Multilevel Analysis of Determinants of Cattle deaths in Ethiopia

Abstract Background The Ethiopian economy is highly dependent on agriculture. Despite being more subsistence, agricultural production plays an important role in the economy. In Ethiopia, the agricultural sector is a cornerstone of the economic and social life of the people. The sector employs 80–85 percent of the population and contributes 47% to the total GDP. Livestock contributes to people’s livelihoods through numerous channels: income, food, employment, transport, and draft-over, manure, savings and insurance, social status etc. Ethiopia is believed to have the largest livestock population in Africa. Despite of this productive and reproductive performance is accompanied by very low, poor health care, high disease incidence, poor management conditions, and unpredictable climactic conditions causing a significant cause of cattle death. Method We analyzed data from the 2019/2020 agricultural sample survey implemented by the Central Statistics Agency, Ethiopia (CSA). The response variable of this study was the number of cattle deaths in Ethiopia which is the response variable count response. Before proceeding to the analysis of data using the multilevel approach, the heterogeneity of cattle death about regions was checked by intra-class correlation (ICC) indicating that 14.6% of the variance of the number of cattle death is at the grouping level (Region). It indicated that there is heterogeneity of cattle death between Regions that the multilevel count regression model fits better than single-level count regression models. Multilevel analysis was conducted with the expectation that there would be a difference in the number of cattle deaths per household across the region. We use the multilevel count regression model. Comparing those models based on model comparison methods (criteria) among six multilevel count regression models; the multilevel ZINB regression model was found to be the most appropriate model to fit the cattle death per household fits for cattle death per household. If a household loses cattle, the result may be a loss of assets for a considerable period. This holds for many households in rural parts of Ethiopia. Loss of cattle can have serious consequences for a rural household’s livelihood, giving a pivotal role played by livestock in the farming system of Ethiopia. Livestock death is considered to be one of the main factors contributing to poverty. Results Livestock farming types is increasing the expected number of cattle death by a factor of 2.314 when compared to croup farming types. Cattle that are not treated are increasing the expected numbers of cattle deaths by a factor of 1.384 when compared to Cattle that treated. Cattle that are not vaccinated are increased the expected number of cattle deaths by a factor of 1.123 when compared to Cattle that are vaccinated. Feeding area of cattle, such as grazing on holding, common pasture grazing and others have increased the expected number of cattle deaths by the factors of 1.411, 2.307 and 1.502 respectively when compared to in-house/barn. The informal education level of households increased the expected number of cattle deaths by a factor of 1.167. The land size of households 0.10–0.5 hectares and 5.01-10.0 hectares are decreasing the expected number of cattle deaths by 0.829 and 0.651 respectively when compared to households that have less than 0.1 hectare. The estimated coefficient of household size is positive and had a significant effect on cattle death per household. That means the expected number of cattle death increase by a factor of 1.048 for every one-unit increase in the household size, holding another variable constant in the model. For every one unit increased household age the odds of cattle death increase by a factor of 1.004, considering all other variables constant in the model. Conclusion The intra-class correlation (ICC) = 14.6% of the variance of the number of cattle deaths is at the grouping level (Region). It indicated that there is heterogeneity of cattle death between Regions. The Multilevel count regression model fits better than single-level count regression models. Comparing those models based model comparison methods (criteria) among six multilevel count regression models; the multilevel ZINB regression model was found to be the most appropriate model to fit the cattle death per household and from the three multilevel ZINB regression models, the random-intercept model provide the best fits for cattle death per household. In the positive count part of the random-intercept ZINB regression model, the variables like; farming types, cattle feeding area, treatment, vaccination, household land size, age of household, household size and education level were found to have statistically significant effects on cattle death.