See discussions, stats, and author profiles for this publication at: [Link]

net/publication/4741668

Statistics: Informed Decisions Using Data, Michael Sullivan, III; An Introduction

to Statistical Analysis for Business and Industry: A Problem Solving Approach,
Michael Stuart

Article in The American Statistician · February 2008

Source: RePEc

CITATIONS READS

0 4,969

1 author:

Richard J Cleary
Babson College
52 PUBLICATIONS 722 CITATIONS

SEE PROFILE

All content following this page was uploaded by Richard J Cleary on 28 July 2022.

The user has requested enhancement of the downloaded file.

 This book is a great addition to a SAS reference library. It compliments other or government. It is easy to read and the chapters are organized in a pedagog-
SAS references, and you can never have too many examples. How can you not ically favorable fashion, including pharmaceutical subject matter, one or more
like a book when the author is pictured holding his cat? real-life examples, statistical methodology, SAS code, SAS output, regulatory
issues, and a discussion of the results. Students and faculty will find an insight-
Margie R. G URWIT ful orientation to some of the experimental design and statistical problems in
University of Missouri drug research and early drug development activities, with the option of rein-
forcement through interactive data analysis sessions.

Pharmaceutical Statistics Using SAS: A Practical Guide. Thomas E. B RADSTREET

Alex D MITRIENKO, Christy C HUANG -S TEIN, Ralph D’AGOSTINO (eds.). Merck Research Labs, UG1D-44
Cary, NC: SAS Institute, Inc., 2007, ix+444 pp., $69.95 (P), ISBN: 1-
59047-886-X.
Pharmaceutical Statistics Using SAS R : A Practical Approach is composed A Quick Course in Statistical Process Control.
of 14 chapters written by 42 experienced statistical and software professionals Mick N ORTON. Upper Saddle River, NJ: Pearson Education, Inc., 2005,
who work in the pharmaceutical industry, academia, the Food and Drug Ad- x+397 pp., $51.40 (P), ISBN 0-13-134666-0.
ministration, and the SAS Institute. The chapters (except the introduction) are
constructed around one or more real-life examples with accompanying data; Designed for a practitioner with no exposure to fundamental statistical ideas,
also provided are some of the relevant SAS code including macros and mod- this book describes the core concepts and implementation strategies for statisti-
ules, sample SAS output, and references. Complete SAS code and datasets are cal process control (SPC). The casual conversational tone, the careful descrip-
available at the book’s companion website. tions with almost no mathematical notation, the wealth of plots and figures, and
An impressively wide range of topics are covered in three major areas of the detailed examples using MINITAB make the material accessible to all.
pharmaceutical research and development: drug discovery (Chapters: 2, 3, 4, Chapters 1–4 guide the reader through some statistical preliminaries, meth-
6, 7), safety assessment (Chapter 5), and earlier phase clinical trials (Chapters ods for charting sample means and variation, signals and measures used to as-
6–14). Specifically, the topics include (1) an introduction to nonclinical, clin- sess control and quality, and other control charts. Restricting itself to the very
ical, and emerging statistical opportunities; (2) modern classification methods core of SPC, and doing a thorough treatment of these concepts and charts, the
for large datasets in drug discovery; (3) model building in drug discovery; (4) book will be a good starting point for practitioners. For a more detailed mathe-
validation of analytical methods; (5) in vivo toxicology studies in animals; (6) matical development of the methods and some additional tools, which assumes
nonparametric methods with special attention given to the Behrens–Fisher prob- only undergraduate-level mathematics and introductory statistics, Montgomery
lem; (7) ten nonlinear optimal design problems ranging from logistic functions (2004) is an excellent alternative.
for categorical and continuous responses to beta regression to logit and probit Chapter 5 fills in some of the basic concepts of probability which have been
models to pharmacokinetic models with and without cost constraints; (8) assess- used in the first four chapters, while Chapter 6 provides a cursory overview of
ment of average bioequivalence, dose proportionality, and clearance in humans; the quality movement and some standard problem-solving tools, such as Pareto
(9) randomization procedures including permuted block, stratified, and covari- charts and cause-and-effect diagrams. A small number of exercises are included
ate adaptive; (10) sample size and power in hypothesis testing; (11) a range of in each chapter to help the reader reproduce the results demonstrated and prac-
strategies for evaluating dose ranging and dose response; (12) a debate on how tice the concepts.
to handle incomplete data; (13) assessing the reliability and validity of outcome Overall, the book will be a comfortable read for those new to statistical idea
measures; and (14) decision analysis regarding, among other things, go/no go and statistical process control but will need to be supplemented with other ma-
decisions and project prioritization. Given the contents, perhaps the book’s ti- terials once the reader has gained an understanding of the basics.
tle is too general. Indeed the authors point out in the preface, “Since many of
the statistical issues encountered during late-phase drug development (e.g., sur- Christine M. A NDERSON -C OOK
vival analysis and interim analyses) have been covered in other books, this book Los Alamos National Laboratory
focuses on statistical methods to support research and early drug development
activities.” REFERENCES
Other notable strengths of the book are its easy-to-read expository style and
tutorial structure. Generally, chapter contents follow a logical progression of Montgomery, D. C. (2004), Introduction to Statistical Quality Control, New
presentation: the pharmaceutical area of application, followed by one or more York: Wiley.
real-life examples, statistical methodology, SAS code, SAS output, and a dis-
cussion of the results—not just in terms of the SAS mechanics, but focused
on the relevance to real-life examples—often accompanied by advice from the Statistics: Informed Decisions Using Data.
authors. Regulatory issues and their implications are also part of the mix; reg- Michael S ULLIVAN, III. Upper Saddle River, NJ: Prentice Hall, 2004, xxxv
ulatory and other websites are provided. The text’s readability is enhanced by + 823 pp. plus appendices, $117.35 (H + CD), ISBN: 0-13-046489-9.
parcing AU: Should be “parsing”? data, SAS code, and SAS output (including
graphs) in box frames. An Introduction to Statistical Analysis for Business and Industry: A Prob-
Although the book is targeted at pharmaceutical practitioners, both statistics lem Solving Approach.
students and their faculty who are interested in working, or consulting with, Michael S TUART. London: Arnold, 2003, xiii + 384 pp., $72.29 (P), ISBN:
the pharmaceutical industry will find valuable insight into some of the subject 0-340-80844-6.
matter and accompanying experimental design and statistical analysis problems
encountered by pharmaceutical statisticians on a daily basis. The book is suit- Most of us who teach mathematics and statistics in the United States have
able for classroom use and self study, but along with the companion website, had the experience of a student showing up at our office with a form for taking
it is ideal for guiding interactive computer laboratory data analysis sessions. In a semester abroad. They are seeking our permission to substitute a course in
addition, the book and the references contained within may provide some initial an exotic locale for the similar course at our own school. Young and enthusias-
conceptual motivation and background reading for dissertation research. tic instructors often compare syllabi, find them to be a decent match, sign the
The editors suggest that, since the book does not place much empha- form and send the student happily along only to hear a few months later that
sis on theoretical details, it will benefit a broad group of pharmaceutical re- the student has failed or been forced to drop the course. The overseas version
searchers including biologists, chemists, pharmacokineticists, and pharmacol- proves too tough, too theoretical, too heavily weighted toward a cumulative fi-
ogists. Whether or not this will be the case is unclear. The accessibility to nal, or generally too incomprehensible. Any of us who have lived through this
“canned” software in these subject matter areas is in direct competition with for many years are forced to conclude that our courses in the U.S. are too easy,
SAS, although the wider range of statistical choices and greater flexibility al- or that our pedagogical approach is vastly superior, or our students lose their
lowed for by the SAS environment should be appealing. wits when they go overseas. Perhaps some interaction between those factors
In summary, Pharmaceutical Statistics Using SAS R : A Practical Approach has led to courses with almost identical syllabi to have very little in common in
is a notable collection of 14 multiauthored contributions covering a wide range practice.
of statistical and subject matter topics in pharmaceutical research and develop- I bring this up in the case of the two books under review because they pro-
ment, accompanied by a website with complete SAS code and datasets. This vide a small window into this issue. Each of the books is a perfectly reasonable
book will be useful to practicing biostatisticians, whether in industry, academia, candidate for selection as the text for a first course in business statistics. It is

The American Statistician, May 2008, Vol. 62, No. 2 5

easy to map many of the chapter titles from one onto the other, although the or- real data from government reports or academic journals. Supplemental instruc-
der differs slightly and a few topics that do not always make it into a first course tion for the various forms of technology is included. Stuart’s treatment is almost
are present in one or the other. Stuart, who teaches at Trinity College, Dublin, the same length in the number of (smaller) pages but without the step-by-step
includes chapters on time series and applications in finance while Sullivan, of instructions and much less detailed examples. The remarkable thing about this
Joliet Junior College, has a chapter on nonparametric statistics. The prefaces of exposition is that the Stuart book contains almost no formal probability, so that
the texts make similar claims for learning objectives and the promise of a course almost all of the ideas that are classified as prerequisite reading in Sullivan are
that recognizes (or recognises, in Stuart’s case) the need for statistical thinking developed with this section itself. The Stuart book makes no suggestion for
and conceptual understanding. computation of p-values, simply comparing the computed chi-square value for
Two teachers, each using one of these texts, might indeed have similar syl- the data to the reference value in the table.
labi. If, however, their pedagogical styles differ as much as the style of these In summary, we have here two textbooks aimed at different audiences that
books do, then the courses might appear very different indeed to students. The are, in many ways, closer together than they have ever been but still a world
text by Stuart (Trinity College, Dublin) is tidy, done in black and white only, apart in expectation. Which of the approaches here would do better at improv-
and runs a little less than 400 smallish paperback pages. The approach could be ing our students understanding of statistics? Both packages have appeal to me.
summarized as “Less is more.” There is a companion website, small and appar- The Stuart text, with its economy of scale and arguments, seems to reflect my
ently maintained by the author himself. It has not been updated in some time. goal that students come to see the subject as a few key ideas, not hundreds of
The site offers a little bit of support for instructors in the form of datasets for little formulas. The Sullivan text does a wonderful job of anticipating student
the problems and some modest teaching suggestions. The book fits neatly in the misconceptions and providing a focus on delivering correct answers to specific
pocket of my winter coat. The text by Sullivan, recently out in a second edition, questions that might arise in a business context. While I have not done the re-
encompasses the classic U.S. “more is more” approach, as it comes with a com- search to answer the question of which is more effective, I have at least gained
panion website and a host of instructional resources. The first edition runs more a valuable lesson in advising students bound for a semester abroad
than 800 hard cover pages, not counting appendices, in a multicolor edition with
a huge variety of graphics, icons, reminders, pull-out summaries and pointers. Richard J. C LEARY
This text and my lunch box do not fit in my backpack at the same time. Bentley College
While it seems unlikely to me that any faculty member would ever actu-
ally make a choice between these two particular texts, as each is very much a
product of the academic culture of the authors, the contrast seems to me to pro- Variations on Split Plot and Split Block Experiment Designs.
vide some potential for study in learning styles and assessment. Do the icons, Walter T. F EDERER and Freedom K ING. Hoboken, NJ: Wiley, 2007,
the self-paced learning exercises, and the all the rest justify the added cost and xiv+270 pp., $99.95 (H), ISBN: 0-470-08149-X.
size of the bigger book? How could we design a study to measure their effec-
tiveness? Would a course that uses the sparser text but has more personalized Over the years many significant contributions have been made to the litera-
instructions (smaller classes, or supervised lab time) be as effective as a more ture on split plot designs and their analyses. The objective of the authors was to
all-encompassing text taught in combination with a large lecture? Does a stu- provide comprehensive coverage of the many variations on split plot and split
dent used to one approach have a more difficult time when faced with the other? block designs.
I am not in a position at the moment to carry out such a research project, but This book has 11 chapters. They are titled: (1) The standard split plot exper-
I am intrigued by the interesting design questions it would raise, as well as by iment design; (2) Standard split block experiment design; (3) Variations of the
the funding potential for carrying out educational research with an international split plot experiment design; (4) Variations of the split block experiment design;
dimension. (5) Combinations of SPEDs and SBEDs; (6) World records for the largest anal-
In addition to the large differences in scale between these books, let me iso- ysis of variance table (259 lines) and for the most error terms (62) in one anal-
late some ways in which the texts differ in respects that would be material to an ysis of variance; (7) Augmented split plot experiment design; (8) Augmented
instructor in a first course. The most surprising thing about the Stuart text is that split block experiment design; (9) Missing observations in split plot and split
the testing of means, which arises in various contexts (control charts in Chapter block experiment designs; (10) Combining split plot or split blocks designed
4, one-sample testing in Chapter 5) is done entirely via reference to the Z distri- experiments over sites; (11) Covariance analyses for split plot and split block
bution. Discussion of the t distribution is confined to a paragraph near the end experiment designs. The book focuses most heavily on Federer’s work on split
of the book (Chapter 11, p. 365). Students using the book will have seen, and plot and split block designs.
been asked to interpret, the results of t tests and confidence intervals in Chapter The book is aimed at a variety of audiences including applied statisticians
6 on simple linear regression and Chapter 8 on multiple linear regression. In this from industry, medicine, agriculture, chemistry, and other fields of study. The
case the meaning of these tests is given only obliquely as even the concept of p authors emphasize the use of the correct error term, the effect on the error term
value is mentioned only in a footnote. Interpretation always refers back to the when factors are either random or fixed, and how to deal with missing obser-
graphics and the data, with an emphasis on estimating variability and its impacts vations. Thus, the book is especially useful for researchers using complicated
on the application at hand. There is no t distribution table in the appendix, just split plot and related designs. The book also presents some useful nonstandard
normal and chi-squared tables. designs and analyses such as augmented split block designs, combining experi-
Sullivan takes what I would consider the classic treatment of nearly all the ments over sites, and covariance analysis for these designs.
textbooks that my department would consider, with 18 pages in Section 8.1 This book can also be used as a reference book by graduate students taking
devoted to confidence intervals for a single mean with population standard de- a course in experimental design. It proceeds in a step-by-step, tutorial fashion
viation known, followed by a 15-page section on the same topic with sigma through the various split plot/block designs. There are a lot of practical exam-
unknown. Many of the pictures in the margin showing tail probabilities appear ples, many SAS programs, and also some datasets for the reader to explore. At
in their nearly identical forms in each chapter. The standard normal table and t the end of each chapter, there are some helpful remarks and some good exercises
tables appear in the inside covers of the book, with instructions on performing to make sure that concepts are reinforced. There are also selected references at
the needed table look-ups in Excel, Minitab, and on the TI-83 Plus calcula- the end of each chapter.
tor. Page 483 includes a detailed four-color flowchart for deciding which of the One limitation of this book is that mathematical aspects like least squares
distributions (or neither, if the data are not approximately normal without any derivation is not provided. And it would have been useful had the authors given
outliers) is the right one. To make specific some of the questions I asked ear- the approximate degrees of freedom for the standard error given in equation
lier, is the additional 15-page section, with associated exercises, the right tool (1.7).
to reinforce the concept of interval estimation? In summary, the authors have done an excellent job in putting together a
Let us consider the introduction of the chi-square test for independence, this book that covers the literature on split plot and split block experiments. The
time looking first at the Sullivan text where the topic makes up Section 11.3. book is easy to read, and it demands little statistical background to comprehend
There are just over ten pages of text, with four examples worked in detail. There the major ideas. The authors’ objective has been met. Variations on Split Plot
are several color graphics, and marginal notes that include a “caution” that has and Split Block Experiment Designs is a welcome addition to the book shelves
replaced the more traditional footnote as a way of working in subtle nuances of all researchers involved with experimental design.
in the content. (The caution, for instance, warns students that rejecting a null
hypothesis of independence does not isolate which of several conditional dis- Lakshmi PADGETT
tributions are different.) This is followed by 19 problems, many of which use Johnson & Johnson

6 Reviews of Books and Teaching Materials

View publication stats