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 on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: Observations included in stratified Cox proportional hazards model


From   Steve Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Observations included in stratified Cox proportional hazards model
Date   Fri, 8 Mar 2013 12:54:29 -0500

Paul Davis:

Designating persons as strata is a faulty strategy. In panel studies,
standard errors are based on between-panel variation. In the Cox model,
stratum differences contribute nothing to standard errors. These will be
based on the assumption that observations on a single person are
independent, conditional on the model. The preferred approach is to
declare person as a cluster variable.. See section 3.2 of
http://www.stata.com/support/faqs/statistics/multiple-failure-time-data/.

If there are issues applying the FAQ to your data, we'd want to know
more about your study questions and data, including how you define your
events, whether the start times and censoring times in your -stset- are
exact; and how they relate to the study interview times.

A minor point: "strata" is is an option of -stcox-,
not a command.

As to your original question: Stratification did not drop anybody. You need to
re-examine your reasons for thinking that it should have. 


Steve


> 
> On Wed, 27 Feb 2013 17:25:02 +0000 "Davis, Paul" <pdavis@bus.wisc.edu> wrote:
> 
> I have panel data in which persons are at risk for repeated events. To model hazard, I've chosen to use the Cox model, stratified by person. Using the strata command, my sample should be lower than it is with an unstratified model as the analysis drops certain individuals. However, I've noticed that Stata reports the same number observations and subjects when using the stratified and unstratified models. Is there a command to isolate only those observations that are used (or, alternatively, those that were dropped)? Using "e(sample)==1" gives the same result as that which is reported in the output tables.



*
*   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/


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