One Quality Opinion to Another

Wow… just wow

Anthony Burns’ opinion piece, “Six Sigma Psychology” published in the Oct. 25, 2012, edition of Quality Digest Daily, piqued my interest. I read it in anticipation of discovering a candy house waiting at the end of a long dark path, along with suggestions about how to avoid the oven. Perhaps I’d find a scroll illuminating the psychology of change and the struggles change agents endure every day. Maybe there would be a magic book of remedies describing ways to tame wild things and energize the bashful—What I found was not what I expected.

First, I do agree that psychology is very important to any choice people make. Unfortunately, that is where my agreement stops. The first choices Burns proposes are that quality is related to specification limits, or that it is inherent to the way a product is made. I believe the choice is obvious: Both are incorrect. Product quality is inherent in the design of the product. This lesson was taught to me in engineering school and reinforced in Six Sigma training and lean training, business school, and during a few decades of experience.

…

Want to continue?

By logging in you agree to receive communication from Quality Digest. Privacy Policy.

Create a FREE account

Forgot My Password

Comments

Always Glad to See Taguchi

Interesting counterpoint to Tony Burns' article. I'm always happy to see Taguchi brought into the conversation. One thing that Taguchi shows us is that we need to get the process on target first, then constantly and forever reduce variation.

My only point of contention is the characterization of Average Run Length as synonymous with "drift" (and as a "consultantese" term). ARL is not "the amount of deviation a process can deviate before it is identified." That might be a good working definition of "drift." Average run length, though, is a formal term used by statisticians to describe the number of subgroups between false signals. It is a basis for comparison for different control charting techniques. ARL is no more "consultantese" than Standard Deviation. To quote Don Wheeler's excellent JQT article on the subject:

"Average run length (ARL) values were intended,and are appropriately used, to compare different techniques under the same conditions. Given the same probability model, and the same conditions, and the same signals, we can use the ARL values for two different techniques to make a judgment about which technique is more likely to detect a signal or give a false alarm. If two techniques have theoretical ARL values that differ by a large amount, then the two techniques are likely to work differently in practice. Small differences in ARL values are unlikely to be noticeable in practice."

Reference:

Wheeler, D. (2000). Discussion. Journal of Quality Technology 32(4), 361-363.

Clarification of "drift" and "ARL"

Mr. Stauffer,

Thank you for your observation. I did not intend for "drift" to be synonymous with "ARL". My intention was that the "drift" is the amount a process can change over the interval defined by the ARL. The ARL would be established by the rules used to detect a signal (real or otherwise). Thus, the rules a company chooses to detect signals will be directly related to the amount of drift the company is willing to tolerate. The amount of drift the company will tolerate is then part of the defects the company will tolerate. Tying back to the original piece, that Motorola used 1.5 standard deviations as their "allowable drift", my contention is that is a business decision and the justification for that number is more business driven (thus the link to Taguchi) than conceived from a statistical methodology.

I will concede that calling ARL "consultantese" might be overly critical. To tie back to the original piece, the claim was Six Sigma fundamentals are badly flawed. My contention is the fundamentals are poorly understood and poorly disseminated by virtual armies of consultants that create acronyms and terms to distinguish themselves from the rest of the pack. Since, the mathematics to calculate the ARL is the reciprocal of the probability to detect a signal, e.g. 25% probability of detection = ARL of 4 and the probabilities are just as easily compared as the reciprocal. My belief is giving a name and acronym to the reciprocal creates the opportunity for misunderstanding the true intent. I concede that choosing ARL to try to establish this point was not the best choice, but was very relevant as the original article maligned the concept of detection of process shift.

Hopefully this clarifies my original intent.