Maternal education gradients in early life height: A comparative study of eight Latin American countries


 Background

More than 20% of the children around the world were stunted in 2018. The situation is not any better in Latin American countries, even though stunting prevalence has been declining since 2000. Stunting has adverse consequences on children: severe short- and long-term health and functional effects, poor cognition and educational performance, low adult wages, and productivity loss.
Methods

This study compares maternal education gradients in height-for-age z-scores (HAZs) and stunting prevalence in children between two and four years of age from eight different Latin American countries: Chile, Uruguay, Costa Rica, Paraguay, Mexico, El Salvador, Nicaragua, and Peru.
Results

Results show that the prevalence of stunting varies widely across Latin American countries. Having a mother with tertiary education increases HAZs in every country (except Paraguay), compared to having a mother with primary education or less. In some countries, there is also a difference in HAZs associated with secondary versus primary maternal education. With regard to stunting, maternal education is a crucial determinant to decrease the odds of being stunted in early years in countries with the highest stunting prevalence (≈ 20%); however, this is not the case for countries where the prevalence is low (< 5%).
Conclusions

We found that maternal education is associated with height and stunting in preschool-aged children in our sampled Latin American countries. The association’s magnitude is higher for LMICs than for countries with a higher income per capita in the zone (i.e., Chile and Uruguay). These results suggest that future policies aiming to increase maternal education will improve children’s height-for-age and decrease their risk of stunting.


Introduction
Favorable physical growth, particularly in the early years of life, has been shown to have an extensive impact on population health, enhancing cognitive and socioemotional skills, allowing people to live longer in a healthy condition (Dewey & Begum, 2011;Grantham-McGregor et al., 2007;Guerrant et al., 2013;Martorell, 1999;Reynolds et al., 2017;Stevens et al., 2012;Walker et al., 2007). According to the latest report of the World Health Organization (WHO), childhood nutrition is essential for proper growth In terms of demographic indicators, Table 2 shows economic and public health parameters for each country closer to the year of our infant data (2013) to understand the demographic and economic contexts of the different countries. Table 2 shows that countries with higher health expenditures have lower stunting prevalence. Yet, other indicators, such as infant mortality rate, are also aligned to the prior. The more investment the government puts into health, the better the outcomes for children.
The disparities in HAZs and stunting prevalence have been explained by different factors (del Pilar Flores-Quispe et al., 2019;Frongillo et al., 1997), but we want to pay close attention to a speci c aspect: maternal educational gradients and their role in explaining the HAZs and stunting prevalence across countries. Figure 2 shows maternal education by country, and Figure 3 displays stunted children's prevalence by maternal educational and country. Secondary education is the most frequent maternal education level in most countries, while primary education remains the most prevalent level in countries like Costa Rica and Nicaragua. When looking at both maternal education level and stunting prevalence, a negative correlation is seen in every country (e.g., if education increases, stunting prevalence decreases).
Therefore, these gures raise new questions about how relevant maternal education is in determining HAZs and stunting prevalence in children under the age of 5 years. This paper explores how maternal education gradients affect HAZs and stunting prevalence and how these gradients differ by nation.

Data
This paper uses the "Latin American -Center for Advanced Studies on Educational Justice" data project (LATAM-CJE), which is a multi-sectoral effort aiming to merge early childhood datasets from different countries in Latin America. The main goal of the project is to generate regionally comparative research on child development outcomes. Thus far, the project has included eight Latin American countries. Data were extracted from four surveys. First, early childhood data from Mexico and El Salvador were obtained from the Multiple Indicator Cluster Surveys (MICS), an international project carried out in LMICs by UNICEF that collects information about children and their households every ve years (UNICEF, 2016). Second, data from Costa Rica, Nicaragua, Paraguay, and Peru were accessed through the Regional Project on Child Development Indicators (PRIDI), which was led by the Inter-American Development Bank (IDB) in 2013 (Verdisco et al., 2016). They measured child development indicators in children between 24 and 59 months of age. Third and fourth, data from Chile and Uruguay were extracted from their countryspeci c early childhood longitudinal studies: the Longitudinal Survey on Early Childhood Development (ELPI), and the Nutrition, Child Development, and Health Survey (ENDIS), respectively (Chilean Ministry of Social Development and Family, 2020; Uruguayan Ministry of Social Development, 2020).
We restricted our sample to children aged between 2 and 4 years (24 to 48 months old) for the present analysis. The chosen range represents a crucial developmental stage for the height-for-age marker. Full sample sizes and analytical sample sizes are shown in Appendix Table 1.

Measures
Two dependent variables were used in the models: HAZs and stunting prevalence. Height was objectively measured in each survey using a measuring tape from the oor to the children to get their full size in centimeters. For our analysis, we standardized height-for-age based on WHO standards for age and sex. Stunting is de ned as a HAZ of less than or equal to two height-for-age standard deviations (SDs) from the WHO Child Growth Standards (WHO, 2006).
The main independent variable is maternal education level, which we divided into three categories: primary or less, secondary, and tertiary education level. The exact de nition of each of these categories varies depending on the country. Primary education is de ned as six years of formal schooling in all countries[1]. In terms of secondary level, Chile, El Salvador, Mexico, Paraguay, and Uruguay have six years of secondary school, while Costa Rica, Nicaragua, and Peru have ve years of secondary school. Any years of education after secondary education were categorized as tertiary education.
Independent variables incorporated as covariates included characteristics of the mother, child, and household. Mother characteristics were age (in years) and ethnicity (which indicates whether the mother spoke indigenous language). Child characteristics included age in months, gender, and whether they were breastfed. Household characteristics contained three different variables: the number of people living in the household, the total number of additional children living at home, and wealth or income terciles (computed based on different scales and measures depending on the survey). Finally, to capture parenting style, we added two childhood stimulation variables: whether the parents stimulated their child by singing them a song and whether they regularly told their child stories (both presented as dichotomous variables). Table 3 depicts the descriptive statistics of our sample by country. The table shows that levels of stunting and levels of maternal education differ by country (also shown in Fig. 1 and Fig. 2). In terms of maternal age, the samples are quite similar, with means from 27.5 years in Costa Rica through 32.5 years in El Salvador. Ethnicity and rurality are characteristics that vary greatly between countries: ethnicity ranges from 0.7% in Costa Rica through 37.1% in Paraguay, and rurality ranges from 10.6% in Chile through 43% in El Salvador. Children characteristics are quite similar across countries: age in months, sex, and if they were breastfed.
To understand each national context, we compare our sample of mothers, whose children were 2 to 4 years old, with the standard socio-economic survey of each country to contrast our main independent variable: maternal education level. The results from this exercise are shown in appendix, section B. Our sampled mothers are in general much more educated for all countries than in the whole population. The main explanation behind this is that this is a relatively young section of the population-it represents a younger cohort born mainly after primary and secondary education became universal in these countries.
Thus, we should bear in mind that we work with a non-random sample of a much more educated population than what is observed in the whole population.

Methods
In the rst set of analyses, we run linear regression models for the HAZ separately by country, as indicated in equation (1). The dependent variable corresponds to the height-for-age for the child i, and the main independent variable are dummies for maternal educational level (), where the rst categoryprimary or less-is omitted. Therefore, an interpretation of the coe cients is made in comparison to this category. is a matrix of "j" observed child characteristics, including mother and household characteristics, and is the error term.
See formulas 1 and 2 in the supplementary les.
[1] Paraguay adopted a reform establishing 9 years for primary education, but we used 6 years for primary and 6 years for secondary according to the international equivalent educational standards of the United Nations Educational, Scienti c, and Cultural Organization (UNESCO; UNESCO, 2013UNESCO, , 2015.

Results
Results from the rst set of analyses indicate the importance of maternal education for HAZs among children between 24 and 48 months. Figure 4 shows the coe cients estimated for maternal secondary and tertiary education graphically by country compared to primary or less. Table 4 displays the coe cient estimates by country for the full model. For almost all countries, children of mothers with tertiary education had higher HAZs, compared to children of mothers with primary education or less. The exception is Paraguay, where there is no clear association between maternal educational level and HAZs.
The maternal educational gradient varies by country: the smallest tertiary-primary gap was estimated for Chile at 0.2 SDs, i.e., children of mothers with tertiary education in Chile were 0.2 SDs taller than were children of mothers with primary education. Countries like Uruguay, Mexico, El Salvador, and Peru had a gradient ranging between 0.3 and 0.4 SDs. In contrast, countries like Costa Rica and Nicaragua show a maternal educational gradient from at least half a SD in height for age. The differences are less consistent when comparing children from mothers with secondary education versus primary or less. There is only a difference in HAZs in some countries between children of mothers with secondary and primary education: Uruguay, Mexico, El Salvador, and Peru.
With regard to other covariates, mother's age is relevant to explain HAZs in children under 5 years, but only in some countries: Chile, Mexico, El Salvador, and Peru. Nonetheless, the magnitude of the coe cient is small. Ethnicity was a relevant predictor in countries like Mexico, El Salvador, and Peru. The coe cients were 0.4 SDs for Mexico and El Salvador and 0.8 for Peru, which was of a higher relevance than the maternal educational gradient. Similarly, household wealth terciles were signi cant predictors of HAZs for all countries. The gradient in HAZs between belonging to the wealthier tercile and the poorer tercile varies between countries, being more important for countries like Paraguay (0.7 SDs), El Salvador (0.6 SDs), and Peru (0.5 SDs) and less important for countries like Chile (0.2 SDs) and Uruguay, where the coe cient did not present any meaningful variation. Table 5 and Figure 5 depict the results for the models for stunting. Maternal education level is a crucial predictor for countries with a higher current stunting prevalence. Countries like Chile and Uruguay, where the prevalence is less than 4%, exhibited no differences in their respective stunting prevalence by mother's primary, secondary, or tertiary education level. Costa Rica, which has a prevalence of stunted children of 9%, did not display a maternal education gradient either. Notwithstanding, Paraguay (12% stunting prevalence) showed that mothers with tertiary education have almost a 50% lesser chance of having a stunted child than mothers with primary education level or less. The gradient is well observed in Mexico, where the odds for children of mothers with secondary education are 35% lower than they were for children of mothers with primary education or less. Meanwhile, having tertiary education increases these odds in about 50%. The maternal educational gradient is even more exacerbated in countries like El Salvador, Nicaragua, and Peru, where the stunting prevalence is the highest of our sampled countries (≈15%). Lower odds of stunting prevalence were seen if the mother had a higher education level. With regard to other covariates, an important nding is that with Nicaragua's exception, wealth terciles are vital determinants in predicting stunting prevalence, being even stronger than maternal educational gradients in some of our sampled countries.
Finally, we did an exploratory analysis by looking at the differences by child's gender and urban/rural residence. Even though for some countries sample sizes are small, Tables C1 and C2 in the appendix demonstrate these results. For boys, there is no clear pattern for all countries, but for girls, we see secondary and tertiary education being an important determinant of HAZs in those countries with a higher prevalence of stunting: Mexico, El Salvador, Nicaragua, and Peru. We saw no clear pattern for rurality, and the most important signi cant nding was found for Peru, where we observed a difference between urban and rural residency. For urban residency, the maternal educational gradient is important at the secondary and tertiary educational levels; however, maternal educational level was not a clear driver of children's HAZs in the rural zone.

Conclusions
This research article examined the association between height-for-age and stunting and maternal educational level among children between 2 and 4 years old from eight different Latin American countries. Our results evidence that children whose mothers have tertiary education had an increased HAZ in every country-except for Paraguay-compared to children whose mothers have primary education or less. In addition, maternal education was a crucial determinant in decreasing the odds of stunting in countries with a high stunting prevalence (Mexico, El Salvador, Nicaragua, and Peru), but not in countries where the prevalence of stunting was low (Chile, Uruguay, Costa Rica, and Paraguay). We view these maternal educational gradients as capturing cultural capital rather than economic resources, as we control for wealth terciles.
Our results are consistent with the previous literature on socioeconomic characteristics and how they affect stunting prevalence in the early years in LMICs (Black et al., 2013;Lu et al., 2016). For Latin American countries, studies have found that children with illiterate primary caregivers were 5 times more likely to be stunted compared with their counterparts with literate primary caregivers in Guatemala (Sereebutra et al., 2006), and maternal stature, low body mass index (BMI), low education, and poverty were also predictors of stunting in Uruguay (Bove et al., 2012). For Colombia, child stunting is associated with lower socioeconomic and maternal nutritional statuses . To our knowledge, this is the rst study to examine the in uence of maternal education on HAZs and stunting across a range of Latin American countries. This paper has some limitations. First, we do not have information on nutritional aspects for each country; this could explain some of the heterogeneous results we are nding across countries. Second, as with many comparative papers, we have to limit the number of controls in the estimations to those available in each survey. This may limit the analysis by introducing a potential omitted variable bias.
Third, even though our analyses used several Latin American countries, our results do not necessarily represent other countries across the zone, because of either missing studies or data availability. Fourth, we lacked data on other environmental characteristics affecting early height and stunting, such as a lack of solid waste disposal, a lack of sanitation, solid fuels in the household, and dirt oor and material deprivation, although we did use socioeconomic measures to proxy for these differences.
Despite these limitations, this study advances the literature on early childhood growth in Latin American countries from a comparative perspective by using a combination of national surveys. We harmonized and merged the data on early childhood stunting and sociodemographic variables, which no study has done before, resulting in a study of a considerable sample size. A particular strength is the standardization of the anthropometric measures-HAZ and stunting-allowing a reasonable degree of comparability between estimates. Finally, we use the most recent surveys from countries, most from 2012-2013.
In conclusion, we found that maternal education is associated with height and stunting in preschool-aged children in our sampled Latin American countries. The association's magnitude is higher for LMICs than for countries with a higher income per capita in the zone (i.e., Chile and Uruguay). This could be explained in part because formal education may equip parents to apply successfully knowledge about health, sanitation, and responsive interaction when caring for their children (Pamuk et al., 2011), which could prevent them from stunting in the context of low or moderate income.
These ndings support the importance of improving maternal education as a way of enhancing children's nutritional and environmental conditions during the early ages to ensure proper linear growth throughout childhood. Another approach would be to provide additional supports to families with less educated mothers. Successful interventions are often implemented as multi-sectoral packages anchored in nurturing care and include parenting support, preschool participation, and responsive care (Britto et al., 2016). Interventions should aim at lessening the di culties imposed by the environmental disruptors of child development. This is part of the Sustainable Development Goals (SDGs) for 2030, aiming to tackle stunting to ensure every child can reach their potential over their life course (Richter et al., 2017).

Declarations
Ethics approval and consent to participate: not applicable, data was public.
Consent for publication: not applicable, data was public.
Availability of data and material: the datasets analyzed during the current study are available in the web page for each survey.