2 messages · Kjetil Halvorsen, Henric Nilsson
Hola! Anybody knows if there exists somewhere in R some implementation of robust Poisson regression, where robust is taken in the sense as usen in rlm(MASS). I found something in the package wle, but only for the Poisson distribution, not for regression. For the moment I try to use linear models with the square-root transformation, and rlm. Kjetil
Kjetil Halvorsen.
Peace is the most effective weapon of mass construction.
-- Mahdi Elmandjra
Anybody knows if there exists somewhere in R some implementation of robust Poisson regression, where robust is taken in the sense as usen in rlm(MASS). I found something in the package wle, but only for the Poisson distribution, not for regression.
Take a look at Eva Cantoni's "Analysis of robust quasi-deviances for generalized linear models" paper in JSS 10(4) (http://www.jstatsoft.org/v10/i04/). The tar'ed file contains S-PLUS functions, and IIRC the modifications to make `glm.rob' run under R were quite simple. I can't find my modified version right now, but I remember making changes to lots of `assign' calls... I also remember having trouble with `integrate' calls in the `quasi.rob.xxx' functions, but since I didn't have to use these I never tried fixing it. HTH, Henric