Modelling Technology Adoption in a Social Interactions Framework : Evidence on Improved Cassava Varieties in Colombia.

Abstract

This study analyzes the impact of social interactions on the adoption of a new technology. Based on the basic model of social interactions with non-cooperative choices, we provideempirical evidence of spillovers or neighborhood effects on the individual probability ofadoption of improved cassava varieties in the State of Cauca, Colombia. We use a sample ofsocio-economic characteristics at the household level and a set of covariates at the municipallevel for cassava small-holder farmers to test peer effects on individual choices. We implementa machine learning clustering algorithm that allows us to measure the information channeland the degree of interaction between adopters and non-adopters by using the household’sgeographical coordinates to calculate physical distance using the Global Position System(GPS). The regression results show that the average village adoption rate has a significanteffect on the household’s decision to use modern, improved varieties of cassava seeds. Theresults also suggest that, as the minimum distance in kilometers by road from the non-adopterto an adopter increases, the probability of adoption decreases. Our results are robust to theso-called correlated non-observables identification problem widely recognized in literatureon social interactions models. Finally, this work leads to highlight the role of simultaneity orreflection problem also widely identified in theory. The empirical evidence does not supportthe presence of simultaneity. However, the future analysis requires detailed information atagent’s network level. We also provide suggestive evidence of potential learning effects byconditioning on adopters who are technologies specific-users such as weed control, fertilizers,etc. This is in accord with stylized facts that highlight the importance of knowing moreabout techniques in the process of adoption. The results show no difference in the estimatedcoefficients between the neighborhood effects and the suggestive evidence of learning. Future research needs to build better identification mechanism as part of the analysis onsocial networks, diffusion and learning. This study represents a first step for understandingthe role of peer effects on the improvement of living conditions for rural population viatechnological enhancements.