The first edition of this book was published in 1997 with 719 pages, the second edition in 2000 with 1016 pages, the third edition in 2003 with 1139 pages and now the fourth edition has been published with 1736 pages. It is printed on thin paper to minimize the weight and size but is still heavy. It is an impressive work, not only in its size, but also in the coverage and details. In total in covers more than 160 statistical procedures, divided into 40 different test categories. It is written to provide researchers, teachers and students in the fields of mathematics/statistics, the social, biological and environmental sciences, business and education with a comprehensive reference in the areas of parametric and nonparametric statistics. The intended readership is broad and includes both people who have little or no knowledge of statistics and those who are well-versed in the subject.
The book starts with a lengthy introduction (more than 100 pages) that gives a general overview of basic terminology, concepts and methods in descriptive statistics and experimental design. This includes discussions of descriptive versus in inferential statistics, levels of measurement, measures of central tendency, variability, skewness and kurtosis, visual methods for displaying data, the normal distribution, hypothesis testing and its pros and cons, the foundations of experimental design, sampling methodologies and the basic principles of probability.
The introduction covers, in a condensed form, all the necessary material of an introductory course in statistics. It also gives an outline of the statistical procedures, and guidelines for selecting the appropriate statistical procedure based on the kind of measure used, the level of the data, the number of samples, possible dependence in the data set and the hypothesis to be evaluated. This is in line with the stated goal of the book: to provide the reader with clear and easy-to-follow guidelines for conducting statistical analyses.
The main part of the book consists of a chapter for each of the 40 test categories discussed in the book. These 40 chapters are in turn collected into seven broader parts, namely
- Inferential statistical tests employed with a single sample,
- Inferential statistical tests employed with a two independent samples (and related measures of association/correlation),
- Inferential statistical tests employed with a two dependent samples (and related measures of association/correlation),
- Inferential statistical tests employed with two or more independent samples (and related measures of association/correlation),
- Inferential statistical tests employed with a factorial design (and related measures of association/correlation),
- Measures of association/correlation, and
- Multivariate statistical analysis.
The discussion of each test category follows a consistent format. It is centred around a specific main test and contains nine sections:
- Hypothesis evaluated with the test and relevant background information, which besides giving a general verbal statement of the hypothesis that the test evaluates discusses such topics as the type of data and experimental design for which the test is appropriate, the assumptions underlying the test and the effects of violating these assumptions.
- Example, which describes one or two examples of situations where the test is applied, accompanied with a data set.
- Null versus alternative hypotheses, which gives formal symbolic definition of the null and alternative hypotheses and a lengthier verbal discussion of how these should be interpreted.
- Test computations, which contains a step-by-step description, including all necessary formulas, for computing the relevant test statistics, illustrated with data from the examples in section II. For the multivariate statistical procedures, SPSS is used in this section, at the expense of some formulas.
- Interpretations of the test results, which gives detailed guidelines of how to evaluate the test results, e.g. with the help of tables of critical values, and how to interpret the results in reference to the null and alternative hypotheses.
- Additional analytical procedures for the test and/or related tests, which discusses further aspects of the tests, e.g. computation of power, construction of confidence intervals, normality approximation and related statistical procedures.
- Additional discussion of the test, which discusses e.g. theoretical aspects of the test, relationships with other tests and some practical issues.
- Additional examples illustrating the use of the test, which gives one or more additional examples of situations, for which the test is applicable.
- Addendum, which describes miscellaneous results related to the test, such as additional related statistical procedures, and also some general statistical topics that are not covered in the introduction, and is somehow related to this specific test. This includes e.g. resampling and randomization techniques, specific probability distributions, Bayesian statistics and meta analysis. This section is, however, not included in all chapters.
Finally, all chapters has a comprehensive list of references with primary and secondary sources and most chapters also a section with endnotes containing further clarification or expansion of the material discussed in the chapter.
Besides these 40 main chapters, the book also has a chapter that gives an overview of matrix algebra and an appendix with more than 50 pages of tables with critical values and graphs of power curves.
This book is really an impressive work. It describes all basic statistical procedures that an applied statistician will encounter, covering both practical and theoretical issues in detail. In spite of the detailed discussions, the book is very readable, well written and pedagogical. The encyclopaedic approach makes it very useful as a handbook, where one can jump into a chapter and find all the necessary information about a specific test.
A minor complaint is that in the multivariate analysis part of the book there is a too heavy reliance on SPSS and access to a computer package with precoded procedures. Because of this some necessary formulas are excluded, which would be good to have when one wants to write program code oneself.
This minor complaint notwithstanding, I have to say that this is a must-have book for every applied statistician. If you are working with applied statistics and have to choose only one book to buy, this is the one. It is well worth its price. I wholeheartedly recommend it.
Andreas Rosenblad (firstname.lastname@example.org) is a Ph. D. in statistics graduated from the Division of Statistics, Department of Information Science, Uppsala University. He is currently working as a biostatistician at the Västerås Center for Clinical Research (part of Uppsala University) at Central Hospital in Västerås, Sweden. His primary research interest is in bootstrap methods and quantile regression with applications in biostatistics.