R devel - Nov 2002 - survreg (survival) reports erroneous results for left-censored data (PR#2287)

Full_Name: Tim Cohn
Version: 1.6.1
OS: Macintosh OS X
Submission from: (NULL) (130.11.34.250)


The Mac version of survreg does not handle left-censored data correctly (at
least the results are not what I get doing it other ways, and they are not the
same as I get running R 1.6.1 in Windows 98se;  the Windows 98 results are
correct).  On the windows version of R 1.6.1. 
>
survreg(Surv(c(0,1,-1),c(T,T,F),type='left')~1,dist='gaussian')
Call:
survreg(formula = Surv(c(0, 1, -1), c(T, T, F), type = "left") ~ 
    1, dist = "gaussian")

Coefficients:
(Intercept) 
 -0.2321569 

Scale= 1.161135 

Loglik(model)= -4.1   Loglik(intercept only)= -4.1
n= 3 


================
The Macintosh OS X result for the same command is:
>
survreg(Surv(c(0,1,-1),c(T,T,F),type='left')~1,dist='gaussian')
Call:
survreg(formula = Surv(c(0, 1, -1), c(T, T, F), type = "left") ~ 
    1, dist = "gaussian")

Coefficients:
(Intercept) 
 -0.7373613 

Scale= 1.366282 

Loglik(model)= -5.6   Loglik(intercept only)= -5.6
n= 3 


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

Using R-1.6.1 on OS X (Darwin)

Call:
survreg(formula = Surv(c(0, 1, -1), c(T, T, F), type = "left") ~
     1, dist = "gaussian")

Coefficients:
(Intercept)
  -0.2321569

Scale= 1.161135

Loglik(model)= -4.1   Loglik(intercept only)= -4.1
n= 3

No problemo. And, in fact, I get the same results in
the R-1.6.0 Carbon version.

--- Jan

On Wednesday, November 13, 2002, at 02:05 PM, tim@timcohn.com wrote:
> Full_Name: Tim Cohn
> Version: 1.6.1
> OS: Macintosh OS X
> Submission from: (NULL) (130.11.34.250)
>
>
> The Mac version of survreg does not handle left-censored data  
> correctly (at
> least the results are not what I get doing it other ways, and they are  
> not the
> same as I get running R 1.6.1 in Windows 98se;  the Windows 98 results  
> are
> correct).  On the windows version of R 1.6.1.
>
>>
survreg(Surv(c(0,1,-1),c(T,T,F),type='left')~1,dist='gaussian')
> Call:
> survreg(formula = Surv(c(0, 1, -1), c(T, T, F), type = "left") ~
>     1, dist = "gaussian")
>
> Coefficients:
> (Intercept)
>  -0.2321569
>
> Scale= 1.161135
>
> Loglik(model)= -4.1   Loglik(intercept only)= -4.1
> n= 3
>
>
> ================>
> The Macintosh OS X result for the same command is:
>
>>
survreg(Surv(c(0,1,-1),c(T,T,F),type='left')~1,dist='gaussian')
> Call:
> survreg(formula = Surv(c(0, 1, -1), c(T, T, F), type = "left") ~
>     1, dist = "gaussian")
>
> Coefficients:
> (Intercept)
>  -0.7373613
>
> Scale= 1.366282
>
> Loglik(model)= -5.6   Loglik(intercept only)= -5.6
> n= 3
>
>
> -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 
> .-.-.-.-.-
> r-devel mailing list -- Read  
> http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
> Send "info", "help", or "[un]subscribe"
> (in the "body", not the subject !)  To:  
> r-devel-request@stat.math.ethz.ch
> _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._ 
> ._._._._
>
>
==Jan de Leeuw; Professor and Chair, UCLA Department of Statistics;
Editor: Journal of Multivariate Analysis, Journal of Statistical  
Software
US mail: 9432 Boelter Hall, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw@stat.ucla.edu
homepage: http://gifi.stat.ucla.edu
   
------------------------------------------------------------------------ 
-------------------------
           No matter where you go, there you are. --- Buckaroo Banzai
                    http://gifi.stat.ucla.edu/sounds/nomatter.au
   
------------------------------------------------------------------------ 
-------------------------

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

Thomas,
you were right. It works now. This is what I get with Carbon R and the 
patched survival package.
I will upload a patched version of survivalLib. Can you please fix 
SurvS.h sources?
Stefano


 >
survreg(Surv(c(0,1,-1),c(T,T,F),type='left')~1,dist='gaussian')
Call:
survreg(formula = Surv(c(0, 1, -1), c(T, T, F), type = "left") ~
     1, dist = "gaussian")

Coefficients:
(Intercept)
  -0.2321569

Scale= 1.161135

Loglik(model)= -4.1   Loglik(intercept only)= -4.1
n= 3

On Saturday, November 16, 2002, at 12:59 AM, Thomas Lumley wrote:
>
>  I think it's a problem with the erfc macro in survS.h, which is 
> defined
> as
> #ifndef erfc
> #define erfc(x) 2*pnorm5(-x*M_SQRT2,0,1,1,0)
> #endif
>
> The problem is that in survreg2.c we call erf(-z/ROOT_2), which 
> expands to
> 2*pnorm5(--z/ROOT_2*SQRT2,0,1,1,0), and the -- becomes a predecrement
> operator.   I hate macros.
>
> If so, the fix would be
> #ifndef erfc
> #define erfc(x) 2*pnorm5(-(x)*M_SQRT2,0,1,1,0)
> #endif
>
> 	-thomas
>
>
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._