Boughdiri, A., Josse, J., & Scornet, E. (2024). Quantifying Treatment Effects: Estimating Risk Ratios via Observational Studies. Proceedings of the 28th International Conference on Artificial Intelligence and Statistics (AISTATS) 2025, Mai Khao, Thailand. PMLR: Volume 258.
This paper aims to address the gap in estimating the Average Treatment Effect (ATE) using the Risk Ratio (RR) in observational studies, proposing and analyzing different RR estimators while establishing their theoretical properties.
The authors utilize the potential outcome framework and semi-parametric theory to analyze various RR estimators, including the Risk Ratio Neyman (RR-N), Risk Ratio Inverse Propensity Weighting (RR-IPW), Risk Ratio G-formula (RR-G), Risk Ratio One-step (RR-OS), and Risk Ratio Augmented Inverse Propensity Weighting (RR-AIPW). They derive asymptotic normality, limiting variance, and asymptotic confidence intervals for each estimator. The performance of these estimators is evaluated through simulations using both linear/logistic and non-linear/logistic data generating processes.
The study concludes that RR-AIPW and RR-OS, particularly when employing linear estimators for nuisance components, demonstrate superior performance in estimating the RR in observational studies. The authors emphasize the need for further research to establish guidelines for selecting between linear and non-parametric approaches for estimating nuisance components.
This research contributes significantly to the field of causal inference by providing a comprehensive analysis of RR estimators for observational studies, offering valuable insights for researchers aiming to quantify treatment effects using this measure.
The study primarily focuses on estimating the ATE for the RR and does not directly address the estimation of Conditional Average Treatment Effects (CATE). Future research could explore extending these methods to CATE estimation and investigate procedures for generalizing the RR to broader populations.
To Another Language
from source content
arxiv.org
Viktige innsikter hentet fra
by Ahmed Boughd... klokken arxiv.org 10-17-2024
https://arxiv.org/pdf/2410.12333.pdfDypere Spørsmål