^{1}

^{2}

^{3}

^{‡}

^{1}

^{‡}

^{1}

^{‡}

The authors have declared that no competing interests exist.

‡ These authors also contributed equally to this work.

Under-nutrition in early childhood has harmful impacts on human capital formation in children, with implications for educational, adult health, and labor market outcomes. We investigate the association of linear growth and weight gain in mid-childhood with years of schooling, the Peabody Picture Vocabulary Test score, and math test score during the adolescent age of 14–15 years.

Data were derived from the Young Lives study conducted in four low- and middle-income countries (Ethiopia, India, Peru, and Vietnam). The data had detailed information on the children anthropometry and characteristics of the child, household, and community. Multivariate regression analysis, adjusted for the confounding variables, was used to investigate the association between mid-childhood health, measured by conditional linear growth and relative weight gain, and human capital outcomes in adolescent age.

After controlling for several confounders, one cm increase in conditional linear growth increased years of schooling by 0.034 years and the Peabody Picture Vocabulary Test score and math test score by 0.474 and 0.083 points respectively. Relative weight gain was negatively associated with years of schooling and math test score. There is no evidence of heterogeneous effects by rural, gender, and household wealth. In the quantile regression analyses, the association between conditional linear growth and outcomes is stronger at the lower level of years of schooling and the Peabody Picture Vocabulary Test score.

Our study highlights that mid-childhood nutritional intervention targeted for students at the lower level of education distribution can accelerate the rate of human capital accumulation in low- and middle-income countries.

Human capital accumulation is often considered as the end and means of economic development. Human capital such as education and health can directly contribute to the output growth through increased productivity and technological progress. While exploring the determinants of human capital formation, a number of studies have found that fetal and early childhood malnutrition adversely affect later educational attainment, cognitive and non-cognitive skills, mid-childhood and adult health, labor productivity, and hence economic growth [

Despite substantial evidence on the association between under-nutrition in early-life and subsequent cognitive development, very little is known about the relative importance of different phases of growth in early childhood. The literature on catch-up growth is mixed and unsettled. Some studies show that cognitive development responds more to investments made in the first 2 years of life [

Given the importance of early life conditions on later human capital development, it is important to understand growth trajectories among children and its impacts on human capital. In low- and middle-income countries, growth faltering is very common during the first two years of life [

In addition to examining the direct effect of linear growth and weight gain on schooling and cognitive outcomes, it is important to separate out the effect of weight gain from linear growth on schooling. Understanding which phase in childhood growth is associated with human capital may help policymakers to decide the timing of interventions to reduce the adverse effect of childhood malnutrition. Linear growth or height gain and weight gain may have different impacts on human capital, and therefore, it is very important to disentangle the effect of linear growth from that of weight gain [

A limited number of studies have attempted to separate out the effect of height gain from weight gain [

The objective of the present study is to explore the association between conditional linear growth (CLG) and relative weight gain (RWG) between mid-childhood (7–8 years) and adolescence (14–15 years) on educational outcomes in four low- and middle-income countries (Ethiopia, India, Peru, and Vietnam). We estimate the mean effects in an ordinary least square (OLS) framework, while heterogeneity in the effects is examined by estimating an interaction model as well as quantile regression (QR) model.

Our study uses data from the Young Lives Study (YLS) of four prospective mid-aged (7–8 years) child cohorts from Ethiopia (N = 1000), India (N = 1008), Peru (N = 714) and Vietnam (N = 1000) surveyed first in 2002 and followed up during adolescence (14–15 years) in 2009–10. YLS was conducted by the University of Oxford, UK in collaboration with local institutions from each country and follows a purposive sampling strategy [

The YLS data is publicly available and could be obtained from the YL website (

Kernel Density graphs (Figs

The main outcome of interest in this study is the educational attainment of children during adolescence (age 14–15 years) in 2009. Educational attainment is measured by years of schooling, child’s receptive language skills (PPVT), and performance in standardized math test score, both of which were developed in YLS [

PPVT uses items that consist of a stimulus word and a set of pictures and is commonly used to represent child cognitive and intellectual ability in developing countries. The test requires respondents to select the pictures that best represent the meaning of a series of stimulus words read out by the surveyors. The test consists of 17 sets of 12 words each and the raw score test results can take possible values from 0 to 204. The PPVT of 125 questions was used in Peru, whereas PPVT of 204 questions was used in Ethiopia, India, and Vietnam. The PPVT was adapted and standardized by YL researchers in each country. The math test measures basic quantitative and number notions. It included 30 items on addition, subtraction, multiplication, division, data interpretation, problem-solving, and basic geometry. The total score of children in the math test was obtained from adding the correct responses and ranged from 0 to 30 [

The key explanatory variables are CLG and RWG during mid-childhood and adolescence. We calculated the CLG and RWG separately for males and females. The CLG of an adolescent is the expected adolescent height adjusted to their mid-childhood height and weight. Similarly, RWG is the expected adolescent weight accounting for the present height and mid-childhood weight and height. Empirically, CLG is the residual term of the regression of height at age 15 years on height and weight at age 8 years and observed height at age 15. Similarly, RWG is the residual term of the regression of weight at age 15 years on height and weight at age 8 years and observed weight at age 15 years. In other words, these conditional growth variables, CLG and RWG, are the deviations of the adolescent’s actual height or weight from the predicted conditional height or weight. Studies found that an individual attains the maximum physiological growth between mid-childhood (age 6–8 years) and adolescent (age 14–16 years) [

We also control for social-demographic (age, birth order, and gender of the indexed child; age of the mother at the birth of the indexed child; education of mother and father; caste/social group; religion/ethnicity; height of mother; morbidity at the time of early childhood; morbidity in last three years; number of times meal taken on the previous day of the survey) and household wealth and place of residence (rural versus urban) as these factors are likely to intermediate the effect of health on human capital accumulation [

Early childhood health status of an individual has a cumulative effect on the adolescent’s health and cognition development. Children with poor health experience low progress in height and low weight gain in due course of age advancement. Assume education is a functions of CLG or RWG between mid-childhood and adolescence:

Empirically, we estimate the following models. First, we estimate the CLG and RWG model.

Similarly, RWG is estimated as follows:
_{8} _{8} are height and weight during the mid-childhood period of age 7–8 years, whereas _{15} _{15} are height and weight during the adolescence period of age 14–15 years. _{15} and _{15} are conditional linear growth and relative weight gain at age 14–15 years, respectively. And, finally, the education equation can be estimated as the following Ordinary Least Square (OLS) model:
_{15} denotes years of schooling and test scores (PPVT and math test scores) at adolescence. The confounding variables (C, P, and S) are from the period when children are 7–8 years and 14–15 years old. We include _{v}, village fixed-effects, to control for fixed characteristics of villages. Any time-invariant differences between villages will be adjusted for by _{v}. Standard errors are clustered at the village level. In order to test whether the association between the conditional growth and outcomes differ by household characteristics, we estimate interaction models. We interact the conditional growth variables in

A number of studies have shown that nutritional supplementation among the severely stunted children has a stronger effect on the recovery of their cognition development than the relatively healthy ones [

Therefore, we estimate the quantile regression (QR) model to capture the underlying heterogeneity in the relationship between CLG/RWG and educational outcomes. We estimate the relationship in

Furthermore, there may be a case that, some unobserved characteristics like genetics, maternal characteristics, household or community environment, public program etc. might have caused some children to record higher CLG and RWG than the others. Due to unobserved time-invariant heterogeneity, error term in

We present the sample means of the outcome variables, key independent variables, and control variables in

Mean | Std. Dev. | |
---|---|---|

Years of schooling 2002-Round1 | 2.44 | 1.20 |

Years of schooling 2009-Round3 | 6.76 | 2.94 |

PPVT score 2009-Round3 | 136.38 | 41.61 |

Math test score 2009-Round3 | 11.00 | 7.97 |

Positive conditional linear growth during 2002–2009 | 0.52 | 0.50 |

Positive relative weight gain during 2002–2009 | 0.46 | 0.50 |

Height (cm) 2002-Round1 | 118.30 | 6.35 |

Height (cm) 2009-Round3 | 154.35 | 8.23 |

Weight (kg) 2002-Round1 | 20.34 | 3.60 |

Weight (kg) 2009-Round3 | 43.56 | 8.70 |

Current age (months) 2002-Round1 | 95.28 | 3.65 |

Current age (months) 2009-Round3 | 179.82 | 3.97 |

Illness in the last three years 2002-Round1 | 0.14 | 0.35 |

Illness in the last three years 2009-Round3 | 0.25 | 0.43 |

Wealth index score 2002-Round1 | 0.38 | 0.22 |

Wealth index score 2009-Round3 | 0.51 | 0.21 |

# of times meal was taken on the previous day 2009-Round3 | 5.14 | 1.16 |

Male | 0.51 | 0.50 |

Birth order | 2.17 | 4.95 |

Mother’s height (cm) | 152.85 | 8.63 |

Mother’s age at child’s birth (year) | 24.73 | 6.41 |

Mother completed primary and above education | 0.47 | 0.50 |

Father completed primary and above education | 0.52 | 0.50 |

Rural | 0.65 | 0.48 |

The correlation matrix presented in

Conditional linear growth | Relative weight gain | Years of schooling | PPVT score | Math test score | |
---|---|---|---|---|---|

Conditional linear growth | 1.00 | ||||

Relative weight gain | 0.00(1.00) | 1.00 | |||

Years of schooling | 0.11(0.00) | -0.02(0.19) | 1.00 | ||

PPVT score | 0.13(0.00) | 0.03(0.09) | 0.36 (0.00) | 1.00 | |

Math test score | 0.10(0.00) | 0.00(1.00) | 0.54 (0.00) | 0.48 (0.00) | 1.00 |

Note: Standard errors are in parenthesis.

Years of schooling | PPVT score | Math test score | |
---|---|---|---|

Conditional linear growth (CLG) | 0.034 |
0.474 |
0.083 |

(0.008) | (0.100) | (0.018) | |

Village fixed effect | Yes | Yes | Yes |

R-squared | 0.16 | 0.10 | 0.23 |

Observations | 3,542 | 2,863 | 3,533 |

Relative weight gain (RWG) | -0.026 |
0.026 | -0.048 |

(0.009) | (0.107) | (0.020) | |

Village fixed effect | Yes | Yes | Yes |

R-squared | 0.16 | 0.09 | 0.23 |

Observations | 3,542 | 2,863 | 3,533 |

* p < 0.05,

** p < 0.01,

*** p < 0.001.

The results on RWG demonstrate that RWG has a significant negative effect on years of schooling and math test score. On an average, one kg deviation in RWG results in a decline in years of schooling by 0.026 years and math test score by 0.048 points. The effect of RWG on PPVT score is positive but the estimates are statistically significant. The negative effect of RWG on years of schooling and math test score is consistent with previous findings that show that conditional weight gain is associated with poor health and higher risks of metabolic and cardiovascular disease and thereby adversely affect years of schooling and IQ [

The estimates for heterogeneous analyses, based on the interaction model, are reported in

Conditional linear growth (CLG) | Relative weight gain (RWG) | |||||
---|---|---|---|---|---|---|

Years of schooling | PPVT score | Math test score | Years of schooling | PPVT score | Math test score | |

CLG |
0.033 |
0.264 | -0.0007 | - | - | - |

(0.016) | (0.206) | (0.037) | - | - | - | |

CLG |
-0.001 | 0.149 | -0.12 | - | - | - |

(0.015) | (0.199) | (0.035) | - | - | - | |

CLG |
0.022 | 0.074 | 0.034 | - | - | - |

(0.015) | (0.201) | (0.035) | - | - | - | |

CLG |
-0.0009 | -0.033 | -0.045 | - | - | - |

(0.016) | (0.208) | (0.036) | - | - | - | |

RWG |
- | - | - | -0.012 | 0.007 | 0.035 |

- | - | - | (0.017) | (0.213) | (0.040) | |

RWG |
- | - | - | 0.011 | -0.241 | 0.029 |

- | - | - | (0.017) | (0.208) | (0.039) | |

RWG |
- | - | - | -0.026 | 0.101 | 0.059 |

- | - | - | (0.019) | (0.240) | (0.048) | |

RWG |
- | - | - | 0.009 | 0.181 | 0.004 |

- | - | - | (0.018) | (0.224) | (0.042) |

^{@}

* p < 0.05,

** p < 0.01,

*** p < 0.001.

Quantile regression coefficient of CLG on educational outcome measures for the pooled data is shown in ^{th}) and middle (50^{th}) quantile of years of schooling has a statistically significant effect on years of schooling. The effect size (0.029) at the 25^{th} quantile is approximately 50% larger than the effect size at the 50^{th} quantile (0.014). However, there is no effect at the higher quantiles reflecting a weaker association between CLG and years of schooling at a higher level of schooling. Effect of CLG on PPVT score increases from 0.497 at the 10^{th} quantile to 0.573 at the 25^{th}quantile, thereafter quantile regression coefficient consistently declines up to 90^{th} quantile of PPVT score (0.188). Further, the estimated effects of CLG on math test score consistently increase from 0.060 to 0.093 between 10^{th} and 75^{th} quantiles. The estimated effects of CLG on math test score at the 75^{th} quantile are one and half times the magnitude of the estimated effect at the 10^{th} quantile.

Years of schooling | 0.008 | 0.029 |
0.014 |
0.000 | 0.000 |

(0.01) | (0.01) | (0.006) | (0.005) | (0.002) | |

PPVT score | 0.497 |
0.573 |
0.284 |
0.233 |
0.188 |

(0.201) | (0.142) | (0.105) | (0.098) | (0.087) | |

Maths Test score | 0.060 |
0.067 |
0.083 |
0.093 |
0.062 |

(0.018) | (0.017) | (0.022) | (0.026) | (0.027) |

^{@}

* p < 0.05,

** p < 0.01,

*** p < 0.001.

^{th} quantile (^{th} quantile of the math test score distribution. Overall, there is some evidence of stronger association at the lower distribution of outcome, particularly for CLG.

Years of schooling | -0.028 |
-0.020 |
-0.009 | 0.000 | 0.000 |

(0.011) | (0.005) | (0.001) | (0.002) | (0.002) | |

PPVT score | 0.145 | 0.009 | 0.051 | -0.110 | -0.039 |

(0.142) | (0.105) | (0.081) | (0.069) | (0.070) | |

Maths Test score | -0.010 | -0.040 |
-0.032 | -0.022 | -0.025 |

(0.023) | (0.02) | (0.029) | (0.027) | (0.030) |

^{@}

* p < 0.05,

** p < 0.01,

*** p < 0.001.

The present study is among the handful of studies that investigate the effects of CLG and RWG on educational attainment in four low- and middle-income countries (Ethiopia, India, Peru, and Vietnam). This study adds to the scant literature on the effect of conditional linear growth and relative weight gain on schooling achievement and test scores. With the recent growing evidence on the dual burden of undernutrition and obesity in developing countries, it is important to separate the effect of linear growth on educational attainment from the effect of weight gain on educational attainment. Using quantile regression and interaction models, we examine the heterogeneous effect of CLG and RWG on years of schooling, PPVT score, and math test score. Our study extends previous works by focusing on heterogeneity in the effects by sub-groups and by different quantiles of distribution of the years of schooling, PPVT score and math test score. Our findings will provide empirical evidence for the targeted intervention among the severely disadvantaged section of the population.

The correlation coefficient between CLG and RWG is zero, which shows that we have disentangled the effect of linear growth from the effect of weight gain on educational attainment. The results show that conditional linear growth has a significant positive effect on years of schooling, PPVT, and math-test score. The results show that 30 cm (12 inches) positive deviation in the height of the adolescent from the expected height as per mid-childhood height and weight will result in one more years of schooling, 14.2 and 2.5 points higher value of PPVT and math test score, respectively. Consistent with the previous literature, we also find that relative weight gain is negatively associated with educational measures. On average, one-kilogram positive deviation in adolescent weight as per mid-childhood height, weight and current height result in a decline in 0.026 years of schooling and 0.048 points math-test score. This implies that one-kilogram increase in RWG is associated with an increase of 0.026 and 0.048 in years of schooling and math test score, respectively. Results from the quantile regression model show evidence of heterogeneous effects; the magnitude of effects varies by the distribution and level of the dependent variables. Generally, the positive effects of height on educational outcome measures are more pronounced at the lower quantiles of years of schooling, PPVT, and math test score. However, no consistent pattern emerges for the distributional impacts of weight on educational outcomes.

Furthermore, nutritional supplementation up to three years of age has the maximum effect on educational attainment in adulthood [

Our study has a few limitations, including a lack of causal relationships and lack of information on birth weight and anthropometrics before age 8, and data pooling. First, conditional height and weight gain are likely to be endogenous as they are correlated with changes in household wealth or socio-economic status and with changes in access to health and educational infrastructures as well. It is possible that there may be observed (household income/wealth) or unobserved factors affecting both CLG/RWG and the educational outcomes. For example, any negative health shock that causes health and weight gain to slow may have negative impacts on school attendance and participation as well. In this study, we are unable to address the causality issue and unobserved heterogeneity, primarily due to the lack of suitable instrumental variable that could address the endogeneity in CLG/RWG. Second, previous studies have found that early life nutrition and birth outcomes are important determinants of height/weight gain and educational outcomes. The analytical model in this study does not control for the initial height, birthweight or any other indicator of early-life health stock. Third, the study currently analyzes data pooled from all the four young lives countries. It is likely that some of the variables are not comparable across countries because of the difference in the scale used. More importantly, it is important to test whether a similar association is observed across different countries. Therefore, to check the external validity of our findings, it would be interesting to analyze each country sample separately. However, it is likely that country-level sample may not have sufficient sample size to conduct these analyses, especially the quantile regression modeling.

As our study highlights that mid-childhood nutritional intervention targeted for malnourished children at a lower level of distribution, in rural areas, and among poor children might accomplish a higher level of educational outcome. Our analysis also indicates that the estimation of central tendencies (OLS and median regression) may be misleading and quantile regression estimation may provide more reliable information about the vulnerable target groups. Given that childhood malnutrition in low- and middle-income countries is highly correlated with socioeconomic characteristics, one of the key reforms needed is that the nutritional intervention should focus on children belonging to socio-economic deprived sections, especially in rural areas and children from poor households. Our results suggest that investment on these vulnerable groups would maximize the effect of early life health on educational outcome later in life.

The authors would like to thank the anonymous reviewers for comments and suggestions on earlier draft.