Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

Re: st: Interpreting streg, time dist(weibull) coefficients as a time metric


From   Steve Samuels <[email protected]>
To   [email protected]
Subject   Re: st: Interpreting streg, time dist(weibull) coefficients as a time metric
Date   Sun, 2 Mar 2014 20:54:37 -0500


Your reference point is not the mean of the Weibull distribution. See:
http://en.wikipedia.org/wiki/Weibull_distribution or
http://www.math.uah.edu/stat/special/Weibull.html (Generalized Weibull
distribution).

A better choice of reference point would be the estimated median from
the Kaplan Meier curve, supplemented by the other quartiles
(-stsum-).

Steve
[email protected]


> On Mar 2, 2014, at 8:53 AM, Yvon Pho <[email protected]> wrote:
> 
> 
> If I am understanding the response below, am I correct to state:
> 
> The estimated (mean) survival time is exp(5.20779/1.635698), or 24.14
> days. A one unit change in X1 results in a 290.1
> (100*[exp(2.226613/1.635698)-1]) percentage increase in t, or 70.03
> days (24.14*2.9). Similarly, a one unit change in X2 results in a 16.2
> (100*[exp(-0.2897575/1.635698)-1]) percentage decrease in t, or 3.92
> days (24.14*0.162).



On Sat, Mar 1, 2014 at 3:31 PM, Steve Samuels <[email protected]> wrote:
> A good start would be to read the Manual entry for -streg-, section on
> "Weibull and Exponential Models":
> 
> "The AFT model is written as log(tj) = xj b*  + zj where zj has an
> extreme-value distribution scaled by 𝞂 "
> 
> then note that 𝞂= 1/p in the Weibull output. Also zj does _not_ have
> value mean zero. See:
> http://www.math.uah.edu/stat/special/ExtremeValue.html
> 
> 
> So
> cons* = 5.20779/1.635698
> b1* = 2.226613/1.635698
> b2* = -0.2897575/1.635698
> 
> Finally, figure out how to interpret coefficients when log t = a + b X +z
> 
> Hint: Subtract log(t|X) = a + b X from log(t|X +1)= a + b X + b; and
> conclude that the percentage change in t associated with a one unit
> change in X is 100 [exp(b)-1].
> 
> 
> 
> On Feb 28, 2014, at 12:27 PM, Yvon Pho <[email protected]> wrote:
> 
> Hello.
> 
> I am running an accelerated failure time model with a Weibull distribution
> to estimate the time (in days) to my failure event. The Stata command is
> streg x1 x2, time dist(weibull)
> 
> My coefficient for x1 is 2.226613, for x2 is -0.2897575, constant is
> 5.20779, and p is 1.635698.
> 
> Am I correct to say that the mean survival time (baseline) when x1
> (continuous covariate) and x2 (binary covariate) are equal to 0 is 182.69
> days (or exp(5.20779))? How do I interpret the coefficients for x1 and x2,
> and convert them into a days metric?
> 
> Any help would be greatly appreciated.
> 
> Yvon
> *
> *   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/
> 
> 
> *
> *   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/

*
*   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/


*
*   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 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index