|
Abstract:
|
For clinical trials with binary data of small to medium size, when logistic regression is applied, sometimes separation problem can happen, i.e., the estimate of the parameter of interest is infinite although the likelihood converges. Several methods have been proposed in the literature to tackle this problem. In this talk, I will focus on comparison of several methods by the Firth procedure (essentially equivalent to a Bayesian approach with Jeffery's invariant prior) and Bayesian methods with other informative prior. Both simulation and case study results will be presented.
|
