James Robins, MD
Harvard TH Chan School of Public Health
Doubly Robust Machine Learning (DR-ML) estimators are the current state of the art estimators of causal effects. However, even the use of DR-ML estimators is not guaranteed to provide valid confidence intervals. Thus, This talk will answer the following: Can tests be developed that have the ability to detect whether the bias of a DR-ML estimator is of the same or greater order than its SE? If so, can we construct new estimators that are less biased without: i) refitting, modifying, or even having knowledge of the employed ML algorithms and ii) without making any assumptions about the smoothness or sparsity of the true outcome regression or propensity score function?
Please RSVP: https://bit.ly/2UcBbVK