1
$\begingroup$

I do have two data-sets and I am not shure which version of the KM estimator to use. Both datasets contain the following columns:

ID | Issue Date | Last Recorded Date | Status

Wheras the Status can either be a) Default b) Current c) Fully Payed at the End

A Loan can run for Maximum 36 month such that Status c) only applies if all payments were made as sheduled.


  1. Set: Paymenthistory of Loans for which some are still current. This mean that I do have right cencored data and could therefore use S(t(i))= (1-d(1)/n(1))...(1-d(t(i)/n(t(i)) ?

  1. Set Paymenthistory of only terminated loans. This would imply that Status b) is never observed. Therefore - if I understood it correctly - I oberve the "whole" life of all loans and therefore have no censored data. In this case I use S(t)=# obs. life after t/N, where N denotes the number of observations.

So I am not shure whether I'm on the right track and whether my suggestions are right

$\endgroup$
  • 1
    $\begingroup$ In general, the choice of statistical method is strongly influenced by your question or hypothesis. Why do you want to do this at all? Or why do you think your question might be better answered by one or the other? $\endgroup$ – David Smith May 22 '17 at 18:32
  • $\begingroup$ @DavidSmith Thanks for the quick reply! Since I am a newbie to survival Analysis, I even do not understand whethere my data in the 2. set is censored. This would be a fist question as I stated it obove. In the end I want to estimate a competing risk model and determine the probability of each Event at each Point in time. $\endgroup$ – Jogi May 22 '17 at 19:25
0
$\begingroup$

For survival analysis you should clearly define the event you are interested in. From your description it looks like you have two potential "events" for a loan. A loan could go into default, or it could be fully paid.

Censoring occurs when that loan is lost to observation.

In your first dataset, you are correct that the loans which are still current are right censored for the last time point.

In your second dataset, if you are only interested in whether the loan goes into default then the loans that were paid during the observation period would be "randomly right censored" at their last observation point.

As far as calculating S(t) I would suggest using this resource.

$\endgroup$

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.