Evaluating Emerging Machine Learning and Multiple Imputation Methods for Estimating Individual Treatment Effects

• In Event: AERA Roundtable Session Thursday 1:45 pm Four Seasons Ballroom 4
  In Roundtable Session: Machine Learning Techniques II: Advancing Coding/Essay Scoring and Estimation of Treatment Effects (Table 2)

Abstract

This study evaluated 15 emerging methods for estimating individual treatment effects (ITE) using synthetic and semi-synthetic educational intervention datasets. Methods were categorized as meta-learners, tree-based direct estimation, (deep-learning) representation learning, and multiple imputation approaches. The S-learner with BART, Bayesian Causal Forest (BCF), Causal Forest, and X-learner with BART showed superior performance among analyzed methods. Meta-learners with BART and tree-based direct estimation methods generally outperformed representation learning and multiple imputation techniques. This research contributes to the growing field of ITE estimation, gaining traction across various disciplines. Its popularity stems from the potential to tailor interventions to individual needs and target programs at those who would benefit most, offering promising applications in personalized treatment strategies and efficient resource allocation in diverse sectors.

Author

• Sangbaek Park, University of Louisville