Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Adam Olszewski <adam.olszewski@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: Degrees of freedom in STPM2 |
Date | Mon, 15 Oct 2012 19:48:55 -0400 |
Dear listreaders, I have been wondering about an issue regarding the (user written, SSC-derived) STPM2 command. I am fitting flexible parametric models in datasets (disease subtypes) of different sizes. Some are smaller than the others (ranging from 200 to 2500 in size) and in some the number of events is as low as 25 (out of ~200). Does the "10:1" rule (10 events per degree of freedom) used in Cox modelling apply equally to these models? Since I'm studying a number (about 8) of variables in larger datasets, should I remove some of the variables in a smaller dataset to stick to the 10:1 rule? Do the degrees of freedom related to the spline knots need to be counted? All of this interestingly does not make significant difference in terms of the parameters of interest, however I would like to avoid disputes with potential reviewers regarding different models applied to subtypes of the same disease. For clarity and logical consistency, I could fit the same model in all datasets using all variables expected to affect the outcome, but for the smaller datasets the 10:1 rule would be grossly violated. I wonder if there is a definite guidance with regards to this. This might be even more complex with competing risk models. Thanks for any insight! Best regards, Adam Olszewski * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/