R devel - Nov 2001 - stepwise algorithm step() on coxph() (PR#1159)

Full_Name: Jerome Asselin
Version: 1.3.1
OS: MacOS 9.2
Submission from: (NULL) (142.103.173.46)



The step() function attempts to calculate the deviance of fitted models even if
does not really need it. As a consequence, the step() function gives an error
when it is used with coxph(). (There is currently no method to calculate the
deviance of coxph() fits.) The code below gives an example of how the deviance
is not necessary (at least for the default parameters of step()). Note that the
deviance does not seem necessary even for linear models (lm()).

#########################
library(survival)

set.seed(3)

#Create a fake dataset
y <- rgamma(10,1)
x1 <- rnorm(10)
x2 <- rnorm(10)
x3 <- rnorm(10)

step(coxph(Surv(y) ~ x1+x2+x3)) #Gives an error
step(lm(y ~ x1+x2+x3))
step(glm(y ~ x1+x2+x3))
step(aov(y ~ x1+x2+x3))

#Mask the deviance() function (Temporary patch??)
deviance <- function(...) NA

step(coxph(Surv(y) ~ x1+x2+x3)) #Gives no error
step(lm(y ~ x1+x2+x3)) #Same result as before
step(glm(y ~ x1+x2+x3))#Same result as before
step(aov(y ~ x1+x2+x3))#Same result as before

rm(deviance)


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

This is not an error.  Nobody said step would work with coxph models.
The help page says

  object: an object representing a model of an appropriate class. This
          is used as the initial model in the stepwise search.

and you are jumping to conclusions.  I've added a note to the help file.

There is an extractAIC.coxph method for use with stepAIC in package MASS.
That does work in your example.

I had a report about multinom fits and step() recently: same answer, it is
stepAIC that works, not step.

Think of step() as minimal implementation to provide some compatibility
with that in S (which only offers lm and glm methods).  And as step() in S
works with deviances, the compatibility function does too.


On Mon, 5 Nov 2001 jerome@hivnet.ubc.ca wrote:
> Full_Name: Jerome Asselin
> Version: 1.3.1
> OS: MacOS 9.2
> Submission from: (NULL) (142.103.173.46)
>
>
>
> The step() function attempts to calculate the deviance of fitted models
even if
> does not really need it. As a consequence, the step() function gives an
error
> when it is used with coxph(). (There is currently no method to calculate
the
> deviance of coxph() fits.) The code below gives an example of how the
deviance
> is not necessary (at least for the default parameters of step()). Note that
the
> deviance does not seem necessary even for linear models (lm()).
>
> #########################
> library(survival)
>
> set.seed(3)
>
> #Create a fake dataset
> y <- rgamma(10,1)
> x1 <- rnorm(10)
> x2 <- rnorm(10)
> x3 <- rnorm(10)
>
> step(coxph(Surv(y) ~ x1+x2+x3)) #Gives an error
> step(lm(y ~ x1+x2+x3))
> step(glm(y ~ x1+x2+x3))
> step(aov(y ~ x1+x2+x3))
>
> #Mask the deviance() function (Temporary patch??)
> deviance <- function(...) NA
>
> step(coxph(Surv(y) ~ x1+x2+x3)) #Gives no error
> step(lm(y ~ x1+x2+x3)) #Same result as before
> step(glm(y ~ x1+x2+x3))#Same result as before
> step(aov(y ~ x1+x2+x3))#Same result as before
>
> rm(deviance)
>
>
>
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
> r-devel mailing list -- Read
http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
> Send "info", "help", or "[un]subscribe"
> (in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
>
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
>
-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._