Generalized Least Squares (AGLS) estimator)43. We get structural coefficients on latent
(utility) variable but, as expected, marginal effects on the probability of observing a one in the
probit model is very similar to results of the former linear probability model (for comparison
purpose, two and three-stage linear probability estimators and presented in the same table in
Appendix).
In the first equation of the model (5.1), which explains the adoption of modern seeds, the
dependent variable is equal to one if farm household has adopted the risky HYVs. As
explanatory variables we include farm household characteristics such as household
demographic variables, representative of family labour endowment; the average level of
education of household members, which is not expected to be highly significant with respect
to the relatively well known ‘green revolution’ technology package; the amount of cultivated
land and land tenure, in order to take into account ‘tenurial-insecurity’ effects on risk
management; farm equipment and means of ploughing owned; the quality of land (i.e. the
percentage of irrigated land); pond44 and cattle ownership, to take into account productive
activities diversification; the ratio between land and family-labour endowment (whose
negative sign suggests that HYVs are more labour-intensive crop than traditional ones);
religion and self poor assessment of the household; regional dummy variable (equal to one
whether household lives in Madhupur)45.
Specification of migration equations is based on the first part of our empirical analysis
conducted above about the determinants of temporary, permanent and international migration.
Along with instruments, we include other (exogenous) variables as determinants of migration,
such as demographic characteristics, the educational level of the most educated household
member, the amount of land owned per consumption unit and the family migration ‘network’.
Table 10 reports econometric results of 3sls estimate of the system of equations, results that
are robust to a number of alternative specifications.
43 This method is closely related to the Heckman two-step procedure (Heckman 1978 p. 947) according to which
if the purpose of the analysis is to estimate the final equation of interest (in our case technology adoption), it is
possible to generate instrumental variables for endogenous regressors by estimating a linear probability model
and with plug-in fitted results standard two-stage least squares procedure applies. Though, this procedure would
yield incorrect standard errors and for this reason, Amemiya Generalized Least Squares is used to recover correct
standard error estimates (see Maddala 1983, pp. 247-252 and Newey, 1987).
44 Ponds are very important in rural Bangladesh life. They have multi-purpose usages. They are one of
the water sources for the rural households, livestock and irrigation. Moreover, ponds are used for small-scale
culture fisheries at a household level, thereby contributing to activity diversification.
45 On the adoption of HYVs in Bangladesh see Mendola 2003.
31