Informing the Design of Livestock Insurance Subsidy Programmes in Ethiopia: A Mixed Method Approach Xandru Gorg Gigi Nazju Anglu Cassar ID: 23330705 Supervisors: Dr Tara Bedi - Trinity College Dublin, Dr Nathan Jensen - The University of Edinburgh August 2024 Trinity Collge Dublin, The University of Dublin Master of Science in Development Practice Wordcount: 10,935 2 3 Contents List of Figures ..................................................................................................................... 4 List of Tables ...................................................................................................................... 5 Declaration of Authorship ................................................................................................. 7 Abstract ............................................................................................................................. 8 Acknowledgements ........................................................................................................... 9 List of Acronyms .............................................................................................................. 11 1. Introduction ................................................................................................................. 12 1.1. Research Objectives ............................................................................................. 13 1.2. Outline .................................................................................................................. 14 2. Literature Review ........................................................................................................ 15 2.1. IBLI’s impacts ........................................................................................................ 15 2.2. Subsidies for index-based insurance .................................................................... 16 2.3. Synthesis and motivation ..................................................................................... 22 3. Methodology ............................................................................................................... 24 3.1. Methodological underpinnings ............................................................................ 24 3.2. Research design .................................................................................................... 24 3.3. Study sites ............................................................................................................. 26 3.4. Methods ................................................................................................................ 27 3.7. Ethics and data storage ........................................................................................ 34 4. Results ......................................................................................................................... 36 4.1. Qualitative results ................................................................................................. 36 4.2. Quantitative results .............................................................................................. 44 5. Discussion .................................................................................................................... 55 4 5.1. Intended outcomes and coordination .................................................................. 55 5.2. Subsidy rates ......................................................................................................... 56 5.3. Temporal distribution and adaptation of subsidies ............................................. 59 5.4. Targeting ............................................................................................................... 61 5.5. Insurance product design and bundling ............................................................... 63 5.6. Limitations and future work ................................................................................. 64 6. Conclusion ................................................................................................................... 65 References ....................................................................................................................... 67 Appendix 1 – Interview scheulde and discussion guide .................................................. 79 KII interview schedule.................................................................................................. 80 FGD discussion guide ................................................................................................... 81 Appendix 2 – Supplementary figures and tables ............................................................ 83 Appendix 3 – R packages used ........................................................................................ 93 Appendix 4 – Ethical Approval ........................................................................................ 94 List of Figures Figure 1: International organisations which have funded, implemented, and/or supported index-based livestock insurance subsidy programmes in Ethiopia.......................................... 13 Figure 2: Variation in willingness to pay for IBLI (WTP) by households’ livestock holdings. ... 21 Figure 3: Diagram summarising the mixed methods design adopted in this study. ............... 25 Figure 4: The timing of IBLI contracts sold in Borana. Figure taken from Ikegami and Sheahan (2018). ...................................................................................................................................... 31 Figure 5: The distribution of households’ herd sizes over the course of the panel, measured in TLU. ........................................................................................................................................... 45 5 Figure 6: The proportion of households purchasing insurance in each sales window and those with coverage at the end of each sales window. ..................................................................... 46 Figure 7: Variation in the marginal effect of receiving a discount coupon with the number of livestock owned . ...................................................................................................................... 49 Figure 8: Visual representation of the average marginal effect of receiving insurance discount coupons of different rates. ....................................................................................................... 52 Figure 9: Visual representation of the average marginal effect of receiving insurance discount coupons against the discount rate of coupon received in the previous sales window and that receiving two sales windows prior. .......................................................................................... 54 Figure 10: Plots showing the frequency at which households receiving a given subsidy level cited different reasons for not purchasing IBLI. ....................................................................... 57 List of Tables Table 1: The research questions addressed in this study. ....................................................... 14 Table 2: The three main benefits of IBLI identified by Jensen, Barrett and Mude (2017), Janzen and Carter (2019), Matsuda, Takahashi and Ikegami (2019), and Banerjee, Johnson and Mude (2022). ...................................................................................................................................... 16 Table 3: The intended outcomes of insurance subsidy programmes discussed by Hill et al. (2014), Hazell and Varangis (2020), Hazell, Jaeger and Hausberger (2021), and Kramer et al. (2022). ...................................................................................................................................... 17 Table 4: The codes and affiliations of KII participants disaggregated by stakeholder category, and the location where each interview was conducted. ......................................................... 27 Table 5: The location and composition of FGDs conducted, listed in the order they were conducted ................................................................................................................................. 29 Table 6: The timing of survey rounds and IBLI sales. Each survey round covers the previous two sales windows. ......................................................................................................................... 32 6 Table 7: The intended outcomes identified by KII participants, disaggregated by stakeholder classification and listed in order of frequency. ........................................................................ 36 Table 8: A characterisation of the different approaches to insurance subsidies conceived of by KII participants. ......................................................................................................................... 38 Table 9: Summary statistics of the data. .................................................................................. 44 Table 10: Balance tests of characteristics of purchasing and non-purchasing households when pooling data across sales windows. ......................................................................................... 47 Table 11: The frequency of primary reasons cited for not purchasing an IBLI policy across different rounds. ...................................................................................................................... 48 Table 12: The estimated coefficients and average marginal effects of Model 2, and Model 2 including an interaction term between the number of livestock owned and the receipt of a discount. ................................................................................................................................... 50 Table 13: Average marginal effects of discount coupons on IBLI uptake. ............................... 51 Table 14: Coefficient estimates and average marginal effects of present and previous period discounts. ................................................................................................................................. 53 Table 15: Coefficient estimates and average marginal effects of Model (2) including an interaction between the receipt of a discount and a variable capturing expected rangeland conditions, and an interaction with the gender of the household head. ................................ 61 7 Declaration of Authorship I, Xandru Gorg Gigi Nazju Anglu Cassar, hereby declare that: a) the dissertation has not been submitted as an exercise for a degree at this or any other University; b) the dissertation is entirely my own work; and, c) I agree that the Library may lend or copy the dissertation upon request. This permission covers only single copies made for study purposes, subject to normal conditions of acknowledgement. Signed: Date: 26/08/2024 8 Abstract Subsidies for index-based insurance are widely used across developing countries to protect farmers from increasingly frequent extreme weather events. This mixed-methods study on subsidies for Index-Based Livestock Insurance (IBLI) in Ethiopia evaluates three important yet unexplored aspects of such subsidies, namely (1) their aims as perceived by key stakeholders, (2) the implication of their design, and (3) exogenous factors which influence their effectiveness. Thematic coding was used to analyse primary qualitative data gathered via key informant interviews and focus group discussions conducted in Ethiopia, the results of which were combined with econometric analysis of secondary household survey data in deriving findings. Findings indicate that stakeholders broadly agree on the intended outcomes of subsidies, yet coordination failures were identified to be constraining IBLI subsidies’ effectiveness. Actual and perceived inaccuracies in the index underpinning the livestock insurance products also emerged as factors undermining subsidies’ impact. Furthermore, results show that subsidy rates below 30% are fairly ineffectual at heightening uptake, whereas those exceeding 70% have a disproportionately large impact. Also, the effectiveness of subsidies diminishes as previous-period subsidy rates increase, yet the expectation of adverse climatic conditions had the opposite effect. No significant heterogeneities in responsiveness to subsidies were observed on the basis of livestock wealth or gender. These findings are of practical significance for the myriad of governmental and non- government entities implementing subsidies for index-based insurance, and serve to guide future research. As hundreds of millions of dollars continue to be spent on such programmes, marginal improvements in cost-efficiency can deliver drastic improvements in beneficiaries’ climate resilience and wellbeing; it is such advances which this study seeks to unlock. 9 Acknowledgements First of all, I would like to thank Dr Tara Bedi for her guidance and assistance – both in organising the fieldwork underpinning this research and in supporting my personal and professional development throughout my studies at Trinity College Dublin. My heartfelt thanks also go to Dr Nathan Jensen, who not only served as an inspiration for me in pursuing this work but also generously offered his technical and practical expertise in planning and conducting this research. To Prof Alan Duncan, I would like to say thank you for shaping my personal and professional aspirations over the last three years and supporting me in working towards them, just as he has supported this project. Without a doubt, this research would not have been possible without the calm, kind, and patient support of Mr Chala Gidissa, who was instrumental in organising and coordinating all the activities relating to this work.A special thank you goes to Ms Askale Aderaw, Ms Mestawot Ketema, and Mr Tsedeke Desalegn for welcoming me so kindly to their team, contributing so much to the conceptualisation and framing of this work, and supporting this research over this year. I would also like to thank my colleagues and friends, Mr Guled Ismael and Mr Abdirashid Godana, for their support, companionship, and sense of humour. Special thanks also go to Mr Meseret and Mr Solomon Assefa, who were instrumental in making this work possible. Thanks to Mr Kenneth Macaria, Ms Deirdre McArdle, and Mr Guyo Denge for their kind support, as well as Dr Getachew Ayeha, Mr Mohammed Ebrahim, and Dr Wuletawu Abera from CIAT. I also extend my sincere gratitude to all those individuals who we spoke to in Borana and Dassanach, and all those communities which welcomed us. Equally, I’d like to thank all those individuals and organisations who participated in this research: Ayuda en Acción, CIFA, Global Communities, GOAL Ethiopia, the International Livestock Research Institute, Nyala Insurance Company, Oromia Insurance Company, the Oromia Irrigation and Pastoralist Development 10 Office, and ZEP-RE Reinsurance Company. My thanks also go to the five translators who contributed to this work. This research was made possible thanks to the support of the following entities: AWARD (African Woman in Agricultural Research and Development), the Jameel Observatory, the Trinity College Dublin Trust, and Trócaire/CST Ethoipia. This research is partially funded by the ENDEAVOUR II Scholarships Scheme (Malta), and may be co-funded by the ESF+ 2021-2027. My gratitude also goes to my friends from the MDP, as well as all those whom I met through the AWARD One Planet Fellowship programme. Grazzi lil familti u lil sħabi, lin-Nanna Mary, u lil Santa Liena. 11 List of Acronyms AME – Average marginal effect FGD – Focus group discussion IBLI – Index-Based Livestock Insurance KII – Key informant interview NGO – Non-governmental organisation TLU – Tropical livestock units, whereby 1 TLU is equivalent to 1 cow, 0.7 camels, 10 goats, or 10 sheep WTP – Willingness to pay 12 1. Introduction Climate change disproportionately affects the most vulnerable, among whom are small-scale farmers whose livelihood and food security are threatened by increasingly frequent and severe extreme weather events (Bezner Kerr et al., 2022; Pörtner et al., 2022). Over the last two decades, index-based insurance has emerged as an effective means of mitigating these burgeoning climate risks (Kramer et al., 2022). By issuing compensation payouts during catastrophic weather events, insured farmers are better able to cope during times of crisis and are thus less likely to adopt harmful coping strategies (Janzen and Carter, 2019). Furthermore, studies demonstrate that the security offered by insurance coverage promotes the adoption of riskier but more profitable production strategies, heightening farmers’ incomes (Nathaniel D. Jensen, Barrett and Mude, 2017; Janzen and Carter, 2019; Matsuda, Takahashi and Ikegami, 2019). In view of these benefits, numerous governmental and non-governmental entities have implemented programmes to subsidise index-based insurance, with 80% of such insurance products offered across developing countries to date having been sold at subsidised rates (Hazell, Jaeger and Hausberger, 2021). However, research on such subsidies is scant, so much so that Hazell and Varangis (2020) remark that “very little is really known about the effectiveness of insurance subsidies in achieving their intended purposes” (p.1); Hazell, Jaeger and Hausberger (2021), Nshakira-Rukundo, Kamau and Baumüller (2021) and Kramer et al. (2022) all reach similar conclusions. The body of evidence upon which development practitioners may draw in designing index-based insurance subsidies is thus severely constrained, likely impinging on their efficacy. The case of Index-Based Livestock Insurance (IBLI) in Ethiopia typifies these circumstances. Introduced to the country in 2012 to protect pastoralists from drought’s pervasive effects, multiple organisations have since offered subsidies for the product (Figure 1). Nonetheless, key aspects of such subsidies – including their intended outcomes, design implications, and factors which influence their efficacy – have not been explored in the academic literature to date. This greatly limits the scope for improving IBLI subsidies’ effectiveness and, by corollary, 13 detracts from the wellbeing of the country’s pastoralist farmers, who are among the most vulnerable populations to climate change (Bostedt et al., 2022). Figure 1: International organisations which have funded, implemented, and/or otherwise supported index-based livestock insurance subsidy programmes in Ethiopia (Zewdie, Taye and Fava, 2020; Ayuda en Acción, 2021; ICRC, 2022; World Bank, 2022; GOAL Global, 2023; ZEP-RE, n.d.). 1.1. Research Objectives This research evaluates the aims, design implications, and influencing factors of IBLI subsidies in seeking to inform the design of future index-insurance subsidy programmes. The research questions posed are detailed in Table 1 overleaf. These were conceived of and formulated in collaboration with CST Ethiopia and CIFA – two organisations with extensive experience implementing IBLI subsidies in Ethiopia – to ensure the relevance and applicability of results. A convergent mixed-methods approach is utilised, combining the findings of thematic coding of key informant interviews and focus group discussions, and multivariate regression analysis of secondary household survey data in answering the questions posed. 14 Table 1: The research questions addressed in this study. The third column uses Morse’s system of notation (1991) to delineate the methods used in answering each question. RQ1. What do key stakeholders regard as the intended outcome/s of IBLI subsidies? QUAL RQ2. How does subsidy design – specifically in terms of subsidy rate, targeting, and temporal distribution – influence subsidies’ impact? QUAN+qual RQ3. What factors other than subsidy design influence the attainment of intended outcomes? QUAL+quan 1.2. Outline The following chapter provides a review of the literature pertinent to index-based insurance subsidies. Chapter 3 then details the study’s methodology whereas the fourth provides qualitative and quantitative results in turn. These are then integrated and jointly discussed in Chapter 5. Chapter 6 concludes. 15 2. Literature Review This literature review consists of three sections. The first provides a brief review of IBLI’s impacts, whereas the second section evaluates evidence pertaining to index-based insurance subsidies. The third section then synthesises the review to motivate this research. 2.1. IBLI’s impacts Index-Based Livestock Insurance (IBLI) is a commercial micro-insurance product designed to protect pastoralist households against the loss of livestock to drought. By integrating satellite- measured data on forage availability and historical records of livestock losses, insurance providers are able to estimate livestock mortality rates, paying policyholders compensation once the expected mortality reaches a pre-established threshold (Chantarat et al., 2013). Since the product’s launch in Kenya in 2010 and in Ethiopia two years later, numerous studies have evaluated IBLI’s impact on pastoralists who acquire coverage. The assessment is overwhelmingly positive, with three overarching benefits being delineated in the literature: reduced livestock mortality, avoidance of adverse coping strategies, and augmented incomes (Table 2). Reductions in livestock losses are particularly significant as herds which shrink below 10 Tropical Livestock Units (TLU)1 tend to collapse in size, with households finding it exceedingly difficult to recover lost heads (Lybbert et al., 2004; McPeak, 2006; Barrett and Santos, 2014; Toth, 2015). IBLI coverage therefore drastically alters the fate of pastoralists by improving short-term responses to drought and protecting against dilapidating long-term poverty traps (Janzen, Carter and Ikegami, 2012; Chantarat et al., 2017). 1 1 TLU is equivalent to 1 cow, 0.7 camels, 10 goats, or 10 sheep. 16 Table 2: The three main benefits of IBLI identified by Jensen, Barrett and Mude (2017), Janzen and Carter (2019), Matsuda, Takahashi and Ikegami (2019), and Banerjee, Johnson and Mude (2022). Barrett et al.'s (2023) longer-term study corroborates positive impacts on education and the shift in production strategies, though no impacts on income or wealth were observed. Impact Explanation Reduced livestock mortality during drought IBLI payouts enable pastoralists to purchase forage, water, and medicine which they otherwise might not afford, averting the loss of livestock to drought Asset protection and improved food security and education IBLI stems the adoption of deleterious strategies during drought, such as selling off livestock, pulling children out of school to save on fees, or cutting down on food consumption Higher incomes The protection IBLI offers emboldens pastoralists to adopt production strategies previously deemed too risky (e.g., investing in veterinary services), in turn improving incomes 2.2. Subsidies for index-based insurance 2.2.1. Intended outcomes In view of the above benefits, it is unsurprising that so many organisations have leveraged subsidies to promote IBLI adoption. However, studies note that the outcomes sought are often more specific than simply heightening insurance coverage. As expanded on in Table 3, subsidies are sometimes implemented to overcome initial barriers to commercial sustainability, in other cases being used to improve access among marginalised groups or replace other forms of social protection. No studies to date have evaluated the intended outcomes of IBLI subsidies, such that it is unclear whether and how these apply in this case. 17 Table 3: The intended outcomes of insurance subsidy programmes discussed by Hill et al. (2014), Hazell and Varangis (2020), Hazell, Jaeger and Hausberger (2021), and Kramer et al. (2022). Intended outcomes of insurance subsidy programme Rationale To overcome initial hurdles to sustainable insurance markets Subsidies can diminish initial scepticism by promoting experimentation, and give confidence to providers that the critical mass required for commercial viability will be reached Inclusion/equity Lowering the effective prices of insurance via subsidies grants access to individuals/households who would otherwise be excluded Social protection, possibly to replace other social safety net/disaster relief aid programmes Heavily subsidised agricultural insurance augments/protects incomes during crises at minimal costs to the beneficiary, thus serving the same purpose as other forms of social protection The literature also makes clear that understanding subsidies’ aims is rarely straightforward. Binswanger-Mkhize (2012) highlights that aims are rarely defined explicitly – particularly when these are intertwined with obscured political goals (Hazell and Varangis, 2020) – making it difficult to discern the intended purpose/s of such programmes. Furthermore, Johnson et al. (2019) show that different actors involved in a single programme – in their study the IBLI pilot project in Kenya – may hold competing and incongruent expectations, yet these remain held together by what Mosse (2004) refers to as translation, the process of “read[ing] the meaning of a project into the different institutional languages of its stakeholder supporters” (p.647). Subsidy programmes are presumably not immune from this, suggesting that perceiving aims may also vary among stakeholders in such instances. 18 2.2.2. Subsidy design The literature on agricultural insurance subsidies repeatedly makes reference to ‘smart’ subsidy design2, referring to subsidies which are effective, cost-efficient, and well-targeted (Hill et al., 2014; Cai, de Janvry and Sadoulet, 2020; Hazell and Varangis, 2020). Although the body of evidence remains insufficient to clearly identify how different aspects of subsidies’ design contribute to making them ‘smart’, several insights do emerge from the literature. 2.2.2.1. Subsidy rates The subsidy rate offered is a key determinant of programmes’ cost efficiency as this needs to be sufficiently large to have the intended effect but conservative enough to be affordable by funders (Hazell, Jaeger and Hausberger, 2021). While no studies to date have focussed on the implications of this aspect of subsidies’ design, the literature on IBLI provides multiple findings of relevance. Studies by Timu et al. (2018), Matsuda, Takahashi and Ikegami (2019), and Takahashi et al. (2019) find that the probability of purchasing IBLI increases with the subsidy rate received. Demand for index-based insurance is, however, found to be price inelastic, suggesting that while higher subsidy rates do lead to increased uptake, the rise in demand is generally smaller than the increase in the subsidy rate (Cole et al., 2012; Bageant and Barrett, 2017; Jensen and Barrett, 2017; Cai, de Janvry and Sadoulet, 2020). Further to the above, studies by Takahashi et al. (2016), Bageant and Barrett (2017), and Jensen, Mude and Barrett (2018) demonstrate that over and above any price effect, the receipt of a subsidy itself increases the probability of acquiring coverage. Therefore, in addition to increasing demand by lowering the effective price of insurance, receiving a subsidy per se raises the likelihood of purchasing IBLI by between 4 and 18 percentage points on average3. This is explained on the basis of a ‘reminder effect’ in the latter study, whereby the subsidy – delivered via a physical discount coupon – serves as a “reminder to households of 2 Some publications which make use of this term include Hill et al. (2014), Keno, Diriba and Lemesa (2018), Hazell and Varangis (2020), and Cai, de Janvry and Sadoulet (2020). 3 Estimates taken from Bageant and Barrett (2017) and Jensen, Mude and Barrett (2018), respectively. 19 the availability of insurance” (p.22), rendering them more likely to consider acquiring coverage. In spite of the above evidence as well as ILRI's finding (2023) that “pastoralists perceived that the…subsidy rate of 50% was adequate” (p.6), Castaing and Gazeaud (2022) note that “the optimal level of subsidies remains unclear” (p.7). Further research on the association between the insurance uptake and the subsidy rate offered as well as the broader implications stemming from the choice of subsidy rate are therefore pertinent. 2.2.2.2. Intertemporal effects and duration The literature on index-based insurance also offers insights into the intertemporal implications of the subsidy rate. Most evidence in this regard relates to concerns that beneficiaries may become dependent on large subsidies, rendering subsidies “difficult to phase out or remove once established” (Hazell and Varangis, 2020, p.3) and thus causing programme costs to mushroom (Nshakira-Rukundo, Kamau and Baumüller, 2021). Research by Hill, Robles and Ceballos (2016) and Takahashi et al. (2019) explores whether current index- based insurance uptake is influenced by the subsidy rate received during previous periods, which they find not to be the case. This suggests the absence of price anchoring effects – referring to circumstances where “one-off subsidies…reduce future demand as the reduced price creates a focal point for purchasers who then become unwilling to pay more for the product later” (Takahashi et al., 2016, p.334) – but also implies that the risk of subsidy dependence is low. However, neither of these studies considers whether previous-period subsidy rates alter beneficiaries’ responsiveness to present-period subsidies – a pertinent consideration when determining the rate of subsidies offered over the course of a programme. Cai, de Janvry and Sadoulet's (2020) finding that farmers who received free index- based rice insurance were no more sensitive to the price of insurance during subsequent sales 20 windows suggests this not to be the case4, yet the possibility of subsidies’ impact being mediated by previous subsidies has not been tested. 2.2.2.3. Targeting The targeting of subsidies – one of the key tenets of ‘smart’ design – is also treated in the literature, albeit indirectly. Hazell and Varangis (2020) note that low-income and/or marginalised groups tend to be the target population of such insurance subsidies, particularly where subsidies’ aim is to replace other forms of social protection. According to research by ILRI (2023), this is regarded as an agreeable approach among Ethiopian pastoralists, themselves stating that “people who cannot afford to buy [IBLI]…and vulnerable households” (p.6) should be privileged in being provided access to IBLI. However, the evidence regarding the soundness of such an approach in pastoralist contexts is mixed. Jensen, Barrett and Mude (2017) find that IBLI disproportionately benefits poorer pastoralists who own fewer livestock, suggesting that targeting such low-income cohorts may augment insurance subsidies’ social impact. On the other hand, simulation-based studies by Janzen, Carter and Ikegami (2012), Ikegami et al. (2017) and Chantarat et al. (2017)5 suggest that targeting IBLI subsidies at those who are at risk of poverty – rather than those who are poor – tends to reduce poverty rates to a greater extent, augmenting programmes’ overall social impact. Further to the above, simulations conducted by Janzen, Carter and Ikegami and Chantarat et al. indicate that pastoralists’ willingness to pay (WTP) for IBLI varies according to the number of livestock owned, with both converging on a somewhat U-shaped relationship between WTP and herd size (Figure 2). This suggests that price sensitivity varies according to pastoralists’ livestock holdings, and thus that responsiveness to subsidies also differs according to herd size. However, no studies to date have empirically evaluated whether such heterogeneities 4 Note also that their data is based on just two sales windows and treats a particular circumstance where insurance was offered for free, meaning that the extent to which this result is likely to hold in other contexts is fairly ambiguous. 5 These studies are based on theoretically estimated willingness to pay for IBLI based on simulated herd dynamics over time. The fact that the IBLI demand is found to be price elastic, contrary to that in the empirical literature, means that the applicability of results to real-world situations should be treated with caution. 21 actually do exist, either on the basis of livestock owned or other characteristics on which subsidies may be targeted. Figure 2: Variation in willingness to pay (WTP) for IBLI by households’ livestock holdings. Figure taken from Janzen, Carter and Ikegami (2012). 2.2.3. Influencing factors Studies demonstrate that several factors unrelated to the design of subsidies bear great influence on subsidies’ impact. Amongst the most widely discussed of these is basis risk – the discrepancies between losses predicted by the index underpinning an insurance product and the losses actually experienced by policyholders. This is widely regarded as the “Achilles’ heel” (Jensen, Mude and Barrett, 2018, p.1) of index-based insurance, having been shown to significantly reduce demand for IBLI (Keno, Diriba and Lemesa, 2018) and other index-based products (Mobarak and Rosenzweig, 2012; Clarke and Wren-Lewis, 2013; Hill, Robles and Ceballos, 2016). However, research by Jensen, Mude and Barrett (2018) also demonstrates that high levels of basis risk reduce households’ responsiveness to IBLI subsidies – a result corroborated by Hill, Robles and Ceballos (2016). Basis risk thus doubly impinges on subsidies’ 22 effectiveness; it diminishes demand and quells any impacts of subsidies. A lack of trust in the insurance product – which often stems from basis risk – also has a similar effect according to research by Clarke and Wren-Lewis (2013) and Keno, Diriba and Lemesa (2018). Cai, de Janvry and Sadoulet (2020) and Chakraborty (2023) find that receiving a payout reduces subsidies’ impact, their effect falling by over 75% among payout recipients in the latter study. Crisis events are however observed to have the opposite effect in that households become considerably more responsive to the receipt of an IBLI subsidy following the loss of livestock to drought (Chakraborty, 2023). Furthermore, Jensen et al. (2024) demonstrate that the supply of insurance remains a major constraint on uptake in pastoralist contexts because of the high costs associated with marketing insurance in such remote, sparsely-populated areas (Keno, Diriba and Lemesa, 2018; Johnson et al., 2019; Kramer et al., 2022). It therefore follows that demand-side measures – including subsidies – can prove ineffective if not accompanied by supply-side support such as investment in insurance distribution networks. 2.3. Synthesis and motivation Three overarching conclusions may be drawn from this chapter, which has reviewed the impacts of IBLI and evaluated the literature on index-based insurance subsidies. Firstly, IBLI is effective at improving the wellbeing of pastoralist households, justifying its promotion via subsidies and thus motivating this study’s aim of improving subsidies’ effectiveness. Secondly, in spite of the evidence presented above, “very little is really known about the effectiveness of insurance subsidies in achieving their intended purposes” (Hazell and Varangis, 2020, p.1) for at least two reasons. First, while general motivations for the use of insurance subsidies have been identified in the literature (Table 3), the intended purposes of subsidies as understood by key stakeholders remain unexplored in the case of IBLI and other products. Second, the implications of subsidies’ design have scarcely been evaluated 23 empirically, with no studies to date having researched the influence of the rate, targeting, or temporal distribution of subsidies in detail. The first and second research questions of this study thus address these two lacunae, respectively exploring the aims of subsidies as held by different stakeholders and the significance of subsidies’ design. Thirdly, numerous factors exogenous to subsidies’ design have been shown to influence subsidies’ effectiveness, including basis risk, climatic conditions, and the supply of insurance. However, evidence on how these are experienced, perceived, and navigated by the implementers and beneficiaries of IBLI subsidies remains scarce. The bearing of other potentially significant factors, including intrahousehold dynamics and the bundling of subsidies with other demand-side measures (Timu and Kramer, 2021; Kramer et al., 2022), remain unexplored. This study’s third research question thus evaluates the significance of known influencing factors in greater depth while exploring the bearing of other factors which emerge from the data. 24 3. Methodology 3.1. Methodological underpinnings To achieve the aim of informing the design of index-based insurance subsidies, the ‘pragmatic approach’ advanced by Morgan (2007) is adopted. This constitutes a “practical and applied research philosophy” (Creswell and Clark, 2018, p.40) that avoids choosing between positivist and constructivist paradigms and prioritises answering the research questions posed in a manner “oriented towards…real-world practice”(p.37). Furthermore, in line with this pragmatic orientation, no a priori theoretical framework is adopted. Rather, this research draws upon the existing literature such that the research questions and the study’s design are based on the evidence presented in Chapter 2. 3.2. Research design A fixed convergent mixed methods design was utilised, simultaneously drawing on key informant interviews (KIIs), focus group discussions (FGDs), and panel household survey data (Figure 3). Its mixed methods nature stems from the integration of qualitative and quantitative data, whereas it is fixed since the use of both qualitative and quantitative data was envisaged ahead of commencing with research, and convergent as qualitative and quantitative data were analysed parallelly (Creswell and Clark, 2018). 25 Figure 3: Diagram summarising the mixed methods design adopted in this study. The quantitative data used was a publicly available dataset provided by ILRI. Figure draws on Creswell and Clark (2018). By using qualitative methods, stakeholders and beneficiaries are given voice in the study, encouraging reflexivity and allowing the research to align with Angus Deaton’s position – with which I agree – that: “social plumbing should be left to social plumbers, not outside experimental economists who have no special knowledge, and no legitimacy” (Deaton, 2020, p.26). The quantitative strand, on the other hand, allows for the observation of actual rather than stated behaviour. A mixed methods approach thus pools these benefits whilst allowing for completeness and triangulation (Fenech Adami and Kiger, 2005; Creswell and Clark, 2018). 26 3.3. Study sites The data underpinning this study was collected from two sites in Ethiopia: the Borana plateau6 in the Oromia Region, and the Dasenech district of the South Omo Zone in the Southern Nations, Nationalities and Peoples' Region. The Borana plateau, located 570km south of Addis Ababa, encompasses arid and semi-arid areas covering 48,360km2 (Amare et al., 2019). It is characterised by a bimodal rainfall pattern with rainy seasons running from March to May and October to November (Takahashi et al., 2016). Most of the population depends on transhumant pastoralism, though agropastoralism is practiced in parts (Amare et al., 2019). Borana is also the area where IBLI was first piloted in Ethiopia (ILRI, n.d.) and has been the site of most insurance subsidy programmes implemented to date. Dasanach, located 956km southwest of Addis Ababa, is bordered by South Sudan to its west and Kenya to its south (Fesseha et al., 2022). This hot semi-arid area follows the same rainfall pattern as Borana, though rains are generally scarcer and more irregular, causing the area to be afflicted by flash floods (OCHA, 2023; Tadesse, 2023). Most of the population relies on pastoralism, with flood-recession agriculture being practiced along the banks of the Omo Delta (Adicha, Alemayeh and Darcho, 2022). IBLI was introduced to Dasenech later than Borana, with multiple subsidy projects having since been implemented. Poverty and food insecurity are “widespread and severe” (Teka, Temesgen Woldu and Fre, 2019, p.2) across both areas, with recent droughts having significantly impinged on pastoralists’ wellbeing (Teka, Temesgen Woldu and Fre, 2019; OCHA, 2023; Nurie, 2024). 6 Also known as the Borana zone. 27 3.4. Methods 3.4.1. Qualitative data collection 3.4.1.1. Key informant interviews Interviews were conducted with eleven individuals (Table 4) who were purposefully selected to represent the following pre-identified stakeholder categories7: insurance companies, NGOs, government entities, research entities, and subsidy project donors. Where possible, a minimum of two interviewees from each stakeholder category were recruited to avoid limiting a stakeholder category’s perspective to that of a single individual. Table 4: The codes and affiliations of KII participants disaggregated by stakeholder category, and the location where each interview was conducted. Note that Global Communities is an NGO, though the interviewee was representing and overseeing the interests of donors in an IBLI subsidy project implemented by the NGO. Category KII Code Affiliation Location (Re)Insurance Companies Ins1 ZEP-RE Reinsurance Company Company office, Addis Ababa Ins2 Oromia Insurance Company Company office, Addis Ababa Ins3 Nyala Insurance Company Virtual NGOs NGO1 GOAL Ethiopia NGO office, Addis Ababa NGO2 CIFA NGO office, Moyale NGO3 Ayuda en Acción NGO office, Addis Ababa Government bodies Gov1 Oromia Irrigation and Pastoralist Development Office Public coffee shop, Moyale Gov2 Oromia Irrigation and Pastoralist Development Office Entity office, Addis Ababa Researchers Res1 ILRI Hotel lobby, Yabelo Res2 African Development Bank ILRI Campus, Addis Ababa Donors Don1 Global Communities Virtual 7 These were identified in collaboration with partner organisations, with many of the participants having been recruited via partners’ networks. 28 Semi-structured interviews were used as they allow for an in-depth understanding of participants’ views on the key aspects of this research and permit emergent points of interest to be probed (Ritchie et al., 2013). An interview schedule8 was prepared to facilitate comparison across interviewees, though this was adapted based on interviewees’ background and the time available. Interviews lasted between twenty-five minutes and one and a half hours. They were conducted in English9 and recorded, being subsequently transcribed verbatim (Bryman, 2012). 3.4.1.2. Focus group discussions Twelve FGDs were conducted with pastoralists in Borana and Dasenech between April and May 2024. The use of this format was motivated by practical considerations, primarily the availability of facilitators and translators. Compared to an individual interview approach, FGDs allow for a greater number of participants over a given timeframe, thus enriching the perspectives and experiences documented. Bryman (2012) also notes that this group-based approach allows for individuals’ statements to be challenged, possibly resulting in “more realistic accounts” (p.503), though this may have also dissuaded the expression of atypical views (Ritchie et al., 2013). Discussions were conducted in six locations (Figure 5), selected in view of logistical constraints and on the basis of there having been IBLI sales at the locations in recent years. Participants were not selected individually but a call for participants10 was made, with those opting-in being allocated to either mixed, male-only, or female-only groups. Groups were stratified by gender to allow women to participate more actively given that women may have been hesitant to speak as openly in the presence of men. As far as possible, various age cohorts were represented in each discussion. Participants had to have acquired IBLI coverage at least once to ensure familiarity with IBLI and thus informed perspective about subsidies. In each 8 See Appendix 1 – Interview scheulde and discussion guide 9 The only exception to this was the interview with Gov1, which was conducted in Amharic with the assistance of a live translator. 10 Where possible, discussions were scheduled on the same day as township meetings to facilitate recruitment. 29 site, however, one discussion was organised with individuals who had never acquired insurance to mitigate bias stemming from self-selection into insurance. A pre-prepared discussion guide11 formed the basis of questions posed by Amharic-speaking facilitators. Questions were then translated to local languages, and vice-versa for responses. All discussions were recorded and translated verbatim directly from the local language to English. Table 5: The location and composition of FGDs, listed in the order they were conducted. Codes are created using a system whereby the first letter represents the study side (B=Borana, D=Dasenech), the second character the gender composition (M=male, F=female, B=both/mixed), and the number serves to differentiate discussions in the same site with the same composition. Note that the asterisk (*) demarcates participants who had not previously had IBLI coverage. Discussion code Location (Kebele; Woreda) Participants (Gender [Age]) BF1 Bade; Moyale F(40s)x2, F(50s) BM1 Bade; Moyale M(50s)x2, M(60s) BB1 Bolade; Moyale F(20s), F(30s), F(40s), F(50s) M(30s), M(40s), M(50s), M(60s) BB2* Bolade; Moyale F(20s)*x2, F(50s)* M(30s)*, M(40s)* BM2 Bokola; Moyale M(20s), M(30s), M(40s) BF2 Bokola; Moyale F(20s)*, F(30s), F(40s) DB1 Fejej; Dasenech F(20s), F(30s), F(60s) M(50s)x2, M(60s) DB2* Fejej; Dasenech F(20s)*, F(30s)* M(40s)*x2, M(50s)* DF1 Ocholoch; Dasenech F(20s)x2, F(30s), F(40s), F(50s), F(60s) DM1 Ocholoch; Dasenech M(20s), M(30s), M(40s)x2, M(50s)x2 DF2 Arikol; Dasenech F(20s)x3, F(30s)x2, F(50s) DM2 Arikol; Dasenech M(20s)x2, M(30s)x2, M(60s)x2 3.4.2. Qualitative data analysis Thematic coding of KII and FGD transcripts was carried out using NVivo 14 (QSR International, 2023) on the basis of themes identified in the literature, with additional codes being created 11 See Appendix 1 – Interview scheulde and discussion guide 30 based on themes emerging from the data12. To facilitate structured comparisons across respondents, framework analysis was then conducted whereby responses were tabulated13, with rows representing different participants and columns different themes (Gale et al., 2013). This approach proved particularly suited to this research as the data was fairly structured and homogenous in terms of questions asked and themes covered (Ritchie and Lewis, 2003). 3.4.3. Quantitative data collection A publicly available dataset provided by the International Livestock Research Institute (ILRI), was utilised14. The author was not involved in its collection, so the procedures detailed hereunder are based on the codebook provided by Ikegami and Sheahan (2018). A baseline survey was conducted in March 2012 among households from twenty-four reeras15 across eight districts of Borana16, selected to capture agro-ecological and livelihood variability and in view of logistical constraints. Household sampling was clustered at the reera level by conducting a census of each selected reera, stratifying households into terciles according to livestock holdings, then randomly selecting 15% of households – a third of which hailed from each tercile. Where 15% of households amounted to less than 25 households, neighbouring reeras were combined into a single sampling unit, referred to as a ‘study site’. A total of 515 households were surveyed at baseline17. IBLI is sold during two windows annually – August/September and January/February – with contracts covering the subsequent twelve months (Figure 4). The product was first sold in Borana during Aug/Sept 2012, followed by a second sales window in Jan/Feb 2013, and so 12 The a priori and emergent codes used are presented in Supplementary Table 1. 13 Separate tables were used for FGDs and KIIs given that questions differed across the two. 14 This is accessible at https://data.ilri.org/portal/dataset/ibli-borena-r1. Full surveys for each round are available at https://data.ilri.org/portal/dataset/ibli-borena-r1/resource/1e97501a-5789-4066-aa22- c33c04f1ff64?inner_span=True. 15 A reera constitutes the second smallest administrative unit in Ethiopia, consisting of approximately 100-300 households (Takahashi et al., 2019). 16 Namely Arero, Dhas, Dillo, Dire, Moyale, Teltele, Yabello, and Miyo. 17 528 households were selected, but 13 could not be contacted. 31 forth (Table 6). The first follow-up survey was conducted in March 2013 – after the first two sales windows – with subsequent rounds conducted each March until 2015. Figure 4: The timing of IBLI contracts sold in Borana. Figure taken from Ikegami and Sheahan (2018). For further details, see Ikegami and Sheahan (2018, p.4-6)18. To promote IBLI uptake, discount coupons were randomly distributed among surveyed households. In each window, a tenth of participating households received a 10% discount, a second tenth received 20%, and so forth up to an 80% discount, with 20% of households not receiving any coupon19. These discounts applied to the first 15 TLU insured and were only valid for the subsequent sales window. Coupons were rerandomized and redistributed each IBLI 18 Note that since this data was collected, IBLI has shifted from an asset-replacement to an asset-protection model, whereby indexes are calculated earlier in the year and payouts are paid before or during the dry season, rather than afterwards (Mude, 2017). 19 Ten surveyed households received a 100% subsidy each year for the purposes of a separate study; these are excluded from the sample used in this study. See Supplementary Table 2 for the balance tests between coupon recipients and non-recipients, which shows that the discount coupons were in fact randomly distributed. 32 sales window. Learning kits intended to improve participants’ understanding of IBLI were also distributed randomly prior to the first two sales windows20. Table 6: The timing of survey rounds and IBLI sales. Each survey round covers the previous two sales windows. Table taken from Ikegami and Sheahan (2018). 3.4.2. Quantitive data analysis Households receiving a 100% subsidy (see Footnote 19) were excluded from the sample, as were replacement households, those which only participated in the baseline survey, and those which dropped out and subsequently rejoined the study. Data cleaning and variable creation were subsequently undertaken, with summary statistics then produced as relevant21. Prior to conducting econometric analyses, the top 1% of values for herd size were winsorized to mitigate the influence of outliers. 20 For full details, see Takahashi et al. (2016, p.326). 21 All the stages of quantitative analysis hereunder were carried out using R (R Core Team, 2024) and the packages cited in Appendix 3 – R packages used. 33 The following model was then constructed to evaluate variables’ influence on insurance uptake, Pr(𝑈𝑝𝑡𝑎𝑘𝑒௜௧ = 1) = 𝛷൫𝛼 + 𝛽𝐷௜௧ + 𝛿𝑋௜௧ + 𝛿ଵ𝐿௜௧ + 𝛿ଶ𝐿௜௧ ଶ + 𝜑𝐻௜ + 𝜔 + 𝑡 + 𝜇௜ + 𝜀௜௧൯ (1) where 𝑈𝑝𝑡𝑎𝑘𝑒௜௧ is a binary dependent variable capturing whether or not household i acquired insurance coverage in sales window t, and 𝐷௜௧ is a binary variable representing the receipt of a discount coupon, and 𝑋௜௧ and 𝐻௜ respectively capture time-variant and -invariant household characteristics22. 𝐿௜௧ is one time-variant household characteristic capturing the number of livestock owned by household i in period t, with 𝐿௜௧ ଶ representing the square of this value. 𝜔 and 𝑡 represent study-site and sales window dummies, respectively. The error is composed of both 𝜇௜, the unobserved time-invariant household effect, and 𝜀௜௧, an idiosyncratic error with zero mean, finite variance 𝜎ఌ ଶ, and distributed independently and identically. Estimating this model, however, poses econometric challenges. The use of a pooled probit estimator would produce biased results due to the presence of unobserved household fixed- effects (𝜇௜), yet a standard fixed-effect probit approach would produce inconsistent estimates owing to the incidental parameter problem which arises due to the panel being short compared to the number of households surveyed (Jensen, Mude and Barrett, 2018). Therefore, a variant of the probit pooled estimator developed by Wooldridge (2010, pp.483– 490)23 is used. This controls for household fixed effects via the introduction of within- household means of time-variant characteristics (𝑋ത௜), assuming independence conditional on those means and error terms to be normally distributed. The model estimated is thus: Pr(𝑈𝑝𝑡𝑎𝑘𝑒௜௧ = 1) = 𝛷൫𝛼 + 𝛽𝐷௜௧ + 𝛿𝑋௜௧ + 𝛿ଵ𝐿௜௧ + 𝛿ଶ𝐿௜௧ ଶ + 𝜑𝐻௜ + 𝜔 + 𝑡 + 𝜗𝑋ത௜ + 𝜀௜௧൯ (2) To further explore the implications of targeting subsidies, an interaction term between 𝐷௜௧ and the number of livestock owned by the household (𝐿௜௧, 𝐿௜௧ ଶ) was added to Model (2) given 22 Time-variant and invariant control variables are detailed in Supplementary Table 3, the choice of which is in the expansive body of evidence on the determinants of IBLI uptake. 23 This is referred to as Chemberlain’s random effects probit model. 34 the significance of livestock assets in pastoralist systems24 and the association between herd size and willingness to pay for insurance posited by Janzen, Carter and Ikegami (2012) and Chantarat et al. (2017). Modelling herd size using variables capturing the number of livestock owned (𝐿௜௧) as well as its square (𝐿௜௧ ଶ) allows non-linearities in this association, in line with the U-shape shown in Figure 2. Three other variants of Model (2) were then constructed to study the significance of subsidy rates in detail. In the first, a continuous variable capturing the discount rate of the coupon received was used instead of the binary variable capturing the receipt of a discount, whereas the second incorporated both of these variables. Instead of these two variables, the third model incorporated a set of dummy variables representing the different discount rates distributed. Four further models were then constructed to study intertemporal effects of subsidies. The first incorporated two binary variables representing the receipt of a coupon in the two previous sales windows (𝐷௜௧ିଵ; 𝐷௜௧ିଶ) into Model (2). Both these variables were interacted with 𝐷௜௧ in the second model. Similarly, the third model included two variables capturing the discount rate received in the previous two windows, with these being interacted with 𝐷௜௧ in the fourth. 3.5. Ethics and data storage This research was approved by the School of Natural Sciences at Trinity College Dublin25 and the partner organisations. Written consent was sought from all participants. Given the low literacy levels among pastoralist populations, particular attention was paid to ensuring that the aims of the research and the implications of participating were clear to participants before 24 As discussed by Anderson and Broch-Due (1999) and Little et al. (2008), animals simultaneously serve as a source of income, depository of value, and the primary form of wealth. 25 See Appendix 4 – Ethical Approval 35 soliciting written consent26. Furthermore, local community-based organisations were consulted in determining the compensation to be provided to participants to ensure this was suitable. Regarding KIIs, these were conducted under the condition of anonymity, and all efforts were made to protect participants’ identities. All materials were handled and stored in line with TCD’s data storage policies. 26 Sample information and consent forms are provided in Appendix 4 – Ethical Approval. A fingerprint was collected from participants who were illiterate, with a second individual signing to confirm that the participant wilfully provided their consent in such instances. 36 4. Results 4.1. Qualitative results 4.1.1. Intended outcomes of subsidies Table 7 below gathers the intended outcomes of IBLI subsidies expressed by KII participants. Heightening IBLI uptake was the most frequently cited, with subsidies generally being framed as a “pull factor” to promote this “new product” (NGO2). Both KII and FGD participants noted that subsidies heighten demand on both the extensive and intensive margins, both “encouraging pastoralists to join the scheme” (Don1) and to purchase larger policies when they do. Many participants also linked the aim of increasing IBLI uptake to that of creating awareness about the product, with insurance company representatives generally associating this with the creation of sustained demand: “The objective is to create awareness in all pastoralist areas so that when the time comes for sales they come…and ask for insurance” (Ins1). Table 7: The intended outcomes identified by KII participants, disaggregated by stakeholder classification and listed in order of frequency. Stakeholder category Insurance NGO Government Research Donor Total Respondents (n) 3 3 2 2 1 11 Increased uptake 2 2 2 2 0 8 Awareness creation 2 1 1 2 0 6 Support commercial sustainability 1 3 0 2 0 6 Inclusion 0 3 1 1 0 5 Resilience/risk reduction 0 2 1 0 1 4 Geographical dissemination 1 2 0 0 0 3 Social protection 2 0 0 1 0 2 Intensification 0 1 0 0 1 2 Increase trust, show commitment 1 0 0 0 0 1 Political goals 0 0 0 1 0 1 Donor recognition 1 0 0 0 0 1 37 NGO representatives and researchers consistently underscored that one intended outcome of IBLI subsidies was “to keep…insurance compan[ies] in the market” (Res1). Most interviewees framed these commercial barriers as ‘teething problems’, though others thought these challenges to be structural and argued for continuous subsidies: “making it [IBLI] a commercial product is difficult, which means that continuous subsidy must be there” (NGO1). Facilitating access to livestock insurance for vulnerable/disadvantaged groups27 was another aim of subsidies consistently identified by NGO representatives, this also being widely recognised among FGD participants. Nonetheless, only two KII participants explicitly identified IBLI subsidies as a form of social protection, one of whom remarked: “instead of providing other social security like food transfer or cash transfer, [it is] better to finance their livestock insurance” (Ins3). Multiple participants did however note that IBLI subsidies serve as a form of climate adaptation and means of augmenting pastoralists’ resilience, with one interviewee also highlighting that subsidies “support the pastoral peoples from dropping out from pastoralism” (Don1). Three interviewees mentioned that subsidies are employed to further IBLI’s geographical dissemination. For NGOs, this was thought of in terms of subsidies attracting insurance providers to neglected areas, thus granting access to previously excluded groups. However, promoting geographical dissemination via subsidies was framed by an insurance company representative as a means of reducing companies’ risk, explaining that “the projects so far are concentrated in one region. If risk happens, then there is a lot of payout but…where you do in different location, risk is diversified” (Ins1). Men and women who participated in FGDs conducted in Borana also noted that subsidies also instill confidence and trust in the product – a sentiment echoed in Ins1. The same interviewee also mentioned that subsidies serve to bolster donors’ recognition and status, with another citing the use of subsidies for political reasons28. 27 Women and female-headed households, people with disabilities, the elderly, and youth were identified as forming part of this cohort. 28 More specifically, this interviewee mentioned that subsidies could be used as a means of channelling resources along lines of political allegiance or as a means of winning political favour. 38 4.1.2. Subsidy design Interviewees from across stakeholder groups explicitly employed the notion of ‘smart subsidies’, yet views on subsidies’ design did vary. Prior to evaluating these divergences, it is worth noting that among KII participants, the subsidy design conceived of was generally linked to the perceived aims of subsidies. When highlighting subsidies’ role in heightening demand to overcome initial hesitancy, participants generally spoke of relatively widely-distributed subsidies to be phased out over time, referred to as the ‘start block approach’. Where subsidies were conceived of as a tool for inclusion, resilience, or social protection, continuous subsidies targeted at households/individuals meeting certain criteria tended to be cited, referred to as the ‘protection approach’ (Table 8). While generally conceived of as two entirely distinct interventions, these were not mutually exclusive, with the vast majority of participants advocating for the use of both. One interviewee, for example, said that “there’s no way that you’re going to provide subsidy for long period of time” (NGO2) but subsequently claimed that “the families who cannot afford need to be subsidised, maybe for long time” (NGO2). A third approach to subsidies emerged from conversations with three KII participants from different groupings. These highlighted the value of directly supporting insurance companies via financing for outreach, awareness creation, and supply network development which is here termed a ‘direct finance approach’. Table 8: A characterisation of the different approaches to insurance subsidies conceived of by KII participants. These are not mutually exclusive, with 10 of 11 interviewees explicitly or implicitly advocating for at least two of these. Start block approach Design: Widely distributed and temporary Primary aim/s: Overcome initial hesitancy and increase demand Protection approach Design: Continuous and targeted Primary aim/s: Inclusion, climate resilience, and social protection Direct finance approach Design: NGO/donor pays for or reimburses (a share of) insurance companies’ activities related to insurance offering Primary aim/s: Support product’s commercial viability/sustainability 39 4.1.2.1. Subsidy rate There was no consensus among KII participants regarding an appropriate subsidy rate, with a split emerging between Government and insurance company representatives on the one hand, and NGOs and the donor representative interviewed on the other. The former group advocated for a “minimum of fifty percent subsidy” (Ins1), rising to as much as 85%, whilst the latter generally held that “the subsidy…should not be more than 50%” (Don1). FGD participants tended to agree with the former: “We want the percentage increased to 80 or 85; the 75 is not sufficient” (F:DF2). Three KII participants were, however, quick to mention that there was a gap in evidence on this point, one stating that “the amount [subsidised]…has not been supported with sufficient data…my response will be too subjective” (Ins3). 4.1.2.2. Duration and temporal distribution As noted above, the majority of KII participants advocated for both temporary subsidies (to be offered more broadly) and continuous subsidies (targeted at vulnerable households). In speaking of the former, three participants spoke of a phase-out approach, whereby the subsidy rate is subject to a pre-determined “periodic reduction” (NGO1) such that it “decline[s] gradually” (Ins3) down to 0%, with one government official arguing that this setup served to avoid breeding dependence among beneficiaries. Most interviewees felt that subsidies (not referring to temporary or continuous subsidies specifically) should be adaptive, in that their rate and duration should vary according to circumstances. Two NGO participants cited climatic conditions as one such factor to consider, noting how their organisations had raised subsidy rates amid droughts29, whereas another participant argued that offering large subsidies where “the community has been already investing” (Res1) in IBLI served to undermine the product’s sustainability, suggesting that subsidies should be adapted according to prevailing demand. One Government official spoke of the importance of “decreasing the amount of subsidy…as the people understand more 29 “With drought…and the last drought we feel that there are a lot of changes in the community – people have lost everything, so we feel that there’s a need now to increase ee subsidy. We have discussed this with partners like [partner organisation name]. We are also in this venture. Then we increase it to fifty percent.” (NGO2) 40 about the livestock insurance” (Gov1), with another interviewee noting that their organisation had also raised subsidy rates in view of the introduction of a different organisation’s subsidy project. 4.1.2.3. Targeting KII participants unanimously supported the idea of targeting, particularly in relation to the offering of continuous subsidies (Table 8). A split among interviewees similar to that cited above was nonetheless evident in terms of the approach to targeting. For Government officials and insurance company representatives, targeting was conceived of in terms of differentiating the subsidy rate across recipients, such that those with “better [more] wealth can be subsidised with minimum rate, the others can be subsidised with maximum rate” (Ins2). On the other hand, NGO and donor representatives conceived of offering subsidies solely to vulnerable individuals/households identified via community wealth ranking, making no reference to heterogeneous subsidy rates. GOAL Ethiopia’s approach was unique in that different subsidy rates were offered for different classes of livestock – “cattle fifty percent, sheep and goat eighty percent” (NGO1) – the reasoning being that this would favour women who tend to own smaller ruminants. In the same vein as GOAL, most participants spoke of targeting subsidies on the basis of gender – discussed further below – though it was more often the case that participants thought of targeting based on income or wealth. In FGDs, participants’ sentiments towards targeting were mixed. In only one of the mixed- gender discussions did participants express a preference for targeted subsidies (BB1), with the other mixed groups insisting on subsidies being “distributed equally” (DB2). Men were generally more open to targeting vulnerable, poor, and/or female-headed households, whereas participants in South Omo were more reluctant than those in Borana. Where targeting was opposed, the main reason cited was that “differences in service [i.e., subsidies] promote grievances” (M:BB1) and thus conflict within communities. Two male participants from Borana also argued that “people are the same due to drought…wealthy people that had more cattle have come to poverty by drought so that support has to be given in equal means” (BB1). 41 4.1.2.4. Limits One aspect of subsidies’ design which emerged from KII interviews was the idea of limits on the number of heads subsidised. One insurance company representative noted that an upper limit was important to prevent better-off households from "abusing the benefits" (Ins3) of such programmes, whereas an NGO endorsed a lower limit on the basis of this dissuading the participation of those with few livestock, for whom programmes other than IBLI subsidies were better suited. 4.1.3. Influencing factors 4.1.3.1. Stakeholder coordination All bar one KII participants cited coordination among project stakeholders as a significant determinant of subsidies’ effectiveness, with many highlighting “how significant a lack of coordination can be in hindering programmes [from] being implemented properly” (NGO3). Participants spoke of multiple subsidy projects – generally having different subsidy rates, targeting criteria, etc – being simultaneously implemented “in the same community, the same village…[while] in next village there is no other organisation”. This was “confus[ing] the community” and fostering “competition rather than collaboration”, as organisations were effectively being forced to raise subsidy rates in order to reach donor targets. KII participants also proposed solutions to this “coordination gap”. Researchers and NGO representatives called for a “policy that regulates agricultural insurance” (Res1) and “map[s] out the responsibilities of the private sector, NGOs, research organisations,…Government and the like clearly” (Res2), with one participant highlighting the need for a Government-convened stakeholder forum. Others pointed towards the approach adopted by the Kenyan Government in implementing the Kenyan Livestock Insurance Programme, whereby IBLI subsidies are funded and “institutionalised through the government structures” (NGO1), with one interviewee remarking: “I can say that the Government is of paramount importance here; it has to come in [and] support the programme” (NGO2). The fact that one Government official spoke of “mainstreaming” IBLI subsidies into their office’s work suggests that they too saw scope for an increased Government role. 42 4.1.3.2. Bundling and digitalisation KII participants from across stakeholder groups spoke of the importance of bundling IBLI subsidies with interventions which improve access to complimentary inputs and services, such as agricultural inputs30, financial services (particularly credit), and market information systems. This was often mentioned in conjunction with digital tools, with six interviewees – predominantly from NGOs and insurance companies – speaking of their value vis-à-vis providing pastoralists with information about IBLI and insurance subsidies, whilst also reducing the costs of programme delivery. 4.1.3.3. Gender and intrahousehold dynamics Women were consistently recognised as more vulnerable than men among KII and FGD participants31, with many citing women’s household and childbearing responsibilities as constraining their income-earning opportunities: “women cannot give up her children to going another place for searching money” (W:BB1). However, multiple KII participants also noted that because women in Borana have superior financial management skills – acquired via their engagement in “business” (NGO2) and “petty trade” (Res1)32 – they are able to make better use of subsidies, suggesting that recipients’ gender may mediate subsidies’ impact: “subsidy…mainly benefitted women-headed household” (Res1). Linked to this, one interviewee remarked of Borana that “a husband and…a wife, they buy different contract” (Res1), this being reiterated during one female-only discussion in the same site. While one participant from Dasenech did claim that “it is mostly women who took out 30 Including forage and veterinary services. 31 As one interviewee noted, “in any pastoral community, the most disadvantaged people are the women” (NGO2). 32 Moyale, one of the major cities in Borana, is a particularly active site for trade given that it transverses the Kenya-Ethiopia border. 43 the insurance” (W:DF2), there were no references to husbands’ and wives’ IBLI uptake decisions being made in isolation in this site33. 4.1.3.4. Insurance product quality FGD participants – particularly those in Dasenech – spoke strongly about insurance products which, in their view, had not paid out when they were meant to. One man recalled that when this had occurred, “all the people decided to let go of the insurance” (M:DM2), another rebuked the insurance company for “not honour[ing] their promise” (M:DM2), and a third stated that: “the satellite should not be used, we refuse to agree to satellite feedback” (M:DM2). While not stated explicitly, these statements strongly suggest that subsidies would have done little to heighten insurance uptake of the products in question. 4.1.4. Other findings Despite showing lower levels of understanding about IBLI34, participants in Dasenech – particularly women – were extremely eager to purchase insurance policies, with one woman claiming that “we really want to purchase insurance” (F:DF2), another remarking: “I would have been purchasing more insurances as big as the sky” (F:DF1). However, many spoke of IBLI as a pathway to acquiring livestock rather than a means of protecting those owned. When asked about “the best thing about insurance”, one man responded: “The fact that I was able to purchase an animal through that” (DB2). Linked to this, participants from this site consistently indicated that it was common practice to purchase IBLI policies despite not owning any livestock35; “it’s not only those who have livestock, also people without livestock 33 On the contrary, one man stated that “either the woman can purchase or the man can purchase on behalf of the family. If the payout is done, it is then again a joint decision to purchase whatever is necessary and required” (M; DM1). 34 Many seemed unaware that IBLI policies offered coverage for 12 months, speaking of insurance policies purchased multiple years ago for which they were still awaiting payouts. 35 In contrast, one woman from Borana stated: “we know about insurance…the criteria for membership of insurance is someone who have cattle” (F:BF2). 44 have the insurance” (F:DF1) one woman remarked, with another reiterating that “persons…without cattle also purchased the insurance” (F:DF1). 4.2. Quantitative results 4.2.1. Summary statistics The data consists of 454 unique households from across 17 study sites, of which 438 participated in all four survey rounds36. As shown in Table 9 below, the vast majority of households were headed by men, with the average head being 50 years old and having received less than one year of formal education. The results of a risk preference experiment conducted indicate that an eighth of households were highly risk averse, with 41% showing low levels of risk aversion. Table 9: Summary statistics of the data. The first six variables are time-invariant and calculated based on the first survey round, whereas other values are calculated by pooling data across all four rounds. Variable Mean Median Std. dev. Head age (years) 50.15 48.00 18.21 Head years of formal education 0.59 0 2.20 Head gender = female (% of sample) 22% Risk aversion = low (% of sample) 41% Risk aversion = moderate (% of sample) 46% Risk aversion = high (% of sample) 13% Nomadic (% of sample) 5% Partially settled (% of sample) 38% Fully settled (% of sample) 58% Household members (n) 6.79 6.00 2.66 Livestock owned (TLU) 17.30 9.63 28.57 Annual household income (ETB) 11062.75 6240.00 30486.23 Share of income from livestock (% of income) 73% 96% 0.36 Household cash savings (ETB) 1718.80 0.00 12977.09 Rangeland expectation (1=much worse than average, 5=much better than average) 1.81 1.00 1.02 Livestock lost to drought since last sales window (TLU) 0.14 0.00 0.95 Social groups (n) 0.80 1.00 0.86 36 Tests of balance (see Supplementary Table 4) show that dropouts were more likely to have younger heads, fewer household members, formed part of fewer social groups, and derived a larger share of their income from livestock. Econometric results discussed hereunder were nonetheless unchanged when dropouts were excluded. 45 Households had a median of 6 members and a mean annual income of 6,200 Ethiopian birr37. Nearly all of the median household’s income was derived from livestock-related activities, yet only 24% were nomadic or semi-nomadic38. While average cash savings were 1,700 ETB (~100 USD), the median household had no savings. The mean number of livestock owned across the four rounds was 17.3 TLU39 with the median being just below 10 TLU, though as shown in Figure 5, this tended to increase over the course of the panel. In spite of households generally expecting worse than average rangeland conditions, households lost the equivalent of three sheep to drought each year on average. The expected conditions and loss of livestock to drought were effectively unchanged over the period surveyed. Figure 5: The distribution of households’ herd sizes over the course of the panel, measured in TLU. Sales window 0 refers to the data at the baseline survey. 37 This equates to approximately 60 USD (2012) per capita annually. 38 This refers to instances where certain household members stay in a fixed base camp, and others travel with the majority of households’ herd. 39 Equivalent to roughly 173 goats/sheep, 17 cows, or 14 camels. 46 With regards to IBLI uptake, an average of 15% of households purchased insurance coverage per sales window, though Figure 6 below demonstrates that this level varied considerably over the period covered, declining following the first sales window, before climbing then falling again. Less than a tenth of households’ livestock assets were covered by the average policy. The mean coverage rate over the course of the panel is 26%, though this also fluctuates (Figure 6). Approximately 41% of households did not acquire coverage in any of the six sales windows. Figure 6: The proportion of households purchasing insurance in each sales window (red) and those with coverage at the end of each sales window (blue). Table 10 provides a comparison of the characteristics of households which purchased and did not purchase IBLI in a given sales window when pooling the data across sales windows. Those purchasing IBLI were more likely to have received a discount coupon and tended to have received higher discount rates. Households owning more livestock40 and which formed part 40 Supplementary Figure 1 shows that the pattern of purchasing households tending to have larger herds is consistent across survey rounds, though particularly pronounced during the 3rd, 4th, and 5th sales windows. 47 of a larger number of social groups were more likely to purchase coverage, as were those with fewer members and which expected worse rangeland conditions. There were only two instances of fully nomadic households purchasing coverage. Table 10: Tests for differences in characteristics of purchasing and non-purchasing households when pooling data across sales windows. S.D. stands for standard deviation, F denotes an F-test, and X2 a chi-squared test. Values differ from Table 9 as the first survey round is not included in this table given that IBLI started to be sold after this round. Variables N Mean S.D. N Mean S.D. Diff. Test Purchased IBLI No Yes Received discount coupon (1=yes) 2254 0.794 403 0.931 0.137 X2=41.449*** Discount rate (% of premium price) 2254 35.8 27.4 403 48 25.6 12.2 F=68.76*** Head age (years) 2253 52.1 18.3 402 51.8 17.7 -0.318 F=0.10 Head years of formal education 2247 0.619 2.2 402 0.575 2.4 -0.0444 F=0.14 Head gender (1=female) 2253 0.218 402 0.239 2.04 X2=0.71 Risk aversion 2254 403 X2=5.53* ... high 296 13.1% 36 8.9% ... moderate 933 41.4% 176 43.7% ... low 1025 45.5% 191 47.4% Settlement status 2254 403 X2=16.53*** ... fully settled 1126 50% 231 57.3% ... nomadic 88 3.9% 2 0.5% ... partially settled 1040 46.1% 170 42.2% Household members (n) 2254 7.03 2.75 403 6.69 2.27 -0.343 F=5.58** Livestock owned (TLU) 2238 17.5 27.8 400 20.7 35.7 3.25 F=4.21** Annual household income (ETB) 2254 5691 20634 403 4969 9446 -722 F=0.48 Household cash savings (ETB) 2252 1612 12803 403 2380 14219 768 F=1.19 Share of income from livestock (% of income) 2254 0.635 0.439 403 0.616 0.456 -0.0194 F=0.66 Social groups (n) 2254 0.829 0.864 403 0.923 0.89 0.0939 F=4.00** Rangeland expectation 2252 1.75 1.05 403 1.57 0.945 -0.184 F=10.78*** Livestock lost to drought since last sales window (TLU) 2254 0.184 1.07 403 0.211 1.28 0.0279 F=0.22 * p<0.1; ** p<0.05; *** p<0.01 When comparing households which purchased IBLI at least once over the course of the panel to those which never acquired coverage (shown in Supplementary Table 5), results regarding livestock ownership, household members, and social group membership are consistent with those above. However, results indicate that female-headed households are significantly more 48 likely to have purchased IBLI at least once (p<0.001), whereas the head’s median age is higher among those who never acquired coverage. Where households did not purchase IBLI, they were asked to cite the primary reason for not acquiring coverage. These responses are summarised in Table 11 below, which shows that a lack of sufficient financial resources41 was the main barrier in a quarter of instances. A fifth of respondents each mentioned insufficient understanding of the product or not having had an opportunity to purchase it, though the frequency of the former decreased over the study period. Owning an insufficient number of livestock was mentioned as the main reason for not purchasing IBLI in 17% of instances, though it is unclear why this is the case given that there was no lower limit on the number of livestock that could be insured. Table 11: The frequency of primary reasons cited for not purchasing an IBLI policy across different rounds (Pooled refers to the pooling of responses from rounds 2-4). Round 1 is not included as the first IBLI sales took place following round 1. Pooled Round 2 Round 3 Round 4 Responses (n) 2007 620 683 704 Do not have money 24% 27% 26% 20% Did not have an opportunity to buy it 21% 21% 21% 21% Did not understand insurance enough 20% 25% 19% 17% Do not have enough animals 17% 16% 19% 17% Afraid of uncertainty 6% 2% 7% 8% Waiting to see what happens to others 5% 5% 3% 7% Don't think insurance will help me 4% 2% 3% 6% Don't trust companies 2% 1% 2% 3% Can rely on others 1% 1% 1% 1% 41 Because the response logged in the survey was “Do not have money”, it is not possible to determine with certainty whether responses refer to a lack of liquidity – i.e. households actually lacking cash to pay for the insurance – or simply the inability to afford to purchase coverage. 49 4.2.2. Econometric results Column 1 of Table 12 below presents the full regression results for Model (2). Receiving a discount coupon is shown to increase the probability of purchasing an IBLI policy in a given window by 10.6 percentage points on average. Of all the control variables included in the regression, only two were statistically significant; nomadic households were found to be 10% less likely than fully settled households to acquire coverage, with the likelihood of purchase falling as households’ income rose. The second column of this table presents regression estimates when discount coupons’ effect is interacted with households’ livestock holdings. The coefficients associated with the number of livestock owned as well as its interaction with discount coupons’ effect are statistically insignificant, suggesting that herd size does not influence the probability of acquiring coverage or the impact of insurance subsidies. Plotting the association between households’ livestock holdings and the marginal effect of receiving a discount coupon does, however, yield a U- shaped association (Figure 7). Figure 7: Variation in the marginal effect of receiving a discount coupon with the number of livestock owned (measured in TLU). The blue line is fitted onto plotted points using locally estimated scatterplot smoothing (LOESS). Results are based on the model presented in column 2 of Table 12. 50 Table 12: The estimated coefficients and average marginal effects of Model 2 (in column 1), and Model 2 including an interaction term between the number of livestock owned and the receipt of a discount (column 2). Estimation was carried out using a variant of the probit pooled estimator developed by Wooldridge (2010). Standard errors clustered at the household level are shown in parentheses. For details on control variables, see Supplementary Table 3. (1) (2) Coef estimates AME Coef estimates AME Intercept -2.53** (0.836) -2.477** (0.825) Received discount = Yes 0.721*** (0.128) 0.106*** (0.014) 0.668*** (0.2) 0.107*** (0.014) TLU 0.002 (0.011) 0.001 (0.002) 0.009 (0.017) 0.001 (0.002) TLU² 0 (0) 0 (0) Received discount * TLU -0.009 (0.015) Received discount * TLU² 0 (0) Head gender = Female 0.389 (0.26) 0.03 (0.018) 0.149° (0.085) 0.031° (0.018) Head age -0.001 (0.002) 0 (0) -0.001 (0.002) 0 (0) Risk aversion=Low 0.052 (0.108) 0.01 (0.02) 0.047 (0.107) 0.009 (0.02) Risk aversion=Mod 0.125 (0.106) 0.025 (0.02) 0.124 (0.106) 0.024 (0.02) Skit tape 0.061 (0.151) 0.013 (0.032) 0.048 (0.15) 0.01 (0.031) Cartoon -0.047 (0.139) -0.009 (0.027) -0.045 (0.139) -0.009 (0.027) Income † 0.047 (0.045) -0.013* (0.005) 0.047 (0.046) -0.013* (0.005) Income² † -0.008° (0.004) -0.008° (0.004) Savings † 0.004 (0.013) 0.001 (0.003) 0.003 (0.013) 0.001 (0.003) Ratio of income from livestock 0.097 (0.13) 0.02 (0.026) 0.096 (0.13) 0.019 (0.026) Expected rangeland conditions 0.002 (0.054) 0 (0.011) 0.002 (0.053) 0 (0.011) TLU lost to drought 0.008 (0.031) 0.002 (0.006) 0.016 (0.037) 0.003 (0.007) Household members †† -0.001 (0.015) 0 (0.003) -0.001 (0.015) 0 (0.003) Social groups 0.003 (0.064) 0.001 (0.013) -0.001 (0.065) 0 (0.013) Status=Nomadic -0.67° (0.342) -0.1** (0.035) -0.671* (0.34) -0.1** (0.035) Status=Partially settled -0.025 (0.074) -0.005 (0.015) -0.027 (0.074) -0.006 (0.015) Within-household means of time- variant variables Yes Yes Yes Yes Sales round dummies Yes Yes Yes Yes Study site dummies Yes Yes Yes Yes n 2626 2626 2626 2626 Prob > Chi2 <0.0001 <0.0001 <0.0001 <0.0001 McFadden P.R2 0.147 0.147 0.149 0.149 *** p < 0.001, ** p < 0.01; * p < 0.05; °p < 0.1 † Variables are IHC transformed †† Variables is lagged one round 51 Three further regressions were constructed to evaluate the significance of the subsidy rate offered, the results of which are shown in Table 13. The average marginal effect of the discount rate received is +2.05x10-3, equating to a 2% increase in the uptake probability with each 10 percentage point increment in the discount rate received (column 1). When incorporating both a binary variable for the receipt of a coupon as well as a continuous variable for the discount rate of coupons received (column 2), results indicate that receipt of the coupon itself – i.e. over and above any price reduction effect – increases the likelihood of purchasing IBLI in a given window by 4.4%, though this is only significant at the 90% level. The impact of the discount rate remains highly significant, its effect falling slightly compared to the previous model (+1.70x10-3). These results jointly indicate that while the discount coupon itself may serve to increase uptake, this effect is weakly significant (p<0.9) and dwarfed by coupons’ price-reduction effect. Table 13: Average marginal effects of discount coupons on IBLI uptake. Column 1 includes a continuous variable capturing the discount rate and does not include a binary variable for the receipt of a coupon, whereas Column 2 incorporates both. In column 3, the discount coupon’s effect is modelled as a series of dummy variables. Standard errors are clustered at the household level. See Supplementary Table 3 regarding control variables. (1) (2) (3) Received discount = Yes 0.044° (0.023) Discount Rate (% of premium price discounted) 0.002*** (0) 0.002 *** (0) Discount Rate = 10% 0.032 (0.02) Discount Rate = 20% 0.082 *** (0.022) Discount Rate = 30% 0.074 ** (0.024) Discount Rate = 40% 0.110 *** (0.023) Discount Rate = 50% 0.086 *** (0.024) Discount Rate = 60% 0.109 *** (0.026) Discount Rate = 70% 0.141 *** (0.025) Discount Rate = 80% 0.197 *** (0.026) Time-variant characteristics controls (𝑿𝒊𝒕) Yes Yes Yes Time-invariant characteristics controls (𝑯𝒊) Yes Yes Yes Time dummies (𝒕) Yes Yes Yes Study site dummies (𝝎) Yes Yes Yes Within-household means of time-variant variables (𝑿ഥ𝒊) Yes Yes Yes n 2626 2626 2626 Prob > Chi2 <0.0001 <0.0001 <0.0001 *** p < 0.001, ** p < 0.01; * p < 0.05; °p < 0.1 52 Treating the different discount rates as a discrete set of dummy variables (column 3) shows that the effect of receiving a 10% discount coupon (AME=+0.03) is not statistically significant, whereas a 20% discount increases the likelihood of adoption by 8.2%, this being statistically significant. As shown in Figure 8, further increases in the discount rate up to 60% only had a marginal effect; 40% and 60% discounts both had an average marginal effect of approximately 0.11. The impact of coupons increases markedly thereafter, with 70% and 80% discounts heightening the uptake likelihood by 14% and 20%, respectively. Figure 8: Visual representation of the average marginal effect of receiving insurance discount coupons of different rates on the likelihood of purchasing IBLI, with whiskers showing upper and lower confidence intervals. The blue line is fitted onto plotted points using locally estimated scatterplot smoothing (LOESS) to approximate the shape of the association between discount rate and uptake. Table 14 show regression results when incorporating the effect of previous period discounts. The receipt of a coupon in either of the previous two windows and the discount rates of those coupons had no significant influence on households’ likelihood of purchasing IBLI in the current period (columns 1 and 2). However, previous periods’ discount rates do appear to 53 reduce the effect of receiving a coupon in the current period, with the influence of the prior period’s discount rate being more pronounced (-2.017 x10-2) and significant (p=0.019) than that lagged two windows (AME=-1.696 x10-2, p=0.052). A clear downward trend is also evident when mapping the association between the marginal effect of receiving a coupon in the current period and the discount rate lagged 1 and 2 windows42 (Figure 9). Table 14: Coefficient estimates and average marginal effects of present and previous period discounts. Column 1 includes the estimated effects of receiving a discount in the current period, and the receipt of discounts in the previous two sales windows. Column 2 includes the estimated effects of the receipt of a coupon in the current period, and the discount rate in the previous two sales windows, with column 3 incorporating an interaction between these two sets of variables. Standard errors clustered at the household level. It was not possible to interact the receipt of a subsidy in the current period with receipt in previous windows due to separation issues. Control variables are detailed in Supplementary Table 3. The number of instances varies across models due to the removal of extremely high influence points. (1) (2) (3) Coef estimates AME Coef estimates AME Coef estimates AME Received discount= Y 1.064*** 0.106*** 1.055*** 0.105*** 2.744*** 0.104*** (0.235) (0.013) (0.237) (0.013) (0.799) (0.012) Received discount lag 1 = Y 0.022 0.003 (0.121) (0.017) Received discount lag 2 = Y 0.000 0.000 (0.121) (0.017) Discount rate lag 1 0.001 0.000 0.020* 0.000 (0.002) (0.000) (0.008) (0.000) Discount rate lag 2 -0.002 -0.000 0.014 -0.000 (0.002) (0.000) (0.009) (0.000) Received discount x discount rate lag 1 -0.020* (0.009) Received discount x discount rate lag 2 -0.017° (0.009) Time-variant characteristics controls (𝑿𝒊𝒕) Yes Yes Yes Yes Yes Yes Time-invariant characteristics controls (𝑯𝒊) Yes Yes Yes Yes Yes Yes Time dummies (𝒕) Yes Yes Yes Yes Yes Yes Study site dummies (𝝎) Yes Yes Yes Yes Yes Yes Within-household means of time-variant variables (𝑿ഥ𝒊) Yes Yes Yes Yes Yes Yes n 1733 1733 1731 1731 1731 1731 Prob > Chi2 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 *** p < 0.001; ** p < 0.01; * p < 0.05; °p < 0.1 42 As shown in Supplementary Figure 2, this pattern is consistent across present-period discount coupon rates. Balance tests (not shown in the interests of space but available via R script) also shows that the present period subsidy rate received is randomly distributed and does not vary according to previous subsidies. These demonstrate that this result is in fact driven by an interaction between previous and present subsidies. 54 Figure 9: Visual representation of the average marginal effect of receiving an insurance discount coupon against the discount rate of coupon received in the prior sales window (left) and that receiving two sales windows prior (right). The blue line is fitted onto plotted points using locally estimated scatterplot smoothing (LOESS). The solid red line shows the estimated average marginal effect of receiving a coupon based on the results of column 2 of Table 14, i.e. when this is not interacted with previous period subsidies. 4.2.2.1. Robustness checks Given that participants cited instances where IBLI purchase decisions were made separately within a household, regressions were rerun excluding those households which at any point purchased an IBLI policy in the name of a household member other than the household head (n=14). The size and significance of estimated coefficients and marginal effects were unchanged. Results were also unchanged when dropouts were excluded and when a simple pooled probit estimator is used. 55 5. Discussion 5.1. Intended outcomes and coordination Heightening insurance uptake was, as expected, the intended outcome of IBLI subsidies most commonly cited by KII participants, with many of the more specific aims mentioned aligning with those identified in the literature (Table 3). The pursuit of inclusion, social protection, and political goals via subsidies were common to both the qualitative data here gathered and the work of Hill et al. (2014), Hazell and Varangis (2020), Hazell, Jaeger and Hausberger (2021), and Kramer et al. (2022), with interviewees’ references to subsidies as a means of supporting commercial sustainability, creating awareness, and showing stakeholders’ commitment to insurance products all aligning with the aim of overcoming initial hurdles commercial viability discussed in the literature. The significant divergences in stakeholders’ perspectives observed by Johnson et al. (2019) were, however, not evident during KIIs. As shown in Table 7, the vast majority of intended outcomes cited were shared across multiple stakeholder groupings, with the most salient goals also being recognised by FGD participants. Where differences across groupings were present, these were not particularly stark. For instance, interviewees from insurance companies did not explicitly refer to inclusion or risk reduction, but two spoke at length regarding subsidies’ contribution to “people on the absolute poverty line, all those who are…food insecure” (Ins3). It is, however, worth noting that divergent understandings of terms may have somewhat exaggerated the appearance of consensus. One case of this, evocative of Mosse's (2004) notion of translation (see pg.16), was NGOs’ and insurance companies’ contrasting framing of geographical dissemination as an aim of subsidies, with the former construing this as a pathway to equitable access to IBLI and the latter as a means of mitigating companies’ risk. This broad agreement in intended outcomes suggests that the “coordination gap” presently hampering IBLI subsidy programmes does not stem primarily from gross divergences in values or goals among actors, but rather may be driven by a lack of structured avenues for communication and collaboration. Interviewee’s calls for multi-stakeholder fora and policies 56 to delineate key actors’ roles – as also advanced by Hazell, Jaeger and Hausberger (2021, p.37) – thus appear well-suited to address this pertinent challenge. Another finding of note from the qualitative analysis conducted is that subsidies’ role in supporting IBLI’s commercial viability was cited more frequently than their role in advancing inclusion. This unexpected result underscores that challenges to IBLI’s commercial viability – treated at length by interviewees from insurance companies and in the literature (Keno, Diriba and Lemesa, 2018; Johnson et al., 2019; Lung, 2021; Kramer et al., 2022) – continue to persist. 5.2. Subsidy rates This study simultaneously corroborates and extends the body of evidence on subsidy rates’ influence. The positive association between the subsidy rate and uptake probabi