Bayesian predictive modelling to ascertain factors affecting cattle milk production in Tanzania: Evidence from the national panel surveys 2012 – 2021

Abstract

This study aimed to evaluate and compare Bayesian predictive models to identify and quantify the key household inputs affecting cattle milk production in Tanzania. A sample of 1,266 households with at least one milked cow was extracted from the National Panel Survey datasets, the data were collected in 2012/2013 (wave 3), 2014/2015 (wave 4), and 2020/2021 (wave 5). Two generalized linear and generalized additive mixed models were fitted using the Integrated Nested Laplace Approximation. The fitted models were evaluated by using Root Mean Squared Error (RMSE), Deviance Information Criteria (DIC), and Watanabe Information Criteria (WAIC). The findings indicate that the generalized linear mixed model with a gamma distribution performed best, as it balanced good fit and predictive performance (RMSE=1.47, DIC=5,061.36, and WAIC=5,079.45). Although the generalized additive mixed model with a gamma distribution had a better fit (DIC=5,006.54 and WAIC=5,074.43), it had a higher RMSE (1.89), making it less suitable for prediction. The study identified that household inputs, such as major feeding practices, person-administered tick treatment, and total labour cost significantly affect cattle milk production, as their 95% credible intervals did not include zero. In conclusion, the study provides valuable insights into the relationship between milk production and household inputs while incorporating prior knowledge about the parameters into the analysis and considering mixed effects models. The study's insights into significant household inputs can guide dairy farmers and policymakers in optimising milk production practices by focusing on key variables such as feeding practices, health management, and labour costs.

Keywords

• Mixed effect models
• Integrated nested Laplace approximation
• Milk production