Dear S-Users:
I am analysing some data using parametric survival methods with interval
censoring , by means of S-Plus 2000. First I have to find out which model
fits the data , using different distributions, using censorReg function.
Some authors suggest that one the elements is to inspect the shape of the
hazard function. From the data, this seems to increase fairly linear to some
point in time and them stays constant for the rest of the observation period.
This shape seems to suggest a gompertz distribution.
From the "guide to statistics manual" of S-Plus" (pag 348/349) is suggested
that it can be fitted "....using the truncation argument to specify that each
observation is restricted to (0, infinite) "..
I try this
fit.gomp <-censorReg(censor(Low, Up, censor.codes)~ AG,
truncation=(censor(tlow, tup, tcode)), data=allinoculation, dist='extreme')
where I arrange the data for all observations as tlow=0, lup=NA and tcode=1.
What disturbed me and originate this mail is the fact that I get the same
results using the truncation argument or simple fit a model with and extreme
distribution.
I checked also that extreme value distribution uses a identity link.
So can any one can help me in:
1- find out if there some wrong in my formulation
2- or someone have other suggestion to fit the gompertz distribution.?
Thanks in advance,
Kind regards;
Gustavo Monti
Wageningen University and Research
Department of Animal Sciences
Quantitative Veterinary Epidemiology Group
Marijkeweg 40
P.O.Box 338
6700 AH Wageningen
The Netherlands
|