Person-time still contributed after censorship? (Stata) I feel like I might be going a little crazy here, so I'd appreciate some advice.
I have a multiple-record-per-subject dataset that goes something like this:
 id | day  | failure
----+------+--------
 1  |  1   |    .
 1  |  2   |    .
 1  |  3   |    . 
 1  |  4   |  fail
 1  |  5   |    .
 1  |  6   |    .
    |      |
 2  |  1   |    .
 2  |  2   |    .
 2  |  3   |    .   
 2  |  4   |    .
 2  |  5   |    .
 2  |  6   |  censor
 2  |  7   |    .
 2  |  8   |    .


I know that id==1 should not contribute any more time to the at-risk pool after day 4 (day of failure).
However, should id==2 contribute any more time to the at-risk pool after day 6 (day of censorship)?
I was under the impression that the answer was no, time after censorship is not considered "at-risk," just as time after failure is not considered "at-risk." However, Stata is continuing to count this post-censorship time in the person-time denominator. Perhaps my understanding of survival analysis is flawed?
UPDATE: To provide more info here's how I stset the data (where censorship:failure == 0 or failure == 2; and where failure:failure == 1):
stset day, failure(failure == 1) id(id)

 A: (Answering my own question after digging through the Stata manual and per suggestions of @DimitriyV.Masterov)

I was under the impression that the answer was no, time after censorship is not considered "at-risk," just as time after failure is not considered "at-risk."

As far as I know, this is true. Censorship means exit from the study, just as failure means exit from the study.

However, Stata is continuing to count this post-censorship time in the person-time denominator.

Incredibly, this is a default "feature" of Stata's survival analysis program stset when dealing with multiple-records-per-subject data. I dug through the Stata manual and found an example:

      patno    t    x1    x2    died
          1    4    23    11       1
          2    5    21     9       0
          2    8     7    20       5
          3    7    20     5       0
          3    9    22     5       0
          3   11    21     5       0
          4  ... 
Typing
.    stset t, id(patno) failure(died)
would treat
patno==1 as dying at t==4
patno==2 as dying at t==8
patno==3 as dying at t==11
Intervening records on the same subject are marked as "censored". Technically, they are not really censored if you think about it carefully; they are simply marked as not failing. Look at the data for subject 3:
      patno    t    x1    x2    died
          3    9    22     5       0
          3   11    21     5       0 
The subject is not censored at t = 9 because there are more data on the subject; it is merely the case that the subject did not die at that time. At t = 9, x1 changed from 22 to 21. The subject is really censored at t = 11 because the subject did not die and there are no more records on the subject.

Stata ignored the censorship coding because there were more data listed for this subject. Aside from my irritation that this is a default feature, wasn't hinted at in the help file, and is contradicted elsewhere in the manual, it describes my observations.
The fix for this involves explicitly telling Stata to exit upon failure or censorship by specifying the exit option. In my case, this would mean the following:
stset day, failure(failure == 1) id(id) exit(failure == 0, 1, 2)

