|
Activity Number:
|
27
- SPEED: Causal Inference and Related Methodology Part 1
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, July 28, 2019 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
|
Abstract #306950
|
|
|
Title:
|
Causal Mediation Analysis Using Gradient Boosting Machines: Developing Methods and Software
|
|
Author(s):
|
Brian G. Vegetabile* and Donna L. Coffman and Daniel F. McCaffrey
|
|
Companies:
|
RAND Corporation and Temple University and Educational Testing Service
|
|
Keywords:
|
Mediation Analysis;
Causal Inference;
Machine Learning;
Nonparametric Estimation;
Propensity Scores
|
|
Abstract:
|
Recent years have seen advances in the understanding, and development, of methodology for causal mediation analysis. Traditional mediation analysis assumes there exists a mediator variable on the path between a treatment variable and an outcome variable. This mediator variable is then used to help “explain” the relationship between the treatment and outcome variables by decomposing the overall effect into a “direct” and an “indirect” effect due to the mediator. Our work integrates statistical learning algorithms within the general weighted causal mediation analysis framework, specifically using gradient boosting machines to estimate the both the treatment and mediator assignment mechanisms and create weights for estimating the total treatment effect and natural direct and indirect effects. To control the statistical learning algorithm the number of trees is selected that minimizes a metric of weighted covariate imbalance at each stage of the estimation process. Standard ignorability assumptions are assumed and a simulation study demonstrates the utility of the approach. The approach will be integrated into the “twang” package in R for estimating causal effects.
|
Authors who are presenting talks have a * after their name.
