IMPACT EVALUATION OF A MULTI -INTERVENTION DEVELOPMENT PROJECT : EFFECTS ON ADOPTI ON OF AGRICULTURAL T ECHNOLOGIES AND LEVELS OF TRUST By Maria -Alexandra Peralta -Sanchez A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricultural, Food and Resource Economics - Doctor of Philosophy 2014 ABSTRACT IMPACT EVALUATION OF A MULTI -INTERVENTION DEVELOP MENT PROJECT: EFFECTS ON ADOPTION OF AGRICULTURAL TECH NOLOGIES AND LEVELS OF TRUST By Maria -Alexandra Peralta -Sanchez In this dissertation I conduct an impact evaluation of a complex rural development project in Central America with more than one intervention taking place at the same time, p urposive program placement , and project participant freedom to self -select to project interventions. For this purpose I use quasi -experimental panel data techniques difference -in- difference, propensity score matching difference -in-differences, and propensity score weighted regression to correct for selection bias due to self -selection by project par ticipants and purposive selection of project beneficiaries. Project impacts two years after implementation began are indicated by early behavior changes in the adoption of agricultural technologies and practices , as outcomes to be evaluated after two years of project implementation. The project had impacts on adoption of soil and water agricultural conservation practices, use of improve d storage technologies, and number of households with savings. These outcomes are likely to lead to long -term project impac ts. Project impacts differ according to wealth , as measured by area of cultivated land. Results suggest that the designers of multi -intervention rural development projects should consider targeting different grou ps, based upon beneficiaries™ characteristics , instead of promoting the full set of interventions to all beneficiaries. Impact evaluation s of multi -intervention development projects should also account for how project interventions will differ in the likely time lapse before behavioral changes can generate long term outcomes. In addition, I investigate how participation in group -based rural development project interventions affects levels of trust, a potential indirect outcome of rural development projects . To measure trust effects , I conduct ed a field -based trust experiment with integrated attitudinal trust questions. The results suggest that group -based rural development project interventions are likely to increase trust levels among farmers in the same village. Higher trust levels are expected to contribute to rural development and increase d agricultural income by facilitating market exchange via reduced transaction costs and increased information sharing . iv To my parents Alvaro and Amparo, and my brot hers Andres and Ca rlos. v ACKNOWLEDGMENTS I would like to thank my major professor Scott M. Swinton for his support and guidance over the past seven years we have worked together . I would like to thank my committee members Mywish Maredia, Robert Shupp, Songqin Jin and Jeffrey Wooldridge for their guidance . I would also like to than k Andrew Dillon for his comments and suggestions . I would also like to acknowledge the funding by the Howard G. Buffett Foundation through the Catholic Relief Services C entral America office. T he collaboration from Nitlapán at the Universidad Centroamericana during the data collection and data cleani ng conducted for this research. I also thank Catholic Relief Services (CRS) office in Nicaragua, Caritas and the Foundation for Research and Rural Development (FIDER) for their collaboration during the fieldwork stage of this research. I would like to thank my colleagues Byron Reyes and Valentin Verdier for t heir contribution to my research through their friendship, discussion and encouragement . Finally, I want to thank my colleagues in AFRE, the list is too long to be included here, my family and friends for all their encouragement, to BW for always being there for me, and D.J. Osborn III for his love and support during the last chapter of this journey. vi TABLE OF CONTENTS LIST OF TABLES ..................................................................................................................... viii LIST OF FIGURES ..................................................................................................................... xi KEY TO ABBREVIATIONS .................................................................................................... xii Chapter 1 Introduction ............................................................................................................... 1 Chapter 2 Impact Assessment with Opt -in Interventions: Evidence from a rural development project in Nicaragua .............................................................................................. 5 2.1 Introduction ................................................................................................................... 5 2.2 The Agriculture for Basic Needs (A4N) Project ......................................................... 9 2.3 Conceptual framework ............................................................................................... 11 2.4 Evaluating project impacts ........................................................................................ 13 2.4.1 Propensity score based methods: .............................................................................. 16 2.4.2 Regression based methods. ....................................................................................... 18 2.4.3 Heterogeneity of program impacts. .......................................................................... 19 2.5 Survey data use for evaluation of impacts ................................................................ 20 2.6 Results: A4N project impacts. ................................................................................... 21 2.6.1 Propensity score estimation ...................................................................................... 23 2.6.2 Project impacts on outcomes related to adoption of techn ologies and practices. .... 27 2.6.3 Heterogeneity if project impacts by area of cultivated land. .................................... 33 2.7 Conclusion ................................................................................................................... 36 Chapter 3 Trust and Group Participation in Rural Development Activities ...................... 38 3.1 Introduction ................................................................................................................. 38 3.2 The Agriculture for Basic Needs project (A4N) ....................................................... 42 3.3 Experimental design and procedures ........................................................................ 44 3.4 Results .......................................................................................................................... 53 3.4.1 Subject characteristics .............................................................................................. 53 3.4.2 Stated trust questions. ............................................................................................... 53 3.5 Trust experiment results . ........................................................................................... 56 3.5.1 Overall results. .......................................................................................................... 56 3.5.2 A4N vs. Non A4N group analysis. ............................................................................. 57 3.5.3 Determinants of the proportion sent ......................................................................... 59 3.6 Conclusion ................................................................................................................... 63 Chapter 4 Conclusions .............................................................................................................. 65 APPENDICES ............................................................................................................................. 69 vii Appendix A Study s ite description. ........................................................................................... 70 Appendix B. The Agriculture for Basic (A4N) Needs Project. ............................................... 73 Appendix C. Impact evaluation methods. ................................................................................. 77 C 1. The problem of impact evaluation ................................................................................. 77 C 2. Overview of Program Evaluation methods .................................................................. 80 C 3. Methods used to estimate project impacts .................................................................... 83 C.3.1. Regression based methods: ..................................................................................... 83 C.3.2. Propensity Score based methods. ........................................................................... 85 Appendix D. Sample design and data collection. ..................................................................... 89 Appendix E. List of data collected with the hous ehold and village survey instruments –.–––––––––––––––––––––––––––––––.. 97 Appendix F. Household survey instrument for the A4N impact evaluation panel. .............. 99 Appendix G. Village survey instrument for the A4N impact evaluation panel. ................. 143 Appendix H. Algorithm for estimating the propensity score. ............................................... 149 Appendix I. Pretreatment characteristics of treatment and comparison households ––––––––––––––––––––––––––––––––...150 Appendix J. Description of outcomes to be evaluated by the project. ................................. 152 Appendix K. Difference in difference estimation of project outcomes ................................ 160 Appendix L. Difference in difference treatment effects by area of cultivated land in 2009–––––––––––––––––––––––––––––––––––..165 Appendix M. Impacts on long -term outcomes related with agricultural income and household wealth. ...................................................................................................................... 180 Appendix N. Pretreatment characteristics of villages considered for the trust game –––––––––––––––––––––––––– –––––––........184 Appendix O. Number of groups per village, number of group members in 2011, population and households 2005, considered for the trust game. ............................................................ 185 Appendix P. Consent script for the trust game. ..................................................................... 186 Appendix Q. Game procedures in the field ............................................................................ 187 Appendix R. Trust game instruments ..................................................................................... 191 Appendix S. Survey instrument, for participants in the trust game. ................................... 193 REFERENCES .......................................................................................................................... 202 viii LIST OF TABLES Table 1. Logit model for estimating the propensity score or probability of participation in A4N. ...................................................................................................................................... 24 Table 2. Balancing tests of pretreatment covariates used for estimation of the propensity score. ..................................................................................................................................... 26 Table 3. Definition of intermediate outcome variables and units of measurement. ............. 28 Table 4. Project impacts on construction of agricultural conservation structures and on agricultural conservation practices. .................................................................................. 30 Table 5. Project impacts on storage practices, kitchen gardens, savings and credit and food scarcity. ................................................................................................................................ 32 Table 6. Project impacts by area of cultivated land on outcomes relate d to adoption of practices and technologies. ................................................................................................. 34 Table 7. Village pretreatment characteristics. ......................................................................... 46 Table 8. Session Villages and Number of Participants. ........................................................... 47 Table 9. Socioeconomic characteristics of A4N and non -A4N participants .......................... 54 Table 10. Results: trust questions .............................................................................................. 55 Table 11. Results: trust experiment .......................................................................................... 55 Table 12. Group analysis ............................................................................................................ 58 Table 13. Determinants of the proportion sent ........................................................................ 61 Table D 1. Sample Size for the A4N project evaluation in Nicaragua ––––––––.. 91 Table D 2. Population weights used in sample design. ............................................................ 91 ix Table D 3. Test for equal means for treatment and comparison villa ges. ............................. 92 Table D 4. Comparison between households in the original sample and households in the reduced final sample due to attrition, 2009. ..................................................................... 94 Table D 5. Number of observations collected by village and municipality. .......................... 95 Table I 1. Pretreatment characteristics of A4N and non -A4N households, 2009 ––––151 Table J 1. Outcome and explanatory variables for impact evaluation analysis ––––. 153 Table J 2. Participation of A4N household in different project interventions 2009 -2011. . 159 Table K 1. Project impacts on building of agricultural conservation structures –––– 160 Table K 2. Project impact on agricultural conservation practices ....................................... 161 Table K 3. Impacts on postharvest storage ............................................................................ 162 Table K 4. Project impacts on the number of small livestock .............................................. 163 Table K 5. Project impacts on savings, credit, hunger, crop losses and kitchen garden. ... 164 Table L 1. DID treatment effect on agricultural conservatio n structures, for households with cultivated land >= 1.5 Mz –––––––––––––––––––––––– 165 Table L 2. DID treatment effect on agricultural conservation structures, for households with 1.5Mz3Mz ................................................................................................ 167 Table L 4. DID treatment effect on agricultural conservation practices, for households with Cultivated land<=1.5Mz ................................................................................................... 168 Table L 5. DID treatment effect on agricultural conservation practices, for households with 1.5Mz3Mz ........................................................................................................ 170 Table L 7. DID treatment effect on postharvest management, for households with Cultivated land<=1.5Mz ................................................................................................... 171 Table L 8. DID treatment effect on postharvest management, for households with 1.5Mz3Mz ........................................................................................................................... 173 Table L 10. DID treatment effect on small livestock, for households with Cultivated land<=1.5Mz ...................................................................................................................... 174 Table L 11. DID treatment effect on small livestock, for households with 1.5Mz3Mz ........................................................................................................................... 176 Table L 13. DID treatment effect on saving, credit, hunger, losses and kitchen garden, for households with Cultivated land<=1.5Mz ...................................................................... 177 Table L 14. DID treatment effect on saving, credit, hunger, losses and kitchen garden, for households with 1.53Mz ............................................................................ 179 Table M 1. Project impact on agricultural income and household wealth related outcomes. ––––––––––––––––––––––––––––––––––. 181 Table M 2. Heterogeneity of program impacts in outcomes related to agricultural income and household wealth. ...................................................................................................... 183 Table N 1. Village pretreatment characteristics ––––––––––––––––– 184 Table O 1. Number of groups per A4N village, number of members in 2011, pretreatment population and households in each village 2005 ––––––––––––––––.....185 xi LIST OF FIGURES Figure 1. Impact trajectories of different project interventions. ............................................ 12 Figure 2 . Estimated propensity score or probability of program participation. .................. 25 Figure 3. Percentage of senders by amount sent, A4N and non -A4N participants in the trust game ..................................................................................................................................... 57 Figure A 1. Map with location of study area –––––––––––––––––––.71 xii KEY TO ABBREVIATIONS A4N: Agriculture for Basic Needs (an integrated rural development project) ATE: Average Treatment Effect ATT: Average Treatment Effect on the Treated BNI: Basic Needs Index CRS-LACRO: Catholic Relief Services Œ Latin America and the Caribbean Office CRS: Catholic Relief Services DID: Difference in Difference ECA: Farmer field school, f rom its acronym in Spanish Epan: Epanechnikov FD: First Difference FIDER: Fundac ión de Investigación y Desarrollo Rural GSS: Generalized Social Survey hh: Household(s) INIDE: Instituto Nacional de Informaci ón de Desarrollo (of Nicaragua) IV: Instrumental Variable LLR: Local Linear Regression xiii LSMS: Living Standards Measurement Stud y Mz: Manzana (unit of land area = 1.73 acres) NGO: Non Governmental Organization NN: Ne arest Neighbor OLS: Ordinary Least Squares PS: Propensity Score PSM: Propensity Score Matching PSM-DID: Propensity Score Matching Difference in Difference PSU: Pri mary Sam pling Unit PSW: Propensity Score Weighting qq: quintal (s) (unit of weight =100Kg) RDD: Regression Discontinuity Design ROSCA: Rotating Savings and Credit Association 1 Chapter 1 Introduction Every year, billions of dollars are spent on development projects around the world with the aim of improving the wellbeing of the poor. Yet rural areas in the developing world still lag behind, with high rates of poverty and inequality (International Fund for Agricultur al Development, 2010). In recent years there has been an increasing attention on the role of agriculture in reducing poverty, since a high percentage (more than 70%) of the poor live in rural areas and depend on agriculture for their livelihoods. Adopti on of improved agricultural technologies and practices is likely to reduce poverty (Feder, Just, & Zi lberman, 1985) , as is confirm ed by recent impact evaluation studies (Bravo -Ureta, Almeida, Solís, & Inestroza, 2011; Canavire -Bacarreza & Hanauer, 2013; Cavatassi, Salazar, González -Flores, & Winters, 20 11; Del Carpio, Loayza, & Datar, 2011; Dillon, 2011; E. Duflo, Kremer, & Robinson, 2008; Mendola, 2007; Nkonya, Phillip, Mogues, Pender, & Kato, 2012; Nkonya et al., 2012) . Increase d trust, a form of social capital, is also likely to increase income, because it facilitate s transactions, particularly in environment s where formal institutions are not well developed (Fafchamps, 2006) . Improvements in the levels of trust among rural communities are likely to improve the dissemination of new technologies and reduce transaction c osts, further helping to increa se rural income (Grootaert & Narayan, 2004; Lyon, 2000) . For project implementers and donors, as well as for governments, international organizations and non-governmental organizations ( NGOs), it is important to have answers to questions as to how effective are their poverty reduction interventions and which interventions have the most impact. 2 For impact evaluation of complex, multi -intervention, rural development projects, a rigorous impact as sessment looks into overall impacts on the different outcomes the project aims to change. But the different components of the project will entail different time elapse to achieve project impacts. Promotion of new technologies and practices implies that the successful adoption of those needs to occur before impacts are translated into increases in agricultural income and household wealth. A complete strategy for analyzing impacts first measures changes of behavior measured as adoption of agricultural techn ologies and practices in the short term and overall effects of these changes in the long term. Rural development project impacts are not limited to increases in outcomes such as adoption of technologies, or increases in agricultural income. Rural develo pment projects using the strategy of organizing the beneficiaries in groups or associations, can also contribute with outcomes related to social capital formation. For instance, the interaction among beneficiaries in group -based interventions boosts the le vel of trust among themselves and of other people. This outcome is likely to facilitate future endeavors for common goals. Moreover, due to the link between trust and economic development (Dearmon & Grier, 2009; Fafchamps, 2006, 2006; Fukuyama, 2001; Özcan & Bjørnskov, 2011) , increased trust facilitates transactions, reducing its costs, contributing to rural development projects™ goal of increased income. In this dissertation I conduct an impact ev aluation of the Agriculture for Basic Needs (A4N) proj ect in Nicaragua, a rural development project that promoted more than one intervention at the same time. A4N provide d poor farmers with a set of skills to achieve sustainable production and to increase agricultural income . To assess the impact of this project, I use p anel data econometric techniques for the analysi s of a household survey of project partici pants and non -participants, conducted in 2010 and 2012 . The project promoted agricultural conservati on, post -3 harvest management, vegetable gardens, saving and lending. Since the evaluation took place after two years of project implementatio n, I evaluated changed behavior, measured as adoption of agricultural technologies and practices. The timing of proj ect impacts was considered. Otherwise the results of the evaluation on long -term outcomes at early stages of project implementation could lead to misleading results of no project impacts. The main strategy of promoting interventions by A4N used group forma tion. This strategy of forming producer groups facilitates dissemination of the practices promoted by the project and interaction among project participants to achieve common goals. It also corrects for market failure, such as lack of access to credit (e.g . formation of saving and lending groups). This strategy is prone to achieve the direct effects of adoption of the technologies and practices promoted, but also indirect effects from interaction among beneficiaries promoted by the project. Such interaction is likely to boost trust and social capital within villages, which in turn will make more likely that members of a village to continue working together on efforts to achieve common goals, even after project ends. My goal o f this dissertation is to conduc t an impact evaluation of the interventions of the Agriculture for Basic Needs (A4N) project in Nicaragua, looking into project outcomes related to adoption of agricultural technologies and practices promoted by the project likely to lead to long term impa cts on agricultural income and farm productivity. For this analysis, different methods that use panel data are implemented to correct for participant selection bias due to purposive selection of participants. I also look into indirect impacts of the A4N pr oject, using experimental economics as a tool for impact evaluation to determine impacts on trust levels, due to participation in group -based interventions. 4 Specific research objectives are as follows: 1. To measure the impact of the A4N interventions in Nicaragua on outcomes related to adoption of improved agricultural practices and practices, likely to lead to long -term outcomes such as agricultural income and farm productivity. 2. To evaluate the indirect impacts of the project on project beneficiaries™ trust levels. This dissertation proceeds as follows: Chapter 2 and C hapter 3 were written as self -contained essay s. Chapter 2 evaluates the impact of the A4N project on outcomes related to adoption of agricultural technologies and other selected practices . It uses panel data methods and consider s the timing of project impacts for conducting impact eva luation. Chapter 3 uses experimental methods to explore how participation in group based project interventions affects the levels of trust among members of the same village. Finally, Chapter 4 presents the main conclusions of the dissertation. The dissertation also contains a set of appendices with background information and extensions to the contents of the two essays. 5 Chapter 2 Impact Assessment with Opt -in Interventions: Evidence from a rural development project in Nicaragua 2.1 Introduction In spite of efforts to reduce poverty worldwide , rural areas still lag behind. Of the 1.4 billion people living with less than $1.25 a day in 2005, around 70% lived in rural areas (International Fund for Agricult ural Development, 2010) . Adoption of improved agricultural technologies has the potential to reduce poverty, either directly by increasing production for home consumption, raising revenues from sales, or reducing production costs for the adopters of the technology, and/or indirectly by reducing prices of food, increasing wages in agricultural production, or through linkages with other economic sectors (de Janvry & Sadoulet, 2002; Minten & Barrett, 2008). Questions on how effective are the strategies promoted by d evelopment projects at achieving the goal of poverty reduction are of particular interest to governments, project implementers and donors . Impact evaluations of projects promoting improved agricultural technologies have been con ducted with the goal of answering these questions. Several studies find that improved seed varieties increases household consumption and expenditures (Becerril & Abdulai, 2010; Mendola, 2007) ; technological changes brought by agricultural conservation projects increase technological efficiency (Cavatassi et al., 2011; Solis, Bravo -Ureta, & Quiroga, 2008) ; and the use of improved storage techno logies reduces stored grain losses (Gitonga, D e Groote, Kassie, & Tefera, 2013) . 6 Sometimes rural development projects promote multiple interventions to achieve the goal of poverty reduction. Techniques for evaluating projects with this design are available to determine the impact of each interventi on and some combinations (Cuong, 2009; Lechner, 2001; Wooldridge, 2010) . Data collec tion requires a sample size that allows for meaningful inferences about these effects. Yet when project participants self -select into different program interventions, it is difficult ex ante to forecast levels of participation. These challenges make diffic ult to conduct evaluations of rural development projects with multiple interventions, and may explain why the literature on impact evaluation of these projects is scant. When two or more agricultural technologies are promoted as a package and the elements of the package are divisible, project participants may adopt elements of this package instead of the package as a whole (Byerlee & Hesse de Polanco, 1986; Feder et al., 1985) . To achieve project goals, such as increase in agricultural productivity and agricultural income, increases in adoption rates o f improved technologies is required (Teklewold, Kassie, & Shiferaw, 2013) . But adoption is not automatic upon exposure to a project treatment. Learning a bout the benefits of different technologies does not imply that project beneficiaries will adopt them. This is because of costs associated with adoption (Feder et al., 1985) . Resource c onstraints also affect adoption, so farm households m ay be willing but unable to adopt the recommended technologies (Nowak , (1992) . Different project interventions are also likely to vary in the time horizons for achieving impacts (King & Behrman, 2009; Tjernström, Toledo, & Carter, 201 3). For instance, agricultural conservation practices and structures will take a long time before stabilizing soils can stabiliz e crop yields. In contrast, interventions such as improved storage c an lead to fairly rapid reduction of storage losses. Thes e different periods of elapsed time from project start date to moment of project impact mean that consideration must be given to two issues: 1) what outcomes to 7 evaluate at different stages of project implementation, and 2) how to identify early indicators of project effectiveness. Our objective in this research is to conduct an impact evaluation of a rural development project with multiple interventions . In using data from just two years after project initiation , the evaluation aims to identify early out comes to determine whether the project strategy Œ promoting multiple interventions for all beneficiaries Œ changed behavior , as measure d by impacts on adoption of improved agricultural technologies. We test for heterogeneity of project impacts according to relative wealth, as measured by the area of cultivated land. With this study we contribute to the literature o n impact evaluation of rural development projects with multiple , opt-in interventions. The project to be evaluated, called Agriculture for Basic Needs (A4N), promoted agricultural conservation practices and structures , post -harvest management, nutritious crops in kitchen gardens, and saving and lending groups, among other interventions. Farm households in participating villages had the opportunity to opt in to a set of A4N interventions . A4N was implemented in four countries of Central America. We focus on the evaluation of A4N in Nicaragua , a country characterized by high concentration of the poor in rural areas, and by low levels of agricultural productivity (World Bank, 2008) , which is the case for many developing countries (International Fund for Agricultural Development, 2010) . Project beneficiaries were not randomly assigned . Instead , they self -select ed into project interventions, so selection bias was a concern for impact evaluation. Since experimental design was not feasible, the program evaluation uses quasi -experimental methods. Difference in difference ( DID ), propensity score matching difference in difference (P SM-DID) and propensity 8 score weighting (PSW) are quasi -experimental methods that can be used to control for time invariant , unobservable characteristics and to correct for selection bias on observables (Smith & Tod d, 2005). Our results suggest that the project increase d the adoption of agricultural practices that are likely to translate into longer -term impacts of increase in farm productivity and agricultural income. The results also suggest that project interv entions should be targeted according to the resource constraints that household s face , instead of being promoted to all households. This chapter is organized as follows: section 2.2 presents the project to be evaluated ; section 2.3 describes a conceptual framework for the analysis of project impacts ; section 2.4 describes the survey data used for analysis; section 2.5 addresses the problem of impact evaluation and presents the methods we use for evaluating project impacts; section 2.6 presents results and finally section 2.7 concludes. 9 2.2 The Agriculture for Basic Needs (A4N) Project The A4N project was a three year integrated rural development project implemented in four Central Americ an countries during 2009 -2012. This research focuses on the project in Nicaragua. It was managed by Catholic Relief Services (CRS) and implemented in the field by its partners Caritas and the Foundation for Research and Rural Development (FIDER). Informati on on the study site and location is found in Appendix A , and more detailed description of the A4N project can be found in Appendix B. The A4N project aim ed to provide farmers with a set of skills for achieving sustainable farm production and increased agricultural income , training farmers on farmer field schools, producer groups, and saving and lending groups, as well as providing technical assistance a t the farm. The project promoted agricultural conservation practices and construction of agricultural conservation structures , training in post -harvest management, storage practices, use of metallic silos for storage of grains, and training in small livest ock management (husbandry, feed production, vaccination regimes, manure collection). Participation in farmer innovation groups, implementation of trial plots with improved varieties of maize and beans, improved farming practices, nutritious vegetable crops in kitchen gardens (cabbage, carrots, onion, tomatoes and green leafy vegetables ). The project also addressed market failure by promoting saving and lending groups to establish the habit of saving and to increase access to credit. The project provided beneficiaries with agricultural assets, such as metallic silos, construction material for animal enclosures, water harvesting structures, plastic water tanks and water filters, and small animals, such as poultry, pigs and goats. Projec t interventions were available for all project participants, the project encouraged participants in different project activities to 10 participate on other project interventions; for instance, producer groups were encourage to form saving groups. The project also encouraged members of the same household to participate in multiple project interventions. The A4N project first targeted villages considered poor, in terms of limited access to basic services such as water and sanitation, predominance of small land h oldings and reliance on production of staple grains (maize and beans). These villages are located in areas of natural resource degradation with relatively high vulnerability to natural disasters. Within these villages, in order to be eligible to participa te in the A4N project, households were expected to be characterized by most of the following official eligibility criteria: Cultivated land area less than two manzanas (1 Mz = 1.73 acres). Cultivated land on steep slopes. Lack of access to any of the foll owing public services: piped water, sanitation, and electricity. Materials for house walls not brick or concrete; roof not concrete, zinc or brick; floor not concrete, ceramic or tile. Household experiences hunger during some period of the year. Household head is female. Household includes children younger than five years old. In spite of these formal eligibility criteria, the A4N ™s village -level managers found it difficult to exclude participation of village members . So the program allowed some technical ly ineligible individuals to participate , in the hope that they would help to spread A4N interventions during and after program implementation. 11 Two different processes led to nonrandom participation in specific A4N interventions. First, official eligibili ty criteria that were not evenly enforced, so households permitted to participate in the A4N proj ect vary on observable traits. Second, the self -selection of individuals into specific A4N interventions means that unobservable traits may also affect partici pation assignments. 2.3 Conceptual framework Development projects with multiple interventions like A4N provide treatment in the form of exposure to training and provision of inputs. As beneficiaries, farmer households learn about new technologies and practices , allowing the m to update the information they use for solving an inter -temporal utility maximization process (Besley & Case, 1993; Feder et al., 1985) . The y make decisions on input allocation in each period as part of a process of learning by doing and learning by using (Feder et al., 1985) . Adoption of new technolog ies and practices implie s changes in costs . The se cost s could take the form of labor ( e.g. building agricultural conservation structures), purchased inputs (e.g. high yield seed varieties, fertilizer), or acquiring information about the new technology, both on its use and its ben efits (Sunding & Zilberman, 2001 ). Farmer households that are both willing and able to adopt a given technology will do so. But the time that must elapse for adopters to realize project impacts will differ for different technologies. Figure 1, panel I, illustrates the impact of a technology with benefits that happen long after adoption. Whereas Figure 1, pane l II, shows a technology that leads to impacts in the short term, close to adoption. Practices such as the construction of terraces and stone barriers, which are agricultural conservation structures, imply significant up -front investments by project benefi ciaries for construction and maintenance. Benefits in the form of averted yield decline and 12 reduced yield variability are realized only gradually and unevenly, with the greatest benefits occurring under rare, extreme rainfall conditions. The contrary will occur with the adoption of the use of metallic silos for storage. Once the silos have been provided by the project and farmers trained in their use, the costs are the time that needed to prepare the grain for storage. Reduced storage losses can be realize d in less than a year. Figure 1. Impact trajectories of different project interventions . Adapted from King and Behrman (2009) Impact Impact time time t1 t1 I II 13 If the project is evaluated at an early stage, say time t1 in Figure 1, we are able to observe adoption of the technologies and practices promoted by the project and their early benefits. For a conservation technology like the one on Panel I, early impacts will be small, regardless of the degree of adoption; for a storage technology like the one in Panel II, early impacts tend to be relatively much larger. With this di fference in mind, we evaluate project impacts on the adoption of a range behaviors promoted, including agricultural conservation structures and practices, improved storage technologies, vegetable kitchen gardens, and membership in savings and credit associ ations. We also evaluate early outcomes from these practices, specifically the number of households that experiencing stored grain losses or food scarcity , and households with savings . 2.4 Evaluating project impacts We approach program evaluation though Rub in™s potential outcome framework (Rubin, 1974) . The objective of program evaluation is to determine how the intervention or applied treatment affects a desired outcome, evaluating the treatment effect against a counterfactual. Participation of individual i in the project is referred to as a fitreatmentfl given by w i=1, so w i=0 if the individual has not been exposed to treatment. The observed outcome for individual i is: Equation 1 iiiiiywywy01)1( which means that the outco me for an individual who participates is y 1i and if she does not participate the outcome is y 0i. The treatment effect of the program intervention is: Equation 2 iiiiyyy01 14 But the resulting outcome attributable to a program cannot be observed in an individual participating and not participating in the program at the same time. Therefore, the problem of program evaluation is a problem of missing data, and the program effect c annot be calculated for the same individual, but instead requires constructing a counterfactual to calculate average treatment effects across individuals in a sample from the population. The parameters of interest are the average treatment effect on the p opulation, ATE, and the average treatment effect on the treated, ATT. The ATE is the difference between the expectation of the outcome with and without the program. For an individual, given a vector of characteristics x, it is: Equation 3 )|()|())((01xxxyEyEEATE The ATE measures the effect of the treatment on both participants and non -participants. The average treatment effect on the treated, ATT, is the expected value of the outcome for those who participated in the program, conditional on the individual characterist ics that determine program participation, x: Equation 4 )1,|()1,|()1|)((01wyEwyEwxEATTxx As already mentioned, E(y0|x, w=1) , the expected outcome of the treated if they were not exposed to the treatment, cannot be observed directly. However, we can observe E(y0|x, w=0 ), the expected outcome of the untreated, given that they were not exposed to the treatment. Subject to the assumption of no selection bias, in the absence of the program, those who participated in the program would have had equal outcomes to those who did not: Equation 5 0)0,|()1,|(00wyEwyExx 15 However, if program selection has not been made randomly then selection bias occurs, and individuals exposed to the treatment will systematically differ from those not exposed to the treatment. Hence, program impact appears as a consequence of these differences, distorting the measure of the benefits from the program. Selection bias can be a consequence of difference in characteristics between participant s and non-participants: Some differences can be observed by the researcher, such as housing characteristics, land allocated to agricultural production, and topographical location of fields. These characteristics are by the program, and they determined eli gibility for program participation. Other differences are not observed by the researcher and can be assumed not to change over time, including such individual characteristics as motivation, cognitive learning ability, and attitudes towards innovation. To e stimate the ATE we use difference in difference ( DID ) estimation . In order to estimate program impacts, we compare the ATE to two measures of the ATT, propensity score matching difference -in-difference (PSM -DID) and propensity score weighted regression (PSW) (Heckman, Ichimura, & Todd, 1997; Smith & Todd, 2005) . When eligibil ity of treatment is not random, the ATE and the ATT can differ , but as will be shown later in our estimation results, in this case the ATE and the ATT are identical . In the following sections , we explain how DID, PSM -DID and PSW control for different sour ces of selection bias. A more detail ed description of the problem of impact evaluation, an overview of the different methods for impact evaluation and the methods used in this paper can be found in Appendix C. 16 2.4.1 Propensity score based methods: Propensity score matching (PSM) consists of choosing the comparison group according to the probability of being selected for a treatment, given a set of observable pre -treatment characteristics and outcome values that do not change with program intervention but that affect program placement. The main assumptions for propensity score matching are: 1) Unconfoundedness: Equation 6 x|,10wyy where y0 is the outcome for non -participants and y1 is the outcome for participants, w is participation and x represents a set of variables that may influence participation. P rogram outcomes are independent of program participation, conditional on x. 2) Mathematically, there is common support (overlap) between the probability distributions of program participants and non -participants (Caliendo & Kopeinig, 2008; Imbens & Wooldridge, 2008; Martin Ravallion, 2008) : Equation 7 1)|1Pr( 0xw To estimate the propensity score (PS), we include a rich set of variables that determine both participation in the project and pretreatment outcomes to reduce bias in estimates (Heckman, Ichimura, Smith, & Todd, 1998) . Propensity score ma tching assumes that after controlling for observable characteristics, outcomes are mean independent of participation in the program. But it is likely that there are systematic differences in outcomes for participants and non -participants due to unobservabl e characteristics, known as bias on unobservables. 17 Assuming that unobserved heterogeneity is time invariant and uncorrelated with treatment assignment, we can control for this source of bias using the PSM -DID estimator defined by Smith and Todd (2005) . By using the PSM -DID estimator we control for observable sources of bias by building our comparison group using PSM as well as time invariant characteristics, by taking the difference of outcomes before and after treat ment. The PSM -DID estimator, defined by Smith and Todd (2005) , is as follows , Equation 8 ppSIjititititSIiDID PSM ATT yyjiyyN01))(,()(11‹1010111, As an additional robustness check, we compare the matching estimates with the propensity score weighted (PSW) regression (Wooldridge, 2010) , in the panel data context we take the difference between outcomes before and after treatment: Equation 9 NiiititiiPSW ATT yywN11,))r(P‹1(‹))(r(P‹(1‹xx For Equation 8 and Equation 9 the subscripts 1 and 0 refer to treated and untreated respectively, Sp refers to the common support , t refers to the time period , N to the total number of observations, (.) is a weight that depends on the matching method used, Pr(x i) is the propensity score and refers to the proportion of treated observations in the sample (N1/N). 18 2.4.2 Regression based methods. The main assumption of DID is that the unobserved differences between participants and non -participants are invariant in time. Examples would be particular individual characteristics like motivation and cognitive ability . By taking the first difference we removed time invariant unobservable characteristics . Then obtaining the difference between periods t and t-1, the unobservable characteristics, assumed invariant in time are eliminated, correcting for this source of bias in the program impact estimation (Wooldridge, 2010) : Equation 10 itititituwyx0 where yit=yit-yit-1, xit=xit-xit-1 and uit=uit-uit-1. We obtain the program impact by the regression of the change in the outcome variable y the project participation variable w, and the change in a set of time varying covariates x. The first difference equation will be consistent if E( xituit)=0. The parameter of interest to estimate the A TE is . The difference in difference estimator assumes parallel trends for both treatment and control in the absence of the treatment (Abadie, 2005) . Therefore, correct ing for differences between the two groups require s control ling for covariates related to household characteristics (Abadie, 2005). To take care of possible differences of covariates between treatment and control , we include some time var ying household cha racteristics as in Equation 10 for estimating program impacts . 19 2.4.3 Heterogeneity of program impacts. Our study focuses on t he average effect of a program on the treated. Yet the average can miss program impacts that vary among subsets of individuals or households. Even if our results on the program average effect for some outcomes are not statistically significant, given the wide range of interventions within A4N, households with certain characteristics might have benefited differentially. For example, the poorest groups might have benefited from most of the project interventions, or to the contrary, the better off beneficiaries might hav e gotten the most from the project. This analysis is conducted for different groups identified in the sample, according to a pretreatment indicator of wealth or income generating capacity. We estimate project impact on outcome y for each of group g. Equa tion 11 gitgitgitgituwy,,,0,x Equation 11 is identical to Equation 10, but Equation 11 includes the term g, which designates group g. These groups will be identified according to a pretreatment indicator of wealth or income generating capacity. The parameter g is the parameter of i nterest Šthe estimate of project impact on outcome y for each of group g 1. 1 The heterogeneity of program effects can also be estimated as follows: itigiigititituDwDwyx0 Where the parameter g indicates the different groups, and D a set of dummy variable to identify each of these groups (excluding the one used as reference group). The parameter of interest is given by , the interaction of each group dummy variable with the treatment variable w, which is the equivalent of obtaining g in Equation 11. 20 2.5 Survey data use for evaluation of impacts The dataset was based on two -stage sampling of treatment and non -treatment villages, where fitreatmentfl refers to being offered the package of interventions under the A4N project. We randomly selected villages from the list of beneficiary villages, and chose similar non -participant villages using the population and agricultural census data from Nicaragua . A detailed d escription of the sample design can be found in Appendix D. The sampled villages were selected according to the population weights of each of the municipalities where the project intervened. Non - participant villages were identified according to natio nal census data on poverty levels, as measured by the index of unmet basic needs, the importance of staple crops, small landholdings (Instituto Nacional de Información de Desarrollo, 2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g, 2008h) , and location in the same agrarian zones (Nitlapan, 2001) . From each vi llage we randomly selected 10 treated households in the participant villages and 10 to 15 households in the non -participant villages, depending on village size. In A4N participant villages, CRS provided lists of treated households. We collected data in 201 0, after the project starting date in August 2009. In non -participant villages, sample lists were developed in consultation with village leaders, who were requested to identify households that would meet the eligibility criteria of the A4N program. We foun d no eligible but un treated households in the treatment villages to include in the sample. A baseline survey measured livelihoods and income for the agricultural year 2008 -09, before and during project implementation (project started activities in August 2009) . A follow up round of the survey did the same for the agricultural year 2010 -11, the second year after project implementation. The survey also collected information on the different technologies and practices im plemented by farmers in their plots. The survey was conducted in the departments of 21 Estelí, Jinotega and Matagalpa, located in the northeast of Nicaragua . A list with the information included in the s urvey can be found in Appendix E, the household survey instruments both in English and Spanish can be found in Appendix F, and the village survey can be found in Appendix G . The final balanced panel includes 578 households, 284 in participant villages and 294 in non -participant villages (See Table D5 in Appendix D ). The abandonment rate between the two ro unds of the survey was 7 %, and we did found no evidence of systematic attrition (See Table D 4 in Appendix D ). More non -participant households were interviewed intentionally, in order to permit the trimming of observations when applying propensity score ma tching. A survey of village characteristics was conducted among village leaders in each of the 63 villages . The data set was reduced from the original set of 578 observations due to dropp ing two outliers, for a total of 576 observations. For the PSM -DID and PSW analysis, missing data for the estimation of the PS (11 observations) and the trimming of observations with PS above 0.90 and below 0.10 (11 observations) was conducted (Imbens & Wooldridge, 2008; Wooldridge, 2010) . The total number of observations used for the PSM -DID and PSW analysis is 554 . 2.6 Results: A4N project impacts. The estimation of project impacts starts w ith estimating the probability of participation in the project using a logit model. These estimated probabilities will later be used for propensity score matching. Balancing tests after matching are presented to measure the degree of differences between tr eatment and control households. Then we show the estimated impacts for intermediate outcomes related to the adoption of the technologies and practices promoted by the project. Finally, we estimate project impacts by terciles of area of cultivated land. 22 Project treatment effects were estimated using DID , PSM -DID and PSW. The point estimates are very similar for most of the outcomes across the methods used. We present these results showing first the regression approach with DID and compare these results with PSM-DID and PSW in order to compare regression -based method results with PS based methods results. The DID estimation includes as control variables household size, average of years of education of household members and cultivated land 2. Then we estimate program impacts using PSM -DID kernel Epanechnikov (kernel(epan)) , nearest neighbor with replacement, using five neighbors (NN(5)), and local linear regression with the tricube kernel (LLR), to conduct sensitivity analysis of the matching results. We estima ted program impact using the difference in the outcome variables before and after the project as dependent variable, for both continuous and binary outcomes. Treatment refers to whether the household was exposed to the package of interventions promoted by the project 3. Before presenting the results for the average treatment effects, we present the estimation for the propensity score of probability of participating in the A4N project. 2 We also conducted fixe d effects estimation, and the results did not differ from the DID ones. Therefore we consider that violation of the strict exogeneity assumption is not a concern (Wooldridge, 2010) . 3 Information on participation in other projects was collected in one of the household survey questions. To test for attribution to the A4N project of impacts that are due to other projects, we estimated the correlation of participation in A4N and participation in other development projects. We found no correlation ( =-0.03), so misattribution is n ot a concern. We also estimated DID including a dummy variable for participation in other projects and did not find this variable statistically significant. 23 2.6.1 Propensity score estimation The probability of program participation or propensity score was estimated using a logit model with the data from 272 treated and 282 non -treated households . Upon application of Dehejia and Wahba™s (2002) algorithm for estimating the propensity score (see Appendix H), it was determined that no interaction terms and higher level terms were justified to improve th e estimation, so the logit model was estimated with all covariates entering linearly. The logit model estimates the probability of program participation ( Table 1). Focusing on variables that are statistically significant (p -value les s than 0.10), the A4N households were more likely to be female -headed and to have lower value of farm infrastructure but also less inadequate services as defined by the basic needs index (housing lacking piped water and interior toilet ). A4N households ten ded to be situated in villages closer to markets but with fewer large farms and less likely to have a health facility . These variables reflect some pretreatment differences between treatment and comparison households (see Table I 1 in Appendix I ). A detailed description of the pretreatment characteristics of the trea tment and comparison households can be found in Appendix I. 24 Table 1. Logit model for estimating the propensity score or probability of participation in A4N. Dependent variable: Participation in A4N Explanatory variables Coefficient Standard error Farm characteristics Cultivated land Mz 0.03 (0.03) Steep slope=1 0.18 (0.20) hh characteristics Inadequate services=1 -0.51** (0.22) Inadequate housing=1 0.11 (0.29) Electricity=1 -0.05 (0.22) Hunger=1 0.34* (0.20) Head female=1 1.19*** (0.31) Children under 5 years old (number) 0.06 (0.15) Head age (years) 0.00 (0.01) Head education (years) -0.01 (0.04) Household size (persons) -0.05 (0.06) Persons per room -0.02 (0.06) Value of productive assets Infraestructure (C$/1000 ) -0.09* (0.06) Livestock (C$/1000 ) -0.02* (0.01) Equipment (C$/1000 ) 0.00 (0.02) Village charcteristics Population 2009 0.00 (0.00) Dist. to Market (Km/10 ) -0.05*** (0.01) Dist. to p aved road (Km/10 ) 0.02 (0.01) Health facility=1 -0.82*** (0.26) Farms producing b asic grains 2003 (percentage) -0.18 (0.63) Lanholdings less than 10Mz 2003 (percentag ) 2.25*** (0.50) Constant -0.20 (0.84) Log likelihood -345 n 554 Levels of significance ***1%, **5%, *10% Standard error in parenthesis hh means household 1 Mz=1.73 Acres The exchange rate for 2011 was U$1=C$22.42 Source: A4N Baseline Household Survey 2010. 25 The predicted probabilities of selection into the A4N participant and non -participant groups are presented in Figure 2. The non -participant distribution contains more observations with propensity scores below 0.6, and a disproportionate number of observations with propensity scores below 0.4. In spite of this , there is substantial overlap, so we have comparison observations to match treatment ones. Figure 2. Estimated propensity score or probability of program participation. Matching of participant and non -participant observations using according to the values of the propensity score, was conducted using STATA™s psmatch2 (Leuven & Sianesi, 2012) . The results for the balancing tests (Caliendo & Kopeinig, 2008; Wooldridge, 2010) after matching with replacement are provided in Table 2. Matching improved overlap between the marginal distributions of the covariates. As eviden ce, the percentage bias decreases for the covariates below the benchmark of 25% for covariate balance (Imbens & Wooldridge, 2008) . 0.2.4.6.81Propensity Score Untreated Treated: On support Treated: Off support 26 Table 2. Balancing tests of pretreatment covariate s used for estimation of the propensity score. Before Matching After Matching Mean Mean Variable A4N non -A4N %bias A4N non -A4N %bias Area of c ultivated land (Mz) 3.29 3.50 -2.68 3.32 3.37 -1.6 Steep slope=1 0.32 0.32 -25.87 0.32 0.37 -9.7 Inadequate services=1 0.66 0.79 21.39 0.67 0.66 3.6 Inadequate housing=1 0.88 0.85 60.10 0.88 0.86 4.4 Electricity=1 0.61 0.63 15.32 0.60 0.58 4.9 Hunger=1 0.39 0.32 -17.89 0.38 0.40 -4.2 Head female=1 0.20 0.07 -73.06 0.18 0.22 -12.8 Children less than 5 years old (number) 0.51 0.51 -24.23 0.51 0.41 14.1 Head age (years) 49 48 68 49 49 -1.2 Head education (years) 2.83 3.04 3.13 2.84 2.79 1.7 Household size (persons) 5.20 5.36 49.55 5.20 4.99 9.3 Persons per room 3.82 3.86 40.94 3.84 3.85 -0.6 Infrastructure (C$/1000 ) 0.52 1.48 -11.12 0.53 0.47 3.3 Livestock (C$/1000 ) 6.71 9.07 -17.33 6.80 6.08 5.7 Equipment (C$/1000 ) 1.76 2.08 -49.91 1.80 2.09 -6.2 Population 2009 637 640 16.68 645 678 -5.9 Dist. to Market (Km/10 ) 14.09 16.29 38.19 14.34 14.46 -1.5 Dist. to Paved road (Km/10 ) 9.53 8.95 -0.99 9.56 8.63 10 Health facility=1 0.21 0.28 -37.29 0.21 0.21 0.7 Farms producing basic grains 2003 ( proportion ) 0.86 0.88 71.93 0.86 0.87 -4.9 Landholdings less than 10Mz 2003 ( proportion ) 0.59 0.52 64.34 0.58 0.54 18 1 Mz = 1.73 Acres U$1=C$22.42 100*%*202101jjjjssxxbias Source: A4N Baseline Household Survey 2010, Baseline Village Survey 2010 27 2.6.2 Project impacts on outcomes related to adoption of technologies and practices. With the goal of determining whether there was a project impact in the adoption of promoted practices, the evaluation of intermediate outcomes focuses on six groups of outcomes: (1) agricultural conservation structures, (2) agricultural conservation practices, (3) post -harvest grain storage, (4) kitchen gardens, (5) saving and credit, and (6) food scarcity 4. Table 3 presents detailed definitions of the outcomes to be evaluated (Table 3, for further details, see Table J 1 in Appendix J). Table 4 and Table 5 present the results for the different methods use for estimating program impacts, DID , PSM -DID using three methods Š kernel (epan), NN(5) and LLRmatching Što compare the sensitivity of estimates to different matching methods (Abadie & Imbens, 2008) , and PSW regression. The results are robust to different estimation methods, as can be seen by the similar point estimates and levels of significance obtained for project treatment effects. Overall, the results using DID 5 for estimating the ATE were almost identical to the results using PSM-DID and PSW for estimating the ATT . This was expected because the sampling frame explicitly included a set of control villages and households for comparison with similar characteristics to t he A4N ones according to pove rty and population indicators. The comparison group was similar by construction to the treatment group according to observable characteristics . Additionally, t he ATE and the ATT do not differ in this case because we were able t o collect data on project bene ficiaries for data collection (more on this in Appendix C and Appendix D). 4 We did not conduct impact evaluation on the use of improved maize and beans varieties due to unre liable data on the names of the varieties planted by farmers collected in the survey. 5 For details in the DID estimation results for this section see Appendix K. 28 Table 3. Definition of intermediate outcome variables and units of measurement. Outcome Variables Unit Definition Agricultural Conservation Structures (built between 2009 and 2011) All structures m/Mz Difference length built in agricultural conservation structures 2011 -2009 Stone barriers/terraces m/Mz Difference length built in stone barriers and terraces 2011 -2009 Live barriers m/Mz Difference length built in live barriers 2011 -2009 Ditches m/Mz Difference length built in ditches 2011 -2009 Agricultural Conservation Practices All practices 1=yes, 0=no The household has implemented at least one cons ag practice in one of the plots under its management Minimum tillage 1=yes, 0=no The h ousehold has implemented minimum tillage at least in one plot Zero tillage 1=yes, 0=no The household has implemented zero tillage at least in one of its plots Vermiculture 1=yes, 0=no The household has implemented vermiculture at least in one of its plots Cover crops 1=yes, 0=no The household has implemented cover crops at least in one of its plots Storage Practices hh experienced stored grain losses 1=yes, 0=no The household has experienced stored grain losses. Only for households that stored grain. hh stored grain in metallic silos 1=yes, 0=no The household uses metallic silos for grain storage. Only for households that stored grain Number of metallic silos number Number of metallic silos owned by the household Kitchen Garden hh had a kitchen garden 1=yes, 0=no Household has a kitchen garden Savings and Credit hh has savings 1=yes, 0=no Household had savings on January 1st hh has credit 1=yes, 0=no Household had credit on January 1st Food Scarcity hh experience food scarcity 1=yes, 0=no Household experienced a period of the year when they could not cook one of the daily meals hh means household . 1 Mz = 1.73 acres 29 The construction of agricultural conservation structures and the use of agricultural conservation practices for soil and water conservation increased thanks to the project, as shown in Table 4. Agricultural conservation structures represent significant investments of capital and labor with a gradual payoff. The adoption of their construction under the A4N project was measured by the change in length of rows built structures per unit of cultivated land (meters/ Mz). The information was obtained with a recall question in 2011 on the length of agricultural conservation structures built over the past two years. This question was asked for each of the plots under the management of the household. On average the increase in agricultural co nservation structures was 77m/Mz, measured by first differences ( Table 4); the estimates for PSM -DID and PSW are similar, and all are highly statistically significant . This increase was explained mostly by the increase in area under s tone barriers and terraces (24m/Mz), live barriers (16m/Mz), and ditches (7m/Mz) ( Table 4). Agricultural conservation practices included reduced tillage, vermiculture and cover crops, all three of which are require less capital and la bor than the construction of terraces, barriers, or ditches. The adoption of practices was measured by changes in whether the household was implementing one or more of the practices promoted by A4N on at least one of the plots managed by the household. On average there was not an overall impact in the use of these practices, but there was significant substitution of minimum tillage for zero tillage. The percentage of households using minimum tillage in at least one of their plots decreased by 14%, whereas this percentage increased by 19% for zero tillage ( Table 4). In addition, there was an increase in households implementing vermiculture and cover crops in at least one of their plots. 30 Table 4. Project impacts on construction of agricultural conservation structures and on agricultural conservation practices. PSM -DID Difference outcome variables DID kernel (epan) NN(5) llr (tricube) PSW Agricultural Conservation Structures All structures m/Mz 77*** 76*** 75*** 73*** 72*** (25) (25) (27) (27) (27) Stone barriers or terraces m/Mz 24*** 24*** 23** 22** 24** (10) (10) (10) (11) (10) Live barriers m/Mz 16*** 17*** 17*** 17*** 17*** (5) (5) (6) (5) (5) Ditches m/Mz 7*** 7*** 8*** 7*** 7*** (3) (3) (3) (3) (3) Agricultural Conservation Practices All practices 1 0.04 -0.02 -0.03 -0.02 0.00 (0.05) (0.06) (0.06) (0.06) (0.05) Minimum tillage 1 -0.14*** -0.17*** -0.16** -0.17** -0.15*** (0.05) (0.07) (0.08) (0.07) (0.05) Zero tillage 1 0.19*** 0.19*** 0.20*** 0.18*** 0.18*** (0.0 (0.07) (0.07) (0.07) (0.07) Vermiculture 1 0.05*** 0.05** 0.05** 0.05** 0.04*** (0.02) (0.02) (0.02) (0.02) (0.02) Cover crops 1 0.03*** 0.04* 0.04* 0.04* 0.04* (0.01) (0.02) (0.02) (0.02) (0.02) 1 For binary outcomes the difference takes values -1, 0 and 1 . Levels of significance ***1%, **5%, *10% NN refers to nearest neighbor, LLR to local linear regression untrimmed sample n=567, trimmed sample n=546 A total of 265 pairs formed with PSM -DID Mz = 1.73 acres The project had a significant, positive effect on adoption of metallic silos for grain storage. On average there was an increase of 11% in the share of households using metallic silos for storage (Table 5). Presumably associated with this, the number of households that experienced stored grain losses fell by 11% to 16%, based the four estimates with p -values below 0.15. The 31 increase d use of metallic silos translated into a reduction on stored grain losses wit hin the first two years of the A4N project , and it is possible that project beneficiaries were still in the process of learning how to best apply postharvest management practices to avoid losses. The successful adoption of these practice can lead to furthe r reduction of losses of grain stored for consumption (Gitonga et al., 2013) . The proj ect had a significant impact in the percentage of households with savings, which increased by 14% ( Table 5). This is not an agricultural technology intervention, but this was a very successful intervention of the project that aimed to stabilize income flow over the year and to provide funds in times of household food scarcity. This outcome is mostly a result of the formation of saving and lending groups promoted by the project. S avings gains are likely to reduce vulnerability to asset liquidation in times of food scarcity, and consumption smoothing (Kaboski & Townsend, 2005) . Savings accumulation can also be used for productive investments (e.g., in agricultural assets) (Chowa & Elliott III, 2011) . 32 Table 5. Project impacts on storage practices, kitchen gardens, savings and credit and food scarcity. PSM -DID Difference outcome variables DID kernel (epan) NN(5) llr (tricube) PSW Storage Practices Experienced stored grain losses 1,2 -0.16*** -0.11~ -0.07 -0.13~ -0.11~ (0.06) (0.08) (0.08) (0.09) (0.08) hh stored grain in meta llic silos 1,2 0.11*** 0.10** 0.11** 0.10* 0.09~ (0.04) (0.05) (0.05) (0.05) (0.06) Number of meta llic silos owned 0.14*** 0.13*** 0.12** 0.13*** 0.13*** (0.05) (0.05) (0.06) (0.05) (0.05) Kitchen garden hh had a kitchen garden 1 0.04 0.04 0.04 0.04 0.04 (0.03) (0.04) (0.03) (0.03) (0.03) Savings and credit hh has savings 1 0.14*** 0.13*** 0.13*** 0.12*** 0.13*** (0.04) (0.04) (0.05) (0.05) (0.04) hh has credit 1 -0.01 -0.01 -0.03 -0.03 0.00 (0.04) (0.05) (0.06) (0.05) (0.05) Food scarcity hh experience d food scarcity 1 -0.06 0.04 0.05 0.05 0.04 (0.05) (0.05) (0.06) (0.05) (0.04) 1 For binary outcomes the diff erence takes values -1, 0 and 1 2 Correspond only to the households that stored grain, non-trimmed sample n=476, trimmed sample n=460 Levels of significance ***1%, **5%, *10% , ~ 15%. NN refers to nearest neighbor, LLR to local linear regression hh means household Untrimmed sample n=575, trimmed sample n=554 A total of 265 pairs formed with PSM -DID 33 2.6.3 Heterogeneity if project impacts by area of cultivated land. Continuing with the analysis of project impacts, we look at the distribution of project effects across households of varyin g asset levels. It is possible that even if average treatment effects for the agricultural income and household wealth related outcomes were not statistically significant, some groups benefited more (or less) than others (Khandker, Koolwal, & Samad, 2010) . The sample was divided into approximate terciles using the information on the pretreatment area of cultivated land. Farmland, an important asset, is the key input for agricultural production. The first group is composed of households with less than 1.5 Mz (small area) of cultivated land, the second one with households with between 1.5 Mz and 3 Mz of la nd (medium area) and the third one with households with more than 3 Mz of cultivated land (large area). Table 6 present s the estimated coefficients of average treatment effects for each of the three groups formed using the area of cultivated land in 2009. The DID , PSM -DID and PSW estimates of average treatment effects are all very similar, so for this analysis we simply re port FD, for households in each size category of cultivated land. The DID estimation uses the same explanatory variables as those included in the estimation of overall program effects: household size, average of years of education of household members and cultivated land. 34 Table 6. Project impacts by area of cultivated land on outcomes related to adoption of practices and technologies. <=1.5Mz n=191 1.53Mz n=186 Outcomes Coef se Coef se Coef se Agricultural Conservation structures All structures m/Mz 111 (73) 41*** (16) 74*** (27) Stone barriers m/Mz 3 (27) 27** (12) 31*** (11) Live barriers m/Mz 16 (15) 13*** (5) 18*** (7) Ditches m/Mz 11** (5) 4** (2) 8 (8) Agricultural conservation practices All practices 1 0.20** (0.09) -0.03 (0.08) -0.06 (0.06) Minimum tillage 1 -0.08 (0.09) -0.05 (0.09) -0.30** (0.09) Zero tillage 1 0.20** (0.08) 0.15 (0.08) 0.19* (0.08) Vermiculture 1 0.05** (0.03) 0.02 (0.02) 0.08* (0.04) Cover crops 1 0.03 (0.02) 0.02 (0.02) 0.03 (0.02) Storage Practices Stored grain losses 1 -0.06 (0.12) -0.28*** (0.09) -0.12 (0.09) Stored in metallic silos 1 0.06 (0.06) 0.15** (0.06) 0.10 (0.08) Number of metallic silos owned 0.07 (0.07) 0.16** (0.07) 0.21* (0.10) Kitchen garden hh has a kitchen garden 1 0.12** (0.05) -0.02 (0.04) 0.02 (0.05) Saving and credit Saving 1 0.22*** (0.07) 0.08 (0.06) 0.09 (0.08) Credit 1 0.10 (0.07) -0.03 (0.07) -0.13 (0.09) Food scarcity Experienced period of hunger 1 -0.03 (0.08) -0.06 (0.08) -0.08 (0.08) 1 For binary outcomes the diff erence takes values -1, 0 and 1 Mz = 1.73 acres hh means household Note 1: the total sample of 576 observations was divided in terciles, and for each tercile there was an approximate equal share of treatme nt and comparison observations. Note 2: The heterogeneity of program effects was also estimated by DID for the whole sample including two dummy variables for two terciles of cultivated land and two interactio n terms between those and the treatment variable. The results for the coefficients of the interaction terms and levels of significance were identical to the ones obtained here. 35 The results pointed to notable differences in impact by asset level. Househol ds with large and medium area of cultivated land built higher densities of agricultural conservation structures, whereas households with small area were more likely to increase their use of agricultural conservation practices. On average, households with m edium and large cultivated area built 41m/Mz and 74m/Mz of agricultural conservation structures ( see Table 6). The implementation of agricultural conservation practices in at least one of the plots under the management of the household increased by 20% among the households with small area, and 20% of these households also increased the use of zero tillage. In contrast, 30% of households with larger area decreased their use of minimum tillage, and 19% increased the use of zero tillage ( Table 6). These results are consistent with results of studies about decisions of carrying out agricultural conservation investments, which depend on access to land and labor, as well as land tenure security (Gebremedhin & Swinton, 2003) , indicating that differences in household characteristics matter for household decisions of take up of project interventions. The households with medium cultivated area are the ones most likely to increase adoption of improved grain storage practices and to experience decreased stored grain losses. A total of 30% more of medium area households experienced reduced losses of stored grain, and 16% more of these households stored grain in metallic silos ( Table 6). Households with small cultivated area were the ones to add kitchen gardens and to gain savings. The ATT for households with kitchen ga rdens was not statistically significant for the whole sample, but 12% more households with small land area have kitchen gardens thanks to the project ( Table 6), which in turn helps to improve food security. Also these households are t he ones that take advantage of the creation of savings and lending groups, with a 22% increase in households with savings. 36 These results suggest that household resource constraints may limit adoption of certain practices. Capital is required to undertake the investments in construction of agricultural structures, including the hiring of labor. For households with small cultivated area, practices that do not require this level of investment, such as participation in savings groups or growing small vegetable gardens, constitute practices that they are more likely to adopt. Finally, we also analyzed project impacts on agricultural income and change in household wealth. Looking at both overall project impacts and impacts by area of cultivated land, the project had no statistically meaningful impact on these outcomes. This finding is not surprising just two years after project implementation. If we did not think carefully about the project timing and the time lapse needed for impacts to occur, we might have conc luded that the project had no impact. The detailed analysis of these analyses is provided on Appendix L. 2.7 Conclusion Using different methods, DID , PSM -DID and PSW, we find identical results. Stability of project impact estimates across the methods used was expected , due to careful design of the impact evaluation with comparison households selected to construct a valid counterfactual for analysis. We focused on the adoption of improved agricultural technologies to measure changes in behavior, as early indicators of proj ect impact. We found that adoption did increase for many of the technologies promoted. If these behavioral changes are maintained over time, they are likely to translate into increases in agricultural productivity and agricultural income by several mechani sms: Investments in agricultural conservation structures and adoption of agricultural conservation practices are both likely to lead to long -term stabilization of yields. Adoption of 37 improved storage technologies, the associated reduction in the number of households experiencing stored grain losses, and increases in households with savings should all lead to more stable, rising cash flows and reduced of risks of food scarcity and asset liquidation. However, rates of adoption of project technologies were not the same across households of different asset levels. The analysis of project impacts by farm size reveals that they vary according to the household™s area of cultivated land . Hence, the targeting of project interventions by participant asset level can i ncrease rates of adoption of practices by tailoring interventions to household resources. Such an approach could increase project impacts for different groups of beneficiaries, instead of promoting all the interventions for all the beneficiaries Ša more cos t-effective strategy. An important recommendation from this impact assessment is that the heterogeneity across project interventions of the expected time lapse before participants experience benefits should be considered both for project design and for im pact evaluation. As shown here, the realization of gains for some interventions (e.g. construction of stone barriers and terraces) takes much longer than others (e.g. storage in metallic silos). Therefore, development projects that promote multiple interve ntions may want to set poverty relief objectives that explicitly incorporate the timing of expected benefits from a doption of specific practices. In an environment of donor impatience to see rapid impacts, such an approach would calibrate donor expectation s to a realistic sequence of intermediate impacts that culminate in long -term desired outcomes. 38 Chapter 3 Trust and Group Participation in Rural Development Activities 6 3.1 Introduction Many studies have suggested that important linkages exist between trust and develop ment. At the national level, trust has been shown to have positive robust effects on income (Baliamoune -Lutz, 2011) and has been shown to play a role in increasing investment. Resea rch also indicates that high levels of trust can encourage group formation and improve coordinati on in order to carry out research and innovation projects (Dearmon & Grier, 2009) . Finally, trust has been found to improve human development (Özcan & Bjørnskov, 2011) and is associated with the adoption of enviro nmental sustainability projects (Owen & Videras, 2008) . At the community level, trust has been linked with effective and sustainable management of natural resources (Bouma, Bulte, & van Soest, 2008) . Generalized trust (that is, trust towards strangers) is typically seen as reducing transactions costs via facilitating information sharing and increased efficiency (Fafchamps, 2006) . But another complementary view of trust, called personalized trust, is formed through repeated interaction among non strangers. Putman (2003) notes that studies of rural development have shown that a fivigorous network of indigenous grassroots associations can be as essential to growth as physical investment, appropriate technology, or figetting prices right .fl. Personalized trust impacts economic growth and is require d for sustained economic development. Knack and Zak (2003) develop a general equilibrium model to illustrate that when trust is lower the amount invested by 6 This chapter was written with the collaboration of Professor Robert Shupp ; it is a coauthored work submitted for journal publication. 39 economic agents is low, which affects savings, and that these low levels of savings would not be sufficient for sustained output growt h. In addition, Grootaert & Narayan (2004) find that trust, measured as membership in different groups or organizations increases household welfare and reduces pov erty. Personalized trust appears to be important in improving individual and group quality of life, especially in rural development situations. In fact, development projects frequently use the strategy of group formation to promote project interventions such as producer groups that aim to correct for market failure and savings and lending groups. These groups are formed and encouraged by development projects for two primary reasons: 1) because the success of the intervention in some way relies on group pa rticipation and cooperation (for instance, savings and lending groups or sustainable management of a resource), or 2) because groups make it easier to disseminate the intervention and possibly improve its effectiveness via information sharing among partici pants. While these are good reasons to encourage group formation in development projects, we suggest (as others have before) that there is an additional potential benefit from group formation, which is the encouragement and development of increased trust a nd social capital and its potential benefits. For example, informal microfinance groups such as Rotating Accumulating Savings and Credit Associations, or ROSCAs, have been shown to have impacts beyond correcting for financial market failure s. Etang, Fieldi ng and Knowles (2011) find evidence of the impact of participation in ROSCAs on trust in Cameron. Other benefits generated from these groups extend to the provision of social securit y and insurance, physical and institutional infrastructure, recreation, community development, and health and education (Bouman, 1995) . 40 In general, groups, through increased trust, have the potential to increase incomes, and generate empowerment and political action, which can help the poor escape poverty (Thorp, Stewart, & Heyer, 2005) . Group formation provides the experience of working together , enhancing trust and enabling individuals, not only to work towards current goals, but also to work together in the future towards other personal or community objectives. Enhancing trust and building long lasting capacities should enable development proje ct beneficiaries to continue working jointly towards common goals. If this is the case, then while rural development projects frequently rely on the existing social capital and trust in target areas, they should also focus on generating greater levels of t rust and social capital among said beneficiaries. If accomplished, the beneficiary groups and their community are likely to be less dependent on the presence of an external development agent to continue working together and in this sense, the impacts of th e project Œ both through its interventions and through the more general impacts of increased trust Œ are more likely to last after the project is over. Assuming that increased levels of trust can potentially alter the success and longevity of rural develo pment project impacts, then the question becomes whether the sort of groups typically used and promoted by rural development projects, such as producer groups and savings and lending groups, actually improve levels of trust and potential cooperation and so cial capital among beneficiaries and their communities, or whether development projects should incorporate further interventions specifically focused on improving trust and cooperation. In this study, we measure how group participation in a rural developme nt project affects levels of trust. Trust has traditionally been measured in two different ways: 1) through survey questions, such as the Generalized Social Survey (GSS) trust questions designed to measure generalized trust and 2) via individual behavior i n incentivized trust game experiments. The trust game has been used 41 extensively to measure trust under different settings with different groups of participants (Cardenas & Carpenter, 2008; Danielson & Holm, 2007; Gächter, Herrmann, & Thöni, 2004; Schechter, 2007; Vollan , 2011). In addition, the two methods have been used jointly to compare and contrast the two methods in different settings (Capra, Lanier, & Meer, 2008; Etang, Fielding, & Knowles, 2012; Gächter et al., 2004; Glaeser, Laibson, Scheink man, & Soutter, 2000; Johansson -Stenman, Mahmud, & Martinsson, n.d.) . The GSS trust question asks, fiGenerally speaking, do you consider that most people can be trusted, or that you cannot be too careful in dealing with people?fl In studies conducted by Capra, Lanier and Meer (2008) , Gatcher, Herrmann and Thoni (2004) , Glaeser, Laibson, Sheinkman, and Soutt er (2000) , it has been found that attitudinal questions that make reference to a specific group of people, such as fistrangersfl, rather than to fimost peoplefl, as in the GSS trus t question, tend to be better predictors of behavoir in trust games. That said, these studies also argue that both attitudinal questions and economic experiments are complementary, rather than exclusive methods for measuring trust, since they allow for che cking consistency of answers to survey questions with behavoir. In this study we use both methods. Specifically, we apply the trust game and survey questions to investigate whether farmers involved in group based interventions promoted by the fiAgricultu re for Basic Needsfl (A4N) project in Nicaragua reveal different levels of trust than farmers who were not exposed to the A4N group interventions. We explore these effects on trust levels among farmers in the same village. We use the trust question from the GSS to measure farmers™ levels of trust towards people in general and an additional attitudinal question to measure trust towards people in the same village. Our implementation of the trust game experiment follows Berg, D ickhaut and McCabe (1995) ; it is a one shot, double blinded design 42 where bo th sender and receiver do not know who they are paired with. We use the proportion of the endowment sent as a measure of trust and the proportion returned as a measure of trustworthiness. Overall, we find evidence that participants involved in the group -based interventions of a rural development project (as represented by the A4N groups) have higher levels of personalized trust than participants who were not involved, but the evidence is weak. Our findings suggest; 1) the need for further investigation of group based interventions on trust and 2) that, if rural development projects are interested in increasing levels of trust among project participants and communities, specific interventions designed at increasing trust levels may be required. We also find that women are more trusting than men and that increased levels of education are associated the lower levels of trust as measured by proportion of endowment sent. The rest of the chapter is organized as follows. First we provide a short description of th e A4N project and its strategy of group promotion. Second, we describe our experimental design, and then we present the results. Finally we discuss the results and draw conclusions. 3.2 The Agriculture for Basic Needs project (A4N) Catholic Relief Services ( CRS) and its partners, Caritas and the Foundation for Research and Rural Development (FIDER), implemented the Agriculture for Basic Needs (A4N) project in Nicaragua during 2010 -12. The primary aim of A4N was to provide rural low -income farmers with a set o f skills for achieving sustainable farm production and increased agricultural incomes. The A4N project worked mostly with smallholder farmers. The average participating farmer has about 5.2 acres of land and grows mostly maize and beans. Farms with livesto ck (93% of total) 43 primarily raise poultry and small animals, and on average use 50% or more of their agricultural production for home consumption. To accomplish the program objectives, A4N interventions promote agricultural conservation and nutritious crop s, improved crop varieties, animal husbandry (for poultry and pigs), integrated pest management and practices to diminish post -harvest crop loss. Other program interventions include saving and lending groups, post -harvest processing, expanded participation in markets, and promotion of farmer innovation groups. The A4N project was initiated in August of 2009 and was formally completed in August of 2012. Given the project™s goals as described above, the A4N project targeted villages that are considered poor. As such, the villages involved in A4N were characterized by high levels of unsatisfied basic needs, are typically located in areas of natural resource degradation, and are highly vulnerable to extreme weather events such as landslides, drought and excessi ve rain. The overarching strategies for the A4N project were to promote group organization and interaction, to build capacity in saving and lending, to introduce enhanced agricultural technologies, and to provide technical assistance to farmers. As such, farmers in the A4N villages were invited and encouraged to form groups focused on one or more of the following project supported objectives: saving and lending, learning sustainable agricultural technologies, and innovation and learning. Once a farmer group was form ed, the A4N project provided technical and financial support in the form of training in agricultural technologies or in microfinance, depending on the kind of group, supplies such as agricultural inputs for plot trials, record -keeping books, and financial support for group initiatives such as starting a new business. In addition, each group was assigned a project field officer who regularly attended group meetings to assist with group organization and to provide training as needed. While the field 44 officer w as there to help, it is important to note that each group determined its own direction within the confines of the project, with the hope that the groups would promote interaction between farmers and illustrate the advantages of working together to achievin g joint goals successfully. While group members received training and assistance from the project, members were active in setting group objectives and in determining group needs. In this sense, the A4N groups went beyond getting farmers together for train ing activities by encouraging members to actively participate in setting and achieving goals. 3.3 Experimental design and procedures The main objective of this research is to investigate the possibility that participation in A4N type intervention projects, t hat is, those focused on improving farmer income through improved village level investment and the transfer of enhanced production techniques via direct training and participation in farmer groups, may also lead to increased levels of cooperation and coord ination via increased levels of interpersonal trust. Given this focus, our experimental design involves two treatments that vary only in whether participants were involved in the A4N group based interventions or not. Specifically, we implement trust experi ments (see description below) in eight villages in Nicaragua, half of these villages where involved in the A4N project while the other half were not (Table 1 lists participating villages along with number of participants). This study was conducted under a project that is evaluating the economic impact of the overall A4N project in agricultural incomes and household wealth. Under this project we also collected secondary information on the Nicaragua™s population and Nicaragua™s agricultural census, and 45 prima ry information from a household survey. Using pre -treatment characteristics (2009) from both sources of secondary and primary data, we selected the villages where we conducted the economic experiment. For the A4N villages we used A4N project data to elabor ate the lists of subjects to be invited to the sessions. For the non -A4N villages we elaborated lists with village leaders on eligible subjects to participate in the economic experiment. The non -A4N villages selected to conduct the economic experiments w ere chosen to have similar characteristics to A4N villages. The eight non -A4N and A4N villages were selected such that they had similar socio -economic characteristics. Four A4N villages were randomly selected from a group of 13 villages, and four non -A4N v illages were selected from 10 non -A4N ones. Comparison of the larger groups of 13 A4N and 10 non -A4N villages using data from Nicaragua™s population census 2005 and Nicaragua™s agricultural census 2003 (see Appendix N) shows that the null hypothesis of equal means cannot be rejected for a range of household wealth indicators . The eight villages included in the experiments were located in the Department of Estelí in the same agro -ecological zone. They were similar in terms of access to water, sanitation and electric power, as well as in area of landholdings and production of basic grains. For these eight villages we did not conduct t -tests for equal means due to the small number of observations. The characteristics for the eight villages where the games took place can be seen in Appendix N . Apart from matching A4N and non -A4N villages on general characteristics, we used two other selection criteria. Firs t, we selected villages such that they were geographically separated in order to minimize the possibility that farmers in one village might discuss the activity with farmers in another village. Second , we selected A4N villages that had more than one A4N gr oup 46 formed (see Appendix O), with the goal of avoiding only having participants from a single group in a given session. The correlation coefficient between the number of groups per village and the village population is 0.41, indicating that a higher number of groups is associated w ith a higher population in the village ( Appendix O). Table 7. Village pretreatment characteristics. Village Population 2005 Households 2005 % inadequate housing 2005 % houses no electricity 2005 % houses no piped water 2005 % households produce basic grains 2003 % farms with landholding >10 Mz 2003 A4N: Las Gavetas 156 39 41% 5% 86% 100% 25% Rosario Abajo 613 126 74% 88% 53% 8% 45% Tomabu 585 128 74% 30% 96% 100% 78% Las Cuevas 699 114 60% 18% 95% 91% 52% Non -A4N: Las Puertas 200 38 67% 86% 84% 100% 38% San Lorenzo 296 58 34% 41% 87% 73% 23% Las Lajas - - - - - 73% 40% El Quebracho 125 25 93% 53% 17% 94% 58% Source: Instituto Nacional de Estadisticas y Censos (INIDE) Nicaragua Ministerio Agropecuario y Forestal (MAGFOR) Nicaragua A total of eight sessions (one session in each selected village) were conducted during May of 2012 with between 17 and 22 farmers participating in each session (see Table 8). For sessions in A4N villages, farmer participants were recruited randomly from community lists of farmers participating in groups promoted and supported by the A4N project. Similarl y, for sessions in non-A4N villages, farmer participants were chosen randomly from community lists of farmers with similar demographic characteristics to the A4N farmers. 47 Table 8. Session Villages and Number of Participants. Village Participants A4N Las Gavetas 19 Rosario Abajo 21 Tomabú 20 Las Cuevas 17 Non A4N Las Puertas 17 San Lorenzo 20 Las Lajas 22 El Quebracho 17 The trust game used in this study is based on a version of the trust game developed by Berg, Dickhaut and McCabe (1995) . This version of the trust game is a one shot game, with no communication , where all participants remain anonymous in that they do not know whom they are playing with. As in most trust games, participan ts are di vided into two types (senders and receivers) and each sender is paired with one receiver. In addition, both senders and receivers are given equal initial endowments. The sender is then asked to decide what portion of their endowment they would like to send to the receiver. The sender can send all or none and knows that whatever portion they do not send they will get to keep. The sender, and receiver, also knows that the amount sent (or invested) is, in this case, tripled before it is given to the receiver. For example, if the sender sends $10, the receiver will get $30. In this way, the receiver will now have their endowment plus three times what the sender sent. In the second step of the game, the receiver can return some amount of what they have (endowmen t plus three times what the sender sent) back to the sender. Clearly, the Pareto optimal outcome is for the sender to send all of their endowment and have it tripled as this creates the largest pot of money for receiver and sender to divide (that is, the total would be four times the initial endowment). If the receiver behaves in an equitable fashion, they would return half and both sender and receiver double their money relative to their initial 48 endowment. On the other hand, using backward induction, and assuming each player seeks to maximize his monetary self -interest , the Nash equilibrium involves the sender not sending anything under the assumption that the receiver will be selfish and return zero. As has been shown repeatedly, neither of these outcomes represents actual behavior. Cardenas and Carpenter (2008) report results from trust games conducted in developing countries where senders sent, on ave rage, between 30% and 73% of their endowment and receivers returned between 18% and 50% of what was available to them. This implies that senders are, to some extent, willing to fitrustfl that receivers will be fair (or more accurately not completely selfish) and will return some amount. Typically, the amount sent by senders is interpreted as a measure of fitrustfl Œ that is the greater the proportion of their endowment senders send, the higher their level of trust or confidence. Similarly, the amount returned b y the receiver can be interpreted as a measure of trustworthiness. Conducting lab experiments in the field can present challenges and we used helpers and followed procedures similar to those adopted by Lopez and Ramos (Lopez, Maria C, personal communicatio n May 17 th 2012) and Cardenas and Ramos (2006) . Specifically, experimental sessions were held at houses and schools in the selected vi llages. As participants arrived at a session, they were given a randomly assigned subject number to be used for identification purposes throughout the experiment. Once all participants had arrived, a consent form was handed out and read aloud ( Appendix P). The instructions were explained to the participants and examples of different possible outcomes were provided on a paperboard to help enhance understanding and elicit questions (see Appendix Q for game procedures in the field). The instructions were provided to all the subjects together, to ensure everyone had the same information from th e very beginning of the activity. This could affect our results since subjects 49 were in the same room together before being split up. It was emphasized that the examples provided were not the only possible outcomes, and that each of them could make their ow n decisions. Once the examples were completed, the group was split and pairs were randomly selected via subject numbers. At this point, participants were asked not to talk about the game or their decisions either during the session or after. We explained c learly to subjects that their decisions would remain anonymous (that is, that none of the other participants would know who they were paired with) both during and after the game. The endowment of C$100 Nicaraguan Córdobas (C$), approximately $4.30 , and pot ential earnings from the experiment are not insignificant in that C$100 is approximately the daily agricultural worker wage in the area. To make sure the participants were clear that they would be paid in real money, it was also emphasized that the fake bi lls were just to be use during the game, and that the fake bills would be replaced by real ones of the same denomination at the end of the experiment. As noted earlier, experiment helpers were used during the sessions due to the expected literacy level of participants. The average level of education in the study site is 4.5 years of schooling. In a rural field setting such as this, several explanations of the instructions are required to ensure understanding of the activity. The helpers played three primar y roles. First, they circulated during the explanation of the experimental instructions and examples helping to give further explanations and answer questions in simple language. This helped avoid having participants discuss the activity with each other be fore making their decisions. Second, when senders and receivers were called individually to make their decisions, the helpers sat with them and went over the instructions again and helped them complete their decision if required. Finally, the helpers acted as enumerators while participants filled in a short survey, which we discuss below. 50 It should be noted that these helpers were not from the study site, and not known by villagers. We recognize that having helpers when subjects are making their decisions i s likely to influence their choices, but as mentioned it was required to ensure understanding of the activity. Below are the specific procedures (after providing instructions and examples) for an experimental session (instruments use in the field can be fo und in Appendix R): Groups were divided into senders and receivers. Groups were sent to separate rooms along with a helper. In the sender room, participants were randomly called one at a time (by participant number) to a separate, private area where a helper assisted them while they made their decision about how much from their endowment to send to the receiver. Each sender w as given a total of C$100 ($4.3 ), in 10 fake bills of C$10 and two envelopes Œ one for them to keep while the other would be sent to the receiver. The sender then decided how to split their endowment, they were told to put in a white envelope the amount the y were going to send to the receiver. They were also given blank fake bills, to put together with the amount the ones that account for the amount that they were sending. They were told to put what they were keeping on a color envelope, to take that envelop e with them and were told not to open, share or exchange this envelope with anyone else. After all senders finished making their decision, the envelopes were taken to a separate room and the amounts sent were recorded and tripled and the envelopes were tak en to the private area where now receivers were called for making their decisions. Each receiver was randomly called by participant number, to a separate and private area where a helper handed the enveloped with the tripled amount to the receiver, the helper reminded the 51 receivers that the amount that was in the envelope is what was sent to them by the sender multiplied by three and that now they have a total which was equal to the amount in that envelope plus the endowment of C$100 they already received. The helper reminded the receiver the instructions of the game, and assisted each rec eiver while they were making their decision. They made sure that each receiver knew they could send the amount they wanted and that this decision was going to be kept confidential. Helpers explained to the receiver that from the amount available to them, t hey were to put what they were returning in a white envelope, and keep the rest. After each of the receivers made their decisions, subjects were asked to stay to fill out a short survey on socioeconomic characteristics and farming activities. After filling in the survey they were called one by one to a separate room to receive their earnings of the game. We conducted the survey in two parts (survey instrument can be found in Appendix S). First, in order to make sure that the trust questions included in our survey were answered before both senders and receivers got any feedback on their decisions, we gave the trust question portion of the survey to receive rs while senders were making their decisions. Similarly, while receivers were making their decisions, a helper in the sender™s room distributed and helped senders answer the trust questions. During all parts of the survey, the helper read aloud the questio ns and possible answers while participants marked their answers with an X on an answer form. The questions and answer choices were read several times, and further explanation was given if participants were still in doubt of the meaning of the questions and answers they were presented with. 52 The trust questions on the survey included a trust question taken from the General Social Survey (GSS). The GSS has been conducted yearly since 1972, and it has been used as source of data for studies on societal trends. The same trust question has been asked regularly since the survey was launched and this question has been used extensively to relate state trust with revealed trust measures (Capra et al., 2008; Gächter et al., 2004; Glaeser et al., 2000) . In addition to this GSS trust question, we included a question regarding trust attitudes towards people in their villages. Both questions are listed below. Generally speaking, do you consider that most people can be trusted, or that you cannot be too careful in dealing with people? 1 Most people can be trusted 2 You cannot be too careful when dealing with people People in your village trust most peo ple in your village. 1 Strongly disagree 2 Disagree 3 Neither agree nor disagree 4 Agree 5 Strongly agree The second part of the survey, conducted at the end of the activity, included questions on socioeconomic and demographic characteristics of participa nts in the trust game ( Appendix S). After each participant finished the survey with a helper, they went individually to a separate private area to receive their earnings from the game, they were asked to leave the place quietly and to not to talk with other participants about the game and t heir earnings. 53 In the following sections we will present the results from the survey questions and the trust game. Note that the analysis focuses primarily on sender behavior (measure of trust) and not receiver behavior (measure of trustworthiness) becaus e of the one shot nature of our experimental design. 3.4 Results 3.4.1 Subject characteristics As shown in Table 9, on average, A4N and non -A4N subjects do not differ in most socioeconomic characteristics, implying that both A4N and non -A4N participants were drawn from the same population. However, the groups do differ in terms of gender. The A4N group is 29% men while the non A4N is 47% men. We sent invitations to a balanced proportion of men and women, however it turns out that more women s howed up in the A4N villages in comparison to the non -A4N villages. The other significant difference between the groups was in terms of the percentage of subjects who stated they were members of a group or association (100% in the A4N group vs. 36% in the non A4N group). Of course, this difference was expected due to the fact that for the A4N group we recruited only subjects that had participated in an A4N group based intervention. 3.4.2 Stated trust questions. As shown in Table 10, the ans wers to the GSS trust question indicate that participants in both treatments tend to think people in general cannot be trusted as only 10% and 13%, A4N and non A4N respectively, answer positively to the statement fimost people can be trustedfl, this level of 54 trust seems small, Etang, Fielding & Knowles (2012) report that in the world values survey for 1999-2002, 35% of Americans and 19% of Africans respondents answer tha t they consider fimost people can be trustedfl. The difference between the two treatments is not statistically significant. In contrast, participants mostly agreed with the statement fimost of the people in your village trust other village membersfl (91% of A4 N and 84% of non A4N agree/strongly agree). Again, although the difference is in the expected direction (that is, participation in group based interventions increased stated perceptions of village level trust) and close to p -value=0.20, the difference betw een the two groups is not statistically significant. As such, with regard to stated levels, participation in group based interventions (as represented by the A4N groups here) does not appear to impact stated levels of trust significantly. Table 9. Socioeconomic characteristics of A4N and non -A4N participants A4N n=76 Non -A4N n=75 Variable Mean Std. Dev. Mean Std. Dev. p value* Group or association member 1.00 0.00 0.36 0.48 0.00 Age (years) 44 13 43 15 0.47 Male g ender 29% 45% 47% 50% 0.02 Years lived in village 33 18 34 19 0.71 Education (years) 5.19 3.78 4.84 4.01 0.41 Household size (number of members) 5.11 2.21 4.97 2.09 0.85 Agricultural sales value C$ 2012 7650 5360 6940 5930 0.57 Cultivated land with maize and beans 2012 (Mz) 1.17 0.70 1.04 0.69 0.30 *p values are for test of equal proportions for binary variables, and for the Mann ŒWhitney U test The exchange rate March 2012 was U $1=C$23.21 55 Table 10. Results: trust questions A4N n=76 Non -A4N n=75 Variable Mean Std. Dev. Mean Std. Dev. p value* General Social Survey question GSS_trust Percentage of participants who answer most people can be trusted 10% 31% 13% 34% 0.60 Agreement with "people in your village trust most people in your village" VILL_trust Percentage of participants who answer agree and strongly agree 91% 29% 84% 37% 0.21 *p values are for test of equal proportions for binary variables, and for the Mann ŒWhitney U test Table 11. Results: trust experiment A4N n=39 Non -A4N n=38 Variable Mean Std. Dev. Mean Std. Dev. p value* Trust game results Amount sent P1 to P2 Amount sent in C$ of 2012 51.03 15.35 45.79 19.12 0.10 Proportion returned P1 to P2 Amount returned divided by amount sent multiplied by three plus initial endowment 0.32 0.12 0.35 0.14 0.30 *p values are for test of equal proportions for binary variables, and for the Mann ŒWhitney U test The exchange rate March 2012 was U$1=C$23.21 56 3.5 Trust experiment results . 3.5.1 Overall results. In this section we first focus on the overall results from the trust game experiment. Recall that sender behavior (proportion of the endowment sent) in the trust game experiments is a revealed (as opposed to stated) measure of trust. As shown in Table 11, on average A4N senders sent more than non -A4N senders, and this difference is statistically significant at 10% level. The average amount sent by participants in the A4N treatment was C$5 1 (51% of their endowments ), slightly higher than half of the endowment that was provided to them. A total of 56% of A4N senders sent half of their endowment to recipients (see Figure 3). Non -A4N senders sent on average C$46 (46%) while 47% sent half of their endowment to recipients. Also note in Figure 3 that higher proportions of non -A4N senders sent less than C$50, while higher percentages of A4N senders sent more than C$50 (except for those senders who sent 100%). In terms of proportion returned by receivers (frequently thought of as a measure of trustworthiness), we find no significant difference (see Table 11). A4N receivers returned on average 32% of their available resources, about C$80, and non -A4N receivers returned 35% or about C$84. The proportion sent and returned that we obtained are consistent with previous results, as pointed out by Alesina and La Ferrara (2002) , who note that in most trust games the sender sends about half of their endowment and the receivers returned about 30% of what they received, more or less the same amount that was se nt to them. 57 Figure 3. Percentage of senders by amount sent, A4N and non -A4N participants in the trust game The exchange rate March 2012 was U$1=C$23.21 3.5.2 A4N vs. Non A4N group analysis. In this section we begin to explore the impact of group membership (both A4N -based and non -A4N based) on trust levels via comparisons of unconditional means. We recognize that parsing the data like this leads to relatively small sample sizes, but we belie ve looking at the data this way allows us to illustrate some interesting characteristics and nuances in the data that may indicate interesting future research possibilities and considerations. The following section, (3.5.3), invest igates the significance of these group effects in a more comprehensive multivariate analysi s. As shown in Table 12 (first row), there are non -A4N treatment participants that are also members of groups (36%) not associated with the A4N project. While we expect these groups to be different in terms of their impact on trust levels of members, we can investigate this directly 010 20 30 40 50 60 10 20 30 40 50 60 70 80 100 Percentage of senders Amount sent A4N Non-A4N 58 by comparing the propor tion sent by non A4N subjects who indicate participation in a group, with the proportion sent by A4N subjects who all participated in A4N based groups. As shown in Table 12, non A4N in group subjects sent 43%, whereas A4N subjects se nt 51% and this difference is significant with p=0.07, suggesting that A4N based groups (with their focus on group directed activities) may indeed be different in their impact on group member trust levels. Furthermore, as shown in table 24, the proportion sent by subjects not in a group (47%) is larger than the proportion sent by subjects in a non A4N group (43%) Œ the opposite of what would be expected if being in any type of group induces higher levels of trust. Finally, we can partially and indirectly a ddress the selection bias problem (that is, that more trusting subjects in the A4N villages were the ones who decided to participate in the A4N group interventions) by noting that in the non A4N villages, the less trusting individuals (as measured by propo rtion sent) were members of groups. Table 12. Group analysis *p values are for the Mann ŒWhitney U test for equal means. A4N Non -A4N n mean sd n mean sd p-value* Proportion sent (group) 39 0.51 0.15 12 0.43 0.18 0.07 Proportion sent Group Non Group n mean sd n mean sd p-value* Pooled A4N and non-A4N 51 0.49 0.16 26 0.47 0.2 0.61 Only n on A4N 12 0.43 0.18 26 0.47 0.2 0.34 59 3.5.3 Determinants of the proportion sent In Table 13, we present the results for a set of multivariate regressions to explore determinants of the proportion sent in the trust game. The first regression (column 1), seeks to determine if socioeconomic characteristics of the senders are significant in explaining the proportion sent. We find that gender and education (level in years and its square) are statistically significant, whereas characteristics such as age, and years in the village are not. With regard to gender, we find that women are more trusting than men in that on a verage, men sent less than women (C$8.6 in regression 1). The evidence from the literature on gender and trust in experiments is not conclusive. Using trust experiments with undergraduate students, Buchan, Croson, & Solnick (2008) , Schwieren & Sutter (2008) , Bonein & Serra (2009) and Chaudhuri & Gangadharan (2007) have found th at men act more trusting than women, whereas Capra, Lanier and Meer (2008) and Garbarino and Slonim (2009) find t he opposite. The impact of education on proportion sent is more complicated. The regression (1) suggests a decreasing relationship, at an increasing rate. E ach additional year of education leads to less being sent (approximately C$4) but the squared term indicates that the more educated people are, the more they send Œ at a rate of C$0.2/year. This implies that at the average years of education (5 years), ceteris paribus, game participants send C$2 less than people with no education. Only after 9 years of education (5.3% of our participants) Œ almost double the average education of our sample Œ does the effect of education become positive. This result is consistent with Schechter (2007) who also finds a negative relationship between education and amount sent in Paraguay. However Etang, Fielding and Knowles (2011) find a significant positive effect between education and amount sent in Cameroon. 60 Regressions 2 and 3 add variables to explo re the impact of group participation. Regression 2 simply adds a dummy variable for participation in the A4N group based interventions 7 while regression 3 adds a dummy variable for participation in group (A4N or not). As shown in table 25, and consistent w ith earlier results, participation in the A4N groups, while not statistically significant at traditional levels, borders on significant with a p -value of 0.15. Even after controlling for gender (and other socioeconomic characteristics), we still find that participation in A4N group interventions increases the proportion sent, albeit with lower significance. Given that women sent significantly more than men overall, some of the difference in proportion sent is driven by the higher percentage of women in the A4N sample. While the revealed trust differences (as measured by trust game behavior) in this study are only weakly significant, another similar recent study by Etang, Fielding and Knowles (2011) , found more significant increases in trust among long term self -directed ROSCA (savings and investment) group members. We suspect that the lack of strong significance in our study (relative to Etang, Fielding and Knowles, 2011) may be related to the fact that the A4N groups in our study are relatively young and therefore the process of enhancing trust is at an earlier stage of development. 7 Since we do not have good instruments for these variables, we are not minimizing selection bias of our estimates, and if there are differences in unobservable characteristics between A4N and non -A4N our estimates could be bias. 61 Table 13. Determinants of the proportion sent Dependent variable: propor tion sent, n=77 Independent variables 1 2 3 4 5 Age -0.001 -0.001 -0.001 -0.001 -0.001 (0.002) (0.002) (0.002) (0.002) (0.002) Gender (man=1) -0.086** -0.073* -0.077* -0.092** -0.086* (0.043) (0.042) (0.044) (0.043) (0.043) Years in village 0.001 0.001 0.001 0.002 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) Education -0.043*** -0.046*** -0.042*** -0.045*** -0.043*** (0.014) (0.013) (0.014) (0.014) (0.013) Education square 0.002*** 0.003*** 0.002*** 0.003*** 0.002*** (0.001) (0.001) (0.001) (0.001) (0.001) In A4N Group 0.052~ (Yes=1) (0.036) Group (non -A4N) -0.059 (Yes=1) (0.055) GSS trust 0.05 (0.061) Village trust -.000 (0.041) Constant 0.641*** 0.625*** 0.623*** 0.632*** 0.642*** (0.120) (0.122) (0.120) (0.123) (0.122) R2 0.1663 0.1871 0.1910 0.1751 0.1663 ***significant at 1% **signifi cant at 5% *significant at 10% ~ p-value=0.15 Robust standard error in parenthesis 8. These regressions correspond to OLS results. On the other hand, participation in a group ( non-A4N ), as represented by the Group variable in regression 3, is not statistically significant. Suggesting that the more trusting individuals are not 8 We clustered standard errors at the village level. This procedure made standard errors smaller and variables such as in A4N grou p significant at 10% level. Due to the small sample size and small number of cluster, this procedure is probably not recommendable. We also conducted bootstrapping using wild bootstrap (Cameron, Gelbach, & Miller, 2008) procedure with 400 repetitions. The confidence intervals obtained were narrow er and the p -values were very close to those obtained using OLS with robust standard errors. 62 necessarily the ones who join groups. Note t hat adding these group variables does not impact the sign or significance of the demographic variables relative to regression 1. Regressions 4 and 5 explore whether subject answers to the stated trust questions can explain proportion sent at the individua l level. Two variables, GSS trust (toward people in general) and Village trust (towards people in the same village), are added in regressions 4 and 5 respectively. While the village trust question results are consistent with the behavior in the trust exper iment overall (that is, non A4N participants state lower levels of trust and send less), the answers do not appear to predict individual behavior given that neither variable is statistically significant. Participation in the A4N group based intervention do es not lead to higher levels of stated trust (as measure by the survey questions) and that the answers do not explain behavior in the trust game. This is, however, consistent with previous findings in studies by Gatcher, Herrmann, & Thoni (2004) , Capra, Lanier and Meer (2008) and Glaeser, Laibson, Sheinkman, & Soutter (2000) . 63 3.6 Conclusion Farmers involved in A4N group based interventions reveal different levels of trust than farmers who were not exposed to these interventions, as measured both by participation in a trust game and through survey questions. Initial analysis of the trust game data suggests that on average those participants in the A4N group based interventions sent a significantly higher proportion of their endowm ent. However, further multivariate analysis reveals that this difference is, in part, attributable to other characteristics such as gender, which has been shown to influence trust levels. If group based interventions increase trust, then perhaps rural dev elopment projects should focus more attention on interventions that promote group formation given their potential to positively impact participant and village level welfare and agricultural incomes , which are primary goals of such intervention programs . In addition to confirming the positive impact of group -based interventions on trust, further research needs to investigate and quantify the significance (and to some extent the existence) of the link between increased trust levels and intervention goals su ch as increased income. This study highlights the emerging generalized link between group -based interventions, trust, and development outcomes and we consider it as an initial investigation to further explore this link in rural development projects. Finall y, further research on the emerging generalized link between trust and productivity, could benefit from inclusion of baseline data (that is, pre intervention) and a better way to deal with potential selection bias. If this emerging link between trust and development outcomes is confirmed and significant, we believe that the impact of group based interventions on trust should be considered more formally in development project design and, consequently, impact evaluations . In particular, in rural developing c ountries setting, where formal institutions are not yet developed and economic 64 activity relies on informal institutions, g roup participation is likely to increase localized opportunities in relatively smaller communities , facilitating network formation and information sharing . Furthermore, the enhanced cooperation and communication among individuals participating in groups, due to increased personalized trust should not only benefit participating individuals, but also their community as a whole. Overall, we believe that this paper illustrates the potential of focusing more formally on building trust and makes the case that the enhancement of trust among beneficiaries of rural development projects should potentially be considered a program strategy designed t o achieve increases in agricultural income and household wealth. 65 Chapter 4 Conclusions In this dissertation I use quasi -experimental panel data econometric methods and a field experiment to evaluate the economic impact s of a complex rural development project in Nic aragua . The research contribut es to the literature on impact evaluation of pro -poor rural development projects with multiple interventions. The quasi -experimental panel data econometric approach is relevant for projects with a design that 1) expose s benef iciaries to more than one intervention at the same time, and 2) allow s project beneficiaries to self -select into different project interventions. The field experiment offers a formal means to measure growth in trust among project beneficiaries, recognizing that community -level trust can reduce transactions costs and enhance economic development. The panel data -based impact evaluation addresses the challenge of measuring the effectiveness of multiple project intervention s over a relatively short period . Focu sing sequentially on how beneficiaries must first change behavior (e.g., by adopting agricultural technologies and practices promoted by the project) that ultimately will lead to increases in agricultural income and household wealth. However, the timing of different interventions and their gestation to impact varies. The time elapsed to project impacts is shorter for some interventions (e.g. improved storage) than for others (e.g. construction of stone barriers) (Figure 2). Timing of project impacts should be considered for impact evaluation, otherwise results could be misleading (e.g. no project impacts on agricultural income ), especially when the project is evaluated at an early stage of implementation (see Appendix M). 66 Adoption of improved agricultural technologies, such as investments in agricultural conservation structures and implementation of agricultural conservation practices, can lead to the sta bilization of crop yields that would otherwise decline . Use of metallic silos for grain storage, growing vegetables in small gardens, and household savings accumulation, can increase cash income and reduce the risk of asset liquidation in times of food sca rcity. Such behavioral changes can be expected to lead to increases in agricultural income and asset accumulation, if the adoption is successful and continue s after the project ends. The analysis conducted answered the question on whether the strategy of simul taneous promotion of more than one intervention was successful. The results suggest that indeed it was for some of the technologies and practices promoted. However, impacts differ by terciles of wealth, as measured by area of cultivated land. Househol ds in the lower tercile were most active in adopti ng of conservation agricultural practices, savings and kitchen gardens, the middle tercile adopted the most postharvest management practices, and the middle and high tercile s adopted the most construction o f agricultural conservation structures. The trust experiment shows that the benefits of the project are seen not only in the take up of the technologies and practices promoted, but also in the project strategy of promoting the new technologies to project beneficiaries. The formation of producer groups and saving groups is likely to have an impact in increasing the levels of trust within a village, and contribute to the sustainability of project impacts after the project ended. Results of the trust experime nt suggest that the interaction among farmers in the same village increases personalized trust. Trust increases the odds that beneficiaries of the project keep working together towards common goals, mobiliz ing village resources towards the diffusion of att ractive technologies, and incentiviz ing the formation of saving groups. 67 The results of the trust experiment revea l that group -based interventions lead to higher trust levels among participants. This group -based intervention approach to promoting adoption o f technologies and use of better practices deserves further exploration in other settings. Due to the link between trust and economic development (Dearmon & Grier, 2009; Fafchamps, 2006; Horváth, n.d .), this strategy from rural development projects contributes not only to the diffusion of new and better technologies but also to accomplish long term goals of poverty reduction. Pro - poor rural development projects could employ this strategy for both purposes, to increase income and to increase trust levels. Project designers should consider not only this strategy but also the evaluation of its success via experimental economics methods. In sum, this dissertation has tackled the impact evaluation of a rural development project considering the impacts on outcomes related with the adoption of technologies and practices promoted by the project and on outcomes related to the levels of trust among farmers from the same village. The results suggest that the p roject triggered changes in behavior related to agricultural and non -agricultural practices, and likely changes in trust levels that will likely translate into long -term outcomes such as increases in agricultural income and household wealth. The results al so suggest ways to improve the design of complex rural development projects that promot e more than one intervention at the same time by means of group -based interventions . Specifically, such projects should aim : 1) to target interventions to different gro ups of the population, to increase take up of interventions and adoption of promoted practices and technologies , and 2) to promot e group -based interventions with the aim of increas ing trust to further impact income. The results also suggest ways to improve impact evaluations. I mpact evaluation s of projects with group training around multiple interventions should: 1) account for 68 the varying time lapse s to realize impacts of different interventions and 2) use experimental economics methods to estimate impact s on trust levels for p roject beneficiaries . 69 APPENDICES 70 Appendix A Study site description. Nicaragua is one of the poorest countries in Latin America, with 46% of its total population below the poverty line and 15% in extreme poverty. Out of Nicaragua™s total population, 68% live in rural areas characterize for high poverty rates. A total of 70% of rural households in Nicaragua are poor . Rural areas are also characterize for high levels of inequality, explained by the disparities in agricultural productivity between smallholder farmers and medium and large farmers (World Bank, 2008) . Nicaragua is characterized by severe malnutrition problems (World Bank, 2008) . The staple foods in the study area are beans and maize. Fruits and vegetables are consumed sporadically . On special occasions, meat and other protein s are eaten , but not regularly . It is common for households in the area to experience food scarcity during the months of April to July. Farmers look for off farm labor inco me as their main coping strategy. The main sources of income are sales of staple crops and small animals. The study site is located in the departments of Estelí, Jinotega and Matagalpa in the northwest of Nicaragua (see Figure A 1). The project conducted interventions in 44 communities located in eight municipalities in the study area. Its target was to serve 2,500 smallholder farmers. 71 Figure A 1. Map with location of study area. Source: Google maps, 2013. The biophysical setting is very diverse, depending on whether the communities are located in highlands or lowlands. Heavy rains, rocky soils and steep slopes characterize the former, while the lowlands are semi -arid with eroded soils. Traditional agricultural production practices prevail both in the lowlands and the highlands. T here are two main rainy season period s for production 72 of crops, primera between May and August and postrera , between September and December. Staple crops , such as maize and beans, are produced in both . In the highlands, production of vegetables also takes place. Whe re irrigation is available , crops are grown in an additional season, riego or apante , between Jan uary and May, the dry months of the year. According to the C ensus of 2005, Estelí, Jinotega and Matagalpa are characterized by high incidence of poverty , with 60% or more of its population with at least one unsatisfied basic need (Instituto Nacional de Información de Desarrollo, 2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g, 2008h) . Smallholdings of less than 10 manzanas (1 Mz = 1.73 acres ) are prevalent , where farmers mostly tend annual crops and the breeds of small livestock (i.e., not cattle) (Instituto Nacional de Información de Desarrollo, 2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g, 2008h). 73 Appendix B. The Agriculture for Basic (A4N) Needs Project. The Agriculture for Basic Needs (A4N) project was a three year integrated rural development project implemented in four Central American countries Œ Guatemala, Honduras, El Sal vador and Nicaragua Œ during August 2009 - August 2012. It was managed by Catholic Relief Services (CRS) and implemented in the field by its partners, Caritas Jinotega, Caritas Matagalpa and the Foundation for Research and Rural Development (FIDER). This research focuses in Nicaragua. The project identified as major problems diminishing productivity, declining incomes, hunger and unhealthy diets, and vulnerability to shocks. These problems were seen as the result of limited ability to innovate and adapt, l ow agricultural productivity and environmental degradation, limited access to financial services, limited access to markets for agricultural products, and weak community based organizations (Catholic Relief Services, Latin America and Caribbean R egional Office, 2009) . To overcome these problems and constraints, t he A4N program aim ed to provide farmers with a set of five skills for achieving sustainable farm production and increased agricultural income . The skill sets and the interventions promo ted (Catholic Relief Services, Latin America and Caribbean Regional Office, 2009) were as follows: 1. Group management: Participation on any of the following groups: Saving and lending groups Producer groups Farmer innovation groups and farmer fi eld schools Water user committees and watershed management boards 74 2. Saving and lending: Participation on saving and lending groups. 3. Marketing: Training in marketing skills on farmer field schools and innovation groups. 4. Basic experimentation and innovation skills for accessing new technology: Participation in farmer innovation groups (CIALes), implementation of trial plots with improved varieties (high yielding, drought resistant) and bio -fortified varieties of maize and beans, improved farming practices, nu tritious vegetable crops in kitchen gardens (cabbage, carrots, onion, tomatoes and green leafy vegetables) . 5. Agricultural production and natural resource management skills: training on agricultural conservation practices and on construction of agricultural conservation structures, training on post -harvest management and storage practices, use of metallic silos fo r storage of grains, training on integrated pest management, training in small livestock management (husbandry, feed production, vaccination regimes, manure collection). Beekeeping and seed production. The project provided beneficiaries with agricultural assets, such as metallic silos, construction material for animal enclosures, water harvesting structures, plastic water tanks and water filters, and small animals, such as poultry, pigs and goats. It also provided assets to groups of farmers and help with building and management of farmer cooperatives and farmers groups to facilitate access to agricultural inputs, building of grain milling facilities and provided inputs for seed producer and bee keeping groups. The project also conducted village level inter ventions such as provision of rural aqueducts, construction of water harvesting structures, and legalization of land property (this last item in partnership with local governments). In some cases the project also helped build irrigation systems that benefi ted groups of beneficiaries in the villages. However the village level and group level interventions were not implemented in all the villages and with all 75 the farmers groups. Household level project interventions promoting the 5 skills set were available f or all eligible households participating in the project. The strategy of the project was training farmer households on components of the five skill set s, instead of a deep training on each skill set. Participant farmers were t rain ed in the promoted practi ces in farmer field schools and innovation groups. The project trained promoters from different villages and then these promoters replicated the knowledge at their villages. There also were producer groups involved with other activities of the project, suc h as seed production and bee keeping. The project also provided technical assistant to individual farmers, not necessarily involved with the producer groups. Group formation was not only used for promoting agricultural technology, but also to promote savin g and lending groups. The project encouraged participants in producer groups and saving groups to participate in different project activities, producers group members were encourage to form saving groups, saving groups member were encourage to participate in marketing or agricultural related activities. The A4N project first targeted villages considered poor, in terms of limited access to basic services such as water and sanitation, predominance of small land holdings and production of staple grains (maize and beans). These villages are located in areas of natural resource degradation with relatively high vulnerability to natural disasters. Within these villages, in order to be eligible to participate in the A4N program, households were expected to be chara cterized by most of the following official eligibility criteria: Cultivated land area less than two manzanas (1 Mz = 1.73 acres). Cultivated land on steep slopes. 76 Lack of access to any of the following public services: piped water, sanitation, and electri city. Materials for house walls not brick or concrete; roof not concrete, zinc or brick; floor not concrete, ceramic or tile. Household experiences hunger during some period of the year. Household head is female. Household includes children younger than f ive years old. In the particular case of the A4N program, project managers found it difficult to exclude the participation of village members who are not officially eligible, so the program allowed for technically ineligible individuals to participate in t he hope that they would facilitate spreading the benefits of the interventions during and after program implementation. Once in the program , participants could elect whether to participate in one or more of various program interventions. In the case of A4N program, the impact evaluation must account for potential selection bias from two sources Šselection into the A4N project via official eligibility criteria and self -selection into specific A4N activities by A4N participants. 77 Appendix C. Impact evaluation methods. C 1. The problem of impact evaluation We approach program evaluation though Rubin™s potential outcome framework (Rubin, 1974) . The objecti ve of program evaluation is to determine how the intervention or applied treatment affects a desired outcome, evaluating the treatment effect against a counterfactual. Participation of individual I in the project is referred to as a fitreatmentfl given by wi=1, so wi=0 if the individual has not been exposed to treatment. The observed outcome for individual I is: Equation C 1 iiiiiywywy01)1( which means that the outcome for an individual who participates is y1i and if she does not participate the outcome is y0i . The treatment effect of the program intervention is Equation C 2 iiiiyyy01 But the resulting outcome attributable to a program cannot be observed in an individ ual participating and not participating in the program at the same time. Therefore, the problem of program evaluation is a problem of missing data, and the program effect cannot be calculated for the same individual, but instead requires constructing a cou nterfactual to calculate average treatment effects across individuals in (a sample from) the population. The parameters of interest are the average treatment effect on the population, ATE, and the average treatment effect on the treated, ATT. The ATE is t he difference between the expectation of the outcome with and without the program. For an individual, given a vector of characteristics x, it is: 78 Equation C 3 )|()|())((01xxxyEyEEATE ATE measures the effect of the treatment on both participants and non -participants. The average treatment effect on the treated, ATT, which is the expected value of the outcome for those who participated in the program, conditional on the individual charac teristics that determine program participation, x: Equation C 4 )1,|()1,|())((01wyEwyEEATTxxx As already mentioned, E(y0|x, w=1) , the expected outcome of the treated if they were not exposed to the treatment, cannot be observed direc tly, whereas we can observe E(y0|x, w=0), the expected outcome of the untreated, given that they were not exposed to the treatment. We can define: Equation C 5 E(y1|x,w1)E(y0|x,w0)E(y0|x,w1)E(y0|x,w0)ATT Therefore, Equation C 6 ATT E(y1|x,w1)E(y0|x,w0)E(y0|x,w1)E(y0|x,w0) Subject to the assumption of no selection bias, in the absence of the program, those who participated in the program would have had equal outcomes to those who did not. Equation C 7 )0,|()1,|(00wyEwyExx 79 When eligibility to participate in a program has been randomly assign ed, outcomes are independent of treatment . As a result, the ATE and the ATT are the same, and we can estimate these parameters by simple differences in means. However, if program eligibility has not been randomly assigned, but rather is granted conditional on a given set of individual characteristics, then selection bias occurs, and individuals exposed to the treatment will systematically differ from those not exposed to the treatment. Hence, program outcomes can confound these initial difference s with the effects of program intervention, distorting the measure of the benefits from the program. Selection bias is a consequence of the difference in characteristics between participants and non -participants. It causes ATE and ATT to differ. The resea rcher can observe some characteristics, such as housing features, land allocated to agricultural production, and topographical location of fields. Other characteristics are not observed by the researcher and can be assumed not to change over time, includin g such individual characteristics as motivation, cognitive learning ability, and attitudes towards innovation. Based on the observability of characteristics that underpin selection bias, methods are available to correct for it, allowing the researcher to closely approximate program impacts . Two assumptions about program assignment mechanisms underlie the two major classes of quasi -experimental methods to correct for selection bias used in this research when conducting program evaluation (Imbens & Wooldridge, 2008) . The first is that e xpected values of outcomes, y, conditional on covariates, x, are independent of program assignment w. This is known as the conditional independence assumption, unconfoundedness or selection on observables. T he second is that unobserved characteristics that affect selection are time invariant . This is referred 80 to as the selection on un -observables. The challenge is to correct for these two sources of selection bias when conductin g impact evaluation, t o estimate program impacts correcting for the two sources of selection bias mentioned above. C 2. Overview of Program Evaluation methods There exist different methods to conduct impact evaluation, for determining the treatment effects and to correct for selection bias. This section provides an overview of some of these methods. Randomization has been implemented for the evaluation of anti -poverty programs in certain instances (N. Ashraf, Giné, & Karlan, 2009; Nava Ashraf, 2009; Banerjee, Cole, Duflo, & Linden, 2007; E. Duflo, Kremer, & Robinson, 2009; Kremer, 2003; Lai, Sad oulet, & de Janvry, 2011). The main feature of this method is to draw two random samples, a treatment group and a control group. Since individuals have been randomly assigned to the treatment and control groups, the mean expectations of outcomes on trea tment effect will only depend on their exposure to the treatment. This way there is no selection bias to correct for (Esther Duflo, Glennerster, & Kremer, 2007) . Randomized ex periments require the close participation of the organization that implements the program. As a result, these evaluations have been conducted with close involvement of governments and NGOs, since the method requires application of randomization in the proj ect design. Randomization of exposure to treatments also implies that project participants who are used as controls do not benefit from the program that is being evaluated (Buddelmeyer & Skoufias, 2004; M. Ravallion & Chen, 200 5). As a result, some organizations find randomization of exposure to treatment benefits to be ethically objectionable. 81 When experimental designs are not feasible, program evaluation can be designed using quasi -experimental methods. Among the methods th at correct for selection bias on un -observables, an effective one is panel data regression analysis using difference in difference (DID) estimators. This approach uses a baseline survey and one or more follow up surveys. It calculates an impact estimate by comparing the sample data between program participants and non -participants, calculating the difference between the mean outcomes of each group before and after the intervention, and then calculating the difference between these two differences. By comp aring differences between groups at different points in time , the procedure removes any bias related to unobservable common time trends. When program placement is likely to be correlated with the outcome variable or with the characteristics of the program participants, another method that is used is the instrumental variable (IV) method. This method c onsist s of using a variable or variables that are correlated with program placement but not correlated with unobservable characteristics of program participants, thereby correcting for endogeneity or bias on unobservables. Propensity score matching (PSM) and regression discontinuity design (RDD ) are two of the quasi -experimental methods that can be used to correct for selection bias on observable characteristics. For PSM the characteristics of the comparison group (individuals not participating in the program) prior to program interventions are used to determine their probability of participating in the program. Therefore, eligibility for program participation becomes an exogenous variable. The propensity score, the estimated probability of being selected for program participation, is used to ma tch members of the comparison group and with members of the treatment group and to estimate impacts as the difference in outcomes between these two groups. 82 For RDD, program participants are chosen according to a threshold value for a given characteristic that determines program eligibility. This method is employed to select a comparison group with similar characteristics to the ones of the treated. This could be accomplished by a sharp regression discontinuity design, where the assignment is a determinis tic function of the covariate used for selecting program participants, or by fuzzy regression discontinuity design, where the probability of being eligible does not necessarily have to change from zero to one at the threshold, producing a jump on the proba bility distribution between participants and not participants. Both RDD and PSM are used when there is cross sectional data available for the treatment and the comparison group to conduct the evaluation. The A4N program did not assign participants randomly . Rather , it focused on benefiting poor smallholder farmers, wh o were selected by the program managers . Purposive selection of potential participants took place with the participation of municipality o fficials and community leaders, elaborated list s of h ouseholds who complied with program eligibility criteria (specified in the program description). These farmers were invited to meetings at their communities where the program was presented, after which they decide whether to participate in specific program interventions. In the language of program evaluation, they s elf -select ed into program interventions of interest based on characteristics that are likely to be unobservable, such as personal motivation . For the impact evaluation of the A4N project, we us e two sets of methods based on different assumptions. We use regression based methods and propensity score methods to control for observable characteristics and time invariant unobservable characteristics. The methods used are described in the next section . 83 C 3. Methods used to estimate project impacts To estimate a program impact on intermediate outcomes related to adoption of agricultural technologies and practices, I appl y panel data econometric methods, beginning with regression usin g the difference in difference (DID) estimator. DID is a traditional regression method for impact assessment . With panel data , this method can be used to estimate the ATE, based in the assumption that unobserved differences between participants and non-participants are invariant in time. Examples of such traits include individual characteristics , like motivation and cognitive ability .. I compare the results from simple DID to those from four methods that attempt to correct for selection bias based o n observables: three forms of PSM-DID (using different matching methods ), and PSW to check for robustness. These methods estimate the ATT, based in the conditional independence assumption that outcomes are independent of the treatment when condition ed on a set of observable characteristics. The theory behind these five methods is set forth below. C.3.1. Regression based methods: Following Wooldridge (2010) , assuming a linear relation between the outcome variable, the unobserved heterogeneity and the covariates or characteristics of t he households, we can write: Equation C 8 itiitititucwy0 Where y indicates the outcome variable, w is a binary variable that indicates participation in the project, and x is a matrix of time varying covariates, c is the unobserved heterogeneity and u is the error term. By taking the difference we removed time invariant unobservable characteristics ci. Then obtaining the first difference between periods t and t -1, the unobse rvable characteristics, 84 assumed invariant in time are eliminated, correcting for this source of bias on the program impact estimation . The difference in difference estimation equation could be written as (Wooldridge, 2010) : Equation C 9 itititituwyx0 where yit=yit-yit-1, xit=xit-xit-1 and uit=uit-uit-1. With two time periods it does not matter if we difference w, since participation in the program will be 0 for all the observations in the first time period, and will take values 0 and 1 depending on whether it is a comparison or a treatment observation. We obtain the program impact by the regression of the change in the outcome variable y the project participation variable w, and the change in a set of time varying covariates x. The first difference equation will be consistent if E( xituit)=0. The parameter of interest is . The difference in difference estimator assumes parallel trends for both treatment and control in the absence of the treatment (Abadie, 2005) . Therefore, correct ing for differences between the two groups require s control ling for covariates related to household characteristics (Abadie, 2005). To take care of possible differences of covariates between treatment and control , include some time varyin g household characteristics as in Equation C 9, and use DID for estimating program impacts. 85 C.3.2. Propensity Score based methods. The main assumptions for estimating the impact of the program are for constructing the counterfactual using propensity score matching are: 1) Unconfoundedness: Equation C 10 x|,10wyy where y0 is the outcome for non -partici pants and y1 is the outcome for participants, w is participation and x represents a set of variables that may influence participation. The sign , denoting orthogonality, means that program outcomes are independent of program participation, conditional on x. 2) Mathematically, there is common support (overlap) between the probability distributions of program participants and non -participants (Caliendo & Kopeinig, 2008; Imbens & Wooldridge, 2008; Martin Ravallion, 2008) : Equat ion C 11 1)|1Pr( 0xw Propensity score matching (PSM) consist s of choosing the comparison group according to the probability of being selected for a treatment, given a set of observable pre -treatment characteristics and outcome values that do not change with program intervention but that affect program placement. The expected probability of program participation is Equation C 12 )()|1Pr( xxGw Here, 0 < G (x) < 1, G refers to the probability distribution function, where x represents a vector of explanatory variables and is a parameter vector . In this case, the explanatory variables refer to program eligibility criteria, household characteristics, village characteristics, 86 farm characteristics, and wealth. Including a rich set of variables that determine both participation in the project and pretreatment outcomes reduces bias in estimates (J.J . Heckman, Ichimura, Smith, & Todd, 1998) . With these estimated probabilities we check for the overlap of the probability distributions of selection into the two groups, by plotting the estimated probability distributions of the treated and comparison groups. Overlap is crucial to be able to implement propensity score based methods, the failure of this assumption is a major source of bias in impact evaluation estimates, basically because the counterfactual is not similar to the treatment group to condu ct valid comparison. In addition we trim the observations with an estimated PS above 0.90 and below 0.10 to improve overlap. With this trimmed sample we re -estimate the PS and conduct matching. We conduct balancing tests to check for the similarity of the marginal distributions of the covariates used to estimate the PS. The tests aim to determine whether the matching procedures have served the purpose of making participants and non -participant groups more similar. Covariates are compared via a measure of standardized bias or normalized differences in means defined as follows (Caliendo & Kopeinig, 2008; Wooldridge, 2010) : Equation C 13 100*202101jjjjssxxnormdiff on the numerator of the expression we have the sample averages for x1 and x0 of variable j for the groups of participants (1) and non -participants (0) , and s1 and s0 are estimated standard error s for variable j for participants and non -participants. An absolute value of percent bias above 25 is typically interpreted to mean that the two groups are not similar by those covariates (Wooldridge, 2010) . We also conducted two -sample t -tests for equal means. The advantage of 87 the s tandardized difference of means with respect to the t -test, is that the former does not depend on the sample size. We compare these bias measures before and after matching. To estimate the ATT we match participants to non -participants using the estimated p ropensity scores using four different matching methods. We use two kernel estimators (E panechnikov and normal or Guassian with bandwith 0.06), local linear regression (tricube kernel and bandwith 0.8), and nearest neighbor (NN) with replacement. Bootstrapp ed standard errors are calculated for all four matching estimates to compare the sensitivity of estimates to different matching methods (Abadie & Imbens, 2008) . Kernel and local linear regression (LLR) are non -parametric matching methods. Kernel matching uses a weighted average of all the observations in the comparison group to construct the counterfactual outcome for each treated observation, whereas LLR estimates a nonparametric locally weighted regression using for comparison observations in the neighborhood of the treated ones (Smith & Todd, 2005) . The weights depend on the type of ke rnel function chosen. An advantage of kernel and LLR matching methods is that they reduce the variance of the estimates by using more information. However, a problem arises if there is insufficient overlap between the distributions of the treated and comp arison groups, as poor matches may be used for comparison, resulting in biased estimates. Nearest neighbor matching with replacement consists of matching each treated observation with one or more having the nearest value of estimated propensity score, so a control observation may be used more than once. When using more than one NN, the estimator constructs a counterfactual mean with the closest comparison observations. Matching with replacement using more than one NN reduces bias in the estimates but incre ases its variance (Caliendo & Kopeinig, 88 2008; Smith & Todd, 2005) . Unlike kernel and LLR methods, NN matched observations all have the same weight. NN matching tends to work best with a large sample of comparison observations to match treated ones with. Propensity score matching assumes that after controlling for observable characteristic s, outcomes are mean independent of participation in the program. But it is likely that there are systematic differences in outcomes for participants and non -participants due to unobservable characteristics, known as bias on unobservables. Assuming that un observed heterogeneity is time invariant and uncorrelated with treatment assignment, we can control for this source of bias using the PSM -DID matching estimator, defined by Smith and Todd (2005) . Smith and Todd (2005) compared longitudinal methods with cross -sectional PSM methods and found that PSM -DID perform best in correcting for selection bias, when compared with experimental results. By using the PSM -DID estimator we contro l for both observable sources of bias by building our comparison groups using PSM and time invariant characteristics, by taking the difference of outcomes before and after treatment. As an additional robustness check, we compare the matching estimates with the propensity score weighted (PSW) regression (Wooldridge, 2010) , basical ly using the DID estimator weighting the regression by the PS. 89 Appendix D. Sample design and data collection. The panel data set consist s of survey data covering the 2008-2009 crop year, collected during June to August 2010, and a follow up survey of the same households covering the 2010-2011 crop year, co llected during February to March 2012. Both rounds of the survey asked respondents to recall their agricultural activities and assets during the previous year . Although the project started in August 2009 , the baseline survey asked about asset ownership at the start of 2009 and about activities during the apante, primera and postrera crop seasons, the la st of which ended before the project began . Hence, the baseline dataset covers management and outcomes that were determined before the A4N project began. The data for 2011 corresponds to recall data during the project™s second year. One problem with recall data is that as more time elapses, what respondents remember can be biased . In particular, information on consumption diminishes the longer the time of recall (Deaton, 2009) . It seems that this problem is less significant with other information, such as agricultural production and household assets. Since the period of recall is about one year, recall bias on the information reported is believed not to be severe; farmers participating on the survey felt able to recall the information requested. The sample includes 30 treatment (A4N) villages, randomly selected from the list of 44 villages where the project was active, and 33 comparison (non -A4N) villages that were randomly selected from a list of 40 villages similar to the treatment ones. The statistical primary sampling units (PSU) are the villages. Data was collected for 10 households in each A4N village and between 10 and 15 households in each non -A4N village. Lists of beneficiary hous eholds in the A4N villages were obtained from the project staff in Nicaragua. Since the project began in 90 August 2009 and the baseline survey took place in July 2010, A4N provided us with the list of treated households. We found no eligible households in th e treated villages that were not listed in the beneficiaries list. From the lists of households, those with cultivated land holdings between 0.25 and 4 manzanas (1 Mz = 1.73 acres) were chosen and randomly ordered for the survey. We collected a sample of t he population with access to land, because we were interested in the interventions that promoted agricultural activities. For the baseline survey, the target sample size was of 700 households, with 300 treated households and 400 comparison households, enou gh to permit anticipated attrition without compromising subsequent statistical analyses. The sample included 100 more non -A4N households than A4N households, anticipating the trimming of observations that is carried out when conducting propensity score mat ching ( see Table D 1). In constructing a balanced panel, with observations for the two time periods on the same households, failure to collecting data on the same hou sehold in both periods of time (attrition) was likely. Reasons include migration or refusal to participate again in the survey. Some attrition is inevitable and should be factored into planning sample size. For this survey a probable attrition rate of 10 % was expected. In developed countries, attrition rates between the first and the second year of household surveys have been found to be between 12% and 15%, but the rate is assumed to be lower for developing countries (Deaton, 1997) . 91 Table D 1. Sample Size for the A4N project evaluation in Nicaragua A4N Non A4N TOTAL Villages 30 33 63 Households 10 10 or 15* Total households 300 400 700 *According to village size. The number of villages surveyed in the eight A4N municipalities is proportional to the population weight of each municipality in the total. Using a projection of the population to 2010, from the Nicaragua Census 2005 (Instituto Nacion al de Información de Desarrollo, 2008a, 2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g, 2008h) , the distribution of A4N villages per municipality obtained is shown in Table D 2. Table D 2. Population w eights used in sample design. Municipality Estimated A4N population 2010 Population weights A4N villages per municipality Non A4N villages per municipality Estelí 1590 9% 3 5 La Trinidad 4152 25% 8 10 San Nicolás 822 5% 1 1 Jinotega 5291 31% 9 6 San Rafael del Norte 673 4% 1 4 Esquipulas 2916 17% 5 4 Terrabona 455 3% 1 1 San Isidro 1037 6% 2 2 Total 16935 100% 30 33 Source: Instituto Nacional de Información de Desarrollo (2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g, 2008h) . 92 To compare A4N and non -A4N villages information on total population, number of households, access to public services (water, electricity and sanitation), proportion of agricultural units growing staple grains, and proportion of agricultural units with less than 10 Mz was compared for treated and comparison villages using t -tests for equal means (using a t -test based on samples of unequal variance). The results suggest that hypotheses of equal mean traits between of villages could not be rejected ( Table D 3). Table D 3. Test for equal means for treatment and comparison villages. Sample Average Variables Treatment Comparison Difference p-value* Population 2005 530 430 98 0.54 Number of h ouseholds 2005 104 84 19 0.49 % inadequate housing 2005 53% 58% -6% 0.24 % no electricity 2005 57% 56% 2% 0.83 % no potable water 2005 59% 66% -7% 0.34 % staple grains 2003 87% 84% 3% 0.53 % landholding<10Mz 2003 50% 58% -8% 0.14 *for t -test with unequal variance Source: Instituto Nacional de Información de Desarrollo (2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g). The non A4N villages were selected to meet two criteria: 1) they are located in the same agricultural zones as the A4N villages (Nitlapan, 2001) , and 2) they are located in areas with similar poverty levels (Instituto Nacional de Información d e Desarrollo, 2008a, 2008a, 2008b, 2008c, 2008d, 2008e, 2008f, 2008g, 2008h) . A list of 45 proposed similar villages was vetted with the A4N management team in Nicaragua. They recommended the elimination of villages they considered ineligible for A4N, because of poverty levels, access to water for irrigation and main economic activities different than agriculture. In order to develop lists of non A4N farm households comparable to the A4N households, we hired field officers to visit the villages and elab orate lists of qualifying households, based on farm size (0.25 to 4 Mz), whether the farmers 93 grew basic grains on 2009, whether household head was female, and whether the household had access to public water and electricity and included children younger th an 5 years old. Data was collected via surveys at both the household level and the village level. The first household survey included a total of 302 households in 30 treatment villages and 366 households in 33 comparison villages. After an analysis of the baseline data, 41 comparison observations were excluded for being considered as invalid comparisons (Peralta, Swinton, & Maredia, 2 011), and 8 observations (treatment and comparison) were excluded as extreme cases with more than 40 Mz of cultivated land. We aimed to follow up on all 619 households for the 2012 return survey. However, only 578 households could be reached for the fol low -up, resulting in an abandonment rate of 7 % (41 observations ). In some cases the household members migrated to work in other areas of the country or moved to another village, but we could not learn the new location; in a few cases they just did not want to participate again in the survey. We did not find systematic reasons to consider attrition a problem. Moreover, the characteristics of the original sample of 619 observations did not differ from the final sample with 578 observations (excluding the 41 households with incomplete data) (Sherman, 2000) . Table D 4 shows the descriptive statistics for several household and village characteristics in 2009 for the original and the final sample of households 9. The p -values for t -tests for equal means suggest that the two samples are not statistically different. W e can conclude that there is not a problem of attrition. 9 We were not able to estimate the probability of abandoning the sample, since the number of observations with positive outcome was only 41 in comparison with 578 negative outcomes. 94 Table D 4. Comparison between households in the original sample and households in the reduced final sample due to attrition , 2009. Original sample n=619 Final sample n=578 Variable Mean Std. Dev. Mean Std. Dev. Difference p-value* Household Head characteristics: Female=1 0.14 0.34 0.14 0.34 0.00 0.96 Age (years) 48.48 15.07 48.52 14.94 -0.04 0.97 Education (years) 2.95 2.76 2.93 2.65 0.02 0.89 Household Characteristics: Farm gross margins C$ 6,641 33,319 6,646 34,105 -4.35 1.00 Experienced hunger=1 0.36 0.48 0.35 0.48 0.01 0.79 Experienced crop losses=1 0.82 0.38 0.82 0.38 0.00 0.98 Experienced stored grain losses=1 0.34 0.47 0.33 0.47 0.01 0.84 Household size (persons) 5.23 2.30 5.28 2.27 -0.05 0.72 Inadequate housing=1 0.85 0.35 0.86 0.34 -0.01 0.67 Overcrowding=1 0.51 0.50 0.52 0.50 -0.01 0.86 Inadequate services=1 0.73 0.44 0.73 0.45 0.01 0.74 Dependency ratio 0.74 0.71 0.74 0.71 0.00 0.91 Farm Characteristics: Cultivated land (Mz) 3.39 3.54 3.40 3.41 0.00 0.99 Steep slope=1 0.32 0.47 0.32 0.47 0.00 1.00 Value of productive assets: Infrastructure (C$/1000) 1.22 7.24 1.19 7.38 0.03 0.94 Equipment (C$/1000) 3.35 8.77 3.39 8.93 -0.04 0.94 Livestock (C$/1000) 9.52 18.11 9.31 16.00 0.21 0.83 Village Characteristics: Distance to market (Km/10) 1.51 0.85 1.52 0.84 -0.01 0.80 Distance to paved road (Km/10) 0.92 0.91 0.92 0.92 -0.01 0.89 Population 2009 640 561 639 562 0.91 0.98 Public school=1 0.97 0.16 0.98 0.15 0.00 0.72 Health facility=1 0.25 0.43 0.25 0.43 0.00 0.87 Farms producing basic grains 2003 (percentage) 0.56 0.22 0.55 0.22 0.00 0.83 Landholdings less than 10Mz 2003 (percentage) 0.86 0.18 0.87 0.17 -0.01 0.56 *Corresponds to p -value for a t -test for equal means. Source: Agriculture for Basic Needs Survey, 2010. 95 The final distribution of the sample of 578 households, 282 households in 30 treatment villages and 294 households in 30 control villages can be found in Table D 5. Table D 5. Number of observations collected by village and municipality. Treatment Comparison Municipality Village Freq Village Freq Jinotega Chaguite Grande 9 Corral de Piedra 10 El Mojón 10 El Cacao 13 Hermita de Saraguasca 10 El Yankee 13 San Antonio de Sisle 9 Las Lomas 9 San Gregorio 10 Mancotal 11 Saraguasca 10 Tomatoya 12 Sasle 10 Sisle 1 9 Sisle 2 9 Total 86 Total 68 San Rafael del Norte Los Horcones 10 Los Encuentros de San Gabriel 8 Los Potrerillo 10 Sacaclí 11 Santa Barbara 10 Total 10 Total 39 San Isidro El Llano Boqueron 9 Wiston Castillo 9 San Ramon de las Uvas 9 El Carrizal 8 Total 18 Total 17 Esquipulas Coscuilo 10 El Castillo 9 El Barro 10 El Zapotal 14 La Enea 9 La Sirena 9 Pita Abajo 10 Pita Arriba 9 Total 48 Total 32 Terrabona San José 9 El Arado 7 Total 9 Total 7 96 Table D 5. (c ont™d) Treatment Comparison Municipality Vil lage Freq Village Freq Estelí El Espinal 12 El Quebracho 10 Las Cuevas 10 Isiquí 7 San Antonio 10 Llano Redondo 14 Llanos de Colon 10 Total 32 Total 41 La Trinidad El Hornillo 10 Cebadilla 10 La Concepcion 9 La Laguna 4 Las Cañadas 7 Las Lajas 14 Las Tablas 9 Las Pencas 14 Mechapa 10 Llano Largo 8 Pacaya 6 Mesa de los Espejos 6 Rosario Abajo 8 San Jose de Guasimal 7 Tomabú 10 San Lorenzo 12 Total 69 Total 75 San Nicolas Quebrada De Agua 10 Cuajiniquil 8 Las Puertas 7 Total 10 Total 15 Grand total 282 294 Source: A4N Household Survey 2010 and 2012. 97 Appendix E. List of data collected with the household and village survey instruments. Household survey: Information was collected on: At the household level Housing characteristics Off farm income sources: remittances, off farm labor, small business. Food scarcity Non -farm assets Farm equipment Farm infrastructures Livestock inventory Livestock products, sales and costs Production from disperse trees Land use Crop losses Post -harvest management practices Total costs of agricultural production: agricultural inputs including hired labor. Credit and saving Participation in rural development projects 98 Individual level Household member characteristics: age , gender, education, main economic activity, participation in groups. Plot level: Land ownership Sharecrop arra ngements Plot characteristics Use of agricultural conservation practices, agricultural conservation structures Access to irrigation Agricultural production per season Farm garden Plantations Planting materials use in the plot Village level survey: Population Infrastructure: transportation, distance to paved road and market Access to water, sanitation, electricity, education and health Main agricultural products prices Livestock prices Rural development projects 99 Appendix F. Household survey instrument for the A4N impact evaluation panel . Statement of Consent Program Participation, Economic Impact, and Agricultural Practices among Nicaraguan Smallholder Farmers We are conducting a research study to understand how agricultural practices and projects affect farm incomes in western Nicaragua. The study is conducted by Michigan State University and Nitlapan at the Universidad Centroamericana with funding from Catholic Relief Services. I would like to ask you questions about your farm activities as well as your participation in agricultural projects. I would also like to look at your house, crop fields and yo ur farm, both today and again after two years. Your participation in the survey is voluntary, so you are free not to participate at all and you may terminate the interview at any time with no penalty. However, I want to encourage you to participate. By le arning about how agricultural practices on farms like yours contribute to your income and welfare, our study aims to inform the design of future agricultural development projects. This survey will take two hours and I will be taking note of your answers. Although you will not directly benefit from your participation in this study, however, the lessons from it may help in designing better agricultural projects and we know of no risks associated with this study. Your privacy will be protected to the maxim um extent allowable by law. All the information you provide us will be kept confidential , with the questionnaire identification sheets locked in a cabinet at Michigan State University for three years after the research is completed. This means that no on e except the researchers and the MSU Human Research Protection Program will have access to your answers . We will not identi fy you or your household in any publication from this study . If you have any questions about this study, such as scientific issues, how to do any part of it, or to report an injury, please contact Professor Scott Swinton by email ( swintons@msu.edu ), by telephone at (1) 517 -353-7218, or by postal mail at Michigan State University, East Lansing, M I 48824-1039, USA. If you have questions or concerns about your role and rights as a research participant, would like to obtain information or offer input, or would like to register a complaint about this study, you may contact, anonymously if you wish , the Michigan State University™ s Human Research Prote ction Program at 517 -355-2180, f ax 517 -432-4503, or e -mail irb@msu.edu or regular mail at 207 Olds Hall, M ichigan State University , East Lansing, MI 48824. You indicate your voluntary agreement to participate by beginning this interview. Thank you again for your help with this important research about agricultural welfare . 100 Questionnaire: Program Participation, Economic Impact, and Agricultural Practices am ong Nicaraguan Smallholder Farmers -2011 Cover page to be removed. Strat um: ( ) 1. Participant ( ) 0. Non -participant Cover page Name of household head: Name of respondent: ____________________________________________ Address: Phone number: __________________________________________ Date: ____________________________________ Department: Municipality: Community: __________________________________________ Read consent script before begin. No. 101 Write questionnaire number again: Questionnaire: Program Participation, Economic Impact, and Agricultural Practices among Nicaraguan Smallholder Farmers -2011 Strat um: ( ) 1. Participant ( ) 0. Non -participant 1. General information 1.1 Enumerator name: ____________________________ 1.2 Supervisor name: _____________________________ 1.8 Relationship of respondent with household head (mark with an X): [ ] 1 Head [ ] 2 Spouse [ ] 3 Son/daughter [ ] 4 Father/mother [ ] 5 Brother/sister [ ] 6 Grandson/granddaughter [ ] 8 Son in law/daughter in law [ ] 9 Brother in law/sister in law [ ] 10 Non relative [ ] 99 Other, specify: _______ _____________________ 2. Household and housing characteristics: Answer questions from 2.1 to 2.3 by observation; please mark the answer with an X 2.1 House predominant walls material: [ ] 1 Adobe [ ] 8 Carrizo [ ] 2 Wood [ ] 9 Blocks [ ] 3 fi Minifaldafl [ ] 10 Stones and dirt [ ] 4 Concrete [ ] 11 Stone, dirt and bamboo [ ] 5 fiPlaycemfl [ ] 12 Zinc [ ] 6 Plastic [ ] 13 Bamboo [ ] 7 Bricks [ ] 99 Other, specify ____________ No. 102 2.2 House predominant roof material: [ ] 1 Thatch [ ] 2 Zinc [ ] 3 Tile [ ] 4 Plastic [ ] 5 fiNicalitfl [ ] 99 Other, specify ____________ 2.3 House predominant floor material: [ ] 1 Dirt [ ] 2 Wood [ ] 3 Concrete [ ] 4 Bricks [ ] 99 Other, specify ___________ 103 3. Household members characteris tics: Now we are going to talk about your household members at January 1st of 2012: A household if formed by people who share food from the same pot or who share food expenses. Family i.d. First name Relationship with household head Gender Age Years of education (Maximum level attained) What did the person do most of the time in 2011? Group or association the person was part of in 2011(multiple answers) C1 C2 C3 C4 C5 C6 C9 C10 C11 C12 (99) 1 2 3 4 5 6 The codes for each variable are in the next page. 104 Family i.d. First name Relationship with household head Gender Age Years of education (Maximum level attained) What did the person do most of the time in 2011? Group or association the person was part of in 2011(multiple answers) HH (Without last name) Write code Write code: (In years) 0 Pre -school Write code: Write code: (This i.d. will be used througho ut this questionn aire to identify the househol d member) Identify household members according to their closeness to the household head, like this: head, spouse, son or daughter, other relatives of household head, non relatives 1 Head 0 Male (Use 0 if it is someo ne young er than 1 year old) 1-6 Elementary 1 Work in own farm 0 None 2 Spouse 1Female 7-11 Secondary 2 Agricultural worker 1 Producers group 3 Son/daughter 12-16 University 3 Non agricultural worker 2 Marketing group 4 Father/mother 17 Professional 4 Professional 3 Savings group 5 Brother/sister 20 Adult education 5 Own business 4 Women group 6 Grandson/ 21 Literate 6 Study/attending school 5 Youth group Granddaughter 22 Illiterate 7 Housewife 6 Church group 8 Son /daughter in law 99 Other, specify 99 Other, specify 7 Watershed committee) 9 Brother /sister in law 999 Does not apply 999 Does not apply 8 Community council 11 Step son/daughter 9 Health brigade 18 Non relative 11 Sports or team 99 Other, specify 12 Political organization 13 School committee 99 Other, specify 999 Does not apply 105 4 Migration 4.1 In 2011 did you or any of your household members migrate temporarily to other region of Nicaragua or overseas for work? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 4.2. CF First name (Without last name) Where did the person go? Write code 1.Managua 2. Inside the department 3. Outside the department 4. Costa Rica. 5. Mexico 6. Other Central American country 7. USA 99. Other, specify How long was the person there? What did the person do? During 2011 how much was the person contribution to the household? Write code: 1. US$ 0 -100 2. US$ 101 -200 4. US$ 201 -300 5. US$ 301 -400 6. US$ 401 Œ 500 7. US$ 501 -1000 8. US$ 1001 or more 9. Did not contribute C1 C2 C3 C4 C5 C6 106 4.2 Did any relatives or others who are not members of your household send money to you or any member of your household during 2011? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 5.1 Name of person who sends money (First name witho ut last name) Where is [–] now? Write code: 1.Managua 2. Inside the department 3.Outside the department 4.Costa Rica. 5. Mexico 6. Other Central American country 7. USA 99. Other, specify How much did you receive from this person in 2011? Write code: 1. US$ 0 -100 2. US$ 101 -200 4. US$ 201 -300 5. US$ 301 -400 6. US$ 401 Œ 500 7. US$ 501 -1000 8. US$ 1001 or more C1 C2 C3 107 5 Other income 5.1 During 2011, did you or any of your household members have a permanent job outside the household farm? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 5.2 CF First name (without last name) Activity How many days did the person work on average a month? During which months did the person work in 2011 Whole year 1 Yes How much did the person make per month on average? C$ C1 C2 C3 C4 C5 C6 C7 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 108 5.2 During 2011, did you or any of your household members have a temporary job outside the household farm? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 5.3 CF First name (Without last name) Activity How many days per month did the person work in 2011, on average? How much did the person make per day, on average C$ C1 C2 C3 C4 C5 C6 Month 1 2 3 4 5 6 7 8 9 10 11 12 #Days Month 1 2 3 4 5 6 7 8 9 10 11 12 #Days Month 1 2 3 4 5 6 7 8 9 10 11 12 #Days Month 1 2 3 4 5 6 7 8 9 10 11 12 #Days 109 Own business 5.3 In 2011, did you or any of your household members work self -employed or in their own business individually ? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 5.3a CF First name (Without last name) Type of Business/ activity See code Name of Business/ activity How many days a month did you run this Business? During which months of 2011 did you run this Business? On average, when you run this Business/activity; what are your: Gross revenue C$ Global costs C$ C1 C2 C3 C4 C5 C6 C7 C8 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Write code: 1. TRADE : Activities that imply the resale of products of any kind. The person does not transform inputs to get the products, just buys and sells a finished product. For example: miscellaneous store, food stores, clothes stores, hardware, etc. 2. SERVICES : Activi ties that imply offering services within the community, such as miscellaneous repair services, transportation, beauty, sewing, gardening, laundry and ironing, etc. 3. FOOD PROCESSING : Activity where food, as an input, is transformed to be sold, for example , making tamales for sale, bread and other baked products, cheese, marmalade, pickles, packaging of different products, etc. 4. SMALL INDUSTRY : Activities where the person transforms non -food inputs, for producing outputs such as soap, bricks, blocks, etc. 5. HANDYCRAFT : Activity where inputs are transformed by hand in a using traditional technologies, such as making hammocks, baskets, hats, wood products, clay products, etc. 6. OTHER ACTIVITIES : Other, not defined previously. NOTE: If you are not sure on the classification of the activity, please write the name of the activity and a short description, and ask you field supervisor for help on classifying the activity. 110 5.3.a. In 2011, did you or any of your household members work self -employed or in t heir own business with others ? Only report what corresponded to the individual, not to the entire group. [ ] 1 Yes [ ] 0 No If the answer is No, continue with 6 CF First name (Without last name) Type of Business/ activity See code Name of Business/ activity How many days a month did you run this Business? During which months of 2011 did you run this Business? On average, when you run this Business/activity; what are your: Revenue C$ Costs C$ C1 C2 C3 C4 C5 C6 C7 C8 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 Write code: 1. TRADE : Activities that imply the resale of products of any kind. The person does not transform inputs to get the products, just buys and sells a finished product. For example: miscellaneous store, food stores, clothes stores, hardware, etc. 2. SERVICES : Activi ties that imply offering services within the community, such as miscellaneous repair services, transportation, beauty, sewing, gardening, laundry and ironing, etc. 3. FOOD PROCESSING : Activity where food, as an input, is transformed to be sold, for example , making tamales for sale, bread and other baked products, cheese, marmalade, pickles, packaging of different products, etc. 4. SMALL INDUSTRY : Activities where the person transforms non -food inputs, for producing outputs such as soap, bricks, blocks, etc. 5. HANDYCRAFT : Activity where inputs are transformed by hand in a using traditional technologies, such as making hammocks, baskets, hats, wood products, clay products, etc. 6. OTHER ACTIVITIES : Other, not defined previously. NOTE: If you are not sure on the classification of the activity, please write the name of the activity and a short description, and ask you field supervisor for help on classifying the activity. 111 6. Food scarcity 6.1.During 2011, was there a period of time when you could not cook one of th e daily meals? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 7. If the answer is Yes, in total, for how long did you experience this situation where you could not cook one of the daily meals? [ ] 1 One week or more [ ] 2 Between 1 and 4 weeks [ ] 3 Between 1 and 2 months [ ] 4 More than 2 months [ ] 0 Does not know, does not remember 6.1a. During 2011, How did you do to cope with this situation (could not cook one of the daily meals)? Mark with an X, multiple answers . [ ] 1 Sell livestock [ ] 2 Sell small animals [ ] 3 Sell farm tools and/or farm equipment [ ] 4 Migrate overseas [ ] 5 Migrate to other region in the country [ ] 6 Ask for a loan to relatives or friends to buy food [ ] 7 Received food from rela tives or friends [ ] 8 Received food from World Food Program (WFP) or from the local municipality [ ] 9 Withdraw money from savings [ ] 10 Requested a consumption loan [ ] 99 Other, specify _______________________ Ask the following questions with reference to January 1 st of 2012 7. Access to public services and housing characteristics. 7.1 How many rooms are there in your house? (Note: consider only the ones made of durable materials) _________________ 7.1a How many of these rooms are used as bedrooms?_ _________________ 112 7.2 In January 1 st 2012, the house where you and your household members were living was: [ ] 1 Rented (*7.2.1) [ ] 2 Owned with documentation (**7.2.2) [ ] 3 Owned without documentation (**7.2.2) [ ] 4 Loaned [ ] 5 Shared [ ] 99 Other, specify ___________ 7.2.1 *If rented, how much did you pay for the rent? C$_____________ 7.2.2 ** If owned, for how much would you sell the house? C$___________________ Ask questions with respect to January 1 st 2012 7.3 How did you get water for your house? [ ] 1 River or spring [ ] 2 Pipe inside the house [ ] 3 Pipe outside the house [ ] 4 Public well [ ] 5 Private well [ ] 6 Water harvesting [ ] 7 Brings water from neighbors/relatives houses [ ] 99 Other, specify ______________________ 7.4 Which ki nd of toilet service did you have in your house? [ ] 1 None [ ] 2 Latrine outside the house (with of without treatment) [ ] 3 Flushing toilet [ ] 99 Other, specify ________ _____________ 7.5 Which type of lighting energy did you have in your house? [ ] 1 Candle [ ] 2 Electric power [ ] 3 Electric plant/motor [ ] 4 Solar panels [ ] 99 Other, specify _____________ 113 7.6 Which kind of cooking fuel did you use? [ ] 1 Wood [ ] 2 Coal [ ] 30 Gas [ ] 31 Biogas produced at the farm [ ] 5 Electric po wer [ ] 99 Other, specify ________________ 7.7 ¿ How many of the following articles did you have at your house? Please consider the ones that worked at January 1 st 2012. Articles Quantity on 1-1-2012 (write 0 if did not have) For how much could you have sold it on 1-1-2012? (Total value in C$) C1 C2 C3 Television Refrigerator Bicycle Motorcycle Car Sewing machine Blender Electric iron Cellphone Radio Tape recorder Stereo 114 8 Livestock inventory Big and Small animals. NOTE: C1=(C2+C3+C5+C6) -(C7+C9+C10) Animals How many animals January 1st 2012? How many animals January 1st 2011? In 2011 Out 2011 Quantity purchase Price per animal C$ Births Gifts Quantity sale Price per animal C$ Household consumpt Deaths, gifts C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 1. Oxen 2. Horses 3. Donkeys/mules 4. Bulls 5. Cows 6. Heifers (all ages) 7. Steers (all ages) 8. Calves (0 to 1 year old) 9. Local hogs 10. Improved hogs 11. Local goats 12. Improved goats 115 Livestock inventory continued Animals How many animals January 1st 2012? How many animals January 1st 2011? In 2011 Out 2011 Quantity purchase Price per animal C$ Births Gifts Quantity sale Price per animal C$ Household consumpt Deaths, gifts C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 13. Local sheep 14. Improved sheep 15. Local poultry 16. Improved poultry 17. Local rabbits 18. Improved rabbits 19. Local ducks 20. Improved ducks 21. Local turkey 22. Improved turkey 99. Other, specify 116 8.2 Did you produce or process livestock/small animals™ products? [ ] 1 Yes [ ] 0 No Product Unit of measure January -April 2011 May -October 2011 Quantity produced Quantity consumed Quantity sold Unit price C$ Quantity produced Quantity consumed Quantity sold Unit price C$ C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 1. Milk Lts x day 2. Cream Lb x week 3. Salty cheese Lb x week 4. Cheese Lb x week 5. Butter Lb x week 6. Soft Cheese Lb x week 7. Eggs units x week 8. Manure/organic fertilizer qq x year 9. Beef (uncooked) 10. Pork (uncooked) 99. Other, specify 117 Produce or process livestock/small animals™ products continued Product Unit of measure November -December 2011 Quantity produced Quantity consumed Quantity sold Unit price C$ C1 C2 C11 C12 C13 C14 1. Milk Lts x day 2. Cream Lb x week 3. Salty cheese Lb x week 4. Cheese Lb x week 5. Butter Lb x week 6. Soft Cheese Lb x week 7. Eggs units x week 8. Manure/organic fertilizer qq x year 9. Beef (uncooked) 10. Pork (uncooked) 99. Other, specify 118 8.3 Livestock production costs 2011. Cost Did you do this ? Did you incur in any costs ? Livestock Small animals Observations 1 Yes 0 No 1 Yes 0 No Total value C$ Total value C$ (If the activity took place, but did not incur in costs ) C1 C2 C3 C4 C5 C6 1. Hired labor 2. Feed/forages purchased (e.g. concentrate , manure , molasses ) 3. Medicines , vaccines , veterinary 4. Inseminatio n 5. Infrastructure management 99. Other services (e.g. transportation , processing .) 119 8.4 Other species Did you do aquaculture or apiculture activities individually during 2011? [ ] 1 Yes [ ] 0 No, If the answer is No, continue with 8.3.a Species Quantity Quantity Unit of measure During 2011 Janu ary 1st 2011? Janu ary 1st 2012? UM Purchase Purchase price Gifts deaths stolen other Product Consump Sales UM Sale price UM Total annual manag cost 2011 C$ C$ C$ C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 Fish in ponds Beehives Honey UM: unit of measure 120 8.3.a Did you do aquaculture or apiculture activities with other farmers during 2011? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 9. If the answer is yes, fill the following table, with the information on what corresponded to the indivi dual farmer, and not to the entire group. Species Quantity January 1st 2012? Unit of measure UM Income 2011 C$ Costs 2011 C$ C1 C2 C3 C4 C5 Fish in ponds Beehives Honey 121 9 Did you have the following equipment/infrastructure on January 1st 2012 ? Equipment Number 1-1-2012 (write 0 if did not have it) For how much could you sell it at 1-1-2012? Total value C$ Observations C1 C2 C3 C4 Oxen plough (without oxen value) Tractor Manual knapsack sprayer Motor pump sprayer Plastic containers Metallic containers Metallic silos Barrels Irrigation pump (without accessories) Irrigation motor Sprinkler irrigation accessories Drip irrigation accessories Cart Bio -digesters Grain grinder Manual sheller Mechanic sheller Apiculture equipment Apiculture protection equipment Other, specify 122 Infrastructure Did you have it? 1 Si 0 No Value of infrastructure at 1-1-2012? C$ Observations C1 C2 C3 C4 Water harvesting ponds Fish ponds Poultry house Pig sty Forage silos Feeders Drinkers Barnyard Granary (wood) Greenhouse Other, specify 123 Now let™s talk about agricultural production at your farm. 10 Did you have dispersed fruit trees on January 1st 2012 ? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 11. NOTE: C3=(C5+C6+C7) 11 For managing your farm, in 2011 did you implement any of the following? [ ] 0 None [ ] 1 Farm plan [ ] 2 Keep books of revenues and costs [ ] 99 Other, specify _______________________ 12 Farm sketch, include all the plots under the management of the farmer during 2011 identify each plot with a number and a name, if the farmer has a name for it. Please draw the sketch in the attached blank sheet With the help of the sketch, fill the table about land uses. Dispersed fruit trees (list by order if importance, write the tree species) Number of trees at 1-1-2012 Productio n 2011 Unit of measure ment UM Quantity consumed 2011 Gifts 2011 Sales 2011 Price per UM C$ C1 C2 C3 C4 C5 C6 C7 C8 124 12.2 Land uses 2011, include all the plots u nder the farmer™s management. NOTE: C3=C4+C5+C6+C7+C8 Plot number Plot name Total area in Mz Area in Manzanas (Mz) Annual crops Plantations Grasses/ forages Forests Uncultivated land C1 C2 C3 C4 C5 C6 C7 C8 0 House* PLEASE START THE PLOT INFORMATION SHEET (FILL ONE PER PLOT) 125 Plot information sheet 13.1.Plot name: _________________________ No.:_____________ 13.2.Area in Mz: _______________________ 13.3.Ownership: [ ] 1 Owned * [ ] 3 Rented (owner is someone else) ** [ ] 2 Owned and sharecropped*, *** [ ] 4 Sharecropped (owner is someone else) *** [ ] 5 Own and rented*, ** [ ] 99 Other, specify _______________ 13.4a *If owned, what is the legal ownership situation of the plot? [ ] 0 None [ ] 2 Agrarian reform title [ ] 1 Ownership document [ ] 3 Bill of sale [ ] 99 Other, specify___________________ 13.4aa *If owned, for how much could you have sold it on January 1 st 2012? C$ ________________ 13.4b **If rented, Rent 1. Paid 2. Received Period 1.Apante 2.Primera 3.Postrera 4. All year Payment 1.In kind 2.Cash 99. Other, specify Area Mz How much was the rent? (total)* C$ Quantity paid in kind In kind payment unit of measurement Form of in-kind payment C1 C2 C3 C4 C5 C6 C7 C8 *If payment was in labor, please estimate the value in C$.. 126 13.4c ***If sharecropped, what was the agreement (%)? Item Percentage assumed in the sharecropping agreement Season 1 Apante 2 Primera 3 Postrera C1 C2 C3 Inputs Production Labor Other, specify 13.4.Distance to closest road: ________Km 13.5.Distance to the homestead: ________Km 13.6.Slope: [ ] 1 Flat [ ] 2 Not to steep [ ] 3 Steep 13.7.Soil texture: [ ] 1 Clay [ ] 2 Silt [ ] 3 Sandy [ ] 4 Loam 13.8.Presence of rocks: [ ] 1 None [ ] 2 Few [ ] 3 Many 13.9.Did you have access to irrigation in this plot on January 1 st 2012? [ ] 1 Yes [ ] 0 No. If the answer is Yes, continue with 13.9a y 13.9b, otherwise go to 13.10 13.9a If yes, from where did you get the water for irrigation? [ ] 1 Well [ ] 3 Water harvesting pond [ ] 2 River or spring [ ] 4 Waterhole [ ] 99 Other, specify _________________________ 13.9b. Type of irrigation system: [ ] 1 Gravity irrigation [ ] 3 Sprinkler irrigation [ ] 2 Drip irrigation with pipe [ ] 4 Drip irrigation with bottle [ ] 99 Other specify___________________ 127 13.10 Soil and water conservation structures Structure Did you have this soil/water conservation structure on January 1st 2012? 1 Yes 0 No Did you build any of these structures between January 1 st 2010 and January 1 st 2012? 1 Yes 0 No Area of length built between January 1 st 2010 and January 1 st 2012 Unit of measureme nt of the area/length built C1 C2 C3 C4 C5 1 Stone barriers/terraces 2 Crop residue barriers 3 Live barriers (perennial crops planted in contours) 10 Trees planted to protect waterways and canals. 11 Infiltration trenches 12 Dams 13 Ditches 99. Other, specify 128 13.11 Did you implement any of the following soil and water conservation practices during 2011? Soil and water conservation practices Did you implement any of these practices during 2011? 1 Yes 0 No C1 C2 1 Minimum tillage 2 No tillage 3 No burning 11 Manure 12 Compost 13 Vermicompost 21 Green manure 22 Cover crops 23 Mulch 31 Contour planting 99 Other, specify 129 13.12 Cultivos anuales durante 2011. Si es a medias reportar solamente lo que le correspondió. Season Crop 1.Maize 2.Beans 3.Maize and beans intercropped 4.Other intercropped 99.Other, specify Cultivated area Mz Production Unit of measure UM Consump tion Gift Quantity of grain stored 1-1-2012 Sales Price per UM C$ Did you sell to: 1 Intermediary 2 Consumer 3 Coop 4 Supermarket/ Enterprise 99 Other, specify C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 Apante 2010-2011 Primera 2011 Postrera 2011 130 Annual crops 2010-2011 continued Season Crop 1.Maize 2.Beans 3.Maize and beans intercropped 4.Other intercropped 99.Other, specify How did you take products to the market? 0 On farm 1 Horse 2 Mule 3 Bus 4 Bicycle 5 Motorcycle 6 Truck 99 Other, specify Transportation costs Cost C$/UM Cost per trip/number of trips C1 C2 C12 C13 C14 Apante 2010-2011 Primera 2011 Postrera 2011 131 13.13 During 2011 did you have family garden? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 13.13a Total area of family garden in Mz: _____________ Crop Quantity produced in 2011 Unit of measure UM Quantity used for consumption 2011 Quantity sold Price per UM C$ Observation C1 C2 C3 C4 C5 C6 C7 13.13a. Did you participate in a group garden with other farmers during 2011? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 13.14 Only report what corresponded to the individual farmer . DO NOT report the total for the group. Total garden area in Mz: ___________________ Crop Quantity for consumption 2011 Quantity sold 2011 Price per UM C$ C1 C2 C3 C4 132 13.14 Plantations 2011 Type Write code: 1. Coffee 2. Fruits 3. Musaceae 4. Cocoa 5. Forest 99 Other, specify Area in Mz Year when planted Quantity produced 2011 Unit of measurement Quantity sold 2011 Price per UM C$ C1 C2 C3 C4 C5 C6 C7 13.15 Grasses and forages 2011 Area [ ] 1 Natural grass _________________________________Mz [ ] 2 Improved pasture _________________________________Mz [ ] 3 Improved forage crop _________________________________Mz 133 13.16 Planting materials 2011 (for all the crops plan ted in the Plot in 2011) NOTE: MAKE SURE YOU FILLED ALL THE PLOT SHEETS, ONE PER PLOT UNDER THE FARMER™S MANAGEMENT, BEFORE CONTINUING. Variety Specify Season when used Write code: 1 apante 2 primera 3 postrera 4 whole year How did you get it? Write code 1 Previous harvest 2 Community 3 NGO, write name 4 Government 5 Bought at inputs store 6 Bought at cooperative 7 Produced yourself 99 Other, specify Quantity Unit of measur e UM Equivalence UM Price per UM (Only if the person bought it) C1 C2 C3 C4 C5 C6 C7 Bean Maize Sweet potato Other 134 Production and storage losses 14 During 2011, did you experience any production losses on any of your crops? [ ] 1 Yes [ ] 0 No, if the answer is No, continue with 15, if the answer is yes, fill this table: Crop Percentage of losses (with respect to what was expected) Season 1 Apante 2 Primera 3 Postrera Plot (use the number already defined) C1 C2 C3 C4 The crop losses experienced were a result of: 14.1a Natural phenomena? [ ] 1 Yes [ ] 0 No, if the answer is No, continue with 14.1b, if yes please mark the cause with an X: [ ] 1 Drought [ ] 3 Flood/landslides [ ] 2 Winds [ ] 99 Other,specify__________________ ______ 14.1b Pests and diseases? [ ] 1 Yes [ ] 0 No. If the answer is no No, continue with 15 if yes, indicate on which crops?____________________________ 15 During 2011, did you use any biological or organic pest and disease control on your crops? [ ] 1 Yes [ ] 0 No, If the answer is No, continue with 16, if the answer is Yes, fill the table Crop Type of control used? C1 C2 135 16 ¿Did you store grains produced during 2011? [ ] 1 Yes [ ] 0 No, if the answer is No, continue with 17, if Yes, fill 16.1 16.1.How did you store grains during 2011? Mark with an X [ ] 10 Sacks [ ] 20 Storage community center [ ] 11 Barrels [ ] 29 Other community storage facilities specify_____________ [ ] 12 Metallic silos [ ] 13 Granary ( made of wood) [ ] 19 Other storage facilities at the household, specify___________________________ 17 Did you experience any grain storage losses during 2011? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 17.1, if yes answer 17.a. 17.a Which were the causes of grain storage losses during 2011? [ ] 1 Humidity and fungi [ ]3 Rodents [ ] 2 Insects [ ]4 Birds [ ] 99 Other, specify__________________ 17.1. Do you think that the percentage or stored grain losses has changed between 2009 and 2011? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 18. Crop Change in grain storage loss percentage 2009 to 2011 % (if the percentage of losses has increased right +, if it has decreased, right -) C1 C2 136 18 Total costs of agricultural inputs 2011. (all crops, including the ones grown in the family garden, and non synthetic inputs) Inputs used (Specify type) How much did your purchase 2011? Apante 2011 Primera 2011 Postrera 2011 Quant Unit Equiv Price per unit Qua nt Unit Equiv Price per unit Quant Unit Equiv Price per unit C1 C2 C3 C4 C5 C6 C7 C8 C7 C8 C9 C10 C11 Fertilizers Pesticides Herbicides Fungicides Cost intercrop Cost Group garden Plantations costs (total for 2011) 137 18.1 Total costs of agricultural inputs 2011. (all crops, including the ones grown in the family garden, and non -synthetic inputs) Inputs used (Specify type) How much did you p urchase in 2011? Apante 2011 Primera 2011 Postrera 2011 Quant Unit Equiv Price per unit Quant Unit Equiv Price per unit Quant Unit Equiv Price per unit C1 C2 C3 C4 C5 C6 C7 C8 C7 C8 C9 C10 C11 Other, specify Other annual costs 138 19. Labor hired 2011 Did you hire workers during 2011? [ ] 1 Yes [ ] 0 No 19.1 Daily payment Activity Apante Primera Postrera Wages paid 2011 C$ Wages paid 2011 C$ Wages paid 2011 C$ No. of days paid With food Without food Other No. of days paid With food Without food Other No. of days paid With food Without food Other C1 C2 C3 C4 C5 C6 C7 C7 C8 C9 C10 C11 C12 Soil preparation Planting Applying fertilizer Weeding Harvest Other 19.2 Other activities Payment by productivity (per volume, area, or other unit of measure independently of how long it takes). Activity 2011 Total paid 2011 C$ Activity 2011 Quantity Unit of measure Payment per unit C$ Observations Apante Primera Postrera C1 C2 C3 C4 C5 C1 C2 C3 C4 Oxen plough Tractor Other 139 Credit and savings 20. On January 1 st 2012, did you or any of your household members hold any loans? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 21. Otherwise fill the table: CF First name (Without last name) Use of loan Write code 1 Agricultural inputs 2 Food 3 School supplies 4 Health expenses 5 Migrate to work 6 Non agricultural busines s 7 Buy clothes 8 Housing improvements 99 Other, specify Source Write code: 1 Bank 2 Intermediary 3 Microfinance institution 4 Cooperative 5 Community lender 6 Savings group 7 Relative/friend 30 FDL( Fondo de Desarrollo Local ) 99 Other, specify Year of approval Total amount C$ Duration of loan Pending amount 1-1-2012 C$ C1 C2 C3 C4 C5 C6 C7 C8 140 21. On January 1 st 2012, did you or any of your household members have savings? [ ] 1 Yes [ ] 0 No If the answer is No, continue with 22. Otherwise fill the table: CF First name (Without last name) Amount Write code 1 C$ 0 Œ 500 2 C$ 501 Œ 1000 3 C$ 1001 Œ 1500 4 C$ 1501 Œ 2000 5 C$ 2001 Œ 2500 6 C$ 2501 Œ 3000 7 C$ 3001 or more Where did you have your savings? Write code 1 Bank 2 Microfinance institution 3 Cooperative 4 Saving groups 5 House 99 Other specify C1 C2 C3 C4 141 Participation in rural development projects 22. From 2009 to the end of 2011 , did you or any of your household members participate in rural development project activities? Themes 1 Yes 0 No Activity and/or benefit (multiple answers) write code 1 Attendance to workshops and talks 2 Technical assistance in the field 3 Agricultural inputs 4 Experimental plots 5 Pest management 6 Credit 7 Animals 8 Food 9 Medicines 10 School supplies 99 Other, specify Institution Write code 1 MAGFOR 2 INTA 3 Caritas 4 CRS 5 CARE 6 FIDER 7 Visión Mundial 8 Plan International 9 PROMIPAC 10 TECNOSERVE 11 RAMAC 12 CIAT 13 UNA 14 UCA 99 Other, specify C1 C2 C3 C4 1 Seeds production 2 Seedlings production 3 Integrated pest management 4 Organic management 5 Good agricultural practices 6 Product transformation 7 Forage/grasses improvement 8 Trade and marketing 9 Posthervest management 10 Conservation agriculture From 2009 to the end of 2011 , did you or any of your household members participate in rural development project activities? Continued 142 Themes 1 Yes 0 No Activity and/or benefit (multiple answers) write code 1 Attendance to workshops and talks 2 Technical assistance in the field 3 Agricultural inputs 4 Experimental plots 5 Pest management 6 Credit 7 Animals 8 Food 9 Medicines 10 School supplies 99 Other, specify Institution Write code 1 MAGFOR 2 INTA 3 Caritas 4 CRS 5 CARE 6 FIDER 7 Visión Mundial 8 Plan International 9 PROMIPAC 10 TECNOSERVE 11 RAMAC 12 CIAT 13 UNA 14 UCA 99 Other, specify C1 C2 C3 C4 11 Small animals management 12 Farmer field schools 13 Local agricul -tural research committee (CIAL) 14 Bee keeping 15. Fish farming 16 Reforestation and watershed conservation 18 Savings 20 Health 21 Nutrition 99 Other, specify 143 Appendix G. Village survey instrument for the A4N impact evaluation panel . Community Survey: Program Participation, Economic Impact, and Agricultural Practices among Nicaraguan Smallholder Farmers -2011 Información General Read statement of consent before begin Name: ____________________________________________ Supervisor: ____________________________________________ Department: Municipality: Community: __________________________________________ Total population: ___________________________________ ____ Number of houses: _____________________________________ Number of families: ___________________________________ Sale price for land: ______________________C$/Mz Date: ______/_______/_________ 1. Infrastructure 1.1 Name of closest market ____________________ _ 1.2 Distance to closest market ______________________Km. 1.3 Distance to closest paved road______________Km. 1.4 Bus trip time to closest municipality _________________________________min Walking time to bus stop__________min 1.5 Where do you buy agricultural inputs? (name of closest place) _______________________ 144 2. Access to education and services 2.1 Public School? [ ] 1 Yes [ ] 0 No 2.1.a Scho oling age for children _____________years 2.2 Health facility? [ ] 1 Yes [ ] 0 No 2.3 Electricity? [ ] 1 Yes [ ] 0 No 2.4 Aqueduct/Piped water? [ ] 1 Yes [ ] 0 No 2.5 Sewage system? [ ] 1 Yes [ ] 0 No 3. Agricultural and livestock production 2011 Product Price C$ Unit of measure Average yield Apante Primera Postrera Basic grains Maize Bean Wheat Millet Sorg hum Horticultural crops/tubers Tomato Cab bage Chiltoma Lettuce Pipían Onion Potat o Quequisque Cassava Ayote Malanga Carrots Beets Cucumber Chilla Sweet potato Radish Chayote Garlic 145 3. Agricultural and livestock production 2011 Continued Product Price C$ Unit of measure Average yield Apante Primera Postrera Fruits/ bananas Mango Avocado Orange Sour lemon Sweet lemon Pas sion fruit Water melon Cant elope Mamon Plums Nancite Pitahaya Tamarind Zapote Banana Other Coffee Other livestock Eggs Beef Cheese Pork Poultry Cow milk Goat milk 146 4. Labor prices 2011 Activity Apante Primera Postrera Adjustment /Unit With meal s Without meals With mea ls Without meals With mea ls Without meals Workday Plowing with oxen Tractor plow Coffee harvest Cane sugar cutting Tomato harvest Harvest Seed ing Soil preparation Chapia 147 5. Livestock prices en 2011 Animal Average price C$ Observations 1. Oxen 2. Horses 3. Mule s / Asses 4. Bulls 5. Cows 6. Heifers (all ages) 7. Steers (all ages) 8. Calves (0 to 1 year old) 9. Local hogs 10. Improved hogs 11. Local goats 12. Improved goats 13. Local sheep 14. Improved sheep 15. Local poultry 16. Improved poultry 17. Local rabbits 18. Improved rabbits 99. Other, specify 6. Institutions that implemented projects in the community 2009 y 2011 Institution Project ________________________ ________________________________________ ________________________ ________________________________________ ________________________ ________________________________________ ________________________ ________________________________________ ________________________ ________________________________________ ________________________ ________________________________________ ________________________ ________________________________________ 148 7. Extreme weather events. Where there extreme weather events during 2009 -2011 (rains, drought, winds, others) that affected the community agricultural and livestock production? [ ] 1 Yes [ ] 0 No If the answer is yes, please answer the following: Weather event Year _____________________________ ________ _____________________________ ________ _____________________________ ________ _____________________________ ________ _____________________________ ________ _____________________________ ________ _____________________________ ________ _____________________________ ________ _____________________________ ________ 149 Appendix H. Algori thm for estimating the propensity score. We apply Dehejia and Wahba™s (2002) suggested algorithm for estimating the propensity score to determine whether higher order terms and/or interaction terms need t o be included in the model: 1. Start with a parsimonious logit specification to estimate the score. 2. Sort the data according to estimated propensity score (ranking from lowest to highest). 3. Stratify all observations such that estimated propensity scores within a stratum for treated and comparison units are close (no significant differences); for example, start by dividing observations intro strata if equal range (0 -0.2, 0.2-0.4, –, 0.8-1) 4. Statistical test: for all covariates, differences in means across treated and comparison units within each stratum are not statistically different from zero. a. If covariates are balanced between treated and comparison observations for all strata, stop. b. If covariates are not balance for some stratum, divide the stratum into finer strata and reevaluate. c. If a covariate is not balanced for many strata, modify the logit by adding interaction terms and/or higher -order terms of the covariates and reevaluate. 150 Appendix I. Pretreatment characteristics of treat ment and comparison households. Looking into the pretreatment (2009) characteristics for both A4N and non -A4N households, for the eligibility criteria, the average area of cultivated la nd among the participants is 3.3 Mz (Table I 1), which is greater than the 2.5 Mz maximum area of land proposed as a formal eligibility criterion for participation. In practice, this type of eligibility criterion is difficult to enforce , and since the project was also allowing non -eligible households to participate as part o f its strategy, it is not surprising that we find that not all the eligibility criteria are met. Also, more than 60% of the A4N households live in inadequate housing, and 88% lacked access to piped water and sewage (inadequate services) ( Table I 1 ), indica ting that most of these households were poor as measure by these two components of the index of unsatisfied basic needs (UBN). Based on comparison of individual sample means for pretreatment characteristics using a t -test for samples with unequal variance (Table I 1 ), the A4N and the non -A4N household characteristics and asset endowments do not differ significantly for most cases . A4N households had less access to adequate services (piped water and flushing toilet), higher incidence of hunger, a higher proportion of households with female head, and infrastructure and livestock with lower value in comparison with non -A4N households. The A4N households were also located in villages farther from markets and the proportion of these villages with access to a health facility was lower. Both A4N and non -A4N villages, and A4N and non -A4N households (see Table I 1) are similar in terms of the characteristics used by the project to select villages and households. We can say with confidence that we successfully constructed a valid counterfactual for impact assessment. 151 Table I 1. Pretreatment characteristics of A4N and non -A4N households , 2009. A4N n=282 Non -A4N n=294 Variable Mean Std. Dev. Mean Std. Dev. Difference p-value Farm Characteristics Cultivated land Mz 3.29 (3.42) 3.50 (3.41) -0.20 0.47 Steep slope =1 0.32 (0.47) 0.32 (0.47) 0.01 0.87 Housing Characteristics Inadequate services=1 0.66 (0.47) 0.79 (0.41) -0.12 0.00 Inadequate housing=1 0.88 (0.33) 0.85 (0.36) 0.03 0.26 Electricity access =1 0.61 (0.49) 0.63 (0.48) -0.02 0.69 Household Characteristics Hunger experienced =1 0.39 (0.49) 0.32 (0.47) 0.07 0.09 Head female=1 0.20 (0.40) 0.07 (0.26) 0.13 0.00 Children under 5 years old (number) 0.51 (0.73) 0.51 (0.71) 0.00 0.95 Head age (years ) 49.41 (15.29) 47.67 (14.57) 1.74 0.16 Head education (years) 2.83 (2.70) 3.04 (2.59) -0.22 0.32 Household size ( persons ) 5.20 (2.32) 5.36 (2.23) -0.16 0.39 Persons per room 3.82 (2.03) 3.86 (2.09) -0.04 0.84 Asset values Infra structure (C$/1000 ) 0.52 (1.49) 1.48 (8.64) -0.96 0.06 Livestock (C$/1000 ) 6.71 (9.94) 9.07 (16.30) -2.35 0.04 Equipment (C$/1000 ) 1.76 (4.39) 2.08 (4.85) -0.32 0.40 Village characteristics Population 2009 637 (488) 640 (626) -2.95 0.95 Dist. to mar rket (Km/10 ) 14.09 (6.86) 16.29 (9.59) -2.21 0.00 Dist. to p aved road (Km/10 ) 9.53 (9.65) 8.95 (8.72) 0.58 0.45 Health facility=1 0.21 (0.40) 0.28 (0.45) -0.08 0.03 Source: A4N Household Survey 2010 and Village Survey 2010. 152 Appendix J. Description of outcomes to be evaluated by the project. This impact evaluation focuses on the population of farmers with access to land and expected to benefit from interventions related to agricultural activities. With respect to project outcomes, Table J 1 reports the sample averages for both A4N and non -A4N households before and after treatment , it also presents the definition of the outcomes, and the units of measure. These outcomes correspond to: agricultural conservation structures, agricultural conservation practices, storage practic es, kitchen garden, small livestock, saving and credit, food scarcity and agricultural income and household wealth related outcomes. Most of these outcomes were not statistically different in 2009. There were some differences between the two groups in term s of the use of cover crops in at least one of the plots under the management of the household, the proportion of households experiencing stored grain losses and the value of production of main crops ( Table J 1). On the whole, the gre at similarities between the groups confirms that we successfully built a valid counterfactual for comparison The self -reported participation in different A4N interventions by project beneficiaries, was obtained from a survey question asking if the household received benefits from development projects between 2009 and 2011. The results are shown in Table J 2 for A4N project beneficiaries. About 40% of project beneficiaries participated in A4N agricul tural conservation, good agricultural practices and organic management, and group formation related interventions. Between 23% and 35% participated in seed and seedling production, savings, integrated pest management and farmer field school CIALes, while 1 5% reported participation on postharvest management activities ( Table J 2). Most of these interventions involved training and workshops on the different technologies and practices promoted by the project. 153 Table J 1. Outcome and explanatory variables for impact evaluation analysis. 2009 Outcome Variables A4N n=282 non -A4N n=294 Dif Variables Unit Definition Mean sd. Mean sd. Mean Agricultural Conservation Structures (Length built in meters between 2009 and 2011) All structures m/Mz Length built in agricultural conservation structures 2011-2009. Stone barriers/terraces m/Mz Length built in stone barriers and terraces 2011 -2009 Live barriers m/Mz Length built in live barriers 2011-2009 Ditches m/Mz Length built in ditches 2011 -2009 Agricultural Conservation Practices All practices 1=yes, 0=no hh has implemented at least one cons ag practice in one of the plots under its management 0.71 (0.45) 0.68 (0.47) 0.03 Minimum tillage 1=yes, 0=no hh has implemented minimum tillage at least in one plot 0.18 (0.39) 0.13 (0.34) 0.05 Zero tillage 1=yes, 0=no hh has implemented zero tillage at least in one of its plots 0.15 (0.36) 0.16 (0.36) 0.00 Vermiculture 1=yes, 0=no hh has implemented vermiculture at least in one of its plots 0.01 (0.10) 0.01 (0.10) 0.00 Cover crops 1=yes, 0=no hh has implemented cover crops at leas t in one of its plots 0.00 (0.00) 0.01 (0.12) -0.01** Levels of significance ***1%, **5%, *10% . sd refers to standard deviation. hh refers to household. 1 Mz=1.73 Acres 154 Table J 1. (cont™d ). 2011 Outcome Variables A4N n=282 non -A4N n=294 Dif Variables Unit Definition Mean sd. Mean sd. Mean Agricultural Conservation Structures (Length built in meters between 2009 and 2011) All structures m/Mz Length built in agricultural conservation structures 2011-2009 116 (387) 41 (116) 75*** Stone barriers/terraces m/Mz Length built in stone barriers and terraces 2011 -2009 43 (7) 21 (119) 23*** Live barriers m/Mz Length built in live barriers 2011 -2009 23 (76) 8 (49) 15*** Ditches m/Mz Length built in ditches 2011 -2009 9 (41) 2 (27) 7*** Agricultural Conservation Practices All practices 1=yes, 0=no hh has implemented at least one cons ag practice in one of the plots under its management 0.90 (0.30) 0.84 (0.37) 0.06 Minimum tillage 1=yes, 0=no hh has implemented minimum tillage at least in one plot 0.27 (0.44) 0.36 (0.48) -0.09** Zero tillage 1=yes, 0=no hh has implemented zero tillage at least in one of its plots 0.45 (0.50) 0.26 (0.44) 0.19*** Vermiculture 1=yes, 0=no hh has implemented vermiculture at least in one of its plots 0.06 (0.23) 0.01 (0.08) 0.05*** Cover crops 1=yes, 0=no hh has implemented cover crops at leas t in one of its plots 0.02 (0.13) 0.00 (0.06) 0.01*** Levels of significance ***1%, **5%, *10% . sd refers to standard deviation. hh refers to household.1 Mz=1.73 Acres 155 Table J 1. (cont™d). 2009 Outcome Variables A4N n=282 non -A4N n=294 Dif Variables Unit Definition Mean sd Mean sd. Mean Storage Practices hh experienced stored grain losses 1=yes, 0=no hh has experienced stored grain losses. Only for households that stored grain. 0.41 (0.49) 0.26 (0.44) 0.15*** hh stored grain in metallic silos 1=yes, 0=no hh uses metallic silos for grain storage. Only for households that stored grain 0.16 (0.36) 0.19 (0.39) -0.04 Number of metallic silos number Number of metallic silos owned by the household 0.26 (0.56) 0.27 (0.54) -0.01 Kitchen Garden hh had a kitchen garden 1=yes, 0=no hh has a kitchen garden 0.06 (0.24) 0.04 (0.21) 0.02 Small Livestock Pigs owned number Pigs in livestock inventory on January 1st 0.40 (0.89) 0.36 (0.88) 0.04 Goats owned number Goats in livestock inventory on January 1st 0.02 (0.25) 0.14 (1.84) -0.12 Poultry owned number Poultry in livestock inventory, on January 1st 9.17 (8.50) 8.22 (8.71) 0.94 Savings and Credit hh has savings 1=yes, 0=no hh had savings on January 1st 0.14 (0.35) 0.11 (0.31) 0.04 hh has credit 1=yes, 0=no hh had credit on January 1st 0.20 (0.40) 0.21 (0.41) -0.01 Levels of significance ***1%, **5%, *10% . sd refers to standard deviation. hh refers to household. 156 Table J 1. (cont™d). 2011 Outcome Variables A4N n=282 non -A4N n=294 Dif Variables Unit Definition Mean sd Mean sd Mean Storage Practices hh experienced stored grain losses 1=yes, 0=no hh has experienced stored grain losses. Only for households that stored grain. 0.22 (0.41) 0.25 (0.43) -0.03 hh stored grain in metallic silos 1=yes, 0=no hh uses metallic silos for grain storage. Only for households that stored grain 0.28 (0.45) 0.23 (0.42) 0.05 Number of metallic silos number Number of metallic silos owned by the household 0.46 (0.66) 0.33 (0.57) 0.13** Kitchen Garden hh had a kitchen garden 1=yes, 0=no hh has a kitchen garden 0.10 (0.31) 0.05 (0.21) 0.06*** Small Livestock Pigs owned number Pigs in livestock inventory on January 1st 0.68 (1.18) 0.46 (0.87) 0.23*** Goats owned number Goats in livestock inventory on January 1st 0.05 (0.34) 0.01 (0.17) 0.04* Poultry owned number Poultry in livestock inventory, on January 1st 12.59 (12.01) 11.84 (10.56) 0.75 Savings and Credit hh has savings 1=yes, 0=no hh had savings on January 1st 0.29 (0.45) 0.12 (0.32) 0.17*** hh has credit 1=yes, 0=no hh had credit on January 1st 0.26 (0.44) 0.28 (0.45) 8.00 Levels of significance ***1%, **5%, *10% / sd refers to standard deviation. hh refers to household. 157 Table J 1. (c ont™d ). 2009 Outcome Variables A4N n=282 non -A4N n=294 Dif Variables Unit Definition Mean sd. Mean sd. Mean Food Scarcity hh experience food scarcity 1=yes, 0=no hh experienced a period of the year when they could not cook one of the daily meals 0.39 (0.49) 0.32 (0.47) 0.07 Agricultural income and households wealth Bean production QQ=100 Kg Total bean production in quintals 13.32 (16.35) 12.08 (14.27) 1.23 Maize production QQ=100 Kg Total maize production in quintals 17.44 (17.30) 17.03 (19.67) 0.41 Tropical livestock units TLU Conversion factors are: horses 0.8; cattle and mule 0.7; asses 0.5; pigs 0.2; goat, sheep 0.1; poultry 0.01 1 1.96 (2.03) 2.14 (2.66) -0.18 Farm gross margins C$ 2011 Total revenues minus total costs of all the agricultural and livestock activities 4,938 (32,729) 6,902 (34,873) -1,964 Total agricultural sales C$ 2011 Total revenues: quantity sold X prices per unit sold for all crops and livestock products 12,746 (35,056) 13,416 (40,608) -670 Value main crops C$ 2011 Total production of maize, beans, sorghum and millet X unit prices at village level 12,495 (15,292) 10,437 (10,912) 2,058* Value of livestock and poultry products C$ 2011 Total production from livestock X self -reported sale prices 1,371 (2,573) 1,500 (3,333) -128 1 source:http://www.ilri.cgiar.org/InfoServ/Webpub/fulldocs/X5443E/X5443E04.HTM Levels of significance ***1%, **5%, *10% . sd refers to standard deviation. hh refers to household. U$1=C$22.42 in 2011 158 Table J 1. (c ont™d ). 2011 Outcome Variables A4N n=282 non -A4N n=294 Dif Variables Unit Definition Mean sd Mean sd Mean Food Scarcity hh experience food scarcity 1=yes, 0=no hh experienced a period of the year when they could not cook one of the daily meals 0.21 (0.41) 0.21 (0.41) 0.00 Agricultural income and households wealth Bean production QQ=100 Kg Total bean production in quintals 13.99 (15.30) 16.50 (19.05) -2.52* Maize production QQ=100 Kg Total maize production in quintals 16.44 (16.64) 18.25 (18.66) -1.81 Tropical livestock units TLU Conversion factors are: horses 0.8; cattle and mule 0.7; asses 0.5; pigs 0.2; goat, sheep 0.1; poultry 0.01 1 2.56 (2.29) 2.71 (3.15) -0.14 Farm gross margins C$ 2011 Total revenues minus total costs of all the agricultural and livestock activities 5,682 (42,150) 7,025 (61,161) -1343 Total agricultural sales C$ 2011 Total revenues: quantity sold X prices per unit sold for all crops and livestock products 19,939 (58,820) 23,531 (67,516) -3592 Value main crops C$ 2011 Total production of maize, beans, sorghum and millet X unit prices at village level 16,659 (16,221) 18,505 (18,702) -1845 Value of livestock and poultry products C$ 2011 Total production from livestock X self -reported sale prices 6,869 (11,652) 7,027 (7,675) -157 1 source:http://www.ilri.cgiar.org/InfoServ/Webpub/fulldocs/X5443E/X5443E04.HTM Levels of significance ***1%, **5%, *10% . sd refers to standard deviation. hh refers to household. U$1=C$22.42 in 2011 159 Table J 2. Participation of A4N household in different project interventions 2009 -2011. Variable Number % of total Agricultural conservation 115 41% Good agricultural practices and organic management 113 40% Group formation 109 39% Seed and seedling production 98 35% Savings 93 33% Integrated pest management (IPM) 72 26% Farmer field schools and local research committees (CIAL) 64 23% Postharvest management 42 15% Marketing and product transformation 32 11% Small livestock management 30 11% Note: Total of participant households 282 Source: A4N Household Survey 2012 The evaluation of certain outcomes, such as use of biofortified seeds of maize and beans, was not conducted due to small sample size or data gaps. Only 15 sampled households reported that they grew biofortified beans in 2011 and zero households reported that they grew biofortified maize in the same year. In addition to this, the information on the variables planted by farmers does not allow for analysis, since names of varieties reported by farmers do not coincide with technical names available. Some group level interventions also had a very low take up. Only 12 house holds reported activities related with beekeeping in 2011, only 13 households reported having a group garden and only five households reported that they participated on seed producer groups. The project impact evaluation will focus on the household -level interventions that promoted improved agricultural technologies and practices, and savings; they do not analyze village level interventions. 160 Appendix K. Difference in difference estimation of project outcomes Table K 1. Project impacts on building of agricultural conservation structures. Conserv ag structures m/Mz Stone barriers m/Mz Live barriers m/Mz Ditches m/Mz A4N=1 77.41*** 24.32*** 15.53*** 7.44*** (25.17) (9.85) (5.43) (3.00) Household size 1.95 0.95 -0.67 1.28* (5.71) (4.42) (1.21) (0.76) Average education hh members years -0.80 -2.96 -0.33 0.29 (4.00) (2.78) (1.24) (0.73) Area of cultivated land Mz 3.44* 2.27 0.68 0.01 (2.14) (1.48) (0.45) (0.07) Constant 36.48*** 16.16** 7.04* 2.36 (9.00) (6.07) (2.73) (2.03) R-squared 0.01 0.02 0.01 0.01 N 567 567 567 567 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 161 Table K 2. Project impact on agricultural conservation practices Cons ag practices Minimum tillage Zero tillage Vermiculture Cover crops A4N participant 0.04 -0.14*** 0.19*** 0.05*** 0.03*** (0.05) (0.05) (0.05) (0.02) (0.01) Household size -0.01 -0.01 -0.00 0.00 -0.00 (0.01) (0.02) (0.02) (0.00) (0.00) Average education hh members years 0.01 -0.01 -0.01 0.00 0.00 (0.01) (0.01) (0.01) (0.01) (0.01) Area of cultivated land Mz 0.01 0.00 -0.00 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.15*** 0.22*** 0.10*** -0.00 -0.01* (0.04) (0.04) (0.03) (0.01) (0.01) R-squared 0.01 0.02 0.03 0.02 0.02 N 567 567 567 567 567 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 162 Table K 3. Impacts on postharvest storage Stored grain losses Stored in metallic silos Number of metallic silos owned A4N participant -0.16*** 0.11*** 0.14*** (0.06) (0.04) (0.05) Household size -0.03* -0.00 0.01 (0.02) (0.01) (0.01) Average education hh members years 0.02 0.00 0.03** (0.02) (0.01) (0.01) Area of cultivated land Mz 0.00 0.00 0.00 (0.00) (0.00) (0.00) Constant -0.03 0.05** 0.08*** (0.04) (0.03) (0.03) R-squared 0.02 0.02 0.03 N 476 476 575 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 163 Table K 4. Project impacts on the number of small livestock Number pigs Number goats Number poultry A4N participant 0.18* 0.15 -0.10 (0.10) (0.11) (0.99) Household size 0.06* -0.00 0.41 (0.03) (0.01) (0.31) Average education hh members years 0.03 -0.01 -0.34 (0.03) (0.02) (0.33) Area of cultivated land Mz -0.02 -0.01 0.20 (0.04) (0.01) (0.11) Constant 0.15** -0.12 3.24*** (0.07) (0.10) (0.69) R-squared 0.02 0.01 0.02 N 575 575 575 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 164 Table K 5. Project impacts on savings, credit, hunger, crop losses and kitchen garden. Saving Credit Experienced period of hunger Experienced crop losses Kitchen garden A4N participant 0.14*** -0.01 -0.06 0.04 0.04 (0.04) (0.04) (0.05) (0.04) (0.03) Household size 0.02* 0.01 0.02 0.00 0.00 (0.01) (0.01) (0.02) (0.01) (0.01) Average education hh members years 0.03* -0.01 -0.02 0.01 -0.01 (0.01) (0.01) (0.01) (0.01) (0.01) Area of cultivated land Mz 0.00 0.00 0.01 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.04 0.06* -0.13*** 0.02** -0.01 (0.03) (0.03) (0.03) (0.03) (0.02) R-squared 0.04 0.01 0.02 0.00 0.01 N 575 575 575 575 571 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 165 Appendix L. Difference in difference treatment effects by area of cultivated land in 2009. Table L 1. DID treatment effect on agricultural conservation structures, for households with cultivated land >= 1.5 Mz Conserv ag structures m/Mz Stone barriers m/Mz Live barriers m/Mz Ditches m/Mz A4N=1 110.68 2.69 16.29 11.37** (73.34) (27.10) (15.00) (5.33) Household size -9.58 -9.45* -2.81 0.39 (10.55) (5.18) (3.29) (0.60) Average education hh member years 1.34 -8.63 3.45 -0.97 (14.19) (10.33) (4.22) (1.39) Area of cultivated land Mz 10.84 9.86 1.86 -0.49 (10.20) (7.93) (1.95) (0.33) Constant 60.67*** 28.71* 13.88* 0.40 (24.00) (15.14) (7.73) (0.63) R-squared 0.00 0.04 -0.01 0.01 N 186 186 186 186 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 166 Table L 2. DID treatment effect on agricultural conservation structures, for households with 1.5Mz3Mz Conserv ag structures m/Mz Stone barriers m/Mz Live barriers m/Mz Ditches m/Mz A4N=1 74.39*** 31.49*** 17.83*** 7.72 (26.68) (11.17) (7.16) (7.63) Household size 0.43 -0.16 0.96 2.12* (4.74) (1.55) (1.31) (1.30) Average education hh member years 0.33 1.70 -2.29 2.41 (5.06) (1.53) (2.28) (2.15) Area of cultivated land Mz 0.91 0.25 0.33 0.07 (0.84) (0.27) (0.32) (0.16) Constant 17.30** 3.85** 4.24** 8.04 (8.44) (1.57) (1.98) (6.50) R-squared 0.03 0.04 0.03 -0.00 N 185 185 185 185 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 168 Table L 4. DID treatment effect on agricultural conservation practices, for households with Cultivated land <=1.5Mz Cons ag practices Minimum tillage Zero tillage Vermiculture Cover crops A4N participant 0.20** -0.08 0.20** 0.05** 0.03 (0.09) (0.09) (0.08) (0.03) (0.02) Household size -0.03 0.01 -0.01 0.00 0.00 (0.03) (0.03) (0.03) (0.00) (0.00) Average education hh member years -0.00 -0.03 -0.06** -0.01 -0.03* (0.03) (0.04) (0.03) (0.02) (0.01) Area of cultivated land Mz 0.01 -0.01 0.01 -0.00 -0.00 (0.01) (0.01) (0.01) (0.00) (0.00) Constant 0.18** 0.25*** 0.03 0.01 0.00 (0.07) (0.07) (0.06) (0.02) (0.01) R-squared 0.04 0.01 0.06 0.03 0.05 N 186 186 186 186 186 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 169 Table L 5. DID treatment effect on agricultural conservation practices, for households with 1.5Mz 3Mz Cons ag practices Minimum tillage Zero tillage Vermiculture Cover crops A4N participant -0.06 -0.30** 0.19* 0.08* 0.03 (0.06) (0.09) (0.08) (0.04) (0.02) Household size -0.00 -0.01 0.01 0.00 -0.00 (0.02) (0.02) (0.03) (0.01) (0.00) Average education hh member years 0.01 0.01 0.02 0.01 0.01 (0.02) (0.02) (0.02) (0.01) (0.01) Area of cultivated land Mz 0.00 0.00 -0.00 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.11* 0.26*** 0.11* 0.00 -0.02 (0.05) (0.07) (0.05) (0.02) (0.01) R-squared 0.01 0.06 0.04 0.04 0.05 N 185 185 185 185 185 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 171 Table L 7. DID treatment effect on postharvest management, for households with Cultivated land< =1.5Mz Stored grain losses Stored in metallic silos Number of metallic silos owned A4N participant -0.06 0.06 0.07 (0.12) (0.06) (0.07) Household size -0.01 -0.01 -0.01 (0.04) (0.02) (0.02) Average education hh member years -0.00 -0.03 0.04* (0.05) (0.02) (0.02) Area of cultivated land Mz 0.01 0.01 -0.01 (0.02) (0.01) (0.01) Constant -0.10 0.01 0.13** (0.08) (0.04) (0.06) R-squared 0.003 0.02 0.02 N 147 147 191 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 172 Table L 8. DID treatment effect on postharvest management, for households with 1.5Mz3Mz Stored grain losses Stored in metallic silos Number of metallic silos owned A4N participantA4N participant -0.12 0.10 0.21* (0.09) (0.08) (0.10) Household size -0.03 0.02 0.05 (0.02) (0.02) (0.03) Average education hh member years 0.07* 0.03 0.03 (0.03) (0.02) (0.03) Area of cultivated land Mz 0.00 0.00 0.00 (0.00) (0.00) (0.00) Constant 0.03 0.14** 0.06 (0.06) (0.05) (0.05) R-squared 0.05 0.04 0.06 N 160 160 185 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 174 Table L 10. DID treatment effect on small livestock, for households with Cultivated land <=1.5Mz Number pigs Number goats Number poultry A4N participant 0.06 0.40 0.60 (0.11) (0.34) (1.77) Household size -0.01 0.00 0.53 (0.04) (0.02) (0.64) Average education hh member years 0.03 -0.01 -1.21 (0.03) (0.08) (0.98) Area of cultivated land Mz 0.10 -0.08* 0.17 (0.07) (0.04) (0.20) Constant -0.03 -0.18 1.93 (0.13) (0.24) (1.35) R-squared 0.19 0.03 0.02 N 191 191 191 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 175 Table L 11. DID treatment effect on small livestock, for households with 1.5Mz 3Mz Number pigs Number goats Number poultry A4N participant 0.26 -0.01 0.19 (0.20) (0.04) (1.77) Household size 0.11* 0.00 1.34** (0.06) (0.00) (0.45) Average education hh member years 0.12* 0.01 -0.42 (0.05) (0.01) (0.47) Area of cultivated land Mz -0.04** -0.00 0.13 (0.03) (0.00) (0.11) Constant 0.24 0.03 3.23** (0.16) (0.04) (1.14) R-squared 0.09 0.003 0.06 N 185 185 185 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 177 Table L 13. DID treatment effect on saving, credit, hunger, losses and kitchen garden, for households with Cultivated land<= 1.5Mz Saving Credit Experienced period of hunger Experienced crop losses Kitchen garden A4N participantA4N participant 0.22*** 0.10 -0.03 0.14* 0.12** (0.07) (0.07) (0.08) (0.08) (0.05) Household size 0.01 -0.02 0.05 0.03 0.01 (0.02) (0.02) (0.03) (0.02) (0.02) Average education hh member years 0.02 0.03 -0.02 0.01 -0.03 (0.02) (0.03) (0.03) (0.04) (0.02) Area of cultivated land Mz 0.01 0.01 -0.01 0.00 -0.00 (0.01) (0.01) (0.01) (0.01) (0.01) Constant -0.02 0.03 -0.17** -0.07 0.02 (0.05) (0.05) (0.06) (0.06) (0.03) R-squared 0.06 0.03 0.02 0.03 0.04 N 191 191 191 191 189 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 178 Table L 14. DID treatment effect on saving, credit, hunger, losses and kitchen garden, for households with 1.5 3Mz Saving Credit Experienced period of hunger Experienced crop losses Kitchen garden A4N participantA4N participant 0.09 -0.13 -0.08 -0.05 0.02 (0.08) (0.09) (0.08) (0.07) (0.05) Household size 0.05** 0.04* 0.01 0.02 -0.00 (0.02) (0.02) (0.02) (0.02) (0.02) Average education hh member years 0.04 -0.02 0.02 0.05** -0.02* (0.03) (0.02) (0.03) (0.02) (0.01) Area of cultivated land Mz -0.00 -0.00 0.01*** 0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) Constant 0.03 0.08 -0.06 0.14* -0.06 (0.05) (0.06) (0.05) (0.05) (0.03) R-squared 0.06 0.03 0.00 0.03 0.02 N 185 185 185 185 184 Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 180 Appendix M. Impacts on long -term outcomes related with agricultural income and household wealth. The anal ysis of overall project impacts aims to determine the impact of the adoption of the different practices promoted by the project on outcomes such as production, gross margins and revenues from agricultural activities. The long -term goal of the project is to increase agricultural income and household wealth . Several proxy variables were available to measure these project impacts. For agricultural income, they include : bean yields , maize yields, farm gross margins, total value of production of main crops (maize, bean, sorghum and millet), t otal value of production of livesto ck products (meat, eggs, dairy) . Another long -term outcome of interest is increases in household wealth. An indicator of agricultural assets is the number of tropical livestock units (TLU) . Impacts were measure using the same methods used in Chapter 2 , FD, PSM-DID and PSW. The FD estimates suggest no impacts on bean and maize yields ( Table M 1 ). Since the project was on its earlier stages when this impact evaluation was conducted, is possible that farmers are still learning and experimenting with new practices that are not yet translated into increases in yields for the main crops they produced. In gen eral t he estimates using FD, PSM -DID and PSW suggest that the project did not have an impact in agricultural income and household wealth related outcomes. The estimates for the monetary measures of agricultural income are not 181 statistically significant and very imprecise, as shown by its very high standard errors ( Table M 1)10. Table M 1. Project impact on agricultural income and household wealth related outcomes. PSM -DID Difference outcome variables DID untrimmed kernel (epan) NN(5) llr (tricube) PSW Bean yields qq/Mz 1 -0.76 -0.85 -0.72 -0.86 -2.05 (0.89) (1.32) (1.50) (1.68) (2.57) Maize yields qq/Mz 2 -1.23 -1.98 -2.23 -1.42 -0.36 (1.86) (2.82) (2.95) (3.25) (1.14) TLU 3 0.09 -0.03 -0.02 -0.05 -0.04 (0.17) (0.20) (0.22) (0.22) (0.21) Farm gross margins C$2011 735 3006 2650 2611 3259 (3869) (3142) (3651) (3298) (2690) Value main crops 4 C$2011 -2755* -760 -601 -780 -1547 (1507) (1814) (2015) (1963) (1752) Value of livestock and poultry products 5 C$2011 59 684 835 837 720 (804) (800) (935) (848) (902) Standard errors (se) in parenthesis, DID robust se, PSM -DID and weighed PS se bootstrap with 1000 repetitions Levels of significance ***1%, **5%, *10% U$1=C$22.42 in 2011, qq=100Kg NN refers to nearest neighbor, LLR to local linear regression Untrimmed sample n=575, trimmed sample for PSW -DID and PSW n=554 A total of 265 pairs formed with PSM -DID 1 Excludes intercropping, non-trimmed sample n=335, trimmed sample n=323 2 Exclu des intercropping, non-trimmed sample n=246, trimmed sample n=234 3 TLU refers to tropical livestock units, conversion factors are: horses 0.8; cattle and mule 0.7; asses 0.5; pigs 0.2; goat, sheep 0.1; poultry 0.01 source: http://www.ilri.cgiar.org/InfoServ/Webpub/fulldocs/X5443E/X5443E04.HTM 4 Estimated at village level prices, only for households that produced maize, beans and sorghum. Non trimmed sample n=541, trimmed sample n=521 5 Estimated at self -reported prices 10 Due to this issue, we conducted quantile regression to estimate average treatment effects in the median (not presented here) instead of the mean, but the results were also not statistically significant. 182 The descriptive statistics for project outcomes appear in Table M 2. Farm gross margins, value of main crops and value of livestock products were measured with high variability. Standard errors are high for both 2009 and 2011, which is explained by the effects of extreme weather events on those years in crop production and yields. The variability due to weather shocks incorporates noise into the estimates, making it difficult to draw conclusions from the analysis. Moreover, it i s not surprising that we do not find impacts in income and wealth just after two years of the project. Looking at the analysis of outcomes by area of cultivated land, results suggest that households with small area decreased their production of beans and m aize ( Table M 2). Therefore the overall decrease in bean production that we found ( Table M 2 ) comes from the decreases for this group. During 2009 -2011 the percentage of households in this group that experienced crop losses of at leas t 25% increased 14%, and crop losses for farmers included in the sample were severely felt during the heavy rains in 2011. However, thanks to the project, counting with savings might have helped these groups of farmers to cope with this situation. 183 Table M 2. Heterogeneity of program impacts in outcomes related to agricultural income and household wealth. Terc iles of Cultivated land <=1.5Mz n=191 1.53Mz n=186 Outcomes Coef se Coef se Coef se Bean yields qq /Mz -1.90 (1.99) 0.01 (1.11) -1.05 (1.61) Maize yields qq/Mz -3.43 (4.01) -2.31 (2.76) 1.47 (3.62) TLU 1 0.18 (0.23) -0.01 (0.24) 0.13 (0.37) Farm g ross Margins C$2011 -4770 (3089) -960 (4009) 9543 (10668) Value main crops C$2011 -5523* (2482) -2566 (2285) -2385 (3052) Value of livestock products C$2011 532 (1156) -358 (946) 277 (1996) Robust standard errors (se) in parenthesis Levels of significance ***1%, **5%, *10% 1 Mz = 1.73 acres, 1 qq = 100 Kg U$1=C$22.42 in 2011 1 TLU refers to tropical livestock units, conversion factors are: horses 0.8; cattle and mule 0.7; asses 0.5; pigs 0.2; goat, sheep 0.1; poultry 0.01 source: http://www.ilri.cgiar.org/InfoServ/Webpub/fulldocs/X5443E/X5443E04.HTM 184 Appendix N. Pretreatment characteristics of villages considered for the trust game. Table N 1. Village pretreatment characteristics. Village Population 2005 Households 2005 % inadequate housing 2005 % houses no electricity 2005 % houses no piped water 2005 % households produce basic grains 2003 % farms with landholding > 10 Mz 2003 non-A4N Llano Redondo 168 38 46% 44% 59% 94% 59% El Quebracho 125 25 93% 53% 17% 94% 58% Licoroy 216 49 57% 67% 90% 78% 22% San Lorenzo 296 58 34% 41% 87% 73% 23% Cebadilla 157 33 73% 84% 46% 100% 13% La Laguna 131 25 17% 41% 79% 50% 50% Llano Largo 377 78 19% 30% 30% 82% 44% Las Lajas - - - - - 73% 40% Cuajiniquil - - - - - 94% 47% Las Puertas 200 38 67% 86% 84% 100% 38% A4N San Antonio 160 31 57% 76% 81% 100% 30% El Espinal 655 142 54% 84% 85% 100% 40% Las Cuevas 699 114 60% 18% 95% 91% 52% Las Cañadas 307 67 26% 6% 2% 42% 67% Tomabu 585 128 74% 30% 96% 100% 78% Las Tablas 322 75 45% 3% 80% 92% 48% Rosario (Arriba+ Abajo) 613 126 74% 88% 53% 8% 45% La Pacaya 289 53 63% 85% 85% 2% 100% Concepción 348 79 25% 17% 22% 100% 9% Mechapa 641 135 27% 32% 83% 75% 61% El Hornillo 251 51 49% 65% 57% 75% 25% Las Gavetas 156 39 41% 5% 86% 100% 25% Quebrada de Agua 203 38 60% 84% 88% 83% 83% Average non A4N 209 43 51% 56% 61% 84% 39% Average A4N 402 83 50% 46% 70% 75% 51% P values 0.02 0.02 0.95 0.47 0.50 0.45 0.23 Source: Instituto Nacional de Estadisticas y Censos (INIDE) Nicaragua Ministerio Agropecuario y Forestal (MAGFOR) Nicaragua 185 Appendix O. Number of groups per village, number of group members in 2011, population and households 2005 , considered for the trust game . Table O 1. Number of groups per A4N village, number of members in 2011, pretreatment population and households in each village 2005. Village A4N groups No. Group members Population 2005 Households 2005 Cañada 2 33 307 67 Espinal 2 23 655 142 Hornillo 2 34 251 51 La Concepcion 1 7 348 79 La Laguna 1 8 131 25 La Pacaya 1 7 289 53 Las Animas 1 16 129 25 Las Cuevas 2 33 699 114 Las Gavetas 4 61 156 39 Las Lomas 1 14 95 22 Las Tablas 2 63 322 75 Mechapa 2 27 641 135 Monte Verde 1 13 144 30 Potrerillo 2 21 122 24 Rosario Abajo 3 37 613 126 San Antonio 1 17 160 31 Tablas 1 25 322 75 Tomabu 4 87 585 128 coefficient of correlation between number of groups 0.41 and population 2005 Source: Fundación para la Investigación y el Desarrollo Rural (FIDER) Instituto Nacional de Estadisticas y Censos (INIDE) Nicaragua 186 Appendix P. Consent script for the trust game. Statement of Consent Program Participation, Economic Impact, and Agricultural Practices among Nicaraguan Smallholder Farmers We are conducting a research study to understand trust attitudes among farmers in villages of western Nicaragua. The study is conducted by Michigan State University and Nitlapan at the Universidad Centroamericana. From this study, resea rchers hope to gain methodological insight in experimental economics. You will participate in an economic experiment with us today, consistent in determining monetary amounts you are willing to share with others. At the end of this activity you will be pre sented with a short survey. The survey will first ask you to identify several demographic features about your household, your farm activities and your participation in agricultural projects. To participate in this activity you must be 18 years old or olde r. Your participation in this experiment and the subsequent survey is voluntary, so you are free not to participate at all and you may terminate your participation at any time with no penalty. However, I want to encourage you to participate, since this act ivity will allow us to better design rural development projects. This session will take one hour. At the end of the agenda for the day you will receive the earnings from the experiment. Although you will not directly benefit from your participation in this study, however, the lessons from it may help in designing better agricultural projects and we know of no risks associated with this study. Your privacy will be protected to the maximum extent allowable by law. All the information you provide us will be kept confidential , with the questionnaire locked in a cabinet at Michigan State University for three years after the research is completed. This means that no one except the researchers and the MSU Human Research Protection Program will have access to your answers . We will not identi fy you or your household in any publication from this study . If you have any questions about this study, such as scientific issues, how to do any part of it, or to report an injury, please contact Professor Scott Swinton by em ail ( swintons@msu.edu ), by telephone at (1) 517 -353-7218, or by postal mail at Michigan State University, East Lansing, MI 48824-1039, USA. If you have questions or concerns about your role and rights as a research participant, would like to obtain information or offer input, or would like to register a complaint about this study, you may contact, anonymously if you wish , the Michigan State University™ s Human Research Prote ction Program at 517 -355-2180, f ax 517 -432-4503, or e -mail irb@msu.edu or regular mail at 207 Olds Hall, M ichigan State University , East Lansing, MI 48824. You indicate your voluntary agreement to participate by beginning with the experiment. Thank you again for your help with this important research about trust among farmers in villages in western Nicaragua . 187 Appendix Q. Game procedures in the field Contact local leaders to conduct recruitment. In our case since we wanted to have 20 subjec ts per session, we recruited 30 subjects. Everyday night before each activity, we have to make sure that we have all the lists of recruited subjects, templates (Cardenas & Ramos, 2006), surveys, fake bills and money required for the activities is ready. For the different sessions, find either schools, community centers, churches or big houses, where is possible to use two separate spaces for conducting the sessions. When recruited subjects arrive at the experiment venue, take down their name (either write it down on a participant sheet, or check them off a list you already have), assign them to a role (sender or returner) and give them a participant number. In a piece of paper, or computer spreadsheet, we will have available the code of each player and the codes for each couple, in such a way that when recording time comes up, we will just need to know the participant™s code. Couples will be already be formed, by rando mly assigned number in excel spreadsheet, and nobody but the PI will have this information. Participants™ code will be handed to each participant in big pieces of paper that can be seen by the researchers conducting the experiments. After a number of sub jects of at least 14 and after waiting for reasonable time for subjects to arrive, the session will start. Once the session has started a person will be at the door indicating to subjects arriving late that the session has already started and no more peopl e is allowed. This is done with the aim of having all subjects participating in the sessions have the same information from the very beginning, and for not having disrupt the explanation of the dynamic of the game to subjects who arrived on time. We will introduce ourselves (PI and helpers) read the consent script aloud and provide copies of it to subjects. Then the instructions for the game will be given orally to all subjects. Then explain the dynamic of the game, we will do this adapting Cardenas & Ram os (2006) (see Peralta and Shupp, 2013) manual for economic games with the helpers we will also act the game in front of all the participants in the activity, and provide examples of possible outcomes of the game, to ensure understanding of the activity by everybody attending the sessions. 188 Here in some of the examples to be provided during the explanation of the game: Initial S Initial R Sent S to R Kept S Received R Total R Returned R to S Total earning S Total earning R 100 100 50 50 150 250 50 100 200 100 100 0 100 0 100 0 100 100 100 100 100 0 300 400 200 200 200 100 100 80 20 240 340 160 180 180 100 100 20 80 60 160 20 100 140 100 100 60 40 180 280 120 160 160 100 100 30 70 90 190 50 120 140 S: sender, R: receiver. Separate the two groups in different rooms or spaces and ask them not to talk to each other, having a facilitator in the room or space with each group. Senders will be called one by one to a separate area, away from both the rest of the senders and receivers, to make their decision on how much they will send. While this process is taking place, the helper in the room with the receivers will be asking them a set of trust attitudinal questions, in our case the GSS questions and a set of questions written by our own. Have a desk with a cover area (maybe using cardboard box for that) where we ask the participants to come over, first start with the senders, one of the facilitators will be there with them to repeat the instructions. They will be given two envelopes with a set of blank fake bills, and one color envelope with a set of fake bills with denominations of C$10 each, for a total of C$100, the amount that we will be giving to each participant as endowment. Using two different colors to indicate what is sent and what is received. The senders will be told that: fiHere you have a color envelope with 10 fake bills of C$10, and a blank envelope with blank fake bills. You will make a decision on how much from those fake bills to send to a receiver, the identity of the receiver will not b e reveal to you. Please remember that the amount that you will be sending to the receiver will be tripled. For example, if you send C$10, the receiver will get C$30, of you send C$50, the receiver will get C$150, if you send C$100, the receiver will get C$ 300. Please let me know if you have any questionsfl fiLater on this activity you will receive some money back from the receiver, or the other person that will send money to you. Please let us know writing down on this piece of paper, how much money are you expecting to receive back. You can write the amount down in private and fold this piece of paper when returning it to mefl. 189 After all the senders have made their decisions, and these decisions have been recorded and the amounts sent tripled, we will start w ith the process with the receivers. Receivers are called one by one to make their decisions on the space designated for that. Give the receiver the following instructions: fiHere you have a color envelope with the amount of money you have been sent by the sender, that has been tripled, as explain previously to you. For example, if you were sent C$10, you are receiving C$30, if you were sent C$50, you will received C$150, if you were sent C$100, you will receive C$300, you will receive this amount plus a C$1 00 endowment. All this money will be given to you in bills of C$10fl. fiBefore you make this decision, and open this envelope, please write down on a piece of paper, how much you are expecting to receive from the sender. Please let us know writing down on th is piece of paper. You can write the amount down in private and fold this piece of paper when returning it to mefl. While receivers are being called to learn how much they were sent and to make their decision on how much to return, the helper in the sender s room will be asking them the attitudinal trust questions mentioned before, the GSS questions and the attitudinal questions we wrote. A separate area will be set for data recording. We put the participants code in a piece of paper inside the envelope all owing recording the amount sent once we receive the envelope, this amount will be tripled, we will take out the number of the sender and introduce the number of the receiver in the envelope. After all the senders make their decisions, and all the envelopes for the receivers have been organized, they are put in a bag, where they will be drop randomly by the helper, who will start calling the receivers to make their decisions. We collect all the data and ask for the participants to return their envelopes. The n while we collect the data and organize the payments, the subjects will be filling a survey. Subjects who complete filling the survey, can come to a separate area to receive their earning from the game in cash, they will be asked to leave quietly and t o not to talk about their decisions and the game. 190 Some additional notes: If there are 20 players, 10 will be fisendersfl and 10 will be fireturnersfl. The 10 senders could have participant numbers 1, 2 , 3,–10 while the 10 returners could have participant numbers 11, 12, 13,–20. We could then do a reverse match where 1 is matched with 20, 2 is matched with 19, etc. This would keep people who arrive together from being matched with each other since we should assign roles by alternating as people arrive to mak e sure we have as close to an equal number of senders and returners as possible. We could just assign people participant numbers randomly by handing packets and thus participant numbers out in a quasi -random way as they arrive Œ the only thing you need to be careful with here is keeping track of which numbers you have assigned in case not everybody comes (say 17 of the 20). Then you can adjust your matching scheme to take that into account. We will also probably want to over prepare–that is, have enough stu ff and participant packets etc. for more than 20 in case extra people come and are eligible to participate. What do you do when there are an odd number of participants? Well, if you have been handing out packets/assigning roles such that the odd participa nt (i.e, the 3rd, 5th, 7th, etc arrival) is always assigned the role of fireturnerfl then, while you are opening the fisenderfl envelopes, recording the amount sent, adding the investment return, and putting that into a separate fireturnerfl envelope (with the r eturner™s participant number NOT the sender™s participant number Œ there needs to be this switch so neither participant can figure out who they were paired with by asking later fiso, what participant number were you?fl) you can ficreatefl, by duplicating a ran dom sender™s choice, an extra fireturnerfl envelope for the fireturnerfl participant who doesn™t have a fisenderfl partner. Essentially, you are using one random fisender™sfl decision twice. The extra returner gets to make a decision, but the returned money doesn™ t go to anyone, you intercept it in the process of collecting and recording the fireturnerfl envelopes and decisions. The random sender only gets the fireturnerfl envelope from their assigned partner. Remember, we need to switch the money back into the fisender ™sfl envelope so the sender can™t tell the number of the participant they were matched with. 191 Appendix R. Trust game instruments Trust game Fake bills: Template per couple of players Decisions and results format Couple number: ____________________ (Fulfill a format per each couple that has been formed) Place:___________ Date (day/month/year): ___/___/___ Inicial time: ___:____ (am/pm) A B C D E F G H I Initial amount P1 Initial amount P2 Sent by P1 to P2 Kept P1 Received P2 (Cx3) Total P2 (B+E) Returned P2 Earnings P1 (D+G) Earnings P2 (F-G) Total payment P1: ____________________________ Total payment P2: ____________________________ C$10 192 Register of players decisions Trust game Place: ___________ Date (day/month/year): ___/___/___ Time: ___:____ (am/pm) Enumerator:__________________ Couple Code P1 Code P2 Sent P1 to P2 Kept P1 Received P2 Total P2 Returned P2 Earnings P1 Earnings P2 Templates from Cardenas and Ramos, 2006. 193 Appendix S. Survey instrument, for participants in the trust game. Read consent script before start. [ ] 1 Beneficiary [ ] 0 Non beneficiary Participant ID number: _________________ Date: ____(dd)/____(mm)/________(yyyy) Player type (mark with an X) : [ ] 1 [ ] 2 1. General information: 1.1 Department: __________________________________ 1.2 Municipality: _____________________________________ 1.3 Village: ___________________________________ Question s asked to type 1 player s ONLY : 2 During the ac tivity, did you think that your decisions affected the decisions made by the person you were partnered with? [ ] 1 Yes [ ] 0 No Explain why (briefly) ________________________________________________ __________________________________________________________________ 3 For this kind of activity, do you think most people would think: [ ] 1 The more money people send the more they will get back [ ] 2 People will not send much money, because they are not going to get much money back, regardless of how much money they send. Question s asked to type 2 players ONLY 4 During the activity, did you think that the decision made by the person you were partnered with consid ered what you would be deciding next? [ ] 1 Yes [ ] 0 No Explain why (briefly) ________________________________________________ __________________________________________________________________ 194 5 For this kind of activity, do you think most people would think: [ ] 1 The more money people receive , the more money people will send back. [ ] 2 The more money people receive , the less money people will send back. 6 Demographic characteristics: 6.1 Age? (in years): ____________________________ 6.2 Gender? [ ] 1 Male [ ] 0 Female 6.3 Relationship with household head: [ ] 1 Head [ ] 2 Spouse [ ] 3 Son/daughter [ ] 4 Father/mother [ ] 5 Brother/sister [ ] 6 Grandson/granddaughter [ ] 7 Non relative [ ] 99 Other, specify __________________ 6.4 Number of people in the household? __________________ 6.5 Level of education in years (maximum level approved):____________ 6.6 For how long have you been living in your village? ____________(years/months) 6.7 What do you do most of your time (mark with an X): [ ] 1 Works in own farm [ ] 2 Agricultural worker [ ] 3 Non agricultural worker [ ] 4 Professional [ ] 5 Self employed [ ] 6 Student /attending school [ ] 7 Housewife [ ] 8 Unemployed [ ] 99 Other, specify ________________ 6.8 Participation in groups or associations. [ ] 0 None [ ] 1 Producers group [ ] 2 Marketing group [ ] 3 Savings group [ ] 4 Women™s group [ ] 5 Youth group [ ] 6 Church group [ ] 7 Watershed committee [ ] 8 Village Council [ ] 11 Sports group or team [ ] 12 Political organization [ ] 13 School Committee [ ] 99 Other, specify _________________ 195 6.9 If participate in a group or association, please complete question 6.9, otherwise continue to question 7. In the group the person is part of, the person is: [ ] 1 President [ ] 2 Vice -presid ent [ ] 3 Secretary [ ] 4 Treasurer [ ] 5 Member [ ] 99 Other, specify_____________ 7 Housing characteristics 7.1 Is your house [ ] 1 Owned [ ] 2 Rented [ ] 99 Other, specify________________ 7.1 How do you obtain water for your house? [ ] 1 River, spring [ ] 2 Pipe inside the house [ ] 3 Pipe outside the house [ ] 4 Well [ ] 99 Other, specify_________________ 7.2 What type of hygienic service do you have in your house? [ ] 1 None [ ] 2 Latrine outside the house [ ] 3 Toilet connected to sewage s ystem [ ] 99 Other specify__________________ 7.3 What type of lighting do you have in your house? [ ] 1 Electric power [ ] 2 Candle [ ] 3 Solar panel [ ] 99 Other specify__________________ 8 Land uses 8.1 Distribution of cultivated land in the last cropping season where you grew crops, according the following land uses. Area in Mz Maize Beans Maize and beans intercropped Other annual crops Plantations Forages Other land uses 196 8.2 Land property, please specify the ownership of the plots you grew in the lastcropping season. Plot Area in Mz Ownership 1. Owned 2. Rented 3. Share crop 4. Borrowed 5. Rented 99. Other, specify 1 2 3 4 5 9 Agricultural production 2011 -2012. Please start with the main crops grown in the farm. Crops 1 Maize 2 Beans 99 Other, specify Season Write code 1 Apante 2 Primera 3 Postrera 4 All year Production Unit of measurement Sales Household consumption 197 10 Livestock inventory: Please provide livestock inventory at May 1 st 2012 Animal How many do you have? No. Observations Chicken Turkey Hugs Horses Donkey Mule Cow Bull Steers Calves Goats Sheep Lamb Rabbits Bee hives 11 Agricultural income 11.1 On average, what is the amount of your annual revenues from agricultural activities (annual, perennial crop product sales and livestock, including livestock products)? [ ] 1 C$0 -C$500 [ ] 2 C$501 -C$1000 [ ] 3 C$1001 -C$2000 [ ] 4 C$2001 -C$3000 [ ] 5 C$3001 -C$4000 [ ] 6 C$4001 -C$5000 [ ] 7 C$5001 -C$6000 [ ] 8 C$6001 -C$7000 [ ] 9 C$7001 -C$8000 [ ] 10 C$8001 -C$9000 [ ] 11 C$9001 -C$10000 [ ] 12 C$10001 -C$15000 [ ] 13 15001 and more 198 12 Did you work outside of your farm in other activities duri ng the last 12 months? [ ] Yes [ ] No 12.1 If yes, what did you do? _________________________________ 12.2 How much did you make on this activity, on average, per week?_______________ 12.3 How many weeks? ______________________________ 13 Did you migrate over the past 12 months to work in other part of Nicaragua, or overseas for work? [ ] Yes [ ] No 13.1 If yes, what did you do? _________________________________ 13.2 Where did you go? _____________________________________ 13.3 How much money did you earn in total? _______________U SD/C$ 14 Did you receive any contributions from household members/other people leaving in other parts of Nicaragua or overseas during the last 12 months? [ ] Yes [ ] No 14.1 How much did you receive in total? ___________________USD/C$ 15 Between 2009 and 2012, did you participate in activities with any organization in rural development projects? [ ] Yes [ ] No If yes please mention the activity and the institution, if not continue with question 16. Activity you developed with the Project Institution ________________________________ _______________________ ________________________________ _______________________ ________________________________ _______________________ ________________________________ _______________________ ________________________________ _______________________ 199 The following questions will be asked during the time experiment participants are making their individual decisions. Questions will be asked by helpers and participants will be distributed the answer sheet formats. 16 Generally speaking, do you consider that most people can be trusted, or that you cannot be too careful in dealing with people? [ ] 1 You cannot be too careful in dealing with people [ ] 2 Most people can be trusted [ ] 88 Do not know 17 Do you think most people would try to take advantage of you if they got the chance, or would they try to be fair? [ ] 1 Would take advantage of you [ ] 2 Would try to be fair [ ] 88 Do not know 18 Would you say that most time people try to be helpful, or that they are just looki ng out for themselves? [ ] 1 Just look out for themselves [ ] 2 Try to be helpful [ ] 88 Do not know 19 In general, do you think that people in your village are interested in getting together to work with a common goal? [ ] 1 Very interested [ ] 2 Not very interested [ ] 3 Not interested [ ] 88 Do not know 20 Do you agree with the following statement: People in your community are interested in getting together to work for a common goal. [ ] 1 Strongly agree [ ] 2 Agree [ ] 3 Neutral (not agree, not disagree) [ ] 4 Disagree [ ] 5 Strongly disagree 200 21 Do you agree with the following statement: In general, most people in your village trust people in your village. [ ] 1 Strongly agree [ ] 2 Agree [ ] 3 Neutral (not agree, not disagree) [ ] 4 Disagree [ ] 5 Strongly disagree 22 Do you agree with the following statement: In general, people in your village only cares about themselves most of the time [ ] 1 Strongly agree [ ] 2 Agree [ ] 3 Neutral (not agree, not disagree) [ ] 4 Disagree [ ] 5 Strongly disagree Thank you for your help. 201 REFERENCES 202 REFERENCES Abadie, A. (2005). Semiparametric Difference -in-Differences Estimators. The Review of Economic Studies , 72(1), 1Œ19. doi:10.1111/0034 -6527.00321 Abadie, A., & Imbens, G. W. (2008). On the Failure of the Bootstrap for Matching Estimators. Econometrica , 76(6), 1537 Œ1557. doi:10.3982/ECTA6474 Alesina, A., & La Ferrara, E. (2002). Who trusts others? Journal of Public Economics , 85(2), 207Œ234. doi:10.1016/S0047 -2727(01)00084 -6 Ashraf, N., Giné, X., & Karlan, D. (2009). Finding missing markets (and a disturbing epilogue): evidence from an export crop adoption and marketing intervention in Kenya. American Journal of Agri cultural Economics , 91(4), 973 Œ990. Ashraf, Nava. (2009). Spousal Control and Intra -Household Decision Making: An Experimental Study in the Philippines. The American Economic Review , 99(4), 1245 Œ1277. Baliamoune -Lutz, M. (2011). Trust -based social capital, institutions, and development. The Journal of Socio -Economics , 40(4), 335 Œ346. doi:10.1016/j.socec.2010.12.004 Banerjee, A. V., Cole, S., Duflo, E., & Linden, L. (2007). Remedying Education: Evidence from Two Randomized Experiments in India. The Quarterly Journal of Economics , 122(3), 1235Œ1264. doi:10.1162/qjec.122.3.1235 Becerril, J., & Abdulai, A. (2010). The impact of improved maize varieties on poverty in Mexico: a propensity score -matching approach. World development , 38(7), 1024 Œ1035. Berg, J., Dick haut, J., & McCabe, K. (1995). Trust, Reciprocity, and Social History. Games and Economic Behavior , 10(1), 122 Œ142. doi:10.1006/game.1995.1027 Besley, T., & Case, A. (1993). Modeling Technology Adoption in Developing Countries. The American Economic Review , 83(2), 396 Œ402. doi:10.2307/2117697 Bonein, A., & Serra, D. (2009). Gender pairing bias in trustworthiness. The Journal of Socio -Economics , 38(5), 779 Œ789. doi:10.1016/j.socec.2009.03.003 Bouma, J., Bulte, E., & van Soest, D. (2008). Trust and cooperatio n: Social capital and community resource management. Journal of Environmental Economics and Management , 56(2), 155 Œ166. doi:10.1016/j.jeem.2008.03.004 203 Bouman, F. J. A. (1995). Rotating and accumulating savings and credit associations: A development perspec tive. World Development , 23(3), 371 Œ384. doi:10.1016/0305 -750X(94)00141 -K Bravo -Ureta, B. E., Almeida, A. N., Solís, D., & Inestroza, A. (2011). The economic impact of MARENA™s investments on sustainable agricultural systems in Honduras. Journal of Agricul tural Economics , 62(2), 429 Œ448. Buchan, N. R., Croson, R. T. A., & Solnick, S. (2008). Trust and gender: An examination of behavior and beliefs in the Investment Game. Journal of Economic Behavior & Organization , 68(3Œ4), 466 Œ476. doi:10.1016/j.jebo.2007. 10.006 Buddelmeyer, H., & Skoufias, E. (2004). An evaluation of the performance of regression discontinuity design on PROGRESA (Vol. 3386). World Bank Publications. Retrieved from http://books.google.com/books?hl=en&lr=&id=5tuCP0PMNvMC&oi=fnd&pg=PA32&d q=%2 2set+from+rural+Mexico+collected+for+the+purposes+of+evaluating+the+impact +of%22+%22poverty+alleviation+program+to+examine+the+performance+of+a%22+% 22the+Regression+Discontinuity+Design+(RDD).+Using+as+a+benchmark+the%22+& ots=_M7LHAYqMJ&sig=Cz8PG5zMCke -ko7lULFNgBhKYug Byerlee, D., & Hesse de Polanco, E. (1986). Farmers™ Stepwise Adoption of Technological Packages: Evidence from the Mexican Altiplano. American Journal of Agricultural Economics , 68(3), 519 Œ527. doi:10.2307/1241537 Caliendo, M., & Kopeinig, S. (2008). Some practical guidance for the implementation of propensity score matching. Journal of Economic S urveys , 22(1), 31 Œ72. Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008). Bootstrap -Based Improvements for Inference with Clustered Errors. Revie w of Economics and Statistics , 90(3), 414 Œ427. doi:10.1162/rest.90.3.414 Canavire -Bacarreza, G., & Hanauer, M. M. (2013). Estimating the Impacts of Bolivia™s Protected Areas on Poverty. World Development , 41(0), 265 Œ285. doi:10.1016/j.worlddev.2012.06.011 Capra, C. M., Lanier, K., & Meer, S. (2008). Attitudinal and Behavioral Measures of trust: a NewComparison. Lanier, Kelli and Meer, Shireen, Attitudinal and Behavioral Measures of Trust: A New Comparison . Retrieved from https://www.gate.cnrs.fr/IMG/pdf/Cap ra.pdf 204 Cardenas, J. C., & Carpenter, J. (2008). Behavioural Development Economics: Lessons from Field Labs in the Developing World. Journal of Development Studies , 44(3), 311 Œ338. doi:10.1080/00220380701848327 Cardenas, J. C., & Ramos, P. A. (2006). Manual de J uegos Económicos para el Análisis del Uso Colectivo de los Recursos N aturales. Lima, Peru: Centro Internacional de la Papa (CIP). Catholic Relief Services, Latin America and Caribbean Regional Office. (2009). Agriculture for basic Needs (A4N): A Propo sal for the Howard G. Buffett Foundation . Cavatassi, R., Salazar, L., González -Flores, M., & Winters, P. (2011). How do Agricultural Programmes Alter Crop Production? Evidence from Ecuador. Journal of Agricultural Economics , 62(2), 403 Œ428. Chaudhuri, A., & Gangadharan, L. (2007). An Experimental Analysis of Trust and Trustworthiness. Southern Economic Journal , 73(4), 959 Œ985. doi:10.2307/20111937 Chowa, G. A. N., & Elliott III, W. (2011). An asset approach to increasing perceived household economic stabili ty among families in Uganda. The Journal of Socio -Economics , 40(1), 81Œ87. doi:10.1016/j.socec.2010.02.008 Cuong, N. V. (2009). Impact evaluation of multiple overlapping programs under a conditional independence assumption. Research in Economics , 63(1), 27 Œ54. doi:10.1016/j.rie.2008.10.001 Danielson, A. J., & Holm, H. J. (2007). Do you trust your brethren?: Eliciting trust attitudes and trust behavior in a Tanzanian congregation. Journal of Economic Behavior & Organization , 62(2), 255 Œ271. doi:10.1016/j.jeb o.2004.10.011 De Janvry, A., & Sadoulet, E. (2002). World Poverty and the Role of Agricultural Technology: Direct and Indirect Effects. Journal of Development Studies , 38(4), 1 Œ26. doi:10.1080/00220380412331322401 Dearmon, J., & Grier, K. (2009). Trust and development. Journal of Economic Behavior & Organization , 71(2), 210 Œ220. Deaton, A. (1997). The Analysis of Household Surveys: A Microeconometric Approach to Development Policy . World Bank Publications. Deaton, A. (2009). Instruments of development: Rand omization in the tropics, and the search for the elusive keys to economic development . National Bureau of Economic Research. Retrieved from http://www.nber.org/papers/w14690 205 Dehejia, R. H., & Wahba, S. (2002). Propensity score -matching methods for nonexper imental causal studies. Review of Economics and statistics , 84(1), 151 Œ161. Del Carpio, X. V., Loayza, N., & Datar, G. (2011). Is Irrigation Rehabilitation Good for Poor Farmers? An Impact Evaluation of a Non -Experimental Irrigation Project in Peru. Journa l of Agricultural Economics , 62(2), 449 Œ473. doi:10.1111/j.1477 -9552.2011.00295.x Dillon, A. (2011). The Effect of Irrigation on Poverty Reduction, Asset Accumulation, and Informal Insurance: Evidence from Northern Mali. World Development , 39(12), 2165 Œ2175. Duflo, E., Kremer, M., & Robinson, J. (2008). How high are rates of return to fertilizer? Evidence from field experiments in Kenya. The American Economic R eview , 98(2), 482 Œ488. Duflo, E., Kremer, M., & Robinson, J. (2009). Nudging farmers to use fertilizer: theory and experimental evidence from Kenya . National Bureau of Economic Research. Retrieved from http://www.nber.org/papers/w15131 Duflo, Esther, Glennerster, R., & Kremer, M. (2007). Chapter 61 Using Randomization in Development Economics Res earch: A Toolkit. In T. Paul Schultz and John A. Strauss (Ed.), Handbook of Development Economics (Vol. Volume 4, pp. 3895 Œ3962). Elsevier. Retrieved from http://www.sciencedirect.com/science/article/pii/S1573447107040612 Etang, A., Fielding, D., & Knowles , S. (2011). Trust and ROSCA membership in rural Cameroon. Journal of International Development , 23(4), 461 Œ475. doi:10.1002/jid.1686 Etang, A., Fielding, D., & Knowles, S. (2012). Are Survey Measures of Trust Correlated with Experimental Trust? Evidence f rom Cameroon. Journal of Development Studies , 48(12), 1813Œ1827. doi:10.1080/00220388.2011.649263 Fafchamps, M. (2006). Development and social capital. Journal of Development Studies , 42(7), 1180Œ1198. doi:10.1080/00220380600884126 Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of Agricultural Innovations in Developing Countries: A Survey. Economic Development and Cultural Change , 33(2), 255Œ298. doi:10.2307/1153228 Fukuyama, F. (2001). Social capital, civil society and development. Third World Qu arterly , 22(1), 7 Œ20. doi:10.1080/713701144 206 Gächter, S., Herrmann, B., & Thöni, C. (2004). Trust, voluntary cooperation, and socio -economic background: survey and experimental evidence. Journal of Economic Behavior & Organization , 55(4), 505 Œ531. doi:10.10 16/j.jebo.2003.11.006 Garbarino, E., & Slonim, R. (2009). The robustness of trust and reciprocity across a heterogeneous U.S. population. Journal of Economic Behavior & Organization , 69(3), 226Œ240. doi:10.1016/j.jebo.2007.06.010 Gebremedhin, B., & Swinton , S. M. (2003). Investment in soil conservation in northern Ethiopia: the role of land tenure security and public programs. Agricultural Economics , 29(1), 69 Œ84. doi:10.1111/j.1574 -0862.2003.tb00148.x Gitonga, Z. M., De Groote, H., Kassie, M., & Tefera, T. (2013). Impact of metal silos on households™ maize storage, storage losses and food security: An application of a propensity score matching. Food Policy , 43, 44Œ55. doi:10.1016/j.foodpol.2013.08.005 Glaeser, E. L., Laibson, D. I., Scheinkman, J. A., & Soutter, C. L. (2000). Measuring Trust. The Quarterly Journal of Economics , 115(3), 811 Œ846. doi:10.1162/003355300554926 Grootaert, C., & Narayan, D. (2004). Local Institutions, Poverty and Household Welfare in Bolivia. World Development , 32(7), 1179 Œ1198. doi:10.1016/j.worlddev.2004.02.001 Heckman, J.J., Ichimura, H., Smith, J., & Todd, P. (1998). Characterizing Selection Bias Using Experimental Data. Econometrica , 66(5), 1017 Œ1098. doi:10.2307/2999630 Heckman, James J., Ichimura, H., & Todd, P. E. (1997) . Matching As An Econometric Evaluation Estimator: Evidence from Evaluating a Job Training Programme. The Review of Economic Studies , 64(4), 605 Œ654. doi:10.2307/2971733 Horváth, R. (n.d.). Does trust promote growth? Journal of Comparative Economics . doi:10.1016/j.jce.2012.10.006 Imbens, G. M., & Wooldridge, J. M. (2008). Recent developments in the econometrics of program evaluation . National Bureau of Economic Research. Retrieved from http://www.nber.org/papers/w14251 Instituto Nacional de Información de Desarrollo. (2008a). Esquipulas en cifras. Instituto Nacional de Informacion de Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm 207 Instituto Nacional de Información de Desarrollo. (2008b). Esteli en cifras. Instit uto Nacional de Informacion de Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm Instituto Nacional de Información de Desarrollo. (2008c). Jinotega en cifras. Instituto Nacional de Informacion de Desarrollo. Retrieve d from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm Instituto Nacional de Información de Desarrollo. (2008d). La Trinidad en cifras. Instituto Nacional de Informacion de Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/Cifras Mun/tablas_cifras.htm Instituto Nacional de Información de Desarrollo. (2008e). San Isidro en cifras. Instituto Nacional de Informacion de Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm Instituto Nacional de Infor mación de Desarrollo. (2008f). San Nicolas en cifras. Instituto Nacional de Informacion de Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm Instituto Nacional de Información de Desarrollo. (2008g). San Rafael del No rte en cifras. Instituto Nacional de Informacion de Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm Instituto Nacional de Información de Desarrollo. (2008h). Terrabona en cifras. Instituto Nacional de Informacion d e Desarrollo. Retrieved from http://www.inide.gob.ni/censos2005/CifrasMun/tablas_cifras.htm International Fund for Agricultural Development. (2010). Rural Poverty Report 2011. New realities, new challenges: new opportunities for tomorrow™s generation . Int ernational Fund for Agricultural Development (IFAD). Kaboski, J. P., & Townsend, R. M. (2005). Policies and Impact: An Analysis of Village -Level Microfinance Institutions. Journal of the European Economic Association , 3(1), 1 Œ50. doi:10.1162/15424760532953 31 Khandker, S. R., Koolwal, G. B., & Samad, H. A. (2010). Handbook on Impact Evaluation: Quantitative Methods and Practices . World Bank Publications. King, E. M., & Behrman, J. R. (2009). Timing and Duration of Exposure in Evaluations of Social Programs. The World Bank Research Observer , 24(1), 55 Œ82. doi:10.1093/wbro/lkn009 208 Knack, S., & Zak, P. J. (2003). Building Trust: Public Policy, Interpersonal Trust, and Economic Development. Supreme Court Economic Review , 10, 91Œ107. doi:10.2307/1147139 Kremer, M. (2003). Randomized evaluations of educational programs in developing countries: Some lessons. The American Economic Review , 93(2), 102 Œ106. Lai, F., Sadoulet, E., & de Janvry, A. (2011). The Contributions of School Quality and Teacher Qualifications to Stu dent Performance Evidence from a Natural Experiment in Beijing Middle Schools. Journal of Human Resources , 46(1), 123 Œ153. Lechner, M. (2001). Identification and estimation of causal effects of multiple treatments under the conditional independence assumpt ion. Econometric evaluation of labour market policies , 43Œ58. Leuven, E., & Sianesi, B. (2012). PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance test. Lyon, F. (2000). Trust, Networks and Norms: The Creation of Social Capital in Agricultural Economies in Ghana. World Development , 28(4), 663 Œ681. doi:10.1016/S0305 -750X(99)00146 -1 Mendola, M. (2007). Agricultural technology adoption and poverty reduction: A propensity -score matc hing analysis for rural Bangladesh. Food Policy , 32(3), 372 Œ393. doi:10.1016/j.foodpol.2006.07.003 Minten, B., & Barrett, C. B. (2008). Agricultural Technology, Productivity, and Poverty in Madagascar. World Development , 36(5), 797 Œ822. doi:10.1016/j.world dev.2007.05.004 Nitlapan. (2001). Tipología Nacional de Productores y Zoni cación Socio -económica. Managua: Nitlapan. Nkonya, E., Phillip, D., Mogues, T., Pender, J., & Kato, E. (2012). Impacts of Community -driven Development Programs on Income and Asset A cquisition in Africa: The Case of Nigeria. World Development , 40(9), 1824 Œ1838. doi:10.1016/j.worlddev.2012.04.028 Nowak, P. (1992). Why farmers adopt production technology Overcoming impediments to adoption of crop residue management techniques will be crucial to implementation of conservation compliance plans. Journal of Soil and Water Conservation , 47(1), 14 Œ16. Owen, A. L., & Videras, J. (2008). Trust, cooperation, and implementation of sustainability programs: The case of Local Agenda 21. Ecological Economics , 68(1Œ2), 259 Œ272. doi:10.1016/j.ecolecon.2008.03.006 209 Özcan, B., & Bjørnskov, C. (2011). Social trust and h uman development. The Journal of Socio -Economics , 40(6), 753 Œ762. doi:10.1016/j.socec.2011.08.007 Peralta, M. A., Swinton, S. M., & Maredia, M. (2011). Accounting for selection bias from data collection: an example from the Agriculture for Basic Needs proj ect in Nicaragua. Presented at the Mind the Gap: From Evidence to Impact, Cuernavaca, Mexico. Putman, R. D. (2003). The Prosperous Community: Social Capital and Public Life. The American Prospect No. 13 . Retrieved January 24, 2013, from http://xroads.virgi nia.edu/~HYPER/DETOC/assoc/13putn.html Ravallion, M., & Chen, S. (2005). Hidden impact? Household saving in response to a poor -area development project. Journal of Public E conomics , 89(11), 2183 Œ2204. Ravallion, Martin. (2008). Evaluation in the Practice o f Development (SSRN Scholarly Paper No. ID 1103727). Rochester, NY: Social Science Research Network. Retrieved from http://papers.ssrn.com/abstract=1103727 Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies . Journal of Educational Psychology , 66(5), 688 Œ701. doi:10.1037/h0037350 Schechter, L. (2007). Traditional trust measurement and the risk confound: An experiment in rural Paraguay. Journal of Economic Behavior & Organization , 62(2), 272 Œ292. doi:10.1016/j .jebo.2005.03.006 Schwieren, C., & Sutter, M. (2008). Trust in cooperation or ability? An experimental study on gender differences. Economics Letters , 99(3), 494 Œ497. doi:10.1016/j.econlet.2007.09.033 Sherm an, R. P. (2000). Tests of Certain Types of Ignorable Nonresponse in Surveys Subject to Item Nonresponse or Attrition. American Journal of Political Science , 44(2), 362 Œ374. doi:10.2307/2669317 Smith, J., & Todd, P. (2005). Does matching overcome LaLon de™s critique of nonexperimental estimators? Journal of E conometrics , 125(1), 305 Œ353. Solis, D., Bravo -Ureta, B. E., & Quiroga, R. E. (2008). Technical efficiency among peasant farmers participating in natural resource management programmes in Central Ame rica. Journal of Agricultural Economics , 60(1), 202 Œ219. 210 Sunding, D., & Zilberman, D. (2001). Chapter 4 The agricultural innovation process: Research and technology adoption in a changing agricultural sector. In Bruce L. Gardner and Gordon C. Rausser (Ed.), Handbook of Agricultural Economics (Vol. Volume 1, Part A, pp. 207Œ261). Elsevier. Retrieved from http://www.sciencedirect.com/science/article/pii/S1574007201100071 Teklewold, H., Kassie, M., & Shiferaw, B. (2013). Adoption of Multiple Sustainable Agricultural Practices in Rural Ethiopia. Journal of Agricultural Economics , 64(3), 597 Œ623. doi:10.1111/1477 -9552.12011 Thorp, R., Stewart, F., & Heyer, A. (2005). When and how far is group formation a route out of chronic poverty? World Development , 33(6), 907 Œ920. doi:10.1016/j.worlddev.2004.09.016 Tjernström, E., Toledo, P., & Carter, M. R. (2013). Identifying the Impact Dynamics of a Small -Farmer Development Scheme in Nicaragua. American Journal of Agricultural Economics , 95(5), 1359 Œ1365. doi:10.1093/ajae/aat042 Vollan, B. (2011). The difference between kinship and friendship: (Field -) experimental evidence on trust and punishment. The Journal of Socio -Economics , 40(1), 14 Œ25. doi:10.1016/j.socec.2010.10.003 Wooldridge, J. (2010). Econometric Analysis of Cross Section and Panel Data . Cambridge, Massachusetts: The MIT Press. Wor ld Bank. (2008). Poverty Assessment Nicaragua (No. 39736 -NI) (p. 215). Retrieved from http://web.worldbank.org/WBSITE/EXTERNAL/TOPICS/EXTPOVERTY/EXTPA/0,,co ntentMDK:20207612~menuPK:435735~pagePK:148956~piPK:216618~theSitePK:4303 67~isCURL:Y~isCURL:Y,00.html