Causal decomposition analysis with synergistic interventions

What is it about?

We develop a causal decomposition framework that evaluates sequential, multi-deomain ("synergistic") interventions--e.g., improving access to high-performing schools and early Algebra I--so researchers can estimate how much the interventions would reduce racial disparities in math achievement. Because these settings feature complex interactions among group status, intervening factors, and confounders, we introduce a triply robust estimator (with GLM or machine learning) that is robust to multiple forms of model misspecification. We validate performance via simulation and apply the method to HSLS:09 data on Black, Hispanic, and White students to show how sequential equalization of opportunities can narrow disparities in math achievement.

Featured Image

Photo by Erik Mclean on Unsplash

Photo by Erik Mclean on Unsplash

Why is it important?

1. Realistic policy design: Many disparities won't be reduced with a single intervention. Our method quantifies disparity reduction with sequential interventions. 2. Robust to model specification: The debiased machine learning (triply robust estimator with machine learning and cross-fitting) enables more reliable estimates in complex social data. 3. Practical guideline: Simulation studies offer practical guidance by model complexity and sample size. 4. Generalizability: Although demonstrated in education, the framework applies to other domains where multi-step interventions are needed.