The famous quality gurus W. Edwards Deming and Walter A. Shewhart distinguished between special cause variation and common cause variation. Special cause variation is due to an specifically identifiable event. Common cause variation is due to the natural random variation in the process. Special cause variation is viewed as the signal, and common cause variation is the noise. Deming used his famous red bead experiment to illustrate the futility of treating random common cause variation as a signal.
Deming's conclusion is correct and appropriate for the simple random processes of manufacturing. In manufacturing, each "trial" is independent of the previous "trial." Under such circumstances, reacting to the variation in an individual trial will actually increase the overall variation of a process. Manufacturers have correctly learned not to adjust their process in response to each individual variation in output. Don't chase noise in manufacturing.
However, Deming's advice not to react to common cause variation is terribly wrong for product developers. In a product development process we should react to both the signal of special cause variation, and the "noise" of common cause variation. Why? Common cause noise has a cumulative effect in product development; in a certain sense our "trials" are not truly independent. For example, let's say I assign independent five tasks to an engineer. The completion time of each task will affect the start time of the next one. Even when the work content of the tasks is totally uncorrelated, the completion times of sequential tasks will be correlated.
In a mathematical sense, in product development we deal with the sums of random variables, not individual random variables. These cumulative sums behave differently than simple random variables. The branch of applied statistics that examines this is called random processes. One illuminating random process is a Markov process, where the outcome of a trial is determined by BOTH the current trial AND the state of the system resulting from the previous trial. (This should sound a lot like product development.) A classic Markov process is the drunkard's walk, where a drunk staggers in random directions from a lamppost. Each step is completely random and unbiased, yet the drunkard's position will diffuse further and further from the lamppost. Although the starting point always remains the most probable place to find the drunk, the actual probability of finding the drunk at the lamppost gets increasing small over time. (Expected distance from the lamppost actually increases with the square root of the time period, a recurring theme in diffusion processes.)
Because product development deals with such random processes, we must react the accumulation of random common cause variation. By reacting, we maintain our process at an acceptable operating point. For example, let's say we assign an engineer ten uncorrelated tasks to complete by a deadline. The actual duration of these tasks varies randomly. Yet, if four of the first five tasks taken longer than expected, we must intervene. We should not hope that the remaining five tasks will fortuitously take less time just because the first five took extra time; their durations are independent. Although the individual durations of the first five tasks will not predict the durations of the next five tasks, the cumulative duration of the first five tasks will affect the completion time of the next five tasks.
Reacting to the accumulation of random variance is especially important in product development because economic damage can rise steeply when we deviate from our desired control range. Think of it as a drunk staggering around a lamppost placed on the flat roof of a 100 story building. If we let the randomness of the drunk's walk accumulate, then he will fall of the roof. If we intervene whenever he is halfway to the edge of the roof we can prevent this tragic confluence of alcohol and stupidity. We may not be able to control the random direction of the drunk's next step, but we can prevent him from falling off the roof. This is how we should operate product development processes.
Deming and Shewhart have provided us with great ideas, however they did not give us permission to stop thinking when we use them.
Thrifty shoppers can find FLOW for $28.76 at: Barnes & Noble
Prepublication Reviews: Book Reviews
Overseas readers can order the new book from: Celeritas Publishing Order Page
UK Readers can order the new book from Dave Donelan (+44 (0)1926 336423) at Smallpeice Enterprises
For bulk orders contact: Celeritas Publishing: Special Sales
[Return to Top of Page]
Public courses are shown below. Please contact us directly
for information about in-company courses. We have done such courses
in the United States and 18 other countries.
![]()