Hostname: page-component-594f858ff7-hf9kg Total loading time: 0 Render date: 2023-06-07T14:47:11.668Z Has data issue: false Feature Flags: { "corePageComponentGetUserInfoFromSharedSession": true, "coreDisableEcommerce": false, "corePageComponentUseShareaholicInsteadOfAddThis": true, "coreDisableSocialShare": false, "useRatesEcommerce": true } hasContentIssue false

Childhood stunting and subsequent educational outcomes: a marginal structural model analysis from a South African longitudinal study

Published online by Cambridge University Press:  26 August 2022

Lateef B Amusa
Department of Statistics, University of Ilorin, Ilorin, Nigeria
Annah Vimbai Bengesai*
College of Law and Management Studies, University of KwaZulu-Natal, 4000Durban, South Africa
Hafiz TA Khan
College of Nursing, Midwifery and Healthcare, University of West London, Brentford, UK
*Corresponding author: Email
Rights & Permissions[Opens in a new window]



To examine the association between childhood stunting and grade completion (as educational outcome) in South Africa.


Longitudinal study. Data were obtained using the National Income Dynamics Study over five waves (2008 to 2017). Children were tracked at wave 1 in 2008 until wave 5 in 2017 to determine their total years of schooling. We controlled for time-variant and time-varying confounding with a marginal structural model to estimate the associations between childhood stunting and subsequent grade completion.


Nationally representative study of South African households.


A total of 2629 children aged 2 and 3 years in 2008.


We observed a substantial decrease in the prevalence of stunting between wave 1 (28·2 %) and wave 4 (8·6 %). Our marginal structural model results suggest that childhood stunting was significantly associated with decreased odds (22 % less likely) of grade completion (OR = 0·78; 95 % CI: 0·40, 0·86; P = 0·015), while those who were only stunted during early childhood had a 29 % reduction in the odds of grade completion (OR = 0·71; 95 % CI: 0·51, 0·82; P = 0·020).


These findings underscore the fact that stunting is a significant predictor of academic achievement, whose effects might be long-lasting.

Research Paper
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (, which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
© The Author(s), 2022. Published by Cambridge University Press on behalf of The Nutrition Society

For many years, poor academic achievement has been a significant area of concern in many countries(Reference Wolde and Belachew1), prompting research that seeks to identify its correlates. Several factors that can lead to underperformance have been identified, including schooling quality, poverty, illness, age at enrolment, absenteeism and nutrition(Reference Banerjee2,Reference Spaull3) . However, for young children, nutritional deficiencies are among the significant factors affecting academic achievement(Reference Wolde and Belachew1). One possible manifestation of malnutrition is stunting (or low height-for-age), which some scholars argue is the best indicator of chronic undernutrition and a more accurate reflection of socio-economic inequalities(Reference Casale4,Reference de Onis and Branca5) .

The prevalence of stunting in Africa is approximately 29 %, 8 % higher than the global average of 21 %(Reference de Onis and Branca5,6) . The Eastern and Southern Africa regions account for the increasing share of the global total within the continent, with the rural population being the most vulnerable due to insufficient dietary intakes. Although South Africa has undergone a nutrition transition characterised by the prevalence of obesity and non-communicable disease similar to most developed countries(Reference Said-Mohamed, Micklesfield and Pettifor7), stunting remains prevalent. The 2020 nutrition status report indicated that over 1·5 million (three out of ten) South African children are stunted(8), surpassing countries like Gabon, Libya and Egypt, which have similar Human Development Indices to South Africa. These high stunting rates have also persisted, and efforts to rectify the situation through school-based nutrition programmes have proven ineffective(Reference Said-Mohamed, Micklesfield and Pettifor7). This is not surprising given that by the time children reach school-going age, the effects of stunting are unlikely to be reversible.

Certainly, stunting is a significant public health issue associated with increased morbidity, unhealthy development and mortality from infections(Reference de Onis and Branca5). For instance, greater risk of metabolic syndrome has been found among adults who were stunted in early childhood(Reference Henriques, Teixeira and Cardoso9,Reference Farah, Nour and Endris10) . This is because malnutrition in the early years can trigger ‘permanent epigenetic changes in metabolism’(Reference Prendergast and Humphrey11), which, although necessary for survival, can result in rapid weight gain and obesity, increased risk hypertension and high blood sugar. For South Africa, this association cannot be ignored, as it poses serious public health consequences in a country already grappling with the burden of HIV and AIDS, among other diseases(Reference Tydeman-Edwards, Van Rooyen and Walsh12). Apart from the increased morbidity, it is believed that poor linear growth observed in the first 24 months of life is associated with delayed cognitive development and poor academic achievement through a shared set of determinants (such as poor nutrition, repeated infection and inadequate psychosocial stimulation)(Reference Leroy and Frongillo13), leading to adverse outcomes affecting the entire lifecycle(Reference Dewey and Begum14), and even persisting into the labour force(Reference Currie and Vogl15). For instance, using the Young Lives (YL) from India, Ethiopia, Peru and Vietnam, and a multivariable probit regression approach, Crookston et al.(Reference Crookston, Schott and Cueto16) found an inverse relationship between stunting at the age of 1 year and achievement in mathematics, reading comprehension, and receptive vocabulary. Casale(Reference Casale17), who used data from four waves of the National Income Dynamics Study (NIDS, South Africa), found that children who had a low height-for-age in the early period were likely to complete fewer years of schooling by the age of 14 years than their peers who did not experience stunting. The South African Birth to Twenty Cohort Study also suggests that not just the quantity and quality of nutrition make a difference, but also the timing(Reference Desmond and Casale18,Reference Shisana, Labadarios and Rehle19) . Even if their nutritional status improves, children who have been stunted in early childhood never quite catch up in academic outcomes(Reference Casale and Desmond20). In this way, malnutrition can hinder children from reaching their full potential. Although some studies have shown that this association is counteracted when controlling for psychosocial and environmental factors(Reference Leroy and Frongillo13), for a country like South Africa with one of the most unequal education systems globally(Reference Spaull and Jansen21), stunting adds an additional layer that cannot be ignored.

Undoubtedly, the adverse effects of stunting in early childhood are well documented. However, some studies have questioned the stability of stunting over time, arguing that recovery is possible beyond 2 years and in early adolescence(Reference Prentice, Ward and Goldberg22). Proponents of this view have used different approaches to support the claim of stunting recovery. For instance, Faye et al.(Reference Faye, Fonn and Levin23), who used the height-for-age Z-score (HAZ) slope modelling to estimate changes in linear growth, found that the incidence of recovery from stunting was approximately 45 %. In Casale’s study(Reference Casale4), a sample aged 24–36 months at baseline was used, given that the prevalence of stunting often increases between birth and 2 years, peaking somewhere between the ages of 2 and 3 years. Thus, using a starting point for stunting too early can lead to misreporting of catch-up growth. However, this perspective, referred to in the literature as catch-up growth or recovery from stunting, has become contentious in nutrition research. For instance, Leroy et al.(Reference Leroy, Frongillo and Dewan24) questioned the validity of this claim, arguing that both the methods and the measures often used are not adequate to estimate stunting recovery. Further, they recommend that catch-up growth can only happen under certain conditions, such as when growth-inhibiting conditions are controlled.

Considering these debates and adopting a time-varying confounding approach, this study utilised data from a population-based longitudinal study to measure the time-dependent confounding in the association between childhood stunting and academic achievement. We compared standard longitudinal methods with the marginal structural modelling approach of estimating causal effects using the inverse probability of treatment (IPT) weighting(Reference Robins, Hernan and Brumback25). Marginal structural models (MSM) provide a framework for estimating time-varying exposures while controlling for both time-invariant and time-varying confounders that may be affected by previous exposures.

Given that stunting is an outcome of many factors, most of which are associated with unfavourable socio-economic and household circumstances(Reference Reyes, Pérez-Cuevas and Sandoval26), we also included some of these and examined their interplay with stunting and educational attainment. Risk factors that have been identified in many studies include limited access to nutritional food, comorbidity of stunting and other illnesses(Reference Kinyoki, Kandala and Manda27), parental education, family income, among others(Reference Mansur, Afiaz and Hossain28). For instance, low-income families might have difficulties obtaining food, resulting in nutritional deficiencies in young children. In some cases, linear growth can falter due to other illnesses, and this co-occurrence can lead to profound negative development outcomes(Reference Orsango, Loha and Lindtjørn29). Several studies have also demonstrated the effect of parental education, particularly maternal education, on children’s outcomes, with maternal education considered to have far-reaching effects than any other determinant(Reference Iftikhar, Bari and Bano30). This is because educated parents can access health information, including following recommended feeding schemes. Although these correlates have been observed in many studies, their effect is likely to vary in space and time. Hence, there is a need for context-specific studies to provide deeper insight into their interplay with stunting and educational attainment.



We used data from the five waves (2008–2017) of the NIDS. The NIDS, a nationally representative sample of the South African population, is a biennial longitudinal survey of the dynamic structure of households and changes in their health and well-being(Reference Woolard, Leibbrandt and De Villiers31). The NIDS uses a stratified, two-stage cluster sample design. Stratification was done at the district council level, while clustering was employed using the primary sampling unit. The first wave of the NIDS, which was conducted in 2008, comprised 26 776 successfully interviewed persons (95 % response rate) from 7296 responding households (69 % response rate). The same people were followed and re-interviewed, with new household members being included in the subsequent waves. The overall attrition rate between waves 1 and 2, waves 2 and 3, waves 3 and 4 was 21·95 %, 15·82 %, and 13·75 %, respectively(Reference Chinhema, Brophy and Brown32). Further details regarding the NIDS study design can be found in a published report(Reference Leibbrandt, Woolard and de Villiers33).

This study examines the impact of childhood stunting of children on subsequent educational outcomes, measured by grade completion. We thus followed a cohort of children aged at most 3 years in wave 1 (2008) until wave 5 (2017), when the youngest cohort members were at least aged 9 years and the oldest aged 12 years. This resulted in an eligible sample of 2629 children. Across all five waves of data collection, 830 children were not successfully interviewed or lost to follow-up in at least one of the waves (wave 1: n 67; wave 2: n 454; wave 3: n 374; wave 4: n 366; and wave 5: n 438). Further, we dropped those who were not successfully interviewed in any of the five waves (n 11), reducing the sample size to 2618.


The NIDS data contain a wide range of information on, but not limited to, children’s nutritional, anthropometric and educational outcomes. Respondents provided self-reported and measured data both at baseline and follow-up.

Outcome variable

In this study, grade completion was the outcome of interest. We used the available information on the highest grade completed by wave 5. By South African educational standards, a child must start Grade 1 in the year they turn 7. Thus, when observed in wave 5, all the children in our sample should be enrolled in school, with the youngest cohort (aged 9 years) having completed Grade 3 (and enrolled in Grade 4) if (i) they were on time and (ii) had progressed one grade per year. Thus, we quantified for analysis with a binary indicator of whether or not a child completed Grade 3 or higher in 2017 (wave 5).

Exposure variable

In our analyses, we defined the exposure variable using a binary indicator of stunting of a given child for a particular time point in the study period. We measured childhood stunting in this study by the z-score of the child’s HAZ. HAZ and other anthropometric indicators were calculated using the WHO international child growth standards and references for children aged under 5 years and adolescents(Reference Onis, Onyango and Borghi34,35) . Based on these growth standards, we defined a stunted child as having a HAZ less than two standard deviations (HAZ < –2) below the median value of a healthy reference population. To avoid inflated z-scores, the NIDS z-scores used months as the unit of analysis. This is in contrast to the zanthro macro scores where the default unit of analysis is the year and a 2-year-old is considered to be 2 years 0 months old. Unfortunately, this leads to children being compared to a reference population that might be up to 364 d younger than they are. Rather, in the NIDS, children aged 30 months (2 years and 6 months), for instance, were considered to be 2-year-old(Reference Chinhema, Brophy and Brown32). The NIDS data were also pre-cleaned, and biologically implausible values were coded as missing in line with WHO guidelines(Reference Casale4).


We controlled for possible confounders for the relationship between stunting status and grade completion outcome. The confounders were chosen based on theory and previous research findings, and their potential relationships were depicted in a causal diagram (see Fig. 1). Time-invariant covariates in the analysis include gender, race (African and others) and the child’s birth weight (kg). We further included the highest school grade completed by the parents. Mother and father’s education were combined into a composite variable, parental education, based on who completed a higher number of years of education.

Fig. 1 Simplified directed acyclic graphs of the hypothetical causal relationship between time-dependent stunting on subsequent educational attainment. Numeric subscripts 0, 1, 2 and 3 denote the baseline period (wave 1), 1st, 2nd and 3rd follow-ups, respectively. V represent the time-invariant baseline covariates (gender, race, birth weight and parental education); A represents stunting status; L represents the time-varying covariates (household receipt of government grants, child lived with both parents, child has any illness or disability, type of residential area, household income and household size) at time points other than the baseline. Panel A: stunting effect was captured only for the early childhood years (within the first 5 years); Panel B: stunting effect was captured till the later childhood years (as late as the age of 9 years)

Fig. 2 Educational attainment of cohort members in 2017 (aged 9–12 years)

The following variables were measured throughout the study and included as time-varying covariates: an indicator for whether both parents live with the child, the number of household residents, an indicator for whether a child has/had any serious illnesses or disabilities, type of residential area (rural and urban), natural log of per capita household income (Rands) and an indicator for whether the household receives a child support grant.

Statistical analysis

Marginal structural models are a suitable statistical approach for longitudinal data to control the confounding effect of time-dependent confounders on time-varying treatments or exposures. Instead of applying IPT weights to each subject, MSM build upon the widely used propensity score weighting technique in a two-stage process. In the first stage, IPT weights are estimated for every subject-time point in the study period. These weights create a pseudo-population where treatment is unconfounded by subject-specific characteristics(Reference Hernán, Brumback and Robins36). In the second stage, a weighted pooled regression model including only the exposure history can estimate the causal effect of the exposure of interest on the study outcome. MSM lead to unbiased estimates under the following assumptions: exchangeability (no unmeasured confounders); consistency (no ambiguity in defining exposure); positivity (nonzero probability of each possible exposure value given all combinations of covariates) and correct model specification(Reference Cole and Hernán37).

Since children stunting status is a time-varying variable, we adopted inverse probability weighted estimation of MSM to provide an unbiased causal effect estimate of childhood stunting on grade completion. Further, as shown in Fig. 1, MSM appropriately adjusts the confounding effects that vary over time from one wave to another, including some of these confounders with both a confounding and mediating role in the causal pathway linking the exposure and the outcome.

We applied the stabilised version of the IPT weights (SW) to reduce variability and produce more efficient estimates(Reference Robins, Hernan and Brumback25) as follows:

$${\rm{SW}}\left( t \right) = \mathop \prod \limits_{t = 1}^T {{P\;{\rm{(}}{A_{t\;}} = \;{a_t}{\rm{|}}\;{{\bar A}_{t - 1}}, = \;{{\bar a}_{t - 1}},\;\;\;V\; = \;v)\;\;} \over {P\;\;{\rm{(}}{A_{t\;}} = \;{a_t}{\rm{|}}\;{{\bar A}_{t - 1}}, = \;{{\bar a}_{t - 1}},\;\;{{\bar L}_t} = \;{{\bar l}_t})\;\;}}$$

${A_t}$ denotes the time-varying exposure at time t, ${A_{t - 1}}$ denotes the exposure history before time t, while ${\bar L_t}$ are the time-dependent covariates through time t, and V represents the vector of baseline covariates.

To estimate the denominator of the stabilised weights, we fitted a binary logistic regression of stunting status at time t ( ${A_t}$ ) on previous times stunting status ( ${\bar A_{t - 1}}$ ), $\overline {{L_t}}$ and $V$ . We similarly estimated the numerator, apart from excluding the time-varying covariates. To assess violations of the positivity and model specification assumptions, we verified that the mean of the stabilised inverse probability weights was close to 1·0, with no extreme weights. To further confirm non-violation of positivity, we checked that no particular subgroups in the sample were rarely or never in the exposed group.

We fitted two separate models based on the corresponding hypothetical scenarios and causal pathways depicted in Fig. 1(a) and (b), respectively. We assumed in the first scenario that the stunting impact was only observed in the early years (within the first 5 years), while in the second scenario, we assumed that the stunting effect was captured till the later years (as late as the age of 9 years). Our MSM were logit models, which allow the estimation of OR. We fitted weighted (using the IPT weights) logistic regression models to estimate the probability of grade completion, comparing stunted and non-stunted children. Standard errors were cluster-robust to account for clustering observations within subjects over time(Reference Williams38). For comparison, we estimated the conventional unweighted regression model (both unadjusted and adjusted) to assess the magnitude of time-dependent confounding.

Missing value analysis

There was at least one missing value for some variables, including whole wave missingness. We assumed that the data were missing at random and adopted multiple imputation via chained equations(Reference White, Royston and Wood39) to allow that the missingness mechanisms might depend on measured covariates. We performed five imputations and applied Rubin’s rule(Reference Rubin40) to get combined estimators for the risk ratios, including CI from the imputed data. In a sensitivity analysis, we repeated the analyses of the fitted models with data on complete cases to evaluate how results compare with those from our main analyses (Appendix Table 1, Supplementary file).

Apart from multiple imputations, which were implemented in Stata due to ease of use, all the analyses were carried out in R version 4.0.2(41).


The mean birth weight of the cohort members was 3·1 kg (±0·6 kg), most of them were African (83·8 %), and a little above half were males (50·5 %). Their parents’ number of years of education was 10·3 years on average (±2·8 years). At baseline, 28·2 % of the children were stunted, the prevalence of which decreased substantially over the three follow-ups, decreasing to 8·6 % in wave 4. Household income increased over time, with a mean log of income of 7·7 Rands at baseline increasing to 8·4 Rands at wave 4. The proportion of children whose parents lived with them and those who lived in the rural area decreased slightly over time (Table 1).

Table 1 Sample characteristics, NIDS waves 1–4

NIDS, National Income Dynamics Study.

All non-missing data are presented.

* Values expressed as n (%).

b Values expressed as mean (sd).

At baseline, in comparison to children who were not stunted, children who had stunted growth were more likely to be males (P < 0·01), to have a smaller birth weight (P < 0·001) and belong to households with smaller income (P < 0·001) (Table 2). Figure 1 shows the distribution of educational attainment of the sample at wave 5. The mean years of education completed (range: 0–8) was 4·0 (± 1·4 years). Only a small proportion of the children (12·6 %) had not completed Grade 2 or higher in 2017.

Table 2 Distribution of baseline covariates according to stunting status

All non-missing data are presented.

** P-value < 0·01.

*** P-value < 0·001

Values expressed as n (%).

Values expressed as mean (sd).

Table 3 shows the OR estimates for stunting status on the children’s educational attainment. In all the models fitted, we found that stunting, compared with not stunting, was associated with an overall reduction in the odds of grade completion (completing Grade 5 or higher in 2017). Our MSM indicated that comparing children who had stunted growth with children who were not stunted, there was a 22 % significant reduction in the odds of grade completion (OR = 0·78; 95 % CI: 0·40, 0·86). Similar MSM estimates were observed for stunting effects limited to early childhood stunting only (Model B). Stunting was associated with a 29 % significant reduction in the odds of grade completion (OR = 0·71; 95 % CI: 0·51, 0·82).

Table 3 Associations between stunting and subsequent educational attainment

Model A: controlled for stunting effect from early to late childhood.

Model B: controlled for stunting effect from early childhood only.

* Marginal structural model (MSM) estimated from pooled logistic regression model (inverse probability of treatment weighted), controlling for the baseline and time-varying covariates listed in Table 1.

Shown for comparison purposes only. Unadjusted model included only the time-varying intercept and stunting status.

Shown for comparison purposes only. Adjusted model used the same pooled logistic regression model used in the MSM but without inverse probability of treatment weighting.

Slightly weaker associations were observed for the unweighted pooled logistic regression model not adjusting for time-dependent confounding. The unadjusted OR of grade completion was 0·67 (95 % CI: 0·53, 0·85), and the addition of baseline covariates to the model decreased the OR to 0·73 (95 % CI: 0·57, 0·94), which still suggests a detrimental effect of stunting. Effect estimates were consistent with what was observed for stunting effects limited to early childhood stunting only (Model B): OR 0·63 (95 % CI: 0·49, 0·80) in the crude model and OR 0·68 (95 % CI: 0·53, 0·88) in the multivariable-adjusted model.

Our results for sensitivity analysis of missing values produced identical results to our main analyses across models (Appendix A). However, the effect estimates were more imprecise with relatively wider CI and insignificant estimates. For example, the MSM analysis for the association between stunting and grade completion produced the following estimates for both early and childhood stunting: OR = 0·79 (95 % CI: 0·50, 1·09) and early childhood stunting only: OR = 0·80 (95 % CI: 0·46, 1·31). Results from our conventional unweighted models were further consistent.


Using the NIDS data, this study examined stunting trajectories over time and academic achievement, measured by grade completion. To the best of our knowledge, this is the first study to utilise a marginal structural modelling approach, which controls time-dependent confounding, to examine the relationship between stunting and educational outcomes in South Africa. Although there were minor differences in our MSM estimates and the conventional longitudinal methods used in other studies, the former is recommended when time-dependent confounding is likely, given its theoretically solid appeal.

Using data from five waves of the NIDS study, we found that a substantial proportion of children in our sample were stunted at baseline, and this figure decreased over 6 years. These findings are in line with statistics from other South African studies, which have estimated stunting to be between 26 and 38 %(Reference Desmond and Casale18,Reference Shisana, Labadarios and Rehle19) , including the 2016 South African Demographic and Health Survey, whose estimate for under-5 stunting was approximately 27 %(42). In essence, the findings confirm that stunting in South Africa remains prevalent and continues to be an alarming burden(Reference Said-Mohamed, Micklesfield and Pettifor7).

Regarding the change in stunting, similar trends have been observed in previous South African(Reference Casale4,Reference Shisana, Labadarios and Rehle19) as well as global studies(Reference Georgiadis, Benny and Duc43,Reference Zhang, Undurraga and Zeng44) . For example, using the National Health and Nutrition Examination Survey, Shisana, Labadarios(Reference Shisana, Labadarios and Rehle19) found that stunting changes from 26 % for girls and 27 % for boys aged 0–3 years to approximately 10 % for girls and 14 % for boys aged 4–6 years. Using the NIDS data, Desmond and Casale(Reference Desmond and Casale18) also found evidence of changes in stunting, with over 80 % having recovered from stunting by 4–6 years. In Zhang’s(Reference Zhang, Undurraga and Zeng44) study, the authors found that although there was some evidence of catch-up growth in HAZ, most of the children in their sample remained in the same HAZ categories. However, we take caution of the earlier arguments against interpreting this improvement as evidence of catch-up growth(Reference Leroy, Frongillo and Dewan24). Instead, these changes indicate that the child’s HAZ is less of an outlier than when it was previously measured(Reference Desmond and Casale18). This potential improvement does not overlook the importance of the first 1000 d in child growth; instead, they have implications for interventions targeting the nutritional needs of older children such as the school nutrition programme in South Africa(Reference Shepherd and Mahohlwane45).

Our consistent results across all the considered estimation techniques, including the MSM, reinforced our hypothesis that childhood malnutrition significantly and adversely affected children’s educational outcomes, measured by grade completion in this study. Results from our MSM mainly show that stunted children were 22 % less likely to have completed the number of grades expected for their age. A further restriction of stunting to early childhood (only the first five years) produced even a more substantial effect: a 29 % reduction in the odds of grade completion. These findings are comparable to a recent analysis of the NIDS data(Reference Casale, Desmond and Richter46), which noted lower educational attainment and the risk of grade repetition among stunted children.

There are several reasons why stunting is negatively associated with educational outcomes. Stunted children are more likely to enter school later than their peers due to a combination of nutritional deficits and delayed cognitive development(Reference Leroy and Frongillo13,Reference Daniels and Adair47) . Reporting on Tanzania, Alderman, Hoogeveen(Reference Alderman, Hoogeveen and Rossi48) noted that stunted children are often considered not ready to start school at the minimum age. Apart from delayed enrolment, stunted children are also likely to repeat grades owing to poor academic performance or missing classes due to food shortages(Reference Wolde and Belachew1,Reference Casale, Desmond and Richter46) . Stunting has also been associated with poorer psychosocial functioning, including being less happy, fussy and more apathetic(Reference Walker, Chang and Powell49). These factors have also been inversely associated with educational attainment(Reference Wolde and Belachew1,Reference Casale4) . Thus, although stunting is not part of the ‘mechanistic path leading to poor cognitive development’,(Reference Leroy and Frongillo13) it still increases the educational disadvantage of stunted children from entry and reduce their likelihood of being in the correct grade for age throughout the schooling cycle.

These findings are informative for South Africa as they confirm that stunting fuels educational inequality. In a country with one of the highest educational inequalities globally, stunting adds another layer of inequality that cannot be easily ignored. Although strides have been made to improve childhood nutrition through the school nutrition programme, the COVID-19 pandemic has worsened the plight of undernourished children(Reference Van der Berg, Patel and Bridgman50). The intermittent closure of many schools, including ECDs as the country works towards containing the pandemic, have meant that millions of children who have been beneficiaries of school-based nutrition programmes have been moving in and out of food security during the pandemic(Reference Zhang, Undurraga and Zeng44). Therefore, issues of undernutrition will remain relevant in South Africa for some time.

Although this study has many strengths, for example, the longitudinal nature of the data and the time-varying approach, it is not without limitations. First, given that HAZ is a relative measure of stunting, there is a possibility that this measure might overestimate any height improvements. Second, we acknowledge that other factors might be associated with both stunting and grade completion beyond what we included in our models. Thus, the possibility of omitted variable bias should be taken into consideration, making it challenging to investigate any casual links. As such, only associations were estimated.

Despite these limitations, our findings are still relevant in South Africa and the scholarship because they confirm prior studies that have shown that stunting is still prevalent in South Africa and a cause for concern. Moreover, our findings indicate that even with the possibility of stunting status varying with time, poor linear growth in the first 36 months is still negatively associated with educational outcomes. Therefore, greater attention must be placed on understanding the mechanisms which underlie linear growth failure as well as interventions that prevent its incidence from the onset. Interventions which also target individual and household-level characteristics that make children vulnerable to stunting and poor educational attainment are also needed.


Acknowledgements: The authors wish to thank the Southern Africa Labour and Development Research Unit for permission to use the National Income Dynamics Data. Financial support: This research received no specific grant from any funding agency, commercial or not-for-profit sectors. Authorship: L.B.A. conceived and designed the study. A.V.B. provided the study data. L.B.A. and A.V.B. contributed to the methodology and data analysis. H.T.A.K. played a supervisory role in the review and editing of the manuscript. All authors provided input into the final version of the manuscript. Ethics of human subject participation: The study used secondary data which is publicly available from the Southern Africa Labour and Development Research Unit. However, permission to use the data was sought, and one of the authors is registered to use the data.

Conflict of interest:

There are no conflicts of interest.

Supplementary material

For supplementary material/s referred to in this article, please visit


Wolde, T & Belachew, T (2019) Chronic undernutrition (stunting) is detrimental to academic performance among primary schools of adolescent children: a randomized cross sectional survey in Southern Ethiopia. BMC Res Note 12, 142.CrossRefGoogle ScholarPubMed
Banerjee, PA (2016) A systematic review of factors linked to poor academic performance of disadvantaged students in science and maths in schools. Cogent Educ 3, 1178441.CrossRefGoogle Scholar
Spaull, N (2013) Poverty & privilege: primary school inequality in South Africa. Int J Educ Dev 33, 436447.CrossRefGoogle Scholar
Casale, D (2020) Recovery from stunting in early childhood and subsequent schooling outcomes: evidence from NIDS Waves 1–5. Dev South Afr 37, 483500.CrossRefGoogle Scholar
de Onis, M & Branca, F (2016) Childhood stunting: a global perspective. Matern Child Nutr 12, 1226.CrossRefGoogle ScholarPubMed
Global Nutrition Report (2020) 2020 Global Nutrition Report: Action on Equity to End Malnutrition. Bristol: Development Initiatives.Google Scholar
Said-Mohamed, R, Micklesfield, LK, Pettifor, JM et al. (2015) Has the prevalence of stunting in South African children changed in 40 years? A systematic review. BMC Public Health 15, 110.CrossRefGoogle ScholarPubMed
UNICEF, South Africa (2020) South African Nutritional Brief 2020. (accessed July 2021).Google Scholar
Henriques, A, Teixeira, V, Cardoso, HF et al. (2018) The influence of stunting on obesity in adulthood: results from the EPIPorto cohort. Public Health Nutr 21, 18191826.10.CrossRefGoogle ScholarPubMed
Farah, AM, Nour, TY, Endris, BS et al. (2021) Concurrence of stunting and overweight/obesity among children: evidence from Ethiopia. PLoS One 16, e0245456-e.CrossRefGoogle ScholarPubMed
Prendergast, AJ & Humphrey, JH (2014) The stunting syndrome in developing countries. Paediatr Int Child Health 34, 250265.CrossRefGoogle ScholarPubMed
Tydeman-Edwards, R, Van Rooyen, FC & Walsh, CM (2018) Obesity, undernutrition and the double burden of malnutrition in the urban and rural southern Free State, South Africa. Heliyon 4, e00983.CrossRefGoogle Scholar
Leroy, JL & Frongillo, EA (2019) Perspective: what does stunting really mean? A critical review of the evidence. Adv Nutr 10, 196204.CrossRefGoogle Scholar
Dewey, KG & Begum, K (2011) Long-term consequences of stunting in early life. Matern Child Nutr 7, 518.CrossRefGoogle ScholarPubMed
Currie, J & Vogl, T (2013) Early-life health and adult circumstance in developing countries. Ann Rev Econ 5, 136.CrossRefGoogle Scholar
Crookston, BT, Schott, W, Cueto, S et al. (2013) Postinfancy growth, schooling, and cognitive achievement: young Lives. Am J Clin Nutr 98, 15551563.CrossRefGoogle ScholarPubMed
Casale, D (2017) The Relationship between Early Child Nutrition and Schooling Outcomes: Evidence from South Africa’s First Longitudinal National Household Survey. International Population Conference; September 2017; Cape Town. (accessed July 2021).Google Scholar
Desmond, C & Casale, D (2017) Catch-up growth in stunted children: definitions and predictors. PLoS One 12, e0189135-e.CrossRefGoogle ScholarPubMed
Shisana, O, Labadarios, D, Rehle, T et al. (2014) The South African National Health and Nutrition Examination Survey, 2012: SANHANES-1: The Health and Nutritional Status of the Nation. Cape Town: HSRC Press.Google Scholar
Casale, D & Desmond, C (2016) Recovery from stunting and cognitive outcomes in young children: evidence from the South African Birth to Twenty Cohort Study. J Dev Orig Health Dis 7, 163171.CrossRefGoogle ScholarPubMed
Spaull, N & Jansen, J (2019) South African Schooling: The Enigma of Inequality A Study of the Present Situation and Future Possibilities. New York: Springer.Google Scholar
Prentice, AM, Ward, KA, Goldberg, GR et al. (2013) Critical windows for nutritional interventions against stunting. Am J Clin Nutr 97, 911918.CrossRefGoogle ScholarPubMed
Faye, CM, Fonn, S & Levin, J (2019) Factors associated with recovery from stunting among under-five children in two Nairobi informal settlements. PLoS One 14, e0215488.CrossRefGoogle ScholarPubMed
Leroy, J, Frongillo, E, Dewan, P et al. (2020) Can children catch up from the consequences of undernourishment? Evidence from child linear growth, developmental epigenetics, and brain and neurocognitive development. Adv Nutr 11, 10321041.CrossRefGoogle ScholarPubMed
Robins, JM, Hernan, MA & Brumback, B (2000) Marginal structural models and causal inference in epidemiology. Epidemiology 11, 550560.CrossRefGoogle ScholarPubMed
Reyes, H, Pérez-Cuevas, R, Sandoval, A et al. (2004) The family as a determinant of stunting in children living in conditions of extreme poverty: a case-control study. BMC Public Health 4, 57.CrossRefGoogle ScholarPubMed
Kinyoki, DK, Kandala, N-B, Manda, SO et al. (2010) Assessing comorbidity and correlates of wasting and stunting among children in Somalia using cross-sectional household surveys: 2007 to 2010. BMJ Open 6, e009854-e.CrossRefGoogle Scholar
Mansur, M, Afiaz, A & Hossain, MS (2021) Sociodemographic risk factors of under-five stunting in Bangladesh: assessing the role of interactions using a machine learning method. PLoS One 16, e0256729.CrossRefGoogle ScholarPubMed
Orsango, AZ, Loha, E, Lindtjørn, B et al. (2021) Co-morbid anaemia and stunting among children 2–5 years old in southern Ethiopia: a community-based cross-sectional study. BMJ Paediatrics Open 5, e001039.CrossRefGoogle ScholarPubMed
Iftikhar, A, Bari, A, Bano, I et al. (2017) Impact of maternal education, employment and family size on nutritional status of children. Pak J Med Sci 33, 14011405.CrossRefGoogle ScholarPubMed
Woolard, I, Leibbrandt, M & De Villiers, L (2010) The South African national income dynamics study: design and methodological issues. Studies Econ Econometric 34, 724.Google Scholar
Chinhema, M, Brophy, T, Brown, M et al. (2016) National Income Dynamics Study Panel User Manual. Cape Town: Southern Africa Labour and Development Research Unit.Google Scholar
Leibbrandt, M, Woolard, I & de Villiers, L (2009) Methodology: Report on NIDS Wave 1. Technical Paper. (accessed July 2021).Google Scholar
Onis, MD, Onyango, AW, Borghi, E et al. (2007) Development of a WHO growth reference for school-aged children and adolescents. Bull World Health Organ 85, 660667.Google ScholarPubMed
World Health Organization (2006) WHO Child Growth Standards: Length/Height-for-Age, Weight-for-Age, Weight-for-Length, Weight-for-Height and Body Mass Index-for-Age: Methods and Development. Geneva: World Health Organization; available at (accessed July 2021).Google Scholar
Hernán, , Brumback, B & Robins, JM (2000) Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 1, 561570.CrossRefGoogle Scholar
Cole, SR & Hernán, MA (2008) Constructing inverse probability weights for marginal structural models. Am J Epidemiol 168, 656664.CrossRefGoogle ScholarPubMed
Williams, RL (2000) A note on robust variance estimation for cluster-correlated data. Biom 56, 645646.CrossRefGoogle ScholarPubMed
White, IR, Royston, P & Wood, AM (2011) Multiple imputation using chained equations: issues and guidance for practice. Stat Med 30, 377399.CrossRefGoogle ScholarPubMed
Rubin, DB (2004) Multiple Imputation for Nonresponse in Surveys. New York: Wiley.Google Scholar
R Core Team (2020) R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.Google Scholar
National Department of Health (NDoH), Statistics South Africa (Stats SA), South African Medical Research Council (SAMRC) et al. (2020) South Africa Demographic and Health Survey 2016 Key Findings. Pretoria, South Africa; Rockville, MD: NDoH, Stats SA, SAMRC, and IC.Google Scholar
Georgiadis, A, Benny, L, Duc, LT et al. (2017) Growth recovery and faltering through early adolescence in low- and middle-income countries: determinants and implications for cognitive development. Soc Sci Med 179, 8190.CrossRefGoogle ScholarPubMed
Zhang, R, Undurraga, EA, Zeng, W et al. (2016) Catch-up growth and growth deficits: nine-year annual panel child growth for native Amazonians in Bolivia. Ann Hum Biol 43, 304315.CrossRefGoogle ScholarPubMed
Shepherd, D & Mahohlwane, N (2021) The Impact of COVID-19 in Education – More Than a Year of Disruption. National Income Dynamics Study (NIDS) – Coronavirus Rapid Mobile Survey (CRAM). Cape Town. (accessed July 2021).Google Scholar
Casale, D, Desmond, C & Richter, LM (2020) Catch-up growth in height and cognitive function: why definitions matter. Econ Hum Biol 37, 100853.CrossRefGoogle ScholarPubMed
Daniels, MC & Adair, LS (2004) Growth in young Filipino children predicts schooling trajectories through high school. J Nutr 134, 14391446.CrossRefGoogle ScholarPubMed
Alderman, H, Hoogeveen, H & Rossi, M (2009) Preschool nutrition and subsequent schooling attainment: longitudinal evidence from Tanzania. Economic Development and Cultural Change 57, 239260.CrossRefGoogle Scholar
Walker, SP, Chang, SM, Powell, CA et al. (2006) Effects of psychosocial stimulation and dietary supplementation in early childhood on psychosocial functioning in late adolescence: follow-up of randomised controlled trial. BMJ 333, 472.CrossRefGoogle ScholarPubMed
Van der Berg, S, Patel, L & Bridgman, G (2020) Hunger in South Africa during 2020: results from Wave 3 of NIDS-CRAM.. National Income Dynamics Study (NIDS). (accessed July 2021).Google Scholar
Figure 0

Fig. 1 Simplified directed acyclic graphs of the hypothetical causal relationship between time-dependent stunting on subsequent educational attainment. Numeric subscripts 0, 1, 2 and 3 denote the baseline period (wave 1), 1st, 2nd and 3rd follow-ups, respectively. V represent the time-invariant baseline covariates (gender, race, birth weight and parental education); A represents stunting status; L represents the time-varying covariates (household receipt of government grants, child lived with both parents, child has any illness or disability, type of residential area, household income and household size) at time points other than the baseline. Panel A: stunting effect was captured only for the early childhood years (within the first 5 years); Panel B: stunting effect was captured till the later childhood years (as late as the age of 9 years)

Figure 1

Fig. 2 Educational attainment of cohort members in 2017 (aged 9–12 years)

Figure 2

Table 1 Sample characteristics, NIDS waves 1–4

Figure 3

Table 2 Distribution of baseline covariates according to stunting status

Figure 4

Table 3 Associations between stunting and subsequent educational attainment

Supplementary material: File

Amusa et al. supplementary material

Amusa et al. supplementary materia

Download Amusa et al. supplementary material(File)
File 67 KB