RE: Issue when producing process behaviour chart (SPC)

["Mark Bailey" <Mark.Bailey@sas.com>]

Stephen,

What you are doing makes sense. Some thoughts in a hurry (my apology):

- Plotting in order of request date seems most informative.

- You should try to get more data to establish either the control limits or that the process is not in control.

- Your problem is known as 'censored' data. In this case, 'left' censoring. The automatic extraction of the request and completion dates and the computation of the elapsed time may not differentiate between projects that have actually completed and those that are still open. For example, some systems put today's date in the completion field if it is missing when the extraction is done or the computation inserts a zero for elapsed when the completion date is missing. This result is wrong either way. In your case, it is the request date that is suspect. What did they provide when the date preceded March 31?

Also, there may be self-censoring that occurs because if you look at requests that started yesterday, they do not have a chance to complete in more than a day. (I am not stating this very well in my haste.)

Another helpful analysis for this problem is survival/reliability/event time history analysis. The event is completion of the request. The data is the elapsed time. Censored data is expected so the completion date is always entered, even if it wasn't actually done and another field indicates if the result represents completion or is censored.

The procedures for this kind of analysis are not interested in the sequence of events, so you still want to compliment this analysis with control charts. Use the charts first because the survival analysis is invalid if the process is not in control.

-Mark 

Please copy Stephen on responses at stephen.sorrell@btinternet.com as he is not a subscriber.