I did not get any warnings when I ran your data/model example. From: Fischer, Felix [mailto:felix.fisc...@charite.de] Sent: Wednesday, January 30, 2013 11:19 AM To: Ravi Varadhan Cc: r-help@r-project.org Subject: AW: [R] starting values in glm(..., family = binomial(link =log))
Thanks for your replies! It seems, that I can fit my model now, when I can provide the "right" starting values; however there remain warnings, such as: 1: In log(ifelse(y == 1, 1, (1 - y)/(1 - mu))) : NaNs wurden erzeugt 2: step size truncated due to divergence 3: step size truncated: out of bounds ... That makes me feel uncomfortable and I wonder whether I can "trust" the fitted model. Why is this kind of regression so picky about starting values compared to logistic regression? And is there a way to explain the difference between binomial - quasibinomial to a simple mind like mine? Best, Felix Von: Ravi Varadhan [mailto:ravi.varad...@jhu.edu] Gesendet: Mittwoch, 30. Januar 2013 17:02 An: Fischer, Felix Cc: r-help@r-project.org<mailto:r-help@r-project.org> Betreff: [R] starting values in glm(..., family = binomial(link =log)) Try this: Age_log_model = glm(Arthrose ~ Alter, data=x, start=c(-1, 0), family=quasibinomial(link = log)) Ravi Ravi Varadhan, Ph.D. Assistant Professor The Center on Aging and Health Division of Geriatric Medicine & Gerontology Johns Hopkins University rvarad...@jhmi.edu<mailto:rvarad...@jhmi.edu> 410-502-2619 [[alternative HTML version deleted]] ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.