Policy Research Working Paper 8974 Reducing Hunger with Payments for Ecosystem Services (PES) Experimental Evidence from Burkina Faso Guigonan Serge Adjognon Daan van Soest Jonas Guthoff Development Economics Development Impact Evaluation Group August 2019 Policy Research Working Paper 8974 Abstract Does financial compensation for providing environmen- a year later. The vast majority of the community members tal conservation, improve the food security of the rural participating in the project were farmers, and the timing of poor in the drylands of Sub-Saharan Africa? This paper the payments coincided with the lean season, when most explores this question using data from a randomized con- farmers were at risk of food insecurity. Compared with trolled trial of a large scale reforestation implemented by the control group, the project’s participants’ households the Government of Burkina Faso. Members of communities reported 12 percent higher food consumption expenditures, located around selected protected forests were invited to and a reduction in moderate and severe food insecurity by plant indigenous tree species on degraded areas, and to take 35 percent to 60 percent. The transfers received were spent care of their maintenance. The financial compensation they mostly on cereals, meat, and pulses, with no evidence of would receive depended on the number of trees still alive increased consumption of temptation goods. This paper is a product of the Development Impact Evaluation Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at gadjognon@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Reducing Hunger with Payments for Ecosystem Services (PES): Experimental Evidence from Burkina Faso Guigonan Serge Adjognona,∗, Daan van Soestb , Jonas Guthoffa a Research Department; The World Bank, Washington D.C., USA b Department of Economics and Tilburg Sustainability Center, Tilburg University, Netherlands Keywords: Payments for Ecosystem Services, Food Security, REDD+, Reforestation, Burkina Faso JEL Codes : I38, Q18, Q23, Q53 ∗ Correspondence: gadjognon@worldbank.org, Tel.: +1-202-473-0222 1. Introduction Hunger is still pervasive in the developing world. Food insecurity is dis- proportionately concentrated in the drylands of the Sahel region, where a large share of the population is severely or moderately food insecure (FAO et al., 2019). The dependence on a single agricultural harvest per year, and the increasingly long lean period, translates into seasonal food insecurity. Such seasonal exposure to hunger can exacerbate malnutrition, health is- sues, and compromise early childhood development (Christian and Dillon, 2018). Food insecurity is one important issue in the Drylands of Sub-Saharan Africa; the battle against desertification is another. Payments for Environ- mental Services (PES) schemes offer communities and/or individual landown- ers financial compensation in return for the increased provision of environ- mental services, such as reduced deforestation or increased reforestation. By offering financial compensation conditional on improved environmental ser- vice provision, PES schemes address the fundamental cause of environmen- tal degradation - the fact that the costs of conservation are incurred locally, whereas a substantial share (if not all) of the conservation benefits accrue globally. While Jayachandran et al. (2017) show that PES schemes are effec- tive in promoting conservation, empirical evidence of their welfare impacts remains lacking. Recent reviews of the impact of PES programs document a critical dearth of evidence on welfare outcomes (Borner et al., 2017). In this paper we estimate the impact of a PES scheme, aimed at reforest- 1 ing degraded forest lands, on food security. We do so using a Randomized Controlled Trial (RCT), implemented in arid Burkina Faso, which allows for the causal estimate of PES payments on food security outcomes. As part of this intervention, the communities living in the proximity of the project forests were invited to participate in reforestation campaigns. The campaigns started with about 33,500 new trees being planted in Au- gust 2017 across 62 reforestation sites, in 11 gazetted forests. Subsequently, half of 630 community members, who were both interested in and eligible to participate in the project, were randomly assigned to the PES scheme and the other half were assigned to the control group. Those who ended up participating in the project were randomly matched into teams of five, who were subsequently made responsible for the maintenance of the trees planted on degraded forest land. Contracts were signed between each maintenance group and the implementing government agency stipulating that each group would receive about $0.70 for every tree still alive at verification almost a year later (in June 2018), and we made sure the payment was divided equally among all five members of each group. While such transfers could increase household consumption, the time and effort needed to take care of the trees, and ensure their survival, may displace labor from other productive (farm or non-farm activities), which could then negate the potentially positive effects of the financial compensation. Using primary data collected 13 months after the start of the PES pro- gram among all 630 individuals in the treatment and control group, we find 2 that the participation in the PES scheme induced a net increase in house- holds’ food consumption expenditures by 12%, and a reduction of the preva- lence of moderate and severe food insecurity by 35% to 60%. These results are robust to alternative measures of food (in)security, including the House- hold Food Insecurity Access Scale and Prevalence (HFIA), and the Household Hunger Scale (HHS). We also ran a number of additional robustness tests (including multiple hypothesis bias correction (Anderson, 2008; Kling et al., 2007)), and the results are unaffected. The vast majority of the individuals in our sample are farmers whose families are at risk of food insecurity in the pre-harvest months (June-August) because of the (imminent) exhaustion of the food stock from the previous harvest period. The timing of our payments coincided with that ‘lean period’, and hence our results indicate that the cash transfers from the above-described PES scheme shielded farmers against food insecurity at the time protection is most needed. These insights are interesting and suggest that there is a ‘double dividend’ associated with (conditional) conservation payments – increased environmen- tal protection, and reduced food insecurity. Our results thus contribute to two main strands of the development eco- nomics literature: Payments for Ecosystem Services (PES) as a dual environ- mental conservation and poverty reduction tool, and Cash for Work (C4W) as a social protection tool for vulnerable communities in the drylands. There is a large and still evolving literature across the developing world showing 3 that cash transfers 1 to individuals can relieve capital constraints and lead to positive welfare effects (De Mel et al., 2008; Angelucci and De Giorgi, 2009). For example, Haushofer and Shapiro (2016) found strong improve- ment in consumption, and large increase in psychological well-being of poor households in rural Kenya, as a result of a purely unconditional cash trans- fer intervention. These results were corroborated by Bedoya et al. (2019) in Afghanistan. However, very little of the evidence-base on “cash transfers ” focuses on C4W per se. One of the few exceptions is Beegle et al. (2017) who found no evidence that a labor intensive public work program implemented by the government of Malawi had improved lean season food security among participants. In contrast, Rosas Raffo et al. (2016) found positive effects of a C4W program on food spending, medicine, and assets, in Sierra Leone. Experimental studies of the impacts of C4W programs, let alone envi- ronmental C4W, especially in the drylands of SSA are still lacking critically from the development economics literature. Our study helps fill such gap and seeks to inform social protection and environmental policies across countries with populations still exposed to food insecurity, exacerbated by unstable rainfall patterns, and extreme weather events resulting from climate change. The rest of this paper is organized as follows: Section 2 describes the intervention being evaluated, and the experimental design used for this study. 1 Here, we refer broadly to all types of programs that involve handing cash to people. We recognize that the term “cash transfers” refers to more specific interventions in the development economics literature. 4 Section 3 lays out the empirical strategy for estimating the treatment effects, while Section 4 describes the data used and defines the main outcomes of interest. In Section 5, we present the results and discuss implications for policy. Section 6 concludes the paper. 2. Program Description and Experimental Design The notion that forests represent a cost effective source of carbon seques- tration while providing economic, environmental, and socio-cultural benefits (Canadell and Raupach, 2008; Busch et al., 2019) has received renewed im- petus with the 2015 Paris Agreement, because of the central role the Agree- ment attributed to forest conservation. In this spirit, the Government of Burkina Faso has initiated a Forest Investment Program (FIP) with joint technical assistance from the African Development Bank (AfDB), the World Bank Group (WBG), and with financial support from the Climate Investment Fund (CIF). One of the main goals of the Burkina Faso FIP is to improve the carbon sequestration capacity of gazetted forests while contributing to poverty reduction in forest-dependent communities. The program started with a pilot phase involving 12 of the country’s 77 gazetted forests. 2.1. The Intervention Burkina Faso is a landlocked country in the Sahel region of Africa, and is characterized by dry forests with sparse tree coverage. As a result, land- scape restoration through large scale reforestation appears as one of the most 5 important forests interventions (Adjognon et al., 2018).2 More specifically, one of the main interventions of Burkina Faso’s FIP program is a reforesta- tion campaign, which typically starts with engaging local communities in tree planting activities. These activities are facilitated by the project imple- mentation team as well as by local institutions called Forest Management Committees (Comite de Gestion Forestiere, or CGF). As stated before, the project targets 11 gazetted forests across the country, each of which has one CGF, which coordinates the forest management activities among the various stakeholders in their forest. Growing conditions in Burkina Faso are harsh, and survival rates can be improved not only by watering plants but also by putting up fire breaks, maintaining the holes in which saplings are planted, and by removing dead organic material in the plant’s vicinity. These activities require time and effort, which may displace labor from other productive (farm or non-farm activities). We partnered with the FIP project to implement a scheme aimed at improving tree survival rates for the 2017/2018 campaign, during which more than 33,500 had just been planted across 62 reforestation plots defined within the 11 gazetted forests of intervention. Following the planting phase in July/August 2017, the FIP project team asked the CGF of each of the 11 forests to identify individual members of the neighboring communities who 2 For example, the African Landscape Restoration Initiative (AFR100) is a large scale multi-country initiative to bring 100 million hectares of African land into restoration by 2030 (https://afr100.org/). 6 would be willing and able to engage in these activities. In total, 630 individu- als were recommended by the CGFs (about 10 people per reforestation plot), of which half were randomly selected to participate in the program. Partic- ipants employed in the PES contract were randomly grouped into teams of five and informed that they could collectively earn money based on mainte- nance and survival of the newly planted trees on their assigned reforestation plot. PES contracts assured the groups that they would receive about $0.70 for every tree that was still alive at verification and that every group member would receive an equal share (20%) of the total amount. About 9 months later, in June/July 2018 (just before the start of the next rainy season), all reforestation areas were visited by independent monitors and participants received their payments based on the number of trees that were still alive. From those selected to be part of the PES scheme, more than 95% received a positive payment. The individual payments ranged from about 840 FCFA ($1.50) to 25,620 FCFA (about $45.6) with an average of 8,300 FCFA ($14.8), which represents approximately a week of food consumption for the median rural household in Burkina Faso3 , and about half the monthly agricultural income of households in our study sample. Figure 1 shows the distribution of the payment amounts received by the participants. Our baseline data suggest that the participants in the scheme come from 3 Authors’ calculations from the BF 2014 Living Standards Measurement Survey (LSMS) data estimates the median rural household’s daily consumption expenditures at about 1580 FCFA ($3). In July 2018, 1000 FCFA = US$ 1.78. 7 the country’s poorest and most vulnerable groups. About 90% of them cited farming as their main source of livelihood, from which they had earned an average of $30 over the last 30 days prior to the survey to support a family of, on average, 12 members. During the endline survey period, which was close to the next harvest season, most of the participants were already out of food stocks from the previous harvest, which dates back almost a year before. Therefore, the timing of the PES payments is of particular importance. As shown in Figure 2, the lean period, when food insecurity is most prevalent due to dwindling stocks from the previous harvest among non- participants, spans over the months of May to September. Since payments were made around June/July, it suggests that the cash transfers as part of the PES scheme may act as a safety net for the recipient households during such a lean period, thereby supporting climate resilience. As a counterfactual the level of food insecurity in the months of June and July might have more quickly reached the same level as the month of August. In fact, at the time of payments, the participants were asked what they would do with the money received, and almost 50% of them mentioned food purchases for their family as primary usage plans. 2.2. Research Design This study uses a Randomized Controlled Trial (RCT) to evaluate the impact of an environmental cash transfer program, on participants’ well- being, measured by food security. The intervention of focus is a reforestation 8 campaign, implemented as part of the Burkina Faso FIP, and which ran from August 2017 to July 2018. Figure 3 shows the timeline of the different activities of this study. The 2017 reforestation campaign targeted about 62 reforestation plots distributed across the 11 forests targeted by the intervention, with each plot carrying about 400 newly planted trees. Immediately after the planting activ- ities, the project team asked the CGFs to find volunteers to join PES schemes; if selected, these volunteers would receive a monetary payment based on the number of trees still alive in their reforestation plot 9 months later. Antic- ipating over-subscription, the project team and the research team designed a public lottery to select the participants to the program. This approach was considered the fairest way to enroll people, and preserve the trust of the community members (Gueron, 2017). The lotteries were conducted separately in each reforestation site. Ten eligible and interested community members were recruited for each of the 62 plots, five of whom were randomly selected to form a maintenance group while the remaining individuals ended up in the control group. Those in the control group were not included in the PES scheme, but they were visited for all surveys.4 Field agents representing the research team were present and helped the project teams implement these public lotteries. This “ran- 4 We did not offer any monetary incentives, as was done in Bertrand et al. (2007), for inducing participation in our survey especially by the control group. Doing so would have potentially dampened our estimates of the impacts of the PES cash transfers. Nevertheless our field survey protocols minimized attrition problems, as discussed in later sections. 9 dom rejection” approach is central to our identification strategy to evaluate the impact of the PES intervention on participants’ food security, because randomly rejecting half of the interested individuals from each community implies that the treatment and control groups are likely to be identical in all respects. Following the treatment assignment, we collected baseline information on socio-economic characteristics and food security indicators of the individuals randomly allocated to the treatment group who signed the PES contracts and those who were assigned to the control group. These baseline characteristics allow us to check the validity of the randomization process. 3. Empirical Framework 3.1. Estimation Strategy We seek to estimate the average treatment effect of participation in the PES scheme on households’ food security outcomes. Using the Rubin Causal Model (RCM), this can be represented as: γ = E [YiT − YiC ], (1) where YiT and YiC represent the potential outcomes for individual i under alternative treatment status – treatment (T), or control (C) (Holland, 1986; Rubin, 1974). Our econometric strategy exploits the random assignment of individual members of the communities into a treatment and a control group 10 respectively. With about 315 individuals in each group, random assignment implies that E [YiC ] = E [YjC ], where i refers to individuals from the treat- ment group and j those from the control group. That means that γ can be estimated by comparing E [Y T ] and E [Y C ]. The treatment group received contracts, which guaranteed monetary com- pensation based on the number of trees still alive on their assigned reforesta- tion area at verification. The control group did not receive any contract, and hence received no cash transfers from the project. We estimate an intention- to-treat (ITT) effect of the intervention on households’ food security, using the following ANCOVA estimator (McKenzie, 2012)5 : Yi1 = α + βxi + λYi0 + γTi + i (2) Here, Yi1 and Yi0 , respectively, represent the endline and baseline values of a measure of food (in)security experienced by household i. xi is a vector of control variables, including forests sections indicators to account for the stratified randomization; Ti is the treatment assignment variable taking val- ues 1 if a household was offered the opportunity to participate in the PES scheme, and zero otherwise. Some food security indicators were collected only at endline. For those, a constrained version of Equation 2 is used, which does not include the baseline outcome value. i is an unobserved error 5 McKenzie (2012) shows that this approach can yield significant power gains compared to a difference-in-difference approach, especially when the outcome variables have low autocorrelation. 11 term assumed to be independently and identically distributed (iid). In Equation 2, γ is the main parameter of interest, which, under exo- geneity and Stable Unit Treatment Value assumption (SUTVA), identifies the ITT effects, on beneficiaries’ food security outcomes. The exogeneity as- sumption is a direct implication of the randomized selection of beneficiaries amongst eligible candidates. We also expect zero spillover across treatment and control groups because most of the potential participants came from separate villages, and do not really interact on a regular basis. In addition, given the relatively high success rate in the intervention6 , γ is also a good measure of the Average Treatment effect on the Treated (ATT). We estimate γ by applying Ordinary Least Squares (OLS) to Equation 2. We did not apply any clustering of the standard errors given that the randomized treatment was assigned at the individual level. As a robustness check, we also apply a randomization inference test. This approach, still not very widespread in economic research, has been strongly recommended for the analysis of randomized experiments such as ours (Athey and Imbens, 2017; Imbens and Wooldridge, 2009). To calculate the p-value associated with the estimated treatment effect, the randomized inference approach reassigns the treatment status by keeping the number of individuals in each group fixed and calculating the corresponding difference in means. This process is repeated 6 Of the 303 PES participants interviewed at endline, 291 reported to have participated and received a payment. This suggests a low reservation wage in our context, similarly to other rural poor communities of Sub-Saharan Africa (Goldberg, 2016). 12 many times, and produces a distribution of “fake” treatment effects under the null hypothesis. The p-value is simply the proportion of cases in which the “fake” treatment effect estimated is at least as large in absolute value as the estimate we seek to test. The method is attractive in part because it relies on fewer modeling assumptions encountered in a regression framework. 4. Data and Descriptive Statistics 4.1. Data Sources During the tree planting, 630 individuals were enrolled into the study and equally distributed among a treatment and a control group. In September to October 2017, right after the PES contracts were signed with farmers, we administered a baseline household survey in the villages of the respon- dents. The survey instrument included modules capturing households’ socio- demographic characteristics, asset ownership, and food consumption. Table 1 presents the baseline characteristics of our participants (and their house- holds) in both the treatment and control group, as well as the tests to what extent the distributions of these characteristics in the two groups are sim- ilar. The table indicates that the treatment and control groups are fairly balanced with respect to most characteristics. In particular, there is no sta- tistically significant difference between treatment and control groups in terms of baseline values of food consumption expenditures and household dietary diversity scores – two of our key variables of interest. There are, however, some imbalances with respect to other potentially important variables such 13 as the share of female participants in the treatment and control group, asset ownership, as well as the distance to the reforestation site. Differences are relatively small in absolute terms, but in our regression analysis we try to mitigate any possible selection effects by including them as control variables. In addition, an F-test of the joint significance across baseline covariates fails to reject the null hypothesis of no significant difference between treatment and control groups. This gives us confidence that any observed difference is likely the result of chance and therefore should not affect our treatment effects estimation. An endline survey was administered in November/December 2018, 5 to 6 months after the monetary transfers were disbursed to the PES participants, based on the tree survival rate on the respective reforestation plot. This follow-up survey was similarly structured as the baseline survey with the exception of a more exhaustive food security module. During this phase, we were able to track down 574 out of the 630 households interviewed at baseline, which corresponds to an attrition rate of 9%, which is fairly balanced across treatment and control households.7 Potential biases are thus likely to be small. Nevertheless, we test for balance in between treatment and control groups on baseline characteristics using the endline sample (see Table 2), and confirm that most characteristics remain balanced between treatment and control. 7 The difference in attrition rates between treatment and control groups was just 4.4 percentage points (p=0.053); See Table 2. 14 4.2. Food Security Measures We use the following common household level food security indicators, which capture broadly dietary diversity and consumption behaviors consid- ered to be symptomatic of food insecurity. • The Household Food Consumption Expenditures Food consumption expenditures (FExp) represents a direct indicator of economic access to food, as it captures the value of food consumed by the household over a given time period. Our survey instrument included a detailed food consumption module which asked each individual i in the treatment and control, from a list of food items f ∈ {1,...,F}, the quantities consumed by qif by their household and the total value of the consumption vif , over a 7-day recall period. The total value of food consumed by the household i (FExpi ) is calculated as: FExpi = vif (3) f • The Household Dietary Diversity Score (HDDS) The HDD Score (HDDS) is also an indicator that captures (economic) access to food.8 As a simple count of the number of different food groups consumed by a household during a defined period, it is meant 8 The HDD Score is a modified version of the U.S. Household Food Security Survey Module (HFSSM), adapted for use in a developing country context. 15 to provide an insight into a household’s economic ability to access a variety of foods (Kennedy et al., 2011; Upton et al., 2016). In this study, we calculate HDDS by first categorizing the food items con- sumed by household members over a 7-day recall period into 7 food groups, excluding sugar, alcohol and tobacco. Then we generate a bi- nary indicator Zig that takes value 1 for a food group g , if the household of respondent i had consumed a food item from that group during the recall period, and 0 otherwise. The HDD score for household i (HDDi ) is then calculated as in Equation 4 below: HDDSi = Zig (4) g • The Household Food Insecurity Access Scale (HFIAS) This indicator of consumption behavior infers food security status from people’s responses to questions about consumption patterns that are considered to be correlated with food insufficiency and anxiety - for example, whether and how often the household members had to eat less preferred food based on a lack of resources (Carletto et al., 2013). The indicator has been adapted from the current US approach to es- timate the national food insecurity prevalence.9 . It uses a set of nine questions designed to capture the degree and anxiety associated with 9 FANTA Project 16 households’ access to food. We sum up the frequency of food insecu- rity events to obtain the HFIAS Score. We also compute the HFIA prevalence by splitting our sample into 4 classes based on the severity of food insecurity: (i) the food secure, (ii) the mildly food insecure, (iii) the moderately food insecure, (iv) the severely food insecure. Figure 4 shows the distribution of our sample households into the 4 categories of food insecurity based on the HFIA Scale. A closely related measure is the Household Hunger Scale (HHS). It includes only 3 out of the 9 questions that are used to compute the HFIAS and is considered most appropriate in capturing the most severe forms of food insecurity. The survey instrument includes questions such as, “whether and how often the household members had to go to bed hungry because of lack in food provision ”. 5. Results and Discussion 5.1. Food Consumption Expenditures Figure 5 presents the Kernel density distribution of the reported house- hold food consumption expenditures (in logarithm) during the last 7 days prior to the survey for individuals in treatment and control group respec- tively. The distribution of the weekly food consumption expenditures for the treatment group is shifted to the right compared to the control group. This suggests a positive effect of the Burkina Faso reforestation PES scheme on food consumption expenditures. The point estimates of the treatment effects 17 from Equation 2 are summarized in Table 3. The results indicate an increase in weekly households’ food consumption expenditures of about 12% or 1798 FCFA, significant at the 5% level. These results are robust to the inclusion of covariates as well as baseline outcomes. However, when we run the same results using per capita household consumption expenditures, the average treatment effect drops to 7% with a confidence interval spanning estimates from 6% to 18%. We then use an instrumental variable approach to estimate the effects of the actual amount received on the food consumption expenditures. The results presented in table 4 show that every 100 FCFA received by farm- ers, increases the weekly household food consumption expenditures by 22.49 FCFA on average. This means that about 22 percent of the transfer received was invested in food consumption by the household, thereby contributing directly to improving household food security. 5.2. Household Dietary Diversity Did the Burkina Faso reforestation PES scheme improve the Household Dietary Diversity of participants? In Figure 6, we display a bar graph stack- ing up the food consumption expenditures for each food group, separately for the participants and non-participants in the scheme. This figure thus simul- taneously represents households’ total food consumption expenditure value (represented by the height of the bars) as well as their food consumption diversity (reflected by the shares of total food expenditures allocated to each 18 of the food categories). The graph suggests that, in general, cereals, roots and tubers take up more than 50 percent of the household’s food expendi- tures for both the treatment and the control group. This is consistent with the relatively low HDDS observed at baseline for both groups (HDDS=4.4 at baseline). Depending on the income elasticity of the consumption expen- ditures in the various food groups, we would expect to observe some change in the HDDS, as a result of the PES cash transfers received. To explore this, we compare the distribution of HDDS between treatment and control group individuals and conclude that those are fairly similar (see Figure 7). The p-value for the Kolmogorov-Smirnov test of equality of the two distributions is 0.99 suggesting no evidence of statistical difference between the two. This is corroborated by the regression results in Table 5 showing no evidence of a significant treatment effect on HDDS. This implies that although there is a positive and significant effect of the PES transfers on households’ overall food consumption expenditures, the ef- fect did not translate into an improved dietary diversity. Which food items did households finance with the PES transfers then? When we compare the mean expenditures for each food group between treatment and control, we find that the additional income received through the transfer increased significantly household expenditures on starchy foods (cereals and tubers), but also more meats and pulses, although only on the intensive margin (see Table A.9 in the Appendix). The PES scheme shows no effects on the likeli- hood of households consumption of any of these food groups. These results 19 are also confirmed when we use a multivariate Tobit regression10 to account not only for the multiple zeros recorded in the expenditures by food groups, but also to improve the efficiency of the estimation, taking into account the inter-dependency of these equations.11 Finally, our data also allows us to estimate the elasticity of expenditures in different food groups with respect to overall food consumption expenditures. As mentioned above, we find that for each percentage increase in food budget, only expenditures on starchy foods (cereals and tubers) are affected significantly (See results in Appendix A.10). 5.3. Household Food Insecurity Access Scale Table 6 summarizes the treatment effects estimation results on food secu- rity captured by the Household Food Insecurity Access Scale (HFIAS). The table reports the results for 3 alternative transformations of the HFIAS: the overall score in Columns 1 and 2; the binary indicator for being food secure in columns 3 and 4; and the binary indicator for being severely food insecure in columns 5 and 6. The conclusions are consistent across all those regressions. The results indicate that the Burkina Faso PES payments helped to sig- nificantly reduce food insecurity. The overall HFIA score was reduced by about 40 percent relative to the control group. In Columns 3 and 4, we see a 14.9 percentage points increase in the likelihood of being food secure in 10 See Wooldridge (2010) and Greene (2003) for details on the mvtobit model. 11 mvtobit results available from authors upon request. 20 general, while in columns 5 and 6, we observe a 9 percentage point reduction in the likelihood of being severely food insecure. All reported impacts are significant at the 1% level, or lower. These results are illustrated graphically in Figure 8. Relative to the control group, these estimates amount to 35 and 60 percent reduction, respectively, in food insecurity and severe food insecurity, as a result of the FIP PES transfers. The results on the HFIAS are even stronger when we consider the House- hold Hunger Scale as a food security indicator instead of the HFIAS. As shown in Table A.11 in the appendix, participation in the PES scheme has reduced the proportion of households who experienced hunger during the months prior to the survey by almost 70%. 5.4. Robustness Analyses The results presented above suggest significant and positive impacts of Burkina Faso’s FIP PES scheme on households’ food security outcomes. In this section we perform a series of robustness checks to ensure that those results were not driven by the choice of estimation strategies. 5.4.1. Randomized Inference Tests The results presented so far are derived from regression analysis, with inferences based on asymptotic theory and distributional assumptions. In this section, we use a randomized inference approach to ensure that our main results are robust to relaxing those assumptions (Athey and Imbens, 2017; Imbens and Wooldridge, 2009). In Table 7, we summarize the p-values for 21 each of our outcome variables from the randomized inference tests alongside the t-test used in the main analysis. The t-test and the randomized inference p-values in the last two columns of the table are very similar, which supports our main results. 5.4.2. Multiple Inference Adjustments As explained by Anderson (2008), the probability of incorrectly reject- ing the null increases as the number of hypotheses being tested increases. Failing to correct for multiple inference can lead to misleading policy advice. Given the number of food security indicators under consideration, might it be possible that the results obtained were the product of pure hazard? Overall, it is not very likely that our analysis suffers from (severe) multiple hypothesis bias. In our main analysis, we have tested seven outcomes, all of which are from the same family (food security), fairly correlated, and expected to be affected in the same direction by the treatment. We were able to reject the null hypothesis for all but one of them. Nevertheless, we follow Anderson (2008) and correct for multiple hypothesis testing in two main ways. First, we apply Bonferroni’s adjustment procedure, which estimates the probability that one or more of our outcome “families” of multiple tests is false – the so-called Family Wise Error Rate (FWER). It does so by upwardly adjusting the p-value for each of the tests within the family. The standardized results, summarized in Figure 9, are fully consistent with our main results. 22 Almost all food security indicators have confidence intervals excluding zero, and in the expected side of the zero line. The exceptions are the HDDS and the food consumption expenditure estimates. The drawback of the above approach is that, by adjusting p-values up- wards, it effectively reduces power of any given test. As an alternative, we use the summary index tests which pools all outcomes into a single test and thereby reduces overtesting, increases power, and provides a test on the gen- eral effect of the treatment on the outcomes of interest. We implement Kling et al.’s (2007) approach12 by combining all our food security outcomes into one weighted standardized index, and estimate the mean standardized effect size. To align all variables in the same direction we multiply the consump- tion expenditure variables by -1 and subtract the food secure status binary variables from 1. We obtain a point estimate of -0.2 standard deviation and a p-value of 0.00 for the summary index. This confirms that the Burkina Faso FIP PES scheme has reduced food insecurity for the beneficiary households. 6. Conclusion We use data on a sample of households across 32 communes in Burkina Faso and an experimental design to investigate the causal impacts of partici- pation in a Payments for Environmental Services (PES) scheme on a variety of food security outcomes. The specific intervention of interest is a refor- 12 We use avg effect command in STATA. 23 estation campaign through which randomly selected community members were invited to plant trees in designated areas of the forest and subsequently offered PES contracts according to which they would receive monetary pay- ments based on the number of trees still alive at verification - 9 months after the implementation. Considering several different food security indica- tors, we find consistently significant and positive effects of the PES transfers on food security. Participants’ households reported 12% higher food con- sumption expenditures, and as high as 60% reduction in the likelihood of being classified as severely food insecure. These results suggest that there is a ‘double dividend’ associated with (conditional) conservation payments – increased environmental protection, and reduced food insecurity. Designing multifaceted interventions presents substantial challenges, which are further complicated because of the diversity of the context across which such interventions might be relevant. To gain insights into the impact of such multifaceted interventions on the various outcome variables they wish to affect, more insights are needed in the heterogeneity of the treatment ef- fects across sub-groups in society. Similarly, it is also important to consider equilibrium effects, in the case of scale-ups or implementations on a larger scale (Angelucci and De Giorgi, 2009). Our study is not able to address these issues, as the estimation of these effects would require larger scale RCTs. In addition, it would be interesting to test the long-run effects of this type of PES intervention on food security, but also on other outcome variables. We leave these questions for future research. 24 References Adjognon, G. S., Rivera-Ballesteros, A. and van Soest, D. (2018). Satellite-based tree cover mapping for forest conservation in the drylands of sub saharan africa (ssa): Application to burkina faso gazetted forests. Development Engineering, p. 100039. Anderson, M. L. (2008). Multiple inference and gender differences in the effects of early intervention: A reevaluation of the abecedarian, perry preschool, and early training projects. Journal of the American statisti- cal Association, 103 (484), 1481–1495. Angelucci, M. and De Giorgi, G. (2009). Indirect effects of an aid pro- gram: how do cash transfers affect ineligibles’ consumption? American Economic Review, 99 (1), 486–508. Athey, S. and Imbens, G. W. (2017). The state of applied econometrics: Causality and policy evaluation. Journal of Economic Perspectives, 31 (2), 3–32. Bedoya, G., Coville, A., Haushofer, J., Isaqzadeh, M. and Shapiro, J. P. (2019). No Household Left Behind: Afghanistan Tar- geting the Ultra Poor Impact Evaluation. Tech. rep., National Bureau of Economic Research. Beegle, K., Galasso, E. and Goldberg, J. (2017). Direct and indi- 25 rect effects of malawi’s public works program on food security. Journal of Development Economics, 128, 1 – 23. Bertrand, M., Djankov, S., Hanna, R. and Mullainathan, S. (2007). Obtaining a Driver’s License in India: An Experimental Approach to Studying Corruption*. The Quarterly Journal of Economics, 122 (4), 1639–1676. Borner, J., Baylis, K., Corbera, E., de Blas, D. E., Honey-Ross, J., Persson, U. M. and Wunder, S. (2017). The effectiveness of pay- ments for environmental services. World Development, 96, 359 – 374. Busch, J., Engelmann, J., Cook-Patton, S. C., Griscom, B. W., Kroeger, T., Possingham, H. and Shyamsundar, P. (2019). Po- tential for low-cost carbon dioxide removal through tropical reforestation. Nature Climate Change, 9 (6), 463. Canadell, J. G. and Raupach, M. R. (2008). Managing forests for cli- mate change mitigation. science, 320 (5882), 1456–1457. Carletto, C., Zezza, A. and Banerjee, R. (2013). Towards better measurement of household food security: Harmonizing indicators and the role of household surveys. Global food security, 2 (1), 30–40. Christian, P. and Dillon, B. (2018). Growing and learning when con- sumption is seasonal: Long-term evidence from tanzania. Demography, 55 (3), 1091–1118. 26 De Mel, S., McKenzie, D. and Woodruff, C. (2008). Returns to cap- ital: Results from a randomized experiment. Quarterly Journal of Eco- nomics, 123 (4), 1329–72. FAO, IFAD, UNICEF, WFP and WHO (2019). The State of Food Se- curity and Nutrition in the World 2019. Safeguarding against economic slowdowns and downturns. FAO. Goldberg, J. (2016). Kwacha gonna do? experimental evidence about labor supply in rural malawi. American Economic Journal: Applied Eco- nomics, 8 (1), 129–49. Greene, W. H. (2003). Econometric analysis. Pearson Education India. Gueron, J. M. (2017). The politics and practice of social experiments: Seeds of a revolution. In Handbook of Economic Field Experiments, vol. 1, Elsevier, pp. 27–69. Haushofer, J. and Shapiro, J. (2016). The short-term impact of uncon- ditional cash transfers to the poor: experimental evidence from kenya. The Quarterly Journal of Economics, 131 (4), 1973–2042. Holland, P. W. (1986). Statistics and causal inference. Journal of the American statistical Association, 81 (396), 945–960. Imbens, G. W. and Wooldridge, J. M. (2009). Recent developments in the econometrics of program evaluation. Journal of economic literature, 47 (1), 5–86. 27 Jayachandran, S., de Laat, J., Lambin, E. and Stanton, C. (2017). Cash for carbon: A randomized trial of payments for ecosystem services to reduce deforestation. Science, 357 (6348), 267–273. Kennedy, G., Ballard, T. and Dop, M. C. (2011). Guidelines for mea- suring household and individual dietary diversity. Food and Agriculture Organization of the United Nations. Kling, J. R., Liebman, J. B. and Katz, L. F. (2007). Experimental analysis of neighborhood effects. Econometrica, 75 (1), 83–119. McKenzie, D. (2012). Beyond baseline and follow-up: The case for more t in experiments. Journal of development Economics, 99 (2), 210–221. Rosas Raffo, N., Sabarwal, S. et al. (2016). Can you work it? evi- dence on the productive potential of public works from a youth employment program in Sierra Leone. Tech. rep., The World Bank. Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of educational Psychology, 66 (5), 688. ´, J. D. and Barrett, C. B. (2016). Food security as Upton, J. B., Cisse resilience: reconciling definition and measurement. Agricultural economics, 47 (S1), 135–147. Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press. 28 Figure 1: Distribution of the actual payments received by the PES participants status 25 20 Share of respondents (in %) 15 10 5 0 0 5000 10000 15000 20000 25000 Actual paiements received (FCFA) 29 Figure 2: Seasonality of Food Insecurity (October 2017 to September 2018) 39.1 40 30 Percent 20 15.1 11.1 10 6.6 5.9 3.3 3.0 0.7 0.4 0.7 0.0 0.0 0 Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sept Note : This figure uses data from our control group only, and summarizes the percentage of households with insufficient food for each of the 12 months prior to the endline. Right after harvests (Oct-Feb), hunger rates are the lowest. But from May through September, food insecurity rises because most people run out of food reserves. 30 Figure 3: Timeline of the Randomized Controlled Trial Organization of Baseline HH survey Verification of Endline HH survey PES contract groups Tree Survival rates & payments July ’17 Sept - Oct ’17 June - July ’18, Aug ’18 Nov - Dec ’18 Lean season 31 Figure 4: Distribution of the sample by food insecurity status (HFIAP) Severely food insecure 15.13% Moderately food insecure 14.02% Food secure 59.41% 11.44% Mildly food insecure Note : At endline, 40 percent of the control group sample indicates to have experiencde some form of food insecurity, with around 15% suffering severe food insecurity. 32 Figure 5: (Log) Food Consumption Expenditures during the past 7 days .6 .4 Density .2 0 Median=9.1 6 8 10 12 Food Expenditure (in FCFA) PES non−PES Note: The log food consumption expenditure covers the household’s expenditures on Cereal and Tubers, Pulses, Vegetables, Fruits, Meat and Fish, Milk, Oil, Sugar as well as expenses for other items such as tobacco, alcohol and condiments, for the seven days prior to the survey. 33 Figure 6: Average Food Expenditure by Food Groups 15,000 1124 419 762 315 1029 Cereals and Tubers 10,000 397 Pulses 1625 704 Vegetables 294 11 911 Fruit 1160 6 Meat and Fish 744 771 Milk 534 Oil Sugar 5,000 Other 7032 6200 0 PES non−PES 34 Figure 7: Distribution of Household Dietary Diversity Score (HDDS) by PES participants and non-participants PES non−PES 40 33 Share of the respondents (in %) 30 30 24 20 19 19 17 16 13 10 10 9 6 4 0 0 2 4 6 8 0 2 4 6 8 HDDS Note : The p-value for the Kolmogorov-Smirnov test of equality of the two distributions equals 0.99 suggesting no difference in the distributions of HHDS between the recipients and non-recipients of the FIP PES transfers. 35 Figure 8: PES and Food Security Food Insecure Severely Food Insecure 40.6 40 40 30 30 26.4 Percentage Percentage 20 20 15.1 10 10 6.3 0 0 (N = 271) (N = 303) *** (N = 271) (N = 303) *** Non Participants PES Participants Note: The figure presents the share of PES participants and non-participants that are food insecure at endline, and also the share of those that are severely food insecure, taking into account the treatment effect. While 40% of the non-participants are food insecure, only about 26% of the PES participants are so too. Furthermore, 15% of the non-participants are severely food insecure, compared to about 6% of the participants. This means that we observe treatment effect sizes of 35-60% in the reduction of the prevalence of mild to severe food insecurity. 36 Figure 9: Bonferroni Multiple Hypothesis Adjustment Total (nominal) food consumption expenditures in FCFA (past 7 days) (LOG) Total (nominal) food consumption expenditures in FCFA (past 7 days) Household Food Insecurity Access Scale (HFIAS) Score HFIA_cat==Food secure (0/1) HFIA_cat==Severely food insecure (0/1) Household Hunger Scale (HHS) Score HHScat == little/no hunger in household (0/1) Household Dietary Diversity Score (HDDS) −.8 −.4 0 .4 .8 ← Standardized Effect of → 37 Table 1: Balance Table: Baseline Characteristics (1) (2) (3) T-test PES non-PES Total Difference Variable N Mean/SE N Mean/SE N Mean/SE (1)-(2) Age of respondent 325 39.871 305 38.433 630 39.175 1.438* (0.597) (0.621) (0.431) Female respondent (1/0) 325 0.123 305 0.193 630 0.157 -0.070** (0.018) (0.023) (0.015) Agriculture as primary activity (1/0) 325 0.908 305 0.885 630 0.897 0.022 (0.016) (0.018) (0.012) Annual revenu from primary activity (LOG) 325 3.057 305 3.064 630 3.061 -0.007 (0.411) (0.415) (0.292) Annual total revenu (LOG) 325 9.893 305 9.927 630 9.909 -0.035 (0.526) (0.531) (0.374) Has a secondary activity (1/0) 325 0.277 305 0.262 630 0.270 0.015 (0.025) (0.025) (0.018) Married (1/0) 325 0.905 305 0.862 630 0.884 0.042* (0.016) (0.020) (0.013) Household size 325 13.286 305 12.666 630 12.986 0.621 (0.405) (0.467) (0.308) Some formal schooling (1/0) 325 0.182 305 0.190 630 0.186 -0.009 (0.021) (0.023) (0.016) Member of a forest management group (1/0) 325 0.588 305 0.538 630 0.563 0.050 (0.027) (0.029) (0.020) Homestead far from the reforestation site (1/0) 325 0.877 305 0.944 630 0.910 -0.067*** (0.018) (0.013) (0.011) Agric. Asset Index (PCA) 325 0.132 305 -0.141 630 -0.000 0.273** (0.093) (0.089) (0.065) Household Asset Index (PCA) 325 0.177 305 -0.189 630 0.000 0.366** (0.130) (0.068) (0.075) Household landholdings (ha) 265 15.668 257 10.982 522 13.361 4.685 (4.077) (3.125) (2.579) Land area cultivated (ha) 265 4.923 257 4.686 522 4.806 0.237 (0.233) (0.229) (0.163) Total (nominal) food consumption expenditures in FCFA (past 7 days) 325 11258.837 305 11602.666 630 11425.294 -343.829 (639.395) (729.457) (482.895) (LOG) Total (nominal) food consumption expenditures in FCFA (past 7 days) 311 8.908 300 8.875 611 8.891 0.033 (0.060) (0.068) (0.045) Household Dietary Diversity Score (HDDS) 325 4.425 305 4.466 630 4.444 -0.041 (0.080) (0.082) (0.057) F-test of joint significance (p-value) 0.121 F-test, number of observations 517 Notes : The value displayed for t-tests are the differences in the means across the groups. The value displayed for F-tests are p-values. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 38 Table 2: Balance Table: Attrition and Characteristics of the Endline Sample (1) (2) (3) T-test PES non-PES Total Difference Variable N Mean/SE N Mean/SE N Mean/SE (1)-(2) Respondent found (Yes=1, No=0) 325 0.932 305 0.889 630 0.911 0.044* (0.014) (0.018) (0.011) Age of respondent 303 39.818 271 38.996 574 39.430 0.822 (0.622) (0.653) (0.450) Female respondent (1/0) 303 0.122 271 0.207 574 0.162 -0.085*** (0.019) (0.025) (0.015) Agriculture as primary activity (1/0) 303 0.904 271 0.889 574 0.897 0.015 (0.017) (0.019) (0.013) Annual revenu from primary activity (LOG) 303 3.071 271 2.899 574 2.990 0.172 (0.425) (0.455) (0.310) Annual total revenu (LOG) 303 9.914 271 9.719 574 9.822 0.194 (0.544) (0.583) (0.397) Has a secondary activity (1/0) 303 0.287 271 0.277 574 0.282 0.010 (0.026) (0.027) (0.019) Married (1/0) 303 0.904 271 0.871 574 0.889 0.033 (0.017) (0.020) (0.013) Household size 303 13.323 271 12.690 574 13.024 0.633 (0.410) (0.494) (0.318) Some formal schooling (1/0) 303 0.188 271 0.181 574 0.185 0.007 (0.022) (0.023) (0.016) Member of a forest management group (1/0) 303 0.584 271 0.535 574 0.561 0.049 (0.028) (0.030) (0.021) Homestead far from the reforestation site (1/0) 303 0.868 271 0.945 574 0.904 -0.077*** (0.019) (0.014) (0.012) Agric. Asset Index (PCA) 303 0.134 271 -0.130 574 0.009 0.264* (0.095) (0.095) (0.067) Household Asset Index (PCA) 303 0.200 271 -0.191 574 0.015 0.391** (0.139) (0.070) (0.081) Household landholdings (ha) 251 14.367 233 11.582 484 13.026 2.785 (3.842) (3.443) (2.590) Land area cultivated (ha) 251 5.002 233 4.688 484 4.851 0.314 (0.242) (0.235) (0.169) Total (nominal) food consumption expenditures in FCFA (past 7 days) 303 11207.086 271 12057.793 574 11608.726 -850.708 (664.619) (804.172) (516.794) (LOG) Total (nominal) food consumption expenditures in FCFA (past 7 days) 289 8.900 267 8.913 556 8.906 -0.014 (0.063) (0.073) (0.048) Household Dietary Diversity Score (HDDS) 303 4.406 271 4.491 574 4.446 -0.085 (0.084) (0.084) (0.060) F-test of joint significance (p-value) 0.198 F-test, number of observations 479 Notes : The value displayed for t-tests are the differences in the means across the groups. The value displayed for F-tests are p-values. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 39 Table 3: Regression Results: PES and Food Consumption Expenditures Food Expenditures log(Food Expenditures) (1) (2) (3) (4) (5) (6) PES 1780.8** 1798.1** 1775.4* 0.119** 0.128** 0.126** (2.04) (2.06) (1.74) (2.14) (2.25) (2.01) Baseline outcome 0.0317 0.0634 0.0446 0.0519 (0.80) (1.04) (1.49) (1.43) Constant 7126.2*** 6856.1** 5168.1 8.619*** 8.231*** 7.975*** (2.65) (2.53) (1.10) (50.13) (26.86) (20.34) Observations 574 574 484 574 556 479 R2 0.231 0.232 0.229 0.341 0.336 0.345 Adjusted R2 0.183 0.182 0.148 0.299 0.291 0.275 Covariates included No No Yes No No Yes Bloc fixed effects Yes Yes Yes Yes Yes Yes Control mean 11097 9 t statistics in parentheses The baseline outcomes represent respectively the food expenditure and log-food expenditure at baseline. The covariates include age, gender, occupation, landholdings, revenue, size of the household, education, distance to the reforestation site, and memnership of a forest membership group. * p < 0.10, ** p < 0.05, *** p < 0.01 40 Table 4: IV Regression Results: PES and Food Consumption Expenditures Food Expenditures log(Food Expenditures) (1) (2) (3) (4) (5) (6) PES tranfers received (in 100 FCFA) 22.49** 22.71** 21.97* 0.00151** 0.00159** 0.00154** (2.10) (2.12) (1.83) (2.20) (2.32) (2.12) Baseline outcome 0.0327 0.0648 0.0467 0.0546 (0.85) (1.12) (1.61) (1.58) Constant 7673.7*** 7400.6*** 5643.3 8.656*** 8.253*** 7.989*** (2.99) (2.86) (1.26) (52.60) (27.82) (21.43) Observations 574 574 484 574 556 479 R2 0.229 0.230 0.227 0.337 0.331 0.342 Adjusted R2 0.180 0.179 0.146 0.295 0.285 0.272 Covariates included No No Yes No No Yes Bloc fixed effects Yes Yes Yes Yes Yes Yes Control mean 11097 9 t statistics in parentheses The baseline outcomes represent respectively the food expenditure and log-food expenditure at baseline. The covariates include age, gender, occupation, landholdings, revenue, size of the household, education, distance to the reforestation site, and memnership of a forest membership group. * p < 0.10, ** p < 0.05, *** p < 0.01 41 Table 5: Regressions Results: HDDS (1) (2) (3) (4) PES 0.0218 0.0215 -0.0466 (0.27) (0.27) (-0.53) HDDS at baseline -0.00389 0.0352 0.0349 (-0.11) (0.75) (0.78) PES tranfers received (in 100 FCFA) -0.000577 (-0.56) Constant 1.996∗∗∗ 2.012∗∗∗ 1.725∗∗∗ 1.714∗∗∗ (8.18) (6.95) (3.87) (4.06) Observations 574 574 484 484 R2 0.585 0.585 0.606 0.605 Adjusted R2 0.558 0.558 0.565 0.564 Covariates included No No Yes Yes Bloc fixed effects Yes Yes Yes Yes Estimation approach OLS OLS OLS IV Control mean 4 t statistics in parentheses The covariates include age, gender, occupation, landholdings, revenue, size of the household, education, distance to the reforestation site, and membership of a forest membership group. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 42 Table 6: Regressions Results: Household Food Insecurity Access Scale (HFIAS) Overall Scale ∈ [0, 4] Being food secure(0/1) Being severely food insecure(0/1) (1) (2) (3) (4) (5) (6) PES -0.607*** -0.592*** 0.148*** 0.149*** -0.0934*** -0.0881*** (-3.80) (-3.34) (3.79) (3.41) (-4.16) (-3.53) Constant 0.800 1.436* 0.595*** 0.516** 0.0625 0.201* (1.63) (1.76) (4.96) (2.56) (0.90) (1.74) Observations 574 484 574 484 574 484 R2 0.184 0.223 0.089 0.119 0.283 0.324 Adjusted R2 0.132 0.143 0.032 0.029 0.238 0.255 Covariates included No Yes No Yes No Yes Bloc fixed effects Yes Yes Yes Yes Yes Yes Control mean 1.517 0.594 0.151 t statistics in parentheses The control covariates include age, gender, occupation, landholdings, revenue, size of the household, education, distance to the reforestation site, and memnership of a forest membership group. * p < 0.10, ** p < 0.05, *** p < 0.01 43 Table 7: Randomized inference results: Food Security Outcomes for PES participants versus non-participants (1) (2) Randomization PES non-PES t-test inference Variable N/[Clusters] Mean/SE N/[Clusters] Mean/SE p-value p-value Total (nominal) food consumption expenditures in FCFA (past 7 days) 303 12940.488 271 11097.077 0.041 0.038 (690.208) (655.157) (LOG) Total (nominal) food consumption expenditures in FCFA (past 7 days) 303 9.147 271 9.022 0.033 0.041 (0.047) (0.046) Household Food Insecurity Access Scale (HFIAS) ∈ [0, 4] 303 0.931 271 1.517 0.000 0.000 (0.108) (0.132) HFIAS = 1 (food secure) 303 0.736 271 0.594 0.000 0.000 (0.025) (0.030) HFIAS = 4 (severely insecure) 303 0.063 271 0.151 0.000 0.000 (0.014) (0.022) Household Hunger Scale (HHS) ∈ [0, 3] 303 0.155 271 0.336 0.000 0.000 (0.030) (0.044) HHS = 1 (no hunger) 303 0.964 271 0.882 0.000 0.000 (0.011) (0.020) HHS = 2 (moderate hunger) 303 0.036 271 0.118 0.000 0.000 (0.011) (0.020) Household Dietary Diversity Score (HDDS) 303 3.878 271 3.845 0.781 0.815 (0.082) (0.085) Notes : We show both standard p-values and p-values computed by randomization inference (RI) with 1000 repetitions. 44 Appendix A. Appendix 45 Table A.8: Balancing Test: Attrited - Endline (1) (2) (3) T-test Attrited Endline Total Difference Variable N Mean/SE N Mean/SE N Mean/SE (1)-(2) Age of respondent 56 36.554 574 39.430 630 39.175 -2.877* (1.456) (0.450) (0.431) Female respondent (1/0) 56 0.107 574 0.162 630 0.157 -0.055 (0.042) (0.015) (0.015) Agriculture as primary activity (1/0) 56 0.893 574 0.897 630 0.897 -0.004 (0.042) (0.013) (0.012) Annual revenu from primary activity (LOG) 56 3.784 574 2.990 630 3.061 0.794 (0.816) (0.310) (0.292) Annual total revenu (LOG) 56 10.806 574 9.822 630 9.909 0.984 (1.034) (0.397) (0.374) Has a secondary activity (1/0) 56 0.143 574 0.282 630 0.270 -0.139** (0.047) (0.019) (0.018) Married (1/0) 56 0.839 574 0.889 630 0.884 -0.049 (0.050) (0.013) (0.013) Household size 56 12.589 574 13.024 630 12.986 -0.435 (1.183) (0.318) (0.308) Some formal schooling (1/0) 56 0.196 574 0.185 630 0.186 0.012 (0.054) (0.016) (0.016) Member of a forest management group (1/0) 56 0.589 574 0.561 630 0.563 0.028 (0.066) (0.021) (0.020) Homestead far from the reforestation site (1/0) 56 0.964 574 0.904 630 0.910 0.060 (0.025) (0.012) (0.011) Indice des actifs agricoles (ACP) 56 -0.097 574 0.009 630 -0.000 -0.106 (0.222) (0.067) (0.065) Indice des actifs du mnage (ACP) 56 -0.157 574 0.015 630 0.000 -0.173 (0.161) (0.081) (0.075) Household landholdings (ha) 38 17.632 484 13.026 522 13.361 4.606 (13.054) (2.590) (2.579) Land area cultivated (ha) 38 4.237 484 4.851 522 4.806 -0.614 (0.643) (0.169) (0.163) Total (nominal) food consumption expenditures in FCFA (past 7 days) 56 9545.107 574 11608.726 630 11425.294 -2063.619 (1187.852) (516.794) (482.895) (LOG) Total (nominal) food consumption expenditures in FCFA (past 7 days) 55 8.743 556 8.906 611 8.891 -0.163 (0.139) (0.048) (0.045) Household Dietary Diversity Score (HDDS) 56 4.429 574 4.446 630 4.444 -0.017 (0.200) (0.060) (0.057) Notes : The value displayed for t-tests are the differences in the means across the groups. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 46 Table A.9: Treatment effects on food expenditure shares (1) (2) (3) T-test Treatment Control Total Difference Variable N Mean/SE N Mean/SE N Mean/SE (1)-(2) Consume Cereales, Roots, & Tubers (0/1) 303 0.997 271 0.989 574 0.993 0.008 (0.003) (0.006) (0.003) Consume Pulses (0/1) 303 0.756 271 0.742 574 0.749 0.014 (0.025) (0.027) (0.018) Consume Vegetables (0/1) 303 0.858 271 0.838 574 0.848 0.020 (0.020) (0.022) (0.015) Consume Fruits (0/1) 303 0.013 271 0.011 574 0.012 0.002 (0.007) (0.006) (0.005) Consume Meat & Fish (0/1) 303 0.498 271 0.487 574 0.493 0.011 (0.029) (0.030) (0.021) Consume Milk Products (0/1) 303 0.139 271 0.118 574 0.129 0.021 (0.020) (0.020) (0.014) Consume Oil (0/1) 303 0.568 271 0.605 574 0.585 -0.038 (0.029) (0.030) (0.021) Consume Sugar (0/1) 303 0.508 271 0.539 574 0.523 -0.030 (0.029) (0.030) (0.021) Consume Other - Alcohol, Tobacco, & Spices (0/1) 303 0.868 271 0.911 574 0.889 -0.043* (0.019) (0.017) (0.013) Expenditures on Cereales, Roots, & Tubers (0/1) 303 7031.601 271 6200.461 574 6639.199 831.139 (403.045) (480.380) (311.180) Expenditures on Pulses (0/1) 303 743.937 271 534.225 574 644.927 209.712** (86.191) (40.981) (49.594) Expenditures on Vegetables (0/1) 303 911.211 271 771.402 574 845.204 139.809 (83.138) (50.417) (49.970) Expenditures on Fruits (0/1) 303 10.561 271 6.273 574 8.537 4.288 (5.901) (4.181) (3.686) Expenditures on Meat & Fish (0/1) 303 1624.670 271 1160.055 574 1405.314 464.615** (176.658) (131.700) (112.406) Expenditures on Milk Products (0/1) 303 314.521 271 294.280 574 304.965 20.241 (66.953) (71.768) (48.920) Expenditures on Oil (0/1) 303 761.634 271 704.151 574 734.495 57.482 (56.656) (52.581) (38.853) Expenditures on Sugar (0/1) 303 418.647 271 396.985 574 408.420 21.662 (59.175) (43.091) (37.249) Expenditures on Other - Alcohol, Tobacco, & Spices (0/1) 303 1123.706 271 1029.244 574 1079.108 94.463 (86.570) (66.435) (55.415) Notes : The value displayed for t-tests are the differences in the means across the groups. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 47 Table A.10: Regression Results: Elasticity of consumption expenditures per food group (1) (2) (3) (4) (5) (6) (7) (8) (9) IHS tranformed Household Total Food Concumption Expenditures 1.388∗∗ 0.686 0.568 0.470 1.111 -1.062 -2.696 -1.465 -1.942 (2.25) (0.38) (0.31) (0.80) (0.50) (-0.50) (-1.08) (-0.57) (-1.02) Constant -3.774 -3.363 -1.193 -4.173 -13.13 10.49 25.30 13.20 20.70 (-0.67) (-0.20) (-0.07) (-0.78) (-0.65) (0.54) (1.10) (0.56) (1.18) Observations 484 484 484 484 484 484 484 484 484 R2 0.487 0.423 0.275 0.073 0.535 . 0.266 0.157 0.049 Adjusted R2 0.434 0.364 0.200 -0.022 0.487 . 0.190 0.070 -0.048 Covariates included Yes Yes Yes Yes Yes Yes Yes Yes Yes Bloc fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Control mean 6200 534 771 6 1160 294 704 397 1029 t statistics in parentheses The covariates include age, gender, occupation, landholdings, revenue, size of the household, education, distance to the reforestation site, and memnership of a forest membership group. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 48 Table A.11: Regressions Results: Household Hunger Scale (HHS) Scale ∈ [0, 3] Scale = 1 (no hunger) Scale = 2 (moderate hunger) (1) (2) (3) (4) (5) (6) PES -0.189*** -0.180*** 0.0855*** 0.0852*** -0.0855*** -0.0852*** (-4.16) (-3.58) (4.60) (4.16) (-4.60) (-4.16) Constant 0.123 0.371 0.945*** 0.929*** 0.0549 0.0711 (0.88) (1.60) (16.48) (9.81) (0.96) (0.75) Observations 574 484 574 484 574 484 R2 0.317 0.360 0.336 0.376 0.336 0.376 Adjusted R2 0.274 0.294 0.294 0.311 0.294 0.311 Controls No Yes No Yes No Yes Bloc fixed effects Yes Yes Yes Yes Yes Yes t statistics in parentheses The control covariates include age, gender, occupation, landholdings, revenue, size of the household, education, distance to the reforestation site, and memnership of a forest membership group. * p < 0.05, ** p < 0.01, *** p < 0.001 49