Social
Housing insecurity and adolescent mental health: A data-driven exploration of risk and resilience Sakurako Okuzono* Sakurako Okuzono Natalie Slopen
Background: Millions of U.S. households face difficulty paying rent and housing evictions, with families with children disproportionately affected by housing insecurity. Understanding heterogeneous responses to housing instability can lead to improved and tailored interventions. This study examined heterogeneity in the association between housing instability and subsequent mental health among U.S. adolescents to identify characteristics of at-risk and resilient subpopulations using a machine-learning approach.
Method: We used data from the Adolescent Brain Cognitive Development Study, a longitudinal cohort of 9-10-year-old U.S. adolescents. We included data from baseline to 3-year follow-up without missing exposure and outcome (n= 8,949). Using an ecological model, we selected 73 baseline variables as potential effect modifiers, reflecting a range of individual, family, neighborhood, and state-level characteristics. The exposure was housing instability one or two years after baseline. Mental health was assessed using the Child Behavioral Checklist at age 12. We estimated the average treatment effect via targeted maximum likelihood estimation and its heterogeneity (Conditional Average Treatment Effect: CATE) via the Generalized Random Forest algorithm (GRF), which allowed us to model complex multiple interactive effects. Last, we identified at-risk and resilient groups by taking the top and bottom deciles of the CATE estimate.
Results: 8 percent of participants experienced housing instability. Housing instability is associated with higher behavioral problems, and CATE obtained with GRF ranges from -0.5 to 1.1 (SD: 0.1). Comparing the at-risk and resilient groups, the at-risk group tended to belong to minoritized racial groups, have lower school involvement, and live in worse neighborhoods. We further identified that complex heterogeneity exists based on unique combinations of baseline characteristics.
Conclusion: By leveraging a data-driven approach, our study highlights the multifaceted nature of risk and resilience in response to housing insecurity and offers a framework for further research that can inform the development of targeted interventions.