SOCIAL PROTECTION & JOBS DISCUSSION PAPER No. 2106 | JUNE 2021 Social Assistance Programs and Household Welfare in Eswatini Dhushyanth Raju and Stephen D. Younger © 2021 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: +1 (202) 473 1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. RIGHTS AND PERMISSIONS The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: +1 (202) 522 2625; e-mail: pubrights@worldbank.org. Abstract retro geometric background: © iStock.com/marigold_88 Project 41595 Social Assistance Programs and Household Welfare in Eswatini Dhushyanth Raju Stephen D. Younger Eswatini has notably high levels of poverty and inequality. Recurrent, negative shocks are an important contributing factor. This study assesses the performance of the largest social assistance programs in Eswatini, based on 2016/17 national household survey data. It examines the coverage rates of these programs, and their incidence and effectiveness in reducing poverty and inequality. The study also examines the association between program participation and negative shocks reported by households, in particular, drought and food price shocks associated with the 2015/16 El Niño event. Across programs, benefits are concentrated among poor households. However, the performance of programs in reducing poverty and inequality tends to be limited because of low intended or actual benefit levels and shortfalls in intended or actual coverage of the poor. Households that receive program benefits are more likely to report a drought shock. Except in the case of emergency food aid, which is provided ex post, we interpret this pattern to indicate that programs tend to provide ex-ante coverage to those vulnerable to this shock. At a minimum, enhancing the performance of programs in addressing poverty, inequality, and the adverse effects of shocks would require that actual benefit levels equal intended levels (for example, by procuring sufficient food commodities to meet the needs of the school feeding program) and that intended benefit levels are fully aligned with program aims (for example, by providing grant amounts to schools that are large enough to allow for tuition-free government secondary education for orphaned and vulnerable children). Absent greater budgetary allocations to programs, addressing these benefit-related disconnects may require improving the targeting of select program benefits to poorer households such as by using a proxy means test. We simulate the effects of programs on poverty and inequality reduction from such hypothetical reforms. JEL Codes: H53, H55, I3, O12, O55 Key words: coverage, incidence, effectiveness, social assistance, safety nets, shocks, microsimulation, Eswatini Dhushyanth Raju, World Bank, draju2@worldbank.org; Stephen D. Younger, Commitment to Equity Institute, sdy1@cornell.edu. We thank Paolo Belli, Shanta Devarajan, Stephen Devereux, Ugo Gentilini, Jeffrey Hammer, Marie Françoise Marie-Nelly, Thulani Matsebula, Victoria Monchuk, Aleksandra Posarac, and participants at the May 2021 Eswatini DPMO–UNDP Workshop for Enhanced Social Assistance in the Kingdom of Eswatini for comments and suggestions. We are grateful to (last names in alphabetical order) Alice Alunga, Cissy Byenkya, Moses Dlamini, Russell Dlamini, Sindisiwe Dube, Thobile Lefty Gamedze, Sandile Ginindza, Nomagugu Khumalo, Marko Kwaramba, Indira Bongisa Lekezwa, Khanya Mabuza, Victor Mahlalela, Nozipho Mikhatshwa, Daison Ngirazi, Elizabeth Ninan, Gugu Nxumalo, Afshin Parsi, Precious Zikhali, and Nonhlahla Zindela for facilitation and assistance with information for this study. We are also grateful to Luxi Liu for research assistance. The study was prepared under the World Bank’s advisory services and analytics task Review of the Public Social Protection System in Southern Africa (P172175). I. Introduction Eswatini has notably high levels of poverty and inequality compared with other lower-middle- income countries as well as developing countries more generally. In 2016/17, the overall poverty rate in Eswatini was 58.9 percent, based on the country’s upper poverty line (or 28.3 percent, based on the international poverty line of $1.90 in 2011 purchasing power parity U.S. dollars). The country’s Gini index, a measure of inequality, was 49.3 percent. Compared to other countries worldwide, Eswatini is assessed to be at a “medium and stable” risk of humanitarian crises and natural or human-caused disasters, reflecting primarily its population’s relatively high socioeconomic vulnerability and the country’s weak institutional and infrastructural capacity to respond to negative shocks (Inter Agency Standing Committee and the European Commission 2020). 1 These dynamics further challenge the government’s plans and actions to tackle structural poverty and inequality. As part of its efforts to address poverty and vulnerability and the adverse consequences of these circumstances on households, the Eswatini government administers a mix of social insurance programs, social assistance programs, and social welfare services. The HIV/AIDS epidemic and its negative effects on household welfare is the raison d’être for several social assistance programs and social welfare services; measured in 2016/17, Eswatini had the highest HIV prevalence in the world, with 27 percent of those age 15 and older living with HIV (GOE 2019c). Primary administrative responsibility for many of the main government social assistance programs lies with the Department of Social Welfare (DSW), under the Deputy Prime Minister’s Office (DPMO). International donors have supported some of the government’s social assistance programs and social welfare services through official development aid and supplemented these programs and services with direct humanitarian aid. This study examines the performance of Eswatini’s main social assistance programs. 2 These programs employ varying combinations of categorical targeting and self-targeting to reach the poor and vulnerable. Specifically, the study asks two primary questions. First, how successful are the main social assistance programs in reducing poverty and inequality? We answer this question through standard benefit incidence analysis (Bourguignon and Pereira da Silva 2003; Lustig 2018; Yemtsov et al. 2018). To draw further insights for informing public policy, we also ask an ancillary, counterfactual question: Could the social assistance programs achieve greater reductions in poverty and inequality if benefits were targeted to poor households through a proxy means test (PMT) formula which we construct for the hypothetical exercise? We answer the question by conducting microsimulations (Figari, Paulus, and Sutherland 2015). The second primary question the study poses is: Do the main social assistance programs tend to cover households that experience adverse shocks? This coverage may be an ex-ante result obtained 1 Eswatini has a value of 3.9 in the risk index constructed by the Inter Agency Standing Committee and the European Union, compared to an average value of 3.8 (standard deviation = 1.9) across 191 countries (the index ranges from 0 to 10). This value ranks Eswatini 87 th in the index. 2 An analysis of the performance of social insurance programs would necessarily have to include an actuarial assessment. Hence, the performance of these programs is not examined in this study. 1 through program eligibility criteria not directly linked to shock exposure, or it may be an ex-post result if program eligibility criteria, the timing, or the level of program benefits are adjusted in response to the shock (referred to as adaptive social protection or shock-responsive social protection). We answer this question by examining the types of adverse shocks reported by households; the association between commonly reported shocks and poverty status, food insecurity status, area of residence, and the receipt of social assistance program benefits; and the types of coping responses reported by households. The analysis is primarily based on data from the Swaziland Household Income and Expenditure Survey (SHIES) 2016/17, which is the latest available national household sample survey with relevant data. The SHIES contains dedicated modules on social protection interventions and shocks. Measures of poverty and inequality are based on household consumption data in the survey. The Eswatini government administers and uses the SHIES for its official estimates of poverty and inequality in the country. The reference period for the data on social protection interventions is the year preceding the survey. Of particular relevance for the analysis of shocks and social assistance programs, the survey data span the El Niño-induced drought of 2015/16, which had large, widespread negative effects on crop and livestock production, and on food security. We examine the five social assistance programs with the largest numbers of beneficiaries reported in the SHIES 2016/17: (a) neighborhood care points (NCPs), which are community-run centers that aim to provide free cooked meals (and basic early child care and development services) to needy, young orphaned and vulnerable children (OVC); 3 (b) the government’s school feeding program, through which the government subsidizes cooked lunches for students in government primary, secondary, and high schools; (c) OVC education grants, through which the government subsidizes tuition and exam fees for orphaned and vulnerable children attending government secondary and high schools; 4 (d) emergency food aid , through which the government distributes food commodities to households facing acute food insecurity; and (e) elderly grants, through which the government provides regular cash benefits to those aged 60 and above. NCPs are fully financed through international aid, with the Eswatini government involved in a coordinating capacity. Emergency food aid is financed by the government, often supplemented by 3 The Eswatini government defines a vulnerable child as “a child whose rights, for example, to personal safety, adequate nutrition, health care and schooling, [are] at risk, and whose prospects for healthy growth and development are consequently diminished” (GOS 2010). 4 Government primary schools are free with respect to school fees for all students. 2 international aid. The other three programs are fully financed by the government. While there are other government social assistance programs, most under DSW, they each cover less than 1 percent of the national population. In our analysis, we examine the performance of each individual program, as well as the performance of all five programs together and of the four government-financed programs together, excluding NCPs. The exclusion of NCPs makes little difference to the overall results as the outlay and coverage of NCPs are small. While we examine the important relationship between the social assistance programs and poverty, inequality, and exposure to shocks, we recognize that the primary aims of these programs may be different. For example, the primary aim of NCPs, school feeding, and emergency food aid is presumably to reduce hunger and undernutrition, while the main goal of OVC education grants is presumably to promote increased education attainment through enrollment in secondary schools, among the poor, vulnerable, and otherwise disadvantaged. We are unable to credibly examine the relationship between the social assistance programs and education, health, nutrition, and labor outcomes, among others, due to potential selection bias. Using the SHIES 2016/17, which provides cross-sectional observational data, estimates of program effects on these other outcomes would be biased due to systematic, unobservable differences in characteristics between social assistance program beneficiaries and nonbeneficiaries which are correlated with the outcome of interest. Indeed, the estimates can be negative simply because the programs are more likely to reach the poor and vulnerable than other individuals (which we show is the case in this study) and also because these groups have worse outcomes on average than the population at large. Nevertheless, the preponderance of rigorous evidence, such as from field experiments, suggests that variously designed social assistance programs have appreciable, positive effects on human development outcomes (Hanna and Karlan 2017). 5 Given that household income is a primary determinant of these other outcomes, one basic but key pathway behind the effects of social assistance programs is the added income from cash and in-kind transfers to households. Interestingly, apart from the basic income-effect pathway, emerging evidence suggests that a reduction in poverty, such as through the added income from transfers, can produce physiological and psychological benefits in individuals (Mullainathan and Shafir 2013; Haushofer and Fehr 2014; Haushofer and Shapiro 2016). These benefits can improve economic decision making and enhance health and learning directly, or they can indirectly improve the effectiveness of other investments and interventions aimed at improving health, nutrition, learning, and education attainment. Further, of particular relevance to Eswatini given its high HIV prevalence, international evidence suggests that social assistance programs can discourage undesirable sexual behavior that may lead to HIV infection (de Walque 2020). This is in addition to social assistance programs potentially 5 What’s more, this evidence is more consistently positive than the collective rigorous evidence on the effects of education, health, and labor market interventions on education, health, and labor outcomes, respectively (Glewwe and Muralidharan 2016; Dupas and Miguel 2017; Fryer 2017; Muralidharan 2017). 3 influencing human development outcomes in communities and households adversely impacted by HIV/AIDS. The Eswatini government spent between E458 million and E687 million annually on social assistance programs between 2014/15 and 2017/18. Over the reference period for the SHIES 2016/17, the government spent the most on elderly grants followed by OVC education grants and school feeding. Elderly grants accounted for about one-third of total government social assistance program spending. Its position as the program with the largest outlay has been further strengthened following large increases in grant benefit values in 2017 and 2020. OVC education grants and school feeding each accounted for about one-quarter of total government social assistance program spending. Government spending on emergency food aid tended to be small (less than 5 percent of total government spending on social assistance programs). Likewise, government spending on other social assistance programs under DSW was also small (less than 5 percent of total government spending on social assistance programs). In recent years, annual government spending on social assistance programs has accounted for 2.5 to 3.5 percent of total annual government spending, and has been roughly equivalent to 1 percent of annual GDP. This level of government spending on social assistance programs as a share of GDP is lower than the corresponding average levels in Sub-Saharan Africa and developing countries in general. In each of these groups of countries, annual spending on social assistance programs averages roughly 1.6 percent of annual GDP. 6 The level of spending by Eswatini is also lower than those of its neighbors, specifically other countries in the Southern African Customs Union (SACU), where, for example, recent annual spending on social assistance programs relative to annual GDP has been more than 3 percent in Lesotho and South Africa (Boko, Raju, and Younger 2021; Oosthuizen 2020). 7 Coverage by the five main social assistance programs is reasonably extensive: 41.3 percent of individuals and 52.0 percent of households are covered by at least one program. School feeding and elderly grants cover large shares of their presumed eligible populations (79.3 percent of government primary, secondary, and high school students and 81.1 percent of those aged 60 years or older, respectively), but smaller shares of the general population (22.8 percent and 5.8 percent, respectively). Emergency food aid by government covers 16.3 percent of individuals. NCPs and OVC education grants have the lowest coverage rates, with 4.0 percent each. All of the programs are pro-poor and pro-food insecure. Taking all five programs together, 54.6 percent of the extreme poor and 50.8 percent of the moderate poor are covered. In contrast, 25.9 of the nonpoor are covered. Likewise, taking all five programs together, 48.7 percent of those that are severely food insecure are covered, while 25.7 percent of those that are food secure are covered. The pro-poor nature of the program is further reflected by program concentration coefficients, which vary between –0.255 and –0.152 across programs. These coefficients indicate that the 6 Averages are own estimates based on data obtained from the World Bank’s Atlas of Social Protection Indicators of Resilience and Equity (ASPIRE) database at http://datatopics.worldbank.org/aspire/home. 7 SACU is a customs union among five countries in Southern Africa: Botswana, Eswatini, Lesotho, Namibia, and South Africa. Member countries pool their customs and excise revenues and distribute them according to a revenue- sharing formula. 4 programs are progressive, but less so than the most progressive social assistance programs in other developing countries. For a point of reference, the Commitment to Equity (CEQ) Institute maintains a database of distributional statistics for transfers from many developing countries.8 The average concentration coefficient value for direct transfers of any type among 32 countries is – 0.30, and the lowest value is –0.63. Despite the pro-poor nature of the programs, their marginal effects on poverty and inequality are small: None reduces either the Gini index or the poverty rate by more than one percentage point. The five programs together reduce the Gini index by 2.0 percentage points (from a baseline level of 49.3 percent), the overall poverty rate by 1.5 percentage points (from a baseline rate of 58.9 percent), and the overall poverty gap by 3.0 percentage points (from a baseline gap of 24.9 percent). 9 While small in absolute terms, the marginal effects compare favorably to recent changes in inequality and poverty in Eswatini. The marginal effect of the five programs together on inequality is roughly equivalent to the decline in inequality in Eswatini between 2000/01 and 2016/17. And the marginal effects of the five programs together on the overall poverty rate and poverty gap are equivalent to 17 percent and 54 percent of the decline in the overall poverty rate and poverty gap, respectively, between 2009/10 and 2016/17. Each of the five programs are quite effective at reducing the poverty gap: They are between 73 and 81 percent as effective as transfers of equivalent outlays that are perfectly targeted to the poor would be. But effectiveness at reducing poverty severity is much less for each program: They are between 48 and 54 percent as effective as transfers of equivalent outlays that are perfectly targeted to the poor would be. 10 This indicates that while the programs perform well at reaching the poor, they perform much worse in reaching the poorest of the poor. The programs are also less effective in reducing inequality than the poverty gap. Given sizeable, recurrent fiscal deficits in the recent past, the Eswatini government is constrained in its ability to increase its spending in general. The government aims to undertake a fiscal consolidation, although the coronavirus disease 2019 (COVID-19) crisis has derailed this plan. Given the expected tight fiscal position over the medium term, we conduct simulations to examine whether the performance of social assistance programs can be improved by better targeting them to the poor, and by reallocating savings from the targeting toward larger program benefits for existing beneficiaries. We conduct this analysis for OVC education grants and elderly grants, estimating a PMT model for Eswatini based on the SHIES 2016/17 and splitting the population into the “PMT-poor” and “PMT-nonpoor.” PMT-based targeting is employed here simply as a data-driven approach (using the SHIES 2016/17) to identify the poor for the simulation exercise. Whether PMT-based targeting (in combination or as an alternative to other targeting methods) should actually be implemented in the field when administering programs should be evaluated based on various criteria. These criteria should include targeting accuracy, but also comprise ease and efficiency of implementation, legitimacy, and transparency. 8 CEQ-estimated statistics are available at http://commitmentoequity.org/datacenter. 9 The poverty gap is a measure of the average shortfall in consumption from the poverty line. It reflects the depth of poverty. 10 Poverty severity is the square of the poverty gap. It assigns greater weight to the poorest. 5 Targeting OVC education grants and elderly grants to the PMT-poor would produce significant savings in spending (totaling about E30 million for OVC education grants and about E60 million for elderly grants), but a large percentage (30 percent or more) of existing beneficiaries would lose their benefits. Simply eliminating benefits from these programs for the PMT-nonpoor would lead to slight increases in poverty and inequality because of errors of inclusion and exclusion inherent in PMT-based targeting. However, reallocating savings from such targeting toward higher benefits for PMT-poor beneficiaries would contribute to meaningful declines in poverty and inequality. With respect to shocks, on average, households report a little less than one negative shock in the year preceding the SHIES 2016/17 and a little more than two in the five years preceding the survey. “Drought or floods” and “large rise in price of food” are the dominant shocks, each representing one-third of the shocks reported in the year preceding the survey. These large shares are presumably driven by the El Niño-induced drought of 2015/16. The vast majority of households report receiving no assistance from others. Only 4 percent report receiving assistance from government. Given this, it appears that government and other institutions are not responding to shocks in ways that make their interventions the main coping strategies for an appreciable share of affected households Nevertheless, reporting a drought shock is positively associated with the receipt of benefits from each of the five programs. The association between this shock and emergency food aid is presumably a result of an ex-post response, while the associations for the other programs are presumably an incidental result of ex-ante coverage. Given our data, we cannot credibly estimate the extent to which the receipt of benefits from the social assistance programs moderate the negative welfare effects of shocks. While our analysis is based on standard benefit incidence and microsimulation methods, these methods have well-known analytical limitations (van de Walle 1998; Lanjouw and Ravallion 1999; Ravallion and Chen 2015). We estimate average benefit incidence, not benefit incidence at the margin. The latter may be more relevant when simulating the effects of reforms that adjust the extent of program coverage. In performing our simulations of reforms to social assistance programs, we generally ignore potential behavioral responses in household labor supply, income generation, or consumption. We also ignore potential general equilibrium, lifecycle, or intergenerational effects as well as political economy responses in our simulations. Further, in terms of survey data quality, the correspondence between SHIES 2016/17 estimates and administrative information on beneficiary numbers and outlay is reasonably good for all programs except for emergency food aid. The shortfalls in beneficiaries and outlay for this program in the survey data are substantial when compared to the administrative information. Consequently, estimates of coverage, incidence, and effectiveness for emergency food aid likely have a large downward bias. However, given that the program is small, these underestimates would have little bearing on the overall results. While in line with the state of the art, our analysis of program performance based on the SHIES 2016/17 is correlational and partial. Critical knowledge gaps remain. To address them, future 6 research should include rigorous evaluations that seek to assess the potential impacts of the social assistance programs on welfare, including the potential welfare-protective impacts of the programs in the face of shocks. Causal impacts can be uncovered through creative empirical strategies applied to available observational data. However, they will likely require the collection of new data following appropriate research designs for causal inference. As noted earlier, the social assistance programs are explicitly or implicitly designed to influence: (a) specific outcomes beyond household consumption and food security (for example, individual education, health, and nutrition outcomes); (b) the behaviors and outcomes of specific subpopulations (for example, the poor or vulnerable, and young or adolescent children); and (c) behaviors and outcomes in the near term as well as over a longer timeframe. Given this, future evaluations should seek to examine program impacts in a comprehensive manner. Future research should also include program operational reviews that aim to assess shortcomings in design parameters and design-implementation gaps, covering the foundational elements and links in the entire program delivery chain (Leite et al. 2017; Lindert et al. 2020). The results from the program performance analysis, coupled with information on program design and implementation discussed in the study, suggest that program intensity (rather than scale or reach) is a primary issue. This issue impairs the performance of programs in addressing inequality, poverty, and the negative effects of shocks. In some cases, actual benefit levels are lower than intended benefit levels. For example, the amount of food procured by the government for the school feeding program is inadequate. Similarly, shortfalls in food procured and distributed to NCPs disrupt their operations. In other cases, intended benefit levels are inconsistent with the aims of the program. For example, the OVC education grant amount provided by the government may not fully defray the tuition costs of the beneficiary. Likewise, the assigned portion of the per- student grant under the government’s Free Primary Education initiative may not fully defray the operational costs of school feeding in government primary schools. Public policy should seek to correct these disconnects. Absent additional financing for the programs, correcting these disconnects may require (further) efforts to target the programs to the neediest. Greater program intensity may need to be traded off against reduced program coverage. Notwithstanding, the option of increased financing of social assistance programs merits serious consideration. Compared to many other developing countries, Eswatini’s levels of poverty and inequality are much higher, while its public spending on social assistance programs is markedly less. The government’s fiscal policy is viewed to have a limited bearing on the country’s poverty and inequality challenges (IMF 2020). Of course, the government’s overall spending is not singularly aimed at reducing current consumption poverty and inequality in the most direct way possible (such as through cash or in-kind transfers to poorer households); the government spends to achieve a host of other goals. As a first level of evaluation, the government should weigh the aims of different lines of spending, and any potential trade-offs that would arise from adjusting the composition of overall spending. As a second level of evaluation, the government should assess the effectiveness and efficiency of a given line of spending vis-à-vis the goal that the spending seeks to achieve. 7 With respect to both these levels of evaluation, there appears to be reason for concern. The Eswatini government allocates relatively high levels of its budget to compensation for public employees and transfers to extrabudgetary entities and public enterprises, and suffers from weaknesses in its public financial management and procurement systems (IMF 2020). Effectively addressing these concerns is critical for improving the government’s overall fiscal position and creating the fiscal space for vital social and core capital spending. The administrative system of the government’s social assistance programs requires a major upgrade. The needed improvements span the entire delivery chain and relate to communication and outreach, targeting and enrollment, benefit transfer processes, case and grievance management, and monitoring. A top priority should be the development and implementation of a single integrated social registry and an integrated management information system for implementing and monitoring these programs. The government should also reconsider the mix of social assistance programs. They should seek to ensure that the aims and designs of their programs are consistent with an up-to-date profile of the main welfare risks (including adverse shocks) that are being experienced by the poor and other disadvantaged sociodemographic groups. A lifecycle framework for identifying the welfare risks faced by individuals is one approach that the government could adopt to ascertain the combination of social assistance programs. In determining the combination of social assistance programs, the government should also account for the whole landscape, current and expected, of social welfare services, social insurance programs, labor market programs, and other development interventions in Eswatini. In so doing, the government should aim to address gaps or weaknesses (i.e., uncovered or weakly covered adverse conditions or risks) through its social assistance programs and to forge stronger links and enhance synergies between its social assistance programs and other programs and services. One axis for rethinking the combination of social assistance programs is the balance between cash and noncash transfers. Presently, among the five main social assistance programs examined in the study, four offer food or other in-kind benefits. Only the elderly grants program offers cash benefits. Another axis for rethinking the combination of programs is the balance between nonlabor income transfers and support for labor income. Programs such as productive inclusion and labor- intensive public works fall under the latter type. 11 In view of the current state of Eswatini’s social assistance system, including the constraints faced by the system from various sources (political, financial, administrative), a gradual stage-by-stage approach toward advancing the development of the social assistance system may be the most practical. The first stage would be to address benefit-related disconnects in the existing social assistance programs. The second stage, staggered with the first, would be to strengthen the social assistance administrative system. The third stage would be to rethink the mix of social assistance programs. An integral part of advancing the development of the social assistance system would be to make it more resilient and more responsive to adverse shocks. 11 Productive inclusion programs are also referred to as economic inclusion or graduation programs (Andrews et al. 2021). Labor-intensive public works programs are also referred to as workfare. 8 The COVID-19 pandemic presents a substantial setback to the government’s efforts to address structural poverty and inequality through its social assistance programs and other development interventions. The International Monetary Fund estimated in April 2021 that Eswatini’s real GDP declined by 3.3 percent (IMF 2021). 12 To address the fallout from the pandemic, the government approved a supplementary budget, and provided food and cash assistance to needy and affected households, tax refunds to enterprises, and cash assistance to laid-off or unpaid employees of enterprises, among other actions. Eswatini's fiscal deficit increased in 2020/21, as government revenues from taxes and SACU receipts declined and expenditures increased. The economic recession is thought to have increased the extent and depth of poverty in the country in 2020.13 Such fallout from the pandemic underscores the case for having robust, flexible, and effective social protection programs. The research findings and public policy recommendations made in the study support such an imperative. The rest of the paper is organized as follows. To help contextualize the findings, section 2 presents background information on Eswatini. Section 3 describes the design and institutional and implementation arrangements of the social assistance programs selected for the performance analysis. Section 4 discusses results from the analysis of the performance of the social assistance programs. Section 5 concludes by discussing implications for future data and research initiatives and for future public policy. The paper also includes four appendices. These discuss the construction of key variables for the program performance analysis (appendix 1), analytical concepts for the analysis (appendix 2), supplemental results (appendix 3), and the design of a PMT model used for the program reform simulations (appendix 4). II. Background To help contextualize the findings from our analysis of the performance of the government’s main social assistance programs based on SHIES 2016/17 data, in this section, we present relevant background information on Eswatini, covering population; national income and growth; poverty and food insecurity; inequality; covariate shocks, in particular natural shocks; overall government revenue and expenditure; and the social protection system, and its main actors, legislative underpinnings, and policies and strategies. Population Enumerated in 2017, Eswatini has approximately 1.09 million individuals (Central Statistics Office 2017). Of this population, 97.9 percent reside in regular households and 2.1 percent reside in collective households, where a large group people share common domicile facilities (for example, dormitories). The homeless represent 0.03 percent of the total population. The average 12 In contrast, IMF projected in October 2019 that Eswatini’s real GDP would grow by 0.5 percent in 2020 (IMF 2019). 13 The discussion of projected developments and announced policy responses are based on inputs provided by World Bank staff. 9 size of a regular household is four members. About 7 percent of the population resides outside Eswatini; of this group, 93 percent reside in South Africa (World Bank 2020). Among the country's four regions, Hhohho and Manzini have higher shares of the population (29.3 percent and 32.6 percent, respectively) than Lubombo and Shiselweni (19.4 percent and 18.7 percent, respectively). Between 2007 and 2017, Hhohho and Manzini saw increases in their populations, while Lubombo and Shiselweni experienced little change. In terms of age, 35.6 percent of the national population are 0–14 years, 59.9 percent are 15–64 years, and 4.5 percent are 65 years or older (Central Statistics Office 2017). Economic growth Eswatini is classified as a lower-middle-income country by the World Bank. In 2018, the country’s GDP per capita in 2011 PPP U.S. dollars was US$9,439 (figure 1). Economic growth averaged 1.9 percent annually over the 2010s, but has been volatile. Eswatini’s strong economic integration with South Africa has helped spur economic progress but has also increased its exposure to external shocks emanating from South Africa. This exposure partly explains the volatility in Eswatini’s economic growth. Despite high government spending, average economic growth has been subdued because of depressed private investment and external competitiveness. The contribution of exports to growth has declined in recent years, driven by a loss in competitiveness of the economy. Likewise, the level of private investment has fallen in recent years in Eswatini. Factor decompositions of recent economic growth suggest that the contribution of physical capital to growth is small, and that the contribution of total factor productivity is negative. These patterns are linked to unfavorable business conditions (high costs of labor and other productive inputs), stemming from distortions in regulation and weaknesses in governance (IMF 2020; World Bank 2020). Consumption and income In 2016/17, the average value of monthly consumption by households was E4,242; the median value was E2,953. On average, food constituted the largest share of household consumption, at 33.4 percent. This is followed by housing (which includes imputed rent) and utilities, at 19.4 percent, and transport, at 13.4 percent. As can be expected, across households, food as a share of consumption declines with the overall level of consumption. Among households in the poorest consumption decile, the food share is about 60 percent, while it is less than 10 percent among households in the richest consumption decile. The main source of income for households is wage income, followed by private transfers (both from domestic and international sources) (World Bank 2020). On average, in 2016/17, wage income accounted for 54.2 percent of total income among households, while private transfers accounted for 20.7 percent. Public transfers (cash and in-kind) accounted for 10.1 percent of total income. Poorer households are more dependent on public and private transfers as a source of income, while richer households are much more dependent on wage income. Public transfers accounted for 16.9 10 percent of total income among the poorest 40 percent of households, compared to 5.5 percent for the richest 60 percent of households. Likewise, private transfers accounted for 29.1 percent of total income for the poorest 40 percent, while it accounted for 15.1 percent of total income for the richest 60 percent. Remittances make up a large share of private transfers. Much of the remittances comprise the transfer of wage income obtained by migrant household members. Given this, household income from private transfers can be treated as conceptually tantamount to household income from wage income. Poverty and food insecurity Eswatini has high levels of poverty compared to other countries at its income level. Measured by the international poverty line of $1.90 per person per day in 2011 PPP U.S. dollars, the poverty rate in Eswatini was 28.3 percent in 2016/17. This level is 3.8 percentage points higher than the average for lower-middle-income countries (World Bank 2020). SHIES 2016/17 data also allow for an in-depth examination of the levels and patterns of poverty. Estimated based on the country’s upper and lower poverty lines, overall and extreme poverty rates for 2016/17 were 58.9 percent and 20.1 percent, respectively. 14 The poverty gap and poverty severity estimates based on the upper poverty line were 24.9 and 13.0 percent, respectively, while the poverty gap and poverty severity estimates based on the lower poverty line were 5.0 and 1.8 percent, respectively (figure 2).15 Poverty rates show large differences across subgroups (figures 3 and 4). Poverty rates are much lower in urban areas than rural areas. For example, for 2016/17, the overall poverty rate in urban areas was 19.6 percent, compared to 70.1 percent in rural areas. Poverty rates also vary markedly across administrative regions in the country, ranging from 51.5 percent in Manzini to 71.5 percent in Lubombo. With respect to age groups, 2016/17 poverty rates were higher among young children (0–5 years), older children (6–14 years), and the elderly (60+ years) than among youth (15–35 years) and prime working-age adults (36–59 years). 16 Poverty rates also sharply decline with education attainment (measured in terms of the level of education completed) of the household head. Differences in poverty rates by sex, the sex of the household head, and orphan status are comparatively small.17 Poverty reduction since 2000/01 has been significant in Eswatini. Both the overall and extreme poverty rates fell by roughly 10 percentage points between 2000/01 and 2016/17. Various 14 For definitions of the upper and lower poverty lines, see appendix 1. 15 For definitions of the poverty gap and poverty severity, see appendix 2. 16 Among individuals below age 18 years, 56.5 percent are estimated to be “multidimensionally” poor in 2014, based on an examination of different dimensions of welfare (GOE 2017). The patterns in the rates of multidimensional poverty among children in 2014 are qualitatively similar to the patterns of poverty in the population at large in 2016/17. 17 Orphans are a large subgroup in Eswatini. Among those below age 18 years, 71 percent were classified as orphaned or vulnerable in 2014 (GOE 2018). The main driver behind this status is mortality or morbidity of caregivers due to HIV/AIDS-related causes (GOE 2018); in 2018, an estimated 45,000 of those below age 18 years were classified as orphans due to HIV/AIDS (http://aidsinfo.unaids.org/). Eswatini has the highest HIV prevalence in the world, with the prevalence among those aged 15 years and older estimated at 27.0 percent based on 2016/17 data (GOE 2019c). 11 decompositions of the decline in poverty levels in Eswatini by the World Bank (2020) point to key drivers. Growth-redistribution decompositions suggest that the decline in the overall poverty rate between 2000/01 and 2016/17 was fully driven by the growth in average consumption (holding the distribution of consumption fixed over time). The change in inequality in consumption over the period (holding average consumption fixed over time) would have in fact contributed to an increase in the poverty rate. The dominant role of average consumption growth behind the reduction in poverty in Eswatini applies to both urban and rural areas. Spatial decompositions suggest that the decline in poverty rate between 2000/01 and 2016/17 was mainly driven by a decline in poverty within areas (within urban areas, within rural areas, and within regions) rather than by the movement of people between areas (e.g., rural-to-urban migration), as reflected by the shift of urban and rural population shares. Lastly, decompositions of income sources suggest that the decline in poverty between 2009/10 and 2016/17 was driven by gains in wage income, followed by remittance income. The patterns in food insecurity in Eswatini in 2016/17 are consistent with those of poverty (figure 4). Overall, 40.1 percent of the population was severely insecure, while another 27.8 percent was moderately insecure. 18 Food insecurity was more prevalent in rural areas than urban areas, and more prevalent in Lubombo and Shiselweni than in Hhohho and Manzini. Food insecurity stems from both adverse structural circumstances as well as negative idiosyncratic and covariate shocks. Reflecting this, food insecurity status is strongly correlated with poverty status: While 67.4 percent of the extreme poor were severely food insecure in 2016/17, 19.6 percent of the nonpoor suffered from the same condition. Inequality Inequality in consumption per adult equivalent, measured by the Gini index, was 49.3 percent in 2016/17. This level of inequality places Eswatini among the 20 most unequal countries in the world. It is however lower than the inequality levels in other SACU countries.19 Inequality levels in 2016/17 were similar between urban and rural areas in Eswatini. Across regions, inequality was markedly lower in Shiselweni than in other regions (figure 5). Between 2000/01 and 2016/17, the Gini index declined in Eswatini by 1.9 percentage points, with the reduction driven by urban areas and by Hhohho region (World Bank 2020). Covariate shocks Eswatini is susceptible to economic and natural shocks, a situation exacerbated by its weak institutional and infrastructural capacity to manage the effects of these shocks. Because of the openness of Eswatini’s economy and its integration with South Africa’s, Eswatini has suffered adverse effects from periods of economic weakness in South Africa. Eswatini also experiences extreme weather events, including droughts. Most of its population, particularly in rural areas, depends on rainfed agriculture and livestock farming for income and consumption. Extreme 18 For definitions of the food insecurity categories, see appendix 1. 19 Based on statistics from the World Bank’s PolcalNet databank (http://iresearch.worldbank.org/PovcalNet/povOnDemand.aspx). 12 weather events have led to substantial crop and livestock losses, worsening an already high rate of food insecurity among the population. The El Niño phenomenon of 2015/16 induced the worst drought Eswatini had experienced in more than three decades, producing large crop and livestock losses and acute shortages of water used in homes and in commercial buildings (Swaziland VAC 2016). Maize production in 2015/16 fell to 33,460 metric tons from an average of 85,172 metric tons annually over the previous four years (2011/12–2014/15), a drop of roughly 60 percent (Swaziland VAC 2016). The government declared a drought emergency in February 2016. Roughly 260,000 individuals in rural areas (30 percent of the rural population) were estimated to require urgent food assistance during the period April–June 2016 (Swaziland VAC 2016). 20 The government, partnering with the international development aid community, responded by mainly providing food, water, and animal feed to affected communities and households. 21 These efforts were coordinated by the government’s National Disaster Management Agency (NDMA). Rainfall returned to a normal level in 2016/17, and agricultural production rebounded (annual maize production rose to 84,344 metric tons) (Swaziland VAC 2017). While the reference periods for the data are not perfectly aligned, compared to 2016, the size of the rural population estimated to require urgent food assistance declined sharply in 2017 and 2018, to 137,000 individuals in July–October 2017 and 123,000 individuals in June–September 2018 (Swaziland VAC 2017; Eswatini VAC 2018). Fiscal position Annual total government revenue ranged between 25 and 28 percent of GDP between 2015/16 and 2018/19 (figure 6). Government revenue mainly comes from taxes (personal income, corporate income, and value-added). A sizeable share of government revenue also comes from SACU receipts, which accounted for between 9 and 15 percent of GDP over the period. Volatility in the latter source of revenue was the main cause of volatility in the government’s overall revenue (UNICEF 2018a; IMF 2020; World Bank 2020). Annual total government expenditures ranged between 33 and 36 percent of GDP between 2015/16 and 2018/19, producing fiscal deficits equivalent to between 6 and 11 percent of GDP. The deficits have been driven by high spending across multiple areas, including personnel costs, administrative costs, capital outlays, extrabudgetary transfers to public entities and enterprises, and debt service payments. Recurrent fiscal deficits in recent years have led to a doubling of the public debt from 17.7 percent of GDP in 2015/16 to 33.4 percent in 2018/19. In addition, domestic arrears have accumulated, although the government is trying to contain personnel costs and slow the 20 Specifically, these individuals were classified as in phases 3 or 4 (crisis or emergency) based on the Integrated Phase Classification (IPC) scheme. 21 International donors also responded to the drought emergency through parallel, independent food distribution efforts. The United States Agency for International Development (USAID), the World Food Programme (WFP), and World Vision Eswatini reached 34,567 households in 2016/17, distributing 14,354 metric tons of food commodities (maize, pulses, vegetable oil), according to NDMA administrative information. Efforts by international donors also included the provision of temporary cash benefits. The Baphalali Eswatini Red Cross Society and WFP distributed a total of E77.2 million in cash benefits to 38,680 households in 2016/17, according to NDMA administrative information. 13 accumulation of arrears. The government needs to implement wide-ranging and drastic reforms on both the revenue and expenditure sides if it is to restore fiscal sustainability (UNICEF 2018a; IMF 2020; World Bank 2020). Social protection While the Eswatini government administers multiple social protection programs, it does not formally tie these programs together into a social protection system. Indeed, the government does not distinguish between social assistance programs and social welfare services (UNICEF 2018b). Many of the social assistance programs and social welfare services, in large measure, owe their genesis and continuation to the HIV/AIDS epidemic in the country (Dlamini 2007; Blank, Mistiaen, and Braithwaite 2012). Illness, morbidity, and mortality due to the disease have created widespread and acute needs among households, arising mainly from the loss of financial and other support from prime-age adults for themselves and their dependents, including children and the elderly. Most of the government’s social assistance programs are administered by the Department of Social Welfare (DSW) under the Deputy Prime Minister’s Office (DPMO). 22 (The next section describes the largest of these programs as defined by outlay and beneficiary numbers.) According to the Government’s Programme of Action 2013– 18 (GOS 2013), which includes DPMO’s action plan up to 2022, DPMO’s mission is the “improved quality of life and the wellbeing of Swazis through the provision of comprehensive social welfare and development systems, gender mainstreaming, children’s issues and proactive disaster management using social safety nets and social protection delivery through theory informed and data driven policy making.” DSW’s mandate, according to the Programme, is to “provide social development services that are people centered through theory informed and data driven policy making, advocacy and community strengthening using social safety nets and relevant social protection mechanisms.” Other ministries and agencies involved in administering or coordinating social assistance programs include the Ministry of Education and Training (MOET), the Ministry of Health (MOH), the Ministry of Tinkhundla Administration and Development (MTAD), the National Disaster Management Agency (NDMA), and the National Emergency Response Council on HIV and AIDS (NERCHA) (Blank et al. 2012). None of the government’s existing social assistance programs is covered by legal provisions. The main social insurance programs provide old-age, disability, survivor, and funeral benefits for public and formal private employees and compensation to individuals for injury, disability, or death due to road accident. These programs are covered by legal provisions and are administered by statutory parastatal organizations.23 The Eswatini government also legally mandates that public and formal private employers finance and provide sickness and maternity benefits and workmen’s compensation (disability, medical care, survivor, and funeral benefits) to employees. 24 Formal 22 Originally situated under the former Ministry of Health and Social Welfare (now Ministry of Health), DSW was relocated to DPMO in 2009 (Blank et al. 2012; UNICEF 2018c). 23 The legal provisions include the Swaziland National Provident Fund Order of 1974, the Pensions Act of 1968, the Motor Vehicle Accidents Act of 1991, the Retirement Funds Act of 2005, and regulations and amendments of these Orders and Acts. 24 The legal provisions include the Employment Act of 1980, the Workmen’s Compensation Act of 1983, and regulations and amendments related to these Acts. 14 private employers are also legally mandated to provide a lump-sum severance allowance to specified types of employees who are dismissed. 25 Several government policies, strategies, and plans underscore the critical role played by social assistance programs in protecting and increasing socioeconomic welfare at the community, household, and individual levels. Important policies and plans in this regard include the National Development Strategy (GOS 1999), the National Food Security Policy (GOS 2005), the National Poverty Reduction Strategy and Action Plan (GOS 2007), the National Disaster Risk Management Policy (GOS 2011), the National Social Development Policy (GOS 2010), the National Development Plan 2019/20–2011/22 (GOE 2019a), and the Eswatini Strategic Road Map: 2019– 2022 (GOE 2019b). The government’s national social development policy provides policy directions to DSW in particular, mainly focusing on the provision of social welfare services. DPMO’s action plan up to 2022 (GOS 2013) suggests an emphasis on the development and implementation of (1) legal provisions for the government’s main social assistance programs; (2) a more comprehensive set of social assistance programs to cover additional subpopulations and risks; (3) a monitoring and evaluation system for social assistance programs; (4) a national disaster early warning system; and (5) national disaster preparedness and response plans. The government has drafted a National Social Assistance Policy and a National Social Security Policy. 26 Work is currently underway to update and validate the national social assistance policy. Several international organizations have provided or currently provide technical, operational, and financial assistance to the government for its social assistance programs and social welfare services. These organizations include the European Union (EU), the United Nations Children’s Fund (UNICEF), the United Nations Development Programme (UNDP), the United Nations Population Fund (UNFPA), the World Bank, the World Food Programme (WFP), and World Vision International. The International Labour Organization (ILO) provides support to the government for its social security programs. III. Programs We examine the five social assistance programs with large numbers of beneficiaries reported in the SHIES 2016/17: neighborhood care points (NCPs), school feeding, food aid, secondary school grants to needy orphaned and vulnerable children (OVC education grants), and elderly grants. All five are non-contributory social assistance programs. NCPs, school feeding, OVC education grants, and elderly grants aim to provide benefits year-round, while the food aid program aims to provide emergency benefits during drought periods. All five programs employ varying combinations of categorical and self-targeting to reach the poor and vulnerable. The SHIES 2016/17 data also indicate that “work retirement pension” has nontrivial coverage, reaching 1.5 percent of individuals (5.5 percent of households). Presumably, these benefits are 25 Severance allowances are mandated under the Employment Act of 1980. 26 The drafts of these policies were first completed in 2018. 15 provided through the National Provident Fund, the Public Service Pension Fund, and pension programs offered by formal private employers. We exclude these social insurance programs from our performance analysis as they are conceptually distinct in important ways from the social assistance programs we examine. Further, the analysis of the performance of these programs would be incomplete without an actuarial assessment. The survey also gathered information on other government social assistance programs such as military (or ex-servicemen) pensions and disability grants. 27 However, the extent of coverage of the other social assistance programs is small, at less than 1 percent of households for each program. As a result, we exclude such programs from the analysis, given that estimates of their performance will be unreliable. The main design features of the five social assistance programs are presented in table 1. The information pertains to the designs in effect during 2015/16 and 2016/17, which overlap with the reference period for the SHIES 2016/17. Neighborhood Care Points Targeting orphaned or vulnerable children of preprimary age, Neighborhood Care Points (NCPs) aim to provide cooked meals (one meal per weekday) following World Food Programme (WFP) standards and the nutrition guidelines of the Swaziland National Nutrition Council. NCPs also aim to provide early childhood care and development (ECCD) services. These services are provided in fixed, physical centers staffed by community volunteers across the country. Over the reference period for the SHIES 2016/17, the program was virtually fully financed by WFP and in effect administered by the organization, although the Ministry of Tinkhundla Administration and Development (MTAD) was responsible for coordinating the program.28 Current implementation arrangements for the program remain the same. In principle, NCPs offer a broad range of childcare services. Caregivers are expected to be trained so that they can offer psychosocial support to children through interactions, ensure that child protection rights are abided by, and foster development through early learning, play, and recreation in accordance with standards established by the Ministry of Education and Training. NCPs are expected to be child-friendly and safe, to have toilets, safe drinking water, and hygiene amenities, and to provide play materials and spaces. Caregivers are also expected to facilitate the enrollment and retention of NCP school-age beneficiaries in nearby schools (GOS 2012). In practice, NCPs either exclusively or mainly provide cooked meals. WFP distributed food commodities to NCPs and provided technical assistance and capacity development support for NCP operations, while Save the Children, contracted by WFP, provided additional monitoring and other support to NCPs (WFP 2016). 27 Both military pensions and disability grants provide monthly cash benefits. Administered by DSW, military pensions are provided to qualifying individuals who served in World War II (or, upon their death, to their widows), while disability grants are provided to qualifying needy persons with disabilities. 28 A small percentage of NCPs at times received food commodities from other donors. In 2018, 4.2 percent of NCPs received food commodities from other donors, coordinated by WFP. 16 Based on data provided by NCP staff to WFP, in 2015, the program provided benefits to 50,845 children age 18 or younger (among which, 40,676 children were younger than age 5) in 1,594 NCPs across the country (WFP 2015), and, in 2016, to 51,899 children age 18 or younger (among which, 41,519 children were younger than age 5) in the same number of NCPs (WFP 2016) (figure 7). The distribution of NCPs by region between 2015–19 is provided in figure 8. NCPs are disproportionately located in the poorer, less-populated regions of Lubombo and Shiselweni. The vast majority of NCPs (97 percent) are in rural areas. Roughly 3 percent of NCPs are in peri-urban areas. The SHIES 2016/17 gathered information on the number of children that received NCP program benefits in the past year. Based on this information, in each relevant household, we assigned NCP program benefit receipt status to children starting from the youngest child up until the reported number of beneficiary children was reached.29 The survey data indicate there were 45,200 NCP program beneficiaries, which is quite close to the numbers from WFP administrative data for 2015 and 2016. However, while administrative data indicate that 80 percent of beneficiaries were younger than age 5, the survey data suggest that this share was significantly lower, at 35 percent. The share estimated using survey data can be interpreted as an upper-bound value given that, by construction, the NCP program receipt status indicator was biased toward classifying the youngest children as beneficiaries. Over the reference period for the SHIES 2016/17, the provision of meals to beneficiaries dominated NCP operations, serving as the main pull for beneficiaries (WFP 2017). However, NCPs experienced major disruptions to this service. The actual distribution of food by WFP to NCPs fell substantially short of planned distribution (figure 9a), resulting in cutbacks in the content of meals, disruptions to regular provision of meals, or temporary closure of NCPs altogether (WFP 2015, 2016, 2017). For 2015, the annual food procurement cost per beneficiary is estimated at E272, while the annual overall program cost per beneficiary, which includes administrative and operational costs, is estimated at E406 (figure 9b). The corresponding estimates decrease to E34 and E109 in 2016 because of the particularly large shortfall in food distribution by WFP to NCPs that year. Averaged across the two years, the annual food cost per beneficiary is estimated at E153 and the overall program cost per beneficiary at E258. We treat the averaged overall program cost per beneficiary as the annual NCP benefit received by the beneficiary over the reference period for the SHIES 2016/17. This benefit value is equivalent to 4.6 percent of the annual lower poverty line and 2.2 percent of the annual upper poverty line in January 2017 prices. Imputing the overall program benefit per beneficiary into the SHIES 2016/17, the average NCP benefit received by beneficiary households as a proportion of their total consumption is estimated to be 1.9 percent. School feeding program Administered by the Ministry of Education and Training (MOET), the government’s school feeding program provides one cooked meal every school day for each student in government primary, secondary, and high schools. Over the period of our analysis, the school feeding program 29 For full details on the construction of the program benefit receipt indicator, see appendix 1. 17 covered 844 schools. Currently, it covers 860 schools. Out of this number, 810 schools are part of the traditional school feeding program, while 50 schools were recently transitioned into a pilot. home-grown school feeding program.30 The government began fully financing school feeding in government primary schools in 2009/10, and began doing the same in government secondary schools in 2014/15.31 All students in these schools are eligible for the program. MOET estimates that beneficiary numbers totalled 350,533 and 368,078 in school years 2015 and 2016, respectively. 32 Estimated beneficiary numbers from school years 2014 to 2019 are presented in figure 10a. Schools are expected to prepare meals in accordance with guidelines and standards developed and distributed by MOET. 33 The program stipulates a ration size per student per school day of maize grain or rice, pulses, and vegetable oil. MOET finances the procurement of food commodities (maize grain and rice, pulses, cooking oil, and peanut butter) from large private food suppliers, while the National Emergency Response Council on HIV and AIDS (NERCHA) makes the actual purchases on behalf of the Ministry. MOET is responsible for distributing procured commodities to schools by the start of each school term, by transporting them to schools from regional warehouses of the food suppliers. 34 The annual per-student grant of E560 provided by MOET to government primary schools for the Free Primary Education initiative includes an amount of E150 (per student per year) which is intended to cover the nonsalary operational costs of the feeding program at the school and another amount of E150 (per student per year) which is intended to cover the salaries of school support staff, including cooks, for the school feeding program. 35 The per-student grant amount of E150 to cover the nonsalary operational costs of the program has remained fixed since 2009/10 when the Free Primary Education Act came into effect and when the government fully took over the school feeding program in government primary schools. Students in government secondary and high schools pay school meal fees to cover program operational costs (salary and nonsalary) at the school. 36 A recent operational evaluation of the program by WFP documented several shortcomings (WFP 2019). Government financing is considered acutely inadequate for procuring the required quantity of food commodities for the program that is in line with the ration sizes per student per school day 30 The pilot program sources food commodities from local smallholder farmers. UN FAO and WFP support the pilot program. 31 School feeding was first introduced in government primary schools in Eswatini in 1963, financed and administered by Save the Children. Before WFP handed over the program to the Eswatini government in 2010, it financed and administered school feeding in government primary schools in two phases (1970–1991 and 2002–09). School feeding was first introduced in government secondary schools in 2007/08 by Claypotts Trust, and was financed by the Global Fund between 2009–13 (Blank et al. 2012; WFP 2019). 32 The beneficiary estimates for the program are the same as school enrollment levels estimated by MOET. 33 These guidelines include the Inqaba Implementation Manual and the National Framework for Food Security in Schools. 34 The school year is composed of three terms. 35 Nonsalary operational costs comprise kitchen operational costs and the purchase of kitchen, storage, and serving equipment and utensils and additional meal ingredients. 36 The SHIES 2016/17 did not gather separate student-level information on household spending on school meal expenses. 18 stipulated by the government. Based on administrative data on released budgets for food procurement and estimated school enrollment in the last few years provided by MOET, we find that the cost of food procured ranged between E130 and E170 per student between 2014/15 and 2017/18 (figure 10b). Consistent with this range, WFP (2019) estimates a cost of E145 per student in 2018 for food procured, but also estimates that this cost would have to be about E300 per student per year (roughly double the current cost) in order to meet the required quantity of food commodities for the program. Notwithstanding, the nutrition content of the ration per student per school day stipulated by the government is found to fall short of international standards for daily nutrition requirements for school feeding rations. 37 Other issues include delays in the release of funds to MOET for the program, which contributes to delays in procurement and in the timely provision of food commodities to schools for the program. Schools, particularly more remote ones, may have to collect food commodities from warehouses on their own because MOET fails to provide (timely) transportation. Finally, the annual per- student grant of E150 for nonsalary operational costs for the program at the school level is considered too low to fully cover these costs. The food procurement cost per beneficiary was E171 and E161 in 2015/16 and 2016/17, respectively. We assign a value of E300 to each primary school student as the upper-bound estimate of the grant benefit per student toward defraying the operational costs of the program at primary schools (secondary school students do not receive this benefit). Based on these values, we estimate a total school feeding program benefit per primary school student of E471 and E461 in 2015/16 and 2016/17, respectively. Averaging the values across the two years, we impute annual program benefit amounts of E466 for a government primary school student and E156 for a government secondary school student into the SHIES 2016/17. Based on this survey, the average total school feeding program benefit received by beneficiary households as a proportion of their total consumption is estimated to be 2.9 percent. Emergency food aid Food aid is distributed by the Eswatini government during periods of, and in areas experiencing, drought. The benefits are administered by the National Disaster Management Agency (NDMA). The agency procures food (cereals, pulses, and cooking oil) from a domestic private food distributor. Local government officials, community leaders, appropriate community committees and structures, and nongovernmental organizations (NGOs) work together to identify households in need of food aid, following selection criteria and procedures jointly developed by the NDMA and a consortium of NGOs. The selection criteria comprise several indicators of household socioeconomic disadvantage and distress in relation to household composition, health and disability status of members, coping strategies for food insecurity, farming output, assets, livelihoods, and income sources, and are accompanied by guidance on the use of the criteria to categorize households into groups considered predictive of the severity of their food insecurity. Food aid is expected to be distributed to households categorized as experiencing the severest food insecurity, until the food aid supply brought to the community is exhausted. Households are 37 Nutrition requirements in international standards for school feeding programs vary by child age, while the daily ration amounts per student stipulated by the Eswatini government do not. 19 formally registered for aid after community leaders validate their food insecurity status. Registered households collect their benefits at food distribution points administered by local government officials and NGOs. 38 Household benefit amounts—equal to a month’s food-commodity requirements—vary by household size. Between April 2016 and December 2017, NDMA distributed 4,834 metric tons of food commodities (87 percent of which was maize) to 72,745 households across the country in response to the 2015/16 drought. 39 Of the total amount of food handed out during this period, 67 percent was distributed in two months, in August and September 2016. Using total spending on benefits by NDMA in 2015/16 obtained from government administrative information, we estimate a food aid benefit per beneficiary household of E658 in response to the 2015/16 drought. This amount is 11.8 percent of the annual lower poverty line and 5.6 percent of the annual upper poverty line in 2017 prices. The SHIES 2016/17 reports the monetized amount of government food aid received by beneficiary households. The average annual amount received by beneficiary households was E620, which is close to the estimate based on administrative information.40 The average total amount received by beneficiary households as a proportion of their total consumption was 2.6 percent, based on the SHIES 2016/17. Orphaned and vulnerable children (OVC) education grants Administered by DSW, OVC education grants are provided to needy orphaned and vulnerable children to support their education in grades 8–12 in 260 government secondary and high schools. 41 Grants are delivered directly to schools on behalf of beneficiary students, to help defray their tuition and exam costs. To apply for the grant, prospective beneficiaries must first secure certification by local community leaders and then approach social welfare offices or the schools they are attending. Social welfare offices review and approve applications and share beneficiary information with the appropriate schools. Schools confirm beneficiary eligibility, including the child’s school enrollment status, and transmit beneficiary lists to DSW through MOET. DSW transfers grants to the schools’ bank accounts in two installments: one-third of the grant amount at the start of the first school term and the remaining two-thirds at the start of the second term. 38 Partnering NGOs included Adventist Development and Relief Agency, Africa Cooperative Action Trust, Caritas, Save the Children, and Baphalali Eswatini Red Cross Society. 39 Apart from assisting NDMA in distributing food aid on the ground, international donors, specifically USAID, WFP, and World Vision Eswatini, also distributed food aid directly to households in need through their own logistical arrangements. Reporting by the donors to NDMA indicates that these donors distributed 14,354 metric tons of food commodities to 34,257 households in response to the 2015/16 drought. 40 For details on the benefit variable construction, see appendix 1. 41 The program was initiated in 2003 and administered by MOET. It was transferred to DSW in 2009. 20 In 2015/16 and 2016/17, the number of program beneficiaries totaled 53,564 and 52,632, respectively. Beneficiary numbers between 2009/10 and 2019/20 are presented in figure 11a. The per-student grant amount going toward tuition fees was E1,950 for students in grades 8–10 and E2,500 for students in grades 11–12. The per-student grant amount going toward exam fees was up to E2,000 for students in grade 12 only. Figure 11b presents the total grant amounts, grant amounts going toward tuition fees, and grant amounts going toward exam fees per beneficiary based on DSW administrative data. The total grant amount per beneficiary averaged E2,535 in 2015/16 and E2,609 in 2016/17 The average amount in 2016/17 is equivalent to 47 percent of the annual lower poverty line and 22 percent of the annual upper poverty line in January 2017 prices. The average grant amount per beneficiary student in the SHIES 2016/17 is somewhat lower than the value from administrative data, at E2,286. 42 Based on the same survey data, the average total grant amount received by beneficiary households as a percentage of their total consumption is 10.1 percent. SHIES 2016/17 data indicate that 47 percent of beneficiary students paid tuition fees, averaging about E1,500 per beneficiary student in the reference year. This suggests that a large share of beneficiary students continued paying some tuition fees despite receiving the grant. Peer nonbeneficiary students who reported paying tuition fees were found to pay on average about E3,700 in such fees in the reference year. Based on the same data, sizeable shares of program beneficiary students reported paying for uniforms (60 percent of beneficiary students), learning materials other than textbooks (20 percent), and transportation to school (17 percent), among other expenses.43 Average annual payments for schooling-related expenses other than tuition and exam fees averaged about E800 per program beneficiary student in the reference year. These other expenses are not covered by OVC education grants. Elderly grants Administered by DSW, elderly grants are offered to citizens aged 60 and older. To qualify, the prospective individual must file an application, along with stipulated supporting documentation, at the local government office. In principle, elderly individuals who receive benefits of a stipulated minimum monthly amount from other pension programs (such as the Eswatini National Provident Fund or the Public Service Pension Fund) are ineligible to receive the elderly grants; however, this restriction does not appear to be applied. In 2005, when the program was launched, there were 43,860 beneficiaries (Blank et al. 2012); in early 2020, there were 58,216 beneficiaries. The total number of beneficiaries in 2017 and 2018 were 66,378 and 69,697, respectively (UNICEF 2017, 2018b); these years are slightly removed from the reference period for the SHIES 2016/17. Under the elderly grants program, beneficiaries receive cash on a quarterly basis through the local government office or post office, or on a monthly basis through electronic transfer to the 42 For details on the benefit variable construction, see appendix 1. 43 Expenses for school meals were not gathered separately in the SHIES 2016/17. 21 beneficiary’s personal bank account. 44 Most beneficiaries receive their grants through the former method. 45 A survey of program beneficiaries conducted in 2010 finds that the vast majority of beneficiaries are able to register for the program without any issue and collect benefits by personally traveling to local government offices. While they are generally aware of the various available collection options, most beneficiaries prefer collecting benefits from local government offices than from postal offices or banks or ATMs mainly because of greater convenience, proximity, or confidence in the former collection option (UK DFID, HelpAge International, and UNICEF 2010). Over the reference period for the SHIES 2016/17, the benefit amount was E240 per month, which is equivalent to 52 percent of the lower poverty line and 24 percent of the upper poverty line in January 2017 prices. 46 The average annual grant amount per beneficiary in the SHIES 2016/17 was E2,474. 47 The average total grant amount received by beneficiary households as a proportion of their total consumption was 11.7 percent. While the grants are not indexed to price inflation, benefit amounts have been raised at irregular intervals since the program’s inception in 2005 (figure 12). While the nominal value has increased more than six times between 2005–20 (from E80 to E500 per month), the real value has increased 2.5 times (from E80 to about E200 in January 2005 prices) over this period. IV. Results We discuss findings from our analysis of the performance of the selected social assistance programs in six main parts. In the first part, we discuss the levels of spending on the programs in absolute terms and relative to total government expenditure and GDP. In the second through fourth parts, we discuss findings for program coverage, incidence, and effectiveness. In the fifth part, we discuss the hypothetical results on program incidence and effectiveness of simulated reforms in benefit amounts and the eligibility for elderly grants and OVC education grants based on applying a PMT model we construct for the exercise. In the sixth and final part, we discuss the findings from profiling reported shocks, examining the association of main shocks with poverty status, food insecurity status, area of residence, and the receipt of benefits from the social assistance programs. In this part, we also discuss findings from profiling reported assistance received in response to shocks. The findings in the first part are based on government and WFP administrative data, while the findings for the other parts are based on SHIES 2016/17 data. Apart from discussing the performance results for individual social assistance programs, we also examine results for two aggregations: all five programs together and the four government-financed programs together (excluding NCPs). The exclusion of NCPs in the aggregation matters little. Hence, we discuss results for the aggregation of all five programs Spending on program benefits 44 DSW pays service fees of E25 and E30 per beneficiary per electronic and manual transfers, respectively. 45 Currently, about 20 percent of beneficiaries receive their elderly grants through electronic transfer. 46 The elderly grant amount of E240 per month became effective on April 1, 2015 (Swaziland Budget Speech 2015/16). 47 For details on the benefit variable construction, see appendix 1. 22 How much was spent on program benefits? Information on spending on NCPs between 2015 and 2017 was obtained from WFP and is discussed separately given that the program is not financed by the government. Spending on NCPs fluctuated significantly, driven essentially by the amount of food procured for the program. Spending on food procurement for NCPs was E13.8 million in 2015, dropping to E1.8 million in 2016 (figure 13a). Commensurately, total spending on NCPs (comprising food procurement, operational, and administrative costs) declined from E20.6 million in 2015 to E5.7 million in 2016. In 2017, spending on food procurement rebounded somewhat. Official government budget reports do not separately specify the amount spent on school feeding. Instead, the various components of spending on the program are subsumed within different budget line items under the MOET's Office of the Chief Inspector, Primary Education. Data on food procurement spending were obtained from MOET. We estimate total spending on the school feeding program based on the assumption that up to E300 of the E560 per-student grant for Free Primary Education was applied toward defraying the operational costs of the school feeding program in primary schools. Our calculation also incorporates information on food procurement spending and total primary enrollment, as provided by MOET. Between 2014/15 and 2017/18, spending on food procurement for the program ranged between E47 million and E60 million, while estimated total spending on the program ranged between E127 million and E139 million (figure 14a). Transfers by DPMO to NDMA to cover benefits and operational and administrative expenses were specified in official government budget books. We assume that the transfers to NDMA for benefits fully constitute food aid benefits. Except for 2016/17, spending on food aid between 2014/15 and 2017/18 ranged from E9 million to E23 million and total spending ranged from E16 million to E26 million (figure 14b). In 2015/16, as a result of the El Niño-induced drought emergency, spending on food aid rose sharply to E47.9 million and total spending to E77.2 million. Spending by DSW on OVC education grants and elderly grants are specified in official government budget reports. Spending on administrative and operational expenses of DSW and other benefits by DSW are also specified in these reports. We examine the distribution of government spending on social assistance program benefits across the four programs (school feeding, emergency food aid through NDMA, OVC education grants, and elderly grants) and other social assistance program benefits under DSW taken together (figure 15). The Eswatini government spent between E458 million and E687 million annually on social assistance programs between 2014/15 and 2017/18. Spending was highest for elderly grants. Between 2014/15 and 2016/17, spending on these grants ranged between E165 million and E185 million annually (equivalent to 35–37 percent of annual total spending on social assistance program benefits) before jumping to E340 million in 2017/18 (equivalent to 50 percent of total spending). The jump is mainly due to a large increase in the monthly benefit amount from E240 to E400 per beneficiary. Spending was second-highest for OVC education grants, followed closely by spending on school feeding. Between 2014/15 and 2017/18, spending on OVC education grants ranged between E137 23 million and E161 million annually (equivalent to 24–31 percent of annual total spending on social assistance program benefits). Over the same period, spending on school feeding ranged between E127 million and E139 million annually (equivalent to 19–30 percent of annual total spending). For food aid, as noted earlier, apart from in 2015/16, when spending reached E37.9 million (9 percent of total spending on social assistance program benefits that year) due to the El Niño- induced drought emergency, spending accounted for less than 5 percent of total spending on social assistance program benefits in each of the other years. Likewise, for all of the other social assistance program benefits combined, spending was 5 percent or less of total spending on social assistance program benefits in each of the years between 2014/15 and 2017/18. Between 2014/15 and 2017/18, annual total government spending on social assistance benefits represented 2.5–3.5 percent of annual total government spending, and was equivalent to 0.9–1.2 percent of annual GDP (figure 16). In comparison, the average level of annual spending on social assistance programs among developing countries and in Sub-Saharan Africa in recent years has equalled roughly 1.6 percent of annual GDP (figure 17). Program coverage What is the extent of coverage of individuals and households by the main social assistance programs in Eswatini? Determining benefit receipt status based on the SHIES 2016/17 is not straightforward for some of the programs. The rules that we applied to assign benefit receipt status are detailed in appendix 1. For NCPs, school feeding, OVC education grants, and elderly grants, benefit receipt status is assigned at the individual level. For food aid, status is assigned at the household level (that is, all household members are considered beneficiaries), because the survey captured information on the receipt of government food aid at the household level. Following our variable constructions, the estimated numbers of program beneficiaries based on the SHIES 2016/17 correspond reasonably well with the numbers from government and WFP administrative data (table A1.1) for all programs except food aid. The shortfall in beneficiaries in the survey data is substantial for food aid. This suggests coverage by the program is likely to be underestimated by a large amount. Likewise, estimates of the incidence and effectiveness of the program, discussed later, are likely to have a large downward bias. We examine social assistance program coverage rates for individuals (including presumed eligible individuals, based on applying some of the eligibility criteria for program benefit receipt to the SHIES 2016/17 data) and for households. We examine coverage rates separately by poverty status (extreme poor, moderate poor, and nonpoor) and by food insecurity status (food secure, mildly insecure, moderately insecure, and severely insecure). To allow for international comparisons, we further estimate coverage rates separately by consumption quintiles. Apart from at the national level, coverage rates are also estimated separately for rural and urban areas. The full set of coverage results is reported in tables A3.1–A3.3. The discussion that follows focuses on program coverage results at the national level for individuals and households, disaggregated by poverty status and by food insecurity status. 24 In terms of individual coverage rates, school feeding registers the highest rate, at 22.8 percent, followed by food aid, at 16.3 percent (figure 18a). Elderly grants, OVC education grants, and NCPs cover 5.8 percent, 4.0 percent, and 4.0 percent of individuals, respectively. The five programs combined cover 41.3 percent of the population. Individual coverage rates by poverty status indicate pro-poor coverage across all the programs (figure 18b). The five programs together cover 54.6 percent of the extreme poor and 50.8 percent of the moderate poor, with the rate falling steeply to 25.9 percent for the nonpoor. The population coverage rates of some of the social assistance programs are low because not all individuals satisfy the stipulated eligibility rules for program benefits. We apply some basic eligibility rules to the SHIES 2016/17 data to approximate eligible individuals, such as restricting eligibility for elderly grants to individuals age 60 years or above, eligibility for OVC education grants to orphans in government secondary schools, and eligibility for food aid to individuals in severely food-insecure households. Individual coverage rates for these presumed eligible individuals shoot up as a result, with a reordering of the ranking of the programs in terms of the extent of coverage. Elderly grants cover 81.1 percent of those aged 60 years or older; school feeding 79.3 percent of government primary school students and 60.8 percent of government secondary and high school students; and OVC education grants 54.7 percent of orphans in government secondary schools (figure 19a). Although rising somewhat, the coverage rate of the presumed eligible population for food aid remains limited, at 23.1 percent of severely food-insecure individuals. The pattern of coverage rates for presumed eligible populations by poverty status remains pro-poor for all of the programs (figure 19b). The pro-poor pattern in coverage rates is particularly strong for NCPs, food aid, and OVC education grants. A household is defined as receiving a given program benefit if any member receives the benefit. By construction, household coverage will be more extensive than individual coverage. Household coverage rates are estimated without applying any benefit eligibility conditions. The extent of household coverage is highest for school feeding (38.7 percent), followed at some distance by elderly grants (19.7 percent) (figure 20a). Food aid and OVC education grants each cover 11–12 percent of households, while NCPs cover 7.7 percent of households. The five programs together cover 52.0 percent of households. Household coverage rates are also pro-poor, with most poor households covered when all five programs are taken together. The programs collectively cover 85.9 percent of extreme-poor households and 74.5 percent of moderate-poor households, with the rate dropping to 33.2 percent for nonpoor households (figure 20b). At the household level, each of the programs exhibits a strong pro-poor pattern in the extent of coverage. Given the high correlation between poverty and food insecurity among households, the pattern of program coverage rates for individuals and households by food insecurity status mirrors the pattern of program coverage rates by poverty status (figure 21). Looking at all five programs together, in terms of individual coverage rates, 48.7 percent of those who are severely food insecure are covered, while 25.7 percent of those who are food secure are covered. Likewise, in terms of 25 household coverage rates by all five programs together, 63.0 percent of those that are severely food insecure are covered, while 33.6 percent of those that are food secure are covered. Individual and household coverage rates estimated separately for urban and rural areas reveal that, across programs, coverage is higher in rural than urban areas and that coverage rates are pro-poor and pro-food insecure in both types of areas (tables A3.1–A3.3). Program incidence Program incidence for the poor is measured by accounting for who receives a social assistance program benefit and the value of the benefit. NCPs, school feeding, and OVC education grants provide in-kind benefits to individuals, monetary values for which are not provided in the SHIES 2016/17. Consequently, we impute benefit values based on government and WFP administrative information for the reference period of the survey. Food aid by government is also in-kind. The survey provides the monetary value of food aid received by each household, which we divide equally among all household members. The survey also provides the value of elderly grants received by each household, which we divide equally among reported program beneficiaries within the household.48 As a first way to assess program incidence, figure 22 shows the distributions of program benefits and beneficiaries by poverty status. The figure also shows the distributions of total household consumption and the population by poverty status, which serve as points of reference. The extreme poor account for 5.1 percent of total household consumption and for 20.1 percent of the population. The poor (extreme poor and moderate poor together) accounts for 24.0 percent of total household consumption and for 58.9 percent of the population. In contrast, program benefits are disproportionately received by poorer households, and program beneficiaries are disproportionately poorer households. Furthermore, the pro-poor patterns in the distribution of benefits and beneficiaries are remarkably similar across programs. The extreme poor receive between 27 percent and 31 percent of benefits across programs, whereas the extreme poor and the moderate poor together receive between 71 percent and 79 percent. Taking all five programs together, the extreme poor receive 28.0 percent of total benefits, and the extreme poor and the moderate poor together receive 73.5 percent of total benefits. Likewise, across programs, between 29 percent and 36 percent of beneficiaries are extreme poor, and between 78 percent and 85 percent of beneficiaries are either extreme poor or moderate poor. Taking all five programs together, 25.2 percent of beneficiaries are extreme poor and 70.7 percent of beneficiaries are either extreme or moderate poor. Concentration coefficients offer another way to assess program incidence (figure 23). 49 A coefficient value closer to minus one indicates that the benefits from a program are more concentrated among poorer households, while a value closer to one indicates that they are more concentrated among richer households. Coefficients range from –0.255 for NCPs to –0.152 for 48 For details on the construction of program benefit value variables, see appendix 1. 49 For a definition of the measure, see appendix 2. 26 elderly grants, which indicate that benefits for all the programs are concentrated among poorer individuals. The concentration coefficient for the five programs together is –0.193. Still another way to assess program incidence is through their marginal effects on inequality and poverty (table 2, columns 1–3). 50 These effects are defined as the change that each program produces in the Gini index, the poverty rate, or the poverty gap, with the sign reversed so that a positive value indicates a reduction in inequality or poverty. The poverty rate and the poverty gap are measured using the upper poverty line. Marginal effects depend on who receives the program benefit and the value of the benefit received, reflected by the concentration coefficient and the size of the program, defined as total program benefits divided by total household consumption (table 2, column 4). Even though NCPs and food aid have more negative concentration coefficients among the programs, their effects on inequality and poverty are small (0.2 percentage points or less) because the sizes of the programs are small. The larger-size programs—school feeding, OVC education grants, and elderly grants—have larger effects on inequality and poverty. School feeding has the largest effect on the poverty rate, reducing it by 0.8 percentage point, followed by elderly grants (0.5 percentage point) and OVC education grants (0.3 percentage point). While elderly grants and OVC education grants have smaller effects on the poverty rate than school feeding, they have larger effects on the poverty gap. All three programs also reduce inequality by similar amounts (by between 0.5 and 0.7 percentage point each). The five programs together, which account for 5.0 percent of total household consumption, reduce inequality by 2.0 percentage points, the poverty rate by 1.5 percentage points, and the poverty gap by 3.0 percentage points. We also examine the sensitivity of marginal effects on poverty to the choice of the poverty line (upper or lower) (table A3.4). OVC education grants and elderly grants have larger marginal effects on the poverty rate when measured using the lower poverty line than when using the upper poverty line, while school feeding shows the opposite pattern. NCPs and food aid have small marginal effects on the poverty rate and poverty gap irrespective of the choice of poverty line, a result driven by their small program sizes. The five programs together reduce the poverty rate to a greater extent when measured using the lower poverty line than when using the upper poverty line (5.2 percentage points, compared to 1.5 percentage points), while the reduction in the poverty gap is comparable across the two poverty lines (about 3.0 percentage points each). Program effectiveness One characteristic of marginal effects is that, all else being equal, programs with larger outlays have larger marginal effects. 51 A way to check the robustness of the incidence results based on marginal effects is to calculate a program’s effectiveness—roughly, its bang for the buck. As with 50 For a more detailed definition of the measure, see appendix 2. 51 This is not quite true, as explained in Lustig (2018), but the intuition serves our purposes here. 27 marginal effects, information on who received a given social assistance program benefit and how much was received is used for estimating effectiveness. Impact effectiveness is defined as the ratio between the inequality or poverty reduction that a program actually achieves and the inequality or poverty reduction that it could achieve if all its outlay were distributed in the most progressive way possible. This perfect distribution of a program’s spending would give enough of a benefit to the poorest person to bring his or her consumption up to the level of the second-poorest person, then give enough to both to bring their consumption up to the level of the third-poorest person, and so on until the outlay is exhausted. Unlike the estimation of impact effectiveness, which holds the program’s outlay constant, the estimation of spending effectiveness holds the inequality or poverty level constant and asks how much less we could spend and still attain that inequality or poverty level if we were to achieve the perfect distribution of program benefits as described above. Both effectiveness measures are based on a scale from 0 to 100 percent.52 We examine program impact and spending effectiveness with respect to inequality as measured by the Gini index and with respect to poverty as reflected by the poverty gap (by construction, the impact and spending effectiveness for the poverty gap are identical), and by poverty severity (table 2, columns 5 – 9). The poverty gap and poverty severity measures are based on the upper poverty line. The effectiveness results are similar across programs. This is despite the higher marginal effects on inequality and poverty for school feeding, OVC education grants, and elderly grants than for NCPs and food aid, indicating that the larger marginal effects for the former programs are due to their larger outlays. The programs produce between 49 percent and 55 percent of the reduction in inequality that a perfect transfer of equivalent outlay would produce. Similarly, the outlay for a perfect transfer that reduces inequality by as much as the marginal effect of actual program spending would be between 48 percent and 54 percent of the actual outlays of the programs. All the programs have impact and spending effectiveness estimates that are higher for the poverty gap than for poverty severity. The programs produce between 73 percent and 81 percent of the reduction in the poverty gap and between 46 percent and 54 percent of the reduction in poverty severity that a perfect transfer of equivalent outlay would produce. This indicates that, while most of the benefits of these programs go to the poor, they perform less well at reaching the poorest of the poor. The five programs together have effectiveness estimates that are quite similar to the estimates for the individual programs, which is not surprising given that the latter are so similar across programs. The programs together produce 54.3 percent of the reduction in inequality, 75.3 percent of the reduction in the poverty gap, and 57.4 percent of the reduction in poverty severity that a perfect transfer of equivalent outlay would produce. We also examine the sensitivity of effectiveness in poverty reduction to the choice of poverty line (upper or lower) (table A3.5). Effectiveness is much lower when measured using the lower poverty line than using the upper poverty line. This pattern indicates that all programs are less effective at 52 For a more detailed discussion of these measures, see appendix 2. 28 reaching the poorest of the poor than they are at reaching the poor. Put another way, the benefits of these programs go mostly to the poor, not the extreme poor. This same point can be seen in the lower impact and spending effectiveness that the programs have in reducing poverty severity than in reducing the poverty gap. But it is also interesting that the effectiveness of OVC education grants and elderly grants in reducing poverty severity tends to be higher than for the other programs, and is clearly highest among the programs when measured using the lower poverty line. This indicates that OVC education grants and elderly grants perform better than the other programs in reaching the poorest of the poor. Hypothetical effects from simulated reforms to program targeting and benefit levels Social assistance program benefits in Eswatini are targeted to subpopulations based on characteristics assumed to be correlated with poverty or, more precisely, to be indicative of need for social assistance: for example, the elderly or orphaned and vulnerable children. Yet many of those subpopulations are not, in fact, poor. To better focus program benefits on the poor and on poverty reduction, it would be best to target them based on income, but this is difficult when income is difficult to observe because many workers are engaged in informal employment. Many countries where this is a problem use a proxy means test (PMT) model to target program benefits. A PMT model uses a few easily identifiable characteristics of people or households to identify those most likely to be poor. The best such characteristics should be: (a) highly correlated with poverty status; (b) easily identifiable by a social welfare worker; and (c) difficult for households to falsify. In appendix 4, we discuss our estimation of a PMT model for Eswatini based on the SHIES 2016/17. Here, we explore the extent to which the Eswatini government could improve the targeting of elderly grants and OVC education grants through use of this PMT model. We use PMT-based targeting in this study simply as a data-driven approach to simulate the budgetary, poverty, and inequality effects of improved targeting of program benefits to the poor. The Eswatini government currently uses a combination of categorical-based targeting and self- targeting in its main social assistance programs. Before the government decides whether to adopt PMT-based targeting to identify poor beneficiaries for its programs, it should weigh the pros and cons of PMT-based targeting vis-à-vis other possible methods. The criteria for assessing different methods should go beyond targeting accuracy to include ease and efficiency of implementation, legitimacy, and transparency. Such an assessment may require testing alternative (combinations of) methods in the field. We return to this discussion near the end of this subsection. We perform two general types of reform simulations. The first type eliminates benefits from those who the PMT model indicates are not poor—the “PMT-nonpoor.” The PMT-nonpoor are those who score above the 59 th percentile of the PMT score distribution, which coincides with the percentile where the upper poverty line intersects the consumption distribution. This reform is designed to save the government money, an important consideration in the current fiscally constrained environment. The second type of reform similarly eliminates benefits for the PMT- nonpoor, but reallocates the savings to larger benefits for beneficiaries who are PMT-poor. For 29 elderly grants, we also simulate the elimination of benefits for those who receive another (larger) pension, eliminating “double dipping.” Table 3 reports the results. PMT-based targeting reduces the concentration coefficients of the programs greatly, to about –0.4 for both programs. These estimates are now within the range of concentration coefficient values found for direct transfer payments in other developing countries. 53 For the reform simulation that eliminates double dipping on pensions, the concentration coefficient for elderly grants declines to about – 0.2. PMT-based targeting of the programs eliminates benefits from large shares of beneficiaries, specifically 30.1 percent of households receiving OVC education grants (3.4 percent of all households), and 39.7 percent of households receiving elderly grants (7.8 percent of all households). Elimination of double dipping on pensions removes benefits from 12.7 percent of households receiving elderly grants (2.5 percent of all households). PMT-based targeting generates large budgetary savings: E28.6 million from targeting OVC education grants and E62.2 million from targeting elderly grants. However, these savings are accompanied by increases in poverty and inequality due to errors of inclusion and exclusion inherent in PMT-based targeting (figure 24). In the case of OVC education grants, the reform increases poverty and inequality slightly, by 0.2 percentage points or less depending on the specific measure. In the case of elderly grants, the reform increases poverty and inequality somewhat more, by 0.1 – 0.4 percentage point depending on the specific measure. On the other hand, if the savings from PMT-based targeting are returned to existing beneficiaries who are PMT-poor so that the reform is budget neutral, this reduces the overall poverty gap, extreme poverty (both the rate and the gap), and inequality. For both programs, the reform still increases the overall poverty rate slightly. 54 The reduction in extreme poverty from the reform to elderly grants is larger than other changes: The extreme poverty rate falls by a full percentage point while the extreme poverty gap decreases 0.4 percentage point. While the implementation of these reforms is a choice the Eswatini government must weigh, our simulations suggest that better targeting of programs, such as through the use of PMT-based targeting, could either generate large budgetary savings or improve the poverty and inequality effects of these programs if these savings are reallocated as increased benefits to PMT-poor beneficiaries. As noted earlier, whether PMT-based targeting is implemented by the Eswatini government will require careful consideration of the potential advantages and disadvantages of this and alternative targeting methods. Little available rigorous evidence (such as from field experiments) exists on the relative performance of alternative targeting methods, across various evaluation criteria. 53 The average concentration coefficient value for direct transfers of any type among 32 countries is–0.30, and the lowest coefficient value is –0.63 (http://commitmentoequity.org/datacenter). 54 This may seem counterintuitive. The removal of the benefit from the PMT-nonpoor still has the effect of increasing poverty, while the additional benefits received by the PMT-poor—a 37-percent increase in the benefit from the reform to OVC education grants and a 61-percent increase in the benefit for the reform to elderly grants—are not large enough to move many of those beneficiaries above the overall poverty line. But the larger benefit they receive contributes to a reduction in the poverty gap. 30 Evidence from Indonesia shows that community-based targeting (or community-based targeting combined with PMT-based targeting) performed marginally worse than PMT-based targeting in identifying the poor (Alatas et al. 2012). This result for community-based targeting does not appear to be due to elite capture but rather due to communities defining poverty more broadly than simply low consumption as the PMT aims to predict. Importantly, community-based targeting resulted in fewer reported complaints, and program delivery was more efficient in communities where such targeting was applied. In sum, community satisfaction was higher with community-based targeting than PMT-based targeting. In contrast, evidence from Niger shows that PMT-based targeting performed better than community-based targeting in identifying the poor, and that a cash transfer program had a bigger impact on the welfare of recipients under PMT-based than community-based targeting (Premand and Schnitzer 2020). The study also finds that communities viewed PMT-based targeting and another formula-based targeting approach to identify food-insecure households as having greater legitimacy than community-based targeting. Community-based targeting was seen by communities as subject to greater manipulation and bias than the formula-based targeting methods. Notwithstanding, small design tweaks to the process of PMT-based targeting may produce appreciable gains in program targeting performance. Evidence from Indonesia suggests that imposing a small cost to potential beneficiaries, such as requiring them to go to the local welfare office to take the PMT survey (instead of being administered the survey at home), can improve the accuracy of program targeting as the nonpoor self-select out from going to the office (Alatas et al. 2016). Shocks and social assistance Providing a list of specific shocks, the SHIES 2016/17 asks households to report a maximum of the three most severe ones experienced in the past five years. The survey also asks how long ago the shock occurred in months and years. We record all shocks reported by households and when they occurred based directly on household responses to these questions. On average, households report a little less than two shocks over the past five years and a little more than one in the past year. Aside from investigating the time pattern in the reports of shocks using data on reported shocks in the past five years, we restrict our analysis to shocks reported to have occurred in the past year, under the assumption that the recall of these shocks is more accurate. As a caveat to the results that follow, the information on shocks, reported by households, may be biased. Households that experience a large negative effect on their socioeconomic status due to a shock may be more likely to report the event than households that successfully prevent the shock from producing a negative effect. This would contribute to an undercount of households experiencing the shock. “Drought or floods” (hereafter, drought shock) 55 and a “large rise in price for food” (hereafter, food price shock) dominate reported shocks. Among all reported shocks across households for the past year, drought shock accounts for 33 percent and food price shock for another 33 percent. Their 55 Information from the various reports of the Swaziland Vulnerability Assessment Committee suggests that the extent of households reporting flood shocks was negligible over the 2010s. 31 high shares in household responses are not surprising given the El Niño-induced drought of 2015/16, which had large, widespread negative effects on crop and livestock production, drinking water, sanitation and hygiene, and food security (Swaziland VAC 2016, 2017). The share for each of the other shocks (15 specific shocks and an “other” shock option) is 7 percent or less (the modal share is 1 percent). Hence, we combined these shocks into an other shock category, constituting 34 percent of all reported shocks. 56 The strong association between the likelihood of reports of drought shock or food price shock and the 2015/16 El Niño event is indicated in figure 25, which plots the distribution of reported shocks by month and year of occurrence. The shares of reports of drought and food price shocks rise sharply between mid-2015 and mid-2016, a period which straddles the affected 2015/16 agricultural season. To understand who might be more susceptible to shocks, we explore the association between the likelihood of reporting a shock and four household characteristics: area of residence (urban or rural), household poverty status, household food insecurity status, and receipt of government social assistance program benefits, separately by program. While all the results we present relate to the likelihood of reporting a shock, for brevity, we sometimes contract this to “the likelihood of a shock.” Table 4 reports coefficients from bivariate regressions of whether the household experiences a given shock type on rural residence, poverty status, or food insecurity status. Food price shock is associated with area of residence but not drought shock or other shock. Rural households are 18 percent less likely to report a food price shock than urban households. Poor households are more likely to report a drought shock or other shock than nonpoor households but are less likely to report a food price shock. Food-insecure households are more likely to report a drought shock or other shock, but not a food price shock. Examination of the association between shocks, poverty status, and food insecurity status separately in urban and rural areas indicates that the positive association between poverty status and drought shock or other shock is only observed among rural households, and that the negative association between poverty status and food price shock is due to urban households. Drought shock and other shock are positively associated with food insecurity status among both urban and rural households, while the positive association between food price shock and food insecurity status is driven by urban households. Several reasons can be posited for why we might expect shocks to be associated with the receipt of social assistance program benefits. As one explanation, social assistance programs may be adjusted to provide relief when a shock occurs, that is, they exhibit ex-post responsiveness frequently referred to as adaptive social protection or shock-responsive social protection. As another explanation, there may be household or individual characteristics that cause households to both receive program benefits and be vulnerable to shocks—thus, resulting in ex-ante social assistance program coverage. Table 5 reports coefficients from bivariate regressions of whether a household experienced a given shock type on whether the household received a given social assistance program benefit. In principle, any documented positive associations could result from ex-ante coverage, ex-post 56 For details on the construction of the shock indicators, see appendix 1. 32 responsiveness (through, for example, adjustments to coverage, benefit levels, or the timing of benefits to help households cope with the shock), or the combination of the two. Available reports indicate that, in practice, the government extended and intensified its food aid program in response to the drought and food price shocks, while the other programs we examine were not adjusted to help households cope with these and other shocks. In general, across programs, households that report receiving benefits are more likely to report a drought shock. In addition, across some programs, namely NCPs, school feeding, and food aid, we find the same result with respect to other shock. Except in the case of NCPs, households that report receiving program benefits are less likely to report a food price shock. The positive associations between drought shock and program receipt status are driven by rural households (table A3.6). Significant positive associations between drought shock or other shock and program receipt status are more common among the nonpoor than the poor (table A3.7) and among the food-secure than the food-insecure (table A3.8). The negative association between food price shock and program receipt status is driven by the fact that, while programs are more likely to cover the poor, the nonpoor are more likely to report a food price shock than the poor as we noted earlier. The survey also asks households that report shocks to indicate their main coping strategies. Based on this information, we construct four indicators of assistance from different sources: no assistance, assistance from family or friends, assistance from NGOs or religious institutions, or assistance from government.57 A response by a government social assistance program to a shock is presumably captured by the indicator for assistance from government. Across the three shock types (drought, food price, and other), 71–92 percent did not receive any assistance (figure 26). When households did receive assistance, it almost always came from family or friends. Assistance from family or friends is more common for other shock (23.8 percent) than drought shock (7.7 percent) or food price shock (6.8 percent), presumably because other shock subsumes different idiosyncratic shocks, which family members or friends may be able to help the affected household cope with. Across shock types, less than 4 percent of households report receiving government assistance. An examination of these patterns separately by area of residence, poverty status, and food insecurity status (figure A3.1) shows that in its response to other shock, government assistance is concentrated in urban areas (where services such as government health services for illness or injury shocks are more easily accessible) while it is concentrated in the response to drought shock among rural households. Further, in the response to drought shock, government assistance is concentrated among the poor and the food-insecure. Available reports indicate that assistance provided by government to households and communities acutely affected by the 2015/16 drought mainly included the distribution of food, water, and animal feed (NDMA 2015, Swaziland VAC 2016, Swaziland VAC 2017). 58 Nevertheless, based on the 57 For details on variable construction, see appendix 1. 58 Coincidentally, the Eswatini government initiated a pilot program in 2016 to provide unconditional cash benefits to poor households with orphaned and vulnerable children in four Tinkhundlas (one in each region), financed by the 33 survey data, it appears that government and other institutions are not responding to shocks in ways that make their interventions the main coping strategies for an appreciable share of affected households V. Conclusion Based on recent observational data, specifically government administrative and household sample survey data, this study assesses the performance of Eswatini’s main social assistance programs. It examines the levels of program spending; program coverage rates, including in relation to poverty status and food insecurity status; program incidence and effectiveness in relation to poverty and inequality; the poverty and inequality effects of simulated reforms in programs; and the association of program participation with household exposure to negative shocks. Below, we discuss recommendations for future data, research, and policy. Recommendations for future data and research Undertaking evaluative research. Given the nature of the data and empirical strategies employed, the analysis is correlational. Causal evidence is vital for government decisions on which programs to retain, which to scale up, and which to reform for greater potential impact. Such evidence on the effects of programs will require evaluative research based on appropriate data and identification strategies that allow for causal inference. To address this critical knowledge gap, the Eswatini government should evaluate key social assistance programs. 59 Criteria for selecting a program for an impact evaluation could be related to size of budgetary allocation, extent of program coverage, potential for having a sizable impact, demand from implementing ministries and agencies for increased funding, or a proposed critical reform in program design. Where possible, impact evaluations should include certain design features. Specifically, evaluations should investigate: (a) longer-term program impacts; (b) differences in program impacts by population subgroup such as by gender, age group, poverty status, and geographic location; (c) independent and combined impacts of key program subcomponents; (d) pathways behind program impacts; (e) impact spillovers, negative or positive, on program nonparticipants; (f) strategic behavioral responses (or “gaming”) by program participants that influence impacts; (g) impacts on local labor markets, prices, and communities; and (h) cost-effectiveness. Undertaking diagnostic operational research . Program operational reviews should be conducted in tandem with impact evaluations, to potentially provide insights into design and implementation factors behind the presence or absence of program impacts (including through the investigation of European Union and the World Bank. In May 2016, the program made the first quarterly payment, reaching 5,518 beneficiary children in 1,312 households (World Bank 2016). Given the timing of its launch, beneficiary households may have viewed the program as a response by the government to the drought emergency. 59 Rigorous impact evaluations are rare in Eswatini but not absent. A case in point is the evaluation of the Sitakhela Likusasa intervention, based on a field experiment (Gorgens et al. 2020). The intervention offered cash incentives to adolescent girls and young women to attend school and to test negative for sexually transmitted infections. 34 pathways behind program impacts, as noted in the previous paragraph). Such operational reviews, covering the foundational elements of and links in the entire program delivery chain (Leite et al. 2017; Lindert et al. 2020), are also useful independent of impact evaluations, because they can help to identify shortcomings in program design parameters and design-implementation gaps. Assessing the potential welfare effect of adverse shocks on households and the potential ameliorative countereffect of programs. An important matter that we do not explore in the study is whether shocks—especially drought and food price shocks—have a negative effect on consumption and whether receipt of social assistance program benefits moderate this effect. As noted earlier, self-reported shocks data may be biased, which complicates any analysis of such effects based on the SHIES 2016/17. A potentially more credible approach involves linking spatially fine data on rainfall and market food prices to the SHIES 2016/17 data (or another appropriate household survey) to assess the effects of extreme changes through these more objective measures. Of course, if the data-generating process for the evaluation appropriately intersects with the occurrence of a shock over space and time, rigorous evaluations can also be used to uncover whether a social assistance program moderates the negative impact of the shock. Strengthening the design and implementation of social protection modules in national household sample surveys. While the SHIES 2016/17 is fairly well designed and appears to have been well administered, the collected data are not especially optimized for an analysis of the performance of social protection programs. In Eswatini and elsewhere, the main purpose of income and expenditure surveys or living standards surveys is to estimate income or consumption for poverty and inequality analysis, not to analyze the performance of social protection programs. Yet small changes in questionnaire design (or in the programming for computer-assisted personal interviewing [CAPI]) could make social protection program data from these surveys more reliable. Specifically, the SHIES could make the following changes to strengthen the collection of social protection program data: (a) list the full set of social protection programs in existence at the time of the survey; (b) specify each social protection program individually when collecting data on the receipt of program benefits; (c) ask survey respondents about their awareness of a specific social protection program (providing all common formal and informal names) before asking whether they received benefits from that program; (d) ask survey respondents well-framed questions to precisely identify program beneficiaries (when beneficiaries are individuals, the questionnaire should ask who has benefited, allowing for the possibility that more than one household member could be a beneficiary; when the beneficiary is the household, the questionnaire should include an option for household rather than individual beneficiaries); and (e) ensure that survey interviewers are trained and have detailed information readily accessible on key design and implementation features of each social protection program and that the CAPI software for the survey also incorporates this information. It would be important to ensure coordination between the ministries and agencies involved in administering social protection programs and the Central Statistical Office (CSO) when preparing the SHIES questionnaire, such as for the next round of the survey. The ministries and agencies could ask the CSO to include questions aimed at soliciting information that would be useful in assessing their respective programs. They should provide the CSO guidance and training on program design and implementation characteristics that would improve the accuracy of the data 35 collected. Some of the external financial assistance for Eswatini’s social protection programs could go toward promoting this coordination between ministries and agencies involved in administering social protection programs and the country’s statistics authority. Recommendations for future policy Addressing benefit-related disconnects. An analysis of program performance, coupled with information on program design and implementation discussed in the study, suggests that a primary challenge facing Eswatini's social assistance programs centers around program intensity rather than scale or reach. This challenge hampers the ability of programs to effectively address inequality, poverty, and the negative effects of shocks. Benefit-related disconnects currently abound in the government’s social assistance program portfolio. In some cases, actual benefit levels are lower than intended benefit levels. For example, the amount of food procured for the school feeding program is inadequate. Similarly, shortfalls in food procured and distributed to NCPs disrupt their operations. In other cases, intended benefit levels are inconsistent with the aims of the program. For example, the OVC education grant amount may not fully defray the tuition costs of the beneficiary. Likewise, the assigned portion of the per-student grant under the Free Primary Education initiative may not fully defray the operational costs of school feeding in primary schools. Some benefits may not be sufficiently encompassing. For example, estimates based on the SHIES 2016/17 show that out-of-pocket expenses, including for transportation, uniforms, textbooks, and other materials, are significant for OVC education grant beneficiaries. While OVC education grants aim to defray the cost of tuition and exam fees, these other important schooling-related costs may discourage poor orphans and vulnerable children from attending school. Improving the targeting of programs. Absent additional financing for the programs, correcting benefit-related disconnects may require the government to improve targeting of the programs to reach the neediest populations. Greater program intensity may need to be traded off against reduced program coverage. The analysis in the study uses PMT-based targeting to demonstrate potential gains from targeting OVC education grants and elderly grants to the PMT-poor. These gains include expenditure savings and the additional positive effects on poverty and inequality from reallocating expenditure savings to higher program benefits for the PMT-poor. Which targeting method is actually used by the government to administer programs should be evaluated based on multiple criteria. These criteria should include targeting accuracy, but also comprise ease and efficiency of implementation, legitimacy, and transparency. Increasing the financing of programs. Notwithstanding improved program targeting to generate expenditure savings, the government should consider increasing financing for social assistance programs. Compared to many other developing countries, Eswatini spends markedly less on social assistance programs, even as its levels of poverty and inequality are much higher. The government’s fiscal policy is viewed as having a limited bearing on the country’s poverty and inequality challenges. Reforms of public financial management and procurement systems and the rationalization of personnel compensation, extrabudgetary entities, and public enterprises can create the fiscal space needed to increase social spending (IMF 2020). 36 Strengthening the administration of programs. The administrative system of the government’s social assistance programs requires a major upgrade. Improvements needed span the entire delivery chain, from communication and outreach to targeting and enrollment to benefit transfer processes, case and grievance management, and monitoring. A key part of the upgrade should include the development and implementation of a single integrated social registry, along with an integrated management information system (MIS) to implement and monitor these programs. To help promote efficiency, integrity, and coordination, the registry and MIS should be used (a) by all government ministries and agencies that administer social assistance programs; (b) by relevant tiers of government (and supporting private actors) from national to local levels, that are engaged in the implementation and monitoring of social assistance programs; and (c) for interventions implemented outside of government by international and national donors and NGOs. It may also be useful to integrate the government’s social insurance programs and social welfare services into these systems. Related to strengthening program administration, the DPMO action plan up to 2022 (GOS 2013) emphasizes the development and implementation of a system for program monitoring. The lack of such a system is seen as a major weakness (UNICEF 2018c). Indeed, strong performance in collecting, managing, and using monitoring data would be instrumental in ensuring the government has actionable information on whether to maintain a program, to adapt it to address any unintended, adverse effects, or to augment intended effects. In a recent review of DPMO’s programs, UNICEF proposes a stronger link between social assistance programs and social welfare services, so that, for example, social welfare workers could proactively reach beneficiaries of elderly grants or disability grants and provide them with needed social welfare services (UNICEF 2018c). If well financed, well designed, and well implemented, these links can raise the joint performance of social assistance programs and social welfare services. The orphaned and vulnerable children cash transfer (OVC-CT) program in some ways can be treated as proof-of-concept of key building blocks toward a next-generation social assistance administrative system. The OVC-CT program was a small-scale, two-year pilot program administered by DPMO. 60 It provided unconditional cash transfers to needy households with orphaned and vulnerable children ages 0–18 years, reaching 15,920 children by the end of the program’s implementation period in 2018. The administration of the program had more contemporary features, such as PMT-based targeting of households, benefit transfers to households through mobile payments, and a digital MIS that encompassed the entire program delivery chain, including case management, grievance redressal, and monitoring (World Bank 2019). The 60 The OVC-CT program was financed by the European Union and the World Bank. It was administered in four Tinkhundlas (one per region). The program aimed to make quarterly cash payments to qualifying households, via mobile payments. Seven rounds of payments were made in total over the life of the pilot, with the first round in May 2016 (made to 5,518 OVCs) and the seventh round in December 2017 (made to 13,506 OVCs); a final, special payment was made to 2,414 additional OVCs in June 2018, bringing the total number of program beneficiaries to 15,920 OVCs (data obtained from various World Bank implementation status and results [ISR] reports for the Swaziland Health, HIV/AIDS, and TB Project, under which the pilot program was financed). Program benefit amounts were a function of the age group of the OVC and the number of OVCs in the household, with three possible amounts disbursed to each OVC every quarter: E300, E450, or E600. The actual average quarterly payment per OVC ranged between E450 and E500 across payment rounds (DPMO 2018). This range is 32–36 percent of the lower poverty line (quarterly, per adult equivalent) and 15–17 percent of the upper poverty line (quarterly, per adult equivalent), in January 2017 prices. The program was discontinued after donor financing for it ended. 37 program was accompanied by a rigorous impact evaluation (DPMO 2018). Indeed, the experience with mobile payments under the program has informed the adoption of this approach for other interventions in Eswatini (World Bank 2019). Rethinking the mix of programs. The Eswatini government may want to reconsider its current mix of social assistance programs. This reconsideration should seek to ensure that programs (their aims and designs) are consistent with an up-to-date profile of the main welfare risks, including adverse shocks, experienced by the poor and other disadvantaged sociodemographic groups. One approach that the government could use to determine the appropriate combination of programs is by applying a lifecycle framework to identify welfare risks faced by individuals. In rethinking the combination of programs, the government should take into account the whole landscape of current and expected social welfare services, social insurance programs, labor market programs, and other development interventions in Eswatini. It should aim to address gaps or weaknesses (i.e., uncovered or weakly covered adverse conditions or risks) through its social assistance programs and forge (stronger) links and enhance synergies between its social assistance programs and the other programs and services. Further, in rethinking the combination of social assistance programs, the government should reconsider the balance between cash versus noncash transfers. Four of the five main social assistance programs examined in the study offer food or other in-kind benefits. Only the elderly grants program offers cash benefits. 61 The government should also consider the balance between programs that offer nonlabor income transfers versus programs that support labor income. Programs such as productive inclusion and labor-intensive public works fall under the latter type. Advancing the development of the system in stages. Given the current state of Eswatini’s social assistance system and the political, financial, and administrative constraints the system faces, it may be most practical to introduce improvements in stages. In the first stage, the government could address benefit-related disconnects in the existing social assistance programs. The second stage, staggered with the first stage, would involve strengthening the social assistance administrative system. In the third stage, the government could rethink the mix of social assistance programs. Strengthening the system’s resilience and responsiveness to adverse shocks. An integral part of advancing the development of the social assistance system would be to make the system more resilient and responsive to adverse shocks. DPMO’s action plan up to 2022 (GOS 2013) calls for a national disaster early warning system and national disaster preparedness and response plans. Given the expected frequency, scale, and severity of covariate natural and other shocks in the country, the Eswatini government should prioritize these initiatives. A basic prerequisite for effective shock responsiveness is sufficient, timely, and well-executed emergency government spending when a major shock occurs. The Eswatini government is fiscally constrained in large part due to public financial management weaknesses (IMF 2020). These weaknesses may be especially debilitating when a major shock occurs and the government needs to respond. The government needs to improve its capability to reprogram planned spending; to activate and earmark additional financing from its own and external sources; to streamline budget execution processes for fast disbursement; and to find innovative ways to track the integrity of emergency spending. 61 Disability grants and military pensions are also cash transfer programs, but these programs have small outlays and beneficiary numbers. 38 In addition to the usual shocks the country may experience, the COVID-19 pandemic presents an unprecedented challenge for Eswatini and the globe. It also presents an opportunity for self- reflection by the Eswatini government on the future form and substance of its social protection programs and the social protection system at large. The Eswatini government reported its first confirmed COVID-19 case on March 14, 2020. As of May 31, 2021, there have been 18,591 confirmed cases of COVID-19 with 673 deaths in Eswatini, reported by the government to the World Health Organization. 62 The government introduced infection prevention and control measures, including social distancing, limits on public gatherings and public engagements, the closure of schools and universities, and a partial lockdown of society. These actions have adversely affected overall economic activity as well as social assistance program administration and delivery. The poor and vulnerable are believed to have experienced significant income losses due to the measures introduced by the government, while also experiencing a rise in their consumption burden (for example, due to the cessation of school meals when schools were closed). Hygiene guidelines are also believed to have imposed an additional burden on the poor and vulnerable, who lack access to improved water, sanitation, and hygiene amenities. A COVID-19 response plan was developed by Eswatini’s social protection cluster, led by DPMO and co-led by UNFPA. To date, the main social assistance-related actions taken by the Eswatini government include the emergency provision of food assistance to poor, vulnerable, and affected households by NDMA following categorical targeting of households (in form and substance similar to the government’s emergency food aid program examined in the study). The food assistance was supplemented by the provision of cash assistance to other target households by the Baphalali Eswatini Red Cross Society and WFP. The actions also include cash assistance to unpaid, laid-off workers from formal enterprises through the Eswatini National Provident Fund. The pandemic has also prompted government deliberations on whether to introduce an unemployment insurance program on a permanent basis. The COVID-19 pandemic reinforces the case for establishing a robust, flexible, and effective social assistance system. The findings from the program performance analysis in the study provide insight on how Eswatini can achieve this goal. Indeed, the social assistance and social security policies and associated implementation plans, when finalized and approved, can serve as a basis for building such a system. 62 See https://www.who.int/countries/swz/ (Accessed: May 31, 2021). 39 References Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, Ririn Purnamasari, and Matthew Wai-Poi. 2016. “Self-Targeting: Evidence from a Field Experiment in Indonesia.” Journal of Political Economy 124 (2): 371–427. Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, and Julia Tobias. 2012. “Targeting the Poor: Evidence from a Field Experiment in Indonesia.” American Economic Review 102 (4): 1206–40. Blank, Lorraine, Emma Mistiaen, and Jeanine Braithwaite. 2012. “Swaziland: Using Public Transfers to Reduce Extreme Poverty.” Social Protection and Labor Discussion Paper 1411, World Bank, Washington, DC. Andrews, Colin, Aude de Montesquiou, Ines Arevalo Sanchez, Puja Vasudeva Dutta, Boban Varghese Paul, Sadna Samaranayake, Janet Heisey, Timothy Clay, and Sarang Chaudhary. 2021. The State of Economic Inclusion Report 2021: The Potential to Scale. Washington, DC: World Bank. Boko, Joachim, Dhushyanth Raju, and Stephen Younger. 2021. “Welfare, Shocks, and Public Spending on Social Protection Programs in Lesotho.” Social Protection and Jobs Discussion Paper 2102, World Bank, Washington, DC. Bourguignon, François, and Luiz A. Pereira da Silva, eds. 2003. The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools. Washington, DC: World Bank. Central Statistics Office. 2017. The 2017 Population and Housing Census: Preliminary Results. Mbabane: Central Statistics Office, The Government of the Kingdom of Swaziland. Central Statistical Office and UNICEF (United Nations Children’s Fund). 2016. Swaziland Multiple Indicator Cluster Survey 2014. Final Report. Mbabane: Central Statistical Office and UNICEF. Coates, Jennifer, Anne Swindale, and Paula Bilinsky. 2007. Household Food Insecurity Access Scale (HFIAS) for Measurement of Food Access: Indicator Guide, version 3. Washington, DC: Food and Nutrition Technical Assistance Project, Academy for Educational Development. de Walque, Damien. 2020. “The Use of Financial Incentives to Prevent Unhealthy Behaviors: A Review.” Social Science Medicine. doi: 10.1016/j.socscimed.2020.113236. Dlamini, Armstrong. 2007. “A Review of Social Assistance Grants in Swaziland: A CANGO/RHVP Case Study on Public Assistance in Swaziland.” Case study. Available at: http://www.rodra.co.za/images/countries/eswatini/research/A%20review%20of%20social %20assistance%20grants%20in%20Swaziland.pdf. DPMO (Deputy Prime Minister’s Office). 2018. Swaziland Orphans and Vulnerable Children Cash Transfer Programme (OVC-CT): Endline Study 2018 Report. Mbabane: DPMO, Government of the Kingdom of Swaziland. Dupas, Pascaline, and Edward Miguel. 2017. “Chapter 1 - Impacts and Determinants of Health Levels in Low-Income Countries.” In Handbook of Field Experiments, Vol. 2, edited by Abhijeet Banerjee and Esther Duflo, 3–94. Amsterdam: Elsevier/North Holland. Enami, Ali. 2018. “Measuring the Effectiveness of Taxes and Transfers in Fighting Inequality and Poverty.” In Commitment to Equity Handbook: Estimating the Impact of Fiscal Policy on Inequality and Poverty, edited by Nora Lustig, 207-218. Washington, DC: Brookings Institution Press. 40 Eswatini VAC (Vulnerability Assessment Committee). 2018. Kingdom of Eswatini Annual Vulnerability Assessment and Analysis Report 2018. Mbabane: Eswatini VAC. Figari, Francesco, Alari Paulus, and Holly Sutherland. 2015. “Microsimulation and Policy Analysis.” In Handbook of Income Distribution , Vol. 2B, edited by A. B. Atkinson and F. Bourguignon: 2141–221. Amsterdam: Elsevier/North-Holland. Foster, James, Joel Greer, and Erik Thorbecke. 1984. “A Class of Decomposable Poverty Measure.” Econometrica 3 (52): 761–6. Fryer, Roland. G. 2017. “Chapter 2 - The Production of Human Capital in Developed Countries: Evidence From 196 Randomized Field Experiments.” In Handbook of Field Experiments, Vol. 2, edited by Abhijeet Banerjee and Esther Duflo, 95–322. Amsterdam: Elsevier/North Holland. Glewwe, Paul, and Karthik Muralidharan. 2016. “Chapter 10 - Improving Education Outcomes in Developing Countries: Evidence, Knowledge Gaps, and Policy Implications.” In Handbook of the Economics of Education , Vol. 5, edited by Eric A. Hanushek, Stephen Machin, and Ludger Woessmann, 653–743. Amsterdam: Elsevier/North Holland. GOE (Government of the Kingdom of Eswatini). 2013. His Majesty’s Government Programme of Action 2013 – 18: Ministries’ Action Plans to 2018 and 2022 . Mbabane: GOE. –––. 2017. Multi Dimensional Child Poverty in Kingdom of Eswatini. Mbabane: Ministry of Economic Planning and Development, The Government of the Kingdom of Eswatini. –––. 2018. The National Multisectoral HIV and AIDS Strategic Framework (NSF) 2018–2023. Mbabane: The Government of the Kingdom of Eswatini. –––. 2019a. National Development Plan 2019/20–2011/22: Towards Economic Recovery. Mbabane: Ministry of Economic Planning and Development, the Government of the Kingdom of Eswatini. –––. 2019b. The Kingdom of Eswatini Strategic Road Map: 2019–2022 . The Government of the Kingdom of Eswatini. –––. 2019c. Swaziland HIV Incidence Measurement Survey 2 (SHIMS2) 2016– 2017. Final Report. Mbabane: The Government of the Kingdom of Eswatini. Gorgens, Marelize, Andrew F. Longosz, Sosthenes Ketende, Muziwethu Nkambule, Tengetile Dlamini, Mbuso Mabuza, Kelvin Sikwibele, et al. 2020. “Evaluating the Effectiveness of Incentives to Improve HIV Prevention Outcomes for Young Females in Eswatini: Sitakhela Likusasa Impact Evaluation Protocol and Baseline Results.” BMC Public Health 20. GOS (Government of the Kingdom of Swaziland). 2005. National Food Security Policy. Mbabane: Ministry of Agriculture and Cooperatives, Government of the Kingdom of Swaziland. –––. 2007. The Swaziland Poverty Reduction Strategy and Action Plan (PRSAP). Mbabane: Ministry of Economic Planning and Development, Government of the Kingdom of Swaziland. –––. 2010. National Social Development Policy. Mbabane: Deputy Prime Minister’s Office, Government of the Kingdom of Swaziland. –––. 2011. National Disaster Risk Management Policy. Mbabane: Government of the Kingdom of Swaziland. Hanna, Rema, and Dean Karlan. 2017. “Chapter 7 – Designing Social Protection Programs: Using Theory and Experimentation to Understand How to Help Combat Poverty.” In Handbook of Field Experiments, Vol. 2, edited by Abhijeet Banerjee and Esther Duflo, , 515–53. Amsterdam: Elsevier/North Holland. 41 Haushofer, Johannes, and Ernst Fehr. 2014. “On the Psychology of Poverty.” Science 344 (6186): 862–7. IMF (International Monetary Fund). 2019. Sub-Saharan Africa Regional Economic Outlook: Navigating Uncertainty. Washington, DC: IMF. –––. 2020. Kingdom of Eswatini: 2019 Article IV Consultation-Press Release; Staff Report; and Statement by the Executive Director for the Kingdom of Eswatini. IMF Country Report No 17/274. Washington, DC: IMF. –––. 2021. Regional Economic Outlook for Sub-Saharan Africa: Navigating a Long Pandemic. Washington, DC: IMF. Inter Agency Standing Committee and the European Commission. 2020. INFORM Report 2020. Shared Evidence for Managing Crisis and Disaster. Publications Office of the European Union, Luxembourg. Kremer, Michael, Gautam Rao, and Frank Schilback. 2019. “Behavioral Development Economics.” In Handbook of Behavioral Economics – Foundations and Applications 2, Vol 2., edited by Douglas Bernheim, Stefano DellaVigna, and David Laibson, 345–458. Amsterdam: Elsevier/North Holland. Lanjouw, Peter, and Martin Ravallion. 1999. “Benefit Incidence, Public Spending Reforms, and the Timing of Program Capture.” World Bank Economic Review 13 (2): 257–73. Leite, Phillippe, Tina George, Changqing Sun, Theresa Jones, and Kathy Lindert. 2017. “Social Registries for Social Assistance and Beyond: A Guidance Note & Assessment Tool.” Social Protection & Labor Discussion Paper 1704, World Bank, Washington, DC. Lindert, Kathy, Tina George Karippacheril, Ines Rodriguez Caillava, and Kenichi Nishikawa Chavez. 2020. Sourcebook on the Foundations of Social Protection Delivery Systems. Washington, DC: World Bank. Lustig, Nora, ed. 2018. Commitment to Equity Handbook: Estimating the Impact of Fiscal Policy on Inequality and Poverty. New Orleans: CEQ Institute at Tulane University; Washington, DC: Brookings Institution Press. McBride, Linden, and Austin Nichols. 2018. “Retooling Poverty Targeting Using Out-of-Sample Validation and Machine Learning.” World Bank Economic Review 32 (3): 531–50. Mullainathan, Sendhil, and Eldar Shafir. 2013. Scarcity: Why Having Too Little Means So Much. New York: Henry Holt and Company. Muralidharan, Karthik. 2017. “Chapter 3 - Field Experiments in Education in Developing Countries.” In Handbook of Field Experiments, Vol. 2, edited by Abhijeet Banerjee and Esther Duflo, 323–85. Amsterdam: Elsevier/North Holland. NDMA (National Disaster Management Agency). 2015. Swaziland Comprehensive National Drought Response Plan: April 2016 to March 2017 . Mbabane: NDMA. Oosthuizen, Morne. 2020. “South Africa: Social Assistance Programs and Systems Review.” Manuscript. Premand, Patrick, and Pascale Schnitzer. 2020. “Efficiency, Legitimacy, and Impacts of Targeting Methods: Evidence from an Experiment in Niger.” World Bank Economic Review: 1–29. Raju, Dhushyanth, and Stephen D. Younger. 2020. “Assessment of the Government of Ghana’s Current Proxy Means Test Model: Technical Note.” Manuscript. Ravallion, Martin, and Shaohua Chen. 2015. “Benefit Incidence with Incentive Effects, Measurement Errors and Latent Heterogeneity: A Case Study for China.” Journal of Public Economics 128: 124–32. 42 Swaziland VAC (Vulnerability Assessment Committee). 2016. Swaziland Annual Vulnerability Assessment and Analysis Report 2016. Mbabane: Swaziland VAC. –––. 2017. Swaziland Annual Vulnerability Assessment and Analysis Report 2017 . Mbabane: Swaziland VAC. UK DFID (United Kingdom Department for International Development), HelpAge International, and UNICEF (United Nations Children’s Fund). 2010. Swaziland Old Age Grant Impact Assessment. Mbabane: UK DFID, HelpAge International, and UNICEF. UNICEF (United Nations Children’s Fund). 2017. Social Protection Budget: Swaziland 2017/18. Mbabane: UNICEF. –––. 2018a. Fiscal Space for Children: An Analysis of Options in Eswatini. Mbabane: UNICEF. –––. 2018b. Kingdom of Eswatini: Social Assistance Budget Brief 2018/2019. Mbabane: UNICEF. –––. 2018c. Quantitative Assessment of the Social Assistance System in the Kingdom of Eswatini. Mbabane: UNICEF. van de Walle, Dominique. 1998. “Assessing the Welfare Impacts of Public Spending.” World Development 26 (3): 365–79. WFP (World Food Programme) in Swaziland. 2015. Assistance to Orphaned and Vulnerable Children at NCPs and Schools: Standard Project Report 2015. Mbabane: WFP. –––. 2016. Assistance to Orphaned and Vulnerable Children at NCPs and Schools: Standard Project Report 2016 . Mbabane: WFP. –––. 2017. Assistance to Orphaned and Vulnerable Children at NCPs and Schools: Standard Project Report 2017 . Mbabane: WFP. –––. 2019. Decentralized Evaluation: Evaluation of National School Feeding Programme in Eswatini 2010 – 2018: Final Evaluation Report. Mbabane: WFP. World Bank. 2016. Swaziland Health, HIV/AIDS and TB Project: Implementation Status & Results Report (June 2016). Washington DC: World Bank. –––. 2019. Swaziland Health, HIV/AIDS and TB Project: Implementation Status & Results Report (June 2019). Washington DC: World Bank. –––. 2020. The Kingdom of Eswatini: Toward Equal Opportunity: Accelerating Inclusion and Poverty Reduction in Eswatini: Systematic Country Diagnostic. Washington DC: World Bank. Yemtsov, Ruslan, Maddalena Honorati, Brooks Evans, Zurab Sajaia, and Michael Lokshin. 2018. Measuring the Effectiveness of Social Protection: Concepts and Applications. Streamlined Analysis with ADePT Software series. Washington, DC: World Bank. 43 Figure 1. National Income Trends between 2010 – 18 Source: World Bank’s World Development Indicators databank (https://databank.worldbank.org/source/world- development-indicators) Note: GDP = Gross Domestic Product. PPP = purchasing power parity. 44 Figure 2. Poverty Levels Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Poverty rate, poverty gap, and poverty severity, and the upper and lower poverty lines, are defined in appendix 1. 45 Figure 3. Poverty Rates, by Subgroups a. By area and region b. By other subgroups Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Overall and extreme poverty rates are defined in appendix 1. Statistics by orphan status are restricted to individuals below age 18. In panel a, within subgroup, observations are organized in ascending order by the extreme poverty rate. hh = household. 46 Figure 4. Food Insecurity Rates, by Subgroups a. By area and by region b. By poverty status Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Categories are based on the Household Food Insecurity Access Scale (HFIAS). For the definition of the food insecurity categories and HFIAS, see appendix 1. In panel a, within subgroup, observations are organized in ascending order by the incidence of severe insecurity. 47 Figure 5. Gini Index, by Area and by Region, 2016/17 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Within subgroup, observations are organized in ascending order by Gini index. 48 Figure 6. Fiscal Position of the Central Government of Eswatini, 2015/16 – 2018/19 Source: IMF 2020. Note: SACU = Southern African Customs Union. FY = government’s fiscal year. Values for FY2018/19 are preliminary. 49 Table 1. Key Design Features of Main Social Assistance Programs in Eswatini, 2015/16 and 2016/17 Program Implementing agency Year initiated Geographic coverage Benefit structure Benefit eligibility criteria (1) (2) (3) (4) (5) (6) Neighborhood Care • Administrator: The • 2001 • National (1,500–1,700 NCPs • Type: In-kind. One cooked • Self-selection. Targeted at needy, Points (NCPs) World Food in recent years. 97% of NCPs meal per day and basic early young orphan or vulnerable Programme with are rural, and 3% are peri- child care and development children in the community who coordination support urban services at community- are out of school from the Tinkhundla organized and -run centers Department of • Periodicity: Daily Administration and (weekdays) Development and • Beneficiary: Individual monitoring support from Save the Children • NCPs are staffed by community volunteers • Financer: World Food Programme School feeding • Primary • 2010 (start of • National (844 schools) • Type: In-kind. One cooked • Enrolled in a government primary, administrator: full financing meal per day, with meal secondary, or high school Ministry of Education of the program ration amounts per student and Training by the per school day stipulated by • Financer: Government government in MOET. Cereals, cooking oil, of the Kingdom of government pulses, and peanut butter Eswatini primary procured and distributed to schools) schools. Funds transferred electronically to the school’s bank account; government primary schools also receive an annual per-student amount of E150, which is expected to cover school operational costs for the feeding program and another annual per-student amount of E150, which is expected to cover the salaries of school support staff, including cooks for the program. These amounts are part of the annual per- student grant of E560 provided by MOET to 50 Table 1. Key Design Features of Main Social Assistance Programs in Eswatini, 2015/16 and 2016/17 Program Implementing agency Year initiated Geographic coverage Benefit structure Benefit eligibility criteria (1) (2) (3) (4) (5) (6) government primary schools for the Free Primary Education initiative • Periodicity: Every school day (200 school days per year) • Beneficiary: Individual Food aid • Primary • 2011 • National, in select areas • Type: In-kind. Pulses, • Food-insecure households as administrator: experiencing high food cooking oil, cereals determined by local government National Disaster insecurity • Periodicity: During food officials (Tinkhundla or local Management Agency insecurity emergencies social welfare officials), • Financer: Government • Beneficiary: Household community disaster committees, of the Kingdom of and partnering NGOs/CSOs in the Eswatini field OVC education grant • Primary • 2003 • National • Type: In-kind. Electronic • Needy single or double orphan, administrator: transfer to school bank needy vulnerable child, Department of Social account for each beneficiary unemployed parents with no Welfare, Deputy student source of income, child without Prime Minister’s • Purpose: To defray the cost parents who is staying with needy Office of tuition and exam fees grandparents, child-headed • Financer: Government • Amount: Form I-III (grades household, mentally or terminally of the Kingdom of 8–10) = E1,950 per year; ill parents, disabled parents, Eswatini Form IV (grade 11) = disabled child who is able to E2,500 per year; attend school; Form V (grade 12) = E2,500 • Certification of OVC status by per year + up to E2,000 for local community leadership; exam fees • Submission to local social welfare • Periodicity: Annual. office, review, and approval of Transfers made in two application and supporting installments during the documentation school year • Verification of OVC and • Beneficiary: Individual enrollment status by school Elderly grant • Primary • 2005 • National • Type: Cash, through local • 60 years or older administrator: government offices (social • Submission to local government Department of Social welfare offices and office, review, and approval of 51 Table 1. Key Design Features of Main Social Assistance Programs in Eswatini, 2015/16 and 2016/17 Program Implementing agency Year initiated Geographic coverage Benefit structure Benefit eligibility criteria (1) (2) (3) (4) (5) (6) Welfare, Deputy Tinkhundla centers) or post application and supporting Prime Minister’s offices or electronic transfer documentation Office to personal bank account • Amount: E240 per month • Financer: Government per person of the Kingdom of • Periodicity: Quarterly if Eswatini collected at local government or post offices; monthly if electronic transfer to personal bank account • No survivor benefit • Beneficiary: Individual Source: Ministry of Education and Training, Department of Social Welfare, and Eswatini WFP country office staff. Note: OVC = orphaned and vulnerable children. NGO = nongovernmental organizaton. CSO = civil society organization. 52 Figure 7. Beneficiary Numbers, Neighborhood Care Points Source: WFP 2015, 2016, 2017. 53 Figure 8. Distribution of Neighborhood Care Points Across Regions a. Absolute distribution b. Relative distribution Source: Eswatini WFP country office staff. 54 Figure 9. Food Procurement and Program Cost Per Beneficiary, Neighborhood Care Points a. Volume of food procurement b. Food procurement and total program cost per beneficiary Source: WFP 2015, 2016, 2017; Eswatini WFP country office staff. Note: Total program cost comprises food procurement, administrative, and operational costs. US dollar amounts of costs provided by WFP converted to emalangeni using exchange rates provided in IMF (2020). 55 Figure 10. Estimated Beneficiary Numbers and Program Benefit Per Beneficiary, School Feeding Program a. Estimated beneficiary numbers b. Food procurement cost per beneficiary Source: Ministry of Education and Training. Note: Estimated beneficiaries is based on enrollment figures. 56 Figure 11. Beneficiary Numbers and Benefit per Beneficiary, OVC Education Grants a. Beneficiary numbers b. Benefit per beneficiary Source: Department of Social Welfare, Deputy Prime Minister’s Office. Note: OVC = orphaned and vulnerable children. The average grant per beneficiary for exam fees is obtained by dividing total grants toward exam fees by program beneficiaries in grades 8–12 (not just beneficiaries in grade 12 only). 57 Figure 12. Trend in Elderly Grant Benefit Levels, Nominal and Real Source: Budget speeches to the Parliament of the Kingdom of Eswatini, various years; UK DFID, HelpAge International, and UNICEF 2010; Blank et al. 2012. 58 Figure 13. Spending on Neighborhood Care Points Source: WFP Eswatini country office staff. Note: Total = spending on food procurement and operational and administrative costs. 59 Figure 14. Government Spending on School Feeding and Food Aid a. School feeding b. Food aid Source: Ministry of Education and Training; official budget books of the Government of the Kingdom of Eswatini. Note: In panel a, estimated grant portion for school feeding, primary schools = E300 multiplied by government primary school enrollment. 60 Figure 15. Government Spending on Social Assistance Program Benefits, 2014/15– 2017/18 a. Absolute spending across programs b. Relative spending across programs Source: Official budget books of the Government of the Kingdom of Eswatini; Ministry of Education and Training. Note: OVC = orphaned and vulnerable children. Other grants and benefits include disability grants, military pensions, benefits for foster children, and transfers to CARITAS Swaziland and Baphalali Eswatini Red Cross Society. 61 Figure 16. Annual Government Spending on Social Assistance Program Benefits Relative to Annual Total Government Spending and Annual GDP, 2014/15–2017/18 Source: Official budget books of the Government of the Kingdom of Eswatini; Ministry of Education and Training; Ministry of Economic Planning and Development 2018. 62 Figure 17. Social Assistance Program Spending: Eswatini in International Comparison Cumulative distribution of annual social assistance program spending as a percentage of GDP 1.0 0.8 Cumulative share 0.6 0.4 0.2 0.0 0 ESW 2 4 6 8 10 average(SSA, ALL) Percent All developing countries Sub-Saharan African countries Source: Data from the World Bank’s Atlas of Social Protection Indicators of Resilience and Equity (ASPIRE) database at http://datatopics.worldbank.org/aspire/home. Statistic for Eswatini is own estimate based on government administration information for 2016/17. Note: For countries in the global data, the average year for data is 2014. Sample size for all developing countries is 124; sample size for Sub-Saharan African countries is 45. The yellow vertical line labeled “average(SSA, ALL)” indicates the average value for all developing countries and for Sub-Saharan African countries, and the orange vertical line labeled “ESW” the value for Eswatini. 63 Figure 18. Program Coverage Rates of Individuals a. Overall b. By poverty status Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For definitions of the poverty groups, see appendix 1. 64 Figure 19. Program Coverage Rates of Presumed Eligible Individuals a. Overall b. By poverty status Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children; govt. = government. For definitions of the poverty groups, see appendix 1. Secondary schools comprise secondary and high schools. Orphans comprise those below age 18 who have lost one or both parents to any cause of death. 65 Figure 20. Program Coverage Rates of Households a. Overall b. By poverty status Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For definitions of the poverty groups, see appendix 1. 66 Figure 21. Program Coverage Rates by Food Insecurity Status a. Individuals b. Presumed eligible individuals c. Households Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children; govt. = government. Secondary schools comprise secondary and high schools. For definitions of the food insecurity categories, see appendix 1. Orphans comprise those below age 18 who have lost one or both parents to any cause of death. 67 Figure 22. Distribution of Program Benefits and Beneficiaries by Poverty Status a. Benefits b. Beneficiaries Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For definition of poverty groups, see appendix 1. 68 Figure 23. Concentration Coefficients Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. 69 Table 2. Marginal Effects and Effectiveness Marginal effects Program Effectiveness (percentage points) size (percent) (percent) Gini Poverty Poverty severity gap Gini Poverty Poverty Impact Spending Impact/ Impact rate gap Spending Spending Program (1) (2) (3) (4) (5) (6) (7) (8) (9) NCP 0.1 0.0 0.1 0.1 51.2 50.9 80.5 45.8 43.9 School feeding 0.5 0.8 0.8 1.1 49.1 48.0 74.4 48.4 45.6 Food aid 0.1 0.1 0.1 0.3 50.6 50.2 77.5 46.0 44.0 OVC education grants 0.6 0.3 0.8 1.1 54.9 53.9 77.7 53.8 51.0 Elderly grants 0.7 0.5 1.1 2.3 49.4 47.9 73.1 51.0 47.5 All programs (except NCPs) 1.9 1.4 2.8 4.8 54.3 51.5 75.3 57.4 52.2 All programs 2.0 1.5 3.0 5.0 54.4 51.7 75.5 57.8 52.4 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For the poverty gap, impact and spending effectiveness are the same. All poverty-related measures are estimated in relation to the upper poverty line. Program size = total spending on the program divided by total household consumption, estimated based on SHIES 2016/17 data. 70 Table 3. PMT-based Targeting of OVC Education Grants and Elderly Grants, Simulation Results Statistic OVC education grants Elderly grants Current Simulated reform Current Simulated reform Targeted Targeted Benefit Targeted to Targeted to to the to the eliminated the PMT- the PMT- PMT-poor PMT- poor from those poor poor and and receiving savings savings another reallocated relocated pension toward toward benefit higher higher benefits benefits (1) (2) (3) (4) (5) (6) (7) Concentration coefficient –0.242 –0.403 –0.403 –0.152 –0.212 –0.395 –0.395 Change in inequality (Gini index) (pp.) 0.1 –0.1 0.0 0.1 –0.3 Change in overall poverty rate (pp.) 0.2 0.2 0.2 0.4 0.1 Change in overall poverty gap (pp.) 0.1 –0.1 0.1 0.3 –0.3 Change in extreme poverty rate (pp.) 0.1 –0.1 0.0 0.2 –1.0 Change in extreme poverty gap (pp.) 0.0 –0.2 0.0 0.1 –0.4 Savings (Emalangeni, millions) –28.6 0.0 –22.1 –62.2 0.0 Share of all households losing benefit (percent) 3.4 3.4 2.5 7.8 7.8 Share of beneficiary households losing benefit (percent) 30.1 30.1 12.7 39.7 39.7 Percent increase in benefit (percent) 0.0 36.9 0.0 0.0 61.4 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: PMT-poor = those below the 59th percentile of the PMT score distribution; this threshold corresponds to the upper poverty line in the consumption distribution (which generates an overall poverty rate of 58.9 percent). pp. = percentage points. OVC = orphaned and vulnerable children. PMT = proxy means test. 71 Figure 24. Actual Household Consumption versus Proxy Means Test Score Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: The horizontal line denotes the 59 th percentile of the PMT score distribution, while the vertical line denotes the 59 th percentile of the consumption distribution, at where the upper poverty line intersects the consumption distribution. For details on the PMT model, see appendix 4. 72 Figure 25. Distribution of Total Reports by Households of Selected Shocks over Time, May 2011–January 2017 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Other shock includes all shocks except drought and food price shocks. 73 Table 4. Coefficients from Bivariate Regressions of Shock Indicators on Area of Residence, Poverty Status, and Food Insecurity Status Shock Rural Moderate Moderately or extreme poor or severely food insecure Coef. p -value Coef. p -value Coef. p -value (1) (2) (3) (4) (5) (6) a. National Drought 0.036 0.297 0.076 0.001 0.101 0.000 Food price –0.181 0.000 –0.101 0.000 0.021 0.420 Other 0.042 0.113 0.041 0.053 0.114 0.000 b. Urban Drought –– –– –0.015 0.806 0.093 0.044 Food price –– –– –0.071 0.168 0.132 0.011 Other –– –– –0.036 0.477 0.121 0.002 c. Rural Drought –– –– 0.096 0.000 0.101 0.001 Food price –– –– –0.029 0.304 0.021 0.449 Other –– –– 0.044 0.064 0.105 0.000 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Information on shocks refers to the 12 months before the survey. Other shock includes all shocks except drought and food price shocks. For definitions of poverty and food insecurity status, see appendix 1. 74 Table 5. Coefficients from Bivariate Regressions of Shock Indicators on Social Assistance Program Benefit Receipt Indicators Program Drought shock Food price shock Other shock Coef. p -value Coef. p -value Coef. p -value (1) (2) (3) (4) (5) (6) NCPs 0.110 0.001 –0.015 0.629 0.113 0.005 School feeding 0.056 0.007 –0.084 0.001 0.058 0.005 Food aid 0.126 0.000 –0.103 0.002 0.157 0.000 OVC education grants 0.087 0.004 –0.055 0.057 0.017 0.548 Elderly grants 0.078 0.002 –0.110 0.000 0.032 0.157 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. Information on shocks refers to the 12 months before the survey. Other shock includes all shocks except drought and food price shocks. 75 Figure 26. Distribution of Assistance Received, Conditional on Reporting a Given Shock Type Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Information on shocks refers to the 12 months before the survey. Other shock includes all shocks except drought and food price shocks. 76 Appendix 1 Data and Variable Construction The analysis of the performance of Eswatini’s main social assistance programs is based on data from the Swaziland Household Income and Expenditure Survey (SHIES) 2016/17. The SHIES is the official source of data on the living conditions and wellbeing of the country’s population. The SHIES 2016/17 is the third round in a series that began in 2001/02. The survey uses the 2007 Population and Housing Census as its frame, and is representative at the national level and at the region-by-urban/rural levels. The survey was administered between March 2016 and February 2017. The final number of households for the analysis from the survey is 3,355 (from 288 enumeration areas), and the final number of individuals is 14,410. Poverty Throughout the analysis, using the SHIES 2016/17, we measure household welfare as total household consumption per adult equivalent, in January 2017 prices. Our poverty measures are based on the national poverty lines for extreme poverty and overall poverty. Extreme poverty is measured based on the lower poverty line (or extreme or food poverty line) of E463.4 per month per adult equivalent. Overall poverty is measured based on the upper line of E975.3 per month per adult equivalent. The two poverty lines are fixed in January 2017 prices. Extreme poor as used in this study refers to those living below the lower poverty line, while moderate poor refers to those living below the upper poverty line and above the lower poverty line. Food insecurity We measure food insecurity with a standard variable, the Household Food Insecurity Access Scale (HFIAS) (Coates et al. 2007). The HFIAS is based on households’ answers to nine questions: 1. “In the past four weeks, did you worry that your household would not have enough food?” 2. “In the past four weeks, were you or any household member not able to eat the kinds of foods you preferred because of a lack of resources?” 3. “In the past four weeks, did you or any household member have to eat a limited variety of foods due to a lack of resources?” 4. “In the past four weeks, did you or any household member have to eat some foods that you really did not want to eat because of a lack of resources to obtain other types of food?” 5. “In the past four weeks, did you or any household member have to eat a smaller meal than you felt you needed because there was not enough food?” 6. “In the past four weeks, did you or any household member have to eat fewer meals in a day because there was not enough food?” 7. “In the past four weeks, was there ever no food to eat of any kind in your household because of lack of resources to get food?” 8. “In the past four weeks, did you or any household member go to sleep at night hungry because there was not enough food?” 77 9. “In the past four weeks, did you or any household member go a whole day and night without eating anything because there was not enough food?” After each question, the respondent is asked “how frequently?” with possible responses of “rarely," "sometimes," or "often.” These are coded 1, 2, or 3, respectively, and the HFIAS simply sums these responses over the nine questions, with a maximum possible score of 27. To obtain a categorical variable, the Household Food Insecurity Access Prevalence (HFIAP) status classifies households as: (1) severely food insecure if their answer to questions 5 or 6 are “often” or their answers to questions 7 through 9 are at least “rarely”; (2) moderately food insecure if the answer to question 3 or 4 is at least “sometimes” or the answer to question 5 or 6 is at least “rarely” (and they are not severely food insecure); or (3) mildly food insecure if the answer to question 1, 3, or 4 is at least “sometimes” or the answer to question 2 is at least “rarely” (and they are not severely or moderately food insecure). Food secure, the fourth category in the variable, is the residual category. Shocks The SHIES 2016/17 questionnaire asks households to rank the three most significant shocks they suffered in the five years prior to the survey date using a list of 18 shocks: 1. “Drought or floods” 2. “Crop disease or crop pests” 3. “Livestock died or were stolen” 4. “Household business failure (non-agric.)” 5. “Loss of wage employment or non payment of wage” 6. “End of regular assistance, aid or remittances from outside household” 7. “Large fall in sale prices for crops” 8. “Large rise in price for food” 9. “Large rise in agric. input prices” 10. “Chronic/severe illness or accident of household member” 11. “Birth in the household” 12. “Death of household head” 13. “Death of working member of household” 14. “Death of other family member” 15. “Break-up of the household” 16. “Jailed/arrested” 17. “Fire/storm” 18. “Other (Specify)” For most of our analysis, we treat “drought or floods” and “large rise in price for food” as reported, and combine all the other shocks under a constructed indicator, “other shock,” as these remaining shocks have relatively few responses. 78 Even though the recall period is for five years, 46 percent of the reported shocks occurred in the past 12 months, which may reflect some telescoping of recall into the most recent year. For much of our analysis, we focus only on the shocks recorded in the past year. Responses The SHIES questionnaire asks households to report up to three actions they took in response to the shocks listed above from a list of 24 options: 1. “Started to save with cash” 2. “Sent children to live with relatives” 3. “Sold assets (tools, furniture, car etc.)” 4. “Sold farmland” 5. “Rented out farmland” 6. “Sold livestock or poultry” 7. “Sold harvested crops e.g maize” 8. “Worked more or longer hours” 9. “Other household members who were not working went to work” 10. “Started a new business” 11. “Removed children from school to work” 12. “Went elsewhere for more than a month to find work” 13. “Borrowed money from relatives” 14. “Borrowed money from money lender” 15. “Borrowed money from institutions e.g. a bank or a co-operative” 16. “Received help from religious institutions “ 17. “Received help from NGO” 18. “Received help from government” 19. “Received help from family/friends” 20. “Reduced food consumption” 21. “Consumed lower cost but less preferred foods” 22. “Reduced nonfood expenditures” 23. “Did nothing” 24. “Other (specify__________)” Based on this information, we construct indicators for assistance from family or friends (response options 2, 13, and 19), assistance from NGOs or religious institutions (response options 16 and 17), assistance from government (response option 18), and no assistance (all other response options). Social assistance programs Part A of Section B (questions 801–806) in the household survey questionnaire asks for information on the following social protection programs (provided verbatim): (1) Neighborhood Care Point, (2) School Feeding Scheme (Primary or Secondary), (3) Community Assistance Fields (Emasimu Endlunkhulu), (4) Food for Work Assistance Programme, and (5) Food aid from Government (Mshamndane). 79 Neighborhood Care Points. We assign program beneficiary status to the number of children indicated in question 803 (“How many children age 18 or younger in your household receive this benefit?”), starting from the youngest and working up. If the number of beneficiaries indicated in question 803 exceeds the number of children age 18 or younger, we reduce the response to match the number of children in the household. Based on administrative data provided by WFP on total program spending and beneficiaries for 2015 and 2016, we assign the per-beneficiary value of spending averaged over those two years as the annual benefit received by each NCP program beneficiary. This value is E257. School feeding . We assign school feeding program beneficiary status to students in public schools only, starting with the youngest until we reach the number reported in question 803. If the response to question 803 exceeds the number of public primary and secondary school students in the household, we next assign the benefits to school-age children who do not report being enrolled. We estimate an annual per-student benefit value of E466 for public primary school students and E166 for public secondary school students. These values are based on administrative data provided by MOET on spending on food procurement in fiscal years 2015/16 and 2016/17 and on public primary and secondary school enrollment in 2015 and 2016, and using a value of E300 as the upper-bound estimate of the grant benefit per student toward defraying the operational costs of the program at public primary schools. The benefit values are assigned to the school feeding program beneficiaries, as appropriate. Food aid from the government. For food aid from the government, we assign program beneficiary status at the household level based on the response to question 801 (“In the last 12 months, did your household, or any of its members, receive any payments, in cash or in any other form, from [SOURCE]?”). For the benefit value, we use the monetary value of food aid benefits reported by the household in question 804 (“Estimate the value that your household received in-kind during the last 12 months from this program?”) and divide the value equally among all household members. Part B of Section 8 in the survey questionnaire asks for information on the following social protection programs (provided verbatim): (1) Military Pension (Umsizi), (2) Work Retirement Pension, (3) Elderly Grant, (4) OVC Education Grant, (5) Youth Development Fund, (6) Public Assistance Fund, (7) Disability Grant, (8) Workmen’s Compensation Fund, (9) Motor Vehicle Accident Fund, (10) Regional Development Fund, (11) Child Welfare Grant, and (12) Other (specify). Elderly grants. Program beneficiary status is assigned based on the response to question 816 (“Who in the household were the principal beneficiaries of the benefit?”). In a few cases, the reported beneficiary is not age 60 or older, but another household member is. In those cases, we transfer the benefit to the elderly household member. 80 The statutory annual payment effective in fiscal years 2015/16 and 2016/17 was E2,880 per beneficiary. We use the amount reported in question 814 (“What was the value of the benefit received in total during the last 12 months?”) divided equally among beneficiaries in the household. If the benefit is greater than E2,880 per year per beneficiary, we assume that the benefit is another kind of pension and exclude the reported amount from our calculation of elderly grant benefits. Secondary education grants for orphans and vulnerable children (OVC education grants). We assign program beneficiary status to public secondary school students who responded in the affirmative to question 241 in Section 2 (Did [NAME] receive a government scholarship for the present school year?” Per information provided by DSW, the amount of annual OVC education grants varies by grade of secondary school: E1,950 for students in forms I to III (grades 8–10); E2,500 for students in form IV (grade 9); and E2,500 plus up to E2,000 in exam fees for students in form V (grade 12). For students in forms I to IV, we use the statutory amount. For students in form V, we assign the maximum possible benefit of E4,500. We compare estimates of the number of beneficiaries and total spending on benefits obtained from the SHIES 2016/17 to corresponding statistics obtained from government and WFP administrative data (table A1.1). The statistics match up reasonably well for NCPs, school feeding, OVC education grants, and elderly grants when we compare the SHIES 2016/17 estimates to administrative statistics averaged over fiscal years 2015/16 and 2016/17 (calendar years 2015 and 2016 for NCPs) (figure A1.1). For food aid, we estimate 33,516 beneficiary households in the survey data, whereas NDMA reports that the government reached 72,745 households with food aid in 2016/17. (No beneficiary number was provided by NDMA for 2015/16.) Hence, the survey estimate is only 46.1 percent of the administrative value. According to NDMA information, over 80 percent of the distribution of government food aid in 2016/17 occurred from August 2016 onward, covering virtually the entire country. The survey was conducted from March 2016 to February 2017, so only for those interviewed from September 2016 through February 2017 (about half of the survey period) would the question of receipt of government food aid be relevant. About 52 percent of survey respondents were interviewed before September 2016; in addition, by September 2016, survey implementation progress was further along in Lubombo and Shisweleni, the poorer, more food-insecure regions, than in Hhohho and Manzini. Given this, we surmise that a large share of the shortfall in the beneficiary estimate in the SHIES 2016/17 is due to the more condensed timing of the food aid distribution vis-à-vis the survey fielding period. 81 Table A1.1. Program Beneficiary Numbers and Total Spending on Benefits, SHIES 2016/17 Estimates versus Administrative Information Program Level Beneficiaries (in thousands) Benefits (in millions of emalangeni) SHIES 2016/17 Administrative data SHIES 2016/17 Administrative data 2015/16 2016/17 2015/16 2016/17 (1) (2) (3) (4) (5) (6) NCPs Individual 45.2 50.8 51.9 11.6 20.6 5.7 School feeding Individual 259.6 350.5 368.1 100.9 135.5 138.5 Food aid Household 33.5 -- 72.7 20.8 47.9 13.2 OVC education grants Individual 45.3 53.6 52.6 106.1 137.3 141.6 Elderly grants Individual 66.4 66.4 69.7 163.3 185.0 165.0 Source: Administrative information: DSW, MOET, NDMA, WFP, and official budget books of the Government of the Kingdom of Eswatini. SHIE. Survey values are Authors’ estimates based on the Swaziland Household Income and Expenditure Survey 2016/17. Note: This table presents estimates from the SHIES 2016/17 of the number of beneficiaries and total spending on benefits, by program. It also presents corresponding statistics from program administrative information for fiscal year 2015/16 and 2016/17 for government programs and information for 2015 and 2016 for NCPs. NCP = Neighborhood Care Point. OVC = orphaned and vulnerable children. 82 Figure A1.1. Beneficiaries and Total Spending on Benefits, Shortfalls in SHIES 2016/17 Estimates Relative to Administrative Statistics in 2015/16 and 2016/17, by Program Source: Survey values are Authors’ estimates based on the Swaziland Household Income and Expenditure Survey 2016/17. Administrative information on the NCP program was provided by WFP, and school feeding by MOET, food aid by NDMA and from official government budget books, and OVC education grants and elderly grants by DSW and from official government budget books. Note: This figure shows the shortfall (in percent terms) in survey estimates for 2016/17 compared to administrative statistics averaged over fiscal years 2015/16 and 2016/17 in the number of beneficiaries and total spending in benefits, by program. NCP = Neighborhood Care Point. OVC = orphaned and vulnerable children. 83 Appendix 2 Description of Measures While there are many measures of inequality, poverty, and social protection program benefit incidence, this review uses only a few. For inequality, we use the Gini coefficient. For poverty, we use the Foster-Greer-Thorbecke (FGT) poverty measures. For social protection program benefit incidence, we use concentration coefficients and the marginal effect of a benefit on poverty and inequality. Finally, for effectiveness, we use measures developed in Enami (2018) that calculate how much inequality or poverty reduction a government gets from a program relative to the size of its budget. In describing these measures, this appendix borrows from Boko, Raju, and Younger (2021). Program coverage Coverage measures the total number of beneficiaries of a social protection program divided by the target population for that program. This denominator may be the entire population or some subset. For example, for elderly grants, we take the number of people receiving such benefits divided by the population age 60 and older. Inequality measure: Gini coefficient The Gini coefficient is the most common measure of inequality. The easiest way to understand this measure is to first construct a Lorenz curve: Order the data by our welfare measure— consumption per adult equivalent (from here on, consumption )—from poorest to richest and then plot the cumulative share of the sample on the horizontal axis against the cumulative share of consumption on the vertical axis. Table A2.1 provides some illustrative data for 10 people. Because there are 10 people, the first person represents 10 percent of the sample, the first 2 people represent 20 percent, and so on, as shown in column 2. Column 3 reports each person’s consumption, and we can see that the data are ordered from poorest to richest. The sum of all consumption is 1,000, so we get the consumption shares (column 4) by dividing each person’s consumption by 1,000. Finally, column 5 gives the cumulative consumption shares. Table A2.1. Illustrative Data for a Lorenz Curve and the Gini Coefficient Cumulative Cumulative population Consumption consumption Observation share Consumption share share (1) (2) (3) (4) (5) 1 0.100 1 0.001 0.001 2 0.200 3 0.003 0.004 3 0.300 7 0.007 0.011 4 0.400 13 0.013 0.024 5 0.500 20 0.020 0.044 6 0.600 30 0.030 0.074 7 0.700 60 0.060 0.134 84 8 0.800 100 0.100 0.234 9 0.900 250 0.250 0.484 10 1.000 516 0.516 1.000 Total 1,000 The Lorenz curve then graphs column 5 against column 2, as in figure A2.1. Because both axes are shares, they range from zero to one. And because the data are ordered from poorest to richest, the Lorenz curve must be convex: The poorest person’s consumption share cannot be greater than his or her population share, similarly for the poorest two people, and so on. Indeed, the more convex the Lorenz curve, the more unequal the distribution of consumption. In the extreme case in which one person has all the consumption, the Lorenz curve would be a right angle, running along the horizontal axis until it reaches the last person in the sample because everyone but that person would have zero consumption. Conversely, if everyone has exactly the same consumption—representing complete equality—the cumulative consumption share would be equal to the cumulative population share and the Lorenz curve would be a 45-degree line. Figure A2.1. Illustrative Lorenz Curve The Gini coefficient is the area between the 45-degree line and the Lorenz curve, multiplied by two. With complete equality, where the Lorenz curve is on the 45-degree line, the Gini coefficient would be zero. With complete inequality, the Gini coefficient would be twice the triangle below 85 the 45-degree line, or one. Gini coefficients are observed to be around 0.25 in the countries with the least inequality and around 0.65 in those with the greatest inequality (World Bank 2019b). Poverty measures: poverty rate and poverty gap The Foster-Greer-Thorbecke measures are the most common measures of poverty (Foster, Greer, and Thorbecke 1984). The first FGT measure, denoted by FGT(0), is the poverty rate (or headcount ratio), which is the share of the population that is poor. The second FGT measure, denoted by FGT(1), is the poverty gap , which is the average difference between the poverty line and a person’s consumption, usually scaled by the poverty line itself. The poverty gap is expressed by the equation  1   z − ci  poverty gap =   ∑   ,  N  i  z + where z is the poverty line, ci is the ith person’s consumption, and N is the sample size. The plus sign indicates that we include the difference only if it is positive. The third FGT measure, denoted by FGT(2), is the squared poverty gap, sometimes called poverty severity. It is calculated in the same way as the poverty gap but with the term in parentheses squared.63 Table A2.2 illustrates the calculation of the poverty measures using the same data with which we derive the Lorenz curve and assuming that the poverty line is 21. Column 3 indicates whether or not a person is poor. Summing this column and dividing by the sample size gives the headcount ratio: 0.50. So, half the population is poor. Column 4 reports the absolute poverty gap for each person: the difference between the poverty line (21) and his or her consumption. This has an interesting interpretation because it shows how much money would be needed to bring every person’s consumption to the poverty line. Column 5 scales each person’s absolute poverty gap by the poverty line. Averaging that over the sample gives the poverty gap: 0.29. Finally, column 6 squares each person’s scaled poverty gap. Averaging that over the sample gives the poverty severity: 0.22. Table A2.2. Illustrative Calculation of FGT Poverty Measures Scaled Absolute Scaled poverty Poverty poverty poverty gap Observation Consumption status gap gap squared (1) (2) (3) (4) (5) (6) 1 1 1 20 0.952 0.907 2 3 1 18 0.857 0.735 3 7 1 14 0.667 0.444 63 In fact, the family of FGT measures can be written concisely as α  1   z − ci  FGT (α ) =   ∑    N  i  z + If α = 0 , the term in parentheses is one if a person is poor, zero otherwise, so that gives us the poverty rate. If α = 1 , we get the poverty gap, and if α = 2 , we get poverty severity. 86 4 13 1 8 0.381 0.145 5 20 1 1 0.048 0.002 6 30 0 0 0 0 7 60 0 0 0 0 8 100 0 0 0 0 9 250 0 0 0 0 10 516 0 0 0 0 Sum 5 2.90 2.23 Average (FGT) 0.50 0.29 0.22 Poverty line 21 Program benefit incidence Concentration curve and coefficient A concentration curve is similar to a Lorenz curve. We order the data from poorest to richest and then plot the cumulative share of a benefit against the cumulative share of the population. Table A2.3 provides illustrative data for a benefit, and figure A2.2 shows the corresponding curve graphing column 6 against column 2. Table A2.3. Illustrative Data for a Concentration Curve Cumulative Cumulative population Benefit Benefit benefit Observation share Consumption received share share (1) (2) (3) (4) (5) (6) 0.00 0.00 1 0.10 1 30.0 0.300 0.30 2 0.20 3 20.0 0.200 0.50 3 0.30 7 15.0 0.150 0.65 4 0.40 13 10.0 0.100 0.75 5 0.50 20 5.0 0.050 0.80 6 0.60 30 5.0 0.050 0.85 7 0.70 60 5.0 0.050 0.90 8 0.80 100 5.0 0.050 0.95 9 0.90 250 3.0 0.030 0.98 10 1.00 516 2.0 0.020 1.00 Total 1,000 100 Unlike a Lorenz curve, a concentration curve can be either convex or concave. A concave curve would occur if poorer people receive a higher share of a benefit than the rich (which may well be the case for explicitly targeted benefits). 87 Figure A2.2. Illustrative Concentration Curve Like the Gini coefficient, the concentration coefficient for a social protection program benefit is the area between its concentration curve and the 45-degree line. This can range from negative one (the poorest person receives all the benefit) to one (the richest person receives all the benefit), with zero representing a benefit spread evenly across the population. Benefits with a concave concentration curve (negative concentration coefficient) are usually referred to as pro-poor or absolutely pro-poor. Those with a concentration curve below the Lorenz curve (concentration coefficient greater than the Gini) are considered to be regressive. And benefits with a concentration curve between the Lorenz curve and the 45-degree line (concentration coefficient less than the Gini, but positive) are relatively pro-poor. Marginal effect Another way to assess the incidence of a social protection program benefit is to see how it alone changes poverty or inequality. We refer to this as its marginal effect. To calculate it, we estimate a measure of poverty or inequality for household consumption per adult equivalent and then estimate the same poverty or inequality measure after either adding or removing the social protection program benefit. The difference between the two estimates is the marginal effect. One important difference between the marginal effect and the concentration coefficient is that the marginal effect depends on the overall size of the social protection program benefit while the 88 concentration coefficient does not. A larger program benefit, for example, can have a larger impact on inequality and especially poverty than a smaller one that is distributed to exactly the same people and so has the same concentration curve. Program effectiveness The last measure we consider is the effectiveness of a social protection program. Enami (2018) presents two measures: impact effectiveness and spending effectiveness. The impact effectiveness of a program benefit is the ratio of its marginal effect on a distributional statistic (we use the Gini coefficient and the poverty gap in this study) to the marginal effect of a perfectly targeted program benefit of the same size, that is, with the same total program budget. A perfectly targeted program benefit is one that transfers enough money to the poorest person in the country to bring his or her consumption up to that of the second-poorest person, then transfers enough to both of them to bring their consumption up to that of the third-poorest person, and so on until the program budget runs out. Impact effectiveness is expressed by the equation π with actual benefit − π without actual benefit impact effectiveness = , π with perfect benefit − π without actual benefit where π is some distributional statistic. Since the actual benefit cannot improve the distributional statistic by more than the perfect benefit does, the measure is bounded above by one. And if the actual benefit does not improve the distributional statistic at all, the measure is bounded by zero.64 Enami’s second measure, spending effectiveness, is the ratio between the amount that would have to be spent on a perfect transfer to achieve exactly the same reduction in the distributional statistic as is actually achieved by a benefit, and the amount spent on the actual benefit. Spending effectiveness is expressed by the equation budget needed to achieve (π with actual benefit − π without actual benefit ) with a perfect transfer spending effectiveness = . budget needed to achieve (π with actual benefit − π without actual benefit ) with the actual benefit This measure is also bounded by zero and one. If the actual benefit does not improve the distributional statistic of interest, then the numerator is zero: No spending is needed to achieve an equally good result. And if the actual benefit is distributed perfectly, the numerator and denominator are equal. It is inappropriate to apply either of these effectiveness measures to the poverty rate. The most effective way to reduce the poverty rate is to give additional consumption to the poor person who is closest to the poverty line—that is, the least-poor poor person—until his or her consumption reaches the poverty line, then transfer additional consumption to the next-poorest poor person until that person’s consumption reaches the poverty line, and so on. Because this “perfect” transfer favors the least poor of the poor, it is not ethically defensible. On the other hand, the perfect transfer 64 We ignore cases where the benefit actually worsens the distribution of consumption (per adult equivalent) and simply set their impact effectiveness to zero. 89 discussed above, which favors the poorest of the poor, may well fail to bring anyone above the poverty line and so be judged as completely ineffective on the basis of the two effectiveness measures—because while it would benefit only the poor, it would not reduce the headcount ratio. 90 Appendix 3 Supplemental tables and figures Table A3.1. Program Coverage Rates of Individuals, by Area, and by Poverty, Welfare Quintile, and Food Insecurity Status Group NCPs School feeding Food aid OVC education grants Elderly grants All programs All programs (except NCPs) Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban In percent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) a. By poverty status Extreme poor 5.7 5.5 12.8 28.8 28.5 37.5 24.3 24.9 5.5 6.0 6.0 7.4 7.9 7.9 5.2 48.9 49.2 37.7 54.6 54.7 50.5 Moderate poor 5.2 5.3 4.6 27.7 28.3 21.2 21.3 23.0 5.4 4.7 4.8 3.5 6.4 6.8 2.7 45.6 47.7 26.2 50.8 52.9 30.8 Nonpoor 2.0 2.6 1.1 15.2 19.3 10.0 7.5 13.1 0.3 2.3 3.0 1.3 4.3 5.6 2.5 24.0 32.6 12.7 25.9 35.2 13.8 b. By welfare quintile Poorest 5.6 5.4 12.8 28.8 28.5 37.5 24.4 25.0 5.5 6.1 6.0 7.4 7.9 8.0 5.2 49.0 49.3 37.7 54.6 54.7 50.5 Second 6.1 6.4 2.9 29.0 29.0 28.1 23.5 25.3 2.0 4.7 4.5 7.6 6.0 6.3 2.5 47.4 48.6 32.2 53.5 55.1 35.0 Third 4.4 4.2 5.4 26.2 27.6 15.5 19.1 20.7 7.2 4.6 5.2 0.7 6.8 7.3 2.6 43.5 46.6 20.8 47.9 50.8 26.2 Fourth 3.0 3.3 2.2 19.9 22.4 13.6 12.2 16.8 0.6 3.3 3.7 2.5 5.0 5.9 2.7 31.8 38.1 16.0 34.7 41.3 18.2 Richest 0.7 1.0 0.5 10.1 12.7 8.4 2.1 4.8 0.2 1.1 1.7 0.8 3.5 5.1 2.4 15.2 21.0 11.3 15.9 22.0 11.9 c. By food insecurity status Food secure 1.2 1.3 1.1 16.3 20.8 9.7 6.4 10.8 0.0 2.8 4.0 1.1 3.9 5.5 1.6 24.5 33.6 11.4 25.7 34.9 12.5 Mildly insecure 3.2 2.8 4.4 23.5 27.4 12.9 12.2 16.7 0.0 3.9 5.1 0.7 5.1 5.9 2.8 35.4 43.2 14.2 38.6 46.0 18.6 Moderately insecure 4.3 4.8 1.7 25.6 27.8 15.5 16.1 19.1 2.6 4.0 4.3 2.4 6.7 7.5 2.9 40.6 45.2 19.5 44.9 50.1 21.2 Severely insecure 5.5 6.0 2.7 24.6 25.9 15.1 23.1 26.0 3.2 4.7 4.9 3.2 6.5 6.9 4.1 43.2 46.5 20.8 48.7 52.4 23.4 All 4.0 4.5 1.9 22.8 25.7 12.6 16.3 20.5 1.3 4.0 4.6 1.8 5.8 6.7 2.6 37.4 43.6 15.7 41.3 48.1 17.6 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For definitions of welfare, poverty status, and food insecurity status, see appendix 1. 91 Table A3.2. Coverage Rates of Presumed Eligible Individuals, by Area, and by Poverty, Welfare Quintile, and Food Insecurity Status Group NCPs School feeding Food aid OVC education grants Elderly grants Orphans, younger than age 8 years Government primary, secondary, Severely food insecure Orphans in government secondary Age 60 years or older and high school students schools Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban In percent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) a. By poverty status Extreme poor 11.5 11.9 0.0 73.4 73.0 85.6 27.7 28.4 6.6 77.1 79.4 0.0 86.6 87.7 52.9 Moderate poor 16.4 15.7 19.8 76.0 76.2 74.1 24.0 25.8 5.7 53.0 51.2 71.4 84.6 84.8 81.4 Nonpoor 6.0 9.6 0.0 69.1 71.3 63.9 13.4 20.6 1.4 45.8 49.9 38.9 72.6 78.6 59.6 b. By welfare quintile Poorest 11.7 12.1 0.0 78.3 77.6 100.0 27.8 28.6 6.6 77.1 79.4 0.0 86.6 87.7 52.9 Second 22.2 20.9 36.1 80.9 80.2 91.5 24.8 27.0 0.0 48.4 39.8 100.0 83.9 84.1 79.3 Third 8.9 7.8 12.8 81.7 82.3 76.2 23.6 25.0 11.6 56.7 59.1 0.0 85.9 86.0 82.7 Fourth 8.8 9.8 0.0 75.7 77.5 68.0 15.1 20.7 2.3 51.6 52.1 50.6 77.7 82.2 60.1 Richest 3.6 11.4 0.0 78.0 74.5 81.8 4.3 9.4 0.3 31.4 39.6 25.5 65.0 69.9 59.3 c. By food insecurity status Food secure 3.0 4.4 0.0 72.2 72.0 72.9 -- -- -- 52.7 66.7 29.1 66.9 74.9 44.6 Mildly insecure 11.7 11.7 0.0 87.0 88.0 80.2 -- -- -- 49.6 57.9 21.8 80.6 83.1 69.3 Moderately insecure 10.7 13.5 0.0 86.2 86.6 82.7 -- -- -- 60.5 58.6 70.0 84.9 86.0 73.9 Severely insecure 16.8 15.8 22.7 76.5 75.9 84.5 23.1 26.0 3.2 53.1 52.2 62.0 84.7 85.9 73.5 All 12.4 13.3 8.2 79.3 79.2 79.6 23.1 26.0 3.2 54.7 57.0 44.4 81.1 83.9 62.2 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For definitions of welfare, poverty status, and food insecurity status, see appendix 1. 92 Table A3.3. Coverage Rates of Households, by Area, and by Poverty, Welfare Quintile, and Food Insecurity Status Group NCPs School feeding Food aid OVC education grants Elderly grants All programs All programs (except NCPs) Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban Overall Rural Urban In percent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) a. By poverty status Extreme poor 16.6 16.3 23.2 65.4 64.6 86.4 23.8 24.4 9.3 24.8 24.8 23.4 42.9 43.5 26.1 84.6 84.5 86.4 85.9 85.9 86.4 Moderate poor 12.8 12.8 12.9 56.9 59.5 40.0 18.7 20.7 5.9 17.6 19.2 7.3 27.1 29.7 9.9 71.5 75.4 45.6 74.5 78.0 50.9 Nonpoor 3.1 4.7 1.4 23.7 30.8 16.3 5.7 10.9 0.3 5.4 7.6 3.1 11.0 15.8 6.0 32.3 43.7 20.6 33.2 45.5 20.6 b. By welfare quintile Poorest 16.5 16.2 23.2 65.3 64.5 86.4 23.9 24.4 9.3 24.8 24.9 23.4 43.0 43.6 26.1 84.5 84.5 86.4 85.8 85.8 86.4 Second 14.8 15.2 10.4 60.4 61.7 48.4 20.8 22.7 3.1 19.1 19.4 16.0 27.5 29.5 9.7 73.8 76.6 48.4 76.5 78.6 56.1 Third 11.4 11.0 13.3 53.4 57.4 33.2 17.1 19.2 6.9 15.9 18.6 2.6 26.4 29.8 9.3 69.0 74.5 41.2 72.3 77.7 45.0 Fourth 5.4 6.7 2.9 34.0 40.8 21.1 10.7 16.1 0.4 9.2 10.7 6.4 15.4 19.4 7.8 46.6 56.8 27.2 48.0 58.9 27.2 Richest 1.2 1.8 0.8 16.0 18.2 14.6 1.8 4.2 0.3 2.6 3.8 1.8 7.6 11.0 5.4 21.8 27.2 18.2 22.3 28.4 18.2 c. By food insecurity statu Food secure 2.4 3.3 1.5 26.1 34.1 17.1 5.0 9.4 0.0 6.4 10.4 2.0 10.6 16.6 3.9 33.4 46.2 19.1 33.6 46.5 19.4 Mildly insecure 6.5 6.6 6.4 40.7 51.7 22.8 9.1 14.6 0.0 9.3 14.0 1.8 16.4 22.0 7.3 49.9 62.8 28.8 50.7 64.1 28.8 Moderately insecure 8.6 10.5 3.5 45.9 54.0 24.0 12.7 16.7 2.1 13.2 15.6 6.9 23.1 28.6 8.4 57.2 67.9 28.5 59.3 70.5 29.4 Severely insecure 11.8 14.1 4.1 43.9 50.8 20.7 17.9 22.5 2.7 15.0 17.9 5.4 26.1 30.9 10.2 60.5 70.0 28.6 63.0 72.9 30.1 All 7.7 9.9 3.1 38.7 47.8 20.1 11.8 17.1 1.1 11.4 15.1 3.8 19.7 26.1 6.8 50.4 63.2 24.5 52.0 65.2 25.1 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. For definitions of welfare, poverty status, and food insecurity status, see appendix 1. 93 Table A3.4 Sensitivity of Marginal Effect Estimates to Alternate Poverty Lines Lower poverty line Upper poverty line Poverty rate Poverty gap Poverty rate Poverty gap In percentage points Program (1) (2) (3) (4) NCPs 0.1 0.1 0.0 0.1 School feeding 1.4 0.6 0.8 0.8 Food aid 0.2 0.1 0.1 0.1 OVC education grants 1.6 0.8 0.3 0.8 Elderly grants 1.8 1.1 0.5 1.1 All programs (except NCPs) 5.2 2.8 1.4 2.8 All programs 5.2 2.9 1.5 3.0 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. The extreme and upper poverty lines are E463.4 and E985.3 per month per adult equivalent in January 2017 prices. 94 Table A3.5 Sensitivity of Effectiveness Estimates to Alternative Poverty Lines Program Lower poverty line Upper poverty line Poverty Poverty Poverty Poverty gap severity gap severity Impact/ Impact Spending Impact/ Impact Spending spending spending In percent (1) (2) (3) (4) (5) (6) NCPs 28.6 17.2 12.4 80.5 45.8 43.9 School feeding 27.2 22.8 13.9 74.4 48.4 45.6 Food aid 28.0 16.9 12.3 77.5 46.0 44.0 OVC education grants 36.7 30.4 20.8 77.7 53.8 51.0 Elderly grants 31.7 30.9 18.4 73.1 51.0 47.5 All programs (except NCPs) 33.9 49.9 21.9 75.3 57.4 52.2 All programs 34.0 51.4 22.1 75.5 58.0 52.4 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. The extreme and upper poverty lines are E463.4 and E985.3 per month per adult equivalent in January 2017 prices. 95 Table A3.6. Coefficients from Bivariate Regressions of Shock Indicators on Social Assistance Program Benefit Receipt Indicators, by Area of Residence Program Drought shock Food price shock Other shock Coef. p -value Coef. p -value Coef. p -value (1) (2) (3) (4) (5) (6) a. Rural NCPs 0.092 0.009 0.009 0.785 0.096 0.024 School feeding 0.069 0.007 –0.037 0.144 0.045 0.067 Food aid 0.114 0.001 –0.053 0.104 0.142 0.000 OVC education grants 0.088 0.008 –0.007 0.809 0.006 0.832 Elderly grants 0.092 0.001 –0.054 0.051 0.026 0.285 b. Urban NCPs 0.173 0.114 0.111 0.211 0.162 0.128 School feeding –0.009 0.802 –0.049 0.342 0.069 0.095 Food aid 0.271 0.370 0.176 0.002 0.382 0.000 OVC education grants 0.026 0.779 –0.039 0.681 0.009 0.920 Elderly grants –0.047 0.443 –0.133 0.038 –0.006 0.921 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. Information on shocks refers to the 12 months before the survey. Other shock includes all shocks except drought and food price shocks. 96 Table A3.7. Coefficients from Bivariate Regressions of Shock Indicators on Social Assistance Program Benefit Receipt Indicators, by Poverty Status Program Drought shock Food price shock Other shock Coef. p -value Coef. p -value Coef. p -value (1) (2) (3) (4) (5) (6) a. Extreme poor NCPs 0.150 0.025 0.097 0.070 0.069 0.313 School feeding 0.080 0.134 0.035 0.493 0.027 0.621 Food aid 0.083 0.167 0.024 0.618 0.186 0.001 OVC education grants 0.059 0.300 0.041 0.418 –0.044 0.392 Elderly grants 0.068 0.154 –0.003 0.941 0.039 0.388 b. Moderate poor NCPs 0.039 0.437 0.027 0.611 0.090 0.112 School feeding 0.041 0.236 –0.014 0.690 0.016 0.609 Food aid 0.117 0.017 –0.048 0.279 0.115 0.008 OVC education grants 0.042 0.383 –0.013 0.750 –0.029 0.455 Elderly grants 0.075 0.028 –0.074 0.032 –0.003 0.921 c. Nonpoor NCPs 0.099 0.095 –0.053 0.457 0.157 0.013 School feeding 0.009 0.733 –0.104 0.002 0.078 0.010 Food aid 0.097 0.046 –0.174 0.000 0.175 0.000 OVC education grants 0.090 0.087 –0.076 0.166 0.084 0.082 Elderly grants 0.026 0.510 –0.132 0.001 0.033 0.414 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. Information on shocks refers to the 12 months before the survey. Other shock includes all shocks except drought and food price shocks. For definitions of poverty status, see appendix 1. 97 Table A3.8. Coefficients from Bivariate Regressions of Shock Indicators on Social Assistance Program Benefit Receipt Indicators, by Food Insecurity Status Program Drought shock Food price shock Other shock Coef. p -value Coef. p -value Coef. p -value (1) (2) (3) (4) (5) (6) a. Food secure NCPs 0.196 0.051 –0.104 0.290 0.185 0.035 School feeding 0.087 0.033 –0.054 0.237 0.146 0.000 Food aid 0.072 0.287 –0.208 0.001 0.185 0.009 OVC education grants 0.185 0.008 –0.029 0.678 0.135 0.055 Elderly grants 0.132 0.009 0.004 0.937 0.159 0.005 b. Mildly insecure NCPs 0.348 0.004 –0.043 0.746 –0.037 0.739 School feeding 0.048 0.420 –0.222 0.000 0.074 0.172 Food aid 0.282 0.003 –0.167 0.078 0.132 0.147 OVC education grants 0.175 0.082 –0.118 0.222 0.105 0.237 Elderly grants –0.012 0.893 –0.181 0.006 0.096 0.202 c. Moderately insecure NCPs 0.113 0.114 –0.022 0.671 0.088 0.209 School feeding –0.020 0.588 –0.124 0.002 –0.044 0.229 Food aid 0.100 0.080 –0.106 0.040 0.105 0.038 OVC education grants 0.048 0.360 –0.049 0.348 –0.049 0.288 Elderly grants –0.007 0.867 –0.141 0.001 –0.060 0.154 d. Severely insecure NCPs –0.002 0.968 –0.005 0.922 0.082 0.103 School feeding 0.049 0.147 –0.059 0.096 0.013 0.710 Food aid 0.080 0.102 –0.081 0.073 0.135 0.001 OVC education grants 0.014 0.744 –0.071 0.085 –0.051 0.187 Elderly grants 0.083 0.016 –0.144 0.000 –0.027 0.391 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: NCP = Neighborhood Care Point; OVC = orphaned and vulnerable children. Information on shocks refers to the 12 months before the survey. Other shock includes all shocks except drought and food price shocks. For definitions of poverty status, see appendix 1. 98 Figure A3.1 Distribution of Assistance Received, Conditional on Reporting a Given Shock Type, by Area of Residence and by Poverty Status a. By area of residence b. By poverty status c. By food insecurity status Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Other shock includes all shocks except drought and food price shocks. For definitions of poverty and food insecurity status, see appendix 1. 99 Appendix 4 Development of a Proxy Means Test Model for Eswatini This appendix discusses the development of a Proxy Means Test (PMT) model for Eswatini based on the SHIES 2016/17. A PMT model uses a few easily identifiable characteristics of people or households to identify those most likely to be poor. Government uses a PMT model, alone or in combination with other information, to determine eligibility for a targeted social assistance program or social service. The motivation for using a PMT model is that determining a household’s poverty status is time-consuming and impractical. A PMT model allows the few easily identifiable characteristics to proxy household welfare and thus poverty status. The best such characteristics should be: (d) highly correlated with poverty status; (e) easily identifiable by a social welfare worker; and (f) difficult for households to falsify. The general idea is to use a rich source of data such as the SHIES 2016/17, which has information both on household welfare (measured by consumption per adult equivalent in our case) and on characteristics to use as proxies. We use the survey to estimate the correlations between actual welfare and the proxies, and then generate a prediction equation for household welfare comprised of several characteristics and weights for each of them. Armed with these weights, a social welfare worker can then collect only the proxy information on potential beneficiaries and use that information plus the estimated weights to predict household welfare for the potential beneficiaries. Based on that prediction, the social welfare worker can determine eligibility for a social assistance program or social service. The desirability of criterion (1) should be obvious: The proxy characteristics must be highly correlated with actual welfare to predict it well. Criterion (2) is important because a social welfare worker needs to be able to determine eligibility quickly and easily. Criterion (3) is important because potential beneficiaries who come to understand the purpose of the PMT model and the questions a social welfare worker asks about the proxy characteristics have an incentive to falsify information to increase their eligibility. A further practical restriction on the characteristics is that if an existing household survey such as the SHIES 2016/17 is to be used to identify potential proxy characteristics, the survey must have collected information on those characteristics, as well as on household welfare itself. The set of proxy characteristics is limited to information collected in the survey. Approach We take a standard approach to estimating a PMT score, which is predicted household welfare: We regress the log of household consumption per adult equivalent on a set of household characteristics. It is possible to use statistical techniques to select these characteristics, 65 but select 65 Many PMT model developers use some variant of stepwise regression. McBride and Nichols (2018) use machine learning algorithms. 100 them directly based on the information available in the SHIES 2016/17 and our general sense of characteristics likely to satisfy the three criteria outlined above. We compare five models: 1. Model (1): This model includes only characteristics we are certain can be both easily observed by a social welfare worker and difficult to falsify by a potential beneficiary. We will call this a “minimum” model. The characteristics include the area (region and rural/urban) where a household lives, the number of rooms in its dwelling, and a set of indicators (dummy variables) for the type of roof, floor, drinking water supply, and toilet facility a household has. 2. Model (2): This model includes the characteristics in model (1) plus a set of dummy variables for household size (limited to a maximum of seven). It seems that the number of household members should be certainly verifiable and thus included in the minimum model (1), but households have been known to “borrow” children or the elderly from their relatives or neighbors to improve their PMT score. In countries with universal registries of births and deaths or where social welfare workers have good knowledge of potential beneficiaries, this is not a problem. 3. Model (3): This model includes the characteristics in model (2) plus a set of indicators of ownership for a list of durable goods. 66 Many PMT models use such indicators because poorer households should have fewer possessions. But most of these durable goods are more easily falsified than the characteristics in models (1) and (2). Goods that are small (radios, stoves, computers, telephones) or mobile (bicycles, motorcycles, cars, trucks, tractors) can be concealed from a social welfare worker if the potential beneficiary understands that these contribute to the household’s PMT score. Nevertheless, we include them here because it is a common practice. 4. Model (4): This model includes the characteristics in model (2) plus some variables that are potentially verifiable by cross-referencing other databases. These include whether the household includes someone working in the public sector; whether it includes someone drawing a retirement pension; what the household spends each month on electricity; and what is its monthly expenditure for piped water. In theory, this information could be collected from databases available from the civil service, the pension funds, and the power and water utilities, but only if those accounts are identified with information such as a national identification number that can be tied to potential social assistance program beneficiaries. This is not currently possible in Eswatini, but it is something that could be pursued in the future. One advantage of using electricity and water expenditures is that they are continuous variables. Predicting a continuous variable, household consumption per adult equivalent, with only discrete characteristics produces “lumpy” predictions as there is a finite number of possible combinations of those characteristics. Using some continuous characteristics can smooth out the predicted values. 66 The list includes all the assets that Section 9 of the SHIES questionnaire asks about: radio cassette/CD player, television, refrigerator/freezer, stove/hot plate, bicycle, motorcycle, car, van/bakkie/truck, tractor, computer, fixed- line telephone, cellular telephone, internet access, and grinding machine. 101 5. Model (5): This model includes the characteristics in model (2) plus both the asset indicators from model (3) and the potentially verifiable variables from model (4). This model should give the most accurate prediction of household welfare, but at the cost of including some variables that may be easily falsified or may not be readily observable without further work to cross-reference databases. In general, any model will be more accurate when tested on the same data used to estimate it (“within-sample”) because both the estimates and the predictions benefit from any spurious correlations found in the estimating data but unlikely to be similar in other data (“out-of-sample”). To avoid an overly generous assessment of each model’s accuracy, we estimate and assess the models with the following steps: 1. Step (1): randomly select two-thirds of the sample; this subset is sometimes called the “training” data; 2. Step (2): estimate a regression of the log of household welfare on household characteristics using that randomly selected sub-sample; 3. Step (3): predict household welfare for the remaining one-third of the sample; this subset is called the “test” data; and 4. Step (4): calculate all the performance assessment statistics on the test data. We bootstrap these steps 100 times and use the results to estimate standard errors for the mean value of all the statistics calculated in step 4. Note that in step (1), we stratify the selection by the same strata used in the SHIES 2016/17 sample: rural and urban areas in each of Eswatini’s four regions. Results Figure A4.1 compares the average predicted household welfare to actual household welfare at each percentile of the household welfare distribution. 67 A well-performing model should yield predictions as close to the actual distribution (in black) as possible. All of the models make predictions that are too flat: They overpredict household welfare at the low end of the distribution and underpredict it at the high end. 68 But there are differences in the models. Model (1) is noticeably flatter than the rest. Model (2) is significantly better, but not as well-performing as the remaining three models, which are difficult to distinguish from one another. There are many other ways to assess the accuracy of a PMT model, 69 almost all of which support the conclusion drawn from figure A4.1: Model (1) is significantly less accurate than model (2), 67 This and all statistics in this section are for the test data only. 68 This is a standard feature of regression. A model with no predictive power would yield a flat line at the mean of the distribution. 69 See Raju and Younger (2020) for an exposition and examples. 102 which is somewhat less accurate than the other three models, all of which yield similar results.70 Overall, the results from comparing the performance of the five PMT models suggest the following conclusions: • Being able to include household size as a predictor of household welfare improves a minimal PMT’s performance significantly. The government should discuss the likelihood that potential beneficiaries could falsify this information and possible means of limiting such falsification. • Being able to include indicators of assets owned is also helpful, so the same suggestions apply. • Being able to include indicators from potential cross-checks against other databases such as the civil service payroll, pension fund rolls, electricity billing, and water billing are also helpful, but not quite as useful as the asset indicators. But if the government doubts the veracity of the asset indicators, it may be able to substitute them with verifiable indicators that perform almost as well. • On the other hand, if the government has confidence in the asset indicators, including the variables that might be cross-checked adds little to the model. • The rank correlation of the PMT model with actual household welfare is very high for models (2) to (5) (ranging from 0.79 to 0.86) so a PMT model would be useful for ranking potential beneficiaries from the most to least needy. However, this correlation drops significantly if we focus only on the poorest half or quarter of the population, so a PMT model performs less well in ranking only (extremely) poor households. • Most importantly, social assistance program benefits targeted with the worst of these models would be as progressive or more so than any existing social assistance program in Eswatini, perhaps much more so if the best of the models can be used. Preferred PMT model Based on the estimation results and the performance criteria laid out at the beginning of this appendix, we prefer model (2), estimated based on ordinary least squares. While this model is not quite as accurate as those that include more variables, the few variables it includes are easily and quickly observed and difficult to falsify. Table A4.1 reports the estimated coefficients for the preferred model along with their standard errors and the means for each variable. Figures A4.2 and A4.3 report actual and predicted poverty rates based on model (2) by welfare decile and by region-by-area. For the comparison, we have used a “matched” cutoff for the PMT score that produces poverty rates equal to the actual overall and extreme poverty rates at the national level. 70 Results available from the authors upon request. 103 Figure A4.1. Actual and Predicted Welfare, by Model Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Welfare = consumption per adult equivalent. 104 Table A4.1. Preferred PMT Model, Estimates Variable type Variable Mean Coefficient Standard error (1) (2) (3) Hhohho, rural 0.177 –0.276 0.039 Manzini, urban 0.160 0.173 0.038 Region and area Manzini, rural 0.215 –0.179 0.038 (Reference category is Shiselweni, urban 0.019 –0.224 0.073 Hhohho, urban) Shiselweni, rural 0.144 –0.216 0.042 Lubombo, urban 0.059 –0.371 0.049 Lubombo, rural 0.135 –0.497 0.042 Floor Cement 0.816 0.161 0.045 (Reference category is earth or dung) High quality 0.130 0.460 0.055 Roof High quality 0.860 0.248 0.047 (Reference category is grass or traditional) Metal 0.091 0.282 0.059 Piped, indoor 0.202 0.388 0.041 Water source Piped, yard 0.298 0.191 0.028 (Reference source is Public tap 0.113 0.012 0.034 surface water or other Tube well 0.087 0.096 0.037 source) Protected spring 0.044 0.015 0.048 Delivered 0.045 0.295 0.048 Flush toilet, sewer, or septic 0.220 0.533 0.042 Toilet facility Flush toilet, elsewhere 0.013 0.502 0.086 (Reference category is no VIP latrine 0.021 0.162 0.068 toilet facility) latrine with slab 0.555 0.148 0.025 Number of rooms in home Number of rooms in home 2.838 0.103 0.006 Two person household 0.128 –0.339 0.032 Three person household 0.141 –0.553 0.032 Household size Four person household 0.137 –0.750 0.032 (Reference category is Five person household 0.107 –0.893 0.036 single person household) Six person household 0.087 –1.064 0.039 Seven+ person household 0.168 –1.245 0.033 Constant 1.000 6.876 0.064 Adjusted R-squared statistic 0.659 Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: VIP = ventilated pit latrine. 105 Figure A4.1. Actual and Predicted Poverty Rates, by Welfare Decile a. Overall poverty b. Extreme poverty Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Predicted poverty rates are based on the preferred PMT model, namely PMT model (2). Predicted rates are obtained by applying the matched cutoff, that is, the percentile of the PMT score distribution equal to the actual overall or extreme poverty rate at the national level. Welfare = consumption per adult equivalent. 106 Figure A4.2. Actual and Predicted Poverty Rates, by Region and Area a. Overall poverty b. Extreme poverty Source: Authors’ estimates based on data from the Swaziland Household Income and Expenditure Survey 2016/17. Note: Predicted poverty rates are based on the preferred PMT model, namely PMT model (2). Predicted rates are obtained by applying the matched cutoff, that is, the percentile of the PMT score distribution equal to the actual overall or extreme poverty rate at the national level. Welfare = consumption per adult equivalent. 107 Social Protection & Jobs Discussion Paper Series Titles 2019-2021 No. Title 2106 Social Assistance Programs and Household Welfare in Eswatini by Dhushyanth Raju and Stephen D. Younger June 2021 2105 The Coal Transition: Mitigating Social and Labor Impacts by Wendy Cunningham and Achim Schmillen May 2021 2104 Social Protection at the Humanitarian-Development Nexus: Insights from Yemen by Yashodhan Ghorpade and Ali Ammar April 2021 2103 Review of the Evidence on Short-Term Education and Skills Training Programs for Out-of-School Youth with a Focus on the Use of Incentives by Marguerite Clarke, Meghna Sharma, and Pradyumna Bhattacharjee January 2021 2102 Welfare, Shocks, and Government Spending on Social Protection Programs in Lesotho by Joachim Boko, Dhushyanth Raju, and Stephen D. Younger January 2021 2101 Cash in the City: Emerging Lessons from Implementing Cash Transfers in Urban Africa by Ugo Gentilini, Saksham Khosla, and Mohamed Almenfi January 2021 2011 Building the Foundation for Accountability in Ethiopia by Laura Campbell, Fitsum Zewdu Mulugeta, Asmelash Haile Tsegay, and Brian Wampler January 2020 2010 Safety nets, health crises and natural disasters: Lessons from Sierra Leone by Judith Sandford, Sumati Rajput, Sarah Coll-Black, and Abu Kargbo December 2020 2009 A Reforma do Bolsa Família: Avaliação das propostas de reforma debatidas em 2019 by Matteo Morgandi, Liliana D. Sousa, Alison Farias, e Fabio Cereda November 2020 2008 The Role of Social Protection in Building, Protecting, and Deploying Human Capital in the East Asia and Pacific Region by Harry Edmund Moroz October 2020 2007 Boosting the Benefits of Cash Transfer Programs During the Early Years: A Case Study Review of Accompanying Measures by Laura Rawlings, Julieta Trias, and Emma Willenborg October 2020 2006 Expansion of Djibouti’s National Family Solidarity Program: Understanding Targeting Performance of the Updated Proxy Means Test Formula by Vibhuti Mendiratta, Amr Moubarak, Gabriel Lara Ibarra, John van Dyck, and Marco Santacroce August 2020 2005 Assessing the Targeting System in Georgia: Proposed Reform Options by Maddalena Honorati, Roberto Claudio Sormani, and Ludovico Carraro July 2020 2004 Jobs at risk in Turkey: Identifying the impact of COVID-19 by Sirma Demir Şeker, Efşan Nas Özen, and Ayşenur Acar Erdoğan July 2020 2003 Assessing the Vulnerability of Armenian Temporary Labor Migrants during the COVID-19 pandemic by Maddalena Honorati, Soonhwa Yi, and Thelma Choi July 2020 2002 Getting it Right: Strengthening Gender Outcomes in South Sudan by Samantha de Silva, Abir Hasan, Aissatou Ouedraogo, and Eliana Rubiano-Matulevich July 2020 2001 The Science of Adult Literacy by Michael S. C. Thomas, Victoria C. P. Knowland, Cathy Rogers January 2020 1936 Moving forward with ALMPs: Active labor policy and the changing nature of labor markets by Jose Manuel Romero and Arvo Kuddo November 2019 1935 Unbundled: A framework for connecting safety nets and humanitarian assistance in refugee settings by Karin Seyfert, Valentina Barca, Ugo Gentilini, Manjula Luthria, and Shereen Abbady September 2019 1934 Decentralization’s effects on education and health: Evidence from Ethiopia by Jean-Paul Faguet, Qaiser Khan, and Devarakonda Priyanka Kanth September 2019 1933 Extending Pension Coverage to the Informal Sector in Africa by Melis Guven July 2019 1932 What Employers Actually Want - Skills in demand in online job vacancies in Ukraine by Noël Muller and Abla Safir May 2019 1931 Can Local Participatory Programs Enhance Public Confidence: Insights from the Local Initiatives Support Program in Russia by Ivan Shulga, Lev Shilov, Anna Sukhova, and Peter Pojarski May 2019 1930 Social Protection in an Era of Increasing Uncertainty and Disruption: Social Risk Management 2.0 by Steen Lau Jorgensen and Paul B. Siegel May 2019 1929 Developing Coherent Pension Systems: Design Issues for Private Pension Supplements to NDC Schemes by William Price April 2019 1928 Pensions in a Globalizing World: How Do (N)DC and (N)DB Schemes Fare and Compare on Portability and Taxation? by Bernd Genser and Robert Holzmann April 2019 1927 The Politics of NDC Pension Scheme Diffusion: Constraints and Drivers by Igor Guardiancich, R. Kent Weaver, Gustavo Demarco, and Mark C. Dorfman April 2019 1926 Setting Up a Communication Package for the Italian NDC by Tito Boeri, Maria Cozzolino, and Edoardo Di Porto April 2019 1925 Sweden’s Fifteen Years of Communication Efforts by María del Carmen Boado-Penas, Ole Settergren, Erland Ekheden, and Poontavika Naka April 2019 1924 Information and Financial Literacy for Socially Sustainable NDC Pension Schemes by Elsa Fornero, Noemi Oggero, and Riccardo Puglisi April 2019 1923 Communicating NEST Pensions for “New” DC Savers in the United Kingdom by Will Sandbrook and Ranila Ravi-Burslem April 2019 1922 Harnessing a Young Nation's Demographic Dividends through a Universal NDC Pension Scheme: A Case Study of Tanzania by Bo Larsson, Vincent Leyaro, and Edward Palmer April 2019 1921 The Notional and the Real in China’s Pension Reforms by Bei Lu, John Piggott, and Bingwen Zheng April 2019 1920 Administrative Requirements and Prospects for Universal NDCs in Emerging Economies by Robert Palacios April 2019 1919 Bridging Partner Lifecycle Earnings and Pension Gaps by Sharing NDC Accounts by Anna Klerby, Bo Larsson, and Edward Palmer April 2019 1918 The Impact of Lifetime Events on Pensions: NDC Schemes in Poland, Italy, and Sweden and the Point Scheme in Germany by Agnieszka Chłoń-Domińczak, Marek Góra, Irena E. Kotowska, Iga Magda, Anna Ruzik-Sierdzińska, and Paweł Strzelecki April 2019 1917 Drivers of the Gender Gap in Pensions: Evidence from EU-SILC and the OECD Pension Model by Maciej Lis and Boele Bonthuis April 2019 1916 Gender and Family: Conceptual Overview by Nicholas Barr April 2019 1915 Labor Market Participation and Postponed Retirement in Central and Eastern Europe by Robert I. Gal and Márta Radó April 2019 1914 NDC Schemes and the Labor Market: Issues and Options by Robert Holzmann, David Robalino, and Hernan Winkler April 2019 1913 NDC Schemes and Heterogeneity in Longevity: Proposals for Redesign by Robert Holzmann, Jennifer Alonso-García, Heloise Labit-Hardy, and Andrés M. Villegas April 2019 1912 Annuities in (N)DC Pension Schemes: Design, Heterogeneity, and Estimation Issues by Edward Palmer and Yuwei Zhao de Gosson de Varennes April 2019 1911 Overview on Heterogeneity in Longevity and Pension Schemes by Ron Lee and Miguel Sanchez-Romero April 2019 1910 Chile's Solidarity Pillar: A Benchmark for Adjoining Zero Pillar with DC Schemes by Eduardo Fajnzylber April 2019 1909 Sweden: Adjoining the Guarantee Pension with NDC by Kenneth Nelson, Rense Nieuwenhuis, and Susanne Alm April 2019 1908 The ABCs of NDCs by Robert Holzmann April 2019 1907 NDC: The Generic Old-Age Pension Scheme by Marek Góra and Edward Palmer April 2019 1906 The Greek Pension Reforms: Crises and NDC Attempts Awaiting Completion by Milton Nektarios and Platon Tinios April 2019 1905 The Norwegian NDC Scheme: Balancing Risk Sharing and Redistribution by Nils Martin Stølen, Dennis Fredriksen, Erik Hernæs, and Erling Holmøy April 2019 1904 The Polish NDC Scheme: Success in the Face of Adversity by Sonia Buchholtz, Agnieszka Chłoń-Domińczak, and Marek Góra April 2019 1903 The Italian NDC Scheme: Evolution and Remaining Potholes by Sandro Gronchi, Sergio Nisticò, and Mirko Bevilacqua April 2019 1902 The Latvian NDC Scheme: Success Under a Decreasing Labor Force by Edward Palmer and Sandra Stabina April 2019 1901 The Swedish NDC Scheme: Success on Track with Room for Reflection by Edward Palmer and Bo Könberg April 2019 To view Social Protection & Jobs Discussion Papers published prior to 2019, please visit www.worldbank.org/sp. ABSTRACT Eswatini has notably high levels of poverty and inequality. Recurrent, negative shocks are an important contributing factor. This study assesses the performance of the largest social assistance programs in Eswatini, based on 2016−2017 national household survey data. It examines the coverage rates of these programs, and their incidence and effectiveness in reducing poverty and inequality. The study also examines the association between program participation and negative shocks reported by households, in particular, drought and food price shocks associated with the 2015−2016 El Niño event. Across programs, benefits are concentrated among poor households. However, the performance of programs in reducing poverty and inequality tends to be limited because of low intended or actual benefit levels and shortfalls in intended or actual coverage of the poor. Households that receive program benefits are more likely to report a drought shock. Except in the case of emergency food aid, which is provided ex post, we interpret this pattern to indicate that programs tend to provide ex-ante coverage to those vulnerable to this shock. At a minimum, enhancing the performance of programs in addressing poverty, inequality, and the adverse effects of shocks would require that actual benefit levels equal intended levels (for example, by procuring sufficient food commodities to meet the needs of the school feeding program) and that intended benefit levels are fully aligned with program aims (for example, by providing grant amounts to schools that are large enough to allow for tuition-free government secondary education for orphaned and vulnerable children). Absent greater budgetary allocations to programs, addressing these benefit-related disconnects may require improving the targeting of select program benefits to poorer households such as by using a proxy means test. We simulate the effects of programs on poverty and inequality reduction from such hypothetical reforms. ABOUT THIS SERIES Social Protection & Jobs Discussion Papers are published to communicate the results of The World Bank’s work to the development community with the least possible delay. This paper therefore has not been prepared in accordance with the procedures appropriate for formally edited texts. For more information, please contact the Social Protection Advisory Service via e-mail: socialprotection@ worldbank.org or visit us on-line at www.worldbank.org/sp