When gender-biased parental investments amplify human capital gaps: results from a structural model of skills formation in India

Romaine Loubes adapts a structural model of skill formation (Attanasio, Meghir and Nix, 2020) to uncover sex-specific dynamics in health and cognition acquisition. She finds that the same determinants determine skill formation between sexes, except for parental investment whose optimal timing is different, and who is gender-biased in the form of a discrimination towards girls.

Romaine Loubes is a M2 Development Economics alumnus (year 2019-2020). She is now a research assistant for Pr. Nathan Nunn at Harvard University.

In the Program for International Student Assessment (PISA) 2018 round, Vietnam ranked 13th in reading, 24th in mathematics and 4th in science out of 78 competing nations. This score places it as an outlier with respect to its South-Easten Asian neighbours and to its per-capita income. (Dang et al., 2021). Although India did not participate in those tests, reports on the Young Lives education levels assessments conduced in both countries find Indian children levels to be considerably lower than Vietnamese ones in both mathematics and reading - 55 % of Indian children aged 8 in the younger cohort could read, versus more than 95 % in Vietnam - despite similar primary education rates (World Bank). To the extent to which education is only a means towards human capital formation, this stresses upon the importance of its institutional form and the volume of potential consequences of institutional change.
In my paper, I tried to examine the difference in cognition and health production between girls and boys aged 1 to 12 in 3 Southern states of India, through the use of the Attanasio et al. (2020) model and the Young Lives database. My goal is to determine the early causes for the secondary schooling gender gaps in India: there is a 10-percentage points enrolment gap at age 12 according to Muralidharan and Prakash (2017). In cognitive terms, the Young Lives data also shows that girls, although not markedly so, fare consistently worse than boys at each round in maths and vocabulary standardized tests. This could of course also be a cause for future occupational differences, as well as a missed opportunity for the country to create more cognitive capital.
In order to examine the causes for this gap, I first adapted the dynamic structural model of skills production from Attanasio et al. (2020), splitting the sample by gender to allow for the interaction of each parameter with this new dimension. I complemented my adaptation with an Oaxaca-Blinder decomposition of parental investments between genders. This decomposition allows one to look at whether a gap in a particular outcome is due to differences in groups’ characteristics (in the seminal adaptation, that the cause for women’s inferior hourly wage could be, say, lower education levels), or if it’s rather due to the differences in the returns to these characteristics (in our example, that the additional income associated with one more year of education is inferior for a woman, which also has incentive implications). From these insights, I drew simulations of what outcomes could be achieved were parents to invest in their children in a gender-neutral way.
My results indicate that, while cognitive and health developments generally seem to be determined by the same parameters no matter the gender, this is not true of parental investment, for which optimal investment timings seem to differ. I also find a gender bias in parental investment, in the form of a deficit of valuation of their girls’s characteristics compared to boys at age 8, which could be the source for vicious circles in skill production. At all ages, parental investment towards girls seems to be more sensitive to the prices of goods relevant for children such as notebooks and school uniforms. In addition, they seem to react to cognition and health levels more if the child is a boy. 

Adapting my seminal model into two, gender-split ones allows one to consider the possibility that girls’ and boys’ developments respond to different dynamics. This is justified empirically, especially in India where son preference has largely been documented : Jayachandran et al. (2011) estimates that 14 % of the difference in girls’ mortality rate with respect to boys to be due to deprivation in inputs such as breastfeeding. Dropping this assumption was also justified by the results of the Oaxaca-Blinder decomposition and the neutral parental investment simulations, which estimates that girls would have 11 % more investments at age 12 if by age 5, parents had started investing in them as if they were boys. Normatively and in terms of efficient policy design for human capital development in India, I think that these results constitute an interesting contribution as, to my knowledge, gender had not been integrated as a dimension of its own into a dynamic structural model of skill formation before.
The robustness of my results was ensured by the use of the model from Attanasio et. al., whose quality lies in its flexibility: it gets rid of several assumptions that have been overruled by recent research, such as the one that parents know their children’s skills production functions, and tends to the endogeneity problem between skills and parental investment (the fact that, for example, parents might invest more in their child if they see that their cognition or health productions have suffered an adverse shock on a particular year) . The use of factor analysis allows for the study of health and cognition, which are by definition untangible, rather than anthropometric measurements and test scores. In the context of my sample splitting strategy, factor study has an additional advantage: considering different health, cognition and parental investment factors according to gender avoids assuming that health or cognition are shown the same way across genders. When considering separate investment factors by gender, they appeared to be more efficient in the skills production function estimates: this shows that, alongside its monetary value, the qualitative composition of parental investments is also of crucial importance (as well as time investments, which were unfortunately not available in the data). Moreover, the standardized tests in the Young Lives database were administered at home, solving gender-biased enrolment issues at later ages, such as the ones we already mentioned.

Adapting this model, I find several results that I think are interesting in the development and gender literature: first, there is no skill production difference between genders - neither in skills’ persistence from one period to the other nor on the other components they depend on (except parental investments). I find, for both genders, effects of the same magnitude as my seminal paper, which found for example that an increase of 10 % of cognition at age 8 had an impact of +7 % on cognition at age 12.

This is an excerpt from my master thesis’ estimates for skill production functions. I selected the most important variables - theoretically as well as in terms of statistical significance – and I highlighted the estimates for which the bootstrapped confidence intervals (below the point estimates, which are from 100 bootstrap samples) did not overlap with zero.
The results read in terms of elasticity: for example, one can say from my table that a 10 % increase of girls’ cognition factor at age 5 has an impact of +4.9 % of their cognition at age 8 (cognition has no value at age 1, which explains the NA for the first cell). We find Attanasio et al (2020)’s results back, notably on the influence of cognition and health (lagged) on themselves, and of health’s impact on cognition, although to a lesser extent than in the original paper. My conclusions with respect to different optimal timings of parental investment depending on the child’s sex, in otherwise similar production processes, are also visible through these results.

Investments, however, seem to be most useful for girls at age 8 (which is also the age at which they have a parental investment deficit), when they are the most inefficient for boys. They also seem to obey to different logics according to gender, as prices seem to matter less for boys-related investment, and cognition and health seem to matter more, creating different incentives for both parents and children. Finally, in my decomposition, I show that girls suffer a 4 % deficit in the returns, in terms of parental investments, to their characteristics (notably their health and cognition levels) at age 8. Considering this both knowing that investments seem to be most useful for girls at age 8 and that they are very persistent throughout stages of development, makes it an important insight in the explanation of gender test scores (or later schooling) gaps.

In studying such a complex nexus as gender gaps in human capital, establishing its causes is important to guide policy. But comparing solutions and designing policies and actions are crucial and constitute a logical next step for this research. Integrating the much-needed parental time investments into this model, and performing international comparisons with the other countries of the Young Lives database (Peru, Ethiopia and especially Vietnam) is another promising field of research.