Title: Propensity score regression and G-estimation
Authors: David Stephens - McGill University (Canada) [presenting]
Abstract: The links between g-estimation and regression adjustment are reviewed by using the propensity score. G-estimation is well established as a doubly robust and locally efficient semiparametric inference procedure under certain assumptions, but has not been as widely used as other adjustment procedures perhaps due to it being less transparent in its construction, and more difficult to implement. The binary outcome case has proved particularly challenging for g-estimation, although procedures have now been developed for this case. We will show that using a regression construction inference in the binary case can be implemented very straightforwardly using standard tools, and that the regression-based procedure constructs the semiparametric efficient estimator. Finally, we will demonstrate the use of g-estimation in the context of dynamic treatment rule construction utilizing a semiparametric model selection criterion.