Research

INTRODUCTION

Which practices farmers choose to use has large implications for the environment and food security at both local and global scales (Cui et al. 2018, Harvey et al. 2018, Reincke et al. 2018, Wekesa et al. 2018, Bezner Kerr et al. 2019). In smallholder farming, change of practice is usually not mandated and enforced top-down because smallholders are typically not anyone’s direct employees and may operate relatively informally on their plots (Balehegn et al. 2020). Nevertheless, agricultural and development agencies have been trying to influence what smallholders do and, with different degrees of success, nudging them into adopting various promoted practices (Phuong et al. 2018, Isaac et al. 2021). Orchestrating change among smallholder farmers is complicated by their large numbers and low accessibility in remote regions, so it is normally not possible for external agencies to work with every smallholder farmer in their jurisdiction directly (Matous 2023). The outcomes of most programs depend on success in two steps: (1) some members of the targeted communities adopt formal recommendations; and (2) these direct impacts are further scaled up to other community members via social influence channeled through interpersonal networks (Nakano et al. 2018).

Although research on which personal network structures correlate with individuals’ adoption of recommended practices has a long tradition, the role of network structures of entire communities in this process is not clear (Bandiera and Rasul 2006, Varshney et al. 2022). This is a problem. Although we may expect from available research that, for example, network brokers may adopt recommended practices first (Zhang et al. 2020), this does not mean that communities with abundance of brokerage should be collectively more effective in managing required transitions. Findings about individuals’ networks do not directly aggregate to collectives (Lazega and Snijders 2016a). However, the progress in empirically addressing this discrepancy has been hampered by the large demands on data required for comparisons of entire networks. This is particularly difficult to address in resource-constrained settings where most of the world’s smallholders operate. Empirical research of individuals’ links and practices can be conducted in a single network, which can be one village. To quantitatively examine the role of collective structures requires a large number of entire networks, and ideally all individuals within them, which is significantly more challenging (Bodin et al. 2019).

We focus here on the collective level and examine the adoption of recommended cocoa farming practices promoted by a program of an international development organization in communities of smallholder farmers in Sulawesi, Indonesia. The program aimed to reduce greenhouse gas emissions from cocoa farming, while sustaining productivity. We analyze (1) which community network structures are associated with adoption of a broad range of practices recommended by this sustainability program and (2) which community network structures are associated with farmers’ adoption of similar practices as their peers within the same community?

THEORETICAL AND CONTEXTUAL BACKGROUND

Network structures and environmental practices

The role of social networks in managing environmental challenges has been theoretically tackled by Bodin and Crona (2009). Since their seminal theorization, many of the expected relationships still await empirical confirmation (Bodin et al. 2019). We overview some of the key highlighted network characteristics and point to recent studies.

Abundant communication relations within communities have been associated with pro-environmental behavior and participation in climate change adaptation (Saptutyningsih et al. 2020, Hua et al. 2021). Network ties can facilitate farmers’ access to resources (Abdul-Rahaman and Abdulai 2020) and information-sharing about novel approaches (Isaac et al. 2021). Some authors found cohesive bonding conducive to high adoption rates of conservation practices in agriculture (Datoon et al. 2023). However, dense networks can also homogenize knowledge in a community, which can reduce its adaptive capacity (Bodin and Norberg 2005, Little and McDonald 2007). Dense internal links may even crowd out cross-level links with the outside world (Di Gregorio et al. 2019).

Distribution of links may matter more than their number. In some networks, links can be concentrated in separate components, cliques, or modules with limited bridges in between, while other networks are more broadly and cohesively interconnected. Network cohesion at its best may help communities build endogenous capacity for sustainable environmental management (Musavengane and Simatele 2016) and accelerate the diffusion of novel practices (Varshney et al. 2022). However, broadly interconnected and internally cohesive networks are not superior in all socio-environmental contexts. A certain degree of separation between subgroups in a network into distinct cliques or modules may enhance complementary developments of diverse aspects of knowledge (Crona 2006).

Other network characteristics of interest relate to the presence and relative positions of highly connected actors (Bodin and Crona 2009, Matous and Bodin 2024). In the Indonesian rural context, the presence of prominent “patrons” can be both a source of social capital for the entire village as well as a sign of possible oppression (Ferse et al. 2012, Nurdin and Grydehøj 2014). Diverse studies have suggested that networks connected through central actors may facilitate coordination (Angst et al. 2018) and collective action (Ernstson et al. 2008). Such networks may also diffuse information or simple innovations that do not require deep deliberation but may be less effective for complex issues (Leavitt 1951, Mbaru and Barnes 2017, Ramirez et al. 2018) and may constrain collaboration (Ernstson et al. 2008). Networks with a single core have been observed in smallholder communities in various parts of the world (Isaac et al. 2007) and innovations have been found in simulations (Abrahamson and Rosenkopf 1997) and some empirical studies (Albizua et al. 2020; Rockenbauch et al. 2019) to “flow” efficiently from the core to periphery in such structures. On the other hand, polycentric networks (i.e., networks with highly connected actors distributed in different parts of the network) and networks fragmented into distinct components ostensibly allow network members more freedom to experiment in adapting practices to their local needs (Pamuk et al. 2014). In polycentric structures, farmers have been found to proactively participate in agri-environmental conservation without “paternalistic” oversight (Marshall 2009, Zulkafli et al. 2017).

“It takes a village”: on the importance of understanding collective-level structures and processes

The term “ecological fallacy” captures the fact that statistical findings about individuals may not apply to collectives (Robinson 1950). Unfortunately, findings about individuals’ networks and practices are not often tested whether they apply beyond individual level before making policy recommendations (Lubell et al. 2023). Because of practical constrains, network studies of remote communities tend to include prohibitively small numbers of village networks for a collective-level statistical analysis. Consequently, collective issues are examined through an individualistic lens. It is paradoxical that individuals still prevail as the dominant analytical unit in network inquiry that emerged in response to atomistic approaches to social systems (Freeman 2004, Snijders 2016). Collectives shape individuals and individuals shape collectives (Breiger 1974). Community network structures reflect these processes, which may help or hinder individuals’ management of local problems.

Theorizations of what constitutes effective social capital diverge between individual (Burt 1995) and collective level (Putnam 1995). We have evidence on which relational structures are effective for individuals (Lin 2001) but we know less about effective relational structures for collectives within different contexts (Lazega and Snijders 2016b). According to Coleman (1988), network cohesion is a source of bonding social capital, enabling the emergence of social norms. Concepts like social cohesion do not exist at the individual level, so its examination requires a collective approach. Relational cohesion across a collective entity, such as a cocoa farming village, may give rise to perceptions of what is “normal” to do on a farm, or establish collective perception of certain practices being locally rejected. Especially in situations when “right” or “wrong” is not clear, one relies on social validation (Coleman et al. 1959). Exploring whole networks of cocoa growing communities can allows us to better appreciate social environments that create and maintain local social norms under which farmers make decisions.

The case of the Sustainable Cocoa Production Programme

Cocoa cultivation needs to become more climate-friendly and socially sustainable to protect the global environment and support growers’ communities (Vervuurt et al. 2022). Farmers are also under pressure to be more efficient. Low productivity per hectare can drive expansion of production areas, deforestation, and greenhouse gas emissions. There is not one single practice that can make a cocoa farm climate friendly and socially sustainable but better general farm management can reduce the need for further growth of local cultivation areas and reduce pressures on the environment (Vervuurt et al. 2022).

What is required are overall good practices of organic matter management for carbon storage, appropriate shading, sustaining soil nutrient levels by a correct application of agro-inputs, and minimization of losses due to pests, waste, or poor tree health (Vervuurt et al. 2022). We examine the role of community network structures for the adoption of such practices promoted in Indonesia through the Sustainable Cocoa Production Programme. The program was implemented during 2012–2020 by a non-governmental non-profit organization (Swisscontact) in collaboration with their partners and support from the Indonesian Ministry of Home Affairs. The program has engaged 165,000 farmers across 10 Indonesian provinces who produce cocoa for export to international markets. The implementation resembled other common initiatives that aim to influence smallholder’ practices through capacity building, field demonstrations, and direct outreach (Norton and Alwang 2020). The official goals of the program were to reduce greenhouse gas emissions from cocoa production while sustaining productivity.

A key component of the program was a “Farmer Coaching Plan” that promoted 22 reportedly underutilized or often incorrectly applied practices in the region. The practices are generally considered to be good practices in cocoa production, leading to lower GHG emissions per unit of produced cocoa, by promoting plant and soil health and reducing waste, while supporting yields and thus growers’ livelihoods (Vervuurt et al. 2022). The practices relate to (1) side grafting, (2) usage of certified clones, (3) application of organic matter, (4) application of organic and inorganic fertilizer, (5) cocoa tree stem maintenance, (6) chupon treatment, (7) preventing and removing mumification, (8) preventing and removing dead branches, (9) use of shade trees, (10) limiting the tree height, (11) implementation of “leopard pruning” and (12) “glass pruning,” (13) prevention and control of black pod disease, (14) cocoa pod borer management, (15) helopeltis sanitation, (16) insecticide timing, (17) insecticide fit, (18) fungicide timing, (19) fungicide fit, (20) weed control, (21) harvest timing, and (22) bean drying. Further information about the practices and their measurement is provided in the methods section and Appendix 2.

METHODS

Here, we describe how we conceptualized and measured networks as explanatory variables in our models and how we measured and analyzed the adoption and similarity of farmers’ practices as dependent variables. Studies in the diffusion of innovations tradition tend to focus on individual technologies or practices (Rogers 2003). In this exploratory study, we include a larger collection of diverse practices (listed above and described in Appendix 2) to see if there exist any broader general trends regarding the role of whole networks across practices with different characteristics. In appendix 4, we further test different subsets of practices based on their expected private competitive advantage, observability, and communal benefits.

Each “network,” i.e., each observation, in this study is a village of cocoa growers in Sulawesi, Indonesia (see examples in Fig. 1). The boundary of each network is the administrative boundary of the village. Each “network actor” is a cocoa grower. The “network links” represent self-reported agricultural advice sharing among the community members. The small number of theoretically possible links between the villages in the sample is not considered in the analysis. The following subsections and appendices provide more details for the chosen approaches and the applied methods.

Analyzing the role of networks at the collective level

In addition to the conceptual reasons for our chosen level of analysis, another advantage of a collective-level analysis is that it mitigates some data reliability issues. Social network surveys can involve high levels of noise (Bernard et al. 1982), especially in large-scale surveys in remote settings. Respondents do not accurately recall communication links (Bernard and Killworth 1977) but a finding that some communities are split into a large number of network components because no one nominates peers beyond their own neighborhood is unlikely to change from day to day. Even if the dyads are captured with noise, the overall picture reflects the social reality of the community. If such whole-network network topologies are consistently associated with a community adopting similar or dissimilar practices, it is unlikely a coincidence but a footprint of real underlying processes. Such processes could be missed in an analysis of node-to-node contagion if these are driven by community characteristics (e.g., cohesive or not), rather than fluid-like flows through the pipework of imprecisely elicited links that are assumed in studies adopting an oversimplified diffusionist perspective (see Bernard et al. 2023 for a related discussion).

Data collection

The data used in this study was collected between May and December 2018 by a non-governmental development organization (Swisscontact) and their partner organization (Koltiva). The staff of the implementing agency (not the authors) surveyed farmers’ peer-to-peer information-seeking relationships and practices listed in Appendix 2. The specific prompt for gathering the network data was (as translated from Indonesian): “Please mention people outside of this household, where you get advice, you can learn from, or who can provide information and knowledge related to farming practices, especially about cocoa.” The survey that this study is based on comprised network data collection by this prompt and farm inspections of adoption of the recommended practices. The farmers consented to this data being used for research in an anonymized form and this was approved by the ethics committee of the University of Sydney. The implementation agency has collected further data from the farmers over the years, but this could not be shared for this study because of confidentiality reasons. Thus, we do not have any data on village geography, socio-demographic variables, and other potentially confounding characteristics that may affect both networks and practice adoption but we understand that all surveyed villages rely to a high degree on cocoa production.

The data collection that this paper is based on covered 73 villages in total. The villages are distributed across Sulawesi. It is possible that there could be some links between the villages but that was not the focus of this study, which revolved around internal community structures. In three of these villages, there were less than 10 cocoa farmers. For comparability, we decided to remove these very small villages, considering the possibility that some network processes in tiny communities may operate differently than in large ones (Matous and Bodin 2024). Also, some network metrics are highly sensitive to network size, which complicates comparisons between networks of very different sizes (Wasserman and Faust 1994; we have subsequently also conducted analyses with the full sample and obtained qualitatively similar results, see Appendix 4.) In total, 5001 peer-to-peer social ties of 4573 farmers in the remaining 70 villages, collected by 31 enumerators, were used in the analyses presented here. In these villages, we know of 5011 cocoa farmers who were registered with the implementing organization and who participated in the Sustainable Cocoa Production Programs. This corresponds to 91% coverage of the target population by the network survey. The adoption levels for each of the 22 recommended practices were assessed during inspections of the farmers’ fields by agricultural professionals working for the implementing organization. The purpose was to see whether a recommendation was effectively implemented on the farm, rather than only surveying self-reports of adoption. Self-reports alone would be less reliable in this context especially because the surveying organizations were also the implementing organizations of the program thus there would be a possibility of farmers’ bias toward self-reporting adoption, if not verified by observation.

It was neither feasible not desirable, to examine every tree on every farm but the decision was to examine two non-adjacent parts of each farm. Based on the experience of the local staff, this was sufficient to assess whether a practice is correctly implemented. On each farm, the inspectors designed with a string two square plots of 10x10 meters distanced 20 to 50 meters apart (depending on the farm size) and examined all trees in the designated areas for their maintenance condition and presence of any disease or pest. Based on this examination and interviews with the farmers, the inspectors scored the farmers’ practices following a template developed by the implementing organization and summarized in Appendix 2.

Measures

Community-level network characteristics

The prevalence of communication links is a fundamental network characteristic that can be quantified as the number of links per actor or as network density. Social network density is defined as the proportion of existing ties in a network relative to the number of theoretically possible ties (Wasserman and Faust 1994). Thus, density increases linearly with links per farmer and decreases quadratically with the size of network. The degree to which networks are composed of separate or loosely connected components, modules, and cliques is quantified respectively by the metrics of number of network components, modularity, and global triadic closure.

Centralization, as defined by Freeman (2004), quantifies whether a single actor is significantly more central than everyone else (Wasserman and Faust 1994). Assortativity quantifies the network actors’ tendency to connect to similar others, and thus relates to the concept of homophily (McPherson et al. 2001). It can be computed on any node property, e.g., centrality measured by “degree,” which is the number of actors’ links. High-assortativity in our data identified networks in which high-degree actors connected to other high-degree actors to form a cohesive network core, rather than being distributed in different parts of the network or separated into fragmented polycentric star structures and surrounded only by their low-degree followers. The measures are described in Appendix 1 and their details are explained in (Wasserman and Faust 1994).

Community-level practice adoption

A vector of length 22, composed of 0s and 1s, describes the practices of each farmer. Manhattan distance between the vectors of practices of two farmers equals the number of practices that one of the farmers adopted and the other did not adopt. Mean Manhattan distance (i.e., the mean of the sum of absolute differences) has been calculated for all pairs of these vectors at the village level, quantifying overall differences between practices of all farmers in each village. The additive inverse of Manhattan distance reflects practice similarity at a village level, which we view as a possible trace of social influence. We use this measure as one of our two main outcome variables to capture the village-level homogeneity of adoption and non-adoption across the full set of recommended practices.

The other analyzed outcome variable assesses broader adoption trends of a wider selection of recommended practices. We developed an “internally consistent scale” of practices (Streiner 2003). We considered that a simple sum of all surveyed practices would not necessarily be most reliable because some of the practices may have substitutive function (e.g., farmers might deliberately choose only one of two or three similar practices to achieve a similar aim). Using psych package in R (Revelle 2023), we developed this adoption scale inductively by gradually dropping variables that were most detrimental to the consistency of the scale in each step until an acceptable threshold was reached (Cronbach alpha > = 0.7; Tavakol and Dennick 2011). The resulting scale included 10 highly diverse but internally consistent set of practices. The practices span grafting techniques and tree pruning, maintaining soil health and preventing pest, fungi and disease.

The adoption scores for these 10 variables were summed to indicate a general propensity of whole collectives toward adopting a wide range of recommended practices and used as a dependent variable in multiple linear regressions (ordinary least squares: OLS). In Appendix 4, we also test a specification where all 22 practices are summed to form an overall index, even if that does not amount to an “internally consistent” scale that would ostensibly capture the same underlying construct (Streiner 2003). The p-values slightly increased in this potentially less reliable specification, but the qualitative patterns prevailed.

The reason a simple OLS regression could be applied to this data without violating independence assumptions is because the units of analysis are entire separate networks and significant interrelationships between the villages across the region are not expected. The presented analysis includes 70 observations, i.e., networks, which was sufficient for basic multiple regression models with two or three of explanatory variables (Green 1991). However, it is to be expected that small p-values may not be achieved with such minimal numbers of observation, unless the relationships between the predictors and the dependent variables is strong and there is little noise in the measures (Green 1991). The OLS does not control for potential geophysical and socio-demographic differences among the villages. Such data was not available to the research team because of confidentiality agreement and research ethics approval conditions. We tested various subsets of the selected theoretically relevant network characteristics. The presented models were chosen based on their fit and parsimony, guided by adjusted R² and Akaike Information Criterion (Cavanaugh and Neath 2019). Collinearity assessments were conducted with variance inflation factors and tolerance scores (Miles 2014). The data analysis was implemented in R using the igraph package for network analysis and mass and olsrr packages for the multiple linear regression and collinearity diagnostics (Csárdi and Nepusz 2006, Hebbali 2020).

RESULTS

Descriptive statistics and associations between measured concepts

Table 1 shows descriptive statistics of the researched communities. The network sizes range from 13 to 169 with a mean of 65 cocoa farmers per village. The farmers reported on average one agricultural information-sharing relationship. All villages displayed positive modularity scores. The networks were predominantly disassortative (only 4 of the 70 networks had positive assortativity scores and these were all small in magnitude). The theoretically possible minimum assortativity (-1) and the theoretically possible maximum centralization (1) were reached in one network, where every farmer reported the same individual as their agricultural information source. On the developed 10-point general adoption scale, the lowest adoption average for a village was 0.5 and the maximum was 6. Manhattan distance indicating the heterogeneity of adopted practices in each village ranged from 0.4 to 8.4, and so its additive inverse, which we use as a measure of practice homogeneity, is between -8.4 to -0.4.

Villages with more heterogeneous practices display a higher average prevalence of recommended practices (r = 0.67, p = 0.000; Appendix 3). In other words, in communities where farmers are united in their practices, fewer of these practices tend to be the recommended ones. Practices are more heterogeneous when farmers have fewer links (r = -0.46, p = 0.000) and their village networks are sparser (r = -0.305, p = 0.010), more modular (r = 0.43, p = 0.000), less transitive (r = -0.25, p = 0.038), have more components (r = 0.529, p = 0.000), and are less centralized (r = -0.25, p = 0.039). Recommended practice adoption is higher in networks where farmers have fewer links (r = -0.30, p = 0.011), more modular networks (r = 0.316, p = 0.008), with more components (r = 0.40, p = 0.001) that are less centralized (r = -0.26, p = 0.031). The following OLS analysis attempts to untangle which combinations of these network characteristics jointly predict outcomes at the village level.

Network predictors of collective outcomes

Among theoretically meaningful specifications, the network metric combinations in Table 2 predict best practice adoption and practice homogeneity. The models account for approximately one quarter of variation in overall practice adoption and one third of variation in practice homogeneity (R² = 0.26, adjusted R² = 0.23, AIC = 241.5 and R² = 0.34, adjusted R² = 0.31, AIC = 266.04, respectively). The best set predictors for recommended practice adoption were network density, number of components, and modularity. Density is determined by links per farmer and network size. When these two variables are included in the model separately instead of density, the fit deteriorates (AIC = 249.31 from 241). Links per farmer coefficient is positive with p = 0.774. Number of nodes coefficient is negative with p = 0.12. When links per farmer is removed from this specification, the fit improves (247.40) and the p-value for the node number decreases below 10%. Conversely, when links per farmer is kept in the model and the node number variable is removed, AIC increase (249.83) and links per farmer remain insignificant (p = 0.496). (Furthermore, Appendix 4 includes tests of adoption outcomes for different subsets of practices and shows that not all types of practices are associated with networks in the same way. See that Model 4 in Appendix 4 for a subset of practices related to limiting the spread of weeds, diseases, and pests has almost no explanatory power.) The best combined set of predictors for practice homogeneity is: (low) number of components, (low) modularity, and assortativity. Community practices are more similar if the entire village forms a cohesive whole that is not split into numerous disconnected components or loosely connected modules and when dissassortativity tendencies are not too strong. In other words, the community is not fractured into islands of followers of popular farmers (or “patrons”) who do not communicate with other popular farmers.

In summary, villages displaying high adoption of recommended practices tend to have more heterogeneous practices (r = 0.67; p = 0.000). An archetypical example of a village network structure with a high adoption of recommended practices and a high heterogeneity of practices is illustrated in Figure 1A (which is an actual community in the sample). The absence of connections between high-degree nodes contributes to community fragmentation into separate components and modules. (General adoption score = 4.42, adoption homogeneity = -8.38, modularity = 0.84, assortativity = -0.47, number of components = 14.) Figure 1B illustrates an archetypical village network in our sample with a high homogeneity of practices and low adoption of recommended practices. High degree nodes share information within the network core and most actors are interconnected into a single cohesive component. (General adoption score = 0.29, adoption homogeneity = -0.54, modularity = 0.46, assortativity = -0.16, number of components = 3.)

DISCUSSION

Food production is an important contributor to climate change and this challenge is particularly difficult to tackle in smallholder agriculture. In addition to environmental issues associated with agricultural production in general, there are many millions of smallholders scattered around the world, including very remote areas, and we often do not know, or are able to ethically influence, what happens on small farms that can operate quite informally. However, any small issues on such numerous farms can aggregate to large environmental impacts globally.

We have examined which community network characteristics are associated with smallholders’ adoption of practices recommended through a sustainability program. Analyzing a unique dataset of 70 village networks, the network topologies showed some similarities and some differences. All villages in our data displayed positive modularity scores, which means that their networks were more segregated into siloed clusters than what would be expected by chance. This is not surprising as social relationships are commonly constrained within local cliques by demography, geography, and social settings (McPherson et al. 2001).

It is also understandable that the heterogeneity of applied practices was higher in networks that were more fragmented into distinct clusters and separate components. However, it is interesting that the overall adoption of recommended practices was also higher in such networks. Furthermore, after fragmentation into components and modules was controlled for, assortativity (i.e., connecting to those with a similar degree) was associated with a higher homogeneity of practices. Specifically, networks and components in which high-degree individuals connect to each other and form cohesive cores appear to provide strong conditions for practice similarity, possibly because of emerging social norms and social influence channelled through such networks. In such villages, similar practices seem to diffuse and prevail against new alternatives that might not have the acceptance of the majority (yet). In the present case, the homogeneously adopted practices in cohesive village networks were not necessarily those recommended by the studied program.

The overall lower adoption trends of cohesive villages were found for a broad mix of practices when considered jointly. However, further tests in the appendices suggest, as could be expected, that different network mechanisms are likely more pronounced for adoption of different types of practices. Previous studies have identified main determinants of agricultural practice and technology adoption in terms of the characteristics of what was being adopted (Kuehne et al. 2017). These innovation characteristics do not necessarily all matter for which community network patterns affect adoption. The way in which networks influence adoption might depend on some less noted practice characteristics, for example, whether the benefits of a practice are relatively more private or public (Matous 2022). Communal pressures toward adoption may arise through cohesive networks when innovation provides non-negligible community benefits or non-adoption poses risks to the community. In the tests reported in the appendices, we do not observe negative effects of cohesive networks on adoption of practices related to limiting the spread of weeds, diseases, and pests, which are most likely to provide positive externalities to the community beyond the adopters’ plot. Possible pro-adoption effects of cohesive networks that may channel neighbors’ interests in pest-free community may have cancelled out, in the present case, the possible anti-adoption effects of bonding network-induced inertia toward status quo or distrust toward external advice. Examining divergent network influences for different types of practices in more depth would be a worthwhile subject of further research. In this study, the main focus was on comparing communities, rather than practices, to explore which whole network characteristics may affect adoption in general.

Networks that are divided into distinct components and polycentric networks where most influential actors are not closely intertwined seem to come with more opportunities to adopt something new and different from others rather than maintaining the same old collectively accepted local practice. There are various possible interpretations of this finding. Less cohesive networks perhaps better nurture diverse knowledge. Lower cohesion may be effective in adaptation to changing environmental conditions as long as there are opportunities to collectively learn from these different pools of knowledge and experience. The network links in our study capture only significant information exchange partners. Within these smallholder communities, there is likely a visibility of the conditions on the plots of other farmers beyond the reported links to the most important advisors. Therefore, farmers may learn and eventually benefit from experimentation in separate network components.

Previous studies finding substitution between intra-community and external information sharing and influence (Bandiera and Rasul 2006) also suggest why cohesive community networks could block externally-induced change. Strong networks within a limited circle of partners were found to support old locally validated practices rather than practices promoted through agri-environment schemes in the name of sustainable adaptation to environmental change (Arnott et al. 2021, Cofré-Bravo et al. 2019). In the presence of strong network bonding, innovation champions were reported to need to “break through the constraint of over-connectedness” to get their message across (King et al. 2019:131). Bonding can come at the expense of bridging and linking social capital, which matter for sustainable technology adoption (Datoon et al. 2023). Bridging and linking was not examined in our study. Our data comes from the final periods of an 8-year program. It would be another question for a future study, who kept the adopted practices after the program was completed, when external professionals departed, and with them a potential source of bridging and linking social capital?

The cross-sectional nature of the network and adoption data does not allow us to directly address causality and the inherent dynamics of the above-implied diffusion and normative social influence processes. We could only observe at one point in time the aggregated results of farmers’ past networking choices and their previous adoption or disadoption decisions. It would have been ideal to empirically prove the channels of influence but causality is extremely difficult to establish in ways that are both internally and externally valid in networked contexts with interdependent actors, even in longitudinal experimental studies (Pratiwi and Matous 2024).

Nevertheless, despite the lack of direct evidence of exact causal pathways and a lack of control for numerous unknown factors that could be confounding the reported quantitative associations, the present findings demonstrate which network structures predict achievement of intended program outcomes. We believe it is useful to know in general in which types of communities this type of program might work and where other approaches should be considered. Even if we cannot prove causality behind the observed regularities, networks are at least a measurable marker, and perhaps a footprint, of unobserved but consequential community characteristics and social processes.

Other studies in similar contexts that engaged more directly with farmers’ lived experience, and employ qualitative inquiry and naturalistic participant observation, further help to make sense of our quantitative findings (Bray and Neilson 2018, Vicol et al. 2019). Such studies remind us that sometimes “sustainability programmes may be inadvertently attempting to encourage household investment in a particular kind of agriculture, which is intended to assist sustainability of supply, but is poorly aligned with prevailing processes of poverty alleviation” (Bray and Neilson 2018:368). We do not examine the environmental and socioeconomic consequences of the recommended practices. The practices have been researched and endorsed by others in works cited in Appendix 2. However, we understand that interventions can come with unintended consequences (Isaac et al. 2021). Hence, despite the potential virtues of well-implemented evidence-based agricultural training and greenhouse gas reduction programs, we would like to avoid our results being interpreted as cohesive community networks being universally detrimental to sustainability of smallholder agriculture. This study took an implicit “pro-adoption” perspective (stemming from a long tradition in the diffusion of innovations literature; Rogers 1976) but we understand that local people have intimate knowledge of their land and may know better what works in their landscape than external experts. This may particularly be the case for people living their whole lives in communities with abundant community-wide exchange of experience, which may be a factor why members of such cohesive communities may be less dependent on external advice.

CONCLUSION

Networks present both opportunities and constraints (Kadushin 2012), with the latter perhaps not as often acknowledged as the former. Leveraging a uniquely broad dataset for its kind, this study subjected to empirical scrutiny some rarely tested theoretical expectations on the role of community networks in environmental conservation. Although literature review showed that strengthening community networks and social capital is a well-regarded strategy in sustainability programs, our findings illustrate that cohesive community network structures may not always support the expected outcomes of such programs.

Community network structures that form one cohesive whole and where popular individuals communicated with each other displayed, perhaps unsurprisingly, a higher homogeneity of practice, but this was not necessarily in line with recommendations. By contrast, and perhaps surprisingly, in disconnected community networks, a higher proportion of farmers used practices that diverged from the community mainstream and aligned with expert recommendations. Although cohesive networks may homogenize practices, they do not automatically diffuse what is “implanted” into them from outside. Cohesive networks may rather cement in what is already there and make it harder for different practices to break through.

This study points to the social fabric of agrarian villages as one possible marker as to why the same program may produce expected outcomes in some communities and not others. There is no reason why stronger community networks should support societal or environmental goals per se (Rothstein and Stolle 2008). Social capital of cohesively interconnected villages is not waiting there to be leveraged by others. Excessive “bonding” social capital with peers from the same social group can obstruct larger societal goals without trustful “bridging” or “linking” to the external world (Putnam et al. 1993, Woolcock 1998, Aldrich and Meyer 2015). For good or bad, cohesive communities where influential farmers communicate with each other might be harder to convince to leave established customs and follow official recommendations.

When bonding social capital dominates over other types, trust is rarely extended to institutions beyond the community (Dale and Newman 2010, Kant and Vertinsky 2022). In turn, distrust and disconnect from formal institutions may encourage more bonding (Rothstein and Stolle 2008). External interveners need to be sensitive to such issues in their search for appropriate points of entry into closed networks to repair trust and broaden the communities’ social opportunities (Varshney et al. 2022). Especially, they need to be open to the possibility that the members of cohesive communities may have more effective solutions at their disposal, know better what works in their local context, and intervenors need to be ready to listen and learn. We hope that the questions raised in this study can be further empirically explored with experimental, larger number, whole-network studies, more longitudinal data as well as deeper ethnographic insights, which will help us better understand community norms, and farmers’ motivations and cultural practices.

LITERATURE CITED

Abdul-Rahaman, A., and A. Abdulai. 2020. Social networks, rice value chain participation and market performance of smallholder farmers in Ghana. African Development Review 32(2):216-227. https://doi.org/10.1111/1467-8268.12429

Abrahamson, E., and L. J. O. S. Rosenkopf. 1997. Social network effects on the extent of innovation diffusion: a computer simulation. Organization Science 8(3):289-309. https://doi.org/10.1287/orsc.8.3.289

Albizua, A., E. Bennett, U. Pascual, and G. Larocque. 2020. The role of the social network structure on the spread of intensive agriculture: an example from Navarre, Spain. Regional Environmental Change 20(3):99. https://doi.org/10.1007/s10113-020-01676-9

Aldrich, D. P., and M. A. Meyer. 2015. Social capital and community resilience. American Behavioral Scientist 59(2):254-269. https://doi.org/10.1177/0002764214550299

Angst, M., A. Widmer, M. Fischer, and K. Ingold. 2018. Connectors and coordinators in natural resource governance: insights from Swiss water supply. Ecology and Society 23(2):1. https://doi.org/10.5751/ES-10030-230201

Arnott, D., D. R. Chadwick, S. Wynne-Jones, N. Dandy, and D. L. Jones. 2021. Importance of building bridging and linking social capital in adapting to changes in UK agricultural policy. Journal of Rural Studies 83:1-10. https://doi.org/10.1016/j.jrurstud.2021.02.001

Balehegn, M., A. Duncan, A. Tolera, A. A. Ayantunde, S. Issa, M. Karimou, N. Zampaligré, K. André, I. Gnanda, P. Varijakshapanicker, E. Kebreab, J. Dubeux, K. Boote, M. Minta, F. Feyissa, and A. T. Adesogan. 2020. Improving adoption of technologies and interventions for increasing supply of quality livestock feed in low- and middle-income countries. Global Food Security 26:100372. https://doi.org/10.1016/j.gfs.2020.100372

Bandiera, O., and I. Rasul. 2006. Social networks and technology adoption in Northern Mozambique. Economic Journal 116(514):869-902. https://doi.org/10.1111/j.1468-0297.2006.01115.x

Bernard, C., X. Poux, I. Herzon, J. Moran, T. Pinto-Correia, D. E. Dumitras, M. I. Ferraz-de-Oliveira, F. Gouriveau, D. Goussios, M. I. Jitea, Y. Kazakova, R. Koivuranta, F. Lerin, M. Ljung, A. Lomba, V. C. Mihai, M. Puig de Morales Fusté, and G. Vlahos. 2023. Innovation brokers in High Nature Value farming areas: a strategic approach to engage effective socioeconomic and agroecological dynamics. Ecology and Society 28(1):20. https://doi.org/10.5751/ES-13522-280120

Bernard, H. R., and P. D. Killworth. 1977. Informant accuracy in social network data II. Human Communication Research 4(1):3-18. https://doi.org/10.1111/j.1468-2958.1977.tb00591.x

Bernard, H. R., P. D. Killworth, and L. Sailer. 1982. Informant accuracy in social-network data V. An experimental attempt to predict actual communication from recall data. Social Science Research 11(1):30-66. https://doi.org/10.1016/0049-089X(82)90006-0

Bezner Kerr, R., J. Kangmennaang, L. Dakishoni, H. Nyantakyi-Frimpong, E. Lupafya, L. Shumba, R. Msachi, G. O. Boateng, S. S. Snapp, A. Chitaya, E. Maona, T. Gondwe, P. Nkhonjera, and I. Luginaah. 2019. Participatory agroecological research on climate change adaptation improves smallholder farmer household food security and dietary diversity in Malawi. Agriculture, Ecosystems & Environment 279:109-121. https://doi.org/10.1016/j.agee.2019.04.004

Bodin, Ö., S. M. Alexander, J. Baggio, M. L. Barnes, R. Berardo, G. S. Cumming, L. E. Dee, A. P. Fischer, M. Fischer, M. Mancilla Garcia, A. M. Guerrero, J. Hileman, K. Ingold, P. Matous, T. H. Morrison, D. Nohrstedt, J. Pittman, G. Robins, and J. S. Sayles. 2019. Improving network approaches to the study of complex social-ecological interdependencies. Nature Sustainability 2(7):551-559. https://doi.org/10.1038/s41893-019-0308-0

Bodin, Ö, and B. I. Crona. 2009. The role of social networks in natural resource governance: What relational patterns make a difference? Global Environmental Change 19(3):366-374. https://doi.org/10.1016/j.gloenvcha.2009.05.002

Bodin, Ö, and J. Norberg. 2005. Information network topologies for enhanced local adaptive management. Environmental Management 35(2):175-193. https://doi.org/10.1007/s00267-004-0036-7

Bray, J., and J. Neilson. 2018. Examining the interface of sustainability programmes and livelihoods in the Semendo highlands of Indonesia. 59(3):368-383. https://doi.org/10.1111/apv.12205

Breiger, R. L. 1974. The duality of persons and groups. Social Forces 53(2):181-190. https://doi.org/10.2307/2576011

Burt, R. S. 1995. Structural holes. Harvard University Press, Cambridge, Massachusetts, USA. https://doi.org/10.4159/9780674029095

Cavanaugh, J. E., and A. A. Neath. 2019. The Akaike information criterion: background, derivation, properties, application, interpretation, and refinements. WIREs Computational Statistics 11(3):e1460. https://doi.org/10.1002/wics.1460

Cofré-Bravo, G., L. Klerkx, and A. Engler. 2019. Combinations of bonding, bridging, and linking social capital for farm innovation: how farmers configure different support networks. Journal of Rural Studies 69:53-64. https://doi.org/10.1016/j.jrurstud.2019.04.004

Coleman, J., H. Menzel, and E. Katz. 1959. Social processes in physicians’ adoption of a new drug. Journal of Chronic Diseases 9(1):1-19. https://doi.org/10.1016/0021-9681(59)90134-1

Coleman, J. S. 1988. Social capital in the creation of human capital. American Journal of Sociology 94(Supplement: Organizations and Institutions: Sociological and Economic Approaches to the Analysis of Social Structure):S95-S120. https://doi.org/10.1086/228943

Crona, B. I. 2006. Supporting and enhancing development of heterogeneous ecological knowledge among resource users in a Kenyan seascape. Ecology and Society 11(1):32. https://doi.org/10.5751/ES-01712-110132

Csárdi, G., and T. Nepusz. 2006. The igraph software package for complex network research. InterJournal Complex Systems.

Cui, Z., H. Zhang, X. Chen, C. Zhang, W. Ma, C. Huang, W. Zhang, G. Mi, Y. Miao, X. Li, Q. Gao, J. Yang, Z. Wang, Y. Ye, S. Guo, J. Lu, J. Huang, S. Lv, Y. Sun, Y. Liu, X. Peng, J. Ren, S. Li, X. Deng, X. Shi, Q. Zhang, Z. Yang, L. Tang, C. Wei, L. Jia, J. Zhang, M. He, Y. Tong, Q. Tang, X. Zhong, Z. Liu, N. Cao, C. Kou, H. Ying, Y. Yin, X. Jiao, Q. Zhang, M. Fan, R. Jiang, F. Zhang, and Z. Dou. 2018. Pursuing sustainable productivity with millions of smallholder farmers. Nature 555(7696):363-366. https://doi.org/10.1038/nature25785

Dale, A., and L. Newman. 2010. Social capital: a necessary and sufficient condition for sustainable community development? Community Development Journal 45(1):5-21. https://doi.org/10.1093/cdj/bsn028

Datoon, R. F., J. V. Camacho, A. V. Lapitan, and J. M. Gapas. 2023. Bridging and bonding social capital for sustainable technology adoption, the Landcare program in the resource poor rural town of Claveria, Philippines. Research in Globalization 7:100152. https://doi.org/10.1016/j.resglo.2023.100152

Daymond, A. J., A. Prawoto, S. Abdoellah, A. W. Susilo, N. C. Cryer, F. Lahive, and P. Hadley. 2020. Variation in Indonesian cocoa farm productivity in relation to management, environmental and edaphic factors. Experimental Agriculture 56(5):738-751. https://doi.org/10.1017/S0014479720000289

Devine, G. J., and M. J. Furlong. 2007. Insecticide use: contexts and ecological consequences. Agriculture and Human Values 24(3):281-306. https://doi.org/10.1007/s10460-007-9067-z

Di Gregorio, M., L. Fatorelli, J. Paavola, B. Locatelli, E. Pramova, D. R. Nurrochmat, P. H. May, M. Brockhaus, I. M. Sari, and S. D. Kusumadewi. 2019. Multi-level governance and power in climate change policy networks. Global Environmental Change 54:64-77. https://doi.org/10.1016/j.gloenvcha.2018.10.003

Djuideu, C. T. L., H. D. B. Bisseleua, S. Kekeunou, and F. C. Ambele. 2021. Rehabilitation practices in cocoa agroforestry systems mitigate outbreaks of termites and support cocoa tree development and yield. Agriculture, Ecosystems & Environment 311:107324. https://doi.org/10.1016/j.agee.2021.107324

Effendy, N. Hanani, B. Setiawan and A. W. Muhaimin. 2013. Effect characteristics of farmers on the level of technology adoption side-grafting in cocoa farming at Sigi Regency-Indonesia. Journal of Agricultural Science 5(12): 72-77. https://doi.org/10.5539/jas.v5n12p72

Ernstson, H., S. Sörlin, and T. Elmqvist. 2008. Social movements and ecosystem services—the role of social network structure in protecting and managing urban green areas in Stockholm. Ecology and Society 13(2):39. https://doi.org/10.5751/ES-02589-130239

Ferse, S. C. A., L. Knittweis, G. Krause, A. Maddusila, and M. Glaser. 2012. Livelihoods of ornamental coral fishermen in South Sulawesi/Indonesia: implications for management. Coastal Management 40(5):525-555. https://doi.org/10.1080/08920753.2012.694801

Freeman, L. C. 2004. The development of social network analysis: a study in the sociology of science. Empirical Press, New York, New York, USA.

Green, S. B. 1991. How many subjects does it take to do a regression analysis. Multivariate Behavioral Research 26(3):499-510. https://doi.org/10.1207/s15327906mbr2603_7

Guehi, T. S., I. B. Zahouli, L. Ban-Koffi, M. A. Fae, and J. G. Nemlin. 2010. Performance of different drying methods and their effects on the chemical quality attributes of raw cocoa material. International Journal of Food Science and Technology 45(8):1564-1571. https://doi.org/10.1111/j.1365-2621.2010.02302.x

Guest, D. 2007. Black pod: diverse pathogens with a global impact on cocoa yield. Phytopathology 97(12):1650-1653. https://doi.org/10.1094/PHYTO-97-12-1650

Harvey, C. A., M. Saborio-Rodríguez, M. R. Martinez-Rodríguez, B. Viguera, A. Chain-Guadarrama, R. Vignola, and F. Alpizar. 2018. Climate change impacts and adaptation among smallholder farmers in Central America. Agriculture & Food Security 7(1):57. https://doi.org/10.1186/s40066-018-0209-x

Hebbali, A. 2020. Tools for building OLS regression models. https://cran.r-project.org/package=olsrr

Hoffmann, M. P., J. Cock, M. Samson, N. Janetski, K. Janetski, R. P. Rötter, M. Fisher, and T. Oberthür. 2020. Fertilizer management in smallholder cocoa farms of Indonesia under variable climate and market prices. Agricultural Systems 178:102759. https://doi.org/10.1016/j.agsy.2019.102759

Hua, Y., F. Dong, and J. Goodman. 2021. How to leverage the role of social capital in pro-environmental behavior: a case study of residents’ express waste recycling behavior in China. Journal of Cleaner Production 280:124376. https://doi.org/10.1016/j.jclepro.2020.124376

Hulme, M. F., D. Salliss, M. S. Konneh, P. Dauda, E. Witcutt, and F. J. Sanderson. 2018. Improving cocoa harvest can mitigate for crop damage by wildlife in a forest-agriculture matrix. Agriculture, Ecosystems & Environment 265:236-243. https://doi.org/10.1016/j.agee.2018.06.019

Isaac, M. E., B. H. Erickson, S. Quashie-Sam, and V. R. Timmer. 2007. Transfer of knowledge on agroforestry management practices: structure of informal social networks. Ecology and Society 12(2):32. https://doi.org/10.5751/ES-02196-120232

Isaac, M. E., H. Nyantakyi-Frimpong, P. Matouš, E. Dawoe, and L. C. N. Anglaaere. 2021. Farmer networks and agrobiodiversity interventions: the unintended outcomes of intended change. Ecology and Society 26(4):12. https://doi.org/10.5751/ES-12734-260412

Kadushin, C. 2012. Understanding social networks: theories, concepts, and findings. Oxford University Press, Oxford, UK.

Kant, S., and I. Vertinsky. 2022. The anatomy of social capital of a Canadian indigenous community: implications of social trust field experiments for community-based forest management. Forest Policy and Economics 144:102822. https://doi.org/10.1016/j.forpol.2022.102822

King, B., S. Fielke, K. Bayne, L. Klerkx, and R. Nettle. 2019. Navigating shades of social capital and trust to leverage opportunities for rural innovation. Journal of Rural Studies 68:123-134. https://doi.org/10.1016/j.jrurstud.2019.02.003

Kuehne, G., R. Llewellyn, D. J. Pannell, R. Wilkinson, P. Dolling, J. Ouzman, and M. Ewing. 2017. Predicting farmer uptake of new agricultural practices: a tool for research, extension and policy. Agricultural Systems 156:115-125. https://doi.org/10.1016/j.agsy.2017.06.007

Lazega, E., and T. Snijders. 2016a. Multilevel network analysis for the social sciences: theory, methods and applications. Springer, Cham, Switzerland. https://doi.org/10.1007/978-3-319-24520-1

Lazega, E., and T. A. B. Snijders. 2016b. General conclusion. Pages 355-361 in E. Lazega and T. A. B. Snijders, editors. Multilevel network analysis for the social sciences: theory, methods and applications. Springer International, Cham, Switzerland. https://doi.org/10.1007/978-3-319-24520-1_15

Leavitt, H. J. 1951. Some effects of certain communication patterns on group performance. Journal of Abnormal Social Psychology 46(1):38-50. https://doi.org/10.1037/h0057189

Lin, N. 2001. Social capital: a theory of social structure and action. Cambridge University Press, Cambridge, UK. https://doi.org/10.1017/CBO9780511815447

Little, L. R., and A. D. McDonald. 2007. Simulations of agents in social networks harvesting a resource. Ecological Modelling 204(3-4):379-386. https://doi.org/10.1016/j.ecolmodel.2007.01.013

Lubell, M., P. Matous, L. Klerkx, and C. Barahona. 2023. The population ecology of sustainable agriculture knowledge networks: insights from California. Ecology and Society 28(2):15. https://doi.org/10.5751/ES-14179-280215

Marshall, G. R. 2009. Polycentricity, reciprocity, and farmer adoption of conservation practices under community-based governance. Ecological Economics 68(5):1507-1520. https://doi.org/10.1016/j.ecolecon.2008.10.008

Matous, P. 2022. Community resilience under rural development projects and technocratic interventions: agenda for social network research. Pages 213-224 in E. Lazega, T. A. B. Snijders and R. P. M. Wittek, editors. A research agenda for social networks and social resilience. Edward Elgar, Cheltenham, UK. https://doi.org/10.4337/9781803925783.00020

Matous, P. 2023. Male and stale? Questioning the role of “opinion leaders” in agricultural programs. Agriculture and Human Values 40:1205-1220. https://doi.org/10.1007/s10460-023-10415-9

Matous, P., and Ö. Bodin. 2024. Hub-and-spoke social networks among Indonesian cocoa farmers homogenise farming practices. People and Nature 6(2):598-609. https://doi.org/10.1002/pan3.10578

Mbaru, E. K., and M. L. Barnes. 2017. Key players in conservation diffusion: using social network analysis to identify critical injection points. Biological Conservation 210:222-232. https://doi.org/10.1016/j.biocon.2017.03.031

McPherson, M., L. Smith-Lovin, and J. M. Cook. 2001. Birds of a feather: homophily in social networks. Annual Review of Sociology 27:415-444. https://doi.org/10.1146/annurev.soc.27.1.415

Miles, J. 2014. Tolerance and variance inflation factor. Wiley StatsRef: Statistics Reference Online. https://doi.org/10.1002/9781118445112.stat06593

Mulia, S., P. J. McMahon, A. Purwantara, H. Bin Purung, F. Djufry, S. Lambert, P. J. Keane, and D. I. Guest. 2019. Effect of organic and inorganic amendments on productivity of cocoa on a marginal soil in Sulawesi, Indonesia. Experimental Agriculture 55(1):1-20. https://doi.org/10.1017/S0014479717000527

Musavengane, R., and D. M. Simatele. 2016. Community-based natural resource management: the role of social capital in collaborative environmental management of tribal resources in KwaZulu-Natal, South Africa. Development Southern Africa 33(6):806-821. https://doi.org/10.1080/0376835X.2016.1231054

Nakano, Y., T. W. Tsusaka, T. Aida, and V. O. Pede. 2018. Is farmer-to-farmer extension effective? The impact of training on technology adoption and rice farming productivity in Tanzania. World Development 105:336-351. https://doi.org/10.1016/j.worlddev.2017.12.013

Norton, G. W., and J. Alwang. 2020. Changes in agricultural extension and implications for farmer adoption of new practices. Applied Economic Perspectives and Policy 42(1):8-20. https://doi.org/10.1002/aepp.13008

Nurdin, N., and A. Grydehøj. 2014. Informal governance through patron–client relationships and destructive fishing in Spermonde Archipelago, Indonesia. Journal of Marine and Island Cultures 3(2):54-59. https://doi.org/10.1016/j.imic.2014.11.003

Pamuk, H., E. Bulte, and A. A. Adekunle. 2014. Do decentralized innovation systems promote agricultural technology adoption? Experimental evidence from Africa. Food Policy 44:227-236. https://doi.org/10.1016/j.foodpol.2013.09.015

Phuong, L. T. H., A. Wals, L. T. H. Sen, N. Q. Hoa, P. Van Lu, and R. Biesbroek. 2018. Using a social learning configuration to increase Vietnamese smallholder farmers’ adaptive capacity to respond to climate change. Local Environment 23(8):879-897. https://doi.org/10.1080/13549839.2018.1482859

Pratiwi, A., and P. Matous. 2024. Using randomized control trials in socio-ecological systems: adoption of agrisilvicultural practices in Indonesia. In M. Barnes and Ö. Bodin, editors. Social networks and the environment. Edward Elgar, Cheltenham, UK, in press.

Putnam, R. D. 1995. Bowling alone: American’s declining social capital. Journal of Democracy 6(1):65-78. https://doi.org/10.1353/jod.1995.0002

Putnam, R. D., R. Leonardi, and R. Y. Nonetti. 1993. Making democracy work: civic traditions in modern Italy. Princeton University Press, Princeton, New Jersey, USA. https://doi.org/10.1515/9781400820740

Ramirez, M., P. Bernal, I. Clarke, and I. Hernandez. 2018. The role of social networks in the inclusion of small-scale producers in agri-food developing clusters. Food Policy 77:59-70. https://doi.org/10.1016/j.foodpol.2018.04.005

Reincke, K., E. Vilvert, A. Fasse, F. Graef, S. Sieber, and M. A. Lana. 2018. Key factors influencing food security of smallholder farmers in Tanzania and the role of cassava as a strategic crop. Food Security 10(4):911-924. https://doi.org/10.1007/s12571-018-0814-3

Revelle, W. 2023. psych: Procedures for psychological, psychometric, and personality research.

Robinson, W. S. 1950. Ecological correlations and the behavior of individuals. American Sociological Review 15(3):351-357. https://doi.org/10.2307/2087176

Rockenbauch, T., P. Sakdapolrak, and H. Sterly. 2019. Do translocal networks matter for agricultural innovation? A case study on advice sharing in small-scale farming communities in Northeast Thailand. Agriculture and Human Values 36(4):685-702. https://doi.org/10.1007/s10460-019-09935-0

Rogers, E. M. 1976. Communication and development: the passing of the dominant paradigm. Communication Research 3(2):213-240. https://doi.org/10.1177/009365027600300207

Rogers, E. M. 2003. Diffusion of innovations. Fifth edition. Free Press, New York, New York, USA.

Rothstein, B., and D. Stolle. 2008. The state and social capital: an institutional theory of generalized trust. Comparative Politics 40(4):441-459. https://doi.org/10.5129/001041508X12911362383354

Ruseani, N. S., W. Vanhove, A. W. Susilo, and P. Van Damme. 2022. Clonal differences in nitrogen use efficiency and macro-nutrient uptake in young clonal cocoa (Theobroma cacao L.) seedlings from Indonesia. Journal of Plant Nutrition 45(20):3196-3211. https://doi.org/10.1080/01904167.2022.2057328

Saptutyningsih, E., D. Diswandi, and W. Jaung. 2020. Does social capital matter in climate change adaptation? A lesson from agricultural sector in Yogyakarta, Indonesia. Land Use Policy 95:104189. https://doi.org/10.1016/j.landusepol.2019.104189

Shaner, D. L., and H. J. Beckie. 2014. The future for weed control and technology. Pest Management Science 70(9):1329-1339. https://doi.org/10.1002/ps.3706

Snijders, T. A. B. 2016. The multiple flavours of multilevel issues for networks. Pages 15-46 in E. Lazega and T. A. B. Snijders, editors. Multilevel network analysis for the social sciences: theory, methods and applications. Springer International, Cham, Switzerland. https://doi.org/10.1007/978-3-319-24520-1_2

Streiner, D. L. 2003. Starting at the beginning: an introduction to coefficient alpha and internal consistency. Journal of Personality Assessment 80(1):99-103. https://doi.org/10.1207/S15327752JPA8001_18

Tavakol, M., and R. Dennick. 2011. Making sense of Cronbach’s alpha. International Journal of Medical Education 2:53-55. https://doi.org/10.5116/ijme.4dfb.8dfd

Tosto, A., P. A. Zuidema, E. Goudsmit, J. B. Evers, and N. P. R. Anten. 2022. The effect of pruning on yield of cocoa trees is mediated by tree size and tree competition. Scientia Horticulturae 304:111275. https://doi.org/10.1016/j.scienta.2022.111275

Vanhove, W., Y. Feng, M. Yu, I. O. Hafiz, N. Vanhoudt, P. L. J. Van Damme, and A. Zhang. 2020. Evaluation of attract-and-kill strategy for management of cocoa pod borer, Conopomorpha cramerella, in Malaysia cocoa plantation. International Journal of Pest Management 66(2):155-162. https://doi.org/10.1080/09670874.2019.1578003

Varshney, D., A. K. Mishra, P. K. Joshi, and D. Roy. 2022. Social networks, heterogeneity, and adoption of technologies: evidence from India. Food Policy 112:102360. https://doi.org/10.1016/j.foodpol.2022.102360

Vervuurt, W., M. A. Slingerland, A. A. Pronk, and L. G. J. Van Bussel. 2022. Modelling greenhouse gas emissions of cacao production in the Republic of Côte d’Ivoire. Agroforestry Systems 96(2):417-434. https://doi.org/10.1007/s10457-022-00729-8

Vicol, M., N. Fold, B. Pritchard, and J. Neilson. 2019. Global production networks, regional development trajectories and smallholder livelihoods in the Global South. Journal of Economic Geography 19(4):973-993. https://doi.org/10.1093/jeg/lby065

Wartenberg, A. C., W. J. Blaser, J. M. Roshetko, M. Van Noordwijk, and J. Six. 2020. Soil fertility and Theobroma cacao growth and productivity under commonly intercropped shade-tree species in Sulawesi, Indonesia. Plant and Soil 453(1):87-104. https://doi.org/10.1007/s11104-018-03921-x

Wasserman, S., and K. Faust. 1994. Social network analysis: methods and applications. Cambridge University Press, Cambridge, UK. https://doi.org/10.1017/CBO9780511815478

Wekesa, B. M., O. I. Ayuya, and J. K. Lagat. 2018. Effect of climate-smart agricultural practices on household food security in smallholder production systems: micro-level evidence from Kenya. Agriculture & Food Security 7:80. https://doi.org/10.1186/s40066-018-0230-0

Wielgoss, A., Y. Clough, B. Fiala, A. Rumede, and T. Tscharntke. 2012. A minor pest reduces yield losses by a major pest: plant-mediated herbivore interactions in Indonesian cacao. Journal of Applied Ecology 49(2):465-473. https://doi.org/10.1111/j.1365-2664.2012.02122.x

Woolcock, M. 1998. Social capital and economic development: toward a theoretical synthesis and policy framework. Theory and Society 27(2):151-208. https://doi.org/10.1023/A:1006884930135

Zhang, A. J., P. Matous, and D. K. Y. Tan. 2020. Forget opinion leaders: the role of social network brokers in the adoption of innovative farming practices in North-western Cambodia. International Journal of Agricultural Sustainability 18(4):266-284. https://doi.org/10.1080/14735903.2020.1769808

Zulkafli, Z., K. Perez, C. Vitolo, W. Buytaert, T. Karpouzoglou, A. Dewulf, B. De Bièvre, J. Clark, D. M. Hannah, and S. Shaheed. 2017. User-driven design of decision support systems for polycentric environmental resources management. Environmental Modelling & Software 88:58-73. https://doi.org/10.1016/j.envsoft.2016.10.012