[R] Residuals from binomial glm

Petar Milin pmilin at ff.uns.ac.rs
Sun May 13 19:21:43 CEST 2012


Dear all!
Sometimes, when using plain lm(), if I have highly correlated numeric
predictors (covariates), I firstly ran lm() to get residuals of one from
the other. This way, they became uncorrelated (orthogonal), and I can
check if the residualized one contributes to prediction, over and above
the other. I do that if and only if there is "theoretical" sense/ground
for such step.
Now, I would need to run binomial glm, but I think that those residuals
may "trick" me. Logic of glm's various types of residuals tells me that
the "response" should be a match to those which I would get from simple
lm(). However, I do not want to wander blindly into this.
Please, does anyone have experience with using glm.residuals on the
right-hand side of main lm() model? What to use, how and why?
Any advice on literature?

Best,
Petar



More information about the R-help mailing list