Why is stata generating _d = . when my failures are all recorded

[Martin Imelda Borg]

Hi, I'm plotting my survival curves.
For My time variable (survivalt) - I don't have missing values) - this is adjcanet to timetodeath variable
For my failure variable (revised) - I don't have missing values

Yet, stata generates a few missing variables _d = . and _t = missing.

Why is stata coding them as missing when actually they are not missing.

And perhaps this may explain why I can not get my 25% 50% 75% survival times? Unless someone else can offer another explanation for this

stsum, by(tkronly)

Please note - I know I've submitted screenshots which goes against statalist <3 but I needed you to see the data rather than sample data to check if I'm doing anything glaringly wrong, because I can't see a problem.

[Maarten Buis]

My eyes are not as good as they used to be (and they were never very good to begin with). So those screenshots are completely useless to me. Also what is shown on screen is not necessarily the same as what is in your data. So we really really do need sample data and not screenshots. Also we really need the exact command you use to stset the data. That is probably where your problem originates.

[Martin Imelda Borg]

That's a pity. As I have said I am unable to provide a real copy of the data as it is stored on a remote platform which has no internet access. I normally play around with sample data and use this as examples. But in this case, I can not as I'm quite sure I'm using the right code. I thought someone on statalist would know why missing variables are generated following stsum

stset survivalt, failure(revised==1)

[Leonardo Guizzetti]

From the screenshot you show, I can just make out what the problem is (or one problem), assuming you have used the -stset- statement in #3.

If survival time is 0, these observations are necessarily excluded from the analysis population because they have not survived any amount of time, by definition.

Also, as you have pointed out in another related thread, you know that you can post fake data that replicates some feature of your data. Please keep that in mind when asking for help. We don’t need your real data, just realistic or toy data from which to work.

[Clyde Schechter]

To add to what Leonardo Guizzetti has said, I'll respond to your other question:

And perhaps this may explain why I can not get my 25% 50% 75% survival times? Unless someone else can offer another explanation for this

The reason that the times for 50% and 75% failiure (and, in the case of TKR, 25%, too) are missing is that your rate of failure is so low that the times to these proportions of failure are longer than anything observed in your data. In other words, even at the end the longest observation period in your study, you still have not had 25% of the people/firms/whatever they are fail in the TKR group. In the THR group, you do achieve 25% of the units failing by t = 16.22466, but the failure percentage never does get up to 50% even by the end of your data.

[Martin Imelda Borg]

To add to what Leonardo Guizzetti has said, I'll respond to your other question:

The reason that the times for 50% and 75% failiure (and, in the case of TKR, 25%, too) are missing is that your rate of failure is so low that the times to these proportions of failure are longer than anything observed in your data. In other words, even at the end the longest observation period in your study, you still have not had 25% of the people/firms/whatever they are fail in the TKR group. In the THR group, you do achieve 25% of the units failing by t = 16.22466, but the failure percentage never does get up to 50% even by the end of your data.

This is really interesting. Thanks for this perspective. Do I just accept that the procedure is so successful there is no 50% failure rate?
How does one normally address this?

[Leonardo Guizzetti]

This is really interesting. Thanks for this perspective. Do I just accept that the procedure is so successful there is no 50% failure rate?
How does one normally address this?

Pretty much, yes. You could wait longer to accumulate more failures, but that’s not usually feasible, especially with such a low incidence rate of failure.