Have you ever asked yourself just what control limits actually measure? Oh sure, there are the standard statistical answers: central tendency, process dispersion, capability and so on. But what do control limits measure fundamentally?

Ignorance is really the only reason we need SPC, or any statistics for that matter. Statistics help us quantify unexplained variation. If we possessed complete understanding of a process or other phenomenon, we wouldn’t need statistics at all; we’d know what the result would be.

Usually our measure of ignorance extends as far as the resolution of our measuring system. If I measured a 1-inch part cut on a precision lathe using a tape measure, the control chart would show a boring series of 1s. But if I measured the same part with an indicating micrometer, I’d get interesting patterns of variation. Interesting in the sense that I could learn something from them. Interesting because there is ignorance. Measurable, quantifiable ignorance.

At what point have we learned “enough” so that we can (or should) stop using statistics? The answer is based on two concerns. First, there is our species’ fundamental thirst for knowledge for its own sake, which prevents us from ever learning enough. Thus we have physicists pursuing ever smaller variances in their measurements. Second, for quality professionals, our thirst for knowledge often is overridden by economics. We usually stop pursuing knowledge when it is no longer economical to do so.

Control charts help us decide when we have reached that point. An operational control chart used on-line helps the process operator decide when knowledge can be obtained economically. Shewhart found, through trial and error for the most part, that it was relatively easy to identify a single important source of variation if a data point varied from the process mean by more than three sigma. This assumes that the investigation takes place quickly, rather than after the trail has cooled.

However, Deming discovered that off-line study of the control charts, often by teams, also proved economical. Off-line analysis looks at more subtle patterns   between the control limits. With both on-line and off-line analysis, we try to reduce our ignorance of the process.

If the on-line investigation reveals the special cause of the variation, then the control limits usually are not adjusted. This is because we already know what the process is capable of doing, and the special cause was, well, special. For example, a wave solder process might produce excessive defects due to a contaminated batch of solder. In such a case, we wouldn’t change the process.

On the other hand, when off-line study teaches us something new, we usually use the knowledge to improve the process permanently, e.g., by computing a new set of control limits. Using the same example, we may learn that changing the wave-solder preheat temperature results in fewer defects, and consequently we write the new knowledge into the process control plan.

When we change the control limits based on the new process, we are saying, in effect, “I have learned something new about this operation.” If the control limits go from 100 defective solder joints per million to 75, we are saying that we formerly understood enough to explain all but 100 parts-per-million defects, but can now explain all but 75. Our goal is, in a sense, the same as the physicist chasing quarks and muons: perfect understanding. If we achieve it, we know what it takes to have zero defects.

What does this mean to those of us working in the quality field? Quite simply, that quality represents the quest for new knowledge. We can only improve quality if we know more about the process under improvement than we do now. Of course, knowledge in and of itself offers little value in the working world–we need to use the knowledge to make it valuable. People who work on quality improvement must have the opportunity to act on what they learn.

Few things are more frustrating than studying a process, making an exhilarating breakthrough, then discovering that implementing the improvement requires an endless cycle of approvals, budget changes and overcoming administrative barriers. If  organizations are going to do SPC, then they should do it right: Set up the mechanism for changing the process ahead of time.