Harnessing Double-Robust Machine Learning for Causal Evaluation: The Impact of Teacher Induction Supports on Teacher Burnout and Attrition

This study applies Targeted Maximum Likelihood Estimation (TMLE) with Super Learner, a machine learning-based causal inference method, to evaluate the impact of induction supports on teacher burnout and attrition using data from the Teacher Prep to Career (TP2C) Study. Findings reveal that mental health and well-being components significantly reduce burnout, while mentorship and professional development unexpectedly increase emotional exhaustion, likely due to implementation challenges. By leveraging advanced causal inference techniques, this study provides actionable insights to optimize induction programs, ensuring they effectively support teacher retention. It also demonstrates how machine learning enhances evaluation rigor in observational data settings, offering a replicable framework for assessing complex interventions. Future research should apply causal mediation analysis to investigate how certain induction supports contribute to first-year teacher stress, potentially influencing early career attrition.

Amota Ataneka – Ph.D Student, University of Cincinnati; Benjamin Kelcey, Dr – Professor, University of Cincinnati; Jean Baptiste Habarurema