Customer Reviews

Design of Experiments: A Realistic Approach

5 star:		(1)
4 star:		(1)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

I use this text to teach an applied, but advanced, course in Design of Experiment for (mostly) engineers and physical scientists. If you want an introductory text on DOE, don't buy this book. If you run experiments on a regular basis and have a reasonable comfort level with statistics, this book will help you design and analyze better experiments. The "no-name" in the title will likely be appreciated by experimentalists--most textbooks in DOE provide basic good designs, ones that have names (factorial, Latin square, RCBD,...), but most interesting experiments require designs that are too complex to be named. These authors show general design principles that are used to build designs of any complexity within a certain class (complete balanced designs). They also show how to generate designs outside of this class, such as fractional factorial designs. This covers a very broad spectrum of designs for industrial experiments. If you have been surprised that the designs in the standard texts don't seem to be quite right for your problem, this book is for you. I would suggest the uninitiated begin with a simpler text, and buy this one next.

Most books and articles on experiment design spend very few, if any, pages on experiment design. Instead, they describe a myriad of existing experiments and leave the reader to select from some sort of catalogue. After that, they move directly into the analysis of experimental data. Well, no amount of analysis is going to salvage data collected from an inappropriate design, and a more frequent problem, analyzing data with techniques that don't reflect how the experiment was actually conducted can lead to misleading, if not irreproducible results. With Dr. Lorenzen's approach, the practitioner can create an initial design, review its utility as well as short-comings and then proceed through enough iterations to converge at the best set of trade-offs for their situation. As one example, classic texts discuss "Completely Randomized Designs" in one chapter and then "Split Plots" in another, without addressing why one might choose one over the other. Without this guidance, people tend to overwhelm true experimental signals with controllable errors or trade off diagnostic ease for difficulties in changing configurations in real-world settings like manufacturing plants and test facilities.

You're better off understanding the building blocks of variable types, nesting, restrictions on randomization and so forth as described in this text. After that, you're well equipped to both design and accurately analyze the data you've taken. Now that's efficiency!