Change in population

[Anne Cathrine Gjærde]

How do I treat a situation with a change in population. I analyse failure data for electrical metering equipment, and sometimes a large group of them are scrapped, from different reasons, without having failed.

[Pantelis]

I presume you are doing life data analysis and that the change in population you refer to is the fact that you have non-failed equipment in the mix.

So, in building the model you have a specific time to ‘event happened data’ (failure for some reason) and time to ‘event NOT happened data’ (it did not fail based on your set of criteria). This data is then treated as censored data (right censored data) and included in the analysis.

How do you do this (or need more info)? See links that follow (as well as links included within these):
Learn more about data and censoring schemes: http://www.weibull.com/LifeDataWeb/data_classification.htm

Analysis of Right Censored (Suspended) Data
http://www.weibull.com/LifeDataWeb/right_censored_suspended_data.htm

Obviously Weibull++ can be used for the analysis. http://reliasoft.com/Weibull/

[Anne Cathrine Gjærde]

It is correct that the items haven't failed, but they are still taken out of "action" - so if I startéd with a population of lets say 1000 items, then with time some of them will fail, and I can do a Weibull plot with censored data for the ones that haven't failed yet. But what if I had 1000 items and none of them have failed after a time t, but for different reasons it was necessary to take e.g. 100 items out of the test, then the test continues, items fail - and I want to do my Weibull plot. But how do I treat the fact that the population have changed during my test time?

[Pantelis]

I am not sure I understand the difference. You state:

“…But what if I had 1000 items and none of them have failed after a time t, but for different reasons it was necessary to take e.g. 100 items out of the test, then the test continues, items fail - and I want to do my Weibull plot.”

So lets say you started a test with a 1000 units. After 50 (t=50) hours none failed but you remove 100. After another 50 hours, (t=100), none fail again and you again remove 100. Now lets also say that at t=300, 310, 320, 350 you have a failures. If at t=350 you want to build a Weibull model then your data is as follows:


100 items Suspended at t=50
100 items Suspended at t=100
1 Failed at 300
1 Failed at 310
1 Failed at 320
1 Failed at 350
796 items Suspended at t=350