Estimating heterogeneous treatment effects in a randomized controlled trial of a nurse-home visiting intervention on child outcomes: A causal forest approach

Heterogeneous treatment effect estimation is ubiquitous in psychosocial studies involving an intervention. While conventional methods most often examine population averaged estimates, it is important to understand the more nuanced effect of an intervention varying from person to person by individual characteristics and circumstances. Traditional regression approaches with interaction effects used for treatment moderating variable identification may perform poorly in certain datasets. In this project, we use a machine learning approach to investigate treatment effect heterogeneity. We use causal random forests to estimate individual treatment effects in a child health data set from a randomized controlled trial evaluation of a nurse-home visiting intervention program exploring child injury, mental health and learning outcomes. We then identify potential moderators using generalized additive models. Our results show evidence of treatment effect moderation by baseline variables such as age, income and education. Our findings may provide guidance for future evaluations of early intervention programs.