Findings from a cluster randomised trial of unconditional cash transfers in Niger

Abstract Unconditional cash transfers (UCTs) are used as a humanitarian intervention to prevent acute malnutrition, despite a lack of evidence about their effectiveness. In Niger, UCT and supplementary feeding are given during the June–September “lean season,” although admissions of malnourished children to feeding programmes may rise from March/April. We hypothesised that earlier initiation of the UCT would reduce the prevalence of global acute malnutrition (GAM) in children 6–59 months old in beneficiary households and at population level. We conducted a 2‐armed cluster‐randomised controlled trial in which the poorest households received either the standard UCT (4 transfers between June and September) or a modified UCT (6 transfers from April); both providing 130,000 FCFA/£144 in total. Eligible individuals (pregnant and lactating women and children 6–<24 months old) in beneficiary households in both arms also received supplementary food between June and September. We collected data in March/April and October/November 2015. The modified UCT plus 4 months supplementary feeding did not reduce the prevalence of GAM compared with the standard UCT plus 4 months supplementary feeding (adjusted odds ratios 1.09 (95% CI [0.77, 1.55], p = 0.630) and 0.93 (95% CI [0.58, 1.49], p = 0.759) among beneficiaries and the population, respectively). More beneficiaries receiving the modified UCT plus supplementary feeding reported adequate food access in April and May (p < 0.001) but there was no difference in endline food security between arms. In both arms and samples, the baseline prevalence of GAM remained elevated at endline (p > 0.05), despite improved food security (p < 0.05), possibly driven by increased fever/malaria in children (p < 0.001). Nonfood related drivers of malnutrition, such as disease, may limit the effectiveness of UCTs plus supplementary feeding to prevent malnutrition in this context. Caution is required in applying the findings of this study to periods of severe food insecurity.


| INTRODUCTION
Wasting (defined as weight for height < −2 Z scores) accounts for 14.6% of mortality among children under 5 years old (McDonald et al., 2013). Nearly a quarter of the world's nations have a prevalence of acute malnutrition (defined as weight for height < −2 Z scores or nutritional oedema) of at least 10% (United Nations Childrens' Fund et al., 2012), which is an "emergency requiring immediate intervention" (World Health Organization, 2000). The highest prevalence of acute malnutrition is in south Asia and sub-Saharan Africa (United Nations Childrens' Fund, World Health Organization, & The World Bank, 2012), regions with frequent shocks such as natural disasters (Centre for Research on the Epidemiolgy of Disasters, 2015) and complex emergencies (Spiegel, Le, Ververs, & Salama, 2007), which often require humanitarian assistance.
Effective prevention of acute malnutrition requires addressing its causes, including food insecurity, deficiencies in the social/care environment, disease, and inadequacies in environmental health.
However, in many humanitarian settings, food-based approaches such as general food distributions and supplementary feeding, predominate (Bailey & Hedlund, 2012) despite limited evidence of effectiveness (Sguassero, Onis, Bonotti, & Carroli, 2012), implementation challenges (Hall, Oirere, Thurstans, Ndumi, & Sibson, 2011; US Centers for Disease Control and Prevention, 2013) and concerns about cost-effectiveness (Puett et al., 2013). These limitations have contributed to an increase in short-term Unconditional Cash Transfer (UCT) and voucher interventions (Harvey, Proudlock, Clay, Riley, & Jaspars, 2010;Overseas Development Institute, 2015;World Food Programme, 2010). However, although cash/vouchers have the potential to address multiple causes of malnutrition in contexts with functional markets and adequate supply (Leroy, Ruel, & Verhofstadt, 2009), evidence for nutritional impact in humanitarian contexts is inconclusive. Most studies have focused on linear growth among beneficiaries of longer-term Conditional Cash Transfers (CCTs) in development contexts (Manley, Gitter, & Slavchevska, 2013;Bastagli et al., 2016). A recent review of 14 studies of CCTs and UCTs in low and middle income countries that assessed anthropometry found that 9 had no impact on anthropometric outcomes. Furthermore, only five measured wasting and of these, only one (a CCT in a mixed rural/urban setting in Bangladesh) documented significant improvement (Bastagli et al., 2016).
In humanitarian settings in particular, cash is unlikely to impact on health unless access to quality services is only limited by poverty (Bailey & Hedlund, 2012).
Given the consensus that cash transfers are unlikely to affect nutritional status when implemented in isolation (Bailey & Hedlund, 2012;Bhutta et al., 2013;de Groot et al., 2015), studies in contexts prone to shocks have mostly tested the effect of cash combined with supplementary feeding or nutrition-related education, finding combinations more effective than cash or food alone (Ahmed, Quisumbing, Nasreen, Hoddinott, & Bryan, 2013;Langendorf et al., 2014). The lack of compelling evidence regarding the impact of cash transfers on nutrition-related indicators is likely due to differences in programmatic context, implementation, and design (Bastagli et al., 2016), including targeting, amount, duration, and timing (Arnold et al., 2011;Manley et al., 2013). More studies are therefore required to build the evidence base on where, when, and how cash-based interventions are effective against acute malnutrition.
Seasonal UCTs have been implemented by international nongovernmental organisations in Niger since 2008 with humanitarian funding to coincide with the preharvest lean season. However, there is inconclusive evidence of their nutritional impact and questions about how to optimise their design remain (Aker, Boumnijel, McClelland, & Tierney, 2011;Aker & Nene, 2012;Bliss et al., 2016;Bliss & Golden, 2013;Fenn, Noura, Sibson, Dolan, & Shoham, 2014;Langendorf et al., 2014;Poulsen & Fabre, 2011;Save the Children, 2009). Furthermore, monthly trends in the admission of acutely malnourished children to feeding programmes indicate that acute malnutrition incidence may not coincide with the lean season, as admissions may rise before and/or after ( Figure S1.1). There was also an absence of studies on the nutritional impact of this targeted intervention at the population level, despite the assumption of operational agencies that socio-economic targeting will prevent rising acute malnutrition prevalence.
We tested whether starting the UCT 2 months earlier, but providing the same total amount of cash over 6 months instead of 4,

Key messages
• Starting the UCT earlier and providing the same amount of cash over 6 months instead of 4, alongside 4 months supplementary feeding, temporarily increased beneficiary food security, but did not impact on children's nutritional status at end line.
• GAM prevalence remained static and elevated, despite improved food security; probably due to a deteriorating health situation.
• Strengthening interventions to tackle malaria, as well as providing seasonal cash and food, may better protect children from acute malnutrition.
• Future studies should test combined health and food security interventions and explore the assumption that the targeting of low income households leads to population level impact.
alongside supplementary food for pregnant and lactating women and children 6-<24 months between June and September, would reduce the prevalence of acute malnutrition in children in targeted households and in the general population. The study protocol was published in 2015 (Sibson et al., 2015).

| Setting
The study setting was the rural communes of Affala and Takanamatt in the department of Tahoua, southwest Niger. Hausa is the largest ethnic group, followed by Tuareg and Fulani. The sedentary, agropastoral communities rely on the single, unpredictable rainy season between June and September. This is also the "lean" season, when stores of crops harvested the previous year begin to run out and the prices of goods with lower availability increase in the market ( Figure S1.2). Rain-fed millet, sorghum, and cow peas are harvested September-November. However, most households produce insufficient cereals for subsistence, and also undertake daily labour, labour migration, petty trade, and borrow and sell assets to maintain food access (Anonymous, 2012). Livestock holdings are small, typically a few sheep/goats, fowl, and a donkey for the poorest, whereas the better-off may own cattle and/or camels (Anonymous, 2012).
Besides food insecurity, challenges in the public health, social, and caring environments also exist. There is a chronic shortage of water in most areas, poor hygiene practices, and few latrines . Malaria is endemic during the rains (Blanford, Kumar, Luo, & MacEachren, 2012) and diarrhoea and acute respiratory infection (ARI) rates are also high . Despite a policy of free health care for children under 5 and pregnant women, there is low utilisation due to facility inaccessibility (Blanford et al., 2012) and household prioritisation of livelihood preserving activities over health-seeking  internationally agreed indicator for the latter). Testing of hypotheses with anthropometric outcomes was conducted using adjusted regressions models, whereas a difference in difference approach was used to describe changes in potential mediating factors and nonanthropometric secondary outcomes. We also conducted a process evaluation to describe the context and fidelity of intervention implementation, to aid interpretation of the results and assess generalisability. This involved the collation of routinely collected programme and health facility M&E data, interviewing key programme staff and conducting field visits.

| Population
Our main study population of interest was children, aged 6-59 months, living in villages that had been selected to receive the UCT by humanitarian targeting criteria. All children in this age range and living in these villages were eligible for inclusion. We exhaustively sampled all beneficiary households in the targeted villages (n = 2,073), and included all children 6-59 months within. We used simple random sampling (probability proportional to cluster size) to sample 500 nonbeneficiary households in these same villages in which we also exhaustively sampled children 6-59 months old. We used Concern's village-level household census and calculated an average of 1.79 eligible children per household. We collected household level data whether or not the household had any eligible children.
Eligibility for the UCT was determined by Concern according to their usual three-step procedure: (a) selection of villages that had received cash in 2014 and/or 2013 and had a forecasted production deficit in 2015; (b) wealth ranking using household survey data (drawing on the principles of the Household Economy Approach that categorises households as "very poor," "poor," "middle," and "better off"; Boudreau et al., 2008); and (c) selection of all the very poor and as many of the poor as possible, until funds were fully allocated. The poorest households typically had less than 10 members, possessed nine chickens, five sheep/goats, no large ruminants, cultivated fewer than 4 ha, forecast fewer than 5 months yield from their last harvest and earned less than 35,000FCFA/month. This led to the UCT being targeted to 42% of households in the study villages. Concern's household definition was a group of people preparing and eating food together. There were no exclusion criteria for households but we excluded disabled children.

| Randomisation
The unit of randomisation was a cluster of villages assigned to receive UCT at the same cash distribution point (CDP), which was within 5 km of the village. We used a clustered design because the intervention was delivered at the village level. There were 25 CDPs in total and to ensure reasonable balance in sample size between arms, we grouped CDP serving smaller populations into a single cluster prior to randomisation. We also limited cluster size by excluding the four largest villages prior to randomisation. This resulted in 20 clusters containing 39 villages. Clusters were randomised in a 1:1 ratio to either the "standard" (June UCT initiation) or the "modified" (April UCT initiation) intervention. The randomisation was undertaken in a public meeting of village leaders managed by the study coordinator. The names of the villages in each cluster were written on papers, one paper per cluster, and placed into identical envelopes that participants blindly selected oneby-one for sequential allocation of clusters to the two arms.

| Interventions
The standard UCT intervention consisted of a monthly transfer of 32,500FCFA (equivalent to £36), for 4 months between June and September 2015; total 130,000FCFA (equivalent to £144). The transfer was designed to allow purchase of a food basket similar to the World Food Programme household ration (cereal, pulse, and vegetable oil) that would meet 75% of the daily energy needs of a seven-person household. The modified UCT intervention consisted of 21,500FCFA/month (£24) in April, May, July, August, and September and 22,500FCFA in June; total also 130,000CFA. In both arms, cash-in-hand was given to female household representatives.
At each distribution beneficiaries first had to attend an education session, which included suggestions on buying food for children, following which women and children were screened for acute malnutrition and provided referral slips as necessary. After this, beneficiaries received their cash. In addition, between June and September, beneficiary households in both study arms were also given 200 g/day Super Cereal Plus for each child 6-<24 months (providing 820 kcal/day) and 250 g/day Super Cereal and 75 g/day vegetable oil for each pregnant/ lactating woman (providing 1,613 kcal/day).

| Sample size
The required sample of beneficiary children, aged 6-59 months, was calculated using the clustersampsi command in Stata (Hemming & Marsh, 2013), to allow detection of a difference in endline GAM prevalence between the arms of 7% points. Based on previous data from the area, we assumed an average cluster size of 176 children with 1.8 children per household, a baseline GAM prevalence of 21% (Bliss & Golden, 2013), an ICC = 0.0138, and allowed for a 5% Type 1 error risk and 80% power. The total calculated sample size required was 3,520 children in 1,956 households, in 20 clusters. A sample of 500 nonbeneficiary households was determined pragmatically, being the maximum additional number of households we could enumerate with our available resources. This sample was allocated between clusters according to the number of households in each cluster. During data collection, 100% of beneficiary households and 17.6% of nonbeneficiary households were sampled from each cluster.

| Data collection
We collected quantitative data in March-April 2015, before the intervention (baseline), and in October-November 2015, after the intervention (endline), using structured questionnaires in Samsung Galaxy G2 7.0-inch tablets running PSI Fusion software. We trained study staff over 2 weeks before both rounds, including anthropometry standardisation tests. Data were collected at the respondents' homes.
Children's anthropometric measurements (weight, height, MUAC) were taken and recorded twice and presence/absence of oedema checked at each time point. Weight was measured to 100 g using an electronic scale (SECA model 870). Length in children <24 months and height in children ≥24 months were measured to 1 mm using a stadiometer (Infant/child/adult ShorrBoard). MUAC was measured on the left arm using a TALC-UK insertion tape to 1 mm. Peripheral blood was collected from a finger prick using a safety lancet and haemoglobin concentration was assessed to 0.1 g/dl precision, using a portable photometer HemoCue® 301 analyser. Acutely malnourished and/or anaemic children were referred for treatment. Unfortunately, we are not able to report the results of the haemoglobin analysis in this paper due to concerns about the reliability of the data.
Four-week retrospective morbidity among children was collected by caregiver recall, for a range of common symptoms and/or diseases.
In addition to primary data collection, we compiled Concern's routine quantitative monitoring data and undertook interviews with programme staff for the process evaluation. More specifically on the evaluation of intervention implementation, we used monitoring data made available by Concern that described accessibility, acceptability, timeliness, and coverage. Data on accessibility and acceptability came from distribution monitoring undertaken by Concern staff on the day of distribution (between June and September only), in which a small randomly selected sample of beneficiary women were administered a short questionnaire. Data on timeliness came from interviewing the deputy programme manager. Data on coverage came from monthly programme implementation reports.

| Data processing and analysis
We used Stata version 14 (StataCorp, 2015) for data management and analysis. Thirty-day expenditure was estimated by adding all recalled spends for the last month (mostly food stuffs) and the average spend over 30 days on additional items such as rent and health care, calculated from recalled spends made during the previous 6 months. An asset index was created using principal components analysis (Vyas & Kumaranayake, 2006) using a list of 33 assets considered neither too rare nor too common; for this we used a 5% cut-off value; for example, >95% of households had access to land so this was too common a commodity to be included in the index. The household food insecurity access scores, reduced coping strategies index scores, food consumption scores, household dietary diversity score, and months of adequate household food provisioning were calculated using standard procedures Coates, Swindale, & Bilinsky, 2007;Maxwell & Caldwell, 2008;Swindale & Bilinsky, 2006;World Food Programme, 2008). Improved water sources included piped water, taps, boreholes, protected wells, and rain water. Improved latrines included flush to septic system or pit, pit, and composting latrines. Infant and young child feeding practice indicators were calcu- The primary outcome, endline GAM prevalence in children 6-59 months old at baseline, was analysed at individual level using mixed effects multilevel logistic regression, accounting for variation at the cluster level using random effects. Covariates included baseline WHZ, sex, and age and those found to differ between arms at baseline (tested using chi square/t tests on transformed data as necessary). We tested for baseline differences between arms because we anticipated that the small number of clusters would produce some significant variation by chance (Hayes & Moulton, 2009). Secondary nutrition outcomes at the individual child level were also analysed using adjusted mixed effects multilevel regressions. We did investigate whether commune had any predictive value for our outcomes and it did not. We did not collect village level data because of resource constraints but we do not think that this would have impacted on the results. Endline differences between arms in possible determinants of undernutrition were analysed using a difference in differences approach, with the exception of infant and young child feeding practices for which the age-limited groups for each indicator rendered difference in difference analysis impossible. Instead, to test for any differences in these variables at endline, we performed comparative cross-sectional analyses.
It should be noted that predictors and outcome measures were analysed using different approaches because the odds ratio allowed for an adjusted model to be constructed to test the impact of the intervention on the outcome, whereas the difference in differences provided a more descriptive and understandable way to look at potential changes in mediating variables.
For population level analyses, the samples of beneficiaries and nonbeneficiaries were combined. To account for the relative undersampling of nonbeneficiaries and ensure that the results were representative of the general population, a population weight of 5.68 was calculated and applied to nonbeneficiaries.
3 | RESULTS There were no discernible differences in implementation or uptake between arms.  There were no differences between arms in average child age or sex ratio (Table 2). Over 10% of children were wasted at baseline and, when combined with the small number of oedema cases, the GAM prevalence was 13.5% (14.1% standard arm, 12.9% modified arm, p = 0.685); that is, at emergency levels (World Health Organization, 2000). The prevalence of low MUAC was small, and significantly greater in the standard arm, whereas the mean WHZ and HAZ scores were below zero and over a third of children were stunted. In addition, nearly 1/3 children were sick in the previous month, of which more than half had fever/malaria and the rest had ARI or diarrhoea. ARI was more common among beneficiary children in the modified arm. Bed net use was very low, as may be expected given the hot/dry conditions with few mosquitoes. Most of the children under 2 years old had been breastfed and a similar proportion of 12-15 month olds were still being breastfed. However, very few children were receiving a minimum adequate diet or achieving minimum dietary diversity or meal frequency. Children 24-59 months of age consumed only 2/7 food groups on average the previous day, which is consistent with the poor household dietary diversity.

| Population sample
Baseline characteristics of the population sample were similar to the beneficiaries (Tables S1 and S2)   Household expenditure and food security improved for all beneficiaries; that is, falling household food insecurity access scores and coping strategies index scores, and rising diet diversity and food consumption scores. We also observed significant increases in land access and ruminant ownership. However, the prevalence of child sickness increased, because of a large increase in fever/malaria despite an increase in bednet use and a fall in ARI.
We also performed cross-sectional analyses to test for any difference in the prevalence of infant and young child feeding practices at endline; however, we found none, despite all indicators showing a trend towards improvement from baseline.

| Population sample
There were also no significant differences between the arms at endline among the population sample, with the exception of a significantly greater (though small) reduction in the size of small ruminant holdings in the modified arm (Table S3). Baseline to endline changes in the arms combined showed similar trends to those observed among beneficiaries.
To explore differences in food security over time, we examined recalled data on the months of adequate household food provisioning. p < 0.001; Figure 2). However, there were no differences between the arms in any other months (p > 0.05).

| Nutrition outcomes
We did not observe any difference in nutritional impact between the modified and standard interventions, either among beneficiary children (Table 4) or in the population (Table S4). The endline odds of a child having GAM were almost equal by arm, for both beneficiaries and the general population. The endline adjusted mean WHZ was also no different by arm in either sample. Stratifying by child age and household asset index did not reveal any differences. All other anthropometric outcomes were also no different by arm, both among beneficiaries and the population. We also observed no change in the prevalence of GAM in children 6-59 months old. Among beneficiaries aged 6-59 months, the GAM at baseline was 13.5% (95% CI [10.8, 16.8]) and at endline among beneficiary children of the same Large ruminants owned (mean ± SD, range) 0.5 ± 0.8 (0-8) 0.7 ± 1.0 (0-13) 0.6 ± 0.9 (0-13) 0.141 Small ruminants owned (mean ± SD, range) 1.7 ± 2.4 (0-30) 2.1 ± 2.5 (0-18) 1.9 ± 2.5 (0-30) 0.087

| DISCUSSION
This is the first trial in a humanitarian context comparing two equalvalue seasonal cash interventions of different durations alongside supplementary feeding for PLW and children 6-<24 months in households receiving cash. It is also the first study to assess nutritional impact at population level, as well as among beneficiaries.
We found no differences in the endline prevalence of GAM, mean WHZ, or other anthropometric indicators, between children in households given cash from April compared with those given cash from June (among which, children 6-<24 months old also received  Access to land (% points, 95% CI) 1.7 (−0.5, 3.8) 6.6 (2.5, 10.7) 4.1 (1.6, 6.7) p < 0.001 5.0 (0.3, 9.6) .037 Large ruminants owned (mean ± SD) Food security Household food insecurity access score (mean ± SD) (0 best-27 worst) −3.93 ± 6.51 −3.46 ± 6.84 −3.71 ± 6.67 p < 0.001 0.47 (−1.57, 2.50) .636 Coping strategies index score (mean ± SD) (0 best-56 worst) Water and sanitation Use of improved water source (% points, 95% CI  supplementary food between June and September), and there was also no impact on the general population. The absence of a differential impact on endline anthropometry is consistent with the lack of differences in food security and health indicators between arms. It is interesting to note the temporary improvement in perceived food access among beneficiaries of the modified intervention in April and May compared with beneficiaries of the standard intervention, who during these 2 months received nothing. But evidently, this improvement in food security was insufficient to translate into a decreased risk of undernutrition by the end of the lean season. We did not test for differences in the prevalence of SAM as the study was not adequately powered to detect clinically important differences.
Unexpectedly, we found that the anthropometric status of children in households receiving either UCT or supplementary food, as well as other interventions and the harvest, was unchanged by endline and remained above the 10% emergency threshold (World Health Organization, 2000). This was surprising given prior studies in Niger that have documented postintervention/harvest improvements among beneficiary children (Aker & Nene, 2012;Bliss & Golden, 2013;Fenn et al., 2014).

| Possible reasons for the lack of impact of the modified intervention
First, the smaller monthly transfers of the modified UCT may not have been sufficient; at 21,500FCFA/month (£24), they were a third less than the standard 32,500FCFA/month (£36) and were not accompanied by supplementary rations in April and May. Transfer size has been suggested as an important factor in determining the impact of cash on child outcomes (de Groot et al., 2015). However, analysis of expenditure data per capita indicated suggested that transfer size was not a limiting factor. Analysis of Concern's market monitoring data indicates that the UCT would have permitted purchase of food to meet between 88% (modified intervention) and 141% (standard intervention) of the average household's kilocalorie requirements, if solely spent on food and regardless of whether the household also received any supplementary food ( Figures S1.3 and S1.4). A larger transfer may have increased spend on health-seeking and thereby improved child nutritional status. This, however, seems unlikely given health service inaccessibility (Bailey & Hedlund, 2012;Blanford et al., 2012) and because barriers to uptake are rooted in chronic rather than transient poverty or seasonal food insecurity .
A second possible reason is that the UCT may not have been sufficiently early. Acute malnutrition admissions have been seen to rise from March onwards and we found more than half of households reporting food insecurity by the same month ( Figure 2). Initiation of UCT prior to April has been suggested in Niger (Fenn et al., 2014) and some UCT interventions intended to prevent deterioration in food security and/or nutrition have even been implemented during/postharvest (Langendorf et al., 2014;Tumusiime, 2015).

| Possible reasons neither intervention succeeded in reducing undernutrition
For both arms, there was a positive change over time in expenditure and food security. This was expected due to the harvest and is, plausibly, also attributable to the UCT and supplementary food as well as other, government-led, interventions. These included distribution of 493 metric tonnes of grain in Affala and Takanamatt between April and September (with no discernible imbalance by arms, this was equivalent to the cereal needs of 22% of households in the studied villages for 6 months), and a social protection scheme providing 10,000FCFA/ month (£11) to 840 households (also not discernibly different by arm, this was equivalent to 17% of households). As well as food availability, accessibility is also likely to have improved, particularly for beneficiary households for whom the transfers were theoretically enough to meet nearly all kilocalorie needs ( Figure S1.4). Food utilisation also improved, as indicated by increased diet diversity and food consump- However, although food security increased, we also saw a typical deterioration in health over the season. Among beneficiaries, child morbidity rose by 10% points by endline and of those who were sick,  The model is a mixed effects multilevel logistic regression accounting for variation at the cluster and household level using random effects.
c The final models are only partially adjusted, as the two household level variables (household lifestyle and land access) had to be included at the individual level to allow the model to converge. The fully adjusted model did not converge. over 80% had fever/malaria, despite increased bed-net use facilitated by seasonal distributions (insecticide-treated bed-net use at endline for children 6-59 months old was 97.4% (95% CI [96.0, 98.3])). A malaria peak is an annual norm, but the timing is perhaps later than typically characterised, with incidence remaining elevated several months after the rains, in early October (Fenn et al., 2014;Vaitla, Devereux, & Swan, 2009; Figure S1.5). It is also worth noting that improved latrines were accessible to only 2% of households, 35% did not use improved water sources and these indicators remained unchanged following receipt of either UCT. As is typical among children from UCT beneficiary households in Niger (Bliss et al., 2016;Fenn et al., 2014), we found a significant association between recent sickness and lower mean WHZ and higher GAM among beneficiary children. Infections, including malaria, are important determinants of malnutrition (Black et al., 2008;Bliss et al., 2016)

| Strengths and limitations
Strengths of our study include the cluster-randomised controlled trial design and high response and follow-up rates of our cohort. We aimed for a sample size of 3,520 beneficiary children but found only 1,831 (52.0%), due to an overestimation of average household size and number of children per household. However, we also found a much lower prevalence of baseline GAM than we had anticipated (14% vs. 21%). A post hoc calculation using the clustersampsi command in Stata 14 indicates that, with 80% power and an alpha risk of 5%, we would have been able to detect a difference in endline GAM prevalence of 6% points with both our achieved samples of 1,831 beneficiary children and the combined sample of 2,292 beneficiary and nonbeneficiary children. Although the achieved sample size was less than planned, our results indicate that the modified intervention did not result in any major differences in the prevalence of GAM. Limitations are that we did not measure the incidence of outcomes during the intervention to understand any transitory impacts, particularly in the first 2 months of the modified UCT, and the limited generalisability of findings to periods of acute food insecurity, given the absence of crisis indicators during 2015.
To conclude, although we observed a temporary increase in food security for beneficiaries in the pre-lean season, there was no evidence that starting the UCT 2 months earlier and providing the same amount of cash over a longer period, together with 4 months of supplementary feeding, would be beneficial to children's nutritional status. There is already consensus that cash usually needs to be combined with complementary interventions to impact on nutrition (Bailey & Hedlund, 2012;de Groot et al., 2015). Probably because past recommendations have focused heavily on the food security-related drivers of malnutrition (Vaitla et al., 2009) and there is a pre-existing recommendation that the UCT/ supplementary food package is implemented where health needs are met (Langendorf et al., 2014). We observed an elevated prevalence of GAM in both arms at baseline and endline, despite improved food security, and there was an association between nutritional status and infection. Therefore, in the future, it would be pertinent to assess any supply-side shortcomings in the health system and barriers to health-seeking, as well as availability and barriers to use of safe water and latrines, in addition to ensuring household food security. More specifically, we suggest that seasonal interventions to treat and prevent malaria, a major cause of death (Langendorf et al., 2014) and an important driver of acute malnutrition in this population (Bliss et al., 2016), in addition to the standard UCT/nutrition supplement intervention, may help protect children from acute malnutrition. Lastly, it is also recommended that future studies test the assumption that the targeting of low income households with UCTs leads to population level impact (de Groot et al., 2015).

ACKNOWLEDGMENTS
We thank Leila Bourahla who contributed to the design of the study, oversaw the implementation of study activities, and secured the intervention funding; Joanna Dadie who supported the implementation of study activities; and all other staff and hired consultants in Concern's Niamey and Tahoua offices who were directly or indirectly involved in the study. We also thank Kate Golden and other Concern head office staff who supported the implementation of the study. Lastly, we thank the people of Tahoua who facilitated and participated in the study.