[R] anova.coxph with subsets of data

[David Winsemius]

On Jan 28, 2014, at 10:32 AM, Oscar Rueda wrote:

> Dear list, 
> I'm using the rms package to fit some Cox models. I run anova() on them to obtain sequential p-values, but I'm getting strange results when I run it on a subset of the data. 
> 
> Following the example on the help page of anova.coxph:
>> library(rms)
>> data(ovarian)
>> fit <- coxph(Surv(futime, fustat) ~ resid.ds *rx + ecog.ps, data = ovarian)
>> anova(fit)
>> fit2 <- coxph(Surv(futime, fustat) ~ resid.ds +rx + ecog.ps, data=ovarian)
>> anova(fit2,fit)
> 
> would give me the same result, as expected. 
> But If I do
> 
>> fit <- coxph(Surv(futime, fustat) ~ resid.ds *rx + ecog.ps, data = ovarian, subset=ovarian$age>50)
>> anova(fit)
>> fit2 <- coxph(Surv(futime, fustat) ~ resid.ds +rx + ecog.ps, data=ovarian, subset=ovarian$age>50)
>> anova(fit2,fit)
> 
> The first p-value seems to be wrong. 

Wrong ... in what way?

> Would anybody please explain to me why?

Perhaps because anova is a generic function and you were expecting anova.cph to be used but the coxph function is not from pkg:rms but rather from pkg:survival. 

methods(anova)  # with both rms and survival loaded

>  methods(anova)
 [1] anova.coxmelist*   anova.coxph*       anova.coxphlist*   anova.glm         
 [5] anova.glmlist      anova.glmmPQL*     anova.lm           anova.loess*      
 [9] anova.loglm*       anova.mlm          anova.negbin*      anova.nls*        
[13] anova.polr*        anova.rms*         anova.rq           anova.rqlist      
[17] anova.survreg*     anova.survreglist*

> 
> Cheers, 
> Oscar
> 
> PS. I'm using R 3.0.1.
> 
> Oscar M. Rueda, PhD.
> Postdoctoral Research Fellow, Caldas Lab, Breast Cancer Functional
> Genomics.
> University of Cambridge. Cancer Research UK Cambridge Institute.
> Li Ka Shing Centre, Robinson Way.
> Cambridge CB2 0RE
> England
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

David Winsemius
Alameda, CA, USA