In this video, we'll look at common and special cause variation. Our learning objective is to differentiate between common cause and special cause variation. As we saw in the previous video, all processes are going to exhibit inherent variation. This variation is created by the sum of many small sources of variability, and these sources of variability produce an observable distribution in the characteristic studied during a short time period. Let's assume that there are 150 consecutive items taken from a process, and assume that these 150 items may be modeled with a normal distribution with an estimated mean and standard deviation. Why aren't these 150 consecutive items all the same? Well, the reason is variation. You can't really explain exactly why one item might be different from another, and the variability often can't even be measured with the measurement systems we have. Although the items are not exactly alike, they do come from a common process and it produces a distribution. The sources of variation that produce this distribution are referred to, in statistical process control, as common causes of variability. Over time, changes in some variables can alter the distribution. These changes may appear unpredictably as changes in the shape of the distribution, spread, or location of the distribution. The causes which move or alter the common cause distribution are known as special causes of variability. When only common causes of variation are present, the process or processes are said to be in a state of statistical control. Processes that are affected by both common and special causes of variation are said to be out of statistical control. Walter Shewhart, the inventor of statistical process control, wrote that, "A phenomenon will be said to be controlled when, through the use of past experience, we can predict at least within limits, how that phenomenon may be expected to vary in the future." Process output affected only by common causes of variation is considered to be both stable and predictable. Process output affected by common and special causes of variation is considered unstable and therefore, by definition, unpredictable. For example, suppose we have a process and we're taking samples from that process through time. Well, suppose for the sake of the example that the sample size is equal to 15. I'll take my first sample and I notice that I have a distribution, and the distribution appears to be symmetric with the center at Mu and a standard deviation. If the process is predictable through time, I would expect that the next sample would come from a distribution that is similar with respect to location and with respect to the standard deviation and the shape. Now, that location could be over here somewhere, or perhaps over here. However, the underlying distribution from which that sample is from is the same as the previous distribution. If I were to take subsequent samples, if the process is predictable and stable and only affected by common cause variation, I would expect that underlying distribution to be the same through time. Let's now consider a process affected by special causes of variation. I may take an initial sample and have a distribution that is affected only by common cause variation. However, if I were to take another sample that was affected by both common and special cause variation, I might see something different. In this case, we see that the spread is larger, indicating that it's not the same distribution that these samples come from. I may see the location shifted. Again, I may see bigger dispersion or larger spread, or even a change in the shape. That would indicate that there is not only common cause variation, but special cause variation causing the underlying distribution to change in some way and also indicating that the process is unstable. Now, special causes can actually be good or bad. If we have a special cause of variability, we must implement some countermeasures to guard against something that's going to move the distribution in an unfavorable direction. However, in some cases, we may have a beneficial special cost such as the reduction of variation. If that happens to be the case, we should incorporate that into the process. In fact, make it common to the process. The function of a control plan is to implement these countermeasures. Now, the good news is that if you observe a special cause of variation which has a negative effect on the process, and you fail to implement an effective countermeasure, you usually get a second chance to do so. The bad news is, is that if you observe a special cause of variation which has a negative effect on the process and failed to implement an effective countermeasure, you usually get a second chance. In fact, Dr. W. Edwards Deming said that, "Special causes of variation left alone and unattended, become part of the process. " Which means if we detect a special cause, it is important to do something about it.
