program, the more they are able to appreciate its real (vs.
expected) advantages and disadvantages. This relationship
generates modelling problems related to sample selection, as
the decision to continue participating is relevant only to those
farmers who adopted it in the first place. Therefore, independent
probitmodels can potentially provide inefficient estimates
of the parameters of adoption and continuation models, as they
ignore the fact that the decision to continue participating is
contingent upon the decision to adopt. To test the possibility of
such a bias, in a second approach, we used an alternative
modelling using a (bivariate) probit with sample selection
(Amsalu and de Graaff, 2007; Greene, 2007; Tura et al., 2010).