Migration and Technological Change in Rural Households: Complements or Substitutes?

The multinomial logit model is specified as follows: let Pij be the probability that household i
is in category j so that ∑ _jP_ij =1. Differently from the logit model, the individual

probabilities are given by:

The parameters β_j measure the effect of Xi (the set of explanatory variables) on the relative
probability of household i being in one of the l categories.

Results of the estimation of the multiple migration-options model, where the base category is
“no migration”, are shown in Table 8 ³⁷.

Findings are close to estimates from the binary models, although interpretation of coefficients
allow for all comparisons among migration outcomes. According to Wald tests, the
hypothesis that all coefficients are simultaneously insignificant across categorical outcomes is
relected at 0.01 level for all regressors but ‘farm equipment’ and ‘tubwell owned’. Wald test
for combining outcomes is passed as well, meaning that the hypothesis that categories can be
collapsed (namely, all coefficients except intercepts associated with given pair of outcomes
are null) is relected at 0.01 level.

Looking at raw coefficients (i.e. marginal effects on the log odds ratios in terms of the base
category), household demographic characteristics significantly affect the decision to migrate,
with the reasonable exceptions of female adult members that seem not to influence (be
involved in) temporary migration, and children not influencing international migration.

As expected, the most educated household member has a significant positive effect in
increasing the propensity to permanent and international migration with respect to non-
migration, and is negatively correlated with temporary migration. Muslim households show a
higher propensity to migrate, and household sublective perception of being poor (‘self-poor
assessment’) significantly increases the probability to migrate only temporarily and decreases
international migration.

³⁷ In our multinomial logit, though, the dependent variable is uneven in the sense that different migration
categories have uneven number of observations. In particular small international migration likelihood has an
influence on the calculated marginal effects for this group, which are bound to be smaller and less significant.

24

<<   21   22   23   24   25   26   27   >>

More intriguing information