RE: st: some questions regarding stcox

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Fri, 4/6/10, XU Moqi wrote privately:
> I'm only standing on my brain regarding the goodness of fit
> -  Imagine I have stcox a and stcox b, how do I do the test?

Note: <http://www.stata.com/support/faqs/res/statalist.html#private>

It all depends on what a and b are:

if a is something like x1 x2, and b is something like x1 x2 x3 x4, than
I would do something like this:

stcox x1 x2 x3 x4
test x3 x4

This test the hypothesis that the coeficients fo x3 and x4 are zero.
This hypothesis is equivalent to the model -stcox x1 x2-, so it is a
test of model 2 versus model 1.

If a and b are completely different variables then it is very hard
(i.e. impossible) to compare the models, as the two models are not
nested. Stata does report BICs and AICs. However, that surprises
me a bit, as I always thought of Cox regression as not being a full
likelihood (but a partial likelihood technique) due to the fact that
we are not estimating the baseline hazard function. This would mean
that the reported likelihood isn't really a likelihood, which means
that you shouldn't be able to compute the BICs and AICs. Maybe
somebody else can confirm that the AICs and BICs are OK after
-stcox-?

-- Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------



      

*
*   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/