Early Childhood Developmental Status in Low- and Middle-Income CountriesNational, Regional, and Global Prevalence Estimates Using Predictive Modeling

Eighty million children. Not eighty million in poverty, not eighty million who are malnourished — eighty million three- and four-year-olds whose cognitive or socioemotional development is already falling behind before they ever set foot in a classroom. That number comes from a 2016 study by McCoy, Peet, Ezzati, and colleagues. And here's the part that should stop you: until this work, nobody had actually measured it. They were guessing, using height charts and poverty rates as stand-ins for what was happening inside a child's mind. For decades, the standard approach to estimating developmental risk in low and middle-income countries was to count stunted children and children living in poverty. A widely cited two thousand seven paper by Grantham-McGregor and colleagues used exactly that method to estimate two hundred nineteen million children under the age of five were failing to reach their developmental potential. Those indicators matter enormously. But stunting and poverty only partially explain the variation in children's actual cognitive and socioemotional functioning. The two things are not the same. Population-level measurement of cognitive and socioemotional development had lagged behind because of the practical difficulty of assessing complex psychological processes across dozens of diverse, low-resourced countries. The measurement gap was real, and it had consequences: without knowing how many children were struggling with attention, self-regulation, or social skills specifically, policymakers couldn't accurately target programs that would actually address those deficits. The instrument McCoy and colleagues used to close that gap is the Early Childhood Development Index, or ECDI — a ten-item, caregiver-reported screening tool designed for children aged 36 to 59 months. It was administered as part of two major survey programs: UNICEF's Multiple Indicator Cluster Survey and the Demographic and Health Surveys. The items are straightforward yes or no questions that parents answer about their child. On the cognitive side: does your child follow simple directions correctly? Can they do a task independently? On the socioemotional side: does your child kick, bite, or hit other children? Do they get easily distracted? Do they get along well with other kids? The ten items were winnowed from an original pool of one hundred fifty-eight through a multistage pilot and validation process, and the authors confirmed the domain structure using confirmatory factor analysis across the full dataset — a statistical technique that tests whether the items actually group the way the theory predicts. The fit was good: a root mean square error of approximation of 0.04, and a comparative fit index of 0.99. The structure held up when replicated country by country. A child is classified as having low development in a domain if they fail more than one item within it. The analytic sample was ninety-nine thousand two hundred twenty-two children across thirty-five low and middle-income countries, collected between two thousand five and two thousand fifteen. In that pooled sample, fourteen point six percent of children had low cognitive scores, twenty-six point two percent had low socioemotional scores, and thirty-six point eight percent performed poorly in either or both domains. Those numbers alone are striking. But what's equally striking is the range. Montenegro came in at four point three percent. Chad came in at sixty-seven percent. Sierra Leone at fifty-four percent. Pakistan at forty-eight percent, Kazakhstan at fourteen percent. The variation across countries is not noise — it reflects real differences in the conditions children are growing up in. The demographic patterns within that variation are consistent and meaningful. Boys scored lower than girls. Rural children scored lower than urban children. Stunting and low household wealth were both associated with worse outcomes — at the country level, the correlation between stunting prevalence and low ECDI scores was 0.72, and the correlation between low ECDI prevalence and the Human Development Index was negative 0.84. That is a very strong relationship. But here's where the data get genuinely important for policy: at the individual level, fifty-five point eight percent of stunted children were developing normally on the ECDI. And roughly a third of non-stunted children — thirty-three point eight percent — had low ECDI scores. The overlap between stunting and developmental delay is real, but the two populations are not the same. Nutrition alone won't find all the children who need help. That insight shapes everything that follows, including how the team built their global estimate. McCoy and colleagues had direct ECDI data from thirty-five countries. They needed to project to all low and middle-income countries. So they built a series of ordinary least squares regression models — essentially asking: which country-level characteristics best predict the prevalence of low ECDI scores in the countries we do have data for? They tested stunting prevalence from the Nutrition Impact Model Study, the Human Development Index, and a combination of both. To decide between models, they used cross-validation: repeatedly holding out portions of their thirty-five country sample and testing whether the model could predict those held-out countries accurately. The metric for success was the root mean square error of prediction. The results were clear. The stunting-only model explained about forty-seven percent of variance between countries. Adding the Human Development Index pushed that to seventy percent — and the combined model added almost nothing over the Human Development Index alone. The Human Development Index coefficient in the winning model was negative 1.06, meaning countries with higher composite scores on income, education, and health had markedly lower rates of developmental delay. The cross-validation root mean square errors were 0.10 and 0.07 for the Human Development Index model, compared to 0.13 and 0.09 for stunting alone. A single composite index — one number summarizing a country's health, education, and income — captured most of the between-country variation in early childhood cognitive and socioemotional scores. The team then applied that model to all low and middle-income countries, multiplied predicted prevalences by two thousand ten population counts from the World Population Prospects, and derived their global estimate: eighty point eight million three- and four-year-olds with low cognitive or socioemotional development. The ninety-five percent confidence interval runs from forty-eight million to one hundred thirteen million — wide, as the authors acknowledge, but unavoidably so given the modeling approach. The global prevalence was thirty-two point nine percent, with a confidence interval of roughly twenty to forty-six percent. The regional breakdown tells you where the burden is concentrated. Sub-Saharan Africa: twenty-nine point four million children, forty-three point eight percent of all three- and four-year-olds in the region. South Asia: twenty-seven point seven million children, thirty-seven point seven percent. Those two regions together account for the majority of the affected children globally. By contrast, Latin America and the Caribbean came in at eighteen point seven percent, and North Africa, the Middle East, and Central Asia at eighteen point four percent. East Asia and the Pacific at twenty-five point nine percent. The gradient follows the Human Development Index gradient — which is precisely what the model predicted. McCoy and colleagues are candid about what this study cannot do. The Early Childhood Development Index is brief by design, which means it can't capture fine-grained subdomains like memory or executive function. The cutoff for "low" development hasn't been fully clinically validated as a diagnostic threshold. Caregiver reports may vary systematically across cultures — a parent in one context may assess the same behavior differently than a parent in another. The thirty-five sampled countries represent just over twenty-one percent of the under-five population in low and middle-income countries, and the extrapolation carries real uncertainty. The confidence intervals are wide for a reason. But what this study does that nothing before it managed is provide a direct, domain-specific count of children falling behind in cognitive and socioemotional development — not children who are short for their age, not children living below a poverty line, but children whose caregivers report they cannot follow directions, control their behavior, or get along with other kids. That distinction is the precondition for acting on the problem. You can make progress on stunting rates and still miss a third of the children who need developmental support. You can reduce poverty and still leave millions of three-year-olds without the attentional and social foundations they'll need in school. McCoy and colleagues argue that addressing this burden requires interventions that work on multiple fronts simultaneously — nutrition, parenting quality, poverty reduction, and access to early education, all at once. No single lever is sufficient. The finding that the Human Development Index predicts developmental delay better than stunting alone suggests that development happens at the intersection of health, education, and economic conditions, not in any one of those channels independently. Counting the problem accurately, the paper concludes, is the first step toward fixing it. Eighty million children is where that counting starts. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.