Heterogeneous treatment effects analysis for social scientists: A review

Area of Science:

• Social sciences
• Statistics
• Machine Learning

Background:

• Social scientists study varying intervention responses, driving the need for heterogeneous treatment effects (HTE) analysis.
• Over 50 years, HTE methods have evolved from linear models to advanced machine learning techniques.

Approach:

• This article systematically reviews major HTE methods, including interaction modeling, generalized additive models, propensity-score methods, marginal treatment effect, and various machine learning techniques.
• Methods are presented chronologically to illustrate developmental trends, discussing strengths and limitations.
• An illustrative example is provided, followed by reflections on future methodological advancements.

Key Points:

• Covers multiplicative interaction modeling, generalized additive modeling, propensity-score-based methods, marginal treatment effect, separate LASSO constraints, causal trees, causal forests, Bayesian additive regression trees.
• Includes meta-learners: S-learner, T-learner, X-learner, and R-learner.
• Highlights the evolution and comparative advantages/disadvantages of diverse HTE methodologies.

Conclusions:

• The review provides a comprehensive overview of HTE methods, aiding researchers in selecting appropriate techniques.
• It identifies current limitations and suggests future research directions for developing more robust HTE analysis tools.