# Cash Transfers, Negative Rainfall Shocks and Child Welfare in Ethiopia

, Volume Advance Article – Dec 7, 2021
26 pages

/lp/oxford-university-press/cash-transfers-negative-rainfall-shocks-and-child-welfare-in-ethiopia-uXUGvIawDv
Publisher
Oxford University Press
ISSN
0963-8024
eISSN
1464-3723
DOI
10.1093/jae/ejab029
Publisher site
See Article on Publisher Site

### Abstract

Abstract This study examines the role of cash transfers in mitigating the welfare impact of negative rainfall shocks on children in rural households. Household-level panel data, obtained from areas where Ethiopia’s Social Cash Transfer Pilot Programme operated, are merged with available climate data. The results from a two-way fixed effects model reveal that cash transfers significantly reduce the negative effect of drought on food consumption Z-score of children in beneficiary households. As the magnitude of drought increases, however, no difference in children’s FCS Z-score is found between beneficiary and non-beneficiary households. The study provides evidence for household food consumption-based coping strategies as a mechanism. As such, beneficiary households are able to avoid food consumption-destabilising coping strategies as long as the droughts they experience are not of high magnitude. The findings of this study offer policy-relevant insights into the extent to which cash transfers can buffer the adverse welfare impact of drought on children. 1. Introduction In recent years, there has been a growing policy emphasis towards cash transfers as an appropriate social policy instrument for poverty reduction and socio-economic development in sub-Saharan Africa (SSA).1 The objectives of most of the cash transfer programmes focus mainly on human well-being indicators such as food and nutrition security, health and educational status. As would be expected, the accompanied empirical evaluations of cash transfers have primarily focused on measuring these dimensions of programme impact. However, in the context of SSA where weather shocks are common phenomena2 , little has been done to incorporate objective metrics for climate variables to adequately measure the role that cash transfers play in mitigating the impacts of drought3 on household welfare. To fill the existing research gap, this study takes the case of a cash transfer programme in Ethiopia to examine the extent to which the transfers enhance rural households’ capacity to manage the impact of drought without sacrificing their children’s welfare. Specifically, the paper provides insight into the role of cash transfers in protecting food security of children, whose ages are 12 years or below, against the adverse effects of drought. The study also identifies variations in the effect of drought on rural households’ capacity to smooth consumption conditional on their beneficiary status. To this end, climate data were merged with household-level panel data collected from recipients and non-recipients of cash transfers. The study uses a two-way fixed effects (FEs) model to examine differences in the welfare effect of drought on children in rural households that are experiencing drought. The outcome variables of interest are food consumption score (FCS), a composite measure based on food frequency and the relative nutritional importance of different food groups consumed by children over a 7-day reference period, and coping strategy index (CSI), an indicator assessing the extent to which households use consumption-destabilising coping strategies. This study uses standardised outcome variables (i.e., FCS Z-score and CSI Z-score) to facilitate interpretation of effect sizes in terms of how many standard deviations apart households are, pivoting around the mean. The results show that the adverse effect of drought on FCS Z-score of children in beneficiary households is around one standard deviation lower compared with the effect that drought has on children in comparison households. Similarly, the effect of drought on CSI Z-score is on average 0.5 standard deviations lower on beneficiary households compared with that of comparison households. Hence, in the presence of negative rainfall shocks, cash transfers enhance rural households’ capacity to take coping measures without detrimental consequences on the welfare of children. However, children in the beneficiary households are no longer protected against the adverse welfare effects of drought as its magnitude increases. Therefore, cash transfers are not a panacea to mitigate the negative welfare consequences of severe droughts on children in the context of Ethiopia. There are few rigorous studies that examine the linkages between cash transfers and the capacity of households and communities to manage income shocks. At meso level, there is mixed evidence about the role of cash transfer programmes in mitigating the impact of climate risk-induced income shocks on crime and suicides (Baysan et al., 2019; Christian et al., 2019). At the household level, a conditional cash transfer programme in the Philippines induces its beneficiary households to allocate smaller portion of their income on temptation goods (such as alcohol and tobacco), compared with non-beneficiary households, when exposed to various idiosyncratic and covariate shocks (Flaminiano, 2021). In a closely related research works to this paper, studies by Hou (2010) in Mexico, Asfaw et al. (2017) in Zambia, Knippenberg & Hoodinott (2017) in Ethiopia and Premand & Stoeffler (2020) in Niger reveal the adverse impact of drought and significant mitigating role of cash transfers on households’ food security. These studies, however, only focus on the welfare effects of cash transfers at the household-level and hence implicitly assume that cash transfers are equally redistributed among household members based on need. However, such assumption of equal intrahousehold resource allocation has long been proven unrealistic in many previous empirical studies (Findlay & Wright, 1996; Haddad & Kanbur, 1990). In the case of unconditional cash transfers, the demographic characteristics of the recipients make a significant difference in the intrahousehold allocation of the transfers resulting in varying welfare outcomes on children (Leroy et al., 2009). In these circumstances, the presumption that all household members in different demographic groups have the same level of food security status may not hold. It has also been shown that the incidence of food insecurity between adults and children is often not equivalent (Kuku et al., 2011). Thus, equating food security status of children with the average food security of the household could not accurately capture the effect of cash transfers in households with vulnerable children. The remaining sections of the paper are organised as follows. Section 2 presents the conceptual framework that provides explanation on the pathways for the expected effects of cash transfers. Section 3 entails a description of the cash transfer programme. Section 4 describes the type and source of data, and the panel data model employed for the analyses. Section 5 presents results of the descriptive statistics and econometric model analyses. A discussion of the results is presented in Section 6. Section 7 provides concluding remarks. 2. Conceptual framework Cash transfers primarily aim to alleviate transitory income poverty over the short term and chronic and intergenerational poverty over the long term (Garcia & Moore, 2012). In the past decade, countries in SSA have experienced a rise in the number, scope and reach of cash transfer programmes that target poor and vulnerable households (Niño-Zarazúa et al., 2012). Cash transfer programmes are increasingly considered as having an appropriate role in protecting the well-being of vulnerable households by enabling them to maintain their spending on essential goods (e.g., food) and services (e.g., health and education) (Davies et al., 2008; Devereux & Sabates-wheeler, 2004; Johnson et al., 2013). One major explanation is the recognition that some other types of aid and humanitarian assistance were not effectively achieving their goals (Garcia & Moore, 2012). Moreover, cash transfers can foster investments in productive assets in addition to raising immediate consumption among the poorest households in rural Africa (Stoeffler et al., 2020). The recent growing interest in cash transfers is premised on the assertion that they could mitigate households’ liquidity constraints generated by imperfect credit and insurance markets (Dercon, 2011; Ravallion, 2009), thereby supporting expenditures on consumption (Alderman & Yemtsov, 2012; Barrientos, 2012; Tirivayi et al., 2016). As such, cash transfer schemes may play a role in protecting livelihood of rural households—previously considered credit-unworthy and uninsurable—against weather-related consumption losses. Therefore, by providing additional income to poor rural households, cash transfers are expected to capacitate them to avoid adverse consumption-based coping measures that have detrimental immediate and life-long impact on the most vulnerable household members. In this respect, cash transfers are envisaged as having ex-post protection to rural households, who have faced rainfall shocks, by providing an income floor to maintain or improve their consumption and consequently buffer their human capital against the effects of weather shocks. This study empirically tests the role of the transfers as a climate risk management instrument. 3. Description of the cash transfer programme In 2011, with the support from the United Nations Children’s Fund Ethiopia, the Social Cash Transfer Pilot Programme (SCTPP) was introduced by the Bureau of Labour and Social Affairs (BOLSA) of Tigray regional state in Ethiopia. A key feature of the SCTPP is the incorporation of the concept of vulnerability to poverty into the targeting criteria. The SCTPP aims to improve the quality of life for vulnerable children (OVC), the elderly and persons with disabilities (PWD) in households that do not benefit from existing major social protection interventions by other programmes in the area. Therefore, the first step was identifying communities with high prevalence of extreme poverty, food insecurity and adverse living circumstances (i.e., large number of OVC, female-headed households, PWD and the elderly) (Berhane et al., 2012). Accordingly, the SCTPP was implemented in tabias4 of Abi Adi, a small market town, and Hintalo Wajirat, a rural district, in Tigray region. Eligible households were identified through a community-based targeting process at tabia-level. A crucial component of the targeting process is identifying the most vulnerable households that lack economically active household members, own low levels of wealth proxied by land and livestock holdings and face severe liquidity constraints.5 A ranked list of eligible households, comprising all households that appeared to meet the targeting criteria, was then prepared from each kushet/ketena (Berhane et al., 2012). Households selected for inclusion in the programme constitute the population from which the ‘treatment’ sample is drawn. Households that met the targeting criteria but were not selected constitute the population from which the ‘comparison’ sample is drawn.6 Beneficiary households received 155 Birr (8.50 in 2012 USD values) on top of which extra grants were awarded based on the demographic features of the household. An additional 25 Birr for each child under the age of 16 years, 10 Birr if the child is enrolled in school for a maximum of four children, 40 Birr disabled child younger than 18 years, 50 Birr a disabled adult and 60 Birr dependent elderly persons over 60 years of age. Since January 2016, after the pilot programme phases out, the BOLSA of the region in collaboration with the Productive Safety Net Programme (PSNP) took over the administrative and financial responsibility to ensure the continuity of the social cash transfer in the rural district—Hintalo Wajirat. In the transition process, beneficiaries of the SCTPP are grouped under the permanent direct support component of the PSNP. Unlike the PSNP public work participants, who receive their transfers for 6 months, permanent direct support clients will receive 12 months of support. Their basic household grant for one or two adults was raised to 300 Birr per person. However, the additional payments mentioned above based on the demographic features of the household were discarded. 4. Methodology 4.1 Data source The evaluation of the SCTPP generated rich panel data set that extends between 2012 and 2014 after the implementation of the programme in 2011. The rationale to conduct an additional survey in 2016 is to adequately measure the welfare effect of an El Niño-induced rainfall shock that had occurred in 2015 and extended to early 2016. In order to avoid seasonal variation, this paper utilised the 2013 and the 2016 waves of panel data, which both were fielded in the exact same month—in November. By conducting the survey before the harvest of the 2013 and 2016 production seasons, the effect of the negative rainfall shock during the previous production year was captured. In 2013, the total number of households that had been surveyed from Hitalo Wajirat district was 2,159.7 For financial reasons, the whole sample from the previous wave could not be re-interviewed in 2016. A probability to size sampling was employed to randomly draw 223 and 202 households from the list of beneficiary and comparison households, respectively, in a total of 25 kushets (villages) in 6 tabias. 4.2 Main variables of interest 4.2.1 Outcomes variables FCS: the FCS was developed by the WFP’s Vulnerability Analysis and Mapping (VAM) unit in 2008 to capture both diet quantity and quality of household food consumption. The panel data of this study captured consumption of 14 food groups by children whose ages are 12 years or below within a reference period of 7 days. The FCS is a weighted sum of weekly consumption frequencies of nine separate food groups computed following a simple two-step procedure.8 First, the number of days that households consume a given food group is multiplied by the weight assigned to it. Then, the products computed in the first step are summed up to yield a single continuous variable measuring children’s FCS of a given household.9 To measure effect sizes as how far households are from the mean in terms of standard deviations, this study uses FCS Z-score of children.10 CSI: there are a number of responses by rural households to drought shocks that would certainly compromise their food consumption. We asked our sample households ‘how often’ they utilised among the five adverse consumption-based coping measures listed in Appendix Table A3 within a reference period of 30 days. The statistics on the responses obtained from rural households were aggregated to construct a CSI, which ranges between 5 and 20. Higher values of the index show that the household utilised more adverse consumption-based coping measures more frequently within that month. The CSI Z-scores are used to analyse household-level consumption destabilising behavioural responses to drought. 4.2.2 Key explanatory variables Beneficiary status is a binary time-invariant variable indicating whether a household is a recipient of the cash transfer programme (binary; 1= recipient of cash transfers, 0 otherwise). Climate variables: monthly rainfall data from January to December between 1987 and 2016 were obtained from National Oceanic and Atmospheric Administration’s Climate Prediction Centre (NOAA-CPC) with 0.1 degrees spatial resolution.11 This study generated the historical 30 years average annual rainfall and the standardized precipitation index (SPI) at the village level to capture variations within tabias. SPI values are based on gamma distribution to measure deviations in total precipitation from historical mean. Various indexes were calculated on time scales of 4 and 12 months using the SPI calculator software package (SPI SL 6) released by the National Drought Mitigation Center. The SPI values indicate the number of standard deviations that the observed precipitation deviates from the long-term mean. The study followed the thresholds set by Svoboda et al. (2002) for the classification of wet and dry periods. A binary rainfall shock variable was created by taking the SPI values that are less than −0.5 in the previous production year12 . By doing so, this study captured the occurrence of at least ‘abnormally dry’ weather conditions in the previous production year in the villages of the study area. \begin{align*} & Negative Rainfall shock = \begin{cases} \text{1}&\quad\text{if SPI} <-0.5\\ \text{0}&\quad\text{if SPI}\ge-0.5\\ \end{cases} \end{align*} For robustness check, the study also considers a cut-off of −0.8 to examine sensitivity of the moderation effect of cash transfers on food security of children when the magnitude of negative rainfall shock increases. Based on drought classification of Svoboda et al. (2002), the cut-off captures the presence of at least a moderate drought. This enables analysis of the extent to which the effect of cash transfers in mitigating the adverse effect of drought remains sizable and statistically significant despite an increase in the magnitude of drought. 4.3 Control variables Section 3 explains that SCTPP is a targeted intervention towards the most vulnerable households in the study area. The control variables, listed in Appendix Table A1, include household characteristics that were used as main criteria for targeting beneficiary households of the cash transfer programme. The time-variant control covariates include dependency ratio, landholding in hectare, livestock holdings in tropical livestock unit, number of crops grown, credit (binary; 1=borrowed during the production year, 0 otherwise), private transfer (binary; 1=received private transfer during the production year, 0 otherwise), walking minute distance of the homestead from village market and time FEs (n-1 survey year dummies). 4.4 Empirical estimation strategy The aim of this paper is to estimate the protective role of cash transfers against the effects of negative rainfall shocks on food security of children as \begin{align}& Y_{it}=\alpha T_{i} + \beta_1 D_{it-1} + \theta_1(T_{i}\times D_{it-1}) + \delta_1 R_{it} + \lambda_1 X_{it} + \tau_1 v_t + \mu_{i}+\varepsilon_{it}, \end{align}(1) where subscripts indicate variation over rural households (i = 1,..., N) and time (t = 1,..., T) and relate the observed outcome |Y_{it}$| (FCS Z-scores of children or CSI Z-scores) to a binary time-invariant variable for receiving the cash transfers (⁠|$T_i$|⁠), time-varying climate and household attributes (i.e., |$D_{it-1}$| is a binary variable for negative rainfall shock during the previous production year, |$R_{it}$| is long-term average annual rainfall and |$X_{it}$| is a vector of observable covariates that are listed in Appendix Table A1, survey year FEs (⁠|$v_t$|⁠) to factor out all time-varying factors that are similar to all households as long as these effects are linear and a disturbance term composed of time-invariant household-level unobserved characteristics (⁠|$\mu _{i}$|⁠) and normally and independently distributed time-varying error term (⁠|$\varepsilon _{it}$|⁠). In the absence of baseline data, FE models provide a means for controlling time-invariant unobserved characteristics (household FEs), which are allowed to be correlated with the covariates included in the model. The use of the FE model is justified whenever one is interested in consistently estimating the effects of only those variables that vary over time (Baltagi, 2005; Wooldridge, 2010). Since the main explanatory variable of interest is the interaction term between beneficiary status and negative rainfall shock, which is time-variant, FE model is more appropriate for the estimation.13 Equation 2 presents the two-way FE, within-household variation, estimation which differences out the effect of household specific fixed unobserved heterogeneity (⁠|$\mu _{i}$|⁠) and captures a linear time trend (⁠|$v_t|⁠), such that \begin{align}& \begin{aligned} (Y_{it}-\overline{Y}_{i})= & \beta_2 (D_{it-1} - \overline{D}_{i}) + \theta_2 ((T_{i}\times D_{it-1}) - (\overline{T \times D})_i) + \\ & \delta_2 (R_{it}-\overline{R}_{i}) + \lambda_2 (X_{it} - \overline{X}_{i}) + \tau_2 (v_t - \overline{v}) + (\varepsilon_{it} - \overline{\varepsilon}_{i}), \end{aligned} \end{align}(2) where |Y_{it}$|⁠, |${T_i}$|⁠, |$D_{it-1}$|⁠, |$R_{it}$|⁠, |$X_{it}$|⁠, |$v_t$| and |$\varepsilon _{it}$| are as described above. |$\overline {Y}_{i}$|⁠, |$\overline {D}_{i}$|⁠, |$(\overline {T \times D})_i$|⁠, |$\overline {R}_{i}$|⁠, |$\overline {X}_{i}$| and |$\overline {v}$| are cluster means for the time-varying variables. The within-household effect of drought on food security of children may vary depending on whether or not the households are receiving cash transfers. After including |$(T_{i}\times D_{it-1})$|—an interaction term between beneficiary status (⁠|$T_i$|⁠) and negative rainfall shock (⁠|$D_{it-1}$|⁠) variables—a Wald (statistical significance) test on |$\theta _2$| reveals whether the effect of cash transfers on food security of children significantly varies between households that have experienced negative rainfall shocks and those that have not. The standard errors are clustered at village (kushet)-level to take into account possible correlations between observations (households) within the same village.14 Heteroskedasticity-consistent standard errors are also computed by allowing correlations between households within 10 KM radius. 5. Results 5.1 Descriptive statistics Table 1 presents the mean differences in the outcome and explanatory variables between beneficiary and comparison households. The mean values of the outcome variables of interest do not significantly vary between the beneficiary and comparison households in 2013. This shows that the pre-period values of the FCS Z-score of children and CSI Z-score are on average similar for beneficiary and comparison households. Moreover, the pre-period averages show that the mean proportion of economically inactive members in beneficiary households is significantly higher than that of in the comparison households. Similarly, the proportion of beneficiary households that receive private transfers are on average larger than the comparison households. Table 1 Mean Differences of Variables in 2013 based on Beneficiary Status . 2013 . . . . Comparision HHs . Beneficiary HHs . Mean Diff. . p-value . Variables . N . Mean . N . Mean . . . FCS Z-score 127 −0.169 73 −0.146 −0.023 0.8803 CSI Z-score 193 0.222 213 0.268 −0.046 0.6987 Drought 202 0.876 223 0.888 −0.0120 0.7104 Long-term average rainfall 202 390.6 223 388.8 1.794 0.2337 Dependency ratio 197 0.446 217 0.602 −0.156 0.0000 Landholding 193 0.662 214 0.493 0.169 0.0000 TLU 193 1.286 214 0.357 0.929 0.0000 Number of crops grown 202 1.752 223 1.426 0.326 0.0001 Credit 201 0.438 222 0.297 0.141 0.0026 Private transfer 201 0.174 222 0.243 -0.069 0.0819 Distance to village market 202 61.10 223 59.79 1.310 0.7755 . 2013 . . . . Comparision HHs . Beneficiary HHs . Mean Diff. . p-value . Variables . N . Mean . N . Mean . . . FCS Z-score 127 −0.169 73 −0.146 −0.023 0.8803 CSI Z-score 193 0.222 213 0.268 −0.046 0.6987 Drought 202 0.876 223 0.888 −0.0120 0.7104 Long-term average rainfall 202 390.6 223 388.8 1.794 0.2337 Dependency ratio 197 0.446 217 0.602 −0.156 0.0000 Landholding 193 0.662 214 0.493 0.169 0.0000 TLU 193 1.286 214 0.357 0.929 0.0000 Number of crops grown 202 1.752 223 1.426 0.326 0.0001 Credit 201 0.438 222 0.297 0.141 0.0026 Private transfer 201 0.174 222 0.243 -0.069 0.0819 Distance to village market 202 61.10 223 59.79 1.310 0.7755 Note: The number of sample households in the comparison and beneficiary households are 202 and 223, respectively. However, the number of observations in the table deviates from those figures due to missing values in the variables. There are many households without children below the age of 12 years, and thus these households have missing FCS values. The F-statistic indicating the joint significance of household characteristics in predicting beneficiary status is F= 5.58, Prob > F = 0.0000. Open in new tab Table 1 Mean Differences of Variables in 2013 based on Beneficiary Status . 2013 . . . . Comparision HHs . Beneficiary HHs . Mean Diff. . p-value . Variables . N . Mean . N . Mean . . . FCS Z-score 127 −0.169 73 −0.146 −0.023 0.8803 CSI Z-score 193 0.222 213 0.268 −0.046 0.6987 Drought 202 0.876 223 0.888 −0.0120 0.7104 Long-term average rainfall 202 390.6 223 388.8 1.794 0.2337 Dependency ratio 197 0.446 217 0.602 −0.156 0.0000 Landholding 193 0.662 214 0.493 0.169 0.0000 TLU 193 1.286 214 0.357 0.929 0.0000 Number of crops grown 202 1.752 223 1.426 0.326 0.0001 Credit 201 0.438 222 0.297 0.141 0.0026 Private transfer 201 0.174 222 0.243 -0.069 0.0819 Distance to village market 202 61.10 223 59.79 1.310 0.7755 . 2013 . . . . Comparision HHs . Beneficiary HHs . Mean Diff. . p-value . Variables . N . Mean . N . Mean . . . FCS Z-score 127 −0.169 73 −0.146 −0.023 0.8803 CSI Z-score 193 0.222 213 0.268 −0.046 0.6987 Drought 202 0.876 223 0.888 −0.0120 0.7104 Long-term average rainfall 202 390.6 223 388.8 1.794 0.2337 Dependency ratio 197 0.446 217 0.602 −0.156 0.0000 Landholding 193 0.662 214 0.493 0.169 0.0000 TLU 193 1.286 214 0.357 0.929 0.0000 Number of crops grown 202 1.752 223 1.426 0.326 0.0001 Credit 201 0.438 222 0.297 0.141 0.0026 Private transfer 201 0.174 222 0.243 -0.069 0.0819 Distance to village market 202 61.10 223 59.79 1.310 0.7755 Note: The number of sample households in the comparison and beneficiary households are 202 and 223, respectively. However, the number of observations in the table deviates from those figures due to missing values in the variables. There are many households without children below the age of 12 years, and thus these households have missing FCS values. The F-statistic indicating the joint significance of household characteristics in predicting beneficiary status is F= 5.58, Prob > F = 0.0000. Open in new tab On the contrary, the average landholding of the beneficiary households is significantly lower than that of the comparison households. Related to landholding, the number of crops that beneficiary households grow is significantly lower than that of their comparison counterparts. Beneficiary households also have fewer livestock holdings than the comparison households. Since SCTPP intends to benefit households with high dependency ratio, low asset holdings and low access to liquidity sources, the mean differences in the variables show a clear pattern that signals the effectiveness of the SCTPP targeting. Berhane et al. (2014) also reported the absence of inclusion errors in the programme targeting. However, households do not significantly vary based on access to market and the climatic conditions (such as drought)15 they experience. Figure 1 depicts the association of FCS Z-score and CSI Z-score values with magnitude of negative rainfall shock conditional on households’ beneficiary status.16 Accordingly, for a lower magnitude of negative rainfall shock, the FCS Z-scores of children in the beneficiary households are visibly higher than that of in the comparison households. Figure 1 Open in new tabDownload slide Non-Parametric Relation between Drought and the Outcome Variables by Households’ Beneficiary Status in the Same Panel Figure 1 Open in new tabDownload slide Non-Parametric Relation between Drought and the Outcome Variables by Households’ Beneficiary Status in the Same Panel Therefore, children in cash transfers recipient households are no longer more advantaged than their counterparts in the comparison households in the presence of higher magnitudes of drought. Similarly, the initial gap in the CSI Z-score narrows as the magnitude of negative rainfall shock increases. As such, food consumption of children and ability to smooth consumption in the beneficiary households decline as the magnitude of drought increases. Appendix Figure A2 depicts that negative rainfall shocks are becoming frequent and severe in the study area. 5.2 Estimation results This section presents the effects of key explanatory variables of interest on the outcome variables—FCS Z-score of children and CSI Z-score. To control for biases that may arise from observable household-level heterogeneity, this study uses the conventional covariate adjustment method by including household characteristics in the final regression model. The covariate adjustment method has the same effect as matching methods (Elze et al., 2017) and thus as having balanced covariates at the baseline (Bruhn & Mckenzie, 2009). Moreover, the interaction terms between the treatment variable and those covariates with notable statistically different means between beneficiary and comparison households are also included in the analysis. This endeavour allows checking whether the heterogeneous welfare effect of drought based on beneficiary status is confounded by household characteristics that are not balanced in the pre-period. For brevity, the interpretation of the results is limited to the main effects of drought and the moderation effect of cash transfers. 5.2.1 The role of cash transfers in mitigating the effects of negative rainfall shocks Table 2 presents the results of two-way FE linear models. Accordingly, Columns 1 and 3 show the extent to which cash transfers mitigate the average within-household effects of negative rainfall shocks on children’s food security and households’ coping strategy. Columns 2 and 4 in the same table show the robustness of the effect of cash transfers in moderating the effects of negative rainfall shocks on the outcome variables of interest after adjusting for the effects of interaction terms between beneficiary status and household characteristics employed for targeting beneficiary households. Table 2 The Varying Effects of Drought on FCS Z-Score and CSI Z-Score based on Beneficiary Status . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought 0.8940 1.0373 −0.5211 −0.5402 (0.3390)*** (0.3367)*** (0.3034)* (0.2865)* [0.0801]* [0.0300]** [0.0891]* [0.0701]* {0.3533}** {0.3126}*** {0.1714}*** {0.1791}*** Drought −0.3011 −0.3735 0.3200 0.3029 (0.1497)** (0.1477)** (0.2083) (0.2135) [0.1121] [0.0581]* [0.1762] [0.1972] {0.1248}** {0.1223}*** {0.1696}* {0.1660}* Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −19.1063 −14.8851 −5.8893 −4.3596 (10.6201)* (10.7047) (10.9031) (12.0710) R-squared 0.1589 0.2073 0.1614 0.1900 Observations 370 370 370 370 . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought 0.8940 1.0373 −0.5211 −0.5402 (0.3390)*** (0.3367)*** (0.3034)* (0.2865)* [0.0801]* [0.0300]** [0.0891]* [0.0701]* {0.3533}** {0.3126}*** {0.1714}*** {0.1791}*** Drought −0.3011 −0.3735 0.3200 0.3029 (0.1497)** (0.1477)** (0.2083) (0.2135) [0.1121] [0.0581]* [0.1762] [0.1972] {0.1248}** {0.1223}*** {0.1696}* {0.1660}* Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −19.1063 −14.8851 −5.8893 −4.3596 (10.6201)* (10.7047) (10.9031) (12.0710) R-squared 0.1589 0.2073 0.1614 0.1900 Observations 370 370 370 370 Note: |$*** p<0.01, ** p<0.05, * p<0.1$|⁠. Robust standard errors: clustered at household level in parentheses and spatial correlation kernal cutoff 10 KM in curly brackets. Wild bootstrapped p-values clustered by village are in square brackets. The analysis is based on rural households that have children whose ages are 12 years or below. Accordingly, the total number of observations across the two time periods are 370 (Appendix Table A1 shows the number of households with children under 12 years of age in each time period). The outcome variable in Columns 1 and 2 is FCS Z-score. Whereas, in Columns 3 and 4 is CSI Z-score. Drought is a binary variable based on SPI calculated using a time-scale of 12 months and takes the value of 1 for SPI values less than −0.5 and 0 otherwise. The interaction terms are between beneficiary status and those covariates with statistically significant mean differences between beneficiary and comparison households (which are dependency ratio, landholding, TLU, number of crops grown and access to credit). Open in new tab Table 2 The Varying Effects of Drought on FCS Z-Score and CSI Z-Score based on Beneficiary Status . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought 0.8940 1.0373 −0.5211 −0.5402 (0.3390)*** (0.3367)*** (0.3034)* (0.2865)* [0.0801]* [0.0300]** [0.0891]* [0.0701]* {0.3533}** {0.3126}*** {0.1714}*** {0.1791}*** Drought −0.3011 −0.3735 0.3200 0.3029 (0.1497)** (0.1477)** (0.2083) (0.2135) [0.1121] [0.0581]* [0.1762] [0.1972] {0.1248}** {0.1223}*** {0.1696}* {0.1660}* Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −19.1063 −14.8851 −5.8893 −4.3596 (10.6201)* (10.7047) (10.9031) (12.0710) R-squared 0.1589 0.2073 0.1614 0.1900 Observations 370 370 370 370 . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought 0.8940 1.0373 −0.5211 −0.5402 (0.3390)*** (0.3367)*** (0.3034)* (0.2865)* [0.0801]* [0.0300]** [0.0891]* [0.0701]* {0.3533}** {0.3126}*** {0.1714}*** {0.1791}*** Drought −0.3011 −0.3735 0.3200 0.3029 (0.1497)** (0.1477)** (0.2083) (0.2135) [0.1121] [0.0581]* [0.1762] [0.1972] {0.1248}** {0.1223}*** {0.1696}* {0.1660}* Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −19.1063 −14.8851 −5.8893 −4.3596 (10.6201)* (10.7047) (10.9031) (12.0710) R-squared 0.1589 0.2073 0.1614 0.1900 Observations 370 370 370 370 Note: |$*** p<0.01, ** p<0.05, * p<0.1$|⁠. Robust standard errors: clustered at household level in parentheses and spatial correlation kernal cutoff 10 KM in curly brackets. Wild bootstrapped p-values clustered by village are in square brackets. The analysis is based on rural households that have children whose ages are 12 years or below. Accordingly, the total number of observations across the two time periods are 370 (Appendix Table A1 shows the number of households with children under 12 years of age in each time period). The outcome variable in Columns 1 and 2 is FCS Z-score. Whereas, in Columns 3 and 4 is CSI Z-score. Drought is a binary variable based on SPI calculated using a time-scale of 12 months and takes the value of 1 for SPI values less than −0.5 and 0 otherwise. The interaction terms are between beneficiary status and those covariates with statistically significant mean differences between beneficiary and comparison households (which are dependency ratio, landholding, TLU, number of crops grown and access to credit). Open in new tab Negative rainfall shocks have adverse effects on food security of children and household’s capacity to smooth consumption in the comparison group. On average, FCS Z-score of children in a given comparison rural household decreases by 0.3 to 0.4 standard deviations when experiencing negative rainfall shocks—SPI values less than −0.5 in the lag production year. Negative rainfall shocks also increase adoption of consumption-destabilising coping responses by rural households in the comparison group. The inclusion of the interaction term between beneficiary status and a binary negative rainfall shock variable allows examining whether the effects of drought on the outcome variables vary depending on whether or not households receive cash transfers.17 The results show that the adverse welfare effects of negative rainfall shocks are significantly blunted in the cash transfers recipient households. For a given rural household, the effect of drought on FCS Z-score is on average around one standard deviation lower on children in the beneficiary households compared with its effect on children in the comparison households. Put differently, in the presence of drought, children in a given beneficiary household realised on average around one standard deviation higher FCS than their counterparts in a comparison household. Similarly, drought has a significantly lower effect in aggravating adverse consumption-based coping measures in beneficiary households than in their comparison counterparts. The effect of drought on a given households’ CSI Z-score is on average around 0.5 standard deviations lower in beneficiary households than in comparison households. Appendix Table A4 Columns 1 and 3 present similar findings using the original (unstandardised) FCS and CSI variables in the estimation. Moreover, I also ran a random forest estimation to see if the algorithm18 picks SCTPP beneficiary status as an important predictor of children’s food security after taking into account the relative importance of other household-level demographic, socio-economic and environmental factors.19 As expected, Figure 2 shows that being a recipient of cash transfers is the most important predictor of children’s food security. The relative importance of negative rainfall shocks and long-term average annual rainfall are important variables in determining food security of children next to households’ livestock holdings. In general, variables with values close to one in the figure have high importance in predicting FCS of children. By contrast, variables with lower variable importance values have lower predictive power on households’ food consumption decisions that determine food security status of their children. Figure 2 Open in new tabDownload slide Variable Importance Plot Figure 2 Open in new tabDownload slide Variable Importance Plot 5.2.2 Cash transfers and higher magnitudes of negative rainfall shocks This section tests whether the important role of cash transfers in protecting the welfare of children remains robust for higher magnitudes of negative rainfall shocks. Table 3 presents the results from two-way FE linear model specification, which is the most favoured under this study. As negative rainfall shocks become severe, the resulting adverse impacts on children’s food security and households’ coping capacity in the comparison group increase in magnitude and remain statistically significant. For a given household in the comparison group, on average, FCS Z-score of children decreases by around 0.4 to 0.6 standard deviations when the household faces a higher magnitude of negative rainfall shocks—SPI values of less than −0.8. Experiencing stronger negative rainfall shocks also increases the CSI Z-score of a non-beneficiary household on average by around 0.4 standard deviations. Table 3 The Protective Role of Cash Transfers against Higher Magnitudes of Negative Rainfall Shocks . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought of high mag. 0.0056 0.2461 −0.2942 −0.2346 (0.3144) (0.3412) (0.2523) (0.2644) [0.9870] [0.6146] [0.2432] [0.3824] {0.2618} {0.3018} {0.1758}* {0.1879} Drought of high magnitude −0.4198 −0.5587 0.3868 0.3476 (0.2422)* (0.2404)** (0.2583) (0.2746) [0.0651]* [0.0230]** [0.0861]* [0.1191] {0.1545}*** {0.1565}*** {0.1675}** {0.1727}** Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant 2.4059 10.2953 −18.7166 −16.0701 (16.5250) (17.0069) (17.2910) (18.8202) R-squared 0.1137 0.1550 0.1534 0.1780 Observations 370 370 370 370 . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought of high mag. 0.0056 0.2461 −0.2942 −0.2346 (0.3144) (0.3412) (0.2523) (0.2644) [0.9870] [0.6146] [0.2432] [0.3824] {0.2618} {0.3018} {0.1758}* {0.1879} Drought of high magnitude −0.4198 −0.5587 0.3868 0.3476 (0.2422)* (0.2404)** (0.2583) (0.2746) [0.0651]* [0.0230]** [0.0861]* [0.1191] {0.1545}*** {0.1565}*** {0.1675}** {0.1727}** Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant 2.4059 10.2953 −18.7166 −16.0701 (16.5250) (17.0069) (17.2910) (18.8202) R-squared 0.1137 0.1550 0.1534 0.1780 Observations 370 370 370 370 Note: |$*** p<0.01, ** p<0.05, * p<0.1$|⁠. Robust standard errors: clustered at household level in parentheses and spatial correlation kernal cutoff 10 KM in curly brackets. Wild bootstrapped p-values clustered by village are in square brackets. The analysis is based on rural households that have children whose ages are 12 years or below. Accordingly, the total number of observations across the two time periods are 370 (Appendix Table A1 shows the number of households with children under 12 years of age in each time period). The outcome variable in Columns 1 and 2 is FCS Z-score, whereas in Columns 3 and 4, is CSI Z-score. Drought is a binary variable based on SPI calculated using a time-scale of 12 months and takes the value of 1 for SPI values less than −0.8 and 0 otherwise. The interaction terms are between beneficiary status and those covariates with statistically significant mean differences between beneficiary and comparison households (which are dependency ratio, landholding, TLU, number of crops grown and access to credit). Open in new tab Table 3 The Protective Role of Cash Transfers against Higher Magnitudes of Negative Rainfall Shocks . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought of high mag. 0.0056 0.2461 −0.2942 −0.2346 (0.3144) (0.3412) (0.2523) (0.2644) [0.9870] [0.6146] [0.2432] [0.3824] {0.2618} {0.3018} {0.1758}* {0.1879} Drought of high magnitude −0.4198 −0.5587 0.3868 0.3476 (0.2422)* (0.2404)** (0.2583) (0.2746) [0.0651]* [0.0230]** [0.0861]* [0.1191] {0.1545}*** {0.1565}*** {0.1675}** {0.1727}** Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant 2.4059 10.2953 −18.7166 −16.0701 (16.5250) (17.0069) (17.2910) (18.8202) R-squared 0.1137 0.1550 0.1534 0.1780 Observations 370 370 370 370 . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS Z-score . FCS Z-score . CSI Z-score . CSI Z-score . Beneficiary statusXDrought of high mag. 0.0056 0.2461 −0.2942 −0.2346 (0.3144) (0.3412) (0.2523) (0.2644) [0.9870] [0.6146] [0.2432] [0.3824] {0.2618} {0.3018} {0.1758}* {0.1879} Drought of high magnitude −0.4198 −0.5587 0.3868 0.3476 (0.2422)* (0.2404)** (0.2583) (0.2746) [0.0651]* [0.0230]** [0.0861]* [0.1191] {0.1545}*** {0.1565}*** {0.1675}** {0.1727}** Control variables Yes Yes Yes Yes Beneficiary statusXControl variables No Yes No Yes Household FEs Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant 2.4059 10.2953 −18.7166 −16.0701 (16.5250) (17.0069) (17.2910) (18.8202) R-squared 0.1137 0.1550 0.1534 0.1780 Observations 370 370 370 370 Note: |$*** p<0.01, ** p<0.05, * p<0.1$|⁠. Robust standard errors: clustered at household level in parentheses and spatial correlation kernal cutoff 10 KM in curly brackets. Wild bootstrapped p-values clustered by village are in square brackets. The analysis is based on rural households that have children whose ages are 12 years or below. Accordingly, the total number of observations across the two time periods are 370 (Appendix Table A1 shows the number of households with children under 12 years of age in each time period). The outcome variable in Columns 1 and 2 is FCS Z-score, whereas in Columns 3 and 4, is CSI Z-score. Drought is a binary variable based on SPI calculated using a time-scale of 12 months and takes the value of 1 for SPI values less than −0.8 and 0 otherwise. The interaction terms are between beneficiary status and those covariates with statistically significant mean differences between beneficiary and comparison households (which are dependency ratio, landholding, TLU, number of crops grown and access to credit). Open in new tab The adverse welfare effects of experiencing higher magnitude negative rainfall shocks are similar to both beneficiary and comparison households.20 This implies that cash transfers do not have a significant role in moderating the effect of high magnitude drought conditions on food security of children. When rural households experience stronger negative rainfall shocks, children in beneficiary households are at risk of low food consumption similar to their counterparts in the comparison households. Similarly, the effect of cash transfers in enabling rural households to avoid adverse coping responses during shock periods disappears as the magnitude of drought increases. Thus, the desirable effects of cash transfers in mitigating the negative consequences of drought on food security of children and households’ coping capacity dissipate with the increase in the magnitude of drought. To explore this further, a coefficient plot for the interaction coefficients with varying drought thresholds in Figure 3 shows that children in beneficiary households are no longer better off than their counterparts in comparison households as the magnitude of drought increases. As such, the desirable effect of cash transfers start to dissipate for SPI values of −0.7 and lower. Figure 3 Open in new tabDownload slide The Effects of Varying Drought Threshold. Note: The coefficient plots are based on 90% confidence interval and clustering the standard errors at kushet level. Figure 3 Open in new tabDownload slide The Effects of Varying Drought Threshold. Note: The coefficient plots are based on 90% confidence interval and clustering the standard errors at kushet level. 6. Discussion The presence of negative rainfall shocks poses serious consequences on the welfare of rural households (Asfaw et al., 2017; Gao & Mills, 2018; Knippenberg & Hoodinott, 2017). In the smallholder rainfed farming system, the direct welfare effects of negative rainfall shocks are related to loss of food and farm income (Brown & Funk, 2008; Saronga et al., 2016; Shumetie & Alemayehu Yismaw, 2018). The indirect effects mainly depend on the way rural households respond to manage rainfall shocks. Some of their responses to negative rainfall shocks could seriously destabilise food consumption of their children. The main consumption destabilising responses—adverse coping strategies that affect households’ food consumption—are described in Section 4.2.1. This study provides evidence suggesting that the desirable welfare effect of cash transfers on children is through improving households’ capacity to cope with the effects of negative rainfall shocks. Social protection policies such as cash transfers provide income and consumption floors to rural households and play a significant consumption smoothing role in the era of recurrent negative rainfall shocks. The transfers neutralise the detrimental welfare effects of negative rainfall shocks that children in the beneficiary households would have otherwise faced. Hence, social protection policies and programmes in general and cash transfers in particular can improve children’s food consumption by enhancing the capacity of rural households to access food and avoid adverse consumption-based coping responses in the presence of negative rainfall shocks. However, the effectiveness of cash transfers as a climate risk mitigating instrument depends on the magnitude of negative rainfall shocks. As droughts become more severe, cash transfers can no longer protect the welfare of children as households rely on coping measures that destabilise food consumption. Thus, cash transfers are not a panacea in mitigating the negative welfare effects of severe weather shocks on children in rural households in the context of Ethiopia. In this respect, recent developments in ‘adaptive social protection’ (see Davies et al., 2013, 2008, and the references therein) may unveil practical solutions to bring synergy between social assistance programmes and climate change adaptation and mitigation efforts. 7. Conclusion This paper evaluates the SCTPP in Tigray regional state of Ethiopia to offer insight into the extent to which cash transfers improve food security of children in rural households that face recurrent negative rainfall shocks. The study finds that the adverse welfare effect of negative rainfall shocks on children is smaller for SCTPP beneficiaries compared to non-beneficiary households. The observed protective role of cash transfers on food consumption of children plausibly emanates from the positive effect of the transfers on households’ coping capacity—responses to drought without negative consequences on household food consumption. Even though the programme is not explicitly designed as an instrument for weather risk management, the cash transfers may provide income and consumption floors for rural households to secure stable food access and avoid adverse coping responses. However, these desirable effects of cash transfers are achieved as long as negative rainfall shocks are mild. As the magnitude of drought increases, children in cash transfer recipient households are no longer in a better position in their food consumption than those in the non-recipients. Regardless of beneficiary status, rural households are unable to avoid adverse coping responses, which plausibly destabilise food consumption of their children, with the increase in the magnitude of drought. In conclusion, the role of cash transfers in protecting the welfare of children against the adverse effect of drought is contingent on the magnitude of drought. This highlights cash transfers are not a panacea to mitigate the welfare effects of negative rainfall shocks of high magnitude. Thus, there is an urgent need for cross-sectoral collaboration to establish complementarities between social protection and climate risk management policies so that rural households may effectively mitigate the detrimental welfare impacts of severe and frequent negative rainfall shocks. Future evaluation studies in this direction will reveal the effectiveness of such ‘adaptive social protection’, which integrates social protection and climate change adaptation and mitigation initiatives. Funding Netherlands Fellowship Programme (NFP). Supplementary material Supplementary material is available at Journal of African Economies online. Footnotes 1 For a comprehensive review and analysis of the existing cash transfer program in SSA, the readers are referred to the recent book published by FAO in 2016 titled From Evidence to Action: The Story of Cash Transfers and Impact Evaluation in Sub Saharan Africa, which is based on evaluations of cash transfer programmes undertaken in eight SSA countries. 2 The countries of the Sahel, the Greater Horn and Southern Africa are the most vulnerable to the recurrent and sometimes prolonged risk of negative rainfall shocks (Shiferaw et al., 2014). The devastating weather shocks triggered by El Nino climate events have affected over 50 million people in rural parts of East and Souther Africa (WFP, 2016). 3 Studies show that drought determines human capital (Haile et al., 2019; Shah & Steinberg, 2017; Thai & Falaris, 2014), individual lifetime earning capacities (Abiona, 2017; Maccini & Yang, 2009), violence (Hsiang et al., 2013, 2011) and suicides (Burke et al., 2018; Carleton, 2017). 4 Tabia (kebele) is an administrative unit in Ethiopia’s federal government structure and comprises three to four smaller administrative regions known as ketenes in Abi Adi and kushets (villages) in Hintalo-Wajirat. 5 The descriptive statistics for these household characteristics are presented in Appendix Table A1. 6 Both beneficiary and comparison households were drawn from the same tabias. 7 Berhane et al. (2012) provided a detailed explanation of the choices made for the selection of the survey sites for the programme implementation, the choice and content of the survey instruments and the sample size calculations. 8 To measure children’s FCS, similar food items among the 14 food groups were aggregated to construct the weighted nine food groups. Appendix Table A2 presents the food groups along with their varying weights. This method of quantifying household food consumption is presented in the 2008 WFP VAM technical guidance sheet (WFP, 2008). 9 Accordingly, the maximum value that FCS can take is 112. Instead of categorising children’s food security status by applying an arbitrary threshold, this study examines food consumption of children using FCS as a continuous outcome variable to avoid loss of information during aggregation. Moreover, a recent study by Marivoet et al. (2019) shows that arbitrary thresholds, including the one proposed by WFP, are marginally relevant to different regions in a country and less sensitive to local contexts. 10 Z-scores 11 In the NOAA-CPC website, the available rainfall data extends up to 1983. However, for the years between 1983 and 1986, rainfall data are missing for one or more of the months constituting the major rainy season in Ethiopia (from June to September). Ignoring the amount of rainfall in any of these months will bias the annual rainfall measure. Hence, we dropped those years from the computation. 12 The agricultural production year in Ethiopia extends between January and December and is divided in to two production seasons—belg and meher. The rainfall shock variables for the 2013 and 2016 survey waves were constructed based on the SPI values in a time scale of 12 months for the months January–December in 2012 and 2015, respectively. 13 Moreover, a causal inference based on the parameter estimate of drought variable (⁠|$\beta _1$|⁠) after using (pooled) cross-sectional variation estimators can only be made if drought is random to communities and households therein (Di Falco & Vieider, 2018). It is highly unlikely that all communities and households in the study area have the same probability of experiencing drought. 14 As the total number of villages for clustering is 25 (see Subsection 4.1), wild cluster bootstraps are appropriate for hypothesis testing when the number of clusters are relatively few (Cameron et al., 2008). 15 The F-statistic indicating the joint significance of household characteristics in predicting negative rainfall shocks is F= 1.14, Prob > F = 0.3365. Thus, negative rainfall shocks are independent of observable characteristics of households. 16 Appendix Figure A1 shows the same fact using separate panels to add the 95% confidence intervals. 17 The coefficient of the interaction term captures the differential welfare effect of drought based on households’ beneficiary status—the difference in the effect of drought between beneficiary and comparison households. 18 The random forest algorithm is described in Appendix C. 19 I thank the anonymous reviewer for this insightful suggestion. 20 Appendix Table A4 Columns 2 and 4 present similar findings using the original (unstandardised) FCS and CSI as the outcome variables. References Abiona O. ( 2017 ) ‘ Adverse Effects of Early Life Extreme Precipitation Shocks on Short-Term Health and Adulthood Welfare Outcomes ’, Review of Development Economics , 21 ( 4 ): 1229 – 54 . Google Scholar Crossref Search ADS WorldCat Alderman H. , Yemtsov R. ( 2012 ) ‘ Productive Role of Safety Nets ’, Social Protection & Labour Discussion Paper No. 1203 . Asfaw S. , Carraro A., Davis B., Handa S., Seidenfeld D., Carraro A., Davis B., Handa S. ( 2017 ) ‘ Cash Transfer Programmes, Weather Shocks and Household Welfare: Evidence from a Randomised Experiment in Zambia ’, Journal of Development Effectiveness , 9 ( 4 ): 419 – 42 . Google Scholar Crossref Search ADS WorldCat Athey S. , Imbens G. W. ( 2019 ) ‘ Machine Learning Methods that Economists Should Know About ’, Annual Review of Economics , 11 : 685 – 725 . Google Scholar Crossref Search ADS WorldCat Baltagi ( 2005 ) Econometric Analysis of Panel Data , 3rd edn. Chichester, England : Wiley . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Barrientos A. ( 2012 ) ‘ Social Transfers and Growth: What Do We Know? What Do We Need to Find Out? ’ World Development , 40 ( 1 ): 11 – 20 . Google Scholar Crossref Search ADS WorldCat Baysan C. , Burke M., González F., Hsiang S., Miguel E. ( 2019 ) ‘ Non-Economic Factors in Violence: Evidence from Organized Crime, Suicides and Climate in Mexico ’, Journal of Economic Behavior and Organization , 168 : 434 – 52 . Google Scholar Crossref Search ADS WorldCat Berhane G. , Devereux S., Hoddinott J., Tegebu F. N., Roelen K., Schwab B. ( 2012 ) ‘ Evaluation of Social Cash Transfer Programme, Tigray Region, Ethiopia. Inception Report ’, technical report . Berhane G. , Devereux S., Hoddinott J., Tegebu F. N., Roelen K., Schwab B., Taffesse A. S. ( 2014 ) ‘ Evaluation of Social Cash Transfer Programme, Tigray Region, Ethiopia. Endline Report ’, technical report . Breiman L. ( 2001 ) ‘ Random Forests ’, Machine Learning , 45 : 5 – 32 . Google Scholar Crossref Search ADS WorldCat Brown M. E. , Funk C. C. ( 2008 ) ‘ Food security under climate change ’, Science , 319 : 580 – 1 . Google Scholar Crossref Search ADS PubMed WorldCat Bruhn M. , Mckenzie D. ( 2009 ) ‘ In Pursuit of Balance: Randomization in Practice in Development Field Experiments ’, American Economic Journal: Applied Economics , 1 ( 4 ): 200 – 32 . Google Scholar OpenURL Placeholder Text WorldCat Burke M. , González F., Baylis P., Heft-Neal S., Baysan C., Basu S., Hsiang S. ( 2018 ) ‘ Higher Temperatures Increase Suicide Rates in the United States and Mexico ’, Nature Climate Change , 8 : 723 – 9 . Google Scholar Crossref Search ADS WorldCat Carleton T. A. ( 2017 ) ‘ Crop-Damaging Temperatures Increase Suicide Rates in India ’, Proceedings of the National Academy of Science , 114 ( 33 ): 8746 – 51 . Google Scholar Crossref Search ADS WorldCat Christian C. , Hensel L., Roth C. ( 2019 ) ‘ Income Shocks and Suicides: Causal Evidence from Indonesia ’, The Review of Economics and Statistics , 101 ( 5 ): 905 – 20 . Google Scholar Crossref Search ADS WorldCat Cameron A. C. , Gelbach J. B., Miller D. L. ( 2008 ) ‘ Bootstrap-Based Improvements for Inference with Clustered Errors ’, Review of Economics and Statistics , 90 ( 3 ): 414 – 27 . Google Scholar Crossref Search ADS WorldCat Davies M. , Béné C., Arnall A., Tanner T., Newsham A., Coirolo C. ( 2013 ) ‘ Promoting Resilient Livelihoods through Adaptive Social Protection: Lessons from 124 Programmes in South Asia ’, Development Policy Review , 31 ( 1 ): 27 – 58 . Google Scholar Crossref Search ADS WorldCat Davies M. , Guenther B., Leavy J., Mitchell T., Tanner T. ( 2008 ) ‘ ‘Adaptive Social Protection’: Synergies for Poverty Reduction ’, IDS Bulletin , 39 ( 4 ): 105 – 12 . Google Scholar Crossref Search ADS WorldCat Dercon , S. ( 2011 ). ‘ Social Protection, Efficiency and Growth ’, Centre for the Study of African Economies Working Paper WPS/2011-17 . Devereux , S. , Sabates-wheeler R. ( 2004 ). ‘ Transformative Social Protection ’, Institute of Development Studies Working Paper, issue 232 . Di Falco S. , Vieider F. M. ( 2018 ) Shocks and Risk Preferences Revisited: Causal Inferences from Panel Data versus Cross-Sections . Washington, DC : Agricultural & Applied Economics Association Annual Meeting . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Elze M. C. , Gregson J., Baber U., Williamson E., Sartori S., Mehran R., Nichols M., Stone G. W., Pocock S. J. ( 2017 ) ‘ Comparison of Propensity Score Methods and Covariate Adjustment: Evaluation in 4 Cardiovascular Studies ’, Journal of the American College of Cardiology , 69 ( 3 ): 345 – 57 . Google Scholar Crossref Search ADS PubMed WorldCat Findlay J. , Wright R. E. ( 1996 ) ‘ Gender, Poverty and the Intra-Household Distribution of Resources ’, Review of Income and Wealth , 42 ( 3 ): 335 – 51 . Google Scholar Crossref Search ADS WorldCat Flaminiano J. P. ( 2021 ) ‘ Cash Transfers, Shocks, and Temptation Goods: Evidence from the Philippines ’, Southeast Asian Journal of Economics , 9 ( 2 ): 1 – 38 . Google Scholar OpenURL Placeholder Text WorldCat Gao J. , Mills B. F. ( 2018 ) ‘ Weather Shocks, Coping Strategies, and Consumption Dynamics in Rural Ethiopia ’, World Development , 101 : 268 – 83 . Google Scholar Crossref Search ADS WorldCat Garcia M. , Moore C. M. T. ( 2012 ) The Cash Dividend: The Rise of Cash Transfer Programs in Sub-Saharan Africa . Washington, DC : The World Bank . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Genuer R. , Poggi J.-M., Tuleau-malot C. ( 2010 ) ‘ Variable Selection Using Random Forests ’, Pattern Recognition Letters , 31 ( 14 ): 2225 – 36 . Google Scholar Crossref Search ADS WorldCat Haddad L. , Kanbur R. ( 1990 ) ‘ How Serious is the Neglect of Intra-Household Inequality? ’ Economic Journal , 100 ( 402 ): 866 – 81 . Google Scholar Crossref Search ADS WorldCat Haile K. K. , Tirivayi N., Nillesen E. ( 2019 ). ‘ Climate Shocks, Coping Responses and Gender Gap in Human Development ’, UNU-MERIT Working Paper Series #2019-052 . Hastie T. , Tibshirani R., Friedman J. ( 2008 ) The Elements of Statistical Learning. Data Mining, Inference, and Prediction , 2nd edn. New York : Springer Science & Business Media . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Hou X. ( 2010 ) ‘ Can Drought Increase Total Calorie Availability? The Impact of Drought on Food Consumption and the Mitigating Effects of a Conditional Cash Transfer Program ’, Economic Development and Cultural Change , 58 ( 4 ): 713 – 37 . Google Scholar Crossref Search ADS WorldCat Hsiang S. M. , Burke M., Miguel E. ( 2013 ) ‘ Quantifying the Influence of Climate on Human Conflict ’, Science , 341 ( 6151 ): 1235367 . Google Scholar Crossref Search ADS PubMed WorldCat Hsiang S. M. , Meng K. C., Cane M. A. ( 2011 ) ‘ Civil Conflicts are Associated with the Global Climate ’, Nature , 476 ( 7361 ): 438 – 41 . Google Scholar Crossref Search ADS PubMed WorldCat Johnson C. , Dulal H. B., Prowse M., Krishnamurthy K., Mitchell T. ( 2013 ) ‘ Social Protection and Climate Change: Emerging Issues for Research, Policy and Practice ’, Development Policy Review , 31 ( S2 ): 2 – 18 . Google Scholar OpenURL Placeholder Text WorldCat Knippenberg E. , Hoodinott J. ( 2017 ) ‘ Shocks, Social Protection, and Resilience: Evidence from Ethiopia ’, The Ethiopia Strategy Support Program (ESSP) Working Paper No. 109, International Food Policy Research Institute (IFPRI) . Kuku O. , Gundersen C., Garasky S. ( 2011 ) ‘ Differences in Food Insecurity between Adults and Children in Zimbabwe ’, Food Policy , 36 ( 2 ): 311 – 7 . Google Scholar Crossref Search ADS WorldCat Leroy J. L. , Ruel M., Verhofstadt E. ( 2009 ) ‘ The Impact of Conditional Cash Transfer Programmes on Child Nutrition: A Review of Evidence Using a Programme Theory Framework ’, Journal of Development Effectiveness , 1 ( 2 ): 103 – 29 . Google Scholar Crossref Search ADS WorldCat Maccini B. S. , Yang D. ( 2009 ) ‘ Under the Weather: Health, Schooling, and Economic Consequences of Early-Life Rainfall ’, American Economic Review , 99 ( 3 ): 1006 – 26 . Google Scholar Crossref Search ADS WorldCat Marivoet W. , Becquey E., van Campenhout B. ( 2019 ) ‘ How Well Does the Food Consumption Score Capture Diet Quantity, Quality and Adequacy across Regions in the Democratic Republic of the Congo (DRC)? ’ Food Security , 11 ( 5 ): 1029 – 49 . Google Scholar Crossref Search ADS WorldCat Niño-Zarazúa M. , Barrientos A., Hickey S., Hulme D. ( 2012 ) ‘ Social Protection in Sub-Saharan Africa: Getting the Politics Right ’, World Development , 40 ( 1 ): 163 – 76 . Google Scholar Crossref Search ADS WorldCat Premand P. , Stoeffler Q. ( 2020 ) ‘ Do Cash Transfers Foster Resilience? Evidence from Rural Niger ’, Policy Research Working Paper No. 9473 , World Bank . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Ravallion M. ( 2009 ) ‘ Bailing Out the World’s Poorest ’, Challenge , 52 ( 2 ): 55 – 80 . Google Scholar Crossref Search ADS WorldCat Saronga N. J. , Mosha I. H., Kessy A. T., Ezekiel M. J., Zizinga A., Kweka O., Onyango P., Kovats S. ( 2016 ) ‘ “I Eat Two Meals per Day” Impact of Climate Variability on Eating Habits among Households in Rufiji District, Tanzania: A Qualitative Study ’, Agriculture & Food Security , 5 ( 14 ): 1 – 7 . Google Scholar OpenURL Placeholder Text WorldCat Shah M. , Steinberg B. M. ( 2017 ) ‘ Drought of Opportunities: Contemporaneous and Long-Term Impacts of Rainfall Shocks on Human Capital ’, Journal of Political Economy , 125 ( 2 ): 527 – 61 . Google Scholar Crossref Search ADS WorldCat Shiferaw B. , Tesfaye K., Kassie M., Abate T., Prasanna B. M., Menkir A. ( 2014 ) ‘ Managing Vulnerability to Drought and Enhancing Livelihood Resilience in Sub-Saharan Africa: Technological, Institutional and Policy Options ’, Weather and Climate Extremes , 3 : 67 – 79 . Google Scholar Crossref Search ADS WorldCat Shumetie A. , Alemayehu Yismaw M. ( 2018 ) ‘ Effect of Climate Variability on Crop Income and Indigenous Adaptation Strategies of Households ’, International Journal of Climate Change Strategies and Management , 10 ( 4 ): 580 – 95 . Google Scholar Crossref Search ADS WorldCat Stoeffler Q. , Mills B., Premand P. ( 2020 ) ‘ Poor Households’ Productive Investments of Cash Transfers: Quasi-Experimental Evidence from Niger ’, Journal of African Economies , 29 ( 1 ): 63 – 89 . Google Scholar OpenURL Placeholder Text WorldCat Svoboda M. , Lecomte D., Hayes M., Heim R., Gleason K., Angel J., Rippey B., Tinker R., Palecki M., Stooksbury D., Miskus D., Stephens S. ( 2002 ) ‘ The Drought Monitor ’, American Meteorological Society , 83 ( 8 ): 1181 – 90 . Google Scholar Crossref Search ADS WorldCat Thai T. Q. , Falaris E. M. ( 2014 ) ‘ Child Schooling, Child Health, and Rainfall Shocks: Evidence from Rural Vietnam ’, Journal of Development Studies , 50 ( 7 ): 1025 – 37 . Google Scholar Crossref Search ADS WorldCat Tirivayi N. , Knowles M., Davis B. ( 2016 ) ‘ The Interaction between Social Protection and Agriculture: A Review of Evidence ’, Global Food Security , 10 : 52 – 62 . Google Scholar Crossref Search ADS WorldCat WFP ( 2008 ) ‘ Food Consumption Analysis: Calculation and Use of the Food Consumption Score in Food Security Analysis ’, report submitted to the Vulnerability Analysis and Mapping Branch (ODAV) . Google Scholar WFP ( 2016 ). ‘ WFP El Niño 2015–2016 Preparedness and Response. Situation Report #4 ’, technical report . Wooldridge J. M. ( 2010 ) Econometric Analysis of Cross Section and Panel Data , 2nd edn. Cambridge, MA : The MIT Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Appendix A. Appendix tables Table A1 Summary Statistics of Variables based on Survey Years . All years (pooled) . 2013 . 2016 . Variable . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Outcome variables FCS 370 34.74 10.87 10.50 70.50 200 33.00 11.35 10.50 64.50 170 36.80 9.914 17 70.50 CSI 831 9.455 3.247 5 20 406 10.25 3.834 5 20 425 8.692 2.322 5 20 Climate variables Drought 850 0.738 0.440 0 1 425 0.882 0.323 0 1 425 0.593 0.492 0 1 Long-term aver. rainfall 850 392.6 15.40 370.6 423.5 425 389.7 15.50 370.6 423.5 425 395.6 14.74 375.1 422.7 Control variables Dependency ratio 839 0.500 0.306 0 1 414 0.528 0.308 0 1 425 0.473 0.301 0 1 Landholding 832 0.582 0.383 0 2.250 407 0.573 0.357 0 2.125 425 0.591 0.406 0 2.250 TLU 832 0.781 1.332 0 9.100 407 0.797 1.364 0 9.100 425 0.765 1.301 0 7.090 No. of crops grown 850 1.646 0.943 0 6 425 1.581 0.865 0 5 425 1.711 1.011 0 6 Credit 848 0.283 0.451 0 1 423 0.364 0.482 0 1 425 0.202 0.402 0 1 Private transfer 848 0.123 0.328 0 1 423 0.210 0.408 0 1 425 0.0353 0.185 0 1 Distance to village mk’t 850 60.41 47.17 1 300 425 60.41 47.19 1 300 425 60.41 47.19 1 300 . All years (pooled) . 2013 . 2016 . Variable . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Outcome variables FCS 370 34.74 10.87 10.50 70.50 200 33.00 11.35 10.50 64.50 170 36.80 9.914 17 70.50 CSI 831 9.455 3.247 5 20 406 10.25 3.834 5 20 425 8.692 2.322 5 20 Climate variables Drought 850 0.738 0.440 0 1 425 0.882 0.323 0 1 425 0.593 0.492 0 1 Long-term aver. rainfall 850 392.6 15.40 370.6 423.5 425 389.7 15.50 370.6 423.5 425 395.6 14.74 375.1 422.7 Control variables Dependency ratio 839 0.500 0.306 0 1 414 0.528 0.308 0 1 425 0.473 0.301 0 1 Landholding 832 0.582 0.383 0 2.250 407 0.573 0.357 0 2.125 425 0.591 0.406 0 2.250 TLU 832 0.781 1.332 0 9.100 407 0.797 1.364 0 9.100 425 0.765 1.301 0 7.090 No. of crops grown 850 1.646 0.943 0 6 425 1.581 0.865 0 5 425 1.711 1.011 0 6 Credit 848 0.283 0.451 0 1 423 0.364 0.482 0 1 425 0.202 0.402 0 1 Private transfer 848 0.123 0.328 0 1 423 0.210 0.408 0 1 425 0.0353 0.185 0 1 Distance to village mk’t 850 60.41 47.17 1 300 425 60.41 47.19 1 300 425 60.41 47.19 1 300 Note: The number of observations for each variable is based on non-missing values. FCS has the most missing values since there are many households in the sample without children below the age of 12 years. Livestock holding is measured using tropical livestock unit (TLU) based on Jahnke (1982) conversion factors as Camel 1.0, horse 0.8, cattle and mule 0.7 each, donkey 0.5, pig 0.2, sheep and goat 0.1 each and chicken 0.01. Open in new tab Table A1 Summary Statistics of Variables based on Survey Years . All years (pooled) . 2013 . 2016 . Variable . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Outcome variables FCS 370 34.74 10.87 10.50 70.50 200 33.00 11.35 10.50 64.50 170 36.80 9.914 17 70.50 CSI 831 9.455 3.247 5 20 406 10.25 3.834 5 20 425 8.692 2.322 5 20 Climate variables Drought 850 0.738 0.440 0 1 425 0.882 0.323 0 1 425 0.593 0.492 0 1 Long-term aver. rainfall 850 392.6 15.40 370.6 423.5 425 389.7 15.50 370.6 423.5 425 395.6 14.74 375.1 422.7 Control variables Dependency ratio 839 0.500 0.306 0 1 414 0.528 0.308 0 1 425 0.473 0.301 0 1 Landholding 832 0.582 0.383 0 2.250 407 0.573 0.357 0 2.125 425 0.591 0.406 0 2.250 TLU 832 0.781 1.332 0 9.100 407 0.797 1.364 0 9.100 425 0.765 1.301 0 7.090 No. of crops grown 850 1.646 0.943 0 6 425 1.581 0.865 0 5 425 1.711 1.011 0 6 Credit 848 0.283 0.451 0 1 423 0.364 0.482 0 1 425 0.202 0.402 0 1 Private transfer 848 0.123 0.328 0 1 423 0.210 0.408 0 1 425 0.0353 0.185 0 1 Distance to village mk’t 850 60.41 47.17 1 300 425 60.41 47.19 1 300 425 60.41 47.19 1 300 . All years (pooled) . 2013 . 2016 . Variable . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Obs . Mean . SD . Min . Max . Outcome variables FCS 370 34.74 10.87 10.50 70.50 200 33.00 11.35 10.50 64.50 170 36.80 9.914 17 70.50 CSI 831 9.455 3.247 5 20 406 10.25 3.834 5 20 425 8.692 2.322 5 20 Climate variables Drought 850 0.738 0.440 0 1 425 0.882 0.323 0 1 425 0.593 0.492 0 1 Long-term aver. rainfall 850 392.6 15.40 370.6 423.5 425 389.7 15.50 370.6 423.5 425 395.6 14.74 375.1 422.7 Control variables Dependency ratio 839 0.500 0.306 0 1 414 0.528 0.308 0 1 425 0.473 0.301 0 1 Landholding 832 0.582 0.383 0 2.250 407 0.573 0.357 0 2.125 425 0.591 0.406 0 2.250 TLU 832 0.781 1.332 0 9.100 407 0.797 1.364 0 9.100 425 0.765 1.301 0 7.090 No. of crops grown 850 1.646 0.943 0 6 425 1.581 0.865 0 5 425 1.711 1.011 0 6 Credit 848 0.283 0.451 0 1 423 0.364 0.482 0 1 425 0.202 0.402 0 1 Private transfer 848 0.123 0.328 0 1 423 0.210 0.408 0 1 425 0.0353 0.185 0 1 Distance to village mk’t 850 60.41 47.17 1 300 425 60.41 47.19 1 300 425 60.41 47.19 1 300 Note: The number of observations for each variable is based on non-missing values. FCS has the most missing values since there are many households in the sample without children below the age of 12 years. Livestock holding is measured using tropical livestock unit (TLU) based on Jahnke (1982) conversion factors as Camel 1.0, horse 0.8, cattle and mule 0.7 each, donkey 0.5, pig 0.2, sheep and goat 0.1 each and chicken 0.01. Open in new tab Table A2 Food Consumption Groups for Computing FCS No. . Food items in the survey questionnaires . Food groups . Weight . 1 Injera; other foods made with cereals: bread, biscuits, cookies, macaroni; roots Main staples 2 and tubers: potatoes, carrots, beets, cassava 2 Legumes: horse beans, lentils, cow, field or chick peas Pulses 3 3 Leafy dark green vegetables: spinach, cabbage; other vegetables: onion, tomato Vegetables 1 4 Fruit: mango, banana, orange, pineapple Fruit 1 5 Meat: beef, mutton, lamb, goat, chicken; egg; fish: tuna, Nile perch Meat and fish 4 6 Dairy product: milk, yoghurt, cheese or other milk products Milk 4 7 Sugar: sugar, honey, sweets Sugar 0.5 8 Fats and oil: foods made with oils, fats, ghee, butter Oil 0.5 9 Coffee and/or tea Condiments 0 No. . Food items in the survey questionnaires . Food groups . Weight . 1 Injera; other foods made with cereals: bread, biscuits, cookies, macaroni; roots Main staples 2 and tubers: potatoes, carrots, beets, cassava 2 Legumes: horse beans, lentils, cow, field or chick peas Pulses 3 3 Leafy dark green vegetables: spinach, cabbage; other vegetables: onion, tomato Vegetables 1 4 Fruit: mango, banana, orange, pineapple Fruit 1 5 Meat: beef, mutton, lamb, goat, chicken; egg; fish: tuna, Nile perch Meat and fish 4 6 Dairy product: milk, yoghurt, cheese or other milk products Milk 4 7 Sugar: sugar, honey, sweets Sugar 0.5 8 Fats and oil: foods made with oils, fats, ghee, butter Oil 0.5 9 Coffee and/or tea Condiments 0 Source: Adapted from WFP (2008). Open in new tab Table A2 Food Consumption Groups for Computing FCS No. . Food items in the survey questionnaires . Food groups . Weight . 1 Injera; other foods made with cereals: bread, biscuits, cookies, macaroni; roots Main staples 2 and tubers: potatoes, carrots, beets, cassava 2 Legumes: horse beans, lentils, cow, field or chick peas Pulses 3 3 Leafy dark green vegetables: spinach, cabbage; other vegetables: onion, tomato Vegetables 1 4 Fruit: mango, banana, orange, pineapple Fruit 1 5 Meat: beef, mutton, lamb, goat, chicken; egg; fish: tuna, Nile perch Meat and fish 4 6 Dairy product: milk, yoghurt, cheese or other milk products Milk 4 7 Sugar: sugar, honey, sweets Sugar 0.5 8 Fats and oil: foods made with oils, fats, ghee, butter Oil 0.5 9 Coffee and/or tea Condiments 0 No. . Food items in the survey questionnaires . Food groups . Weight . 1 Injera; other foods made with cereals: bread, biscuits, cookies, macaroni; roots Main staples 2 and tubers: potatoes, carrots, beets, cassava 2 Legumes: horse beans, lentils, cow, field or chick peas Pulses 3 3 Leafy dark green vegetables: spinach, cabbage; other vegetables: onion, tomato Vegetables 1 4 Fruit: mango, banana, orange, pineapple Fruit 1 5 Meat: beef, mutton, lamb, goat, chicken; egg; fish: tuna, Nile perch Meat and fish 4 6 Dairy product: milk, yoghurt, cheese or other milk products Milk 4 7 Sugar: sugar, honey, sweets Sugar 0.5 8 Fats and oil: foods made with oils, fats, ghee, butter Oil 0.5 9 Coffee and/or tea Condiments 0 Source: Adapted from WFP (2008). Open in new tab Table A3 Adverse Consumption-based Coping Responses . In the past 30 days, how often did this happen... . 1. Never 2. Rarely (1--2 times in the past 30 days) 3. Sometimes (3--10 times in the past 30 days) 4. Often ( |$>10$| times in the past 30 days) . a) Did you or any household member have to EAT SOME FOODS THAT YOU NORMALLY REALLY DID NOT WANT TO EAT (WILD FOODS)? b) Did you or any household member EAT A SMALLER MEAL than you felt you needed because there was not enough food? c) Did you or any household member EAT FEWER MEALS IN A DAY because there was not enough food? d) Did you or any household member GO TO SLEEP AT NIGHT HUNGRY because there was not enough food? e) Did you or any household member GO A WHOLE DAY WITHOUT EATING because there was not enough food? . In the past 30 days, how often did this happen... . 1. Never 2. Rarely (1--2 times in the past 30 days) 3. Sometimes (3--10 times in the past 30 days) 4. Often ( |$>10$| times in the past 30 days) . a) Did you or any household member have to EAT SOME FOODS THAT YOU NORMALLY REALLY DID NOT WANT TO EAT (WILD FOODS)? b) Did you or any household member EAT A SMALLER MEAL than you felt you needed because there was not enough food? c) Did you or any household member EAT FEWER MEALS IN A DAY because there was not enough food? d) Did you or any household member GO TO SLEEP AT NIGHT HUNGRY because there was not enough food? e) Did you or any household member GO A WHOLE DAY WITHOUT EATING because there was not enough food? Source: Own survey questionnaire. Open in new tab Table A3 Adverse Consumption-based Coping Responses . In the past 30 days, how often did this happen... . 1. Never 2. Rarely (1--2 times in the past 30 days) 3. Sometimes (3--10 times in the past 30 days) 4. Often ( |$>10$| times in the past 30 days) . a) Did you or any household member have to EAT SOME FOODS THAT YOU NORMALLY REALLY DID NOT WANT TO EAT (WILD FOODS)? b) Did you or any household member EAT A SMALLER MEAL than you felt you needed because there was not enough food? c) Did you or any household member EAT FEWER MEALS IN A DAY because there was not enough food? d) Did you or any household member GO TO SLEEP AT NIGHT HUNGRY because there was not enough food? e) Did you or any household member GO A WHOLE DAY WITHOUT EATING because there was not enough food? . In the past 30 days, how often did this happen... . 1. Never 2. Rarely (1--2 times in the past 30 days) 3. Sometimes (3--10 times in the past 30 days) 4. Often ( |$>10$| times in the past 30 days) . a) Did you or any household member have to EAT SOME FOODS THAT YOU NORMALLY REALLY DID NOT WANT TO EAT (WILD FOODS)? b) Did you or any household member EAT A SMALLER MEAL than you felt you needed because there was not enough food? c) Did you or any household member EAT FEWER MEALS IN A DAY because there was not enough food? d) Did you or any household member GO TO SLEEP AT NIGHT HUNGRY because there was not enough food? e) Did you or any household member GO A WHOLE DAY WITHOUT EATING because there was not enough food? Source: Own survey questionnaire. Open in new tab Table A4 The Role of Cash Transfers in Protecting Welfare of Children and Households’ Coping Capacity . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS . FCS . CSI . CSI . Beneficiary statusXDrought 9.7141 0.0605 −1.6916 −0.9553 (3.6838)*** (3.4159) (0.9848)* (0.8190) [0.0801]* [0.9870] [0.0891]* [0.2432] {3.8391}** {2.8447} {0.5566}*** {0.5707}* Drought −3.2720 −4.5619 1.0389 1.2556 (1.6263)** (2.6313)* (0.6764) (0.8385) [0.1121] [0.0651]* [0.1762] [0.0861]* {1.3564}** {1.6791}*** {0.5506}* {0.5437}** Control variables Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −172.8700 60.8877 −9.6649 −51.3086 (115.4015) (179.5658) (35.3969) (56.1353) R-squared 0.1589 0.1137 0.1614 0.1534 Observations 370 370 370 370 . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS . FCS . CSI . CSI . Beneficiary statusXDrought 9.7141 0.0605 −1.6916 −0.9553 (3.6838)*** (3.4159) (0.9848)* (0.8190) [0.0801]* [0.9870] [0.0891]* [0.2432] {3.8391}** {2.8447} {0.5566}*** {0.5707}* Drought −3.2720 −4.5619 1.0389 1.2556 (1.6263)** (2.6313)* (0.6764) (0.8385) [0.1121] [0.0651]* [0.1762] [0.0861]* {1.3564}** {1.6791}*** {0.5506}* {0.5437}** Control variables Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −172.8700 60.8877 −9.6649 −51.3086 (115.4015) (179.5658) (35.3969) (56.1353) R-squared 0.1589 0.1137 0.1614 0.1534 Observations 370 370 370 370 Note: |$*** p<0.01, ** p<0.05, * p<0.1$|⁠. Robust standard errors: clustered at household level in parentheses and spatial correlation kernal cutoff 10 KM in curly brackets. Wild bootstrapped p-values clustered by village are in square brackets. The analysis is based on rural households that have children whose ages are 12 years or below. Accordingly, the total number of observations across the two time periods are 370 (Appendix Table A1 shows the number of households with children under 12 years of age in each time period). The outcome variable in Columns 1 and 2 is FCS of children, whereas in Columns 3 and 4, is CSI of households. Drought is a binary variable based on SPI calculated using a time-scale of 12 months. In Columns 1 and 3, it takes the value of 1 for SPI values less than −0.5, and 0 otherwise. In Columns 2 and 4, it takes the value of 1 for SPEI values less than or equal to −0.8, and 0 otherwise. Open in new tab Table A4 The Role of Cash Transfers in Protecting Welfare of Children and Households’ Coping Capacity . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS . FCS . CSI . CSI . Beneficiary statusXDrought 9.7141 0.0605 −1.6916 −0.9553 (3.6838)*** (3.4159) (0.9848)* (0.8190) [0.0801]* [0.9870] [0.0891]* [0.2432] {3.8391}** {2.8447} {0.5566}*** {0.5707}* Drought −3.2720 −4.5619 1.0389 1.2556 (1.6263)** (2.6313)* (0.6764) (0.8385) [0.1121] [0.0651]* [0.1762] [0.0861]* {1.3564}** {1.6791}*** {0.5506}* {0.5437}** Control variables Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −172.8700 60.8877 −9.6649 −51.3086 (115.4015) (179.5658) (35.3969) (56.1353) R-squared 0.1589 0.1137 0.1614 0.1534 Observations 370 370 370 370 . (1) . (2) . (3) . (4) . . FE . FE . FE . FE . Variables . FCS . FCS . CSI . CSI . Beneficiary statusXDrought 9.7141 0.0605 −1.6916 −0.9553 (3.6838)*** (3.4159) (0.9848)* (0.8190) [0.0801]* [0.9870] [0.0891]* [0.2432] {3.8391}** {2.8447} {0.5566}*** {0.5707}* Drought −3.2720 −4.5619 1.0389 1.2556 (1.6263)** (2.6313)* (0.6764) (0.8385) [0.1121] [0.0651]* [0.1762] [0.0861]* {1.3564}** {1.6791}*** {0.5506}* {0.5437}** Control variables Yes Yes Yes Yes Year dummy Yes Yes Yes Yes Constant −172.8700 60.8877 −9.6649 −51.3086 (115.4015) (179.5658) (35.3969) (56.1353) R-squared 0.1589 0.1137 0.1614 0.1534 Observations 370 370 370 370 Note: |$*** p<0.01, ** p<0.05, * p<0.1$|⁠. Robust standard errors: clustered at household level in parentheses and spatial correlation kernal cutoff 10 KM in curly brackets. Wild bootstrapped p-values clustered by village are in square brackets. The analysis is based on rural households that have children whose ages are 12 years or below. Accordingly, the total number of observations across the two time periods are 370 (Appendix Table A1 shows the number of households with children under 12 years of age in each time period). The outcome variable in Columns 1 and 2 is FCS of children, whereas in Columns 3 and 4, is CSI of households. Drought is a binary variable based on SPI calculated using a time-scale of 12 months. In Columns 1 and 3, it takes the value of 1 for SPI values less than −0.5, and 0 otherwise. In Columns 2 and 4, it takes the value of 1 for SPEI values less than or equal to −0.8, and 0 otherwise. Open in new tab Appendix B. Appendix Figures Figure A1 Open in new tabDownload slide Non-parametric relation between drought and the outcome variables by households' beneficiary status in the same panel Figure A1 Open in new tabDownload slide Non-parametric relation between drought and the outcome variables by households' beneficiary status in the same panel Figure A2 Open in new tabDownload slide (a)–(g) SPIs using a time scale of 4 (meher: June-September) and 12 (annual: January-December) months and averaged over the villages in each of the seven tabias of Hintalo Wajirat district. Figure A2 Open in new tabDownload slide (a)–(g) SPIs using a time scale of 4 (meher: June-September) and 12 (annual: January-December) months and averaged over the villages in each of the seven tabias of Hintalo Wajirat district. Appendix C. Random forest algorithm A random forest (RF) algorithm identifies the relative importance of variables in predicting FCS of children. RF method is a machine learning (Breiman, 2001) with the goal of determining the predictive power of a particular variable rather than reporting its point estimates and statistical significance levels (Athey & Imbens, 2019). In the method, many independent trees are learned from the same training data. The statistical framework by considering a learning set |$L = {(X_1, Y_1),..., (X_n, Y_n)}$| made of n independently and identically distributed (i.i.d) observations of a random vector (X, Y) (Genuer et al., 2010). Vector |$X = (X^1,..., X^p)$| contains |$p$| explanatory variables and Y is the outcome variable. The principle of RF is to combine many binary decision trees built using several bootstrap samples coming from the learning sample L and choosing randomly at each node a subset of explanatory variables X. Following Hastie et al. (2008), equation 1 shows that the RF algorithm aims to reduce the correlation between the trees (⁠|$\rho $|⁠) without increasing the variance (⁠|$\sigma ^2|⁠). \begin{align}& \rho\sigma^2+\frac{1-\rho}{B}\sigma^2 \end{align}(A.1) The |m$| explanatory variables, |$m \leq p$|⁠, are randomly selected as candidates for splitting. As such, reducing |$m$| will reduce the correlation between any pair of trees in the ensemble and hence reduce the variance of the average of B identically distributed, but not necessarily independent, random trees. After growing B trees, the RF regression to make a prediction at a new point |$x| is \begin{align}& \hat{f}_{rf}^B(x)=\frac{1}{B}\displaystyle\sum_{b=1}^{B} T(x;\Theta_b), \end{align}(A.2) where |\Theta _b$| in equation 2 characterises the |$b^{th}\$| RF tree in terms of split variables, cut-points at each node and terminal-node values. The variable importance is evaluated using the out-of-bag (oob) data, which are not used for constructing the current tree. © The Author(s) 2021. Published by Oxford University Press on behalf of the Centre for the Study of African Economies, all rights reserved. For Permissions, please email: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

### Journal

Journal of African EconomiesOxford University Press

Published: Dec 7, 2021

### References

Access the full text.