Most previous studies evaluating agricultural technology adoption focus on estimating homogeneous average treatment effects across technology adopters. Understanding the heterogeneous effects and drivers of impact heterogeneity should enable interventions to be better targeted to maximise benefits. We apply machine learning using data from a randomised controlled trial to estimate the heterogeneous treatment effect of fruit fly IPM practices (i.e., parasitoids, orchard sanitation, use of food bait, biopesticides, male annihilation technique, and their combinations) in Central Kenya. Results suggest significant heterogeneity in the effect of IPM practices conditioned on household characteristics. The most important covariates explaining differences in treatment effects are wealth, distance to the mango fruit market, age of the household head, labour and experience in mango farming. Results further indicate that those with fewer mango trees benefit more from most IPM practices. Additional analysis across other covariates shows mixed results but generally suggests significant differences between households benefiting the most and those benefiting the least from IPM practices.

中文翻译：

---

一刀切：肯尼亚芒果果蝇害虫综合防治实践的异质经济影响——一种机器学习方法

之前大多数评估农业技术采用的研究都侧重于估计技术采用者的同质平均处理效果。了解异质效应和影响异质性的驱动因素应该能够使干预措施更有针对性，以实现利益最大化。我们利用随机对照试验的数据应用机器学习来估计肯尼亚中部果蝇 IPM 实践（即寄生蜂、果园卫生、使用食物诱饵、生物农药、雄性消灭技术及其组合）的异质治疗效果。结果表明，因家庭特征而异的 IPM 实践效果存在显着异质性。解释处理效果差异的最重要的协变量是财富、距芒果水果市场的距离、户主的年龄、劳动力和芒果种植经验。结果进一步表明，芒果树较少的人从大多数 IPM 实践中受益更多。对其他协变量的进一步分析显示了不同的结果，但总体表明从 IPM 实践中受益最多的家庭和受益最少的家庭之间存在显着差异。