The statistical process control chart is designed to help differentiate between what are usually called “common” causes of variation and “special” causes. But this is very misleading.
Typically, six-sigma consultants and various authors advise you to ignore data between control limits because it is “random.” They say to only make process adjustments when the process falls outside the control limits. In most cases, the premise behind this advice is false.
The justification given goes as follows:
Consider all possible measurements that we could get from this process over the long term (whatever that is). The control limits show how far we would expect 99% of the measurements to wander from the mean of the process if the process does not change. Ignore what you think you have learned over the years about what causes the process to wiggle within the control limits because it is just random variation (whatever that is).
This explanation is irrelevant. It gives the same weight to future measurements—that have not yet occurred—as to the current measurement. It requires you to throw away information you have learned over a long time. And it is not what you need to know. What you need to know is:
The probability that the process has changed, given the measurement you just took, and given the background information about the process.
The statistical process control and six-sigma dogma tells you instead:
The probability that future measurements will fall between the control limits if the process does not change and if you ignore your background information about the process.
These are not at all the same probabilities. They can be very different, depending on circumstances. Proponents of statistical process control treat them as if they were the same. Therein lies the misleading indicator.
More to come….