Determination of Subgroup Size for P Charts

[PAD01]

Hi,

Using P-Charts to monotor the number of defective units from a process. Only interested in the yield (or defect rate). How many points in a subgroup should be used when 'calculating parameters'?
I have different 'stages' when the process was changed - should the subgroup 'recommence' at the start of each stage for x points?

All should all points always be used - an just observe the last datasets.

Thanks in advance
Paul.

[Steve Prevette]

I usually find it best if there are at least 20 in each subgroup. This keeps the p chart control limits "reasonable" (doesn't work well if the LCL is less than 0%and UCL is greater than 100%).

Generally speaking I would not try to artificially align the subgroups to changes in the process - I'd let the data speak for itself as to if a shift is found.

Yes, you could do random sampling and then plot that data if it really is too hard to deal with all of the data.

[PAD01]

Cheers Steve - so when would you start to use a new subgroup, for calculating control limits, if not when a process change is made?

Paul

[Steve Prevette]

Generally in my case it is a time interval, such as weekly or monthly. Or if it is at a higher rate, lot sizes of 100 for example.

[PAD01]

Thanks again Steve