Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Small strata and test of proportional hazard assumption


From   Steven Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Small strata and test of proportional hazard assumption
Date   Thu, 16 Dec 2010 11:48:39 -0500

--

With strata so small, you can't check the PH assumption, not with plots and not with residuals. The checks require a reasonable estimate of the survival curve in each stratum, and you can't get one. The purpose of stratum checks is to assess whether a categorical variable can be a predictor in a Cox model, but a variable such as yours can't be (unless it is ordinal), because the parameters/failures ratio is too high for valid inference. (The max should probably be < 1:10) If you have many small groups with possibly different survival curves, use a frailty (random-effects) model. In -stcox--, frailty models are implemented with the -shared()- and (optionally) the -vce(cluster)- options.

Steve

Steven J. Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
Voice: 845-246-0774
Fax:    206-202-4783

On Dec 16, 2010, at 4:24 AM, Grethe Søndergaard wrote:

Hello,

I am analysing my data using a stratified cox-regression and am trying
to find out if the proportional hazards assumptions in my model is
violated. I wanted to test this using the command estat phtest.
However, I have become aware that you cannot do this, when the stratas
are very small (each of the stratas contains 2-6 individuals).
I haven’t been able to find any information about how to test the PH
assumption when you have very small strata. Does anyone know of such
test? (Otherwise I am thinking about first running the full model.
Afterwards I could stsplit my data in two, and run these models. By
lrtest I could probably test whether there is better goodness of fit
in the two time-specific models compared to the full model).

Grethe

*
*   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/


*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index