Quality Consulting and Training

Quality improvement assistance

 

Variation Question

I studied statistics in college briefly. We recently hired a new quality manager and he is really pushing statistical quality control. I noticed you list this as one of your areas of specialization and have a question. We have started using control charts extensively but I don't agree with the new quality manager's views on statistics and how to use them in the workplace. For example, we all know that variation will always exist but what I call outliers that should be discarded from the data, he calls cause for immediate action!

I also feel that statistics serve us best as a way to predict and understand our processes. That's the whole point isn't it? To understand the capability of our processes? The quality manager says that stats are one of our most effective tools for improvement and that these outliers are possibly the most valuable info we have. What is our stance on this?

Anonymous Submission August, 2006 : Posted with consent


Common Cause and Special Cause Variation

I'm going to have to side with your new Quality Manager. Many, even in the field of statistics, are too hasty to just drop outliers. Myself I see them as points outside of our control limits when using control charts. These points that fall outside the Upper Control Limits (UCL) and Lower Control Limits (LCL) are assignable or special cause variation that must be investigated.

Anything within the control limits is common cause variation and must be left alone! Common cause variation is inherent in the process. Whereas special cause variation indicates that something happened that, for an instant, through the process out of control. Control charts are very valuable. In fact, they are great tools for both understanding the capability of your processes as you so wisely stated AND to help us identify problems. Whether a point is barely outside the control limits or way out of bounds, they should still be investigated.

Focusing on these out of control points enables us to reduce variation which is key to managing quality. I'm reminded of a project in which I was working side by side with an engineer on a quality improvement project. After gathering 50 observations/measurements we calculated control charts. However the engineer automatically erased 4 out of 50 points, stating that they were "outliers". If I had agreed to this the control charts would not have been accurate. They were not outliers. They were special cause variation and did help us identify problems that needed immediate attention.

I honestly cringe a little when I hear the word outliers. To me, it seems a way to toss aside data and make the processes appear more "in control" than they actually are. Use all data you have collected for constructing control charts. Focus strictly on all points that fall outside of the UCL and LCL. Eliminate the root cause of those out of control points and you will be reducing variation in the entire process. By doing so, you will have less variation, a more predictable process, and improved quality.

There are even gurus that share your opinion. Then again there are equally as many that share the opinion of your Quality Manager; myself included. This data can be very valuable and discarding data as outliers can actually allow problems to continue and root causes to never be identified. All in all, it is worth the effort to keep the data and work through the process of eliminating all causes of out of control points. Another reason for this is that statistics are based on eliminating assumptions. Are we not making assumptions when we discard an outlier? Assuming it is an outlier rather than assignable variation; assuming we know the cause of that data point? I "rarely" discard anything as an outlier in my SPC and SQC efforts, and our track record is excellent. I would recommend giving your Quality Manager your support. It seems that his efforts are in the right place. There are formulas that statisticians use for identifying and eliminating outliers, but I always feel more comfortable using the "pure data" rather than toying with it in order to make it look more the way we like it. After all, what can be more accurate in research of any kind than pure, unaltered data?

Great question though. Thank you!

Feel free to contact us if you have further questions.

Thank you for your interest in Lean Consulting and Training LLC.

Brian Lean CEM of Lean Consulting and Training

 
   
[Home] [Back To Top]

 Organizational Development and Process Improvement Specialists
~
Consulting and Training 2022 ~

Consulting and Training Services | Search Our Site