Simulating log-logistc or log-normal survival data.

[Joepants]

Hello,

I am currently try to simulate some log-normal and log-logistic distributed survival times in SAS. I have done it for Weibull and Exponential and have attempted to get a log normal however comes up with errors. If anyone could offer me any insight I would greatly appreciate it.

Regards,

Joey.

[PeterClemmensen]

How have you done it for Weibull and Exponential? And how have you done it with log normal?

Show us your code and your log?

[Joepants]

/* Weibull and Exponential */

%Let Lambda = 1000 ;
%Let Shape = 1 ; /* value of 1 for exponential */
%Let Time_Censor = 99999999999999;
%Let Beta0 = 1;
%Let Beta1 = 1.2;
%Let Sn=10;

Data Sim1A
( Drop = Linear_Predict Time_Censor Beta1 Lambda Shape Beta0);
Lambda = &Lambda. ;
Shape = &Shape.;
n= 10;
Time_Censor = &Time_Censor.;
Do Sim = 1 To &Sn. ;
Do n = 1 To n ;
ID + 1 ;
x1=rand('BERN',.5);
Beta0=&Beta0.;
Beta1=&Beta1.;
Linear_Predict = Exp(-(beta0* + (Beta1 * x1))) ;
Time = Rand( "WEIBULL", Shape ,Lambda* Linear_Predict) ;
If Time_Censor < Time
Then
Do ;
Censored = 1 ;
Time = Time_Censor ;
End ;
Else Censored = 0 ;
Output ;
End ;
End ;
Run ;

[ballardw]

And the code for your attempt at lognormal or log-log?

If you get errors copy the code and error messages from the log and paste into a code box opened with the forums {i} icon to preserve formatting of any error diagnostics that appear. It is a good idea to paste code in either the code box or the box opened with the "running man"  to preserve formatting as the message windows will remove a lot of white space.

[Joepants]

I have fixed it now thank you, see below, I think I have done it correctly.

data Sim3

 ( Drop = Linear_Predict Beta1 Lambda Shape Beta0 Beta1);

  Time_Censor = 999999999999999999;

 Lambda = 1 ;

 Shape = 2;

 Do Sim = 1 To 1;

 Do n = 1 To 10000 ;

 ID + 1 ;

 x1=rand('BERN',.5);

r=RAND("uniform");

Beta0 = 1;

Beta1 = 2;

Linear_Predict = Exp(-(beta0* + (Beta1 * x1))) ;

Time=(((1/r)-1)**(1/Shape))*(Linear_Predict);

If Time_Censor < Time

 Then

 Do ;

 Censored = 1 ;

 Time = Time_Censor ;

 End ;

 Else Censored = 0 ;

output;

End ;

End ;

run;

[FreelanceReinh]

@Joepants:

Why do you write

-(beta0* + (Beta1 * x1)) ?

This is equivalent to

-(beta0 * Beta1 * x1)

i.e., a model with intercept 0, unlike the more plausible

-(beta0 + Beta1 * x1)

Also, it would seem more familiar to me to omit the minus sign, so that the estimates of beta0 and beta1 are the "Intercept" and "x1" parameters in the output of a generic analysis like 

proc lifereg data=sim3;
model time*censored(1)=x1 / dist=llogistic;
run;

But, of course, this is only a matter of parameterization.

When I perform simulations I always specify an initial seed value using the CALL STREAMINIT routine in order to get reproducible results.