[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

AW: st: survival analysis - iterations "not concave"

From   "Simon Oertel" <>
To   <>
Subject   AW: st: survival analysis - iterations "not concave"
Date   Sat, 18 Aug 2007 14:09:29 +0200

Thanks Paul for your reply and your answer. I suggested that "not
concave" means sth. like this but I was not sure about it. So again,
thanks for your help. 


-----Ursprüngliche Nachricht-----
[] Im Auftrag von E. Paul
Gesendet: Freitag, 17. August 2007 13:23
Betreff: Re: st: survival analysis - iterations "not concave"

Maximizing likelihood is like climbing a hill to the top. "Not concave" 
simply means that when Stata calculates the numerical derivatives around

the current point, it might appear to be in a valley or a saddle, with 
no clear direction to the top.


Simon Oertel wrote:
> Dear Stata list members,
> I am using parametric models for a survival analysis. To figure out 
> which model fits my data best, I used the common procedures 
> (theoretical implications, graphical plot of the cumulative 
> hazard-rate against the cox-snell residuals, AIC values). For 
> calculating the AIC values of the different (standard) models 
> (exponential, weibull, gompertz, log-logistic, log-normal, gamma) I 
> first used the "streg" command and fit each of them to my data. 
> Finally, I found the model that will fit my data best (weibull) - but 
> while running the iteration-processes for the log-logistic, log-normal

> and gamma model, STATA tells me for the first (and for the 
> log-logistic also for the second) iteration that it is "not concave".
> Also if I will not use a log-logistic, log-normal or gamma model for 
> my data, I am really interested in figuring out what this message 
> implicates. I suggested that it is not a big problem, as long as the 
> following iterations (especially the last one) will be concave. I 
> think that the message "not concave" only tells me that during the 
> iteration process STATA found a local maximum (what depending on the 
> local maximum could still be a problem)- but again I am not sure about

> that.
> I appreciate any answer and help.
> Thanks
> Simon
> *
> *   For searches and help try:
> *
> *
> *

E. Paul Wileyto, Ph.D.
Assistant Professor of Biostatistics
Tobacco Use Research Center
School of Medicine, U. of Pennsylvania
3535 Market Street, Suite 4100
Philadelphia, PA  19104-3309

Fax: 215-746-7140 

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index