Policy Research Working Paper 10541 Seasonal Deprivation in the Sahel Is Large, Widespread, and Can Be Anticipated Jonathan Lain Stephanie Brunelin Social Protection and Jobs Global Practice & Poverty and Equity Global Practice August 2023 Policy Research Working Paper 10541 Abstract Shocks and seasonality may have profound effects on poor implying that seasonality brings about extreme forms of households’ wellbeing, especially in contexts like the Sahel deprivation. Welfare losses may begin early in the lean where livelihoods depend on rainfed agriculture and pas- season, even as early as April. When the data were collected toralism. Understanding how seasonal variation affects in 2018/19, the climatic conditions were relatively benign Sahelian households is therefore essential for guiding poli- and the security situation was more stable than today, so cies that jointly seek to address chronic poverty, seasonality, the effects of seasonality shown in this paper likely represent and unexpected shocks. This paper uses harmonized house- a lower bound. On policy, although initiatives currently hold survey data from Burkina Faso, Chad, Mali, Niger, focus on responding to unpredictable shocks, seasonal food and Senegal, collected in two distinct waves in 2018 and insecurity could be better tackled by expanding social pro- 2019, to examine the extent of seasonal deprivation in the tection and providing regular transfers early in the lean Sahel. These data reveal significant seasonal variation in season, when prices are lower and fewer households have poverty and wellbeing. Mean real monetary consumption succumbed to extreme deprivation. Seasonal variation hap- is around 10.5 percent lower in the lean season. Moreover, pens every year and more can be done to support Sahelian rather than representing a reduction in dietary diversity, households if there is information on how it will perennially this drop is concentrated in staple foods (especially cereals), threaten their wellbeing. This paper is a product of the Social Protection and Jobs Global Practice and the Poverty and Equity Global Practice. 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 jlain@worldbank.org and sbrunelin@worldbank.org. 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 Seasonal Deprivation in the Sahel Is Large, Widespread, and Can Be Anticipated Jonathan Lain and Stephanie Brunelin1 (JEL: I38, O12, Q18) Keywords: Seasonality, Poverty, Welfare, Food Insecurity, Sahel. 1 Both authors are with the World Bank and can be reached through jlain@worldbank.org and sbrunelin@worldbank.org. The authors are extremely grateful to Sharad Tandon, who worked on the policy brief (https://openknowledge.worldbank.org/handle/10986/37725) to which this paper is related. Vital comments were also received from Christophe Rockmore, Christian Bodewig, and many others in the World Bank Social Protection and Jobs, Western and Central Africa unit. This work was made possible through support from the Sahel Adaptive Social Protection Program Trust Fund. Section 1. Introduction Sahelian households face both unpredictable shocks and regular seasonal variation, which could have dramatic effects on their wellbeing. Shocks may be ‘covariate’ – hitting all community members at the same time (like droughts or floods) – or ‘idiosyncratic’ – affecting only certain households (like ill health or injury). Either way they can reduce households’ current and future welfare, as they may resort to negative coping strategies in response (Brunelin, Ouedraogo, & Tandon, 2020). Yet on top of these unpredictable shocks, livelihoods in the Sahel – most of which are in agriculture or pastoralism – also revolve around seasonal variation in rainfall, temperature, and other climatic conditions. In particular, households have to endure a ‘lean season’ when output from the agricultural cycle is low or herds may be lacking pasture.2 Understanding how seasonal variation affects Sahelian households is essential for guiding policies that jointly seek to address chronic poverty, seasonality, and unexpected shocks. Poverty is widespread in the Sahel. Across the five countries considered in this paper – Burkina Faso, Niger, Mali, Senegal, and Chad – around one-third of people live on less than 1.90 USD 2011 PPP per day.3 Social protection programs can help protect poor and vulnerable Sahelians from shocks and seasonality while addressing chronic poverty. With the development of government-led safety net systems in the region, there has also been growing body of evidence on the impacts of social safety nets on poverty reduction, food security, and livelihood opportunities for the poor and vulnerable. Recent impact evaluations of cash transfer programs conducted in Niger and Senegal have shown that regular cash transfers help poor households mitigate the adverse effects of shocks, including those that are climate related (Premand & Stoeffler, 2020; Bossuroy, et al., 2022). Yet adaptive safety net systems need to know how to weight efforts to address chronic poverty versus unpredictable shocks versus seasonality and what the best modalities are for providing households with assistance. This paper uses new household survey data to examine Sahelian households’ susceptibility in more detail. In particular, the paper draws on the ‘Enquête Harmonisée sur les Conditions de Vie des Ménages’ (EHCVM), which collected harmonized data from Burkina Faso, Mali, Niger, Senegal, and Chad during two distinct ‘waves’ throughout 2018 and 2019. These waves correspond approximately to the lean and non- lean seasons in Burkina Faso, Mali, Niger, and Senegal, so comparing them reveals the impact of seasonality in those countries. In Chad the timing of data collection was different, so those sections of the paper making cross-season comparisons will not include Chad.4 The 2018 rainy season, which would determine the extent of the 2019 lean season, was regarded as ‘above average’ in terms of the timing and reliability of rainfall, making it possible to isolate the additional impact of ‘regular’ seasonality over and above the effects of unpredictable weather shocks.5 Descriptive statistics from the EHCVM data immediately hint at Sahelian households’ exposure to seasonality. The majority of households rely primarily on agriculture or pastoralism, and their income 2 Further details on seasonal variation in temperature, rainfall, livelihoods, and food prices can be found in Annex 2. 3 Using the 2018 poverty numbers and the 2020 population numbers reported in the World Bank ’s Poverty and Inequality platform, 27.5 percent of people in Burkina Faso, Niger, Mali, Senegal, and Chad, live below the international poverty line of 1.90 USD 2011 PPP per person per day. 4 Chad is retained for the descriptive statistics on strategies for managing and coping with seasonality in Section 3. 5 See Annex 2 and in particular analysis by Action Contre La Faim (2018) and FEWS NET (2019) for information on the 2018 rainy season. 2 sources are not well diversified, especially in rural areas. Across the five countries, 55.5 percent of the population live in a household in which the main income earner engages primarily in agriculture, raising animals, or fishing. At the same time, relatively few Sahelians hold savings, and neither borrowing nor selling assets is widely used as a coping mechanism: 28.9 percent of the population live in a household in which any member held an account with a financial institution, just 16.1 percent of the population live in a household in which any member actually held savings, while livestock sales were not clearly associated with seasonality in the EHCVM sample. Despite this exposure and widespread poverty, government support for households is insufficient, with just 10.5 percent of Sahelians living in a household that received food during the 12 months prior to being interviewed, and 3.6 percent living in a household that received cash transfers.6 Comparing the waves of the EHCVM data demonstrates the dramatic effects of seasonality on Sahelian households’ welfare. The share of the population living below the national poverty line in the wave corresponding to the lean season was 13.7 percentage points higher in Burkina Faso, 6.6 percentage points higher in Niger, and 8.1 percentage points higher in Senegal compared to the wave corresponding to the non-lean season. Pooling the data from these countries, mean real monetary consumption is estimated to be around 10.5 percent lower in the lean season. Controlling for region fixed effects and other household-level controls has little effect on these estimates. Moreover, this drop is concentrated in reduced consumption of staple foods (such as cereals), rather than the more ‘typical’ reduction in dietary diversity that might be observed as households try to adapt. Seasonality therefore brings about extreme forms of deprivation. Suggestive evidence also indicates that the welfare losses begin relatively early in the lean season, even as early as April. Turning to heterogeneity, the effects of seasonality appear to be concentrated in rural households that rely on agriculture or pastoralism for their livelihoods, but other households may still be affected. Pooling the data from Burkina Faso, Niger, and Senegal, mean real monetary consumption is estimated to be about 13.5 percent lower in the lean season than the non-lean season in rural areas, a difference that is statistically significant at the 1 percent level. In urban areas, the difference is just 2.4 percent and is not statistically significant, even at the 10 percent level.7 Within rural areas, it also appears that households cultivating fields are more affected by seasonality. Yet even households whose main income earners are engaged in services or industry appear to be at least partly impacted by seasonality, suggesting that leaving agriculture is not sufficient to escape the effects of seasonality fully. The extent of seasonal swings in welfare, especially in rural areas, has clear implications for policy, and especially for Adaptive Social Protection (ASP) programs. Numerous studies have demonstrated that social safety nets boost consumption and reduce poverty.8 Safety nets also help poor households to invest in productive assets and livestock holdings, diversify income-generating activities, and bolster savings. Yet despite safety nets’ demonstrated impact on poverty reduction and resilience building, their coverage is limited. Except for Mauritania and Senegal – where coverage of regular cash transfer programs reaches about 40 percent of the poor – coverage of poor households is below 10 percent in the rest of the Sahel. 6 In each country, the shares receiving such support are generally only slightly higher among those living below the national poverty line. 7 The difference between the effects of seasonality in rural and urban areas is formally tested using a regression with interaction terms. This is explained in detail in Section 2. 8 See, for example, Beegle, Coudouel, and Monsalve (2018). 3 As such, overall coverage of regular social assistance could be broadened to help households cope with seasonality. More specifically, while much emphasis is currently placed on responding to unpredictable shocks, seasonal food insecurity could be better tackled by providing regular transfers early in the lean season, when prices are lower and fewer households have fallen into extreme deprivation. Seasonal variation happens every year and more can be done to support Sahelian households if we know it will perennially threaten their wellbeing. The paper echoes many existing studies that show how household welfare depends directly on shocks and seasonality. As Dercon’s (2002) review demonstrates, consumption responds to seasonal variation in or shocks to household income in a wide range of low- and middle-income countries. These swings in income and consumption often leave households at risk of falling into poverty (Dercon & Krishnan, 2000). The paper also builds on a large literature that describes at least five reasons why households’ ability to manage and cope with shocks and seasonality better differs. First, households with more diversified livelihoods, or incomes that depend less on agriculture, may be less exposed to the effects of seasonality (Alderman & Paxson, 1994; Barrett, Reardon, & Webb, 2001; Harrower & Hoddinott, 2005; Kaminski, Christiaensen, & Gilbert, 2016). Second, households may try to borrow in seasons when incomes or access to food are reduced, but such strategies will be more tenable for those with better access to formal or informal financial markets (Pitt & Khandker, 2002; Khandker, 2012). Third, households can save, store output, or accumulate assets during surplus seasons which can then be sold or consumed when needed, although often households are reluctant to sell their livestock during difficult times if they can stay above some subsistence level of consumption (Hoogeveen, 2003; Fafchamps & Lund, 2003; Kazianga & Udry, 2006). Fourth, certain households may have access to informal insurance whereby communities pool risks and provide support to those in need, although such insurance mechanisms tend to be most relevant to idiosyncratic shocks rather than covariate shocks or seasonal variation (Mace, 1991; Townsend, 1994; Ravallion & Chaudhuri, 1997). Fifth, government policy – and, in particular, access to social protection – influences the base level of income to which households have access: such social protection may be adapted to respond explicitly to shocks and seasonality (Bodewig, 2019). The rest of this paper is organized as follows. Section 2 describes the data and methodology. Section 3 presents some descriptive statistics, showing the limitations on strategies for coping with and managing seasonality in the Sahel. Section 4 presents the main results, comparing key markers of welfare between the lean and non-lean season. Section 5 describes heterogeneity in the main results, including comparing rural and urban areas. Section 6 concludes and describes the policy implications. Section 2. Data and methodology The main data source used in this paper is the ‘Enquête harmonisée sur les conditions de vie des ménages’ (EHCVM), which captures detailed household-level microdata from Burkina Faso, Mali, Niger, Senegal, and Chad. The EHCVM includes a wide range of socioeconomic indicators, including consumption, employment and agricultural activities, food security, assets, education, and health. The EHCVM data are representative at the national and regional levels for each of the five countries. The timing of the EHCVM makes it possible to assess the impact of seasonality in four of the five Sahelian countries surveyed as the data were collected in two distinct waves during the same agricultural year; the first wave corresponding to the harvest period and the second one to the lean season. The waves collected information on different households, so the EHCVM data do not have a panel structure. However, the 4 data are representative at the wave level, so it is possible to compare the two waves in each country to assess the impact of seasonality. In the absence of widespread irrigation systems, cereals are harvested only once a year from October to December; the agricultural lean season, which coincides with the rainy season, therefore runs from June to September while the pastoral lean season starts earlier and takes place between April and June. For Burkina Faso, Mali, Niger, and Senegal, the bulk of the interviews in the first wave took place between October 2018 and December 2018, although interviewing began in Burkina Faso and Senegal in September 2018 (see Annex 1). The second wave for these four countries was carried out between April 2019 and July 2019. Thus, for Burkina Faso, Mali, Niger, and Senegal, the first wave roughly corresponds with the harvest and immediate post-harvest (non-lean) season, especially as the 2018 rains arrived on time and resulted in above average biomass production (Action Contre La Faim, 2018). By contrast, the second wave captures the 2019 pastoral lean season and the first part of the agricultural lean season.9 Since the timing of the EHCVM is similar in Burkina Faso, Mali, Niger, and Senegal, results from these countries can be pooled as well as presented separately. Annex 2 provides more detail on the timing of seasons in the Sahel. The cross-wave comparisons for Chad will not be pooled with the other countries, given the timing of the EHCVM data collection there. In Chad, the first wave of the EHCVM data was collected between June and September 2018, which corresponds to the lean season of the 2017/2018 agricultural year, while the second wave was collected between January and April 2019. This differs too much from the data collection schedule in Burkina Faso, Mali, Niger, and Senegal for the data from all five countries to be combined when comparing the waves. However, Chad is retained for Section 3 to provide additional descriptive evidence on potential strategies for managing and coping with seasonality. While useful, further temporal disaggregation to isolate how welfare differs for particular months of the lean season should be approached carefully. The timing of the second wave in the EHCVM data collection approximately coincides with the 2019 agricultural lean season for Burkina Faso, Mali, Niger, and Senegal, but the correspondence is not perfect. Indeed, the second wave starts and ends slightly too early in these four countries to capture the full extent of the 2019 agricultural lean season, and is better aligned with the pastoral lean season. Focusing on month-specific results may help to ascertain how welfare changes as the agricultural lean season evolves. However, the EHVCM is not designed to be representative at the month level, so these additional month-specific results should be interpreted with some caution. Ramadan overlapped with the lean season in 2019, which could lead the analysis to underestimate the effects of seasonality. In 2019, Ramadan took place from 5th May to 3rd June, falling directly in the middle of the lean season wave of the EHCVM data collection. All of the countries in the sample are majority Muslim – with Mali, Niger, and Senegal being almost exclusively Muslim – so this may impact consumption patterns. All other things equal, consumption typically rises for Muslim households in Ramadan and reported food insecurity declines (see, for example, MVAM (2021)). Thus, it may be that the overall lean season losses estimated in this analysis represent an underestimate of what might happen when Ramadan does not correspond with the lean season. Notwithstanding the caveats outlined above, looking at month- 9 The rains in 2019 arrived late in some western parts of the Sahel region, which may have led the 2019 lean season to arrive early as well, at least for pastoralists. However, the full effects of the late 2019 rains would be expected in 2020. 5 by-month results could help to understand how Ramadan influences the effects of the lean season on household consumption. The main variables used to measure the effects of seasonality on welfare come from the EHCVM’s detailed consumption module. The EHCVM questionnaire collects data on up to 138 food items, recording both own-produced food and purchased food consumed over the previous seven days. Information is also recorded on expenditures on food items over the past 30 days. This can be combined with information on spending on education, health, housing, and many other non-food items to produce an overall consumption aggregate that can be used to proxy welfare; however, specific elements of the consumption aggregate may also be considered separately. This consumption aggregate is spatially and temporally deflated, using prices collected within the EHCVM itself. This allows different households to be compared and allows poverty to be calculated using a single national poverty line. The consumption levels are also converted to 2011 USD using Purchasing Power Parities (PPPs) to facilitate comparisons between countries, where necessary.10 The detailed consumption data are also complemented by additional information on food security, subjective poverty, and employment to assess the impacts of seasonality fully.11 The comparisons between the lean and non-lean season waves can be enhanced using multivariate regressions. Since the EHCVM data are representative at the wave level, the raw differences in means between the waves for key outcome variables should be sufficient for estimating seasonal variation in welfare across the 2018/19 agricultural and pastoral cycle.12 To strengthen these results, however, and to ensure that they do not arise due to confounding differences in the sample between the two seasons, it is also be important to test whether any differences in the outcome variables change when controlling for stable household and location characteristics, that is, household and location characteristics that would not be expected to change dramatically season to season. 13 Specifically, for household , in community , in region , in country , we run a regression of the form:14 (1) = + ′ ′ + + where is the outcome variable of interest, is a dummy variable capturing the season in which the household was interviewed, is a vector of household controls, including the age, sex, and education of the household head, and the size and number of dependents in the household, is a series of region fixed effects, and is the error term. The specific set of household and location controls and 10 Leaving consumption in nominal terms does not significantly affect the main results. 11 The time horizons for questions on non-monetary poverty indicators, such as access to health and education, are too long in the EHVCM questionnaire to allow for seasonal variation to be captured. 12 Section 4 also shows consumption differences across the entire distribution. 13 If remains stable as variables in and the fixed effects are added incrementally, this may suggest that unobserved confounding variables are less likely to be biasing the results (Altonji, Elder, & Taber, 2005; Oster, 2019). We therefore report the results with and without controls and region fixed effects. 14 This approach follows a wide range of descriptive work looking at the impact of shocks and seasons on household welfare, see for example Dercon, Hoddinott, and Woldehanna (2005), Harrower and Hoddinott (2005), and World Bank (2005). 6 their functional form is described in Annex 3. The coefficient can be interpreted as the effect of seasonality.15 To examine heterogeneity in the impacts of seasonality, the regressions can be augmented with interaction terms. Simply splitting the sample into different segments of the population provides a first check on heterogeneity in the results. However, to formally test whether the effects of seasonality differ for certain sub-populations, it is also helpful to run regressions of the form: (2) = + ( × ) + + ′ ′ + + where all variables are defined as above, gives the ‘cutting variable’ (such as urban-rural or different livelihoods), and can be interpreted as the differential impact of seasonality on certain types of households.16 Section 3. Strategies for managing and coping with seasonality in the Sahel Building on the literature described above, descriptive statistics from the EHCVM data immediately reveal three key characteristics of Sahelian households, which could limit their strategies for managing and coping with seasonality. First, their livelihoods are concentrated in agriculture and rearing livestock and are not well diversified. Second, financial inclusion appears to be low; holdings savings and taking out loans are relatively rare phenomena, even accounting for mobile money. Third, social assistance that could help households weather both seasonality and shocks does not appear to be widespread. Section 3.1. Livelihoods in the Sahel Sahelian households, especially those in rural areas, depend mainly on rainfed agriculture, which is especially exposed to the effects of seasonality. In the absence of widespread irrigation systems, agriculture in the Sahel is mainly rainfed, so yearly swings in temperature and rainfall are crucial for determining output (see Annex 2 for further details). Pooling data from the EHCVM countries, 50.5 percent of the population live in a household where the head’s primary sector of work was agriculture and a further 5.0 percent of the population live in a household where the head worked primarily in raising animals or fishing (Figure 1). 17 Nevertheless, there is substantial variation between the five countries, with work in agriculture and livestock being relatively more common in Burkina Faso, Mali, 15 When is a binary variable, the regression will still be estimated using Ordinary Least Squares, yielding a Linear Probability Model. This means the coefficients can directly be read as marginal effects. The standard errors will be robust to the inherent heteroskedasticity introduced into the model from this approach, as they will be clustered at the ‘grappe’ (or cluster) level for each regression. The results are qualitatively similar if a probit model is estimated and the marginal effects at the mean are calculated. 16 The regressions with interaction terms focus only on continuous outcome variables, principally deflated consumption per capita. This avoids the challenges associated with estimating interaction terms with a binary outcome variable, which may arise even in Linear Probability Models (Holm, Ejrnæs, & Karlson, 2015). 17 The statistics shown in Figure 1 focus on the household head’s primary activity over the last 12 months. Similarly large shares of household members work in agriculture, if looking at the activities in which they engaged during the last seven days. Across the five countries, 52.8 percent of households had a head engaged in farming in the previous seven days and 58.2 percent of all working-age individuals were engaged in farming in the previous seven days. 7 Niger, and Chad than in Senegal. Work in agriculture and livestock is also about five times more prevalent in rural areas compared to urban areas, pooling the data from all five EHCVM countries. Figure 1. Share of the population living in households where the head is primarily engaged in different employment sectors over the last 12 months by country and rural-urban 100 Share of population living in a household with a head working primarily in each sector over the last 12 90 80 70 60 months (percent) 50 40 30 20 10 0 Rural Rural Urban Rural Urban Rural Urban Rural Urban Rural Urban Urban Total Total Total Total Total Total Burkina Faso Mali Niger Senegal Chad All countries Agriculture Livestock/fishing Industry/construction Services/other Not working Note: Individual-level sampling weights applied. Primary occupation is the one in which the household head spent the most time working in the previous 12 months. Source: EHCVM and World Bank estimates. Cultivating fields and owning animals were also widespread across the EHCVM sample, especially in rural areas. Pooling all five countries, some 66.4 percent of households own animals and 68.1 percent of households had cultivated fields in the previous agricultural season (Figure 2). Animal ownership is 2.3 times more common in rural areas than urban areas, while cultivating fields is about 4.5 times more common. There is also variation between the countries, with Senegalese households being far less likely than households in the other four countries to cultivate fields, in line with the statistics presented on household head occupation. 8 Figure 2. Share of households owning animals and cultivating fields, by country and urban-rural Share of households (percent) 100 80 60 40 20 0 Urban Urban Urban Urban Urban Urban Total Total Total Total Total Total Rural Rural Rural Rural Rural Rural Burkina Faso Mali Niger Senegal Chad All Any animals Any fields Note: Household-level sampling weights applied. ‘Any animals’ measured at the time of the interview. ‘Any fields’ refers to whether fields were cultivated in the previous agricultural season. Source: EHCVM and World Bank estimates. We also observe that diversification of income sources is relatively low among Sahelian households. Across the EHCVM sample, just under half of households (46.8 percent) had any working-age members primarily engaged in activities outside of agriculture, livestock, or fishing, with households as a whole engaging in 1.5 different sectors on average (Figure 3). Income-generating activities also appear to be less diversified in rural areas than urban areas. Figure 3. Diversification of household employment sectors, by country and urban-rural 2.5 100 primarily engaged outside of agriculture or Mean number of sectors of primary work over the last 12 months in the household livestock in the last 12 months (percent) 90 Share of households with a worker 2.0 80 70 1.5 60 50 1.0 40 30 0.5 20 10 0.0 0 Rural Urban Rural Urban Rural Urban Rural Urban Rural Urban Total Total Total Total Total Burkina Faso Mali Niger Senegal Chad All Number of sectors (LHS) Any non-agriculture/non-livestock (RHS) Note: Household-level sampling weights applied. Only working-age household members (those aged 15-64 years or more) that reported being employed or in unpaid family work in the previous 12 months used to assess the sectoral diversification of the household. Households without any working members coded as zero for number of sectors. Source: EHCVM and World Bank estimates. 9 Section 3.2. Financial inclusion in the Sahel Penetration of financial institutions is generally low, and most households do not hold savings, although there is sizeable variation between the five EHCVM countries. Access to financial markets may help households cope with the effects of seasonality through credit or insurance products, while savings may enable households to smooth any losses in consumption that arise in the lean season. Pooling data from across the EHVCM sample, around 28.9 percent of the population live in a household in which any member held an account with a financial institution, and just 16.1 percent of the population live in a household in which any member actually held savings (Figure 4).18 Nevertheless, financial inclusion differs significantly between the five countries, with more than half of the population of Burkina Faso and Senegal living in a household where someone holds an account compared to less than 1 in 10 in Niger and Chad. This higher financial inclusion in Burkina Faso and Senegal seems to be driven largely by the penetration of mobile banking, the most prevalent form of account in the two countries.19 Figure 4. Share of the population with access to financial institutions and savings, by country 60 Share of the population living in a household with such an account 50 40 30 (percent) 20 10 0 Regular bank Postal Rural savings Mobile Stored-value Any Any (with association / banking / prepaid savings) microfinance card Burkina Faso Mali Niger Senegal Chad All countries Note: Individual-level sampling weights applied. Statistics capture whether any household member had an account or had savings. Source: EHCVM and World Bank estimates. There is no clear evidence that households take out loans or are able to sell livestock to cope with the lean season. Across the EHCVM sample, around 16.2 percent of households had outstanding loans at the time of the survey, but the dates when these loans were agreed do not reveal any clear seasonal patterns. Additionally, despite the hypothesis that households may use sales of livestock as a buffer to insulate against losses induced by shocks or seasonality, existing evidence from both the Sahel and other countries suggests that this mechanism is not frequently observed in practice.20 In-keeping with this, in the EHCVM data, the number of animals that households own in fact appears to be very slightly higher in the wave corresponding to the lean season, taking Burkina Faso, Mali, Niger, and Senegal together. These results persist even when disaggregated by animal type – large ruminants, small ruminants, pigs, poultry, and 18 The financial institutions considered are regular banks; postal banks; rural savings associations or microfinance institutions; mobile banking; and stored-value or prepaid cards. 19 This may partly reflect patterns of mobile phone ownership. According to the EHCVM assets module, around 95.3 percent, 93.7 percent, and 98.8 percent of households own at least one mobile phone in Burkina Faso, Mali, and Senegal respectively, compared to 72.7 percent and 64.4 percent of households in Niger and Chad. 20 See Hoogeveen (2003), Fafchamps and Lund (2003), and Kazianga and Udry (2006). 10 rabbits. Therefore, livestock sales do not appear to be a coping mechanism that is widely available to Sahelian households either. Section 3.3. Social assistance in the Sahel Few households receive food or cash transfers, although in-kind benefits are more widespread. Distribution of food, cash, or in-kind benefits may further help to smooth consumption between the lean and non-lean season. However, for the survey period, only around 10.5 percent of Sahelians live in a household that received food during the previous 12 months, and 3.6 percent live in a household receiving cash transfers (Figure 5). Yet a much larger share of the population receives in-kind support in the form of care for children under 5 and bed nets. These in-kind benefits may help boost longer-term efforts to build human capital, but they may be less likely to support households in coping with short-term shocks and seasonality. Figure 5. Share of the population receiving support from different social safety nets, by country 60 50 40 Percent 30 20 10 0 Cash transfers from the Food for work Any food Program for pregnant Cereals Flour from cereals Care for children under 5 Distribution of bed nets Any non-food Food for school students Public works malnourished children Supplements for government women Food Non-food Burkina Faso Mali Niger Senegal Chad All countries Note: Individual-level sampling weights applied. Statistics capture whether any household member received that type of support during the previous 12 months. Source: EHCVM and World Bank estimates. These descriptive statistics show that Sahelian households’ mechanisms for managing and coping with the effects of seasonality may be limited. All other things equal, this could leave households exposed to large seasonal swings in welfare. We now turn to the main results to assess whether this is the case. Section 4. Main results In this section we present the main results, showing how welfare differs between the lean and non-lean seasons in the EHCVM sample. We first present the effects of seasonality on overall consumption and poverty. We then disaggregate these changes in consumption, showing how the data are consistent with 11 more extreme forms of deprivation where households reduce their intake of staple foods. Finally, we consider other markers of welfare, including subjective measures of poverty. 4.1. Overall monetary consumption and poverty We begin by estimating Equation 1 with the log of consumption – in 2011 PPP-adjusted USD – as the independent variable. As described above, Chad is excluded from this analysis due to the timing of the EHVCM waves. Mali is also excluded from any specifications involving monetary consumption due to limitations on the consumption data.21 Pooling the data from Burkina Faso, Niger, and Senegal, monetary consumption is approximately 10.5 percent lower in the lean season according to the specification without controls, a difference which is statistically significant even at the 1 percent level (Table 1).22 At the mean level of consumption across those three countries, this represents a difference of 0.37 USD 2011 PPP per person per day. This estimate only drops to 9.9 percent when a full set of controls are included and remains statistically significant. Therefore, seasonal swings in welfare appear to be sizeable in the EHCVM sample. Table 1. Difference in the log of total monetary consumption per capita between the lean season wave and non-lean season wave Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season -0.1793*** -0.1397*** -0.0605* -0.0671*** -0.0899** -0.0888*** -0.1049*** -0.0988*** wave (0.0529) (0.0268) (0.0359) (0.0217) (0.0437) (0.0184) (0.0305) (0.0135) N 7017 6936 6020 5919 7153 7107 20190 19962 R-squared 0.0183 0.5474 0.0030 0.4726 0.0058 0.5607 0.0066 0.5858 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is the log of total monetary consumption per capita. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as region fixed effects. Source: EHCVM and World Bank estimates. Moreover, monetary consumption appears to be lower in the lean season across the consumption distribution. Looking at the probability density functions for the log of consumption in the lean season and non-lean season in Burkina Faso, Niger, and Senegal demonstrates that it is only around the 90th percentile that the two distributions begin to converge (Figure 6). Thus, poorer households – including 21 Including the consumption data that are available for Mali would, if anything, strengthen the results. Estimating Equation 1 with the data for Burkina Faso, Mali, Niger, and Senegal, consumption was 11.5 percent lower in the lean season than in the non-lean season in the specification without controls and 10.2 percent lower in the specification with full household and location controls. 22 Here we are reading the coefficient in the log-linear regression directly as a percentage change. Exponentiating this coefficient does not dramatically alter the results. It should also be noted that these results differ slightly from those presented in the previous policy brief using the EHCVM data. There, the difference in means were calculated using a linear model rather than a log-linear model. 12 below or those vulnerable to falling below the poverty line – appear to be susceptible to seasonal changes in welfare. Figure 6. Kernel density charts showing the difference in total monetary consumption per capita between the lean season and non-lean season waves across the full consumption distribution Note: Epanechnikov kernel density function used. Bandwidth set to 0.1. Individual-level sampling weights applied. Source: EHCVM and World Bank estimates. Given these movements in consumption, seasonality appears to have profound effects on poverty in Burkina Faso, Niger, and Senegal. To show this, we estimate Equation 1 with poverty status according to the national poverty line as the dependent variable; this is a Linear Probability Model. Since these estimates use each country’s national poverty line, the results cannot be pooled across countries. In the specifications without controls, the lean-season poverty headcount rate is 13.7 percentage points higher in Burkina Faso, 6.6 percentage points higher in Niger, and 8.1 percentage points higher in Senegal than the non-lean-season poverty headcount rate (Table 2). These differences are all statistically significant at least at the 5 percent level and the magnitude does not change substantially when controls are added. Thus, the seasonal movements in monetary consumption are sufficient to push large portions of the Sahelian population below their national poverty lines. 13 Table 2. Difference in share of the population living below the national poverty line between the lean season and non-lean season waves Burkina Faso Niger Senegal No controls Full controls No controls Full controls No controls Full controls Lean season 0.1369*** 0.1158*** 0.0659** 0.0697*** 0.0810*** 0.0816*** wave (0.0332) (0.0206) (0.0274) (0.0199) (0.0306) (0.0164) N 7017 6936 6020 5919 7153 7107 R-squared 0.0198 0.3093 0.0045 0.2421 0.0071 0.3441 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level are in parentheses. Dependent variable is a dummy variable taking 1 if a household’s consumption level is below the national poverty line and 0 otherwise. Individual-level sampling weights are applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as an urban-rural dummy and region fixed effects. Source: EHCVM and World Bank estimates. Month-by-month results suggest that the effects of seasonality may be even larger than the estimates above, which simply compare the lean and non-lean season waves of the EHCVM. To show this, we modify Equation 1 but rather than including a single dummy variable for the lean season as a regressor, we include four dummies that capture whether or not the date of the interview was in April, May, June, or July of 2019. The base category remains the non-lean season wave. For this model, the specification with household and location controls is preferred because the EHCVM is not designed to be representative at the month level, so there could be systematic differences in the sub-samples from April, May, June, and July. Pooling the data from Burkina Faso, Niger, and Senegal, the drop in consumption compared to the non-lean season wave increases between April and July, although this is driven principally by Burkina Faso and Niger (Table 3). This corroborates the notion that the agricultural lean typically continues to September or October each year and could still be yet to reach its peak in July; the effects of seasonality may therefore be understated by simply comparing the lean and non-lean season waves. 14 Table 3. Regression results showing the difference in total monetary consumption per capita between the non-lean season and different months in the lean season All except Chad and Burkina Faso Niger Senegal Mali April -0.0263 -0.0253 -0.1441*** -0.0908*** (0.0409) (0.0418) (0.0297) (0.0237) May -0.1428*** -0.0657** -0.0556*** -0.0942*** (0.0321) (0.0309) (0.0208) (0.0175) June -0.1579*** -0.0714** -0.0865*** -0.1027*** (0.0338) (0.0281) (0.0283) (0.0181) July -0.1793*** -0.1549*** -0.0780 -0.1613*** (0.0523) (0.0272) (0.0826) (0.0266) N 6936 5919 7107 19962 R-squared 0.5486 0.4738 0.5626 0.5861 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is the log of total monetary consumption per capita. Individual-level sampling weights applied. All specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as an urban-rural dummy and region fixed effects. Source: EHCVM and World Bank estimates. 4.2. Disaggregated differences in consumption We now disaggregate the results by different parts of the consumption basket to characterize more fully the deprivation implied by the overall changes in consumption and poverty described in Section 4.1. We begin by breaking down the results by food and non-food consumption; that is, we re-estimate Equation 1 with the log of food consumption per capita and the log non-food consumption per capita as the main regressors. Pooling the data from Burkina Faso, Niger, and Senegal, food consumption was 11.4 percent lower in the lean season wave while non-food consumption was 7.4 percent lower (Table 4). The difference between these two effects is even larger at 11.3 percent for food consumption and 6.3 percent for non-food consumption when full controls are included.23 This in itself suggests that seasonality may have more effect on food consumption. Nevertheless, these results may partly be an artifact of the way in which the EHVCM data were collected; the recall period for the consumption of food items generally covered the previous seven days, while non-food items were recorded for the previous seven days, 30 days, three months, six months, and 12 months, which may be less sensitive to seasonal swings. It is therefore important to dig deeper into the consumption basket. 23 We use Stata’s suest command to formally test the difference between the effects of seasonality on food and non-food consumption. When pooling Burkina Faso, Niger, and Senegal, the p-value on the formal test for the difference between the food and non-food coefficients is 0.1190 without controls and 0.0003 with controls. 15 Table 4. Difference in food and non-food consumption per capita between the lean season and non- lean season waves Panel A: Food Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season -0.1976*** -0.1544*** -0.0839** -0.0877*** -0.0823** -0.0799*** -0.1140*** -0.1133*** wave (0.0478) (0.0321) (0.0353) (0.0270) (0.0349) (0.0188) (0.0279) (0.0160) N 7016 6935 6020 5919 7153 7107 20189 19961 R-squared 0.0223 0.3853 0.0055 0.2656 0.0052 0.3961 0.0079 0.4252 Panel B: Non-food Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season -0.1380** -0.1053*** -0.0023 -0.0150 -0.0937* -0.0939*** -0.0739* -0.0630*** wave (0.0642) (0.0265) (0.0430) (0.0209) (0.0556) (0.0219) (0.0378) (0.0136) N 7017 6936 6020 5919 7153 7107 20190 19962 R-squared 0.0079 0.5973 0.0000 0.5778 0.0043 0.5817 0.0023 0.6411 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is the log of total monetary consumption per capita; for food items in Panel A and for non-food items in Panel B. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as region fixed effects. Source: EHCVM and World Bank estimates. Considering the specific food items in which the lean-season drop in consumption was concentrated indicates particularly extreme forms of seasonal deprivation. In particular, the difference between the lean and non-lean season waves in terms of expenditure on staple cereal goods was particularly large. Pooling the data from Burkina Faso, Niger, and Senegal, the seasonal difference in expenditure on cereal foods was as much as 15.2 percent without controls and 15.8 percent with controls (Table 5). Similar effects are seen when looking at the quantities of cereals consumed in the lean and non-lean season. At the same time, any differences in dietary diversity – captured by comparing the World Food Programme’s Food Consumption Score (FCS) – between the lean and non-lean season are far less clear cut.24 These season-to-season changes in consumption do no match the more ‘typical’ ways in which households would cope with price and income shocks, whereby consumption of more expensive or nutritious food 24 The FCS measures dietary diversity by counting the number of days out of the last seven days in which a food from each food group was consumed, then aggregating across food groups using a weighted sum (WFP, 2008). Households with a FCS of less than 42 are said to have ‘poor or borderline’ food security. When Equation 1 is estimated with a dummy variable for households having poor or borderline food security as the dependent variable – using the pooled data from Burkina Faso, Niger, and Senegal – the difference between the lean season is only around 1.9 percentage points and is not statistically significant at the 5 percent level, with or without controls. For Burkina Faso, Niger, and Senegal, around 1 in 5 people live in households with poor or borderline food security as per the FCS, when pooling data from the two waves. 16 would be reduced but consumption of staple foods would be maintained. Instead, households are cutting the consumption of the most fundamental food items in their consumption basket. Table 5. Difference in expenditure on cereals and quantities of cereals consumed between the lean and non-lean season waves Panel A: Expenditure Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season -0.1898*** -0.1720*** -0.1383*** -0.1430*** -0.1450*** -0.1425*** -0.1517*** -0.1575*** wave (0.0419) (0.0333) (0.0316) (0.0267) (0.0266) (0.0216) (0.0230) (0.0165) N 6898 6824 5918 5832 6944 6910 19760 19566 R-squared 0.0237 0.1938 0.0166 0.1587 0.0163 0.1286 0.0165 0.2064 Panel B: Quantities consumed Burkina Faso Niger Senegal All except Chad and Mali No Full Full No No controls No controls Full controls Full controls controls controls controls controls Lean season -0.1347*** -0.1298*** -0.1272*** -0.1280*** -0.2164*** -0.2102*** -0.1524*** -0.1555*** wave (0.0408) (0.0285) (0.0313) (0.0240) (0.0259) (0.0212) (0.0229) (0.0145) N 6903 6829 5920 5834 6946 6912 19769 19575 R-squared 0.0140 0.2114 0.0137 0.2030 0.0371 0.1324 0.0173 0.2364 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is the log of total monetary expenditure on cereals per capita in Panel A and the log of total quantity of cereals consumed, in kilograms, in Panel B. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as an urban-rural dummy and region fixed effects. Source: EHCVM and World Bank estimates. 4.3. Subjective poverty Sahelian households’ self-perceptions seem to echo the finding that welfare is lower in the lean season. In the EHCVM questionnaire, households were asked to classify themselves on a scale from ‘very poor’ to ‘very rich’; we can therefore re-estimate Equation 1 with a dummy variable for whether households classified themselves as ‘poor’ or ‘very poor’ as the dependent variable. Pooling the data from Burkina Faso, Mali, Niger, and Senegal, the share of people that were subjectively poor by this metric was 4.4 percentage points higher in the lean season than in the non-lean season (Table 6). This difference was statistically significant at the 1 percent level and was largely unchanged by the introduction of controls. 17 Table 6. Difference in subjective poverty between the lean and non-lean season waves Burkina Faso Mali Niger Senegal All except Chad No Full No Full Full No Full No Full No controls controls controls controls controls controls controls controls controls controls Lean season 0.0068 -0.0004 0.0622** 0.0465** 0.0561*** 0.0499*** 0.0558** 0.0567*** 0.0435*** 0.0405*** wave (0.0253) (0.0216) (0.0243) (0.0203) (0.0214) (0.0172) (0.0256) (0.0192) (0.0133) (0.0097) N 6993 6912 6582 6550 5999 5898 7102 7057 26676 26417 R-squared 0.0001 0.1438 0.0046 0.1256 0.0048 0.1512 0.0031 0.1627 0.0022 0.1776 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is a dummy variable taking 1 if a household classified itself as ‘poor’ or ‘very poor’ and 0 otherwise. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, ag e, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as an urban-rural dummy and region fixed effects. Source: EHCVM and World Bank estimates. Section 5. Heterogeneity This section explores heterogeneity in the main results. First, we demonstrate that non-farm enterprise work appears to be a coping mechanism deployed by the Sahelian households during the lean season; this in itself suggests there may be heterogeneity in the results, because such opportunities outside of agriculture may not be available to everyone.25 Second, we show that the negative effects of the lean season on consumption appear to be concentrated in rural areas and, specifically, among those households that cultivate fields. Third, however, we show that even when households’ main income earners work outside of agriculture that may not offer insulation from the knock-on effects of seasonality. 5.1. Non-farm enterprise work as a coping mechanism Before assessing heterogeneity in the main results directly, we first consider whether typical mechanisms for coping with seasonality – which may be more widely available to certain types of households – show up in the data. It emerges that work in non-enterprise work, which households could use to bolster waning agricultural production, is indeed more prevalent in the lean season. Pooling all EHCVM countries except Chad, both household heads and all household members are about 3.6 percentage points more likely to be engaged in non-farm enterprise work in the lean season than the non-lean season (Table 7). This difference is statistically significant at (at least) the 5 percent level and is robust to the inclusion of controls. Nevertheless, altering the mix of activities in this way is unlikely to be possible for all Sahelian households. Looking across the countries, the effects are much stronger in Senegal, where household heads are as much as 5.8 percentage points more likely to engage in non-farm enterprise work in the lean season. Yet the descriptive statistics in Section 3 also demonstrate how non-agricultural activities are generally far less prevalent in rural areas; rural households may therefore struggle to find such activities when faced with the effects of the lean season. 25 For a detailed description of some additional income-generating activities undertaken by young people in Central Mali, see Von Der Goltz et al. (2022). 18 Table 7. Difference in the share of household heads or any household members engaged in non-farm enterprise activities between the lean season and non-lean season waves in the last seven days Panel A: Household head Burkina Faso Mali Niger Senegal All except Chad No Full No Full No Full No Full No Full controls controls controls controls controls controls controls controls controls controls Lean season 0.0114 0.0209 0.0337 0.0383* 0.0397 0.0400** 0.0576*** 0.0539*** 0.0358*** 0.0356*** wave (0.0244) (0.0164) (0.0257) (0.0219) (0.0241) (0.0192) (0.0189) (0.0156) (0.0122) (0.0096) N 7016 6936 6600 6568 6020 5919 7153 7107 26789 26530 R-squared 0.0002 0.1829 0.0015 0.0974 0.0018 0.1176 0.0038 0.0974 0.0016 0.1150 Panel B: Any household member Burkina Faso Mali Niger Senegal All except Chad No Full No Full No Full No Full No Full controls controls controls controls controls controls controls controls controls controls Lean season 0.0247 0.0350 0.0131 0.0521** 0.0466* 0.0497** 0.0547** 0.0587*** 0.0357** 0.0474*** wave (0.0313) (0.0227) (0.0324) (0.0250) (0.0277) (0.0214) (0.0216) (0.0160) (0.0156) (0.0113) N 7017 6936 6600 6568 6020 5919 7153 7107 26790 26530 R-squared 0.0006 0.1986 0.0002 0.1411 0.0022 0.1377 0.0034 0.1386 0.0013 0.1768 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is a dummy variable taking 1 if the household head was engaged in non-farm enterprise work in the last seven days and 0 otherwise in Panel A. Dependent variable is a dummy variable taking 1 if any working-age household member (aged 15-64 years) was engaged in non-farm enterprise work in the last seven days and 0 otherwise in Panel B. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as an urban-rural dummy and region fixed effects. Source: EHCVM and World Bank estimates. 5.2. Rural-urban differences We now examine whether the sharp differences in potential strategies for managing and coping with seasonality between rural and urban areas observed in Section 3, translate to rural-urban differences in welfare. It turns out that overall impacts on consumption observed in Section 4 are significantly concentrated in rural areas. Pooling the data from Burkina Faso, Niger, and Senegal, real per capita consumption is about 12.1 percent lower in the lean season in rural areas, a difference which is statistically significant at the 1 percent level and is robust to the inclusion of controls (Table 8). By contrast, there is no statistically significant difference in consumption levels between the lean and non-lean season in urban areas. Moreover, when the difference in the impacts of seasonality between rural and urban areas is assessed using a regression like in Equation 2, the interaction term (between the urban dummy and the lean season dummy) is statistically significant at the 1 percent level in the specification including full controls. This confirms the stark differences between rural and urban Sahelian households in terms of the effects of seasonality. 19 Table 8. Difference in total monetary consumption per capita between the lean season and non-lean season waves, split by rural and urban households Panel A: Rural Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season -0.1861*** -0.1871*** -0.0866** -0.0991*** -0.0911** -0.0900*** -0.1211*** -0.1305*** wave (0.0444) (0.0331) (0.0346) (0.0243) (0.0406) (0.0251) (0.0268) (0.0168) N 3863 3802 4447 4359 3214 3194 11524 11355 R-squared 0.0272 0.3721 0.0081 0.3625 0.0081 0.4274 0.0123 0.4362 Panel B: Urban Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season -0.0668 0.0150 0.0553 0.0735* -0.0917* -0.0891*** -0.0422 -0.0238 wave (0.0782) (0.0355) (0.0790) (0.0419) (0.0544) (0.0260) (0.0445) (0.0194) N 3154 3134 1573 1560 3939 3913 8666 8607 R-squared 0.0025 0.6005 0.0018 0.5446 0.0064 0.5255 0.0011 0.5779 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Dependent variable is the log of total monetary consumption per capita. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as region fixed effects. Source: EHCVM and World Bank estimates. Within rural areas, we observe that it is crop-cultivating households where the effects of seasonality appear to be largest; non-crop-cultivating households in rural areas appear to be impacted far less. For crop-cultivating households, the difference in consumption between the lean and non-lean season wave is around 13.1 percent, statistically significant at the 1 percent level (Table 9). By contrast, there is no statistically significant difference between the two waves for rural households that do not cultivate crops. Again, using Equation 2 to test the difference between these two effects, the coefficient on the interaction term (between the crop cultivation dummy and the lean season dummy) is statistically significant at the 1 percent level, in the specification with full controls. Therefore, it is agriculture that exposes rural households to the effects of seasonality. 20 Table 9. Difference in total monetary consumption per capita between the lean season and non-lean season waves for rural households, split by whether households cultivated fields in the previous agricultural season Panel A: Did not cultivate fields Burkina Faso Niger Senegal All except Chad and Mali No Full No Full No Full No Full controls controls controls controls controls controls controls controls Lean season 0.0241 -0.1871* -0.0478 -0.0643 -0.0212 -0.0167 -0.0126 -0.0527* wave (0.1549) (0.0946) (0.0826) (0.0432) (0.0538) (0.0383) (0.0519) (0.0276) N 269 255 1024 966 920 904 2213 2125 R-squared 0.0003 0.5862 0.0018 0.4525 0.0005 0.3924 0.0001 0.4847 Panel B: Cultivated fields Burkina Faso Niger Senegal All except Chad and Mali Full No Full Full Full No controls No controls No controls controls controls controls controls controls Lean season -0.2001*** -0.1959*** -0.0841** -0.0994*** -0.1086*** -0.1095*** -0.1310*** -0.1416*** wave (0.0429) (0.0335) (0.0325) (0.0253) (0.0381) (0.0258) (0.0253) (0.0177) N 3594 3547 3423 3393 2294 2290 9311 9230 R-squared 0.0334 0.3490 0.0084 0.3213 0.0129 0.3859 0.0160 0.3858 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Sample restricted to rural households. Dependent variable is the log of total monetary consumption per capita. Individual- level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as region fixed effects. Source: EHCVM and World Bank estimates. 5.3. Differences by household head’s income-generating strategies Even though welfare losses appear to be concentrated among rural, agricultural households, having household heads working outside of agriculture does not appear to guarantee avoiding the effects of the lean season. To verify this, we split the sample according to the main occupation of the household head and re-estimate Equation 1 for each sub-sample with the log of real per capita consumption as the dependent variable. While the coefficient on the lean season dummy variable is larger for households where the head primarily engages in agriculture or in livestock/fishing it is still signed negatively even when the household head is primarily engaged in industry/construction or in services/other sectors (Table 10). These effects in households where the head is engaged in industry/construction or in services/other sectors are only statistically significant at the 5 percent level in the specifications that include full controls, although we prefer these specifications here given the sample splits applied; the controls may help to address observed sample selection. These results may arise because, even if the household head engages in non-agricultural work, they may have secondary activities linked to agriculture, or other household members may engage in agriculture as their primary or secondary activity too. Using the household head’s occupation alone to determine who faces the effects of seasonality may therefore be misleading. 21 Table 10. Difference in total monetary consumption per capita between the lean season and non-lean season waves, split by the household head’s main sector of work Not working Agriculture Livestock/fishing Industry/construction Services/other No Full Full No Full No No controls No controls Full controls Full controls controls controls controls controls controls controls Lean season -0.0812 -0.0532** -0.1365*** -0.1429*** -0.1767** -0.1314*** -0.0787 -0.0698** -0.0713* -0.0568*** wave (0.0510) (0.0261) (0.0246) (0.0176) (0.0712) (0.0447) (0.0510) (0.0284) (0.0392) (0.0203) N 2720 2639 7848 7773 1114 1090 2023 2017 6485 6443 R-squared 0.0037 0.6208 0.0187 0.3612 0.0184 0.5938 0.0043 0.6019 0.0030 0.5724 Note: * p<0.10, ** p<0.05, *** p<0.01. Standard errors clustered at the ‘grappe’ level in parentheses. Sample comprises Burkina Faso, Niger, and Senegal. Dependent variable is the log of total monetary consumption per capita. Individual-level sampling weights applied. ‘Full controls’ specifications include controls for the gender, age, and education level of the household head, household size, the dependency ratio, household walls material, household roof material, household floor material, access to electricity, access to improved drinking water, and access to improved sanitation as well as an urban-rural dummy and region fixed effects. Source: EHCVM and World Bank estimates. Section 6. Conclusion and policy discussion This paper demonstrates the dramatic effects that seasonality has on Sahelian households’ welfare, especially in rural areas. Household consumption is approximately 10.5 percent lower in the lean season than the non-lean season, as captured by the two waves of the EHCVM sample. Since households across the entire consumption distribution are affected by these welfare losses, this leaves poverty – measured at each country’s national line – 13.7 percentage points higher in Burkina Faso, 6.6 percentage points higher in Niger, and 8.1 percentage points higher in Senegal in the lean season. Moreover, the types of consumption losses that households incur appear to be consistent with particularly extreme forms of deprivation, as the quantities of cereals, or key staple goods, that are consumed are reduced. In-keeping with these results, subjective poverty appears to be 4.4 percentage points higher in the lean season too. The large seasonal swings in welfare observed in the EHCVM data are unsurprising given the limited strategies that Sahelian households have to manage and cope with seasonality. Households’ income sources are concentrated in agriculture and raising livestock and are not especially well diversified. There is a shift towards work in non-farm household enterprises when the lean season arrives, but this option may not be available to all households, especially in rural areas. Additionally, financial penetration is low, so smoothing consumption through saving or borrowing may be difficult. Finally, social protection is rare compared with the extent of poverty and vulnerability, with only 10.5 percent of Sahelians living in a household that received food assistance during the previous 12 months, and 3.6 percent living in a household receiving cash transfers. The results in this paper have three clear policy implications; first, expanding the coverage of regular safety net programs – which aim to ensure that even the poorest and most vulnerable households reach a minimum level of consumption and can cover their basic needs – could help tackle seasonal deprivation. Extensive evidence demonstrates that such programs can improve resilience and reduce poverty (Bastagli, et al., 2016; Premand & Stoeffler, 2020). Yet in most Sahelian countries, regular safety nets’ coverage of poor households is extremely limited. In turn, considerable effort is dedicated to trying to track unpredictable shocks and estimate the number of households that will be food insecure when they hit. This includes cases where shocks compound seasonal variation. Every year, annual response plans from 22 both humanitarian agencies and governments are presented soon after the second cycle of the Harmonized Framework (or ‘Cadre Harmonisé’) in March. Yet some form of seasonal variation affects household welfare, even in relatively benign years, like the 2018/19 agricultural and pastoral cycle. Deprivation and food insecurity could therefore be better tackled by expanding coverage of regular cash transfers to poor and vulnerable households, especially in rural areas. Second, and relatedly, temporary social safety nets that are designed to provide relief to poor and vulnerable households during the lean season could be triggered earlier. The month-by-month results suggest that households face welfare losses as early as April each year, and that these losses get worse as the lean season draws on. Pre-empting the lean season welfare losses and responding earlier could have at least two key benefits. First, responding earlier would be more efficient as households would be able to buy food more cheaply; during the lean season itself, food prices tend to be higher (see Figure 9 and Figure 10. Seasonal variation in staple crop prices in Sahelian countries according to World Food Programme data 23 Panel A: Sorghum 350 300 Price (XOF/kilogram) 250 200 150 100 50 0 Apr-17 Apr-19 Apr-15 Apr-16 Apr-18 Apr-20 Oct-15 Oct-16 Oct-17 Oct-18 Oct-19 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Panel B: Millet 350 300 Price (XOF/kilogram) 250 200 150 100 50 0 Apr-17 Apr-19 Apr-15 Oct-15 Apr-16 Oct-16 Oct-17 Apr-18 Oct-18 Oct-19 Apr-20 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Burkina Faso Mali Niger Senegal Chad Note: All prices are retail prices. Prices shown are national averages. Source: World Food Programme Vulnerability Analysis and Mapping Food Security Analysis tool and World Bank estimates. in Annex 2). Second, responding earlier could reduce the overall number of households that need support. This could prevent major drops in food consumption, curtail the use of detrimental coping mechanisms, help households to plan their expenses, and even support investments in other income- generating activities to manage and cope with the effects of seasonality better. This echoes a growing literature pointing to the role of shock-responsive cash transfers to mitigate the impacts of climate-related shocks and the importance of the timing of the intervention (Gros, et al., 2019; Pople, Hill, Dercon, & Brunckhorst, 2021). 24 Third, the results emphasize the importance of integrating efforts to tackle chronic poverty, seasonality, and unpredictable shocks in a holistic way. This requires strengthening ASP systems to enable them to cover the chronic poor, reach households facing seasonal food insecurity, and expand during crises. This, in turn, hinges on the availability of regular resources, not just for extreme situations; strengthening delivery mechanisms; and developing adaptive social registries so that chronically-poor households as well as households in transitory need of assistance can be identified quickly. This paper also demonstrates the potential of cross-country microdata for improving our understanding of the effects of seasonality. By harmonizing both the survey instrument and the data collection schedule, it is possible to pool data from Burkina Faso, Mali, Niger, Senegal and – to some extent – Chad, improving statistical power but also allowing the results from one country to be benchmarked against the others. Efforts to collect such consistent data in other countries in the future could help guide social protection and other policies if the welfare effects of shocks and seasonality are to be overcome and chronic poverty is to be reduced. 25 Annex 1. Timing of data collection for the 2018/19 EHCVM Table 11. Timing of interviews in the EHCVM surveys 2018 2019 Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Burkina Faso 795 1,153 1,224 342 259 1,889 1,221 133 Mali 1 522 1,723 662 1 932 1,777 958 24 Niger 1,022 1,291 667 432 1,183 1,230 195 Senegal 358 1,100 1,398 710 917 1,541 1,067 62 Chad 340 1,421 1,311 634 688 1,258 1,379 461 Source: EHCVM and World Bank estimates. 26 Annex 2. Seasonality in the Sahel Rainfall varies dramatically across the year in Sahelian countries. The rainy season typically lasts from June to September, although rains may begin in April or May and continue until October or November (see Figure 7). The remainder of the year is extremely dry, underlining the importance of the rainy season for agricultural and pastoralist activities. Figure 7. Seasonal variation in rainfall and temperature in Sahelian countries Panel A: Rainfall 50 Rainfall over five-day period 40 30 (milimeters) 20 10 0 Apr-17 Apr-16 Apr-18 Apr-19 Apr-20 Oct-15 Jan-16 Oct-16 Jan-17 Oct-17 Jan-18 Oct-18 Jan-19 Oct-19 Jan-20 Jul-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Panel B: Temperature 40 Daily mean temperature (degrees 30 20 Celsius) 10 0 Apr-16 Apr-17 Apr-18 Apr-19 Apr-20 Oct-15 Oct-16 Oct-17 Oct-18 Oct-19 Jan-16 Jul-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-20 Jul-15 Jul-17 Jul-18 Jul-19 Burkina Faso Mali Niger Senegal Chad Note: Monthly averages taken for penta-daily rainfall and daily mean temperatures. Source: Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) data for rainfall, Climate Forecast System Reanalysis (CFSR) data for temperature, and World Bank estimates. The agricultural lean season corresponds to the rainy season. The harvest season for millet and sorghum, the main coarse grains on which Sahelian households rely, ordinarily centers around October and November of each year, coming directly after the rainy season (FAO, 1995; FEWS NET, 2018). This may be complemented by off-season cultivation of rice, maize, and other crops – primarily between January and March – which may be supplied with water through irrigation or undertaken in areas where flood waters recede (FEWS NET, 2017; FEWS NET, 2017). By the time the rains arrive again, agricultural households face a lean season in which they will not only have drawn down their main season and off-season harvest but also face the physical demands of planting their principal rainfed crops. 27 The pastoralist lean season typically arrives before the rainy season has begun. Pastoralists rely on the rainy season between June and September to replenish pasture and water points, and hence the condition of their animals. Animals’ condition influences milk production and the price for which they can be sold. The lean season for pastoralists is therefore in April to June each year, just before the rains arrive (FEWS NET, 2018). Figure 8, taken from FEWS NET, shows a more detailed representation of the seasonality for agriculture and pastoralist activities in Burkina Faso, Mali, Niger, Senegal, and Chad. Figure 8. Seasonality in agricultural and pastoralist activities in five Sahelian countries Panel A: Burkina Faso Panel B: Mali 28 Panel C: Niger Panel D: Senegal Panel E: Chad Source: FEWS NET. Prices for agricultural output and for livestock also vary with the seasons in the Sahel. For rainfed agricultural products – including millet and sorghum – prices drop as supply rises heading into the main harvest in October or November of each year (see Figure 9 and Figure 10. Seasonal variation in staple crop prices in Sahelian countries according to World Food Programme data 29 Panel A: Sorghum 350 300 Price (XOF/kilogram) 250 200 150 100 50 0 Apr-17 Apr-19 Apr-15 Apr-16 Apr-18 Apr-20 Oct-15 Oct-16 Oct-17 Oct-18 Oct-19 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Panel B: Millet 350 300 Price (XOF/kilogram) 250 200 150 100 50 0 Apr-17 Apr-19 Apr-15 Oct-15 Apr-16 Oct-16 Oct-17 Apr-18 Oct-18 Oct-19 Apr-20 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Burkina Faso Mali Niger Senegal Chad Note: All prices are retail prices. Prices shown are national averages. Source: World Food Programme Vulnerability Analysis and Mapping Food Security Analysis tool and World Bank estimates. ). As such, prices of agricultural products are at their highest during the agricultural lean season between June and September. By contrast, livestock prices tend to be lowest in the pastoralist lean season between April and June each year, as households are forced to sell animals at ‘distress prices’ in order to cope (FEWS NET, 2017). Prices therefore provide an early indicator of the seasonal stresses on consumption and welfare that Sahelian households may be facing. 30 Figure 9. Seasonal variation in staple crop prices in Sahelian countries according to Food and Agriculture Organization data Panel A: Sorghum 0.6 Price (USD/kilogram) 0.5 0.4 0.3 0.2 0.1 0 Apr-15 Apr-16 Apr-17 Apr-18 Apr-19 Oct-19 Apr-20 Jan-15 Oct-15 Jan-16 Oct-16 Jan-17 Oct-17 Jan-18 Oct-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Panel B: Millet 0.6 Price (USD/kilogram) 0.5 0.4 0.3 0.2 0.1 0 Apr-15 Apr-16 Apr-17 Apr-18 Apr-19 Apr-20 Oct-18 Oct-15 Oct-16 Oct-17 Oct-19 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Burkina Faso Mali Niger Senegal Chad Jul-19 Note: Prices taken from the capital city market for each country. Sorghum prices are wholesale for Burkina Faso, Mali, and Niger and retail for Senegal and Chad. Millet prices are wholesale for Burkina Faso and Mali and retail for Niger, Senegal, and Chad. Source: Food and Agriculture Organization Food Price Monitoring and Analysis Tool and World Bank estimates. 31 Figure 10. Seasonal variation in staple crop prices in Sahelian countries according to World Food Programme data Panel A: Sorghum 350 300 Price (XOF/kilogram) 250 200 150 100 50 0 Apr-17 Apr-19 Apr-15 Apr-16 Apr-18 Apr-20 Oct-15 Oct-16 Oct-17 Oct-18 Oct-19 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Panel B: Millet 350 300 Price (XOF/kilogram) 250 200 150 100 50 0 Apr-17 Apr-19 Apr-15 Oct-15 Apr-16 Oct-16 Oct-17 Apr-18 Oct-18 Oct-19 Apr-20 Jan-15 Jan-16 Jan-17 Jan-18 Jan-19 Jan-20 Jul-15 Jul-16 Jul-17 Jul-18 Jul-19 Burkina Faso Mali Niger Senegal Chad Note: All prices are retail prices. Prices shown are national averages. Source: World Food Programme Vulnerability Analysis and Mapping Food Security Analysis tool and World Bank estimates. The ‘consensual’ approach for measuring food insecurity used by the ‘Cadre Harmonisé’ also suggests large swings in seasonal swings in Sahelian countries. The Cadre Harmonisé is a framework that brings together food and nutrition security experts from a wide range of international agencies and non- governmental organizations (NGOs) to help synthesize a number of data sources, including food consumption surveys, nutrition surveys, the Household Economy Approach, or other information provided by agricultural surveys and market monitoring (Cadre Harmonisé, 2019). This framework is used 32 to determine the distribution of food assistance in Sahelian countries. As Figure 11 shows, the share of the population that is classified as food insecure by this methodology – specifically those in the ‘crisis’, ‘emergency’, or ‘disaster/famine’ phase of the Cadre Harmonisé – typically rises in the period from June to August, which corresponds directly with the agricultural lean season in Sahelian countries. Thus, seasonality is also captured by the tools that development practitioners use to monitor welfare and food security in the Sahel. Figure 11. Seasonal variation in food insecurity according to the Cadre Harmonisé 16 Share of the population that are food 14 12 insecure (percent) 10 8 6 4 2 0 Burkina Faso Mali Niger Senegal Chad Note: Food insecure corresponds to any populations in Phases 3 (‘Crisis’), 4 (‘Emergency’), and 5 (‘Disaster/Famine’) of the Cadre Harmonisé. Population numbers used to calculate share of the food insecure population taken from the Cadre Harmonisé itself. Source: Cadre Harmonisé and World Bank estimates. The 2018 rainy season, which would determine the extent of the 2019 lean season on which this paper focuses, was regarded as relatively benign, enabling high agricultural productivity across the Sahel. For many areas in Burkina Faso, Mali, Niger, and Chad, biomass production during the 2018 rainy season was well above the 1998-2018 average, although in the Western Sahel – including in Senegal – rainfall was lower, leaving biomass somewhat depleted (Action Contre La Faim, 2018; FEWS NET, 2019). The reduced scarcity of agricultural output across most Sahelian countries is corroborated by the relative stability or even slight decline in the prices of key staple crops, which at least partially emerges in Figure 9 above (RPCA, 2019). As such, this paper covers a period of relative ‘normality’, which enables the analysis to try and isolate the effects of seasonality per se, rather than the compounding effects of large covariate shocks. 33 Annex 3. Household and location controls Table 12. List of household and location controls The following variables are included in the specifications with full controls, unless they are knocked out by specified interaction terms. Variable Description Household head Dummy variable taking 1 if the household head is male, and 0 if the household gender head is female. Household head The age in years and its square are included. age Household head Dummy variables that take 1 if the household head has that level of education education and 0 otherwise are included at the following levels: primary, secondary, post- secondary. No/less than primary education is the excluded category. Household size Log of the number of people in the household. Dependency ratio Ratio of those aged less than 15 or more than 64 to those aged 15 or more and 64 or less. Walls Dummy variable taking 1 if the household has adequate housing material for the walls, as per the United Nations standards, and 0 otherwise. Roof Dummy variable taking 1 if the household has adequate housing material for the roof, as per the United Nations standards, and 0 otherwise. Floor Dummy variable taking 1 if the household has adequate housing material for the floor, as per the United Nations standards, and 0 otherwise. Access to Dummy variable taking 1 if the household has access to electricity and 0 electricity otherwise. Access to improved Dummy variable taking 1 if the household has access to improved drinking drinking water water, as per the United Nations standards, and 0 otherwise. Access to improved Dummy variable taking 1 if the household has access to improved sanitation, as sanitation per the United Nations standards, and 0 otherwise. Urban Dummy variable taking 1 if the household is in a ‘grappe’ classified administratively as urban, and 0 otherwise. Region Fixed effects specified at the Administrative 1 level. Captured through Stata’s areg command. 34 References Action Contre La Faim. (2018). 2018 Rain Mid-Season Biomass and Surface Water Over Sahel Report. Dakar: Action Contre La Faim. Récupéré sur https://sigsahel.info/wp- content/uploads/2018/09/ACF-2018-Mid-season-biomass-and-surface-over-Sahel-report.pdf Alderman, H., & Paxson, C. (1994). Do the Poor Insure? A Synthesis of the Literature on Risk and Consumption in Developing Countries. Dans E. Bacha, Economics in a Changing World (pp. 48- 78). London: International Economic Association Series, Palgrave Macmillan. doi:10.1007/978-1- 349-23458-5_3 Altonji, J., Elder, T., & Taber, C. (2005). Selection of Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools. Journal of Political Economy, 113(1), 151-184. doi:10.1086/426036 Barrett, C., Reardon, T., & Webb, P. (2001). Nonfarm income diversification and household livelihood strategies in rural Africa: concepts, dynamics, and policy implications. Food Policy, 26, 315-331. doi:10.1016/S0306-9192(01)00014-8 Bastagli, F., Hagen-Zanker, J., Harman, L., Barca, V., Sturge, G., & Schmidt, T. (2016). Cash Transfers: What Does the Evidence Say? London: Overseas Development Institute. Beegle, K., Coudouel, A., & Monsalve, E. (2018). Realizing the Full Potential of Social Safety Nets in Africa. Washington DC: World Bank. Récupéré sur https://openknowledge.worldbank.org/handle/10986/29789 Bodewig, C. (2019). Climate change in the Sahel: How can cash transfers help protect the poor? Washington DC: Brookings. Récupéré sur https://www.brookings.edu/blog/future- development/2019/12/04/climate-change-in-the-sahel-how-can-cash-transfers-help-protect- the-poor/ Bossuroy, T., Branders, N., Brunelin, S., Coudouel, A., Ferre, C., Gueye, A., . . . Thiam, D. (2022). Rapport d'évaluation d'impact du programme de bourses de sécurité familiales. Brunelin, S., Ouedraogo, A., & Tandon, S. (2020). Five Facts About Shocks in the Sahel. Washington DC: World Bank. Cadre Harmonisé. (2019). Manual Version 2.0: Identification and analysis of areas at risk and populations affected by food and nutrition insecurity. Cadre Harmonisé. Récupéré sur https://www.fsinplatform.org/sites/default/files/resources/files/CH20- Manual_En_finalWeb.pdf Dercon, S. (2002). Income Risk, Coping Strategies, and Safety Nets. World Bank Research Observer, 17(2), 141-116. doi:10.1093/wbro/17.2.141 Dercon, S., & Krishnan, P. (2000). Vulnerability, seasonality and poverty in Ethiopia. Journal of Development Studies, 36(6), 25-53. doi:10.1080/00220380008422653 Dercon, S., Hoddinott, J., & Woldehanna, T. (2005). Shocks and Consumption in 15 Ethiopian Villages, 1999-2004. Journal of African Economies, 14(4), 559-585. doi:10.1093/jae/eji022 35 Fafchamps, M., & Lund, S. (2003). Risk-sharing networks in rural Philippines. Journal of Development Economics, 71(2), 261-287. doi:10.1016/S0304-3878(03)00029-4 FAO. (1995). Sorghum and millets in human nutrition. Rome: Food and Agriculture Organization. Récupéré sur http://www.fao.org/3/T0818E/T0818E00.htm FEWS NET. (2017). Burkina Faso: Staple Food and Livestock Market. Famine Early Warning Systems Network. Récupéré sur https://fews.net/sites/default/files/documents/reports/FEWS%20NET%20BurkinaFaso%20MFR _final_20170929_0.pdf FEWS NET. (2017). Niger: Staple Food and Livestock Market. Famine Early Warning Systems Network. Récupéré sur https://fews.net/sites/default/files/documents/reports/FEWS%20NET%20Niger%20MFR_final_ 20170929.pdf FEWS NET. (2018). West Africa Special Report: Pasture deficients in the Sahel to continue until at least July 2018. Famine Early Warning System Network. Récupéré sur https://fews.net/sites/default/files/documents/reports/Sahel%20Special%20Report_20180326_ 1.pdf FEWS NET. (2019). West Africa Seasonal Monitor. Famine Early Warning Systems Network. Récupéré sur https://fews.net/sites/default/files/documents/reports/WEST_AFRICA_Seasonal_Monitor_Octo ber%202019%20.pdf Gros, C., Bailey, M., Schwager, S., Hassan, A., Zingg, R., Mamtaz Uddin, M., . . . Coughlan de Perez, E. (2019). Household-level effects of providing forecast-based cash in anticipation of extreme weather events: Quasi-experimental evidene from humanitarian interventions in the 2017 floods in Bangladesh. International Journal of Disaster Risk Reduction, 41. doi:10.1016/j.ijdrr.2019.101275 Harrower, S., & Hoddinott, J. (2005). Consumption Smoothing in the Zone Lacustre, Mali. Journal of African Economies, 14(4), 489-519. doi:10.1093/jae/eji007 Holm, A., Ejrnæs, M., & Karlson, K. (2015). Comparing linear probability model coefficients across groups. Quality & Quantity: International Journal of Methodology, 49(5), 1823-1834. doi:10.1007/s11135-014-0057-0 Hoogeveen, J. (2003). Evidence on Informal Insurance in Rural Zimbabwe. Journal of African Economies, 11(2), 249-278. doi:10.1093/jae/11.2.249 Kaminski, J., Christiaensen, L., & Gilbert, C. (2016). Seasonality in local food markets and consumption: evidence from Tanzania. Oxford Economic Papers, 68(3), 736-757. doi:10.1093/oep/gpw013 Kazianga, H., & Udry, C. (2006). Consumption smoothing? Livestock, insurance and drought in rural Burkina Faso. Journal of Development Economics, 79, 413-446. doi:10.1016/j.jdeveco.2006.01.011 Khandker, S. (2012). Seasonality of income and poverty in Bangladesh. Journal of Development Economics, 97, 244-256. doi:10.1016/j.jdeveco.2011.05.001 36 Mace, B. (1991). Full Insurance in the Presence of Aggregate Uncertainty. Journal of Political Economy, 99(5), 928-956. doi:10.1086/261784 MVAM. (2021, May 19). Understanding Outliers: Ramadan & Insufficient Food Consumption. Récupéré sur MVAMBLOG: https://mvam.org/2021/05/19/understanding-outliers-ramadan-insufficient- food-consumption/?utm_source=rss&utm_medium=rss&utm_campaign=understanding- outliers-ramadan-insufficient-food-consumption Oster, E. (2019). Unobservable Selection and Coefficient Stability: Theory and Evidence. Journal of Business and Economic Statistics, 37(2), 187-204. doi:10.1080/07350015.2016.1227711 Pitt, M., & Khandker, S. (2002). Credit Programmes for the Poor and Seasonality in Rural Bangladesh. Journal of Development Studies, 39(2), 1-24. doi:10.1080/00220380412331322731 Pople, A., Hill, R., Dercon, S., & Brunckhorst, B. (2021). Anticipatory cash transfers in climate disaster response. London: Center for Disaster Protection. Premand, P., & Stoeffler, Q. (2020). Do Cash Transfers Foster Resilience? Evidence from Rural Niger. Washington DC: World Bank. Ravallion, M., & Chaudhuri, S. (1997). Risk and Insurance in Village India: Comment. Econometrica, 65(1), 171-184. doi:10.2307/2171818 RPCA. (2019). Concertation technique du dispositif régional de prévention et de gestion des crises alimentaire (PREGEC). Bamako: Le Réseau de Prévention des Crises Alimentaires au Sahel et en Afrique de l’Ouest. Récupéré sur https://ec.europa.eu/newsroom/know4pol/document.cfm?doc_id=60646 Townsend. (1994). Risk and Insurance in Village India. Econometrica, 62(3), 539-591. doi:10.2307/2951659 Von Der Goltz, J., Karlen, R., Leite, P., Saidi, M., Sylla, K., & Taptue, A.-M. (2022). Emplois pour les Jeunes du Centre du Mali : Rapport d’Évaluation. Washington DC: World Bank. WFP. (2008). Food consumption analysis: Calculation and use of the food consumption score in food security analysis. Rome: World Food Programme. Récupéré sur https://documents.wfp.org/stellent/groups/public/documents/manual_guide_proced/wfp1972 16.pdf World Bank. (2005). Well-Being and Poverty in Ethiopia: The Role of Agriculture and Agency. Washington DC: World Bank. Récupéré sur https://documents.worldbank.org/en/publication/documents- reports/documentdetail/386211468032101729/ethiopia-well-being-and-poverty-in-ethiopia- the-role-of-agriculture-and-agency 37