Statalist


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

RE: st: Multilevel modelling of survival data


From   "Justin B Echouffo Tcheugui" <[email protected]>
To   <[email protected]>
Subject   RE: st: Multilevel modelling of survival data
Date   Wed, 25 Mar 2009 15:17:57 -0000

Dear Marteen, 
I am using the population averaged Cox model in my analysis of clustered data form a trial:  stcox with the option-cluster (). 
Not being in possession of the book you advised, I thought that I could ask you another question.  
The marginal (Cox) model uses the sandwich estimator to obtain standard errors. It seems that with a total number of clusters below 40 as it is my case, one has to correct the confidence interval as with less than 40 clusters the sandwich estimator is biased downwards. 
I am not sure about how to correct for the confidence intervals in general and in Stata in particular. How do I do that in Stata?  
I came across an article suggesting theoretical approaches (use of a t distribution or jacknife standard errors) to do the correction, but it does not really tell how to implement the suggested approaches in a package. 
What about the following ways of implementing the approaches below?  
 
1- Using the t distribution instead of the Z distribution to derive CI: Would the derivation of the confidence interval using the robust SE on the log scale and the value of the t distribution with n-2 degree of freedom instead of the usual 1.96 value, where n is the total number of clusters randomised, be a sensible approach?  I would then do and exponentiation of the CI limits to have the corrected standard errors. I am thinking of something like this: Estimates ± t30 x robust SE on the log scale. Does that sounds like a sensible application of this correction. 

2- Using a jacknife estimator instead of the sandwich estimator:  

xi: jacknife: stcox  i.randomgp, cluster(clinic) 

What do you think? I hope that I am not completely off track 
Many thanks 
Justin B.   

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Justin B Echouffo Tcheugui
Sent: 16 March 2009 16:17
To: [email protected]
Subject: RE: st: Multilevel modelling of survival data

Dear Marteen, 
I tried the command stcox with the option -shared () as you advised. As
you can see below I am not having the desired output 
xi: stcox  i.randomgp, shared(clinic)
i.randomgp   _Irandomgp_0-1     (naturally coded; _Irandomgp_0 omitted)

failure _d:  event
analysis time _t:  followup_time

Fitting comparison Cox model:

Estimating frailty variance: 
numerical derivatives are approximate flat or discontinuous region
encountered 
Iteration 0:   log profile likelihood = -2482.4152 
could not calculate numerical derivatives flat or discontinuous region
encountered r (430);

I tried adding the option - difficult, hoping that it will help but it
did not 
xi: stcox  i.randomgp, shared(practice) difficult
i.randomgp       _Irandomgp_0-1     (naturally coded; _Irandomgp_0
omitted)

failure _d:  event
analysis time _t:  followup_time

Fitting comparison Cox model:

Estimating frailty variance:
numerical derivatives are approximate flat or discontinuous region
encountered
Iteration 0:   log profile likelihood = -2482.4152  
could not calculate numerical derivatives flat or discontinuous region
encountered r(430);

Could you please advise on this?  
Many thanks 
Justin B. 

-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Maarten buis
Sent: 16 March 2009 10:58
To: stata list
Subject: RE: st: Multilevel modelling of survival data


--- On Mon, 16/3/09, Justin B Echouffo Tcheugui wrote:
> > in this case the option - cluster() in this case does
> > not fit the clinic into the model as a random
> > intercept 

--- On Mon, 16/3/09, Maarten buis wrote: 
> That is correct. 

A point on terminology again: When discussing the 
distrinction between these models, the models estimated 
with the -cluster()- option are sometimes known as 
population averaged models, while the random intercept 
models are sometimes known as individual specific models.

-- Maarten

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

http://home.fsw.vu.nl/m.buis/
-----------------------------------------





      

*
*   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/

*
*   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 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index