The potential influence of social networks on the adoption of breeding strategies

P N Pali, L Zaibet*, S K Mburu, N Ndiwa and H I Rware

International Livestock Research Institute (ILRI),
P.O. Box 30709, Nairobi 00100, Kenya
* Ecole Supérieure d’Agriculture de Mograne, University of Carthage, Tunisia
pnpali@yahoo.com

Abstract

Smallholder livestock farmers face challenges such as lack of appropriate fora and networks which can motivate and influence the adoption of breeding strategies in West Africa. Efforts to ensure participation of livestock owners in breeding programs such as performance recording, the use of village herd stocks for breeding programs have been documented as has the creation of livestock producer organizations to enhance effective participation of farmers in breeding programs. The study was conducted in southern Mali where livestock such as the N’dama cattle are endemic. We explored baseline characteristics from a household and community survey to determine the potential effects of individual, household and network characteristics on the knowledge and use of livestock breeding strategies. We assume that interaction amongst stakeholders results in transfer of knowledge between stakeholders. Results from the stakeholder analysis of the types of stakeholders present at the site level were used in the Probit and Seemingly Unrelated Regression (SUR) models.

Farmers’ organisations were commonly found at the village level. However, despite the presence of these organizations across sites, they do not interact with the technical organisations as often as they should hence a lower incidence of use of livestock breeding technologies across sites. Results from the models confirm the importance of networks as key determinants to improve the adoption of breeding technologies.

Key words: N’dama cattle, potential adoption, seemingly unrelated regression (SUR), social capital, stakeholders

Introduction

Social networks are a structural form of social capital that are constructed through network linkages and practices between individuals and organisations within and between communities (White 2002). Linkages of agricultural producers to social networks provide them with an environment for social learning processes which initiate and accelerate adoption decisions (Bandiera and Rasul 2006; Hogset 2005; Conley and Udry 2010) because information is passed on informally between and among network elements but also within the network a concentration of knowledge and identity resources is embedded (Falk and Kilpatrick 2000). Specific information such as the best ways of applying new, improved knowledge and technologies and judging their usefulness (Kilpatrick 2003) is learnt and this information is linked to the producers’ capacity to change (Monge Hatwich and Halgin 2008:1). Social learning is associated with network benefits such as informal finance (Hogset 2005), facilitation of collective action (Isham 2000, Hogset 2005) which benefits the co-ordination of adoption required to circumvent technological externalities but from an economic standpoint, also affects economic decisions through the reduction in transaction costs of information exchange (Isham 2000). In addition to facilitating information and innovation dissemination, networks also facilitate communication, coordination with in the network and information about character traits of network members (e.g trust) (Tatlonghari et al 2012).

Bandiera and Rasul 2006 identify that literature about social networks and their influences on the adoption of technologies report this after technologies have been adopted however, their research focusses on whether social learning does in fact lead to the initial adoption decisions. While it is widely recognized that information about agricultural innovation diffuses mostly through household level affiliations to networks of relatives and community networks (see Bandiera and Rasul 2006; Isham 2000; Katungi et al 2007; Tatlonghari et al 2012), it is being increasingly recognized that information exchange on a diversity of subjects can be ensured between a wider diversity of actors at village level network in a nonlinear manner (See Hartwich et al 2007). For example, in social network analyses conducted by Fungo et al (2011) at the household level in South Western Uganda, farmers were mostly linked to other farmers and at the village level, local governments and national programs had a larger number of networks to the farmer organizations and extension services and whilst Isham (2000) acknowledges that villages with more extension activity have higher levels of adoption; farmer linkages to a wider diversity of actors (network) such as research, development and the private sector organizations at the village level results in a higher level of adoption (Hartwich et al 2007). Based on this hypothesis, in this article we address the question of baseline stakeholder composition at the village level and whether it affects the current use of breeding strategies at the household level. This is important to assess whether other types of networks can influence adoption but also to track changes in the composition of these stakeholders over time with respect to the information and technology transfer services and benefits provided by these networks to households at the village level.

As part of an initiative to ensure the conservation of endemic ruminant livestock in Mali, an initial assessment to understand the livestock production systems, natural resource management, marketing and policy issues surrounding these production systems was conducted. As a mechanism for community and stakeholder engagement in this initiative, the need to understand the factors including social networks that influence the potential adoption of livestock breeding strategies in three sites based in southern Mali is crucial. This study investigates the potential role of social network systems that influence the adoption of the livestock breeding technologies in Mali. Because social networks are a form of social capital and are an unobservable asset, we have used the existence of social networks at the village level as a proxy of social capital, which is critical to engage community level stakeholders who are associated with the breeding technologies that are being promoted.

Specifically, the three objectives of this study are:

1. To identify the extent to which selected breeding strategies are known and used by livestock owners in Mali
2. To examine the types of interactions that exist between livestock owners and the social networks that provide these breeding services and other services at the village level
3. To identify the influence of social networks and other socio economic factors have on the knowledge and use of livestock breeding technologies

Background, theoretical framework and hypotheses

In this section we provide a background into the importance of cattle in Mali. We specifically focus on endemic ruminant livestock (ERL), with particular reference to the N’dama cattle and the efforts to improve the genetic make-up of that breed. We then develop the theoretical framework from which we derive the hypothesis in three broad categories, effects of individual, household, village characteristics on the potential adoption of breeding strategies.

Endemic ruminant livestock in Mali

Mali is one the largest livestock producing countries of West Africa (DNS 2000) where livestock contributes up to 30% of the agricultural and mining sector GDP and 10 % of National GDP (IER-Mali 2005). The livestock population was estimated at 7 Million cattle and 16 Million sheep/goats in 2005. Due to rare veterinary care, feeding deficiencies and the low genetic potential for milk production of local breeds, milk production in traditionally managed herds is only 0.5-2 litres/cow/day (Debrah et al 1995). The main feed source is still unimproved pasture, but more supplementary alimentation (0.25-5.5 kg/day) is provided, mainly consisting of cotton residues from the oil producing industry and harvest residues (Coulibaly 2002).

Endemic ruminant livestock constitute 70% of the ruminant population and are indigenous to many parts of West Africa. A significant area of West Africa is also highly infested with tsetse flies. Tsetse flies are a vector of trypanosomiasis, which affect both livestock and humans and consequently, livelihoods. Endemic ruminant livestock are known to be resistant to trypanosomiasis. Exploitation of genetic resistance to trypanosomiasis through ERL conservation and their habitat has been identified as an optimal and sustainable option to trypanosomiasis control and increased production. This vector and disease control option is known to foster a balanced ecosystem health because it does not involve the use of chemicals that are destructive to the environment. Also, it circumvents the challenges of resistance of trypanosomes to the available trypanicides, scarcity of drugs and prevents the consumption of valuable foreign exchange from their purchase (Adgyemang et al 1997). By 1985, the Ndama cattle formed the largest population of trypanotolerant breed of cattle, however, the proportion of trypanotolorant cattle to the total cattle is on the decline due to increased numbers of imported cattle. For example, between 1985 and 1998, the proportion of N’dama declined by about three and a half percent from 13.10% in west Africa (Thévenon and Belemsaga 2005). The N’dama cattle, which number about 7 million head in West Africa are one such trypanotolerant and hardy breed and are particularly dominant in Guinea, Guinea Bissau, The Gambia and Sierra Leone. In Mali, the N’dama are numbered at about 11,300 head of cattle.

Genetic improvement of N’dama cattle in Mali

There has been considerable effort placed on the genetic improvement of cattle in Mali that began as early as 1960 at the Centre National de Recherches Zootechniques de Sotuba (Tamboura et al 1982) where crossbreeding programs with European breeds began. The objectives of these efforts have been primarily to improve the tolerance of zebu cattle to trypanosomaiasis and to improve the genetic make-up of the N’dama cattle breed due to its trypanotolerant properties. While cross breeding to improve resistance to trypanosomaiasis between trypanotolerant N’Dama cattle and the trypanosusceptible Zebu cattle have been reported, a recent survey by Corsius (2012) established that crossbreeding in the south of Mali is based on social-economic breeding objectives, market price and multiple-use of cattle rather than animal health. More specifically, the genetic improvement of N’dama cattle for the internal market and eventually the export market began in 1975 (Planchenault and Traoré 1993). At a more regional level, one such initiative was designed and has been implemented since 1994 at the International Trypanotolerance Centre (ITC) in The Gambia to genetically improve the N'Dama cattle breed in order to meet future market demands and changes in production circumstances.

In Mali, there is some level of controlled breeding applied in most flocks primarily through sire selection (Ejlertsen, Poole, and Marshall 2012). Currently, livestock owners of modernized managed herds use crossbred bulls for natural insemination as the current practice. In Mali artificial insemination (AI) is a widely available technology however the use of AI is not as widespread. It has been restricted to "exploratory" purposes mainly by research institutions. The AI technology is used to upgrade indigenous stock and also as a service to a limited number of commercial farmers keeping exotic dairy cattle breeds (Lebbie and Kagwini 1996). For the local Zebu and N’dama cattle however, artificial insemination exists but is complicated by their low heat intensity which hampers recognition of the appropriate moment for insemination (Traore and Bako 1984).

Knowledge about breeding strategies and social networks in Mali

A leading constraint amongst smallholder farmers is the knowledge about how to produce more efficiently (McDermott et al 2010). In West Africa, smallholder farmers face challenges of adopting breeding strategies including lack of appropriate fora and networks through which they can exert their influence. Past efforts to ensure participation of livestock owners in breeding programs include performance recording, the use of village herd stocks for breeding programs and the creation of livestock producer organizations (Fall 2000; Bosso 2006), that function as an institutional link through which other internal and external social networks are formed, these networks can subsequently contribute to community based breed development (Bosso 2006). Producer organisations increase the interest in local breeds (Ramsey et al 2003), but also have the potential to influence decision making processes in locations where breeding schemes are being implemented (Fall 2000). Limited knowledge at the national breeding program and farm levels has been identified as a major constraint to effective genetic improvement of local indigenous breeds. At the program level, knowledge limitations of the genetic potential of the local genetic resources and sustainable utilization methods of these resources has been identified (Bosso 2006).

The potential influence of social networks on technology adoption

Single equations assume that the explanatory variables are exogenous, that there is one-way causality relationship between the dependent variable y and the independent explanatory variables (x’s). However, in the case where y=f(x) but also x=f(y), it is not allowed to use a single-equation model for the description of the relationship between y and x. Instead; it is recommended to use a multi-equation model, which would include equations in which y and x would appear as endogenous variables in one equation, although they might appear as explanatory variables in other equations of the model (Koutsoyiannis 2003). In our case where the dependent variables are qualitative, single equation Probit models assume that random disturbances that affect the adoption of technology and knowledge of technology are not correlated. However, it is possible that the random disturbances that affect the adoption of technology and knowledge of the technology are correlated and need to be tested. Therefore, we estimate jointly both equations using simultaneous Probit.

Empirical framework

A simultaneous system model was used to estimate the influence of the socio-economic, farm level and the social network indicators on the knowledge of breeding technologies and the rate of use or adoption of these technologies. To empirically estimate the simultaneous equations, the seemingly unrelated regression (SUR) methodology is used. A seemingly unrelated regression system comprises several individual equations that are linked by the fact that their disturbances are correlated. There are two main motivations for use of SUR. The first one is to gain efficiency in estimation by combining information on different equations and the second is to impose and or to test restrictions that involve parameters in different equations. Zellner (1962) provided the seminal work in this area, and a thorough treatment is available in Srivastava and Giles (1987). A recent survey can be found in Fiebig (2001). Maddala (1983) showed that simultaneous equation estimation is the best estimator in existence for endogenous and dichotomous explanatory variables in the limited dependent model.

To model the adoption of breeding technologies, we consider the following system of simultaneous equations:

(1) T₁=a₀*T₂+ a₁*X₁ + a₂*K₁ + a₃*SN + e₁;

(2) T₂ =b₀*T₁+ b₁*X₂ + b₂*K₂ + b₃*SN + e₂.

(3) K_i = c₀ + c₁*X₃ + c₂*SN + e₃,

Where:

T₁ is a binary variable for use of technology 1

T₂ is a binary variable for use of technology 2

K_i is a binary variable for the knowledge of the technology i

X_i are socio-economic and farm level characteristics

SN   refers to social network variables

The variables T₁,T₂ and  K are not observed but are latent binary variables.

In the empirical model described above; the variables are defined as follows;

Endogenous variables are (i) Technology 1 (T₁ ) defined in a binary form (1=used strategies to avoid breeding from inferior animals, 0= otherwise);  (ii) Technology 2 (T₂ )  defined in a binary form (1=used controlled mating for breed selection, 0=otherwise) and (iii) Knowledge of technology (K) defined in a binary form (1=has knowledge on livestock improvement, 0= otherwise.). We did not include the variable ‘use of specific males for mating’ due to insufficient data. The exogenous variables are composed of socio-economic, farm level and social network variables (see the section on The hypotheses).

The hypotheses

In the absence of evidence to show the effect of social networks on livestock technology adoption including breeding, we consider adoption studies within social network literature and the factors that influence the adoption of livestock technology adoption at the individual, village and network levels in the next section to provide a hypothesis for the determinants of the adoption of livestock breeding strategies.

In their assessment of the individual determinants of adoption of sun flower seed in Mozambique, Bandiara and Rasul (2003) established results that were congruent with conventional literature about early adopters (Besley and Case 1994, Feder et al 1985, Rogers1995), namely that in networks, literate, older and non-religious farmers were more likely to adopt technologies. But contrary to conventional literature about female headed households and poverty and male tendency to have a larger (family) social network than female headed households (See Tatlonghari et al 2012; Katungi 2008) they found that female heads of households associated within networks were more likely to adopt technologies.

Further exploration of livestock technology adoption literature which may not necessarily be associated with social networks shows that other household level characteristics influence the adoption of livestock management characteristics. Family size was found to positively influence adoption of pasture conservation technologies by Rezvanfar and Arabi (2009) but also adoption of livestock technologies increased with older, more educated farmers, who were trained in animal health and who had access to off farm employment (Musaba 2009, Rezvanfar and Arabi 2009). It is assumed that off-farm income provides supplemental income to finance technology expenditures purchase of salt block, urea, mineral lick, hay and small tools for dehorning and castration. On the other hand there was an inverse relationship between adoption of livestock technologies and distance from the extension office which implied that farmers located further away from the extension offices are less likely to adopt improved livestock technologies. Chebil et al (2009) found that age and education negatively affect technology adoption leading us to the hypothesis of a non-directional influence of these variables on the adoption of technologies.

Based on the theoretical framework, we expect a number of variables to influence the dependant variable as seen in Table 1.

An individual’s decision to adopt a new technology is influenced by the characteristics of the networks that they belong to. Networks are defined here as a set of linked organizations and groups of individuals working together around areas of common interests. The size of the network for example has been found to be inversely related to adoption (Bandiera and Rasul 2003) while the types of networks including informal (Hamre, 2008; Crona and Bodin, 2006) and family networks (Bandiera and Rasul 2003) positively influence adoption. Homogeneity or heterogeneity of the network members affects adoption in different ways. Homogeneity within groups may lead to faster knowledge transfers, but as Crona and Bodin (2006) note if a group is too homogeneous, knowledge may not spread, due to inaccessibility of the homogenous group by outside sources. Matuschke (2008) counter argues that learning takes place along geographical lines, and asserts that this assumption does not apply due to socio - cultural stratification of villages which may be heterogeneous with respect to resource flows, trust, norms, or other institutional factors.

In one of limited studies that employ network characteristics in an estimation framework, Hartwich et al (2007) aimed to compare how different knowledge management schemes influence innovation behavior of smallholder farmers in Bolivia by comparing both top-down approach and bottom-up approaches that promote innovation via a network of technology providers, farmers, and private sector agents. They found that farmers who participated in network-related extension schemes had higher adoption rates of modern technologies than did farmers who participated in more traditional extension systems. Yet farmers’ networks were defined somewhat widely to include other farmers, researchers, extension agents, nongovernmental organizations, input buyers, and transporters. Such a wide definition makes it difficult to interpret estimation results and to pin down the actual impact of each network agent on adoption.

Methodology

Site selection and description of sites

Three sites were selected in Mali (Sagabary, Madina Diassa and Manankoro). The primary focus of site selection was the endemicity and breed purity of the ERL specifically N’dama cattle, Djallonke sheep and the West African Dwarf goats. These livestock breeds were identified as those that demonstrate adaptation to a wide variety of eco systems due to their hardiness and disease resistance and whose conservation would have beneficial impacts and replicability for their conservation. In addition, these sites were considered as centres of diversity and geographical distribution of pure populations of endemic livestock, and sub-regional biodiversity hotspots for native flora and fauna. Other selection criteria included presence of diversified production systems, state of the natural environment, degree of threats on the ecosystems, level of tsetse challenge, and priority given to trans boundary sites. A range of agro-climatic, market access and other variables were used to characterise the sites (Table 2).

Sampling and data collection

Household and community surveys were conducted to collect cross sectional data. Participatory rural appraisal (PRA) data were collected at the community level to characterize the context, and community perspectives of the current situation with respect to livestock production including information about stakeholders involved in livestock production. Different sampling structures were used for each survey. Within each of the three sites, PRA data were collected during 3 workshops using a focus group discussion interview schedule. Each site comprised three villages each with different sizes. One PRA was conducted for each village size category (Table 3). Small villages had 40 households or less, medium villages had between forty and 100 households and large households with over 100 households. The workshops were attended by men, women, livestock and non-livestock keepers. During the PRAs’, a stakeholder analysis was conducted to map out the actors involved in the value chain including stakeholders involved in breeding activities. This baseline study was conducted with the aim of understanding and keeping track of changes in interaction among actors, before, during and after the project interventions and to understand the relationship between project interventions, stakeholders interaction and the adoption of livestock management strategies including breeding strategies. Data from the stakeholder analyses were analyzed to generate network maps which were aggregated at the village level within the three sites.

During the household baseline surveys, quantitative data were collected on livelihood strategies, around livestock production systems in ten villages in each site using an interview schedule. For both surveys, the sites were pre-determined based on selection criteria outlined in section  REF _Ref349443842 \r \h 3.1, while the villages and households were randomly selected from lists that were provided. Between 5-10 households were interviewed by enumerators in each village depending on the number of households in each village. For each household, interviews were held with heads of households, and in their absence, adults within these households.

Data analysis

Descriptive statistics

The unit of analysis for the study was the household for the household modelling and the village for the social networks. The statistical package for social sciences SPSS (17) was used for data entry however, these data were exported to STATA to conduct the exploratory data analysis (EDA) before the final analysis was conducted. Several analyses were conducted including descriptive statistics such as means and percentages used to estimate, understand and compare various household characteristics including household size, land size, household ownership and the social network characteristics at the site level. Overall means and standard deviations were calculated to give an indication of centrality measures.

At the village level, the social network analysis was conducted to give an indication of the interaction patterns between stakeholders particularly with breeding stakeholders. Measures of centrality, such as degree, closeness and betweenness centrality were computed at village levels. In social network analysis, measures of centrality make implicit assumptions about the flow of traffic through the network – in our case the flow of knowledge as a result of interaction between different stakeholders at different levels including the site and village levels. More specifically we use this to explain the effect of multi-stakeholder interaction on the adoption of technologies as a result of knowledge that flows between these networks. We assume that knowledge flows through the networks as a result of interactions between stakeholders. Degree centrality as a measure of the number of direct connections a node has (Hannemann  and Riddle 2010). It is a measure of actor activity as determined by the number direct ties the village has with the other stakeholders in the network. Villages with many direct links are assumed to have better opportunities and high influence than those with less links. The betweeness centrality is measure of links among actor clusters in a network (Newman 2005). In this study, the betweeness centrality was measured as the number of villages that a single actor was found in, for each site. According to Fungo et al (2011) a node with high betweenness has great influence over what flows and what does not in the network. According to (Newman 2005; Hannemann and Riddle 2010), closeness centrality defines the pattern of partners’ direct and indirect ties that allows them to access all the stakeholders in the network more quickly compared to other members. It is a measure of how fast an actor can make direct and indirect contact with the other actors. Actors with high closeness centrality can pass or access information faster and in most cases are less dependent; hence closeness centrality is inversely related with the ease of accessing a given partner in the network (Hannemann and Riddle 2010).

Results and discussion

Sample characteristics

Two hundred and ninety four households were interviewed. More households in the sample owned cattle than small ruminants with an average of 20 head of cattle being owned per household. However, for all livestock, the highest mode of ownership was between one and five heads and for poultry between 11 and 20 (Figure 1). Seventy percent of the households owned goats compared to sheep (60%) but no household owned more than 80 heads of small ruminants. By contrast, the importance of cattle and poultry is shown by the ownership of these animals across all modes, however, less than ten percent of the households owned more than 80 heads of cattle and poultry.

All heads of households that were sampled in all the sites were males. Overall 82% of the heads had never been to school and only 8% had completed primary education (Table 5).

In our hypothesis, age was found to have a non-directional effect; however, all other household characteristics were as hypothesized. For example, respondents from Sagabary had the least knowledge about livestock improvement but the household heads were younger and the family and land sizes were smaller, as was the number of animals and incomes compared to other sites (Table 4). Madina Diassa reportedly showed the highest family and land sizes, households’ incomes and average number of animals owned per household across sites, but also had higher statistics of knowledge about breed improvement and use of the breeding strategies. The discrepancy between knowledge of livestock improvement and use (of controlled mating and the use of a strategy to avoid breeding with other animals) was lower for Madina Diassa than Manankoro. In Manankoro a higher percentage of respondents reportedly had a higher knowledge of livestock improvement, but the actual use was low, for example, the use of improved breeding strategies was the lowest in Manakoro compared to the other two sites.

At the village level, we considered the social capital characteristics including network characteristics. Household membership to a group did not show much variation across sites with between 75 and 80% of the respondents belonging to groups. With regard to the measures of centrality, Manankoro had a relatively lower measure of closeness and Sagabary had the highest (0.816). The site with the highest closeness centrality score is the site which goes through the fewest number of ties to reach everyone else in the network. Sagabary as a site has the shortest connections to stakeholders compared to the other sites; we assume that this would translate to higher levels of knowledge; however the results show the contrary. In Sagabary, the closeness centrality was high but the level of knowledge was the lowest compared to other sites; this could be associated with the lower number of livestock owned by respondents from this site and the younger age of respondents who could be engaged in off farm employment. Hence, lack of demand for information about livestock improvement in this site.

Madina Diassa had the highest betweeness centrality described as the indirect relationship between direct nodes or the number of villages that are linked to stakeholders. In this site, higher use of breeding strategies to control in breeding such as the use of controlled mating or the use of another strategy to control mating with the other animals were reported. This could be associated with the larger number of the stakeholders that are present in all villages compared to the other sites. We explore which of these stakeholders are present at the village level and their functions in the sections on Site level social networks.

Site level social networks

Stakeholder types across sites

A breakdown of stakeholders across sites shows that (Figure 2) there were about 15 stakeholders that provided technical support mainly the veterinary officer, Centre Malien de Développment du textile; (CMDT) and Action against hunger (Action Contre la Faim; (ACF)). Despite the presence of a veterinary officer in all nine villages, only one village based in Madina Diassa had a veterinary clinic. There were also almost as many farmer groups or associations in each district with five in each of site which comprised women’s groups, livestock owners associations and farmers groups or associations. Only one credit organization or NRM organization was found in Manankoro and Madina Diassa but none was found in Sagabary. Sagabary also had the fewest number of stakeholder types (6) than all other sites which had 7 types of stakeholders, but input dealers were found in Sagabary only.

Stakeholder composition for site level social networks

There was representation of at least one type of institution from eight institutional categories but within each site, no village had a representation of all eight institutional categories. The eight institutional categories were traders, input dealers, technical support and extension, livestock farmers or farmers associations, policy, cattle herders, natural resource management and credit organisations (Figures 3- 5). In these three maps, the square shaped nodes represent villages in each site where the community surveys (PRA) were conducted. Their size represents a measure of the number of direct connections each village has with different types of institutions, called the degree centrality. For example, Sienré and Sanankouroun (Figure 3) in Sagabary had the same number of linkages (8) to different types of institutions; hence they are the same size. The circle shaped nodes represent the names and types of organizations present in each village. The size of each dot represents the indirect relationship between direct nodes or the number of villages that are linked to the institutional category (betweeness centrality). The smallest sized dots shows stakeholder representation at only one village in the site.

The veterinary agent reportedly responsible for the eradication of contagious diseases and the cattle herder were the most common actors found in eight and seven of nine villages respectively. The cattle herder associations carry out joint grazing and therefore almost village has such associations.

In all three sites of Sagabary, all villages were linked to between four and five actor organisations. For example, Gadougou1 village in Sagabary, was linked to five types of organizations or actors such as the cattle herders, livestock traders, the CMDT, veterinarian and Action against hunger (ACF)) for technical support, Agro breeders, (a farmers’ association), and the village associations. In the Sagabary site, all villages had two technical support organisations (the veterinarian agent, CMDT). Sienré was linked to more farmers’ organisations such as the women’s and village associations who provided support and advisory services on livestock husbandry and general management- an indication of the potential for the bridging social capital, but these were only associated with one village in the site (Sienré), more so, the village associations were not linked to the stakeholders providing technical support (veterinarian and CMDT) in any of the three villages of Sagabary which could also explain the low level of knowledge and use of livestock improvement technologies.

In Manankoro and Sagabary, there were fewer institutions found across all the three villages including ACF input dealers and the veterinarian agent and the municipal authorities, cattle herders and CMDT.

In Madina Diassa, the municipal authorities, Shepard, the livestock traders and agro breeders (farmer association groups that assist in information access and livestock marketing) and the veterinary agent were present in all three villages. Also in Madina Diassa, uncommon to all other sites there were organisations such as Coordination de associations et ONG fémiminines de Mali CAFO Jiginew which provided credit, a water and forest, natural resource management organisation reportedly protected the natural resources, and a hunters association which conducted village patrols in search for lost and stolen animals.

Noteworthy from all three network maps for all three sites is the sparse network density and the fact that none of the stakeholder was linked to one another which further limits potential for information and knowledge sharing.

Determinants of knowledge and use of livestock improvement practices

Results from the household and community surveys were input into the individual probit and simultaneous models to understand the factors (and the levels at which these factors) that influenced the knowledge and use of livestock improvement strategies. For each Probit model, the Inverse Mills Ratio (IMR) was estimated to test for potential sample selection bias which resulted from a small sample size of sixty two respondents who knew about and used livestock improvement technologies. For all three models, the IMR was not significant leading us to the conclusion that there was no sample selection bias resulting from the small sample (Annex 1). In the results we display both models (Probit and SUR) although SUR method provides more consistent results.

Individual level variables such as age of the household head were significant for the model on strategies to avoid breeding from inferior animals only for both SUR and Probit results. Several strategies were used to avoid breeding from inferior animals and unproductive animals including castration of inferior males and culling or selling inferior males and females. From the descriptive statistics, this practice was used by between 48% and 54% percent of the respondents. The probability of using strategies to avoid breeding from inferior animals decreased with the age of the head; this was probably because of the effort required to conduct castration, culling and other such activities. Chebil et al (2009) also shows that age is negatively associated with technology adoption.

For the household level variables, ownership of assets was found to influence knowledge about and use of livestock improvement strategies. Respondents with more land used controlled breeding less, probably due to the in ability to control cattle herd movement in households with large land sizes. Cattle ownership was found to increase the probability of knowledge about livestock improvement strategies (Table 5). This variable was significant for both the probit and SUR models. On the contrary more sheep owned by the household was negatively associated with the number of respondents reporting knowledge about livestock improvement practices. Households in the sample owned between ten and 11 head of sheep which could explain why less attention was paid to knowledge about improvement of the management of sheep but also a higher focus is being placed on the knowledge about livestock improvement for cattle.

The probability of use of strategies to avoid breeding with inferior animals increased with knowledge about livestock improvement. This is an important finding to recommend the increase of stakeholders at the lower levels of administration to enhance capacities on improved technologies and their use.

Knowledge about livestock improvement was not an important factor required to increase the probability of use of controlled mating probably because this is a specialized form of breeding and requires specialized knowledge. Several control mating methods were reported including mating best males with best females, avoiding mating of close relatives and ensuring that each sire has desired number of progeny. Asset ownership played an important role in the use of this livestock improvement methodology.

Network characteristics were found to have an influence on the knowledge and use of livestock improvement practices in all but one model. The presence of a stakeholder in all villages was directly associated with the number of respondents who knew about livestock improvement and who used control mating as was the rate at which the stakeholders in the network could reach other stakeholders due to more direct (shorter) linkages. This finding confirms the notion that in cattle keeping communities the presence of stakeholders will increase knowledge about livestock production aspects (Falk and Kilpatrick 2000). The most common stakeholders that were found in most villages included the veterinary officers, technical support institutions and herders. Veterinary officers were reportedly responsible for eradication of diseases hence frequent consultation could be sought from these stakeholders.

Network characteristics were different for the model on strategies used to avoid breeding with inferior animals. The probability to use strategies to avoid breeding with inferior animals decreased with the linkage of a stakeholder to all villages in a site on the other hand this probability increased with more indirect linkages to the stakeholders. From the descriptive statistics, about half of the respondents used strategies to avoid breeding with inferior animals, and although at the household level, knowledge about livestock improvement was important to avoid breeding from inferior animals, village level networks were not an important factor in explaining the use of these breeding strategies.