When a recipe calls for adding salt and pepper "to taste," it's crying out for sampling. Even without prior
training in sampling methodology, every cook is empowered to respond to this direction enthusiastically. Intuitively, the cook also knows that only a small sample is necessary to perform the
taste test, although repeating the sampling process is called for each time more seasoning is added. As Hartford Simsack learned last month, it wouldn't be productive to take repeated samples
when the process hasn't changed. Ten tastes will not yield a better sense of the seasoning than one. In fact, it may actually desensitize the pallet and render further sampling inaccurate.
Additionally, too much sampling may significantly diminish the output, perhaps to the point of rendering the yield unworthy of the investment.
What good cooks understand
intuitively has required research on the part of Simsack and Franklin Benjamin at Greer Grate & Gate. Last month, Simsack (with his ghost source, his professor Dr. Stan Deviation) helped
Benjamin wade through the issue of sampling. Benjamin, you may recall, was enthusiastically endorsing more frequent sampling, since he was certain that this would ensure higher product quality.
He learned, however, that increasing sampling frequency is not a change to be implemented capriciously, with a "the more the merrier" approach, but rather depends on two factors: how often a
process is likely to change and how much the sampling process itself will cost.
Somewhat chastened by this setback, Benjamin has been pursuing other ways to draw attention to
the process and measure its output. "Aha," he says to himself as he emerges from a long shower. "If we can't sample more often, we can at least select a bigger sample!" He feels certain that
selecting a dozen samples from each of the three shifts will guarantee an understanding of the process. From the sidelines, we can see that he intends to use a soup bowl rather than a teaspoon
for his sampling.
When he offers this suggestion to Quality Manager Simsack, he finds his boss slightly more wary than he had been last month. Deviation had actually laughed
at the suggestion that more frequent sampling alone might yield better understanding of the process. This time, Simsack ponders Benjamin's well-meant suggestion and then says: "Let me do some
calculations, Frank. I'll get back to you on this as soon as I can."
Of course, "get back to you" means that the ill-prepared quality manager intends to consult his textbook
or call Deviation immediately in order to maintain his facade of expertise in this area. He takes the latter option, mentioning to his mentor that the company would like to consider larger sample
sizes in gathering data for process improvement.
"I thought we had this all figured out," Deviation replies, with just a hint of irritation. The professor had already
painstakingly gone through the rules for sampling frequency with Simsack but had neglected to emphasize that, as is the case with the frequency of sampling, the size of the sample is subject to
rules. But Simsack is not an intuitive cook; he has to rely on exact recipes.
What will Deviation say about taking larger samples in order to support analysis of the process
and improve its quality? Will the fact that Simsack has never in his life made a pot of soup or a batch of chocolate chip cookies be a liability to his understanding the process?
Although he is unwilling to share his family's recipe for carrot soup, Deviation will spell out exact instructions for sampling size in process improvement. It must be
remembered that determining sampling size is a complex process. Among the considerations are questions such as:
What are critical factors that one needs to learn? For example, these factors
might be related to weight or volume.
Are there differences among different shifts? Months? One would want to group each shift's data about a critical factor (for example, weight), and view the data for
each shift. For months, data for each month is aggregated and then compared.
What type of chart has been selected to analyze the data? If a p chart is required,
sample sizes of 25, 50 or 100 are usually specified, while for X-bar and R charts, samples of four or five are suggested for shop charts or engineering studies of an
ongoing process. For np charts, the sample size must be constant.
The homogeneity of the sample can be as critical as the sample size.
There are far too many factors to make this a simple process. A respected
statistics textbook, such as Donald J. Wheeler's Advanced Topics in Statistical Process Control (SPC Press, 1995), must be consulted before scooping up an arbitrary number of products to sample.
Hartford Simsack has met his Waterloo, because he can't list even a few of the complex questions involved. While he had relished an opportunity to be right once
again and to send Benjamin back to the shower to rethink his ideas, he has taken a bath on this issue.
About the author
Michael J. Cleary, Ph.D., is founder and president of PQ Systems Inc. He has published articles on quality management and statistical process control in a variety
of academic and professional journals. E-mail Cleary at firstname.lastname@example.org.