Making early childhood programs work where they are hardest to run

The first thousand days set the trajectory for language, cognition, and schooling, yet in rural China nearly half of young children in our study sites show risk of cognitive or language delay. My earliest work with Stanford's Rural Education Action Program synthesized the evidence base for this problem: a systematic review and meta-analysis across twenty-nine English- and Chinese-language studies establishing both the scale of delay and the fact that parental training programs meaningfully move cognition, language, and social-emotional outcomes.

Since then I have worked on the mechanism rather than the headline. Using Language Environment Analysis (LENA) recorders in rural households, I study the home language environment directly — how much adults actually talk to children, how many conversational turns a day contains, and how those measured realities diverge from what caregivers believe is happening. That gap between the measured home environment and the perceived one turns out to be systematic, and it matters for how the field designs both interventions and evaluations.

The through-line is a question that belongs as much to education as to public health: what are the conditions under which a child arrives at school ready to learn, and which of those conditions are actually within the reach of policy?

Nearly a quarter of the pregnant women and new mothers we surveyed in rural Sichuan showed symptoms of depression, anxiety, or stress. What predicted those symptoms was not primarily individual pathology but household structure: decision-making power, family conflict, and the presence or absence of social support. Perinatal mental health, in other words, is a social and institutional fact before it is a clinical one.

This has direct consequences for intervention design. In a cluster-randomized trial across one hundred villages, village parenting centers that improved children's environments produced no overall improvement in caregivers' mental health — with heterogeneous effects by socioeconomic status and by whether the primary caregiver was a mother or a grandmother. The lesson is not that parenting programs fail, but that adding a child-facing component does not automatically support the adult delivering the care, and that grandparent caregivers are a distinct and understudied population.

A continuing thread in my work is the causal version of this question: whether and how caregiver mental health moderates who benefits from early childhood interventions — compliance, dose, responsive stimulation, and ultimately child outcomes.

I help lead the design and evaluation of the Thinking Healthy Extended Program (THEP), an adaptation of the WHO's Thinking Healthy curriculum for caregivers of infants and toddlers. THEP is being evaluated in a four-arm, 125-village factorial cluster-randomized trial that crosses parenting stimulation with caregiver mental health support — designed from the outset to be delivered and supervised by local agents of the All-China Women's Federation, the nationwide government organization already responsible for women's and children's welfare.

That design choice is the research question. An intervention that only works when delivered by trained psychologists is not a policy; it is a demonstration. So alongside the effectiveness trial I run process evaluation and human-centered design work: what fidelity looks like when non-specialists deliver a mental health curriculum, where facilitators adapt the material and why, what dose families actually receive, and which frictions — time, childcare, stigma, distance — determine whether a mother attends at all.

Complementary qualitative work asks the same question of a different delivery platform: in-depth interviews with fifty mothers in Shanghai on whether community and township health centers are an acceptable and appropriate place to receive postnatal mental health support, and what would have to change for them to use it.

Much of what the field knows about children in low- and middle-income countries rests on instruments developed elsewhere and on what caregivers report about their own children. Both assumptions deserve scrutiny. An early validation study examined the concurrent validity of the Ages and Stages Questionnaire against the Bayley Scales of Infant Development III in China — necessary groundwork for any claim built on ASQ scores in this setting.

More recently I have documented a systematic perception bias: caregivers experiencing depressive and anxiety symptoms tend to overestimate both their children's language development and their own verbal input to those children. Because caregiver mental health is itself correlated with the outcomes we care about, self-report measurement can bias estimates in exactly the populations interventions are meant to serve.

Taking measurement seriously is not a methodological footnote. It determines which programs look effective, which populations look well-served, and where public money goes.