9:00-11:00 明德主楼1016
Doubly Robust Survival Trees and Random Forests
Professor Liqun Diao
University of Waterloo
Estimating a patient’s mortality risk is important in making treatment decisions. Tree-based methods are useful tools to identify risk groups and conduct prediction by employing recursive partitioning to separate patients into different risk groups. Existing “loss based" recursive partitioning procedures that would be used in the absence of censoring have previously been extended to the setting of right censored outcomes using inverse probability censoring weighted estimators of loss functions. We propose new "doubly robust" extensions of these loss estimators motivated by semiparametric efficiency theory for missing data that better utilize available data. We realized such extensions by imputation and extended this single tree method to doubly robust survival random forests (ensemble methods). Simulations and a data analysis demonstrate strong performance of the doubly robust survival trees and random forests compared to previously used methods.
Propensity Score Methods for Causal Inference
Professor Hongwei Zhao
ScD, School of Public Health, Texas A&M University
In this talk, I willpresent fundamental theory for performing causal inference using propensity score methods in secondary/observational data. I will first introduce the potential outcome model for conducting causal inference, and the necessary assumptions involved. Next, the definition of propensity score will be introduced, and the propensity score based methods will be discussed, together with the advantages and disadvantages of each approach. Alternative ways for obtaining propensity scores will be considered. Finally we will illustrate methods for conducting causal inference with observational data using Stata software.