I have computed and plotted the survival function for a subscription-based service and the following is the result.

enter image description here

The problem is that there does not seem to be enough data to get a full curve. This is because most of the oldest accounts are still active. So my question is would it still be useful to compare survival curves for different segments of accounts given that there is clearly not enough data for a full curve.


It is still useful - I wouldn't make really bold claims about what happens past 600 days if I were you, but seeing a clear departure in one category or the other, even if they don't eventually hit zero, is still useful.

Consider this: The ideal randomized clinical trial of a perfect, flawless drug will have one of the curves not only never going to zero, but never going below ~1.00. The fact that a curve doesn't drop in a meaningful time horizon is, in and of itself, useful information. And since all accounts that have not yet closed are censored, the techniques you're using are already accounting for "They'll close someday far in the future".

  • $\begingroup$ Thanks. That makes sense. I guess that due to the nature of subscription-based services the curve will get fuller as time goes on. There must be some limiting time where all accounts cancel but I guess we don't know what it is yet... $\endgroup$ Oct 15 '13 at 14:42
  • $\begingroup$ @user1893354 The day you go out of business ;) And the curve should never get "fuller" (by which I assume you mean go up) because your time variable is not calendar time, but time since the account was opened. $\endgroup$
    – Fomite
    Oct 15 '13 at 14:45
  • $\begingroup$ Sorry, by "fuller" I meant that the curve should get closer to zero (as older accounts eventually start cancelling) $\endgroup$ Oct 15 '13 at 15:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.