The Effect of Social Networks on Smallholder Farmers’ Decision to Join Farmer-Base Seed Producer Cooperatives (FBSc): The Case of Hararghe, Oromia, Ethiopia

Abstract

The paper presents the results of analysis conducted to empirically establish the role of social network in smallholder farmers’ decision to join seed producer cooperative, in Hararghe, Oromia, Ethiopia. We used a ‘random matching within sample’ technique to generate data on social links and resources shared like information among their links. Average treatment effects were used to limit ‘non-awareness’ or ‘selection bias’ on participation rate. An econometric strategy was employed to isolate endogenous effect from correlated and contextual social network effects and analyzed by Probit model. Rainfall data was obtained from NMA for 1986–2018 years and analyzed by coefficient of variation (CV) and standardized anomalies index (Z). The result CV and Z shows intra- and inter seasonal variability of rainfall and, the severity and frequency of drought have increase which works against a single optimum seed source. Rate of membership to FBSc was 75% for exposed sub-sample against 70% for the full sample. This shows the existence of exposure bias. The size of social network, linkage with relatives and extension agents influence farmers’ decision to join FBSc and thus, the presence of endogenous effects of social networks. Thus, participation decision is more likely to be affected by characters and structure of social network. Access to off-farm income and perception on profit influences farmer participation decision where weather-indexed seed insurance can service as an effective strategy to ensure sustained membership. Therefore, any intervention in promotion of FBSc should consider the characters’ and structure of social network and emphasis on indexed-based seed insurance and climate information services as strategy to promote seed producers cooperatives.

1. Introduction

Seed is a basic crop production units and most vital input of agriculture which is threated by climate change [1] but service as a technology transfer agent indispensable for climate change adaptation [2]. It is a means of promoting productively and resiliency of agriculture. Recent empirical evidence has identified climate change as a significant stressor of seed supply especially for informal source [1,2]. The share of formal seed supplies is also unable to meet the seed demand of majority of smallholder and subsistence farmers’ [3,4]. For instance, for crops such as sorghum and Haricot bean where public and private seed companies have less interest and often left to the informal source. To supply quality seeds amidst such challenge the farming community needs alternative seed source such as FBSc.

In Ethiopia, with the changing climate neither informal nor formal alone meet the seed demand of smallholder farmers’. Promoting FBSc is considered as pro-poor institutional options to fill seed shortages found in Ethiopia. FBSc is specialized cooperative on seed production and marketing [5] and classified as intermediary seed system [5,6]. Policymakers, NGOs and the government have been working for the well-function of FBSc especially in remote and climate ‘hotspot’ areas such as Hararaghe to achieve various development goals. However, despite the supports and promotions made by different organizations smallholder farmers’ participation in FBSc is quiet small as compared to its potential. Beaman et al. [7] argue that lack of information sharing is one of the impediments in technological adoption. Thus, social network can serve as channel for exposure which enhances the adoption of technological and institutional innovation. FBSC is rather institutional innovation though it can promote technological innovation.

Numerous studies in Ethiopia have found that household level characteristics are important factors that affect the participation of smallholder farmers in FBSc [3,8,9]. It is evident that participation in training, field day, and experience sharing significantly influence participation behavior in collective action including seed-producer cooperatives. Farmers’ networks are not only channels for information dissemination and knowledge spillover but also provide spaces for the negotiation of innovations. This suggests that social connections could be important factors in influencing their participation in FBSc.

However, empirical evidence demonstrated that peer effects have contributed in influencing most individual and economic outcomes [10,11], but there is a gap in knowing the social interactions effect on farmers’ participation in FBSc. Empirical evidences suggest that many individual decisions, such as access to credit and income diversification decisions [12], information and resources exchange decision [13,14], and agricultural technology adoption [15,16,17,18] are influenced by the behavior of the social group the individual belongs to. Thus, available evidence largely ignores the role social interactions in farmers’ decision to join collective action. Information and knowledge shared by social network can be driven and determined by the nature of social network structure thereby influence behavioral changes. Therefore, the important role of farmers’ social networks in promoting FBSc needs to be rigorously analyzed empirically in Ethiopian context.

2. Material and Methods

2.1. Selection of the Study Area

The study was conducted in east and west Hararghe zones of Ethiopia. Fedis and Kersa were selected from east Hararghe Zone representing lowland and midland. While Chiro and Oda Bultum from west Hararghe representing lowland and midland respectively (Figure 1). Rainfalls in the two zones are characterized by a bimodal and erratic distribution pattern, which gives rise to two cropping seasons’ spring and summer. The short rainy season which run from March–April (spring) followed by a long rainy season from June–August (summer). The study areas are known by mixed farming system. The major crops grown in the area include cereals, pulses, vegetables, root crops, ‘khat’, and coffee. Crop production is the main sources of livelihood while livestock serves as a source of cash income and insurance against uncertainty. The dominant feature of the crop production system is mainly the ‘khat’ based intercropping with sorghum and Haricot bean. Sorghum is the first crop from cereals while beans are the first among pulses grown in terms of area coverage. The intercropping system makes the farming system complex in Hararghe. This area is characterized with erratic rainfalls resulting in recurrent drought leaving over at least ten of thousands if not hundred thousands of people to rely on relief assistance every year.

2.2. Sampling Method and Measurement Issues

Our selection of data collection method(s) was informed by the debates around the concept and empirical measurement of social network. Numerous studies used different sampling approaches to generate data required for social network analysis. Egocentric inference which was used in this study needs modest data requirements and is easily adapted to large scale survey research [19]. The data generated by this approach are also more comprehensive in terms of nodes and links and can be quantitatively analyzed using standard statistical tests. Usually this approach focuses on the presence or absence of tie but also considers issues relating to characteristics of social ties and information that passes through the ties.

There are different approaches in how to establish which actors constitute an individual’s network. Among the different approaches we employed ‘random matching within sample’. This approach can address both strong and weak network links [20] and thus, this approach can produce network parameters that represent the real network more efficiently. Based on ‘random matching within sample’ technique, we paired individual respondents with six others in the sample: three from the respondents’ village and three from neighboring villages which make up the respondents’ village cluster. In the first place the interviewed households where asked whether they know each of their matched households, conditional on knowing the matched household, we produced the details of their relations between the interviewed household and the matched households. The information includes whether they share information on FBSc, discuss on FBSc, and the households’ knowledge about the matched households’ status regarding membership to FBSc.

Although ‘random matching within sample’ has good performance in capturing important information, the presence of certain structural nodes in the network needs to be carefully managed. Thus, we also allowed our respondents’ to list two other individuals they share experience and produced details of their relations with the expectation that it minimizes the chance of omitting a key network node due to the ‘random matching within sample’. However, if the named household was out of the sample, we automatically dropped such link as we don’t have important information about attributes and characters of the nodel. This made a maximum of eight matches per households. Thus, the paper focuses on exploring network effects under the actual contact availability measure rather than the endogenous potential contact availability measure.

2.3. Model Specification

Numerous literature depends on classical approaches to estimate the determinants of farmers’ participation in FBSc where many scholars used econometric model such as logit [9], probit [8,21] and a Double Hurdle [22] models. Previous studies, like on determinants of smallholder farmers’ participation rate in technology adoption and adaptation, simply computed percentage of participants from the sample. The current analysis of participation in FBSc as an innovation has recognized that exposure is a necessary condition for participation which confronts the application classical models. As pointed out by Diagne and Demont [23] classical models used to estimate adoption rate like participation rates suffer from ‘non-awareness’ bias or ‘selection bias’. Thus, classical models would reach biased and inconsistent estimates even when based on a randomly selected sample.

The assumption of this framework is that farmers’ exposure to quality seed production through FBSc, which is a precondition for participation in FBSc, is not necessarily random in the population. This is due to the reason that farmers may be targeted by innovation promoter for awareness and experiences into these collective action or they may self-select themselves into exposure. In addition, unobserved factors that influence exposure in turn affect farmers’ participation rates. The selection bias makes sample participation rate among the aware farmers systematically overestimate or underestimate the true population participation rates. The non-awareness bias make observed sample participation rate, also underestimates the population potential participation rates, even if the sample is random. These biases are not avoided except the whole population is expose of the innovation or it is disseminated randomly in the population. Thus, selection and non-awareness biases contribute to the coefficients of classical participation models to be inconsistent.

Knowing at least one characteristics of FBSc is used as ‘treatment’ while unaware on the characteristics of FBSc is ‘untreated’ variable. Participation gap is defined as the difference between the populations mean potential participation outcome and observed participation outcome and is known as non-exposure bias [23]. The actual population participation rate is considered as the average treatment effect (ATE), which measures the effect of a ‘treatment’ on a respondent who randomly selected from the population [23]. Similarly, the participation rate among the aware farmers corresponds to the average treatment effect on the treated (ATT) and the potential participation rate among the non-aware-farmers to the average treatment effect on the untreated (ATU).

In analyzing social networks effect, is equivalent to estimating neighborhood or peer effects, since individuals behavior if the result of group behavior within link. But the effect of group behavior on individual s within the group is limited by design. Since ‘random matching within sample’ overcome some of the limitation in identification of endogenous (peer) effects separately from correlated and contextual effects [20,24]. This ensure that households do not choose network members to whom they have strong links and the error term in FBSc participation regression equation is limited by design. The social interactions are also specified in more accurate way since we have full information about respondents matched composition. All are contributing to minimize the limitation in measurement error.

Furthermore, social networks naturally have endogeneity problems. Manski [25] identifies three categories as to why network members behave in a similar fashion: (i) endogenous network effects, (ii) exogenous network effects, and (iii) correlated (non-network) effects. It is the endogenous network effect that is usually referred to as a network effect, i.e., an effect generated by the network as a result of direct interaction between network members. Exogenous network effects are characteristics of social networks that are not a result of the internal dynamics of the network, but rather a result of factors that are at play during the network formation process, i.e., how people get selected into networks. Correlated effects are external to networks, the common behavior of peers in the chosen network due to prior similarities owing to exposure to common environment e.g., environmental factors, institutional environment. All of these can make members of the same networks to choose similar in participation decision to FBSc decisions. However, endogenous network problem is minimized by design.

Mokenon et al. [16] stated that it is possible to limit contextual network effects. Thus, we averaged the values of observed exogenous characteristics of the responds’ in the sample using subsample drawn from respective study districts and sub-districts. The average of age, average education, average of sex, average of livestock in TLU and average of land size in Ha are used as the exogenous characteristics of the interaction. We used district dummies in the model to limit correlated network effect problems.

To examine the effects of social networks in FBSc participation, we assume that farmer’ make binary decisions whether to participate or not to participate in FBSc, depending on his/her characteristics. This decision is determined by comparing the expected utility from the participation, ($U{i}^p)$ and the expected utility from non-participation, ($U{i}^N)$. Thus, a farmer decides to participate in FBSc if the utility difference ($VC{i}^{*})$ is positive, that is, $VC{i}^{*}=U{i}^p-U{i}^N>0$, which implies that the utility the farmer derives from participating in the FBSc outweighs the utility derived from non-participation. However, i’s likelihood of participation also depends on the precision of his/her beliefs on innovation such as in FBSc and the information he/she receives from his/her network. At the same time, $VC{i}^{*}$ is a latent variable, and cannot be directly observed. In this context, what is observed is the actual decision by the farmer to participate in FBSc. Therefore, we specify it as a function of observable farm, household, ecological, and social network characteristics as follows:

$$VC{i}^{*}={\mathrm{Z}}_{\mathrm{i}}\mathsf{\delta }+{Ɗ}_{\mathrm{i}}ɸ+ȵ\mathrm{i}, {\mathrm{VC}}_{\mathrm{i}}=\{\begin{array}{c}1, if VC{i}^{*}>0\hfill \\ 0, if VC{i}^{*}\le 0\hfill \end{array}$$

(1)

where $VC{i}^{*}$ is a binary indicator variable that takes one if a farmer participates in a FBSc, and zero otherwise; Z is a vector of farm, ecological related and household characteristics believed to influence FBSc participation decision. These include age, education, sex, farm size, TLU, access to credit, extension and irrigation, training and profit, climate change perception; Di denotes variables representing social networks; δ is a vector of parameters capturing the direct effects of exogenous observable characteristics in Z while φ measures the endogenous, contextual, and correlated network effects on farmer’s decision to participate in the FBSc; and ηi is an error term, with zero mean and variance σ2. We measure the information available to individual i by first focusing on the major source of information such as friends, relatives, local administrators’, and extension agents. Because the network variable is individual specific and defined within a given zones, we use zone fixed effects to control for zonal determinants to FBSc.

2.4. Analytical Framework

Social network theory can be explained in a different ways to analyze the linkage data. Social interactions develop due to links between actors/individuals, which may arise because of kinship, affection, cognitive relational characteristics or acquaintance between them [26]. Social network has been shown to foster the capacity to buffer, adapt to, and shape change [27] by providing resources needed to manage with external stresses and disorders [20,28], and fostering humans’ ability to initiate social innovations and enhance in collective action. Social interaction service as channel for information flow and resource sharing thereby facilitate innovation dissemination in farming community. Borgatti et al. [19] in developing countries, in the absence of well-functioning of formal information institution social networks plays an important source of information.

Social Network Analysis (SNA) is an approach that offers conceptual and methodological tools with which to visualize, measure, and analyze the relationships and how the patterns of interactions between individuals contribute to explain social life [26]. The SNA examines both the patterns of interaction and the content of social interaction. The pattern and content of social interaction tells us how and what resources flow from one actor to another actor. It focused on the kinds of things that ‘flow’ through the network and the measurable structural properties of the network that affect how they ‘flow’. Thus, link outcomes is driven and influenced by the characteristics and linkages of social actors. SNA as tool used to measure networks structures such as the number of links each actor has and network level measures such as network density [29].

In SNA, there are many measurements of networks [29] which are mainly descriptive in nature giving the similarities, centralities and positional measures. It has been proven to investigate individuals’ decision in FBSc participation since it enables farmers to find interaction with different actors and agents in the network. We expect the individual farmer’s participation decision to be affected by the participation decision of other farmers in their network because farmers anticipate that they will share information and knowledge with others. Farm households seek information and knowledge from small networks of friends, relatives, and neighbors or households with similar socio-cultural backgrounds [30,31]. From this, the formation of social ties is driven and influenced by network structure, nodal attributes and external contextual factors.

Social network based on its relation can be strong or weak ties. Weak ties can be characterized as common exchange of information and knowledge spillover among distant connections [32]. In weak ties it is possible to access new and private information. The strength of tie is affected by the period of tie, emotional strength and mutual services that characterize a relationship. Similarly, Fu et al. [33] and Son and Lin [34] revealed that tie strength is function of the type of relationship, period of familiarity and frequency of contact. These in turn affect the quality of information and knowledge shared and transferred in their link. Thus, we examine whether the strength of ties influences the effect of networks on farmers’ participation to FBSc.

The type of network also matters in explaining its effect on the participation decision to collective action. For instance, geographic proximity in which number of network link out of match within and outside village were used in this study. It is a proxy for the social networks and generates peer effects on technology adoption behavior. In this case we consider network link with public extension agent and local administrators, where farmers’ interaction is less frequent, no invested in their relationship, and are not influenced by one another. However, strong ties are linked with close friends, kinship, or neighbors whose interactions are frequent and affectively charged.

This study employed coefficient of variation (CV) and percentage of departure from the mean (Anomalies) to check the degree of variability of rainfall. According to Solomon et al. [35] (when CV, (CV < 20), (20 < CV < 30) and (CV > 30) is classified as less, moderate, and highly variable respectively. Standardized rainfall anomaly (Z) was used to examine the pattern of rainfall that exhibits dry and wet years over time. This characteristic of the Z has contributed to its popularity for the application of drought monitoring and also makes it possible to a determination of the dry and wet years in the record [36].

3. Results and Discussion

3.1. Actual Rainfall Variability and Its Effect on FBSc

Long-term annual mean rainfall variability ranges from 38%, 33%, 25%, and 22% for Chiro, Kersa, Fedis, and Oda Bultum respectively. This implies Chiro and Kersa are highly variable while Oda Bultum and Fedis are moderately variable (

Table 1. Rainfall variability for study districts for 1986–2018.

Rain	Oda Bultum			Chiro			Kersa			Fedis
(mm)	µ	σ	CV	µ	σ	CV	µ	σ	CV	µ	σ	CV
Spring	347.5	137	39	254.2	153.5	60	277.8	149.2	53	344.3	149.9	43
Summer	397.8	104.1	26	301	152.6	38	336.6	132.4	39	356.9	80.9	22
Annual	1078.3	235.2	22	770	295.3	38	900.5	304	33	989.7	251.6	25

). For all cases, the CV of spring is higher than summer and annual rainfall. Thus, rainfall varies spatially and temporary with intra-seasonally and inter-annually variability. Rainfall amount with a CV greater than 30% is imply the area is vulnerable to drought [35]. This argument is also supported by respondents’ perceptions.

Survey respondents ranked their perceived climate factors based on the frequency of occurrence and level of impact on crop production. Drought and delayed spring rain were ranked as the most important climate factors affecting farmers’ participation in FBSc. The result of Standardized Anomalies Index (Z) which demonstrates the intensity and frequency of drought indicates that for Kersa and Fedis 53% and 62% of the years are negative anomalies while it accounts 43% and 50% for Chiro and Oda Bultum for the year 1986 to 2018 (

Table 2. Standardized rainfall indices with the frequency and percentages of occurrence over the period of 1986 to 2018.

Drought Category	Standard Anomaly Index Value Range (Z)	Frequency of Occurrence (Years)
		Fedis	Kersa	Chiro	Oda Bultum
Extremely drought	−2.0 or less	1	6	0	1
Severe drought	−1.5 to −1.99	3	4	4	1
Moderate drought	−1.0 to −1.49	0	3	4	3
Mild drought	−0.99 to 0	13	13	7	11
Normal	+0.01 to +1.49	13	5	16	16
Sever wet	+1.5 to +1.99	3	1	2	0
Extremely wet	+2.0 or more	0	1	0	1

Source: computed from data 2018/9.

). A substantial numbers of cases where also reported on prolonged drought and delayed rainfall in the past few years by FGD respondents. Indeed, farmers’ perceptions are based on their knowledge to capture the intensity and extent of drought. However, climate records capture average conditions at large spatial scales and diverse topographical environments within the district that often do not depict conditions perceived by farmers [36]. Generally, successive drought and shortening of drought return periods were observed where farmers’ perception supports this argument.

All study sites are located in the long cycle crop growing region of Ethiopia where major crop production is mainly correlated with spring season rainfall. The spring season rainfall is crucial for the farmers as the season determines the preparation, planting activities and choice of seed and seed sources. These problems are often exacerbated by farmers’ inadequate access to very-localized early warning information on climate variability conditions. FGD claim that smallholder farmers’ might plant long-maturing varieties of sorghum in a season projected to have a short rainy season.

FGD participants also claim that climate variability affect seed supply from informal and FBSc because of delays in onset and early cessation of spring rain which may force farmers to miss the optimum sawing time. This has interrupted own farm-saved seed supply, limit the operation of social network, share of FBSc and, local markets. FAO [1] argues that climate change as seed security stressors operating within an agricultural setting are usually exhibited by its effects on crop production, seed supply, and disrupt local market functionality. Similarly, Diriba et al. [37] argued that adequacy of seed supply in Hararghe for various sources were mainly challenged by climate change and drought along with other factors. Thus, the changing climate works against a single optimal seed supply where farmers tended to shift between different seed sources.

3.2. Effect of Social Network on Farmers’ Exposure to FBSc

Farmers’ social network can affect the probability of farmers’ exposure/awareness about FBSc in many ways. Due social interactions farmers share information, advice and knowledge and this indicate what flows through interaction. Farmers who are exposed to information about FBSc through their networks are more likely to have a better understanding about the seed producers’ cooperatives and associated benefits, and hence decide in favour of joining it. This in turn affects participation decision. The exposure variable in this study shows knowledge of respondents’ on the characteristics of FBSc as a new innovation. FGD participants claim that they get information about FBSc from multiple sources such as fellow farmers who can be of relatives’ vs non-relatives, local, and extension agents. Farmers’ interaction with such group indicates network structure. Exposure to FBSc is fundamentally correlated to the structure of social interaction which is pre-condition for behavior changes [14].

Social network as channel for learning and behavior change on innovation determines the nature and quality of information and knowledge shared in their social tie. This network supports exposure and participation by creating enabling environment for individuals to acquire new information, share their past experiences and knowledge with others on FBSc. Learning from peers may support potential members to make informed decisions. The learning can takes place by interacting and/or observing the actions of peers decisions and experiences. Linkage with external sources such as public extension agents and local administrators also viewed as one way of getting information about FBSc as a new innovation.

The result of current analysis (

Table 3. Farmers’ participation and exposure in FBSc (%).

Exposure/Participation	%	N
Exposure (% of sample)	40.29	112
Exposure (% of FBSc)	56.76	105
Participated in FBSc (% of sample)	66.55	185
Participated (% Exposure)	65.18	73

Source: computed data 2018/19.

) reveals awareness/exposure to cooperatives in social network is affected by the characteristics of social network. It is the characteristics of two different farmers that create social closeness based on social contagion theory in which both farmers’ are getting exposed to FBSc and enable them socially learn from each other. The nature and intensity of interaction between the matches varies with linkage based on kinship, friendship, geographic proximity, personal attributes such as education. In addition, it also varies with network structure showing the size of linkage. Most of the interaction reported on network size implying diversity of information available for each farmer households. These characteristics of network and attributes in turn affect the flows of resources that support smallholder farmers’ decision to join FBSc. The flows that support farmers’ decision to join FBSc include information on characters’ of FBSc, discussion and advice on quality seed production and management, and beliefs and resources necessary for expose to cooperatives. Ramirez et al. [11] highlighted that extension services such as seed as technologies introduced to model farmers in a given village with the expectation that other farmers would observe the benefit conditional on farmers’ networks.

Focused group discussants revealed that in all study sites social learning is a powerful means in promoting FBSc as an innovation and is more effective than learning from public extension services. Farmers’ participation in FBSc is a continuous process involving farmers’ handling information from a variety of sources. FGD and key informant discussants revealed that there are different sources of information that influences farmers’ participation decision in FBSc. The major sources are the experiences of fellow farmers who were involved in seed production either through FBSc, FRG, NGOs-based seed multiplication, cluster based-seed multiplication, and other projects. In addition, farmers reported that they sought further information from external link with public extension agents and local administrators. FGD participants’ across all study sites frequently responded that farmers’ usually hit their links for further information regarding the economic viability of FBSc based on other farmers’ experiences making participation decision. Beaman et al. [7] argue that lack of information is one of the impediments to agricultural technology adoption where social interactions can serve can as an important channel through which individuals can access information and learn about and eventually adopt new technologies and other innovations.

Organizational level networks serve as spaces for the negotiation of innovations where actors endeavor to engage others in the implementation of an innovation. For instance, weather-index insurance has been promoted by insurance company in collaboration with farmers’ cooperatives. Cooperatives and farmers organization are a natural focus for any new innovation as delivery channel. In Ethiopia, the adoption of weather-indexed insurance is higher when insurance is target to cooperatives or group-based informal insurance schemes [38].

Chercher Oda Bultum and Afran Kallo unions in collaboration with Oromia insurance company has been working on crop insurance. For example, Misoma Gudina as primary seed producer cooperatives has been promoting weather-indexed seed insurance in Oda Bultum District. Most of FGD participants’ in all study sites argued that crop diversification and intercropping has been used as adaptation to climate change. Similarly, Shimelis and Kühl [39] reported crop diversification is one of adaptation strategies used by smallholder farmers’ in Ethiopia. However, seed production through FBSc is basically monoculture in its practice where weather-indexed insurance services used as adaptation strategy to ensure continuity of seed production under FBSc. These activities also further support climate information services and strengthen the work of adaptation strategies.

In fact, accurate and timely weather data are crucial for successful adoption of weather-indexed crop insurance products. Weather-index insurance product enhances the adoption of adaptation practices. Million et al. [40] reported that in Ethiopia, weather-index insurance allowed farmers adoption of improved seed. The use of climate information services drives the adoption behavior of farmers for climate-smart agricultural practices such as the adoption of water management and multiple cropping practices [41,42].

The FGD confirmed that farmers decided to participate in FBSc because of information and knowledge obtained from friends, relatives, local administrators, and extension agents. Out of the total respondents about 40% were exposed to FBSc. About 57% out of FBSc members are exposed to FBSc by social network (Table 3). Similarly, available empirical studies indicated that social network serve as an important determinant of technology exposure and adoption decision making [15,16,18] and decision regarding participation in farmer-organization such as cooperatives and cluster-based seed production [22].

Table 4. Description and comparisons of the means of explanatory variables between categories.

Variables	Definition and Measurement	Participants		Non-Participants		Diff. (t-stat.)
		Mean	SD	Mean	SD
Relative	Having link with relatives 1 = yes, 0 = no	0.35	0.48	0.29	0.45	7.7 **
Friend	Having link with friend 1 = yes, 0 = no	0.64	0.48	0.46	0.50	6.3 **
Local Admin	Frequency of meetings with local administrator at least once per month, 1 = yes, 0 = no	0.35	0.48	0.32	0.47	0.9
Age	Age of household (years)	45.43	8.27	48.76	8.56	0.5
Avage	Average age farmers’ in sample	46.07	3.53	45.97	3.44	0.2
AveHa	Average value of farm size of farmers in the sample	1.43	0.24	1.42	0.23	1.1
TLHa	Size of farm in Ha	1.46	0.78	1.36	0.69	0.9
TLU	Livestock ownership in TLU	4.02	1.95	3.2	1.62	0.1
AveTLU	Average value of TLU of farmers in the sample	3.76	0.28	3.73	0.26	0.9
Sizenet	Size of social network(no.)	5.08	1.14	4.78	1.27	4.1 **
Irrigation	1 = access to irrigation 0 otherwise	0.16	0.37	0.25	0.43	10.9 ***
EDHH	Education of respondent (years)	1.07	1.97	1.08	1.92	0.9 **
AveHH	Average value of education of farmers in the sample	1.06	0.26	1.06	0.24	0.9
SexHH	1 = male 0 = female	0.86	0.34	0.85	0.36	0.5
Aversex	Average value of sex of farmers in the sample	0.86	0.05	0.86	0.05	0.9
credit	Access to credit 1 = yes 0 = no	0.61	0.49	0.37	0.48	0.6
Climate	Farmers’ perception of climate change 1 = yes 0 = no	0.88	0.32	0.75	0.43	28.9 ***
Training	Received training 1 = yes 0 = no	0.19	0.39	0.20	0.41	0.4
profit	Perceived profit 1 = yes 0 = no	0.31	0.46	0.08	0.28	114.3 ***
Off-farm income	Access to off-farm income 1 = yes, 0 = no	0.51	0.50	0.18	0.38	122.5 ***
Extension	1 = if >4 frequent of contact 0 = otherwise	0.58	0.49	0.12	0.32	190.1 ***
zone	Zone 1 = east 0 = west	0.49	0.50	0.48	0.50	0.2
Nokin	Number of kinship out of match	2.49	1.31	2.35	1.33	0.1
Nolink	number of links out of random matches within the village	2.79	0.52	2.69	0.60	5.08 **
NOlink2	number of links out of random match outside the village	2.26	0.84	2.11	0.88	0.001

Note: ** and *** represent significance at p < 0.05, & p < 0.01, respectively.

presents the variable definition and comparisons of the means of explanatory variables between categories of participants and non-participants.

The average age of household is 45 and 49 years for participants and non-participants of FBSc. The average land holding size is 1.46 and 1.36 Ha for participants and non-participants of FBSc respectively. The average livestock ownership in size is 4.02 and 3.2 TLU. Thus, this shows that most individuals’ characteristics such as age, TLHa, and TLU are statistically insignificant differences between participants and non-participants in FBSc. However, there are differences between participants and non-participants in FBSc in education level of households and the difference is significant at a significant level of 0.05. Education is likely to have a positive influence on participation because well-educated farmers are more likely to possess the skills and networks necessary to initiate and manage an association [43,44].

The mean size of social network is 5.08 and 4.78 for participants and non-participants of FBSc, and the difference is significant at a significant level of 0.05. Network size is positively associated with membership in FBSc showing that the size of network is an essential driver of participation decision behavior towards collective action. When farmers receive the same information about the benefit of FBSc from many farmers, they are more likely to understand and decide to join FBSc than those who receive information from few or one person.

Having links with relatives and friends affect farmers’ participation in FBSc. There are significant differences in having links with relatives (t = 7.7, p < 0.05) and friends (t = 6.3, p < 0.05) between participant and non-participant in FBSc. The stronger the links the higher probability of being membership in FBSc. Songsermsawas et al. [31] found that caste-based network generated peer effects in technology adoption.

Strength of tie is also measured by geographic proximity. Participation in FBSc as an innovation in social networks is influenced by tie strength, which is a reflection of the closeness and frequency of interactions among individuals. The result of network links on inter-village and intra-village also shows an association with farmers’ participation in FBSc. The mean number of links out of the random matches within the village (t = 2.79) and (t = 2.69) for participants and non-participations in FBSc. There is mean differences in intra-village network between participants and non-participants in FBSc and statistically significant at a significant level p < 0.05. This implies that network within the village is stronger than that of outside village. This implies exposure to FBSc is not promoted by having link out-side village as it requires continues interactions and reliable actors from whom the household believes.

FBSc participants found to have a higher proportion of farmers who have frequent contact with extension agent, access to irrigation and off-farm income and the result is statistically significant at p < 0.001. Farmers who perceive climate is changing and perceived high benefit of FBSc than other are participating in FBSc and the result is statistically significant at p < 0.00l. Smallholder farmers’ participation decision is also shaped by the strength of ties and their importance to the information being pursued. Having a link with external source such as public extension agents and local administrators is often viewed as weak tie.

3.3. Effect of Social Network to Farmers’ Participation Rate in FBSc

The result of average treatment effects (ATE) framework model on participation rate in FBSc is presented in

Table 5. Estimated participation rates in FBSc.

Predicted (Treatment Effect)	Coef.	Std. Err
Population participation rate/potential (ATE)	0.6959 ***	0.0467
Participation rate among exposed subsample (ATE1)	0.7481 ***	0.0382
Participation rate among non-exposed subsample (ATE0)	0.6621 ***	0.0647
Classic participation rate—joint exposure and participation/actual (JEA)	0.6359 ***	0.0427
Non-exposure bias (participation gap)	−0.0607	0.0686
Population selection bias (PSB)	0.1121 **	0.0520

*** & ** is significant at p < 0.01 & p < 0.05.

. The participation rates in the full sample are about 70% which are low as compared to participation rate among exposed sub-sample 75%. This indicates full sample participation rate suffer from predisposal bias. As indicated in our analysis not all farmers have exposure to FBSc. Thus, the observed sample participation rate undervalues the true population participation rate.

The participation gap is 6.7% with negatively sign but statistically insignificant imply that there is potential for significantly increasing in participation rates in Hararghe, Ethiopia. The population selection bias was 11.2% and statistically significant at significant level of 0.05. This implies that the probability of participation by a farmer belonging to network of exposed sub-population is more likely to participate in FBSc than to any other farmers randomly selected from general population. The result demonstrated that estimated participation rates conditional on exposure might still suffer from selection bias. The limited availability in type, quantity and high costs of basic seed limit the opportunity of farmers’ involvement in FBSc. In addition, land fragmentation also may contribute inspection problem and reduce quality of seed. Promoters of FBSc such as public extension agents, research Centre and University-projects target only limited geographic areas. Hence, these are among the sources of such a positive selection bias.

3.4. Effect of Social Network on Farmers’ Participation to FBSc

Econometric (probit regression results on determinants of farmers’ participation in FBSc are presented in

Table 6. Social network effect on smallholder farmers’ participation decisions to FBSc.

Variables	Parametric		Classical
	Coef.	Std. Err.	Coef.	Std. Err.
Average value of peers’ characteristics
Avage	−0.0454	0.0894	−0.4394	21.82
AveHa	1.0888	1.3479	−1.758	97.58
AveTLU	0.4483	0.4060	0.3144	8.279
AveHH	−0.7289	0.488	−0.0306	97.873
Avesex	−4.1583	1.884	−56.910	2458.5
HH characteristic
Age	−0.0119	0.0152	−0.0123	0.0158
Sizenet	0.2529 **	0.0914	0.4155 ***	0.1321
TLHa	0.2152	0.1766	0.0814	0.1968
TLU	0.0569	0.0678	0.0556	0.0742
Nokinship	−0.0031	0.0802	−0.0118	0.0990
Nolink	0.1413	0.1823	0.0034	0.2575
NOlink2	−0.2232	0.1480	−0.1759	0.1893
EDHH	0.1037	0.0633	0.1029	0.0734
SexHH	−0.1555	0.2955	−0.4301	0.3662
Local administrators’	0.4861	0.3302	0.4144	0.3335
Relative	0.8665 **	0.3253	0.7191 **	0.3317
Friends	0.3218	0.3169	0.2839	0.3957
Irrigation	−0.2801	0.2423	−0.4568 *	0.2767
	0.3509	0.2229	0.4146 *	0.2295
Training	0.2406	0.2349	0.3844	0.2796
Profit	−0.5143 **	0.2349	−0.8126 ***	0.2775
Frequency Extension	1.281 ***	0.2373	1.035 ***	0.2692
Climate	0.3766	0.2681	0.3655	0.2689
Off-farm	1.357 ***	0.2423	1.481 ***	0.2867
zone	0.0953	0.2358	1.229	54.71
cons	−3.739	3.154	−2.813 **	1.163
Pseudo R2	0.4073	0.5327
LR chi2(25)	143.43	187.60
Log likelihood	−104.3667	−82.2848
Prob > chi2	000	000
Number of obs	277	277

*, ** & *** represent significant at p < 0.1, p < 0.05 & p < 0.01 respectively.

. The chi-square results demonstrate that likelihood ratio statistics are highly significant (p < 0.000), suggesting the model has the power to reliably explain behavior that leads to participation in FBSc. The parametric model shows result of the Probit regressions estimated for the sub-sample of exposed participants only, while the classic model shows result for the full sample of participants in FBSc, including those who are not exposed to FBSc. As indicated in Table 6 non-exposure bias is insignificant. Thus, we present the result of both parametric and classic models, but we discuss only the result of the parametric model.

The study indicates that from all specifications participation decisions are positively and significantly associated with information network. This is in line with the theory and empirical evidence of network tie on technology adoption. In the context of new technological innovation practice a number of studies have found a positive relationship between information network and participation in collective action.

The results of endogenous social network variables such as size of social network, having links with relatives and frequency of contact with extension agents have positive and statistically significant effect on participation decision of farmers. Farmers who have large size of social network are more interested to participate in FBSc. Farmers who are members of FBSc share information and knowledge among themselves; this increases their inclination for further participation in FBSc. Thus, receiving FBSc information from multiple sources indicates large size of social network likely influence exposure to the characteristics of FBSc thereby increase the likelihood of participation decision of smallholder farmers’ to collective action. Moreover, when it comes to collective action, information received from relatives and extension agents are seen as powerful network links which increases the likelihood of participation decision in FBSc. This finding suggests the presence of endogenous effects of social networks on farmers’ participation in FBSc. Endogeneity of social network on collective action has been demonstrated by vast number of literatures [42,43]. Since social network and participation in collective action have reverse causality relationships.

Social network also demonstrate endogenous effect on technology adoption. Empirical evidence suggests that many individual adoption decisions such as in agricultural technology [16,17,18] are positively correlated with the endogenity of social network. Similarly, Negi et al. [18] reported that the effect of social networks on the adoption of technologies varies depending on the size of the network, the complexity of the technology, and heterogeneity in the agro-ecological and socio-economic conditions. Thus, the effects of networks on participation behavior in FBSc could be heterogeneous, depending on the types of network, the technology in question, and the characteristics of the members.

The individual characteristics of the peers (exogenous) (e.g., age, sex, education, livestock and land) in the social networks do not have association with farmers’ likelihood of participating in FBSc. This reveals that farmer’s participation in FBSc decision is not correlated with the exogenous characteristics of his/her network members. These shows there are no exogenous effects of social network on farmers’ participation in FBSc and the result is supported with literature [16,18]. However, Abdula-Rahaman and Abdulai [45] revealed that average sex is significantly influence farmers value chain participation decision. Similarly, Song and Chang [46] found that average education of network members positively influences the frequency of health information seeking. This suggests that the effect of exogenous characters’ of social networks varies depending on the structure of networks and characteristics of innovation under study.

The result of the study reveals that all the average exogenous characteristics of the farmers’ peers were not statistically significant implying absence of contextual effects. The result also reveals that the number of links out of the random matches within and outside the village and location of the study were statistically insignificant. This implies that no evidence of correlated network effects. Songsermsawas et al. [31] reported that geographical closeness is not a good proxy for the social networks, and the caste-based networks produce peer effects.

All household-level characteristics such as gender, age, livestock ownership, land holding size, and education level of household head do not demonstrate associated with the likelihood of participation decision in FBSc. Similar result are found by Abdula-Rahaman and Abdulai [45] who report individual characters’ of household did not influence participation in value chain. Mokenon et al. [47] reported most household characters’ except education and land holding size have no association with adoption of row planting in Ethiopia. Under changing climate the effect of household-level characteristics on farmers’ participation behavior varies substantially depending on nature and characteristics of collective action. Similarly, the exogenous effects of social networks on varietal adoption are not significant for the peers’ individual characteristics, but jointly these characteristics are significant and enhance the endogenous effects of social networks [18].

Social network size: Size of social network is positively and significantly related to farmers’ participation in FBSc which means that household having larger size of social network is more likely to participate in FBSc (Beta = 0.2529, p < 0.05) than others. If household network sizes increase by a unit, the probability of participation in FBSc would increase by 0.89% keeping other variable constant (

Table 7. Marginal effect of Probit.

Variables	Parametric		Classic
	dy/dx	Std. Err.	dy/dx	Std. Err.
AveHa	−0.2725	0.3310	−0.2905	16.12
Avage	0.0322	0.0227	−0.0933	4.632
AveEd	−0.2031	0.1559	−0.0065	20.778
Aversex	−8.344	406.24	−9.404	406.24
AveTLU	0.0589	0.1065	0.0519	1.368
Friend	0.0376	0.0473	0.0476	0.0673
Local administrator	−0.3106	0.0947	0.0855	0.0659
Age	0.0066	0.0043	−0.0026	0.0033
Relative	0.4377 ***	0.1007	0.1600 **	0.074
TLHa	0.0108	0.0479	0.0135	0.0325
TLU	−0.0141	0.0145	0.0092	0.0123
Sizenet	0.0089 **	0.0252	0.0687 ***	0.0204
Irrigation	0.0129	0.0772	−0.0782	0.0478
EDHH	0.0071	0.0154	0.0218	0.0155
Sex HH	−0.0281	0.0815	−0.0680	0.0548
credit	0.0275	0.064	0.0896 *	0.0498
Climate	−0.0023	0.0918	0.0801	0.0600
Training	−0.076	0.0707	0.0623	0.0439
Profit	−0.1526 **	0.0732	−0.1354 ***	0.0445
Off-farm	0.0566	0.0675	0.2579 ***	0.0421
Frequency Extension	0.0616 ***	0.0729	0.1880 ***	0.0486
zone	0.0986	0.0661	0.2415	9.126
Nokinship	0.0004	0.0212	−0.0019	0.0164
Nolink	−0.0108	0.0563	0.0006	0.0426
NOlink2	−0.0482	0.0444	−0.0291	0.0310

Note: dy/dx for factor levels is the discrete change from the base level. *, ** & *** represent significant at p < 0.1, p < 0.05 & p < 0.01 respectively.

). Social network size is an indicator of social resources in which farmers’ participation decision to FBSc varies with size of social network. Household who have more links have diverse sources of information thereby increases the trust on the information. Diverse source of information increases farmer level of awareness about social /institutional innovation such as FBSc which can enhance the participation of farmers. A household who has linkage with many number of actors are socially more integrated than someone who has few linkages [48]. Similarly, the effect of social networks on the adoption of technologies varies depending on the size of the network, complexity of technology and socio-economic condition [18,49]. Mokenen et al. [16], in their study used rigorous econometric techniques and further confirmed that technology adoption is positively influenced by the size of the social network with which information is exchanged. Our findings emphasize that the effect of social networks varies depending on their structure and the characteristics of the innovations considered.

Social network with relatives: Network with relatives is positively and significantly related to farmers’ participation in FBSc which means that farmers who have contact with relatives of their network out of match have a higher likelihood of participating in FBSc (Beta = 0.8665, p < 0.05). Relational ties between family members/relatives and friends often viewed as bonding and strong ties. It suggest that having link with relatives increases the likelihood of participating in FBSc by 43.7% as compared to who have link with non-relatives keeping other variables constant. Homophily based on same socio-economic and demographic characteristics creates strong and bonding ties [48,49] where structural characters of social networks support the flow of information on collective action.

The type of network matters in illustrating its effect on farmers’ participation in agricultural innovation practices such as quality seed production through cooperatives. Our finding shows that network with relatives out of match is viewed as a denser network in which continuous exchange of information and knowledge, and hence, is considered as strong link. According to Molina and Martinez [32] strong tie has the advantage of getting high-quality of information and creating strong social norms as well as sanctions generated in this process. Farmers’ participation in FBSc is not a simple agricultural practice in which information from multiple sources and strong tie reinforces farmers to join FBSc. Similarly, in Ethiopia studies by Todo et al. [15] highlight that adoption of complex technologies requires strong social tie.

Frequency of extension contact: Frequency of extension contact with extension agent is positively and significantly related to farmers’ participation in FBSc this implies that a farmer who have higher frequent of contact to extension agent is more likely to participate in FBSc. The result is statistically significant at significant level of 0.001. Frequency of interaction is a key feature of social networks [50] and is a component of the betweenness centrality measure. If the frequency of extension contact increase by one unit, the probability of farmers’ households’ participation in FBSc would increase by 6.2%, keeping other variables constant (Table 7). Extension agents and local administrators’ are heterophilic actors who spread the information vertically through weak bridging ties. Weak ties are formed by dissimilar groups either within or outside their groups’ members which serve as bridge tie. Extension agent is among the actors from formal institution that promote innovative practices such as farmer-based seed production. Farmers’ will be more exposed to be influenced from those that they believe to be a realm expert and with whom they feel a strong measure of trust. University, research centers and often employ public extension agent and local administrators in seed multiplication. For instance, practical demonstration of new crops and/or varieties for adaptation trial (e.g., FRG, cluster-based community seed multiplication) that involve farmers has been promoted by public extension agents and local administrators’. Official information obtained from extension agents may help to lessen risks, uncertainties, and distorted information and thereby play a key role in increasing cooperative actions [39]. Frequency of contact with extension agent have effect on participation in FBSc in which information from reliable extension agent can be trustworthy and is more likely to be absorbed and implemented by farmers’. Todo et al. [15] demonstrated that adoption of complex agricultural knowledge in low developing countries requires strong external ties and flows of the same information from multiple sources.

Access to off-farm income: Access to off-farm income is positively and significantly related to farmers’ participation in FBSc (Beta = 1.357, p < 0.001). If sample household had the probability of participating in FBSc would increase by 5.7% compared to those household who lack access to off-farm income keeping other variables constant (Table 7). Farmer household who have off-farm income gating the purchasing power of input needed for seed production. Quality seed production through seed-producers cooperatives requires proper supplementary inputs such as basic seed, fertilizer and chemicals for herbicides. Thus, income obtained from off-farm can be re-invested to purchase supplementary input which enhance farmers’ desire to join FBSc. Similarly, recent empirical evidence identified that the financial resources of off-farm income determine farmers’ participation in agricultural cooperatives [42,51] if farmers’ face legal constraints or additional investments related to agricultural production. However, other studies finding stated that off-farm income reduce the likelihood of farmers’ participation to seed production. For example, Rubyogo et al. [52] and Tebeka et al. [8] revealed that household heads who have access off-farm income opportunity might face high opportunity cost of time when obliged to attend training and meetings on seed production techniques supported by the project. This might be true if the opportunity of off-farm income is higher than farm income.

Perceived profit: Farmers who perceive the higher benefit of seed than grain is less willing to participate in FBSc (Beta = −0.514, p < 0.05). If sample household perceive higher profit of seed than grain, the probability of participating in FBSc decrease by 15.3% as compared to those households who don’t perceive higher profit keeping other variables constant. All FBSc members are liable to all costs of climate related production risks and rejections by seed inspectors. In addition the limited the availability in type, amount, and high costs of basic seed minimize profit margin. In climate ‘Hot-Spot’ areas such as Hararghe smallholder farmers’ mainly focuses on risk aversion than opting for profit. As observed by Teferi [53], and Tebeka et al. [8] resource constrained farmers are generally driven by food security objectives and might not respond to profitable opportunities. Thus, weather-indexed seed insurance is serving as coping strategy and helping seed producers’ farmers’ vulnerability who are totally rainfall dependent and thereby ensures the sustainability of seed business. In fact, the success of weather-indexed seed insurance services dependent on effective climate information services. The adoption of weather-index seed insurance also further support climate information services and strengthens the work of adaptation strategies. According to news story of ISSD and Benefit project of Haramaya University, in Ethiopia in the year 2014 immense decline of seed production trends for seed producers’ cooperatives due to El. The use of climate information services drives the adoption behavior of farmers for climate-smart agricultural practices [30,31,32,33,34,35,36,37,38,39,40] (where FBSc is also cited as climate-smart agricultural practices.

4. Conclusions

This paper examined the effect of social networks on smallholder farmers’ participation in FBSc in Hararghe, Oromia, Ethiopia. Social network facilitate farmers’ participation to FBSc as an alternative source of seed for sorghum and Haricot bean crops. From the result of CV and Z analysis, successive drought and shortening of drought return periods and irregular rainfall are common in all study sites. This affect the amount of seed farmers saved for subsequent planting seasons. Climate variability and drought have led to a narrowing of seed choices farmers can make for adaptation to climate change-induced seed insecurity. Thus, it is concluded that the unpredictability and variability of climate change work against a single optimal seed source supply. Therefore, any intervention on seed should promote diverse seed sources as an adaptation to seed insecurity.

To supply quality seeds amidst the challenges of climate change effect on seed supply, the farming community needs alternative seed source (e.g., FBSc) which gives farmers a chance to evolve adaptive pathways that can sustain their seed security. The success of farmers’ membership in joining FBSc as alternatives seed source option is contingent on social network services. Therefore, smallholder farmers’ participation in FBSc may be effective through localized social network and building connection between formal and informal sectors in promotion of cooperatives.

The result of ATE model reveals that participation rate among exposed subsample was higher than that of full sample participation rates; this shows exposure/predisposal bias. Both exposure and influence are closely correlated to the underlying social network structure and as a result network structures can be powerful in determining behavioral change as the content of information that is being spread. Not all exposure is equally impactful on the participation decision, as certain network connections have a stronger social influence than others. Thus, information received from a socially powerful network ties will have a greater likelihood of affecting participation decision than information received from a non-influential tie.

The statistical econometric model results demonstrated the existence of peer effect on farmers’ participation in cooperatives. The result of endogenous social network variables such as size of social network, having links with relatives and frequency of contact with extension agents are positively and significantly affect participation decision behavior of farmers in FBSc. Implying the presence of endogenous effects of social networks on farmers’ participation in FBSc. Our findings suggest that the effect of social networks on participation decision behavior in FBSc varies depending on the structure such as tie strength and composition social network. Therefore, any intervention in promotion of FBSc as seed source options efforts should consider the characters’ of social network.

Access to off-farm income is positively and significantly influences farmers’ participation decision in FBSc. But farmers’ perception on profit was negatively and significantly influences farmers’ participation to FBSc. With changing climate, indexed-based seed insurance can be used as an effective strategy to ensure sustained farmers membership in FBSc and this practice can be scalable. Therefore, the stakeholders promoting FBSc and seed system in climate ‘Hot-Spot’ areas need to adopt this practice. This strategy also further demand effective climate information services to strengthen the work of adaptation strategies. Therefore, the stakeholder working in agriculture and seed system should work on climate information services via expanding the number of stations account for diverse topographical differences at district level.