Policy Research Working Paper 11138 The Earlier the Better? Cash Transfers for Drought Response in Niger Ashley Pople Patrick Premand Stefan Dercon Margaux Vinez Stephanie Brunelin Social Protection and Labor A verified reproducibility package for this paper is Global Department & available at http://reproducibility.worldbank.org, Development Impact Group click here for direct access. June 2025 Policy Research Working Paper 11138 Abstract Climatic shocks exacerbate consumption volatility and regular transfers throughout the year. The results show that seasonality. When facing fluctuations in labor needs and large early transfers yield greater net benefits on economic prices, low-income rural households may rationally increase welfare and psychological well-being before and during the their consumption during specific seasons rather than main- lean season compared to a traditional humanitarian lean taining it constant throughout the year. This makes the season response. The large early transfers also tend to have optimal timing of policy responses to shocks ambiguous. larger effects than the year-long transfers. These welfare This paper examines the impact of varying the timing of differences do not persist after the lean season and nine cash transfers in response to drought in Niger, leveraging months later. The early timing of transfers shifts borrowing satellite-based triggers for a faster response before the lean behavior but has no discernible impact on livelihoods. The season. A randomized controlled trial compares large early findings demonstrate the value of sufficiently large early transfers delivered before the lean season, a traditional transfers in mitigating the effects of a severe drought in humanitarian response during the lean season, and smaller presence of seasonal fluctuations in labor needs and prices. This paper is a product of the Social Protection and Labor Global Department and the Development Impact Group, Development Economics.. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at apople@worldbank.org or ppremand@worldbank.org. A verified reproducibility package for this paper is available at http://reproducibility.worldbank. org, click here for direct access. RESEA CY LI R CH PO TRANSPARENT ANALYSIS S W R R E O KI P NG PA The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Earlier the Better? Cash Transfers for Drought Response in Niger Ashley Pople∗ Patrick Premand† Stefan Dercon‡ Margaux Vinez§ Stephanie Brunelin¶ Keywords: Food security, climate change, seasonality, drought, cash transfers, timing, field experiment, social protection. JEL Codes: D12, O12, Q54 ∗ World Bank: apople@worldbank.org, corresponding author. † Development Impact, Development Economics, World Bank: ppremand@worldbank.org ‡ University of Oxford: stefan.dercon@economics.ox.ac.uk § World Bank: mvinez@worldbank.org ¶ World Bank: sbrunelin@worldbank.org Acknowledgments: This study is based on a collaboration between the Niger Cellule Filets Sociaux (CFS), which manages the Niger Adaptive Safety Nets Project, the World Bank Sahel Adaptive Social Protection program (SASSP), the World Bank Development Impact Group and the University of Oxford. The study was financed by the Sahel Adaptive Social Protection Program. We are grateful to Moussa Bouda, Bassirou Karimou, Kadi Aboubacar, Moumouni Moussa, and the entire CFS staff, as well as Snjezana Plevko, Felix Lung, Mahamane Maliki Amadou and the World Bank Sahel Adaptive Social Protection Program team for a fruitful collaboration. We appreciate the helpful comments from Jenny Aker, Josh Blumenstock, Thomas Bossuroy, Paul Christian, Karen Macours, Prabh- meet Kaur Matta, Hyuk Harry Son, seminar and conference participants at Bor- deaux, DENS, LSE, Stanford Doerr School of Sustainability and Oxford. Special thanks to Rongmon Deka, Dieynab Diatta, Desmond Fairall, Michael Green and Phyllisa Joseph for excellent research assistance, Mariana Martinez for her strong project management, Karim Pare for overseeing the field work and IFHRA for data collection. This study was registered in the American Economic Association RCT Registry as part of a larger multi-country study under trial number 10097 (https://doi.org/10.1257/rct.10097-1.1). The Niger results are issued first as the program did not initially trigger in Mauritania and Senegal. The study received ethics approval by the Economic Department’s Research Ethics Committee (DREC) at the University of Oxford (Protocol No. ECONCIA21-22-25). Computational re- producibility has been verified by DIME analytics. The findings, interpretations, and conclusions of the paper are those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of its Executive Directors or the governments they represent. 1 Introduction Climate change is intensifying the frequency and severity of extreme weather events, with long-lasting implications for welfare and poverty (Hallegatte et al., 2016; Rohde, 2023). For low-income rural households dependent on rain-fed agriculture, weather shocks such as drought exacerbate the effects of seasonality. When facing fluctuations in labor needs and prices, these households may rationally increase consumption during specific periods of the year. This makes the optimal timing of responses to shocks a priori undetermined. Cash transfers are often used to respond to shocks and support consumption smooth- ing. However, many interventions tend to target the period when food insecurity peaks. This risks reaching households late after they have reduced their consumption or used cop- ing strategies with lasting adverse consequences (Dercon, 2004; Birkmann et al., 2022). Recent advancements in the capacity to predict crises have spurred governments and humanitarian actors to experiment with early response interventions, for instance using remote sensing or satellite-based triggers. Yet while the effectiveness of regular cash transfer programs against shocks is well documented (Jensen et al., 2017; Adhvaryu et al., 2022; Premand and Stoeffler, 2022), experimental evidence on the impact of vary- ing the timing of cash transfers relative to the trajectory of these shocks remains thin, particularly for slow-onset shocks (Jeong and Trako, 2022). This study assesses the value of early cash transfers in response to a severe drought – a slow-onset event – and in anticipation of the resulting food insecurity crisis in Niger, one of the world’s most climate-vulnerable countries (World Bank, 2022). In late 2021, a satellite-based trigger was activated when rainfall levels in designated areas fell 25% below seasonal average during the growing period – a level of scarcity typically seen only once every ten years. This drought caused significant damage, reducing annual cereal production nationally by 36% compared to the five-year average (Brouillet et al., 2022). In response, the Government of Niger implemented a program delivering cash transfers amounting to 180,000 West African CFA francs – approximately USD PPP $885 – to villages in the most drought-affected communes. This program piloted early cash transfer modalities relative to a traditional response provided during the following lean season when food insecurity is expected to peak and a humanitarian crisis unfolds. Using a three-arm randomized controlled trial in collaboration with Niger’s safety net agency, we evaluate the impact of adjusting the timing of post-drought cash transfers relative to the lean season. Specifically, we compare large temporary transfers provided early during the four months before the lean season, smaller year-long transfers starting at the same time, and large, temporary cash transfers provided during the lean season (the traditional humanitarian response). By holding household targeting and total transfer amounts constant across the three modalities, we isolate the impact of timing relative to drought and the agricultural cycle. 1 There are several reasons why the timing of cash transfers might matter for welfare in the aftermath of a severe drought. Cash transfers received during the lean season are intended to help households cope with the effects of the shock at a time when they need it the most. In other words, this traditional response targets vulnerable households when they are most likely to reduce their food consumption and sell their productive assets before their next harvest. In contrast, providing liquidity before the lean season may help households to adopt pre-emptive behaviors and adjust their livelihoods to mitigate the worst impacts of drought. Furthermore, in a highly deprived and food insecure environment, providing resources when returns to better nutrition are highest due to labor needs may boost productivity (Behrman, 1993; Banerjee and Duflo, 2007). However, early cash support may be suboptimal if households are unable to save or smooth their consumption throughout the crisis period. There are also other behavioral reasons why earlier cash transfers may or may not be preferable, such as greater mental scarcity as food insecurity increases (Mani et al., 2013) or social pressures to share the transfers with other community members (Baland et al., 2016; Boltz et al., 2019; Di Falco et al., 2019). More regular albeit smaller cash transfers may serve to smooth consumption over a longer time horizon throughout the agricultural cycle, but their size may nevertheless limit spending on durable goods or other lumpy expenditures (Crosta et al., 2024). Our findings demonstrate the net welfare benefits of an early response to a drought for poor rural households facing seasonal movements in prices and in labor needs. Large early cash transfers delivered in the months before the lean season yield greater net benefits before and during the lean season compared to the traditional humanitarian lean season response. Indeed, we find that households are already highly food insecure before the lean season when prices are beginning to move up but when labor needs are already high. Specifically, the large early transfers improve food security by 8% and increase monthly food consumption by 17.6% relative to the traditional response. The early transfers also lead to immediate improvements in life satisfaction (17.8%) and mental health (0.29 standard deviation). Crucially, some welfare benefits persist in the lean season even after the transfers cease. While the traditional response boosts food consumption during the lean season relative to the early intervention, the large early cash transfers show positive net benefits: higher levels of food security and psychological well-being are sustained into the lean season. In contrast, the smaller year-long cash transfers show more modest pre-lean season benefits than the larger early transfers, with similar effects compared to the traditional response during the lean season. The timing of cash transfers significantly alters financial behaviors. Both early trans- fer modalities reduce borrowing by 24%-38% in the pre-lean season. The traditional response reduces borrowing by similar amounts, but only during the lean season when transfers are delivered. Even with early interventions, households still take on substantial debt, underscoring the severe economic stress caused by the drought. Beyond reduced indebtedness, we find minimal differential impacts on livelihood activities across transfer 2 modalities. After the lean season, we find that the welfare effects tend to converge across all three transfer modalities, both in the immediate post-lean season period and nine months after the harvest, during the subsequent year’s pre-lean season. These results suggest that the timing of transfers provides important short-term benefits in supporting households to manage their consumption and adjust their financial behaviors after a severe drought shock, in a setting with seasonal price and labor needs fluctuations. However, the various modalities do not lead to differential medium-term impacts as households enter the next agricultural cycle. Importantly, in the absence of a control group, we are unable to comment on the overall impact of the cash transfers themselves, as already documented in Niger (e.g. Premand and Stoeffler, 2022), only on the relative effects of different transfer modalities. Our paper provides the first experimental evidence on how cash transfer timing and frequency affect household welfare in a high-seasonality context affected by a drought. As such, it contributes to the literature on seasonality and on the role of cash transfers in mitigating extreme weather impacts - particularly slow-onset events. Cash transfer and livelihood interventions rolled out prior to shocks have been shown to support house- holds in mitigating their adverse repercussions through consumption smoothing, saving, and livelihood diversification (Jensen et al., 2017; Adhvaryu et al., 2022; Bossuroy et al., 2022). Evidence from Niger demonstrates that multi-year cash transfers mitigated drought impacts for recipients (Premand and Stoeffler, 2022). In Nicaragua, post-drought bi-monthly cash transfers combined with training or an investment grant had persistent effects on livelihood diversification and resilience to future shocks (Macours et al., 2022). Nevertheless, the overall effectiveness and optimal timing of short-term responses to ex- treme weather events, such as a humanitarian interventions, has remained an open ques- tion.1 Recent advances in prediction technologies have spurred research on the value of early response, focusing primarily on sudden onset events, where households either receive cash or guaranteed access to credit ahead of floods to induce ex-ante behavioral change (Balana et al., 2023; Pople et al., 2024; Lane, 2024; Dunsch et al., 2025).2 In contrast, our study experimentally varies the timing and frequency of cash transfers relative to the impacts of a drought – a slow-onset event. Unlike floods that are immediately destruc- tive to assets and crops, the welfare effects of a drought manifest gradually, impacting agricultural production first and food security later, creating a window of opportunity to intervene early, but raising questions on how this interacts with seasonality. Our findings show that a sufficiently large early response produces better outcomes than the traditional 1 A large body of literature has evaluated design parameters of transfers other than timing, such as cash versus in-kind transfers, physical versus electronic payments, or lump-sum versus monthly install- ments (e.g. Aker, 2017; Skoufias et al., 2013; Cunha, 2014; Hidrobo et al., 2014; Aker et al., 2016; Haushofer and Shapiro, 2016; Hoddinott et al., 2018; Cunha et al., 2019). 2 These new approaches are sometimes referred to as “anticipatory action”, “forecast-based financing” or “shock-responsive transfers”, although there are some nuances between these concepts. 3 lean season response in the period before and during the lean season. Our study also provides new insights to the seasonality literature by demonstrat- ing how early interventions enhance household consumption during the lean season in a drought-stricken year with large seasonal movements in food prices and labor require- ments. The negative effects of seasonality, with food insecurity peaking during the lean season, are well-established (Reardon and Matlon, 1989; Paxson, 1993; Kaminski et al., 2014) and similar patterns have been observed in the context of drought (Kazianga and Udry, 2006). Partial consumption smoothing has been attributed to numerous market failures, including imperfect credit or insurance markets and limited saving capacity (Mor- duch, 1995). Recent interventions have successfully boosted agricultural production by providing liquidity to rain-dependent agricultural households during different moments of the agricultural cycle: before the lean season (Beaman et al., 2023), during the lean season (Fink et al., 2020) or at harvest (Burke et al., 2019).3 In contrast, Beegle et al. (2017) find that varying the timing and payment structure of a public works program in Malawi between the lean season and harvest period did not improve food security. Nevertheless, households hold preferences over the timing and frequency of cash transfers relative to the agricultural cycle (Kansikas et al., 2023), suggesting that timing matters for economic decision making. We uniquely test the timing of unconditional cash trans- fers by experimentally varying whether households receive cash before or during the lean season, or throughout the year following a drought and in a context of intense seasonality. Our findings confirm earlier evidence on liquidity constraints: the control group takes on substantial debt soon after the drought-affected harvest. However, we also show that large, early cash transfers received before the lean season improve welfare by boosting food consumption and limiting the extent of borrowing, thereby mitigating the negative effects of seasonality. The paper is organized as follows. Section 2 presents the context and intervention. Section 3 introduces the experimental design and sample. Section 4 discusses data, time- line and outcomes. Section 5 provides a theoretical framework in which an agricultural household handles consumption decisions in the presence of seasonality in price and labor requirements, generating plausible predictions on the impact of shocks on welfare. Section 6 summarizes the empirical strategy. Section 7 presents results. Section 8 concludes. 3 Bazzi et al. (2015) show that unexpected delays in Indonesia’s cash transfer program reduced house- hold consumption, highlighting how timing misalignment can undermine the effectiveness of cash transfers in credit-constrained populations. 4 2 Context and intervention Located in the Sahel region, Niger is one of the poorest and most climate-vulnerable countries in the world (World Bank, 2022). It is subject to high rainfall variability, resulting in particularly severe droughts. This vulnerability is exacerbated by heavy reliance on rain-fed agriculture, which employs 80% of the workforce. These factors contribute to acute and recurrent food insecurity crises. By 2050, Niger’s annual GDP could decline by 2.2% to 11.9% under different climate scenarios (World Bank, 2022). Niger’s semi-arid southern region follows a distinctive agricultural cycle dominated by a brief West African monsoon from end of June to September, followed by an ex- tended dry period (see Figure 1). Agriculture is almost exclusively rain-fed, with millet and maize as the primary crops. There is high seasonality in both prices and farm labor requirements. First, food prices rise significantly during the annual lean season and agri- cultural households often lower their consumption during that period (Aker, 2012).4 The annual lean season corresponds to the rainy season, where high humidity and temper- atures result in high food stock spoilage and rodent infestations (Christiaensen, 2017). With poorly integrated food markets, food prices tend to peak, resulting in lower con- sumption and poor nutrition due to depleted food stocks (Reardon and Matlon, 1989). Weather shocks affecting agricultural production in the preceding rainy season will likely translate into larger price movements and deeper consumption adjustments during the subsequent lean season (Kakpo et al., 2022). Second, the agricultural calendar also creates distinct seasonal labor demands through- out the year (see Figure 1). From March to May, farmers prepare land and begin plant- ing millet and maize in anticipation of late June rains. Labor requirements remain high throughout the lean season (late June to September), as weeding and crop maintenance are crucial for ensuring a successful October harvest. Following harvest, an off-season period begins, when farmers often pursue local off-farm activities or engage in tempo- rary migration – locally termed “L’Exode” – seeking wage opportunities in nearby cities and neighboring countries. The lean season thus presents significant challenges to rural households due to demanding farm labor under harsh weather conditions, deteriorating food stocks from previous harvests, and elevated food prices throughout this period.5 To address recurrent droughts, the government and humanitarian actors jointly de- velop annual response plans based on projected food insecurity during the upcoming lean season, as estimated in the Harmonized Framework (commonly known as “Cadre Harmo- e” in francophone countries). Initial planning meetings occur in November, with final nis´ 4 Even though there was considerable seasonality in food prices in this year in line with historical trends (see Appendix Figures 4 and 5), annualized inflation in 2022 was only about 3%, as the CFA franc is pegged with the euro. 5 A further factor is disease burden, with malaria the most important one in Niger, and mainly prevalent during the rainy season. In 2023, about 5.9% of global deaths from malaria were in Niger, while its population is only 0.34% of the world population (World Health Organization, 2023). 5 plans established by the following March. The majority of Niger’s response, and more broadly throughout the Sahel, is thus determined in March and primarily targets the lean season – what we refer to as “the traditional response” throughout this paper.6 This pa- per examines whether alternative transfer modalities can improve upon this established approach. A drought in 2021 provided an opportunity to test alternative response modalities. In response to this drought, the Government of Niger through the Safety Net Unit (Cel- lule Filets Sociaux or CFS) piloted Niger’s first shock-responsive cash transfer program leveraging satellite-based triggers. The program employed the Water Requirement Satis- faction Index (WRSI) to monitor the occurrence of a drought – defined as either moderate or severe – in pilot communes7 at the end of the agricultural season.8 Upon trigger activa- tion, unconditional cash transfers were to be delivered to affected communities, reaching 22% or 44% of a commune’s population for moderate or severe droughts, respectively. The 2021 agricultural season was characterized by long dry spells at critical points of the cropping cycle. Consequently, 3.7 million people (14% of the population) were esti- mated to face acute food insecurity during the next lean season in 2022.9 At the end of the growing season in late October 2021, satellite-based triggers activated in four pilot communes, signaling that rainfall levels had dropped to 25% below the average – a level of scarcity typically seen only once every ten years. This prompted the government to deliver unconditional cash transfers to approximately 15,400 households in villages not previously covered by the national safety nets program (Figure 3). This paper evaluates three distinct cash transfer modalities across 171 villages in three communes where severe drought was detected, as detailed in Section 3. In contrast to 2021, the subsequent agri- cultural season in 2022 was fairly productive, although flooding impacted over 300,000 people, including more than 10% of our sample. 6 The government’s national safety net program provides chronically poor households with regular monthly transfers of 15,000 CFA francs, combined with human capital and economic inclusion measures, over a 24-month period. This program started in 2012 and has covered around 120,000 households since then. Its funding has not been sufficient to cover all extreme poor households in Niger. 7 A commune in Niger is a third-tier administrative division, responsible for local governance, and comprised of many villages. 8 The trigger design process involved three phases. First, historical climate, production, and price data analysis identified drought-vulnerable communes and established millet as the critical food security crop. Second, the Water Requirement Satisfaction Index (WRSI) – a measure of crop performance based on estimated water availability during the rainy season – was selected as the optimal predictor of mil- let production and price shocks. Third, cost analyses of various scale-up scenarios informed triggering parameters. Eligible communes were selected based on their vulnerability to shocks, (lack of) partici- pation in the national safety net, low average historical WRSI values, and significant millet production. Thresholds were established for moderate (-10%) and severe (-25%) droughts relative to median WRSI in the third dekad of October, with corresponding population coverage targets. A secondary trigger using November’s Harmonized Framework data was incorporated to enable activation based on the decisions of the Steering Committee if remote sensing failed to capture other significant production-affecting shocks. 9 Drought, combined with crop infestations and increasing insecurity, caused national cereal produc- tion to drop by 36% compared to the five-year average (Brouillet et al., 2022). As a result, per capita growth declined from 3.6% in 2020 to 1.4% in 2021 (World Bank, 2022). 6 3 Experimental design 3.1 Three-arm randomized controlled trial We study the effects of adjusting the timing of cash transfers delivered to drought-affected villages. Treatment group 1 (T1): “Traditional response” intervention. Eligible households within target villages receive four rounds of cash transfers of 45,000 CFA francs each – approximately USD PPP $22010 – from July to September 2022 during the lean season and before the next harvest. This treatment arm replicates how most human- itarian actors would typically respond to a food insecurity crisis, with the aim to support vulnerable households when they are most at risk of reducing food intake and selling assets. As discussed in Section 2, it is also a period of high labor requirements on the farm, while food prices peak and remaining food stocks are depleting and deteriorating. Treatment group 2 (T2): “Early short” intervention. The early short re- sponse intervention mirrors the traditional response intervention, except that the cash transfers are made four months earlier and before the lean season, rather than during. Eligible households within target villages receive four rounds of cash transfers (45,000 CFA francs per transfer) from March to June 2022 in advance of the lean season. These transfers are thus made after the failed rainy season, but a few months ahead of the expected peak of food insecurity during the subsequent lean season. Food prices will not have peaked yet, but labor demands on the farm are already high in this period. Provid- ing liquidity before the lean season may help households to adopt pre-emptive behaviors, support their labor productivity, and adjust their financial and livelihood decisions to mitigate the worst impacts of drought. Treatment group 3 (T3): “Early long” intervention. Eligible households within target villages receive smaller, but more regular cash transfers (12 payments of 15,000 CFA francs or approximately USD PPP $74 per transfer) for roughly 10 months, starting from March 2022 to January 2023.11 Thus, these transfers are made before and during the lean season, and continue for several months after the harvest. For simplicity, we characterize them as “year-long”. The size and frequency of transfers may support consumption smoothing more effectively over a longer time horizon, but also limit spending on durable goods or other lumpy expenditures, as suggested by Crosta et al. (2024). The total value of the cash transfers is the same across the three modalities (180,000 CFA francs or approximately USD $885 in PPP terms), approximately one-third of yearly 10 We use the 2022 PPP conversion factor of 203.29 for private consumption from the World Bank’s World Development Indicators database. 11 The intended start date was January 2022, but the first payment was delayed due to operational and procurement considerations. 7 food consumption for the average household in our sample. The early short and tradi- tional response arms entail relatively larger, but fewer transfer payments, compared to the early long group with its relatively smaller, more frequent transfer payments. The early long cash transfer payments coincide with the early short transfers in the pre-lean season period and the traditional response during the lean season. Households across all three groups were simultaneously targeted and registered for cash transfers between January and February 2022. Thus, households may have antici- pated receiving cash transfers, but were not told about the exact timing or amount of the cash transfers before the payments started. Households received no additional messaging about drought response. The result should thus be interpreted as impacts net of any potential anticipation effects. 3.2 Sample Our sample comprises 4,144 households across 171 villages in three communes in Niger. We identified our target population through a three-step process. First, a subset of com- munes were selected for weather monitoring using a satellite-based trigger mechanism. Of these, four communes were identified as affected by severe drought in late October 2021, as illustrated in Figure 3. We chose three of these communes (Tagazar, Imanan and Dantchandou) for the study, as the national multi-year social safety net program had yet to be expanded to these areas and thus households were not already receiving government cash transfers. Second, in the absence of a pre-existing social registry cover- ing these communes, the Niger safety nets unit (Cellule Filets Sociaux) collected proxy means test (PMT) scores for all households in target villages between December 2021 and January 2022. Households were deemed eligible if they fell in the bottom 44% of the PMT distribution in each commune. Finally, households in target villages were sampled for the study in January 2022. We randomly sampled 25 eligible households from each target village, or all eligible households if a village had fewer than 25 households. Nearly all village clusters in the sample have 24 to 25 households. We sampled households based on the PMT scores, and before the government im- plemented a verification of the list of households eligible for the program, due to time constraints. In practice, only a small number of households considered eligible based on the PMT score are excluded after the field verification process,12 so the differences are marginal, with just over 2% of households sampled not receiving any cash transfers. We thus estimate intent-to-treat effects, as outlined in Section 6. Moreover, two villages were completely inaccessible throughout the study due to security concerns, including at baseline.13 We drop these two villages from our main analysis and use the baseline survey 12 Community leaders are consulted to ensure that the final list of households receiving cash transfers within a village are indeed poor. 13 The survey firm conducted monthly assessments to determine whether it was safe enough to enter 8 to define our final sampling frame. We show that the experiment is well-balanced across treatment groups in the next section. 4 Data and measurement 4.1 Five rounds of household surveys We collect five rounds of household survey data at high frequency during the year following the drought and throughout the intervention period. Figure 1 illustrates the timing of each data collection round in relation to the seasonal calendar and cash transfer payments. This design allows us to capture the effects of the different cash transfer modalities on household outcomes before, during and after the lean season, both immediately and in the medium term. To minimize noise in our food consumption measures, we ensured that the scheduling of data collection did not conflict with key festivals, such as Ramadan and Tabaski. We also conducted a questionnaire with a community leader during each survey round to measure price fluctuations within villages over time. We begin with a baseline survey in February 2022, capturing households’ socio- economic and demographic characteristics and pre-treatment outcome measures. The survey took place after identifying target households, but prior to the government-led registration of participants for all three cash transfer interventions. The pre-lean season round takes place in the months before the lean season (March to June 2022), split across four high frequency surveys using an approach adapted from Christian et al. (2022).14 We collect high-frequency surveys to capture the short-term and potentially transitory effects of cash transfers on welfare. We divide the sample equally into four groups, ensuring balance across our randomization strata, and collect one survey for each household within the round. These surveys follow immediately after the early cash transfers are made, capturing immediate impacts on food security, consumption, and household behaviors. For the main analysis, we pool these four surveys and refer to them collectively as the “pre-lean” survey. Our midline survey, conducted during the lean season in three waves (July to Septem- ber 2022), follows the same approach as the high frequency surveys but includes more comprehensive measures of food consumption to capture the impact of the transfer modal- ities at the peak of the food insecurity crisis. The post-lean season round (“post-lean” survey) comprises four high frequency sur- a village to collect data. 14 See Premand, Christian, et al., 2024 for more details. One difference is that we collect high- frequency data for all our sample within each period and randomly rotate which households are surveyed each month. This maintains our study statistical power to make comparisons for each season, in this case the pre-lean season. In contrast, Christian et al. (2022) and Premand, Christian, et al. (2024) collect more frequent observations but only for a subset of their sample. 9 veys in the months immediately after the harvest (October 2022 to January 2023), fol- lowing the smaller, regular cash transfers. Finally, we conduct an endline survey with the full sample approximately nine months after harvest and six months after the last early long cash transfer payment (May to June 2023). The endline survey takes place just before the following year’s lean season, coinciding with the timing of the first pre-lean season survey. Both rounds thus capture similar phases of the agricultural cycle when households are planting for the up- coming harvest. The endline survey aims to capture medium-term effects of the different cash transfer modalities. We implemented thorough data collection and tracking protocols, achieving an av- erage response rate of 98.2% across all survey rounds. Non-response is balanced across treatment groups within each survey round, with no statistically significant differences.15 4.2 Outcome measures We pre-specified three primary outcomes – food security, food consumption, and psy- chological well-being – to capture welfare among low-income rural households and their ability to respond to drought-induced seasonal fluctuations. Food security is captured by the Food Consumption Score (FCS), which measures a household’s dietary diversity and nutritional intake by weighting the frequency of eight food groups consumed in the past week by their nutritional importance, e.g. a weight of 2 for cereals and 4 for pro- tein (World Food Programme, 2009). A score above 35 is considered acceptable, a score between 21 and 35 is borderline, and between 0 and 21 is poor. The measure has been validated as correlated with caloric intake; however, the cut-offs are low, so low scores capture more severe lack of basic caloric intake than the labels may suggest (Weismann et al., 2009). Food consumption measures the monthly value of household food consumption, in- cluding purchases, own production, and gifts. For high-frequency and midline surveys, we collect data on the 12 most commonly consumed items, while baseline and endline surveys cover 36 food items. Values are price-adjusted using a monthly deflator based on the three main cereals (millet, maize, rice) that comprise over 80% of food consumption, with baseline as the reference period. Psychological well-being is measured for the household head using Cantril’s 10-point ladder of life satisfaction, where respondents rate their current life position from worst (1) to best possible (10). Our secondary outcomes include mental health, financial behaviors and livelihood activities. Mental health is assessed using the CESD-R-10 depression scale administered 15 In our analysis of endline data, we use recollected food consumption information for approximately 300 households where initial data entry was incorrect, controlling for these cases with a dummy variable. 10 to the household head, which measures the frequency (0-7 days) of experiencing 10 depres- sive symptoms during the previous week. The index is standardized for each treatment arm within each survey round relative to the baseline mean and standard deviation.16 For financial behaviors, we measure monthly borrowing, savings and remittances aggregated to the household level. We also examine livelihood outcomes through indices covering off-farm business activities, wage employment, livestock rearing, and agricultural inputs and output (the latter at endline only). These livelihood indices aggregate several indi- cators to address multiple hypothesis testing concerns within each family of outcomes. The high-frequency and midline surveys use monthly recall periods for these measures; the endline survey captures the full rainy season for agricultural measures. We augment the agriculture and livestock indices using additional measures from our more detailed endline survey. The indices are standardized against the mean and standard deviation of the traditional response group for each survey round.17 Appendix Tables A1 to A3 provide complete variable definitions. 4.3 Baseline descriptives and balance Table 1 describes the study sample and shows that baseline characteristics are well- balanced across the three treatment arms for all outcome measures and other household characteristics. The study sample comprises highly impoverished rural households de- pendent on rain-fed agriculture and operating in an environment where humanitarian crises are recurrent. Our sample reports an average household size of 7.5 members and that 97% of household heads have not completed primary school. Livelihood activities are primarily agricultural, with 93% of households cultivating plots. Additionally, 53% operate an off-farm business and 37% own livestock, but very few households engage in wage employment. This is not only a poor population, but also highly vulnerable to extreme weather shocks. 89% of our sample reported experiencing a drought in the past year, reflecting the event that triggered the response. Food insecurity is prevalent: Half the sample have a mean Food Consumption Score below 21, classified as seriously deficient food intake by WFP standards, despite baseline measurements occurring well before the lean season. Monthly household food consumption averages 45,135 CFA francs (approximately USD $222 in 2022 PPP terms), while monthly non-food consumption accounts for only 5,070 CFA francs (approximately USD $25). With large household sizes, this translates to per capita consumption of barely USD PPP $1, well below the $2.15 extreme poverty threshold (2017 PPP), noting that our partial consumption measure captures approxi- 16 The CESD-R-10 is considered a screening tool. Since there is no locally validated threshold to directly derive the prevalence of depression, we use the standardized measure and do not comment on levels in the sample. 17 Not all livelihood variables were measured at baseline. 11 mately 80% of typical household expenditures. While only 5% of households report having any savings, 64% took out loans in the past year. In line with other work in Niger (e.g. Aker et al., 2016, Hoddinott et al., 2018), these are exclusively informal loans from local networks, moneylenders and traders. These loans are typically short-term: of the 31,390 CFA francs (USD $154) borrowed in the last year, more than half had already been repaid. Despite the high prevalence of borrowing, only 8% of households report being confident to be able to raise 40,000 CFA francs in case of need. Combined with very low savings, this underscores the population’s limited financial buffers in the immediate aftermath of the 2021 rainfall shortages. Finally, not only were the rains relatively poor, food prices exhibited pronounced seasonal movements during the year of our study as well: maize and millet prices in July and August 2022 (during the peak of the lean season) were about 40% higher than in November 2022, after the harvest (see Figures 4 and 5). We do not observe significant price differences across treatment arms. 5 Conceptual framework To understand how the timing of cash transfers may affect welfare outcomes, it is help- ful to consider how resource-constrained rural households allocate their minimal means across the agricultural cycle within a given year. This is particularly pertinent in an extreme drought context, where the intensity or frequency of the shock exceeds house- hold expectations (for instance, due to climate change) and consumption needs by far surpass precautionary savings. As discussed in the section above, our study population has poor nutritional outcomes, relies heavily on agriculture, lacks savings and has limited access to longer-term loans, while facing high seasonal fluctuations in labor needs and food prices. Although smoothing consumption levels might seem intuitively optimal, this strategy may not maximize welfare when food prices fluctuate strongly or when returns to better nutrition increase during peak labor demand periods. Both factors alter the opportunity cost of consumption across different periods of the agricultural cycle, despite concave utility offering gains from smoothing. We present a simple framework of household decision making for consumption and smoothing, explicitly accounting for food storage and seasonality. Figure 1 summarizes the agricultural cycle with its distinctive labor periods, as discussed in Section 2. There are three phases: a pre-lean dry season when land preparation and sowing occurs before the start of the rainy (lean) season; the rainy lean season when weeding and other tending take place as crops grow; and the post-lean season after harvest when agricultural activity is limited. We model this as an inter-year optimization problem with four “seasons” corresponding with these agricultural stages and our data collection periods, starting with an initial post-lean/harvest period, followed by the pre-lean and lean seasons with 12 considerable labor needs, and a second post-lean/harvest season where the returns of this labor are realized. Since these farm households face severe poverty and food insecurity, they have incentives to increase food intakes during periods of peak labor demand when the returns to nutrition or the real value of the cash transfer are highest. Food prices fluctuate significantly throughout this cycle, reaching their lowest during the point post- lean/harvest period and their highest during the lean season, as illustrated in Figures 4 and 5. We make five assumptions in this simple framework. First, households hold no savings in cash but only in kind as food.18 Second, food can be stored between seasons, but not across years.19 Third, we ignore intertemporal discounting for simplicity. This only imposes that, without differential returns to nutrition and seasonality in food prices, households would try to keep consumption levels constant over the seasons. Fourth, informal credit is typically short-term, in line with our data. Households also report not being able to raise emergency cash, so we assume that households are credit constrained between periods.20 Finally, we assume that agricultural output increases as a function of (food) consump- tion during labor-intensive periods – specifically during land preparation (in period 1) and weeding (in period 2) – with returns realized in the post-lean or harvest period (period 3). We assume that this function is increasing in consumption but at a decreasing rate in each period, so that output equals f (c1 , c2 ) with fi′ > 0 and fi′′ < 0 for each period i. This assumption on nutritional returns during peak labor periods creates an intertemporal link in decision making across years. With these assumptions, let us analyze a household’s decision making over four sea- sons t = 0, 1, 2, 3, i.e., from first post-lean/harvest period to the second post-lean/harvest period a year later, and up to T years with ty =τ,...,T . Consider that the household maximizes intertemporal utility across the seasons and years with instantaneous utility defined as u(ct ) in which ct is consumption in season t. We assume that this function is increasing and concave (u′ > 0 and u” < 0) as follows: T 3 V = u(cty t ) (1) ty t=0 We formulate budget constraints for each season within our model. Let At represent 18 Given the extent of seasonality, it is rational to save between the post-harvest and the lean season in kind, as cash does not earn a return, while food prices are up to 40% higher 9 months after the harvest. 19 We assume that storage after the rainy season and harvest is not possible, as humidity and pest infestations would ruin crop quality. Inter-year storage also does not make sense as prices collapse during the harvest period, even without considering depreciation. 20 We assume no inter-year savings, technological change, or access to credit markets. The planning horizon is therefore restricted to one agricultural year, regardless of the number of years over which households could theoretically optimize. 13 the food stock at the end of each season that can be carried over to the next period, subject to some depreciation or spoilage rate d. We define q0 as the quantity of harvest remaining in the first post-lean/harvest season after accounting for consumption in that period. CTt represents any additional income sources in season t, in particular the cash transfers from the program. We denote the three treatment arms as defined in Section 3 by j = 1, 2, 3. Assuming that full depreciation prevents stocks being carried over beyond the lean season, the budget constraint for each season within a year can be defined as follows, where we omit the year subscripts ty for simplicity: Period 0 (First Post-Lean/Harvest season): p 0 A 0 = p 0 · q 0 − p 0 · c0 (2) Period 1 (Pre-Lean): p1 A1 = p1 · A0 · (1 − d) − p1 · c1 + CT1j (3) Period 2 (Lean): 0 = p2 · A1 · (1 − d) − p2 · c2 + CT2j (4) Period 3 (Second Post-Lean/Harvest): p3 · A3 = p3 · f (c1 , c2 ) − p3 · c3 + CT3j (5) First-order conditions lead to the following Euler equations, governing how consump- tion will be allocated between seasons: u′ (c0 ) = (1 − d) · [u′ (c1 ) + fc1 · u′ (c3 )] (6) u′ (c1 ) + fc1 · u′ (c3 ) = (1 − d) · [u′ (c2 ) + fc2 · u′ (c3 )] (7) How to interpret (6) and (7)? In period 1 and 2, consumption generates utility both directly and indirectly through productivity returns realized in period 3. Therefore, as seen in equation (6), in period 0 consumption will be kept lower (marginal utility higher) than it would have been without this additional positive effect: households have incentives to consume more in period 1 than in period 0 to benefit from nutritional returns. Depreciation between periods will attenuate this effect. Similar factors are at play in equation (7) that captures allocation between periods 1 and 2, where the optimal consumption decision depends on the relative marginal productivity of nutrition in each period, adjusted for depreciation. Households will adjust consumption between these two periods until the marginal utility is equalized, after accounting for the productivity effect and depreciation. These Euler equations determine how cash transfers CTtj are allocated towards con- sumption across periods. Prices will matter too, even if they do not enter the Euler equations: transfer timing affects real purchasing power for food that can be bought and stored. For example, while the the traditional response cash transfer CT3,1 received in period 3 (treatment group 1) is equal to the early short response transfer CT2,2 received in period 2 (treatment group 2) in nominal terms, the real value differs substantially: CT2,2 p2 > CT 3,1 p3 , as p3 > p2 . 14 Importantly, this framework implies that rational households would not necessar- ily be expected to equalize consumption levels across seasons. Furthermore, nominally equivalent cash transfers may generate different welfare impacts due to three reasons. First, there is limited incentive to save food after the harvest and the early cash transfer for lean season consumption in monetary form due to the risk of spoilage of food stocks and varying prices. Second, seasonally varying prices mean that the real value of trans- fers will differ by modality, as their purchasing power and ability to accumulate food stocks is different at the time they are received. Third, productivity returns to nutrition generate incentives to allocate relatively more consumption to high labor demand periods – particularly the pre-lean season when land needs to be prepared and during the rainy lean season for crop tending. Which effect will dominate is an empirical matter, as is the question about whether the nutritional and labor requirements are highest before or during the lean season. What could one expect from the different cash transfer modalities? First, households are very food insecure, as the baseline illustrated. We could therefore expect a boost of consumption as soon as the transfers from each modality are received, even if there will be incentives to smooth consumption in the remaining periods before the next harvest. Second, accumulating food stocks is possible, although depreciation may limit smoothing into the lean season from the “early long” and “early short” modalities. Nevertheless, by getting cash earlier, households could buy food at lower prices and carry over stocks. In fact, using transfers to increase food stocks rather than consuming can be a high return investment. Third, households may attempt to boost consumption when labor effort is most intense in the pre-lean and lean season, implying that “early long” and “early short” will boost consumption earlier than just trying to buffer the lean season. This model helps with the interpretation of the finding discussed below. To evaluate the overall welfare effects across cash transfer modalities, we will assess the evolution of outcome levels and volatility over time, assessing the degree to which households smooth their welfare between seasons. 6 Empirical strategy Our empirical strategy assesses how early cash transfers impact household welfare out- comes compared to traditional lean season response in the year following a severe drought. We estimate intent-to-treat effects within each survey round (pre-lean, lean, post-lean/harvest, and endline coinciding with the subsequent pre-lean season) using the following specifi- cation: Yhi = β0 + β2 · T 2i + β3 · T 3i + γ · Xhi + δ + εhi (8) 15 Where Yhi is the outcome for household h in village i. The treatment indicators T 2i and T 3i denote assignment to “early short” and “early long” cash transfer treatments respectively, with the traditional response cash transfer modality (T 1i ) serving as the control group. δ represents randomization strata fixed effects (based on village size and PMT strata) used in treatment assignment and cohort fixed effects to account for the group in which households were surveyed within each round of high-frequency surveys.21 For robustness, we present specifications with additional baseline controls (Xhi ) selected using the post-double-selection LASSO procedure of Belloni et al. (2014). εhi is a mean zero error term. We report robust standard errors clustered at the village level. To address multi- ple inference concerns, we adjust for the false discovery rate across our three primary outcomes within each survey round using sharpened q -values following Benjamini et al. (2006) two-stage procedure, implemented as in Anderson (2008). The coefficients β2 and β3 capture the intent-to-treat effect of the early short cash transfers (T2) and early long cash transfers (T3), relative to the traditional response modality (T1). Given that this was a response to a severe drought shock, it was not possible to include a pure control group with no cash transfers. As such, we document impacts in each round relative to the default traditional response, with negative coeffi- cients indicating periods where the traditional response proved more effective. 7 Results 7.1 Transfer timing matters for welfare Our primary research question examines the impact of the timing of cash transfers on household welfare in response to drought and seasonality. Using Equation (8), Table 2 presents results for our two main economic welfare outcomes: food security, measured by the Food Consumption Score (FCS) (top panel) and monthly household food consumption (bottom panel). We analyze these outcomes across four distinct time periods, represented in each column: pre-lean season, lean season, post-lean season and endline (nine months after the 2022 lean season and late in the 2023 pre-lean season). When interpreting results, recall that this study follows a drought year with poor baseline food security: the mean FCS is below the defined ”poor” threshold of 21, with only half the sample above the threshold (Table 1). 21 We also control for a dummy variable indicating whether households were surveyed twice. Some households were surveyed twice due to one of two reasons: i) we initially designed three pre-lean season survey cohorts but expanded to four cohorts after collecting data for the first cohort to match the frequency of early payments, which required resampling some households, and ii) we needed to resurvey approximately 300 households at endline, because their food consumption data was incorrectly collected in the first survey. In both cases, we retain the second survey. 16 7.1.1 Pre-lean and lean season effects We find that large, early cash transfers (“early short”) significantly improve welfare before the lean season relative to other modalities (Table 2, column 1). Compared to the tradi- tional response group (mean FCS of 22.8), the early short group reports an 8% increase in FCS (+1.85 FCS points) and 17.6% higher monthly food consumption (+8,192 CFA francs). The early short group also has 11 percentage points fewer households report- ing poor food security (Table A5). Furthermore, transfer size matters: the large early transfers have larger effects than the smaller year-long transfers delivered at the same time during the pre-lean season period. While the “early long” modality with smaller, year-long transfers increases monthly food consumption by 2,475 CFA francs compared to the traditional response, this effect is only about one-third that of the early short modality, which provides transfers three times larger. The early long modality also does not significantly improve food security relative to the traditional response.22 During the lean season, the traditional response induces larger effects on welfare compared to both early interventions (Table 2, column 2). Importantly, however, the early short modality still yields greater net welfare benefits when considering the pre- lean and lean seasons combined. Compared to early short which no longer receives transfers, the traditional response increases food security by 4.6% (+1.17 FCS points) and reduces households with poor food security by 6 percentage points (Table 2 and Table A5, column 2). Monthly food consumption is 5.6% higher (+3,400 CFA francs). Similarly, the traditional response has significantly higher monthly food consumption (+4,150 CFA francs) compared to the early long group, although there is no significant difference in food security.23 We also cannot reject that both early interventions have similar effects during the lean season. We consider not only relative welfare impacts within a period, but also changes in levels over time in line with the conceptual framework. We find that large early transfers yield greater net benefits before and during the lean season compared to the traditional response. Notably, the gains in pre-lean season consumption for the early short group compared to the traditional response (+8,192 CFA francs) is significantly larger (p =0.06) than what the traditional response gains during the lean season (+3,398 CFA francs). The point estimate for food security is also larger for the early short group in the pre- lean season, although we cannot reject that the estimates are of a similar magnitude for both groups across the two periods (p =0.41). Moreover, the early short group achieves 22 Before receiving any transfers, the “traditional response” group shows slight improvements in food security and consumption compared to the baseline, for instance with a mean pre-lean season FCS of 22.8 (Table 2 and Table A5, column 1). This is consistent with higher nutritional intake during periods of increased labor needs. We cannot simply state that this would be equivalent to a pure control group without any intervention, as this group would have been aware that some support was upcoming. 23 During the lean season, the traditional response modality induces higher consumption of rice, pasta, salt, and peanut oil relative to the early interventions (Table A7, second panel), with improvements in dietary diversity driven by oil (Table A6, second panel). 17 smoother food security across pre-lean and lean seasons: we cannot reject equal mean FCS across these two periods using a nested regression (p =0.45; Table A4), while we strongly reject smoothing for other modalities.24 Our findings remain consistent when we examine the proportion of households above the poor food security threshold (Table A5): the early short group maintains a more stable share of households above the threshold (61%), whereas the other modalities experience larger fluctuations. The pre-lean season gains for the early short group also tend to be larger than the improvements in the traditional response group during the lean season, although the difference is not statistically significant (p =0.28). 7.1.2 Post-lean season and endline effects Beyond the lean season, there is no differential welfare effect between the early inter- ventions and the traditional response in the immediate months after harvest (Table 2, column 3) and 9 months later at endline (column 4). Food security improves between the lean and post-harvest seasons by approximately 14% for the early short group (to 29 FCS points), and there are no significant differences across modalities. Food consump- tion also increases for all groups. The early long group that still receives small transfers experiences a slightly larger effect on food consumption compared to early short (+3,512 CFA francs; p =0.06). By endline, all differences between modalities have disappeared. The endline survey is conducted approximately 6 months after the last early long transfer and during a similar phase of the agricultural cycle as the pre-lean season survey – right before the next lean season. Average outcomes have returned to levels similar to baseline and the pre-lean season without transfers, suggesting substantial constraints on inter-year consumption smoothing. All modalities deliver relatively high transfer amounts in total, yet adjusting the timing or frequency of these transfers does not lead to relatively better or more resilient livelihoods, as discussed in Section 7.3 below. 7.1.3 Robustness Our results remain robust to multiple hypothesis corrections (Table 2), the inclusion of baseline controls (Table A8) and more detailed food consumption measures used during the lean season and at endline (Table A9, column 2). We find no evidence that transfers differentially affect village-level prices of key agricultural products (Figures 4 to 6).25 24 We reject a difference in mean monthly food consumption between the pre-lean and lean season periods for the early short group (smoothing test: p =0.06). When we account for the post-lean and endline periods, we reject differences in means for both measures, in line with our conceptual framework. 25 These figures show village-level prices of millet, maize and rice for each month and treatment arm. While prices do vary during the year, notably with lower price for maize and millet following the 2022 harvest, we do not find that the cash transfer modalities differentially affect prices. 18 Tables A10 and A11 provide disaggregated results by survey rounds for completeness, showing largely consistent, albeit noisier results. Going beyond averages, we find no systematic differences for sub-groups based on baseline access to liquid assets or diversified labor market activities, as specified in our pre-analysis plan.26 7.1.4 Discussion Overall, our findings demonstrate that large early cash transfers yield greater net welfare benefits in responding to drought and seasonality relative to both the traditional lean season response and smaller year-long transfers. This holds when considering the pre- lean and lean seasons together, even without lasting differential effects. Our results are consistent with the conceptual framework, which highlights how households may rationally increase consumption levels during critical agricultural periods with high nutritional and labor needs. This underscores the role of the timing of cash transfers. Throughout the experiment, households consistently experience FCS below the “adequate” food security threshold of 35 (Weismann et al., 2009), indicating immediate food needs. The early short group receives transfers during the pre-lean period when labor requirements are high yet food prices remain below lean season levels. Thus, our results on the benefits of large, early cash transfers are consistent with both high marginal returns to nutrition during intensive labor periods and higher real value of transfers when received at lower pre-lean season prices, facilitating intertemporal consumption smoothing into the lean season. We find that households receiving early cash transfers intensify consumption of both usual cereals (millet) and more preferred, expensive grains (rice), while diversifying towards other energy-rich food groups, in line with our conceptual framework and other research (Aker et al., 2016; Hoddinott et al., 2018; Webb et al., 1994; Subramanian and Deaton, 1996; Banerjee and Duflo, 2007; Aker et al., 2016).27 Moreover, the early short group achieves greater smoothing in consumption levels across both pre-lean and lean seasons compared to other modalities. Conversely, the results show that the sole expectation of future transfers is not sufficient for households in the traditional response group to improve their food security and food consumption before 26 In our pre-analysis plan, we hypothesized that households with differing asset positions at baseline might respond differently to the cash transfers in the aftermath of the drought shock. We had also hypothesized that households with more diversified non-agricultural livelihood activities at baseline might be better positioned to intensify these activities using the cash transfers. We do not find robust evidence of heterogeneous treatment effects by baseline access to liquid assets or diversified labor market activities, as specified (Tables A28 and A29). Since most households in the sample are extremely poor and faced a common drought shock before the baseline, there may not be sufficient variation between households to detect heterogeneity. 27 Tables A6 and A7 further unpack food security and food consumption results by food group. Dif- ferential impacts from the large early transfers on food security before the lean season stem from more frequent consumption of cereals, dairy products, oil and sugar (Table A6, top panel). This is consistent with results on food consumption, showing higher consumption of millet, rice, sugar, salt, palm and peanut oil, along with a small increase in meat consumption (Table A7, top panel). 19 the lean season.28 While the consumption patterns before and during the lean season appear consistent with differential labor needs, our findings indicate that the modalities have limited differ- ential impacts on welfare after harvest and into the next agricultural cycle. Post-harvest outcomes improve for all modalities with higher food consumption and food security than any other period, with the early long group able to sustain more stable levels of consumption. Inter-year credit constraints appear binding: our sample of low-income rural households has difficulty in smoothing consumption across years and is unable to borrow against uncertain future harvests. All groups have similar welfare outcomes by endline. The benefits from large early transfers thus likely stem from their timing dur- ing a pre-lean season with high labor demands following drought, enabling households to smooth consumption into the lean season, without generating differential productivity ef- fects or sustained improvements in lifetime wealth trajectories or food security. Without a pure control group, we cannot determine whether transfers improved outcomes relative to receiving no assistance. Nevertheless, our findings are consistent with other studies finding that liquidity provided for seasonality has limited effects on consumption after harvest (Burke et al., 2019; Fink et al., 2020). 7.2 Early cash boosts psychological well-being To what extent are these patterns reflected in other dimensions of well-being? Table 3 presents the effects of transfer timing and frequency on psychological well-being, specif- ically life satisfaction (a third pre-specified outcome) and a mental health index based on the widely-used Center for Epidemiological Studies Depression Scale (CESD-R-10) (Radloff, 1977). The results for psychological well-being align closely with our findings on food se- curity and consumption. Before the lean season (Table 3, column 1), both early inter- ventions lead to immediate improvements in life satisfaction compared to the traditional response – the early short group by 17.8% (+0.57 points from the traditional response group mean of 3.22) and the early long group by 9.8% (+0.32 points).29 The effect is sig- nificantly higher for the larger early transfers compared to the small year-long transfers, by 0.25 points (p =0.01). During the lean season (Table 3, column 2), the traditional response group experi- ences the largest increase in life satisfaction, resulting in a significantly higher level than 28 Since these households are receiving cash transfers from the national safety nets agency for the first time, trust in receiving cash transfers may also be low ex ante. 29 It is worth noting that the mean level observed in each of these groups, despite cash transfers, is very low by international standards, and even for Niger. The World Happiness Report (2023) based on the Gallup Poll (Helliwell et al., 2023) reported an average for 2022 for Niger of 4.50 while the global average is 5.23. 20 the early short group (at p ≤0.10)). However, the early short transfers appear to have lasting benefits across the pre-lean and lean seasons, with sustained psychological well- being during the lean season even after transfers cease. Although the traditional response group reports higher life satisfaction compared to the early short group during the lean season, the relative gain is significantly smaller than the pre-lean season impact of early cash (0.18 point difference in column 2 versus 0.57 point difference in column 1; p =0.01 in column 5). Moreover, there is no significant drop in average life satisfaction for the early short group between the pre-lean and lean seasons, suggesting these transfers help mitigate fluctuations in well-being across periods of high labor demand and heightened seasonal price pressures.30 The results of a standardized mental health index follow a similar pattern, which indicates robust results, as reported in Table 2 (bottom panel).31 Before the lean season, both the early short and the early year-long groups show immediate significant improve- ments in mental health by 0.28 and 0.22 standard deviation, respectively, compared to the traditional response group. During the lean season, these early mental health ben- efits persist, with no significant differences between modalities despite the traditional response group now receiving transfers and experiencing a similar boost in psychological well-being. Looking at means across the pre-lean and lean season periods, these results align with earlier findings that the early short group sustained psychological well-being throughout the stressful lean season without additional transfers, whereas the traditional response and early year-long groups maintain similar levels, despite receiving transfers. In other words, the early transfer modality helps households avoid large fluctuations in well-being, with net gains compared to the traditional response.32 Beyond the lean season, the differences in psychological well-being across transfer modalities dissipate - consistent with the economic welfare results. By endline, mean life satisfaction for the traditional response group is not far from the levels observed a year earlier during the pre-lean season before transfers started. Similarly, the mental health index converges between modalities, meaning that there is no differential effect by transfer timing and frequency in the medium term (columns 3 and 4). By the next pre-lean season at endline, indicators have reached similar levels for all, albeit somewhat higher than at baseline.33 30 Using a nested regression, we cannot reject a difference in mean life satisfaction between the pre-lean and lean season periods for the early short group (smoothing test: p =0.16), but do reject a difference for the two other modalities (see Table A4). 31 The index is standardized using the baseline, so the effects are changes expressed in percentages of the standard deviation, while levels can be compared across periods with a mean of 0 at baseline. 32 The pre-lean season mental health benefit of the early short group over the traditional response is significantly larger than the traditional response’s boost during the lean season (p =0.07 in Table 3, column 5), indicating a net gain from early transfers across these critical periods. Moreover, using a nested regression, we cannot reject a difference in the mean mental health index between the pre-lean and lean season periods for the early short group (smoothing test: p =0.58), but do reject a difference for the two other modalities (see Table A4). 33 To complement results on psychological well-being, we also consider proxies of social well-being. 21 Our findings suggest that early cash transfers provide households greater “peace of mind” as they enter the lean season, mirroring our findings on food security and consumption.34 Next, we explore changes in livelihoods and financial behaviors behind the economic welfare and psychological well-being results. 7.3 Transfer timing has no effect on livelihoods The temporary cash transfers we study were primarily designed to improve welfare and food security in response to shocks, and not necessarily to boost livelihoods or income- generating activities. Nevertheless, other studies, including in Niger, have shown that cash transfers can induce changes in livelihood activities that help households mitigate the welfare effects of rainfall shortages, either through more diversified income from off-farm activities or an intensification of agricultural production (Macours et al., 2022; Premand and Stoeffler, 2022). We thus test whether alternative transfer modalities have differential impacts on livelihoods. For instance, early transfers that are provided before the main agricultural season may facilitate the purchase of inputs, whereas the traditional response may enable critical investments during the season itself. Transfer size may also play a role, with larger transfers potentially fostering greater investments in income-generating activities compared to smaller, regular streams (Crosta et al., 2024). The net effects on livelihood outcomes across transfer modalities are thus ambiguous a priori. Figure 2 presents treatment effects for livelihood indices capturing agricultural in- puts, agricultural output (measured at endline only), livestock, off-farm business activities and wage employment35 for the two early cash interventions, standardized relative to the traditional response group.36 We find no differential effect on any of the livelihood indices across transfer modalities throughout the study period. Most point estimates are very close to zero, and all point estimates are below 0.07 standard deviation. Of course, the absence of a pure control group means that we cannot rule out the possibility that all On the one hand, targeted cash transfer programs can induce positive effects on social dynamics within communities (Premand and Barry, 2022; Bossuroy et al., 2022). On the other hand, a priori, they have ambiguous effects on engagement in conflict (Premand and Rohner, 2024). Table A13 presents results for social cohesion and perceptions of violence. We find no significant difference in these social outcomes across modalities over time, with one exception: the early long group shows a 0.08 standard deviation improvement in perceptions of violence at endline compared to the traditional response group. High levels of social cohesion persist across all cash transfer groups throughout the study period, with at least 94% of households reporting no tension between community members at any given period. While this measure is crude and likely subject to misreporting, the high cohesion levels across all cash transfer groups suggest that targeting transfers to the most vulnerable households was not perceived as inequitable. 34 The life satisfaction results remain consistent under various robustness checks, including multiple hypothesis corrections and the inclusion of baseline controls (Table A8). Table A12 provides disaggre- gated results on life satisfaction by survey rounds for completeness, again showing largely consistent results with our pooled results. 35 Note that with only 6% of households engaged in any wage employment at baseline, including both agricultural and non-agricultural wage employment, such activity remains rare in our rural study setting. 36 Tables A16 - A21 provide the point estimates, and results for the subcomponents are presented in Tables A22 - A27. 22 modalities had similar impacts on livelihood outcomes. Still, the lack of differential effect explains why we observe no significant difference across transfer modalities in welfare out- comes during the post-lean (harvest) period and at endline. Our findings are consistent with the conceptual framework: it shows that the additional liquidity provided by the cash transfer was used to boost food intake and thus meet labor (productivity) needs at critical periods in the agricultural cycle, rather than altering investment decisions in the immediate aftermath of a severe climatic shock. 7.4 Timing of cash affects financial behaviors Although shifting the timing of cash transfers has no impact on livelihood investments, it significantly alters financial behaviors, particularly debt management in the aftermath of the drought. Rural households in Niger typically make extensive use of short-term informal credit from local networks, moneylenders and traders (see Aker et al., 2016 and Hoddinott et al., 2018). At baseline, households had borrowed 31,390 CFA francs (USD $154) in the past year, with more than half already repaid. Nevertheless, few households (only 8%) report being able to raise 40,000 CFA francs in case of need. Notably, households already take on substantial debt within a few months after the drought shock, with the traditional response group borrowing over 8,000 CFA francs per month on average in the absence of cash transfers (Tables 4). In fact, 56% of households in the traditional response group take out loans in the four months before the lean season. Again, this underscores the severe financial constraints households face, requiring loans to meet consumption needs even before the lean season. Before and during the lean season, cash transfers are used to partially substitute for borrowing. In the pre-lean season period, both early transfer modalities lead to large reductions in monthly borrowing incidence by 18% to 29% (10 to 16 percentage points) and lowered monthly amounts by 24% to 38% (3,130 CFA francs and 1,958 CFA francs) for early long and early short groups respectively compared to the traditional response group. However, even recipients of early transfers still take on substantial debt, pointing to a highly stressed setting. During the lean season, the pattern again reverses: the traditional response group is less likely to borrow relative to the early long and early short groups (by 11 and 13 percentage points, respectively) and borrowed smaller amounts (962 to 2,035 CFA francs or 22% to 34%, respectively). After the lean season, differences in financial behaviors between the cash transfer modalities diminish, with the early long group showing slightly lower borrowing immedi- ately post-harvest. By endline, all modalities again converge.37 37 While the high-frequency and midline surveys capture borrowing activity during the preceding month, the endline survey covers a 12-month period, and measures are subsequently converted to monthly averages. The magnitudes at endline are thus not comparable with earlier survey rounds. 23 Unlike borrowing, the impacts of transfer timing on savings and remittances are lim- ited. Very few households report savings, and the amounts remain small (Table A14). The large early transfers modestly increase the likelihood of saving in the pre-lean season, from a low base of 3%.38 We detect no differential effect on saving behavior by trans- fer modality during the lean season, although the traditional response households save marginally more than the early short group post-lean season.39 By the endline, no signif- icant difference persists. Monthly remittances also show limited changes across transfer modalities throughout the study (Table A15). However, at endline, the early long group that receives smaller year-long cash transfers reports significantly lower remittances over the 6-month recall period (by 560 CFA franc) compared to the traditional response group, suggesting some substitution between informal support and government assistance. 8 Conclusion This paper shows that the effectiveness of crisis response can be improved by optimizing the timing and frequency of cash transfers. We find that there is value in intervening early before the lean season to mitigate the adverse effects of seasonality and drought. Large early cash transfers have more pronounced positive impacts on food security, con- sumption, and psychological well-being before the lean season compared to the impacts induced by a traditional response during the lean season. Some of the effects of large early transfers persist throughout the lean season, with households using cash to pur- chase grain and diversify consumption. This suggests that large early transfers better support households to smooth in the short term and immediately after drought shocks. Large early transfers also tend to have larger effects than smaller, year-long transfers, highlighting the importance of a sufficient transfer size. Our findings shed light on how households adjust their consumption in a context of high seasonality and large price fluctuations. Large early cash transfers generate effects consistent with two key mechanisms in our framework: enhanced productivity from better nutrition during periods of high labor demand, and greater real value of transfers when food prices are seasonally low, both facilitating intertemporal consumption smoothing. The results show that early shock-responsive cash transfers fulfill their primary ob- jective of supporting food security and welfare after shocks. However, varying transfer timing and frequency has limited welfare effects beyond the harvest and in the months leading to the next agricultural season. Consistent with this, adjustments in transfer 38 Compared to both the traditional response group (where 2.7% of households save) and the early long group, the proportion of households who save increases by approximately 3 percentage points in the early short group. The amount saved monthly also doubles (by 260 CFA francs), but the increase is not statistically significantly different from zero. 39 This could indicate either a persistent positive effect on savings from the traditional response inter- vention, or that the early short group had already built up adequate savings earlier in the year. 24 modality do not significantly impact livelihood decisions throughout the year, suggesting that modifying the timing and size of cash transfers alone cannot alter patterns of produc- tive investments after drought. These limited differential effects post-lean season show that varying transfer modalities is unlikely to induce longer-term effects, while noting the limitations of not having a pure control group. For building resilience, longer-term programs established ex-ante before the onset of shocks may be more effective, such as multi-year safety nets (Premand and Stoeffler, 2022) or those incorporating economic inclusion or livelihood components to promote economic diversification (Bossuroy et al., 2022; Macours et al., 2022). While our findings demonstrate the value of early response, whether an even earlier intervention would be more effective remains an open question. The program we study already significantly accelerated the response by using satellite data to detect rainfall shortages at the end of the agricultural season. However, monitoring and forecasting technologies to anticipate drought onset are still underdeveloped in Niger and many other countries. In contrast, flood forecasting is more advanced, with anticipatory cash transfers for floods showing short-term benefits, albeit with limited medium-term effects (Balana et al., 2023; Dunsch et al., 2025). Similarly, guaranteed credit schemes with ex- ante contributions and ex-post payouts tested for floods (Lane, 2024) could prove valuable in drought contexts where the shock can only be observed in real time. Future research should weigh the benefits of earlier interventions against the risk of targeting errors and examine the long-term effects of shock-responsive early transfers in settings where pure control groups can be introduced. 25 References Adhvaryu, Achyuta, Nyshadham, Anant, Molina, Teresa, and Tamayo, Jorge (2022), ‘Helping children catch up: Early life shocks and the PROGRESA experiment’, Eco- nomic Journal 132(645). Aker, Jenny (2012), ‘Rainfall shocks, markets and food crises: the effect of drought on grain markets in Niger’, Center for Global Development, working paper. Aker, Jenny (2017), ‘Comparing Cash and Voucher Transfers in a Humanitarian Context: Evidence from the Democratic Republic of Congo’, World Bank Economic Review 31(1), pp. 44–70. Aker, Jenny, Boumnijel, Rachid, McClelland, Amanda, and Tierney, Niall (2016), ‘Pay- ment mechanisms and antipoverty programs: Evidence from a mobile money cash transfer experiment in Niger’, Economic Development and Cultural Change 65(1), pp. 1–37. Anderson, Michael L. (2008), ‘Multiple inference and gender differences in the effects of early intervention: A reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects’, Journal of the American Statistical Association 103(484), pp. 1481– 1495. Balana, Bedru, Adeyanju, Dolapo, Clingain, Clare, Andam, Kwaw, Brauw, Alan de, Yohanna, Ishaku, Olarewaju, Olukunbi, and Schneider, Molly (2023), ‘Anticipa- tory cash transfers for climate resilience: Findings from a randomized experiment in Northeast Nigeria’, IFPRI Working Paper. Baland, Jean-Marie, Bonjean, Isabelle, Guirkinger, Catherine, and Ziparo, Roberta (2016), ‘The economic consequences of mutual help in extended families’, Journal of Devel- opment Economics 123, pp. 38–56. Banerjee, Abhijit V. and Duflo, Esther (2007), ‘The Economic Lives of the Poor’, Journal of Economic Perspectives 21(1), pp. 141–168. Bazzi, Samuel, Sumarto, Sudarno, and Suryahadi, Asep (2015), ‘It’s all in the timing: Cash transfers and consumption smoothing in a developing country’, Journal of Economic Behavior Organization 119, pp. 267–288. Beaman, Lori, Karlan, Dean, Thuysbaert, Bram, and Udry, Christopher (2023), ‘Selec- tion Into Credit Markets: Evidence From Agriculture in Mali’, Econometrica 91 (5), pp. 1595–1627. Beegle, Kathleen, Galasso, Emanuela, and Goldberg, Jessica (2017), ‘Direct and indirect effects of Malawi’s public works program on food security’, Journal of Development Economics 128, pp. 1–23. Behrman, Jere R (1993), ‘The economic rationale for investing in nutrition in developing countries’, World Development 21(11), pp. 1749–1771. 26 Belloni, Alexandre, Chernozhukov, Victor, and Hansen, Christian (2014), ‘High-dimensional methods and inference on structural and treatment effects’, Journal of Economic Perspectives 28(2), pp. 29–50. Benjamini, Yoav, Krieger, Abba M., and Yekutieli, Daniel (2006), ‘Adaptive linear step- up procedures that control the false discovery rate’, Biometrika 93(3), pp. 491–507. Birkmann, J., Liwenga, E., Pandey, R., Boyd, E., Djalante, R., Gemenne, F., Leal Filho, W., Pinho, P. F., Stringer, L., and Wrathall, D. (2022), ‘Poverty, Livelihoods and Sustainable Development’, in: Climate Change 2022: Impacts, Adaptation and Vul- nerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, ed. by H.-O. P¨ ortner, D. C. Roberts, M. Tignor, E. S. Poloczanska, K. Mintenbeck, A. Alegr´ ıa, M. Craig, S. Langsdorf, S. L¨oschke, V. M¨ oller, A. Okem, and B. Rama, Cambridge, UK: Cambridge University Press, pp. 1171–1274. Boltz, Marie, Marazyan, Karine, and Villar, Paola (2019), ‘Income hiding and informal redistribution: A lab-in-the-field experiment in Senegal’, Journal of Development Economics 137, pp. 78–92. Bossuroy, Thomas, Goldstein, Markus, Karimou, Bassirou, Karlan, Dean, Kazianga, Harounan, Parient´ e, William, Premand, Patrick, Thomas, Catherine C, Udry, Christo- pher, Vaillant, Julia, and Wright, Kelsey A (2022), ‘Tackling psychosocial and capital constraints to alleviate poverty’, Nature 605. Brouillet, Audrey, Sultan, Benjamin, Barnes, Clair, Zachariah, Mariam, Otto, Friederiek EL, Vahlberg, Maja, Singh, Roop, Guigma, Kiswendsida, Heinrich, Dorothy, Kane, Cheikh, et al. (2022), Food crisis in Central Sahel in 2022 driven by chronic vulnera- bility with uncertain role of climate change, Imperial College London: World Weather Attribution Scientific Report. Burke, Marshall, Bergquist, Lauren Falcao, and Miguel, Edward (2019), ‘Sell Low and Buy High: Arbitrage and Local Price Effects in Kenyan Markets’, The Quarterly Journal of Economics 134 (2), pp. 785–842. Christiaensen, Luc (2017), Agriculture in Africa–Telling myths from facts: A synthesis. Christian, Paul, Kelley, Erin, Holmlund, Marcus, Premand, Patrick, Trako, Iva, Lane, Gregory, Heirman, Jonas, Paulose, Hanna, Fernandez, Chloe, Jospe, Eric, and Jeong, Dahyeon (June 2022), Climate and Resilience Impact Evaluation Window: Experi- mental Evidence from Several Countries, AEA RCT Registry. Crosta, T., Karlan, D., Ong, F, R¨ ohler, F., and Udry, C. (2024), Unconditional uschenp¨ Cash Transfers: A Bayesian Meta-Analysis of Randomized Evaluations in Low and Middle Income Countries, NBER Working Paper No. 32779. Cunha, Jesse M. (2014), ‘Testing paternalism: Cash versus in-kind transfers’, American Economic Journal: Applied Economics 6(2), pp. 195–230. 27 Cunha, Jesse M., De Giorgi, Giacomo, and Jayachandran, Seema (2019), ‘The Price Ef- fects of Cash Versus In-Kind Transfers’, Review of Economic Studies 86(1), pp. 240– 281. Dercon, Stefan (2004), ‘Growth and shocks: Evidence from rural Ethiopia’, Journal of Development Economics 74(2), pp. 309–329. Di Falco, Salvatore, Lokina, Razack, Martinsson, Peter, and Pin, Paolo (2019), ‘Altruism and the pressure to share: Lab evidence from Tanzania’, Plos one 14(5), e0212747. Dunsch, F., Adusumalli, N., Balantrapu, T., Batmunkh, O., Christian, P., Heirman, J., Kelley, E., Kondylis, F., Lane, G., Malhotra, K., Moreno, S., and Paulose, H. (2025), Impact of Anticipatory Action for Floods in Nepal, Impact Evaluation Policy Report, Rome: World Food Programme Office of Evaluation. Fink, G¨ unther, Jack, B. Kelsey, and Masiye, Felix (2020), ‘Seasonal liquidity, rural la- bor markets, and agricultural production’, American Economic Review 110 (11), pp. 3351–3392. Hallegatte, Stephane, Bangalore, Mook, Bonzanigo, Laura, Fay, Marianne, Kane, Tamaro, Narloch, Ulf, Rozenberg, Julie, Treguer, David, and Vogt-Schilb, Adrien (2016), Shock Waves: Managing the Impacts of Climate Change on Poverty, Washington, DC: World Bank. Haushofer, Johannes and Shapiro, Jeremy (2016), ‘The short-term impact of uncondi- tional cash transfers to the poor: Experimental evidence from Kenya’, Quarterly Journal of Economics 131(4), pp. 1973–2042. Helliwell, John F., Layard, Richard, Sachs, Jeffrey D., Neve, Jan-Emmanuel De, Aknin, Lara B., and Wang, Shun (eds.) (2023), World Happiness Report 2023, New York: Sustainable Development Solutions Network. Hidrobo, Melissa, Hoddinott, John, Peterman, Amber, Margolies, Amy, and Moreira, Vanessa (2014), ‘Cash, food, or vouchers? Evidence from a randomized experiment in northern Ecuador’, Journal of Development Economics 107, pp. 144–156. Hoddinott, John, Sandstr¨ om, Susanna, and Upton, Joanna (2018), ‘The Impact of Cash and Food Transfers: Evidence from a Randomized Intervention in Niger’, American Journal of Agricultural Economics 100(4), pp. 1032–1049. Jensen, Nathaniel D., Barrett, Christopher B., and Mude, Andrew G. (2017), ‘Cash trans- fers and index insurance: A comparative impact analysis from northern Kenya’, Jour- nal of Development Economics 129, pp. 14–28. Jeong, Dahyeon and Trako, Iva (2022), Cash and In-Kind Transfers in Humanitarian Settings: A Review of Evidence and Knowledge Gaps, World Bank Policy Research Working Paper 10026, World Bank. Kakpo, Ange, Mills, Bradford F, and Brunelin, St´ ephanie (2022), ‘Weather shocks and food price seasonality in Sub-Saharan Africa: Evidence from Niger’, Food Policy 112, p. 102347. 28 Kaminski, Jonathan, Christiaensen, Luc, and Gilbert, Christopher L (2014), The End of Seasonality? New Insights from Sub-Saharan Africa, World Bank Policy Research Working Paper 6907. Kansikas, Carolina, Mani, Anandi, and Niehaus, Paul (2023), Customized Cash Transfers: Financial Lives and Cash-flow Preferences in Rural Kenya, Working Paper 30930, National Bureau of Economic Research. Kazianga, Harounan and Udry, Christopher (2006), ‘Consumption smoothing? Livestock, insurance and drought in rural Burkina Faso’, Journal of Development Economics 79 (2), pp. 413–446. Lane, Gregory (2024), ‘Adapting to Climate Risk With Guaranteed Credit: Evidence From Bangladesh’, Econometrica 92(2), pp. 355–386. Macours, Karen, Premand, Patrick, and Vakis, Renos (2022), ‘Transfers, Diversification and Household Risk Strategies: Can Productive Safety Nets Help Households Manage Climatic Variability?’, The Economic Journal 132(647), pp. 2438–2470. Mani, Anandi, Mullainathan, Sendhil, Shafir, Eldar, and Zhao, Jiaying (2013), ‘Poverty impedes cognitive function’, Science 341(6149), pp. 976–980. Morduch, Jonathan (1995), ‘Income Smoothing and Consumption Smoothing’, Journal of Economic Perspectives 9, pp. 103–114. Paxson, Christina (1993), ‘Consumption and Income Seasonality in Thailand’, Journal of Political Economy 101 (1), pp. 39–72. Pople, Ashley, Hill, Ruth, Dercon, Stefan, and Brunckhorst, Ben (2024), The Importance of Being Early: Anticipatory Cash Transfers for Flood-affected Households, CSAE Working Paper. Premand, P, Christian, P, Fernandez, C, Heirman, J, Jospe, E, Waidler, J, Bajpai, P, Eltoukhi, O, and Rajaram, H (2024), Impact Evaluation of an Integrated Resilience Programme in Niger. World Food Programme Office of Evaluation, Impact Evalua- tion Policy Report, Rome: World Food Programme Office of Evaluation. Premand, Patrick and Barry, Oumar (2022), ‘Behavioral change promotion, cash trans- fers and early childhood development: Experimental evidence from a government program in a low-income setting’, Journal of Development Economics 158. Premand, Patrick and Rohner, Dominic (2024), ‘Cash and Conflict: Large-Scale Exper- imental Evidence from Niger’, American Economic Review: Insights 6(1), pp. 137– 53. Premand, Patrick and Stoeffler, Quentin (2022), ‘Cash transfers, climatic shocks and resilience in the Sahel’, Journal of Environmental Economics and Management 116. Radloff, Lenore Sawyer (1977), ‘The CES-D Scale: A Self-Report Depression Scale for Re- search in the General Population’, Applied Psychological Measurement 1 (3), pp. 385– 401. Reardon, Thomas and Matlon, Peter (1989), ‘Seasonal food insecurity and vulnerability in drought-affected regions of Burkina Faso’, in: S. David (ed.), Seasonal Variability 29 in Third World Agriculture: The Consequences for Food Security, Baltimore, MD: Johns Hopkins University Press. Rohde, M.M. (2023), ‘Floods and droughts are intensifying globally’, Nature Water 1, pp. 226–27. Skoufias, Emmanuel, Unar, Mishel, and Gonzalez de Cossio, Teresa (2013), ‘The poverty impacts of cash and in-kind transfers: experimental evidence from rural Mexico’, Journal of Development Effectiveness 5(4), pp. 401–429. Subramanian, Shankar and Deaton, Angus (1996), ‘The Demand for Food and Calories’, Journal of Political Economy 104(1), pp. 133–162. Webb, Patrick, Braun, Joachim Von, and Yohannes, Yisehac (1994), ‘Famine in Ethiopia: Policy Implications of Coping Failure at National and Household Levels’, Food and Nutrition Bulletin 15(1), pp. 1–2. Weismann, Doris, Bassett, Lucy, Benson, Todd, and Hoddinott, John (2009), Validation of the World Food Programme’s food consumption score and alternative indicators of household food security, IFPRI Working Paper, Washington, DC. World Bank (2022), G5 Sahel Region Country Climate and Development Report, Wash- ington, DC: World Bank. World Food Programme (2009), Comprehensive Food Security and Vulnerability Analysis Guidelines, Rome: World Food Programme. World Health Organization (2023), World malaria report 2023, World Health Organiza- tion. 30 Tables and Figures Figure 1: Study and Intervention Timeline Notes: The timeline depicts the timing of the drought, cash payments and data collection relative to the agricultural cycle. Following drought trigger activation in late 2021, the Government of Niger implemented three cash transfer modalities in 2022: four early transfers (March-June), four traditional lean season trans- fers (July-September), and twelve smaller monthly transfers (March 2022-January 2023). The government first collected proxy means test data to target vulnerable households in December 2021-January 2022, fol- lowed by our baseline survey (February 2022). We then collected high-frequency surveys during pre-lean (surveys 1-4), lean (5-7), and post-lean (8-11) periods, with households surveyed once per period. We conducted the endline survey in June 2023, just before the next lean season. 31 Table 1: Balance at baseline (1) (2) (3) (4) (5) Full Trad. response Early short Early long p-val Household size 7.50 7.58 7.47 7.44 0.73 Female Head of Household (HHH) 0.19 0.20 0.18 0.19 0.57 Polygamous HHH 0.18 0.18 0.18 0.16 0.21 HHH completed primary school 0.03 0.03 0.03 0.02 0.16 Monthly food consumption 45135.21 45619.31 46388.48 43270.29 0.19 Monthly non-food consumption 5069.88 4725.70 5458.24 5018.05 0.36 Total monthly consumption (food and non-food) 50190.95 50355.31 51853.09 48224.89 0.20 Total monthly per capita consumption (adult equiv) 18095.72 18225.26 18517.61 17503.42 0.41 Food Consumption Score 20.97 20.92 21.28 20.68 0.88 Food Consumption Score less than 21 0.51 0.51 0.50 0.52 0.84 Life Satisfaction 2.82 2.74 2.87 2.84 0.54 Mental health index 31.17 30.73 31.07 31.75 0.52 Household assets index -0.00 -0.06 0.04 0.02 0.46 Cultivated land in past 12 months 0.93 0.93 0.94 0.93 0.74 Area cultivated (in ha) 2.21 2.15 2.20 2.29 0.72 Agricultural inputs index 0.00 0.00 -0.01 0.01 0.94 Currently owns livestock 0.37 0.37 0.39 0.34 0.19 Number of livestock 2.23 2.16 2.49 2.03 0.27 Business in past 12 months 0.53 0.49 0.53 0.56 0.10 Number of businesses 1.50 1.50 1.49 1.51 0.96 Annual business revenues 125787.24 120314.21 133894.96 122586.73 0.75 Wage-employed in past 12 months 0.06 0.06 0.06 0.06 0.94 Has savings 0.05 0.05 0.06 0.05 0.63 Total savings 966.24 666.04 1083.70 1157.63 0.33 Took loan in past 12 months 0.64 0.65 0.63 0.63 0.85 Amount borrowed 31390.08 31944.59 32330.39 29799.17 0.69 Outstanding loans 16905.69 17329.56 17310.74 16025.04 0.73 Confident of raising 40000 FCFA 0.08 0.08 0.07 0.09 0.45 Drought in past 12 months 0.89 0.90 0.89 0.89 0.90 Notes: This table presents baseline characteristics across treatment arms. Columns 1-4 report means for the full sample and each treatment arm. Column 5 reports p-values from joint F-tests of equality of means across all treatment arms. Variables are measured at the household level. Monetary values are reported in local currency units. The household assets index, agricultural inputs index, and mental health index are standardized relative to the traditional response group. 32 Table 2: Economic welfare (food security and food consumption) (1) (2) (3) (4) (5) Pre-lean + Lean Pre-lean Lean Post-lean Endline =0 Food Security Early short 1.85*** -1.17** -0.41 0.22 0.41 (0.51) (0.52) (0.54) (0.60) [0.00] [0.05] [1.00] [1.00] Early long 0.12 -0.58 -0.04 0.70 0.57 (0.49) (0.56) (0.58) (0.59) [0.16] [0.17] [1.00] [1.00] Early short = Early long 0.00 0.27 0.53 0.41 Trad. response mean 22.83 25.67 29.23 21.50 Early short mean 24.66 24.47 28.81 21.65 Early long mean 22.93 25.07 29.21 22.19 Observations 3918 4071 4080 4131 Food Consumption Early short 8192.21*** -3397.97** -2192.22 202.18 0.06 (1538.85) (1418.99) (1549.16) (1365.20) [0.00] [0.05] [1.00] [1.00] Early long 2474.94* -4149.97*** 1319.88 806.56 0.49 (1422.94) (1358.67) (1941.43) (1346.66) [0.04] [0.02] [1.00] [1.00] Early short = Early long 0.00 0.61 0.06 0.67 Trad. response mean 46462.57 55976.21 56932.00 43440.65 Early short mean 54651.00 52452.16 54744.31 43466.13 Early long mean 48869.82 51678.00 58307.96 44201.31 Observations 3915 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for two primary outcome variables. Food security measured by the Food Consumption Score (FCS) captures dietary diversity and nutritional intake as a weighted sum of eight food groups consumed in past 7 days, with weights reflecting nutritional value (e.g., cereals=2, meat/fish=4, oils=0.5). Monthly food consumption captures the total value of household food consumption (in CFA francs) of 12 food items consumed through purchases, own production, and gifts. Quantities are valued based on purchase prices and adjusted using a seasonal price deflator for the top three cereals consumed (maize, millet, rice). All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). In the panel below the coefficients, we report p -values testing equality between Early Short and Early Long treatments. Column 5 reports p -values testing joint significance of treatment effects across pre-lean and lean seasons. Sharpened q -values adjusting for multiple inference across all three primary outcomes (food security, food consumption and life satisfaction) within each time period are shown in brackets, following the two-stage procedure developed by Benjamini et al. (2006) and implemented with code by Anderson (2008). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 33 Table 3: Psychological well-being (life satisfaction and mental health) (1) (2) (3) (4) (5) Pre-lean + Lean Pre-lean Lean Post-lean Endline =0 Life Satisfaction Early short 0.57*** -0.18* -0.03 -0.00 0.01 (0.10) (0.11) (0.11) (0.10) [0.00] [0.08] [1.00] [1.00] Early long 0.32*** -0.12 0.05 -0.00 0.12 (0.08) (0.10) (0.11) (0.08) [0.00] [0.17] [1.00] [1.00] Early short = Early long 0.01 0.54 0.49 0.98 Trad. response mean 3.22 3.85 3.49 3.42 Early short mean 3.79 3.67 3.47 3.42 Early long mean 3.54 3.73 3.55 3.41 Observations 3918 4071 4080 4131 Mental health index Early short 0.28*** -0.10 -0.09 -0.01 0.07 (0.06) (0.07) (0.08) (0.05) Early long 0.22*** -0.07 -0.02 0.03 0.14 (0.06) (0.07) (0.08) (0.05) Early short = Early long 0.27 0.59 0.42 0.37 Trad. response mean 0.18 0.55 0.54 0.28 Early short mean 0.47 0.45 0.45 0.26 Early long mean 0.41 0.48 0.52 0.31 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for two psychological outcome variables. As one of the three pre-specified primary outcome variables, life sat- isfaction is measured using Cantril’s ladder where respondents rate their current life from 1 (worst possible) to 10 (best possible). The mental health index sums the frequency (0-7 days) of experiencing 10 symptoms in the past week using the widely-used Center for Epidemiological Studies Depression Scale (CESD-R-10), with higher values indicating fewer depressive symptoms reported in the past week. The mental health index is standardized relative to the baseline mean and standard deviation. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). In the panel below the coefficients, we report p -values testing equality between Early Short and Early Long treatments. Column 5 reports p -values testing joint significance of treatment effects across pre-lean and lean seasons. Sharpened q -values adjusting for multiple infer- ence across all three primary outcomes (food security, food consumption and life satisfaction) within each time period are shown in brackets, following the two-stage procedure developed by Benjamini et al. (2006) and implemented with code by Anderson (2008). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 34 Figure 2: Livelihood outcomes (a) Pre-lean (b) Lean Business index Business index Wage index Wage index Livestock index Livestock index Early short Early short Ag inputs index Ag inputs index Early long Early long -.2 -.1 0 .1 .2 .3 -.2 -.1 0 .1 .2 .3 Treatment effect on standardised indices, relative to traditional response Treatment effect on standardised indices, relative to traditional response (c) Post-lean (d) Endline Business index Business index Wage index Wage index Livestock index Ag inputs index Livestock index Ag inputs index (endline) Early short Early short Ag inputs index Ag outputs index Early long Early long -.2 -.1 0 .1 .2 .3 -.2 -.1 0 .1 .2 .3 Treatment effect on standardised indices, relative to traditional response Treatment effect on standardised indices, relative to traditional response Notes: Each panel in these figures shows treatment effects of the early interventions relative to the traditional response for a distinct time period, comparing standardized livelihood indices with 95% confidence intervals. The business index includes asset value, monthly revenue, number of businesses (past month in high-frequency/midline, past year at endline), and days worked in business (past month). The wage employment index captures monthly agricultural and non-agricultural earnings and days worked. The livestock index combines Tropical Livestock Units (using standard conversion factors) and days worked tending livestock (past month), with endline adding total market value and monthly product revenue. The agricultural inputs index includes monthly input expenses and days worked in cultivation, with endline adding plot size, number of crops, and value of agricultural assets (past rainy season). Separate endline-only indices capture agricultural inputs in the past month and harvest value from the 2022 rainy season. All indices are standardized relative to the traditional response group mean and standard deviation. Estimates include commune, village size, and PMT strata fixed effects, with standard errors clustered at village level. 35 Table 4: Loans (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Borrowed (0, 1) Early short -0.16*** 0.13*** -0.02 0.05 (0.03) (0.02) (0.02) (0.03) Early long -0.10*** 0.11*** -0.05** 0.04 (0.02) (0.02) (0.03) (0.03) Early short = Early long 0.02 0.44 0.20 0.77 Trad. response mean 0.56 0.32 0.40 0.56 Early short mean 0.40 0.45 0.38 0.61 Early long mean 0.46 0.43 0.35 0.60 Observations 3918 4071 4080 4131 Amount borrowed Early short -3130.34*** 2035.31*** -222.05 47.86 (565.52) (457.18) (463.84) (115.57) Early long -1958.34*** 961.51** -824.57* -54.70 (585.74) (431.61) (482.36) (113.09) Early short = Early long 0.03 0.03 0.23 0.35 Trad. response mean 8328.76 4428.01 4550.62 1316.86 Early short mean 5173.45 6479.95 4343.36 1374.85 Early long mean 6323.67 5390.79 3737.90 1264.90 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for borrowing behavior. The borrowing indicator equals one if the household took a loan from any source in the past month during high-frequency and midline surveys or past year at endline. Monthly borrowing is denominated in CFA francs. While high-frequency and midline surveys captured borrowing activity during the preceding month, endline measurements covered a 12-month period and were subsequently converted to monthly averages. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). In the panel below the coefficients, we report p -values testing equality between Early Short and Early Long treatments. ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 36 Appendix Figure 3: Map of pilot activation Notes: The map shows the pilot communes where satellite-based triggers were activated for drought condi- tions in October 2021, out of a pre-selected list of 8 communes. Red-shaded areas indicate severe drought (defined as Water Requirement Satisfaction Index falling 25% below median at the third dekad of October), yellow indicates moderate drought (WRSI 10% below median), and green shows no activation. Based on these triggers, the Government of Niger through the the Safety Net Unit (Cellule Filets Sociaux or CFS) delivered unconditional cash transfers to approximately 15,400 households in four communes. The three study communes activated for severe drought near Niamey (western Niger) were selected for the study, as they were not previously covered by the national safety nets program. 37 Figure 4: Millet prices by month and treatment (average of retail and wholesale) Trad. response Early short 400 Early long Price per kg (CFA) 350 300 250 200 Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Jun 23 Notes: This figure shows monthly millet prices by treatment arms across the study period from February 2022 to January 2023 and then June 2023, representing the average of retail and wholesale prices per kilogram in CFA francs collected through monthly surveys with village leaders. Figure 5: Maize prices by month and treatment (average of retail and wholesale) 400 Trad. response Early short Early long 350 Price per kg (CFA) 300 250 200 Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Jun 23 Notes: This figure shows monthly maize prices by treatment arms across the study period from February 2022 to January 2023 and then June 2023, representing the average of retail and wholesale prices per kilogram in CFA francs collected through monthly surveys with village leaders. Figure 6: Rice prices by month and treatment (average of retail and wholesale) Trad. response Early short Early long 500 Price per kg (CFA) 450 400 Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Jun 23 Notes: This figure shows monthly rice prices by treatment arms across the study period from February 2022 to January 2023 and then June 2023, repre- senting the average of retail and wholesale prices per kilogram in CFA francs collected through monthly surveys with village leaders. 38 Table A1: Variable definition and construction: Primary outcomes Primary outcome variables Variable Definition Food security Food security is measured using the Food Consumption Score (FCS), a weighted sum of the frequency of household consumption across eight food groups over a 7-day period (World Food Programme, 2009). The weights re- flect the relative nutritional importance of each food group: FCS = 2*(cereals/grains/roots/tubers) + 3*(legumes/pulses/nuts) + 1*(veg- etables/leaves) + 1*(fruits) + 4*(meat/fish/eggs) + 4*(milk/dairy products) + 0.5*(oil/fat/butter) + 0.5*(sugar/sweet) Food consumption Food consumption is measured as the total value of household food consump- (monthly, CFA francs) tion in the past month. For each food item, households report total quantity consumed (including purchases, own production, and gifts) and the value and quantity of their most recent purchase. We standardize consumption and pur- chase units using household-provided conversion factors where available. For items consumed but not purchased, we impute prices using median purchase prices at the village level (requiring at least 5 observations) or, if insufficient observations exist, at the commune level. We winsorize unit quantities and unit prices at the 98th percentile for each food item before calculating consumption values as the product of quantity and price. We multiply weekly values by 4 to obtain monthly consumption. The final consumption measure sums across all food items and adjusts for seasonal price fluctuations using a monthly price deflator based on the three most consumed cereals (millet, maize, and rice), which comprise over 80% of food consumption. The price deflator uses the baseline survey as the reference period. Our baseline and endline surveys cover 36 food items. Due to survey time constraints, our high-frequency and midline surveys include only the 12 most commonly consumed items at baseline. Main tables report results for this reduced list, while appendix tables show results using the full list at endline. Life satisfaction Psychological well-being of the household head is measured using the 10-point Cantril’s ladder of life satisfaction (as in Bossuroy et al., 2022), where respon- dents rate their current life satisfaction from worst possible (bottom) to best possible life (top). 39 Table A2: Variable definition and construction: Secondary outcomes Variable Definition Psychological well-being (Household head) Mental health index Sum of 10 questions from CESD-R-10 measuring depressive symptoms (Radloff, 1977). For each item below, respondents indicate number of days (0-7) experiencing the symptom in past week: (1) felt bothered by things that usually don’t bother you, (2) had trouble keeping your mind on what you were doing, (3) felt that everything you did was an effort, (4) felt confident about the future (reversed), (5) felt sad, (6) felt nervous or tense, (7) had trouble sleeping, (8) felt happy (reversed), (9) felt alone, and (10) felt too tired to do anything. Index is standardized relative to traditional response group mean and standard deviation. Higher score indicates fewer depressive symptoms. Financial behaviors (Household level) Borrowed (0,1) Indicator for loan taken from any source in past month (HFS/midline) or past year (endline). Amount borrowed Total monthly amount borrowed in the past month (HFS/midline). Endline measurements covered a 12-month period and were subsequently converted to monthly averages. Saved (0,1) Indicator for any formal or informal savings in past month (HFS/midline) or past 3 months (endline). Amount saved Total amount saved across all formal and informal sources in past month (HFS/midline) or past 3 months (endline). Received remittances Indicator for whether household received any remittances from non-household (0,1) members in past month (HFS/midline) and past six months (endline). Remittance amount Total value of remittances received in past month (HFS/midline) and past six months (endline). Social outcomes (Household head) Perceptions of violence Standardized index averaging two components: (1) respondent estimate to index “Out of 10 women, how many experience tensions in their household?” and (2) response to “In this village, is it common for husbands to beat their wives if they burn food?”. Index reversed so higher values indicate less perceived violence. Index is standardized relative to traditional response group mean and standard deviation. Social cohesion (0,1) Indicator for absence of tension between household and community members when asked “In the past 30 days, have you or your household experienced tensions with other members of your community?” Notes: Monetary values are measured in CFA francs and winsorized at 98th percentile. 40 Table A3: Variable definition and construction: Sub-components of livelihood indices Variable Definition Business index Business asset value Total value of assets used in non-farm business activities. Business revenue Total revenue from all business activities in past month. Number of businesses Count of distinct non-farm businesses operated by household members in past month (HFS/midline) and past year (endline). Days worked in business Total number of days in which household members worked at least one hour in a non-farm business in past month. Wage employment index Agricultural wage earnings Total earnings from agricultural wage labor in past month. Days in agricultural work Total number of days in which household members worked at least one hour in agricultural wage labor in past month. Non-agricultural earnings Total earnings from non-agricultural wage labor in past month. Days in non-agricultural work Total number of days in which household members worked at least one hour in non-agricultural wage labor in past month. Livestock index Tropical Livestock Units Weighted sum of livestock, owned or rented, using conversion factors: (TLU) camels (1.1), cows/calves (0.7), horses/mares/donkeys (0.8), bulls (0.5), pigs (0.2), sheep/goats/mutton (0.1), guinea fowl (0.03), chicken (0.01). Days worked in livestock Total number of days in which household members worked at least one hour tending livestock in past month. Value of livestock (endline) Total market value of all livestock owned. Revenue (endline) Revenue from livestock products (milk, eggs, etc.) in past month. Agricultural inputs index Input expenses Total expenditure on seeds and fertilizer in past month (HFS/midline) or 2022 rainy season (endline). Days worked in cultivation Total number of days in which household members worked at least one hour in crop cultivation in past month. Plot size (midline & endline) Total area of all plots cultivated in square meters. Number of crops (endline) Count of distinct crops planted across all plots. Value of agricultural assets Total value of agricultural tools and equipment. (endline) Agricultural inputs index (endline only) Input expenses Total expenditure on seeds and fertilizer in past month. Number of plots Count of plots cultivated in past month. Total plot size Total area of plots cultivated in past month in square meters. Agricultural outputs index (endline only) Value of harvest Total value of all crops harvested during 2022 rainy season. Notes: All indices are constructed by averaging the standarized sub-components aggregated to the household level. Sub-components are standardized relative to the traditional response group mean and standard deviation. Monetary values are measured in CFA francs and winsorized at 98th percentile. We aggregate days worked across household members, so the total may exceed 30 days. 41 Table A4: Smoothing test, equality of means between periods Pre-lean = Lean Food Security Trad. response 0.000 Early short 0.454 Early long 0.000 Food Consumption Trad. response 0.000 Early short 0.057 Early long 0.005 Life Satisfaction Trad. response 0.000 Early short 0.162 Early long 0.021 Mental health index Trad. response 0.000 Early short 0.576 Early long 0.062 Table A5: Proportion of households above “poor” food security (FCS>21) (1) (2) (3) (4) (5) Pre-lean + Lean Pre-lean Lean Post-lean Endline =0 Food Security Early short 0.11*** -0.06* 0.02 -0.01 0.28 (0.03) (0.03) (0.02) (0.03) Early long 0.03 -0.03 0.02 -0.01 0.87 (0.03) (0.03) (0.03) (0.03) Early short = Early long 0.00 0.38 0.94 0.94 Trad. response mean 0.50 0.67 0.71 0.49 Early short mean 0.61 0.61 0.73 0.48 Early long mean 0.52 0.63 0.73 0.48 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for the proportion of households above the “poor” food security threshold, defined as having a Food Consumption Score above 21. All specifications include village size, PMT and cohort strata fixed effects, with standard errors clustered at the village level shown in parentheses. Column 5 reports p-values testing joint significance of treatment effects across the pre-lean and lean seasons. ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 42 Table A6: Food security, number of days HH ate each food group Cereals Legumes, nuts Vegetables Fruits Meat, fish, eggs Milk, dairy Oil, fats, butter Sugar, sweets Early short 0.13* 0.06 0.11 0.01 0.01 0.16* 0.77*** 0.43*** (0.07) (0.04) (0.16) (0.02) (0.02) (0.09) (0.18) (0.15) Early long -0.03 -0.03 -0.13 0.03 0.00 0.04 0.31* 0.13 (0.10) (0.03) (0.17) (0.03) (0.02) (0.08) (0.17) (0.15) Trad. response mean 6.63 0.18 5.02 0.08 0.06 0.42 2.63 1.40 Observations 3918 3918 3918 3918 3918 3918 3918 3918 Cereals Legumes, nuts Vegetables Fruits Meat, fish, eggs Milk, dairy Oil, fats, butter Sugar, sweets Early short -0.02 -0.03 -0.23 0.02 -0.00 -0.16 -0.34*** -0.12 (0.05) (0.07) (0.17) (0.02) (0.02) (0.10) (0.11) (0.12) Early long 0.01 -0.07 -0.01 -0.00 0.00 -0.07 -0.21* -0.08 (0.04) (0.07) (0.17) (0.01) (0.02) (0.11) (0.13) (0.12) Trad. response mean 6.76 0.37 5.38 0.03 0.05 0.71 4.20 0.98 Observations 4071 4071 4071 4071 4071 4071 4071 4071 Cereals Legumes, nuts Vegetables Fruits Meat, fish, eggs Milk, dairy Oil, fats, butter Sugar, sweets Early short -0.06 0.03 -0.05 -0.05** -0.02 -0.04 -0.02 -0.11 43 (0.05) (0.15) (0.14) (0.02) (0.01) (0.12) (0.17) (0.09) Early long -0.02 0.04 0.10 -0.05*** -0.00 -0.04 0.05 -0.01 (0.06) (0.15) (0.16) (0.02) (0.01) (0.13) (0.18) (0.09) Trad. response mean 6.81 1.69 4.80 0.10 0.04 0.89 3.07 0.70 Observations 4080 4080 4080 4080 4080 4080 4080 4080 Cereals Legumes, nuts Vegetables Fruits Meat, fish, eggs Milk, dairy Oil, fats, butter Sugar, sweets Early short -0.02 0.08 0.11 -0.01 -0.02 -0.01 -0.02 -0.09 (0.10) (0.05) (0.17) (0.02) (0.03) (0.06) (0.16) (0.11) Early long 0.06 0.04 0.36** -0.00 0.05 -0.02 -0.03 0.02 (0.10) (0.05) (0.17) (0.03) (0.03) (0.06) (0.16) (0.12) Trad. response mean 6.29 0.40 4.54 0.06 0.12 0.27 2.25 0.87 Observations 4131 4131 4131 4131 4131 4131 4131 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for the number of days households consumed different food groups in the week prior to each survey round. These food groups are the components used to calculate the Food Consumption Score (FCS). Each panel represents a survey round, with the first panel showing pre-lean season results, followed by lean season, post-lean season, and endline. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A7: Monthly food consumption by food group Maize Millet Rice Sorghum Pasta Mutton Salt Sugar Palm oil Peanut oil Kola nuts Dry beans Early short 1174.10 3611.84*** 2227.79*** 19.05 40.14 182.72* 110.71*** 155.28*** 394.57*** 243.21*** -7.17 108.72 (779.01) (1102.29) (448.01) (149.49) (33.15) (94.29) (33.98) (57.53) (113.75) (85.56) (12.37) (78.83) Early long -585.93 1092.30 1512.04*** 96.55 63.40 -14.63 33.61 33.96 265.17*** 26.83 8.48 -28.50 (750.57) (1135.92) (459.98) (157.30) (39.41) (59.77) (31.02) (57.14) (94.19) (74.46) (14.38) (53.69) Trad. response mean 19592.68 19195.03 3949.12 493.59 68.87 181.83 742.72 439.28 780.81 745.72 64.02 138.35 Observations 3918 3918 3918 3918 3918 3918 3918 3918 3918 3918 3918 3918 Maize Millet Rice Sorghum Pasta Mutton Salt Sugar Palm oil Peanut oil Kola nuts Dry beans Early short -630.78 -99.44 -2229.79*** 249.62 -124.56** -55.03 -61.88** -55.25 -14.13 -325.41*** -9.29 -74.64 (673.88) (1223.66) (535.04) (314.47) (56.02) (69.99) (30.82) (36.30) (37.10) (83.23) (20.58) (102.30) Early long -1814.12** -1239.85 -744.63 128.50 -120.67** -41.39 -72.78** -41.72 -17.63 -189.96** -18.68 -30.85 (710.21) (1207.28) (570.61) (308.70) (56.40) (80.42) (30.08) (39.45) (38.94) (88.40) (16.43) (98.87) Trad. response mean 22777.50 16869.99 10085.00 1335.94 286.90 227.37 901.25 267.02 157.14 2336.85 100.83 630.43 Observations 4071 4071 4071 4071 4071 4071 4071 4071 4071 4071 4071 4071 Maize Millet Rice Sorghum Pasta Mutton Salt Sugar Palm oil Peanut oil Kola nuts Dry beans Early short -638.09 -599.40 -756.89* -22.86 16.39 -90.86 44.01 -56.91** -24.77 -48.79 -11.73 -2.33 44 (580.66) (1090.62) (431.93) (107.72) (22.16) (66.55) (27.29) (24.92) (20.30) (94.02) (16.44) (384.27) Early long 507.33 120.00 618.84 -62.71 3.36 -2.49 -0.02 2.74 21.49 62.67 -4.40 48.60 (643.08) (1230.39) (473.64) (96.39) (22.31) (82.44) (28.00) (29.36) (23.32) (94.42) (16.18) (437.16) Trad. response mean 14770.08 29352.30 4700.42 387.30 43.77 183.36 756.21 180.50 82.45 1575.31 66.84 4833.45 Observations 4080 4080 4080 4080 4080 4080 4080 4080 4080 4080 4080 4080 Maize Millet Rice Sorghum Pasta Mutton Salt Sugar Palm oil Peanut oil Kola nuts Dry beans Early short -224.99 395.61 -39.61 -73.28 -37.12 45.27 6.16 -27.46 -0.38 -26.07 -0.53 36.54 (734.93) (855.65) (349.27) (205.28) (23.29) (141.92) (55.60) (33.33) (43.02) (81.35) (13.99) (96.00) Early long -310.72 1325.78 -288.13 -49.78 20.76 71.93 -41.82 19.12 28.09 -44.67 -3.08 71.60 (735.83) (824.77) (367.92) (208.87) (32.77) (133.76) (48.12) (36.88) (44.93) (80.55) (13.14) (90.95) Trad. response mean 19039.34 16672.55 3638.80 793.95 80.93 288.97 742.61 217.38 268.44 1056.18 58.73 582.78 Observations 4131 4131 4131 4131 4131 4131 4131 4131 4131 4131 4131 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for monthly food consumption (in CFA francs), by food group. Each panel represents a survey round, with the first panel showing pre-lean season results, followed by lean season, post-lean season, and endline panels. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A8: Primary outcomes, with baseline controls (1) (2) (3) (4) (5) Pre-lean + Lean Pre-lean Lean Post-lean Endline =0 Food Security Early short 1.78*** -1.27** -0.47 0.10 0.41 (0.50) (0.51) (0.50) (0.56) [0.00] [0.03] [1.00] [1.00] Early long 0.12 -0.64 -0.21 0.83 0.57 (0.47) (0.53) (0.54) (0.57) [0.16] [0.16] [1.00] [1.00] Early short = Early long 0.00 0.24 0.64 0.20 Trad. response mean 22.83 25.69 29.29 21.46 Early short mean 24.66 24.50 28.83 21.66 Early long mean 22.94 25.09 29.15 22.25 Observations 3814 3935 3939 3946 Food Consumption Early short 7649.59*** -3686.30*** -2499.68 -218.17 0.06 (1400.69) (1347.58) (1522.95) (1306.56) [0.00] [0.03] [1.00] [1.00] Early long 2996.07** -3537.20*** 1838.32 1179.89 0.49 (1381.16) (1337.43) (1884.10) (1311.40) [0.01] [0.03] [1.00] [1.00] Early short = Early long 0.00 0.91 0.02 0.28 Trad. response mean 46593.29 55965.61 57249.23 43462.53 Early short mean 54882.64 52705.98 54969.18 43542.26 Early long mean 49098.77 52002.98 58875.86 44209.16 Observations 3812 3935 3939 3946 Life Satisfaction Early short 0.59*** -0.17 -0.00 0.01 0.01 (0.10) (0.11) (0.10) (0.09) [0.00] [0.09] [1.00] [1.00] Early long 0.31*** -0.11 0.08 0.01 0.12 (0.08) (0.10) (0.11) (0.07) [0.00] [0.17] [1.00] [1.00] Early short = Early long 0.00 0.53 0.44 0.99 Trad. response mean 3.21 3.86 3.50 3.41 Early short mean 3.80 3.68 3.48 3.43 Early long mean 3.53 3.73 3.55 3.41 Observations 3814 3935 3939 3946 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for our three primary outcome variables, with baseline controls selected using the double-selection lasso method proposed by Belloni et al. (2014). Column 5 reports p -values testing joint significance of treatment effects across pre-lean and lean seasons. Sharpened q -values adjusting for multiple inference across all three primary outcomes (food security, food consumption and life satisfaction) within each time period are shown in brackets, following the two-stage procedure developed by Benjamini et al. (2006) and implemented with code by Anderson (2008). ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 45 Table A9: Monthly food consumption using extended measure at midline and endline (1) (2) Lean Endline Food consumption Early short -3747.96** 554.59 (1486.52) (1465.92) Early long -4434.18*** 889.64 (1414.43) (1430.57) Early short = Early long 0.66 0.82 Trad. response mean 60585.98 46857.79 Observations 4071 4131 Notes: This table reports intent-to-treat estimates comparing the ef- fects of the early interventions to the traditional lean season response on monthly food consumption using more comprehensive measures at midline and endline. During these survey rounds, consumption was measured using 36 food items (compared to 12 items in high- frequency surveys), capturing a broader range of household food con- sumption patterns. Food consumption is measured as the total value in CFA francs of household food consumption in the past month, including purchases, own production, and gifts. Values are price- adjusted using a monthly deflator based on the three main cereals (millet, maize, rice), which comprise over 80% of food consumption, with baseline as the reference period. All specifications include vil- lage size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 46 Table A10: Food security by survey round Pre-lean Lean Post-lean Endline (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Early short 2.18** 1.34 1.89* 1.99*** 0.55 -1.23* -2.48*** -0.18 -1.45 0.14 -0.13 0.22 (0.96) (0.84) (0.98) (0.69) (0.78) (0.64) (0.91) (1.01) (1.04) (0.84) (0.72) (0.60) Early long 0.40 -0.03 -0.37 0.54 0.26 -0.35 -1.40* -0.32 -0.59 0.92 -0.14 0.70 (1.09) (0.82) (0.72) (0.70) (0.77) (0.80) (0.83) (0.87) (1.07) (0.84) (0.80) (0.59) Trad. response mean 21.60 22.45 24.50 22.72 23.73 24.95 27.76 31.02 33.80 26.87 25.27 21.50 Observations 981 1000 973 964 1220 1273 1578 1018 1023 1009 1030 4131 Notes: This table reports intent-to-treat estimates of food security as measured by the Food Consumption Score, disaggregated by individual survey rounds within each pre-lean, lean, and post-lean season study period, plus the endline survey. All specifications include village size, PMT and cohort strata fixed effects, with standard errors clustered at the village level shown in parentheses. ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A11: Monthly food consumption by survey round 47 Pre-lean Lean Post-lean Endline (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Early short 3849.52 12497.61*** 7857.07*** 8448.35*** 1916.13 -6590.87*** -5117.98** -4926.96** -6091.98** -2526.24 4683.53* 202.18 (2501.74) (3217.11) (2782.88) (2794.17) (2135.27) (2414.75) (2389.59) (2197.18) (2588.95) (2714.53) (2441.12) (1365.20) Early long 1238.67 5252.75* 2321.43 1107.42 -604.49 -6556.20*** -5134.18** -2494.36 38.64 3035.78 4675.13* 806.56 (2742.41) (2767.07) (2606.38) (2709.76) (1954.62) (2320.20) (2087.59) (2527.63) (2708.48) (3127.27) (2623.36) (1346.66) Trad. response mean 45816.43 43385.85 47246.82 49490.29 50906.49 57876.43 58315.28 60818.88 62674.37 57055.47 47374.84 43440.65 Observations 980 1000 973 962 1220 1273 1578 1018 1023 1009 1030 4131 Notes: This table reports intent-to-treat estimates of monthly food consumption disaggregated by individual survey rounds within each pre-lean, lean, and post-lean season study period, plus the endline survey. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A12: Life satisfaction by survey round Pre-lean Lean Post-lean Endline (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Early short 0.70*** 0.79*** 0.26 0.55*** -0.16 -0.14 -0.21 0.07 -0.12 -0.09 0.04 -0.00 (0.17) (0.20) (0.17) (0.12) (0.13) (0.15) (0.13) (0.16) (0.16) (0.13) (0.10) (0.10) Early long 0.35** 0.66*** 0.12 0.13 -0.15 -0.06 -0.13 0.16 -0.14 0.12 0.07 -0.00 (0.18) (0.19) (0.15) (0.13) (0.13) (0.15) (0.12) (0.15) (0.16) (0.14) (0.11) (0.08) Trad. response mean 3.08 3.33 3.15 3.30 3.79 3.83 3.92 3.51 3.76 3.39 3.31 3.42 Observations 981 1000 973 964 1220 1273 1578 1018 1023 1009 1030 4131 Notes: This table reports intent-to-treat estimates of life satisfaction, disaggregated by individual survey rounds within each pre-lean, lean, and post-lean season study period, plus the endline survey. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 48 Table A13: Social outcomes (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Perception of violence index Early short 0.04 -0.00 -0.05 0.07 (0.04) (0.04) (0.03) (0.04) Early long -0.02 -0.00 -0.01 0.08** (0.04) (0.04) (0.03) (0.04) Early short = Early long 0.18 0.99 0.31 0.69 Trad. response mean -0.00 0.00 0.00 -0.00 Observations 3918 4071 4080 4131 Social cohesion Early short 0.01 0.00 0.00 -0.01 (0.01) (0.00) (0.00) (0.01) Early long 0.01 0.00 0.01 0.00 (0.01) (0.00) (0.00) (0.01) Early short = Early long 0.49 0.98 0.21 0.33 Trad. response mean 0.96 0.99 0.99 0.94 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for social outcomes. The perceptions of violence index is a standardized index averaging two components: (1) respondent estimate to “Out of 10 women, how many experience tensions in their household?” and (2) response to “In this village, is it common for husbands to beat their wives if they burn food?”. The index is reversed so higher values indicate less perceived violence and standardized relative to the traditional response group mean and standard deviation. Social cohesion is measured as an indicator for absence of tension between household and community members when asked “In the past 30 days, have you or your household experienced tensions with other members of your community?”. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 49 Table A14: Savings (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Saved (0, 1) Early short 0.03*** -0.00 -0.01 -0.00 (0.01) (0.01) (0.01) (0.01) Early long 0.01 0.00 -0.00 0.01 (0.01) (0.01) (0.01) (0.01) Early short = Early long 0.01 0.49 0.80 0.30 Trad. response mean 0.03 0.04 0.04 0.07 Early short mean 0.06 0.04 0.03 0.07 Early long mean 0.03 0.04 0.04 0.09 Observations 3918 4071 4080 4131 Amount saved Early short 259.84 -78.10 -123.45* -148.41 (166.01) (279.25) (65.56) (132.90) Early long -2.03 -167.72 -36.65 -95.72 (108.56) (281.55) (82.93) (131.03) Early short = Early long 0.10 0.75 0.17 0.62 Trad. response mean 240.47 601.34 219.28 476.98 Early short mean 499.63 525.02 98.85 325.34 Early long mean 237.26 442.45 186.76 385.60 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for savings behavior. The savings indicator equals one if the household saved any amount formally or informally in the past month during high-frequency and midline surveys or past three months at endline. Monthly savings is denominated in CFA francs. While high-frequency and midline surveys captured saving amounts during the preceding month, endline measurements covered a 3-month period and were subsequently converted to monthly averages. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). In the panel below the coefficients, we report p -values testing equality between Early Short and Early Long treatments. ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 50 Table A15: Remittances (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Received remittance? Early short -0.03 0.01 -0.03 -0.01 (0.02) (0.02) (0.02) (0.03) Early long 0.02 0.03* -0.03 -0.04 (0.02) (0.02) (0.02) (0.03) Early short = Early long 0.03 0.27 0.82 0.38 Trad. response mean 0.22 0.18 0.17 0.32 Observations 3918 4071 4080 4131 Remittance amount Early short -686.21 653.49 -600.42 -193.15 (596.30) (674.31) (532.54) (239.32) Early long -25.88 731.54 -206.31 -560.82*** (666.63) (716.00) (543.74) (213.38) Early short = Early long 0.28 0.91 0.44 0.07 Trad. response mean 4936.83 4135.05 3077.14 2202.07 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for remittance outcomes. The remittance indicator equals one if the household received any remittances from non-household members in the past month (high-frequency/midline surveys) and past six months (endline). Monthly remittance amounts is denominated in CFA francs. While high-frequency and midline surveys captured remittance activity during the preceding month, endline measurements covered a 6-month period and were subsequently converted to monthly averages. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 51 Table A16: Business index (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Early short 0.00 -0.02 -0.02 0.09* (0.05) (0.04) (0.05) (0.05) Early long 0.02 -0.02 -0.02 0.02 (0.05) (0.04) (0.05) (0.05) Early short = Early long 0.78 0.93 1.00 0.19 Trad. response mean -0.00 -0.00 -0.00 0.00 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early inter- ventions to the traditional lean season response for the business index. The business index includes asset value, monthly revenue, number of businesses (past month in high-frequency/midline, past year at endline), and days worked in business (past month). All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A17: Wage index (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Early short -0.02 0.03 -0.01 0.02 (0.03) (0.04) (0.04) (0.04) Early long -0.02 0.03 -0.01 0.02 (0.03) (0.05) (0.03) (0.03) Early short = Early long 0.83 0.99 0.93 0.97 Trad. response mean 0.00 0.00 -0.00 -0.00 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early inter- ventions to the traditional lean season response for the wage employment index. The wage employment index captures monthly agricultural and non-agricultural earnings and days worked (past month). All specifications include village size, PMT and co- hort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 52 Table A18: Livestock index (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Early short 0.05 0.03 -0.03 0.03 (0.05) (0.05) (0.04) (0.04) Early long 0.01 0.04 -0.01 0.02 (0.04) (0.04) (0.04) (0.05) Early short = Early long 0.44 0.78 0.74 0.85 Trad. response mean -0.00 0.00 -0.00 -0.00 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early inter- ventions to the traditional lean season response for the livestock index. The livestock index combines Tropical Livestock Units (using standard conversion factors) and days worked tending livestock (past month), with endline adding total market value and monthly product revenue. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A19: Agriculture inputs index (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Early short 0.05 0.03 0.05 0.05 (0.04) (0.05) (0.05) (0.04) Early long -0.03 -0.01 0.05 0.04 (0.04) (0.05) (0.05) (0.05) Early short = Early long 0.10 0.35 0.94 0.89 Trad. response mean 0.00 0.00 -0.00 0.00 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early inter- ventions to the traditional lean season response for the agricultural inputs index. The agricultural inputs index includes monthly input expenses and days worked in cultivation, with endline adding plot size, number of crops, and value of agricultural assets (past rainy season). All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 53 Table A20: Agriculture inputs in last 30 days (endline) index (1) Pre-lean Early short 0.06 (0.06) Early long 0.02 (0.06) Early short = Early long 0.48 Trad. response mean -0.00 Observations 4131 Notes: This table presents intent-to-treat estimates comparing the two early inter- ventions to the traditional lean season response for the agricultural inputs index in the past 30 days at endline, i.e. before the next lean season. All specifications include village size, PMT and cohort strata fixed effects, control for households sur- veyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 Table A21: Agriculture outputs (endline) index (1) Pre-lean Early short -0.05 (0.07) Early long 0.07 (0.09) Early short = Early long 0.19 Trad. response mean -0.00 Observations 3948 Notes: This table presents intent-to-treat estimates comparing the two early inter- ventions to the traditional lean season response for the agricultural outputs index at endline, which captures the total value of crops harvested during the 2022 rainy season. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 54 Table A22: Business index components (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Business assets Early short 566.92 -277.73 127.80 1591.69* (593.58) (489.01) (532.32) (955.78) Early long 675.25 -707.02 -2.79 1038.28 (589.76) (528.94) (513.61) (1075.65) Early short = Early long 0.87 0.40 0.81 0.65 Trad. response mean 3125.64 2789.75 2252.75 6427.96 Observations 3918 4071 4080 4131 Business revenue Early short -632.71 -698.72 -570.54 205.76 (828.43) (632.86) (606.76) (1623.76) Early long -624.11 -597.36 -74.14 -557.62 (918.58) (751.72) (639.58) (1765.04) Early short = Early long 0.99 0.87 0.38 0.66 Trad. response mean 6780.54 4845.46 4014.66 13783.09 Observations 3918 4071 4080 4131 Number of businesses Early short 0.00 -0.01 -0.01 0.11** (0.04) (0.03) (0.03) (0.05) Early long 0.02 0.00 -0.01 0.03 (0.04) (0.04) (0.03) (0.05) Early short = Early long 0.68 0.70 0.93 0.09 Trad. response mean 0.45 0.35 0.35 0.56 Observations 3918 4071 4080 4131 Days worked Early short -0.09 0.12 -0.11 2.30** (0.88) (0.82) (0.75) (1.16) Early long 0.15 0.01 -0.38 0.38 (0.95) (0.78) (0.67) (1.11) Early short = Early long 0.80 0.89 0.68 0.12 Trad. response mean 8.72 6.20 6.23 9.20 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for the components of the business index. The components include: i) total value of assets used in non-farm business activities; ii) total business revenue from all business activities in past month; iii) number of non-farm businesses operated by household members in past month (high- frequency/midline surveys) and past year (endline); and iv) number of days in which household members worked at least one hour in a non-farm business in past month. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 55 Table A23: Wage index components (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Ag wage Early short -238.76 422.87 -80.65 -11.62 (179.37) (381.53) (134.69) (125.72) Early long -138.40 449.89 -122.76 42.57 (204.93) (429.75) (133.57) (117.15) Early short = Early long 0.56 0.95 0.68 0.68 Trad. response mean 803.73 2360.22 418.60 315.66 Observations 3762 3949 3925 4131 Days in ag employment Early short -0.08 0.26 0.10 -0.02 (0.17) (0.28) (0.12) (0.13) Early long -0.13 0.20 0.10 0.01 (0.18) (0.29) (0.14) (0.10) Early short = Early long 0.73 0.86 0.99 0.86 Trad. response mean 0.76 1.71 0.37 0.29 Observations 3918 4071 4080 4131 Non ag wage Early short -482.90 199.64 -218.34 28.04 (498.37) (375.43) (276.75) (144.91) Early long -227.95 231.77 -156.46 121.57 (514.11) (443.80) (253.79) (147.50) Early short = Early long 0.44 0.94 0.79 0.56 Trad. response mean 1568.10 1190.12 1002.35 378.68 Observations 3847 3475 4024 4131 Days in non ag employment Early short 0.01 -0.06 -0.05 0.23 (0.29) (0.29) (0.32) (0.16) Early long 0.03 -0.05 -0.12 0.05 (0.29) (0.31) (0.25) (0.11) Early short = Early long 0.95 0.98 0.79 0.30 Trad. response mean 1.24 1.02 1.04 0.32 Observations 3918 4071 4080 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for the components of the wage employment index. The components include: i) total earnings from agricultural wage labor in past month; ii) number of days in which household members worked at least one hour in agricultural wage labor in past month; iii) total earnings from non-agricultural wage labor in past month; and iv) number of days in which household members worked at least one hour in non-agricultural wage labor in past month. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 56 Table A24: Livestock index components (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Livestock TLU Early short 0.15 0.04 -0.02 0.04 (0.12) (0.16) (0.18) (0.07) Early long 0.09 0.07 -0.05 0.01 (0.11) (0.15) (0.17) (0.06) Early short = Early long 0.67 0.86 0.88 0.59 Trad. response mean 0.76 1.01 1.34 0.40 Observations 3918 4071 4080 4131 Days worked Early short 0.43 0.66 -0.89 0.62 (1.36) (0.83) (0.85) (1.40) Early long -0.87 0.91 -0.23 -0.85 (1.26) (0.88) (0.88) (1.40) Early short = Early long 0.30 0.78 0.45 0.32 Trad. response mean 29.88 31.95 35.87 21.88 Observations 2863 3182 3320 4131 Value of livestock Early short 11224.68 (20149.41) Early long 389.44 (18078.62) Early short = Early long 0.57 Trad. response mean 80643.74 Observations 4131 Revenue Early short 9.81 (11.71) Early long 31.17 (39.08) Early short = Early long 0.58 Trad. response mean 12.06 Observations 4131 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional response for the components of the livestock index. The components include: i) Tropical Livestock Units (weighted sum of livestock owned/rented using conversion factors); ii) number of days in which household members worked at least one hour tending livestock in past month; iii) total market value of all livestock owned (endline only); and iv) revenue from livestock products in past month (endline only). All specifica- tions include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 57 Table A25: Agricultural inputs index components (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Input expenses Early short 590.46 587.28 360.53 99.83 (543.42) (446.69) (611.40) (394.69) Early long 99.38 -413.78 346.10 249.91 (526.59) (390.61) (590.84) (366.36) Early short = Early long 0.34 0.03 0.98 0.73 Trad. response mean 1585.80 2989.14 2365.18 1393.10 Observations 3918 4071 4080 4131 Days worked Early short 0.77 -0.43 1.53 0.71 (1.12) (1.53) (1.57) (0.74) Early long -1.08 0.29 1.37 0.16 (1.00) (1.54) (1.43) (0.93) Early short = Early long 0.11 0.65 0.92 0.55 Trad. response mean 12.54 37.64 21.02 8.38 Observations 3918 4071 4080 4131 Total plot size Early short 961.67 1501.65 (2184.07) (1403.06) Early long 295.26 1973.04 (2290.75) (1524.61) Early short = Early long . 0.74 . 0.77 Trad. response mean 25520.79 23258.85 Observations 3959 3990 # of crops Early short 0.03 (0.05) Early long 0.03 (0.05) Early short = Early long . . . 0.88 Trad. response mean 2.22 Observations 3948 Value of assets Early short 28357.09 (32203.02) Early long 6747.84 (32661.20) Early short = Early long . . . 0.50 Trad. response mean 390251.49 Observations 4131 Notes: This table presents intent-to-treat estimates for the components of the agricultural inputs index. The components include: i) total expenditure on seeds and fertilizer in past month (high-frequency/midline) or 2022 rainy season (endline); ii) number of days in which household members worked at least one hour in crop cultivation in past month; iii) total area of all plots cultivated (midline and endline); iv) count of distinct crops planted across all plots (endline only); and v) total value of agricultural tools and equipment (endline only). All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 58 Table A26: Agricultural inputs index in last 30 days (endline) components (1) Endline Input expenses Early short 330.29* (177.00) Early long 182.77 (191.88) Early short = Early long 0.51 Trad. response mean 724.70 Observations 4131 # of plots Early short 0.05 (0.08) Early long -0.00 (0.08) Early short = Early long 0.51 Trad. response mean 1.03 Observations 4131 Total plot size Early short 407.04 (1265.54) Early long -53.59 (1360.89) Early short = Early long 0.74 Trad. response mean 14224.77 Observations 4131 Notes: This table presents intent-to-treat estimates compar- ing the two early interventions to the traditional lean sea- son response for the components of the agricultural inputs index in the past 30 days at endline. The components in- clude: i) total expenditure on seeds and fertilizer in past month; ii) number of plots cultivated in past month; and iii) total area of plots cultivated in past month. All spec- ifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 59 Table A27: Agricultural outputs index components (1) Endline Harvest value Early short -6319.66 (8707.31) Early long 9044.34 (11473.77) Early short = Early long 0.19 Trad. response mean 108156.76 Observations 3948 Notes: This table presents intent-to-treat estimates compar- ing the two early interventions to the traditional lean season response for the agricultural outputs index at endline, which captures the total value of crops harvested during the 2022 rainy season. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 60 Table A28: Heterogeneity by baseline liquid asset holdings (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Food Consumption Score Early short (β1 ) 1.36* -1.34* -1.17 0.24 (0.77) (0.72) (0.79) (0.78) Early long( β2 ) -0.02 -0.02 0.41 0.36 (0.76) (0.86) (0.79) (0.77) Early short × Liquid (β3 ) 0.76 0.27 1.17 0.00 (0.84) (0.77) (0.88) (0.84) Early long × Liquid (β4 ) 0.32 -0.86 -0.90 0.68 (0.80) (0.90) (0.89) (0.81) β1 + β 3 = 0 0.00 0.07 1.00 0.73 β2 + β4 = 0 0.58 0.14 0.47 0.14 Observations 3833 3954 3958 3964 Food Consumption Early short (β1 ) 9045.78*** -2668.52 -4446.54** 793.73 (2423.91) (2137.19) (1848.26) (1947.51) Early long( β2 ) -545.93 -4219.36* -93.36 -661.33 (2192.05) (2184.44) (2295.31) (1819.66) Early short × Liquid (β3 ) -1654.16 -990.10 3430.46 -971.46 (2677.09) (2445.46) (2118.48) (1985.47) Early long × Liquid (β4 ) 4853.35* 685.10 2401.10 2823.76 (2833.56) (2399.98) (2500.67) (1969.65) β1 + β 3 = 0 0.00 0.04 0.59 0.91 β2 + β4 = 0 0.02 0.02 0.31 0.18 Observations 3831 3954 3958 3964 Life Satisfaction Early short (β1 ) 0.55*** -0.25* -0.07 0.06 (0.14) (0.14) (0.14) (0.12) Early long( β2 ) 0.35*** -0.19 0.14 -0.04 (0.13) (0.12) (0.14) (0.11) Early short × Liquid (β3 ) 0.06 0.11 0.09 -0.08 (0.15) (0.14) (0.12) (0.13) Early long × Liquid (β4 ) -0.06 0.11 -0.13 0.08 (0.15) (0.14) (0.12) (0.12) β1 + β 3 = 0 0.00 0.24 0.88 0.87 β2 + β4 = 0 0.00 0.50 0.93 0.66 Observations 3833 3954 3958 3964 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for our three primary outcome variables. The interaction terms “Early short x Liquid” and “Early long x Liquid” capture the differential effects of the treatments based on the household’s baseline liquid asset holdings. Baseline liquid asset holdings is defined as an index constructed using the household’s total savings, livestock holdings measured in Tropical Livestock Units, and the total value of agricultural stocks; the index is then split at the median to identify those with high versus low liquid asset holdings. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). In the panel below the coefficients, we report p -values from tests of whether the sum of the main treatment effect and the interaction effect is equal to zero for each early intervention. ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 61 Table A29: Heterogeneity by baseline off-farm employment or business (1) (2) (3) (4) Pre-lean Lean Post-lean Endline Food Consumption Score Early short (β1 ) 1.89** -1.43** -0.39 -0.21 (0.78) (0.68) (0.75) (0.78) Early long( β2 ) 0.08 -0.30 -0.41 0.08 (0.72) (0.75) (0.74) (0.75) Early short × Off-farm work (β3 ) -0.06 0.48 -0.10 0.88 (0.77) (0.67) (0.84) (0.84) Early long × Off-farm work (β4 ) 0.10 -0.51 0.35 1.29* (0.75) (0.75) (0.81) (0.75) β1 + β3 = 0 0.00 0.09 0.44 0.33 β2 + β4 = 0 0.74 0.18 0.94 0.05 Observations 3833 3954 3958 3964 Food Consumption Early short (β1 ) 8288.32*** -869.36 -733.34 984.27 (2068.39) (1655.13) (2108.35) (1855.99) Early long( β2 ) 1293.22 -1072.47 1392.02 1230.99 (1822.38) (1750.97) (2019.08) (1918.13) Early short × Off-farm work (β3 ) -32.73 -4201.07* -2899.92 -1354.27 (2526.61) (2197.68) (2457.32) (2231.40) Early long × Off-farm work (β4 ) 2114.98 -5092.01** -248.13 -282.36 (2252.15) (2106.74) (2297.32) (2320.79) β1 + β3 = 0 0.00 0.01 0.06 0.83 β2 + β4 = 0 0.06 0.00 0.64 0.58 Observations 3831 3954 3958 3964 Life Satisfaction Early short (β1 ) 0.55*** -0.18 -0.05 0.06 (0.15) (0.14) (0.13) (0.14) Early long( β2 ) 0.32** -0.13 0.08 0.10 (0.14) (0.13) (0.13) (0.11) Early short × Off-farm work (β3 ) 0.07 0.00 0.05 -0.08 (0.16) (0.15) (0.11) (0.14) Early long × Off-farm work (β4 ) -0.03 0.01 -0.04 -0.15 (0.17) (0.13) (0.12) (0.13) β1 + β3 = 0 0.00 0.15 0.97 0.86 β2 + β4 = 0 0.00 0.31 0.79 0.56 Observations 3833 3954 3958 3964 Notes: This table presents intent-to-treat estimates comparing the two early interventions to the traditional lean season response for our three primary outcome variables. The interaction terms “Early short x Off-farm work” and “Early long x Off-farm work” capture the differential effects of the treatments based on whether the household had any off-farm business or non-agricultural wage employment at baseline. All specifications include village size, PMT and cohort strata fixed effects, control for households surveyed twice, and cluster standard errors at the village level (shown in parentheses). In the panel below the coefficients, we report p -values from tests of whether the sum of the main treatment effect and the interaction effect is equal to zero for each early intervention. ∗ p < 0.10, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01 62