To empirically estimate the different adoption steps and
reveal the impact of different types of program interventions,
we use cross-sectional data from 412 smallholder
farmers in two territories in South-Kivu. We use different
estimation techniques, including univariate probit models,
Heckman selection probit models, and bivariate probit
models, to understand and control for non-exposure bias,
selection bias, and possible endogeneity bias. Our results
indicate that the impact of program interventions varies
over the different adoption steps, and entail important
implications for the design of policies to stimulate technology
adoption in areas where technology exposure and
adoption is extremely low