Learning What Works: Uncertainty and Selective Inference
*Sharon Lise Normand, Harvard University
Keywords: dimension reduction, sparsity, propensity score, model averaging
The growing interest in estimation of the effectiveness of competing treatment strategies has been driven by many factors, including several technical advances: the evolution of data acquisition and storage technologies, the escalation in the processing power of computers, and the proliferation of information. Big health data systems present opportunities to learn what works but are associated with unique statistical challenges. For example, with hundreds of potential confounders readily available, how do researchers select a particular subset to use in causal estimation? In this talk, selection inference for dimensional reduction strategies and associated implications for estimation of treatment effect parameters are described. Effectiveness of cardiovascular treatments illustrate approaches.
Important Dates & Deadlines
- October 9 - 11, 2013