Thomas Lumley <[EMAIL PROTECTED]> writes: > Also, the algorithm in glm.fit, while not perfect, is a little smarter > than a simple IRLS. It uses step-halving to back away from the edge, > and when the parameter space is convex it has a reasonable chance of > creeping along the boundary to the true MLE.
Hmm. That wasn't my experience. I had a situation where there was like a (virtual) maximum outside the boundary, and the algorithm would basically stay on the path to that peak, banging its little head into the same point of the wall repeatedly, so to speak. (If you make a little drawing of an elliptical contour intersecting a linear boundary, I'm sure you'll see that this process leads to a point-of-no-progress that can be quite far from the maximum along the edge.) > I think better glm fitting is worth pursuing: computational > difficulties with the log-binomial model have forced many > epidemiologists turn to estimators other than the MLE (or contrasts > other than the relative risk), which is a pity. Yup... -- O__ ---- Peter Dalgaard Blegdamsvej 3 c/ /'_ --- Dept. of Biostatistics 2200 Cph. N (*) \(*) -- University of Copenhagen Denmark Ph: (+45) 35327918 ~~~~~~~~~~ - ([EMAIL PROTECTED]) FAX: (+45) 35327907 ______________________________________________ [EMAIL PROTECTED] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
