5. Estimation strategy and empirical results
The empirical analysis we carry out is twofold and aims at answering to the following
questions: What determines the decision to participate in the migration process? Is it always a
“profitable” - in that constraints-alleviating - household strategy as suggested by the NELM
insights? In particular, given the income uncertainty farm households typically face, does
migration have any importance in risk-taking behaviour in agricultural production?
The first step of our empirical strategy is estimating the determinants of household choice of
having a migrant member in the household. Since there are different types of migration -
which yield extremely different levels of (net) remittances as we saw above - we estimate
household behaviour with respect to all types of moving, i.e. permanent, temporary and
international migration, throughout binomial and multinomial logit models24.
In the second step of our empirical analysis, we estimate the impact of the three different
typologies of migration on the adoption of high-productive varieties of rice, technology
relatively more risky but higher yielding than traditional seeds. We do so through three-stage
least squares (3sls) estimation of linear probability models, in order to solve the problem of
simultaneous determination of migration and adoption decisions at household level.
Understanding household migration behaviour in the first step is needed in order to estimate
the economic effect of this endogenous process on the propensity to invest in agricultural
activities in source farm households in the second step.
The estimation strategy of a simultaneous linear probability equations aims at sorting out
problems of both endogeneity of migration choices and cross-correlation of household
decisions towards technology and migration. Linear probability models have the advantages
that are generally more tractable for assessing causation and applicable to data with limited-
dependent outcome variable and dummy endogenous regressors (Angrist, 2001; see also
below). Moreover, included explanatory variables shaping technology and migration actual
investment decisions are often of greater analytical and policy interest than are latent index
structural coefficients. Though, since the migration selection process is endogenous and
shaped by many of the same characteristics that determine technology adoption in each
regime, correct identification of the model depends on finding instrumental variables (IV) that
affect technology adoption solely through their impact on migration choices.
24 See below for further explanations on this. It should be noted that we estimated linear probability models as
well and results from the latter are similar to marginal effects predicted by non-linear binary models.
16