Review By: Robert Bruce Kelsey, Ph.D.
06/23/2010This book shows software managers, quality managers, and software process improvement specialists how to use statistical analysis to solve real-world problems. Why do software projects cost so much and take so long? How can we improve time to market and cut maintenance costs? In six very carefully constructed chapters, the author explains how to analyze project data and how to develop and evaluate models that can answer these questions.

The first chapter introduces a “recipe” for analysis: data validation, preliminary analysis of variables and statistical models, analysis of variance, and model validation techniques. The emphasis is on critical thinking, not on number crunching. What are we trying to learn about software projects? What factors (variables) do we need to analyze to achieve this understanding? How do these variables affect one another? By the end of the first chapter, the reader appreciates the need for correlation analyses, stepwise regression, and analysis of variance. Then Maxwell puts these tools to work.

The next four chapters provide case studies in developing models for productivity, time to market, development cost, and maintenance cost. Each study builds upon the previous one, reinforcing basic concepts while also introducing new analysis techniques. By the end of the rather lengthy but lucid final study on maintenance cost, the reader has learned how to create and improve a practical, statistically well-founded software measurement program.

The final chapter presents the mathematics and theory behind the graphs, charts, and tables that are used so effectively and extensively throughout the book. A knowledge of algebra is all one needs to follow the explanations, although the author is quick to point out that you’ll need a statistics or SPC software package for any significant analysis effort. It might seem odd to place a tutorial on statistics at the end of a book on applied statistics, but in fact that’s perfectly in line with the author’s intent. She wants to teach software managers only “what they need to know” about statistics. According to this author, what they need to know most is how to make logically sound and statistically valid judgments about data and analysis models.

That’s what sets this book apart from the many books and articles on software metrics. Corporate board rooms everywhere are littered with presentations on project size, cost, and quality that don’t tell the whole story—or don’t even tell the right story. That’s because too often metrics efforts are based on untested assumptions about data relationships or are predisposed to use certain models rather than others. Writing in first person and from personal experience, the author is clearly determined to make sure that her readers won’t make the same mistakes. She explains how to handle incomplete or dirty data, how to verify your choice of dependent variables, how to choose between three and five variable models when each has about the same apparent estimation accuracy, and how to determine which factors (variables) can be used to improve cost, schedule, or productivity performance. It’s this level of detail and truly practical advice that makes this book a “must have” for anyone involved in software engineering management, quality assurance, and process improvement.