Book Review by Ray Rosenblatt: Metrics and Models in Software Quality Engineering
Stephen H. Kan
Addison-Wesley, 2002

The following is what Capers Jones had to say about Metrics and Models in Software Quality Engineering:

“This is the single best book on software quality engineering and metrics that I’ve encountered.”

Coming from the software-measurement-industry icon himself, that is a huge endorsement. And IMHO it’s very well-deserved.

Metrics and Models in Software Quality Engineering by Dr. Stephen H. Kan is superb. It packs an enormous amount of solid information into its 528 pages, ostensibly aiming for a broad audience by offering a combination of content you might not expect to see in a single volume. By this I mean it includes high-level and strategic topics (such as the chapter entitled Do’s and Don’ts of Software Process Improvement) that will be of interest primarily to process architects and managers, yet it also thoroughly pours over the mathematical and algorithmic details needed by software implementation developers.

The book begins with the obligatory overview of the basics, but even here the material is first rate due to the sprinkling of useful tips and caveats designed to help the reader avoid many of the traps that software measurement novices often fall into. A good example is the introductory treatment of the concept of lines of code (LOCs). While the book presents LOC as a useful metric in the context of defect rate calculation (i.e. how many defects per thousand LOC?), it strongly urges us before we plunge in to beware the pitfalls. For instance:

(a) Depending on the counting method used, final counts may differ by as much as 500%! Therefore any initiative involving the counting of LOCs must begin with a precise operational definition of what constitutes a countable line of code.

Or,

(b) Although some organizations may try to use LOCs for productivity studies, the practice is fraught with problems and should be avoided altogether. One oh-so-very-interesting little snag with using LOCs as a measure of productivity is that the number of LOCs in a software program is negatively correlated with design efficiency! Put another way, if you evaluate your developers based on how many LOCs they produce, you are effectively rewarding sloppy design.

The meat of the book however is the collection of quality models and metrics covered in chapters 6 through 14. In total, they analyze software quality from four perspectives: 1) the development process, 2) the software product, 3) system availability, and 4) customer satisfaction. Some models are retrospective, some are predictive, and some provide a snapshot of present reality. In each case, model composition, application, and interpretation are lucidly explained.

Be forewarned that some fraction of the material here may be too statistically technical for many organizations. The Rayleigh model used for predictive quality analysis, for example, gets into probability distribution and density functions; and usage of the model requires special tools and statistical modeling expertise. Rest assured however that such mathematical sophistication is the exception. Those who prefer to keep things simple will especially appreciate that most of these chapters conclude with a short section entitled Recommendations for Small Organizations where you often find down-to-earth advice for implementing a simplified substitute for the model just discussed. This is one more example I suppose of how the book aims to satisfy a broad audience.

All in all, if you are looking for guidance in developing a software metrics capability with quality improvement as your objective, then Metrics and Models by Dr. Stephen H. Kan is positively the book you want. Once you finish reading it, you may very well decide it is the only book you need.