I am dealing with survey data from firms to conduct survival analysis. I am going to estimate with Kaplan - Meier and a Cox Regression. I face rigth censored data as usual but I have to deal with the different starting operation year of each firm.

The survey covers 2006 - 2015.

Some firms start operations on 2006 and survive until end or close before they get censored. But then some others starts on 2007, 2008 or even 2014.

An easy way to deal with would be to just consider firms that started on 2006 and followed them until 2015. But then I would lose 80% of data.

The literature I says it shouldnt be a problem if I take in consideration the less exposed time to risk. As happens with medical survival analysis when you have patients information. I wonder if I should also control in my model the effect of the year they started operations. Or that wouldn´t be rigth at all.

Thanks in advance!


1 Answer 1


It doesn't seem to me that you have a problem. As it looks from your description, you have different companies starting to operate at different times. In such a case, you observe their 'birth' and follow them until their 'death' or until they leave your data (right censored). So unless there is a reason to believe that the time of their 'birth' affects the event, or that you have an immortal time bias (which I do not know by the data) - use a regular cox model and you do not need to worry.

If you have time varying covariates, you can state all as starting in time 0, without regard to the actual entry year. I think you can add start year covariates to see if starting time has an independent effect on event hazard ratio.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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