Spoiler alert! Spillovers in the context of a video intervention to maintain seed quality among Ugandan potato farmers

Pervasive use of poor quality seed remains an important reason for low yields throughout the developing world. We explore cost-effective ways to increase the quality of the local stock of seed tubers among a sample of Ugandan potato farmers. We do this by providing agricultural extension information on (i) how to select the best seed tubers and (ii) how to properly handle and store seed tubers until the next planting season. The information is conveyed in the form of engaging videos, shown to individual farmers on mobile devices. The relative effectiveness of the information interventions is tested using an individually randomized controlled trial with a 2x2 factorial design. However, such interventions are prone to spillovers, and there are indications that control farmers might have learned about seed selection, storage and handling through their proximity to, or contacts with, farmers in the treatment group. Therefore, we explicitly model spillovers ex post using a randomization-based framework and use both farmers' GPS locations as well as survey data on actual interactions between treatment and control farmers to define the social networks through which information travels. After accounting for spillovers, we find evidence that especially the video containing information on seed selection translated into a higher awareness and adoption of recommended practices, a higher probability of using improved inputs as well as higher consumption.