Control Charts

Emine

Starting to get Involved
Hello everyone,

We are ISO 17025 accredited testing laboratory. We create control charts and monitor them regularly. I have just realized that the size of the points are very variable in different test methods (some labs calculating 60 points, other 25 points etc). Are there an optimum number of points in creating control charts that will give a correct statistical analysis?

Thank you very much, any thoughts are very much appreciated,
Emine
 

Miner

Forum Moderator
Leader
Admin
When you say the "size of the points", do you mean the subgroup size or the number of subgroups used to calculate control limits?
 

Emine

Starting to get Involved
When you say the "size of the points", do you mean the subgroup size or the number of subgroups used to calculate control limits?
Hi Miner,
Thank you for your response, the points representing averages of measurements of QC s (CRM) that are run together with the samples. And as you have mentioned we calculate the control limits from standard deviation. We are a bit confused on the optimum points should be used to calculate warning/action limits, I realized that sometimes when there are a lot of QC measurements are included the control limits are wider...

Many Thanks,
Emine
 

Emine

Starting to get Involved
And what are you plotting?

Hi BD,
Thank you for your reply,
We are plotting our QC measurements from day to day analysis in a time sequence. And we calculate average and action and warning limits from the standard deviation.

Emine
 

Miner

Forum Moderator
Leader
Admin
Hi Miner,
Thank you for your response, the points representing averages of measurements of QC s (CRM) that are run together with the samples. And as you have mentioned we calculate the control limits from standard deviation. We are a bit confused on the optimum points should be used to calculate warning/action limits, I realized that sometimes when there are a lot of QC measurements are included the control limits are wider...

Many Thanks,
Emine
I am interpreting this to mean that it is your subgroup size that varies from 25-60. This is probably too large and is making your chart overly sensitive to small process changes. The subgroup size balances the risks of type 1 and type 2 errors. Too small a subgroup size and you risk missing a process change, and too large a subgroup size and you see changes that are too small to be practical. Most subgroup sizes should fall between 1 and 9.

The linked article contains a formula for calculating an appropriate size, but you can also determine an appropriate size experimentally. Start with a midrange size such as 5. Is it responding to the size of process shift that you need? If not, increase it to 6 and try again. Likewise, if it is too sensitive, reduce it to 4 and try again.

Your control limits should get tighter with larger subgroup sizes. Your experience makes me suspect that you do not have rational subgroups (i.e., you may have mixtures, or or other issues).
 
Last edited:

Bev D

Heretical Statistician
Leader
Super Moderator
If you are plotting QC results then your subgroup size should be 1. Two articles that cover SPC for QC results are: Link to article 1 and link to article 2

if your situation is not what I think it is please elaborate.
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
I usually worked with keeping a minimum of 25 points displayed on any control chart, which was based upon Dr. Shewhart's original declaration of do not declare a process stable without 25 stable points. The rule of thumb always worked for me. Visually, I usually ended up keeping the past two years plus the current year on the chart of monthly data.
 
Top Bottom