From safety to productive net: Unconditional Cash Transfer and agricultural outcomes in Cameroon

Transcription

1 From safety to productive net: Unconditional Cash Transfer and agricultural outcomes in Cameroon Diana Cheung*, Soazic Elise Wang Sonne ** and Quentin Stoeffler *** Draft, March 2017 (Please do not cite) Abstract This paper analyzes the short-term impact of a cash transfer intervention on productive activities in rural Cameroon. It addresses an important knowledge gap regarding the potential of cash transfers, in a very poor environment, to raise well-being in the long term by stimulating investments. Using baseline and endline data collected in January 2013 and August 2016 in the Far North region of Cameroon, we evaluate the impact of the transfers on agricultural production, land access and livestock. We also assess the spillover effects of the intervention on non-beneficiary households of the community (Soulédé-Roua) that has benefited from the program. To the best of our knowledge, this paper is the first to study productive spillover effects of cash transfers in Sub-Saharan Africa. In order to measure accurately the causal effects of the intervention, we compare treatment and control communities using quasi-experimental methods (Propensity Score Matching combined with Difference-in- Difference) and control for attrition and unbalanced characteristics at the baseline. We find that the cash transfer intervention manages to improve land access and value, agricultural production, main subsistence crops cultivation (corn, cassava, sorghum, and tomato) and livestock (cows, chickens) of beneficiary households. However, the transfers fail to help them mitigate agricultural shocks such as production losses or animal deaths. Importantly, we also find that the intervention has large spillover effects on non-beneficiary households, as these latter also experienced greater land access, agricultural production, sales and livestock. These findings stress the essential role of cash transfers to alleviate poverty, in particular against chronic food insecurity. Monetary transfers do not only improve social welfare (health, education, consumption) but also provide empowerment. They increase the capacity of households to support themselves in the future by enhancing income generating activities, such as agricultural production. The spillover effects on nonbeneficiaries also suggests a local economy impact, and have important policy implications in a context of targeted transfers in areas with very high chronic poverty rates. JEL Codes: O1, Q1, I3, D91 Keywords: Cash transfer, agricultural outcomes, Difference-in-Difference, Propensity Score Matching, Spillovers * Assistant professor, Poitiers University, diana.cheung@univ-poitiers.fr ** PhD Fellow, UNU-MERIT, wangsonne@merit.unu.edu *** Assistant professor, Istanbul Technical University, stoeffler@itu.edu.tr

2 1. Context and Motivation Unconditional and conditional cash transfer programs have increasingly become powerful tools for poverty reduction in developing countries over the past two decades (Gentilini et al., 2014). If lot of evidence has already been found on the likelihood of those safety net programs to improve health, education and consumption outcomes, 1 there is still little empirical evidence on whether such interventions could also enhance poorest households accumulation of agricultural productive assets (Gertler et al., 2012; Stoeffler et al., 2015). Yet, accumulation of productive assets and increase in agricultural productivity are key for rural households to move out of poverty traps (Carter and Barrett, 2006). However, whether or not cash transfers are best suited to support poor households asset accumulation is an open and debated question (Ghatak, 2015; Janzen et al., 2013). According to the FAO (2016), agriculture contributes on average to 15% of the total GDP of Sub-Saharan African countries. It also constitutes the main source of income in rural areas with the majority of men and women depending on agriculture for their livelihoods (FAO, 2015). Nonetheless, empirical evidence on the link between cash transfer programs and agricultural outcomes is still sparse in Sub-Saharan Africa, especially in Central Africa 2 where more than 80% of the labor force in rural areas rely on agriculture. 3 In Cameroon, the most populated country in Central Africa, agriculture is considered as the secular source of growth and jobs; holding a share of 22.3% of the GDP in 2014 and employing around 61.3% of the labor force. 4 The Cameroonian agriculture is dynamic, as it does not only contribute in curbing national food insecurity but also provides food to neighboring countries (Chad, the Central African Republic), especially to those with a weak production system as Gabon and Equatorial Guinea. However, this agricultural dynamism has not yet been translated into better social welfare of the population with 37.5% of households still living below the poverty line. 5 The level of poverty is even more acute in rural areas, especially in the Far North region with more than 74.3% of households being poor. 6 Nonetheless, the Far North with a share of 17.4 % of the Cameroon s overall population devote 23% of land to agriculture, holding over 35% of the country national livestock. Besides, this region is the main cotton 1 See e.g Baird, et al. 2014, Saavedra and García 2013, Sadoulet et al. 2004, Schultz 2004, Glewwe and Olinto 2004, Maluccio and Flores 2005 Gertler 2004, Attanasio et al Cameroon, Congo, Gabon, Chad and Central African Republic build together the Economic Community of Central African States (CEMAC). 3 Sub-Regional Platform of Farmers Organisations in Central Africa (PROPAC) in UN Data Statistics, millions of people out of a total population of 21.6 millions of inhabitants in The official poverty line in Cameroon is CFA (almost 701 USD) per year, implying 931 CFA (almost 2USD ) per day. 6 Fourth Cameroon Household Survey (ECAM IV). 1

3 producing area of the country, a crop which represents 22% of national agricultural exports (pushing Cameroon among the top five producers in Africa). Cotton is also contributing to social welfare and political stability for rural communities in the northern part of the country, recently plagued by Boko-Haram attacks. In order to fight against poverty and chronic food insecurity, the Government of Cameroon in partnership with the World Bank launched early in 2013, a cash transfer intervention in the Far North part of the country. Considering the importance of agriculture in that region specifically and for the country and the sub-region as a whole, we seek to understand whether positive income shocks like cash transfers to poorest households can alleviate their liquidity constraints and enhance their productive capacity. Using baseline and endline data from a two-year Unconditional Cash Transfer (UCT) intervention (from January 2013 to August 2016), we assess whether giving cash to the poorest households of the Far North part of Cameroon improves their agricultural outcomes. We focus mainly on agricultural production, land access and value, livestock, as well as agricultural shocks mitigation. In doing so, we broaden the literature on UCT impact evaluation which mainly studies the effect of cash transfers on human capital and poverty, to productive activities. We also seek to assess the effect of cash transfers in the local economy by measuring the spillover effects of cash transfers on the agricultural outcomes of non-beneficiary households. The remainder of the paper is structured as follows. Section 2 presents a review of the literature on the nexus between cash transfer and agricultural outcomes (inputs and outputs). Section 3 gives a detailed presentation of the design and implementation of the intervention. Section 4 describes the data and the estimation strategy leading to the main findings in Section 5. Section 6 concludes with key policy implications, discussions and directions for future research. 2. Cash Transfer and agricultural production: What does the evidence say? The literature on cash transfers tend to focus on the income effects of cash transfers, namely the effects of the transfers on health, education and consumption outcomes. However, cash transfers also have positive investment effects that are much less studied in the literature. According to Stoeffler et al. (2016), Alderman et al. (2013) and Barrientos (2012), regular monetary transfers to poorest households could also impact their productive investments by increasing income predictability and altering liquidity and credit constraints to perform optimal investments. The small existing literature on the effects of cash transfer on agriculture considers three core type of outcomes: i) productive assets (hoes, sickles, hammers, shovels, ploughs, etc.), ii) agricultural inputs (e.g. fertilizer) and iii) livestock ownership and value (goats, chicken, 2

4 sheep, fowl, cattle, poultry, etc.). The effect of cash transfers on these three types of outcomes is ambiguous. According to the 2016 report of the Overseas Development Institute (ODI), only two studies out of eight find a positive effect of cash transfers on agricultural productive assets with a statistically significant effect ranging from 3 to 32 percentage points increase (Covarrubias et al., 2012 in Malawi; AIR, 2014, Daidone et al., 2014b in Zambia). Six studies out of eight report a significant and positive effect of cash transfers on agricultural inputs' (purchase on seeds, fertilizers and pesticides); with a magnitude varying between 4 and 18 percentage points (Daidone et al., 2014a in Lesotho; Daidone et al., 2014b in Zambia; Handa et al., 2014; Karlan et al., 2014 in Ghana; Todd et al., 2010 in Mexico; Asfaw et al., 2014 in Kenya). And finally, 12 studies out of 17 find that cash transfers increase ownership of production animals (goats, chicken, sheep, fowl, cattle, poultry, etc. ownership) (Covarrubias et al., 2012 in Malawi, Daidone et al., 2014b and AIR, 2014 in Zambia; Merttens et al. and Blattman et al., 2015 in Uganda; Haushofer and Shapiro (2013) and Merttens et al., 2013 in Kenya; Gertler et al., 2012, Evans et al., 2014 and Todd et al., 2016 in Mexico; Stoeffler et al., 2015 in Niger). While the original seminar study was conducted on the Progresa program in Mexico, all other studies mentioned in the report were conducted in Sub-Saharan Africa. This may be due to the potential of cash transfers to stimulate important productivity gains in very poor, creditconstrained and risky environments of African rural areas. Among this already small literature, only a few studies investigate the impact of cash transfer on ultimate agricultural outcomes such as yields and productivity but also on agricultural spending or investments. These studies generally find a positive average impact on all of these outcomes (Gertler et al., 2012; Seidenfeld and Handa, 2011; Daidone et al., 2014; Davis et al., 2002; Karlan et al., 2014). Some of these studies also go further and study the existence of a gender bias on the effect of cash transfer on agricultural outcomes. In Kenya, a significant impact on livestock ownership (sheep, goats) is found only for female-headed households (Asfaw et al., 2012). This finding is confirmed in Malawi where beneficiaries' female-headed households accumulate more agricultural productive assets and livestock than male headed ones (Covarrubias et al., 2012). However, opposite results are shown in Tanzania (Evans et al., 2014) and Bolivia (Martinez, 2004) with men more likely to have higher livestock and agricultural inputs than their female counterparts. Overall there is not a clear consensus on whether cash transfer can improve all the different type of agricultural outcomes or not. Besides, the studies cited above tend to focus on Eastern and Southern African sub-regions. Today, few is known for Central African countries and some intel would be valuable to inform and back up the external validity of those findings. This paper intends to fill this gap in the literature. 3

5 3. The intervention: The Cameroon Safety Net Project With the objective of alleviating chronic poverty in Cameroon, the government launched an Unconditional Cash Transfer pilot program (UCT) in The intervention provides cash transfers to poor and "very" poor households in the far north part of the country, an area plagued by chronic food insecurity and since 2013, by terrorist attacks of the group Boko Haram. The main aim of the UCT program is to help poor households satisfy their short-term consumption needs and stimulate production and entrepreneurial activities to reduce poverty. Overall, while promoting investment in human capital, the program key objectives are also to: (i) increase the number of poor and vulnerable households with access to social safety nets, (ii) establish an efficient and effective targeting and management information system, (iii) improve behavior of beneficiary households in relation to health, nutrition and education and last but not least, (iv) increase their productive assets Description of the project The Social Safety Net Pilot Project enable 2,000 chronic and vulnerable poor households to benefit from 360,000 CFA Franc (about 752 USD) during 24 months (an average of 15,000 CFA Franc per month). This amount is equivalent to about 19% of the average monthly living expenditure of household living at the poverty line (Stoeffler and Nguetse- Tegoum, 2012). More specifically, each household received 12 payments between November 2013 and January 2016, including 10 payments of 20,000 CFA Franc (about 42 USD) and 2 payments of 80,000 CFA Franc (about 167 USD) on the 12th and 24th months). This combination of small and large regular payments is meant to stabilize the general consumption of beneficiaries over the period, but also to foster investments in income-generating activities in two ways: transfers address risk by increasing income predictability (small transfers), while also addressing the cash constraint and allowing relatively large lump-sum investments (large transfers). The beneficiary households were selected in rural areas, in the Far North community of Soulédé-Roua (1,500 households in 15 villages), as well as the urban areas of the Ndop 9 In order to estimate a causal impact, the RCT creates a counterfactual, a «twin» to the beneficiary in order to know what the beneficiary would have become if there was no program. We can evaluate the impact of the program by measuring the difference of the evolution of the living conditions between a beneficiary and his twin. Moreover, it is suitable that the twin does not see his living conditions changed indirectly by the transfers received by the beneficiaries or the causal estimation would be biased. We talk about no contamination of the control group. 4

6 town in the North-West (500 households in seven districts). Transfers were primarily given to women, being considered as better household goods' managers. Indeed, evidence in Africa and other regions revealed that giving money to women increases the likelihood that all members of the household will benefit from the transfer, especially the most vulnerable, such as children and the elderly (Quisumbing, 2003; Schmidt, 2012) Program implementation The selection of beneficiary households was made on the basis of geographical targeting, community targeting and statistical targeting: (i) Geographic targeting (ii) Choice of potentially beneficiary households through community-based targeting (CBT), (iii) statistical targeting with the Proxy Means Test (PMT). The choice of beneficiary regions is based on the poverty rate (proportion of the population and households living below the national monetary and/or non-monetary poverty line), in particular on poverty map rates based on socio-economic data from the Third General Population and Housing Census (RGPH3) of 2005 and the Third Cameroon Household Survey (ECAM3) in Thus, ten (10) departments in five regions (Adamaoua, East, North-West, Far-North, North) and the cities of Douala and Yaoundé were selected. The choice of beneficiary communities is based on three criteria: the poverty rate, the distribution of the community's population by village (districts) and physical accessibility (see a map of the intervention area in Figure 1, Annex 6). The choice of beneficiary villages is based on a list of criteria that define the institutional, socio-economic and cultural context of the communities (districts) predetermined by the Community Working Group ("Groupes de Travail Communal" GTC). According to the criteria adopted, the GTCs rank the villages or neighborhood of the communities, from the poorest to the least poor, and then select the poorest. The GTCs selected the 15 beneficiary villages of Soulédé-Roua in July The choice of potentially beneficiary households by the community has followed a rigorous process. It was made on the basis of a list of poverty criteria defined by the GTC. These criteria were then discussed in village forums (neighborhoods) bringing together all the inhabitants. Poverty criteria discussed include basic infrastructure (access to drinking water, road conditions, etc.), housing conditions, health, education, physical assets and economic activity in the village, as well as the population density, access to land and geographical access. Local Targeting Groups ("Groupes Locaux de Ciblage": GLCs) are responsible for selecting eligible households according to poverty criteria and this selection is verified and validated by the Local Citizen Monitoring Group (In French, "Groupe Local 5

7 de Contrôle Citoyen": GLCC). If any disagreement arise, the GLCC may file a complaint with the GTC. Potential beneficiary households were selected in Soulédé-Roua in October Statistical targeting uses the Proxy Means Test (PMT) approach. PMT targeting is based on statistical methods to construct a score that predicts household well-being (usually in terms of consumption). The PMT survey, which aims to collect the data necessary to establish the score for the classification of potentially beneficiary households, was carried out in Soulédé-Roua between November and December The 2000 beneficiary households were randomly selected among the designated beneficiary households by both PMT and community targeting. A combination of PMT and community targeting has been used for the Social Safety net Program intervention. 4. Impact evaluation of the Safety Net pilot program 4.1. Evaluation design Given the importance of the Safety Net Project (SNP) pilot program as a cornerstone to develop a coordinated safety net system in Cameroon, it was essential to lead a rigorous impact evaluation to assess the program s effectiveness in achieving its objectives. The Government of Cameroon with the support of the World Bank designed a randomized controlled trial (RCT) to study the impact of the unconditional cash transfers (UCT). RCT estimates compare the evolution (before and after receiving the transfers) of the outcomes of interest between treated and control groups. For instance, it compares the evolution of the average living conditions of the group of households who receive the transfers (treated group) to the evolution of the average living conditions of the group of households who do not receive the transfers (control group). As social programs tend also to benefit non-beneficiary households living in villages where the program takes place, using them as a control group could reduce a potential positive impact of the program and lead to the underestimation (or the absence) of impact. The comparison of households living in distinct villages allows to avoid estimation bias due to contamination of the control group. As a result, in this paper, we compare a group of beneficiary households to a group of households who do not benefit from transfers because they are living in a village where the program has not been implemented, but who are similar. Among the two beneficiary communities of the pilot program, the poorest (also the poorest in Cameroon), Soulédé-Roua, has been selected as the beneficiary group for the impact evaluation. The control group has been chosen to provide a perfect counterfactual, that is to 6

8 say insuring similarity between treatment and control and most importantly, further away geographically to avoid contamination of the control group by transfers given to beneficiaries. 9 Hina, which is located in the same department as Soulédé-Roua, Mayo-Tsanaga, has been selected as the control group. The two communities are indeed similar in terms of culture, behaviour and the level of poverty in Hina was not very different than the one in Soulédé- Roua when we set the RCT. Hina is also geographically far away from Soulédé-Roua. They do not share the same markets which allows to avoid contamination (see the map of the intervention area in Figure 1, Annex 6). We should also mention that two years for a program is usually considered enough to observe short-term effects but not necessary longer ones. We will thus focus on short-term agricultural outcomes as well as those that are more likely to improve living conditions in the long-run Design and implementation of the survey Survey data In order to measure the living conditions of households before and after the transfers, a baseline survey (BS) was conducted in November and December 2013, and an endline survey (ES) in July Beyond its contribution to the impact evaluation, the BS was also used to study the efficiency of the targeting method, in particular the efficiency of Community-Based Targeting (CBT) and statistical targeting with Proxy Means Testing (PMT) Sampling The BS and ES were conducted by the National Institute of Statistics (NIS) in 20 villages of the far-north region, among which 15 in Soulédé-Roua and 5 in Hina In order to estimate a causal impact, the RCT creates a counterfactual, a «twin» to the beneficiary in order to know what the beneficiary would have become if there was no program. We can evaluate the impact of the program by measuring the difference of the evolution of the living conditions between a beneficiary and his twin. Moreover, it is suitable that the twin does not see his living conditions changed indirectly by the transfers received by the beneficiaries or the causal estimation would be biased. We talk about no contamination of the control group. 10 The sampling frame used to draw the sample of households to interview comes from the PMT survey realised in 2012 by NIS in Soulédé-Roua and Hina. It is a census made with a PMT questionnaire. This database contains information on almost all the households in the 19 villages. 7

9 To enable the utilization of BS data for both the impact evaluation (IE) and the assessment of targeting efficiency, we consider 4 groups of households according to their eligibility status : (i) beneficiary households of the 15 villages of Soulédé-Roua, (ii) non-beneficiary households of the 15 villages of Soulédé-Roua who were selected by the community but not the PMT, (iii) non-beneficiary households of the 15 villages Soulédé-Roua who were not selected by the community, (iv) control households chosen among the poorest in the selected villages in Hina. The precision of the IE is also insured by statistical power calculations that indicate the optimal size of the sample. The calculations indicated that we had to interview 628 households per group, namely 2512 households in total. 11 ES data is only used for the IE. As a result, only 3 groups of households are considered: (i) beneficiary households of the 15 villages of Soulédé-Roua, (ii) non-beneficiary households of the 15 villages of Soulédé-Roua who were selected and non-selected by the community, (iii) control households chosen among the poorest in the selected villages in Hina. Thus, the second group of the ES gathers the second and third groups of the BS. The households to be interviewed in the ES are drawn among the households effectively interviewed in the BS, that is to say, 2350 households among which all the beneficiary households in Soulédé-Roua (611), all the control households in Hina (537), as well as a sample of 611 non-beneficiary households among the 1172 non-beneficiary households (targeted or not by the community) surveyed in the BS Data 2317 households 12 were surveyed in the BS and 1900 in the ES. Among the 2317 households interrogated in the BS, 610 are beneficiaries, 582 are non-beneficiaries targeted by the community, 563 are non-beneficiaries not targeted by the community, and 562 are control households. Among the 1900 households interrogated in the ES, 627 are beneficiaries, 329 are non-beneficiaries targeted by the community, 316 are nonbeneficiaries not targeted by the community, and 628 are control households (see table A below). Table 1. Distribution of households according to their eligibility status in the baseline and endline surveys 11 The sampling consists in drawing units of a population that allows to estimate the characteristics of the population. The larger the samples, the more precise the estimates of the population characteristics are. The estimates based on too small samples are less precise and the results less reliable (see annex 6 for more details on the calculation of the initial sample). 12 More precisely 2350 households were surveyed but only the data was exploitable only for 2317 ménages. 8

10 Group 1 : Beneficiary households in the 15 villages of Soulédé-Roua Group 2 : Nonbeneficiary households targeted in Soulédé-Roua Group 3 : Households not targeted by the community in Soulédé-Roua Group 4 : Control households chosen among the poorest in the selected villages of Hina Total Baseline survey Endline survey Source: 2013 baseline and 2016 endline survey The final sample used for the IE gathers all the households that are followed between 2013 and 2016, namely the households that are surveyed both in the BS and in the ES, that is to say 1683 households, among which, 580 beneficiaries, 583 non-beneficiaries and 520 control households (see table 2 below). Table 2. Distribution of households according to eligibility status in the impact evaluation sample Beneficiaries Non-beneficiaries Soulédé-Roua Non-beneficiaries Hina Total Baseline survey Endline survey Followed households Source: 2013 baseline and 2016 endline surveys 4.3. Data analysis Balancing-test: similarity between treated and control groups Although eligibility to the program has been randomly assigned across households, it is suitable to verify that the treated group, the beneficiaries of Soulédé-Roua, and the control group, the non-beneficiaries of Hina, are alike. Table 1 in annex 2 presents the results of the balance tests 13 on a wide set of characteristics (household, human capital, head of households, housing, socio-economic). Results suggest that the treated and the control groups are relatively but not completely similar. Columns 1 and 2 give the average value for the considered characteristics for the 13 For more information on balance tests see annex 6. 9

11 control group and the treated group, respectively, the third column indicate whether the two average values are statistically different or not. Globally, the two groups of households are identical in terms of education and health outcomes, whether we consider kids, the head of household or the whole household. They are also similar in terms of employment and their main activity, agriculture. A few differences, however, are worth to note about the household composition, entrepreneurship and revenues from agricultural products. Households are smaller in Hina and their age structure differs slightly. There are more single and Muslim household heads in Hina but less of them own an identity card. The PMT score of the control households in Hina is much smaller than the PMT score of beneficiaries in Soulédé-Roua, suggesting that control households are initially less poor than treated households. There are also differences in cattle ownership and consumption between the two groups at the BS. We account for these differences at the baseline survey by adding the corresponding characteristics as controls in the double difference equation Attrition analysis Among the 2317 households surveyed in the BS, 634 households have not been properly followed in the ES. In total, only 1683 questionnaires were exploitable. 14 It is therefore essential to study the consequences of this loss of households between the two surveys to ensure the precision of the program impacts' estimates. A consequent part of attrition, namely 27,4% is due to a reduction of non-beneficiary in Soulédé-Roua. School vacation and migration can explain this loss. If we consider the sample of households used for the IE, beneficiaries in Soulédé-Roua and control in Hina, among the 1172 households interviewed in BS, 1100 have been surveyed in ES. We lose 72 households between the two surveys, 42 in Hina and 30 in Soulédé-Roua, that is to say, 7.5% and 4.9% respectively. We will check whether the attrition is random or not in order to make sure that the lost of households do not differ significantly on average from the households that remained in ES. Table 2 in annex 2 provides the results of the comparison. Attrited households and the remaining households are statistically different in terms of poverty (attrited households are less poor), gender of the head of household (more female heads in attrited households), household size (attrited households are smaller), level of questionnaires were exploitable but 61 households saw their eligibility status change between the two surveys and have been excluded from the analysis. 10

12 education (more attrited households have achieved secondary education), and age structure within the household. We control for these characteristics in our IE estimations. 4.4 Intention to Treat (ITT) estimation Difference-in-Difference A relevant way to estimate the causal impact of transfers is to apply a double difference estimation method. By definition, the double difference allows to account for group fixed effects and times trends. Knowing that the ES has been conducted 32 months after the BS in November 2013, we estimate the effect of UCT using the temporal dimension of the data with the following equation: Y!" = β! + β! Post! + β! Treatment! + δ Treatment! Post! + γ X!" +!" 2 where i refers to the household. Treatment! is the indicator for the eligibility status of the household, Post! is the time variable indicating the period of observation (equal to 1 and 0 if the observations come from the endline and the baseline survey, respectively). X!" gathers the characteristics to take into account for which treated and control households are not similar in the BS due to factors we cannot control such as attrition for instance. The coefficient δ associated to the interaction term Treatment! Post! gives the average effect of the treatment (ATE). To account for potential correlated errors across villages, standard errors are clustered at the village level Difference-in-Difference and Propensity Score Matching Given the differences in the characteristics between the treated and the control households at the baseline underlined in section 4.3.2, the mere application of the double difference estimation is not appropriate in this study. This estimation method does not allow taking into account all the factors that may explain the trend differences between the two groups. It is thus wise to combine a double difference estimation with a matching method to account for those factors. The Propensity Score Matching (PSM) method builds a comparison group based on the probability to be treated conditionally to a set of observable characteristics (X). Let T be the dummy variable indicating whether an individual receives the transfers or not. The propensity score is given by the probability to participate in the program considering the observable characteristics X. The vector of variables X should not be affected by the probability to participate in the program and these variables have to be collected ideally before the implementation of the program. 11

13 In the SNP framework, the matching is done only based on the PMT score. Indeed, the treatment assignment is based on the PMT score collected in 2012, so before the intervention and this score takes into account poverty and living conditions aspects. The probability of treatment assignment which allow us to build the common support between the treated and the controls is estimated by the following equation : where X is the PMT score. P(X) = Pr(T = 1 X ) = α + βx! + ε! (3) Cash transfers program spillovers on the non-beneficiaries in Soulédé-Roua Like in most RCTs, spillovers of the treatment to the non-beneficiary households are to be considered. The control group of households in Hina was purposely chosen in the evaluation design to avoid any bias in the causal estimation due to the spillover effects that may occur in Soulédé-Roua. 15 Incidentally, field observations suggest that non-beneficiary households in Soulédé-Roua do benefit indirectly from the transfers through employment. It is thus appropriate to measure the spillovers effects in order to evaluate the extent of the benefits of the UCT pilot program, beyond the beneficiaries, at the village level. Following Clarke (2015), we estimate spillover effects by adding a dummy variable for the nontreated households living near the treated households, that is to say the non-beneficiaries living in Soulédé-Roua. Y!" = β! + β! Post! + δ! Treatment! + δ! close! + δ! Treatment! Post! + δ! close! Post! + γ X!" +!" 5 where δ! et δ! give, respectively, the average treatment effect and the effect of being close to the treatment ; X is the vector of control variables that include the characteristics for which the beneficiary households and the non-beneficiary households are different at baseline and those for which the attrition is not random. 15 The study by Kremer and Miguel (2004) which evaluates the impact of the distribution of deworming treatments to children in Kenyan schools provides an interesting illustration of spillovers. Worms are parasites that can be transmitted across individuals in contact with contaminated fecal matters. When a child takes deworming medicines, he is less infected by worms. The children living in the same environment are in turn less exposed to worms. As a result, in the Kenyan experience, the distribution of deworming medicines in some schools benefits the children of these schools (direct effect of treatment) but also to the children of neighboring schools (indirect effect). The positive impact of deworming by comparing the children in the treated schools to the children in the (close) non-treated schools would thus be smaller compare to its real impact. Hence, the necessity to do the comparison with schools that are located further away. 12

14 5. Results 5.1. Impact of the intervention on the beneficiaries We find that the intervention has increased the likelihood of beneficiary households to own a land by 14.9 percentage points. It has also increased by 30.3 percentage points the probability of households to hire labor force to work in the farm. Beneficiary households tend to demand more credits to buy fertilizers (+ 19.8%), which translates into an increase in the use of fertilizers (+ 464 CFA Franc). The sales of their agricultural production is also greater on average (+ 151 CFA Franc). On average, beneficiary households grow about 2.6 more agricultural products (see tables 3.1 and 3.2 in annex 3). This first set of results clearly reveals that by uplifting their liquidity constraints, poorest communities of the Far North part of Cameroon can improve their agricultural outcomes by increasing their access to land, improving their production and diversifying their agricultural portfolios. Therefore, the cash transfer intervention almost increases the likelihood of producing a diverse set of subsistence culture, contributing to reduce chronic food insecurity in the region 16, which is also the highest in the country. Specifically, the cash transfer program improves the likelihood of beneficiaries to produce key subsistence crops such as Tomato (by 1.2 percentage points); Taro (by 13.4 percentage points); Yam (by 7.3 percentage points); Potato (by 2.3 percentage points); Corn (by 41.4 percentage points); Rice (by 19.6 percentage points); and Cotton (by 14.4 percentage points). It is worth underlining that among these categories of staple foods, corn is the crop for which the program has the largest impact (+ 41 percentage points). This larger effect on corn production suggests that the training that household received to cultivate corn (along with cash transfers) can multiply the effect of the CT program. The combined effect of the training and cash transfers will be assessed in a forthcoming IDA project. The program also develops livestock-raising activities, in particular breeding (see table 3.4 in annex 3). Specifically, the probability of breeding as the main activity increases by almost 21 percentage points among the beneficiaries (column 1). The number of bovine animals increased by almost one unit (0.66 on average), while the probability of owning a cow increased by 24.6 percentage points (columns 2 and 3). The overall livestock value 16 The probability to suffer from insufficient amount of food during the 4 past weeks decreases significantly (by 53.4 percentage points); food insecurity, measured by the Fanta diversity scale (2007) also decreases subsequently by points (scale varying from 0 to 27, with a median at 6 at the baseline); the probability or begin malnourished is reduced by 28 percentage points (results from the report of the evaluation of the cash transfer program, World Bank, 2017). 13

15 measured by the Tropical Livestock Unit (TLU) has also increased over time as a result of monetary transfers. The current livestock stock value and the one over the past twelve months increase by almost 1.2 and 0.95 unit respectively. The difference between these two key measures translating the net effect on livestock increases by points (see table 4.4 columns 8 to 10). However, although households were also trained in chicken raising, the intervention slightly decrease the likelihood of beneficiary households to raise chicken by 13.3 percentage points. Knowing that the intervention increases the number of chickens raised by almost 3 units and that the sales of chicken improved by almost 2K FCFA Franc, it seems that the training convinced some households not to raise chicken or encourages them to switch to other livestock. Finally, we investigated whether the intervention has the ability to enhance the households' protection ability against agricultural shocks and risks. We fnd that the cash transfers have increased financial losses due to agricultural activity shocks by almost 27.6 thousand of CFA Franc (see table 3.5 in annex 3). This could potentially be explained by the increase in the area of land owned by beneficiaries' households induced by the intervention, which, a fortiori, would imply a greater loss in production shocks. Thus, households do not seem to have succeeded in neither protecting themselves against agricultural shocks, nor against livestock or any other household's losses or deaths. However, it is essential to stress that the CT may have reduced their aversion towards risk leading them to engage in (risky) productive activities that can help them enrich in the longer term Heterogeneity of the effects of the cash transfer intervention We also look at the potential heterogeneity of the cash transfer intervention by the gender 17 and the literacy level of the head of household (see Figures in annex 6). We found distinct effects of the CT program on land ownership across those characteristics but not on agricultural production. Illiterate households are able to own almost 1.83 more hectares of land compared to literate ones as the result of the intervention. We also found that women are 18.2 percentage points more likely to own land as compared to men after the intervention. These results are interesting as they show that the impact of the intervention is greater for the households that are more in needs. On the contrary, the results suggest no program heterogeneity across gender and literacy for all the different outcomes used to proxy agricultural production (value of agricultural production, number of agricultural products, recruitment of qualified labor force, etc). 17 While "De facto" Female Headed Households in Cameroon are very uncommon, there is a fair amount of "De Jure" female headed ones (almost 20.3% in our sample). These are mainly widows and divorced/ separated women. 14

16 However, when considering the production of specific crops, our results show that female headed households are 12.9 percentage points more likely to produce Cotton than male headed ones. They are also as compared to men, 24.9 percentage points more likely to produce corn. While no gender bias on rice production has been found, our estimates reveal that female headed households are 12 percentage points more likely to produce Sorghum than male. However, we found no statistical significant difference between literate and illiterate head of households for all those cereals. Further, while no gender bias has been found on the production of other subsistence culture, especially Tomato and Taro, we found that female headed households are 4.37 percentage points less likely to produce Potato but also 5.75 percentage points less likely to produce Yam. Our heterogeneity analysis on livestock show than even though Male headed households have a higher livestock value as compared to female, this difference is not statistically significant. However, we find that female headed households are 15 percentage points more likely to practice breeding as their main economic activity than male Spillover effects on non-beneficiaries In this paper, we also evaluate whether untreated households from the same community of recipients, have indirectly benefited from the intervention through spillover effects. As this was the case for other social welfare outcomes such as poverty and food insecurity, the dissemination effects of the cash transfer program are also very important on agricultural activities and outputs of non-beneficiary households. Indeed, the area of land owned by non-beneficiaries increases by almost 2.8 ha (see table 4.1 in annex 4, column 3). Besides, alike beneficiary households, the number of agricultural products grown by nonbeneficiary households increase by almost 2 units on average and the estimated value of agricultural production by almost 230 thousands of CFA Franc. The probability that the household sold all or part of its production also increases by 11.5 percentage points (see tables 4.2 in annex 4). The cash transfer program also supports the development of several crops: Tomato, Taro, Yam, Potato, Corn, Cotton, Sorghum and Millet (see table 4.3 in annex 4) among untreated households. Similarly to beneficiaries, Corn and Millet / Sorghum were the most grown as the result of the intervention (+21 percentage points). As this was the case for beneficiary households, the program also had a positive effect on livestock activity of non-beneficiary households (see table 4.4 in annex 4) and on its diversity (chickens, other poultry and cattle). The probability of breeding as a main activity increases by 17 points (column 1). The number of bovine animals increased by almost half a unit (0.54 on average), while the probability of owning a cow increased by 13 percentage points (columns 2 and 3). The current (actual) livestock value and the one over the past 12 months has also increased for non-beneficiary households (around 0.67). On the other hand, although the intervention decreases the likelihood of raising chickens among non- 15

17 beneficiaries, it increases their purchase of chicken the year prior to the intervention by almost 3 units (column 5), as well as their chickens' sale of about 1750 (almost 3.6 USD) and 1800 CFA Franc (almost 4 USD) respectively (columns 6 and 7). 6. Conclusion and policy implications The aim of this paper is to shed some light on the link between monetary unconditional transfers and agricultural outcomes in Sub-Saharan Africa. Applying Diff-in-Diff and Propensity Score Matching methods on data collected between January 2013 and August 2016 in the Far North region of Cameroon, we find that while the cash transfer intervention manages to improve land access and value, agricultural production, main subsistence crops cultivation (corn, cassava, sorghum, tomato) and livestock (cows) of beneficiaries households, it fails to help them mitigate agricultural shocks such as losses in their farm or animal deaths. These results are consistent with the prevailing literature. We shall, however, precise that the greater vulnerability to shocks can be explained by a greater exposure to shocks. We also find that the intervention has important spillover effects on non-beneficiary households of the same community (Soulédé-Roua) who have also greatly benefited from the transfers, especially through employment and an access to more diversified food. These findings stress the essential role of cash transfers to fight against poverty, in particular against chronic food insecurity. Monetary transfers do not only improve social welfare (health, education and consumption) but also the capacity of households to support themselves in the future by enhancing income generating activities, such as agricultural production. We also find that the intervention has larger effects for the most needy households with differences in the magnitude of the intervention impact according to the gender and literacy status of the household head. However, some caveats to our results should be emphasized. First, concerning the empirical strategy used, it is worth to mention that we make use of a Diff-in-Diff estimation coupled with a Kernel Propensity Score Matching under a Linear Probability Model (LPM). However, an Augmented Inverted Probability Weighting (AIPW), a Regression Adjustment or a combination of the two with the Diff-in-Diff could have well been used under a non-linear regression (Probit or Logit) more suitable for some of our outcomes that are dummy variables. A mediation analysis to empirically identify channels or mechanisms (mediators) could add to our research. In an upcoming version of the paper, we aim to perform a causal 16

18 mediation analysis to better inform why the intervention has improved agricultural outcomes of the poorest in the Far North part of Cameroon. More precisely, we plan to investigate the extent to which the positive effects on agricultural outcomes translate into better health and education outcomes of under five years old children. 17

19 References American Institutes for Research (2014) Zambia s Child Grant Program: 36-month impact report. Washington, DC: American Institutes for Research. Asfaw, S., Davis, B., Josh, D., Handa, S. and Winters, P. (2014) Cash transfer programme, productive activities and labour supply: evidence from a randomised experiment in Kenya Journal of Development Studies 50(8); Blattman, C., Green, E.P., Jamison, J., Lehmann, M. C. and Annan, J. (2015) The returns to cash and microenterprise support among the ultra-poor: a field experiment. Rochester, NY: Social Science Research Network (SSRN). Carter, M. R., & Barrett, C. B. (2006). The economics of poverty traps and persistent poverty: An asset-based approach. The Journal of Development Studies, 42(2), Clarke, D. (2015) Estimating Difference-in-Differences in the Presence of Spillovers: Theory and Application to Contraceptive Reforms in Latin America. University of Oxford. Unpublished. Covarrubias, K., Davis, B. and Winters, P. (2012) From protection to production: productive impacts of the Malawi Social Cash Transfer scheme, The Journal of Development Effectiveness 4(1): Daidone, S., Davis, B., Dewbre, J. and Covarrubias, K. (2014a) Lesotho s Child Grant Programme: 24-month impact report on productive activities and labour allocation. Lesotho country case study report. Rome: FAO. Daidone, S., Davis, B., Dewbre, J., González-Flores, M., Handa, S., Seidenfeld, D. and Tembo, G. (2014b) Zambia s Child Grant Programme: 24-month impact report on productive activities and labour allocation. Rome: Food and Agriculture Organization of the United Nations (FAO). Davis, B., Handa, S., Ruiz Arranz, M., Stampini, M. and Winters, P. (2002) Conditionality and the impact of programme design on household welfare: comparing two diverse cash transfer programmes in rural Mexico. Working Paper Rome: FAO. Evans, D.K., Hauslade, S., Kosec, K. and Reese, N. (2014) Community-based conditional cash transfers in Tanzania: results from a randomized trial. World Bank Study. Washington, DC: World Bank. FAO (2015) The State of Food and Agriculture Social protection and agriculture: breaking the cycle of rural poverty. Rome: Food and Agriculture Organisation (FAO). Gentilini, Ugo, Maddalena Honorati, and Ruslan Yemtsov. (2014). The state of social safety nets Washington, DC : World Bank Group. 18

20 Gertler, P.J., Martinez, S.W., Rubio-Codina, M. (2012) Investing Cash Transfers to Raise Long-Term Living Standards, American Economic Journal: Applied Economics 4(1): Ghatak, M. (2015). Theories of poverty traps and anti-poverty policies. The World Bank Economic Review, lhv021. Haushofer, J. and Shapiro, J. (2013) Household response to income changes: Evidence from an unconditional cash transfer program in Kenya. Princeton University (mimeo). Janzen, S. A., Carter, M. R., & Ikegami, M. (2013). Valuing asset insurance in the presence of poverty traps. UC-Davis working paper. Karlan, D., Osei, R., Osei-Akoto, I. and Udry, C. (2014) Agricultural decisions after relaxing credit and risk constraints, Quarterly Journal of Economics 129(2): Miguel, E. and Kremer, M. (2004), Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities. Econometrica, 72: ODI (2016) Cash transfers: what does the evidence say? A rigorous review of programme impact and of the role of design and implementation features, ODI report Stoeffler Quentin, Bradford Mills, Carlo del Ninno (2015) Reaching the Poor: Cash Transfer Program Targeting in Cameroon, 2015 Agricultural & Applied Economics Association and Western Agricultural Economics Association Annual Meeting Stoeffler Quentin, Bradford Mills and Patrick Premand (2015) Poor Households Productive Investments of Cash Transfers: Quasi-experimental evidence from Niger Todd, J.E., Winters, P. and Hertz, T. (2010) Conditional cash transfers and agricultural production: lessons from the Oportunidades experience in Mexico, Journal of Development Studies 46(1):

21 Annex 1 Intervention Design and Implementation Table 1. List of beneficiary regions and departments Région Department** Population Overall Poverty rate (%) Chronic Poverty rate (%) Number of chronic poor Population in the region Number of chronic poor in the region Djerem ,6 28, Faro et Deo ,7 32, Mayo Banyo ,9 60, Mbere ,8 37, Vina ,8 30, Boumba et Ngoko ,7 16, Haut Nyong ,7 29, Kadey ,3 56, Lom et Djerem ,3 38, Diamare ,6 50, Logone et Chari ,7 22, Mayo Danay ,6 61, Mayo Kani ,8 60, Mayo Sava ,2 64, Mayo Tsanaga ,0 68, Benoue ,3 45, Faro ,3 66, Mayo Louti ,7 53, Mayo Rey ,1 62, Boyo ,9 47, Bui ,7 26, Donga Mantung ,5 38, Menchum ,2 35, Mezam ,2 10, Momo ,4 29, Ngo Ketunjia ,5 20, Source: manual of implementation of the Social Safety Net Project funded by the Cameroonian Government (*) Excluded Yaoundé and Douala (**) Beneficiary departments are those highlighted. Adamaoua East Far North North Western north 20

22 Annex 2 Selection and Attrition Table 1. Balance test between Treatment and control (Hina) (1) Control (2) Treatment (3) Difference (4) Variables mean mean (1)-(2) N Household Characteristics - Composition Household size ** 1172 (0.586) (0.118) (0.534) Number of persons between (0.146) (0.065) (0.145) Number of persons between ** 1172 (0.328) (0.073) (0.301) Number of persons between * 1172 (0.185) (0.065) (0.176) Number of persons > (0.020) (0.031) (0.036) A household member can read * 1172 (0.031) (0.025) (0.037) - Education Highest level of Education: primary (0.057) (0.041) (0.065) Highest level of Education: secondary (0.027) (0.025) (0.034) Highest level of Education: secondary (0.060) (0.026) (0.060) Total education expenditures * 1172 (2.598) (1.561) (2785) A child doesn't go to school (between 6 and 13) (0.006) (0.007) (0.008) At least one household member is illiterate * 1172 (0.031) (0.025) (0.037) - Health At least one household member has a bad health (0.028) (0.037) (0.044) Composite health index (0.018) (0.029) (0.033) At least one household member suffer from malnutrition (0.000) (0.002) (0.002) At least one household member is stunting (0.004) (0.002) (0.004) - Household head characteristics Age of household head (0.431) (0.785) (0.869) Female household head (1=Yes, 0=No) (0.024) (0.021) (0.030) Muslim household head (1=Yes, 0=No) *** 1172 (0.160) (0.002) (0.143) Christian household head (1=Yes, 0=No) *** 1172 (0.064) (0.044) (0.072) 21

23 Animist household head (1=Yes, 0=No) *** 1172 (0.063) (0.036) (0.067) No religion (1=Oui, 0=Non) (0.081) (0.027) (0.077) Marital status separated (0.009) (0.010) (0.013) Marital status single *** 1172 (0.009) (0.005) (0.010) Marital status married polygamy (0.034) (0.024) (0.039) Household head has National ID *** 1158 (0.035) (0.006) (0.032) Household is registered (0.044) (0.028) (0.048) Housing characteristics The household has no water ** 1172 (0.054) (0.035) (0.060) The household has no electricity (0.005) (0.005) (0.007) The household has no toilet *** 1172 (0.049) (0.021) (0.049) The household head put waste in the nature (0.015) (0.003) (0.014) Socio-economic characteristics -Income activity Income from agricultural activities (in KCFA Franc) *** 1157 (0.020) (0.067) (0.069) - Employment Worked as an employee in the past 12 months (0.009) (0.004) (0.009) At least one household member is unemployed Enterprise Owns a micro-enterprise *** 1172 (0.045) (0.025) (0.047) Equipment value of the micro-enterprise (in CFA F) ** 1172 (10.625) (0.420) (9.467) - Land access The household owns land (0.071) (0.055) (0.083) Total estimated value of land ( ) (80.675) ( ) No household member works in the land (0.006) (0.006) (0.008) The household raises animal (0.051) (0.036) (0.058) *** p<0.01, ** p<0.05, * p<0.1 Source: 2013 Baseline Survey 22

24 Table 2: Test of Attrition 23 (1) Non attrition (Mean) (2) Attrition (3) Difference non attritionattrition (Mean) N Household Characteristics - Composition Household size *** 1172 (0.352) (0.575) (0.719) Number of persons between *** 1172 (0.076) (0.147) (0.185) Number of persons between *** 1172 (0.204) (0.232) (0.286) Number of persons between *** 1172 (0.113) (0.211) (0.262) Number of persons > (0.016) (0.126) (0.119) A household member can read *** 1172 (0.024) (0.040) (0.048) - Education Highest level of Education: primary *** 1172 (0.030) (0.082) (0.062) Highest level of Education: secondary * 1172 (0.019) (0.046) (0.036) Highest level of Education: secondary (0.031) (0.065) (0.050) Total education expenditures (0.005) (0.018) (0.019) A child doesn't go to school (between 6 and 13) (0.015) (0.047) (0.050) - Health At least one household member has a bad health (0.022) (0.074) (0.071) Composite health index (0.018) (0.037) (0.035) At least one household member suffer from malnutrition (0.001) (0.000) (0.001) At least one household member is stunting (0.002) (0.000) (0.002) - Household head characteristics Age of household head (0.484) (2.199) (2.128) Female household head (1=Yes, 0=No) ** 1172 (0.018) (0.041) (0.045) Muslim household head (1=Yes, 0=No) (0.160) (0.002) (0.143) Christian household head (1=Yes, 0=No) (0.056) (0.084) (0.048) Animist household head (1=Yes, 0=No) (0.050) (0.054) (0.028) No religion (1=Yes, 0=No) (0.081) (0.027) (0.077) Marital status separated (0.007) (0.020) (0.018) (4)

25 Marital status single (0.007) (0.039) (0.036) Marital status married polygamy (0.021) (0.063) (0.066) Household head has National ID *** 1158 (0.025) (0.064) (0.057) Household is registered (0.016) (0.126) (0.119) Housing characteristics The household has no water (0.043) (0.084) (0.086) The household has no electricity (0.003) (0.013) (0.012) The household has no toilet ** 1172 (0.030) (0.082) (0.074) The household head put waste in the nature (0.004) (0.046) (0.042) Socio-economic characteristics - Income revenues Income from agricultural activities (in K CFA Franc) (0.056) (0.062) (0.044) - Employment Worked as an employee in the past 12 months (0.004) (0.027) (0.026) At least one member of the household unemployed (0.003) (0.017) (0.017) -Enterprise Owns a micro-enterprise (0.039) (0.048) (0.062) Equipment value of the micro-enterprise (in CFA Franc) ** 1172 (5.133) (1.990) (3.593) - Land access The household owns land (0.047) (0.087) (0.081) Total estimated value of land * 1172 (0.147) (0.050) (0.143) No household member works in the land ** 1172 (0.003) (0.035) (0.035) The household raises animal ** 1172 (0.028) (0.082) (0.075) *** p<0.01, ** p<0.05, * p<0.1 Source: 2013 Baseline Survey. 24

26 Table 3.1: Impact of the cash transfer intervention on land access Annex 3: Results of the impact of the program on beneficiaries (1) Owns land (2) Total value of land (in K CFA Franc) ( 0.304) (0.282) * (0.058) * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) (3) Total surface of land (in ha) 3.033*** (0.626) *** (0.692) *** (0.100) (4) Land rented out (in K CFA Franc) ( 0.018) Impact of the intervention 0.149* (0.077) Difference at baseline * (0.082) (0.012) Difference at endline (0.015) (0.017) N R

27 Table 3.2: Impact of the cash transfer intervention on agricultural inputs and outputs (outcomes measured over the past 12 months) (1) Wage paid to labor force (in K CFA Franc) (2) No household member work in the farm (3) Number of agricultural products cultivated * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) (4) Value of total agricultural production (in K CFA Franc) (57.752) *** (37.439) *** (76.673) (5) Value of sales of agricultural products (in K CFA Franc) 0.151** (0.064) *** (0.068) *** (0.035) (6) Credit to buy fertilizers (7) Spending in fertilizers (in K CFA Franc) Impact of the intervention 0.303*** (0.071) (0.006) 2.575*** (0.408) 0.198** (0.082) 0.464*** (0.076) Difference at ** *** ** *** baseline ( 0.071) (0.005) (0.362) (0.076) (0.070) Difference at *** endline (0.080) (0.002) (0.433) (0.074) (0.061) N R

28 Table 3.3: Impact of the Cash transfer intervention on subsistence culture (1) Tomato (2) Cassava (3) Taro (4) Yam (5) Potato * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) (6) Corn (7) Rice (8) Cotton (9) Millet/ Sorghum (0.084) 0.195** (0.91) 0.416** (0.171) Impact of the intervention 0.012* (0.006) (0.005) 0.134** (0.036) 0.073** (0.030) 0.023** (0.009) 0.414*** (0.061) 0.196*** (0.059) 0.144* (0.078) (0.006) Difference at ** ** 0.042** *** ** *** baseline (0.007) (0.003) (0.072) (0.019) (0.018) (0.077) ( 0.050) (0.099) (0.006) Difference at 0.024*** ** 0.064*** endline 0.007) (0.004) (0.078) (0.027) (0.017) (0.110) (0.049) (0.104) (0.006) N R (10) Onion 27

29 Table 3.4: Impact of the Cash transfer intervention on livestock (1) Breeding as the main activity of the household 0.207** (0.251) (0.062) 0.204*** (0.047) (2) Number of cows (3) No cow (4) Raise chickens * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) ª : TLU stands for Tropical Livestock Unit. It measures the overall livestock value. (5) Number of chicken (6) Value of chicken consumed (in K CFA Franc) (0.761) (0.761) (.) (7) Value of sales of chicken (in K CFA Franc) 2.021*** (0.432) *** (0.410) (0.102) (8) Actual TLU ª Index value (9) TLU Index value Over the past 12 months 0.953** (0.385) *** (0.256) ** (0.189) (10) Difference (Actual TLU - TLU past 12 months) Impact of the intervention 0.666** (0.296) *** (0.059) ** ( 0.056) 3.091** (1.174) *** (0.304) 0.165*** (0.098) Difference at ** ** ** 0.243*** baseline (0.243) (0.065) (0.058) (1.165) (0.220) (0.054) Difference at *** *** endline (0.126) (0.045) (0.029) (0.037) (0.134) (0.071) N R

30 Table 3.5: Impact of the Cash transfer intervention on agricultural shocks Impact of the intervention Difference at baseline Difference at endline (1) Total value of loss due to shocks (in K CFAF) (43.239) (43.626) *** (9.893) (2) Total value of animal loss (in K CFAF) (10.199) *** ( ) *** (4.186 ) * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) (3) Total value of loss in the farm (in K CFAF) *** (9.420) *** ( 8.691) *** (11.344) (4) Total value of loss at the household level (in K CFAF) (4.482) ** (7.616) *** (3.914 ) N R

31 Annex 4. Results of the impact of the program on non-beneficiaries in Soulédé-Roua Table 4.1: Spillover effects of the intervention on land access (1) (2) Owns land Total value of land (in K CFA Franc) (3) Total surface of land (in ha) (4) Land rented out (in K CFA Franc) Beneficiaries *** (0.0805) (0.330) (0.560) (0.0269) Non-beneficiaries *** (0.0756) (0.370) (0.536) (0.0258) Baseline Mean Baseline SD Diff pval N Number of clusters Adj. R * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) 30

32 Table 4.2: Spillover effects of the intervention on agricultural production (1) Wage paid to labor force (in K CFA Franc) (2) No household member work in the farm (3) Number of agricultural products (4) Value of total agricultural production (in K CFA Franc) (5) Value of sales of agricultural products (in K CFA Franc) cultivated Beneficiaries *** *** *** ** (0.0819) ( ) (0.417) (52.82) (0.0647) Non-beneficiaries *** *** ** (0.0709) ( ) (0.378) (52.06) (0.0455) Baseline Mean Baseline SD Diff pval N Number of clusters Adj. R * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) 31

33 Table 4.3: Spillover effects of the intervention on subsistence culture (1) Tomato (2) Cassava (3) Taro (4) Yam (5) Potato (6) Corn (7) Rice (8) Cotton (9) Millet/Sorghum (10) Onion Beneficiaries * * *** ** *** *** *** * ** ( ) ( ) (0.0375) (0.0274) ( ) (0.0571) (0.0599) (0.0737) (0.0863) ( ) Non-beneficiaries *** ** *** ** ** ( ) ( ) (0.0389) (0.0247) (0.0198) (0.0545) (0.0540) (0.0683) (0.0851) ( ) Baseline Mean Baseline SD Diff pval N N of clusters Adj. R * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) 32

34 Table 4.4: Spillover effects of the intervention on livestock (1) Breeding as the main activity of the household (2) Number of cows (3) No cow (4) Raise chickens (5) Number of chicken (6) Value of chicken consumed (in K CFA Franc) (7) Value of sales of chicken (in K CFA Franc) (8) Actual TLU ª Index value (9) TLU Index value Over the past 12 months (10) Difference (Actual TLU -TLU past 12 months) Beneficiaries ** *** *** *** ** *** *** *** (0.0837) (0.205) (0.0453) (0.0508) (1.256) (0.613) (0.706) (0.220) (0.258) (0.0592) Non-beneficiaries * ** *** * ** *** ** *** ** (0.0941) (0.212) (0.0443) (0.0485) (1.259) (0.303) (0.722) (0.223) (0.282) (0.0950) Baseline Mean Baseline SD Diff pval N N of clusters Adj. R * p<0.10, ** p<0.05, *** p<0.01; Standard errors (in parentheses) clustered at the village level. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) ª : TLU stands for Tropical Livestock Unit. It measures the overall livestock value. 33

35 ANNEX 5: STATISTICAL APPENDIX Sampling As the Impact Assessment of the Social Nets Project was a randomized controlled trial with householdlevel treatment assignment, a power calculation was performed to determine the sample size of the baseline survey by considering three of the main Variables of interest: the poverty rate, the chronic under- 5 years old malnutrition rate, and the primary school enrollment rate. The calculation carried out by the National Institute of Statistics considers the following assumptions: -A reduction in the poverty rate from 54% to 44% with a power of 80%, type I error of 0.05, and a twosided alternative hypothesis; -A reduction in the under-five chronic malnutrition rate from 45% to 33% with a power> 90%, type I error of 0.05, and an unilateral alternative hypothesis; -An increase in the primary net enrollment from 52% to 72% with power> 90%, type I error of 0.05, and a one sided alternative hypothesis; -A non-response rate of 2%. These assumptions yielded an optimal sample size of 628 households for each of the 4 Strata considered, for a total of 2512 households to be surveyed. During the data collection of this sample, 2,350 households were surveyed, implying a gap of 162 households in relation to the target, specifically a gap of 17 beneficiary households, 30 non-beneficiary households targeted by the community, 54 Non-beneficiary households not targeted by the community, and 61 control households. Balance test The balance test is the basic one to verify balance at baseline in the key characteristics of treated and untreated households such as anthropometric values of children, education, consumption, saving, agricultural productivity, multidimensional poverty index, etc.). It consists of an ordinary least squares using baseline data (OLS) regression: Y! = β! + β! Treatment! +! With β! the mean value in the beneficiary households and β! that in the control households in the initial survey. A coefficient equality test judges the significant difference of the variable of interest Y! 18 between the two groups of households. Household characteristics that turn out to be not identical to the original survey are included in our basic regression. We apply the test of equality of coefficient to the following control variables: household size, type of head of household, education level of the household head, roof ceiling material, ownership of specific household assets (iron, radio, television, refrigerator) and cattle (goats, sheep, chickens, oxen). 18 i is the household 34

36 ANNEX 6: FIGURES Figure 1: Map of the intervention area Source: United Councils and Cities of Cameroon 35

37 Figure 2: Heterogeneity of the impact of the cash transfer intervention on land access according to the household head literacy rate and gender. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) 36

38 Figure 3: Heterogeneity of the impact of the cash transfer intervention on agricultural production according to the household head literacy rate and gender. Source: 2013 Baseline Survey (EESR) and 2016 Endline Survey (EESF) 37

39 Figure 4: Heterogeneity of the impact of the cash transfer intervention on the production of main culture (cereals and cotton) according to the household head literacy rate and gender. Source: 2013 Baseline Survey (EESR) and 2016 Endline survey (EESF) 38

40 Figure 5: Heterogeneity of the impact of the cash transfer intervention on the production of other subsistence culture according to the household head literacy rate and gender. Source: 2013 Baseline Survey (EESR) and 2016 Endline survey (EESF) 39