Like father, like son: does migration experienced during child schooling affect mobility?

ABSTRACT

Using a sample of 39,297 father-son pairs from Indian Human Development Surveys (IHDS), we examine whether migration experienced during child schooling affects the relationship between parent and child education. We relax the co-residency restriction for father-son pairs to obviate coresident sample selection bias in our mobility estimates. The panel structure of data enables us to identify children who were enrolled in school at the time when their families migrated. We find that migration experienced during schooling increases downward mobility. In particular, those children who were young at the time of migration tend to have poor educational outcomes. We show that widely cited aggregate measures of mobility provide an incomplete representation of intergenerational persistence in India, as most of the persistence originates from the tails of the educational distribution. The sons of the least educated fathers have poor prospects of upward mobility and face a glass ceiling in higher education. We use heteroscedasticity-based identification and Rosenbaumʻs sensitivity analysis to account for unobserved heterogeneity.

KEYWORDS:

• Schooling
• education
• unobserved heterogeneity
• heteroscedasticity-based identification

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Some studies observe that use of sampling weights in regressions may induce bias in the parameter estimates (Gary, Haider, and Wooldridge 2015; Winship and Radbill 1994). We estimate both weighted and unweighted regression models and both sets of results are qualitatively similar.

2 40.5% of the non-migrants belong to other backward castes (OBC) whereas the corresponding proportion for migrants is 35.6%. This difference is significant at the 1% level.

3 The Breusch-Pagan/Cook-Weisenberg tests reported in the paper are based on the set of all exogenous variables. To investigate the source of heteroscedasticity in our data, we also conducted the test with each individual variable (not reported). We find that most of the heteroscedasticity comes from variables for age, household assets and the rural dummy.

4 $\hat{\rho },$ are slightly different from those reported in Table 3. This is because the decomposition procedure of Checchi and Flabbi (2007) does not include additional socioeconomic controls, which are included in the regression models reported in Table 3..

5 A temporal increase in education level obscures positional mobility. The primary educated sons of uneducated fathers may be viewed as decline in intergenerational persistence in terms of absolute attainment. However, these pairs may still contribute positively to ICC if both fathers and sons have similar positions in the education distribution of their generations.