Schwab 2013.

Study characteristics
Methods	Study design: cRCT Study grouping: parallel group How were missing data handled? Authors reported that data were analysed as ITT. However, due to discrepancies in timing of implementation of the interventions in both groups, with the food group receiving the last instalment 15 days before the endline survey whereas in the cash group the transfer was received 49 days before, some HHs were excluded from the analysis: for outcomes that rely on a recall period including the week before the survey (e.g. days in the last 7 the HH consumed meat), the analysis excluded those HHs that received their transfer in the 8‐day period before the survey. Other data were also excluded. This analysis conducted throughout this report was restricted to 3353 treatment and comparison HHs for whom consistent data from both the baseline and endline surveys exists. In addition, the remaining HHs not considered here had extensive incomplete, missing or unreliable data for key sections. Randomisation ratio: 1:1. The 136 FDPs within the sample area were randomised into equal numbers of cash or food transfers. Recruitment method: NR. Initial meetings with beneficiaries were held in June 2011 before the first transfer of the 2011 cycle was distributed to sensitise beneficiaries to the programme objectives and logistics. A follow‐up meeting for cash beneficiaries was held in November 2011 during the first disbursement of cash transfers. Sample size justification and outcome used: NR. Based on the distribution of clusters in the treatment groups and the required sample sizes, 15 intervention HHs and 11 non‐beneficiary HHs were randomly selected to be interviewed in each FDP. In total, 3536 HHs were included in the baseline sample. Approximate sample size calculations were conducted across countries at the inception of the study and are found in Ahmed et al. (2010). (Note: we were unable to find this reference.) Sampling method: 14 governorates were chosen to implement the ESN based on the classifications of ≥ 10% of the population as severely food insecure, with the end objective of reaching ≥ 75% of this population at the governorate level. The governorates of Hajjah and Ibb were chosen to be the sites of the cash and voucher pilot based on several criteria. These governorates are second‐ and third‐ranked among the 14 governorates implementing the ESN in terms of absolute numbers of food‐insecure people. In addition, Hajjah and Ibb had high percentages of the food‐insecure (Hajjah: 46.3% and Ibb: 44.0%, according to the 2009 CFSS), as well as relative stability and implementation feasibility (WFP‐CO Yemen 2011a). The 136 FDPs within the sample area were randomised into equal numbers of cash or food transfers. Taking into consideration the context of the project area, the study authors stratified the randomisation of clusters at the governorate level due to the distinct socioeconomic and geographic characteristics of Hajjah and Ibb. Based on the distribution of clusters in the treatment groups and the required sample sizes, 15 intervention HHs and 11 non‐beneficiary HHs were randomly selected to be interviewed in each FDP. In total, 3536 HHs were included in the baseline sample. Unclear who the non‐beneficiary HHs were; they were not mentioned elsewhere. Study aim or objective: in order to provide rigorous evidence on the relative impact and cost‐effectiveness of cash and food transfers, the study authors analysed the results of a cRCT of a seasonal safety net programme implemented by the WFP in rural Yemen. The analysis focused on the relative effectiveness of food and cash transfers. The analysis focused primarily on the differential impacts of these transfer types on food security outcomes. Study period: 7 months. Start and end date of study not clearly reported but it started sometime in 2011, with the first cash disbursement in November 2011. Initial meetings with beneficiaries were held in June 2011 before the first transfer of 2011. A follow‐up meeting for cash beneficiaries was held in November 2011 during the first disbursement of cash transfers. Unit of allocation or exposure: clusters; FDPs (villages)
Participants	Baseline characteristics Intervention or exposure Age: HH head, mean, years: 47.59. HH members aged 0–5 years, mean, n: 1.2. HH members aged 6–17 years, mean, n: 3.89 Place of residence: proportion in Hajjah governorate, %: 51 Sex: proportion of female headed HHs, %: 21 Ethnicity and language: NR Occupation: NR Education: proportion of HHs where HH head attended primary school or higher, %: 27 SES: Standardized Wealth Index, mean: 0.07 SDs (Note: other SES indicators included in Table 2 of publication) Social capital: NR Nutritional status: NR Morbidities: NR Concomitant or previous care: NR Control Age: HH head, mean, years: 47.06. HH members age 0–5, mean, n: 1.23. HH members age 6–17 years, mean, n: 4.00 Place of residence: proportion in Hajjah governorate, %: 49 Sex: proportion of female‐headed HHs, n: 17 Ethnicity and language: NR Occupation: NR Education: proportion of HHs where HH head attended primary school or higher, %: 25 SES: Standardized Wealth Index, mean: –0.02 SDs (Note: other SES indicators included in Table 2 of publication) Social capital: NR Nutritional status: NR Morbidities: NR Concomitant or previous care: NR Overall: NR Inclusion criteria: both treatment groups, only need‐eligible HHs, as determined by a PMT, received benefits. Authors mentioned criteria for which governorates were selected: governorates are second‐ and third‐ranked among the 14 governorates implementing the ESN in terms of absolute numbers of food‐insecure people. In addition, Hajjah and Ibb have high percentages of food‐insecure people (Hajjah 46.3% and Ibb 44.0%, according to the 2009 CFSS), as well as relative stability and implementation feasibility (WFP‐CO Yemen 2011a). Exclusion criteria: NR Pretreatment: comparing the food and cash treatment groups, the randomisation appeared to function reasonably well. In terms of HH demographics, food HHs appeared relatively more likely to be headed by females and singles, although the education levels of the HH head did not significantly differ. In terms of assets, cash HHs appeared to be slightly more likely to have more telephones and own their plot of land, and they had a Wealth Index Level 0.09 SDs higher than food HHs. These differences were relatively small in magnitude, but significant at the 10% level, implying that controlling for baseline SES in the main analysis would improve the accuracy of estimated treatment effects. Note: the baseline characteristics below were taken from Table 2. (Table 3 could also have been chosen: subsample only. However, text with respect to Table 3 seemed not in line with what is represented in Table 3.) Furthermore, comparisons of means of key outcome variables at baseline were only presented for treatment (food and cash) vs comparison. Attrition per relevant group: data for 183 HHs excluded from analysis (5% of baseline sample): of the 183 HHs in the original sampling frame not included in this analysis, only 26 were omitted due to pure attrition. These 26 HHs had moved away from their location during the baseline survey, and were unable to be interviewed for the endline survey. The majority of these HHs originally resided in the Al‐Wahdah FDP in Hajjah, and were forced to move due to ongoing tribal violence, but these HHs were not included because multiple beneficiaries lived in the same HH. The remaining HHs not considered here had extensive incomplete, missing or unreliable data for key sections. Attrition per group was NR. Furthermore, for outcomes that relied on a recall period including the week before the survey (e.g. days in the last 7 the HH consumed meat), the analysis excluded those HHs that received their transfer in the 8‐day period before the survey. However, they stated that as a whole, the summary statistics did not suggest that selecting the subsample of food HHs who received the transfers > 8 days from survey time introduces discernible bias into the analysis. Description of subgroups measured and reported: dietary diversity in infants and young children: aged 6–23 months and 24–59 months Total number completed and analysed per relevant group: 1581 HHs analysed. Numbers per group NR. Total number enrolled per relevant group: total: 1983 HHs; cash group: 1001 HHs; food group: 982 HHs. Total number in baseline table 1 of publication showed 3355 HHs (this may have included a comparison group that is sometimes mentioned but this was unclear). Total number randomised according to text was 3536. Total number randomised per relevant group: total: 1983 HHs; cash group: 1001 HHs; food group: 982 HHs. Total number in baseline table 1 of publication showed 3355 HHs (this may include a comparison group that is sometimes mentioned but this was unclear). Total number randomised according to text was 3536.
Interventions	Intervention characteristics Intervention or exposure Food access intervention category: increase buying power Intervention type: food transfer Description: HHs in food FDPs received 3 food transfers, each consisting of 50 kg of wheat and 5 L of oil. The bi‐monthly food ration to cover this gap for a mean HH size of 7 people is 50 kg of wheat flour and 5.0 L of vegetable oil. Duration of intervention period: 7 months. However, the timing of transfers/disbursements was not the same in both groups. The first food disbursement began in August prior to the baseline survey, and the second transfer began in late October. The final transfer, however, did not occur until April. Frequency: transfer every 2 months Number of study contacts: 2; baseline and endline surveys of both beneficiaries and those with proxy mean scores just above the qualifying threshold were conducted in all clusters. Providers: HH‐level transfers were distributed in co‐ordination with local partners: the Ministry of Education (MoE) in the case of food transfers. Transfers were given out at district branches of the PPSC in each governorate (see Annexe 1, PPSC branches in Hajjah and Ibb). Delivery: the food transfers were stored in warehouses outside of Sana'a and distributed through local government‐run primary schools with the assistance of a FDC (see Annexe 2, FDPs in Hajjah and Ibb). The FDC is comprised of approximately 3 individuals per FDP including a school teacher from each primary school, a local council administrator and a guard. Each individual beneficiary held an WFP ration card containing a unique ID number, photograph, and other identifying information, and presented the card at the time of transfer pickup. Because beneficiaries may not always have been able to travel due to physical disability or other reasons, other family members can collect transfers on behalf of the beneficiary if they have the ration card, national ID of the beneficiary and self‐identification. Initial meetings with beneficiaries were held in June 2011 before the first transfer of the 2011 cycle was distributed to sensitise beneficiaries to the programme objectives and logistics. Co‐interventions: NR Resource requirements: the FDC is comprised of approximately 3 individuals per FDP including a school teacher from each primary school, a local council administrator and a guard. Economic indicators: food incurred higher costs for distribution and those costs associated with in‐country transport, as well as warehousing and other associated costs for commodity storage. Ocean freight, port operations and other external shipping expenses were excluded from this analysis. However, internal transportation and labour costs were included to accurately reflect the cost of food distribution in country. Cash modality was less expensive per beneficiary (USD 162.65) than the food modality (USD 181.49). These costs included beneficiary verification and the cost of the transfers itself during the 3‐cycle intervention period. On a per‐transfer basis (Figure 3), excluding the cost of the transfer, the modality‐specific cost of cash (USD 4.09) was approximately half as expensive compared to food (USD 10.37). Incorporation of the beneficiary cost to collect transfer raised the per‐transfer cost (excluding the value of the transfer) of cash to USD 8.22. For food transfers, addition of beneficiary costs raised the per‐transfer cost (excluding the transfer value) to USD 11.35. Thus, including the beneficiary costs reduces the per‐transfer cost gap from USD 6.28 to USD 3.13. Control Food access intervention category: increase buying power Intervention type: UCT Description: HHs in cash FDPs received 3 cash transfers of an amount equivalent to the local value of the food basket (about USD 50). The total value of the cash transfer was approximately USD 49 (YER 10,500) per transfer per HH, a figure based on the mean equivalent price of the food ration on local markets. Cash transfer HHs could collect cash at any time up to 25 days after disbursement. In the case of cash transfer FDPs, a second resensitisation campaign was held between 22 and 25 November after funds were transferred to PPSC to reinforce messages from the first campaign. Duration of intervention period: 7 months. However, the timing of transfers/disbursements was not the same in both groups. The first food disbursement began 3 August prior to the baseline survey, and the second transfer began in late October. The final transfer, however, did not occur until April. Frequency: 3 cash transfers; every 2 months Number of study contacts: 2; baseline and endline surveys of both beneficiaries and those with proxy mean scores just above the qualifying threshold were conducted in all clusters. Providers: HH‐level transfers are distributed in coordination with local partners: the Yemen PPSC in the case of cash transfers. Transfers are given out at district branches of the PPSC in each governorate (see Annexe 1, PPSC branches in Hajjah and Ibb). Delivery: initial meetings with beneficiaries were held in June 2011 before the first transfer of the 2011 cycle was distributed to sensitise beneficiaries to the programme objectives and logistics. A follow‐up meeting for cash beneficiaries was held in November 2011 during the first disbursement of cash transfers. Cash transfer HHs could collect cash at any time up to 25 days after disbursement. Co‐interventions: NR Resource requirements: HHs had to invest income in significant travel to receive the cash transfer. Economic indicators: a primary cost driver for cash was the 3% fee of total cash transferred each cycle as incurred by the post office. Cash modality was less expensive per beneficiary (USD 162.65) than the food modality (USD 181.49). These costs included beneficiary verification and the cost of the transfers itself during the 3‐cycle intervention period. On a per‐transfer basis (Figure 3), excluding the cost of the transfer, the modality specific cost of cash (USD 4.09) was approximately half as expensive as compared to food (USD 10.37). Incorporation of the beneficiary cost to collect transfer raises the per‐transfer cost (excluding the value of the transfer) of cash to USD 8.22. For food transfers, addition of beneficiary costs raises the per‐transfer cost (excluding the transfer value) to USD 11.35. Thus, including the beneficiary costs reduces the per‐transfer cost gap from USD 6.28 to USD 3.13.
Outcomes	Food expenditure Food security: days in the past week that HHs/adults/children were required to reduce the amount of food consumed at or frequency of meals consumed; months in the previous 6 that HHs had difficulty satisfying their food needs Dietary diversity: HDDS (number of distinct food categories consumed by the HH in the previous 7 days); probability of having a low FCS (< 28.5)
Identification	Sponsorship source: government of Spain. Country: Yemen Setting: poor HHs in rural communities in poorest districts in Yemen Author's name: Benjamin Schwab Email: b.schwab@cgiar.org Declarations of interest: NR Study or programme name and acronym: NR Type of record: report
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (Selection bias)	Unclear risk	Not described how randomisation was conducted other than (quote) "In the intervention under study, 136 village clusters (known as Food Distribution Points) were randomly assigned to receive either food or cash assistance."
Allocation concealment (Selection bias)	Low risk	Allocation was at cluster level, by FDP.
Baseline characteristics similar (Selection bias)	Unclear risk	There were many disparities in baseline characteristics between the groups. Unclear whether these had been adjusted for in all analyses. Comparing the food and cash treatment groups, the randomisation appeared to function reasonably well. In terms of HH demographics, food HHs appeared relatively more likely to be headed by a females and singles, although the education levels of the HH head did not significantly differ. In terms of assets, cash HHs did appear to be slightly more likely to have more telephones and own their plot of land, and they have a Wealth Index Level 0.09 SDs higher than food HHs. These differences were relatively small in magnitude, but significant at the 10% level, implying that controlling for baseline SES in the main analysis will improve the accuracy of estimated treatment effects. With respect to the subgroup: comparing those included and those excluded within the food treatment group (Table 3), it becomes clear that while the excluded group was slightly more likely to come from Hajjah, differences in other indicators were generally not significant economically and statistically. The lone exception was that the excluded group had higher motor vehicle ownership rates (5% vs 1%), and slightly higher wealth index levels (not statistically significant). As a whole, the summary statistics did not suggest that selecting the subsample of food HHs who received the transfers > 8 days from survey time introduced discernible bias into the analysis.
Baseline outcome measurements similar (Selection bias)	Unclear risk	Despite the fact that the main analysis was comparing the food and cash transfers, the comparison of baseline outcome measures was only presented for treatment (food and cash) vs comparison (Table 5). The baseline data were not true baseline as the food transfer group had already received the first transfer.
Blinding of participants and personnel (Performance bias)	Low risk	No blinding was possible, but this was unlikely to influence the delivery of the intervention.
Blinding of outcome assessment (Detection bias)	High risk	Blinding was not possible. Unclear who the outcome assessors were. Some outcomes were self‐reported and could thus have been influenced by the knowledge of treatment allocation, especially as preferences for cash vs food among participants varied from baseline to endline, with majority preferring cash at endline.
Protection against contamination (Performance bias)	Low risk	Communities were randomised. Could only get the relevant intervention when having the correct ID.
Incomplete outcome data (Attrition bias)	High risk	183 HHs were excluded from the analysis (5% of baseline sample). At the HH level, there was high attrition: the number analysed was 1581/1983 at baseline (20%). The numbers missing/excluded per group were not provided.
Selective outcome reporting (Reporting bias)	Unclear risk	The paper outlining the methods could not be accessed. Authors did clearly report all the outcomes of interest in the methods section of the report.
Other bias	High risk	There might have been bias introduced due to the differences in the timing of the interventions. Quote: "changes in timing of the survey and distribution schedule resulted in the loss of a pure pre‐intervention survey, as the baseline survey occurred after the first food transfer (but before the first cash transfer). Ideally, the disbursement schedules should be identical so that differences in impact can be attributed to difference between the modalities rather than differences in seasonal or other environmental factors influencing budgeting and resource flows within the household, or discrepancies in the period between transfer receipt and survey measurement." Misclassification bias of exposure: low risk. Measurement bias: unclear. It is unclear who collected data and how it was done. Incorrect analysis: low risk. Analyses were adjusted for clustering. Misclassification bias of exposure: low risk. Measurement bias: unclear. Unclear who collected data and how it was done. Incorrect analysis: low risk. Analyses were adjusted for clustering.