A predictive model for assigning the sample farms into the three dairy farming systems was built

The purpose of classification of farming systems is to develop strategies and interventions relevant to the various systems which may vary in the types and degrees of severity of constraints, resource bases and enterprise patterns. Blanket recommendations of technologies and improved management practices could be one of the reasons for low adoption of interventions by agricultural systems which are highly diverse in agro-ecological and socio-economic conditions.Dairy farming systems in Ethiopia have been extensively characterized. Since dairy cattle genetic improvement strategies in Ethiopia target the peri-urban/urban system and the major milk sheds for introduction of exotic crossbred dairy cattle, the major distinguishing characteristics of the three highland systems is the herd genetic structure, the urban and peri-urban systems are characterized as cross bred-based systems keeping high grade exotic crossbred cows whereas the rural or traditional system is a local cattle-based system.

The studies cited above have however focused on the systems in and around the big cities and the major milk sheds. On the other hand, studies on regional towns have indicated that the crossbred herd composition in the urban/peri-urban system is less than reported in the above cited studies. Despite the extensive characterization of dairy farming systems, a comprehensive characterization of dairy systems in the highlands across the value chain supported with valid statistical analyses is not available in the literature.Secondly, the topology of farms within systems and the underlying determinants for within-system variation/diversity in herd genetic composition are not known to confirm if farms within systems could be considered as a uniform unit to target development interventions. In this study, we analyzed farms sampled form of two states in the wet highlands of Ethiopia to validate the classification of the smallholder dairy farming systems, establish the genetic structure of the three systems and identify within-system determinants of variations in herdgenetic structure based on a series of statistical analyses taking the small holder system in Ethiopia as a case study.

Questionnaire survey on herd genetic composition, herd size, production resources,and household characteristics based on farmers’ recall were collected from February to September 2015 using structured questionnaires. The data from West Shoa and West Gojam zones were analyzed separately to serve as are plication of the study. The genotypes of animals were classified as local, low grade, medium grade and high grade crossbreds based on the exotic blood level of the animals. Low grades were defined as those having about 25% exotic blood resulting from first cross cows backcrossed to local breed sire, medium grade crosses were first crosses with 50% exotic blood resulting from crossing of local cows and exotic bulls , and high grades were animals with about 75% or above exotic blood resulting from first cross cows crossed with pure exotic sire commonly through AI. The level of exotic blood level for cows supplied by government ranches and research institutes was established from the cow certificate provided by the institute. For animals that were born on the farm, exotic blood levels were estimated from the dam and sire breed type as recalled by the owner and from records of AI service providers.Discriminant function analysis was used to reclassify the 180 farms in each zone into their original rural, peri-urban and urban categories.

The model was composed of discriminant functions based on linear combinations of four predictor variables which represented the genetic compositions of herds. The predictor variables were local, low grade, medium grade and high grade crossbred animals in the herds. The variables that provided the best discrimination between the groups were selected based on their statistical significance. The classifications were cross-validated applying leave-one-out analysis where each farm was classified by the functions derived from all farms other than the farm to be reclassified. Multinomial logistic regressions were conducted to estimate the probability of keeping the various genotypes across the three farming systems. Factors that would determine adoption of the various genotypes by farmers in the three farming systems were identified through generalized linear regression analysis fitting Poisson distribution with log transformation of the number of animals as a dependent variable. All analyses were conducted using SPPS version 20 .

Using discriminant function analyses based on herd genetic structures as predictor variables, 180 farms in each of West Shoa and West Gojam zones were reclassified into rural, peri-urban and urban systems . The predictor variables that contributed significantly to the reclassification were number of local, low grade,medium grade and high grade crossbred animals in the herds in West Shoa,and number of local and high grade crossbred animals in West Gojam. In West Shoa, the classification function coefficients for number of local animals were marginally higher for the rural and peri-urban system than for the urban system, the coefficients for the number of low grade crossbreds was higher for the rural system than for the peri-urban and urban systems, whereas the coefficients for the high grades for the rural system were lower than for the peri-urban and urban systems.