Merttens 2013.

Study characteristics
Methods	Study design: cRCT Study grouping: parallel group How were missing data handled? (quote) "Column 8 shows the number of observations at follow‐up (FU1) which is 2,867 (the sample of households comprising the treatment plus control panel) minus any missing values." Therefore, where there were missing data, the sample size for that outcome was reduced accordingly. Probability of retention was estimated using regression, and weighting of HHs was adjusted by inverse retention probability to adequately represent attrition HHs. Randomisation ratio: 1:1 Recruitment method: evaluation covered the 4 former districts of Mandera, Marsabit, Turkana and Wajir, in 12 randomly selected sublocations in each district. The sublocations that were covered by the evaluation were referred to as the evaluation sublocations. The HSNP applied a staggered roll‐out, with sublocations being brought into the Programme on a month‐by‐month basis. Sample size justification and outcome used: calculated based on the expected sampling error for point estimates, differences and DID for key indicators. These key indicators were not explicitly reported. A sample attrition buffer of 10% was factored in to compensate for the high risk of sample attrition. Sampling method: intervention and control HHs were sampled from HSNP administrative records. From these, 66 HHs were sampled from each sublocation using simple random sampling. In the event of non‐response, a replacement HH which had not yet been drawn was randomly selected from the administrative record. The sequence in which the sampled evaluation sublocations were targeted and surveyed was determined randomly. The evaluation sublocations were selected from a sample frame of all secure sublocations in each district. The evaluation sublocations were sorted within new districts by population density and paired up, with 1 of the pair being control and 1 being treatment. The reason sublocations were sorted (within each new district) by population density before pairing them up was to ensure that similar sublocations were matched to reduce as far as possible significant variations between the characteristics of the control and treatment groups. For both the treatment and control sublocations there were an equal number of CBT, SP and DR sublocations. Study aim or objective: to evaluate the HSNP programme's impact on consumption expenditure and poverty reduction, food security, and increased asset retention and accumulation. Secondary impacts namely uptake of health and education services, stabilised food prices, supplies of key commodities, diversity of livelihood activities, financial savings, vulnerability to shocks, empowerment of women, and well‐being of the young and elderly people were also assessed. Study period: August 2009 to November 2012 Unit of allocation or exposure: sublocations (geographical areas within counties)
Participants	Baseline characteristics Intervention or exposure Age: HHs containing ≥ 1 child aged < 18 years, %: 91.9; HHs containing ≥ 1 orphan child, %: 21.6; HHs containing ≥ 1 chronically ill member, %: 11.7; HHs containing ≥ 1 disabled member, %: 8.4% Place of residence: HHs in communities with, %: a primary school: 52.5; a health facility: 29.8 Sex: HHs that are female‐headed, %: 34.0 Ethnicity and language: NR Occupation: adults in productive work, %: main activity 53.8; main or secondary activity; 58.5; children aged 5–17 years whose main activity was paid or unpaid work, %: 22.4 Education: children currently attending school, %: all children aged 6–17 years: 63.2; female children aged 6–17 years: 57.5; male children aged 6–17 years: 68.3; all children aged 6–12 years: 63.9; all children aged 13–17 years: 62.0; children whose main activity was education, %: 69.1 SES: proportion of HHs below absolute poverty line, %: 88.0; mean HH consumption expenditure: KES 1941; mean HH food consumption expenditure: KES 1446; proportion of HHs, %: food insecure in worst recent food shortage 61.8; receiving food aid 70.5; receiving school feeding 57.2; receiving supplementary feeding 16.5; owning agricultural land 9.5; owning livestock 61.5; mean number of children aged < 18 years per HH: 3.4; mean HH size: 6.0 Social capital: NR Nutritional status: mean HH DDS: 6.7; children aged < 5 years with, %: moderate stunting 26.7; severe stunting 11.6; moderate wasting 25.3; severe wasting 6.8; moderate underweight 30.7; severe underweight 9.8; proportion of HHs which in the last 30 days, %: had a reduced number of meals 77.5; had smaller meals 74.5; skipped eating for entire days 57.7. Morbidities: children aged 0–17 years who were ill or injured in previous 3 months, %: 20.0; adults aged ≥ 55 years who were ill or injured in previous 3 months, %: 36.6 Concomitant or previous care: HHs that received informal in‐kind transfers in previous 3 months, %: 41.3; HHs that received informal cash transfers in previous 3 months, %: 45.6 Control Age: HHs containing ≥ 1 child aged < 18 years, %: 91.5; HHs containing ≥ 1 orphan child, %: 22.7; HHs containing ≥ 1 chronically ill member, %: 15.4; HHs containing ≥ 1 disabled member, %: 9.0 Place of residence: proportion of HHs (%) in communities with: a primary school 52.0; a health facility 24.2 Sex: proportion female‐headed HHs, %: 30.9 Ethnicity and language: NR Occupation: proportion adults in productive work, %: main activity 58.5; main or secondary activity 63.5; children aged 5–17 years whose main activity was paid or unpaid work, %: 29.2 Education: children currently attending school, %: all children aged 6–17 years: 42.6; female children aged 6–17 years: 37.5; male children aged 6–17 years: 47.1; all children aged 6–12 years: 42.0; all children aged 13–17 years: 43.4; children whose main activity was education, %: 58.3 SES: proportion of HHs below absolute poverty line, %: 93.2; mean HH consumption expenditure: KES 1753; mean HH food consumption expenditure: KES 1385; proportion of HHs, %: food insecure in worst recent food shortage 74.8; receiving food aid 88.7; receiving school feeding 53.7; receiving supplementary feeding 10.6; owning agricultural land 7.1; owning any livestock 85.1; mean number of children aged < 18 years per HH: 3.0; mean HH size: 5.5 Social capital: NR Nutritional status: mean HH DDS: 6.1; children aged < 5 years with, %: moderate stunting 35.6; severe stunting 15.2; moderate wasting 24.2; severe wasting 8.0; moderate underweight 33.7; severe underweight 10.9; proportion of HHs which in the last 30 days, %: had a reduced number of meals 89.0; had smaller meals 87.8; skipped eating for entire days 72.7 Morbidities: children aged 0–17 years who were ill or injured in previous 3 months, %: 20.1; adults aged ≥ 55 years who were ill or injured in previous 3 months, %: 36.6 Concomitant or previous care: HHs that received informal in‐kind transfers in previous 3 months, %: 42.4; HHs that received informal cash transfers in previous 3 months, %: 39.1. Overall: NR Inclusion criteria: selected for the HSNP programme (whether currently receiving or on the waiting list), which included: community identification as a HH in need of cash transfer, or HHs with proportion of its members aged < 18 and > 55 years exceeding a prespecified dependency ratio, or individuals aged ≥ 55 years. Exclusion criteria: ineligible for the HSNP programme. Pretreatment: lower proportion of intervention HHs were food insecure in the worst recent food shortage when compared to control HHs. Higher proportions of intervention HHs were receiving food aid when compared to control HHs, and a lower proportion of intervention HHs had children aged < 5 years with moderate stunting. A lower proportion of intervention HHs reported having skipped eating for entire days in the past 30 days. School attendance for all ages and both sexes was higher in intervention HHs. Significantly more control HHs owned livestock when compared to intervention HHs. Attrition per relevant group: follow‐up (November 2011): 9% (137/1571) for intervention and 7% (103/1536) for control HHs. From baseline at second follow‐up (November 2012): 22% (347/1571) for intervention and 21% (324/1536) for control HHs. Description of subgroups measured and reported: NR. Total number completed and analysed per relevant group: for the final impact analysis there were 20 treatment and 20 comparison (control) sublocations, with 1224 intervention HHs and 1212 control group HHs. Total number enrolled per relevant group: total of 1571 intervention and 1536 control HHs at baseline. Total number randomised per relevant group: just over 5000 HHs were randomly selected at baseline (prior to programme roll‐out) for interview on an annual basis in 48 evaluation sublocations (24 treatment and 24 control), also selected at random.
Interventions	Intervention characteristics Intervention or exposure Food access intervention category: increase buying power Intervention type: UCT transfer Description: unconditional transfer of KES 2150 (at commencement) which increased to KES 3500 by the end of the intervention period. The value of the transfer was determined as 75% of the WFP food aid ration in 2006 (when the value was determined). Some HHs had multiple nominated beneficiaries; the effective value of the transfer per HH member was smaller for larger HHs. Duration of intervention period: 2 years (November 2010 to November 2012) Frequency: every 2 months Number of study contacts: impact evaluation data collected over 3 rounds comprising a baseline round (August 2009 to November 2010), follow‐up 1 round (November 2010 to November 2011), and follow‐up 2 round (February 2012 to November 2012). Providers: operated under the Ministry of State for the Development of Northern Kenya and Other Arid Lands and was delivered by several contracted service providers. Delivery: cash was loaded onto a biometric smart card that could be used to collect the cash transfer from a range of pay points (usually small shops). Co‐interventions: NR Resource requirements: NR Economic indicators: for two‐thirds of beneficiary HHs the transfer had a per capita value of KES 350–700. Control Food access intervention category: delayed‐intervention (waiting list) control (non‐active control at the time of data collection) Intervention type: N/A Description: control HHs only began to receive cash upon completion of the final round of data collection (follow‐up 2 survey), i.e. 2 years after the baseline survey. Duration of intervention period: N/A Frequency: N/A Number of study contacts: baseline data collection (November 2010) and 2 follow‐up interviews (November 2011 and 2012) Providers: N/A Delivery: N/A Co‐interventions: NR Resource requirements: NR Economic indicators: NR
Outcomes	Proportion of HH expenditure on food: proportion of consumption expenditure spent on food; monthly food consumption expenditure per AE Food security: proportion of HHs food insecure in worst recent food shortage period Dietary diversity: HDDS (0–12) Anthropometry: moderate and severe wasting; moderate and severe stunting; moderate and severe underweight Morbidity: proportion of people ill/injured in previous 3 months
Identification	Sponsorship source: DfID. Country: Kenya Setting: impoverished rural HHs Author's name: Fred Merttens Email: fred.merttens@opml.co.uk; admin@opml.co.uk Declarations of interest: NR Study or programme name and acronym: Hunger Safety Net Programme (HSNP) Type of record: impact evaluation report
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (Selection bias)	Low risk	Quote: "Forty‐eight programme sub‐locations were then randomly selected from the pool of all programme sub‐locations and then from those each pair were randomly assigned between treatment and control at a public lottery event (bahati na sibu) facilitated by the HSNP Secretariat and attended by officials from the district and the two sub‐locations in question."
Allocation concealment (Selection bias)	Low risk	The unit of allocation was 'sublocations', stratified by district. Allocation was performed on all units at the start of the study.
Baseline characteristics similar (Selection bias)	High risk	Important differences between the number of children going to school (intervention group: 63.2%; control group: 42.6%), and the number of HHs owning any livestock (intervention group: 61.5%; control group: 85.1%) were present and not adjusted for in analyses.
Baseline outcome measurements similar (Selection bias)	High risk	Some outcomes were not equivalent at baseline: intervention HHs were significantly more food secure in the worst recent food shortage and significantly more received food aid (both P = 0.05). Intervention HHs also had significantly fewer moderately stunted children aged < 5 years (P = 0.05) and significantly less reported skipping eating for entire days (P = 0.10). Proportion of children attending school (for all ages and both sexes) was significantly higher for intervention HHs (P = 0.05). The HHs for which the baseline characteristics were reported differed from the HHs that were analysed in the year 2 analysis, so it is unclear whether outcome measurements of those analysed have been similar.
Blinding of participants and personnel (Performance bias)	Low risk	Participants were not blinded, but this was unlikely to have influenced participant behaviour.
Blinding of outcome assessment (Detection bias)	High risk	Unclear whether the data collectors were blinded towards the group allocations, which may or may not have influenced the measurement of outcomes across groups. Self‐reported outcomes may have been prone to reporting bias.
Protection against contamination (Performance bias)	Low risk	Given the geographical separation between intervention and control sublocations, the risk of contamination was low. Participants were also selected from a list that identified current recipients and waiting list recipients. According to the report itinerant/mobile people were more likely to attrite from the sample, and, therefore, unlikely to contaminate conditions due to their migration.
Incomplete outcome data (Attrition bias)	High risk	Quote: "The reduction in the number of sublocations surveyed at follow‐up 2 was the result of decisions made by the programme and its stakeholders, rather than a technical decision by the evaluation team. This reduction in sample size is unfortunate for a number of reasons …" Comment: in the intervention group, 1224/1571 (77.9%) and in the control group 1212/1536 (78.9%) HHs were analysed. Although the percentage attrition per group was similar, the total attrition was > 20%.
Selective outcome reporting (Reporting bias)	Unclear risk	No protocol or trial registration reported. All expected outcomes were prespecified in methods section and addressed in results section.
Other bias	Unclear risk	Misclassification bias: high risk. Loss of 8 clusters (sublocations) – a programme and stakeholder decision. Measurement bias: incorrect analysis: unclear. It is unclear whether these findings were comparable to individually randomised trials. Low risk of bias due to clustering, as adjustment for clustering was performed. Seasonality bias: high risk. The 8 sublocations that were dropped were scheduled to be surveyed in the end and beginning of the calendar year. Recruitment bias: unclear. Could not be ruled out as it is not explicitly stated whether randomisation, in the form of a public lottery, preceded recruitment.