[R] R coxph Method=df (Question on methods)

[David Parker]

Dear R help,

## Example from
http://stat.ethz.ch/R-manual/R-patched/library/survival/html/frailty.html

# Random institutional effect
coxph(Surv(time, status) ~ age + frailty(inst, df=4), lung)

I am trying to understand what exactly happens when method=df is used ? The
R documentation pages says that if /df/ is used then the degrees of freedom
for random effects is fixed to that number. I have few questions regarding
this

1)  How do I specify the correct df? Is it simply any number that leads to
model convergence? Is there any sensible way to specify this? Dr. Terry
Therneau once said that he chose the df=4 out of nowhere, just like that for
this example.

https://stat.ethz.ch/pipermail/r-devel/2012-April/063883.html


2) My main question is regarding the methods. What exactly the method=df is
doing? I tried to search the vignette but there were very limited details.
Does using method=df fixes the scale parameter of my frailty distribution?


3) Finally, a practical question. I am trying to fit a frailty model that
have convergence problem. When I use method=df option and adjust the degrees
of freedom then it leads to model convergence which otherwise was not
possible to achieve? 
Is it statistically correct to use this option of method=df where I fix the
df to a certain number so that my model converges? 

Suggestions (and criticisms :) ) are all  welcome. Thank you very much.

Regards,
D








--
View this message in context: http://r.789695.n4.nabble.com/R-coxph-Method-df-Question-on-methods-tp4640841.html
Sent from the R help mailing list archive at Nabble.com.