Industrial Statistics – Research vs Practiceby John Hunter
2013 American Statistical Association (ASA) Deming Lecture: Industrial Statistics – Research vs Practice by Vijay Nair, University of Michigan.
In his talk Dr. Nair referenced the article, W. Edwards Deming: The Story ofa Truly Remarkable Person, by Robert B. Austenfeld. He also referenced Dr. Deming becoming famous in the USA overnight, at the age of 80, after the broadcast of If Japan Can, Why Can’t We? in 1980.
This presentation is focused statistical process control (SPC) and some emerging statistics issues in industry. This presentation is probably of interest mainly to those very interested in more advanced statistical ideas. That isn’t a surprise for a presentation at ASA but I just want to mention this is likely of interest to a smaller sub-set of our blog readers than some of the other Deming Lectures.
At about 48 minutes of the webcast Dr. Nair discusses an interesting idea with processes failure examples (without watching this portion of the lecture this paragraph may be difficult to appreciate). Essentially there are various patterns that are indications of specific types of issue. So in his example a loose tie rod results in one pattern a warn bearing results in another pattern. His example is manufacturing but the same idea could be applied to any in-process measures.
In his example they intentionally set the conditions (for example a warn bearing) and learned how the process would respond (and what in process indications would show). That same idea could be used to determine in advance how the process would behave in various situations and thus give you a quick idea of where to look if you find a matching pattern.
Today we often rely on someone remembering the last time they saw such a pattern it was caused by x. Or we rely on someone having the insight to draw that conclusion based on their deep understanding of the process. This recall or insight is much more difficult in highly complex processes with many interrelated factors. For simple processes often recall or insight may provide acceptable results but in cases where they won’t work having an appreciation for what results point to specific causes could be very useful information to have.
Another issue is to capture such knowledge in a way that is useful. How do you standardize the process so when certain results are seen the organization responds in the right way? This is true even when recall or insight might be enough (if the right person is responding). It is likely a weakness if you rely on one, or a couple, of people to be responding in order to take effective action. For a few critical responses it often isn’t hard to make someone work but as the amount of knowledge to make available grows making sure there is a good process to keep information updated and easily accessible becomes a problem that many organizations struggle with.
Related: Process Behavior Charts are the Secret to Understanding the Organization as a System – Deming Lecture by Dr. Jeff Wu on Quality Improvement from Autos to Nanotechnology – A Historical Look at Deming’s Career: Lecture by J. Stuart Hunter – Statistical Techniques Allow Management to do a Better Job