Kurdi 2019.

Study characteristics
Methods	Study design: cRCT Study grouping: N/A How were missing data handled? A conservative ITT approach was taken for determining impact estimates. 6 HHs, which had identification matching problems at follow‐up, were dropped entirely from analysis. Randomisation ratio: 1:1. Data for the evaluation came from a HH survey of 2000 HHs with indirect relatives of Social Welfare Fund beneficiaries in the 3 targeted districts in Al Hodeidah, half of which were in communities where indirect relatives were included in the programme (treated) and half of which were in communities where they were not included (control). (Kurdi 2019 policy brief) Recruitment method: programme targeted the poorest and the most vulnerable HHs in the country by restricting recipients to Social Welfare Fund beneficiary HHs in 3 eligible districts in Al Hodeidah governorate: Marawi'ah, Bayt Al Faqiah and Zabid (Kurdi 2019, Introduction – Program description). Beneficiaries were divided into 2 separate priority groups. Women in the pilot districts who were direct family members of Social Welfare Fund beneficiaries were in the first priority group and were automatically included in the Cash for Nutrition programme (received the intervention). Indirect family members, such as daughters‐in‐law, of Social Welfare Fund beneficiaries were added to a second priority list. Because the programme was not large enough to include all of these second priority women, inclusion of these indirect family member beneficiaries was randomised at the community level (Kurdi 2019, Sample and Methodology – Randomization). The primary survey respondent was the woman identified as a second‐priority potential beneficiary by the programme, which meant that all respondents were pregnant or mothers of children aged < 2 at the time of the baseline survey (Kurdi 2019, Sample and Methodology – Survey). Manner in which HHs were approached and invited to participate NR. Sample size justification and outcome used: NR Sampling method: purposive sampling. Programme targeted the poorest and the most vulnerable HHs in the country by restricting recipients to Social Welfare Fund beneficiary HHs (Kurdi 2019, Introduction). Because the pilot programme was limited to 4800 beneficiaries, Social Welfare Fund beneficiaries were divided into 2 separate priority groups. Women in the pilot districts who were direct family members of Social Welfare Fund beneficiaries were in the first priority group and were automatically included in the Cash for Nutrition programme (received intervention). Indirect family members, such as daughters‐in‐law, of Social Welfare Fund beneficiaries were added to a second priority list. Because the programme was not large enough to include all of these second priority women, inclusion of these indirect family member beneficiaries was randomised at the community level (Kurdi 2019, Sample and methodology – Randomization). The cRCT population consisted of the HHs of these women on the 'second priority list.' Study aim or objective: to measure the impact of the Cash for Nutrition programme on eligible HHs; to describe the degree to which HH characteristics, details of programme implementation and the external environment increased or decreased programme impact; and to describe the functioning of the programme and changes in HH welfare in the current conflict environment. Study period: December 2014 to August 2017 Unit of allocation or exposure: HHs
Participants	Baseline characteristics Cash for Nutrition Intervention Age: mother mean, years: 27.6 (SD 6.86); mother's age at marriage: mean, years: 17.6 (SD 2.94) Place of residence: NR Sex: NR Ethnicity and language: NR Occupation: NR Education: HHs with illiterate mother, %: 79.7 SES: Asset Index mean –0.1349 (SD 1.768); HHs owning a television, %: 33.6; HHs owning a washing machine, %: 5.9; HH owning livestock, %: 52.3; rooms in house, mean: 1.30 (SD 0.581); people in house, mean: 6.21 (SD 3.54) Social capital: HHs with husband, %: 72.9 Nutritional status: mean calories consumed per adult male equivalent mean coefficient: 1820.4 (SD 1081.8); proportion of HHs without enough food in past 7 days: 0.358 Morbidities: NR Concomitant or previous care: NR Control Age: mother, mean, years: 28.1 (SD 6.81); mother's age at marriage, mean, years: 17.6 (SD 2.86) Place of residence: NR Sex: NR Ethnicity and language: NR Occupation: NR Education: HHs with illiterate mother, %: 73.4 SES: Asset Index mean 0.1352 (SD 1.930); HHs owning a television, %: 44.0; HHs owning a washing machine, %: 9.2; HH owning livestock, %: 52.7; rooms in house mean: 1.31 (SD 0.636); people in house mean: 6.50 (SD 3.26) Social capital: HHs with husband, %: 74.6 Nutritional status: mean calories consumed per adult male equivalent mean coefficient: 1662.8 (SD 888.9); proportion of HHs without enough food in past 7 days: 0.407 Morbidities: NR Concomitant or previous care: NR Overall: NR Inclusion criteria: for cRCT (not overall Cash for Nutrition programme): residents in 1 of the 3 targeted districts in Al Hodeidah (Marawi'ah, Bayt Al Faqiah or Zabid); mothers of children aged < 2 years and pregnant women; indirect relatives of Social Welfare Fund beneficiaries Exclusion criteria: NR Pretreatment: intervention HHs had a lower Asset Index mean than control HHs; fewer intervention HHs owned TVs and washing machines compared to controls; and more mothers in intervention HHs were illiterate compared to mothers in control HHs. Attrition per relevant group: at baseline, communities that included indirect relatives of Social Welfare Fund beneficiaries were randomly assigned to the intervention (95 communities, 1001 HHs) or control (95 communities, 999 HHs) groups. The randomisation only applied to HHs of INDIRECT relatives of Social Welfare Fund beneficiaries – DIRECT relatives all received the intervention, and were not included in the impact analysis. In the intervention group, 935 HHs in 95 communities were resurveyed at follow‐up (attrition rates: HHs 6.59%, community 0.0%). In the control group, 915 HHs in 93 communities were resurveyed at follow‐up (attrition rates: HHs 8.41%, communities 2.1%). Total HH attrition rate was 7.5%, slightly higher among the treatment sample, but not significantly different between treatment and control (Kurdi 2019, Sample and Methodology – Sample size and attrition). Description of subgroups measured and reported: heterogeneity of impacts for baseline HH wealth, baseline sources of information, women's position in the HH, women's educational level. Total number completed and analysed per relevant group: with replacement HHs at follow‐up: 935/1001 (93.4%) intervention HHs; 915/999 (91.6%) control HHs. Without replacement HHs at follow‐up: 898/1001 (89.7%) intervention HHs; 857/999 (85.8%) control HHs (communities: 95/95 (100.0%) intervention communities; 93/95 (97.9%) control communities). Total number enrolled per relevant group: 1001 women from intervention HHs (in 95 communities) and 999 women from control HHs (in 95 communities). Total number randomised per relevant group: number of HHs randomised (indirect family members of Social Welfare fund beneficiaries): 2000; intervention group: 1001 HHs in 95 communities; control group: 999 HHs in 95 communities
Interventions	Intervention characteristics Cash for Nutrition intervention Food access intervention category: increase buying power Intervention type: CCT Description: first part of intervention: CCT started as a pilot in Al Hodeidah in January 2015 (Kurdi 2019, Introduction). Originally intended to last for 2 years, with payments of YER 3000 per month (Kurdi 2019, Introduction). The pilot programme provided CCTs to mothers of children aged < 2 years and pregnant women to motivate attendance at nutritional training sessions and compliance with health centre referrals (Kurdi 2019, Introduction). The pilot programme was suspended in late 2015 due to financing challenges related to the civil conflict in Yemen (Kurdi 2019, Introduction). The conditionality of the cash transfers was based on the attendance of the beneficiaries of the programme at nutrition‐focused training sessions and compliance with child monitoring and treatment of malnutrition (Kurdi 2019, Executive summary). The monthly sessions covered topics on infant and young child feeding practices, including exclusive breastfeeding, for children aged ≤ 6 months, complementary feeding for children aged 6–24 months, the importance of balanced meals, use of iodised salt, proper hygiene and sanitation, appropriate treatment of drinking water, and treatment of diarrhoea. Additional quarterly sessions targeted pregnant and lactating women and covered breastfeeding initiation, the importance of colostrum and no prelacteal feeds, as well as the consequences of consuming the stimulant qat (Catha edulis), smoking during pregnancy, hygiene and sanitation and treatment of drinking water. Pregnant women were also referred to the nearest health centre for antenatal care. In addition, under Yemen's Social Fund for Development, periodic screening sessions during home visits were carried out to detect and refer cases of malnutrition to health centres for treatment (Kurdi 2019, Introduction). Cash for Nutrition programme beneficiaries were required to attend these sessions and attendance was tracked, although the conditionality was not strictly enforced (Kurdi 2019, Introduction). Second part of the intervention: in the last quarter of 2017 an expanded version of the pilot programme was included in the World Bank funded Yemen Emergency Crisis Response Project. Duration of intervention period: January–December 2015; October 2016 to August 2017 Frequency: quarterly transfers during January–December 2015, monthly transfers during October 2016 to August 2017, monthly nutritional training sessions from January–December 2015 and nutritional training sessions for 12 months from October 2016 to August 2017 Number of study contacts: surveys completed at baseline (December 2014 to January 2015) and follow‐up (July–August 2017) Providers: Yemen Social Fund for Development in coordination with the Ministry of Public Health and Population, and the Yemen Emergency Crisis Response Project (funded by the World Bank) Delivery: local women with at least a high school education were selected as community health volunteers and received basic training in health and nutrition education and malnutrition screening. The volunteers were employed to provide monthly educational sessions and monitor the children of participating HHs (Kurdi 2019, Introduction). Data collection for the baseline survey was done by an independent survey organisation, Prodigy, while for the follow‐up survey the data collection was managed directly by the Social Fund for Development due to the challenges of conducting survey fieldwork during the conflict (Kurdi 2019, Sample and methodology – survey). Co‐interventions: unspecified other food distribution programmes. Resource requirements: transfers of YER 3000 per HH per month for January–December 2015, and YER 10,000 (about USD 30 at the time of conflict) per HH per month for October 2016 to August 2017. Requirements for nutritional training NR. Economic indicators: NR Control: no intervention Co‐interventions: unspecified other food distribution programmes.
Outcomes	Diet diversity: HDDS (0–12) WHZ; HAZ
Identification	Sponsorship source: managed by the World Bank and funded by the Nordic Trust Fund. Data collection funded by World Bank and United Nations Development Programme. Country: Yemen Setting: poor and vulnerable HHs in 3 eligible districts in Al Hodeidah governorate: Marawi'ah, Bayt Al Faqiah and Zabid (Kurdi 2019, Introduction). Comments: N/A Author's name: Sikandra Kurdi Institution: N/A Email: s.kurdi@cgiar.org; ifpri@cgiar.org Address: N/A Declarations of interest: NR Study or programme name and acronym: Cash for Nutrition programme Type of record: impact evaluation report
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (Selection bias)	Unclear risk	Reported that communities were randomised to either the intervention or control groups, but the method was not specified.
Allocation concealment (Selection bias)	High risk	Allocation of communities of HHs to intervention and control groups was done for indirect family members of Social Welfare Fund beneficiaries. However, direct beneficiaries in all the communities all received the intervention. Therefore, it was not possible (and there does not seem to have been attempts made) to conceal group allocation.
Baseline characteristics similar (Selection bias)	High risk	Intervention HHs had significantly lower Asset Indices (P = 0.001), and owned significantly fewer televisions (P < 0.01) and washing machines (P = 0.007). Intervention HHs also had a significantly higher percentage of illiterate mothers (P < 0.001).
Baseline outcome measurements similar (Selection bias)	Unclear risk	No formal hypotheses to test baseline outcome values reported.
Blinding of participants and personnel (Performance bias)	Low risk	Given the nature of the study design, it was unlikely that blinding of participants and training volunteers was possible. Some outcomes, such as health and nutrition behaviour and spending on food, may have been prone to performance bias. It was not clear, however, whether participants and volunteers were aware that the survey was recording their behaviour.
Blinding of outcome assessment (Detection bias)	High risk	It is NR whether people administering the survey and taking anthropometric measurements were blinded to assignment. It is also unclear whether participants, the outcome assessors for self‐reported measures, were blinded.
Protection against contamination (Performance bias)	High risk	The authors reported high levels of contamination. During the survey, 16% of intervention HHs reported not receiving the transfer at baseline and 7% at follow‐up. 23% of control HHs reported receiving the cash transfer at baseline and 24% follow‐up. It is not clear how large the role of contamination was for non‐recipient HHs located close to recipient HHs in the same communities.
Incomplete outcome data (Attrition bias)	High risk	Attrition 7.5% overall and fairly balanced across intervention (6.59%) and control groups (8.41%). The use of replacement HHs for those who could not be resurveyed at follow‐up is reported, but it was assumed that these HHs are in addition to those HHs 're surveyed at follow‐up'. However, 2 entire control clusters were lost.
Selective outcome reporting (Reporting bias)	Unclear risk	No protocol available for review
Other bias	Low risk	Incorrect analysis: low risk for SEs for model coefficients clustered at community level; recruitment bias: low risk for recruitment prior to randomisation.