CAF - Improving food security through community participation results from a randomized field experiment

C A F - W O R KI N G PA P E R # 2 0 2 6 / 0 1

F i r s t v e r s i o n : D e c e m b e r 1 8 , 2 0 2 5 ( c u r r e n t ) Improving food security through community participation: results from a randomized field experiment in rural Nicaragua

Pablo A. Celhay1 1Escuela de Gobierno and Instituto de Economía, Pontificia Universidad Católica de Chile. pacelhay@uc.cl This paper evaluates a community based development program designed to promote climate-smart agriculture and improve food security in rural Nicaragua. Using a within-community randomized controlled trial, we estimate shortand medium-term impacts on agricultural practices, production, and welfare. The program combined productive asset transfers, technical assistance, and training delivered through local solidarity groups. Results show significant increases in the adoption of improved inputs—such as certified seeds, biofertilizers, and post-harvest technologies—along with higher maize and bean yields, greater crop diversification, and expanded participation in producer organizations. Beneficiaries also report better food security and higher satisfaction with their quality of life. Because randomization occurred within communities, spillovers likely make these estimates conservative. The findings suggest that community-based delivery can effectively scale up CSA practices and strengthen food security in vulnerable rural areas.

K E Y W O R D S

Community Development Programs; Food Security; Agricultural Production; Impact Evaluation. Small sections of text that are less than two paragraphs may be quoted without explicit permission as long as this document is acknowledged. Findings, interpretations and conclusions expressed in this publication are

the sole responsibility of its author(s) and cannot be, in any way, attributed to CAF, its Executive Directors or the countries they represent. CAF does not guarantee the accuracy of the data included in this publication and is not, in any way, responsible for any consequences resulting from its use. ©2026 Corporación Andina de FomentoC A F - D O C U M E NT O D E T R AB A J O # 2 0 2 6 / 0 1

E s t a v e r s i ó n : 1 8 d e d i c i e m b r e d e 2 0 2 5

Mejorando la seguridad alimentaria mediante la participación comunitaria: resultados de un experimento de campo aleatorizado en zonas rurales de Nicaragua Pablo A. Celhay1 1Escuela de Gobierno and Instituto de Economía, Pontificia Universidad Católica de Chile. pacelhay@uc.cl Este estudio evalúa un programa de desarrollo basado en la comunidad orientado a promover la agricultura climáticamente inteligente y mejorar la seguridad alimentaria en zonas rurales de Nicaragua. A partir de un experimento aleatorio dentro de comunidades, estimamos los efectos de corto y mediano plazo sobre las prácticas agrícolas, la producción y el bienestar. El programa combinó transferencias en especie, asistencia técnica y capacitación impartidas a través de grupos solidarios locales. Los resultados muestran aumentos significativos en la adopción de insumos mejorados —como semillas certificadas, biofertilizantes y tecnologías de poscosecha—, mayores rendimientos de maíz y frijol, mayor diversificación de cultivos y una participación

ampliada en organizaciones productivas. Los beneficiarios también reportan mejoras en seguridad alimentaria y calidad de vida. Los hallazgos sugieren que los modelos comunitarios pueden escalar eficazmente la CSA y fortalecer la resiliencia rural. K E Y W O R D S Cooperativas, desarrollo local, crecimiento inclusivo, políticas basadas en el territorio. Pequeñas secciones del texto, menores a dos párrafos, pueden ser citadas sin autorización explícita siempre que se cite el presente documento. Los resultados, interpretaciones y conclusiones expresados en esta publicación son de exclusiva responsabilidad de su(s) autor(es), y de ninguna manera pueden ser atribuidos a CAF, a los miembros de su Directorio Ejecutivo o a los países que ellos representan. CAF no garantiza la exactitud de los datos incluidos en esta publicación y no se hace responsable en ningún aspecto de las consecuencias que resulten de su utilización. ©2026 Corporación Andina de FomentoCELHAY 2 1 | INTRODUCTION Community-based development (CBD) emphasize the role of local participation in the design and implementation of development programs.1 These approaches seek to leverage local knowledge, increase community ownership, and improve the targeting and sustainability of development interventions (Mansuri and Rao, 2004, 2012). While the promise of CBD has spurred widespread adoption—especially in rural, fragile, or low-capacity settings—their effectiveness remains a subject of ongoing debate. Community-based development (CBD) programs have shown promise in improving agricultural livelihoods by using local knowledge and better targeting poor and marginalized groups (Nkonya et al., 2012; Park and Wang, 2010). However, evidence across countries remains mixed. While some interventions improve participation and service delivery (Beath

et al., 2017; Madajewicz et al., 2021), others find limited or no effects on income or governance outcomes (Humphreys et al., 2019; Saguin, 2018). Overall, the literature suggests that CBD effectiveness depends on strong institutional design, continuous support services, and safeguards against local elite capture.2 This paper studies the shortand medium-term effects of a community-based development (CBD) program that promoted climate-smart agricultural (CSA) practices in rural Nicaragua. The analysis relies on a randomized controlled trial implemented in 64 rural communities, where productive assets and CSA training were provided to randomly selected small farmers within each village. The program aimed to increase the adoption of sustainable farming practices, raise agricultural yields, and improve household food security through a participatory delivery model. Raising crop yields is essential for sustainable growth in low-income countries, where many households depend on agriculture for their livelihoods (Antón et al., 2013). Climatesmart agriculture (CSA) has emerged as a key strategy to address the dual challenges of climate adaptation and food insecurity. Defined by the FAO as an approach to make agri-food systems more climate-resilient,3 CSA includes practices such as conservation agriculture, improved seeds, and agroforestry (Lipper et al., 2014; Jayne et al., 2018). Although CSA has been incorporated into many national policy agendas (Campbell et al., 2016), adoption among smallholders remains low because of high upfront costs, uncertain returns, and short-term risks to yield and consumption (Amadu et al., 2020; Arslan et al., 2015; Knowler and Bradshaw, 2007). Stronger evidence is needed on delivery models that can

effectively promote CSA adoption and measure its impacts on productivity, food surplus, and household food security (Steenwerth et al., 2014; Battisti and Naylor, 2009). We use data from the Proyecto Apoyo para el Incremento de la Productividad, Seguridad Alimentaria y Nutricional en la Costa Caribe Nicaraguense (PAIPSAN), a program designed to reduce food insecurity among rural communities in this region. To achieve this goal, the program encouraged beneficiaries to form cooperative groups known as solidarity groups. These groups served two main purposes: first, to identify the needs of each participant and prepare an Innovative Development Plan (Plan de Desarrollo Innovador, 1Community-based development (CBD) – often termed community-driven development (CDD) – refers to aid strategies that delegate decision-making and resource allocation to local communities. See Bank (2025). 2Studies that explore the effects of CBD projects include, for instance, Alatas et al. (2012, 2019); Araujo et al. (2008); Avdeenko and Gilligan (2015); Beath et al. (2017); Bernard et al. (2008); Borisova et al. (2024); Casey et al. (2012); Casey (2018); Cooke and Kothari (2001); Dasgupta and Beard (2007); Heß et al. (2021); Humphreys et al. (2019); Kahsay and Bulte (2021); Krishna et al. (2025); Labonne and Chase (2011); Lund and Saito-Jensen (2013); Madajewicz et al. (2021); McNamara et al. (2020); Nguyen and Rieger (2017); Nkonya et al. (2012);

Olken (2010); Park and Wang (2010); Platteau (2004); Pradhan et al. (2014); Saguin (2018); Salinger et al. (2024); Stiglitz (2002); White et al. (2018); Wong and Guggenheim (2018). 3See https://www.fao.org/climate-smart-agriculture/en/.CELHAY 3 PDI) that guided the intervention; and second, to facilitate implementation by organizing the purchase and distribution of inputs, as well as workshops, demonstrations, and consultation meetings. This group-based structure represents the community component of the PAIPSAN program, reflecting its community-driven development (CDD) approach, where planning and implementation decisions were made collectively with each community rather than at a centralized level. The intervention provided in-kind agricultural inputs—such as seeds, tools, and fertilizers—alongside technical assistance and training. Our analysis combines administrative and survey data to evaluate the program’s effects on smallholder farmers’ livelihoods. In particular, we study changes in training participation, livestock ownership, and involvement in productive organizations, and we assess how these interventions affected food security, perceived quality of life, and overall satisfaction among participants. Households within each of the participating communities were randomly assigned to two groups: 908 participants in the treatment group, who began the project in June 2017, and 903 participants in the control group, who started in November 2018. Because the assignment was random, both groups were similar in their baseline characteristics, ensuring that any later differences can be attributed to the PAIPSAN-CCN intervention rather than pre-existing factors. Data were collected from both groups between March 2017 and December 2018 through three survey rounds: a baseline survey (March–May 2017), a midline follow-up (March 2018),

and a final survey (November 2018). The surveys covered 1,811 households—representing 8,583 individuals—across the 64 communities in the project area. Our results show that the PAIPSAN-CCN intervention had a positive impact on the adoption of improved agricultural inputs among beneficiary households. Survey data indicate significant increases in the use of certified and fortified seeds, bio-fertilizers, pest control measures, and post-harvest practices. Treated farmers reported planting about 20 percent more basic grains than the control group, reflecting broader progress in the adoption of techniques promoted by the program. The program also strengthened nutrition-sensitive practices, particularly in hygiene and food preparation. Beneficiaries in the treatment group reported performing more food preparation and preservation activities, using safer handling practices, and adopting natural ingredients and water filters more often. These behavioral changes were accompanied by higher rates of staple crop production: maize cultivation rose from 55 to 65 percent and bean cultivation from 68 to 84 percent among treated households. In terms of production, treated farmers achieved higher yields across key crops. Maize yields increased from 9.7 to 15 quintals per manzana, while bean yields rose from 12 to 16.7 quintals over the same period. Similar gains were observed for other products, and the share of farmers producing fruit or perennial crops grew from 1 to 9 percent by late 2018. These results translated into higher satisfaction with harvests, quality of life, and household food security, as measured by the Food Insecurity Experience Scale (FIES). However, given the relatively short exposure period, it is too early to assess lasting effects on health or nutritional outcomes. Our results contribute to the literature on community-based development (CBD) in two distinct ways. First, a central concern in this literature is that elite capture and local political

dynamics can bias program impacts, making it hard to separate program design from local power relations (Platteau, 2004; Alatas et al., 2019). The within-community randomized design of our study helps address this issue by isolating treatment assignment from local political influence. This approach minimizes the scope for capture and provides a clean estimate of CBD effects on agricultural outcomes such as yields, technology adoption, and food security (Beath et al., 2017; Madajewicz et al., 2021; Humphreys et al., 2019).CELHAY 4 Second, unlike previous studies that focus on improving participation in decisionmaking, committee transparency, or trust in local authorities, our analysis evaluates a CBD intervention centered on economic outcomes—specifically agricultural production and participants’ welfare (Beath et al., 2017; Humphreys et al., 2019). In doing so, it contributes to the debate on whether community control can enhance service delivery and livelihoods when elite dominance is limited (Madajewicz et al., 2021). The result is a rigorous and internally valid estimate of CBD effectiveness in rural production settings, separate from the confounding effects of local political selection. In addition, we provide rare experimental evidence from Latin America on how CSA can be integrated into a CBD model. The intervention aimed to reduce coordination, information, and risk barriers that often limit smallholders from adopting productive technologies. While most rigorous studies on agricultural community programs come from Africa and Asia—and often find short-lived or mixed results without complementary support (Nkonya et al., 2012; Park and Wang, 2010)—our study measures multi-season effects on crop yields, food insecurity, and adoption of improved inputs within villages. By combining in-kind assets, technical assistance, and collective implementation, the program tests key CSA

mechanisms such as local learning, risk-sharing, and coordination at the community level (Lipper et al., 2014; Campbell et al., 2016). In this way, the paper helps explain why some CBD projects improve participation but not welfare (Beath et al., 2017; Humphreys et al., 2019; Saguin, 2018), identifying design features—household randomization, community platforms, and bundled support—that can make participatory delivery more effective in improving rural livelihoods. 2 | CONTEXT We study the effects of climate-smart agriculture promoted by theProyecto de Apoyo para el Incremento de la Productividad, Seguridad Alimentaria y Nutricional en la Costa Caribe Nicaragüense (PAIPSAN), coordinated by theMinisterio de Economía Familiar, Comunitaria, Cooperativa y Asociativa (MEFCCA). The main goal of the program was to improve food and nutritional security in rural communities on Nicaragua’s Caribbean Coast, which includes two autonomous regions—the Northern (Región Autónoma de la Costa Caribe Norte, RACCN) and the Southern (Región Autónoma de la Costa Caribe Sur, RACCS). The project operated in 246 communities across 15 municipalities: eight in the north ( Waslala, Bonanza, Siuna, Rosita, Mulukukú, Waspam, Prinzapolka, and Puerto Cabezas) and seven in the south (La Cruz de Río Grande, Paiwas, El Tortuguero, Desembocadura de Río Grande, Laguna de Perlas, Kukra Hill, and Bluefields). The coordination and implementation of PAIPSAN-CCN are the responsibility of the Ministry of Family, Community, Cooperative, and Associative Economy (MEFCCA) and

are carried out with financial support from a grant provided by the Global Agriculture and Food Security Program (GAFSP). This area covers around 43% of the national territory (see Figure A.1) and was prioritized due to its high rates of poverty, malnutrition, and indigenous population. In 2012, 23.3% of children under five in the Northern region were stunted—6 percentage points above the national average—and, in 2014, the poverty rate was 9.4 percentage points higher than the national average of 29.6%. Approximately 35% of the population identified as indigenous (INIDE, 2016), mostly from the Miskito ethnic group, which represents over half of Nicaragua’s indigenous population. Beneficiaries were selected through community meetings with local authorities, based on poverty status and primary occupation. A key feature of the program was its participatoryCELHAY 5 structure: households organized themselves into solidarity groups of 30–50 members. These groups identified local needs, preparedPlanes de Desarrollo Innovador(PDIs), and participated in the planning and implementation of the activities financed by the program. Each PDI combined three main components:

1. Financial support, provided in kind—such as seeds, tools, fertilizers, machinery, or livestock.

2. Technical assistance, delivered by ministry field officers who demonstrated farming techniques, supported orchard setup, and followed up on each group’s progress.

3. Training, offered through workshops on topics such as crop diversification, fertilizer use, irrigation, storage, marketing, livestock care, food hygiene, and healthy diets. Training also included sessions on cooperative management and associative organization.

Solidarity groups facilitated coordination with the ministry, helped distribute inputs, and organized local activities. They functioned similarly to cooperatives—sharing responsibilities, solving problems collectively, and promoting the program within their communities. Participation brought clear benefits through access to inputs and knowledge but also required time, organization, and sustained community engagement. In the survey, the term “community organizations” refers broadly to local economic and community groups linked to the implementation of the PDIs, such as producer associations, production committees, or work groups created under the project to coordinate purchases, sales, or training activities. These organizations were either created or strengthened as part of the intervention. Solidarity Groups are one specific organizational form used by the project to structure beneficiary participation within a PDI. Thus, the communityorganization outcomes in Section 5.4 capture a wider set of associative arrangements than Solidarity Groups alone and reflect the broader institutional environment fostered by the program. 2.1 | Formation and Role of the Planes de Desarrollo Innovador(PDIs) Within PAIPSAN–CCN, the Planes de Desarrollo Innovador (PDIs) are the core instrument that structures how resources are allocated and how the intervention is implemented on the ground. PDIs are not spontaneous proposals. They are formulated through a structured process facilitated by the Ministry of Family, Community, Cooperative and Associative Economy (MEFCCA), in coordination with institutions in the National System for Production, Consumption and Commerce (SNPCC). The process begins with a joint diagnostic of local productive potential, food insecurity, and market opportunities conducted by technical facilitators together with beneficiary households. Based on this assessment, each PDI defines a package of investments and services consistent with the project’s objective of improving food security and nutrition in the Caribbean Coast. PDIs are organized into four broad types: family agriculture, agroindustry, artisanal fishing, and rural non-agricultural enterprises, reflecting the main livelihood strategies in the region. Within family agriculture and fishing, PDIs are further differentiated by orientation (subsistence/autoconsumption versus surplus/commercial production). This typology responds primarily to productive criteria (type of activity and market orientation),

but PDIs are also shaped by territorial conditions such as access to water, infrastructure, and markets. Beneficiaries are not required to meet minimum technical or managerial criteria to design a PDI. Instead, the project embeds capacity building in the planning cycle: MEFCCA facilitators support the formulation, costing, and sequencing of activities and ensure that each PDI respects predefined budget ceilings, eligible expenditure categories, and technicalCELHAY 6 norms. Functionally, PDIs operate as the main coordination device between beneficiary households and implementing agencies. Each PDI specifies (i) the mix of productive assets to be financed (for example, seeds, tools, small equipment), (ii) the schedule and content of trainings and technical assistance (including climate-smart agriculture, post-harvest management, and nutrition-sensitive practices), and (iii) the services to be provided by specialized public entities. In practice, INTA, MINSA, IPSA, MARENA, SERENA, and INPESCA4 are responsible for different components of the intervention: INTA for agricultural technologies and field schools; MINSA for nutrition education; IPSA for phytosanitary surveillance; MARENA and SERENA for environmental safeguards; and INPESCA for fishing and aquaculture, all coordinated by MEFCCA’s project unit. In this way, PDIs concentrate local autonomy in setting priorities and choosing the combination of activities, while execution and compliance are anchored in a clear institutional framework. From the perspective of the CBD literature, this institutional design matters for two reasons. First, local participation is exercised through a structured planning instrument (the PDI) supported by continuous technical assistance, rather than through ad hoc community demands. Second, standardized eligibility rules, investment ceilings, and multi-agency review introduce procedural safeguards that limit arbitrary allocation or capture, while

still allowing plans to be tailored to local needs. Our impact estimates should therefore be interpreted as the effects of a community-based planning and delivery model in which PDIs embody the main institutional innovation. 3 | DATA AND STUDY SAMPLE For the impact evaluation, a random sample of beneficiaries was drawn from the population. The sample size required to detect an impact was estimated to be 1,900 individuals, drawn from a total population of 14,000 people. Randomization was conducted at two levels: community and individual. Out of a total of 256 villages, 64 were selected to participate in the impact evaluation. Subsequent to this selection, program officials held assemblies in each village to assign individuals to treatment and control groups through a public lottery. Treatment assignment was phased in, meaning that both groups were provided the opportunity to participate in the program. The program commenced for the treatment group in June 2017, while it began for the control group in June 2018. As illustrated in Figure A.2, baseline data were collected during April and May of 2017. Although the estimated sample size was 1,900, due to challenges in the implementation of the baseline survey, the total number of households interviewed was 1,810, with 905 belonging to the treatment group and 905 to the control group. Two follow-up surveys were subsequently conducted: the first approximately one year after the baseline, and the second between December 2018 and March 2019. The attrition rate in these follow-ups was minimal; of the 1,810 families included in the baseline survey, 1,804 were interviewed in the first follow-up, and 1,802 in the second. Descriptive characteristics.– For the data analysis, the sample was restricted to non-missing values of the variables across the three surveys, resulting in a total of 1,672 observations from an initial 1,810. Table B.1 presents the mean and standard deviation of the demographic

characteristics of the individuals and the outcome variables at baseline. A significant proportion of recipients are illiterate, nearly 30%, with a median age of approximately 39 4INTA: Instituto Nicaragüense de Tecnología Agropecuaria; MINSA: Ministerio de Salud; IPSA: Instituto de Protección y Sanidad Agropecuaria; MARENA: Ministerio del Ambiente y los Recursos Naturales; SERENA: Secretaría de Recursos Naturales de los gobiernos regionales; INPESCA: Instituto Nicaragüense de la Pesca y Acuicultura.CELHAY 7 years. Additionally, around 25% of respondents identify as belonging to an indigenous group. The gender distribution in the sample is nearly balanced, with women slightly outnumbering men, comprising 51.26% of the observations. Certain dwelling characteristics highlight the poverty conditions of the beneficiaries: 50% of individuals live in homes with dirt floors, and only 53.67% have electricity. Furthermore, the majority of houses (90%) are constructed with wooden walls. The primary occupations of program recipients are agriculture and cattle raising; 95.6% report agriculture as their primary activity, and 79.55% engage in cattle raising, while only a small percentage (0.6%) is involved in forestry. Additionally, production decisions vary by season, with 71.83% of individuals participating in the sowing of any crop during the apante season, which is the most important growing season for this region, and 37.32% during the postrera season. As in any RCT, a fundamental condition for estimating the impact of the PAIPSAN project is that the beneficiaries in the treatment group are statistically identical to those in the control group. This implies that, on average, the characteristics of the control group and the treatment group are equal. Table B.1 shows the results for a series of demographic

indicators and project outcomes. These show the simple comparison of means between treatment protagonists and control group protagonists. The p-values (p-value) are calculated after adjusting the differences for sample stratification by estimating equation (1). In summary, the results show that the random assignment worked appropriately as there are no significant differences between the treatment group and the control group. Table B.2 examines whether attrition is systematically related to treatment status or baseline household characteristics. The dependent variable is an indicator for remaining in the longitudinal sample, and all specifications include community fixed effects. The estimates show that the treatment indicator is small and statistically insignificant in all columns, indicating that treated and control households are equally likely to be observed in all survey waves. Among the baseline covariates, only the age of the household head is weakly associated with panel retention, and the magnitude of this effect is negligible. Taken together with the very low overall attrition rates, these results suggest that differential attrition is unlikely to bias our impact estimates in a meaningful way. Finally, Not all outcomes were collected in every survey wave. Some variables were included from the outset, while others were added later in response to operational constraints and evolving priorities of the implementing agency. For example, the indicators reported in Table B.3 were measured only in the two follow-up surveys, once the project activities were underway. Similarly, food insecurity (FIES) was collected only in the first and second follow-up rounds, after additional funding from GAFSP placed greater emphasis on food security outcomes. We make this timing explicit in the tables and focus our interpretation on the periods for which data are available. 4 | EMPIRICAL FRAMEWORK The experimental design of this study uses a stratified approach by communities where farms within communities are assigned to a treatment and control group. . The estimation of the program’s impact is represented by the following equation:

Yij = a + dDij + fj + eij In this equation, Dij equals one if individual i from community j belongs to the treatment group and zero if the individual belongs to the control group. Moreover, given theCELHAY 8 characteristics of the sampling design—specifically, the selection of participants within the communities—it is essential to adjust the regression for community-level fixed effects, denoted as fj, to account for the effects of sample stratification. A typical analysis in experimental impact evaluation, particularly in its simplest form, involves comparing means between a treatment group and a control group that have been randomly assigned. Due to the randomness, it is expected that any observed differences after the program’s implementation will result from the participation of one group in the program while the other does not. However, complications arise when the control group also benefits from the program in some way, such as through the exchange of information with the treatment group. In such cases, a simple comparison may not yield an accurate assessment of the program’s impact. Under the presence of spillovers we interpret our results as a lower bound effects of the program. Future work will analyze spillovers and control for these. The treatment indicator captures random assignment to start the PDI early (treatment group) versus with a delay (control group). All households in our sample are enrolled in subsistence family agriculture PDIs, so there is no full compliance in take-up of the program. Measures of intensity such as the number of trainings attended combine both supply-side and demand-side factors and do not represent a single causal channel of the intervention. For these reasons, we report the differences between being randomized into the program or out of the program, which can be interpreted as the average treatment effect of being enrolled in the PDI, which we view as the most transparent and policy-relevant estimand in this context. 5 | RESULTS

5.1 | Implementation and take-up Before comparing treatment and control groups, it is important to describe how the program was implemented in practice. This section shows to what extent the activities and transfers planned by the project actually reached the beneficiaries. Understanding the scale of inputs, train