MENA REGIONAL PROGRAM WORKING PAPER 40 SEPTEMBER 2022 Impact Evaluation Report: Egypt’s Takaful Cash Transfer Program Second Round Report Hoda El Enbaby, Dalia Elsabbagh, Dan Gilligan, Naureen Karachiwalla, Bastien Koch, and Sikandra Kurdi 1 CONTENTS Executive Summary .......................................................................................................... vii Background And Methodology ....................................................................................... vii Takaful Impact Results .................................................................................................. viii Recommendations .......................................................................................................... ix 1. Introduction ................................................................................................................... 11 1.1 Context For The Initiation Of The Takaful And Karama Program ............................. 11 1.2 Evidence Of Impact From The First-Round Evaluation Of The Takaful And Karama Program ........................................................................................................................ 12 1.3 Covid-19 And The Government Of Egypt’s Response ............................................. 12 1.4 The Second Round Of Impact Evaluation Of The Takaful And Karama Program .... 13 1.5 Outline Of This Report ............................................................................................. 14 2. Takaful And Karama Program ...................................................................................... 14 2.1 Program Description ................................................................................................ 14 2.2 Targeting ................................................................................................................. 14 2.3 Program Experience ................................................................................................ 15 3. Impact Evaluation Design ..................................................................................... 18 3.1 Motivation For The Regression Discontinuity Approach ........................................... 18 3.2 Sample Selection .................................................................................................... 19 3.3 Heterogeneity Analysis ............................................................................................ 19 3.4 Regression Discontinuity Validity ............................................................................. 20 4. Sample And Survey Data ...................................................................................... 25 4.1 Sample .................................................................................................................... 25 4.2 Data Collection ........................................................................................................ 26 4.3 Survey ..................................................................................................................... 27 5. Summary Statistics For The Impact Analysis Sample ........................................ 28 5.1 Household Characteristics ....................................................................................... 28 6. Impact Of The Takaful Program ............................................................................ 29 6.1 Variables And Outcomes ......................................................................................... 29 6.2 Household Total Expenditure And Poverty .............................................................. 31 ii 6.3 Components Of Household Consumption ............................................................... 33 6.3.1 Household Composition ....................................................................................... 37 6.4 Assets, Savings, And Debt ...................................................................................... 39 6.5 Household Labor ..................................................................................................... 44 6.6 Child Schooling And Child Labour ........................................................................... 45 6.7 Household, Mother, And Child Dietary Diversity ...................................................... 48 6.8 Mother And Child Anthropometry, Overweight And Wasting, And Morbidity And Treatment ...................................................................................................................... 49 6.9 Antenatal And Postnatal Care And Infant And Young Child Feeding (Iycf) Practices52 6.10 Women’s Decision-Making And Gender Norms ..................................................... 53 6.11 Mental Health ........................................................................................................ 58 6.12 Shocks And Coping Strategies .............................................................................. 59 6.13 Covid-19 ................................................................................................................ 62 6.14 Robustness To Alternate Definition Of Beneficiary Status ..................................... 63 7. Conclusions And Recommendations ............................................................................ 1 About The Authors.............................................................................................................. 5 Acknowledgments .............................................................................................................. 5 References .......................................................................................................................... 5 TABLES Table 2.3.1. Self-reported Transfer Amounts………………………………………………. 16 Table 2.3.2. Challenges in Applying to Takaful…………………………………………….17 Table 2.3.3. Challenges in Receiving Transfers…………………………………………….17 Table 3.4.1. Beneficiary status as determined by the PMT threshold…………………..21 Table 4.2.1 Distribution of Sample by Governorate...……………………………………...26 Table 5.1.1 Summary Statistics of Household Demographic Characteristics…………29 Table 6.1.1. Current beneficiary versus ever beneficiary status…………………………30 Table 6.2.1 Impacts of Takaful program on household consumption expenditure…..32 Table 6.2.2. Impacts of Takaful program on household poverty measures……………33 iii Table 6.3.1. Impacts of Takaful program on food consumption expenditure by category…………………………………………………………………………………………….35 Table 6.3.2. Impacts of Takaful program on non-food consumption expenditure by category…………………………………………………………………………………………….36 Table 6.3.3. Household Composition………………………………………………………….38 Table 6.4.1. Impacts of Takaful program on asset indices………………………………...40 Table 6.4.2. Impacts of Takaful program on individual agricultural assets …………….41 Table 6.4.3. Impacts of Takaful program on livestock…………………………………….42 Table 6.4.4. Impacts of Takaful program on household savings and debt…………… .44 Table 6.5.1. Impacts of Takaful program on employment outcomes…………………….45 Table 6.6.1. Impacts of Takaful program on enrolment and attendance………………..47 Table 6.7.1. Number of food groups consumed by households …………………………48 Table 6.7.2 Impacts of Takaful program on dietary diversity outcomes……………………………………………………………………………………………49 Table 6.8.1. Impacts of Takaful program on mother anthropometrics outcomes……..50 Table 6.8.2. Impacts of Takaful program on child anthropometrics outcomes………..51 Table 6.8.3. Impacts of Takaful program on child stunting and wasting……………….52 Table 6.9.1. Impacts of Takaful program on antenatal care (ANC) outcomes…………53 Table 6.9.2. Impacts of Takaful program on Infant and Young Child Feeding (IYCF) practices……………………………………………………………………………………………53 Table 6.10.1 Impacts of Takaful program on women's decision-making, all women…55 Table 6.10.2. Impacts of Takaful program on women's decision-making for women with some formal education……………………………………………………………………..56 Table 6.10.3. Impacts of Takaful program on women's decision-making for women with no formal education…………………………………………………………………………57 Table 6.10.4. Impacts of Takaful program on gender norms ………………………………58 Table 6.11.1. Impacts of Takaful program on mental health indicators………………….59 Table 6.12.1 Impacts of Takaful program on shocks experienced by household……..60 Table 6.12.2. Impacts of Takaful program on coping methods for shocks experienced by the household………………………………………………………………………………….61 Table 6.13.1. Impacts of Takaful program on experiences with the COVID-19 pandemic……………………………………………………………………………………………63 Table 6.14.1. Impacts of Takaful program duration (self-reported)………………………65 iv FIGURES Figure 2.3.1. Program Satisfaction…………………………………………………………….16 Figure 3.1.1 Intuition Behind Regression Discontinuity…………………………………...19 Figure 3.4.1: Probability of having received a Takaful transfer in the past two months, by PMT score (Self-Reported)…………………………………………………………………..23 Figure 3.4.2: Probability of ever having received a Takaful transfer, by PMT score (Self-Reported)…………………………………………………………………………………….24 Figure 3.4.3: Number of months transfers were received (Self-Reported)……………..24 v ACRONYMS AEU Adult Equivalent Unit BMI Body Mass Index CAPI Computer-Assisted Personal Interview CAPMAS Central Agency for Public Mobilization and Statistics CCT Conditional cash transfer CPI Consumer Price Index DHS Demographic Household Survey GAD Generalized Anxiety Disorder GDP Gross Domestic Product HH Household HIECS Household Income, Expenditure, and Consumption Survey IFPRI International Food Policy Research Institute IRB Internal Review Board IYCF Infant and Young Child Feeding IYCN Infant and Young Child Nutrition MoSS Ministry of Social Solidarity PCA Principal components analysis PMT Proxy Means Test RD Regression Discontinuity RDD Regression Discontinuity Design TKP Takaful and Karama Program WHO World Health Organization WHZ Weight-for-Height Z-score vi EXECUTIVE SUMMARY BACKGROUND AND METHODOLOGY Egypt introduced the Takaful and Karama Program (TKP), a pair of targeted cash transfer schemes in March 2015. Takaful and Karama was designed as a conditional cash transfer program providing income support targeted to the poor and most vulnerable; namely poor families with children (under 18 years of age), poor elderly (aged 65 years and above) and persons with severe disability. Originally implemented as an unconditional cash transfer, the program is now a conditional cash transfer program, but the conditionalities have yet to be monitored. Starting July 2017, households received EGP60 for each child under 6 years old, EGP80 for each child in primary education, EGP100 for children in preparatory educa- tion, and EGP140 for secondary education. As of June 2017, 90% of TKP beneficiaries were women. In 2018, the International Food Policy Research Institute (IFPRI) completed the first round of impact evaluation of TKP, based on household survey data collected after the first 15 months of the program. The evaluation found that TKP substantially improved wellbeing for poor households, increasing household consumption per adult equivalent by 8.4 percent. and reducing the probability that a beneficiary household is poor (< USD1.90 per capita per day) by 11.4 percentage points, which is comparable to several of the well- known, large-scale programs in Latin America where consumption impacts are on the order of 7-8 percent. In the period between the first-round evaluation and the data collection for this report in January 2022, Egypt faced an enormous economic shock in the COVID-19 pan- demic with a complete loss of tourism, which before the crisis was responsible for 12% of GDP and 10% of employment (IMF, 2021). Partial lockdowns and restrictions on large public gatherings further reduced economic activity. Tax revenues decreased substan- tially, and international investors withdrew. The Ministry of Social Solidarity expanded target- ing of Takaful and Karama, adding an additional 411,000 households to the beneficiary list. Egypt undertook several measures to respond to the economic crisis posed by the pan- demic, including expansion in some safety net programs and introduction of some new so- cial protection programs. This follow-up evaluation was designed to assess whether impacts estimated from the first round have been sustained and whether longer duration of treatment has led to impacts on additional outcome variables. The follow-up evaluation focuses only on Takaful. The evaluation assesses the program’s impact on indicators that have been cov- ered by the first evaluation and adds new insights on some outcomes, such as COVID-19 related impacts and responses. Data from a household survey collected in January and Feb- ruary 2022, and administrative data on registration into the program are used for the impact evaluation. This follow-up impact evaluation was conducted using a regression discontinuity (RD) design similar to the first round but using a different sample of households much more tightly concentrated around the 4500 threshold to address concerns about the smaller discontinuity. Regression discontinuity is the most scientifically rigorous vii methodology available given the program’s approach to targeting and timing of the evalua- tion. TAKAFUL IMPACT RESULTS Households invested in assets, particularly productive assets. There are large and sig- nificantly positive effects on household asset holdings. Beneficiary households own more as- sets in general, and importantly, they own more productive and livestock assets. Specifically, households invested in tractors, plows, and irrigation. With regards to livestock, households purchased buffaloes, cattle, goats, and sheep, and reduced their investments in smaller ani- mals such as chickens, geese, pigeons, and ducks. These large and lumpy productive and livestock investments are important for future income generation. Many households cannot afford to make these larger investments and the Takaful program may have given house- holds the ability to redirect funds for these investments. Takaful households reduced their debt burdens. We see statistically significant differ- ences between beneficiaries and non-beneficiaries with beneficiary households carrying substantially less debt. This is also a positive result as it frees households from being forced to pay down loans for specific items rather than using the money in more beneficial ways, and because it reduces the burden of interest. There were no measurable impacts of the Takaful program on household consump- tion or poverty. Compared to the first round, statistical power of the analysis was more limited with results not being statistically distinguishable from zero, although the confidence intervals do rule out that the impacts on consumption could have been as large as those found in the first round. Importantly, we note that the consumption aggregates are calcu- lated per adult equivalent unit (AEU), which adjusts for the amount of consumption needed by adult and child household members (children are given less weight). In particular, house- holds reduced their consumption of grains, eggs, oils and fats, and fruits. Households also spent less on construction and communications. Changes in household demographics may partly explain the lack of impact on con- sumption. We find that Takaful beneficiaries had more household members, and particu- larly more children 0-5 years old and 6-11 years old. The higher number of household mem- bers ages 6-11 years old is likely an artifact of the PMT score construction which was not completely near the 4500 cutoff. We also see some suggestive evidence that Takaful in- duced increases in number of children born in the past 5 years, including among households that already had at least 2 children at the time of registration. Takaful changed household labor patterns. We see a small difference in the types of oc- cupations among beneficiaries. Beneficiaries are significantly more likely to be employed in the informal sector. Takaful improved school enrollment and attendance. Children of primary school age were almost 9 percentage points more likely to be enrolled in school and children of prepara- tory school age were 21 percentage points more likely to be enrolled in school. Further, we see improvements in attendance rates among secondary school children. Secondary school children who are enrolled in school attend more regularly, and the result is driven by in- viii creased participation among girls. This positive result will have implications for future gener- ations – increased education among women is associated with several positive outcomes for their children (better nutrition, higher learning levels, higher earning potential, etc.). Impacts on nutrition outcomes are weak and mixed. Household level dietary is reduced household level dietary diversity in one specification. One the other hand, beneficiary house- holds had lower rates of wasting among children 6-23 months of age. Mother’s anthropomet- ric outcomes were also not affected. Women’s ability to make decisions within their households did not differ between beneficiaries and non-beneficiaries. While the first-round evaluation found decreases in measures of women’s decision-making power, particularly among women with no formal ed- ucation, we do not find this same pattern in our analysis. Overall, there are no impacts in any domain of decision-making. For the sample of women who have some formal education, there is a slight negative effect on decision-making regarding taking children to the doctor, and for women without any formal education there is a slight positive impact on deciding what food to cook. However, overall, Takaful did not have any substantive impact on women’s decision-making within their households. We also see some evidence of higher lev- els of gender positive norms among beneficiary households. We do not find any effects on mental health – worries, generalized anxiety, or self-esteem. Takaful contributed to households’ ability to cope with shocks. When faced with shocks in the past five years, the predominant method Takaful households used to cope was selling gold/jewelry. We also see a reduction in borrowing to cope with shocks. This is a positive re- sult since it is coping strategy with less potential for long-term negative impacts than others such as child labor or having daughters marry early. RECOMMENDATIONS Takaful should be continued and even possibly extended. The program enabled house- holds not to resort to coping with shocks in negative ways. Particularly considering increas- ingly frequent global shocks like COVID-19 and the Russian invasion of Ukraine, social pro- tection programs, including cash transfer programs like Takaful, could be an effective way to protect against large-scale shocks since the infrastructure to reach people is largely in place. Proceed with plans for recertification and graduation of beneficiaries who have achieved self-sufficiency while using a generous cut-off for self-sufficiency given that many house- holds have not managed to substantially increase their consumption despite increased pro- ductive assets. Improve communication regarding exclusion restrictions, program length, and recerti- fication so that beneficiaries understand that they will not be excluded from the program for formal sector work with income below a certain threshold and to ensure that beneficiaries are not surprised by sudden changes in program status or unnecessarily worried about the short-term continuity of the transfers. Consider greater coordination with communication campaigns related to family plan- ning if the behavioral response by families of having more children is confirmed and seen as in conflict with other national policy goals. ix Continue to work towards a comprehensive social protection strategy that helps to continue protecting the poor as well as contributing to achieving longer-term developmental goals. Coordinating with the Ministry of Education to provide high quality public service deliv- ery will magnify the impacts of increased school enrollment and with the Ministry of Health regarding diets and nutrition. Complementary programming would also be beneficial. In general, complementary pro- gramming on issues such as nutrition practices or financial training need to be quite inten- sive to be impactful. There are currently programs that are implemented by the Government of Egypt on these topics, particularly a nation-wide nutrition campaign. However, it would be worth considering pairing these programs and intensifying them by leveraging Takaful to link to already vulnerable households. x 1. INTRODUCTION 1.1 Context for the initiation of the Takaful and Karama Program Since 2014, Egypt has undertaken a broad program of macroeconomic reforms designed to reduce government spending, liberalize the economy, promote employment and economic growth, and strengthen the social safety net. These reforms included a devaluation of the ex- change rate, substantial reductions in energy subsidies, changes to the tax code, and freez- ing of public sector hiring (World Bank 2019). The government also initiated a series of changes to social programs, in part to help smooth the effects of the economic reforms and to improve the functioning of the social safety net through changes in program designs and improved targeting. Egypt has long provided substantial social support. Major programs include a large social solidarity pension and a system of broad food subsidies initiated after World War II. These reach millions of households, but they are costly and inefficient as a form of redistribution. The food subsidy alone cost 1.6% of GDP and reached 70% of the Egyptian population (Ministry of Finance, 2017). In 2015, the government reformed food subsidies, instituting a ration card that served as a voucher for discounted purchases of selected food items (Moselhy, 2017; Ecker et al. 2016). During the macroeconomic reforms, the government in- creased the size of voucher payments to help offset the negative impacts of the reforms (Breisinger et al. 2018). In the context of these reforms, Egypt introduced the Takaful and Karama Program (TKP), a pair of targeted cash transfer schemes in March 2015. Takaful and Karama was designed as a conditional cash transfer program providing income support targeted to the poor and most vulnerable; namely poor families with children (under 18 years of age), poor elderly (aged 65 years and above) and persons with severe disability. The introduction of the program repre- sented a significant step on behalf of the Egyptian government to increase the share of so- cial spending reaching poor households. Takaful and Karama is implemented by the Ministry of Social Solidarity (MoSS), and co-financed by the Government of Egypt and the World Bank. Takaful (Solidarity), the larger of the two programs, is a family income support scheme. The program was initially rolled out as an unconditional cash transfer, but planned education and health conditionalities were introduced in 2018. Continued receipt of Takaful transfers was conditioned on school children aged 6-18 years maintaining attendance of at least 80% of the school days, and on mothers and children below 6 years completing three visits to health clinics per year, in addition to maintaining child growth monitoring records, and attending nutrition awareness sessions. Takaful transfers start from a basic amount of EGP325 per household, per month, which increases depending on the number of children in the households and their educational level. At the beginning of the program, household re- ceived EGP60 for each child in primary education, EGP80 for each child in preparatory edu- cation and EGP100 in secondary education. The nominal value of the transfers was in- creased over time. For example, starting July 2017, households received EGP60 for each child under 6 years old, EGP80 for each child in primary education, EGP100 for children in preparatory education, and EGP140 for secondary education. Households can receive bene- fits for up to three children only, who are usually the oldest three children in the households. Karama (Dignity) is an unconditional income support scheme targeted at the poor elderly 11 and persons with severe disability, and orphans. Orphans were added as Karama beneficiar- ies in 2017 and receive EGP350 per beneficiary. Karama monthly transfers for poor elderly and person with disability started at EGP350 per beneficiary and were increased in July 2017 to EGP450 per beneficiary. Karama also has a maximum of three beneficiaries per household (World Bank, 2015). Families can be entitled to both Takaful and Karama bene- fits. As of June 2017, 90% of TKP beneficiaries were women. 1.2 Evidence of impact from the first-round evaluation of the Taka- ful and Karama Program In 2018, the International Food Policy Research Institute (IFPRI) completed the first round of impact evaluation of TKP, based on household survey data collected after the first 15 months of the program. The evaluation was conducted using a regression discontinuity de- sign, a methodology that provides rigorous estimates of program impact for programs like TKP targeted using a proxy means score with a fixed threshold for household eligibility. The evaluation found that TKP substantially improved wellbeing for poor households, increasing household consumption per adult equivalent by 8.4 percent. and reducing the probability that a beneficiary household is poor (< USD1.90 per capita per day) by 11.4 percentage points. The consumption impact of Takaful is comparable to several of the well-known, large-scale programs in Latin America. A review of conditional cash transfer programs in Latin America (Fizbein et al, 2009) found that impacts on household expenditure ranged from 7-10 percent among four programs in Brazil, Mexico, Colombia, and Honduras that providing transfers of similar size to Takaful as a share of household expenditure. The first-round impact evalua- tion also showed that Takaful increased the value of household food consumption and im- proved the quality of household diets. The first round of evaluation also identified some limitations in the program’s impact. For ex- ample, there were no significant impacts of Takaful on school enrollment or heath care utili- zation, which may be explained by the absence of conditionalities at the time of the evalua- tion. Also, estimates showed a negative and significant impact of the program on women’s control over decision making, which was driven primarily by households in Lower Egypt and by women with less than primary education. The first-round evaluation did not find any measurable impacts of the Karama program, due to methodological challenges introduced by variation in the eligibility threshold as program leaders sought to expand its coverage. 1.3 COVID-19 and the Government of Egypt’s response In the period between the first-round evaluation and the data collection for this report in Jan- uary 2022, Egypt faced an enormous economic shock in the COVID-19 pandemic. The pan- demic had several immediate, severe negative effects on the economy. For example, Egypt experienced a complete loss of tourism, which before the crisis was responsible for 12% of GDP and 10% of employment (IMF, 2021). Partial lockdowns and restrictions on large public gatherings further reduced economic activity. In addition, tax revenues fell and the country experienced capital outflows of more than $15 billion as investors withdrew from emerging markets. Despite these challenges, Egypt was one of the few emerging markets to experi- ence positive, though modest, economic growth in 2020. Economic growth was projected to further accelerate in 2021, to 5.5% (World Bank 2022). 12 Egypt undertook several measures to respond to the economic crisis posed by the pan- demic, including expansion in some safety net programs and introduction of some new so- cial protection programs. These changes included an increase in the social pension and new transfers targeted to pregnant women, women with young children, individuals with disabili- ties, the elderly, and informal and self-employed workers, such as in the tourism and agricul- ture sectors (Gentilini, 2022). In addition, the Ministry of Social Solidarity expanded targeting of Takaful and Karama, adding an additional 411,000 households to the beneficiary list. With this expansion, through July 2021, the project has reached 3.37 million direct project benefi- ciaries (including 75% women), while 11.85 million direct and indirect beneficiaries benefited from the project. Not surprisingly, these changes in eligibility rules used to expand the TKP created some challenges in measuring impacts of the program, which we discuss below. 1.4 The second round of impact evaluation of the Takaful and Karama Program In 2018, the World Bank contracted IFPRI to conduct the first round of impact evaluation of the Takaful and Karama Program, in conjunction with the Ministry of Social Solidarity and with funding from the United Kingdom Foreign and Commonwealth Office (UK FCO). For the second round, IFPRI has been contracted by the World Bank again, this time with funding from the United States Agency for International Development (USAID) for the data collection. As with the first-round evaluation, IFPRI has worked in conjunction with the Ministry of Social Solidarity in conducting this second-round impact evaluation of the Takaful program. The follow-up evaluation focuses only on Takaful, which is the largest of the two programs, and where more human capital accumulation is expected with the implementation of the con- ditionalities. The main objective of conducting a second round of evaluation is to assess whether impacts estimated from the first round have been sustained and whether longer du- ration of treatment has led to impacts on additional outcome variables. This evaluation round also follows the implementation of conditionalities, which were not in place when the first evaluation was conducted. The second round of evaluation rigorously assesses the pro- gram’s impact on indicators that have been covered by the first evaluation, such as house- hold consumption, poverty, asset ownership, as well as other measures of well-being, such as the prevalence of overweight and obesity in adult women and children, as well children’s education and health. This second round of evaluation also adds new insights on some out- comes, such as COVID-19 related impacts and responses. Data from a household survey collected in January and February 2022, and administrative data on registration into the pro- gram are used for the impact evaluation. The Takaful program continues to be targeted using a Proxy Means Test (PMT), which is an index of well-being based on household demographics, income, housing quality, assets, and other characteristics. Households with a PMT score below a preset threshold are considered eligible for the program, while those above it are not. The PMT cut-off point as well as the PMT function used by MOSS have been changed between both evaluations. The design of the program creates a quasi-experiment which IFPRI researchers utilized to rigorously as- sess the impact of the program. 13 1.5 Outline of this Report The remainder of this report is organized as follows Chapter 2 provides a brief overview and updates about the Takaful Program. Chapter 3 presents the impact evaluation design. Chap- ter 4 describes the evaluation survey and sample. Chapter 5 summarizes beneficiary and non-beneficiary household characteristics, providing context to the study. Chapter 6 presents the impact estimates for Takaful. Finally, Chapter 7 concludes and provides policy recom- mendations. 2. TAKAFUL AND KARAMA PROGRAM 2.1 Program Description Egypt launched the Takaful and Karama Program in March 2015. The initiation of the pro- gram followed the implementation of a series of economic reforms that were implemented in Egypt starting 2014. The program aimed to protect the most vulnerable population groups from the impacts of the economic reforms, as well as to improve the targeting of social pro- tection in Egypt (World Bank, 2015). Takaful and Karama is a conditional cash transfer pro- gram that seeks to provide income support to the poor and most vulnerable; namely poor families with children (under 18 years of age), poor elderly (aged 65 years and above) and persons with severe disability. The program has two main sub-programs: Takaful, and Karama. “Takaful” (Solidarity) is a family income support scheme aimed at protecting poor families with children. The program is de jure conditioned on school attendance and health outcomes but de facto, the conditionalities are not enforced. Receiving the cash transfers is conditional on attendance of at least 80% of the school days by children aged 6–18 years, and on re- ceiving three health monitoring visits per year, by mothers and children below 6 years in ad- dition to maintaining child growth monitoring records and attending nutrition awareness ses- sions. Takaful transfers start from a basic amount of EGP 325 per household, per month, and increase depending on the number of children in the households and their educational level, with a maximum amount of EGP 625 Transfers were originally delivered on a quarterly basis but were shifted to monthly transfers in 2017. With the outbreak of COVID-19 and the global economic crisis that followed, MOSS has been expanding Takaful to 411,000 more households to shield more households from the loss of employment and the inflationary pressures. 2.2 Targeting When the program was first launched, targeting beneficiaries combined geographical target- ing with a Proxy Means Test (PMT) mechanism. With respect to the geographical targeting, the program was first launched in the poorest districts within the poorest governorates in Egypt. Currently, the program is available in all of Egypt’s governorates. The PMT is used to identify the poor, based on selection criteria and a set cutoff score, based on the poverty line derived from Egypt’s Household Income, Expenditure and Consumption Survey (HIECS). The cutoff score for Takaful has been changed several times throughout the life of the pro- gram but the cutoff faced by the overwhelming majority of applicants is 4,500 points. Any 14 household can apply to the program if they meet the following criteria: 1) the household head is 35 years or older, 2) the household’s monthly income is less than EGP 400 per month from the public or private sector, 3) the household does not benefit from social insurance, 4) the household has children, and 5) the household resides in Egypt (Al-Masry Al-Yom, 2022). The PMT uses poverty predictors from the 2012–2013 Household, Income, Expenditure and Consumption Survey (HIECS). A PMT formula was developed based on the following crite- ria:  Household head characteristics (e.g., gender, marital status, age, education, work status etc.)  Household members characteristics (e.g., number of elderly, children, members en- rolled in education, illiterate, employment situation)  Money transfers received by household (e.g., pensions, remittances, allowances, etc.)  Housing unit characteristics (e.g., type of unit, ownership, ceiling, flooring, water con- nection, etc.)  Ownership of assets (e.g., private car, internet connection, refrigerator, etc.) The PMT formula is based on a regression with the logarithm of per capita annual expendi- ture (adjusted to regional differences, price variations and age differences of HH members) as the dependent variable (El-Sheneity, 2014). Different PMT models are used for the differ- ent geographic regions in Egypt; namely urban Upper Egypt, rural Upper Egypt, urban Lower Egypt, rural Lower Egypt, Metropolitan, and Frontier governorates, however the same PMT eligibility threshold is used for all regions. 2.3 Program experience Satisfaction from Takaful program seems to have declined between 2018 and 2022. Most applicants seem to have a positive experience and are satisfied with the program. Yet, in 2018, 68.1% of surveyed households were very satisfied from the program and 89.2% were either very satisfied or somewhat satisfied (from a nationally representative sample). Mean- while, in 2022, only 48.7 percent of our surveyed households were very satisfied with Taka- ful, and 76.1 percent were either very satisfied or somewhat satisfied, see Figure 2.3.1. 15 Figure 2.3.1. Program Satisfaction Program Satisfaction 11.6 7.5 4.9 48.7 27.4 Very satisfied Somewhat satisfied Neither satisfied nor dissatisfied Somewhat unsatisfied Very unsatisfied Source: Authors’ calculations The average transfer amount reported by beneficiaries in the past 2 months was EGP 432 per month. On average, administrative data shows that households have been receiving payments for 3.8 years. Table 2.3.1. Self-reported Transfer Amounts Mean Standard devi- ation Average amount last 2 months – EGP 431.5 (65.5) No of months since first transfer 42.2 (20.9) No of years since first acceptance 3.5 (1.7) Note: Standard deviations are reported in parentheses The evaluation investigated the challenges that households experienced during the registra- tion process, as well as in receiving the transfers. Table 2.3.3 shows the challenges that reg- istrant households faced while applying for Takaful. The most common challenge among registrants was that queues for program application were long, which was reported by 20 percent of Takaful beneficiaries and 30 percent of non-beneficiaries. The second most com- mon challenge was preparing the needed documents, which required time and effort. Regis- trants who did not end up as Takaful beneficiaries were more likely to report this issue (20 percent), compared to Takaful beneficiaries (14 percent). Meanwhile, 64 percent of Takaful beneficiary household reported no challenges, while 43 percent of non-beneficiaries also re- ported no challenges. 16 Table 2.3.2. Challenges in Applying to Takaful Variable Takaful Benefi- Takaful Non-Benefi- ciaries ciaries The social workers were not helpful in explaining the 0.13 0.25 needed documents (0.33) (0.43) Getting the needed documents require a lot of money 0.10 0.18 (0.30) (0.38) Getting the needed documents require traveling 0.08 0.10 (0.27) (0.30) Getting the needed documents is difficult and time con- 0.14 0.20 suming (0.35) (0.40) There were long queues for program application 0.20 0.30 (0.40) (0.46) The time to travel to apply was prohibitive 0.04 0.05 (0.20) (0.22) The application is time consuming 0.04 0.09 (0.21) (0.29) The application form was too difficult 0.02 0.06 (0.15) (0.23) Other challenges 0.01 0.04 (0.07) (0.19) No challenges met during application to TKP 0.64 0.43 (0.48) (0.49) Number of households 2,543 3,932 Note: Standard deviations are reported in parentheses In terms of receiving the transfers, the only commonly reported issue is that the working hours of payment delivery units were not convenient, which was reported by 15 percent of Takaful households. Table 2.3.3. Challenges in Receiving Transfers Variable Takaful Beneficiaries Requires traveling for long distances 0.02 (0.14) Travel costs are costly 0.05 (0.23) Informal facilitation fees need to be paid to receive payment 0.01 (0.10) Did not know where or how to receive it 0.01 (0.08) Payments are regularly delayed 0.07 (0.25) Do not know when the payment should be received 0.01 (0.11) Lost the card and found difficulty in renewing it 0.01 (0.07) Lost the pin code and found difficulty in renewing it 0.01 (0.07) Working hours at payment delivery units are not convenient 0.15 (0.35) Other challenges 0.01 (0.09) Number of households 2,543 Note: Standard deviations are reported in parentheses 17 3. IMPACT EVALUATION DESIGN 3.1 Motivation for the Regression Discontinuity Approach Impact evaluation is a valuable tool for policy makers because it allows them to understand the causal impact of programs - i.e., the amount of difference in household welfare caused by the program being evaluated rather than any other factor. As recognized in the baseline impact evaluation of Takaful and Karama (Breisinger et al., 2018), the ideal impact evaluation approach to use to measure the impacts of the Takaful program is regression discontinuity. Other common strategies that economists use to iden- tify impact such as a randomized control trial or differences-in-differences were not feasible due to inability to collect pre-program data or randomly assign participants to treatment and control groups. A third common approach, matching, was also discarded because of the sig- nificant role of self-selection in determining which households applied to the program and the large number of observable characteristics already included in the PMT score. By contrast, regression discontinuity works well, as it takes advantage of the program’s targeting ap- proach which uses a strict cutoff in the PMT score. Figure 3.1.1 shows the intuition behind the regression discontinuity evaluation approach: households just above and just below the cutoff are compared, while controlling for the un- derlying relationship between the outcome and the PMT score variable. As highlighted in the baseline evaluation, this approach is recognized as a rigorous impact evaluation strategy in the international literature. One disadvantage, however, is that the impact estimated using the regression discontinuity approach should be interpreted as the average impact of the program specifically for households in the neighborhood of the cutoff. In practice, this means that the estimated impacts of Takaful are for the relatively better off among all Taka- ful beneficiary households. Note that another option considered was defining the “treatment” group as those currently receiving Takaful transfers, and the “control” group as those not currently receiving transfers. However, there are many factors that determine whether a household is receiving transfers. The first is of course whether they meet the eligibility criteria: primarily, a PMT score below 4500 points when the household applied. That criterion is exogenous, meaning that the household cannot do anything to influence it (and we can check whether that is the case in the data) so there are no confounding factors that would influence the results. Accordingly, it is considered exogenous and impacts using this eligibility criterion allows for a causal inter- pretation of results. Since our sample comprises households who applied to Takaful several years ago, they were eligible at the 4500 point cutoff. However, the cutoff has changed since then and household status has also changed for many households. Some households grad- uated from the program and are no longer receiving transfers, some households were newly eligible and added to the program and are now receiving transfers even though they were previously ineligible, and some households who stopped receiving transfers applied again under a new ID and are now receiving transfers. These factors are not exogenous – they are factors that households can affect, and household who can affect these changes are differ- ent from those who cannot, making them an invalid comparison group. Consequently, using a definition of whether the household is a current beneficiary would not result in a causal in- terpretation of estimates. 18 Figure 3.1.1 Intuition Behind Regression Discontinuity Source: Authors’ illustration 3.2 Sample Selection In order to measure the medium-term impact of Takaful, we constrained our sample for this evaluation round only to households that had registered in May 2016-December 2016. Eligi- ble households in this sample – i.e., households with PMT scores below the 4500 cutoff- started receiving Takaful transfers in late 2016 or early 2017 given the several month delay between registration and program entrance. This means that households still receiving transfers as of the second-round evaluation survey had been in the program for approxi- mately five years. Not all households continued in the program, however, and some households that had origi- nally been excluded from the program eventually re-registered and were included. However, we define our “treatment” and “control” groups based on the original eligibility in 2016 in or- der to maintain a causal identification strategy. Whether households met the eligibility crite- ria( a PMT score below 4500 points) when the household applied is exogenous, meaning that the household cannot do anything to influence it (and we can check whether that is the case in the data) so there are no confounding factors that would influence the results. Ac- cordingly, impacts using this eligibility criterion allows for a causal interpretation of results. Some households graduated from the program and are no longer receiving transfers, some households were newly eligible and added to the program and are now receiving transfers even though they were previously ineligible, and some households who stopped receiving transfers applied again under a new ID and are now receiving transfers. These factors are not exogenous – they are factors that households can affect, and household who can affect these changes are different from those who cannot, making them an invalid comparison 19 group. Consequently, using a definition of whether the household is a current beneficiary would not result in a causal interpretation of estimates. 3.3 Heterogeneity analysis In recognition of the limitation of the regression discontinuity approach, an associated com- plementary analysis of administrative data is planned to estimate the degree to which house- holds farther from the cutoff point may have benefited differently from the Takaful program than households in the immediate neighborhood of the cutoff. The heterogeneity analysis portion of the evaluation report is expected to be available by December 2022 depending on the timing of the administrative data collection. 3.4 Regression Discontinuity Validity The remainder of this chapter provides details about the regression discontinuity specifica- tion used for this impact evaluation and shows tests used to ensure that the regression dis- continuity estimate is valid for estimating an impact in this particular sample. Readers less interested in the technical details may skip to Chapter 4. The general justification for use of a regression discontinuity design based on the Proxy Means Test (PMT) was presented in the baseline evaluation report and remains relevant. The use of a Proxy Means Test to create an indicator of well-being for program registrants and use of a threshold PMT score to determine program eligibility provides the conditions needed to measure impact using a regression discontinuity (RD) design. While the eligibility cutoff changed several times in recent years, our sample was selected to include only households that initially registered for Takaful between May 2016 and Decem- ber 2016. This sample was specifically selected such that the 4500 cutoff was relevant for determining household eligibility. The 4500 cutoff was applied from September 2016 to April 2017 and during this period in the Takaful program, registrants who applied in May 2016-Au- gust 2016 and who were not eligible according to the previous cutoff of 4296 were automati- cally included in the program when the cutoff increased. Nevertheless, the 4500 cutoff is not perfectly predictive of program participation. In contrast to the baseline evaluation when approximately 90% of registrants below the threshold were current beneficiaries, in round 2, only approximately 50% of registrants who were below the threshold are still listed in the MoSS database as beneficiaries. This is because MoSS used data from other departments to ensure that all eligibility criteria were met (for example, not receiving remittances, not owning a car, etc.). Based on the original household registration, the strictness of the threshold was very strong with less than 2% of applicants with PMT scores above the 4500 cutoff being beneficiaries, and more importantly, we learned that many households that were initially rejected re-registered later for the program under a dif- ferent application number. Nevertheless, the household PMT score at the time of registra- tion relative to the 4500-cutoff remained a substantial determinant of program participation. 20 Table 3.4.1 demonstrates the probability of ever having been a beneficiary, being a current beneficiary (defined as receiving a transfer in the past two months), and the average amount received if a household’s PMT score is less than or equal to 4500. When the PMT score is less than or equal to 4500, the household is almost twice as likely to have ever received transfers from the Takaful program, was 26% more likely to report a transfer in the past two months (indicating that some households who were beneficiaries are no longer beneficiaries) and received 109 EGP more than non-beneficiary households. Additionally, beneficiary households received transfers for almost two years longer than non-beneficiary households (non-beneficiary households are coded as zero months). Another potential definition of hav- ing been “treated”, is to consider households who are currently beneficiaries but and have also received at least 24 transfers.1 While we know the number of transfers received, we do not know whether a household began receiving transfers, stopped receiving them, and then started receiving them again, or whether they have not recently received transfers. We do know, however, that 17% of households reported ever having ever received a transfer but are no longer receiving them. No households who received ever having received a transfer reported that they do not currently receive transfers. This alternative definition captures the combination of more recent transfers and what could be considered a “sufficient” history sub- stantial period of receiving transfers equal to at least 40% of the maximum possible period of transfer receipt. Using this definition, households whose PMT score fell below the 4500 point threshold were 32% more likely to be treated compared to those below the threshold. Over- all, the PMT threshold is a very strong predictor for all three outcomes, meaning that RD is a valid strategy. As a result, as in the baseline evaluation, we employ a ‘fuzzy’ regression dis- continuity design. We estimate local linear regressions on either side of the cutoff with robust bias-corrected confidence intervals (Calonico et al, 2014). 1 25% of households have a PMT score below 4500, are still receiving transfers, and have had 36 or more months of transfers. 26% of households have a PMT score below 4500, are still receiving transfers, and have had 24 or more months of transfers. This indicates that the choice of 24 months versus a longer period would not change the treated sample very much. 21 Table 3.4.1. Beneficiary Status as Determined by the PMT Threshold (1) (2) (3) (4) HH ever re- HH reported a Taka- Average amount of Number of monthly ceived trans- ful transfer in the Takaful transfers re- Takaful transfers fers from Taka- past two months ceived in the last two ever received ful months PMT score 0.481*** 0.264*** 108.5*** 23.17*** ≤ 4500 (0.0128) (0.0133) (5.973) (0.699) Mean Dep. 0.568 0.393 169.2 23.97 Var. R2 0.236 0.470 0.065 0.196 F-statistic 293.56*** 82.28*** 67.82*** 226.93*** N 6449 6475 6475 6449 The RD estimation strategy, as described above, is a local linear regression model that iden- tifies impacts around the threshold of participation in the program. The estimating equation is as follows. The treatment effect is estimated as the difference in the expected value of the outcome conditional on the PMT score on either side of the threshold relative to the change in proba- bility of participation (P) at the cutoff: lim 𝐸[𝑌|𝑠𝑐𝑜𝑟𝑒]− lim 𝐸[𝑌|𝑠𝑐𝑜𝑟𝑒] 𝜏 = 𝑠𝑐𝑜𝑟𝑒↑4500 𝑠𝑐𝑜𝑟𝑒↓4500 (1) lim 𝐸[𝑃|𝑠𝑐𝑜𝑟𝑒]− lim 𝐸[𝑃|𝑠𝑐𝑜𝑟𝑒] 𝑠𝑐𝑜𝑟𝑒↑4500 𝑠𝑐𝑜𝑟𝑒↓4500 where 𝜏 represents the impact of the Takaful program on a particular outcome. To estimate E[Y|score] and E[p|score] we use local unweighted linear regressions including strata fixed effects to account for the minor differences in the formula used for calculating the PMT score variable in different regions and cluster standard errors at the community level. We estimate two separate specifications: one based on current beneficiary status, and one based on having ever received Takaful transfers. We define a household as a current benefi- ciary if they reported receiving a Takaful transfer in the past two months, and as ever having been a beneficiary if they reported ever receiving Takaful transfers. As a robustness check, we also use the number of months that a household reports receiving transfers as the “treat- ment”. We do this for a select number of our main outcomes only, since the estimates for program participation on average are easier to interpret and participation in the program is our main object of interest rather than intensity of participation. Figure 3.4.1 shows the probability of being a Takaful participant in the past two months, among those who registered, by PMT score. The vertical line denotes the 4500 threshold. While approximately 45% of households above the threshold are receiving transfers, nota- bly, almost 25% of households above the threshold are also receiving transfers. This finding is important in interpreting our results because the size of the difference in probability of re- ceiving transfers is relatively small at around 20%. In the first-round evaluation of Takaful, the equivalent difference was about 55% and that larger difference enabled us to estimate effects more precisely. The small size of this discontinuity compared to the baseline evalua- tion means that our statistical power is still lower than ideal, despite the sampling approach 22 in the second-round evaluation being designed to maximize statistical power. The implica- tion of this reduced statistical power for our impact evaluation is that while larger impacts are still discernable, the confidence intervals on our estimates are large enough that there may be small but meaningful impacts that we are not able to statistically distinguish them from zero. Figure 3.4.2 shows the discontinuity in whether a household has ever been a Takaful recipient. While 80 percent of households below the threshold report that they have received Takaful transfers, approximately 25 percent above the threshold also report ever having re- ceived transfers. Finally, Figure 3.4.3 shows that there is also a discontinuity in the duration of transfers received by households. The outcome variable is the number of months that a household reported receiving Takaful transfers. Below the 4500 point threshold, the average number of months is approximately 35 months, while above the threshold the average is about 10 months. This is a fairly large gap in the duration of transfers. Figure 3.4.1: Probability of having received a Takaful transfer in the past two months, by PMT score (Self-Reported) Notes: Estimated using rdplot with uniform kernel and 30 equally spaced bins. 23 Figure 3.4.2: Probability of Ever Having Received a Takaful Transfer, by PMT Score (Self-Reported) Notes: Estimated using rdplot with uniform kernel and 30 equally spaced bins. Figure 3.4.3: Number of Months Transfers Were Received (Self-reported) 24 3.5 Regression Discontinuity Specification To estimate the treatment effect described in equation (1), we use the ‘rdrobust’ package in Stata, developed by Calonico, Cattaneo, and Titiunik (2014). There were two main decisions to be made when estimating this equation: the choice of bandwidth and the kernel. As will be described in Chapter 4, we chose to sample from the beginning from a very narrow window around the 4500 cutoff to maximize statistical power. Consistent with the baseline evaluation in which we found that when the initial sample is al- ready concentrated around the cutoff, data-driven bandwidth selection is not feasible, we chose to set the bandwidth such that all data collected was used. Equally consistent with our choices in the baseline evaluation, we used a uniform kernel. Following the suggestion of Calonico et al. (2014), we report conventional, bias-corrected, and robust bias-corrected estimates. The conventional results approximate a linear regres- sion both above and below the cutoff. The concern, however, is that if the true underlying relationship is non-linear, the conventional estimates may be affected by leading bias caused by the non-modeled curvature of the relationship. The bias-corrected point estimate allows for quadratic regressions on each side of the cutoff and the robust estimate shows the bias-corrected point estimate together with recentered confidence intervals. The confidence interval in the robust bias-corrected estimates is more robust to the choice of bandwidth than the conventional approach. To the extent that we are confident in the assumption that the underlying relationship between the PMT score, and the outcome variable is linear, however, the conventional estimates remain valid. 4. SAMPLE AND SURVEY DATA 4.1 Sample The sampling strategy was designed to provide a sample of Takaful households that had had the opportunity to receive transfers for approximately five and half years with PMT scores as close as possible to the 4500 threshold to maximize the statistical power. To make the data collection feasible, we targeted 12 households per village. We began with administrative data on all registrants in the program. After extracting all households that registered in the period of May 2016-December 2016 and removing any overlaps with ongoing Forsa or Haya Karima interventions, we selected villages in which it was possible to find 16 Takaful registrant households within the smallest possible distance of the 4500 cutoff. To identify 540 villages for a sample size comparable to the baseline evalua- tion, we ended up using a window of PMT scores from 4448 to 4562 (for geographic repre- sentativeness, we also allowed the inclusion of two villages in the frontier region for which 14 Takaful registrant households with PMT scores from 4444 to 4564 could be found). The ini- tial sample included an extra four households per village as replacement households, to ac- count for anticipated difficulties with locating households based on registration data. Compared to the baseline evaluation, the sample for the second-round evaluation has a dra- matically narrower bandwidth: 62 compared to 600. This narrower bandwidth is expected to increase the precision of our estimates, partially making up for the large loss in statistical power resulting from the much smaller discontinuity in program participation at the cutoff. 25 4.2 Data Collection Data was collected by the survey firm El-Zanaty and Associates between January 8 and February 13, 2022. The field staff consisted of 8 teams of 1 male interviewer and 4 female interviewers each. Households were interviewed by female interviewers, with male interview- ers collecting the community questionnaire. The main enumerator training took place in July 2021, but as the data collection was consid- erably delayed due to security clearance issues, a refresher training was held for four days in January 2022 prior to the start of the data collection. Our initial sample for the Takaful and Karama analysis components consisted of 6,480 households. Due to the anticipated difficulties in locating all households listed in the regis- tration data, the replacement households in each community were used. Within the main sample, 77% of households were located and surveyed. Considering only households which could be located and including the replacement households used in these cases, the overall response rate was 94.6%. The final sample size was 6,475 households. Table 4.2.1 shows the distribution of our sample by governorate. Table 4.2.1 Distribution of Sample by Governorate Governorate Number Percent Cairo 72 1.11 Alexandria 72 1.11 Suez 12 0.19 Kalyubia 96 1.48 Kafr El-Sheikh 132 2.04 Gharbia 24 0.37 Menoufia 36 0.56 Behera 552 8.53 Ismailia 96 1.48 Giza 276 4.26 Beni Suef 1,105 17.07 Fayoum 828 12.79 Menya 1,311 20.25 Assuit 336 5.19 Souhag 780 12.05 Qena 420 6.49 Aswan 192 2.97 Luxor 120 1.85 New Valley 15 0.23 Total 6,475 100 26 4.3 Survey Data was captured in CAPI (computer assisted personal interview). The household survey instrument consisted of the following modules: a) Household Roster: ages, educational attainment, and disability status b) Children’s Schooling: enrollment, grade level, school type, and tuition payments c) Employment: time spent in small business or agriculture, unemployment, time spent in wage employment, sector, and average monthly wage d) Housing Conditions: number of rooms, building materials, water and sanitation facili- ties e) Household Assets, Debt, and Income from Sources other than Wages f) Household Program Participation: including both Takaful and Karama participation and transfers from other government programs g) Shocks: description of type and severity of shocks in past 3 years and coping strat- egy employed h) Food Consumption: 7 day recall period i) Nonfood Consumption: 30 day recall period j) Dietary Diversity: for mother/ caretaker, one child age 6-23 months, and one child age 24-59 months k) Women’s Use of Antenatal and Postnatal Care: only for women with children under 5 years • Infant and Young Child Feeding Practices: only for women with child under 2 years l) Mental Health and Preferences: depression, anxiety, risk, ambiguity, and time prefer- ences m) Intrahousehold Decision-Making: Who makes decisions on various aspects of house- hold affairs n) Anthropometry: for woman or caretaker of children under 18, one child 6-23 months, and one child 24-59 months o) COVID-19 p) Cognition: Digit span forwards and backwards q) Mental Health of Mother: depression Compared to the baseline evaluation, four modules were dropped (agriculture, health, infant and young child feeding knowledge, and Raven’s test) and four modules were added (P, S, T and V), while there was also substantial revision of modules B and M to include additional outcomes and reflect lessons learned from the baseline evaluation. Households were read an informed consent statement in which the purpose of the data col- lection was explained, respondents were told that they did not have to participate in the inter- view and could stop at any time, and we emphasized that individual-level data would not be shared with MoSS or impact program participation and that it would be stored on an en- crypted, password protected file. All COVID-19 protocols set out by the Government of Egypt were strictly adhered to. The community questionnaire included responses informed leader of the community on shocks faced recently by the community as a whole as well as services offered at the near- est health center. 27 Internal Review Board (IRB) approval was sought and received through the International Food Policy Research Institute’s IRB. Questionnaires and a protocol for data collection and protection were submitted along with certificates in data collection for human subjects for all PIs and researchers involved. 5. SUMMARY STATISTICS FOR THE IMPACT ANALYSIS SAMPLE 5.1 Household Characteristics This section will provide a picture of the characteristics of households within our sample. Ta- ble 5.1.1 displays means and standard deviations (in parentheses) of several demographic characteristics of Takaful beneficiaries (column 1) and non-beneficiaries (column 2). The total number of household members in both samples is just under five. Of these, about 2.5 are children between the ages of 0-18, with beneficiary households having slightly more children. This result is expected, of course, since Takaful targets households with children. The PMT score is also as expected – below 4,500 for beneficiaries and above 4,500 for non- beneficiaries. Household heads are also younger in the beneficiary sample at 40 years old versus 42 years old for non-beneficiaries. Heads of households are almost exclusively male and the propor- tions are the same in the two samples. The main household demographics are extremely similar between the two groups: levels of education of the household head and their spouse hardly differ and there is no consistent pattern to show that some households may be much worse off in these pre-determined characteristics. This result is important and encouraging because it means that the non-beneficiary households serve as a valid control group for the beneficiary households. Characteristics that are pre-determined and would not be affected by the program should indeed not be substantially different between the two groups. In examining assets as a measure of wealth, however, it is clear that there are differences between beneficiaries and non-beneficiaries. We examine indices of livestock assets, pro- ductive assets, household durables, and total assets, and in all cases, Takaful beneficiaries have lower index scores. This result is also expected since the beneficiaries are indeed poorer – they fall below the 4,500 threshold whereby they are eligible to receive the pro- gram. 28 Table 5.1.1 Summary Statistics of Household Demographic Characteristics Variable (1) (2) Takaful Beneficiar- Takaful Non-Beneficiar- ies ies Number of members in the household 4.97 4.82 (1.12) (1.26) Total children 0-18 in the household 2.77 2.53 (1.19) (1.38) PMT Score 4,492.91 4,502.27 (22.03) (21.90) Age of household head 40.19 42.29 (8.22) (9.17) Household head is male 0.96 0.96 (0.20) (0.21) Household head did not attain any education 0.34 0.33 (0.47) (0.47) Household head attained primary education level 0.14 0.11 (0.34) (0.31) Household head attained preparatory education 0.10 0.08 level (0.30) (0.27) Household head attained secondary education 0.39 0.44 level (0.49) (0.50) Spouse did not attain any education 0.39 0.37 (0.49) (0.48) Spouse attained primary education level 0.10 0.09 (0.30) (0.28) Spouse attained preparatory education level 0.14 0.13 (0.35) (0.33) Spouse attained secondary education level 0.32 0.34 (0.46) (0.47) Index of Livestock Assets -0.06 0.04 (0.94) (1.57) Index of Productive Assets -0.01 0.01 (1.27) (1.41) Index of Durable Assets -0.09 0.06 (1.39) (1.39) Index of all household assets -0.05 0.03 (1.48) (1.66) Number of households 2,543 3,932 Note: Standard deviations are reported in parentheses. Asset index is based on principal component analysis of reported household assets, segregated into livestock assets, productive assets, household durable goods, and livestock. 6. IMPACT OF THE TAKAFUL PROGRAM 6.1 Variables and Outcomes In this chapter, we report the impact estimates of the Takaful program. We report on our main outcomes of household expenditure, poverty, household composition, labor supply, as- sets, savings, and debt, child schooling, mother’s and children’s anthropometry, dietary di- versity, household, mother, and child dietary diversity, antenatal care and infant and young child feeding (IYCF) practices; maternal overweight and obesity, women’s control over deci- sion-making and shocks and coping strategies, including COVID-19. 29 In each table, we report three estimates. The first row reports the standard coefficients esti- mated without any corrections, i.e., the base, standard specification. The second row reports estimates that have been bias-corrected. Finally, the third row reports estimates that are bias-corrected and robust. Estimates marked with stars are statistically significant, meaning that we are confident that there is a non-zero impact for these results. For other results, the estimate is not measured precisely enough to determine whether there is an impact. As stated in Chapter 3, we define two separate treatment variables. The first is an indicator variable equal to one if the household has received a transfer from Takaful in the past two months. This variable is interpreted as indicating the household is a current beneficiary. Due to the way that the sample was drawn (i.e., households who applied in 2017), these are households who were recipients in 2017 and reported being current recipients. The second variable is an indicator variable equal to one if the household has ever been a Takaful bene- ficiary, which can be thought of as more of a persistence interpretation of the program. Both variables are self-reported by the household. The correct variable to use differs depending on the outcome. For outcomes that would only be affected by the current state of household liquidity such as consumption, schooling, or diets, the correct treatment variable is whether the household is currently a Takaful beneficiary. For outcomes that take time to change, such as assets holdings and anthropometry, the correct treatment variable is whether the household has ever been a Takaful participant. Consequently, for most outcomes we report only the treatment variable of currently being a Takaful participant. For other outcomes, we report both and compare the consistency of the two. Table 6.1.1 shows the concordance between the two variables. We see that, as expected, all households who have received transfers in the past two months are also recorded as ever having received a transfer. For those who are not currently receiving transfers, 28.6% of households report having received a Takaful transfer in the past. This means that these households stopped receiving the program at some point. Table 6.1.1. Current Beneficiary Versus Ever Beneficiary Status Has never received Has received Takaful trans- Total Takaful transfers fers at some point Has not received Takaful trans- 2,788.0 1,118.0 3,906.0 fers in the past 2 months (71.4%) (28.6%) (100.0%) Has received Takaful transfers 0.0 2,543.0 2,543.0 in the past 2 months (0.0%) (100.0%) (100.0%) Total 2,788.0 3,661.0 6,449.0 (43.2%) (56.8%) (100.0%) Because we are using a regression discontinuity impact evaluation strategy, all results should be interpreted as the program impact on a household near the cutoff point. We first note that there are outcomes directly related to spending of the transfers received, and other outcomes that are related to the way in which the transfers and other income are spent. The first type of outcomes are decisions that households make in how to spend trans- fers. Households can either increase (or decrease) consumption, save the transfers, pay down debts, or invest in assets (durables or productive assets). We examine each below. 30 The second type of outcomes are the result of the decisions made in spending. For example, if a household spends more on food, dietary diversity could increase, or antenatal and post- natal outcomes could improve. If a household decreases debt, perhaps households can af- ford to keep children in school for longer. We will examine these types of outcomes in the second part of Chapter 6. 6.2 Household Total Expenditure and Poverty We do not detect statistically insignificant impacts of Takaful and Karama on household ex- penditure or on moving households over the poverty line. Table 6.2.1 reports total monthly consumption, monthly food consumption, and monthly non-food consumption in Adult Equiv- alent Units (AEU) using the standard definition in the literature.2 While two coefficients are negative and statistically significant at the 10% level, we do not interpret this as an overall negative effect. For the same outcome, the other estimates are not statistically significant, and the overall effect size is quite small. Additionally, the standard errors are large, indicat- ing that we do not have sufficient power to detect precise effect sizes. Comparing these estimates to the first round of the Takaful evaluation, these are in the op- posite direction. In the first round, consumption increased by 7% overall, and this was pri- marily driven by an increase in food consumption (an increase of 8%). Recall that this sam- ple of respondents comprises households who may have graduated from the program but reapplied and are now receiving transfers even though their PMT score is recorded in our application data as being below the threshold. Also recall that the difference in the probability of receiving Takaful at the threshold based on the PMT score is only 20%, reducing our abil- ity to detect smaller effect sizes. It is also important to note that estimated impacts on total consumption value (not in AEU) are also consistent with Table 6.2.1; we do not detect any effects on consumption. Importantly, estimating no effect on consumption is not an outlier in the literature. There are not many studies estimating long-term effects on consumption, but the studies that do show mixed results with a couple of studies showing continuing improvements, and others show- ing no effects. In a review of cash transfer studies conducted by Bastagli et al (2016), only 2 out of 7 evaluations that studied long term consumption effects showed positive results. In the Egyptian context, inflation may have played a role. More recently, Haushofer and Shapiro (2018) show that while a cash transfer program in Kenya had positive short-term ef- fects on consumption, the effects did not persist in the longer term. Instead, households in- vested in assets. Investment in assets is common in the literature on cash transfers and will be discussed further below. We note that current consumption is indicative of where house- holds might fall in the distribution of basic versus longer-term needs. Once households be- come less poor, they do not have to put most of their transfer income into consumption but rather, can put more money towards productive investments, which would increase their fu- ture income. This is a natural progression that is a very positive potential impact of cash transfer programs. 2 Adult equivalent units in our analysis are defined as giving a weight of 1 to the first adult, 0.7 to additional adults and, 0.3 to children under age 18 (Hagenaars, de Vos, and Zaidi, 1994). 31 Table 6.2.1 Impacts of Takaful Program on Household Consumption Expenditure Treatment variable: household received Takaful transfers in the past two months (1) (2) (3) Monthly food consump- Monthly non-food con- Monthly total (food & non- tion expenditure sumption expenditure food) consumption ex- penditure Conventional -0.0176 -0.0555 -0.0338 (0.0717) (0.0744) (0.0608) Bias-corrected -0.105 -0.143* -0.106* (0.0717) (0.0744) (0.0608) Robust -0.105 -0.143 -0.106 (0.0924) (0.0974) (0.0785) Mean Dep. 6.791 6.526 7.396 Var. N 6475 6475 6475 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. Consumption aggregates shown are winsorized at the 2nd and 99th percentiles, calculated as Adult Equivalent Units (AEU), and transformed using Inverse Hyperbolic Sine (IHS). * p < 0.10, ** p < 0.05, *** p < 0.01. Next, we turn to poverty. Table 6.2.3 shows impact estimates on poverty outcomes using whether a household received a Takaful transfer in the past two months as our treatment in- dicator. We examine whether the household is living under USD 1.90 per day and under USD 3.20 per day (the World Bank definitions of poverty and extreme poverty, respectively). We also examine whether the household is under the Egyptian 2017/2018 poverty line (de- fined at the regional level, whereby each region has a different poverty line that is relevant to that context). In interpreting our results on poverty, it is necessary to keep in mind first that our results are sensitive to the exact poverty line chosen and secondly to realize that our sample comprises a population that is already near the regional poverty line (in fact, that is how the threshold was selected). All the coefficients are positive, but only one is marginally statistically signifi- cant. We do not interpret this as strong evidence that there was an increase in poverty, how- ever. These findings are consistent with the findings on consumption. 32 Table 6.2.2. Impacts of Takaful Program on Household Poverty Measures Treatment variable: received a transfer in the past two months (1) (2) (3) Household living Household living un- HH living under under US$1.90 per der US$3.20 per per- 2017/2018 regional person per day son per day poverty line Conventional 0.0156 0.0350 0.0866 (0.0340) (0.0734) (0.0610) Bias-corrected 0.0422 0.0833 0.106* (0.0340) (0.0734) (0.0610) Robust 0.0422 0.0833 0.106 (0.0468) (0.0995) (0.0801) Mean Dep. Var. 0.0604 0.413 0.811 N 6475 6475 6475 Standard errors in parentheses. Linear Trend on PMT Score, Uniform Kernel, Bandwidth=63, Strata dummies included (Fron- tier and Upper Egypt (rural) combined to single stratum); excluded category: Metropolitan. (1) and (2) show the World Bank's International Poverty Line at US$1.90 (2011 PPP) and the Lower Middle Income Class Poverty Line at $3.20 (2011 PPP), re- spectively. The national-level core CPI was used to express 2011 PPP figures in 2022 EGP values. Similarly, the regional pov- erty line in (3) uses the latest regional poverty lines available (2017/2018) and was compared to our consumption survey data collected in 2022 which was deflated to 2017/2018 values. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.3 Components of Household Consumption We next examine the impact of the Takaful program on the value of (log) expenditure of spe- cific food groups. There are 12 food groups (grains, potatoes, vegetables, fruits, meat, eggs, fish, legumes, dairy, oils and fats, sweets, and other foods) and we also examine impacts on expenditures incurred on food consumed outside of the household. The treatment indicator is whether the household is a current Takaful beneficiary. Table 6.3.1 reports the results of log expenditures per AEU. We find that there are statisti- cally significant negative effects on four of the food groups: grains, fruits, eggs, and dairy. Unfortunately, these food groups have high nutritional value, though we note that expendi- tures on meat, fish, vegetables, and legumes are not affected by the program, and these items are also high in nutritional content. For fish, the evidence is weakly suggestive of a positive impact. Encouragingly, there are no effects of the program on the consumption of sweets and salty snacks or drinks and beverages, which are low in nutritional content. There is also no impact on expenditures on food consumed outside the household. There is a neg- ative impact on the amount spent on oils and fats, which could be interpreted as a positive result, depending on what types of oils and fats are being consumed (processed versus un- processed). The magnitude of these differences is large. There is a more than 20% decrease in the con- sumption of grains, 77% for fruit, 66% for eggs, and 57% for dairy. Since many of the coeffi- cients are negative (even though not statistically significant), these patterns are consistent with the effects on food consumption expenditure, with a negative but statistically insignifi- cant coefficient. We also examine the impact of the Takaful program on the log expenditure per AEU of spe- cific non-food groups. Table 6.3.2. presents the results, again using current beneficiary sta- tus as the treatment variable. We do not see many changes in patterns of non-food con- sumption as a result of the Takaful program. There is weak and suggestive evidence that ex- penditures were lower on construction and communications/entertainment, but the results 33 are not consistently statistically significant. Schooling expenditures, while not statistically sig- nificant, have a large coefficient. We will return to this result in subsequent sections below but recall the previous result that the number of school-aged children in the household in- creased. Consequently, the amount spent per child may have decreased. The effect on ex- penditure on clothes is also large and statistically insignificant. Note again that these are in AEU so the spending per person would have decreased. Notably, the magnitudes of the co- efficients on healthcare and medicines are large and are also negative. It is possible that household members required less healthcare and medicine as a result of being healthier. We will also return to this result in later sections. Encouragingly, there are no effects on smoking expenditures as a result of the program. 34 Table 6.3.1. Impacts of Takaful Program on Food Consumption Expenditure by Category Treatment variable: household received Takaful transfers in the past two months (1) (2) (3) (4) (5) (6) (7) Monthly Expendi- Monthly Expendi- Monthly Expenditures Monthly Expendi- Monthly Expendi- Monthly Expendi- Monthly Expendi- tures on Grains per tures on Potatoes on Vegetables per tures on Fruits per tures on Meat per tures on Eggs per tures on Fish per AEU- log values per AEU - log values AEU - log values AEU- log values AEU - log values AEU- log values AEU -log values (screened) (screened) (screened) (screened) (screened) (screened) (screened) Conventional -0.111 -0.0823 0.129 -0.467* 0.00396 -0.0234 0.355 (0.101) (0.122) (0.0925) (0.255) (0.308) (0.245) (0.245) Bias-corrected -0.221** -0.191 0.0478 -0.775*** -0.480 -0.660*** 0.474* (0.101) (0.122) (0.0925) (0.255) (0.308) (0.245) (0.245) Robust -0.221* -0.191 0.0478 -0.775** -0.480 -0.660** 0.474 (0.133) (0.160) (0.118) (0.333) (0.405) (0.333) (0.330) Mean Dep. 4.056 2.683 3.941 2.111 3.829 1.762 0.909 Var. N 6475 6475 6475 6475 6475 6475 6475 (8) (9) (10) (11) (12) (13) Monthly Expenditures on Monthly Expenditures Monthly Expenditures Monthly Expenditures on Monthly Expenditures on Monthly Expenditures Legumes per AEU - log on Dairy per AEU - log on Oils and Fats per Sweets/Salty Snacks per Drinks/Beverages per Outside the HH per values (screened) values (screened) AEU- log values AEU- log values AEU- log values AEU- log values (screened) (screened) (screened) (screened) Conventional 0.355 -0.148 -0.408* 0.118 0.0139 0.000164 (0.245) (0.236) (0.225) (0.102) (0.0893) (0.121) Bias-corrected 0.474* -0.0221 -0.570** 0.0587 -0.0347 -0.116 (0.245) (0.236) (0.225) (0.102) (0.0893) (0.121) Robust 0.474 -0.0221 -0.570* 0.0587 -0.0347 -0.116 (0.330) (0.309) (0.299) (0.131) (0.116) (0.154) Mean Dep. 0.909 2.194 2.676 3.855 3.171 3.081 Var. N 6475 6475 6475 6475 6475 6475 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. Consumption aggregates shown are winsorized at the 2nd and 99th percentiles, calculated as Adult Equivalent Units (AEU), and transformed using Inverse Hyperbolic Sine (IHS). * p < 0.10, ** p < 0.05, *** p < 0.01. 35 Table 6.3.2. Impacts of Takaful Program on Non-food Consumption Expenditure by Category Treatment variable: household received Takaful transfers in the past two months (1) (2) (3) (4) (5) (6) Monthly Expenditures Monthly Expendi- Monthly Expendi- Monthly Expendi- Monthly Expendi- Monthly Expendi- on school (total) per tures on Transpor- tures on Rent and tures on Communi- tures on Personal tures on Smoking AEU - log values tation per AEU - log Utilities per AEU- cations and Enter- Care and Hygiene per AEU - log val- (screened) values (screened) log values tainment per AEU - Items per AEU - log ues (screened) (screened) log values (s values (sc Conventional 0.379 0.115 -0.0183 -0.0460 0.0247 0.0284 (0.305) (0.139) (0.0755) (0.184) (0.109) (0.336) Bias-corrected 0.200 0.262* -0.0303 -0.332* -0.116 -0.284 (0.305) (0.139) (0.0755) (0.184) (0.109) (0.336) Robust 0.200 0.262 -0.0303 -0.332 -0.116 -0.284 (0.400) (0.191) (0.0994) (0.239) (0.141) (0.445) Mean Dep. Var. 4.152 0.346 4.667 2.305 3.779 1.645 N 6475 6475 6475 6475 6475 6475 (7) (8) (9) (10) (11) (12) Monthly Expenditures Monthly Expendi- Monthly Expendi- Monthly Expendi- Monthly Expendi- Monthly Expendi- on Clothes per AEU - tures on Construc- tures on HH Dura- tures on Occasions tures on Healthcare tures on Medicine log values (screened) tion per AEU - log bles per AEU - log per AEU - log val- per AEU - log val- per AEU - log val- values (screened) values (screened) ues (screened) ues (screened) ues (screened) Conventional 0.263 -0.0739* 0.0257 -0.00615 -0.0362 -0.146 (0.224) (0.0439) (0.0334) (0.0475) (0.222) (0.160) Bias-corrected -0.287 -0.0747* 0.0269 -0.0451 -0.150 -0.219 (0.224) (0.0439) (0.0334) (0.0475) (0.222) (0.160) Robust -0.287 -0.0747 0.0269 -0.0451 -0.150 -0.219 (0.307) (0.0564) (0.0433) (0.0643) (0.292) (0.210) Mean Dep. Var. 2.553 0.0554 0.0350 0.0648 1.949 3.231 N 6475 6475 6475 6475 6475 6040 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. Consumption aggregates shown are winsorized at the 2nd and 99th percentiles, calculated as Adult Equivalent Units (AEU), and transformed using Inverse Hyperbolic Sine (IHS). * p < 0.10, ** p < 0.05, *** p < 0.01. 36 6.3.1 Household Composition To dig deeper into the consumption effects we estimate, we examine household composition since the consumption results are in AEU, meaning that the denominator depends on house- hold size. Consequently, we investigate whether changes in household composition could also be affecting our estimates. We estimate the impacts of Takaful on the total number of household members, and split the number of household members into age groups as well (children 0-5, children 6-11, and adolescents 13-18 years old). The results are presented in Table 6.2.2. In Panel A our treatment variable is an indicator equal to one if the household received transfers from Takaful in the past two months, and in Panel B our outcome variable is an indicator equal to one if the household has ever received a Takaful transfer. The latter can be interpreted as more of a cumulative effect; the effect even if a household is not cur- rently receiving transfers. Results are consistent across panels. In Panel A Column 1, we find that the number of household members has increased by ap- proximately 0.3 members in Takaful households (0.2 in Panel B). The increase in household members is primarily driven by an increase in household members aged 6-11 (Column 3). These results raise the question of whether Takaful households are more likely to have a child (extensive margin) or are having more children as a result of the program (intensive margin). To dig into this question further, we restrict the sample to households who already had two or more children at the time of registration and examine whether, among that sam- ple, there are increases in the number of children aged 0-3 (Column 5) or 0-5 years old (Col- umn 6). Takaful beneficiaries who already had two or more children at registration are in- deed more likely to have more children. However, this result should not be too strongly inter- preted because households just under the threshold at registration we more likely to have 2 or more children (the discontinuity is significant at the 10 percent level). The interpretation of these results is not straightforward. Increases in fertility could possibly be driven by different factors: the desire to have more children coupled with the newfound ability to afford them, or a mistaken view that further transfers will be received if there are more children in the household (i.e., households do not know that Takaful pays transfers for a maximum of two children). We do not have data on the latter, but we do see that the aver- age desired number of children is 3.52 and the average number of children delivered is 3.68, which is neither a large difference nor indicative of the transfers enabling households to fulfill their desire for more children. 37 Table 6.3.3. Household Composition Panel A: Received Takaful transfers in the past two months Number of household members Number of household members (HHs who had 2 or more children at regis- tration) (1) (2) (3) (4) (5) (6) Number of total 0-5 years old 6-11 years old 12-18 years old 0-3 years old 0-5 years old household members Conventional 0.314* 0.178 0.221* 0.0385 0.165* 0.194 (0.187) (0.134) (0.131) (0.102) (0.0996) (0.134) Bias-corrected 0.450** 0.252* 0.287** 0.0610 0.244** 0.378*** (0.187) (0.134) (0.131) (0.102) (0.0996) (0.134) Robust 0.450* 0.252 0.287 0.0610 0.244* 0.378** (0.251) (0.180) (0.175) (0.141) (0.138) (0.181) Mean Dep. Var. 4.876 0.811 1.156 0.542 0.844 0.914 N 6475 6475 6475 6475 3895 3895 Panel B: Ever received Takaful transfers Conventional 0.182* 0.0996 0.122* 0.0223 0.203** 0.239** (0.0988) (0.0710) (0.0691) (0.0540) (0.0909) (0.112) Bias-corrected 0.259*** 0.138* 0.156** 0.0362 0.229** 0.288** (0.0988) (0.0710) (0.0691) (0.0540) (0.0909) (0.112) Robust 0.259** 0.138 0.156* 0.0362 0.229* 0.288* (0.132) (0.0957) (0.0929) (0.0746) (0.124) (0.153) Mean Dep. Var. 4.876 0.811 1.156 0.542 0.844 0.914 N 6449 6449 6449 6449 2770 2275 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 38 6.4 Assets, Savings, and Debt In this section, we examine three other choices households can make with regards to spend- ing transfer and other income. In addition to consumption, households can either invest in assets, increase their savings, or reduce their debts. Assets, savings, and debt are cumula- tive and reflect a history of past investments. Consequently, we use the treatment status in- dicator equal to one if the household reports that they have ever received a Takaful transfer in addition to the indicator for being a current beneficiary. We construct four indices using principal components analysis: 1) an index of total house- hold assets including durable assets (such as furniture, television), productive assets (those that can enable increased income generation), and livestock assets (such as cows, chick- ens) comprising 37 items; 2) household durable assets comprising 20 items, 3) an index of productive assets comprising 8 items, and 4) an index of livestock assets comprising 9 types of livestock. Table 6.4.1 displays the impact estimates on these outcomes, using the treat- ment indicator of receiving a Takaful transfer in the past two months in Panel A, and using whether the household ever received a Takaful transfer in Panel B. In Panel A we see that there is a statistically significant impact on total assets and this effect appears to be driven by investments in productive assets (using the conventional estimates). While the bias-cor- rected and robust specifications show a negative value for total assets and for durable as- sets, the magnitudes are extremely small and not economically meaningful so we do not in- terpret this as a negative effect. For productive assets, the conventional estimate is positive and statistically significant and the bias-corrected and robust estimators are not significant but they are also positive and relatively large. In Panel B we see that the estimates are con- sistent, but the coefficients are smaller. We conclude that households used Takaful transfers to invest in productive assets, and that there is suggestive evidence that more recent partici- pation in the program matters more for asset ownership. 39 Table 6.4.1. Impacts of Takaful Program on Asset Indices Panel A: Household received Takaful transfer in past two months Asset index for (1) (2) (3) (4) Total Durable Productive as- Livestock as- assets assets sets sets Conventional 0.517** -0.0514 0.467** 0.349 (0.218) (0.226) (0.189) (0.230) Bias-corrected -0.00418 -0.0254 0.167 0.0643 (0.218) (0.226) (0.189) (0.230) Robust -0.00418 -0.0254 0.167 0.0643 (0.310) (0.300) (0.261) (0.356) N 6474 6474 6474 6474 Panel B: Household ever received a Takaful transfer Conventional 0.278** -0.0111 0.249** 0.191 (0.115) (0.120) (0.0991) (0.122) Bias-corrected -0.00703 -0.00594 0.0838 0.0346 (0.115) (0.120) (0.0991) (0.122) Robust -0.00703 -0.00594 0.0838 0.0346 (0.163) (0.160) (0.137) (0.189) N 6449 6449 6449 6449 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. All indices are constructed based on the first principal component from principal component analysis (PCA). The index that includes all assets uses dummies for assets ownership. The same for the durables and productive assets indices. The livestock index is composed using a count of the livestock owned by the household. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The fol- lowing strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. Note that the mean dependent vari- able is not reported as the outcomes are indices constructed to have mean zero. We next delve further into the specific types of assets that households were investing in – agricultural assets in Table 6.4.2 and livestock assets in Table 6.4.3. A substantial proportion of households invested in large and lumpy assets such as tractors, machine and animal plows, and drip irrigation. These particular investments enable households to make their ag- ricultural production more efficient and diversify their labor into other income generating ac- tivities. As with the asset indices, it appears that more recent transfers are associated with higher investments in these productive assets. Livestock investments were also lumpy, with larger animals being purchased. Households decreased the number of small animals they owned (chickens, geese, pigeons, and ducks) and increased their stock of buffaloes, cattle, goats, and sheep. The purchase of buffaloes and cattle is consistent with the result that households invested in animal plows. These larger animals can also result in increased efficiency of agricultural production, further free- ing up time for other income-generating activities. We will explore labor outcomes in the next section. 40 Table 6.4.2. Impacts of Takaful Program on Individual Agricultural Assets Panel A: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) (6) (7) (8) Tractor Machine pulled Animal pulled Mechanical wa- Animal or man- Drip irrigation Rice winnower Ox cart or don- plow or har- plow ter pump ual powered network key cart rower water pump Conventional 0.0107** 0.00887* 0.0149* 0.00941 0.000739 0.00609* -0.000397 0.0139 (0.00452) (0.00526) (0.00867) (0.00877) (0.00350) (0.00359) (0.000402) (0.0113) Bias-corrected 0.0114** 0.00969* 0.00728 -0.00398 0.000243 0.00840** -0.00167*** -0.0118 (0.00452) (0.00526) (0.00867) (0.00877) (0.00350) (0.00359) (0.000402) (0.0113) Robust 0.0114* 0.00969 0.00728 -0.00398 0.000243 0.00840 -0.00167 -0.0118 (0.00582) (0.00731) (0.0121) (0.0122) (0.00482) (0.00547) (0.00166) (0.0153) Mean Dep. Var. 0.000927 0.00124 0.00402 0.00386 0.00124 0.000463 0.000154 0.00834 N 6474 6474 6474 6474 6474 6474 6474 6474 Panel B: household ever received a Takaful transfer Conventional 0.00570** 0.00474* 0.00794* 0.00503 0.000402 0.00325* -0.000210 0.00745 (0.00238) (0.00278) (0.00458) (0.00465) (0.00186) (0.00191) (0.000212) (0.00596) Bias-corrected 0.00592** 0.00505* 0.00371 -0.00222 0.000128 0.00439** -0.000885*** -0.00639 (0.00238) (0.00278) (0.00458) (0.00465) (0.00186) (0.00191) (0.000212) (0.00596) Robust 0.00592* 0.00505 0.00371 -0.00222 0.000128 0.00439 -0.000885 -0.00639 (0.00308) (0.00387) (0.00641) (0.00650) (0.00257) (0.00291) (0.000887) (0.00810) Mean Dep. Var. 0.000927 0.00124 0.00402 0.00386 0.00124 0.000463 0.000154 0.00834 N 6449 6449 6449 6449 6449 6449 6449 6449 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Model details: Linear Trend on PMT Score; Uni- form Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 41 Table 6.4.3. Impacts of Takaful Program on Livestock Panel A: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) (6) (7) Buffaloes/ Cat- Goats/Sheep Donkeys/ mules Horses Chicken/geese/ Turkeys Rabbits tle pigeons/ducks Conventional 0.0823* 0.152** 0.0252 0.00136 -0.436 0.0133 0.00741 (0.0441) (0.0740) (0.0204) (0.00136) (0.955) (0.0155) (0.0310) Bias-corrected 0.0163 0.0537 -0.00855 0.000959 -1.956** 0.0233 0.0121 (0.0441) (0.0740) (0.0204) (0.00136) (0.955) (0.0155) (0.0310) Robust 0.0163 0.0537 -0.00855 0.000959 -1.956 0.0233 0.0121 (0.0618) (0.0824) (0.0307) (0.000912) (1.336) (0.0260) (0.0479) Mean Dep. Var. 0.0513 0.0539 0.0188 0.000154 2.599 0.00154 0.00618 N 6474 6474 6474 6474 6474 6474 6474 Panel B: household ever received a Takaful transfer Conventional 0.0442* 0.0795** 0.0135 0.000723 -0.154 0.00707 0.00396 (0.0233) (0.0387) (0.0108) (0.000721) (0.503) (0.00823) (0.0165) Bias-corrected 0.00807 0.0264 -0.00471 0.000494 -0.976* 0.0122 0.00640 (0.0233) (0.0387) (0.0108) (0.000721) (0.503) (0.00823) (0.0165) Robust 0.00807 0.0264 -0.00471 0.000494 -0.976 0.0122 0.00640 (0.0327) (0.0430) (0.0163) (0.000485) (0.706) (0.0138) (0.0255) Mean Dep. Var. 0.0513 0.0539 0.0188 0.000154 2.599 0.00154 0.00618 N 6449 6449 6449 6449 6449 6449 6449 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 42 The literature has shown strong precedent for cash transfer programs increasing invest- ments in assets, and productive assets in particular. Bastagli et al (2016) find that out of 8 impact evaluations of cash transfer programs, 3 had positive and statistically significant ef- fects on agricultural productive asset accumulation. The rest either had positive but statisti- cally insignificant effects, or null effects. Additionally, 6 of the studies also show increases in agricultural input adoption, which are also considered productive investments. Finally, 12 out of 17 studies showed that cash transfers increased livestock holdings. Accordingly, there is substantial evidence that cash transfers are used to build up productive assets as invest- ments in the future. One consideration in this context is whether assets were received by other programs rather than the Takaful program necessarily leading to asset investments. There are many pro- grams that provide productive assets to the poor. For our results to be biased, it would need to be the case that the asset transfers are also targeted to Takaful participants, or at least with similar characteristics to beneficiaries whose PMT score is below 4500. To check whether this might be a concern, we run the same specification as above but include a con- trol variable for whether the household received any other source of income from: family members or friends, religiously motivated support, divorce allowance, cash transfers from other organizations, a private insurance fund, or rental earnings. Unfortunately, we do not have specific information on whether the household received asset transfers, so this variable serves as a proxy. When we include the control variable, the results barely change. Households also reduced the amount of debt they carried. Table 6.4.4 shows that there are no differences in rates of saving among beneficiary households (column 1). In column 2, the outcome is a dummy variable equal to one if the respondent says that the amount they had in savings increased since July 2020. The coefficients are positive but not statistically signifi- cant. We do see that beneficiary households have lower levels of debt and the effect is sta- tistically significant and meaningful in size. There is no evidence that beneficiary households had different levels of debt with regards to instalment payments (column 4). 43 Table 6.4.4. Impacts of Takaful Program on Household Savings and Debt Treatment variable: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) Amount of Amount of sav- Total amount of Amount owed Total savings ings increased debt currently of debt owed amount of (EGP) or remained the owed to anyone - for purchases debt owed same between formal/informal on credit to informal July 2020 and lenders (EGP), (EGP) lenders or Feb 2022 IHS owed for purchases on credit Conventional -209.6 0.340 -1.151 -332.4 -4326.0* (157.6) (0.701) (0.716) (748.4) (2359.7) Bias-cor- -255.0 1.030 -1.835** -218.7 -5479.2** rected (157.6) (0.701) (0.716) (748.4) (2359.7) Robust -255.0 1.030 -1.835* -218.7 -5479.2 (241.0) (0.934) (0.965) (966.5) (3342.1) Mean Dep. 53.53 0.909 3.219 827.7 4430.5 Var. N 6475 88 6475 6475 6475 Treatment variable: household ever received a Takaful transfer (1) (2) (3) (4) (5) Amount of Amount of sav- Total amount of Amount owed Total savings ings increased debt currently of debt owed amount of (EGP) or remained the owed to anyone - for purchases debt owed same between formal/informal on credit to informal July 2020 and lenders (EGP), (EGP) lenders or Feb 2022 IHS owed for purchases on credit Conventional -111.4 0.122 -0.640* -174.1 -2315.4* (83.83) (0.228) (0.379) (398.1) (1254.0) Bias-cor- -132.4 0.345 -0.977** -101.4 -2855.7** rected (83.83) (0.228) (0.379) (398.1) (1254.0) Robust -132.4 0.345 -0.977* -101.4 -2855.7 (127.9) (0.322) (0.511) (513.9) (1770.0) Mean Dep. 53.53 0.909 3.219 827.7 4430.5 Var. N 6449 88 6449 6449 6449 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.5 Household Labor In this latter part of this chapter, we examine outcomes that are the product of the decisions made as to how to spend transfer and other income. We begin with labor supply. Depending on how households are choosing to use their transfers, labor supply patterns may be differ- ent. In Table 6.4.1 we report impact estimates on the likelihood that a household member aged 5 or older engages primarily in unpaid work, participates in agricultural activities for household 44 consumption, works in the formal sector, and works in the informal sector. The share of household members who participate primarily in unpaid work is already very low at 0.4% and we see no impacts on this outcome. Only 4% of household members engage in agricultural activities for household consumption and there are also no impacts of the program on this outcome. While there is a negative and statistically significant coefficient on the likelihood that beneficiaries are engaged in formal labor, the evidence is not strong. There is, however, a higher likelihood that beneficiary households are engaged in informal labor. However, we do see a difference in formal and informal labor participation, with a movement away from the formal sector to the informal sector. Column 5 shows that the average monthly wages of all household members slightly decreased as well. Table 6.5.1. Impacts of Takaful Program on Employment Outcomes Treatment variable: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) Household Household member Household Household Total monthly member par- participates in agri- member par- member partic- wage income ticipates in un- cultural activities for ticipates in ipates in infor- (EGP) all HH paid work own household con- formal work mal work members sumption Conventional 0.00627 0.0126 -0.0396 0.0246 -366.5* (0.00650) (0.0170) (0.0272) (0.0216) (191.2) Bias-corrected 0.00840 -0.0119 -0.0570** 0.0546** -225.4 (0.00650) (0.0170) (0.0272) (0.0216) (191.2) Robust 0.00840 -0.0119 -0.0570 0.0546* -225.4 (0.00846) (0.0221) (0.0384) (0.0297) (255.1) Mean Dep. Var. 0.00410 0.0398 0.0408 0.973 1848.1 N 26096 26096 7213 7213 6475 Standard errors clustered at the village level. Models run at the household member-level. Details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. Samples in (1) and (2) in- clude all household members over age 5. Samples in (3) and (4) include household members who participate in economic ac- tivities. The sample in column (5) includes individuals who report earning a wage, either from formal or informal employment. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.6 Child schooling and Child Labour In this section, we examine the impacts of the Takaful program on children’s schooling and labour. Depending on the amount of liquid cash and the extent to which children are needed to help at the home or family business, school enrolment and attendance could be affected. Table 6.6.1 first presents results on children’s enrollment in school by level (nursery, pri- mary, preparatory, secondary, and university or higher).3 In column 1 the sample is all chil- dren of school going age, and in columns 2-6 the samples are children who are of the age range for that level. There are no impacts of Takaful on enrollment in school of 6-18 year olds. However, 96% of 6-18 year olds are enrolled in some level of schooling so we do not expect to see much of an impact on such a high base. While very few children attend nursery, 87% and 84% of children attend primary and preparatory school, respectively. For both levels, we see positive and significant impacts on enrollment. The probability of a child of primary school age being enrolled in primary school increased by about 6-8 percentage points and the probability of a child of preparatory school age being enrolled in preparatory school increased by 2-3 percentage points. These are economically meaningful effects – 3 Primary school comprises grades 1-6, preparatory comprises grades 7-9, secondary comprises either grades 10-12 (general secondary) or grades 10-14 (technical secondary). 45 they suggest that almost all primary school aged children in the beneficiary group were at- tending primary school and most were attending preparatory school. There are no significant effects on enrollment at the secondary school level, and there are actually decreases in en- rollment at the university level or above. However, this decrease can be explained by the changing demographics of the household. With younger households in the beneficiary sam- ple, there are fewer university-age students residing in these households. With regards to attendance, we see that across all levels of schooling there is no impact of the program. However, there is suggestive evidence that secondary school students attend school more regularly as a result of the program. In particular, it is girls who are driving the result (see column 10). This is an encouraging result as secondary school participation for girls is a substantial driver of income, empowerment, and their children’s outcomes (Glewwe 2002; Hanushek and Zhang 2009). 46 Table 6.6.1. Impacts of Takaful Program on Enrolment and Attendance Treatment indicator: household received Takaful transfers in the past two months Currently enrolled in: (1) (2) (3) (4) (5) (6) Any level Nursery Primary Preparatory Secondary University or higher Conventional 0.0296 -0.0191 0.0870** 0.210*** -0.0156 -0.0647 (0.0218) (0.0306) (0.0439) (0.0801) (0.136) (0.0701) Bias-corrected 0.0179 -0.0374 0.0615 0.315*** 0.0391 -0.148** (0.0218) (0.0306) (0.0439) (0.0801) (0.136) (0.0701) Robust 0.0179 -0.0374 0.0615 0.315*** 0.0391 -0.148 (0.0286) (0.0381) (0.0576) (0.112) (0.186) (0.0964) Mean Dep. Var. 0.969 0.0257 0.873 0.840 0.647 0.0403 N 11314 4700 8354 2699 1936 818 Currently attending secondary: (7) (8) (9) (10) Regularly attending All Boys Girls (all levels) Conventional 0.0337 0.161 0.148 0.101 (0.0506) (0.127) (0.151) (0.177) Bias-corrected 0.0471 0.266** 0.152 0.338* (0.0506) (0.127) (0.151) (0.177) Robust 0.0471 0.266 0.152 0.338 (0.0696) (0.172) (0.207) (0.231) Mean Dep. Var. 0.882 0.838 0.802 0.886 N 10967 1253 717 536 Standard errors clustered at the village level. Models run at the household member-level. Details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. Sample in (1) includes all 6-18 year-old children. Samples in (2)-(7) include children who fall within the age range of that level of education. Samples in (8)-(10) include children currently attending secondary school. * p < 0.10, ** p < 0.05, *** p < 0.01. 47 6.7 Household, mother, and child dietary diversity In this section, we examine household, mother, and child dietary diversity. These outcomes can be affected by spending decisions on food, non-food, and other expenditures and in- vestments. We first look at the entire household, and then dietary diversity of the mother and children. Table 6.7.1 shows that most households consumed between 9 and 11 food groups out of 12, which is reasonably high. In that respect, we may not actually expect to see much more improvement. Table 6.7.1. Number of Food Groups Consumed by Households <8 8 9 10 11 12 Total Number of households 329 670 1,365 1,966 1,814 331 6,475 Percentage of households 5.1 10.3 21.1 30.4 28.0 5.1 100.0 In Table 6.7.2 we present impact estimates on the dietary diversity of the household, of mothers in the household, as well as the dietary diversity a randomly selected child aged 6- 23 (up to 2 years old) months and one aged 24-59 months (between 2 and 5 years old). The outcome variables are sums of the number of different food groups consumed in the past 7 days. For households there are 12 groups,4 for mothers, there are 9 groups,5 for children aged 6-23 months there are 7 food groups,6 and for children aged 24-59 months there are 8 food groups.7 In general, dietary diversity at the household level is reasonable – on average, 10 groups out of 12. However, for mothers and young children, it is relatively low. Out of 9 food groups, mothers consume an average of 4, children 6-23 months old consume an average of 3.3 out of 7 groups, and children aged 24-59 months consume an average of 5 of 8 groups. The impacts on household dietary diversity are negative, but not significant except when us- ing the robust specifications. Also, while there is not strong evidence of statistical signifi- cance, there is a negative coefficient for dietary diversity of children aged 24-59 with a large effect size at 0.8 food groups. 4 The 12 food groups are: cereals, potatoes and tubers, vegetables, fruits, meat, eggs, fish, legumes nuts and seeds, dairy, oils and fats, sweets, and spices condiments and beverages. 5 The 9 food groups are: starchy foods, dark green leafy vegetables, vitamin A rich fruits, other fruits and vegetables, organ meat, meat and fish, eggs, legumes nuts and seeds, and milk and milk products. 6 The 7 food groups are: grains, roots, and/or tubers, legumes, nuts and/or seeds, milk and/or milk products, flesh foods, eggs, vitamin A rich fruits and/or vegetables, and other fruits and/or vegetables. 7 The 8 food groups are: grains, roots, and/or tubers, legumes, nuts and/or seeds, milk and/or milk products, flesh foods, eggs, vitamin A rich fruits and/or vegetables, other fruits and/or vegetables, and foods cooked in oil/fat. 48 Table 6.7.2 Impacts of Takaful Program on Dietary Diversity Outcomes Treatment variable: household received Takaful transfers in the past two months (1) (2) (3) (4) Household die- Mother's dietary Children's dietary di- Children's dietary di- tary diversity diversity score (0- versity score (0-7), 6- versity score (0-8), 24- score (0-12) 9) 23 months old 59 months old Conventional -0.287 -0.267 0.684 -0.273 (0.226) (0.293) (1.039) (0.370) Bias-corrected -0.707*** -0.419 0.133 -0.811** (0.226) (0.293) (1.039) (0.370) Robust -0.707** -0.419 0.133 -0.811 (0.294) (0.401) (1.436) (0.494) Mean Dep. Var. 9.772 3.842 3.344 4.909 N 6475 3340 483 2468 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Sample in (1) is all households, sample in (2) includes adult women who have children under 5 years (60 months) of age, and samples in (3) and (4) are chil- dren of that age range. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.8 Mother and child anthropometry, overweight and wasting, and morbidity and treatment In this section, we examine anthropometry measures for mothers and children health out- comes, which are influenced by dietary diversity. We first look at mother’s body mass index (BMI), then at child anthropometry, including overweight and wasting. Since anthropometric outcomes reflect a series of prior decisions and investments, we again present results using both treatment variables: whether the household received a Takaful transfer in the past two months and whether the household ever received a Takaful transfer. We first examine anthropometry outcomes of mothers, estimating impacts the mother’s body mass index (BMI), as well as indicators for whether the mother is considered overweight (BMI between 25 and 30) or obese (BMI above 30).8 Table 6.8.1 displays the results. There are no statistically significant impacts on women’s BMI, nor on the likelihood of a mother be- ing overweight or obese with either treatment variable. It is reassuring that women are not becoming more overweight as a result of the program. However, we do note that the aver- age BMI for all mothers in the sample is 28.96, which means that the average mother is in- deed overweight and is bordering on obese. This finding is important for integrated program- ming within Takaful and for any complementary programming on diets and weight. 8 Body Mass Index (BMI) is calculated as a person’s weight in kilograms (or pounds) divided by the square of height in meters (or feet). 49 Table 6.8.1. Impacts of Takaful Program on Mother Anthropometrics Outcomes Panel A: household received a Takaful transfer in the past two months (1) (2) (3) Mother's Body Mother is overweight Mother is obese (BMI Mass Index (BMI) (BMI between 25-30) above 30) Conventional 0.160 -0.109 0.0926 (0.736) (0.0788) (0.0774) Bias-corrected -0.681 -0.0516 -0.0119 (0.736) (0.0788) (0.0774) Robust -0.681 -0.0516 -0.0119 (0.978) (0.105) (0.103) Mean Dep. Var. 28.96 0.486 0.350 N 5711 5711 5711 Panel B: household ever received a Takaful transfer (1) (2) (3) Mother's Body Mother is overweight Mother is obese (BMI Mass Index (BMI) (BMI between 25-30) above 30) Conventional 0.0454 -0.0548 0.0459 (0.390) (0.0416) (0.0410) Bias-corrected -0.397 -0.0219 -0.0101 (0.390) (0.0416) (0.0410) Robust -0.397 -0.0219 -0.0101 (0.518) (0.0553) (0.0544) Mean Dep. Var. 28.96 0.486 0.350 N 5690 5690 5690 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. In Table 6.8.2 we look at children’s anthropometry for children aged 6-23 months old 24-59 months old. We examine BMI, whether the child is overweight, and the height-for-age Z- score for the two randomly selected children in the household. There are no program im- pacts on any of these outcomes. Children are much less likely to be overweight than moth- ers. Next, we examine stunting and wasting, in Table 6.8.3. Stunting is an indicator equal to one if the height-for-age Z-score is two standard deviations below the population mean, defined by the WHO. It is a measure of whether a child is considered “short”. Wasting an indicator equal to one if the weight-for-height Z-score is two standard deviations below the population mean and is a measure of whether a child is considered too “thin”. Both height and weight are indicators of whether the child receives proper nutrition in turn determines many out- comes like cognition and schooling performance, and recovery from illness. There are no program impacts on any of the outcomes when the treatment variable is that the household is currently a Takaful beneficiary. However, when we look at whether the household has ever been a beneficiary, we see a reduction in wasting for children aged 6-23 months. The effects are also large in magnitude and are economically meaningful. This result is an en- couraging finding of the program – sustained investments in people can lead to meaningful longer-term benefits. 50 Table 6.8.2. Impacts of Takaful Program on Child Anthropometrics Outcomes Panel A: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) (6) BMI BMI Overweight (0- Overweight (24- HAZ score HAZ score (0-23 months) (24-59 months) 23 months) 59 months) (0-23 months) (24-59 months) Conventional 0.341 -0.567 -0.205 -0.0337 -0.586 -0.104 (0.986) (0.479) (0.180) (0.0731) (0.972) (0.360) Bias-corrected -0.670 -0.806* -0.289 -0.0397 -0.399 -0.0430 (0.986) (0.479) (0.180) (0.0731) (0.972) (0.360) Robust -0.670 -0.806 -0.289 -0.0397 -0.399 -0.0430 (1.397) (0.611) (0.256) (0.0985) (1.340) (0.475) Mean Dep. Var. 0.353 0.358 0.125 0.0672 -0.108 -0.639 N 904 2408 905 2412 886 2392 Panel B: household ever received a Takaful transfer (1) (2) (3) (4) (5) (6) BMI BMI Overweight (0- Overweight (24- HAZ score HAZ score (0-23 months) (24-59 months) 23 months) 59 months) (0-23 months) (24-59 months) Conventional 0.164 -0.266 -0.0965 -0.0162 -0.276 -0.0620 (0.443) (0.233) (0.0802) (0.0360) (0.438) (0.177) Bias-corrected -0.257 -0.396* -0.146* -0.0196 -0.213 -0.0376 (0.443) (0.233) (0.0802) (0.0360) (0.438) (0.177) Robust -0.257 -0.396 -0.146 -0.0196 -0.213 -0.0376 (0.629) (0.297) (0.115) (0.0485) (0.601) (0.233) Mean Dep. Var. 0.353 0.358 0.125 0.0672 -0.108 -0.639 N 900 2399 901 2403 882 2383 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Models run at the household member-level. Details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 51 Table 6.8.3. Impacts of Takaful Program on Child Stunting and Wasting Panel A: Household received Takaful transfers in the past two months (1) (2) (3) (4) Child Stunted Child Stunted (24- Child Wasted (6- Child Wasted (6-23 months) 59 months) 23 months) (24-59 months) Conventional -0.0411 -0.0440 -0.225 0.00879 (0.190) (0.0838) (0.145) (0.0532) Bias-corrected -0.180 -0.0673 -0.236 -0.0166 (0.190) (0.0838) (0.145) (0.0532) Robust -0.180 -0.0673 -0.236 -0.0166 (0.257) (0.112) (0.202) (0.0696) Mean Dep. Var. 0.134 0.105 0.0673 0.0374 N 886 2392 892 2377 Panel B: Household ever received Takaful transfers Conventional -0.0212 -0.0208 -0.0996* 0.00176 (0.0847) (0.0411) (0.0590) (0.0258) Bias-corrected -0.0869 -0.0325 -0.116** -0.0108 (0.0847) (0.0411) (0.0590) (0.0258) Robust -0.0869 -0.0325 -0.116 -0.0108 (0.115) (0.0551) (0.0822) (0.0341) Mean Dep. Var. 0.134 0.105 0.0673 0.0374 N 882 2383 888 2368 Standard errors clustered at the village level. The treatment variable in Panel A is an indicator variable equal to one if the household reported a Takaful cash transfer in the two months prior to being interviewed, and the treatment variable in Panel B is an indicator variable equal to one if the household reported ever having received a transfer. Models run at the household member-level. Details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are in- cluded as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.9 Antenatal and postnatal care and infant and young child feeding (IYCF) practices This section explores outcomes related to antenatal and postnatal care and infant and young child nutrition (IYCN) practices. These outcomes are also influenced by spending decisions and cash liquidity since they have cost implications. We first explore whether the program caused more mothers to get antenatal and postnatal care during their last pregnancy. The sample includes adult females with children born between July 2020 and February 2022. Ta- ble 6.9.1 reports the results. There are no impacts of the program on whether the mother of the index child received ante- natal care (ANC) during her last pregnancy, the number of ANC sessions received, whether she took iron supplements, gave birth in a safe place, or received postnatal care within two days of giving birth. Many of the coefficients on receipt of antenatal and postnatal care are positive but none are statistically significant. We interpret this as suggestive evidence of a positive impact, but we are not able to reject that the coefficients are zero. 52 Table 6.9.1. Impacts of Takaful Program on Antenatal Care (ANC) Outcomes Treatment variable: household received Takaful transfers in the past two months (1) (2) (3) (4) (5) Mother received Number of ANC Took iron sup- Gave birth in a Received post- ANC during last sessions re- plements during save place natal care pregnancy ceived during last pregnancy (public or pri- within 2 days of last pregnancy vate facility) giving birth Conventional 0.184 2.123 -0.0145 0.0425 0.0461 (0.136) (1.873) (0.188) (0.137) (0.206) Bias-corrected 0.157 1.968 -0.175 -0.0606 0.0192 (0.136) (1.873) (0.188) (0.137) (0.206) Robust 0.157 1.968 -0.175 -0.0606 0.0192 (0.184) (2.632) (0.273) (0.190) (0.300) Mean Dep. Var. 0.899 6.640 0.766 0.900 0.654 N 829 745 820 829 829 Standard errors clustered at the village level. This sample includes adult women with children born between July 2020 and Feb- ruary 2022. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01 Next, we examine IYCN practices, specifically, whether someone helped put the baby to breast just after birth, whether anything but breast milk was given to the baby within the first three days, whether the child was given colostrum, and the age at which the mother stopped breastfeeding the child. The sample includes women with a baby less than two years old. Table 6.9.2 contains the results. Table 6.9.2. Impacts of Takaful Program on Infant and Young Child Feeding (IYCF) Practices Treatment variable: household received Takaful transfers in the past two months (1) (2) (3) (4) Someone helped Baby received only Mother gave Age at which mother put the baby to breast milk during baby colostrum mother stopped the breast after birth the first 3 days breastfeeding (in months) Conventional 0.193 -0.272 0.0427 -1.983 (0.196) (0.181) (0.102) (1.378) Bias-corrected 0.112 -0.0987 0.158 -1.214 (0.196) (0.181) (0.102) (1.378) Robust 0.112 -0.0987 0.158 -1.214 (0.281) (0.254) (0.146) (2.032) Mean Dep. Var. 0.466 0.718 0.937 4.420 N 1205 1204 1204 1099 Standard errors clustered at the village level. This sample includes adult women with children <2 years. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.10 Women’s decision-making and gender norms Women’s control over decision-making within the household was measured using a series of 9 questions asking women to describe their ability to influence decisions on a scale of 1-4, corresponding to ability to influence decisions “to a great extent” (4), “a medium extent” (3), 53 “a small extent” (2), or “not at all” (1). The domains were: wage employment, major and mi- nor household expenditures, how to use cash transfers, what food can be cooked, getting medical treatment and buying clothes for herself, taking a child to the doctor, and children’s schooling. Table 6.10.1 reports the results of indicator variables equal to one if the woman can influence decisions “to a great extent” on these domains in columns 1-9. We see that there is no impact of the program on any of these outcomes. Some coefficients are positive and some are negative, so there are no clear patterns in the results. While one coefficient is statistically significant, give the number of tests for significance being conducted, we should not interpret this as a negative effect. The main conclusion is that the program did not, on average, have any impact on women’s ability to make household decisions. In addition, an index for women’s control over decision-making was constructed using princi- pal component analysis (PCA). Column 10 in Table 6.10.1 reports this result and we see that there is no impact of the program on the index either, which is not surprising given that there were no impacts on any of its components. In the first round Takaful evaluation, we found negative and statistically significant impacts of the program on women’s decision-making, both in the overall index and in the domains of using government subsidies, taking a child to the doctor, and children’s schooling. These im- pacts are further explored in detail in Elenbaby et al (2021), where four findings emerge. First, the decreases in women’s decision-making are concentrated among women who did not have any formal education. Second, men began to participate more in household deci- sions; decision-making became more joint. Third, women who were working out of necessity stopped working as a result of the transfers. Fourth, qualitative evidence found that the pro- gram was generally well-perceived and that the pattern of results can be explained partly by a relaxation of the household’s budget constraint (there were more decisions to make about what to spend money on) and partly by the context. In Egypt, the relevant contextual factor is that men are perceived and expected to be primary decision-makers in the household, so the joint decision-making resulting from the transfers is not perceived poorly. 54 Table 6.10.1 Impacts of Takaful Program on Women's Decision-Making, All Women Treatment variable: household received a Takaful transfer in the past two months Woman can make own decisions to a great extent on: (1) (2) (3) (4) (5) Wage employ- Major house- Minor house- How to use What food can ment hold expendi- hold expendi- cash transfers be cooked tures tures every day Conventional 0.00336 0.100 0.00547 0.0958 0.0710 (0.0607) (0.0654) (0.0744) (0.0780) (0.0621) Bias-corrected 0.0342 0.0901 -0.102 -0.0271 0.0477 (0.0607) (0.0654) (0.0744) (0.0780) (0.0621) Robust 0.0342 0.0901 -0.102 -0.0271 0.0477 (0.0819) (0.0885) (0.0983) (0.105) (0.0817) Mean Dep. Var. 0.218 0.286 0.605 0.516 0.742 N 6473 6473 6473 6473 6473 Woman can make own decisions to a great extent on: (6) (7) (8) (9) (10) Getting medical Buying Taking a Children's school- Women’s de- treatment for her- clothes for child to a ing cision- making self herself doctor index Conventional 0.0905 0.0631 -0.00451 -0.00961 0.0446 (0.0714) (0.0712) (0.0725) (0.0752) (0.163) Bias-corrected 0.0444 0.0886 0.0332 -0.170** -0.135 (0.0714) (0.0712) (0.0725) (0.0752) (0.163) Robust 0.0444 0.0886 0.0332 -0.170 -0.135 (0.101) (0.0958) (0.0977) (0.104) (0.215) Mean Dep. Var. 0.507 0.450 0.572 0.478 -0.0866 N 6473 6473 6473 6055 6055 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. Consequently, we next split the sample into women who have some formal education and those who have no formal education to examine whether the pattern uncovered in Elenbaby et al (2021) still holds. There is no evidence that the pattern was maintained – there are no effects of the program on the decision-making of women when they have some formal edu- cation. However, for women with no formal education, decision-making regarding what foods to cook every day increases by about 20 percentage points, which is large. There are no systematic patterns in the results, so we interpret these findings as the program having a small positive impact on this one domain of women’s decision-making among women who have no formal education. 55 Table 6.10.2. Impacts of Takaful Program on Women's Decision-Making for Women with Some Formal Education Treatment variable: household received Takaful transfer in the past two months Woman can make own decisions to a great extent on: (1) (2) (3) (4) (5) Women deci- Wage employ- Major house- Minor house- How to use sion-making in- ment hold expendi- hold expendi- cash transfers dex tures tures Conventional 0.0807 0.0402 0.151* 0.0286 0.0884 (0.195) (0.0735) (0.0820) (0.0895) (0.0937) Bias-corrected -0.179 0.0604 0.120 -0.114 -0.0896 (0.195) (0.0735) (0.0820) (0.0895) (0.0937) Robust -0.179 0.0604 0.120 -0.114 -0.0896 (0.260) (0.101) (0.111) (0.119) (0.126) Mean Dep. Var. -0.0866 0.218 0.286 0.605 0.516 N 4365 4540 4540 4540 4540 (6) (7) (8) (9) (10) What food can Getting medical Buying clothes Taking a child Children's be cooked treatment for her- for herself to a doctor schooling every day self Conventional 0.00806 0.114 0.122 -0.0441 -0.0131 (0.0793) (0.0884) (0.0893) (0.0881) (0.0910) Bias-corrected -0.00997 0.0153 0.147 0.0582 -0.133 (0.0793) (0.0884) (0.0893) (0.0881) (0.0910) Robust -0.00997 0.0153 0.147 0.0582 -0.133 (0.104) (0.123) (0.122) (0.117) (0.125) Mean Dep. Var. 0.742 0.507 0.450 0.572 0.478 N 4540 4540 4540 4540 4365 Standard errors clustered at the village level. Sample includes women who have some formal education. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 56 Table 6.10.3. Impacts Of Takaful Program on Women's Decision-Making for Women with No Formal Education Treatment variable: household received Takaful transfers in the past two months Woman can make own decisions to a great extent on: (1) (2) (3) (4) (5) Women deci- Wage employ- Major house- Minor house- How to use sion-making in- ment hold expendi- hold expendi- cash transfers dex tures tures Conventional -0.0373 -0.0829 -0.0156 -0.0458 0.107 (0.277) (0.104) (0.113) (0.122) (0.132) Bias-corrected 0.00505 -0.0217 0.0119 -0.0806 0.123 (0.277) (0.104) (0.113) (0.122) (0.132) Robust 0.00505 -0.0217 0.0119 -0.0806 0.123 (0.368) (0.140) (0.154) (0.167) (0.184) Mean Dep. Var. -0.0866 0.218 0.286 0.605 0.516 N 1690 1933 1933 1933 1933 (6) (7) (8) (9) (10) What food can Getting medical Buying clothes Taking a child Children's be cooked treatment for her- for herself to a doctor schooling every day self Conventional 0.207** 0.0415 -0.0640 0.0830 -0.00492 (0.105) (0.130) (0.124) (0.122) (0.132) Bias-corrected 0.195* 0.124 -0.0410 0.0119 -0.243* (0.105) (0.130) (0.124) (0.122) (0.132) Robust 0.195 0.124 -0.0410 0.0119 -0.243 (0.138) (0.180) (0.169) (0.175) (0.187) Mean Dep. Var. 0.742 0.507 0.450 0.572 0.478 N 1933 1933 1933 1933 1690 Standard errors clustered at the village level. Sample includes women who have no formal education. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. Another aspect of women’s empowerment concerns gender norms. In the questionnaire, we ask several questions regarding gender norms, including whether men should make the im- portant household decisions, whether men should help with housework if a woman works outside the home, whether a husband should allow his wife to work outside the home, whether a woman has the right to express her opinion even if she disagrees with a man, and whether a woman must accept if her husband beats her. We also ask the female respondent whether she has money that she can use for whatever she wants, how often she talks to a non-household family member, and about her mobility. We construct a mobility index using PCA with four questions on how often the woman can go to the market, to a friend or family member’s house that is an hour away, to a place outside the village, and to see a healthcare provider. Table 6.10.4 shows the results. We see that Takaful beneficiaries seem to have more gender-equal norms compared to non-beneficiaries. Using both definitions of participa- tion, the impact estimates for the gender norms index is positive and statistically significant using conventional estimates. Other variables are not statistically significant. We take this as some evidence, but not strong evidence, that Takaful improved gender norms. 57 Table 6.10.4. Impacts of Takaful Program on Gender Norms Panel A: Household received Takaful transfers in the past two months (1) (2) (3) (4) Norms and attitudes Do you have your own How often do you Mobility in- around gender money that you can use talk to a family mem- dex (higher index=more for what you want to use ber not living in the liberal) it? HH? Conventional 0.465** 0.0191 0.106 0.0712 (0.224) (0.0482) (0.195) (0.185) Bias-corrected 0.176 -0.0632 0.232 -0.0874 (0.224) (0.0482) (0.195) (0.185) Robust 0.176 -0.0632 0.232 -0.0874 (0.297) (0.0635) (0.254) (0.242) Mean Dep. Var. 2.76e-09 0.105 6.021 -0.157 N 5832 6473 6473 6203 Panel B: Household ever received Takaful transfers Conventional 0.245** 0.00916 0.0605 0.0257 (0.115) (0.0256) (0.104) (0.0988) Bias-corrected 0.0900 -0.0336 0.126 -0.0632 (0.115) (0.0256) (0.104) (0.0988) Robust 0.0900 -0.0336 0.126 -0.0632 (0.153) (0.0336) (0.135) (0.129) Mean Dep. Var. 2.76e-09 0.105 6.021 -0.157 N 5813 6449 6449 6179 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.11 Mental Health In this section, we examine program impacts on the mental health of the survey respondent. This was almost always the head of the household, who is predominantly male. It is now widely documented that cash transfers have the potential to improve mental health out- comes (Ridley et al 2020). The poor face a lot of uncertainty and stressors and face constant worry. These issues can then lead to poor mental health in the form of depression and anxi- ety. The evidence base is mixed, with some transfer programs having large effects on men- tal health and others having no impacts (negative impacts are very rare). In Table 6.11.1 we examine three outcomes related to mental health. The first is a worry scale calculated from 12 questions regarding the extent of worry about having enough money to purchase essentials, having enough money to purchase other items, being able to find consistent work, having the ability to sufficiently feed the family, being able to repay debt, being able to repay installment payments, the death of a household member, illness of a family member, being able to pay medical bills, being able to pay education expenses, a natural disaster, or that the family’s economic situation will become worse. The choices were not worried at all (1 point), a bit worried (2 points), quite worried (3 points), or extremely wor- ried (4 points). We summed responses over all 12 questions and the outcome is represented in Column 1. In Column 2 we present results on general anxiety using the Generalized Anxi- ety Disorder (GAD-7) scale (Spitzer et al, 2006). The GAD-7 includes questions regarding various indications of generalized anxiety rather than specific worries. In Column 3, we esti- mate impacts on self-esteem. Self-esteem is important in feeling optimistic about the future and in feelings of agency, helping people to make decisions in a better way. We use the 58 Rosenberg Self-Esteem Scale as our measure (Rosenberg, 1965). In all three measures, we see that there are no significant impacts of the Takaful program. Table 6.11.1. Impacts of Takaful Program on Mental Health Indicators (1) (2) (3) Worry score General Anxiety Disorder Rosenberg Self-Esteem (GAD-7) score score Conventional 0.546 -0.212 0.356 (1.413) (0.789) (0.888) Bias-corrected 1.412 -0.316 -0.146 (1.413) (0.789) (0.888) Robust 1.412 -0.316 -0.146 (1.900) (1.044) (1.157) Mean Dep. Var. 17.63 7.309 37.27 N 6473 6473 6473 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Worry score sums the level of worry over 12 items, ranges between 0 and 36. General Anxiety Disorder (GAD-7) score sums anxiety level over 7 items, ranges between 0 and 21. Rosenberg Self-Esteem score sums 10 self-esteem items, ranges between 0 and 50. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01 6.12 Shocks and coping strategies In this section we examine the impacts of the Takaful program on shocks experienced by households, and the types of ways that households cope with shocks. We examine shocks in general and shocks specifically caused by COVID-19. This type of social protection pro- gram may help shield households from certain types of shocks, and/or may help households to better cope with shocks. We first look at the number of shocks experienced by the house- hold in the past five years in Table 6.12.1. We report whether the household experienced any shock, the number of shocks experi- enced, and whether any of the shocks experienced were a result of COVID-19. Shocks are relatively common, with 61% of households having experienced a shock in the past 5 years and 23% of households reporting that a shock experienced in 2020 or 2021 was due to the pandemic. The probability of experiencing a shock, the number of shocks, and the probabil- ity of a shock being due to the pandemic do not differ between recipient and non-recipient households. This result is expected because we do not expect that the program would change the likelihood of a shock, but rather, the way that households cope with shocks. 59 Table 6.12.1 Impacts of Takaful Program on Shocks Experienced by Household Treatment variable: household received Takaful transfer in the past two months (1) (2) (3) Household experienced Number of shocks Household experienced a any shocks in past 5 in past 5 years shock in 2020/21 due to years COVID-19 Conventional 0.0567 0.0515 0.0397 (0.0805) (0.119) (0.0659) Bias-corrected 0.0714 0.0559 0.0691 (0.0805) (0.119) (0.0659) Robust 0.0714 0.0559 0.0691 (0.105) (0.152) (0.0871) Mean Dep. Var. 0.605 0.764 0.231 N 6473 6475 6475 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. We then look at potentially detrimental ways that households tend to cope with shocks: sell- ing productive assets, household goods or jewelry, borrowing money from relatives, a bank, or NGO, eating less or lower quality food, reducing spending on school or healthcare, chang- ing occupations, using savings, or having children work outside the home. The impact esti- mates for these outcomes are reported in table 6.12.2. There are no statistically significant impacts on most of these outcomes. The main coping strategy was selling gold or jewelry, but the magnitude of the impact is quite small. There is also weak evidence that borrowing from relatives, traders, and banks was lower. Unfortunately, there is also weak and sugges- tive evidence that households responded to shocks by eating less food and eating less nutri- tious food to reduce food expenditures. Reassuringly, very few households engaged in extremely harmful processes like having their daughters marry early (this is very rare in the data so not reported in the table), spending less on health and education, or having children work outside the home. 60 Table 6.12.2. Impacts of Takaful Program on Coping Methods for Shocks Experienced by the Household Treatment variable: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) (6) (7) Sold productive Sold household Sold gold/ jew- Borrowed money Borrowed Took a loan from Ate less food to asset good elry from relatives money from NGO/institution reduce expenses trader or bank Conventional 0.00307 0.000491 0.0218* -0.109 -0.0124 0.0139 0.0207 (0.00512) (0.00735) (0.0119) (0.0865) (0.0266) (0.0213) (0.0824) Bias-corrected 0.00807 -0.0136* 0.0245** -0.159* -0.00343 0.0150 0.144* (0.00512) (0.00735) (0.0119) (0.0865) (0.0266) (0.0213) (0.0824) Robust 0.00807 -0.0136 0.0245* -0.159 -0.00343 0.0150 0.144 (0.00697) (0.0113) (0.0144) (0.116) (0.0368) (0.0296) (0.112) Mean Dep. Var. 0.00102 0.00204 0.00587 0.321 0.0263 0.0128 0.310 N 3914 3914 3915 3919 3917 3914 3922 (8) (9) (10) (11) (12) (13) (14) Ate lower qual- Reduced Reduced Adult household Forced to Used savings Children started ity food to re- spending on ed- spending on member temporarily change occupa- working outside duce expenses ucation health care took job elsewhere tion the home Conventional 0.0680 -0.0399 0.00139 -0.00471 0.00272 0.0113 -0.00317 (0.0733) (0.0367) (0.0166) (0.00730) (0.00787) (0.0213) (0.00825) Bias-corrected 0.0409 0.00783 -0.0103 -0.00422 0.00338 -0.0135 -0.0106 (0.0733) (0.0367) (0.0166) (0.00730) (0.00787) (0.0213) (0.00825) Robust 0.0409 0.00783 -0.0103 -0.00422 0.00338 -0.0135 -0.0106 (0.101) (0.0511) (0.0206) (0.0104) (0.0109) (0.0311) (0.0115) Mean Dep. Var. 0.221 0.0470 0.0102 0.00204 0.00204 0.0135 0.00255 N 3921 3914 3914 3914 3914 3916 3914 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 61 6.13 COVID-19 The COVID-19 pandemic has disrupted life around the globe since early 2020 when its exist- ence was first announced. Millions of people globally have died or become severely ill, lost their livelihoods, gone hungry, lost years of learning, and suffered mental health problems, among many other negative effects. In this section, we examine the ways in which house- holds in our sample suffered due to the pandemic, as well as the ways in which they coped with the negative effects. Table 6.13.1 presents the results. On average, one third of households were negatively im- pacted by COVID-19 in some way, and there were many negative shocks experienced. More than half of households experienced job loss and/or a pay cut, and 16% of households expe- rienced were actually infected with COVID-19.9 Of those infected, two thirds sought medical care, and 44% needed to borrow money to cover the associated medical costs. There are no impacts of the program on the likelihood of being negatively affected by the pandemic, nor the type of negative effect suffered, nor the coping strategies used to address negative ef- fects. The interpretation is that the Takaful program did not lead to less suffering from COVID-19, and also did not particularly cushion any negative effects experienced by house- holds. 9 We should not take this figure at face value as there may be substantial misreporting. The availability of testing for COVID-19 was generally sparse in most of the country. 62 Table 6.13.1. Impacts of Takaful Program on Experiences with The COVID-19 Pan- demic Treatment variable: household received a Takaful transfer in the past two months (1) (2) (3) (4) (5) Household Household experi- Household Household Household was affected enced job loss experienced was in- borrowed to by COVID-19 pay cut fected with cope with COVID-19 COVID-19 Conven- 0.0280 -0.0115 -0.0975 0.0679 -0.0000475 tional (0.0798) (0.153) (0.157) (0.110) (0.0637) Bias-cor- -0.102 -0.0413 -0.311** 0.0722 -0.0445 rected (0.0798) (0.153) (0.157) (0.110) (0.0637) Robust -0.102 -0.0413 -0.311 0.0722 -0.0445 (0.106) (0.202) (0.211) (0.152) (0.0844) Mean Dep. 0.321 0.555 0.590 0.161 0.206 Var. N 6473 2075 2076 2075 6472 (6) (7) (8) (9) Household lent money to help another Household Household Household household cope with COVID-19 member(s) member in- borrowed migrated for fected with money to work before COVID-19 cover COVID- the pan- sought 19 medical demic medical care costs care Conven- -0.00151 -0.00239 0.0379 0.0199 tional (0.0103) (0.00422) (0.357) (0.364) Bias-cor- 0.00286 -0.00636 -0.284 0.495 rected (0.0103) (0.00422) (0.357) (0.364) Robust 0.00286 -0.00636 -0.284 0.495 (0.0152) (0.00479) (0.487) (0.494) Mean Dep. 0.00556 0.000927 0.673 0.435 Var. N 6472 6470 352 352 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (ur- ban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01. 6.14 Robustness to alternate definition of beneficiary status In this report, we have focused on two ways of defining a Takaful beneficiary: whether the household reports ever having received a Takaful transfer and whether the household has received a Takaful transfer in the past two months. We believe that these are the two most relevant definitions of beneficiary status. The first definition can be interpreted as whether any program participation had an impact on the outcomes. The second definition can be in- terpreted as whether recent transfers have had an impact on the outcomes. For some out- comes, one is more relevant than the other and for other outcomes, both definitions could be relevant. Accordingly, we have presented our results as such. 63 Another way to think about beneficiary status is intensity of treatment, and rather than desig- nate a household as a beneficiary or non-beneficiary, focus on the number of months that the household received transfers. Those who have never received a transfer are coded as zero months. The interpretation here is whether a higher intensity of treatment, a longer du- ration of transfers, has an effect on the outcomes considered. We consider this definition as less informative because the interpretation is difficult. The transfers may have been received for several years and then stopped, or they could have been received for the same number of months but more recently. The “treatment” would be the same, but we could not distin- guish between the two. Consequently, our main results focus on the first two definitions. Nonetheless, in Table 6.14.1 we use this alternate definition and estimate impacts on our main outcomes: total consumption, non-food consumption, food consumption, savings, debt, and investments in durable, productive, and livestock assets. We see that all of the results are in the same direction and of similar statistical significance. Accordingly, all three defini- tions result in consistent estimates and the results should be considered robust. In Table 6.14.2, we consider another alternative definition: whether a household is a current beneficiary (received transfers in the past 2 months) and received at least 24 transfers. This definition captures recent transfers and a longer duration of transfers. Once again, the re- sults are robust to this definition and suggest that our results are robust to this alternative definition as well. It also reflects the fact that few households (17%) experienced disruptions in transfers and that stopping to receive transfers during the past few years did not substan- tially affect household behaviour. 64 Table 6.14.1. Impacts of Takaful Program Duration (Self-reported) (1) (2) (3) (4) (5) (6) (7) (8) Monthly food Monthly non- Monthly total Amount of Total amount of debt Durables Productive Livestock consumption food con- (food & non-food) savings currently owed to infor- assets expenditure sumption ex- consumption ex- (EGP) mal lenders or for pur- penditure penditure (IHS) chases on credit (IHS) Conventional -0.000202 -0.000671 -0.000419 -0.00235 -0.0133 -0.000237 0.00534** 0.00408 (0.000814) (0.000848) (0.000692) (0.00146) (0.00813) (0.00257) (0.00212) (0.00261) Bias-corrected -0.00124 -0.00172** -0.00129* -0.00271* -0.0177** -0.000124 0.00173 0.000691 (0.000814) (0.000848) (0.000692) (0.00146) (0.00813) (0.00257) (0.00212) (0.00261) Robust -0.00124 -0.00172 -0.00129 -0.00271 -0.0177 -0.000124 0.00173 0.000691 (0.00105) (0.00111) (0.000891) (0.00185) (0.0111) (0.00341) (0.00293) (0.00404) Mean Dep. Var. 6.791 6.526 7.396 0.0956 3.717 2.51e-09 6.72e-09 -3.06e-09 N 6449 6449 6449 6449 6449 6449 6449 6449 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Consumption aggregates shown are winsorized at the 2nd and 99th percentiles, calculated as Adult Equivalent Units (AEU), and transformed using Inverse Hyperbolic Sine (IHS). Asset indices are constructed based on the first principal component from principal component analysis (PCA). The index that includes all assets uses dummies for assets ownership. The same for the durables and productive assets indices. The livestock index is composed using a count of the livestock owned by the household. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01 65 Table 6.14.2. Impacts of Takaful program receipt in last 2 months, and having received at least 24 transfers overall (1) (2) (3) (4) (5) (6) (7) (8) Monthly food Monthly non- Monthly total Amount of Total amount of debt Durables Productive Livestock consumption food consump- (food & non- savings currently owed to in- assets expenditure tion expenditure food) consump- (EGP) formal lenders or tion expenditure (IHS) owed for purchases on credit (IHS) Conventional -0.0146 -0.0459 -0.0280 -0.171 -0.937 -0.0425 0.386** 0.289 (0.0594) (0.0614) (0.0503) (0.106) (0.592) (0.187) (0.155) (0.190) Bias-corrected -0.0868 -0.118* -0.0872* -0.202* -1.316** -0.0203 0.132 0.0483 (0.0594) (0.0614) (0.0503) (0.106) (0.592) (0.187) (0.155) (0.190) Robust -0.0868 -0.118 -0.0872 -0.202 -1.316 -0.0203 0.132 0.0483 (0.0765) (0.0805) (0.0650) (0.135) (0.809) (0.249) (0.214) (0.294) Mean Dep. Var. 6.791 6.526 7.396 0.0956 3.717 2.51e-09 6.72e-09 -3.06e-09 N 6475 6475 6475 6475 6475 6474 6474 6474 Standard errors clustered at the village level. The full sample (N=6475) consists of 2539 households who reported a Takaful cash transfer in the two months prior to being interviewed, and 3936 households who did not. Consumption aggregates shown are winsorized at the 2nd and 99th percentiles, calculated as Adult Equivalent Units (AEU), and transformed using Inverse Hyperbolic Sine (IHS). Asset indices are constructed based on the first principal component from principal component analysis (PCA). The index that includes all assets uses dummies for assets ownership. The same for the durables and productive assets indices. The livestock index is composed using a count of the livestock owned by the household. Model details: Linear Trend on PMT Score; Uniform Kernel; RD Bandwidth=63. The following strata indicators are included as covariates: Lower Egypt (urban), Lower Egypt (rural), Upper Egypt (urban), Upper Egypt/Frontier (rural), excluded category: Metropolitan. * p < 0.10, ** p < 0.05, *** p < 0.01 66 7. CONCLUSIONS AND RECOMMENDATIONS This report has presented the findings from a second-round evaluation of the Takaful cash transfer pro- gram. The first-round evaluation reported on data collected in 2017, and this second round evaluation uses data collected in 2021. The goal of this second evaluation is to assess whether there may have been medium term and sustained impacts for Takaful beneficiaries who have been in the program for several years. Takaful is a cash transfer program that provides income support to the poor and most vulnerable; namely poor families with children (under 18 years of age) and the poor elderly (aged 65 years and above). It is implemented by the Ministry of Social Solidarity (MoSS) and co-financed by the Govern- ment of Egypt and the World Bank. Targeting for the program uses a combination of geographical tar- geting and application of a Proxy Means Test (PMT), an index of well-being based on household de- mographics, income, housing quality, assets and other characteristics. In poor districts, potentially eli- gible households were registered and interviewed to collect information for the PMT. Households with a PMT score below a preset threshold, 4,500 points, were considered eligible for the program and re- ceived transfers. A household survey for the impact evaluation was conducted in early 2022 by the firm El-Zanaty and Associates. The sample for the evaluation includes 6,473 households. The evaluation sample was se- lected from the administrative database of registrants for the program and the sample was first re- stricted to households who had been in the program for the longest. Subsequently, households who fell within 63 points above or below the 4500 threshold were sampled. The impact evaluation was designed using a regression discontinuity (RD) methodology, which is effec- tive for measuring the impact of programs that use a threshold level of a continuous measure of well- being, like a PMT score, to determine access to the program. The RD approach compares outcomes for beneficiaries just below the threshold for eligibility to outcomes for non-beneficiaries just above the threshold. Because the specific level of the eligibility threshold is not within the control of program appli- cants, whether households near the threshold end up below it or above is nearly random and cannot be affected by their actions. Consequently, the application of the threshold PMT score creates a quasi-ex- periment locally around the threshold that is used to measure impact of the program. Because the PMT score is not a perfect predictor of program participation (some households above the threshold participate in the program and some households below the threshold do not participate) we use a ‘fuzzy’ regression discontinuity design. The fuzzy RD adjusts for the fact that the threshold PMT score does not perfectly predict participation by estimating the model in two stages: the first stage pre- dicts the probability of participating in the program as a function of being below the eligibility threshold on the PMT score and the second stage measures impact as the change in the level of the outcome variable that is due to the difference in predicted probability of participating in the program as a result of the use of the threshold level of the PMT score. Households have four major ways to spend transfers and income: consumption, savings, debt reduc- tion, or investments in assets. We first examine these four choices. We then examine outcomes that 1 may depend on the choice of where transfers and income were used, such as dietary diversity, school- ing, infant and young child feeding practices, ante- and post-natal care, and anthropometry. We also examined women’s empowerment, gender norms, mental health, and shocks. Given the prominence of the COVID-19 pandemic and the resulting severely negative impacts on poor households, we also col- lected data on how households were affected by the pandemic and the strategies they used to cope with it. We do not detect effects on food and non-food consumption expenditures per adult equivalent unit (similar to per capita) compared to non-beneficiaries. This result contrasts with the results in the first- round evaluation, which found a statistically significant increase in the value of monthly food consump- tion per AEU by 8.3 – 8.9%. It is worth highlighting that this is not a precise estimate as the statistical power of the second-round evaluation is reduced compared to the first round, so it is possible that there is a positive but smaller impact on consumption than in the first round. The difference in food consumption is driven by a few food categories: grains, fruits, eggs, oils, and fats. Some of these groups are nutritious (grains, fruits, and eggs) but the reduction in oil and fat con- sumption should not have a negative nutritional impact. Encouragingly, consumption of unhealthy snacks and beverages, as well as food consumed outside of the household, has not increased. The decrease in non-food consumption expenditures is due to a decrease in construction expenditures and a decrease in expenditures on communications, such as phones and televisions. Rather than pur- chasing these items, it is possible that households instead were investing in their capacity to generate future income. We see increases in asset holdings, and in particular, productive assets. These assets include both agricultural technologies and livestock. Beneficiary households invested in large items such as tractors, plows, irrigation, buffaloes, and cows. These are lumpy expenditures that many households may not be able to save for and should generate a persistent increase in income. Benefi- ciary households also had lower levels of debt than non-beneficiary households. We find that beneficiary households had more children 6-11 years old due to a slight imbalance across the 4500 cutoff at the time of registration. Additionally, we find suggestive evidence that beneficiary households had more children in the past 5 years as a result of receiving the transfers. This result may explain part of the null effect on consumption. Beneficiary households did not save differentially compared to non-beneficiary households, but they did lower their debt owed to informal lenders and paid off their installment payments. Additionally, they in- vested in assets, primarily productive assets in the form of items for small businesses and livestock. These behaviors indicate that households may have moved out of the stage of needing to increase consumption because minimum needs are not being met and may have transitioned into the next phase where they are able to build their assets and improve their future income. In terms of labor supply, we see a difference between beneficiary and non-beneficiary households. Beneficiary households are significantly more likely to be engaged in informal labour and significantly less likely to be engaged in formal labor This result may possibly be driven by the increased productive asset holdings of households. The assets purchased may have enabled households to start their own businesses. 2 There are very encouraging results with regards to children’s schooling. The likelihood that children are enrolled in both primary and preparatory school increased substantially, by between 4 to 5 percentage points. This is also an investment in households’ future – these children should go on to earn more and may further push their families out of poverty. Girls attending secondary school are also more likely to attend school regularly as a result of the program. There are mixed results when it comes to child nutrition, and the results depend on the age of the child. Dietary diversity is lower for 2-5 year olds among beneficiaries, but wasting is reduced for children aged 6-23 months for those who have ever received transfers. It is possible that the increased number of children led households to focus their attention on fewer foods. The previous evaluation of Takaful found a decrease in women’s decision-making power among women who had no formal education. We test whether this pattern still holds, and we find that it does not. There continues to be no impact of the program on women who have some formal education, but now, beneficiary women with no education are more likely to be able to influence decisions regarding what food can be cooked every day. We do not interpret these results as a large positive shift in women’s decision making, but it is encouraging that women’s decision-making was not reduced by the program. We also see some evidence of higher levels of gender positive norms among beneficiary households. We do not find any effects on mental health – worries, generalized anxiety, or self-esteem. The households in our sample were indeed exposed to shocks, including COVID-19 and its resulting repercussions on movement and availability of foods. However, households did not use harmful coping strategies such as pulling children out of school and having them work, having daughters marry early, or reducing food consumption. The only response from beneficiary households is that they are more likely to sell gold/jewelry to cope with shocks. This evaluation has shown that the Takaful program caused several positive shifts in households. There were investments in physical and human capital. The overall message is that while some as- pects of household behavior were either not affected by the program or even had negative effects, these investments may reflect shorter term decisions that households believe will pay off in the future. Several policy recommendations emerge from these findings: Takaful should be continued and even possibly extended. The program enabled households not to resort to coping with shocks in negative ways. Particularly in light of increasingly frequent global shocks like COVID-19 and the Russian invasion of Ukraine, social protection programs, including cash transfer programs like Takaful, could be an effective way to protect against large-scale shocks since the infra- structure to reach people is largely in place. Proceed with plans for recertification and graduation of beneficiaries who have achieved self-suf- ficiency while using a generous cut-off for self-sufficiency (and generous duration of exposure to the program) given that many households have not managed to substantially increase their consumption in spite of increased productive assets. Improve communication regarding exclusion restrictions, program length, and recertification so that beneficiaries understand that they will not be excluded from the program for formal sector work with in- come below a certain threshold and to ensure that beneficiaries are not surprised by sudden changes in program status or unnecessarily worried about the short-term continuity of the transfers. 3 Consider greater coordination with communication campaigns related to family planning if the behavioral response by families of having more children is seen as in conflict with other national policy goals. Continue to work towards a comprehensive social protection strategy that helps to continue pro- tecting the poor as well as contributing to achieving longer-term developmental goals. Coordinating with the Ministry of Education to provide high quality public service delivery will magnify the impacts of in- creased school enrollment. Complementary programming would also be beneficial. In general, complementary programming on issues such as nutrition practices or financial training need to be quite intensive to be impactful. There are currently programs that are implemented by the Government of Egypt on these topics, partic- ularly a nation-wide nutrition campaign. However, it would be worth considering pairing these programs and intensifying them by leveraging Takaful to link to already vulnerable households. 4 ABOUT THE AUTHORS Hoda El Enbaby is a PhD Candidate at Lancaster University. Dalia Elsabbagh is a Senior Re- search Assistant at IFPRI, based in Cairo, Egypt. Dan Gilligan is a Senior Research Fellow and Dep- uty Director of the Poverty, Health, and Nutrition Division at IFPRI, based in Washington, DC. Naureen Karachiwalla is a Research Fellow in the Poverty, Health, and Nutrition Division (PHND) at IFPRI, based in Washington, DC. Bastien Koch is a Research Analyst at IFPRI, based in Washington, DC. Sikandra Kurdi is a Research Fellow in the Development Strategy and Governance Division at IFPRI. ACKNOWLEDGMENTS This impact evaluation was managed by the World Bank and funded by the United Kingdom Foreign and Commonwealth Office (UK FCO) and the United States Agency for International Development (USAID). We gratefully acknowledge very helpful consultations on the design and conduct of the evalu- ation with the Ministry of Social Solidarity of the Government of Egypt, in particular Her Excellency Dr. Nivine El-Kabbag, Minister of Social Solidarity, Mr. Raafat Shafeek, Minister of Social Solidarity Advisor for Strengthening Social Safety Net Programs and Executive Director of Takaful and Karama Condi- tional Cash Transfer Program, and Eng. Amal Helmy, Database Administration and Quality Assurance Specialist, and advice from Dr. Heba Ellaithy, Professor of Statistics at Cairo University, and Dr. Hania Sholkamy, Professor of Anthropology at American University of Cairo. The impact evaluation has also benefitted substantially from valuable comments from the World Bank team led by Ms. Nahla Zeitoun and including Dr. Imane Helmy and Ms. Matuna Mostafa. We are grateful to Dr. Fatma El-Zanaty and the research firm El-Zanaty and Associates for their role in collecting the household survey data. We would also like to thank Prof. Matias D. Cattaneo, for his consultation and support. Finally, we thank the participants of the study for their time. Address for correspondence: Naureen Karachiwalla, n.karachi- walla@cgiar.org. REFERENCES Alderman, H. (2017). Key Steps for Nutrition-Sensitive Social Protection. World Bank. Retrieved from https://the- docs.worldbank.org/en/doc/559041529969957669-0160022017/render/4pmMarch13NutritionSensitive2017.pdf Al-Masry Al-Yom. (2022). Egypt Independent. Retreived from https://www.almasryalyoum.com/news/details/2566957 Bastagli, F., Hagen-Zanker, J., Harman, L., Barca, V., Sturge, G., Schmidt, T., & Pellerano, L. (2016). Cash transfers: what does the evidence say. A rigorous review of programme impact and the role of design and implementation features. London: ODI, 1(7). Breisinger, C., Mukashov. A., Raouf, M., Wiebelt, M. (2018). Phasing Out Energy Subsidies as Part of Egypt’s Economic Reform Program: Impacts and Policy Implications. IFPRI MENA Working Paper 07. Calonico, S., Cattaneo, M., Titiunik, R. (2014). Robust Nonparametric Confidence Intervals for Regression-Discontinuity Designs. Economet- rica, 82, 2295-2326. Ecker, O., Al-Riffai, P., Breisinger, C., & El-Batrawy, R. (2016). Nutrition and Economic Development: Exploring Egypt's Exceptionalism and The Role of Food Subsidies. International Food Policy Research Institute. Retrieved from http://dx.doi.org/10.2499/9780896292406. El-Enbaby, H., Gilligan, D. O., Karachiwalla, N., Kassim, Y., & Kurdi, S. (2021). “Cash Transfers, Gender Norms, and Women’s Control over Decision-Making in Egypt”, working paper. El-Garhy, A. (2017). Egypt and IMF Chart Path to Long-term Prosperity. The Financial Times. Retrieved from https://www.ft.com El-Sheneity, S. (2014). Proxy Means Testing (PMT), Documentation of model-building process. MoSS internal document. Cairo. Fiszbein, A., and Schady, N. (2009). Conditional Cash Transfers: Reducing Present and Future Poverty. World Bank. Retrieved from https://openknowledge.worldbank.org/handle/10986/2597 Glewwe, Paul. (2002). Schools and Skills in Developing Countries: Education Policies and Socioeconomic Outcomes. Journal of Economic Literature, 40 (2): 436-482. 5 Hagenaars, A., K. de Vos and M.A. Zaidi (1994), Poverty Statistics in the Late 1980s: Research Based on Micro-data, Office for Official Publi- cations of the European Communities. Luxembourg. Hanushek, E. A., & Zhang, L. (2009). Quality-consistent estimates of international schooling and skill gradients. Journal of Human capital, 3(2), 107-143. Haushofer, J., & Shapiro, J. (2018). The long-term impact of unconditional cash transfers: experimental evidence from Kenya. Busara Center for Behavioral Economics, Nairobi, Kenya. Hidrobo, M., Hoddinott, J., Kumar, N., & Olivier, M. (2018). Social protection, food security, and asset formation. World Development, 101, 88- 103.IMF, 2015 IMF, 2015. Arab Republic of Egypt 2014 Article IV Consultation—Staff Report. International Monetary Fund. Washington, D.C. IMF (International Monetary Fund) Country Focus. (2021). “Egypt: Overcoming the COVID Shock and Maintaining Growth”. Retrieved from https://www.imf.org/en/News/Articles/2021/07/14/na070621-egypt-overcoming-the-covid-shock-and-maintaining- growth#:~:text=Egypt%20was%20one%20of%20the%20few%20emerging%20market%20coun- tries%20that,the%20face%20of%20the%20pandemic. Kondylis, F., & Loeser, J. (2021). Intervention Size and Persistence. World Bank Policy Research Working Paper No. 9769. Ministry of Finance. (2017). Monthly Bulletin, October 2017. Government of Egypt. Cairo. Moselhy, A. (2017). Key Note Speech. Speech presented at Book Launch: The 1.5 Billion People Question: Food, Vouchers, or Cash Trans- fers? in Conrad Hotel, Cairo. Rosenberg, M. (1965). Rosenberg self-esteem scale. Journal of Religion and Health. Spitzer, R. L., Kroenke, K., Williams, J. B., & Löwe, B. (2006). A brief measure for assessing generalized anxiety disorder: the GAD-7. Ar- chives of internal medicine, 166(10), 1092-1097. World Bank. (2015). Strengthening Social Safety Net Project. Retrieved from http://documents.worldbank.org/cu- rated/en/944911468023332501/pdf/PAD6110PAD0P14010Box385454B00OUO090.pdf. World Bank. (2019). Understanding Poverty and Inequality in Egypt: Background Papers. Washington DC: The World Bank. World Bank. (2022). “Egypt’s Economic Update – April 2022”. Washington, DC. The Egypt Strategy Support Program (EgSSP) is managed by the International Food Policy Research Institute (IFPRI) and is financially supported by the USAID through the “Evaluating Impact and Building Capacity (EIBC)" project. This publication has been prepared as an output of the EIBC project and has not been independently peer reviewed. Any opinions expressed here belong to the author(s) and are not necessarily representative of or endorsed by IFPRI. INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE A world free of hunger and malnutrition IFPRI is a CGIAR Research Center 1201 Eye Street, NW, Washington, DC 20005 USA | T. +1-202-862-5600 | F. +1-202-862-5606 | Email: ifpri@cgiar.org | www.ifpri.org | www.egypt6ss p.ifpri.info © 2022, copyright remains with the author(s). All rights reserved.