Estimating the Returns to Mobile Technology in Smallholder Agriculture: A Double Machine Learning Approach

Digital technologies are widely promoted as tools for raising smallholder agricultural productivity in Sub-Saharan Africa, yet existing causal evidence is geographically fragmented and rarely accounts for high-dimensional selection into adoption, leaving the magnitude of productivity returns uncertain. This study estimates the causal effect of mobile phone ownership on both farm production value and maize yield using the Malawi Integrated Household Survey (IHS5, 2019-2020) and a Double Machine Learning (DML) framework. We implement partially linear regression (PLR) and interactive regression models (IRM) with a suite of machine learners to flexibly control for high-dimensional confounders. Our stacked DML estimates indicate that mobile phone ownership raises log farm production value by 10.7 percent and log maize yield by 5.9 percent, with consistent results across learners and IV specifications using a leave-one-out community mobile ownership instrument. Pathway analysis identifies fertilizer use and formal credit access as the dominant mechanisms, with secondary contributions from ICT-based extension and improved seed adoption. Productivity gains are concentrated among youth-headed households and in the Centre and Southern regions, where complementary market infrastructure amplifies the returns to digital connectivity. We contribute new causal evidence that mobile phones function as productive agricultural infrastructure and document that the yield and production channels operate through distinct but reinforcing pathways.