Panel Paper:
Childhood Mental Health: Latent Risk, Missed Diagnoses, and Long-Term Effects
Friday, July 24, 2020
Webinar Room 4 (Online Zoom Webinar)
*Names in bold indicate Presenter
One in five children receives a mental health diagnosis before the age of 18, with significant adverse effects on education and later life earnings. Population-based treatment studies have shown limited pathways to mitigate this negative gradient. Still, selection into diagnosis complicates the current understanding of treatment effects. This paper combines data-driven methods to predict latent mental health risk with casual identification to address this issue. Using a nationally representative longitudinal child health survey linked to current day administrative tax files, we estimate childhood mental health risk and its heterogeneous long-term welfare implications by treatment status. We predict mental health risk using gradient boosted decision tree modelling and find sex, rurality and ethnicity drive deviations in machine-predicted versus realized diagnosis. Low-risk diagnoses are evident but rare, while missed or delayed diagnoses in the top deciles of the risk distribution pose a more significant issue. To assess if misdiagnosis is occurring, we evaluate marginal diagnoses, induced by a child's school starting age, by predicted risk decile. These diagnoses appear throughout the risk distribution, suggesting misdiagnosis is occurring. Long-term, while diagnosis negatively affects educational attainment and income, these effects are concentrated in low-risk individuals. Diagnosis and medical treatment effects are, in fact, positive and increasing for children in the top risk decile. This work highlights the nuance required in assessing long-term treatment effects in mental health. Identification of heterogeneous long-term treatment effects by risk profile has potential policy implications for targeted treatment to ease mental health-based socioeconomic inequality.
