The World Bank Economic Review, 39(2), 2025, 439–472 https://doi.org10.1093/wber/lhae023 Article Food Transfers, Cash Transfers, Behavior Change Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Communication and Child Nutrition: Evidence from Bangladesh Akhter Ahmed, John Hoddinott , and Shalini Roy Abstract This paper reports the results of two 2-year randomized control trials in two poor rural areas of Bangladesh. Treatment arms included monthly cash transfers, monthly food rations of equivalent value to the cash transfers, and mixed monthly cash and food transfers, and treatment arms—one with food and one with cash—that combined transfers with nutrition-behavior communication change (BCC). This design enables a comparison of transfer modalities within the same experiment. Intent-to-treat estimators show that cash transfers and nutrition BCC had a large impact on nutritional status, a 0.25 standard deviation increase in height-for-age z-scores and a 7.8 percentage point decrease in stunting prevalence. No other treatment arm affected anthropometric outcomes. Mechanisms underlying these impacts are explored. Improved diets—particularly increased intake of animal source foods in the cash plus BCC arm—are consistent with the improvements observed in this paper. JEL classification: O10, I38, D13 Keywords: cash transfers, food transfers, behavior change communication, child nutrition, social protection, Bangladesh Akhter Ahmed is a senior research fellow at the International Food Policy Research Institute, Dhaka Bangladesh; his email address is A.Ahmed@cgiar.org. John Hoddinott, corresponding author, works in the Division of Nutritional Sciences, at the Charles H. Dyson School of Applied Economics and Management, the Department of Global Development, Cornell University, Ithaca, NY, USA, and is a nonresident senior fellow at the International Food Policy Research Institute, Washington, DC, USA; his email address is Hoddinott@cornell.edu. Shalini Roy is a senior research fellow at the International Food Policy Research Institute, Washington, DC, USA; her email address is S.Roy@cgiar.org. The study builds on research funded by the German Ministry for Economic Cooperation and Development (BMZ), the Swiss Agency for Development and Cooperation (SDC), the United Nations Development Programme (UNDP), and the United States Agency for International Development (USAID). Additional support came from the CGIAR Initiative on Gender Equality; the CGIAR Research Program on Policies, Institutions, and Markets; and the UK Department for International Development (now FCDO). The authors thank the World Food Programme (WFP) for their partnership, Christa Rader for her support throughout this study, DATA for careful data collection, and Wahid Quabili for superb research assistance. The paper has benefited from comments received from referees, Harold Alderman, Emanuela Galasso and seminar participants at the Cornell University, Ethiopia Public Health Institute IFPRI, Florida International University, NEUDC 2019, Gottingen University, University of Minnesota, University of Pennsylvania, UNICEF, USAID, and WFP (Dhaka). The study received ethical approval from the institutional review board of the International Food Policy Research Institute, Washington, DC. The study was also reviewed in Bangladesh by the Ministry of Food and Disaster Management who issued Letters of Authorization to conduct the surveys. The study was registered with ClinicalTrials.gov (Study ID: NCT02237144) and the American Economic Association social science registry (AEARCTR- 0000247). The authors thank Data Analysis and Technical Assistance (DATA) for careful data collection, Janaki Narayan, Wahid Quabili for excellent research assistance, and seminar participants in Dhaka, Florida International University, the C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 440 Ahmed, Hoddinott, and Roy 1. Introduction Improving the nutritional status of children in early life is of intrinsic and instrumental value. Sustainable Development Goal Target 2.2 (United Nations 2021) to end all forms of malnutrition captures this intrin- sic value. Reducing undernutrition of children in utero and in the first two years of life (the “1,000 days window”) is also of instrumental value. Doing so reduces premature mortality, reduces avoidable illness Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 that impairs quality of life while additionally saving on resources used to treat those illnesses, and averts irreversible neurological damage that leads to cognitive or behavioral deficits (Grantham-McGregor et al. 2007; Galler et al. 2021). Thus, nutritional status during this period of life contributes to longer- term health, cognitive and noncognitive development, schooling, economic productivity, and incomes. In turn, these benefits pass on to subsequent generations, as higher levels of household income and maternal schooling are associated with reductions in child undernutrition (Black et al. 2013; Hoddinott et al. 2013; Victora et al. 2021). While global improvements in child nutrition have occurred over the last 20 years, undernutrition remains widespread. UNICEF, WHO, and World Bank (2023) estimate that in 2022,148 million chil- dren under the age of five were stunted (that is, had a height-for-age z-score less than −2, a commonly- used marker for chronic undernutrition), and 45 million children were estimated to be wasted (that is, had a weight-for-height z-score less than −2, a commonly-used marker for acute malnutrition). Growing understanding of the importance of early childhood nutrition has thus sparked extensive efforts to de- velop and implement interventions to accelerate reductions in stunting and wasting. Systematic reviews of “nutrition-specific” interventions aimed at reducing different forms of malnutrition (for example, Bhutta et al. 2013 and Keats et al. 2021) show that there are a range of interventions that appear to effectively reduce forms of micronutrient deficiency as well as severe acute malnutrition. However, there are fewer promising approaches to reducing chronic undernutrition, and those that exist are challenging to scale. While linking efforts to other sectors such as social protection could accelerate progress, reviews of social- protection interventions’ effects on child nutritional status meanwhile show mixed results (Manley et al. 2022). Specific knowledge gaps include the roles of different transfer modalities and complementary pro- gramming, due in part to few studies that can make rigorous comparisons across these design variations. This paper assesses the impacts on early childhood nutritional status of the Transfer Modality Research Initiative (TMRI), a novel social protection intervention that was designed as two randomized control tri- als varying transfer modality and complementary programming, in two poor rural areas of Bangladesh. The TMRI was a two-year intervention, with the objective of improving household food security and reducing infant and young child malnutrition. Treatment arms included different transfer modalities: namely, monthly cash transfers, monthly food rations of equivalent value to the cash transfers, and mixed monthly cash and food transfers; TMRI’s treatment arms also varied the inclusion of complementary programming that aimed to improve knowledge and practices around infant and young child nutrition, through behavior change communication (BCC); one arm combined food with BCC, while another com- bined cash with BCC. Thus, TMRI’s design permits a direct comparison of the impact of different transfer modalities and inclusion of nutrition BCC on child nutritional status within the same experiment. Using the experimental design and rich longitudinal data, the paper estimates intent-to-treat impacts of TMRI’s treatment arms on six measures: height-for-age z-score (HAZ), a stunting indicator (HAZ←2), International Food Policy Research Institute (IFPRI), NEUDC 2019, UNICEF New York, and the University of Minnesota for useful comments. Emanuela Galasso and two anonymous reviewers provided immensely helpful comments that significantly strengthened the paper. All errors are the authors’ This work was undertaken as part of the CGIAR Research Program on Policies, Institutions, and Markets (PIM) led by IFPRI. The study builds on research funded by the German Ministry for Economic Cooperation and Development (BMZ), the UK’s Department for International Development (DFID), the Swiss Agency for Development and Cooperation (SDC), the United Nations Development Programme (UNDP), and the United States Agency for International Development (USAID). A supplementary online appendix is available with this article at the World Bank Economic Review website. The World Bank Economic Review 441 weight-for-age z-score (WAZ), an underweight indicator (WAZ←2), weight-for-height z-score (WHZ), and a wasting indicator (WHZ←2). The combination of cash transfers and nutrition BCC (Cash + BCC) causes a large increase in HAZ of 0.25 standard deviations, reduces stunting by 7.8 percentage points, and increases WAZ by 0.16 standard deviations. Further, the effects of Cash + BCC on HAZ and WAZ do not differ significantly between girls and boys. By contrast, treatment arms that provided only cash transfers, Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 only food transfers, a mix of cash and food, or food transfers combined with BCC (Food + BCC) had no effect on HAZ, WAZ, or WHZ. Cash + BCC was more effective at improving child nutrition than cash alone, and that Cash + BCC was more effective than Food + BCC. The paper explores mechanisms at the household level and the child level to understand why these findings emerge: first considering child diet and determinants such as maternal knowledge and household diets; then considering child illness and determinants such as hygiene-related behaviors. Both Cash + BCC and Food + BCC led to large increases in maternal knowledge of infant and young child nutrition (rel- ative to arms without BCC), and effects were similar across Cash + BCC and Food + BCC. Nearly all treatment arms increased household caloric acquisition, but the increase was largest and most diversified toward animal-source foods for Cash + BCC. These large impacts of the Cash + BCC arm on household diets were driven by a combination of (a) increased income—both from the transfer itself and from the fact that Cash + BCC led to investment in new income-generating activities; (b) positive expenditure- caloric elasticities, which mean that this higher income translates into increased caloric acquisition, par- ticularly of animal-source foods; and (c) Cash + BCC shifting household caloric acquisition towards a more diverse diet that includes some quantities of pulses and animal-source foods. Consistent with these findings, Cash + BCC was most effective at improving child diets—particularly in terms of animal-source foods—although all arms, including Food + BCC led to some improvements. Relative to children in other treatment arms, children in the Cash + BCC treatment arm consumed a more diverse diet as measured by diet diversity scores; they were more likely to consume legumes, dairy, flesh foods, eggs, and vitamin A–rich fruits and vegetables; and they had higher intakes of calories, protein, zinc, calcium, and choline. All these results on diets are consistent with the finding that only Cash + BCC resulted in improved child nutritional outcomes, with analysis indicating that both increased knowledge and income play a role. In both the North and the South, treatment arms that included BCC led to improvements in the clean- liness of the home environment and in purchases of hygiene-related goods; for some of these impacts, the differences were statistically significant, but not all. The null hypothesis that Cash + BCC and Food + BCC have equal impacts on hygiene-related behaviors is always rejected, with the larger impact found in the region where baseline prevalences of these behaviors were lowest. Children in the Cash + BCC treatment arm were less likely to have experienced a fever, or a cough or cold, but were not less likely to suffer from diarrhea, but these impacts did not differ from those observed in the Food + BCC treatment arm. These findings represent several important contributions to knowledge around improving child nu- trition. Given that social protection programs are widespread (Gentilini et al. 2022), have been shown to be effective in addressing several determinants of child nutritional status such as poverty and food consumption (Bastagli, Hagen-Zanker, and Sturge 2016; Reynolds et al. 2017; Hidrobo et al. 2018), and are generally well-targeted towards poor households where chronically malnourished children are more likely to live (Victora et al. 2021), they are seen as a promising scalable platform for adding nutrition interventions (Ruel and Alderman 2013). However, as noted above, evidence of the effectiveness of social protection programs in reducing chronic undernutrition has been mixed. Several recent reviews assessing the effectiveness of cash transfer and food transfer programs in improving child nutrition (de Groot et al. 2017, Manley et al. 2020 and 2022, Little et al. 2021, Barnett et al. 2022, Olney et al. 2022) find that effects vary substantially across studies and tend to be small. For example, Olney et al. (2022) docu- ment that, of studies published between 2010 and 2020 that assess impacts of cash transfers on children’s anthropometric status, 3 out of 17 find positive impacts on HAZ, and 5 out of 15 report reductions in stunting; among published studies of in-kind transfers over the same period, 2 out of 15 find positive 442 Ahmed, Hoddinott, and Roy impacts on HAZ, and 5 out of 11 find positive impacts on stunting. Manley et al. (2022) meta-analysis shows that cash transfer programs have, on average, statistically significant impacts, but the magnitude of the effect size is small, an increase in HAZ of 0.02 standard deviations. The present stud y, however, finds that, depending on design, impacts on HAZ and stunting can be substantial. Some reviews suggest that complementary nutrition-sensitive components included in social protection Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 programs may strengthen impacts, but again the evidence is mixed, and few studies can shed light on why the complementary components matter.1 For example, Manley et al. (2022) argue that the content of BCC included with cash transfer programming matters for program effectiveness, specifically that BCC providing instruction on water, sanitation, and hygiene is particularly helpful. On the other hand, meta- analysis by Little et al. (2021) states that “there is no added impact above cash alone in Cash + Nutrition BCC for improving anthropometrics.”2 A complication in interpreting these analyses is that many studies assess the effects of a bundled intervention combining transfers with a nutrition-sensitive component such as BCC, but do not assess the effect of transfers alone, thus making it difficult to disentangle effects.3 The present study, however, can do so, as a result of including arms both with and without BCC. Further, while a handful of studies assess mechanisms for intervention effects on child nutritional status (Field and Maffioli 2021 and Carneiro et al. 2021), many do not, limiting their ability to explain why bundling transfers with BCC do or do not improve child nutritional status. The rich data set underlying the present study, however, makes it possible to explore both diet and health-related mechanisms. Lastly, while the existing evidence reviews attempt to assess which transfer modalities (e.g., cash, in- kind, or some combination) are more effective in improving child nutrition, they must rely on looking across different studies across varying settings, study populations, and implementation features.4 For ex- ample, Olney et al. (2022) look at associations between whether programs provided cash, in-kind, or vouchers and their impacts on child nutrition, but many other features of these programs also vary. Little et al. (2021) do a similar assessment of cash transfers relative to interventions that combine cash and food transfers. Although there are several studies that, within the same experimental design, compare the impacts of cash, in-kind (and in some cases vouchers) on household-level measures of food and nu- trient acquisition—for example, Cunha (2014), Hidrobo et al. (2014), Lusk and Weaver (2017), and MacPherson and Sterck (2021)—these do not include measures of child nutritional status. To the best of the authors’ knowledge, this study is the first to directly compare the impacts on child nutrition of mul- tiple modalities (such as cash transfers and food transfers) within the same setting, for comparable study populations, for equivalent transfer value, and with comparable implementation duration and frequency. 1 These nutrition-focused components are often included within conditional cash transfer programs, for example through requirements that children receive basic health services and/or caregivers attend meetings where optimal health and nutrition practices are discussed (Fiszbein and Schady 2009). 2 Levere, Acharya, and Bharadwaj (forthcoming) do not find positive impacts of cash plus BCC activities in Nepal, nor do Khan et al. (2019) in Pakistan. 3 For example, Carneiro et al. (2021) find positive impacts on child anthropometry from a combination of cash transfers and BCC in northern Nigeria, but do not test an arm that provides cash transfers only. An exception is Field and Maffioli (2021), who find positive impacts on child HAZ from adding nutrition BCC over and above cash transfers in Myanmar. In addition to these papers, it should be noted that Fitzsimons et al. (2016) find that the provision of BCC alone improved anthropometric outcomes in Malawi. 4 Bastagli, Hagen-Znaker, and Sturge (2016) report that approximately 130 low- and middle-income countries have at least one cash transfer program. However, Alderman, Gentilini and Yemtsov (2017) note that the use of in-kind pay- ments such as food in social programs remains widespread. Alderman, Gentilini and Yemtsov (2017) and Hirvonen and Hoddinott (2021) provide perspectives on the theoretical merits of both types of modalities. The World Bank Economic Review 443 2. Intervention and Sample Design, Data, and Methods The Transfer Modality Research Initiative The Transfer Modality Research Initiative study operated for 24 months, May 2012 to April 2014. It consisted of two randomized control trials implemented in two regions of Bangladesh: Rangpur division (province) in the northwest (hereafter “North”) and Barisal and Khulna divisions in the south (hereafter Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 “South”). At the time of the study, poverty and food insecurity rates were, on average, higher in Rangpur than elsewhere in Bangladesh; at baseline 77 percent of households in the North sample lived under the $1.90 poverty line in 2012 as did 61 percent in the South. Ahmed et al. (2024) provides additional details on the localities where the study took place. In the RCT fielded in the North, there were four treatment arms in addition to the control group: a cash transfer; a food ration; a half cash payment and half food ration; and a cash transfer plus nutrition BCC. In the South, three of the same treatment arms—Cash, Food, and the Cash & Food mix—and a control group were also implemented. However, the fourth treatment arm in the South was different: a monthly food ration (rather than cash) plus the same nutrition BCC that was implemented in the North. Across all arms, the target beneficiary was the mother of an “index child” aged 0–24 months in March 2012, residing in a poor rural household.5 The cash treatment (“Cash”) consisted of a monthly payment of 1,500 taka (approximately $19) per household.6 Payments were made to mothers via mobile money.7 The amount was about 25 percent of the average monthly household consumption expenditures of poor rural households in Bangladesh as of 2012. The food treatment arm (“Food”) consisted of a monthly food ration of 30 kilograms (kg) of rice, 2 kg of mosoor pulse (a type of lentil), and 2 liters of micronutrient-fortified cooking oil. Several considerations guided the choice of these commodities. First, they were all foods that were familiar to beneficiaries. Second, they collectively provided energy (through the rice and cooking oil), protein (via the pulses), and fats (the cooking oil). Third, they could all be obtained via existing World Food Programme procurement channels. Fourth, for logistical and budget reasons, as food transfers could only be made once a month, and virtually none of the beneficiaries had access to refrigeration, it was necessary to ensure that the foods provided were shelf-stable (that is, nonperishable). The quantities were chosen so that—based on prevailing prices at the time the intervention commenced—the value of the food ration was equal to the value of the cash provided in treatment arms that provided cash. The treatment arm combining cash and food transfers (“Cash & Food”) provided half of each of the above transfers monthly—that is, 750 taka, 15 kg of rice, 1 kg of mosoor pulse, and 1 liter of micronutrient-fortified cooking oil. The BCC component consisted of a suite of activities led by a Community Nutrition Worker (CNW). The CNWs, all women, were from local communities and came from the same villages as TMRI partici- pants.8 The core BCC activity was a weekly, one-hour group session in each village with a CNW. The BCC 5 Poverty was defined as having consumption below the lower poverty line in Bangladesh. 6 The payment amount was chosen to be approximately equivalent to the midpoint between transfer levels of two large government social safety net programs: the Vulnerable Group Development Program and the Rural Maintenance Pro- gram (Ahmed et al. 2010). At baseline, a 1500 taka transfer was equal to a 306 taka transfer per person per month in the north (mean household size in the north was 4.9 persons) and a 277 taka transfer per person per month in the South (mean household size in the north was 5.4 persons). Mean monthly per capita food expenditure in the North was 794 taka; in the South, it was 948 taka. So if the household were to spend their entire transfer on food, it would (at baseline) increase food expenditures by 38.5 percent in the North and 29.2 percent in the South. 7 In order to facilitate payments to cash recipients and maintain comparability across arms, a basic mobile phone was provided to the target mother in all treatment and control groups. 8 On average, they were 25 years old and had completed secondary school. They were recruited, trained, and supervised, by a Bangladeshi NGO, EDSO. 444 Ahmed, Hoddinott, and Roy session materials were derived from material developed for Alive & Thrive (A&T) in Bangladesh, a com- prehensive program aimed at improving breastfeeding and complementary feeding practices and reducing stunting among young children.9 This curriculum is widely used throughout Bangladesh (Nguyen et al. 2014, Hoddinott et al. 2016) and follows World Health Organization and UNICEF guidelines for infant and young child nutrition (IYCN). TMRI also followed A&T in terms of the content of the BCC sessions Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 and an approach that included community engagement, group BCC sessions, and home visits (Haider et al. 2010; Baker et al. 2013). Because many factors underlying the IYCN recommendations—including local food availability and seasonality—were comparable in the North and South of Bangladesh, TMRI used the same approach to the BCC across the two regions. These sessions covered the following topics: (1) the importance of nutrition and diet diversity for health; (2) how hand washing and hygiene improve health; (3) diet diversity and micronutrients; (4) breastfeed- ing; (5) complementary foods for children 6–24 months; and (6) maternal nutrition. A variety of methods were used to deliver this information, including presentations, question and answer, interactive call-and- answer songs and chants, practical demonstrations, and role-playing. One of these sessions per month, with only beneficiaries participating, occurred on the day of the transfer distribution. For the remain- ing group BCC trainings each month, other household members—particularly mothers-in-law, husbands, and other pregnant or lactating women—were invited to attend along with beneficiaries, with the inten- tion of creating a supportive household atmosphere and behavior change at the household level. These combined sessions facilitated women’s ability to participate in the BCC, as household members could see what women were participating in and reduce restrictions on attendance, and to increase uptake of BCC messages as husbands and mothers-in-law are also key decision makers on food purchases, IYCF, and child-rearing in the household. The CNWs also made home visits to beneficiaries twice a month to follow up on topics discussed during the group sessions and to discuss specific concerns that mothers might have. While attendance at these BCC sessions was a condition for receipt of transfers, this was a “soft” condition. When a mother missed a session, CNWs followed up with a home visit to ascertain why the session had been missed, and there were no cases where a beneficiary was dropped from the study for failing to attend sessions. In addition, CNWs and project staff conducted community meetings and met with influential members (village leaders, imams, elders) of the villages in which the BCC took place to explain the purposes of the nutrition training and to provide them with the information being conveyed to study participants. CNWs received training prior to the start of the intervention, with refresher training undertaken 3 and 12 months after the intervention began. In localities where the same payment point was used for both the Cash arm and the Cash + BCC arm, Cash beneficiaries were paid in the morning while Cash + BCC beneficiaries were paid in the afternoon, to minimize the likelihood of information from the BCC activities spilling over to the Cash treatment arm. The BCC activities cost approximately $50 per year per beneficiary (Ahmed et al. 2016). Thus, in both the North and the South, the RCTs were designed to ensure that across many dimensions, treatments were identical. All arms were identical in terms of the value of the payments (1500 taka), the identity of the recipients (mothers of children under age 2), the duration (24 months), frequency (monthly) and timing (second week of each month) as well as the receipt of a basic mobile phone. They differed only in terms of the transfer modality (Cash, Food, or a Cash & Food combination) and whether the beneficiaries received nutrition BCC. Quantitative and qualitative data collected throughout the intervention indicate that implementation fidelity was high (Ahmed et al. 2016). Survey data and WFP records indicate that beneficiaries were paid 9 A&T drew on a variety of methods to design the program, including several behavior-change theories, quantitative and qualitative formative research, trials of improved practice, previous studies in other countries, assessments of media habits, and stakeholder consultations (Baker et al. 2013). These included: the theory of “reasoned action”; models focused on interpersonal interactions, self-efficacy, and learning from role models; and community models emphasizing the diffusion of information through social networks. The World Bank Economic Review 445 in full, with transfers provided in a timely fashion. Pay points were easily accessible with median one- way travel time of about 30 minutes. Payments were made efficiently with the median wait time at the pay point being approximately 30 minutes. Few respondents (< 5 percent) reported problems with using mobile phones for transfers. Among beneficiaries receiving food transfers, it was rare (∼2 percent) that any of this food was sold. The nutrition BCC component was well implemented. Knowledge of CNWs Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 was high; in a 14-question test administered at endline to CNWs on key nutrition messages in the BCC curriculum, the mean score out of 14 was 13.2 in the North and 13.5 in the South. Beneficiaries assigned to a BCC intervention attended on average 48 of the scheduled 52 sessions per year in the North and 49 of the scheduled 52 sessions per year in the South. Sample The TMRI sample consists of 4,992 households at baseline, 2,498 in the North and 2,494 in the South. In each region, 10 households were selected in 250 different villages (clusters) with 50 clusters allocated to each treatment and control arm.10 Participating households lived in districts (Upazilas) with high levels of poverty relative to the region in which they were located. Within these villages, selected households were estimated to have consumption below Bangladesh’s lower poverty line, had at least one child aged 0–24 months when the intervention began, and were not receiving benefits from other safety net interven- tions. Ahmed et al. (2024) and the supplementary online appendix provide further details on how these households were selected, the survey instruments that were fielded, and levels of household attrition. The sample for assessing the impacts of TMRI on child nutritional status—as measured by anthropometry—was informed by the evidence on potential for impact, combined with the specifics of TMRI’s targeting. Global evidence shows that the “first 1,000 days” of life, from conception to age 24 months, is a crucial “window of opportunity” during which improving children’s nutritional status—for which linear growth, that is, height-for-age, is a commonly-used marker—has lasting benefits throughout life (Victora et al. 2010; Black et al. 2013). Several features of the TMRI study shaped which children were exposed to the program and measured during this window: (1) the intervention was designed around providing resources to the same households for a two-year period; (2) no new households were added to the beneficiary list after the intervention began; (3) the presence of a child aged 0–24 months in the selected household at baseline was a precondition for participation in the intervention, but there was no requirement that the transfers be used only for this child; and (4) in all survey rounds, anthropometric measurements were collected for all children less than 60 months of age who were present in the household at the time of interview. Given the timeline of the TMRI intervention, the sample of children with any exposure to TMRI during the window of opportunity are those aged 0–48 months at endline. This sample includes children aged 0–24 months at baseline who, by endline, had been exposed to the intervention for varying lengths of time within the “1,000 days,” as well as the small number of children (495) born during the two-year intervention. Although there was also anthropometric measurement at endline of children 25–36 months at baseline, these children are not in the estimation sample, as they were not exposed to the intervention during the “1,000 days.” Thus, the sample includes children who were exposed to the TMRI intervention in utero and/or after they were born. Note that while the intervention lasted for 24 months, children in the sample were not necessarily exposed during the 1,000-day window of opportunity in its entirety. Instead, their duration of exposure varies based on how old they were when the intervention began; for example, a child who was six months old when the TMRI intervention began was exposed for only 18 months of the 1,000-day window. The results, therefore, reflect an averaging of impacts over all children who had different durations of exposure during their first 1,000 days of life. 10 Three households were not interviewed because, on religious grounds, they had changed their minds about being in- cluded in the study, having previously agreed to be included. It is not known why the remaining five were not interviewed. The supplementary online appendix provides additional details on sample size calculations, selection of treatment units, survey timeline and attrition at the household level. 446 Ahmed, Hoddinott, and Roy Lastly, the sample is restricted sample to biological children of the household head. This leads to an estimation sample of 4,399 children–2,218 in the North and 2,181 in the South. Outcome Variables Three sets of anthropometric measures are used as outcome variables. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 The first includes HAZ.11 HAZ is calculated using the WHO child growth standards (WHO 2006). For HAZ, a value of −1 indicates that, given sex and age, a child’s height is one standard deviation below the median child in her age/sex group reference group. HAZ is a measure of chronic undernutrition. It can be thought of as a summary indicator of many factors that influence growth and development during the first 1,000 days of life, from conception to age two (Leroy and Frongillo 2019). Linear growth is itself causally linked to difficult childbirth and poor birth outcomes for women. However, because many determinants of linear growth retardation—such as suboptimal nutrition, inadequate care, and repeated infections— are also determinants of other functionally important outcomes, such as poor cognition, linear growth is additionally a key predictor of these outcomes. In other words, although improved linear growth does not lead to improved cognition per se, it is an outcome that is easily measured in the field that can predict improved cognition, and thus also predict improved school achievement and progress, increased earnings, and reduced probability of living in poverty in adulthood (Grantham-McGregor et al. 2007; Hoddinott et al. 2013). Given that the adverse consequences of chronic undernutrition worsen as the severity of chronic undernutrition rises (see, for example, Alderman, Hoddinott, and Kinsey 2006), this paper also considers impacts on stunting, which equals 1 if the child has a HAZ less than −2. A second set of outcome variables assesses whether the TMRI treatment arms impacted weight, WAZ, and whether the child was underweight, having a WAZ of less than −2. Because weight can reflect the fact that the child is taller (ceteris paribus, taller children will be heavier), or the impacts of very recent nutrient intakes and/or illness, the study also considers a third set of measures: WHZ, and whether the child is wasted (WHZ < −2). Model Specification Intent-to-treat impacts of each of TMRI’s treatment arms using a single-difference specification are es- timated. Estimations are run separately for the North and for the South. The base model in the North is: yend,iv = βCash × Cashv + βFood × Foodv + βCash & Food × Cash & Foodv + βCash,+BCC × Cash + BCCv + εiv In the South, it is: yend, iv = βCash × Cashv + βFood × Foodv + βCash & Food × Cash & Foodv + βFood,+BCC × Food + BCCv + εiv where yend, iv is the endline outcome for child i, living in village v; Cashv, Foodv, Cash&Foodv, Cash + BCCv and Food + BCCv are the treatment arms described above, the β ’s are coefficient estimates of treatment impact, and ε ihv is an unobservable term. Standard errors are clustered at the unit of randomization, the village. Some outcomes are continuous variables, others are dichotomous. In the case of the latter, linear probability models are estimated; robustness is assessed, in part, by also estimating these as probits. To further assess robustness, extended models are also estimated. These control for child characteristics (age, sex) and maternal characteristics (log age, log height, and dummy variables for completing 1–4 grades of schooling and 5–12 grades of schooling, with no schooling being the omitted category12 ). This extended model also includes union fixed effects—unions being the administrative unit above the (village) unit of 11 For children > 24 months, heights were recorded with children standing; for children < 24 months, recumbent length was measured. Heights were recorded to one decimal place. For simplicity, measures of height and length are referred to as height. 12 No mother in the sample had more than 12 grades of schooling. The World Bank Economic Review 447 randomization. Maternal age and education can be thought of as proxies for knowledge of good care practices, the union fixed effects capture locality wages, prices, and other characteristics (e.g., presence of infectious diseases, sanitation that might affect nutritional status), while child sex and maternal height capture, in part, a child’s genetic endowments. Finally, as an additional robustness check, results are run as an ANCOVA with baseline anthropometric Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 status included as a covariate. Doing so requires restricting the sample to children who were observed at baseline, thus excluding children born after the intervention began. Basic Descriptive Statistics Table 1 provides baseline descriptive statistics, means, and standard deviations for outcome and control variables in the estimation sample, disaggregated by region and by treatment arm. At baseline, children had poor nutritional status relative to the WHO standards for a well-nourished population. The nationally representative 2011 Bangladesh Demographic and Health Survey indicated that the mean HAZ in rural Bangladesh was −1.67; the lower mean HAZ reported in table 1 is consistent with these study localities being relatively poorer when compared to other regions. A mean child age of 13 months reflects the sampling strategy, and the sample is approximately equally divided between girls and boys. Mean maternal schooling levels are low: 2.7 grades in the North and 3.8 grades in the South. Mean maternal age is 27 years. Outcome and control variables are similar across the North and South and similar across treatment arms.13 3. Impacts on Anthropometric Indicators Basic Results Table 2 provides basic results for the North. Only Cash + BCC has statistically significant impacts on any anthropometric outcomes. No other treatment arm has a statistically significant impact on HAZ.14 The Cash + BCC treatment increased HAZ scores by 0.25 standard deviations and weight-for-age z scores by 0.16 standard deviations.15 It reduced the percentage of children who were stunted by 7.8 percentage points. To put this number in perspective, during the period that the TMRI intervention took place, stunting in Bangladesh was falling at the rate of 1.1 percentage points per year (Headey et al. 2015). Annualizing the impact of the Cash + BCC treatment makes it equivalent to a 3.4 percentage point reduction per year—three times the rate of the national reduction, which itself was amongst the highest in the world at that time (Headey et al. 2015). For the outcomes for which statistically significant impacts of Cash + BCC are observed, the null hypothesis that the impacts of Cash and Cash + BCC are equal are rejected. Table 3 provides basic results for the South. In contrast to the North, no treatment arms have a sta- tistically significant effect on any anthropometric outcome. In nearly all cases the point estimates are small. 13 McKenzie (2017) notes that balancing tests on baseline data are not necessary in randomized trials unless, for example, there is a concern that randomization was not correctly undertaken, which does not apply here. Nonetheless, the study constructed omnibus tests of joint orthogonality across the child and maternal characteristics described in table 1. Because there are multiple treatment arms, the analysis uses a multinomial specification and calculates chi squared statistics. These yield p-values of 0.16 in the North and 0.37 in the South, indicating that the sample is balanced. 14 The small, not statistically significant, point estimate for the cash transfers on HAZ, 0.035, is consistent with the findings of Manley, Alderman, and Gentilini (2022); in their meta-analysis of cash transfers, the mean impact of the cash transfer programs they consider on HAZ is 0.024SD. 15 Following Leroy et al. (2014), the analysis also estimated impacts on height-for-age deviations (HAD). It is found that Cash + BCC (and no other treatment arm) reduced this by 0.94cm, a reduction (relative to the control group at baseline) of 12.6 percent. 448 Ahmed, Hoddinott, and Roy Table 1. Baseline Child and Maternal Characteristics, by Region and Treatment Arm Child Mother Age Age Schooling Height Baseline sample % female (months) (years) (grades) (cm) size Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 NORTH Treatment: Cash only Mean 45.6 13.2 27.0 2.7 149.3 404 SD 49.9 6.8 5.8 3.1 5.4 Treatment: Food only Mean 46.5 13.4 27.2 2.6 149.8 410 SD 49.9 6.1 5.9 3.0 5.4 Treatment: Cash & Food Mean 46.9 13.5 27.5 2.4 149.4 404 SD 50.0 6.6 5.9 3.0 5.6 Treatment: Cash + BCC Mean 50.4 13.2 27.5 2.6 149.8 406 SD 50.1 6.5 5.9 3.0 5.3 Control Mean 46.8 13.3 27.1 2.9 149.4 386 SD 50.0 6.1 5.7 3.2 5.3 All Mean 47.2 13.3 27.3 2.7 149.6 2,010 SD 49.9 6.4 5.8 3.1 5.4 SOUTH Treatment: Cash only Mean 55.2 13.7 27.8 3.1 150.7 437 SD 49.8 6.0 5.9 3.2 5.3 Treatment: Food only Mean 47.4 12.5 27.6 3.1 150.5 455 SD 50.0 6.5 6.0 3.0 5.5 Treatment: Cash & Food Mean 49.0 13.3 27.0 3.4 150.9 408 SD 50.1 6.2 5.7 3.0 5.6 Treatment: Food + BCC Mean 47.2 13.5 26.8 3.5 150.5 427 SD 50.0 6.6 5.6 3.0 5.3 Control Mean 48.7 13.2 27.4 3.8 150.7 425 SD 50.0 6.2 5.7 3.2 5.5 All Mean 49.5 13.3 27.3 3.4 150.7 2,152 SD 50.0 6.3 5.8 3.1 5.4 Height-for-age Stunting Weight- Weight- Wasting Z-score for-age Underweight for-height Z-score Z-score NORTH Treatment: Cash only Mean −1.90 0.48 −1.59 0.33 −0.68 0.11 SD 1.54 0.50 1.13 0.47 1.24 0.31 Treatment: Food only Mean −1.89 0.47 −1.55 0.33 −0.69 0.12 SD 1.52 0.50 1.09 0.47 1.19 0.32 Treatment: Cash & Food Mean −1.79 0.46 −1.63 0.35 −0.88 0.16 SD 1.36 0.50 1.00 0.48 1.16 0.37 Treatment: Cash + BCC Mean −1.59 0.40 −1.49 0.32 −0.82 0.14 SD 1.39 0.49 1.10 0.47 1.24 0.35 Control Mean −1.84 0.48 −1.60 0.35 −0.81 0.14 SD 1.37 0.50 1.11 0.48 1.22 0.35 All Mean −1.80 0.46 −1.57 0.33 −0.77 0.13 SD 1.44 0.50 1.09 0.47 1.21 0.34 SOUTH Treatment: Cash only Mean −1.68 0.41 −1.57 0.34 −0.92 0.15 SD 1.45 0.49 1.10 0.47 1.10 0.36 Treatment: Food only Mean −1.59 0.42 −1.51 0.34 −0.85 0.15 Treatment: Cash & Food Mean 1.65 0.49 1.20 0.47 1.26 0.36 SD −1.70 0.42 −1.54 0.34 −0.85 0.15 Treatment: Food + BCC Mean 1.45 0.49 1.08 0.47 1.23 0.36 SD −1.70 0.41 −1.52 0.32 −0.82 0.15 The World Bank Economic Review 449 Table 1. Continued Height-for-age Stunting Weight- Weight- Wasting Z-score for-age Underweight for-height Z-score Z-score SOUTH Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Control Mean 1.42 0.49 1.12 0.47 1.22 0.36 SD −1.64 0.38 −1.52 0.32 −0.86 0.17 All Mean 1.46 0.49 1.09 0.47 1.28 0.38 SD −1.66 0.41 −1.53 0.33 −0.86 0.15 Source: Authors’ calculations based on data from the TMRI study. Note: Sample includes all children who are offspring of the household head and who were 0–48 months at endline when anthropometric data were collected. Table 2. Impact on Anthropometric Outcomes, North (1) (2) (3) (4) (5) (6) Height-for-age Weight-for- Weight-for- Z-score Stunting age Z-score Underweight height Z-score Wasting Cash only 0.035 −0.008 0.019 0.034 −0.013 0.003 (0.08) (0.04) (0.07) (0.04) (0.07) (0.03) Food only 0.048 −0.031 0.087 −0.009 0.090 −0.035 (0.08) (0.03) (0.06) (0.03) (0.06) (0.02) Cash & Food 0.119 −0.039 0.029 0.016 −0.041 −0.001 (0.08) (0.03) (0.07) (0.04) (0.07) (0.03) Cash + BCC 0.248∗∗∗ −0.078∗∗ 0.162∗∗ −0.038 0.022 −0.015 (0.08) (0.03) (0.07) (0.04) (0.06) (0.02) R−squared 0.007 0.003 0.004 0.002 0.002 0.002 Mean: Control group −2.03 0.51 −1.89 0.43 −1.05 0.15 p-values: Cash = Food 0.87 0.52 0.28 0.17 0.13 0.12 Cash = Cash + BCC <0.01 0.06 0.04 0.05 0.65 0.50 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,218. OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline when anthropometric data were collected. Robustness Robustness is assessed in several ways. First, for the dichotomous outcomes of stunting, underweight, and wasting, models are re-estimated models as probits and marginal effects calculated. Results are shown in table S2.1. These show impacts that are nearly identical to those reported in tables 2 and 3. Next, the robustness of the results for HAZ, stunting, and WAZ to the inclusion of additional con- trols. Results are shown in table S2.2 (North) and table S2.3 (South). For each outcome, results from four specifications are reported: (1) a base specification found in columns 1, 5, and 9 that replicate the results reported in tables 2 and 3; (2) a second specification (columns 2, 6, and 10), where child-level con- trols (dummy variables for age-in-months, sex) and maternal controls (log age, log height, and dummy variables for completing 1–4 grades of schooling and 5–12 grades of schooling) are added; (3) a third specification (columns 3, 7, and 11) where controls for union fixed effects are also included; and (4) a fourth specification using an ANCOVA specification with baseline anthropometric status as a covariate and the sample restricted to children observed at baseline. In the North (table S2.2), adding in additional controls has no meaningful effect on the Cash + BCC treatment impacts on HAZ, stunting, or WAZ. In the South (table S2.3), adding controls has no effect on the parameter estimates for the Food + BCC treatment; these remain nonsignificant (and generally 450 Ahmed, Hoddinott, and Roy Table 3. Impact on Anthropometric Outcomes, South (1) (2) (3) (4) (5) (6) Height-for-age Weight-for-age Weight-for-height Z-score Stunting Z-score Underweight Z-score Wasting −0.097 −0.124 −0.088 Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Cash only 0.006 0.067 0.017 (0.08) (0.03) (0.08) (0.04) (0.08) (0.02) Food only −0.100 0.010 −0.090 0.037 −0.044 −0.005 (0.09) (0.04) (0.08) (0.04) (0.08) (0.03) Cash & Food 0.024 −0.052 0.001 0.020 −0.017 0.004 (0.08) (0.04) (0.07) (0.04) (0.08) (0.03) Food + BCC 0.079 −0.053 0.016 0.002 −0.042 −0.007 (0.08) (0.03) (0.07) (0.03) (0.08) (0.03) R-squared 0.004 0.003 0.003 0.003 0.001 0.001 Mean: Control group −1.93 0.48 −1.70 0.37 −0.87 0.14 p-values: Cash = Food 0.96 0.90 0.64 0.40 0.57 0.37 Food = Food + BCC 0.03 0.06 0.14 0.32 0.98 0.91 Cash + BCC = Food + BCC 0.18 0.62 0.10 0.38 0.51 0.81 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,181. OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline when anthropometric data were collected. small in magnitude) for HAZ, stunting, and WAZ. In both the North and the South, these additional specifications provide no evidence of positive impacts of Cash only or Food only payments on child anthropometric status. Note that the Cash & Food treatment impacts are small and nonsignificant across most specifications and outcomes, with two exceptions: in the North, Cash & Food has a significant effect on HAZ (table S2.2, columns 3 and 4), but not stunting; and in the South, Cash & Food has a significant effect on stunting (table S2.3, column 8) but not HAZ. Lastly, the Romano and Wolf (2005) approach to control for the familywise error rate (FWER) is applied and adjusted p-values calculated. Across HAZ, WAZ, and WHZ, applying the Romano-Wolf method with 1,000 bootstrap replications does not alter the patterns of statistical significance reported in tables 2 and 3. Disaggregations Impacts of these treatments were disaggregated by child and maternal characteristics. Using the same specification as in tables 2 and 3, figs. S2.1 and S2.2 show impacts in the North and South respectively, for only the three outcomes where significant impacts were found in aggregate: HAZ, stunting, and WAZ. Each figure shows the point estimates and their confidence intervals for girls and boys separately. Because the samples are divided into roughly two halves, not surprisingly the confidence intervals will widen.16 Figure S2.1 shows visually that in the North, Cash + BCC has equal effects on HAZ and WAZ for girls and boys; p-values (available upon request) confirm that the null hypothesis that the impacts on girls and boys are equal for each treatment and anthropometric measure. Other treatments have no statistically significant impacts when the sample is disaggregated by child sex. Figure S2.2 show that in the South, no treatment has an impact on the anthropometric status (HAZ, stunting, WAZ) of either girls or boys.17 16 Heterogeneity was also assessed by interacting treatment status with the variable on which the sample was disaggregated. This produces similar findings (available on request). 17 For the other anthropometric outcomes that were considered, underweight, WHZ, and wasting, there are no statistically significant impacts when they are disaggregated by child sex. Results are available on request. The World Bank Economic Review 451 The next disaggregation considers whether impacts vary by duration of exposure during the period when a child is less than two years of age and/or the age of the child at baseline. Disentangling the ef- fects of these is not straightforward. If impacts are cumulative, then longer exposure should lead to larger effects. But impacts may also depend on when the exposure takes place; for example, if a limiting fac- tor for child growth is adequate nutrient intake, then an exposure of a given duration when the child is Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 older and nutrient requirements are increasing may have a larger effect. For these reasons, the sample is disaggregated into the following subsamples. One consists of children 30 months or older at endline (“older children”). By construction, these children will have been six months or older when the inter- vention commenced, and thus beyond the age when they were supposed to be exclusively breastfed. On average, they have 9.0 months of exposure. The second subsample consists of children younger than 30 months at endline (“younger children”). By construction, these children were either less than six months old when the intervention commenced or were born after it started. They are exposed to the intervention for longer (16.4 months), but this longer period includes an age window (0–6 months) when they should have been exclusively breastfed, and this might dilute the impacts of the intervention. Results for these age and duration disaggregations are found in fig. S2.3 (North) and fig. S2.4 (South). In both the North and the South, there is no evidence of statistically significant differences between impacts on older and younger children for any treatment. Next, the sample is disaggregated by maternal characteristics. First, it is divided into two groups: women with 0–4 grades of formal schooling, and women with more than four grades (the latter corre- sponding to a minimal level of literacy). In tables S2.4 and S2.5 the null that these impacts are equal is not rejected in either the North or South. Second, mothers are disaggregated by age: older versus younger than median maternal age, 26.1 years. Impacts of Cash + BCC on HAZ and WAZ appear larger in magnitude for children of younger mothers, but as with the other disaggregations, the null that they are equal to those found for children of older mothers (table S2.6) is not rejected. In the South (table S2.7), there is no evidence that any treatment has a positive impact on any anthropometric outcomes when disaggregating by maternal age. 4. Mechanisms The results described in section 4 show robustly that the Cash + BCC arm had impacts on anthropo- metric outcomes, most notably those related to chronic malnutrition. However, they do not reveal why Cash + BCC generated these impacts, nor do they explain why they tend to be larger in magnitude and more precisely estimated than other treatments that were of equal monetary value and were equally well- implemented. This section considers potential mechanisms that relate to inputs into the production of child anthropometric status (Behrman and Deolalikar 1988). These are (1) maternal knowledge of op- timal IYCN practices; (2) impacts on household expenditures and child diets; (3) household sanitation and hygiene as captured by housing quality, sanitation, waste management and hygiene practices; and (4) child illness. Impacts on Knowledge of Infant and Young Child Nutrition (IYCN) At baseline and endline, mothers were administered a test of their knowledge around Infant and Young Child Nutrition (IYCN). This test had 23 items: 7 relating to breastfeeding practices; and 16 relating to complementary feeding.18 Correct responses are converted into percent scores. At baseline, women in the 18 The study does not use four questions where, at baseline, nearly all mothers answered these correctly: how to treat a child who has diarrhea; when mothers should wash their hands; whether vegetables should be cooked in oil; and when mothers should wash children’s hands. 452 Ahmed, Hoddinott, and Roy Table 4. Impact on Maternal Knowledge of Infant and Young Child Nutrition (IYCN), by Region (1) (2) (3) (4) (5) (6) IYCN Knowledge Knowledge: Breastfeeding Knowledge: Complementary Foods North South North South North South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Cash only 1.786∗∗ 0.872 −1.657 0.113 3.284∗∗∗ 1.239 (0.89) (0.97) (1.29) (1.23) (1.01) (1.13) Food only 1.976∗∗ 1.445 −0.040 2.569∗∗ 2.874∗∗∗ 1.010 (0.85) (0.95) (1.24) (1.21) (0.97) (1.11) Cash & Food 1.438 1.231 −1.078 0.927 2.497∗∗ 1.362 (0.86) (0.97) (1.25) (1.23) (0.99) (1.13) Cash + BCC 19.414∗∗∗ 11.899∗∗∗ 22.680∗∗∗ (0.88) (1.29) (1.01) Food + BCC 18.456∗∗∗ 13.310∗∗∗ 20.672∗∗∗ (0.98) (1.24) (1.14) Baseline mean: Control group 54.0 52.0 61.3 63.9 48.5 46.8 p-values: Cash = Food 0.83 0.54 0.21 0.05 0.68 0.83 Cash = Cash + BCC <0.01 <0.01 <0.01 Food = Food + BCC <0.01 <0.01 <0.01 Cash + BCC = Food + BCC 0.56 0.46 0.31 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,193 (North) and 2,151 (South). OLS regressions. Outcome variables are expressed as percent scores with maximum value of 100. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes the IYCN knowledge of mothers of all children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. North control group scored 54 percent and women in the South scored 52 percent. Baseline scores were higher for questions relating to breastfeeding than they were for questions about complementary foods. Table 4 shows the impact of all treatment arms on maternal IYCN knowledge, and on the subsets of questions on breastfeeding and on complementary feeding, controlling for child age and sex; maternal age, education, and height; and union fixed effects. In the North, Cash + BCC increases scores on breastfeeding knowledge by 11.8 percentage points, complementary feeding knowledge by 22.6 percentage points, and all IYCN knowledge by 19.4 percentage points. Other treatment arms have either no significant effect, or effects that, while statistically significant, are tiny in magnitude (one to three percentage points). The null hypotheses that Cash, Food, or Cash & Food have impacts that are equal to the impacts of Cash + BCC, are all rejected. Similar results are found in the South. Food + BCC increases scores on breastfeeding knowledge by 13.3 percentage points, complementary feeding knowledge by 20.6 percentage points, and all IYCN knowledge by 18.4 percentage points. In the South, other treatment arms have no effect on IYCN knowledge, and the analysis rejects the null hypotheses that Cash, Food, or Cash & Food have impacts that are equal to the impacts of Food + BCC. Two additional results are noteworthy. First, the final rows of table 4 indicate that the null hypothesis, that the impact of Cash + BCC on IYCN knowledge is equal to the impact of Food + BCC, is rejected. This is not surprising since the content of the BCC material was the same in both the North and the South, and attendance at BCC sessions was similar across both regions. Second, one might be concerned that information obtained through these BCC sessions might spill over to other treatment groups. Although treatment was at the village level, villages were spatially separated, and attempts were made to ensure that participants in the BCC treatments did not mingle with participants receiving other treatments (for example, where several treatment groups shared a common pay point, women in the BCC treatment arms received their payments in the afternoon, while others received their payments in the morning). While it The World Bank Economic Review 453 is not possible to guarantee that there was never any mingling of participants in different treatment arms, the fact that non-BCC treatment arms show little increase in scores on IYCN knowledge at endline is consistent with such spillovers being minimal. Household Food Consumption Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 The survey instrument included questions on household consumption of approximately 200 food items in the 7 days prior to the interview. Quantities of reported consumption for each item are converted into kilocalories per person per day that are available for consumption and grouped into nine categories: rice; pulses; oils; eggs; dairy; meat and fish; fruit and vegetables; other starchy foods; and all other foods. At baseline, caloric acquisition in the control group was low in both regions and especially in the North, at only 1801 kcal per person per day. These diets were heavily oriented towards rice, accounting for 82 percent of caloric acquisition in the North and 79 percent in the South. Consumption of protein and nutrient-dense foods—pulses, eggs, dairy, and meat and fish—was low in both regions: 56 kcal per person per day (or 3.1 percent of calories available for consumption) in the North; and 94 kcal per person per day (or 4.8 percent of calories available for consumption) in the South. Impact estimates are found in table 5 (North) and table 6 (South). Because there are both zero values and relatively large values for some outcome variables, all are IHS-transformed. In the North and the South—taking the reported parameter estimates and following Bellemare and Wichman (2020) calculating their marginal effects—treatment arms that did not include BCC increased caloric acquisition by between 2.0 (Cash, South) and 5.9 percent (Cash, North). Treatments that included BCC had larger impacts: 7.6 percent in the South and 14.8 percent in the North. In the North and the South, the null hypotheses that treatment arms that did not include BCC, compared to those that did, had equal impacts on caloric acquisition are rejected. Second, treatment arms that included pulses (the Food only, Cash & Food, and Food + BCC) led to increased caloric availability from pulses, but so did Cash + BCC in the North. Third, the Cash + BCC treatment had a very large impact on the consumption of eggs (a 180 percent increase), dairy products (a 252 percent increase), and meat and fish (261 percent increase). For these food groups, the null hypothesis that these impacts are equal to those from treatment arms in the North that did not include BCC are also rejected. Fourth, a similar pattern is found in the South; however, the impacts of Food + BCC—153 percent for eggs, 65 percent for dairy, and 116 percent for meat and fish—are smaller than those observed from Cash + BCC in the North. One reason why the impacts on pulses, eggs, dairy, and meat and fish are so large is that, at baseline, many households simply did not consume items found in these food groups. For example, in the North, only 43 percent of households reported having consumed any pulses in the seven days prior to the survey. Comparable figures for eggs were 51 percent, for milk 33 percent, and for meat and fish, 85 percent. (Other food groups were consumed by at least 95 percent of households at least once in the previous seven days.) Given this, for these food groups, impacts at the extensive margin are reported in fig. 1 (North) and fig. 2 (South). In the North, the Cash + BCC treatment arm increases the likelihood that the household consumed pulses, eggs, dairy, and meat and fish in the last seven days by 36, 25, 30, and 13 percentage points respectively. In all cases, the null that these impacts are equal to the Cash treatment arm are rejected. In the South, Food + BCC also increases the likelihood of the consumption of these food groups, but the magnitudes tend to be smaller—17, 26, 13 and 3 percentage points for pulses, eggs, dairy, and meat and fish respectively—than those found in the North. Taken together, tables 5 and 6, and figs. 1 and 2 show that the largest effects on household caloric availability, at both the extensive and intensive margins, are generated by the Cash + BCC treatment arm. But they do not explain why such a pattern exists. These are explored in tables S2.8, and S2.9. in the supplementary online appendix. Table S2.8 looks at the impact of these treatment arms on the value of per capita household monthly consumption. This is the value of food consumed (both purchased and consumed from own production) 454 Table 5. Impact on per Capita Household Caloric Acquisition, Total and by Food Group, North (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Meat and Fruit and Other starch Other All calories Rice Pulses Oils Eggs Dairy fish vegetables foods foods Cash only 0.058∗∗∗ 0.038∗∗ 0.740∗∗∗ 0.139∗∗∗ 0.436∗∗∗ 0.066 0.696∗∗∗ 0.223∗∗∗ −0.074 0.151∗∗ (0.02) (0.02) (0.10) (0.04) (0.10) (0.14) (0.10) (0.04) (0.07) (0.06) Food only 0.055∗∗∗ 0.039∗∗ 1.615∗∗∗ 0.225∗∗∗ 0.162 0.236 0.432∗∗∗ 0.145∗∗∗ −0.115 0.150∗∗ (0.02) (0.02) (0.10) (0.04) (0.10) (0.14) (0.10) (0.04) (0.06) (0.06) Cash & Food 0.035∗∗ 0.026 1.492∗∗∗ 0.123∗∗∗ 0.215∗∗ 0.282∗∗ 0.455∗∗∗ 0.130∗∗∗ −0.107 0.041 (0.02) (0.02) (0.10) (0.04) (0.10) (0.14) (0.10) (0.04) (0.07) (0.06) Cash + BCC 0.138∗∗∗ 0.032 1.814∗∗∗ 0.246∗∗∗ 1.030∗∗∗ 1.261∗∗∗ 1.284∗∗∗ 0.576∗∗∗ 0.177∗∗∗ 0.577∗∗∗ (0.02) (0.02) (0.10) (0.04) (0.10) (0.14) (0.10) (0.04) (0.07) (0.06) Baseline mean: Control group 1801 1475 14 81 4 10 28 56 101 33 Marginal effects Cash only 0.059 0.038 1.096 0.149 0.547 0.069 1.005 0.250 0.071 0.162 Food only 0.056 0.040 4.025 0.252 0.176 0.267 0.541 0.156 0.109 0.161 Cash & Food 0.035 0.026 3.445 0.131 0.239 0.326 0.576 0.138 0.101 0.041 Cash + BCC 0.148 0.032 5.138 0.279 1.802 2.528 2.610 0.779 0.194 0.780 p-values: Cash = Food 0.86 0.93 <0.01 0.03 <0.01 0.24 <0.01 0.05 0.54 0.99 Cash = Cash + BCC <0.01 0.74 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 <0.01 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size: 2,209. Outcome variables are inverse hyperbolic transformed. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes household caloric acquisition of food in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include baseline levels, child age, sex, maternal age, height and education, and union fixed effects. Ahmed, Hoddinott, and Roy Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Table 6. Impact on per GCapita Household Caloric Acquisition, Total and by Food Group, South (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Meat and Fruit and Other starch Other All calories Rice Pulses Oils Eggs Dairy fish vegetables foods foods Cash only 0.020 0.007 0.098 −0.004 0.398∗∗∗ 0.014 0.283∗∗∗ 0.169∗∗∗ −0.202∗∗∗ 0.178∗∗∗ The World Bank Economic Review (0.02) (0.02) (0.10) (0.04) (0.11) (0.14) (0.09) (0.05) (0.07) (0.05) Food only 0.039∗∗ 0.016 0.879∗∗∗ 0.093∗∗ 0.162 0.004 0.150 0.119∗∗∗ −0.046 0.129∗∗ (0.02) (0.02) (0.10) (0.04) (0.11) (0.13) (0.09) (0.04) (0.07) (0.05) Cash & Food 0.037∗∗ 0.017 0.763∗∗∗ 0.082 0.177 0.152 0.225∗∗ 0.167∗∗∗ −0.080 0.145∗∗∗ (0.02) (0.02) (0.10) (0.04) (0.11) (0.13) (0.09) (0.05) (0.07) (0.05) Food + BCC 0.073∗∗∗ −0.004 0.962∗∗∗ 0.010 0.928∗∗∗ 0.505∗∗∗ 0.774∗∗∗ 0.433∗∗∗ 0.081 0.360∗∗∗ (0.02) (0.02) (0.10) (0.04) (0.11) (0.14) (0.09) (0.05) (0.07) (0.05) Baseline mean: Control group 1940 1533 44 108 4 8 38 71 86 48 Marginal effects Cash only 0.020 0.007 0.103 0.004 0.488 0.014 0.327 0.184 0.182 0.194 Food only 0.040 0.016 1.408 0.097 0.176 0.004 0.162 0.127 0.044 0.137 Cash & Food 0.037 0.018 1.145 0.085 0.194 0.164 0.252 0.182 0.076 0.156 Food + BCC 0.076 0.004 1.617 0.010 1.530 0.650 1.169 0.542 0.084 0.433 p-values: Cash = Food 0.26 0.61 <0.01 0.03 0.03 0.94 0.15 0.27 0.03 0.34 Food = Food + BCC 0.04 0.28 0.41 0.06 <0.01 <0.01 <0.01 <0.01 0.08 <0.01 Cash + BCC = Food + BCC 0.01 0.27 <0.01 <0.01 0.56 <0.01 <0.01 0.07 0.36 0.02 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size: 2,175. Outcome variables are inverse hyperbolic transformed. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at ∗ the village level, the unit of randomization. Sample describes household caloric acquisition of food in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include baseline levels, child age, sex, maternal age, height and education, and union fixed effects. 455 Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 456 Ahmed, Hoddinott, and Roy Figure 1. Impact on Any Household Consumption of Selected Food Groups, North Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size: 2,209. Outcome variables equal 1 if household consumed from that food group in the previous seven days. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes household consumption of food in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include baseline levels, child age, sex, maternal age, height and education, and union fixed effects. as well as expenditures on nonfood groups. Controlling for baseline consumption levels and child and maternal characteristics, table S2.8 shows that treatment arms that did not include BCC lead to increases in per capita consumption that ranged from 6.1 (Cash, South) to 15.4 (Cash, North) percent. By contrast, Food + BCC increased per capita consumption by 20.9 percent and Cash + BCC increased per capita consumption by 30.9 percent. The null that Cash + BCC had impacts equal to the non-BCC treatment arms in the North are rejected as are the null that Food + BCC had impacts equal to the non-BCC treatment arms in the South. But these results generate a further question—why, if the treatment arms were equal in value, do some treatments (Food + BCC, and especially Cash + BCC) have much larger impacts on consumption than those without a BCC component? One possibility is that, at any given level of income, the BCC induces higher consumption. A second complementary possibility is that the addition of BCC increased income. A companion paper (Ahmed et al. 2024) assesses the impacts of the different TMRI treatment arms on household-level assets and income. Ahmed et al. (2024) consider five asset categories (livestock, poultry, productive nonanimal assets, consumer durables, and cash in hand). They find that Cash + BCC has a large impact on livestock holdings, as well as on poultry. Impacts of other treatment arms tend to be smaller in magnitude and, especially in the South, not statistically significant. They also argue that ownership of these assets allows Cash + BCC households to generate higher incomes, with much of that income being derived from own farm labor income or nonfarm self-employment income. These findings are consistent The World Bank Economic Review 457 Figure 2. Impact on Any Household Consumption of Selected Food Groups, South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size: 2175. Outcome variables equal 1 if household consumed from that food group in the previous seven days. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes household consumption of food in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include baseline levels, child age, sex, maternal age, height and education, and union fixed effects. with the fact that the Cash + BCC treatment arm has a larger effect on per capita consumption than the other treatment arms. With these results as background, the analysis proceeds to table S2.9. The outcome variables are (IHS transformed) endline per capita daily caloric availability by food group. Regressors include endline per capita consumption (in taka, IHS transformed) and the dummy variables for treatment status.19 The coefficients on per capita consumption indicate how caloric availability, by food group, changes as con- sumption rises. If the TMRI treatments only work through an income effect, then the coefficients on the treatment variables should not be statistically significant when they are controlled for income. By contrast, a positive coefficient indicates that a treatment is shifting the Engel curve upwards; a negative coefficient indicates that a treatment is shifting the Engel curve downwards. There are two key results found in table S2.9. First, in both the North and the South, the marginal effects of income on caloric acquisition are positive and statistically significant, and the magnitudes are as one would expect: smaller for staples such as rice and oil, higher for “luxury” foods such as eggs, dairy, and meat and fish.20 Second, the coefficient on Cash + BCC has a statistically significant effect on rice, pulses, oils, eggs, dairy, and meat and fish. Conditioning on household consumption levels, Cash + BCC 19 The analysis also controls for child and maternal characteristics. 20 Haushofer and Shapiro (2016) find similar effects in Kenya; these are also discussed in Almås, Haushofer, and Shapiro (2019). 458 Ahmed, Hoddinott, and Roy households shift caloric acquisition away from rice (the marginal effect is negative) and towards pulses, oils, and animal source foods—eggs, dairy, and meat and fish. Similar shifts occur in the Food + BCC treatment arm—with caloric acquisition shifting away from rice and towards pulses and animal source foods—but these shifts are smaller than those observed for the Cash + BCC treatment. In the Cash treatment arm in the North, there is also a shift towards pulses, eggs, and meat and fish (but not dairy); Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 however, these are smaller in magnitude than those for Cash + BCC.21 Thus, the large impacts of the Cash + BCC arm on caloric acquisition are driven by a combination of factors: (a) income increases—both from the transfer itself and from the fact that Cash + BCC leads to investment in new income-generating activities; (b) positive expenditure–caloric elasticities mean that this higher income translates into increased caloric acquisition, particularly of animal source foods; and (c) in addition to these income effects, Cash + BCC shifts household caloric acquisition away from rice and towards a more diverse diet that includes some quantities of pulses and animal source foods (Bennett’s Law). Other treatment arms show some of these effects, but because they provide a smaller income boost and/or because, conditional on income, they shift consumption less than what is seen with Cash + BCC, they lead to smaller changes in caloric acquisition. Child Diets The improvements in household caloric availability and in diet quality can explain the differential impacts of treatment arms on child nutritional status only if there are similar patterns when child diets are assessed. Two approaches are taken. The first entails adapting the Infant and Young Child Dietary Diversity Score (IYCDDS) to assess the impact of TMRI on diet quality, specifically micronutrient density (see WHO 2008; Leroy et al. 2015). The IYCDDS is constructed by asking mothers about types of foods consumed by children during the previous day. The TMRI interviews included a module that had been extensively used in Bangladesh (Ahmed et al. 2013), which asks mothers about 18 types of foods consumed by pre- school children.22 These are aggregated into seven food groups: (1) grains, roots, and tubers; (2) legumes and nuts; (3) dairy; (4) flesh foods (meat such as beef, mutton; chicken, duck, pigeon; liver, heart, kidneys; fish); (5) eggs; (6) Vitamin A–rich fruits and vegetables; and (7) other fruits and vegetables. This produces a count variable ranging from 0 to 7. In addition, following WHO (2008) minimum diet diversity is defined as equaling 1 if the child consumed food from four or more different food groups the previous day, 0 otherwise. Impacts on both IYCDDS and minimum diet diversity are estimated using OLS; estimating IYCDDS using a Poisson estimator and estimating minimum diet diversity using a probit produce similar results. Results are shown in table 7. In the North, Cash + BCC increased the number of food groups consumed by 1.1 and the likelihood that a child met minimum diet diversity requirements by 40.6 percentage points. It is the only treatment arm that had statistically significant impacts. In the South, Food, Cash & Food, and Food + BCC all led to increases in the number of food groups consumed and in minimum diet diversity, but these impacts were larger (0.9 food groups and 35.6 percentage points respectively) for Food + BCC. In both the North and South, the null hypotheses that the impacts of a treatment arm that included BCC were equal to a treatment arm that did not include BCC, are rejected. By contrast, the null hypotheses that the impacts of Cash + BCC and Food + BCC are equal for either IYCDDS or minimum diet diversity, are not rejected. Figure 3 shows impacts on each of these food groups, except for grains that are nearly universally consumed by all children in the sample. In the North and South, a treatment arm that contained legumes (Food, Cash & Food, Food + BCC) increased the likelihood that a child consumed legumes the previous 21 The increase in animal source foods among Cash & BCC households may also be linked to the increase in livestock and poultry ownership among these households. 22 These questions were asked about the youngest child in the household who was 6–23 months at baseline and 6–41 months at endline, and so the sample size is slightly smaller than that used in section 3. The World Bank Economic Review 459 Figure 3. Impact on Any Child Consumption of Food Groups, by Food Group, by Region Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,045 (North) and 1,977 (South). OLS regressions. Outcome variables are dichotomous, equaling 1 if child consumed from that food group during the previous day. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. 460 Ahmed, Hoddinott, and Roy Table 7. Child Consumption of Food during the Previous Day, Food Groups, by Region (1) (2) (3) (4) Number of food groups Minimum dietary diversity North South North South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Cash 0.109 0.086 0.005 0.046 (0.09) (0.09) (0.03) (0.03) Food 0.050 0.200∗∗ 0.037 0.096∗∗∗ (0.08) (0.09) (0.03) (0.03) Cash & Food 0.069 0.210∗∗ 0.023 0.095∗∗∗ (0.09) (0.09) (0.03) (0.03) Cash and BCC 1.112∗∗∗ 0.406∗∗∗ (0.09) (0.03) Food and BCC 0.923∗∗∗ 0.356∗∗∗ (0.09) (0.03) Endline mean: Control group 2.31 2.58 0.16 0.23 p-values: Cash = Food 0.50 0.22 0.32 0.15 Cash = Cash + BCC <0.01 <0.01 Food = Food + BCC <0.01 <0.01 Cash + BCC = Food + BCC 0.15 0.31 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,045 (North) and 1,977 (South). OLS regressions. Number of food groups is a count variable with values 1–7. Minimum Diet Diversity is a 0/1 variable that equals 1 if the child consumed from four or more different food groups the previous day. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. day as did the Cash + BCC treatment arm. In both the North and South, the null that the impacts of a treatment arm that included BCC were equal to a treatment arm that did not include BCC is rejected, but the null hypothesis that the impacts of Cash + BCC (a 24.6 percentage point increase) and Food + BCC (a 26.1 percentage point increase) are equal is not rejected. Cash + BCC is the only treatment arm to increase the consumption of dairy products (a 12.9 percentage point increase), and the null that Cash + BCC and Food + BCC had equal impacts on the likelihood that a child consumed dairy products the previous day is rejected. Third, Cash + BCC and Food + BCC had essentially equal impacts on children’s consumption of flesh foods: increases of 18.1 and 15.8 percentage points respectively. In the North, Cash also increased children’s consumption of these foods, but the null that the impact of Cash and Cash + BCC are equal is rejected. Fourth, Cash + BCC has a large impact on the likelihood that a child consumed eggs, increasing this by 33.6 percentage points. For eggs in the North, the null hypothesis that the impacts of a treatment arm that included BCC were equal to a treatment arm that did not include BCC is rejected, and the study rejects the null hypothesis that the impacts of Cash + BCC and Food + BCC (11.7 percentage points) are equal. Lastly, Cash + BCC and Food + BCC increase the likelihood that children consumed both Vitamin A–rich fruits and vegetables as well as other fruits and vegetables by approximately equal amounts (13.9 to 18.1 percentage points). One might be concerned that BCC could lead to social desirability bias affecting the IYCDDS responses—that is, after two years of nutrition training, mothers might respond to questions about child feeding by over-reporting foods commonly discussed during the group training sessions. While this ex- planation cannot be ruled out completely, the fact that there are differences between what mothers in the North described and what mothers in the South described—for example, that mothers receiving BCC in The World Bank Economic Review 461 the South did not report feeding their children dairy products more frequently than those in the control group—despite their receiving identical BCC gives confidence in these results. The food group impacts indicate what kinds of food were consumed but not the quantities of food consumed. These are assessed using 24-hour dietary recall data collected by female enumerators who interviewed mothers about all foods consumed the previous day. Mothers were asked to list the foods, Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 by meal, that were consumed (the household’s “menu”), the ingredients used to prepare these, and their raw and cooked weights. This also accounted for food consumed outside the home (for example, as a meal provided by an employer) and any consumption of leftovers.23 With the aid of props such as spoons, cups, and plates, the enumerator and mother then discussed who was present at each meal, who consumed each menu item, and how much of each item was consumed. Using food composition tables specific to Bangladesh (Shaheen et al. 2013), the study calculates the nutrient intake on the previous day for each household member. Table 8 shows treatment impacts by region on caloric intakes (in kcal) and protein intake (in grams). It also reports impacts on micronutrients that are particularly important for child growth—zinc, calcium, and choline intake (all measured in mg)—for the sample of children aged 6–48 months at endline. In the North, the Cash, Cash & Food, and Food treatment arms increase intake between 49 and 58 kcal/day. By contrast, Cash + BCC increased caloric intake by 218 kcal/day. This impact—equivalent to a 25.4 percent increase relative to the control group—is statistically significant at the 1 percent level; also at the 1 percent level, the null that this impact is equal to the impacts of the other treatment arms is rejected. In the South, the non-BCC treatment arms have largely similar impacts on caloric intake. In the North, all treatment arms increase protein intake, but Cash + BCC has a larger impact—an increase of 8.2 g/day— significant at the 1 percent level. In the South, the non-BCC treatment arms have positive impacts similar in magnitude to what is seen in the North, but these are not always precisely measured. The null that Cash + BCC and Food + BCC have equal impacts on children’s protein intakes is rejected. Table 8 reports impacts on nutrients important for linear growth. In the North, the only treatment arm that has a statistically significant impact is Cash + BCC. It increases zinc intake by 0.54 mg, calcium intake by 27.9 mg, and choline by 106.7 mg; the latter represents an increase of 91.1 percent relative to the control group. In the North, the study rejects the null that the impacts of a treatment arm that included BCC were equal to a treatment arm that did not include BCC. In the South, Food + BCC increases consumption of zinc, by 0.54 mg, but not calcium or choline. The null that Cash + BCC and Food + BCC have equal impacts on calcium intake (at the 6 percent level) and choline intake is rejected. Impacts on Housing Quality, Sanitation, Hygiene, and Hygiene Practices Changes in housing quality, in household cleanliness and in hygiene could also be a route through which treatment arms affect anthropometric outcomes. Infections such as diarrhea and upper respiratory dis- eases can harm growth through several channels. First, energy that could go towards growth is diverted to fight infection. Second, illness might depress a child’s appetite, thus leading to reductions in food intake. Finally, gastro-intestinal infections such as diarrhea may inhibit the body’s ability to absorb nutrients that are consumed. Improvements in housing quality and increased cleanliness of the home environment, as well as increased spending on hygiene-related items all contribute to creating a healthier environment where children are less exposed to infectious diseases. At baseline and endline, data were collected on the quality of housing stock, including information on the number of households who shared the dwelling; the state of repair of the building (a scale that goes from “no visible sign of damage” to “in a very poor state”); whether the walls were made of con- crete, brick, or tin; whether the roof was made of concrete, brick, or tin; and the size of the dwelling (in 23 This information could be subject to reporting bias as well but is challenging to falsify given how it is asked. 462 Table 8. Child Consumption of Food during the Previous Day, Intakes of Calories, rotein, and Choline, by Region (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Calories (kcal) Protein (g) Zinc (mg) Calcium (mg) Choline (mg) North South North South North South North South North South Cash 58.566∗∗ 41.398 1.700∗∗ 1.364 0.099 0.105 2.137 6.880 −19.510 −25.161 (27.36) (28.67) (0.77) (0.86) (0.11) (0.11) (5.89) (8.63) (17.26) (15.63) Food 51.917∗∗ 34.935 1.890∗∗ 1.485 0.132 0.089 1.306 −5.038 7.710 −9.603 (26.29) (27.94) (0.74) (0.84) (0.10) (0.10) (5.66) (8.41) (16.59) (15.24) Cash & Food 49.285 33.848 1.897∗∗ 0.777 0.132 0.044 2.457 −5.459 5.382 1.146 (26.52) (28.62) (0.74) (0.86) (0.10) (0.10) (5.71) (8.61) (16.74) (15.61) Cash and BCC 218.476∗∗∗ 8.164∗∗∗ 0.545∗∗∗ 27.944∗∗∗ 106.748∗∗∗ (27.28) (0.77) (0.11) (5.87) (17.21) Food and BCC 151.237∗∗∗ 4.742∗∗∗ 0.540∗∗∗ 8.744 14.204 (29.08) (0.88) (0.11) (8.75) (15.86) Endline mean: Control group 857 905 20.0 21.9 2.9 2.8 60.9 87.3 117.0 122.4 p-values: Cash = Food 0.81 0.83 0.81 0.89 0.75 0.88 0.89 0.17 0.12 0.32 Cash = Cash + BCC <0.01 <0.01 <0.01 <0.01 <0.01 Food = Food + BCC <0.01 <0.01 <0.01 0.12 0.13 Cash + BCC = Food + BCC 0.14 0.01 0.98 0.06 <0.01 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,148 (North) and 2,109 (South). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. Ahmed, Hoddinott, and Roy Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 The World Bank Economic Review 463 Table 9. Impact on Housing Quality, by Region (1) (5) North South Cash only 0.293∗∗∗ 0.054 Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 (0.08) (0.08) Food only 0.151 0.020 (0.08) (0.08) Cash & Food 0.321∗∗∗ 0.203∗∗ (0.08) (0.08) Cash + BCC 0.338∗∗∗ (0.08) Food + BCC 0.154 (0.08) Baseline mean: Control group −0.44 −0.45 p-values: Cash = Food 0.08 0.68 Cash = Cash + BCC 0.57 Food = Food + BCC 0.11 Cash + BCC = Food + BCC 0.18 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,200(North) and 2,167 (South). OLS regressions. ∗∗ signifi- cant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. squared meters and logged). These data are pooled, and their first principal component calculated. This is a measure of the quality of housing stock. Table 9 shows that in the North, Cash, Cash & Food, and Cash + BCC, led to comparable improve- ments in housing quality, increases of approximately 0.3 Standard Deviations. In the South, only the Cash & Food treatment improved housing. The null hypothesis that the impacts of Cash + BCC and Food + BCC are equal is not rejected. Next, the study considers whether there were changes in sanitation and waste management. The TMRI survey has data on three dichotomous outcomes: whether the household used a latrine (either unsealed pucca latrine, or a sanitary latrine) as opposed to open defecation; whether the household disposed of garbage in a fixed place such as a container as opposed to dropping it anywhere around the compound; and whether, if the household kept poultry, the animals were kept in a coop or were allowed to roam around the compound. The use of protected water sources is not assessed, as at baseline use of these was already high; nor was how cattle were housed studied because at baseline, nearly all households (>80 percent) kept them in sheds overnight. Table 10 shows24 that treatment arms that included BCC led to increased use of latrines (by 23.8 percentage points in the North and 10.0 percentage points in the South) as did Cash transfers in the North. The null hypotheses that the impact of Cash, Food, and Cash & Food had effects equal to the treatment arms that included BCC are rejected in the North and the South as is the null hypothesis that Cash + BCC and Food + BCC had equal effects. Treatment arms that included BCC led to improved 24 Estimated models are ANCOVA for latrine use and waste management with controls including their baseline values, the child and maternal characteristics used in the IYCN regressions, and union fixed effects. The results for poultry do not include baseline values. 464 Ahmed, Hoddinott, and Roy Table 10. Impact on Sanitation and Waste Management, by Region (1) (2) (3) (4) (5) (6) Use a latrine Improved waste management Improved poultry practice North South North South North South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Cash only 0.066∗∗ 0.019 0.006 0.026 −0.061∗∗ 0.004 (0.03) (0.03) (0.03) (0.04) (0.03) (0.03) Food only 0.029 0.025 0.024 0.017 −0.007 −0.038 (0.03) (0.03) (0.03) (0.03) (0.03) (0.03) Cash & Food 0.011 0.003 −0.002 0.062 0.003 −0.003 (0.03) (0.03) (0.03) (0.04) (0.03) (0.03) Cash + BCC 0.238∗∗∗ 0.056∗∗ 0.001 (0.03) (0.03) (0.03) Food + BCC 0.100∗∗∗ 0.133∗∗∗ 0.008 (0.03) (0.04) (0.03) Baseline mean: Control group 0.56 0.61 0.23 0.33 0.13 0.11 Sample size 2,205 2,167 2,202 2,165 1,574 1,706 p-values: Cash = Food 0.28 0.84 0.55 0.80 0.06 0.12 Cash = Cash + BCC <0.01 0.08 0.03 Food = Food + BCC 0.02 <0.01 0.09 Cash + BCC = Food + BCC 0.02 0.15 0.86 Source: Authors’ calculations based on data from the TMRI study. Note: OLS regressions. Outcome variables are dichotomous. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes sanitation and waste management in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. garbage disposal, by 5.6 percentage points in the North and 13.3 percentage points in the South—the null that Cash + BCC and Food + BCC had equal effects on this outcome is not rejected. No treatment has a positive impact on whether poultry were kept in coops. Table 11 shows impacts on expenditures on hygiene-related items: soap, shampoo, toothpaste; laundry soap; dish soap and other cleaners; and mosquito coils and repellants. At baseline, households in the control group spent very little on these, around 90 taka per household per month. Given the mix of zero values and a few households with relatively high expenditures on these items, these are transformed using the Inverse Hyperbolic Sine function (Bellemare and Wichman 2020). The Cash + BCC treatment increased expenditures on all hygiene-related items by 15.4 percent, but because baseline expenditures were so low, the magnitude of these impacts in terms of taka spent is small (less than 10 taka). The Food + BCC treatment increased expenditures on laundry soap by 16.4 percent but had no impact on other expenditures. Other treatments had no, or limited, impacts on purchases of these items.25 Improvements in the cleanliness of the home environment, in the purchase of hygiene products and so forth may have only limited impacts if infants and children do not directly benefit from them. Table 12 reports impacts, by treatment arm and region, on three hygiene-related practices as reported by the mother: whether the child defecates in a latrine (columns 1 and 2); whether the child is bathed using soap and water (columns 3 and 4); and whether the child’s mother washes hands with soap before feeding the child (columns 5 and 6). Both the Cash + BCC and Food + BCC treatments increase use of these hygiene behaviors, with magnitudes ranging from 5.8 percentage points (child is bathed with soap, North) to 39.5 percentage points (mother washes hands with soap before feeding child), and all these impacts are 25 For example, while the coefficient on Cash & Food on expenditures on mosquito coils and repellants in the South looks large, it is equivalent to an increase of approximately one taka in spending on those items. Table 11. Impact on Expenditures on Hygiene−Related Items, by Region All washing and cleaning Dish soap and other Mosquito coils and expenses Soap, shampoo, toothpaste Laundry soap cleaners repellants North South North South North South North South North South (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) The World Bank Economic Review Cash only 0.070 0.058 0.020 0.048 0.060 0.056 −0.069 0.043 0.371∗∗ 0.194 (0.05) (0.05) (0.05) (0.05) (0.05) (0.06) (0.06) (0.05) (0.16) (0.11) Food only 0.021 0.073 0.046 0.028 0.091∗∗ 0.125∗∗ −0.108 0.063 −0.094 0.160 (0.04) (0.04) (0.05) (0.05) (0.04) (0.05) (0.06) (0.05) (0.16) (0.11) Cash & Food 0.027 0.048 0.084 0.067 0.042 0.089 −0.056 −0.019 −0.169 0.228∗∗ (0.04) (0.05) (0.05) (0.05) (0.04) (0.06) (0.06) (0.05) (0.16) (0.11) Cash + BCC 0.143∗∗∗ 0.140∗∗∗ 0.191∗∗∗ 0.052 0.104 (0.05) (0.05) (0.05) (0.06) (0.16) Food + BCC 0.070 0.150∗∗∗ 0.048 0.028 0.173 (0.05) (0.05) (0.06) (0.05) (0.11) Baseline mean: Control group (taka) 93.4 91.4 34.6 42.6 40.0 43.8 1.8 0.6 17.0 4.4 p-values: Cash = Food 0.29 0.74 0.60 0.71 0.50 0.22 0.49 0.70 <0.01 0.76 Cash = Cash + BCC 0.11 0.02 <0.01 0.04 0.11 Food = Food + BCC 0.95 0.02 0.18 0.13 0.91 Cash + BCC = Food + BCC 0.32 0.89 0.06 0.75 0.78 Source: Authors’ calculations based on data from the TMRI study. Note: Samples size: 2,195 (North); 2,157 (South). OLS regressions. Outcome variables are inverse hyperbolic transformed. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes expenditures on hygiene-related items in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. 465 Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 466 Ahmed, Hoddinott, and Roy Table 12. Impacts on Hygiene-Related Behaviors, by Region (1) (2) (3) (4) (5) (6) Child is bathed using soap and Mother washes hands with Child defecates in latrine water soap before feeding child Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 North South North South North South Cash only 0.094∗∗∗ 0.055 0.072∗∗∗ 0.039 0.042 0.080∗∗ (0.04) (0.03) (0.02) (0.03) (0.04) (0.04) Food only 0.067 0.035 −0.001 0.022 0.033 0.076∗∗ (0.03) (0.03) (0.02) (0.03) (0.04) (0.04) Cash & Food 0.038 0.035 −0.001 −0.029 −0.013 0.128∗∗∗ (0.03) (0.03) (0.02) (0.03) (0.04) (0.04) Cash + BCC 0.313∗∗∗ 0.058∗∗∗ 0.214∗∗∗ (0.04) (0.02) (0.04) Food + BCC 0.138∗∗∗ 0.142∗∗∗ 0.395∗∗∗ (0.04) (0.03) (0.04) Baseline mean: Control group 0.30 0.57 0.89 0.72 0.52 0.37 Sample size 2,003 1,927 2,000 1,935 2,003 1,938 p-values: Cash = Food 0.45 0.57 <0.01 0.63 0.80 0.91 Cash = Cash + BCC <0.01 0.53 <0.01 Food = Food + BCC <0.01 <0.01 <0.01 Cash + BCC = Food + BCC <0.01 0.03 <0.01 Source: Authors’ calculations based on data from the TMRI study. Note: Single difference OLS regressions. Outcome variables are dichotomous. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes hygiene practices in households of children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. statistically significant at the 1 percent level. In nearly all cases, the null hypothesis that the impacts of a treatment arm that included BCC were equal to a treatment arm that did not include BCC is rejected. The null that Cash + BCC and Food + BCC have equal impacts is always rejected, with the larger impact always seen in the region where baseline prevalences of these behaviors were lowest. Child Illness In each survey round, mothers were asked whether their children had the following symptoms in the previous two weeks: fever; cough or cold; diarrhea.26 Results are shown in table 13. In the North, columns (1), (3), and (5) show that Cash + BCC reduces reported fever and coughs/colds but not diarrhea. These effect sizes are large, the 11.5 percentage point reduction for fever is equivalent to a 26.1 percent reduction relative to the control group. No other treatment arm affected reported illness. In the South, no treatment arm—including Food + BCC—affected reported illness. In the North, the null hypothesis that the impacts of a treatment arm that included BCC were equal to a treatment arm that did not include BCC is rejected, but the null hypotheses that Cash + BCC and Food + BCC have equal impacts on the three illnesses considered here is not rejected. Note that prevalence of fever and coughs/colds was relatively high in the control groups (36 to 44 percent) but that the prevalence of diarrhea was low (5 and 7 percent in the 26 Mothers were also asked about children who exhibited difficulties with breathing but as virtually no children were reported with this symptom, this outcome was excluded from the analysis. The World Bank Economic Review 467 Table 13. Caregiver Reported Child Illness in the Previous Two Weeks, by Region (1) (2) (3) (4) (5) (6) Fever Cough or cold Diarrhea North South North South North South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Cash 0.024 −0.009 0.039 −0.030 −0.002 −0.009 (0.04) (0.04) (0.04) (0.04) (0.02) (0.02) Food −0.005 0.044 0.008 0.006 −0.006 0.023 (0.04) (0.04) (0.04) (0.04) (0.02) (0.02) Cash & Food −0.009 0.030 −0.002 −0.037 0.003 0.022 (0.04) (0.04) (0.04) (0.04) (0.02) (0.02) Cash and BCC −0.115∗∗∗ −0.089∗∗ −0.025 (0.04) (0.04) (0.02) Food and BCC −0.011 −0.074 −0.014 (0.04) (0.04) (0.02) Endline mean: Control group 0.44 0.41 0.36 0.40 0.05 0.07 p-values: Cash = Food 0.46 0.17 0.41 0.33 0.81 0.12 Cash = Cash + BCC <0.01 <0.01 0.18 Food = Food + BCC 0.16 0.04 0.08 Cash + BCC = Food + BCC 0.06 0.76 0.70 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,004 (North) and 1,939 (South). OLS regressions. All outcome variables are (0/1) dichotomous variables where = 1 means that the child had these symptoms at least once in the previous two weeks. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. Controls include child age, sex, maternal age, height and education, and union fixed effects. North and South, respectively). The endline survey was fielded in April, a time when the weather was colder but also—especially in the North—drier.27 5. Summary and Discussion This paper analyzes two, linked randomized control trials in rural Bangladesh to contribute to the mixed evidence around the effectiveness of social protection interventions on children’s nutritional status. It addresses two major limitations in the literature: disentangling the effects of transfers alone relative to transfers with complementary nutrition BCC, and directly comparing transfer modalities (cash versus food) within the same experimental design and setting. It is done within the context of an intervention that was well-implemented and thus allows speaking to proof-of-concept, and it assesses mechanisms that make it possible to understand why any differences in impact across treatment arms emerge. The combination of cash transfers and nutrition BCC led to a large increase in HAZ scores (0.25 stan- dard deviations), reduced stunting by 7.8 percentage points, and increased WAZ scores by 0.16 standard deviations. The null hypothesis that the impacts on girls and boys are equal is not rejected. There are no differential impacts when the sample is disaggregated by child age, maternal age, or maternal educa- tion. Patterns of impacts are robust to how the outcome variables are specified (continuous or dichoto- mous), the inclusion or exclusion of control variables, and adjustments for multiple hypothesis testing. By 27 A caveat to these findings is that the BCC training may have affected the reporting of illness, either increasing it (because mothers are more aware of these illnesses and their effect on child nutritional status) or reducing it for social desirability reasons. 468 Ahmed, Hoddinott, and Roy contrast, cash or food transfers by themselves have little impact on children’s nutritional status.28 Food transfers combined with BCC had no effect on any anthropometric outcomes. The exploration of mechanisms at the household level and child level provide explanations as to why these findings emerge, specifically: (a) differences between Cash and Cash + BCC in the North (thus mak- ing it possible to focus on the additional impacts of the BCC over and above the impact of cash transfers Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 in the same region; and (b) differences between Cash + BCC and Food + BCC, focusing on comparing two treatment arms that included BCC, but had different transfer modalities and were implemented in different parts of the country. At endline, in the North, mothers knew much more about IYCN if they were exposed to the BCC activ- ities compared to mothers in the Cash only treatment. Cash + BCC households increased their acquisition of calories, particularly those from animal source foods—and were more likely to consume animal source foods—more than households in the Cash only treatment. This occurs because Cash + BCC households experienced a greater increase in their income, income is positively associated with acquisition of animal- source foods, and the BCC treatment led to shifts in the composition of household caloric acquisition over and above the impact that worked through higher incomes. Cash and Cash + BCC had compara- ble impacts on housing quality. However, Cash + BCC led to greater improvements in the cleanliness of the home environment, on purchases of hygiene-related goods (though the magnitude of this impact was small), and on hygiene-related practices. Comparing Cash + BCC with Food + BCC, both had positive and equal impacts on maternal IYCN knowledge. Both increase caloric acquisition, but the increase was larger for Cash + BCC, most notably for certain animal-source foods such as dairy and meat and fish. Both led to improvements in the cleanliness of the home environment and in purchases of hygiene-related goods; for some, but not all, of these impacts, the differences were statistically significant. Relative to children in the Cash treatment arm, children in the Cash + BCC treatment arm consumed a more diverse diet as measured by the IYCDDS and minimum diet diversity. They were more likely to consume legumes, dairy, flesh foods, eggs, and vitamin A–rich fruits and vegetables. They had higher intakes of calories, protein, zinc, calcium, and choline. They were less likely to have experienced a fever, or a cough or cold. Differences between the Cash + BCC and Food + BCC treatment arms are more subtle. Both improve diet diversity as measured by IYCDDS and minimum diet diversity. While Cash + BCC reduces the in- cidence of fever and of coughs and colds, the null hypothesis that Cash + BCC and Food + BCC have equal impacts on these illnesses, or on diarrhea, was not rejected. Both increase caloric intake; the effect is larger in magnitude for Cash + BCC, but the null that these impacts are equal is not rejected. Relative to Food + BCC, Cash + BCC has a larger impact on the likelihood that a child consumed a dairy product, whether the child consumed eggs, protein intake and on the intakes of calcium and choline; differences in these impacts are statistically significant. These impacts on child diet are consistent with what is known about the biology of child growth. Children in this sample require approximately 30.9 grams of protein per day (Parikh et al. 2022). At endline, 33 percent of children in the Cash + BCC treatment group met this requirement, compared to 26 percent of children in the Food + BCC group. Proteins are not homogeneous; they are made up of various amino acids (Parikh et al. 2022)—some of which cannot be synthesized within the human body and must be obtained via diet (WHO and FAO 2007). These are referred to as essential amino acids, and Semba et al. (2016a, 2016b) report that children who are deficient in essential amino acids are more likely to be stunted. Animal source foods are excellent sources of essential amino acids (although essential amino acids can be contained in plant sources, they are typically in much lower concentrations). In results available 28 While certain specifications indicate that Cash & Food had a positive impact on HAZ, it is not clear how much weight should be put on that finding as it is not robust and does not show consistent impacts in the analysis of mechanisms. The World Bank Economic Review 469 on request, the study finds that Cash + BCC increased intakes of all these essential amino acids—the magnitudes of these increases ranged from 31 to 46 percent. Food + BCC also increased the intake of these essential amino acids, but by smaller magnitudes. Second, relative to Food + BCC, Cash + BCC increased the likelihood that children consumed eggs. Eggs are rich in choline, and so this increased consumption is also reflected in higher choline intakes. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Choline is an essential nutrient that contributes to linear growth (Semba et al. 2016c); this comes about because choline is needed for the synthesis of phosphatidylcholines; this synthesis is needed for bone and cell-membrane formation (Ianotti et al. 2017). Third, relative to Food + BCC, Cash + BCC increased the likelihood that children consumed dairy products. Dairy is unique in stimulating plasma insulin-like growth factor 1 (IGF-1), a growth hormone that acts to increase the uptake of amino acids (FAO 2013). Dairy products are rich in calcium, and so this increased likelihood of consumption is also reflected in higher intakes of calcium. Calcium is needed for bone growth, bone length and strength, and skeletal development (Institute of Medicine 2011). Dairy products are also rich in potassium, magnesium, and phosphorus (Dror and Allen (2014). Dairy products have a higher digestibility-corrected amino acid score than any other food (FAO 2013; Headey 2023)— put differently, essential amino acids consumed from dairy products are more easily absorbed than amino acids from other foods—particular in poorer populations more exposed to infections (Semba 2016b). Thus, these results indicate that the different effects on child diets are likely to drive the different effects seen on child nutritional status—particularly with regard to animal-source foods. That said, it would be incorrect to privilege any one of these mechanisms over another in explaining why Cash + BCC, but not Food + BCC, led to increases in HAZ; all of them may have played some role. These results on the role of BCC and the mechanisms through which it operates suggests that work on interventions seeking to improve children’s nutritional status may well benefit from designs that at- tempt to address multiple constraints—energy, diet quality, maternal knowledge, among others—rather than focusing on only one of these. In this setting, food transfers alone did not improve child nutrition outcomes, possibly because they did not lead to increases in children’s consumption of foods most im- portant for growth. Cash transfers also had limited impacts on child nutritional status. However, these results demonstrate that social protection programs that combine cash transfers with well-implemented nutrition BCC can represent a promising approach to reducing chronic undernutrition in low-income settings. Data Availability Statement The data underlying this article will be shared on reasonable request to the corresponding author. References Ahmed, A., J. Hoddinott, S. Roy, E. Sraboni, W. Quabili, and A. Margolies, 2016. “Which Kinds of Social Safety Net Transfers Work Best for the Ultra Poor in Bangladesh?” Operation and Impacts of the Transfer Modality Research Initiative. IFPRI: Dhaka, Bangladesh. Ahmed, A., A. Quisumbing, M. Nasreen, J. Hoddinott, and E. Bryan. 2010. Comparing Food and Cash Transfers to the Ultra-Poor in Bangladesh. IFPRI Research Monograph 163. Washington, DC: International Food Policy Research Institute. Ahmed, A.U., K. Ahmad, V. Chou, R. Hernandez, P. Menon, F. Naeem, F. Naher, W. Quabili, E. Sraboni, B. Yu, and Z. Hassan 2013. The Status of Food Security in the Feed the Future Zone and Other Regions of Bangladesh: Results from the 2011–2012 Bangladesh integrated household survey. Project report submitted to the US Agency for International Development, International Food Policy Research Institute: Dhaka. Ahmed, A.U., J. Hoddinott, S. Roy, and E. Sraboni, 2024. “Transfers, Nutrition Programming, and Economic Well- Being: Experimental Evidence from Bangladesh.” World Development 173: 106414. 470 Ahmed, Hoddinott, and Roy H. Alderman, and U. Gentilini and Yemtsov, R. eds., 2017. The 1.5 billion people question: food, vouchers, or cash transfers? World Bank Publications, Washington DC. Alderman, H., J. Hoddinott, and B. Kinsey, 2006. “Long Term Consequences of Early Childhood Malnutrition.” Oxford Economic Papers 58(3): 450–74. Almås, I., J. Haushofer, and J.P. Shapiro. 2019. “The Income Elasticity for Nutrition: Evidence from Unconditional Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Cash Transfers in Kenya.” NBER Working Paper No. w25711. National Bureau of Economic Research.” Cam- bridge, MA, USA. Baker, J., T. Sanghvi, N. Hajeebhoy, L. Martin, and K Lapping. 2013. “Using an Evidence-Based Approach to Design Large-Scale Programs to Improve Infant and Young Child Feeding.” Food and Nutrition Bulletin 34(3 Suppl2): S146–55. Barnett, I., J. Meeker, K. Roelen, and N. Nisbett. 2022. “Behaviour Change Communication for Child Feeding in Social Assistance: A Scoping Review and Expert Consultation.” Maternal & Child Nutrition 18: e13361. Bastagli, F., J. Hagen-Zanker, and G. Sturge. 2016. “Cash Transfers: What Does the Evidence Say? A Rigorous Review of Programme Impact and of the Role of Design and Implementation Fea- tures.” ODI. https://odi.org/en/publications/cash- transfers- what- does- the- evidence- say- a- rigorous- review- of- imp acts- and- the- role- of- design- and- implementation- features/. Behrman, J.R., and A.B. Deolalikar. 1988. “Health and Nutrition.” In Handbook of Development Economics. Vol. 1, edited by H. Chenery, and T. N. Srinivasan633–711. Amsterdam: Elsevier. Bellemare, M.F., and C.J Wichman, 2020. “Elasticities and the Inverse Hyperbolic Sine Transformation.” Oxford Bulletin of Economics and Statistics 82(1): 50–61. Bhutta, Z.A., J.K. Das, A. Rizvi, M.F. Gaffey, N. Walker, S. Horton, P. Webb, A. Lartey, and R. E Black, 2013. “Evidence-Based Interventions for Improvement of Maternal and Child Nutrition: What Can Be Done and at What Cost?” The Lancet 382(9890): 452–77. Black, R.E., C.G. Victora, S.P. Walker, Z.A. Bhutta, P. Christian, M. De Onis, M. Ezzati, S. Grantham-McGregor, J. Katz, R. Martorell, and R Uauy., 2013. “Maternal and Child Undernutrition and Overweight in Low-Income and Middle-Income Countries.” The Lancet 382(9890): 427–51. Carneiro, P., L. Kraftman, G. Mason, L. Moore, I. Rasul, and M. Scott. 2021. “The Impacts of a Multifaceted Pre-natal Intervention on Human Capital Accumulation in Early Life.” American Economic Review 111(8): 2506–49. Cunha, J.M., 2014. “Testing Paternalism: Cash versus In-Kind Transfers.” American Economic Journal: Applied Eco- nomics 6(2): 195–230. de Groot, R., T. Palermo, S. Handa, L. Ragno, and A. Peterman, 2017. “Cash Transfers and Child Nutrition: Pathways and Impacts.” Development Policy Review 35(5): 621–43. Dror, D.K., and L.H. Allen. 2014. “Dairy Product Intake in Children and Adolescents in Developed Countries: Trends, Nutritional Contribution, and a Review of Association with Health Outcomes.” Nutrition Reviews 72:68–81. doi:10.1111/nure.12078. FAO. 2013. Milk and Dairy Products in Human Nutrition. Rome: Food and Agriculture Organization. Field, E., and E. Maffioli. 2021. “Are Behavioral Change Interventions Needed to Make Cash Transfer Programs Work for Children?” Experimental Evidence from Myanmar. Working paper 28443. National Bureau of Economic Research. Cambridge, MA, USA. Fiszbein, A., and N. Schady. 2009. Conditional Cash Transfers: Reducing Present and Future Poverty. World Bank Policy Research Report. Washington, DC: World Bank. Fitzsimons, E., B. Malde, A. Mesnard, and M.V. Hernandez, 2016. “Nutrition, Information, and Household Behavior: Experimental Evidence from Malawi.” Journal of Development Economics 122: 113–26. Galler, J., M. Bringas-Vega, Q. Tang, A. Rabinowitz, K. Musa et al. 2021. “Neurodevelopmental Effects Of Childhood Malnutrition: A Neuroimaging Perspective.” Neuroimage 231: 117828. Gentilini, U, M.B.A. Almenfi, Y. Okamura et al. 2022. Social Protection and Jobs Responses to COVID-19: A Real- Time Review of Country Measures. Living paper. World Bank. Washington, DC, USA. Grantham-McGregor, S., Y.B. Cheung, S. Cueto, P. Glewwe, L. Richter, and B. Strupp. 2007. “Developmental Potential in the First 5 Years for Children in Developing Countries.” The Lancet 369(9555): 60–70. Haider, R., S. Rasheed, T.G. Sanghvi, N. Hassan, H. Pachon, S. Islam, and C.S. Jalal. 2010. “Breastfeeding in Infancy: Identifying the Program-Relevant Issues in Bangladesh.” International Breastfeeding Journal 5(1): 21. Haushofer, J., and J. Shapiro. 2016. “The Short-Term Impact of Unconditional Cash Transfers to the Poor: Experi- mental Evidence from Kenya.” Quarterly Journal of Economics 131(4): 1973−2042. The World Bank Economic Review 471 Headey, D. 2023. “Can Dairy Help Solve the Malnutrition Crisis in Developing Countries? An Economic Analysis.” Animal Frontiers 13(1): 7–16. Headey, D., J. Hoddinott, D. Ali, R. Tesfaye, and M. Dereje, 2015. “The Other Asian Enigma: Explaining the Rapid Reduction of Malnutrition in Bangladesh.” World Development 66(2): 749–61. Hidrobo, M., J. Hoddinott, N. Kumar, and M. Olivier. 2018. “Social Protection, Food Security, and Asset Formation.” Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 World Development 101: 88–103. Hidrobo, M., J. Hoddinott, A. Peterman, A. Margolies, and V. Moreira. 2014. “Cash, Food, or Vouchers? Evidence from a Randomized Experiment in Northern Ecuador.” Journal of Development Economics 107: 144–56 Hirvonen, K., and J Hoddinott. 2021. “Beneficiary Views on Cash and In-Kind Payments: Evidence from Ethiopia’s Productive Safety Net Programme.” World Bank Economic Review 35(2): 398–413. Hoddinott, J., N. Karachiwalla, N. Ledlie, and S. Roy. 2016. “Adolescent Girls’ Infant and Young Child Nutrition Knowledge Levels and Sources Differ among Urban and Rural Samples in Bangladesh.” Maternal & Child Nutrition 12(4): 885–97. Hoddinott, J., J. Maluccio, J. Behrman, R. Martorell, P. Melgar, A. Quisumbing, M. Ramirez-Zea, A. Stein, and K. Yount, 2013. “Adult Consequences of Growth Failure in Early Childhood.” American Journal of Clinical Nutrition 98: 1170–8. Iannotti, L. L., C.K. Lutter, W.F. Waters, C.A. Gallegos Riofrío, C. Malo, G. Reinhart, A. Palacios, C. Karp, M. Chap- nick, K. Cox, and S. Aguirre. 2017. “Eggs Early in Complementary Feeding Increase Choline Pathway Biomark- ers and DHA: A Randomized Controlled Trial in Ecuador.” American Journal of Clinical Nutrition 106(6): 1482–9. Institute of Medicine. 2011. Dietary Reference Intakes for Calcium and Vitamin D. Washington, DC: National Academies Press US. Keats, E., J. Das, R. Salam, Z. Lassi, A. Imdad, R. Black, and Z. Bhutta, 2021. “Effective Interventions to Address Maternal and Child Malnutrition: An Update of the Evidence.” Lancet Child & Adolescent Health 5(5): 367–84. Khan, G.N., S. Kureishy, N. Akbar, M. Nasir, M. Hussain, I Ahmed. et al. 2019. “A Stunting Prevention Cluster Randomized Controlled Trial: Leveraging the Social Protection System to Prevent Stunting in District Rahim Yar Khan, Punjab, Pakistan.” Islamabad, Pakistan. Leroy, J.L., and E.A. Frongillo. 2019. “Perspective: What Does Stunting Really Mean? A Critical Review of the Evi- dence.” Advances in Nutrition 10(2): 196–204. Leroy, J.L., M. Ruel, E.A. Frongillo, J. Harris, and T.J Ballard., 2015. “Measuring the Food Access Dimension of Food Security: A Critical Review and Mapping of Indicators.” Food and Nutrition Bulletin 36(2): 167–95. Leroy, J.L., M. Ruel, J.-P. Habicht, and E.A. Frongillo., 2014. “Linear Growth Deficit Continues to Accumulate beyond the First 1000 Days in Low- and Middle-Income Countries: Global Evidence from 51 National Surveys.” Journal of Nutrition 144: 1460–6. Levere, M., G. Acharya, and P. Bharadwaj. forthcoming. “The Role of Information and Cash Transfers in Early Child- hood Development: Short and Long Run Evidence from Nepal.” Economic Development and Cultural Change. Little, M.T., K. Roelen, B.C. Lange, J.I. Steinert, A.R. Yakubovich, L. Cluver, and D.K. Humphreys., 2021. “Effective- ness of Cash-Plus Programmes on Early Childhood Outcomes Compared to Cash Transfers Alone: A Systematic Review and Meta-Analysis in Low- and Middle-Income Countries.” PLoS medicine 18(9): e1003698. Lusk, J.L., and A. Weaver. 2017. “An Experiment on Cash and In-Kind Transfers with Application to Food Assistance Programs.” Food Policy 68: 186–92. MacPherson, C., and O. Sterck. 2021. “Empowering Refugees through Cash and Agriculture: A Regression Disconti- nuity Design.” Journal of Development Economics 149: 102614. Manley, J., H. Alderman, and U. Gentilini. 2022. “More Evidence on Cash Transfers and Child Nutritional Outcomes: A Systematic Review and Meta-Analysis.” BMJ Global Health 7: e008233. Manley, J., Y. Balarajan, S. Malm, L. Harman, J. Owens, S. Murthy, D. Stewart, N.E. Winder-Rossi, and A. Khurshid. 2020. “Cash Transfers and Child Nutritional Outcomes: A Systematic Review and Meta-Analysis.” BMJ Global Health 5:e003621. doi:10.1136/bmjgh-2020-003621 McKenzie, D., 2017. “Should We Require Balance t-Tests of Baseline Observables in Randomized Experiments?” World Bank blog post, accessed July 3, 2019. https://blogs.worldbank.org/impactevaluations/should- we- require- balance- t- tests- baseline- observables- randomized- experiments. 472 Ahmed, Hoddinott, and Roy Nguyen, P.H., K.K. Saha, D. Ali, P. Menon, S. Manohar, L.T. Mai et al. 2014. “Maternal Mental Health Is Associated with Child Undernutrition and Illness in Bangladesh, Vietnam and Ethiopia.” Public Health Nutrition 17(06): 1318–27. Olney, D.K., A. Gelli, N. Kumar, H. Alderman, A. Go, and A. Raza. 2022. “Social Assistance Programme Impacts on Women’s and Children’s Diets and Nutritional Status.” Maternal and Child Nutrition 18:e13378. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Parikh, P., R. Semba, M. Manary, S. Swaminathan, E. Udomkesmalee, R. Bos, B.K. Poh, N. Rojroongwasinkul, J. Geurts, R. Sekartini, and T.T Nga. 2022. “Animal Source Foods, Rich in Essential Amino Acids, Are Important for Linear Growth and Development of Young Children in Low- and Middle-Income Countries.” Maternal & Child Nutrition 18(1): e13264. Reynolds, T., C. Anderson, P. Biscaye, D. Coomes, T. Madsen, E. Ebeling, and A. Favreau 2017. “Review of Long- Term Impacts of Cash Transfer Programs.” Evans School of Policy and Research. University of Seattle. Seattle Washington. Romano, J.P., and M. Wolf 2005. “Stepwise multiple testing as formalized data snooping.” Econometrica 73 (4): 1237–1282. Ruel, M.T., H. Alderman, andMaternal and Child Nutrition Study Group. 2013. “Nutrition-Sensitive Interventions and Programmes: How Can They Help to Accelerate Progress in Improving Maternal and Child Nutrition?” Lancet 382(9891): 536–51. Semba, R.D., M. Shardell, F.A.S. Ashour, R. Moaddel, I. Trehan, K.M. Maleta, M.I. Ordiz, K. Kraemer, M.A. Khadeer, L. Ferrucci, and M.J. Manary. 2016b. “Child Stunting Is Associated with Low Circulating Essential Amino Acids.” EBioMedicine 6: 246–52. Semba, R.D., I. Trehan, M. Gonzalez-Freire, K. Kraemer, R. Moaddel, M.I. Ordiz, L. Ferrucci, and M.J. Manary., 2016a. “Perspective: The Potential Role of Essential Amino Acids and the Mechanistic Target of Rapamycin Complex 1 (Mtorc1) Pathway in the Pathogenesis of Child Stunting.” Advances in Nutrition 7(5): 853–65. Semba, R.D., P. Zhang, M. Gonzalez-Freire, R. Moaddel, I. Trehan, K.M. Maleta, and M.J. Manary. 2016c. “The Association of Serum Choline with Linear Growth Failure in Young Children from Rural Malawi.” American Journal of Clinical Nutrition 104(1): 191–7. Shaheen, N., A.T. MA Rahim, Md. Mohiduzzaman, Md.L. Bari, and SMM. Rahman 2013. “Food Composition Table for Bangladesh.” Dhaka, Bangladesh: University of Dhaka. UNICEF, WHO, and World Bank. 2023. UNICEF/WHO/World Bank Group—Joint Child Malnutrition Estimates 2021 edition. New York: UNICEF. United Nations. 2021. “Sustainable Development Goals.” https://sdgs.un.org/goals. Accessed December 10, 2021. Victora, C., P. Christian, L.P. Vidaletti, G. Gatica-Domínguez, P. Menon, and R.E. Black. 2021. “Revisiting Maternal and Child Undernutrition in Low-Income and Middle-Income Countries: Variable Progress towards an Unfinished Agenda.” Lancet 397(10282): 1388–99. Victora, C.G., M. De Onis, P.C. Hallal, M. Blössnerand and R. Shrimpton, 2010. “Worldwide timing of growth fal- tering: revisiting implications for interventions.” Pediatrics, 125 (3): pp. e473–e480. WHO (World Health Organization). 2006. WHO Child Growth Standards: Length/Height-for-Age, Weight-for-Age, Weight-for-Length, Weight-for-Height and Body Mass Index-for-Age: Methods and Development. Geneva, Switzer- land: WHO. ———. 2008. “Indicators for Assessing Infant and Young Child Feeding Practices: Conclusions of a Consensus Meet- ing Held 6–8 November 2007 in Washington DC, USA.” Geneva, Washington: WHO. WHO, & FAO. 2007. “Protein and Amino Acid Requirements in Human Nutrition: Report of a Joint FAO/WHO/UNU Expert Consultation.” WHO Technical Report Series 935. Geneva, Switzerland: World Health Organization. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Supplementary Online Appendix Food Transfers, Cash Transfers, Behavior Change Communication and Child Nutrition: Evidence from Bangladesh Akhter Ahmed, John Hoddinott , and Shalini Roy S1: Sample Size Calculations, Selection of Treatment Units, Survey Timeline and Attrition at the Household Level Sample size calculations were undertaken to assess the number of clusters (villages) and households needed to detect changes in both household- and child-level outcomes. Using data from an earlier study in Bangladesh (Ahmed et al. 2010), setting significance level at 0.05 and statistical power at 0.80, assuming Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 attrition of 10 percent over the duration of the intervention, and using outcome-specific means, standard deviations and intra-cluster correlations, a sample based on 50 clusters per treatment and 10 households per cluster would provide sufficient statistical power to detect an increase of 12 percent in household per capita total expenditure per month; 7 percent in household per capita calorie intake per day; 16 percent in child height-for-age z-score; and 8 percent in dietary diversity of children 12 to 60 months. Based on these calculations, in the North, five upazilas (subdistricts) were selected using simple random sampling from a list of upazilas where in 2010 the proportion of households living below Bangladesh’s lower poverty line was 25 percent or higher. All villages within these five upazilas were listed. Villages classified as urban or villages with fewer than 125 households were dropped. Using a random number generator, each village was assigned a random number. Villages were then sorted in ascending numerical order with the first 275 retained. Given that in each region, there are four treatment arms and a control group, the first 50 villages were assigned to treatment group 1, the second 50 to treatment group 2, the third 50 villages to treatment group 3, the fourth 50 villages to treatment group 4, and the fifth 50 villages to the control group. The remaining 25 villages were held as a reserve. A complete village census was carried out in each of the 250 selected villages, collecting information on household demographics, a set of poverty indicators, and whether households participated in safety nets and other targeted interventions. Using these data, a list was compiled of households that: (1) were considered poor (i.e., based on the poverty indicators collected, they were estimated to have consumption below Bangladesh’s lower poverty line); (2) would have at least one child aged 0–24 months when the intervention began; and (3) were not receiving benefits from other safety net interventions. These households were eligible to participate in the study. Using simple random sampling, 10 eligible households were selected from each village. The total sample in the North included 250 clusters and 2,500 households. An identical process was used in the South to select upazilas, villages, and households. The baseline survey was carried out in March–April 2012, prior to the first transfer payment in May 2012. The principal survey instrument was a multitopic household survey with modules covering house- hold demographics, income generation, assets, food and nonfood expenditures, measures of food security and food consumption, health and morbidity, women’s status, shocks, anthropometry of all children under 5 years of age and their mothers, and a 24-hour recall module of food groups consumed by children 0–24 months. Modules were split across household heads and their spouses, with relevant sections asked of the most knowledgeable household member, and men being interviewed by male enumerators and women be- ing interviewed by female enumerators. A midline survey was conducted in June 2013, to assess whether the intervention was being implemented as designed from the beneficiary perspective and to provide a first set of outcome measurements. The endline survey was conducted in April 2014 during the final month of transfer payments. In addition to the household survey instrument, community questionnaires were administered at baseline, midline, and endline to capture information on local infrastructure, access to services, and food prices. Qualitative research using a mix of focus groups and key informant interviews was undertaken in October 2012, five months after the intervention began, to assess program imple- mentation, beneficiary perceptions on how transfers had affected livelihoods and wellbeing; and whether cash and food transfers affected the relations between TMRI participants and nonparticipants within the communities. At endline, the study also surveyed the community nutrition workers who implemented the nutrition BCC trainings. The researchers interviewed 4,992 households at baseline, 2,498 in the North and 2,494 in the South.29 In the North, the study re-interviewed 2,410 households at endline, an attrition rate of 3.5 percent; 78 households were not surveyed at endline because they had migrated, another 10 dropped out of study, refused to be interviewed, or could not be found. In the South, the study re-interviewed 2,438 households at endline, an attrition rate of 2.2 percent; 49 households were not surveyed at endline because they had Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 migrated, another 7 dropped out of study, refused to be interviewed, or could not be found. Using probit regressions, the analysis found no evidence that attrition was related to treatment status or household demographic, occupational, or asset characteristics (Ahmed et al. 2016). S2: Additional Figures and Tables Figure S2.1. Impacts on Selected Anthropometric Outcomes, North, by Child Sex Source: Authors’ calculations based on data from the TMRI study. Note: Sample sizes are 1,165 (boys) and 1,053 (girls). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. For each treatment and anthropometric measure, the analysis does not reject the null hypothesis that the impacts on girls and boys are equal; p-values for these tests are available on request. 29 Three households were not interviewed because, on religious grounds, they changed their minds about being included in the study, having previously agreed to be included. The study does not have documentation on why the remaining five were not interviewed. Figure S2.2. Impacts on selected anthropometric outcomes, south, by child sex. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Source: Authors’ calculations based on data from the TMRI study. Note: Sample sizes are 1,113 (boys) and 1,068 (girls). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0-48 months at endline. For each treatment and anthropometric measure, the analysis does not reject the null hypothesis that the impacts on older and younger children are equal; p-values for these tests are available on request. Figure S2.3. Impacts on Selected Anthropometric Outcomes, by Child Age at Endline, North Source: Authors’ calculations based on data from the TMRI study. Note: Sample sizes are 1,668 (older children) and 530 (younger children). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. For each treatment and anthropometric measure, the analysis does not reject the null hypothesis that the impacts on older and younger children are equal; p-values for these tests are available on request. Figure S.4. Impacts on Selected Anthropometric Outcomes, by Child Age at Endline, South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Source: Authors’ calculations based on data from the TMRI study. Note: Sample sizes are 1,651 (older children) and 530 (younger children). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. For each treatment and anthropometric measure, the study does not reject the null hypothesis that the impacts on older and younger children are equal; p-values for these tests are available on request. Table S2.1. Estimating Impacts on Dichotomous Anthropometric Outcomes Using Probits, by Region (1) (2) (3) (4) (5) (6) North South Stunting Underweight Wasting Stunting Underweight Wasting Treatment Cash only −0.008 0.034 0.002 0.006 0.068 0.017 (0.04) (0.04) (0.02) (0.03) (0.04) (0.03) Food only −0.031 −0.009 −0.035∗ 0.010 0.037 −0.005 (0.03) (0.03) (0.02) (0.04) (0.04) (0.03) Cash & Food −0.039 0.016 −0.001 −0.052 0.020 0.004 (0.03) (0.04) (0.02) (0.04) (0.04) (0.03) Cash + BCC −0.078∗∗ −0.038 −0.014 (0.03) (0.04) (0.02) Food + BCC −0.053 0.002 −0.008 (0.03) (0.03) (0.03) p-values: Cash = Food 0.52 0.17 0.12 0.90 0.41 0.37 Cash = Cash + BCC 0.06 0.05 0.51 Food = Food + BCC 0.06 0.32 0.91 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,218 (North) and 2,181 (South). OLS regressions. Estimates are expressed as marginal effects. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. Table S2.2. Treatment Impacts with Additional Controls, Selected Anthropometric Outcomes, North Height-for-age z scores Stunting Weight-for-age z scores (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Cash only 0.035 0.060 0.132 0.106 −0.008 −0.023 −0.052 −0.040 0.019 0.040 0.081 0.047 (0.08) (0.07) (0.08) (0.06) (0.04) (0.03) (0.04) (0.03) (0.07) (0.06) (0.07) (0.05) Food only 0.048 0.030 0.051 0.089 −0.031 −0.024 −0.027 −0.032 0.087 0.069 0.082 0.035 (0.08) (0.08) (0.07) (0.06) (0.03) (0.03) (0.03) (0.03) (0.06) (0.06) (0.06) (0.04) Cash & Food 0.119 0.124 0.145∗∗ 0.127∗∗ −0.039 −0.039 −0.051 −0.035 0.029 0.028 0.041 0.066 (0.08) (0.08) (0.07) (0.06) (0.03) (0.03) (0.03) (0.03) (0.07) (0.06) (0.06) (0.04) Cash + BCC 0.248∗∗∗ 0.250∗∗∗ 0.263∗∗∗ 0.210∗∗∗ −0.078∗∗ −0.079∗∗ −0.079∗∗ −0.056∗ 0.162∗∗ 0.165∗∗∗ 0.170∗∗∗ 0.103∗∗ (0.08) (0.08) (0.07) (0.06) (0.03) (0.03) (0.04) (0.03) (0.07) (0.06) (0.07) (0.05) p-values: Cash = Food 0.87 0.70 0.28 0.78 0.52 0.81 0.96 0.49 0.28 0.63 0.99 0.78 Cash = Cash + BCC <0.01 0.01 0.08 0.07 0.06 0.63 0.11 0.45 0.04 0.06 0.18 0.21 Sample size 2,218 2,214 2,214 2,016 2,218 2,214 2,214 2,016 2,218 2,214 2,214 2,016 Source: Authors’ calculations based on data from the TMRI study. Note: OLS regressions. ∗ ∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample includes all children who are offspring of the household head and who were 0–48 months at endline. Columns (1), (5), and (9) are the base specifications. Columns (2), (6), and (10) include child (age, sex) and maternal (age, education, height) controls. Columns (3), (7), and (11) include child and maternal controls and union fixed effects. Columns (4), (8), and (12) are ANCOVA models that include child baseline anthropometric status, child, and maternal controls and union fixed effects. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Table S2.3. Treatment Impacts with Additional Controls, Selected Anthropometric Outcomes, South Height-for-age z scores Stunting Weight-for-age z scores (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Cash only −0.097 −0.071 −0.085 −0.089 0.006 −0.001 0.004 −0.013 −0.124 −0.106 −0.078 −0.057 (0.08) (0.07) (0.07) (0.06) (0.03) (0.03) (0.04) (0.03) (0.08) (0.07) (0.07) (0.05) Food only −0.100 −0.084 −0.080 −0.043 0.010 0.003 0.003 −0.022 −0.090 −0.073 −0.069 −0.066 (0.09) (0.08) (0.07) (0.06) (0.04) (0.03) (0.04) (0.03) (0.08) (0.07) (0.07) (0.05) Cash & Food 0.024 0.010 −0.012 0.037 −0.052 −0.046 −0.054 −0.072∗∗ 0.001 −0.003 0.043 0.015 (0.08) (0.08) (0.07) (0.06) (0.04) (0.03) (0.04) (0.03) (0.07) (0.07) (0.07) (0.05) Food + BCC 0.079 0.060 0.009 0.032 −0.053 −0.043 −0.041 −0.052 0.016 0.026 0.073 0.064 (0.08) (0.08) (0.07) (0.06) (0.03) (0.03) (0.04) (0.04) (0.07) (0.07) (0.07) (0.05) p-values: Cash = Food 0.96 0.86 0.95 0.46 0.90 0.90 0.99 0.79 0.64 0.63 0.90 0.85 Food = Food + BCC 0.03 0.07 0.23 0.25 0.07 0.16 0.23 0.39 0.14 0.16 0.04 0.01 Sample size 2,181 2,180 2,180 1,958 2,181 2,180 2,180 1,958 2,181 2,180 2,180 1,965 Source: Authors’ calculations based on data from the TMRI study. Note: See Table S2.2. Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Table S2.4. Impacts on Selected Anthropometric Outcomes, North, by Maternal Schooling (1) (2) (3) (4) (5) (6) Height-for-age Z-score Stunting Weight-for-age Z-score Maternal schooling 0–4 grades > 4 grades 0–4 grades > 4 grades 0–4 grades > 4 grades Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Treatment Cash only 0.025 0.077 0.003 −0.040 −0.038 0.141 (0.10) (0.11) (0.04) (0.05) (0.08) (0.11) Food only 0.080 0.002 −0.039 −0.023 0.080 0.107 (0.10) (0.12) (0.04) (0.05) (0.07) (0.10) Cash & Food 0.199∗∗ −0.032 −0.069 0.017 0.046 0.002 (0.10) (0.13) (0.04) (0.06) (0.08) (0.13) Cash + BCC 0.289∗∗∗ 0.188 −0.087∗∗ −0.071 0.133 0.238∗∗ (0.10) (0.11) (0.04) (0.05) (0.08) (0.11) p-values: Cash: 0–4 grades= GT 4 grades 0.71 0.51 0.18 Food: 0–4 grades= GT 4 grades 0.60 0.82 0.82 Cash & Food: 0–4 grades= GT 4 grades 0.12 0.20 0.75 Cash + BCC: 0–4 grades= GT 4 grades 0.48 0.82 0.42 Sample size 1,506 712 1,506 712 1,506 712 Source: Authors’ calculations based on data from the TMRI study. Note: OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of random- ization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Table S2.5. Impacts on Selected Anthropometric Outcomes, South, by Maternal Schooling (1) (2) (3) (4) (5) (6) Height-for-age Z-score Stunting Weight-for-age Z-score Maternal schooling 0–4 grades > 4 grades 0–4 grades > 4 grades 0–4 grades > 4 grades Treatment Cash only −0.184 0.089 0.049 −0.080 −0.166 −0.042 (0.10) (0.12) (0.04) (0.06) (0.09) (0.13) Food only −0.156 0.019 0.049 −0.064 −0.123 −0.026 (0.11) (0.12) (0.04) (0.06) (0.10) (0.11) Cash & Food −0.019 0.096 −0.033 −0.082 −0.040 0.061 (0.10) (0.11) (0.05) (0.06) (0.09) (0.11) Food + BCC 0.068 0.103 −0.020 −0.102 0.016 0.018 (0.10) (0.12) (0.04) (0.06) (0.09) (0.12) p-values: Cash: Younger = Older mother 0.09 0.09 0.43 Food: Younger = Older mother 0.25 0.15 0.48 Cash & Food: Younger = Older mother 0.42 0.51 0.44 Food + BCC: Younger = Older mother 0.82 0.29 0.99 Sample size 1,316 865 1,316 865 1,316 865 Source: Authors’ calculations based on data from the TMRI study. Note: OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of random- ization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Table S2.6. Impacts on Selected Anthropometric Outcomes, North, by Maternal Age at Baseline (1) (2) (3) (4) (5) (6) Height-for-age Z-score Stunting Weight-for-age Z-score Maternal age <26.1 years Older than 26 <26.1 years Older than 26 <26.1 years Older than 26 Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Treatment Cash only −0.015 0.086 0.010 −0.028 −0.070 0.099 (0.10) (0.10) (0.05) (0.04) (0.09) (0.08) Food only 0.163 −0.055 −0.069 0.002 0.100 0.077 (0.11) (0.10) (0.04) (0.05) (0.08) (0.07) Cash & Food 0.133 0.104 −0.038 −0.040 0.064 −0.005 (0.12) (0.09) (0.05) (0.04) (0.10) (0.07) Cash + BCC 0.321∗∗∗ 0.181 −0.075 −0.085 0.200∗∗ 0.125 (0.11) (0.10) (0.04) (0.05) (0.09) (0.09) p-values: Cash: Younger = Older mother 0.42 0.48 0.13 Food: Younger = Older mother 0.11 0.22 0.81 Cash & Food: Younger = Older 0.83 0.96 0.54 mother Cash + BCC: Younger = Older 0.31 0.86 0.53 mother Sample size 1,065 1,164 1,065 1,164 1,065 1,164 Source: Authors’ calculations based on data from the TMRI study. Note: OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of random- ization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Table S2.7. Impacts on Selected Anthropometric Outcomes, South, by Maternal Age at Baseline (1) (2) (3) (4) (5) (6) Height-for-age Z-score Stunting Weight-for-age Z-score Maternal age <26.1 years Older than 26 <26.1 years Older than 26 <26.1 years Older than 26 Treatment Cash only −0.164 −0.036 0.040 −0.024 −0.268∗∗ −0.004 (0.11) (0.09) (0.05) (0.04) (0.11) (0.09) Food only −0.099 −0.109 0.019 0.006 −0.056 −0.138 (0.11) (0.11) (0.05) (0.05) (0.10) (0.10) Cash & Food 0.063 −0.039 −0.073 −0.016 0.042 −0.073 (0.11) (0.10) (0.05) (0.05) (0.10) (0.09) Food + BCC 0.116 0.012 −0.070 −0.030 −0.065 0.079 (0.11) (0.10) (0.04) (0.05) (0.10) (0.09) p-values: Cash: Younger = Older mother 0.32 0.34 0.04 Food: Younger = Older mother 0.93 0.82 0.49 Cash & Food: Younger = Older mother 0.44 0.38 0.36 Food + BCC: Younger = Older mother 0.45 0.50 0.23 Sample size 1,099 1,102 1,099 1,102 1,099 1,102 Source: Authors’ calculations based on data from the TMRI study. Note: OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of random- ization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Table S2.8. Impact on Endline per Capita Household Consumption, by Region North South Cash only 0.143∗∗∗ 0.059∗∗∗ Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 (0.02) (0.02) Food only 0.102∗∗∗ 0.080∗∗∗ (0.02) (0.02) Cash & Food 0.061∗∗∗ 0.077∗∗∗ (0.02) (0.02) Cash + BCC 0.270∗∗∗ (0.02) Food + BCC 0.189∗∗∗ (0.02) Baseline mean (taka): Control group 1256 1414 Marginal effects: Cash only 0.154 0.061 Food only 0.107 0.083 Cash & Food 0.063 0.080 Cash + BCC 0.309 Food + BCC 0.208 p-values: Cash = Food 0.05 0.37 Cash = Cash + BCC <0.01 Food = Food + BCC <0.01 Cash + BCC = Food + BCC 0.04 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,209 (North) and 2,175 (South). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Controls include baseline value of dependent variable, child age, sex, maternal age, height and education, and union fixed effects. Table S2.9. Impact of Treatment on Household Caloric Acquisition with Control for per Capita Expenditure, by food Group and Region (1) (2) (3) (4) (5) (6) Rice Pulses Oils Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 North South North South North South Per capita expenditure (IHS) 0.265∗∗∗ 0.260∗∗∗ 0.797∗∗∗ 0.893∗∗∗ 0.462∗∗∗ 0.494∗∗∗ (0.02) (0.02) (0.10) (0.09) (0.04) (0.04) Cash only −0.000 −0.007 0.648∗∗∗ 0.053 0.076 −0.029 (0.02) (0.02) (0.10) (0.10) (0.04) (0.04) Food only 0.014 −0.003 1.548∗∗∗ 0.836∗∗∗ 0.180∗∗∗ 0.065 (0.02) (0.02) (0.10) (0.10) (0.04) (0.04) Cash & Food 0.011 −0.004 1.465∗∗∗ 0.721∗∗∗ 0.098∗∗∗ 0.050 (0.02) (0.02) (0.10) (0.10) (0.04) (0.04) Cash + BCC −0.035∗∗ 1.620∗∗∗ 0.121∗∗∗ (0.02) (0.11) (0.04) Food + BCC −0.053∗∗∗ 0.809∗∗∗ −0.074 (0.02) (0.10) (0.04) Elasticities Per capita expenditure 0.30 0.29 1.22 1.44 0.59 0.64 Marginal effects: Cash only 0.0004 −0.007 0.91 0.05 0.08 −0.03 Food only 0.014 −0.003 3.70 1.31 0.20 0.07 Cash & Food 0.011 −0.004 3.32 1.06 0.10 0.05 Cash + BCC −0.035 4.05 0.13 Food + BCC −0.051 1.24 −0.07 Baseline mean: Control group 1475 1533 14 44 81 108 p-values: Cash = Food 0.36 0.82 <0.01 <0.01 <0.01 0.02 Cash = Cash + BCC 0.03 <0.01 0.24 Food = Food + BCC <0.01 0.78 <0.01 Cash + BCC = Food + BCC 0.60 <0.01 0.01 (7) (8) (9) (10) (11) (12) Eggs Dairy Meat and fish North South North South North South Per capita expenditure (IHS) 1.231∗∗∗ 1.329∗∗∗ 1.645∗∗∗ 0.952∗∗∗ 2.016∗∗∗ 2.269∗∗∗ (0.09) (0.10) (0.13) (0.12) (0.09) (0.07) Cash only 0.262∗∗∗ 0.323∗∗∗ −0.174 −0.042 0.383∗∗∗ 0.163∗∗ (0.10) (0.10) (0.14) (0.13) (0.09) (0.08) Food only 0.056 0.055 0.089 −0.063 0.225∗∗∗ −0.035 (0.09) (0.10) (0.13) (0.13) (0.09) (0.08) Cash & Food 0.145 0.079 0.209 0.064 0.332∗∗∗ 0.042 (0.09) (0.10) (0.14) (0.13) (0.09) (0.08) Cash + BCC 0.707∗∗∗ 0.830∗∗∗ 0.735∗∗∗ (0.10) (0.14) (0.09) Food + BCC 0.678∗∗∗ 0.345∗∗ 0.332∗∗∗ (0.11) (0.14) (0.08) Elasticities Per capita expenditure 2.42 2.78 4.18 1.59 6.51 8.66 Marginal effects Cash only 0.30 0.38 −0.16 −0.04 0.47 0.17 Food only 0.06 0.06 0.09 −0.06 0.25 −0.04 Cash & Food 0.16 0.08 0.23 0.07 0.40 0.04 Cash + BCC 1.03 1.29 1.08 Food + BCC 0.97 0.41 0.39 Table S2.9. Continued (7) (8) (9) (10) (11) (12) Eggs Dairy Meat and fish North South North South North South Downloaded from https://academic.oup.com/wber/article/doi/10.1093/wber/lhae023/7680205 by The World Bank user on 02 May 2025 Baseline mean: Control group 4 4 10 8 28 38 p-values: Cash = Food 0.03 <0.01 0.06 0.88 0.08 0.01 Cash = Cash + BCC <0.01 <0.01 <0.01 Food = Food + BCC <0.01 <0.01 <0.01 Cash + BCC = Food + BCC 0.87 0.02 <0.01 (13) (14) (15) (16) (17) (18) Fruit and vegetables Other starches Other foods North South North South North South Per capita expenditure (IHS) 0.807∗∗∗ 0.944∗∗∗ 0.480∗∗∗ 0.708∗∗∗ 1.271∗∗∗ 0.966∗∗∗ (0.03) (0.04) (0.06) (0.06) (0.05) (0.04) Cash only 0.106∗∗∗ 0.112∗∗∗ −0.147∗∗ −0.233∗∗∗ −0.020 0.110∗∗ (0.04) (0.04) (0.07) (0.07) (0.06) (0.05) Food only 0.062∗ 0.049 −0.163∗∗ −0.095 0.042 0.055 (0.03) (0.04) (0.06) (0.07) (0.05) (0.05) Cash & Food 0.084∗∗ 0.106∗∗∗ −0.135∗∗ −0.125∗ −0.033 0.084∗ (0.03) (0.04) (0.07) (0.07) (0.05) (0.05) Cash + BCC 0.361∗∗∗ 0.050 0.245∗∗∗ (0.04) (0.07) (0.06) Food + BCC 0.259∗∗∗ −0.050 0.186∗∗∗ (0.04) (0.07) (0.05) Elasticities Per capita expenditure 1.24 1.57 0.62 1.03 2.56 1.63 Marginal effects Cash only 0.11 0.12 −0.14 −0.21 −0.02 0.12 Food only 0.06 0.05 −0.15 −0.09 0.04 0.06 Cash & Food 0.08 0.11 −0.12 −0.12 −0.03 0.09 Cash + BCC 0.43 0.05 0.28 0.20 Food + BCC 0.30 −0.05 Baseline mean: Control group 55.8 71.4 100.9 85.6 32.7 47.9 p-values: Cash = Food 0.21 0.12 0.80 0.05 0.27 0.25 Cash = Cash + BCC <0.01 <0.01 <0.01 Food = Food + BCC <0.01 0.52 <0.01 Cash + BCC = Food + BCC 0.14 0.36 0.92 Source: Authors’ calculations based on data from the TMRI study. Note: Sample size 2,209 (North) and 2,175 (South). OLS regressions. ∗∗ significant at the 5 percent level; ∗∗∗ significant at the 1 percent level. Standard errors are clustered at the village level, the unit of randomization. Sample describes the housing of all children who are offspring of the household head and who were 0–48 months at endline. Controls include baseline value of dependent variable. child age, sex, maternal age, height and education, and union fixed effects. C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.