Parental education, gender preferences and child nutritional status in Peru

ABSTRACT

This paper examines whether the distribution of bargaining power between parents affects nutritional indicators in the early stages of a child’s life, giving evidence that the allocation of household resources varies by the gender of the child and the parents. After accounting for the potential endogeneity of the indicator of power distribution within the household, related to assortative mating in the marriage market, this paper shows that maternal power is more positively associated with girls’ nutrition than boys’. Among households located in rural areas, resource allocation between girls and boys seems to differ. Similarly, some evidence of competition for household resources affecting girls’ nutrition is found.

KEYWORDS:

• Intrahousehold allocations
• nutrition
• children
• Peru

JEL CLASSIFICATION:

• D13
• I12
• J13

Supplementary material

Supplemental material for this article can be accessed here.

Acknowledgements

I would like to thank Arjun Bedi, Mark Bryan, Adeline Delavande, John Ermisch, Marta Favara and Patrick Nolen for their very helpful comments and suggestions. I would also like to thank participants at the COSME-FEDEA workshop in Gender Economics, 2010; the ‘Health, Happiness and Inequality’ conference at the University of Darmstadt, 2010; the International Association for Feminist Economics (IAFFE) conference, 2010; the ‘Women’s Bargaining Power and Economic Development’ workshop at the ISS-Erasmus University, 2010; and the Latin American and Caribbean Economic Association (LACEA) conference, 2011. The data used in this publication come from Young Lives, a 15-year survey investigating the changing nature of childhood poverty in Ethiopia, India (Andhra Pradesh and Telangana), Peru and Vietnam (www.younglives.org.uk). Young Lives is core-funded by UK aid from the Department for International Development (DFID) and co-funded from 2010 to 2014 by the Netherlands Ministry of Foreign Affairs. The views expressed here are those of the author. They are not necessarily those of Young Lives, the University of Oxford, DFID or other funders.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. According to the National Statistics Office of Peru, in the period from 2002 to 2015, the poverty and extreme poverty rates went down in 32 and 20 percentage points, from 54% and 24%, respectively. However, in 2015, the poverty rate in rural areas (45%) rose threefold compared to the one in urban areas, and the extreme poverty rate is 14 times larger (14% versus 1%). Moreover, in 2015, while 5 out of 10 Peruvians living in poverty are indigenous, only 3 out of 10 non-poor are indigenous.

2. YL also collected data of children in Ethiopia, India (Andhra Pradesh and Telangana) and Vietnam. It is worth mentioning that the sampling design followed by the YL team and used in this paper does not aim to be representative at the national level. More detailed information than that presented herein can be found here: www.younglives.org.uk.

3. Appendix I discusses the theoretical framework.

4. The ratio of female to male years of education is used as an additional indicator of power distribution in the household, yielding to similar results to the discrete indicator. Results are available in Appendix II.

5. According to the YL project, a sentinel site corresponds to a geographical area where it is possible to collect individual, household, community, regional and national characteristics. Sentinel sites are also important for YL complementary thematic studies.

6. The YL team in Peru used a multistage, cluster-stratified, random-sampling procedure to select the two cohorts of children, enabling the randomization of households and of sentinel sites, and selecting sentinel sites with a pro-poor bias (Escobal & Flores, 2008).

7. Restricting the sample to households where the two biological parents are present is not strictly necessary, but it rules out additional channels through which the biological relationship with the decision makers within the household may affect investments in children’s human capital. In addition, the proportion of households where at least one of the two adults taking decisions about the child is not a biological parent represents less than 1% of the whole sample of Peru.

8. The indicators of migration status correspond to dummy variables (one for fathers and one for mothers) on whether the parent has always lived in the community surveyed.

9. The variable years of education was constructed using the information on school grades and levels collected in the YL data. Given that YL collected information about the years of education achieved in each level, independently if the individual completed or not that level, there was no need for imputing years of education, except in the following two cases: (1) when parents reported that their highest educational level or the one they were enrolled at the time of the interview was ‘none’, ‘nursery’ or ‘in 1st year of primary education’, zero years of education were imputed; and (2) for one observation (a father), who declared having completed tertiary non-university education, but he did not declare the years achieved, 13 years of education was imputed (11 years of secondary education needed to access most technical degrees in the country and two for the tertiary education level).

10. For a complete discussion of WHO child growth standards and a comparison with the National Center for Health Statistics (NCHS) standards, see WHO Multicentre Growth Reference Study Group (2006). According to this document, the standards depict normal early childhood growth under optimal environmental conditions and can be used to assess children everywhere, regardless of ethnicity, socio-economic status and type of feeding. The YL team calculated these z-score indicators using a STATA macro for the 2006 growth standards that were downloaded from the WHO website: http://www.who.int/childgrowth/standards/en/.

11. For more information, see http://www.who.int/nutgrowthdb/about/introduction/en/index2.html.

12. YL estimates were similar to the ones calculated using the Demographic Health Survey 2000, for children in the same age group at national level. At national level, stunting and wasting rates were 16% and 2%, respectively. However, while the proportion of wasted children was similar between rural and urban areas (3% and 2%, respectively), the proportion of stunted children was much higher in rural than urban areas (30% vs. 11%).

13. Following Barcellos et al. (2014), I restricted the sample to households where the YL child is the youngest and/or only child, which corresponds to 99% of households in the sample.

14. This refers to the effect on child’s nutrition of maternal phenotype (manifested in her height), which is composed by her genotype and the influence of environmental factors and the interactions between the two.

15. In Peru, an important unobserved factor might be the submissive role of women in society. However, Knodel, Loi, Jayakody and Huy (2005) argued that the submissive role of women in traditional societies does not generally affect women’s control over the expenditure on and distribution of food within the household.

16. Reggio (2011) included the age difference between husbands and wives as an additional distribution factor (i.e. a factor affecting the intrahousehold distribution of power that does not directly affect individual preferences). The author finds that wives who are much younger than their husbands have less decision-making power in the household.

17. Full regression results are available on request. Moreover, the discussion of results focuses on models that include the extended set of covariates related to household formation. Relative to models containing only variables in X, adding the set of variables related to household formation slightly improves the models’ fit but does not statistically change the point estimates from the first models.

18. A channel through which maternal preferences might negatively affect children’s health in the short-run is related to weaning. On one hand, as Martorell and Habicht (1986) showed, in developing countries, the most important problem of infection affecting child health is diarrhoeal diseases, which are more likely to occur while a child is being weaned. On the other hand, Jayachandran and Kuziemko (2011) and Chakravarty (2012) found evidence that breastfeeding practices are used for favouring boys in India and Africa. Along this line of thinking, results in Table 4 might be related to the fact that in the YL sample of Peru: (1) girls are weaned later than boys and (2) children of mothers with more bargaining power are weaned before other children.

19. The difference is statistically significant at the 5% level as reflected in the coefficient on Boy × $x_i$.

20. It is not possible to reject the null hypothesis of equality of the coefficients MBP (maternal bargaining power) in the model of child weight (I) and growth in weight (II) (p value = 0.83). Similarly, it is not possible to reject the null hypothesis of equality of the coefficients Boy × MBP (measuring the differential association between maternal bargaining power and the nutritional outcomes of boys and girls) in the model of child weight (I) and growth in weight (II) (p value = 0.80).

21. It is not possible to reject the null hypothesis of the coefficients BP (maternal bargaining power) in the model of child length (I) and growth in length (II) (p value = 0.75). Similarly, it is not possible to reject the null hypothesis of equality of the coefficients Boy × BP (measuring the differential association between maternal bargaining power and the nutritional outcome of boys and girls) in the model of child length (I) and growth in length (II) (p value = 0.95).

22. Households with no missing information on paternal height seem to be poorer than those with missing information on paternal height. This may reflect correlations between the presence of the father during the interview and unemployment status. Households with no missing information on paternal height are statistically different in the following observable characteristics: children show smaller weight-for-length z-scores, parents are less educated, mothers are less able to speak the main language in the community, the household has a (relatively) poorer wealth index, the house has less access to sewers and lacks good-quality floors, parents are not migrants and parents are mainly in rural areas.

23. Results are available upon request.

24. In the specification of children’s length-for-age, adding paternal height significantly improves the fit of the model relative to the one containing only maternal height.

Additional information

Funding

Young Lives is core-funded by UK aid from the Department for International Development (DFID) and co-funded from 2010 to 2014 by the Netherlands Ministry of Foreign Affairs.

Notes on contributors

Rafael Novella

Rafael Novella is a research associate at Department of International Development, University of Oxford and works at the Inter-American Development Bank. His main research fields are development, family and labor economics.