the human capital theory of migration and the argument of selectivity effects of individual’s
skills and education on migration (Sjaastad, 1962). Yet, if education typically promotes rural
out-migration, it does not do so with respect to all potential migrant destinations. Therefore,
the effects of some human capital variables may differ across migrant destinations and this
calls for a better mapping and estimation of different typologies of migration decision.
More in general, as we argued above, household variables (that influence individuals’ income
creation as migrants and/or non migrants) and migration costs significantly affect the decision
to send a household member to work in a different market, so that heterogeneity of household
strategies toward migration needs to be better disentangled. Thus, in the next tables we
present migration probability models for different typologies of migration conceived as
separated household alternatives.
Firstly, we carry out three logit models in order to separately predict the probability to migrate
temporary, permanently or internationally (with respect to “all the other options” respectively,
including non-migration). Secondly, we estimate a multinomial logit in order to estimate more
specifically the relative probability of household participating in one of the three categories of
migration with respect to the option of staying put. We do so because household migration
decision has multiple outcomes, which are not close substitutes for each other, though. Thus,
if on one hand three binary logit models include redundant information, on the other hand the
multinomial logit has some potential weakness30. Still, the latter model provides more
information about the simultaneous effects of independent variables across different migration
outcomes, allowing for comparisons among all combinations of the categorical dependent
variable.
Table 7 show three logit estimation results (for comparison purposes, Table A.1. in Appendix
shows results from linear probability models). Migration types have been identified from the
household survey in the following way:
MTi = 1, if household has at least one temporary migrant; MTi = 0 otherwise;
MPi = 1, if household has at least one permanent migrant; MPi = 0 otherwise;
MIi = 1, if household has at least one international migrant; MIi = 0 otherwise.
30 By assumption, the odds ratios in the multinomial logit model are independent of the other alternatives. On
models for nominal outcomes see Greene 1997, and J. Scott Long, 1997.
20