Recovery from stunting in early childhood and subsequent schooling outcomes: Evidence from NIDS Waves 1–5

ABSTRACT

While an extensive literature documents the negative effects of stunting on children’s developmental potential, there is far less evidence on whether a recovery from stunting in childhood – often referred to as ‘catch-up growth’– helps mitigate the negative effects of early growth retardation. This paper explores the association between catch-up growth in early childhood and subsequent schooling outcomes using data from the first five waves of NIDS. The findings suggest that children who recovered from stunting in early childhood go on to complete fewer years of schooling compared to their non-stunted counterparts, driven in large part by a slower progression through school. However, there also appear to be heterogeneous effects depending on the extent of recovery; the small proportion of children who recovered substantially exhibit similar schooling outcomes to the non-stunted group. These results have important implications for the timing of nutritional (and other) investments in the early childhood period.

KEYWORDS:

• Stunting
• catch-up growth
• early childhood
• schooling outcomes
• South Africa

Acknowledgements

The author would like to thank Chris Desmond and Linda Richter for their collaboration on earlier work on this topic using the Birth to Twenty data, which helped inform this study.

Disclosure statement

No potential conflict of interest was reported by the author(s).

ORCID

Daniela Casale http://orcid.org/0000-0003-1337-1196

Notes

1 While the terms ‘catch up’ and ‘catch-up growth’ are used frequently in the public health literature, their use is not without contention. In addition to debate about how one should define catch-up growth (or in other words, how much growth needs to have taken place for catch-up to be said to have occurred), there is also the issue that researchers do not know the individual child’s genetic growth potential. Definitional issues are discussed further in the methods section.

2 Although they do not explore catch-up growth among stunted children specifically, the work by Glewwe & King (2001) and Adair et al. (2013) is relevant. Glewwe & King (2001) use data on the Philippines to explore the effect of growth in cm at various stages between 0 and 8 years and conclude that the 18–24 month period is the most important for subsequent cognitive function. Adair et al. (2013) use cohort data from five developing countries to examine whether conditional linear growth between 0 and 2 years and between 2 years and mid-childhood is related to various outcomes, including the likelihood of completing secondary schooling (where conditional linear growth is calculated as current height conditional on previous weight and height, and represents faster or slower than expected relative growth). They find that conditional linear growth between 0 and 2 years especially is associated with better schooling outcomes.

3 While linear growth retardation or stunting is commonly used to index nutritional deficits, stunting can be considered to be a marker of more general deprivation in childhood, including a poor sanitary environment and inadequate care.

4 To ensure precision in the measurement of height, fieldworkers took the measurement twice and if there was more than a cm difference between the two estimates, height was measured a third time. Children younger than 24 months (731 days) were measured in the recumbent position, while for children aged 24 months or older, standing height was measured. NIDS used age in days to calculate the z-scores. For children up to the age of five years, the WHO international child growth standards were used (WHO 2006), and for children older than five years, the WHO growth standards for school-aged children and adolescents were used (de Onis et al. 2007). The NIDS data are pre-cleaned, with biologically implausible values set to missing following WHO guidelines (further detail can be found in de Villiers et al. 2013:30–2).

5 One of the key issues is whether changes in HAZ over time should be used to define catch-up growth. HAZ is calculated as the cm difference between the index child’s height and the (age and sex-appropriate) reference population median height, divided by the standard deviation. Because the standard deviation increases with age, it is possible that a child’s cm height deficit can remain the same over time, while the HAZ increases. Some authors have suggested that the cm gap should at least decline for catch-up growth to be considered meaningful. Another issue is that children close to the -2 HAZ cut-off for stunting will be more likely to be classified as ‘caught-up’ at follow-up than those further away from the threshold. Desmond & Casale (2017) apply a range of definitions to the Birth to Twenty cohort data and find that the percentage of children who are classified as having experienced catch -up growth varies substantially depending on definition. The strictest definition they used was a recovery from stunting with HAZ at follow up >–1; all children who recovered to this extent also exhibited a reduction in the cm height deficit.

6 This definition has been applied previously, where the rationale is that under a normal distribution, 15.87% of the population would fall below the –1 HAZ cut-off (Wang & Chen 2012).

7 Casale et al. (forthcoming) show how varying the starting point from which catch-up growth is measured from 2 years to 1 year substantially affects the percentage of children classified as having experienced catch-up growth as well as the association between catch-up growth and cognitive function in 5-year-olds. Indeed, this is likely to be part of the reason for why mixed results have been identified in the literature; many of the studies rely on data from the Young Lives surveys, where data were collected on children who were 6–18 months at baseline.

8 Unfortunately, the sample size is limited because of high rates of missing data on HAZ in some of the waves. Of children aged 6 months to14 years, a valid HAZ was captured for 77% in Wave 1, 55% in Wave 2, 82% in Wave 3 and 90% in Wave 4. This issue is compounded by general attrition between waves because the children also needed to be interviewed in the subsequent wave for catch-up growth to be measured. Of the 668 two-year-olds in Wave 1, 68% have a HAZ value in Wave 1 and 51% have a HAZ value in Wave 2. Of the 794 two-year-olds in Wave 2, 43% have a HAZ value in Wave 2 and 74% have a HAZ value in Wave 3. And of the 801 two-year olds in Wave 3, 79% have a HAZ value in Wave 3 and 80% have a HAZ value in Wave 4.

9 Due to the shorter lag between Waves 2 and 3, 49 children from the Wave 2 sample of 2-year-olds were still either 2 years old (n = 1) or 3 years old (n = 48) in Wave 3. These children were excluded from the analysis so that catch-up growth is measured over the same age range for all the children.

10 For convenience, this variable is referred to as ‘failed in 2016’, as the vast majority of children who attended school in 2016 either passed or failed, with less than a quarter of a percent withdrawing before completing the year.

11 Given small sample sizes and the concern that the model may be over-fitted, the regressions were also estimated with only a basic set of controls (namely, age, female and African). No substantive differences in the size or significance of the main results were found.

12 In addition to the ‘don’t know’ responses, there were a number of missing values due to the skip patterns in the education module of the NIDS child questionnaire. Children who were enrolled in Grade 1 or higher, but who had not yet completed a full grade of school, were not asked the question on the year they started Grade 1. Instead of dropping these observations, I assumed that those who are reported to be enrolled in Grade 1 or higher in 2017 but had not yet completed a grade (or had only completed Grade R), started Grade 1 in 2017. I gained 160 observations by doing this. When I reran the regressions on age started school using the smaller sample of children, however, the results did not change substantively (not shown here).

13 The prevalence of stunting at age 2 for this sample of 840 children (drawn from the first three waves) is 31.3%. This is very close to the 32.4% prevalence recorded in 2008 using the full sample of 2-year-olds from the first wave who had data on HAZ (n = 454), i.e. the sample unaffected by attrition.

14 OLS regressions were used for both the continuous and binary outcomes for ease of interpretation. However, the results were robust to using probit regressions for the binary outcomes. Only the results on the main variables of interest are presented in the regression tables. The appendix shows the full set of estimation results with the controls for the Table 3 regressions for illustrative purposes.

15 Household/sibling fixed effects can be used to account for household/parental-level heterogeneity (if sample size allows). However, to account for child-specific investment allocations, an instrument is required that identifies variation in nutritional status between siblings within a household. There are very few examples of successful instrumentation in the literature. Exposure to civil war, crop loss, drought, and flood shocks, and variations in food prices and rainfall have been used (as siblings would have experienced these factors at different ages), but these studies were generally instrumenting for height at a particular point in time (Glewwe & King 2001; Alderman et al. 2001, 2006, 2009). The paper by Glewwe & King (2001) probably comes closest to addressing the issue of timing: they use price and rainfall data to identify the effects of growth between 0–12 months, 12–24 months and 2–8 years on IQ scores and found that growth between 12–24 months (and particularly between 18–24 months) significantly predicted IQ scores. Nonetheless, there are no studies to my knowledge that have tried to instrument for catch-up growth over time among children who were already stunted.