Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
| From | Henry <jakanyada@gmail.com> |
| To | statalist@hsphsun2.harvard.edu |
| Subject | st: Problem of Monotone Likelihood in Cox Regression |
| Date | Mon, 22 Feb 2010 00:57:26 +0000 |
Dear All, I am fitting a multivariable Cox regression model with shared frailty to help model the within group correlation among practices since my data is based GP prescription database. In fitting the model, the likelihood converges to a finite estimate (although after 45 iterations) but one of the covariates (factors) diverges to infinity. The factor has 4 levels/groups with one of the levels having no "failure" thus giving this infinite estimate. This phenomenon has been referred to by Heinze and Shemper as monotone likelihood. A solution to this problem in Cox regression has been implemented in SAS and SPLUS/R http://www.meduniwien.ac.at/msi/biometrie/publikationen/Separata/Heinze_Ploner_2002_CompMethProgBiom.pdf I have tried to check on statlist but only managed to find a request from Jeff n the same, http://www.stata.com/statalist/archive/2008-02/msg00847.html. My question is, has this been implemented in Stata? If so is it advisable to use this in a model with shared frailty? How does this affect the post estimation test for proportionality assumption and/or goodness-of-fit? If this has not been implemented, can someone advise on the way forward in this situation. Many thanks, Henry * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/