Right censored data

[Ben Turner]

[Originally Posted: 8/16/00--Transferred by ReliaSoft Moderator]

If I have a system that has been running for an unknown period of time, before failure recording commences, is it correct to say that the first recorded failure for the system is right censored data, since I know that the system has been in operation for at least X hours?

[RS Tech Support]

[Originally Posted: 8/18/00--Transferred by ReliaSoft Moderator]

I am a bit confused at your situation: you say you don't know how long the units have been running, yet you say they have been operating X hours. Could you please clarify the situation? Thanks.

[Ben Turner]

[Originally Posted: 8/21/00--Transferred by ReliaSoft Moderator]

Example: I have a system that has been operating for some unknown time. On January 1, I start to monitor the failures of the system. On January 10, I record the first failure. Therefore, I know that the system was in operation at least 10 days but I don't know exactly how long. So, would this first recorded failure be a suspended data point?

[RS Tech Support]

[Originally Posted: 8/22/00--Transferred by ReliaSoft Moderator]

Well, technically, a suspension is a failure that occurs at an unknown period of operation time, which definitely applies in this case. However, the way in which you handle this data in an analysis is anothe matter entirely. In the analysis of a data set with standard suspensions (i.e. units that have run for a known amount of time but not failed) the information from the failed units is used to extrapolate failure times for the suspensions. Since you make no mention of other data points, it is difficult to assess. Overall, the best suggestion would be to estimate the unknown operating time of the equipment and use that to develop an estimated failure time rather than treating this as a suspension.

[Ben Turner]

[Originally Posted: 8/22/00--Transferred by ReliaSoft Moderator]

Even though I may be able to estimate the total operating time, it wouldn't really help. During the unknown period of time, before failure recording started, there were also an unknown number of failures, we assume. There are other failures after this first failure. Would it be best to just take the time for first failure of the system as the starting time? We would lose one data point then, but wouldn't need to use a suspended data point at the start. Thanks.

[Dr. Dave Olwell]

[Originally Posted: 8/31/00--Transferred by ReliaSoft Moderator]

If I started monitoring on 1 Jan of a unit that had been operatining some time previously, and it failed at 10 Jan, I would say it was right censored: its life was at least 10 days.

If I had other units that had failed before 1 Jan, but I had no data on when they started or when they failed or how they failed, I would ignore them. They tell me nothing (except that the unit can fail.)

I would not forget to consider all never failed units (i.e. the others that were still working at 10 Jan) in my analysis. They too are right censored -- they also have been working at leat 10 days. Of course on 11 Jan, they will have been working at least 11 days...

You are not going to have much information in your data initially, until units placed in service after 1 Jan fail and you have exact failure times or at least left censored failure times. Depending on your needs, that can be okay -- depends what you are trying to do.

In my opinion, you have less of a technical support question and more of a data analysis consulting question. Your goals and assumptions will really influence what you do.

Best wishes,

Dr. Dave