Improving Smallholder Agriculture via Video-Based Group Extension (WP-23-06)
Tushi Baul, Dean Karlan, Kentaro Toyama, and Kathryn Vasilaky
Providing technical advice at scale poses operational challenges, particularly with respect to managing a sufficiently large staff. Technology may help, but risks reducing efficacy given reduced customization and human interaction. The researchers tested a video added onto standard human-provided extension services promoting a climate-smart practice, System Rice Intensification in India. Using frequentist statistical methods, they find large but imprecisely estimated treatment effects: The 95% confidence interval is 10kgs to 500kgs and 717Rps to 9650Rps for output and profits, respectively. However, the researchers’ data are not normally distributed: Specifically, key outcomes have fat tails. A Bayesian hierarchical model finds smaller but more precise treatment effects: analogous 95% intervals from -8kgs to 70kgs and -193Rps to 1380Rps. They also test two messaging sub-treatments designed to address commonly cited constraints to adoption: labor needs and self-efficacy. A frequentist analysis shows no added gains, while the Bayesian shows an added benefit when delivered in tandem.