A Bayesian Assessment of Productivity and Risks to Achieve Target Yields from Improved Chickpea and Mung Bean Varieties Using On-Farm Trials in Afghanistan

Abstract

Legumes are essential to meet nutrition need of growing population in Afghanistan but have low productivity under the farmer practices. With an aim to introduce improved varieties of chickpea (Cicer arietinum) and mung bean (Vigna radiata) along with associated cultivation practices, a number of on-farm demonstration trials were conducted in eight district covering three provinces to assess their performance in comparison with local practices. The data or information generation from on-farm trials is routine but remains unutilized under the commonly used analysis based on frequentist approach. This study makes use of such prior information to widen the statistical inference framework using a Bayesian approach for analyzing any future on-farm trials. The analysis of data in 2012 was carried out under a Bayesian framework, where the datasets generated over 2009-2011 were used to determine prior information on the standard deviation. The a priori distributions for standard deviations were positive value truncated normal distribution for chickpea and long-normal distribution for mung bean data. The statistical analysis evaluated the mean performance of the improved and local production packages and risks associated with them to meet a given target. The improved variety, Madad of chickpea (2.25 – 2.40 t/ha) and Mash 2008 of mung bean (1.22 – 1.24 t/ha) yielded higher than their respective farmer or local package (1.93-1.94 t/ha) for chickpea with probability (0.947 – 0.986), and 0.93-0.94 t/ha for mung bean with probability unity at the two districts in 2012. The stochastic dominance analysis showed that the recommended package with improved variety Madad has 95% probability of achieving a target of 2 t/ha while such a target for local practice has around its one third, when cultivated in Baghlan-e-sannati district. Crossover in the risk curves led to know the region specific risk to meet a given target. Similar trend mung beans were also found. Bayesian approach is recommended in the analysis of on-farm trials data.