Agents, coercive learning, and social protection policy diffusion in Africa

generation, advocacy,


Summary Summary
The adoption by numerous African governments of a similar set of social protection policies and programmes is often analysed as a paradigmatic case of policy diffusion or policy transfer. A particular social protection discourse has been actively and successfully promoted since the late 1990s by many international development agencies -bilateral and multilateral donors, United Nations organisations, international financial institutions, and international NGOs.
This paper contributes to this literature. We first introduce the concept of 'coercive learning' to focus on a specific set of mechanisms that have been deployed by international development agencies in support of the diffusion of social protection policies throughout Africa. This concept draws on two of the four commonly agreed transmission mechanisms: learning and coercion. We argue that learning is not a neutral process but is political and shaped by ideology. This is especially significant in the context of learning that is explicitly linked to policy advocacy, where hegemonic ideas are promoted through soft forms of coercion such as donor resource flows.
'Coercive learning' therefore describes the propagation of certain ideas through mechanisms such as: selecting the conceptual frameworks that underpin national social protection policies (ranging from narrowly neoliberal targeted to more progressive and inclusive approaches); selective investment in evidence-building (notably empirical impact evaluations of social protection projects) with the intention of influencing policy adoption or policy reform; research into financing options that aims to challenge perceptions by African policymakers that social protection is unaffordable and that no fiscal space exists in low-income countries; and finally, delivering professional training or capacity building inputs to government officials that reflect the ideas and ideological positions of the agencies that design and deliver this training.
Our second contribution to this literature is to draw a distinction between agencies and agents. We note, based on a pragmatist theoretical framework, that individuals (agents) are the car-riers of ideas that are promoted and financed by institutions (agencies) because only human beings can ultimately make knowledge effective. Since agents of international development agencies (whether policy advisors or consultants) bring not only their technical expertise but also their own positionality and inherent biases when interacting with African governments, the diffusion of social policy is thus not a predictable and linear process by which the intentions of agencies are simply and mechanistically carried out. Instead, policy diffusion is always also marked by personal relationships and the specific experiences and perspectives of all the agents involved.
To illuminate the centrality of agents, we identify three individuals who played major roles in the introduction and institutionalisation of social cash transfers in Zambia, and examine how they exercised policy influence through their personal connections, both to the international community and to domestic politicians and civil servants.
Finally, this paper also makes a contribution to this literature at the methodological level. Given the central role of individual agents in the social protection policy process in African countries, we argue for agents to adopt a critical self-reflexive lens, partly to deepen our understanding of the contingent nature of the policy diffusion process, and partly for agents to acknowledge that inserting themselves into policy processes is in itself an expression of power that derives not from domestic authority or legitimacy but from transnational sources of epistemic authority, underpinned by the soft power of international development finance. To illustrate how such a self-reflexive reflection might be undertaken, the authors revisit and interrogate our own roles as agents engaged, albeit in a more limited way, with the social protection discourse and practice in Zambia. [2]

IntroductIon IntroductIon
The spread of social protection policies across Africa can be considered as a paradigmatic case of policy diffusion. In the late 1990s, a group of influential transnational actors 1 (TAs) began advocating for a particular social protection instrument, social cash transfers (SCTs), as "a new instrument for combating poverty" in Africa (Leisering, 2019). Two indicators capture social protection's rapid rise up the social policy agenda in the new millennium. Firstly, social protection appeared in none of the eight Millennium Development Goals in 2000, but in three of the 17 Sustainable Development Goals in 2015 (United Nations, 2015). Secondly, no African country had a National Social Protection Policy or Strategy in 2000, but 35 of 55 countries had published theirs by 2020 (Devereux, 2020). As we will see, these policy documents were strongly influenced by the involvement of TAs. In this paper, we map this diffusion process in sub-Saharan Africa, and analyse the case of Zambia to make three arguments.
First, based on our reading of diffusion theories and an analysis of social protection in Zambia, we suggest the term 'coercive learning' , to draw attention to the interwovenness of hegemonic ideas, the collective construction of social policy ideals, and concrete processes of cross-country and within-country learning. Constructivist scholars have long established that the power of international agencies in large part rests on their ability to define problems and delineate the realm of imaginable solutions (Barnett & Finnemore, 2004). Since the "socialisation of global politics" from the 1980s onwards (Deacon et al., 1997), learning about social protection programmes -their design, effectiveness, etc. -has been part and parcel of the diffusion process in Africa (Davis et al., 2016). However, while policy diffusion is often presented as a technical undertaking, based on accumulated learning about objectively proven best practices, it is inherently political (Gilardi & Wasserfallen, 2019). Learning is political because the way we 'know' the world has repercussions for how we act towards it, and this has material effects. The creation and propagation of a dominant social protection discourse, underpinned by, and enabling, resource flows and political pressures, can thus be considered a soft form of coercion. Secondly, while pointing to the coercive dimension of policy learning, we make an argument for taking the role of individual agents seriously: agents' agency matters. Especially when studying processes of 'knowing' , it is individual human beings who acquire, transmit, translate, and diffuse knowledge. Kuhlmann et al. (2020, p. 87), too, have argued that "money, people and procedures" are key carriers of diffusion, and that people in particular "play a crucial role for the diffusion of ideas, especially when it comes to the role of knowledge". We draw on practice theory to think this through further. In the theoretical part of this paper, we conceptualise the significance of individuals in terms of their structural embeddedness on the one hand, and creativity on the other. We argue that this view is well suited to account for the structures of the coercive learning context that individuals tap into and reproduce, as well as the possibility for change that lies in those individuals' agency. We demonstrate the interplay of structure and agency through zooming in on several individual diffusion agents in the Zambian context, including through two short cases studies of ourselves.
Our third contribution is thus a methodological one. Given that the lines between policy and academic learning are often impossible to draw and that we as global social policy scholars are in various ways ourselves involved in creating the social protection discourse, we cannot exclude ourselves from a study of the field and should be fully aware of our role as "makers and shapers" (Cornwall & Gaventa, 2001) of policies in foreign countries. While a means of practicing heightened self-awareness, the self-reflexive lens can add to our understanding of the ways in which diffusion is contingent rather than a neutral process. By extending the study of individual agents to ourselves, we are able to shed more light on how agents inevitably reproduce the structural context in which they operate but also how their agency creates room for change. The power structures we tap into, our positionalities and relationships within the field, and not least our own normative motivation make the diffusion of policy knowledge a highly contingent process. Adopting a self-reflexive lens can shed light not only on these influences on our own role in the transnational diffusion process but can be a way of "reflecting about an entire community of which we are a part" (Berten & Wolkenhauer, 2021, p. 5).
In the following, we begin with a theoretical argument for conceiving of diffusion as coercive learning, while attributing central importance to individual human agents (section 2). We then concentrate first on the coercive learning that has taken place in social protection policy diffusion in Africa, break this down into four processes, and argue that agents have mattered (section 3). In the subsequent section, we focus on three individuals who have occupied different positionalities in the social protection diffusion process in Zambia, spanning transnational and national realms. Building on that, we refine our insights about agents through two short pieces of self-reflection, by drawing on our own experiences with coercive policy learning in Zambia (section 4). Finally, section 5 summarises our main contributions and draws out lessons for future diffusion research. Policy diffusion involves the movement of policy ideas, models, and instruments from one context to another (Dobbin et al., 2007;Obinger et al., 2013;Kuhlmann et al., 2020). 2 The concept is commonly agreed to comprise four transmission mechanisms: competition (states vying for global advantages keep track of other countries' policies); emulation (states follow international standards); learning (states adapt their policies based on ideas and experiences from elsewhere); and coercion (states are pressurised into adopting policies they did not freely choose). The latter two especially are of interest here and merit a few remarks. First, coercion can be seen to include "softer" variants, including policy conditionalities, changes in incentives, or the imposition of hegemonic ideas (Dobbin et al., 2007). Hegemonic ideas are not clearly distinguishable from emulation, which Dobbin et al. (ibid.,p. 452) include in "constructivism" as the creation of overriding discourses by epistemic communities, to which states then conform (see also Gilardi & Wasserfallen, 2019). This, in turn, exhibits some proximity to learning, as learning is not a neutral process, but is shaped by one's ideological lens at a given point in time. For instance, the social protection interventions that were "merchandised" in Africa by external agents have been criticised for having been based on a narrow, neoliberal understanding of social policy (Adesina, 2020). The spread of social policies, moreover, needs to be contextualised in what could be considered a coercive situation writ large: the push by TAs for poverty reduction interventions that followed a period of a seemingly contradictory push (also heavily externally induced) for state withdrawal and a shrinking of the public sector. The impact evaluations that constituted a major part of the diffusion process were meant to establish the cost-effectiveness of specific programme designs, trying to reconcile an unintentional process of increasing resemblance of policies across countries with a focus on structures, while the latter is said to involve concerted strategies for transferring policy knowledge and agency (Obinger et al., 2013, p. 113). The former is closely associated with International Relations and the latter with Public Policy literature; while tending to be based on quantitative versus qualitative analyses, respectively. Ultimately, the lines are not clear-cut. By choosing the term "diffusion" as more commonly used in the literature of relevance to our cases, we include also the more conscious and agency-related dynamics sometimes associated with policy transfer.

agentS and
[4] the need to save money with the need for poverty reduction. The resulting targeting of transfers to the poorest groups has since been criticised for falling short of inclusive social policy ideals (Mkandawire, 2005). The promotion of specific social policy ideas can thus be seen as infringing on governments' sovereignty and upholding their dependence on changing trends in international development aid. Overall, we thus argue that learning and coercion are intricately interwoven in the case of social protection policy diffusion in Africa.
The discursive power that TAs possess, due to their role in knowledge creation (Barnett & Finnemore, 2004), comes into being through being enacted by individuals. Some authors have therefore suggested focusing on practices rather than on discourses as such. Practices are defined as "socially meaningful patterns of action which […] simultaneously embody, act out, and possibly reify background knowledge and discourse in and on the material world" (Adler & Pouliot, 2011, p. 6). Bourdieu's notion of habitus captures that individuals internalise and act on the meanings acquired from the professional communities they represent, and that individuals thereby connect structure and agency (Bourdieu, 1977). In other words, discursive structures only come into effect through being enacted by individual human beings, as only they are "provided with corporeality, reflexivity, and the aptitude for abduction" (Franke & Roos, 2010, p. 1069. "Abduction" describes the process by which the world becomes known in an iterative process that reconciles past knowledge with new impressions. This means that agency always has a reproductive component whereby existing structures are re-enacted, as well as a component of change, as new situations require creative solutions for adapting those structures. To elaborate how this works, the individual must be prised open, which can be done by drawing on George Herbert Mead's distinction between an individual's "I" and "me". A basic assumption in Mead's pragmatist theory is that even individual consciousness is socially constituted as it arises from perceiving ourselves through the eyes of others. From young age, a child becomes conscious of her or his body and its ac-tions through the ways in which others respond to it. Awareness of the self is derived from the social surrounding as the child takes over the responses of others and develops an attitude towards her-or himself (Mead, 1962(Mead, [1934). This is what Mead calls the "me", the internalised view of society on the self, while "I" is the pre-social component of the self that carries out actions in an initially unconscious way and is the place of creativity and spontaneity (Franke & Roos, 2010, p. 1069. Rather than rationally determining all actions from the outset, the "me" retrospectively makes sense of these actions (ibid.). Especially in new situations, when routines no longer work or do not yet exist and individuals have to use their creative power of abduction, room opens up for developing new rules for action by applying, adapting, and refining existing knowledge.
Following on this conceptualisation, we argue, therefore, that precisely because TAs generate their power through creating knowledge and, thereby, possibilities for action, one needs to shift the focus to the individual agent in the diffusion process, as the place where previous knowledge (structures) and its creative adaptation (agency) meet. This in no way means that material structures, resources, and power play less of a role. But the only causal effects that they exert is through individual human beings, some of whom will obviously be more influential than others, depending on their structurally defined position.
Adopting a focus on agents moreover underlines that policy diffusion is not realised in a straightforward and linear way but is always mediated by the agency and reflexivity of individuals. Even though highlighting the coercive nature of social protection policy diffusion through hegemonic ideas and an overall context in which TAs exert power in various ways, all individuals involved in the process possess agency. This holds true for those acting within TAs as well as those from domestic contexts. We thereby address also the criticism that diffusion theory often seems to suggest that receiving countries (usually in the Global South) are passive applicants of policy models invented elsewhere (usually in the Global North). Edwards (2020) points out that while our current international institutions, the global economy and what might be called world culture all originate from racialised colonial relations, they are also products of interactive and relational dynamics with actors in the Global South equally impacting on them (what she refers to as "subaltern agency"). While there is certainly a hierarchy that comes from the fact that rich countries often possess better-resourced research infrastructures and long-developed connections between research and policy (Dobbin et al., 2007), agents at the national and local levels are required to make diffusion effective. The more one zooms in on individual stories of diffusion, the less convincing it seems to maintain these dichotomous categories, as agents connect and often cut across those interwoven spheres.

the tranSnatIonal hIStory of the tranSnatIonal hIStory of
coercIve SocIal polIcy learnIng coercIve SocIal polIcy learnIng

In afrIca In afrIca
In the following, we map out various processes through which coercive learning occurs within social protection in Africa, and explain how individuals have mattered.

Coercive learning in shaping national social protection policies
In 2010, only five countries in Africa had a National Social Protection Policy (NSPP) or Strategy (NSPS), but another 30 countries promulgated one between 2010 and 2019. During this decade, a group of TAs that was active in promoting the adoption and institutionalisation of social protection in Africa instigated and often commissioned the process of producing these documents, which were typically drafted by teams led by expatriate consultants and sometimes included national consultants. Government officials were consulted but not necessarily centrally engaged in this policy formulation process. In 2012, for example, the World Bank published a report titled 'Togo: Towards a National Social Protection Policy and Strategy' which "incorporates the Government's comments" (World Bank, 2012, p. i).
The 35 national policy documents display remarkable similarities in terms of their conceptual and programming content. Conceptually, they embody ideas about the appropriate framing of social policy that were produced by TAs. Most NSPPs follow UNICEF's preferred 'life-course' approach, the World Bank's 'social risk management' framework, the ILO's 'social protection floor' , or 'transformative social protection' , from the UK's Institute of Development Studies (IDS). Programmatically, most NSPPs are dominated by social cash transfers for targeted population groups identified as poor and/or vulnerable.
The explanation for this convergence is that the ideas informing these documents derived not from domestic policy priorities or participatory consultation processes, but from an influential cohort of international agencies that shared a common agenda in terms of promoting the uptake of social protection in African countries, even if they disagreed on some details.
Rather than learning from and emulating neighbouring countries or asking their own citizens about their preferences prior to choosing a particular policy direction, African governments learned from agents of organisations that imported specific ideas and approaches from abroad. On some NSPPs, the logos of TAs that were involved in the process of developing the policy are prominently displayed on the cover, alongside the national coat of arms.

Coercive learning through selective investment in evidence-building
As noted above, once the TAs had identified targeted SCTs as their preferred social protection instrument in Africa, they advocated for the adoption of SCTs through tactics of 'coercive learning' and 'policy merchandising' (Adesina, 2020). One tactic was to design and finance small-scale pilot projects, which TAs subjected to rigorous impact evaluations to generate evidence that was intended to persuade governments of the effectiveness of these programmes. Although TAs provided the initial financing, their objective was to hand over responsibility for running, scaling up and financ- ing these programmes to national governments.
Sometimes different design modalities were tested (e.g. cash or food, or alternative targeting mechanisms), but the main purpose was policy advocacy for cash transfers. Summarising the evidence base as of 2011, the UK Department for International Development concluded that cash transfers "have proven potential to contribute directly or indirectly to a wider range of development outcomes" (DFID, 2011, p. i). Two initiatives illustrate the ways that TAs generated and used evidence explicitly for policy advocacy. The first is the Regional Hunger and Vulnerability Programme (RHVP), co-funded by DFID and the Australian Agency for International Development (AusAID), which commissioned studies of 15 social protection programmes in six southern African countries under its Regional Evidence-Building Agenda (Ellis et al., 2009). The purpose was to disseminate good practice and influence social protection policy adoption across the region. RHVP's theory of change was encapsulated as: 'Evidence-building + Capacity-building = Positive policy change'.
The second is the Transfer Project, a joint initiative of two United Nations agencies, the Food and Agriculture Organisation (FAO) and UNICEF, which commissioned impact evaluations of cash transfer programmes in eight African countries. As with RHVP, the intention was partly to contribute to building the evidence base, but also to feed into 'evidence-based policy' choices by the governments of these countries. In the Foreword to an edited book titled 'From Evidence to Action' , the Director-General of FAO and the Executive Director of UNICEF state: "These pages also document the ways in which the Transfer Project has influenced the policy debate in each of the eight countries … This innovative approach transcended impact evaluation and influenced wider social protection policies in each country" (Davis et al., 2016, p. vi).
Against this heavy investment in evaluating the impacts of often tiny cash transfer pilot projects, TAs invested very little into evaluating government-run programmes. This selectivity is one tactic of coercive learning. By drawing attention to projects they supported -through large monitoring and evaluation (M&E) budgets that generated rigorous evaluation reports, policy briefs, national workshops and media attention -TAs ensured that cash transfer projects became synonymous with new and innovative thinking and practice on social protection in African countries. At the same time, less favoured government programmes were neglected and marginalised. The fact that donor-initiated pilot projects are more rigorously evaluated than government-run programmes is, in itself, confirmation of our assertion that evidence-building is both selective and coercive.
It should be noted that the techniques applied by TAs to promote adoption of social protection in Africa are not new but can be traced back to colonial times. Yet the extent to which "evidence" drives public policymaking has significantly grown over the past two decades, and has been attributed both to the end of ideological battles post-Cold War and to New Public Management ideals of efficiency and "value for money" (Eyben, 2013).

Coercive learning through interventions in affordability and financing debates
The initiation and expansion of cash transfer programmes in Africa was expedited by financial support provided by TAs, especially in countries where governments were reluctant to commit public resources of their own, either because of budget constraints, or because they were not yet convinced by the evidence, or because social protection was not a national priority at that time. TAs assumed, and tried to get governments to agree, that programmes initially funded by external actors would eventually be funded from domestic resources. However, in contexts of widespread poverty and limited fiscal space, governments were understandably hesitant about establishing permanent large-scale welfare programmes, and there remains a large "'affordability gap' between what is advocated for African countries and what these countries' governments are willing to spend" (Seekings, 2017, p. ii to subsidies, so they tied their official development assistance (ODA) to cash transfer projects instead. This financial leverage gave TAs substantial power to shape social policies throughout Africa, especially in heavily aid-dependent countries.
One coercive learning tactic that TAs deployed was to try to refute perceptions by governments that social protection at scale is unaffordable in low-income countries. Several TAs -DFID, FAO, ILO, World Bank, and others -commissioned research and microsimulations to demonstrate that cash transfers should in effect be classified as public economic investments rather than social welfare spending, since they can generate income multipliers and contribute to economic growth (Alderman & Yemtsov, 2012;Barrientos, 2010;Taylor, 2012).
The ILO's role in this area has been especially forceful. In the early 2000s, ILO produced a series of publications that addressed the rhetorical question: 'Can low-income countries afford basic social protection?' (Pal et al., 2005;Behrendt & Hagemejer, 2009), sometimes with the added advocacy question in the sub-title: 'Can they afford not to have it?' (ILO, 2008). ILO's advocacy strategy included modelling to show that a package of 'basic benefits' would cost countries only a few percentage points of GDP and could be paid for out of a combination of reallocation of government spending plus transitional financing from donors (Pal et al., 2005). More recently, ILO's advocacy for expanded social protection shifted to advising governments on how they can create more fiscal space to 'expand social investments' (Ortiz et al., 2015).

Coercive learning through professional trainings
The most direct pathway for coercive learning is by building the technical and administrative capacity of government officials to design and deliver social protection programmes, and to institutionalise national social protection systems.  Table 2 sets out a "Tentative sequence of activities for implementing an SPF approach at country level" and section 3.4 focuses on "Integration of the SPF into national, regional and global planning processes" (ibid., pp. 11-12).
Some of these agencies translated their manuals into training workshops that they delivered not only to their own staff but to thousands of government and agency officials. The World Bank Institute has offered an annual 2-week training course in Washington DC since the early 2000s, initially called 'The Design and Implementation of Effective Safety Nets' and now called the 'Social Safety Nets and Delivery Core Course' , with distance learning options in English, French, Russian and Spanish. The ILO has an International Training Centre in Turin, where it offers a 2-week training course each year on social protection systems. ILO also offers online training on 'Building and Managing Social Protection Floors in Africa' , hosted by the 'Virtual Campus' of socialprotection.org. [8] Although these documents and activities might be interpreted as contributing to a neutral process of discursive ideation, we suggest viewing them rather as instruments in the TA toolkit for "policy merchandising" (Adesina, 2020). The coercive learning objective of the ILO's online course is explicit in the explanation of "What you will learn": "By the end of this course, you should be able to understand why and how a Social Protection Floor is beneficial to your specific country context and how it can assist social and economic development" (TRANSFORM, 2020). Useful tools like this can become coercive instruments if they limit choice, and if they exert "hidden and invisible power to determine what knowledge counts" (Eyben, 2013, p. 3).

ZoomIng In on IndIvIdualS In ZoomIng In on IndIvIdualS In
SocIal protectIon dIffuSIon In SocIal protectIon dIffuSIon In

ZambIa ZambIa
In this section we discuss the social protection policy process in Zambia through a study of three individual agents, selected based on variance in their positionalities. This is followed by reflections by each author on our own roles as 'agents' in the ongoing social protection policy diffusion process in Zambia. In combination, both sub-sections allow us to draw conclusions about how structure and agency interact in diffusion practice. But first, we briefly recapitulate our argument about why individual agents matter in policy diffusion processes.

Why agents matter
It is commonly argued that the specific form of social protection that has permeated Africa in the last 20 years has been the product of TAs with particular ideological and ideational perspectives, which they 'export' in the form of policy ideas that they support -and by corollary, other policy ideas that they reject and do not support -in other words, that the idea of social protection is developed and propagated by 'agencies'. We take a slightly different view.
We argue that ideas of social protection were exported to Africa both by institutions (agencies) and by individuals (agents), and that agents and agencies are not indistinguishable, even if individual agents are broadly ideologically aligned with the institutions that employ or contract them. Although agency staff, such as social policy advisors and country desk officers, have their own ideas and preferences about the policies they promote, they often recruit intellectual and technical expertise from outside the agency, as short-term consultants. These agents fill the gap between broad policy orientation and specific policy advice or project design with their own creative ideas and inherent biases, to ensure that they become effective.
This distinction between agencies and their agents matters because it nuances the popular view of donors as monolithic, with predefined ideas and rigid preferences that they roll out in a uniform way, across all countries in which they operate and exert policy influence.

Three agents at the intersection between the transnational and domestic
In the social protection community, the Zambian SCT is well-known. While the case lends itself to showing where individual agents have been influential in the ascent of the programme and the country's wider National Social Protection Policy, it likewise serves to illustrate the variety of roles that agents can play in diffusion. Three agents, who embody in different ways the transnational character of the policymaking process, are singled out in this section -not because they were necessarily the most important, but because they stand for different positionalities on the transnational-domestic spectrum and because they illuminate different ways in which individuals matter for diffusion. The focus is not on domestic policymakers or bureaucrats, who of course play the key roles in accounting for Zambia's social protection policies and their implementation (Siachiwena, 2016;Kapingidza, 2019;Wolkenhauer, 2020 For decades, access to social protection in Zambia was restricted to a small minority of public and private sector workers, who benefited from employment-linked social insurance schemes, especially old age pensions and insurance against work injuries and invalidities. This limited coverage was a legacy of Zambia's colonial history, its copper dependence and its undiversified economy, that generated relatively few formal employment opportunities with associated benefits (Wolkenhauer, 2021). However, in the early 2000s, in line with the global turn towards social protection, several non-contributory social assistance schemes were initiated, most visibly the SCT. The idea of the SCT entered Zambia in 2003 when the German development agency GTZ (now GIZ) hired Bernd Schubert as a consultant, to undertake a poverty profile of Kalomo district to inform GTZ's social sector work.
Schubert, a German national, is widely acknowledged as one of the original instigators of SCTs in Africa. He had worked with GTZ on the delivery of cash transfers in Mozambique before coming to Kalomo. 3 He identified a category of so-called "incapacitated households" where -due mainly to HIV and AIDS -there was nobody of the working age generation, leaving the household unable to meet its basic needs (ibid.). A meeting was subsequently conducted in the Ministry of Community Development where Schubert presented his results and introduced the idea of supporting these "ultra-poor" households -which he defined as households that cannot meet even 80% of their consumption needs despite spending 80% or more of their income on food -with regular cash transfers. This proposal was met with some scepticism, so it was agreed to undertake a trial of such an intervention, funded by GTZ, to test its effectiveness (Schubert, 2005;Kabandula & Seekings, 2016). The Pilot SCT Scheme carried out in Kalomo District almost immediately showed positive results in terms of poverty reduction and food security (Schubert, 2005).
Although it would take several more years, more rigorous impact evaluations, and a new government to significantly expand the programme, the findings from Kalomo became an important political tool for promoting social protection in other countries in the region. In 2006, HelpAge International (an international NGO) and the African Union organised a conference in Livingstone, Zambia, funded by DFID, to promote the model. It was attended by social ministries from 13 African countries, as well as representatives from Brazil, various UN agencies and NGOs (Hagen, 2009). The conference included a tour to Kalomo and resulted in the African Union's 'Livingstone Call for Action' , which urged governments to adopt programmes similar to the Zambian SCT. Schubert subsequently went on to promote the "Kalomo model" in other countries, including Malawi, Sierra Leone, Liberia, Zimbabwe, and South Sudan (Schubert, 2020). When interviewing him more than ten years later, he was about to set off to Eswatini for another such assignment.
An intriguing fact is that the "ultra-poor" approach to targeting social assistance is found only in southern African countries where SCT projects were instigated and designed with inputs from Schubert. In other words, this idea came from an individual agent, it was not devised by the sponsoring agency, nor was it chosen by the government of each country. Even more pertinent, the agency that employed Schubert to initiate the SCT pilot project in the Kalomo District of Zambia was not in favour of social cash transfers. Rather than a Northern TA, in this case GTZ, "merchandising" the idea of SCTs to Zambia, the consultant hired by GTZ had to first convince the agency of his idea, then set up a project to try to convince the government of Zambia. In the early 2000s, GTZ was well-known for promoting community-based health insurance schemes as their preferred anti-poverty instrument. Thanks to Schubert, GTZ became an early funder and one of the first promoters of cash transfers in Africa.
Schubert can clearly be identified as a 'policy pollinator' (Devereux, 2020) -an agent who flies from one country to the next with a standard policy prescription in his briefcase. By himself though, he did not diffuse the SCT into national [10] policy; that would take several more years and other actors. In the mid-2000s, Zambia's Ministry of Community Development began to draft a National Social Protection Strategy (Quarles van Ufford et al., 2016), which would culminate in the adoption of the National Social Protection Policy (NSPP) in 2014, and in which the SCT eventually became one of the flagship programmes. Two transnational agents who played key roles during that period are Charlotte Harland Scott, a development professional from the UK, later active in the Zambian Ministry of Community Development and UNICEF, and Denis Wood, a Zambian consultant and a key figure in donor-funded research and advocacy.
Harland Scott came to Zambia in 1989 after studying development studies in the UK, where she was born, to work for an international humanitarian organisation in Mpika in Northern Zambia (Zimba, 2014). Here she met and married (in 1994) Guy Scott, whose parents had moved to Zambia during the colonial era, and who at the time was campaigning for a parliamentary seat in Mpika together with later-president Michael Sata. Harland Scott knew Sata because she stayed in his father's home village. In 1995 she was hired by the Ministry of Community Development as a consultant assigned with the task of redesigning the government's Public Welfare Assistance Scheme (PWAS) (Siachiwena, 2016). In this role, she tried unsuccessfully to lobby different donors for increased funding to PWAS (Harland, 2011).
Several years later, in 2003, Zambia's Poverty Reduction Strategy Paper resulted in the establishment of a sectoral advisory group (SAG) on social protection, to bring together different government departments, donors, and civil society (Kabandula & Seekings, 2016). The SAG's main function was to draft a strategy on social protection, which became part of the fifth National Development Plan. The advisory group selected Harland Scott as a consultant, paid by DFID, to draft the Ministry of Community Development's strategy (ibid.). In 2007, she became the Chief of Social Policy for UNICEF Zambia, while continuing to work closely with the Movement for Multiparty Democracy government of the time (Siachiwena, 2016). With UNICEF being one of the main international agencies promoting social protection in Zambia, this post has been a key point of influence throughout. Harland Scott remained in this position until 2011, when Guy Scott became vice-president under Michael Sata and her UN position was incompatible with being married to a politician (Zimba, 2014). When Guy Scott served as interim president from 2014 until 2015 upon Sata's death, she served as the first lady of Zambia, and in 2016 ran as the MP for the Lusaka Central seat in the national assembly on a United Party for National Development (UPND) ticket (Rainbow News Zambia, 2020).
Harland Scott has been an influential agent in the SCT's ascent and social protection more widely. Her biography and posts blur the lines between international and national, having had both an influential position in one of the main international agencies as well as direct access and political affiliation to the Sata government, which was responsible for the decisive expansion of the SCT budget in 2013 (Pruce & Hickey, 2017). While being able to tap into powerful structural positions offered by her professional background and the established power of UNICEF, Harland Scott's agency is one crucial component of the "success story" of the SCT.
The third agent that we have singled out to exemplify the diffusion process of social protection in Zambia is Denis Wood, who has facilitated the communication between the transnational and national spheres. Wood, originally an agricultural specialist, was involved in a key piece of research early in the history of the SCT, when DFID funded an enquiry into the political feasibility of social protection in Zambia (Barrientos et al., 2005). Besides being involved with the donor research effort, Wood was also well connected to domestic decision makers. He explained that those who want to influence policymaking, including donors and himself, need to target the ministerial technocrats as well as the inner circle of the president and his loyalists. 4 As he put it: "then, we've got the opportunists, these are the likes of you [Wolkenhauer as interviewer] and me, the World Bank, the CPs, the cooperating partners and so on, who [11] SOCIUM • SFB 1342 WorkingPapers No. 22 would like to influence policy. Now, normally the big problem with regard to state involvement, state engagement, is, how do we influence the centre point here, and get the decision to be made at this particular level" (ibid.). One of the channels for influencing the inner circle was a rather informal one. Wood and Michael Sata went to the same Catholic church, and in the church yard he talked to Sata about the SCT, who later decided to raise support for it in the Ministries of Community Development, Finance and Labour (ibid.).
Wood subsequently kept his role as a facilitator of the ongoing diffusion process by which the NSPP was further refined and expanded. He has been involved in composing reports of the yearly reviews of the Zambian social protection system, that consist of several trips by Ministry officers and TAs through the country, followed by intensive workshops in Lusaka. These tours and workshops form an important part of the collective learning process between government departments and the various TAs. Finally, Wood is also a generous interview partner for academics trying to reconstruct the history of the SCT and thereby diffuses his knowledge back into the transnational academic discourse.
In sum, all three agents exemplify how coercive learning structures need to be "enacted" (policy merchandising through TAs), that this is enabled by existing power structures (linked to resources and legitimacy of involved TAs), and that these become influential through an individual's agency that connects the international and national levels.
Having demonstrated that policy learning becomes politically meaningful by being embedded in power relations (hence our term "coercive learning"), it would be helpful to shed further light on these research activities in particular: how much agency do they leave to agents, how much are they shaped by existing structures? For a tentative answer to these questions, we draw on our own experience as policy researchers in the Zambian social protection space.

Two experiences from a selfreflexive perspective
Both authors of this paper have been involved in social protection policy diffusion in Zambia at different stages, in 2008-09 (Devereux) and in 2017-18 (Wolkenhauer). While we do not consider ourselves particularly significant for the trajectory of social policy in Zambia, our experiences are illustrative of how agencies and their agents interact to co-construct social policy in the Global South by way of coercive learning. This self-reflexivity can therefore further substantiate the arguments made in this paper.
In 2008-09 Devereux was contracted by DFID Zambia, representing the 'Government of the Republic of Zambia and Cooperating Partners in Social Protection' , to provide support to the Technical Working Group (TWG) on 'Social Assistance for Incapacitated Households' (Schubert's term for households with low productive capacity, his preferred target group for SCTs). Background documents provided by DFID included "a draft 'Options paper' prepared by in-country active donors in social protection (DFID, UNICEF, Irish Aid primarily)" (SP-CP Technical Group 2008: 2). ILO endorsed this paper, adding in the margin: "We also strongly agree that the priority is to develop a National Social Protection Programme for Zambia" -though whose priority this was is not clear.
The first written output of this consultancy was a review of Zambia's five SCT pilot schemes, and the final written output was a 'Proposal for the Scaling Up of the Social Cash Transfers in Zambia'. Both documents were co-authored by Devereux and Denis Wood, who was recruited and paid by DFID as the local consultant. The review was prepared for the TWG on Social Assistance, with both authors named on the cover (Devereux & Wood, 2008). However, the scale-up proposal was officially authored by the Ministry of Community Development and Social Services (MCDSS 2009) for submission to the Cooperating Partners for their financial and technical assistance. This proposal from a Zambian government ministry went through at least six drafts, with detailed comments provided on each by the Social Protection [12] Cooperating Partners, and occasional comments provided by MCDSS.
Zambia in the early 2000s was a laboratory for testing alternative design modalities for SCTs, with funding and technical support provided by different TAs with their own mandates and ideologies. For example, conditionalities were applied in one district but not in others, higher benefits were paid in one district to test whether this produced bigger impacts, and the 10% most incapacitated households were targeted in four districts but all persons over 60 (i.e. a universal social pension) were targeted in the fifth. Cooperating Partners objected (in written feedback to the authors) to the draft MCDSS proposal describing these as two distinct programmes: "ONE scheme with TWO targeting mechanisms -please emphasise the cohesion and rationale, rather than repeatedly suggesting that there is some huge difference". By 2008 the Cooperating Partners were campaigning for their pilot schemes to scale up and merge into a national social protection programme -which is why they hired consultants to write this proposal for MCDSS -so the proposal was required to present the five pilots as coherent building blocks towards this vision, rather than as fragmented and fundamentally incompatible projects.
Ten years later, the SCT had become institutionalised and widely accepted but the internationalised learning process continued. Though for the most part government-driven and -funded, the government of Zambia continued to receive technical and financial assistance for the SCT through the UN Joint Programme on Social Protection (UNJP-SP), at the time consisting of UNICEF, ILO, FAO, IOM and WFP (UNICEF, 2019). While in its early years, research revolved around impact evaluations of the SCT, the focus later moved to fine-tuning of the targeting and implementation of the transfer, as well as improving the alignment between different programmes within the NSPP. This was one rationale behind FAO's contracting of Wolkenhauer in 2017 to support the Ministry of Community Development's assessment of the Food Security Pack (FSP).
The FSP is meant to enable participating households to be self-sustaining through improved productivity and food security for which it supplies them with farming inputs (fertiliser, seed, and sometimes livestock). For four years, funded by the Norwegian aid agency, the FSP had been complemented with an electronic voucher and a cash supplement (Hichaambwa et al., 2014), so there was an interest to learn how both versions worked. The assessment was moreover supposed to shed light on possible linkages between the FSP, the SCT and the much larger farming input support programme (FISP). In this case, the main diffusion happened upwards. Practices were assessed at the local level to inform policy reform at the national level, and to feed into FAO's transnational stock of expertise.
For this assessment, Wolkenhauer conducted interviews and focus group discussions with state officers, civil society, and recipients, to learn about their roles, experiences, and daily challenges. She wrote a report, whose main findings were presented to leading bureaucrats in the Ministry of Community Development in early 2018, via skype, as Wolkenhauer had travelled back home by then. Based on her assessment, another consultant, an experienced Zambian academic, was then tasked with the organisation of a new pilot to put the derived best practices to test.
One striking observation is that Wolkenhauer, whose positionality as an outsider should have counted against her, was able to tap into existing structures of epistemic authority, derived from her affiliation with the FAO as an established agency, all the while reproducing (post-colonial) legacies of whose knowledge is considered credible. Provided with sufficient resources, she was able to bridge distances between the capital and remote areas and speak to lower-level bureaucrats who held the relevant knowledge of the programmes. In these ways, existing structures of power and resources were enabling factors.
This policy learning exercise consisted of the collection of existing knowledge from a variety of actors and places, which had not become powerful by itself but needed to be diffused to the policymaking level through condensation and formalisation. Needless to say, all that was supposed to be "learnt" about the programmes was known to all those dispersed individuals involved in their daily implementation, most notably the district and sub-district officers. Besides their stored memories, day-to-day reports of programmes existed in local offices but had remained underutilised. Simply by being given sufficient time, resources, and transport to visit many offices and spend several days talking to officers, the consultancy provided an opportunity for tapping into all those treasures of experience and knowledge. In the "field", the research task provided by the FAO and Ministry needed to be interpreted, such as by choosing who to speak with and what questions to ask. These choices were inevitably informed by the previous knowledge, positionalities, and unconscious instincts of the researcher, and ultimately impacted on the findings derived for the assessment. Far from being neutral, even if transparent and robust, the research outputs were contingent on these factors, too.
As another instance of mediating between different realms, this assessment involved a lot of translation between different knowledges. Apart from having to be detected and synthesised, the dispersed expertise needed to be brought into a format and framework that corresponded with conventions at the transnational and national levels. In this case, the FAO had clear ideas as to the structure of the final assessment report and it took several rounds of revisions for it to adhere to a format that made sense to professionals in Rome. One might say, this sort of bottom-up diffusion was not only about the substantive findings on social protection but also about facilitating the communication between different spheres that run according to different social logics -district offices in rural places and transnational knowledge repositories. Standardised procedures and formats, including for instance recognised methodologies or predefined sub-headings are needed for policy ideas to travel across geographical contexts. A degree of "creativity" is required in fitting what was learned in the field into such grids, while the necessary epistemic authority is provided by the researcher's structural position. Exclusion mechanisms meanwhile ensure a reproduction of knowledge hierarchies: contextual understanding alone does not provide access to policymaking. Post-colonial power relations thus correlate with a hierarchy of knowledge and are ultimately reproduced through these acts of agency.
Taken together, both self-reflexions show that individual agents like us are key in the diffusion of policies across localities. While we have considerable leverage over research procedures and findings, which in turn exert powerful performative effects, we operate within the larger structures of coercive learning. These include the power we are given through our positions in the larger postcolonial structure and our affiliations with professional fields and TAs' resource endowments. They also include having to adhere to predefined agendas as well as formalities and power dynamics that render some types of knowledge more intelligible than others. Hence these reflections have shown that any diffusion process necessarily rests on agents' agency and involves countless moments of latent transformation. But they have also demonstrated that the room for change is curtailed by the overall context of coercive learning.

concluSIon concluSIon
The adoption by many African governments of social protection policies, and specifically social cash transfers, as a coherent set of ideas developed by agencies located in the Global North, must be seen as one of the most remarkably successful cases of transnational policy diffusion in recent times. Between 2010 and 2016, for instance, the number of countries in sub-Saharan Africa with at least one social assistance programme doubled to 40 out of 48 countries, largely driven by "the efforts of transnational actors to promote particular forms of social assistance through a combination of ideational influence and financial leverage" (Hickey et al., 2020, p. 11).
The literature identifies four diffusion pathways, two of which -competition and emulation -are government-led while the other two -coercion and learning -speak directly to the activities of TAs. Our theoretical contribution to this literature is to suggest that learning in this context cannot be separated from coercion, as it entails the construction of policy possibilities and a dominant under- [14] standing of problems and solutions. Despite its objective connotation, 'evidence-based policymaking' is not apolitical. At the same time, discursive structures are not effective in and of themselves but always need to be interpreted and enacted by individuals. Whether policy messages are delivered as technical advice, training activities, or evidence generated from project evaluations, agencies and their agents exert influence over policy choices by the knowledge they select or omit, and how they present this knowledge to policy-makers and politicians. This process, which we term 'coercive learning' , calls for a closer look at human agents, particularly those who straddle and connect the transnational and domestic spheres.
At the empirical level, this paper explored the roles played by three influential individual agents who worked for transnational agencies to expedite the uptake of social protection in Zambia. These agents 'diffused' policy ideas in various ways and their individual biographies transcend national boundaries, affiliations and positionalities. While, initially at least, this was a story of policy diffusion failure, as the TAs experimented with different cash transfer design modalities in five districts, leaving the government confused about which version of social protection it should adopt and scale up, the social protection agenda ultimately took hold. This was mainly due to domestic political developments, specifically a change in government that increased domestic support and enabled effective policy coalitions.
Nonetheless, throughout the Zambian social protection policy diffusion story, the role of research and evidence has played a large part. We included a reflection on our own experiences with such learning and advisory activities to show that the knowledge that accrues and ultimately influences policymaking is far from neutral but always shaped by how we enact given structures as well as interpret them creatively. Individual agents such as consultants exert considerable influence through their coercive learning activities, but the type of knowledge that is understood by transnational agencies and their agents generates its own exclusion mechanisms.
In sum, we propose the adoption of a critical and self-reflexive perspective onto policy diffusion as coercive learning, and to factor in the individual human agents involved in the process -for they are the actors who transport and translate ideas back and forth between different institutional structures and social worlds. This perspective not only transcends common dichotomous conceptions of senders and receivers in diffusion processes but manages to factor in structure and agency, and to account for the reproduction of dominant ideas as well as the perpetual potential for change.
referenceS referenceS