Machine Learning Analysis of Heterogeneity in the Effect of Student Mindset Interventions

We study heterogeneity in the effect of a mindset intervention on student-level performance through an observational dataset from the National Study of Learning Mindsets (NSLM). Our analysis uses machine learning (ML) to address the following associated problems: assessing treatment group overlap and covariate balance, imputing conditional average treatment effects, and interpreting imputed effects. By comparing several different model families we illustrate the flexibility of both off-the-shelf and purpose-built estimators. We find that the mindset intervention has a positive average effect of 0.26, 95%-CI [0.22, 0.30], and that heterogeneity in the range of [0.1, 0.4] is moderated by school-level achievement level, poverty concentration, urbanicity, and student prior expectations.