It is arguably not an exaggeration to say that stunting is a mainstay approach to judging the ill-health of children and cumulative under-investment in childhood. Recently, though, a number of thought-provoking papers (Leroy, Frongillo, LF, among others) in the public health literature have re-examined the meaning of this measure and have sparked an important debate for those of us who study human development. A recent Forman lecture by Harold Alderman at IFPRI in June masterfully brought together a large body of work on the interplay between biology and economics of child development. It is time to take stock.
Why does it matter? There is overall consensus that human capital is multi-dimensional; health and nutrition are only just two broad domains of child development (seminal work by Heckman, and cumulative body of evidence in the LANCET series). And certainly physical growth (and stunting as one indicator of insufficient growth) and cognitive development share many observed and unobserved inputs in early life (Currie, Vogl). But just how interrelated are they? Does nutrition beyond its impact on physical growth matter for long term outcomes or it is only an association? Have we gone too far by excessively proxying human capital with measures of linear growth to understand the economic contribution of nutrition? And if we are over-stating the extent to which stunting is a good stand-in measure for human capital, do we then bias policy direction away from potentially effective interventions.
Here are a few key points raised by LF and Alderman in their excellent reviews:
Stunting as a marker: Linear growth retardation is a used as a marker of the inadequacy of the early environment to which children have been exposed early in life. The interdependence of physical growth and cognitive development during the early years of life accounts for the strong association between stunting in early life and future cognition and productivity in later life (through improved adult height, improved schooling attainment and improved cognition, which are valued in the labor market (Currie, Vogl, Galasso,Wagstaff).
We are ultimately interested in child development, but we track what can be measured: There is probably agreement that we are interested in cognitive and non-cognitive skills, but in the first few years of life we rely on height (linear growth) because it is easy(er) to measure and, as such, we have it in more data sets or can include it when we do our own surveys. The attractiveness of using the 2006 WHO Growth Standards for population assessments is their simplicity and comparability across populations, time, and settings. But maybe some alternates are coming; there are ongoing joint and parallel efforts in establishing and validating population-based indicators of early childhood development that mirror the WHO Growth standards that will allow us to measure trajectories of child development across ages and settings (here). Until then, though, the WHO Growth Standards will still be the best tool for population-level assessments of child development.
When stature falls short: So why not just stick with stunting and linear growth? Because there are nutrition interventions that may improve important nutrition outcomes, such as morbidity or mortality or directly cognition (exclusive breastfeeding in the early months, iodine supplementation, or deworming) without having an impact on linear growth. Focusing solely on program success based on the ability to affect stunting may lead to under-estimates of the potential benefits from integrated maternal and child health and nutrition interventions that impact a broader set of nutritional outcomes but maybe not stunting.
Putting some structure: How strong is the association between stunting and cognition? What can we say about pathways and causality? A recent meta-analysis suggests that the association is significant, but weak (Kowalski et al, Prado et al). LF conclude that “based on our current understanding of mechanisms, it is not likely that they are causally related”.
A recent important paper by Attanasio, Meghir and Nix (AMN hereafter) carefully revisits this question using high quality data from the Young Lives Survey in India. The rich dataset includes measures of health (anthropometrics, self-reported health), and cognition (language, math/literacy tests), material investments (food frequency and diversity, and specific expenditure in children on food, clothes and books) and a rich set of household socioeconomic characteristics. The dataset follows the same cohort of children longitudinally at age 1, 5, 8 and 12: this allows the researchers to model the human capital accumulation at multiple stages of childhood development in a setting where nutritional deficiencies are substantial. They do so with the state of the art methods to estimate a production function for child development (pioneered by Heckman, Cuhna and Heckman, Cuhna and Schennach). Here are a selection of key results that are worth highlighting:
- Proxies and markers: Height-for-age, together with weight-for-age, and self-reported status are used as (imperfect) proxies to capture what we are ultimately interested in, that is an underlying health construct (the latent factor). The same holds for the observed proxies for cognition (as well as for investments and parental endowments): measurement error is ever-present, and needs to be modeled.
- Persistence and dynamics: Both health and cognition are self-producing. Health becomes highly persistent and hard to change past early childhood. As Alderman emphasized, some of the catch-up growth documented in the literature might be capturing measurement error, and its biological and economic significance might have been overemphasized. Conversely, cognition is less persistent at younger ages. This interesting feature may explain why some promising interventions show fade-out in the literature.
- Health and cognition are complements, not substitutes. This is true statically, at a given stage of child development, and, importantly, dynamically. The most novel result of the paper is that early health has a strong effect on subsequent cognitive development. By age 12, this effect no longer matters, but earlier impacts have been embedded in cognition. This dynamic link (dynamic complementarity) and its timing may explain in part the weak association documented in the literature.
The syntheses offered by LF and Alderman and the AMN paper represent an important step forward in our understanding of the interplay between different domains in the process of human capital accumulation and open the venue to more exciting work ahead. A few concluding thoughts:
- Interventions and timing: In line with the literature (here), AMN show that interventions that promote investments in early stimulation have potential larger returns on child development than investments in health/nutrition, but their impact is more elusive at early ages, and thus require sustained interventions over time. Early investments in health in low income settings may be hard to achieve, but, when realized, feed into cognition and their effect does not fade out. These findings have important implications for the design and sequencing of interventions.
- Investment in measurement and high-quality longitudinal data at critical ages of child development pays off. Some key outcomes, such as socio-emotional skills, however, could not be modeled in AMN as they were not present in the Young Lives data. The same holds for parental investments, which captured only material inputs, thus missing an important role of time investment and the quality of parent and child interaction, or data on mediators, such as parental mental health.
- Structural models can be a powerful tool to complement findings of RCTs: they allow to model pathways of impact, and to simulate counterfactual policies. AMN simulate the dynamic impact of two possible interventions (cash transfers and exogenous health improvements). They provide insights on dynamics and returns to investments along the entire distribution of skills, which is critically important if we are interested in reaching the most disadvantaged and tackling inequalities that emerge early in life. The modeling exercise is also clear about its assumptions and scope: household decision-making and simulations of counterfactual policies take preferences, beliefs and expectations about the returns to investments (here and here), and social norms (here) as given. More multidisciplinary work is needed to better understand these powerful drivers of human behavior in order to design more effective interventions.