How do I force the largest observe time to be always failure

[desireatem]

Hello, in the following data generation; I want to change the censored value for the highest observed time to be always 0 '. That is after sorting by time from minumum time to maximum, I want censored=0 for the largest time

data test;

seed=-1;

  alpha1 = 2.50;

  beta1 = 9.50;

  do i = 1 to 100;

   lambdaT = 0.025; *baseline hazard;

   lambdaC= .03;  *light=0.15; *heavy=0.03;

   er=0+sqrt(0.0001)*rannor(1);

    t = rand("WEIBULL", 0.75, lambdaT);  * time of event;

    c = rand("WEIBULL", 1.25, lambdaC) ;* time of censoring;

    time = min(t, c);    * which came first?;

    censored = (c lt t);

      obs=(t lt c); * creating observation variable from censored when

      y= alpha1 + beta1*t + er;

    output;

  end;

run;

[esjackso]

Is this what you need?

Proc sort data=test out=test2; by descending time; run;

data test3;

     set test2;

     by descending time;

     if first.time then censored = 0;

run;

[Astounding · Accepted solution]

I think the intent is simpler than that.  The way I read it, the IF/THEN statement should be:

if _n_=1 then censored=0;

All of the other work is necessary ... sorting, then reading the data back into a DATA step.  There is no easy way to make all of this happen in a single DATA step.

[esjackso]

Yeah in this case this two methods are the same.

EJ

[PGStats]

You can do it this way:

proc sql;

create table testMax as select max(t) as maxt from test;

update test

set censored = 0

where t = (select maxt from testMax);

drop table testMax;

quit;

PG

[esjackso]

I like this method (one correction I believe it not just t but min of t or c that makes up time) .. .you could probably select the obs of the max time into a macro variable and use that in the update where and avoid create the temp table all together.

EJ

[Astounding]

PG,

I suspect you're fine with that ... Not knowing anything about a Weibull distribution, I wasn't sure if a tie would be possible.  I didn't want to allow the possibility of uncensoring two observations.  That's probably impossible, given the use of RAND, but I was too lazy to look it up when I had a working approach!

[PGStats]

Well, OP's request wasn't that clear to me. The Weibull distribution is continuous, so that ties are very unlikely.  I'm waiting for some feedback from OP.

[Reeza]

You probably don't need to be generating a censoring time then do you? About half the lines in your original code could be deleted.