Censored data


No cake for spunky
This is the issue at hand.

"The censoring nature of earning information in national surveys and the administrative case reporting system should be taken into consideration in developing empirical specifications of wages. For example, in RSA-911 data, clients’ earnings after receipt of services are only discernible for those clients whose cases were successfully closed with the code “Exited with Employment Outcome.” Unfortunately, individuals who exited the VR system without any employment outcome are not followed; as a result their earnings after exit from the VR system are not observable to researchers. The main reason for “unsuccessful” case closure for this segment of the VR population as indicated by Nazarov et al. [14] is the inability of VR counselors to locate or communicate with these individuals after their receipt of VR services. Hayward & Schmidt-Davis [7] using Longitudinal Study of the Vocational Rehabilitation Services Program (LSVRSP) data, provide evidence that 37 percent of those who received services but exited the VR system without an employment outcome actually worked three years after separation."

To be clear there are cases that are successful and you know the income at closure for them. And there are cases that are not successful and you don't know the impact (the wages) of them. I have only seen the concept of censoring applied to Cox Proportion models and I am not sure how this issue impacts the results. Or how you deal with it.


Less is more. Stay pure. Stay poor.
I think literature may be more robust if you call this loss-to-follow-up, which falls under selection bias. If you can get a random sample of these individuals there can be weighting or Bayesian methods to correct for it.