R help - May 2003 - To update() or not to update()?

Hi,

Suppose I have:
  # Fit a base model
  d1.ph <- coxph(Surv(start, stop, event)~
               ejec + diavol + score + smoking +
               beta + surg.done,
               data = data.frame(foo))

  summary(update(d1.ph, . ~ . + td1))
  summary(update(d1.ph, . ~ . + td2)) 

As I have many columns in my data frame, foo, called td's.  e.g. td1, td2, 
td3, ....  And I'd like to add one column each time.  What is the 
recommended way to do this?  Whether I should do what I did above, or 
should I do something like:
  td1.ph <- coxph(Surv(start, stop, event)~
                ejec + diavol + score + smoking + 
                beta + surg.done + td1,
                data = data.frame(foo)) 

  td2.ph <- coxph(Surv(start, stop, event)~
                ejec + diavol + score + smoking + 
                beta + surg.done + td2,
                data = data.frame(foo))  

I've done a system.time() on it and update() doesn't seem to be much (if
at all) faster.  Is there an advantage of using update() then (other than 
that the codes look neater)?


-- 
Cheers,

Kevin

------------------------------------------------------------------------------
/* Time is the greatest teacher, unfortunately it kills its students */

--
Ko-Kang Kevin Wang
Master of Science (MSc) Student
SLC Tutor and Lab Demonstrator
Department of Statistics
University of Auckland
New Zealand
Homepage: http://www.stat.auckland.ac.nz/~kwan022
Ph: 373-7599
    x88475 (City)
    x88480 (Tamaki)

On Mon, 19 May 2003, Ko-Kang Kevin Wang wrote:
> Suppose I have:
>   # Fit a base model
>   d1.ph <- coxph(Surv(start, stop, event)~
>                ejec + diavol + score + smoking +
>                beta + surg.done,
>                data = data.frame(foo))
> 
>   summary(update(d1.ph, . ~ . + td1))
>   summary(update(d1.ph, . ~ . + td2)) 
> 
> As I have many columns in my data frame, foo, called td's.  e.g. td1,
td2,
> td3, ....  And I'd like to add one column each time.  What is the 
> recommended way to do this?  Whether I should do what I did above, or 
> should I do something like:
>   td1.ph <- coxph(Surv(start, stop, event)~
>                 ejec + diavol + score + smoking + 
>                 beta + surg.done + td1,
>                 data = data.frame(foo)) 
> 
>   td2.ph <- coxph(Surv(start, stop, event)~
>                 ejec + diavol + score + smoking + 
>                 beta + surg.done + td2,
>                 data = data.frame(foo))  
> 
> I've done a system.time() on it and update() doesn't seem to be
much (if
> at all) faster.  Is there an advantage of using update() then (other than 
> that the codes look neater)?
You are calling update.default on a coxph fit, so it is just neater.  It
need not be for other model-fitting classes.

BTW, addterm (MASS) would automate this for you, adding all the columns 
one at a time and collating the results.

-- 
Brian D. Ripley,                  ripley at 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 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Hi Kevin,
Here there is a possible solution

Extract from the base model the response, the design matrix, (possible)
strata,....Note: fit it with argument x=T and y=T
    o<-list()
    X0<-d1.ph $x
    Y<-d1.ph $y
stra<-d1.ph $strata
....
    for(i in ....){ #what are the positions of "td" variables in the
dataframe?
            X<-cbind(X0,X[,i])
            o[[length(o)+1]]<-
agreg.fit(X,Y,strata=stra,control=coxph.control(eps 1e-04),init=rep(0,ncol(X)),
method="efron",rownames=NULL))
                   }

o is a list with all the fitted models.
agreg.fit() is faster that coxph() (that uses agreg.fit). It is something
like glm() and glm.fit().

Hope this helps you.
best,
vito


----- Original Message -----
From: "Ko-Kang Kevin Wang" <kwan022 at stat.auckland.ac.nz>
To: "R Help" <r-help at stat.math.ethz.ch>
Sent: Monday, May 19, 2003 11:31 AM
Subject: [R] To update() or not to update()?

> Hi,
>
> Suppose I have:
>   # Fit a base model
>   d1.ph <- coxph(Surv(start, stop, event)~
>                ejec + diavol + score + smoking +
>                beta + surg.done,
>                data = data.frame(foo))
>
>   summary(update(d1.ph, . ~ . + td1))
>   summary(update(d1.ph, . ~ . + td2))
>
> As I have many columns in my data frame, foo, called td's.  e.g. td1,
td2,
> td3, ....  And I'd like to add one column each time.  What is the
> recommended way to do this?  Whether I should do what I did above, or
> should I do something like:
>   td1.ph <- coxph(Surv(start, stop, event)~
>                 ejec + diavol + score + smoking +
>                 beta + surg.done + td1,
>                 data = data.frame(foo))
>
>   td2.ph <- coxph(Surv(start, stop, event)~
>                 ejec + diavol + score + smoking +
>                 beta + surg.done + td2,
>                 data = data.frame(foo))
>
> I've done a system.time() on it and update() doesn't seem to be
much (if
> at all) faster.  Is there an advantage of using update() then (other than
> that the codes look neater)?
>
>
> --
> Cheers,
>
> Kevin
>
> --------------------------------------------------------------------------
----
> /* Time is the greatest teacher, unfortunately it kills its students */
>
> --
> Ko-Kang Kevin Wang
> Master of Science (MSc) Student
> SLC Tutor and Lab Demonstrator
> Department of Statistics
> University of Auckland
> New Zealand
> Homepage: http://www.stat.auckland.ac.nz/~kwan022
> Ph: 373-7599
>     x88475 (City)
>     x88480 (Tamaki)
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help