Control charts, probabilities and hypothesis tests
Six Sigma – iSixSigma › Forums › Old Forums › General › Control charts, probabilities and hypothesis tests
 This topic has 12 replies, 7 voices, and was last updated 19 years, 6 months ago by Gabriel.

AuthorPosts

March 18, 2003 at 9:19 pm #31740
Charles HParticipant@CharlesH Include @CharlesH in your post and this person will
be notified via email.Gabriel, Stan, Jamie . . .
I’ve tried positng to the original thread and it’s not working, so I’m trying a new one and we’ll see if it flies . . .
I have found my copy of “Shewhart’s Charts and the Probability Approach” by Neave and Wheeler (May 1996) http://www.spcpress.com/ink_pdfs/Wheeler%20Neave.pdf. I think it does a great job of pointing out why associating probabilities and hypothesis tests with control charts not only goes against the intention of their creators, but limits their use and function in process improvement. Gabriel, I believe that once you read the paper, it will become evident why your analogy is not approapriate. If not – what can I do? :)
Charles H.0March 19, 2003 at 12:41 pm #84006Charles,
Thanks for all your comments on this topic. I have being reading the thread with interest. I decided not to add any more comments because you clearly defined the issues. I have already had this discussion with the individuals involved. You won’t change their perspective – at least not in a couple of months.
They really need to go back to read Dr. Shewhart’s book and learn the basis of the control chart – what it is and what it isn’t. The use of defined probabilities can’t be stated. Unfortunately, so many engineers read primarily engineering professor books. These books also do not understand Dr.Shewhart’s work. Perhaps, they just don’t have the basic education in the other disciplines to fully appreciate his work. But, by being adament and talking models doesn’t change the foundation work of Dr. Shewhart.
Thanks again for adding your thoughts on this topic.
Eileen, Quality Disciplines
0March 19, 2003 at 1:52 pm #84008
Charles HParticipant@CharlesH Include @CharlesH in your post and this person will
be notified via email.Eileen,
Thanks for your words of support. I know I may be pushing a boulder up a steep incline, but hey – it’s my job. Never give up, always push forward, and always try to educate. Dr. Deming’s lament that “there was no one to teach them” does not ring true anymore, and I know there are those who will never listen or learn. But, as the good Doctor might observe, if we can reach only one person, the effort is well worth it.
Charles H.0March 19, 2003 at 2:46 pm #84012Charles,
Interesting paper, but I disagree with the conclusions.
The writers argue that if one attaches (incorrectly I agree) specific probabilities to an out of control condition, they cannot or will not go after continual improvement. This is nonsensense.
They are correct about specific probabilities as we never know exactly where the mean lies, we only know the general area. This is also the reason why the whole question of sigma short term or within and sigma long term or overall is an important diagnostic.
I would contend that the quote attributed to Deming:
If it displays controlled variation then, according to Deming, “it will not be profitable to try to determine the cause of individual variations.”
is wrong. The work of Six Sigma flies in the face of this every day. The value of the control chart, or the comparison of Cp to Ppk (two sides of the same coin) is to know if the first improvement is through better control or a deeper study of the causes of variation.
Gary0March 19, 2003 at 3:20 pm #84016Charles, Dam I had a long reply typed and I hit the back key and lost the entire thing… oh well maybe thats best. So a quick summary… appearently this topic isn’t new the author states “Many statisiticians consider a control chart to simply be a sequential test of hypothesis procedure with an unbiased uncontrolled type I error rate.” Now he does warn against taking this simplistic view, but at least it seems there are others (statisticians) that take the hypothesis analogy a step further by saying it is a true test. I actually feel good I’ve got company.
I did find the discussion on assumptions of normality, mean, and std deviation to be disturbing. From a quick read (and maybe thats the problem), the author states these are never known. We only have estimates so the exact probabilities aren’t known either. So applying them is not appropriate (I might be misparaphrasing). If one accepts this to be true then this would tend to invalidate the use of inferential statistics in process analysis. Yes we violate certain assumptions to apply inferential statistics in process analysis but we knowingly do this accepting the risk, it doesn’t make it useless. But frankly I have never seen an application of statistics where it was certain that the base assumption were not violated.
At the least this has been very interesting, but the bottom line is I’m not sure how changing this small difference in intepretation will change my application of the tools so its probably best to leave things as is.
Regardless, I truly appreciate your research and posting.
Thanks, Jamie
0March 19, 2003 at 4:48 pm #84020
GabrielParticipant@Gabriel Include @Gabriel in your post and this person will
be notified via email.Jamie, once again, we agree (what boring!)
I had also typed a long reply but then I said, “no, it doesn’t add anything new”, so I intentionally pressed the “back” button.
Now, I agree with most of the article, but I don’t feel that it goes against the analogy. I goes against the use of the “hard prerequisites” of the mathematical models to limit the application of the tool. And I agree.
But if you look the objective of the chart (“to see if there is a universe”, i.e. if the process is stable) and read the mistakes 1 and 2 and understand that the risk to make that mistakes exist, even if they can not be quantified, then the analogy with the Hypothesis Test is more clear than ever before.
I disagree with the article in 2 points:
– It is not economic to try to find the causes of the common variation. I think that “REDUCE VARIATION” is appliable to any type of variation.
– You can not use probabilities becuse you don’t know the exact distribution. So let’s erase inferential statistics from any real life case. t test, F test, DOE, surveys, nothing works. Ok, the outcome will not be exact, because the inputs are not exact, but until someone invents something better… And who needs exact after all? Improving is doing better, and not going from exactly a 4.32 level sigma to a 5.54 sigma level.0March 19, 2003 at 5:17 pm #84024
Aldous WongParticipant@AldousWong Include @AldousWong in your post and this person will
be notified via email.For those who are interested,here is additional reference:
Controversies and Contradictions in Statistical Process Control (with discussions) by William Woodall, J. of Quality Technology, vol 32, no. 4, Oct 2000, p. 341 378.0March 19, 2003 at 5:32 pm #84025
Charles HParticipant@CharlesH Include @CharlesH in your post and this person will
be notified via email.Gabriel,
The risk is that by using your analogy, you put the us into the Probability Approach Box that Neave and Wheeler caution us about. And it is a complete misrepresentation of the control chart.
Charles H.0March 19, 2003 at 5:36 pm #84026
Charles HParticipant@CharlesH Include @CharlesH in your post and this person will
be notified via email.This is too funny. I had a long reply ready to send to Gary at it got lost when I went to check some documentation. Gary – I will respond after I recover and redo the post :)
Ya know – do we have a failure mode identified with the website?0March 19, 2003 at 6:29 pm #84031Amen
0March 19, 2003 at 6:59 pm #84037
Charles HParticipant@CharlesH Include @CharlesH in your post and this person will
be notified via email.Gary, Gabriel, Stan
Does anyone out there really go after individual occurences of common cause variation? Reread the quote and you will see this is what Deming was referring to. If you do go after single occurences, you are wasting your money and you are looking at the control chart through a microscope. You are guilty of “The Funnel Experiment” and you may actually induce variation, rather than limit it. I believe it more likely that once you have a process in control, you look at common cause variation in a way that allows you to see the whole and reduce it in a way that improves the process and the system – what Deming would call optimization. If not – “Off we go to the Milky Way!”0March 19, 2003 at 7:42 pm #84038Of course no one goes after individual occurances. But what is seen as common cause can be economically reduced.
0March 19, 2003 at 9:58 pm #84048
GabrielParticipant@Gabriel Include @Gabriel in your post and this person will
be notified via email.SPC is not the right tool to adress variation due to common cause. There are other tools for that.
But certainly working to reduce variation due to common causes is not wasting money.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.