Forget opinion leaders: the role of social network brokers in the adoption of innovative farming practices in North-western Cambodia

ABSTRACT

It has become accepted that social networks influence farmers’ decisions and agricultural development programs routinely support influential farmers in an effort to disseminate recommended practices through their social networks. We have interviewed (1) 120 heads of farming households in one village in North-western Cambodia about their networks and practices; and (2) representatives of organizations managing agricultural development programs in the region. We have constructed an information-sharing network of the village and computed network centrality measures of each farmer in the village indicating their access to information from peers, opinion leadership, brokerage, and the redundancy among their contacts. We have analysed the relation of farmers’ networks and their practices. While the number of links farmers’ had in the community was mainly unrelated with their adoption decisions, the structure of these links mattered. Farmers who were perceived as influential were not necessarily inclined towards recommended practices. In contrast, farmers who had fewer contacts but were in brokerage positions on main channels of information flows between different groups displayed more progressive practices. The results suggest some limitations of excessive reliance on perceived influential farmers in agricultural programs and highlight the role that network brokers play in adoption of innovative farming practices.

KEYWORDS:

• Resource-conserving practices
• social networks
• brokerage
• opinion leadership
• technology adoption
• Cambodia

1. Introduction

Ill-informed agricultural practices in many parts of the world undermine food security and degrade natural resources (Chapin et al., 2011; Foley, 2011). To tackle environmental problems and restore diminishing yields on deteriorated soils, diverse resource-conserving practices have been recommended as a key enabler in achieving economic and environmental goals in the agricultural sector (FAO, 2014). However, large-scale dissemination of sustainable agricultural practices has been challenging among geographically dispersed populations of resource-constrained smallholder famers in lower- and middle-income countries (Sanchez, 2015).

Promotion of novel agricultural technologies has traditionally relied on official institutions (Chapman et al., 2003) and formal information-dissemination networks are still needed for the diffusion of sustainable practices among smallholder farmers in rural regions (Demiryurek, 2010; Isaac, 2012; Sidibé et al., 2014; Wambugu et al., 2011). It is accepted that it is difficult for rural inhabitants to fundamentally change their production systems without institutional support (Kassam et al., 2009).

However, agricultural sustainability cannot be achieved by institutional channels alone. The linear top-down model of information transfer from formal institutions to farmers has long been criticized for its inflexibility, bureaucratic inefficiency, and a failure to meet the changing demands of contemporary agriculture (Ogunlade et al., 2009; Sanyang et al., 2015). Smallholder farmers do not simply obey expert advice from extension agents, nor do they unilaterally adopt new technologies (Rogers, 2003; Schneider et al., 2009). Instead, they are influenced by other farmers’ attitudes and behaviours, which subtly shape their decision-making when they are exposed to new technologies (Benhabib et al., 2011; Ville et al., 2016, 2020). People tend to learn from peers whether it is because of the peers’ superior knowledge or out of conformism (Manski, 2000).

As the importance of the human tendency towards social learning has become widely recognized in studies of agricultural sustainability, an increasing number of researchers have analysed the diffusion of agricultural practices from a social network perspective (Arora, 2012; Isaac, 2012; Matous et al., 2013; Matuschke & Qaim, 2009; Thuo et al., 2014; Van den Broeck & Dercon, 2011; Wossen et al., 2013).

The aim of this paper is to provide empirical evidence regarding the role of diverse social network structures in smallholders’ adoption of resource-conserving practices. We use ‘whole network’ primary data from one village in North-west Cambodia that allows us to study the role of network structure in more depth and to provide practical implication for policies trying to leverage social learning in the promotion of conservation agriculture in remote villages.

2. Conceptual framework

2.1 Farming villages as networked systems

While ‘networks’ have often been mentioned in the research of agricultural systems in a metaphorical sense (Burbi et al., 2016; Kansanga, 2017; Olde et al., 2016), or have been examined using diverse proxy measures (Small et al., 2015; Wossen et al., 2013), empirical analysis of directly measured webs of social relationships has become increasingly common in the agriculture sector in recent years with increasing accessibility of the tools of social network analysis (SNA). Maertens and Barrett (2012) defined a network as a collection of actors (nodes) connected by a set of relations (ties), through which information, knowledge, or tangible resources flow. In the study of diffusion of agricultural innovations, farmers, households, and institutions are usually regarded as actors in a network, with the most frequently investigated network relation being knowledge and advice sharing between farmers (Rockenbauch & Sakdapolrak, 2017).

The study of social networks is central to tackling challenges related to sustainable development, such as fostering of social learning, enhancing cooperation collective action, and participatory innovation (Bauermeister, 2015; Henry & Vollan, 2014; Kishioka et al., 2017; Lamb et al., 2015). Empirical researchers have used often ‘egocentric’ social networks in the study of conservation practices (e.g. Baird et al., 2016; Bourne et al., 2017; Doewd et al., 2014; Manson et al., 2016; Matous et al., 2013; Moxley & Lang, 2006; Warriner & Moul, 1992; Weyori et al., 2017). Egocentric networks¹ are relatively easy to collect by instruments that can be implemented in conjunction with traditional surveys and sampling methods. Egocentric networks take into account only respondents’ direct connections. No attempt is made to connect the separate egocentric networks of each respondent together and thus the overall structure of social networks in the studied location is not known outside of the first step of the direct connections of each respondent.

In contrast, sociocentric network research is more complicated as it requires data collection and analysis tools that are distinct from traditional reductionist methods, which treat all observations as independent. Sociocentric network research connects answers from all respondents within a predefined network boundary (e.g. a village boundary) to reveal the overall social structure of the group, including connections between network actors over several steps (see Figure 2). Despite its advantages, sociocentric network research of agricultural sustainability has been less common because of more demanding data collection (Arora, 2012; Matous & Todo, 2015; Rockenbauch et al., 2019; Ville et al., 2016).

2.3 Diffusion of innovations through opinion leaders and network brokers

Studies of technology diffusion place a strong emphasis on the role of ‘opinion leaders’, i.e. individuals with disproportionate influence over the adoption decisions of others (Rogers, 2003). Opinion leaders are considered to contribute to the development of shared knowledge pools in local social networks (Bodin et al., 2006). According to the diffusion of innovations theory, rapid large-scale adoption of a new technology or practice happens only after locally influential opinion leaders, from whom many others learn and imitate, decide to adopt this practice (Rogers, 2003).

In sociocentric network studies, opinion leadership is typically defined by high centrality in the networks of advice, imitation, and influence (Feder & Savastano, 2006; Matous & Wang, 2018; Valente & Davis, 1999). Network centrality can be measured in various ways. The most basic measure is degree centrality, which counts the number of direct ties of each node. Sociocentric network data enables the analyst to distinguish indegree and outdegree. Outdegree is the number of network partners whom the respondent named as important, while indegree is the number of other respondents who named the focal individual as their important partner (Gamboa et al., 2010). Indegree is considered to be a reliable measure of opinion leadership because it is aggregated based on views of everyone else in the network and thus reflects the individuals’ potential for influence without a bias of their own perception or their cooperativeness with the interviewer (Hoang et al., 2006; Prell et al., 2009; Rogers, 2003).

Opinion leadership is not defined in network studies by someone’s practices or personal attributes but by having a followership within local networks. In other words, opinion leaders are those who many others report as people that influence them. Note that not all actors with many links are opinion leaders. Seeking information, advice, or influence from many others, as captured by outdegree, is not a sign of opinion leadership. The direction of links matter and we distinguish that.

The network definition of opinion leadership suits well the practical purpose of this study. Defining opinion leaders, for example, as people with progressive practices would not be useful if we want to examine whether locally influential individuals (i.e. those who might influence others whether adopting a certain practice is a good idea) actually adopt the practices that we want them to help disseminate. Whether opinion leadership in social networks is correlated with progressive practices is not given and it is likely context-dependent. In our case, we empirically examine this relationship for farming practices in one Cambodian village.

In this paper, the outdegree measure captures the respondent’s all reported sources of information, including those outside of the network boundary and advisors that could not be uniquely identified within the sample (hence it is labelled ‘total outdegree’ in the presented tables). In contrast, indegree is defined only within the network boundary and cannot include links from actors outside of the network boundary who were not interviewed.

Sociocentric data also allows for the measurement of more sophisticated centrality measures that go beyond the simple number of connections but also reflect connections between network partners or even the structure of entire communities. Having three network partners who do not know each other presents different opportunities than being connected to three friends who are also friends with each other. While the latter case may create an enjoyable and supportive environment, the three links in the first case are considered more likely to access diverse information. This notion is captured by centrality measures proposed by Burt (1992).

According to Burt (1992), innovators are not in the centre of things but on the edge of things. They are not the people with the most connections within their community but people who bridge diverse groups and thus have access to a diversity of novel ideas. In the context of agriculture, transformative adaptors were found to link diverse sources of knowledge that helped them tackle uncertainty and change (Doewd et al., 2014). This is consistent with Burt’s more general theory that people in brokerage network positions, who link otherwise unconnected actors, have a superior control over novel information flowing within the networks (Burt, 1992). The degree to which an actor connects other actors within a network has traditionally been quantified by betweenness centrality, a measure that counts the number of shortest pathways that the actor lies on between all pairs of nodes in a network (Freeman, 1978).

Burt (1992) argued that instrumental links come at a cost. Farmers may need to return favours to everyone who helped them out and arguably they may need to spend time and energy to maintain each of the relationships. From an instrumental viewpoint, maintaining relationships is a waste if the information and resources coming from these relationships are redundant. This notion is captured by Burt’s efficiency index, which is low for individuals whose partners are interconnected, because they are more likely to share similar information (Burt, 1992). Specifically, Burt defines the effective size of an individual’s network as the number of the individual’s partners minus the average number of links each of them has with others. Efficiency is the effective size divided by degree.

Although both betweenness and efficiency have been used to quantify brokerage, there are important differences between the measures. Efficiency focuses only on the actor’s immediate network neighbourhood, betweenness takes into account the entire network by counting the shortest pathways between all pairs of nodes in the network. Thus, the latter measure is especially relevant when network links work as conduits that deliver information or resources over several steps in the network. However, betweenness may become less relevant in very large networks where positions on shortest pathways between actors who are too many steps apart loses practical brokerage relevance.

Possibly the most important difference between the two measures is that betweenness does not consider relationships to be a burden. Creating a new link never decreases the actor’s betweenness and can increase it, if it sets the actor on a new shortest pathway between two other nodes. In contrast, new links never increase actor’s efficiency and decrease it they tap to a ‘redundant’ source of information, i.e. to someone who is already connected to other partners of the actor.

3. Agricultural sustainability in North-western Cambodia

Eighty-five percent of Cambodian population is involved in the agricultural sector (Michigan State University, 2017). Cambodian farmers tend to be socio-economically disadvantaged by a low access to education, health services, infrastructure, and institutional support (Farquharson et al., 2008). Figure 1 below shows the approximate location of the studied village in North-western Cambodia. This region has a typical monsoonal climate with a dry season from November to March and a wet season from May to October (Belfield et al., 2013). In the lowlands of this region, most farmers cultivate one harvest of rice crop every year under rainfed conditions, typically using traditional photoperiod sensitive long duration varieties (such as Phka Rumdoul). However, if supplementary irrigation is available, it is possible to grow two rice crops a year with short-duration varieties (such as Sen Kra Ob) (ACIAR, 2015).

Figure 1. Map of Cambodia with a star depicting an approximate location of the surveyed village (background from United Nations public domain map).

Inappropriate cultivation practices have contributed to the deterioration of soil and water resource in this region (Belfield et al., 2013). As a result, smallholder farmers find it increasingly difficult to make a living on the degraded land and move to cities to seek employment in other sectors (ACIAR, 2015). Promoting soil conservation and water restoration is a high priority in this region. However, there are many barriers to their large-scale adoption, including inadequate availability of information and a lack of access to training on these practices (ACIAR, 2015).

Numerous international non-governmental organizations (NGOs) operate in the agricultural sector in every province in Cambodia, aiming to close the information gap (Michigan State University, 2017). In addition to advisory work, many also provide improved seed varieties, fertilizers, pesticides and microfinance. At the village level, NGOs usually partner with local agricultural cooperatives, which are formal organizations of local farmers that meet on a regular basis to facilitate knowledge-sharing and a dissemination of information from the international NGOs and channelling of their resources and subsidies (ACIAR, 2015).

The cooperative in the studied village receives support from the Provincial Department of Agriculture, Forestry, and Fisheries (PDAFF), an international NGO Voluntary Services Overseas (VSO), and a smaller Cambodian NGO Krom Akphiwat Phum (KAWP, meaning Village Development Group).

PDAFF formally registers and documents the activities of all agricultural cooperatives in Battambang Province and informs the cooperatives about changes and updates in relevant government policies.

KAWP was localized from an Australian international organization called Overseas Service Bureau more than 20 years ago. Based on our interviews, KAWP have 15 permanent staff members who work on a rotation basis in 42 villages of Battambang province, including the studied village. KAWP played a key role in forming the village agricultural cooperative and providing training on general management to the cooperative.

VSO provides technical support, training, and new seed varieties to KAWP which supports the local agricultural cooperative. VSO also supports the cooperative directly. Furthermore, KAWP sends their own community development officers to provide farmers with new inputs and training.

The major functions of the agricultural cooperative are to organize monthly community agricultural information-sharing activities, provide microfinance services, and disseminate information from the supporting organizations. In order to join the agricultural cooperative, inhabitants have to purchase its shares and pay a one-time membership fee.

The main analysis in this study focuses on two highly recommended resource conservation agricultural practices in the region, crop rotation and drip irrigation. Although most of the local farmers have not yet adopted these recommended practices, the techniques are sufficiently diffused to allow statistically reliable analysis of the adopters’ characteristics and their networks (Table 1). In the appendix, we provide an expanded analysis that includes other (not necessarily resource-conserving) technologies to test the characteristics of early adopters of a wide range of agricultural methods in general. Brief descriptions of the two main resource-conserving practices used in the main analysis as follows:

1. Crop rotation is sustainable agriculture practice defined as alternately planting different types of crops on the same land in following seasons (Kasu et al., 2019). This practice ensures that organic matter in the soil is preserved, which improves soil structure and nutrient content, and prevents soil erosion. These improvements in soil health also lead to higher yields in the long term (Bullock, 1992). Long-term evidence shows that crop-rotation increases agricultural sustainability by increasing agroecosystem diversity and resilience to weather extremes (Bowles et al., 2020). In the local context, often in combination with drip irrigation, mungbean, watermelon and rice cucumber (also known as muskmelon) are most popular crops grown in rotation with rice. This practice allows the producers to gain additional income in between rice harvest when their fields would be unused otherwise.

2. Drip irrigation is a water-conservation micro-irrigation system to water the crop at the root zone, decreasing water loss by evaporation and surface run off (Narayanamoorthy, 2010). Drip irrigation allows local farmers to diversify their crops and gain additional income, especially in combination with crop rotation. Drip irrigation local installation costs are around US$92.50/ha to produce cucumber with a gross margin of US$5,690/ha compared to rain-fed rice production which only has a gross margin of US$400/ha. Other crops that can be locally grown with drip irrigation are water melon, egg plants, and chilly. Without drip irrigation, these crops would be watered manually with hoses. The materials for installation of drip irrigation can be reused for five seasons.

   Table 1. Variable definition and descriptive statistics.

   Dependent Variables	Definition	Frequency	%
   Adoption of crop rotation	0 (Non-adopter)	72	60%
   	1 (Adopter)	48	40%
   Adoption of drip irrigation	0 (Non-adopter)	77	64.25
   	1 (Adopter)	43	35.8%
   Independent variables		Mean	Std. Deviation	Minimum	Maximum
   Network variables
   Total outdegree		2.78	1.51	0	10
   Indegree		1.06	3.70	0	39
   Betweenness		22.17	55.92	0	261
   Efficiency Binary		0.93	0.23	0	1
   Non-network variables
   Age	Age of household head (years)	51.33	13.36	26	80
   Education	Years of education for household head	4.88	3.49	0	12
   Farming experience	Years of farming for household head	19.65	13.68	0	55
   Household size	Number of people in household	4.31	1.39	1	9
   Land size	Average operated land (ha)	2.26	3.00	0	21
   Net profit	Estimated household net profit made from farming activities in last crop season (USD)	803.02	1279.21	−1150	7500
   Non-network binary variables		Frequency	%
   Land Ownership	0 (No land)	5	4.2%
   	1 (Own land)	115	95.8%
   Member of agricultural cooperative	0 (No AC member)	106	88.3%
   	1 (AC member)	14	11.7%
   Gender	0 (Female)	22	18.3%
   	1 (Male)	98	81.7%

   Note: N = 120.

4. Methodology

4.1 Data collection

Human Research Ethics Committee (HREC) approval (Project number: 2016/882) was obtained prior to conducting the village interviews. A two-phase data gathering was conducted by the first author with trained interpreters and research assistants between January and February 2018 in one village in Battambang Province, Cambodia. The third author has led a long-term project in this location, which further contributed to the contextual knowledge base for this study.

The first phase included semi-structured interviews with local extension agents, local agriculture scholars, village chief, and 20 village members. The topics of the semi-structured interviews covered attitudes toward different agricultural technologies among the farmers in the village, farmers’ strategies for agricultural information gathering and local activities of the cooperative and the external NGOs. The outcomes of this phase were used to inform a fixed-form interview survey in the next phase. The information from the interviews was also useful to understand the operations of the local community organizations summarized in the background section above and to help us interpret the analytical findings.

In the second phase, we interviewed all self-identified household heads in the village about their household characteristics, social networks, and technology adoption. This process included data gathering of household head characteristics including their age, gender, education, and farming experience, as well as the household attributes, such as the number of household members and membership in the local agricultural cooperative. We asked about the households’ annual net profits from farming activities, which indicate the household income, and land ownership (Zeng et al., 2018) and farm size (Uaiene et al., 2009), which indicated wealth.

Information sharing networks were elicited by asking ‘who do you go to for farming advice?’, followed by questions about the nature of the relationship and why this person is considered a useful source of information. These open-ended answers were recorded in verbatim and inductively classified according to most commonly re-occurring issues. This information illuminates the reasons for centrality of certain individuals in the local social networks and the role of external organizations in shaping the structure of the network.

To examine the adoption of agricultural technologies, the respondents were asked whether they had used any of the selected technologies in last season and the reasons for adoption or non-adoption.

There were 188 farming families present in the village but 24 refused to get interviewed. We managed to talk to 164 families and among them 120 households provided complete and reliable information.

4.2 Data analysis

In the main analysis, we examined the adoption of crop rotation and drip irrigation separately through binary logistic regression. As a robustness test, the appendix further presents analysis of a larger number of technologies, which confirm the general trends reported for the main technologies of interest.

In order to find the net effects of network variables on farmers’ adoption behaviour, we tested the potential confounding effects of the following theoretically relevant personal and household characteristics, which may influence adoption of farming practices, as identified in previous research: age (Rogers, 2003), gender (Farnworth et al., 2015; Haug, 1999; Pircher et al., 2012), farming experience (Kebede, 1990; Nhemachena & Hassan, 2008), education (Haug, 1999), wealth proxied by land ownership (Belay & Abebaw, 2004; Hoang et al., 2006; Knowler & Bradshaw, 2007), and household size (Yirga, 2007). However, most of these characteristics were found to be insignificant in the present case. In the main text of the paper, we present parsimonious models, which control only for characteristics that were empirically most relevant. The appendix further reports correlation between all elicited characteristics and the outcomes of interest.

Network centrality measures were computed in UCINET, a social network analysis software (Borgatti et al., 2002), and regression modelling was conducted in SPSS (IBM Corp., 2017).

The statistical results were interpreted and are discussed in combination with qualitative information obtained from the semi-structured interviews and the open-ended survey questions that were recorded in verbatim and helped us to understand the informants’ perceptions of each relationship and their stated reasons for preferring certain popular agricultural advice providers.

5. Results

5.1 Descriptive results

Table 1 below presents descriptive statistics of the variables and the sample. The rate of the adoption of crop rotation and drip irrigation is similar, at 40% and 35.8%, respectively. With respect to socio-economic and socio-demographic characteristics, the average net profit from farming activities per annum per household was approximately US$800. This is well below Cambodia's average income per capita of US$1376 (CEIC Data, 2019). Typical interviewed household heads were men in their early fifties with five years of formal schooling and twenty years of farming experience. Only 18.3% of the self-declared household heads were women, which in the local context typically meant that their husbands had passed away. Women did not have significantly different practices and networks from men (see the correlation table in the appendix). Although 95.8% of households had their own land, the average area was only around 2 hectares. Fourteen households belonged to the local agricultural cooperative.

5.2 Central actors in the village network

An average respondent asked and was asked by one other person from the sample for agricultural advice (Table 1, mean indegree). The difference between average outdegree (2.8) and average indegree (1.1) is 1.7 and means that more than half of reported sources of advice (1.7 out of 2.8) were outside of the network boundary (or could not be identified among the interviewed farmers).

Farmers’ popularity as advisors was unevenly distributed. See Figure 2 of the agricultural information sharing network in the village. Households are represented by circles in the figure, which are connected by advice-sharing links. By traditional measures, Farmer 071 stands out as an opinion leader, i.e. a popular advisor with the highest indegree. Thirty-nine other farmers in the sample see this farmer as an important source of agricultural information. The most common stated reason for mentioning Farmer 071 as an important contact on agricultural matters was his connections to external NGOs that provide him with new technologies and inputs.

Figure 2. Village agricultural advice-sharing network. Arrows point from respondents to their influential advisors. Opinion leaders, who have influence over many peers, display a high number of incoming arrows (e.g. #071). Node size reflects the farmers’ betweenness, i.e. the degree to which each farmer is located in between different groups in the village network.

The respondents seemed to take the fact that Farmer 071 has been consistently chosen by external organizations as the ‘model farmer’ or ‘lead farmer’ for new inputs, and that he gets invited to meetings of international NGOs and the local government, as a proof that he must be a successful farmer. Demonstrating new inputs on someone’s farm and labelling him as a model farmer to be emulated seems to raise and reinforce the status of the individual in a community (even if the farmer’s other practices are not necessarily novel).

The size of nodes in the diagram indicates the value of betweenness centrality: larger nodes have higher betweenness, which means that they lay on many shortest paths in the community networks. Farmer 103 has the highest betweenness, although he has only five links. Farmer 103 is directly connected to several central actors who are linked to other parts of the network but are not connected with each other (084, 102, 129) and thus plays an important role as connecter of different groups within the village. In contrast, numerous connections of Farmer 071 are either ‘dead end’ to farmers with low centrality and presumably low access to information, or are connected to each other, which suggests that information among them is already shared and redundant. Famer 102 and Farmer 105 have even fewer links but were ranked second and third in terms of betweenness.

While 36 of our informants considered Farmer 071 to be the most successful farmer in the village, no informant characterized Farmers 102, 103, or 105 in that way. Interestingly, even Farmer 103 considered Farmer 071 to be the most successful farmer. Farmer 103 explained the reason why he thinks Farmer 071 is the most successful farmer as follows: ‘He works hard. Now, he is working with NGOs and gets good advice and technical support from them, so his income increased’.

5.3 Estimation results

Tables 2 and 3 below show the logistic regression results for the adoption of crop rotation and drip irrigation, respectively. (Further statistical description of the characteristics of adopters and non-adopters is in the Appendix.) All models show strong positive association of net profits with the adoption of the recommended agricultural technologies. However, for both technologies, membership in the local agricultural cooperative is negatively correlated with adoption. The results for education are mixed: negative for drip irrigation, insignificant for crop rotation, and positive for a wider battery of technologies tested in the appendix. Overall, the role of the length of formal schooling during the household heads’ youth did not have a clear and consistent impact on their farming practices today.

Table 2. Binary logistic regression for the adoption of crop rotation.

Explanatory variable	Outdegree centrality n=120	Indegree centrality n=120	Betweenness centrality n=120	Personal network efficiency n=120
	Coefficient Standard error	Coefficient Standard error	Coefficient Standard error	Coefficient Standard error
Non-network variables
Membership of AC	−1.793** 0.829	−1.995** 0.907	−1.902** 0.838	−1.701** 0.824
Net Profit	0.540** 0.217	0.554** 0.220	0.537** 0.221	0.548*** 0.214
Education	−0.013 0.083	−0.014 0.059	−0.016 0.060	−0.003 0.059
Network variables
Total outdegree	0.083 0.143
Indegree		0.047 0.058
Betweenness			0.008** 0.004
Efficiency				−1.736** 0.987
Log likelihood	146.919	146.713	143.013	143.519
Pseudo R^2	0.115	0.116	0.143	0.139

Note: ***^, **^, *Significant at 1%, 5% and 10% probability level, respectively.

Table 3. Binary logistic regression for the adoption of drip irrigation.

Explanatory variable	Outdegree centrality n=120		Indegree centrality n=120		Betweenness centrality n=120	Personal network efficiency n=120
	Coefficient Standard error		Coefficient Standard error		Coefficient Standard error	Coefficient Standard error
Non-network variables
Membership of AC	−1.585** 0.861		−1.876** 0.953		−1.671** 0.868	−1.544** 0.858
Net Profit	0.701*** 0.235		0.721*** 0.241		0.707*** 0.240	0.703*** 0.233
Education	−0.125** 0.064		−0.131** 0.065		−0.132** 0.065	−0.119** 0.063
Network variables
Total outdegree	0.065 0.149
Indegree			0.071 0.059
Betweenness					0.008** 0.004
Efficiency						−0.512 0.863
Log likelihood	137.487	137.487	136.505	136.505	133.516	137.323
Pseudo R^2	0.147	0.147	0.154	0.154	0.175	0.148

Note: ***^,**^,*Significant at 1%, 5% and 10% probabilitylevel, respectively.

Other household characteristics, including age, farming experience, gender, household size, and land area, which were tested in numerous combinations, did not show any significant correlations with the farmers’ adoption behaviour.

With regard to network variables, outdegree is not significant in any of the presented models and is not significant in any of the tested alternative specifications. Note that outdegree is the most basic measure of the number of actor’s links and often the only available measure of network centrality in commonly administered personal (ego) network surveys used in this field. In other words, people who report access to more information sources are not necessarily earlier adopters of the recommended agricultural technologies.

Indegree, another common measure of network centrality that relies on the count of actors’ direct links is also insignificant for the main technologies in focus (Tables 2 and 3). Both indegree coefficients are positive and of similar magnitude as the associated standard errors. Indegree becomes positively significant (p<0.05) when a wider battery of technologies is considered jointly (see the appendix), which suggests that the reason for insignificance in the separate models might be partially due to a lack of statistical power in this small sample. This weak evidence also suggests that indegree (i.e. the number of other respondents who mentioned the focal individual as their network partner in sociocentric data) may often be a more relevant measure for the study of agricultural technology adoption than outdegree, i.e. the number network partners mentioned by the respondent in personal network surveys.

Betweenness centrality is positively significant in all models. This means that farmers who have connections to different parts of their village network were earlier adopters of the studied technologies. All of the models that include betweenness have the best fit (indicated by Adjusted R², Pseudo R², and log likelihood) in all specifications for all technologies compared to models with other network centrality measures. This means that betweenness in the village network explains the farmers’ agricultural adoption choices better than other network measures.

In contrast, efficiency is negatively significant for the adoption of crop rotation and for the wider battery of technologies. It is insignificant for drip irrigation. Overall, the results provide no evidence that redundant links would decrease the likelihood of adoption.

5.4 Summary of stated reasons for adoption or non-adoption of resource-conserving practices

Here we summarize the most common answers to the open-ended questions about the farmers’ reasons for adopting or non-adopting the recommended practices. The informants were invited to explain their motivation and volunteer as many issues as they liked. Issues related to profitability, affordability, availability of land, and access to information emerged as central concerns. These issues were interrelated, for example the perception of profitability, seemed often influenced by direct access to information from others. We provide an overall summary of the prevalence of these issues in terms of frequency at which they were mentioned.

The reasons to adopt the recommended technologies was apparently not out of concern for the environment. As in previous research (Kasu et al., 2019), the most commonly mentioned reason for the adoption of the two technologies was a hope that it would increase the farmer’s profits (43 out of 48 adopters of crop rotation and 28 out of 42 adopters of drip irrigation mentioned this motive in their explanations). The most commonly perceived source of this conviction was based on information provided by the external NGOs operating in the village and also other farmers. (For example, 20 out of 42 adopters mentioned that they expected drip irrigation to result in increased profits based on the experience of this technology by their relatives or friends.)

The main explanation for not adopting drip irrigation was the initial cost of installing it; 33 informants thought they would not be able to afford it or mentioned reasons related to cost for watering their crops in a conventional way.

The most common reason for not using crop rotation, mentioned by 29 informants, was not knowing about this practice. Although many community members were able to describe the benefits of the technology as taught by the NGOs, twenty-nine respondents still did not know anything about it. This highlights communication gaps between different subgroups of the community, especially considering that one third of the village uses this relatively visible practice. This was also the second most common reason for not using drip irrigation (mentioned by 24 informants). In addition to not knowing about the practice at all, some farmers (9 and 12 for crop rotation and drip irrigation, respectively) knew about the techniques but would not know who to implement them.

The third most common topic in farmers’ explanations of why they did not adopt the recommended practices was availability of land (mentioned by 25 and 16 informants for crop rotation and drip irrigation, respectively). The farmers either thought these practices would require more land than they had, or they did not want to use these new practices on the land on which they had been applying conventional techniques.

Finally, some farmers did not believe switching to these techniques would be more lucrative (9 and 12 for crop rotation and drip irrigation, respectively). Additionally, 6 farmers mentioned that they were convinced by someone else that drip irrigation would not pay off.

Overall, the informants’ direct explanations match the estimation results that farmers who have higher net profits are more likely to adopt the recommended technologies. The farmers’ explanations suggest that the adopters’ higher profits are likely to be both a driver and a consequence of the technology adoption. In particular for drop irrigation, those who learnt about the benefits of this technology from external agents and friends and could afford it, installed it and later reported further increases in net profits. These explanations corroborate also previous results from other contexts (Kasu et al., 2019).

6. Discussion and conclusions

It has been recognized that social networks influence people’s thoughts and actions, including farmers’ decisions to adopt sustainable practices. It is essential to understand these networks to design programmes that can leverage them for good. So far, vast majority of network intervention programmes have relied on identifying locally popular opinion leaders (Feder & Savastano, 2006; Keys et al., 2010; Prell et al., 2009; Valente, 2012; Valente & Davis, 1999; Wyckhuys & O'Neil, 2007). Numerous network studies have confirmed that some farmers have many more links than others and, based on such findings, recommended that extension services work even more closely with these ‘focal farmers’ (Weyori et al., 2017). Our study suggests that this approach may also have limitations in some context.

By adopting sociocentric network approach, the results of this study suggest that it is worthwhile to look beyond farmer’s direct influence and popularity when designing network-based intervention programmes. When we look beyond the first step in the networks, i.e. beyond farmers’ immediate network partners, individuals whose networks connected through several steps to diverse parts of the studied village and who were in between the village cliques were more likely adopters of novel agricultural practices. (Naturally, this relationship is not deterministic. There were some high betweenness individuals who did not adopt the recommended practices but overall the positive relationship between betweenness and likelihood of adoption was statistically significant.) That was despite the fact that these brokers might have quite a low profile in the village. Some of the highest betweenness and most progressive individuals were not recognized to be particularly successful farmers in the village (such as Farmer 102 or Farmer 105). However, it did not matter whether these brokers had many links or just a few. Having many direct followers in a network, the most basic common measure of network influence and opinion leadership, does not imply progressivity in a conservative community. (For example, Farmer 071 was named by many as their important contact and a successful farmer but he did not adopt the recommended technologies. Farmers 102 and 105 were named by fewer peers, although they did adopt the recommended technologies.)

In conservative communities, opinion leaders’ may be popular and influential because they are conservative and this behaviour fits the local social norms and values (Rogers, 2003). Moreover, individuals with strong internal connections may be willing to accept free inputs and resources but may be less keen to follow advice from external institutions (Matous et al., 2013).

In the present case, the gathered information suggested that many of the village members followed the most popular opinion leaders because the opinion leaders received resources (such as new seed varieties and fertilizers) provided by external NGOs, either directly or through the local agricultural cooperative. Even several high betweenness farmers who were using the recommended practices saw Farmer 071, who was not using them, as more successful because of his access to NGOs. (Other reasons why some popular individuals were named included that they were hard workers and always present in the neighbourhood.)

In a circular fashion, the external organizations preferred to work with these individuals because they were the local opinion leaders. Arguably, the external institutions could have instead improved the social position of more marginalized community members by offering free resources to them.

Although the local ‘opinion leaders’ were found not to be acting as role models for adoption of the studied locally-recommended practices because they did not adopt them themselves, they were likely influential in terms of adoption of new inputs and could potentially facilitate access to these inputs as well as information on how to apply them. Most farmers were interested in getting access to new improved input varieties. As diffusion of innovations theory explains (Rogers, 2003), innovations that do not require the adopters to change their habits (such as new inputs that are applied in a similar way as old ones) tend to be much more readily accepted than innovations that require fundamental changes in the adopters’ ways of doing things (such as crop rotation or drip irrigation).

While the positive results for betweenness suggested the importance of tapping into diverse knowledge flows across the village network, the negative significant results for efficiency suggests that redundancy among the farmers’ immediate contacts does not hinder technology adoption. Experiments from other contexts have demonstrated how contact redundancy can in fact stimulate diffusion of a new behaviour (Centola, 2010). Risky decisions may require repeated social reinforcement from multiple peers, which is enabled by redundant network structures. The need to see a new technology working for more than one neighbour has also been mentioned by the respondents.

Such positive reinforcement was unfortunately not observed within the local agricultural cooperative. While membership in the cooperative facilitated access to new seeds and fertilizers, it did not lead to increased adoption of practices that save water and conserve soil. Despite positive expectations on the role of community organizations (Uaiene et al., 2009; Katung & Akankwasa, 2010), and the robust finding that the farmers who adopted crop rotation and drip irrigation had higher net profits, the cooperative members were consistently lower adopters of these two technologies that have been strongly recommended in this region (ACIAR, 2015).

In summary, using cooperatives to reinforce information-sharing networks with other farmers who do not adopt recommended practices does not necessarily help to disseminate these practices. Importantly, external organizations’ delivery of agricultural development programmes by targeting locally popular opinion leaders may not always be the best policy. The main points of this research for policy implications are:

1. Popular opinion leaders in a conservative community, who may be influential in promotion and dissemination of novel inputs, are not necessarily also receptive to novel resource-conserving technologies. (This implication is based on a lack of any statistical relationship between a farmer’s adoption of the recommended technologies and the number of times the farmer has been named by others as a useful source of agricultural information.)

2. Opinion leaders may be influential in local networks because they receive resources from external agencies that others need, not necessarily for providing an example for sustainable practices. (This implication is based on the finding that farmers who were most often mentioned as important contacts in the local social networks were considered important because of their connections to NGOs and research organizations that provided them with free or subsidized new inputs and technologies. This explanation was shared by other respondents as the reason for naming these individuals as important contacts and sources of information. However, as mentioned in point (1) above, these influential individuals were not any more likely to adopt the recommended practices than the rest of the village.)

3. External agencies channelling support only through perceived opinion leaders may be exacerbating inequalities in local communities. (The value of information and resources that the selected farmers receive from external organizations is evident from responses of other farmers who connect to these selected farmers to gain at least second-access to these resources. If new programmes select ‘model farmers’ based on their current access to resources and their prominent position in local social networks, the advantage of the selected farmers and their privileged local social status is further reinforced. External agencies may need to weight the relative importance of promoting equity and poverty reduction by their programmes vis-à-vis the organizational and logistical convenience of collaborating with the most accessible farmers who have an established track record of working with external organizations.)

4. Farmers in brokerage positions who are not necessarily widely influential in their communities but tap into diverse knowledge pools may be more receptive to novel practices and more suitable targets for intervention. (This implication is based on the result that farmers whose links span diverse parts of local social networks, as captured by the network measure of ‘betweenness’, were generally more likely to adopt the recommended practices, even if they did not have a high number of links and were not mentioned by many others as important contacts.)

Acknowledgements

The authors are thankful for the ongoing support from Bob Martin, Van Touch, Kirstan Xing, Ratha Rien, and Sophea Yous. They played a vital role in the project design and data collection. Special thanks also go to our team of research assistants from the University of Battambang.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Financial support from the Australian Centre for International Agricultural Research (ACIAR) (Project CamSID: CSE-2015-044) is gratefully acknowledged.

Notes on contributors

Aaron Junjian Zhang

Aaron Junjian Zhang holds a Bachelor of Engineering Honours (Civil Engineering) from the University of Sydney, and is currently a MPhil student in Engineering for Sustainable Development at the University of Cambridge.

Petr Matous

Dr Petr Matous is a senior lecturer and an associate dean in the Faculty of Engineering at the University of Sydney. He has lead social network research projects in numerous farming communities in Asia and Africa.

Daniel K. Y. Tan

Dr Daniel K. Y. Tan is an associate professor at the Sydney Institute of Agriculture, School of Life and Environmental Sciences, Faculty of Science, The University of Sydney. He holds a PhD in Agronomy from the University of Queensland and he is currently the project leader of the Australian Centre for International Agricultural Research (ACIAR) funded project, Sustainable intensification and diversification in the lowland rice system in Northwest Cambodia (CSE-2015-044).

Notes

1 Note there is a variability in the use of these terms within the literature. Some researchers would not use the term ‘egocentric networks’ but ‘personal networks’ for separate respondent networks that were collected without an intent to be connected into one sociocentric network and would reserve the term ‘egocentric’ for individuals’ network neighbourhoods within a sociocentric network.

Appendix

After presenting the detailed descriptive characteristics of adopters and non-adopters of the two main technologies in focus (right below), in the remainder of the appendix, we show overall results for a broader set of technologies, which we did not present in the main body of the paper because of space limitations.

Comparison of adopters and non-adopters of crop rotation

Table

Variables	Definition	Distribution
Dependent variable		Frequency	%
Adoption of crop rotation	0 (Non-adopter)	72	60%
	1 (Adopter)	48	40%
		Adopters		Non-adopters
Independent variables		Mean	Std. Deviation	Mean	Std. Deviation
Network variables
Total outdegree		2.96	1.87	2.69	1.21
Indegree		1.15	1.57	1.01	4.68
Betweenness		36	71.01	12.56	41.35
Efficiency		0.04	0.31	0.96	0.14
Non-network variables
Age	Age of household head (years)	49.58	12.67	52.17	13.72
Education	Years of education for household head	4.91	3.31	4.76	3.58
Farming experience	Years of farming for household head	20.00	12.69	19.84	14.35
Household size	Number of people in household	4.21	1.35	4.4	1.37
Land size	Average operated land (ha)	2.00	3.30	2.09	2.68
Net profit	Estimated household net profit made from farming activities in last crop season (USD)	1191.00	1672.69	504.56	727.30
Non-network binary variables		Frequency	%	Frequency	%
Land Ownership	0 (No land)	3	6.25%	4	5.56%
	1 (Own land)	45	93.75%	68	94.44%
Member of agricultural cooperative	0 (No AC member)	46	95.83%	60	83.33%
	1 (AC member)	2	4.17%	12	16.67%
Gender	0 (Female)	8	16.67%	16	22.22%
	1 (Male)	40	83.33%	56	77.78%

N=120

Comparison of adopters and non-adopters of drip irrigation

Table

Variables	Definition	Distribution
Dependent variables		Frequency	%
Adoption of drip irrigation	0 (Non-adopter)	77	64.25%
	1 (Adopter)	43	35.8%
		Adopters		Non-adopters
Independent variables		Mean	Std. Deviation	Mean	Std. Deviation
Network variables
Total outdegree		2.93	1.86	2.69	1.28
Indegree		1.26	1.59	0.95	4.47
Betweenness		37.07	68.34	13.84	46.03
Efficiency		0.91	0.26	0.94	0.20
Non-network variables
Age	Age of household head (years)	51.12	12.68	51.44	13.80
Education	Years of education for household head	4.28	0.47	5.22	3.66
Farming experience	Years of farming for household head	20.00	12.69	19.09	13.50
Household size	Number of people in household	4.23	1.23	4.35	1.48
Land size	Average operated land (ha)	2	2.15	2.39	3.39
Net profit	Estimated household net profit made from farming activities in last crop season (USD)	1286.22	1765.21	533.17	796.48
Non-network binary variables		Frequency	%	Frequency	%
Land Ownership	0 (No land)	1	2.33%	4	5.48%
	1 (Own land)	42	97.67%	73	94.52%
Member of agricultural cooperative	0 (No AC member)	41	95.35%	65	84.42%
	1 (AC member)	2	4.65%	12	15.58%
Gender	0 (Female)	8	16.67%	15	19.48%
	1 (Male)	40	83.33%	62	80.52%

N=120

The broader technology adoption trends, alternative model specifications, and robustness checks presented below, all support the general conclusion regarding the association of network brokerage with early technology adoption.

The table below presents a set of technologies that are at various stages of diffusion in the studied village, sorted from the most popular to the least popular. Some of the individual technologies have been adopted (or not-adopted) by only a small number of people which would not provide sufficient statistical power for separate analysis. To analyze overall trends, we compose an aggregate scale of the total number of technologies adopted by each farmer.

Table

Adoption of agricultural technologies
Technologies	Number of households (%)
	Using now	Not using now
Using fertilizer	112 (93.3%)	8 (6.7%)
Crop rotation	48 (40.0%)	72 (60.0%)
Using drip irrigation	42 (35.0%)	78 (65.0%)
Using drum seeder	8 (6.7%)	112 (93.3%)
Aquaculture in the rice field	2 (1.7%)	118 (98.3%)

The table below displays results for OLS regression using the aggregate scale. While we tested all potentially theoretically relevant variables introduced in the text, only those that proved significant are presented in the parsimonious models in the paper. Similar regression results presented here as those reported in the main body of the paper generally confirm our previous findings. The model is tested for multicollinearity through the variance inflation factor (VIF). The results show that VIF for all variables is between 1 and 2, which indicates that multicollinearity is not a problem in this model (O’Brien, 2007).

Table

Ordinary least square regression for the number of adopted practices
Explanatory variables	Number of adopted practices
	Model 1		Model 2		Model 3		Model 4
	Coefficients Standard error	VIF	Coefficients	VIF	Coefficients	VIF	Coefficients	VIF
Non-network variables
Membership of AC	−0.232** 0.252	1.110	−0.183** 0.246	1.028	−0.193** 0.237	1.028	−0.174** 0.243	1.031
Net Profit	0.265*** 0.062	1.072	0.278*** 0.064	1.087	−0.014 0.022	1.073	0.285*** 0.062	1.037
Education	−0.026 0.023	1.049	−0.007 0.023	1.060	0.235*** 0.061	1.058	0.006 0.023	1.063
Network variables
Indegree	0.178** 0.022	1.132
Total outdegree			0.026 0.053	1.059
Betweenness					0.264*** 0.001	1.042
Efficiency							−0.154** 0.345	1.012
R^2	0.135		0.107		0.174		0.161
Adjusted R^2	0.105		0.076		0.145		0.130

***, **, * Significant at 1%, 5% and 10% probability level, respectively. N=120.

The last table of the appendix presents correlations between all variables of interest, including the farmer characteristics that were excluded from the presented parsimonious models because of their insignificance.

Pearson Correlations

Pearson Correlations
		Land Ownership	Area Owned	Member of AC	Gender	Age	Education	Household Size	Farming Experience	Net Profit	Out-Degree	In-Degree	Betweenness	Efficiency	Crop Rotation	Drip Irrigation
Land Ownership	Pearson Correlation	1	.158	.076	.009	.159	–.007	–.044	.117	.058	.108	.037	–.063	.120	–.085	.069
	Sig. (2-tailed)		.086	.411	.922	.083	.939	.633	.203	.530	.242	.686	.494	.193	.355	.455
Area Owned	Pearson Correlation		1	.084	.162	.053	.215*	.168	.116	.279**	.186*	.079	.038	.091	.034	–.060
	Sig. (2-tailed)			.364	.077	.567	.018	.066	.207	.002	.042	.391	.681	.323	.712	.518
Member of AC	Pearson Correlation			1	.105	–.110	.154	.088	–.151	.071	–.015	.297**	.053	.079	–.191*	–.163
	Sig. (2-tailed)				.253	.230	.093	.340	.100	.439	.874	.001	.569	.390	.037	.075
Gender	Pearson Correlation				1	–.188*	.300**	.215*	.056	.225*	.115	.078	.165	–.054	.035	.040
	Sig. (2-tailed)					.040	.001	.019	.545	.013	.210	.399	.072	.556	.703	.667
Age	Pearson Correlation					1	–.324**	–.073	.609**	–.106	–.196*	–.042	–.189*	.024	–.107	–.012
	Sig. (2-tailed)						.000	.426	.000	.248	.032	.646	.038	.793	.245	.899
Education	Pearson Correlation						1	–.051	–.256**	.183*	.094	.176	.072	.088	.008	–.130
	Sig. (2-tailed)							.577	.005	.045	.307	.054	.434	.337	.932	.157
Household Size	Pearson Correlation							1	.103	–.004	–.031	.023	.065	–.070	–.059	–.041
	Sig. (2-tailed)								.263	.965	.739	.806	.482	.450	.522	.657
Farming Experience	Pearson Correlation								1	.049	–.018	.034	–.098	.126	.020	.055
	Sig. (2-tailed)									.597	.844	.715	.285	.171	.831	.551
Net Profit	Pearson Correlation									1	.227*	.145	.194*	.025	.249**	.283**
	Sig. (2-tailed)										.013	.113	.033	.783	.006	.002
Out-Degree	Pearson Correlation										1	.097	.456**	.369**	.100	.077
	Sig. (2-tailed)											.291	.000	.000	.279	.402
In-Degree	Pearson Correlation											1	.404**	.019	.019	.040
	Sig. (2-tailed)												.000	.838	.833	.664
Betweenness	Pearson Correlation												1	.018	.203*	.200*
	Sig. (2-tailed)													.842	.026	.029
Efficiency	Pearson Correlation													1	–.186*	–.076
	Sig. (2-tailed)														.042	.408
Crop Rotation	Pearson Correlation														1	.348**
	Sig. (2-tailed)															.000
Drip Irrigation	Pearson Correlation															1
	Sig. (2-tailed)

*. Correlation is significant at the 0.05 level (2-tailed).

**. Correlation is significant at the 0.01 level (2-tailed).