Technometrics, November 1994

Pyzdek's Guide to SPC: Volume Two, Applications and Special Topics, by Thomas PYZDEK, Milwaukee: ASQC Quality Press, 1992, iv + 237 pp.. $34.95.

Statistical process control (SPC) is a practice mixing art, science, and the human touch, Knowing how to calculate control limits is important, as is understanding the processes under investigation and the limitations of what you know, In Pyzdek's Guide to SPC, we see one man's work amid such themes. Pyzdek's Guide has two volumes-Volume 1, on SPC fundamentals, for a broad audience; and Volume 2. applications and special topics. For Volume 2, the target audience is SPC coordinators, supervisors, and engineers involved in SPC applications.

This review focuses on Volume 2, with its 12 chapters ranging from SPC implementation, measurement error and capability analysis, various special (mostly variables) control charts, SPC in service industries, and process planning.

SPC has a host of small, controversial issues. You learn something about Pyzdek's book by hearing some of his opinions:

1. SPC is approached bottom-up, with teams working to reduce variation, emphasizing simplicity. Other points of view top-down deployment and planning, out-of-control action planning, designing experiments-are either absent or short shrifted.

2. Sometimes different machines are assigned the same centerlines and common control limits. Pyzdek's favorite easy-to-use chart is the medians chart. For the individuals chart, moving ranges are plotted too.

3. Standard deviations are estimated by R/d2. The within rational-subgroups variation is used to estimate process-capability indices such as Cp and Cpk, which are to be calculated only for stable processes.

4. Variance components are presented only for measurement error.

5. Attribute charts are de-emphasized, without observing low ppm issues (unlike Fasser and Brettner 1992).

6. No transformations, powers, or logarithms are taught. He does, however, supply probability tables for skewness = ±1.

Of course, no SPC book should be recommended solely for an author's positions. (There is after all the important issue of how he or she presents them.) Still, in using a book as a textbook, one should be comfortable with an author's basic philosophy.

Which brings us to the book itself, and its general format. A sequence of special topics, each chapter is designed to be self-contained, assuming knowledge of Volume 1. Ordering reflects relative importance. Each chapter begins with hollered objectives and concludes with its own references. None has student exercises but most have worked-out examples, dominated by discrete machining and drilling processes. Most tables are in the Appendix. The typeface is clean, but paragraphs begin without any indents or extra vertical space.

On reading closely, I find myself classifying chapters into strong, weak, and mixed ones. In the strong chapters-which include those on process capability, median control charts, cumulative sum (CUSUM) charts, and SPC service applications-Pyzdek writes at his best. His prose is cleanly written and to the point, and his section headings express a logical framework.

For making a chapter strong, the critical factor is Pyzdek's careful step-by-step reduction of the methods for those not fully comfortable with statistical methods. Some of these step-by-step guides are from Pyzdek himself, his 10-step process for process capability analysis (Pyzdek 1985), for example. (I would recommend including probability paper to make this chapter more self. contained.) Others are patterned on the templates of others; the chapter on median charts follows Griffen (1986), that on CUSUM charts, Ewan (1963). I take special delight in his description of how to administer customer surveys, based on Dillman (1983). This includes guidelines for how to fold the questionnaire and how to deal with nonresponse, which climaxes with a third reminder by certified mail!

Among the weaker chapters, I number those on SPC with small runs, continuous and batch process, multiple streams, automated manufacturing, and process-control planning. On continuous and batch processes, except for a brief worksheet for the time-lagged scatterplots, no numerical examples are cited, and I miss explicit formulas for variance components, introduced earlier.

On SPC for automated manufacturing, Pyzdek introduces exponentially weighted moving average (EWMA) charts. He reminds us, but assumes that we know, of Deming's funnel experiment. Yet to embrace EWMA techniques, we must understand the limitations of the funnel experiment, which Pyzdek fails to detail, albeit autocorrelated scatterplots were earlier introduced.

Pyzdek says that the EWMA chart can be used when processes have inherent drift and are adjustable. I would assert that EWMA charts are only appropriate for drifting, adjustable processes and require data from a period for which we have not adjusted. Missing also is Hunter's (1986) advice, that the individual points be plotted with the EWMA.

The chapter on process control planning has four topics-

(1) quality function deployment, (2) process control planning, (3) contingency planning, and (4) process improvement planning.

The quality function deployment section is brief and narrowly focused, the example apt. The other topics, however, receive cursory treatment. The section on contingency planning is half a page, yet this topic, determining what should be done when a process is out of control, is a major focus of many SPC programs with which I am familiar. I have concluded that this chapter is misplaced and should be in that on SPC implementation (and displace the over detailed digressions on control charts).

Which brings us to the chapters I have ranked in the middle, on SPC implementation, measurement error, and linearly drifting processes. The chapter on SPC implementation opens with Pyzdek's would-you-hire-this-consultant presentation, which mostly works. After a section on monthly reviews, however, we take a turn, a multiple-page digression on control charts and rational subgroups, and we never quite return. Management, manifestly the target audience of this chapter, ultimately receives few guidelines for sustaining SPC programs.

The chapter on measurement error follows a similar pattern. Key definitions are laid out, but then we digress to the work of Cimenera and Tukey (1989), as if their main application is determining accuracy, which is not the case. Pyzdek fails to recognize it as the generalization of Clifford's (1959) robust individuals chart, based on the median moving range. Next introduced are variance components of measurement. Some might be irritated by the fact that the standard deviations are estimated by R/d2 others by the typo in the standard deviation formula. My personal regret is that measurement resolution is defined but never developed. I was hoping for the equivalent of a control chart for the slope of a two point calibration.

Deeper into the book is a chaplet on SPC for linearly drifting processes, based on the approach of Taguchi (1986, chap. 3). (Whoops, typo in Formula 20.3.) The example, involving the color index of an ink whose solvent is evaporating. seems simple and apt. What I miss here is Pyzdek's careful step-by-step guide path, and his advice for not-so-capable processes.

So for a book of diverse chapters, what is the recommendation? Would I hire this consultant, or suggest his book as a references Would I teach using it, or suggest it for self-study? Well, I might hire Thomas Pyzdek, for I fully believe he can teach people from his book. And I will use it for a reference. Pyzdek has demonstrated his depth of experience and taken care to list issues for the many techniques he has included here. The absence of student exercises limits the book's usefulness for self-study, but Pyzdek's stronger chapters are at a standard for which other SPC authors should strive.

William D. HEAVELIN

Advanced Micro Devices

REFERENCES

Cimenera, J. L., and Tukey, J. H. (1989), "Control Charting Automated Laboratory Instruments When Many Successive Differences May Be Zero," Journal of Quality Technology, 21, 7-15.

Clifford, P. C. (1959), "Control Charts Without Calculations," Industrial Quality Control, 15, 2-6.

Ewan, W. D. (1963), "When and How to Use Cu-Sum Charts," Technometrics, 5, pp. 1-22.