, several key national bureaucrats attended an international CCT conference in Turkey organized by the Bank, which further solidified their support for piloting the CCT (interviews 15, 19, 21). Subsequently, the pilot was included in the draft annual budget submitted to the Parliament. Upon Parliament’s approval, the piloting began in 2007 in seven provinces covering about half a million families. Its coverage has since been expanding – 10 million families benefitted from the programme in 2018.

Enabling dynamics

Mirroring CCT’s diffusion to Turkey, two dynamics ‘oiled the wheels’ of CCT’s diffusion to Indonesia. The first concerned bureaucrats’ motivation. CCT’s desirability for these bureaucrats was less centred on the mitigation of adjustment’s effects and more centred on the opportunity to introduce a more effective approach to social assistance (interviews 10, 18). Thus, CCT’s diffusion to Indonesia, like Turkey, is not a story about satisficing bureaucrats drawing lessons from other countries haphazardly after a crisis (Rose, 1991) but a story about bureaucrats using fiscal and political opportunities to improve the social assistance system.

The second dynamic concerned the relationships between national bureaucrats and Bank bureaucrats. These relationships influenced CCT’s diffusion to Indonesia, even more than in Turkey, given the intensity of anti-World Bank sentiment in Indonesia (interview 9). Not only national bureaucrats were willing to listen to Bank bureaucrats during at times backdoor meetings, they also trusted the analyses and advice presented. In fact, in some cases, national bureaucrats sought out technical support from individual Bank bureaucrats (interviews 10, 18, 20). These trusting relationships enabled individual Bank bureaucrats’ engagement in policy processes leading to CCT’s diffusion.

Some relationships pre-existed the subsidy adjustment when these individuals wore other hats in different organizations (interviews 9, 18). Similar to Turkey, working together during the emotionally charged period after the 2004 tsunami is seen as strengthening the trust in these relationships (interviews 10, 18). Some relations established during the subsidy adjustment were based on referrals from these older relationships. Several persisted by the time the study interviews were conducted in 2014 (interviews 10, 20).

Findings on the diffusion mechanisms

The literature on policy diffusion typically distinguishes between learning, emulation and coercion as mechanisms of policy diffusion. Accordingly, policy diffusion can emerge as the outcome of ‘lesson-drawing’ and ‘learning’ stages of policymaking (Boehmke and Witmer, 2004; Bouché and Volden, 2011; Rose, 1991; Simmons and Elkins, 2004), ‘emulation’ conducted primarily to send a symbolic signal in quest for legitimacy for a policy with a norm-like status (Elkins and Simmons, 2005; Finnemore, 1993; Finnemore and Sikkink, 1998; Johnston, 2001) or ‘coercion’ (Dobbin et al., 2007; Sharman, 2008).

While these diffusion mechanisms are relatively distinguishable in their ideal forms, that task becomes challenging when empirical observations lie closer to definitional boundaries (Dobbin et al., 2007; Marsh and Sharman, 2009). This article employs four tests to empirically distinguish between learning, emulation and coercion, which build on their defining features and resemble a series of ‘hoop tests’ and ‘straw-in-the-wind tests’, two types of empirical tests used in process tracing (Bennett, 2010; Mahoney, 2012). The analyses using these tests strongly suggest learning, and not coercion or emulation, to be the primary mechanism in both cases.

Test #1 on normative appeal of policy: does the policy have social value or normative appeal at the time of diffusion?

This hoop test for emulation focuses on the nature of the policy. If the answer is yes, then the mechanism might be, but does not have to be, emulation. If the answer is no, then the primary mechanism is not emulation.

The CCT model was in the nascent stages of becoming a globally recognized model in both 2001 and 2005 (see Supplemental Annex 1 on model’s origins and diffusion). In 2001, the model was just starting to find traction in some Latin American countries and inside the World Bank (2005), it gained some popularity in Latin America but was mostly unknown elsewhere, and had limited traction among Bank’s sectoral experts, particularly in country offices. In Indonesia, national bureaucrats were aware of the competing views about the CCT inside the Bank (interview 13). Thus, in 2001 and 2005, CCT had not yet gained its faddish status inside the Bank or reached a tipping point in its global diffusion.

Short of being perceived as a fad by national bureaucrats, it is unlikely that CCT’s diffusion to Turkey and Indonesia was driven by its normative appeal or quest for international legitimacy. It also is unlikely that CCT’s social value drove its adoption as such value is often peer-based (Berry and Berry, 2007; Brooks, 2005; Obinger et al., 2013). Neither Turkey nor Indonesia had substantial contact or ties with the model’s pioneers Mexico and Brazil. Overall, this test suggests it is unlikely that emulation was the primary mechanism of diffusion to Turkey or Indonesia.

The absence of CCT’s normative appeal in 2001 and 2005, however, cannot ignore the ideational forces of the broader social protection paradigm promoted by the Bank in early 2000s. CCT’s alignment with the broader social protection paradigm promoted by the Bank may have laid the groundwork for national bureaucrats’ receptiveness to the model. In other words, while it is unlikely that emulation was the primary mechanism for CCT’s diffusion, it is possible that the dynamics of emulation in the broader social protection sector were indirectly at play in these diffusion episodes.

Test #2 on external actor involvement: is a powerful country/IO playing a leadership role in diffusion?

This hoop test for coercion focuses on external actors’ involvement. If the answer is yes, then the mechanism might be, but does not have to be, coercion. If the answer is no, then the mechanism is not coercion.

In 2001 and 2005, Mexico was the only country promoting the CCT model. Mexico could not have used hard or soft power to coerce Turkey or Indonesia into adopting CCTs. As for IOs, however, the Bank was involved in the model’s diffusion to both countries. The relevant question then concerns the nature of Bank’s involvement.

Explicit conditions were not part of the CCT-related discussions in Turkey or Indonesia (interviews 2, 10, 35). Interviews with national bureaucrats and Bank bureaucrats repeatedly refer to their equal relations and national policymakers extensive ‘latitude for choice’, a concept and finding also put forward by Weyland (2006) in his analysis of policy diffusion in the social sector in Latin America (interviews 1, 2, 3, 4, 5, 9, 15, 22). Indeed, similar proposals introduced earlier by Bank bureaucrats were rejected by national bureaucrats (interviews 1, 3, 10), also suggesting an extensive ‘latitude for choice’. Thus, it is unlikely that coercion was the primary mechanism of diffusion in either case.

Test #3 on domestic policymakers’ motives: are policymakers seeking a solution to a problem or are they trying to send a signal?

This straw-in-the-wind test helps differentiate between emulation and learning. If domestic policymakers are genuinely seeking a solution to a problem, then learning is more likely to be the mechanism. If they are trying to signal, then emulation is more likely to be the mechanism.

Empirical observations suggest that in both cases, CCT appeared on the policy agenda while searching for policy solutions. For policymakers’ motives, however, empirical observations are less conclusive and cannot decipher their precise composition. Policymakers may have been concerned with mitigating the negative effects on poor households or sending a signal to the public to prevent social unrest or both.

Using counterfactual analysis, however, CCT’s adoption in both countries seem primarily motivated by concern about effects on poor households combined with some bureaucrats’ desire to improve the social assistance system. If the primary motive was sending a signal to the public to prevent social unrest, rapidly disbursed and highly visible UCTs would have been the obvious choice. In fact, in both countries, UCTs were carried out immediately and received sizable media coverage (Hastuti et al., 2006; Widjaja, 2009; Zabci, 2006). In contrast, CCT’s piloting came many months after the economic shocks when the immediate risk of social unrest had long diminished (Gemici, 2013). Thus, this test supports the first test’s conclusion: emulation is unlikely to be the primary mechanism.

Test #4 on depth and nature of learning: are key policy actors interested in learning about the model? What is the depth of learning during policymaking?

This straw-in-the-wind test also helps differentiate between emulation and learning. If the answer is yes and deep learning takes places during policymaking, then learning is the more likely mechanism.

In both cases, two consecutive stages of learning are observed. The first stage involves a small circle of bureaucrats and ministers who made the initial decision for CCT’s adoption. Learning at this stage was relatively shallow and mediated by Bank bureaucrats. Yet decision makers still requested information about policy options and experiences from national and Bank bureaucrats (interviews 1, 2, 10, 21, 33), suggesting genuine interest in learning about their options to assess them.

The second stage involves bureaucrats tasked with planning CCT’s piloting. Learning at this stage was deeper. It involved direct learning from other countries through Bank-organized conferences and visits, and indirect learning mediated by Bank bureaucrats. Overall, this test supports first and third tests’ conclusion that the more likely diffusion mechanism is learning and not emulation.

Together, these four tests strongly suggest that CCT’s primary diffusion mechanism in both cases was learning and not coercion or emulation.

Findings on Bank bureaucrats’ role in learning

Leading up to the piloting decision, Bank bureaucrats were primarily involved in identifying and exploring policy options. In Turkey, they presented a policy option paper, in which the CCT option was highlighted positively (interviews 2, 35). ⁵ In Indonesia, they hired a consultant on cash transfers to build their own knowledge, approached national bureaucrats offering technical support for designing a compensation package and facilitated informal meetings between these bureaucrats and the consultant (interviews 10, 20, 21, 22). After the Vice President excluded the CCT from the compensation package, Bank bureaucrats continued to engage with national bureaucrats through informal meetings about CCT’s piloting (interviews 15, 18, 19, 20).

During the policy design period, Bank bureaucrats remained engaged in both countries. In Turkey, they facilitated national bureaucrats’ learning by organizing study visits, providing technical documents and sharing their expertise (interviews 1, 2, 6, 27, 33). Similarly, in Indonesia, they organized study visits, supported participation in an international CCT conference and provided technical documents (interviews 12, 13, 15, 19, 20). In addition, the Bank seconded experts to support the pilot programme’s technical design (interviews 18, 19, 21).

During both periods, Bank bureaucrats’ trusting relationships with national bureaucrats were critical. They facilitated Bank bureaucrats’ engagement in domestic policy processes and influence on CCT’s adoption because national bureaucrats trusted the information and advice they presented. Thus, trust between national and Bank bureaucrats played a ‘mediating role’ in learning about the CCT model (Levin and Cross, 2004), which went beyond making a model salient and available, and watching the ‘availability heuristic’ do its magic. In Turkey, the key national bureaucrat trusted the Bank expert’s advice concerning CCT’s appropriateness and effectiveness despite having limited knowledge about the model’s details. Similarly, in Indonesia, key bureaucrats trusted Bank bureaucrats’/consultant’s advice about compensation schemes and CCTs without having in-depth knowledge about CCTs themselves. Hence, trusting relationships were critical in both cases for facilitating Bank bureaucrats’ access to policy processes and influence on policy decisions (interviews 2, 4, 5, 9, 10, 18, 21). Hence, these trusting relationships were a critical factor in CCT’s early-stage diffusion to Turkey and Indonesia. Without these relationships, it seems unlikely that Bank bureaucrats could engage in and influence policy processes as much as they did. Without their engagement and influence, it seems unlikely that the model would diffuse to Turkey and Indonesia as early as it did.

Contending trust-based interpersonal learning’s critical role in CCT’s diffusion to Turkey and Indonesia does not imply that learning was based solely on Bank bureaucrats’ mediation of information or that cognition had no role in learning. National bureaucrats took into consideration their experiences during both decision and design phases (interviews 19, 27, 33). During the design phase, they learned directly about the CCT model and verified their initial assessments. Thus, while the depth of direct learning varied across bureaucrats and elected officials, they were all influenced by the information and advice relayed by ‘trustworthy’ Bank bureaucrats/consultants.

Findings on political, fiscal and institutional context

Although this article focuses primarily on Bank bureaucrats’ role in CCT’s adoption, this focus does not imply that contextual factors were inconsequential. On the contrary, in both cases, the opening of a ‘policy window’ (Kingdon, 2011) as a result of political and fiscal factors was a necessary precondition to CCT’s adoption. Without the domestic context motivating a search for new social assistance policies, Bank bureaucrats could not have facilitated the subsequent policy learning.

The economic crisis in Turkey and the fuel subsidy adjustment in Indonesia both hold the traits of a ‘problem stream’ ripe for the opening of a ‘policy window’ (Kingdon, 2011) with both triggering the search for large-scale social assistance programmes. These unique periods also hold the traits of a ‘political stream’ ripe for a ‘policy window’ (Kingdon, 2011) – with politicians and bureaucrats highly responsive to social needs. The responsiveness in Turkey was primarily related to a devastating earthquake that showed corruption and lethargy’s deadly results and gave rise to a deep sense of mission among politicians and bureaucrats (interviews 30, 33). The economic crisis came as a second shock to the system and further heightened the high gear activity (interview 30). In Indonesia, the government had just come to power through the first-ever direct presidential elections and responded to the devastating tsunami of December 2004.

Without these policy windows, politicians and bureaucrats are unlikely to have adequate interest in new social assistance models. In fact, an earlier attempt by a Bank bureaucrat for the introduction of a cash programme for schooling was outright rejected (interviews 3, 34). Similarly, the social investment fund, which was introduced as part of the SRMP, was repeatedly rejected or put on the backburner by national bureaucrats despite Bank bureaucrats’ prior efforts for its adoption (interview 1).

The link between a policy window and new social assistance programmes, including the CCT, was strengthened by fiscal factors – namely, the presence of adequate fiscal space. However, contrary to findings about a positive relation between economic recovery from the debt crisis and CCT’s diffusion (Brooks, 2015), in both Turkey and Indonesia, the fiscal space for CCT model’s adoption emerged from economic shocks. For Turkey, it was created by a Bank loan in the midst of a deepening debt crisis and worsening recession risk, when borrowing from the Bank for the CCT programme made economic sense – it both facilitated cash insertion into a contracting economy and demonstrated to lenders Turkish economy’s continuing credibility (interviews 2, 33, 35). In Indonesia, a massive fuel subsidy adjustment created the fiscal space, and new social assistance programmes with fast and wide disbursement schemes partly responded to an economic contraction risk arising from a fiscal space whose size went well beyond the needs of budget balancing (interviews 12, 18) (The World Bank, 2005). Even after a large compensation package was introduced following the fuel subsidy adjustment, the remaining fiscal space was large enough that the pilot CCT programme’s inclusion in the national budget did not raise any eyebrows (interviews 13, 18).

It is probable that without these policy windows and fiscal space, the CCTs would still diffuse to Turkey or Indonesia later given its faddish status in late 2000s. Yet the policy windows and fiscal space were necessary for the CCTs to diffuse as early as it did. The policy window, however, was not a sufficient factor since it does not account for CCT’s emergence onto the policy agenda and its selection over alternative policies. Only when combined with some Bank bureaucrats’ engagement in policymaking process using their trusting relationships, the CCT model ended up diffusing to Turkey and Indonesia when it did.

While the political and fiscal factors created the necessary conditions for CCT’s diffusion, institutional factors pertaining to national actors played a role in shaping the policymaking processes – fast processes influenced greatly by bureaucrats – concerning CCT’s adoption. Elements of a developmental bureaucracy could be observed in both Turkey in 2001 and Indonesia in 2015, including the strong influence of state planning agencies, treasury and ministry of finance over policy decisions and a highly centralized governance structure. The crises in both countries gave rise to even faster and more closed decision processes, which augmented bureaucrats’ influence.

Institutional factors pertaining to the Bank, specifically its decentralized structure, also affected decision processes. In both countries, Bank’s decentralized structure allowed in-country bureaucrats to build close relations with national bureaucrats and provide timely input into time-sensitive decision processes (interviews 4, 10, 18). Unlike their predecessors, senior Bank bureaucrats working on Turkey were based in Ankara, allowing them to engage with national bureaucrats in the immediate aftermath of the crisis (interviews 2, 4). The social protection expert in contact with national bureaucrats about the post-crisis social assistance package was able to exert considerable sway internally because of being on the ground where considerable discretion could be used (interview 7). Similarly, in Indonesia, senior Bank bureaucrats were based in Jakarta and had pre-existing relations with national bureaucrats, which put them in the loop about the imminent fuel subsidy adjustment and subsequent policy discussions. Bank bureaucrats in Jakarta could also hire a consultant quickly and engage in technical conversations with national bureaucrats without any discernible involvement from Washington, DC (interview 10).

Discussion of the findings on learning

The findings underline the importance of trusting relationships in policy diffusion via learning in the two cases studied. By doing so, it draws attention to interpersonal social and emotional aspects of transnational policy learning, which receive limited attention in the pertinent international relations (IR) literature. Empirical studies on policy learning focus predominantly on the limits of cognition (Dolowitz and Marsh, 2000; Elkins and Simmons, 2005; Weyland, 2006); use models built on assumptions about limited time, information and cognitive ability; and examine resulting heuristics and biases in policy learning. They pay less attention to heuristics based on social and emotional factors; when they do, they look at the country or network level and not the individual level. Thus, interpersonal aspects of learning as a policy diffusion mechanism and IO bureaucrats’ roles in these dynamics remain mostly unexplored.

This gap concerning interpersonal social and emotional aspects of learning in the IR literature is puzzling. Several studies of emulation as a policy diffusion mechanism investigate the role of socialization and social concepts like empathy, affection and esteem (Finnemore and Sikkink, 1998; Johnston, 2001). General theories of diffusion pay considerable attention to interpersonal social elements, including Rogers’ (2003) foundational work that describes innovation diffusion as ‘a very social process that involves interpersonal communication relationships’. ⁶ There is also a pertinent strand in the broader IR literature that focuses on individuals, and not organizations, as the ‘real’ learners (Levy, 1994).

The social and emotional aspects of interpersonal learning highlighted by this study also find parallels in a long-standing body of research in social psychology that examines learning’s social nature and credence of social information in decision making (Bonaccio and Dalal, 2006; Collins et al., 2011; Lee and Dry, 2006). The study findings also speak directly to a strand in organizational learning literature that examines interpersonal trust as a moderator variable in knowledge diffusion via learning (Levin and Cross, 2004; Mayer et al., 1995; Rousseau et al., 1998).

Thus, building on the findings about the significance of trusting relationships in CCT’s diffusion to Turkey and Indonesia, and a wealth of parallel findings from studies in IR, innovation diffusion, social psychology and organizational learning literatures, a case can be made for incorporating interpersonal social and emotional elements into analyses of transnational policy diffusion via learning and IO’s roles in them.

Discussion of the findings on Bank bureaucrats’ role

Findings from Turkey and Indonesia cases suggest that Bank bureaucrats influenced domestic policy processes not through ‘powering’ but by participating in the national bureaucracy’s ‘collective puzzlement’ in finding new policies. National and international bureaucrats were together engaged in ‘collective puzzlement’ whereby they brought their personal experiences and convictions to the table, learning from and persuading each other. More concretely, in both cases, national bureaucrats were joined by Bank bureaucrats in their search for a policy for mitigating the effects of a crisis on poor households. In Turkey, a Bank bureaucrat with recent exposure to CCT model put it forward as a solution in this collective puzzlement but did so because of personal conviction and not because of Bank’s organizational position. In Indonesia, Bank bureaucrats were exposed to CCT’s details almost concurrently with their national counterparts. Following this exposure, Bank bureaucrats personally convinced of CCT’s appropriateness, presented it as a solution to an ongoing collective puzzlement. Similar to Turkey, they did so because of their personal conviction and not because of an organizational position. In fact, in both cases, other Bank bureaucrats actively opposed the adoption of the CCT model by Turkey and Indonesia.

These study findings contribute to a continuing debate in the growing body of literature examining the influence of IOs on social policy reforms in Turkey (Agartan, 2016; Güleç, 2014; Yilmaz, 2017a, 2017b). While the study findings about the CCT model’s adoption regarding the trusting and consensual relationships between national and international bureaucrats find close parallels in studies about the health care reform (Agartan, 2016; Yilmaz, 2017b), they differ in their assessment of the intentions and veracity of the claims about the nature of these relationships. Similarly, the findings speak directly to a point of contention in the scholarly debates about the power dynamics underlying donor engagements in Indonesia (Carroll, 2009; Crawford, 2003; Mallarangeng and Van Tuijl, 2004). The study findings closely parallel Mallarangeng and Van Tuijl’s (2004) position, who argue that decision-making processes are located in complex transnational settings, and that the donor–recipient dichotomy assuming internal homogeneity is too narrow to capture such processes. More broadly, these findings about Bank bureaucrats’ role diverge from pertinent studies on IO’s role in policy diffusion via learning where they act as a homogeneous policy diffusion agent via policy framing and marketing (Brooks, 2005), knowledge generation and model standardization (Walt et al., 2004) and top-down advocacy (Ogden et al., 2003; Stone, 2004) in pursuit of a predetermined policy agenda. The study findings also diverge from studies examining Bank’s role in CCT’s regional diffusion and later stage global diffusion, which describe it as a zealot and assertive promoter (Borges, 2018; Brooks, 2015; Clemens and Kremer, 2016; Hall, 2015; Simpson, 2017; Sugiyama, 2011), and from studies highlighting the role of instrument constituencies in CCT’s global diffusion (Béland et al., 2018). This article contends that the findings from two early-stage inter-regional diffusion cases are at odds with other studies examining Bank’s role in CCT’s diffusion because they focus on different time periods and geographic areas. When a Bank bureaucrat presented the CCT model to Turkish bureaucrats in February 2001, the Bank’s general counsel had just approved the Bank’s first-ever loan for the CCTs (Colombia). When a Bank consultant presented the CCT model in Indonesia in July 2005, many Bank bureaucrats knew of the model but did not know its particulars. These historical contexts are critical for resisting the temptation to generalize Bank bureaucrats’ role in CCT’s diffusion through a 15-year-long period – a temptation that may have shaped the generalizations put forward in some of the studies on this topic (Hall, 2015). Hence, this study underlines the importance of time and space in understanding diffusion mechanisms and IOs’ roles in these mechanisms.

Conclusion

Taken together, the findings provide new insights both about learning as a diffusion mechanism and about IOs’ roles in early-stage policy diffusion. Although the World Bank employed different tools in the cases studied (financial services in Turkey and technical assistance in Indonesia), learning was the primary diffusion mechanism in both cases. In both cases, Bank bureaucrats influenced policy processes by joining national bureaucrats in their ‘collective puzzlement’ around important policy problems – as partners, rather than pressuring external actors. Finally, in both cases, trusting relationships between national bureaucrats and Bank bureaucrats played a critical role.

This article agrees with Sharman (2008) in making a case against assumptions of uniformity across time and space when studying diffusion mechanisms and IOs’ roles in them. Neither a policy nor an IO remains static throughout a global diffusion process. Hence, our understanding of IO’s roles in global diffusion processes would benefit from incorporating this constant change into our analytic framework. More broadly, the article parallels Béland and Orenstein’s (2013) insights on IOs as navigators of a route between complex and shifting ideas and interests, for whom internal policy consensus is an exception. Also critical in this respect is a recognition of IOs as non-monolithic actors in the global diffusion of policies. In the cases examined, bureaucrats in the same IO push in different directions regarding a policy’s desirability and promotion, and individual Bank bureaucrats not in senior positions influence CCT’s early-stage policy diffusion. These findings highlight the importance of incorporating individual bureaucrats’ roles into policy diffusion studies to better approximate a complex reality tainted with serendipity.

Finally, this study confirms the need for more studies about the micro-processes of policy diffusion. Studies that investigate both positive and negative cases of diffusion using comparative process tracing could be a fruitful next step in this regard. As expressed by Dobbin et al. (2007), ‘the more evidence we compile that narrows down the possible explanations of the diffusion of particular policies to certain countries in specific time periods, the closer we will be to understanding which mechanisms are at work, when, and where’ (p. 464).

Supplemental Material

sj-docx-1-gsp-10.1177_14680181211012975 – Supplemental material for Trusting relationships, learning bureaucrats: International organizations and early-stage policy diffusion

Supplemental material, sj-docx-1-gsp-10.1177_14680181211012975 for Trusting relationships, learning bureaucrats: International organizations and early-stage policy diffusion by Ozsel Beleli in Global Social Policy