PDA Book Summary

I just completed a book entitled Practical Data Analysis for Designed Experiments, to appear probably in December 1996 with Chapman & Hall: London. This text is aimed at statisticians and scientists who would like to gain practical experience with the design and analysis of experiments, with enough theory to understand the analysis of standard and non-standard experimental designs. This work has been inspired by over a dozen years of statistical consulting with scientists in CALS, augmented by teaching Statistics 850 ``Theory and Practice of Analysis of Variance'' and Statistics 998 ``Statistical Consulting''.

This book is divided into nine parts, begining with an overview which places data in context of experiments in the process of scientific discovery of knowledge. The following parts examine issues in comparing groups, sorting out factor effects, questioning assumptions, regressing with factors, deciding on fixed or random effects, nesting experimental units and repeated measures on subjects.

Part A deals with the interplay of design and analysis in working with data, taking a broad view of the need to understand the context of an experiment, to be aware of the processes which surround data collection and manipulation. Part B develops the comparison of means among groups of data, including issues of multiple comparisons, ordered groups and errors in variables. Sorting out two or more factors in Part C introduces issues of main effects and interactions and model selection, primarily using balanced designs.

The next three parts address complications in a completely randomized design. Unbalanced designs, missing data and the linear model framework are examined in Part D. Assumptions are examined in Part E, with various approaches suggested, including the use of plots, transformations and ranks. Part F concerns analysis of covariance and multiple responses, addressing pragmatic issues of combining quantitative variates and qualitative factors in data analysis.

The remaining parts examine different sized experimental units, building on design issues involving blocking and sub-sampling. Part G examines several source of random variation, distinguishing between fixed effects considered up to this point and random effects. Nested designs in Part H translate random effects into the errors for different sized experimental units. Part I addresses experiments which involve repeated measures on the same subject, and cross-over designs in which subjects may be assigned several treatments in succession.

The book aims for a compromise in practical data analysis for designed experiments between linear models theory in an intuitive framework and pragmatic suggestions on data analysis and interpretation. Accompanying Internet resources at http://www.stat.wisc.edu/~yandell/pda/ include an outline, datasets from the examples, new S-Plus functions, and SAS and S-Plus code samples. The S-Plus function library for figures in the book is in the public domain and available at our FTP site as ftp://ftp.stat.wisc.edu/pub/yandell/pda.tar.gz (see PDA installation instructions).


Last modified: Wed Jun 19 12:35:01 1996 by Brian Yandell