Heterogeneous Treatment Effects of Agricultural Extension Services in Nigeria: Machine Learning Evidence

This study applies machine learning methods to estimate heterogeneous treatment effects of agricultural extension services on productivity and technology adoption in Nigeria, contributing an original methodological contribution to the evaluation of agricultural extension programmes. Standard average treatment effect estimates of extension impact mask important heterogeneity in who benefits most and least from extension services, limiting the ability of programme designers to optimally target extension resources. Machine learning methods including causal forests, Bayesian Additive Regression Trees, and double machine learning offer powerful tools for estimating individual-level and subgroup treatment effects that conventional econometric approaches cannot readily produce. This study applies these methods to a large agricultural household panel dataset augmented by a randomised controlled trial of extension contact in Benue, Kaduna, and Lagos States, comprising 1,200 households. The RCT randomises extension service delivery among eligible households, providing a clean identification strategy. Machine learning algorithms are applied to estimate conditional average treatment effects as a function of pre-treatment household characteristics. Variable importance analysis identifies the household characteristics that most strongly moderate extension impact. Findings reveal substantial treatment effect heterogeneity, with extension impact on productivity ranging from near zero for the most resource-constrained households to over 40 percent for households with adequate land and capital access. Female farmers show larger extension impact heterogeneity driven by differential initial knowledge gaps. The study contributes original machine learning evaluation methods for agricultural extension and recommends adaptive extension targeting using household characteristic scorecards based on the heterogeneity analysis.

Keywords: heterogeneous treatment effects, agricultural extension, machine learning, causal forests, Nigeria.