[R] Coxme penalized log-likelihood mismatch
Therneau, Terry M., Ph.D.
therneau at mayo.edu
Thu Apr 9 14:26:44 CEST 2015
The perils of backwards compatability....
During computation the important quantity is loglik + penalty. That is what is contained
in the third element of the loglik vector.
Originally that is also what was printed, but I later realized that for statistical
inference one wants the loglik and penalty to be separate, e.g. the AIC is based on the
loglik and the degrees of freedom. So I changed the printout to give
loglik at beta=0, integrated loglik at solution, ordinary loglik at solution
(Integrated and ordinary are the same at beta=0).
But, because I was worried that some people would have been reading the results in a
program (like you are), I didn't want to break their code so left the internal variable
alone. It appears that I traded "maybe safer" for "certain confusion".
On 04/09/2015 05:00 AM, r-help-request at r-project.org wrote:
> I need to extract the penalized log-likehood term from coxme objects but I find the values stored whitin the object different than the penalized term given in the summary output of coxme function. Both the Null and the Integrated values are identical but the penalized is always off.?
> Any thoughts on why and how i can extract the right value to compute the AIC myself? (I know an AIC value is given in the output but I need to compute it myself inside a loop)?
> Many thanks,?
> Alexandre Lafontaine
More information about the R-help