Everything varies. We know it happens, and if you can’t see it, the variation may not be that significant to your process. However, it may be that your measurement systems are incapable of detecting significant variation that is important to your process, more about that in another post. Variation leads to production problems, waste and ultimately quality and delivery problems. Control the variation, you control the waste and costs. If waste and costs are a problem in your business, you may be interested in reading on.
There are two types of variation, common cause and special cause. Common cause variation is natural, characteristic of the process and most importantly, predictable. Special cause variation is caused by external factors acting on the process and is not predictable. This is an important distinction because the methodologies for investigating special and common cause variation are different, and if you investigate the wrong sort of variation it can waste a huge amount of time and cause frustration.
Take the process shown above. Just creating a graph of the data isn’t really useful, since it is unclear what should be investigated, or how to proceed. Typically a manager will look at a trend line to see if the process data is trending up or down. If the process is in control and (often) a manager observes an undesirable deviation from target, it is common to ask for that to be investigated. If the investigation focuses on special cause variation which is likely, since the investigator is likely to assume something is “wrong” therefore there must be a root cause. In businesses that do not use process control charts, there is no objective assessment of process performance before launching into seeking the root cause. The problem this creates is that there may not be a root cause. If common cause variation is at work, it is a fruitless exercise.
Where a root cause analysis finds nothing, managers can assume that the investigation is flawed and demand more work to identify the root cause. At this point willing workers are perplexed, nothing they look at can explain what they have seen. Eventually, the pressure leads to the willing worker picking the most likely “cause” and ascribing the failure to this cause. Success! The manager is happy and “corrective action” is taken. The problem is that system tampering will increase the variability in the system, making failures more likely.
The danger is then clear, if we investigate common cause variation using special cause techniques, we can increase variation through system tampering.
What then of the reverse, chasing common cause corrections for special cause variation. The basic performance of the process is unlikely to change, and every time there is a perceived “breakthrough” in performance, as soon as the special cause happens again the process exhibits more variation. The process does not see an increase in variation however, neither is there any improvement in the variation.
The only way to determine if the process is in control, or if a significant process change has occurred is to look at the data in a control chart. Using a control chart we can see which variation should be investigated as a special cause, and where we should seek variation reduction. In this example, the only result that should be investigated is result 8. This is a special cause and will have a specific reason. Eliminate the root cause of that and the process is in normal control. Everything else appears to be in control. Analysing the process data in this way leads to a focused investigation. If after removal of the special cause the process limits are inconsistent with the customer specification, variation reduction efforts should focus on common cause variation.