nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Control Chart postive test percent for each phase & Control Chart standards

chris_G_ttu

Hello,

I have an IMR chart that is seperated into a few phases. I ran all the Nelson test to see which data points test postive for the 8 test. Is there a way to get the percentage of data points that are in each phase that test postive for each Nelson test Or the total points that test postive in the phase. I have many control charts to do this for and would like to know a fast way to do this instead of counting them.I am not sure if this capability is bulit in to jmp 13.

for example I would like a table that has something like this

phase1: test 1, 1% test 2, 0.3%

phase 2: test 4, 3%

phase 3 test 6, 6%

Also, I have serchaing for the standards or rules to follow for control charts in order to determine what is out of contol. I have come acorss several things but nothing has been black or white as far as guidlines for control charting.I would appericate some guidnce on this topic.

Accepted answers

txnelson

You could just create one column called Percent of Total and use the following formula

Col Number(Test,Phase,Test)/ Col Number(Test, Phase)

All comments

chris_G_ttu

I figured out how to get a new jmp table the shows the test that each sample number test postive for but I am having trouble making a formula that counts the percent for each phase. I would like to do something like this using a jmp formula in the column for test 1 percents.

I am still looking for some standards to determine if a process is out of control. Is there a veriable that indicates a process is out of control? I know for capability there is cpk so is there something like this for stability?

txnelson

You could just create one column called Percent of Total and use the following formula

Col Number(Test,Phase,Test)/ Col Number(Test, Phase)

Kevin_Anderson

Hi, chris_G_ttu!

Regarding your question about standards: The Process Control methods invented by Walter A. Shewhart are so elegant that they appear simple. They are elegant and powerful and world-altering, but don't be deceived; they are not simple.

The Nelson control rules (and all variations of control rules) depend on the limits that are applied to the data. The data used to set the limits should be abundant, independent and identically distributed(Normal), and free of Special Cause variation. If appropriate limits are applied to appropriate data, you should have a baseline proportion of 0.0027 Out Of Control...that's 0.27% OOC. This implies a "chicken-and-egg" problem: the only way to assess Special Cause variation is from a control chart with correct limits, but correct limits come from data without Special Cause variation.

Applying all 8 Nelson rules on incorrect limits calculated from small datasets is almost guaranteed to be a train wreck of misapplication. Lloyd Nelson documented his tests in a 1984 Journal of Quality Technology article "...for convenience and uniformity of application...". He also states that "...analysts should be alert to any patterns that indicate the influences of special causes...", since a comprehensive list of standards for OOC would be infinitely large. There cannot be black-and-white standards for OOC.

And all these methods get their real power from being applied on-line, not after the fact (as you appear to be applying them!).

Please consult with a statistician. Good luck!

chris_G_ttu

This worked! Thank you

chris_G_ttu

Hi, Kevin_C_anders

I apperciate the response! That was some great info.

Peter_Bartell

To pile on to @Kevin_Anderson's input and here's the hard part. Try not to think of process monitoring decision making as a hypothesis test. This is what happens when we start making statements such as 'The process IS in control.' Or, 'The process IS NOT in control'. A control chart is NOT a statistical inference tool.

All a control chart can 'tell you' is 'Is there evidence of assignable cause variation?'. Then once we answer that question, we have to decide our next action. It might be 'do nothing' or 'do something'. Which leads to conversation and discussion about consquences of making Type I or Type II errors...which leads back to Shewhart's classic discussion of control charts being intended to economically balance the risks of making either one of these two mistakes. It's here that statistics provides little, if any guidance. So at the end of the day, try not to focus on the 'is it in control or not' question...but the data, your process knowledge, decisions, actions, consequences of those actions.

I try to tell every quality practitioner that one should invest the time in reading the source document from which all of this comes...Shewhart's classic book "Economic Control of Quality of Manufactured Product." Much will hopefully become clear after reading this seminal work...almost 100 years old...but still just as relevant today as it was when first published in 1931.

chris_G_ttu

Thank you for the response and guidence.The six sigma material I have read about control charts uses the is in contol or is out of control phrasing. I will defintly read up on Shewhart's book!