Sources of gender difference in rural to urban migration in Kenya: does human capital matter?

Abstract

Using data from Kenya this article estimates the urban to rural gender gap in the rate of migration and then decomposes the gap into the explained portion and the portion due to gender differences in coefficients. The former is further decomposed to unveil the relative influence of each explanatory variable on the explained portion of the gender gap in the rate of migration. A non-trivial finding suggests that human capital variables may exert the strongest influence on gender differences in migration, partially explaining the higher incidence of male migration.

Notes

The dependent variable is binary and takes the value 1 if the individual is a migrant in the urban area and zero if the individual is a non-migrant in the rural area. The independent variables consist of variables thought to proxy the determinants of rural to urban migration without affecting wages: size of the rural land holding, number of children at the time of migration, and distance to the urban area. Strictly speaking, however, identification of the instrumental variables would have been more satisfactorily resolved if one had a larger number and variety of variables that would shift the probability of migration without affecting wages—indeed this is a perennial problem in the literature; however, the very limited nature of the data precluded such attempts by the authors.

Where ln W_ig,u (the subscript ig,u = the ith migrant in the urban area for each gender) is the log weekly wage for migrant workers in urban areas, for each gender. Similarly, lnW_ig,r (the subscript ig,r = the ith non-migrant in the rural area for each gender) is the log weekly wage for non-migrant workers in rural areas, for each gender. The matrices X_u and X_r consist of individual characteristics for migrant workers in urban areas and non-migrant workers in rural areas for each gender respectively. These variables include age, square of age, marital status and categorical variables for education. For the education variables, the combination of individuals with no education and those with primary education constitute the base group. The vectors β _u and β _r are the regression coefficients for migrant workers in urban areas and for non-migrant workers in rural areas, respectively.

And as Jones (1983) demonstrates, the latter portion (that is the portion due to coefficients) cannot be decomposed further. Therefore, the validity of the interpretation of the portion due to coefficients largely depends on the adequacy in the specification of the status equations.