Causal mediation analysis with double machine learning

Description

Presented by Martin Huber with Helmut Farbmacher,  Lukas Laffers, Henrika Langen and Martin Spindler

This paper combines causal mediation analysis with double machine learning for a data-driven control of observed confounders in a high-dimensional setting.

The average indirect effect of a binary treatment and the unmediated direct effect are estimated based on efficient score functions, which are robust w.r.t. misspecifications of the outcome, mediator, and treatment models. This property is key for selecting these models by double machine learning, which is combined with data splitting to prevent overfitting.

We demonstrate that the effect estimators are asymptotically normal and root-n consistent under specific regularity conditions and provide a simulation study as well as an application to the National Longitudinal Survey of Youth.

SoDa Labs webinar series

The SoDa Labs webinar series provides a platform for researchers around the world to present work that uses novel and alternative data and/or tools from data science and beyond to answer social science questions.