[R] OOT: frailty-multinivel

Thomas Lumley tlumley at u.washington.edu
Wed Nov 24 18:00:11 CET 2004

On Wed, 24 Nov 2004, Kjetil Brinchmann Halvorsen wrote:

> Hola!
> I started to search for information about multilevel survival models, and
> found frailty in R. This seems to be something of the same, is it the same?

More or less. "Shared frailty" models are the same as hierarchical/mixed 
survival models.  R uses a fitting method that is equivalent to maximum 
likelihood only when exp(random effects) has a Gamma distribution.  The 
"survival" package can fit random intercept models; the new "kinship" 
package fits much more general mixed models.

[I will put in my usual objection to the term "multilevel model"
being used to refer solely to models that include unmeasured 

> Then: why the name frailty (weekness?)

The idea is that 'weaker' individuals fail earlier than `stronger' 
individuals for the same values of covariates.  The concept has been used 
both for modelling correlation between survival times and to motivate 
parametric models that give an initially decreasing hazard.  I have a 
vague impression that the term originated in Scandinavia somewhere.


More information about the R-help mailing list