2.2. Assumptions for propensity score matching technique
In this paper, a Probit model was used to estimate the propensity scores that were used to measure
impact of participating in BVIS on net agricultural income. The estimate of the propensity scores depends
on the covariates included in the Probit model. However, there is no consensus on the type of covariates
should be included in the discrete choice models when estimating propensity scores (Austin, 2011). The
variables were chosen that strongly influence participation in the irrigation scheme but weakly influence
net annual agricultural income (Table 1). We also included high order and interaction terms to improve
the balancing of covariates as suggested by Dehejia andWahba (2002). Following Rosenbaum and Rubin’s
(1983) procedure, the propensity scoreswere divided into blocks among the groups. The propensity scores
for the blocks among the groups were not different between participants and non-participants, thereby
satisfying the balancing condition of propensity scores as suggested by Becker and Ichino (2002).
Other important considerations thatwere adhered to are the common support and overlap assumptions
to ensure that farmers with the same X or Z values have a positive probability of being either participants