Dear R helpers,
I am currently trying to analyze data with a cox proportional hazard survival
analysis. For one of my datasets, the proportional hazards assumption is
violated. Reading the literature, it seems that the weighted version of cox PH
(function coxphw() ) is a good alternative in case of non proportional hazards.
However, the function coxphw() does not seem to take categorical explanatory
variables and I cannot figure out the reason why. Is there a statistical reason
for that? Could you help me please?
In a few words, my data represents 6 species of fishes (~16 individuals per
species) tested for their reaction towards novel objects. The response variable
is the time to approach the novel object (the experiments was stopped after 5
minutes, so the data is right censored) ad the explanatory variable is species
identity. When I run a coxph() model in R, it works with no problems (model1
below). If I run the same model but with the coxphw() function instead of the
coxph() it gives me an error message (model2 below). If I coerce the explanatory
variable into a numeric format, it also works (model3 below). However, I am not
sure this last approach is appropriate, as I guess there must be a reason why
coxphw() cannot deal with my response variable in the categorical (factor)
format. Can anyone explain?
> model1 <- coxph(Surv(time=O1_approach_sec, event=O1_approach) ~ species,
data=personew)
> model2 <- coxphw(Surv(time=O1_approach_sec, event=O1_approach) ~
species, data=personew, AHR=T)
Erreur dans weights[, 1] : indice hors limites> model3 <- coxphw(Surv(time=O1_approach_sec, event=O1_approach) ~
as.numeric(species), data=personew, AHR=T)
Best regards
Simon
___________________________
Simon Gingins
PhD student
University of Neuch?tel
Institute of Biology
Department of Behavioural Ecology
Rue Emile-Argand 11
2000 Neuch?tel
Switzerland
Office: +41 32 718 31 09
research: http://www2.unine.ch/ethol/gingins_simon
photos: www.simongingins.com
[[alternative HTML version deleted]]