[R] survreg with numerical covariates

[Richard Mott]

Ales - I have identified the problem. It is caused by missing data. In 
my previous posting I was analysing a dataset which happened to have 190 
missing values in the covariate. I compared two models:

r0 <- survreg(s ~1)
r1 <- survreg(s~x)
anova(r0,r1)

I was using anova to investigate the significance of adding covariates 
to the model, rather than printing out the results of a survreg() 
directly, and this was the problem. The behaviour of anova() is subtly 
different when applied to lm() or survreg().  With lm(), if the models 
compared have different numbers of missing observations, you get a 
warning, but with survreg(), the difference in observation count shows 
up as a large change in the df and deviance, giving the appearence that 
the covariate has been fitted as a factor with 191 levels.

So my fault, although possibly anova(survreg()) should be consistent 
with anova(lm()).


-Richard


Ales Ziberna wrote:

> As you can se from the example bellow, survreg works prefeclty fine 
> with numerical values. (I'm runing R2.0.1 on WinXP(SP2) and 32bit AMD 
> with survival version 2.17.).
>
> As the posting guide asks, plese provide a small example.
>
> Ales Ziberna
>
>
>> library(survival)
>
> Loading required package: splines
>
>> data(cancer)
>> survreg(Surv(time, status)~age,data=cancer)
>
> Call:
> survreg(formula = Surv(time, status) ~ age, data = cancer)
>
> Coefficients:
> (Intercept)         age
> 6.88712062 -0.01360829
>
> Scale= 0.7587515
>
> Loglik(model)= -1151.9   Loglik(intercept only)= -1153.9
>        Chisq= 3.91 on 1 degrees of freedom, p= 0.048
> n= 228
>
> ----- Original Message ----- From: "Richard Mott" 
> <Richard.Mott at well.ox.ac.uk>
> To: <r-help at stat.math.ethz.ch>
> Sent: Tuesday, April 26, 2005 11:32 PM
> Subject: [R] survreg with numerical covariates
>
>
>> Does anyone know if the survreg function in the survival package can 
>> fit numerical covariates ?
>>
>> When I fit a survival model of the form
>>
>> survreg( Surv(time,censored) ~ x )
>>
>> then x is always treated as a factor even if it is numeric (and even 
>> if I try to force it to be numeric using as.numeric(x). Thus, in the 
>> particular example I am analysing, a simple numerical covariate 
>> becomes a factor with 190 levels. Is this the expected behaviour ? Am 
>> I doing something wrong ?
>>
>> I am running R 2.0.1 on a 64bit Debian Linux system, and version 2.17 
>> of the survival package
>>
>>
>> Thanks
>>
>> Richard Mott
>>
>> -- 
>> ----------------------------------------------------
>> Richard Mott       | Wellcome Trust Centre
>> tel 01865 287588   | for Human Genetics
>> fax 01865 287697   | Roosevelt Drive, Oxford OX3 7BN
>>
>> ______________________________________________
>> R-help at stat.math.ethz.ch mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide! 
>> http://www.R-project.org/posting-guide.html
>>
>>
>


-- 
----------------------------------------------------
Richard Mott       | Wellcome Trust Centre 
tel 01865 287588   | for Human Genetics
fax 01865 287697   | Roosevelt Drive, Oxford OX3 7BN