Dear Brian and list members,
Thanks very much for your response Brian!
I applied the adjusted calculation that you advised me to use
[1/(1+exp(-upperlogit))] and as a result I don't get any more NA values in my
upper confidence interval values.
Yet, some outcomes are very akward, since for ver
Possibly your calculation overflows: exp(upperlogit)/(1+exp(upperlogit))
could be replaced by 1/(1+exp(-upperlogit)), or even better by
plogis(upperlogit). This could happen via the Hauck-Donner effect: the
fitted probabilities are very near one and the standard errors are very
large.
As for
Hello all,
I've come across an online posting
http://www.biostat.wustl.edu/archives/html/s-news/2001-10/msg00119.html
that described how to get confidence intervals for predicted values from
predict.glm. These instructions were meant for S-Plus. Yet, it generally seems
to work with R too, b
3 matches
Mail list logo
