Skip to main content

Unpacking Skills at the Cradle: A Machine Learning Approach to Construct Infant Skill Measures

Periodic Reporting for period 1 - SKILLS (Unpacking Skills at the Cradle: A Machine Learning Approach to Construct Infant Skill Measures)

Reporting period: 2019-10-01 to 2021-09-30

Motivated by a growing body of research implying large private and social returns to investments in children at an early age, there has been interest among policy makers in cost-effective programs to improve early-life environments. The returns of these programs are likely to be particularly high in low and middle-income countries where it has been estimated that 43% of children under 5 are at risk of developing cognitive delays that can hinder further skill development. Although several correlates of poverty contribute to this risk, a major factor appears to be that children often lack sufficiently stimulating environments during critical periods of brain development. Due to high levels of neuroplasticity, stimulation during this period it is thought to be critical for longer term cognitive and psychosocial development. One approach that has proven effective in improving child development outcomes is interventions that support caregivers to engage in cognitively stimulating activities with their children. Evidence from multiple randomized evaluations across several countries shows that parenting interventions can significantly improve cognitive of young children in the short run. Less well understood is how these short-term cognitive and language skill improvements for children can be sustained over time to translate into improved outcomes at school, work, health and overall wellbeing.

In a first project, the research fellow aimed to track back children that were part of a parenting intervention in a rural area of China’s Shaanxi province to study whether there were any sustained improvements in parenting behavior and child outcomes after the initial program ended. The original parenting intervention took place between November 2014 and April 2015 and consisted of weekly home-visits to families with children that were around 3 years old at the start of the intervention. During the weekly home-visits parenting trainers trained parents to interact more effectively with their offspring through cognitively stimulating activities using a structured educational curriculum. This initial parenting intervention was successful and improved parenting practices and cognitive skills of young children for the families enrolled in the program. In order to study whether there were any persistent improvements in the lives of these families our enumerators went back two and half years after the initial program ended to trace all families and collected data on parenting behavior and child outcomes. The study found that the initial parenting intervention has persistent benefits for cognitive skills development of children. Parents of treatment children are found to spend considerably more time with their offspring. Beyond parental investments in the home, the study found that the parenting intervention also changed how parents think about the decision to enroll their children in school. Children from original treatment villages were enrolled earlier and in better quality preschools. The study further found that the changes in parents schooling decision reflect an increase in valuation of school quality relative to other attributes like distance and tuition fees.

The second project aimed to develop better measures to track children’s skill development over time as they age. In collaboration with Stanford’s Centre on China’s Economy and Institutions (SCCEI), the research fellow set up a research project titled: “Big Data on Little People: Towards Better Measurement of ECD Data” with as its main aim to develop and validate a shortened version of the Bayley Scales of Infant and Toddler Development (Bayley) test. Although this test is considered the golden standard in the literature when it comes to the measurement of early skills, it is expensive and time consuming to administer. Given the high prevalence of cognitive developmental delay in rural China and other low-income settings, there is a pressing need for shorter assessment tools suitable at identifying cognitive delays on a large scale and in low-income settings.
For both research projects, the fellow analyzed a large amount of survey data on children, parents, and schools in rural parts of China collected by a team of enumerators. For the first research household survey data from approximately 500 households in rural Shaanxi was cleaned and analyzed using econometric techniques to study the impact of the initial parenting program on infant skills and parenting behaviour. In collaboration with leading scholars from Stanford and Yale, the research fellow documented the scientific findings of this project in a research paper currently under peer-review for open access publication: “Parental Investment, School Choice, and the Persistent Benefit of Interventions in Early Childhood”.

For the second research project the fellow collaborated with Stanford’s SCCEI center to collect item level test data from approximately 10,000 young children living in different rural areas in China. The fellow cleaned and standardized all the item test data into one harmonized data set. Using methods from latent factor modelling and machine learning the fellow analyses how informative specific test items are for different age groups and whether age-specific skill-clusters can be identified. Research findings were disseminated to academic audiences through participation in international conferences and workshops and a first research paper summarizing the scientific findings has been submitted for peer review. The research fellow also disseminated scientific findings to non-academic audiences by invited seminars at several international policy institutes.
Scientific evidence from the research project implies an important role for school quality in sustaining longer-term impacts of early childhood interventions, particularly where caregivers have low levels of human capital. The project further contributes to the state of the art by developing a shortened assessment tool to detect cognitive developmental delay in low-income settings. Gains to society are considerable given that this will lessen the burden on test administrators and surveyed households and allow for detection of cognitive delays on a larger scale. This will lower the financial cost to measures skills in low-income setting considerably, as current instruments are patented and expensive. Furthermore, the findings of this research project can provide more insight in the timing of skill specific sensitive periods of development where skills develop at an unusually fast pace and where interventions can be most effective.

The scientific evidence from both research projects is highly relevant for policy makers worldwide who design and implement policies aimed at providing early childhood educational services. The importance of early childhood services, especially in low and middle- income countries is highlighted by its recent inclusion in the 2030 Sustainable Development Goals (SDGs). Improving access to early childhood services has shown to increase infant skills, school performance and leads to improved health and labour market outcomes. Additionally, providing better access to early childhood services is linked to several other goals stated in the SDGs such as poverty reduction and gender equality.
Toddlers in Rural Shaanxi
Parenting Household Visit