Beruflich Dokumente
Kultur Dokumente
net/publication/254331890
CITATIONS READS
10 5,116
1 author:
Ron Snee
Snee Associates, LLC
403 PUBLICATIONS 6,151 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Ron Snee on 22 October 2016.
Book Reviews
(Publishers are invited to send books for review to Dr. Gary to make scientists other than statisticians aware of non-
G. Koch, Department of Biostatist,ics, School of Public parametric methods, and secondarily to produce a handbook
Health, University of North Carolina, Chapel Hill, North of these techniques. Although not mentioned therein as a
Carolina 27514) specific goal, the book is intended for use as a text for a formal
course; it has been class tested extensively with senior students
at Florida State University in the Department of Psychology
Nonparametric Statistical Methods, by M. Hollander and Statistics. The book is nonm&thematical, with the only
and D. A. Wolfe. John Wiley & Sons, 1973, 503 pp., prerequisite an introductory statistics collrse.
$18.95. Siegel’s book has also been used as both a textbook and
While the statistics journals have been replete with articles handbook. However, its intended audience was behavioral
concerning nonparametric techniques for many years, there scientists, while Hollander and Wolfe are apparently trying
has been a dearth of books on the subject until recently. to provide a handbook for persons engaged in quantitative
Fraser’s book [2] was the sole treatise in the theoretical area research in all scientific areas, whether natural, physical,
until Hajek and Sidak [3]. Similarly, Siegel [1956] and Tate medical, behavioral, or human. The authors’ means of enticing
and Clelland [1937] represent the only early attempts to such a broad audience is by inclusion of examples which apply
present nonparametric methods to the applied user. However, the methods to data from actual experiments in the fields
in the past five years there has been a flurry of publications of astronomy, biology, criminology, education, engineering,
in both the theoretical and applied areas. Statisticians lament environmental science, geology, home economics, medicine,
that a book like Siegel [7] (see Savage [5]) seems to persist oceanography, physics, psychology, sociology and space
as the primary handbook scmrce for users, since that book science.
focuses only on significance tests, gives little attention to For each particular statistical technique included, the
their properties, contains erroneous statements, and does methodology discussion includes (when applicable) the step-
not include any of the procedures developed in the last by-step procedure, large sample approximation, means of
eighteen years. dealing with ties, a numerical example, comments, properties,
Conover’s book [l] should probably be considered the references and problems. The fact, that this format is followed
first note-worthy attempt to updat)e and replace Siegel (see consistently throughout the book effectively places it in the
Sen [6]), although some feel it is appropriate for a more category of a useful and valuable handbook. However, the
sophisticated audience. Nonparametric Statistical Methods by resulting sterility and roteness discourages reader interest
Hollander and Wolfe is of the same genre and a welcome and thereby detracts from its appeal as a textbook, partic-
addition to the literature, since its scope is broader and more ularly for undergraduates or for those scientists who consider
current, than any other book presently available at the same statistical methods primarily a necessary evil in their research.
level. While a less formal and concise style would detract from
The stat,istical methods covered in this book include not the book’s value as a handbook, it would elicit more reader
only the familiar nonparametric estimators, tests and con- interest and thereby enhance its contribution to pedagogy.
fidence procedures, but also many techniques which have The total lack of availabilit,y of answers and/or solutions
probably never actually been used in a practical problem to any problems is a detriment to usefulness as either a text-
of data analysis. The experimental situations described in book or a handbook for self-instruction.
sequence over 253 pages are the dichotomous data problem, While the prerequisite knowledge is stated accurately,
the one-sample location problem, the two-sample location the reader needs a great facility for dealing with symbols
problem, the two-sample dispersion problem, the one-way and substituting in formulas. Most introductory courses in
and two-way layont,s (t,he k-sample location problem), the applied statistics simply would not develop enough algebraic
two-variable independence (association) problem, regression sophistication for readers to handle effectively some of the
problems involving slopes, and some miscellaneous methods techniques described and to understand the comments made.
designed for broad alternatives. Even the descriptions of some procedures, while qllite logical,
The remainder of the book consists of extensive Tables accurate and precise, frequently involve notation and
and Charts (176 pages), Glossary (32 pages), Bibliography symbolism which are far from elementary. Without consider-
(24 pages), and Index. The Bibliography includes a large able motivation or the aid of an inst,ructor, the reader may
number of journal articles which expand on the properties of lose sight of or interest in the ultimate procedure.
those procedures included in the text. The Tables and Charts The comments given for all techniques are usually quite
include 27 tables useful for inference plus a square root table varied in level, completeness and difficulty, and some seem
and a table of natural logarithms. The two charts present quite esoteric for the intended audience. Many comments
percentile points for Student,‘s t distribut,ion and t,he chi- are replete with references, but in most cases these references
square distribution in a rat,her atypical format. This is perhaps are to articles which cannot be comprehended without con-
the best collection of tables in any nonparametrics book. siderable formal training in statistics. While t,hese references
Whenever practical and possible, exact tail probabilities are would contribute substantially to an intermediate or advanced
given instead of critical values at selected levels of significance. course for statistics majors, or to a reference book for
Each table is headed by a brief but, concise explanatory note statisticians, they add little with respect to either of the goals
concerning its use. The two charts are probably unfamiliar stated in the Preface.
to most, practitioners and are more difficult to read than the The properties listed for tests usually include only con-
more traditional method of tabulation. sistency and asymptotic efficiency. Little attention is given
The Preface states that the goal of t,his book is primarily to power, small sample properties, accuracy of large sample
477
BOOK REVIEWS
approximations, the consequence of weakening the assump- goal is thereby fully achieved. The selectivity of techniques
tions or the effect of ties on reliability. limits this book’s usefulness as a handbook or reference, and
For each procedure which is fully covered, one numerical its concision hampers classroom effectiveness. On the other
example is worked through and a few problems, all represent- hand, students and research workers interested in applying
ing data from actual experiments, are given. There are ap- nonparametric methods would benefit from a thorough
proximately 45 examples and 126 problems. However, new acquaintance with this volume. This book is for readers who
data are not given for each example, and some problems are want good tables, a large assortment of topics with a wide
more theoretical than applied. There are about 40 actual range of complexities, and who are unconcerned with solved
data situations, rather unevenly distributed among the various problems or a wide range of examples.
sciences. In fact, almost half of the experiments described
are from the biomedical area. Such experiments frequently REFERENCES
involve terms, processes and substances that are unfamiliar [l] CONOVER, W. J., Practical Nonparametric Statistics, New
to scientists in other areas, and thus elicit little interest from York: John Wiley & Sons, 1971.
readers outside the relevant field. The social and behavioral [2] FHASER, D. A. S., Nonparametric Methods in Statistics,
sciences are relatively poorly represented, and yet nonpara- New York: John Wiley & Sons, 1957.
metric statistical methods are particularly useful to such [3] HAJEIC, J. and Z. SIDAK, Theory of Rank Tests, New York:
experimenters. The relatively small number and narrow Academic Press, 1967.
representation of problems and examples, plus the fact that [4] NOETHRR, G. E., “Distribution-Free Confidence Inter-
no answers are given for any problems, detract considerably vals,” The American Statistician, 26 (February 1972),
from the book’s usefulness as a general handbook and text- 3941.
book. [5] SAVAGE, I. R., “Nonparametric Statistics” (Book Review),
The major difficulty with the book is the relegation of Journal of the American Statistical Association, 52 (1%7),
definitions to the Glossary in the Appendix. Reading would 331-344.
be greatly simplified if concepts such as asymptotic relative IS] SEN, P. K., “Practical Nonparametric Statistics” (Book
efficiency or consistency of a test, probably unfamiliar terms Review), Journal of the American Statistical Association,
to most members of the intended audience, were defined in 67 (1972), 246-247.
the text. Barring that, some symbol should be used in the [7] &GEL, S., Nonparametric Statistics for the Behavioral
text to indicate that a new term is being introduced and it is Sciences, New York: McGraw Hill Book Company, 1956.
explained in the Glossary. While the Glossary gives an [S] TATE, M. W. and R. C. CLELLAND, Nonparametric and
example (with page number) of usage in this book for most Shortcut Statistics, Danville, Ill. : The Interstate Publishers
terms defined, no system of reference to the Glossary is & Printers, 1957.
adopted in the text.
Jean D. Gibbons
The scope of any nonparametrics book should rightly be
University of Alabama
considered the prerogative of the author. However, omission
of simple and useful procedures such as the Kolmogorov-
Smirnov tests (goodness-of-fit and equality of two or more
distributions), and chi-square tests (goodness-of-fit tests and Bayesian Inference in Statistical Analysis, by
contingency table analysis), leaves a serious lack in the George E. P. Box and George C. Tiao. Addison-
repertoire of potential users of nonparametric methods. Wesley Publishing Company, 1973, 588 pp., $16.95.
Procedures based on runs, multivariate topics and univariate
sequential methods are likewise omitted. Only certain This book effectively presents a particular view of statistical
procedures are included for those data situations covered. inference which utilizes Bayes’ theorem. Its strength lies
For example, rank correlation and normal scores tests are in the breadth of coverage, and in the many realistic examples
omitted, even though they are easier to apply and understand of the analysis of data.
than some of the techniques selected for inclusion, as for Based upon a notion of ignorance relative to an experiment,
example the bivariate symmetry test. On the other hand, the authors suggest that a prior distribution should be taken
the inclusion of confidence interval procedures and multiple (locally) uniform in a parameter for which the likelihood
comparisons techniques is a most welcome aspect of this function is exactly, if possible, approximately, if not, data
book’s scope. An explanation of P-values or associated prob- translated, in the sense that changes in the data should only
ability as a useful alternative to traditional methods of shift the likelihood funct.ion, not change its shape. This
hypothesis testing where a particular level is selected in eventually leads them to take the prior density proportional
advance is somewhat hidden on page 20, Comment 8. The to the square root of the Fisher information matrix, i.e., to
discussion of choice of levels in discrete null distributions is Jeffreys’ rule. They suggest that in certain situations this
inadequate; for example, nominal level versus exact level is rule be modified, for example, when some parameters are
given little attention. For the two-sample dispersion problem, viewed as independent of others, and in general used with
the Ansari-Bradley test is presented instead of the Siegel- caution. When such a prior distributioh is used it is called
Tukey test. The former procedure requires a special table, noninformative, and described as the distribution of an
the formulas for its mean and variance differ depending on unprejudiced observer, although no attempt is made to
whether the total sample size is odd or even. Further, the justify such language. In the remainder of the book their
only confidence procedure given for dispersion parameters is approach is illustrated in a number of important areas of
based on rank-like tests. Those simpler methods which cor- inference, namely, standard normal theory, linear models,
respond to the Mann-Whitney-Wilcoxon and Sukhatme variance component problems, some general multivariate
test procedures (see Noether [4]) are not memioned. problems, choice of transformation of data, and robustness
In summary, the authors have provided a well-documented, based upon the exponential power distributions. They
well-referenced and concise summary of a number of selected emphasize the effectiveness of Bayesian approaches in assess-
nonparametric procedures. Their proper usage is illustrated ing assumptions by means of broadening an initial model to
by actual experimental situations. The authors’ primary allow additional parameters, and then, numerically, if neces-
sary, integrating out such parameters. There are 10 chapters, inference, for which purpose, when suitably supplemented,
6 tables, and a good reference list. There are no exercises. I find it very well adapted.
The only major weakness in this book stems from the Bruce n/r. Hill
particular, seemingly self contradictory, point of view taken University of Michigan
by the authors in their choice of a prior distribution. Thus
on the one hand they pay lip service to the subjective theory
of de Finetti and L. J. Savage, on the other hand, they try Digital Time Series Analysis, by Robert K. Otnes
to be “objective” in their choice of a prior distribution after and Loren Enochson. John Wiley and Sons, Inc.,
the fashion of Jeffreys. But, despite the fact that a subjectivist 1972, 467 pp., $16.95.
will often find, via a stable estimation argument, that his Otnes and Enochson offer their book first “as a tool and a
posterior distribution is well approximated using Jeffreys’ reference for those actively engaged in digital data analysis”
prior distribution, these approaches are logically incompatible. and second for use in “a graduate level course in time series
Nowhere could this be more clear than in the curious discus- analysis for students in engineering and related fields.” It is
sion on page 45 of the dependence of their prior distribution written from an engineering point of view rather than a
upon the design of the experiment. The authors consider statistical one. Statistical concepts employed (moments of a
Bernoulli data consisting of 21 successes in 24 trials and distribution, stationarity) are defined heuristically rather
observe that their prior distribution for the usual parameter than rigorously. The book is accessible to anyone with a
depends upon whether the sample size n = 24 is fixed in background of calculus and elementary real and complex
advance, yielding a binomial distribution, or whether the analysis.
number of successes r = 21 is fixed in advance, yielding a The authors start by presenting statistical concepts to be
Pascal distribution, for the data. Of couyse the likelihood used in the rest of the book. They discuss data records as
functions are identical, and the system of inference proposed “time histories” which may be either deterministic (a sample
by the authors thus violates the likelihood principle. The of a sine wave, for example) or random (an observation of a
authors implicitly dismiss this principle with a comment stochastic process). Fourier transforms, convolutions, the
to the effect that they do not find it very surprising that autocorrelation function and power spectrum, the sampling
the inference should be different in the two situations. It is a theorem, and resolution in the frequency domain are also
little disappointing that the likelihood principle should meet discussed in Chapter 1. As this list suggests, the title of the
such a fate (not with a bang but a whimper?). It is even more book should perhaps be “Digital Spectral Analysis”. Most.
disappointing that the authors do not acknowledge that they of the book deals with filtering and spectral estimation and
are simultaneously dismissing the snbjectivistic foundations is most applicable to the analysis of vibration, acoustic, or
to which they pay lip service. While it is true that choice of similar data for which a frequency domain description is
experiment sometimes does convey information about a sought.
parameter, for example, choice of the Pascal as opposed to Chapter 2, Preprocessing of Data, and Chapter 10, Prob-
the binomial in the above example suggests that the experi- ability Density Function Computations, are exceptions.
menter views the parameter as small, such information is of Chapter 2, like the first chapter, contains little that will be
a highly subjective nature, and can hardly be routinely new to the experienced time series analyst but is a good
represented in terms of an additional factor p-4 in the prior introduction for t,he novice. Simple explanations of analog-to-
density, as done by the authors. (One is led to wonder whether digital convertors, magnetic tape, and other components of a
they would use a mixture of the two priors if the choice of data acquisition system are given. There are good discrlssions
sampling scheme were made by flip of a coin?) Plainly, the of digitization error and of such practical problems as
notion of ignorance relative to an experiment which the removing bad data points and trends and scaling plots.
authors propose, whatever be its merits, is not compatible Chapter 10 applies to any data, not just time series, and is
with the srtbjectivistic theory. Rather, within the subjecti- largely devoted to recipes for computing histograms and
vistic theory such ignorance may or may not occur, and chi-square test,s for normality.
although life is pleasant when it does, it should not be forced Chapter 3 deals with data filtering, particularly recursive
in when not applicable. The authors are eventually led, digital filtering. There are computational recipes for numerous
incidentally, to t,he llse of “prior” distributions which depend filters (sine and t,angent Butterworth and Chebyshev), ag
upon the data, nor merely upon the design of the experiment, well as a discussion of instability problems due to computer
in connection with their analysis of transformations. Their round-off error.
approach might then best be described as one of modulating Chapter 4 discusses the discrete Fourier transform and
a likelihood function by another frmction, which in certain fast Fourier transform (FFT). Chapters 5 through 8 deal
instances is allowed to depend upon the data, and interpreting with spectral and cross-spectral density computations by
the product as a probability density for the parameter, given three different methods, via the correlation flmctionz the
the data. Such methods may eventually prove fruitful, bllt, direct FT, and using filters. Good discussions of leakage and
since they cannot be justified in terms of maximizing sub- confidence bands for spectral estimates are included. A member
jective expected lltility, or coherent behavior, upon what of efficient met)hods of computing correlation frmctions,
basis do they rest? While explicitly rejecting most non- including the approach via the FFT, are presented.
Bsyesian criteria, the allthors unfortunately fail t,o suggest Chapt,er 9 deals with the estimation of frequency response
alternative ways of judging the performance of t,heir methods, functions of linear systems and of coherence. Computation
apparently (in my opinion mistakenly) relying upon the of confidence limits for coherence estimates is discussed.
us11al sllbjectivistic criteria. Finally, suggestions are offered for applying such mcdti-
variate techniques as principal component analysis to the
I would not like, however, to leave i he reader with a nega-
spectral density matrix when relat iollships among several
tive impression of this book. It is a very valrlable reference
time series are being studied. The final chapters treat non-
source, containing material 1hat is not generally known, stationary time series and present examples of the comprlta-
nor readily available, elsewhere. There seem to be few errors. tions discussed in the book, performed on generated data.
1 am at present using it as a text for a course in Bayesiall This book is more likely to be valuable as a reference than
as a text, though much of the material could be used in the As the practicing statistician looks over these chapter
course envisioned by the authors, and they do include some headings, he might be inclined to conclude that this pnblica-
problems. Its weakness is that it contains more than most tion will be of little use to him. Fortunately, this is not the
people want to know about some aspects of digital time series case. As in Volume 1, a large introduction to the underlying
analysis (FFT algorithms, for example) and neglects others theory and methods of application of the tables is included.
or loses sight of the big picture in all the details. For example, In Volume 2 the introduction covers 141 pages and includes 65
the chapter on filtering includes virtually no discussion examples of the use of the tables. Over 80% of the examples
(except in terms of passing specified frequencies) of why involve real data collected in experimental investigations.
filters should be used on data at all. Many will feel that the introduction alone is wort,h the pur-
A warning to any user: Four pages of errata came with the chase price.
book, and I found a page more. Some errors are minor, but The examples include calculation of confidence limits for
many involve sloppy notation, missing factors of two, etc., several parameters such as the mean of a Poisson distribution,
which would be particularly confusing to the relatively the ratio of two variances (~2/&), the binomial parameter T,
unsophisticated time series analysts who are likely to be the the coefficient of variation, the multiple correlation coefficient,
book’s primary audience. and the median. Other examples illustrate the comparison
Judith Zeh Nelson of means of two exponential distributions, combining evidence
University of Washington from several experiments, testing for randomness and
normality, estimation of the mean and standard deviation
Biometrika Tables for Statisticians, Volume 2, using order statistics, setting of specifications and computation
edited by E. S. Pearson and H. 0. Hartley. of tolerance limits in quality control problems, testing for
Cambridge University Press, 1973, 285 pp., $17.50. homogeneity of variances, comparison of mean vectors and
covariance matrices, power and sample size calculations,
Volume 1 of Biometrika Tablesfor Statisticians, issued in
graduation of empirical distributions, nonparametric analyses,
1954, updated and revised versions of BTS published earlier
by K. Pearson. Volume 1 was planned to include “the most and quanta1 assay and estimation of the LDj, to name a few.
A reading of the examples may suggest applications of statist-
essential tools of the statistician’s trade” and contained
ical methodology which had not occurred to the reader before.
tables published in Biometrika and elsewhere.
Many tables have been published since Volume 1 appeared Another important use of the introduction is that it also
and, while some of these were included in later revisions of serves as a brief review of the literature on the topics covered
Volume 1, there is a need for a book of tables associated with by each of the ten chapters. The amount of the statistical
statistical methodology developed more recently. It is obvious literature is growing exponentially; and while most practicing
that one volume could not summarize all of the new tables statisticians are generalists and need to keep up on all aspects
which have appeared in the recent statistical literature. of statistical methodology, it becomes impossible to do so
The authors point out that “Volume 2 is one of the many and at the same time contribute to the solution of real prob-
possible companions to Volume 1.” lems. If one is confronted with a problem in a certain area,
Some of the tables in Volume 2 supplement tables in Volume and it has been some time since he used this particular statist-
1 by providing additional decimal places, fractional degrees ical methodology, he can refer to this volume and get a quick
of freedom, wider range of arguments, and probability review of the literature.
integrals in place of percentage points. The 69 tables in This reviewer found the sections on testing the form of the
Volume 2 cover 232 pages and are grouped into ten chapters distribution function (i.e., normal, lognormal, exponential),
as summarized below. fitting of univariate frequency distributions, and quanta1
Number Number assay particularly useful. Statisticians using other t,echniques
of of will no doubt find other sections useful to them.
Chapter Title Tables Examples Another strong point of Volume 2 is the chapter on
interpolation in statistical tables. This topic is given little
I The Normal Probability 8 20
attention in the statistical literature and, as the authors
Function and Certain Derived
point out, linear interpolation is inaccurate in many instances.
Tables
Many distribution functions are highly nonlinear even over a
II Tables for Procedures Based 13 12
small range. Interpolation techniques are illustrated in
on the Use of Order Statistics
numerical examples throughout the text.
III Mean Slippage Tests Based 2 3
on Ranks While space does not permit a complete summary of all
IV Tables & Charts for Noncentral 7 the tables, the following should give one a general impression
12
Distribut,ions of the types of tables contained in Volume 2.
V Systems of Univariate Fre- 8 5 Chapter I contains tables associated with the normal
quency Distributions probability function including values of X and Z in terms
VI Tables for Use in Applying 8 4 of P, differential coefficients PZ(X), percentage points of
Techniques of Quanta1 Assay the chi-square and F distributions for integral and fractional
VII Tables for Multivariate 7 3 degrees of freedom, and probability integrals of the extreme
Analysis standardized deviate from the population mean, extreme
VIII Goodness of Fit Tests Based 2 4 standardized deviate from the sample mean, and the mean
on the Empirical Distribution deviation from the sample mean.
Function The tables in Chapter II give expected values, variances,
IX Analysis of Directions on a 10 2 and covariances of normal order statistics, coefficients for
Circle & Sphere estimating the mean and standard deviation from k normal
X Tables to Aid Interpolation 4 0 order statistics, moments and moment ratios of the extreme
__ __ values in normal samples, statistics associated with the
TOTAL 69 65 Shapiro-Wilk W test, for normality, and expected values of
order statistics in samples from negative exponential, gamma, the topics covered include: probability, random variables,
and half-normal distributions. probabilit)y distributions, organization and presentation of
Chapter III gives lower tail critical values for the Wilcoxon dat,a, estimation, hypothesis testing, goodness of fit and
two-sample rank-sum test and the probability integrttl and contingency tables, regression and correlation, analysis of
lower percentage points for the Wilcoxon paired rank test. variance and nonparametric statistics. No material is included
The tables in Chapter IV are associated with the noncentral (and perhaps rightly so) on the “decision analysis” approach
t, F, and chi-square distributions and include values and t,o st,atistics. Each topic has, I think, the right amolmt of
confidence limits for noncentral parameters and coefficients depth of coverage.
to assist the determination of moments. Charts for power This book has a number of strong points which wordd
and sample size calculations are also included. make it suitable for a text at the suggested level. Perhaps
Chapter V contains tables which aid in the fitting of the the most important point is that the book is carefully written
Pearson and Johnson SU and Sg univariate frequency distri- with very few minor slips and half-truths. One often finds
butions. Tables for the maximum likelihood estimator of p in in texts of this nature, numerous slips and other transgressions
the gamma (Type III) distribution and the coefficients in which are presumably permitted so that the student is not
the expansions for the bias and variance of the maximum overburdened with details and the pre-calculus level is re-
likelihood estimator p are also included. tained. Malik and Mullen have done an excellent job of a
The tables in Chapter VI are used in the analysis of quanta1 most difficult task: keeping the book usable at the level
assay data by the techniques of minimum normit chi-square indicated and yet, giving carefld st,atements of definitions,
minimum logit chi-square, and maximum likelihood logistic. theorems and problems. As an example, they are sllccessfrd
Probit analysis was described in Volume 1. in introducing continuous random variables.
Chapter VII contains tables for multivariate anltlysis No doltbt, some of the aforementioned success is due to
techniques such as Wilk’s likelihood criterion, W = /A(/ the numerous examples included: after virtually every defini-
(A + BI. Tables of the percentage points of the largest charac- tion and theorem and in many other situations examples
teristic root of IB - t(A + B)I, the extreme roots of (SZ-l - are given. I liked, particularly, an example (page 171) in
CZI, and the multiple correlation coefficient, n, are included. which twenty (20) different confidence intervals for the normal
Tables used in testing the equality of li covariance matrices mean were given (the student was told /1 = 100) and the
and the hypothesis B = x0 are also given. student could see the end-points varying and collld see which
The tables in Chapter VIII are associated with goodness- intervals included 100 and which did not,. ?rIany problems
of-fit tests and include percentage points for the statistics (with answers in the back of the book) are given. The book
D, V, W2, U2, and A in finite samples of n observations and has a rather complete set of tables which also makes it useful.
the critical values for the Kolmogorov two-sample test). The major short,coming, it seems to me, is that no references
Chapter IX contains tables of percentage points for n/N are given for further reading. It would have been nice to
(on circle), R/N (on sphere), S = Z(cos2 0,)/A’ (on sphere), have some references for the inquisitive student. There are
and Smin and S,,, (on sphere). Charts to determine the also, as in any text, some slips but they are too few to mention.
percentage points of R (on circle) and R (on sphere) and The level of typographical errors also seems low.
critical values for testing the equality of two modal vectors The Authors suggest that the text could also be used for a
(on circle) and two modal vector (on sphere) are included. first collrse in probability. I must disagree with this idea blit
Tables for the estimation of K for dispersion on a circle and any further discussion of the point would be academic since
sphere are also given. it is unlikely that students at this second year level would
The tables in Chapter X aid in int,erpolation and include be studying probability alone wit’hout better preparation
coefficients for Bessel interpolation formula, four-point and in mathematics.
five-point Lsgrangian interpolation coefficients, and Lagrang- In summary, I can recommend this text for a first, course
ian coefficients for use with harmonic argrlments. at the level indicated.
The authors express a concern that some will find these
tables of little use because of ready availability of the com- Ray E. Schafer
puter. Much statistical work is done away from the computer Hughes Aircraft Company
where tables of statistical functions are necessary. It is
doubtful that the computer will make such a pltblication
obsolete.
Problems in Complex Variable Theory, by Jan G.
In conclusion, Volume 2 of Biometrika Tables for Statist-
Krzyz. American Elsevier Publishing Co., Inc.,
icians is an appropriate companion to Volume 1. While Volr~me
1972, 283 pp., $18.50.
2 is not as generally useful as Volrime 1, it, will be helpful to
many statisticians. This book consists of a collection of nearly 1000 different
exercises covering the topics usually presented in a first
Ronald D. Snec
course on the theory of functions of a complex variable. This
E. I. du Pant de lYemows 12 Co., Inc.
field is today recognised as an essential part of the mathe-
matical training of not only mathematicians bllt also of
statisticians, engineers, physicists and other scientists. For
A First Course in Probability and Statistics, by this reason the book will be invaluable both to teachers and
Henrick J. Malik and Kenneth Mullen. Addison to students alike especially since the second part of the book
Wesley, 1973, 361 pp., 59.95. contains solutions to the problems. Actually the scope of
the book is a little wider than the title suggests, as an examina-
The Authors state in the preface that this text is intended tion of the Chapter headings shows:
as an introductory course in probability and statistics for 1. Complex numbers. Linear transformations.
second-year students with only a pre-calculus knowledge of 2. Reg\darit,y conditions. Elementary frmctions.
mathematics and, of course, no previous experience in 3. Complex integration. *
statist its. The material is well organized in “classical” fashion; 4. Sequences and series of analytic f\ulctions.
5. Meromorphic and entire functions. 17. “Stochastic programming models for inventory
6. The maximum principle. control and water storage problems,” by A. Prekopa
7. Analytic continuation. Elliptic functions. 18. “Time series investigations of rainfall as the input of
8. The Dirichlet problem. storage models,” by A. Salamin
9. Two-dimensional vector fields. 19. “Storage availability for a product manufactured
10. Univalent functions. under certain stochastic conditions,” by M. Scott
In a few of the chapters some preliminary material covering 20. “On minimizing water loss by modifying lock dates,”
mainly basic definitions is inserted although this is certainly by J. Stahl
not sufficent for this book to be regarded as a text. The book 21. “Die Proportionierung der Vorrate als strategische
appears to be surprisingly free of typographical errors and Entscheidung der Vorratsoptimierung,” by C. -J.
mistakes. This could be partly due to that fact it is an enlarged Strauss
and revised English edition of a Polish version first published 22. “A first passage problem for integrated renewal
in 1962. In conclusion, it is a useful addition to the library processes,” by J. Tomko
of any teacher of complex variable theory. 23. “Produktion und Vorrate in Betriebsmodellen,”
by L. Uncovsky
Je$rey J. Hunter 24. “Inventory control with unknown demand distri-
University of Auckland bution: A discrete time-discrete level case,” by J.
Wessels
25. “Decentralizability of a stochastic optimization
Inventory Control and Water Storage, edited by A. approach to water resource allocation,” by A.
Prekopa. North-Holland Publishing Company, Whinston
1973, 382 pp., $28.00. 26. “Simple multi-stage inventory models,” by T. M.
Whitin
This publication contains the proceedings of a conference 27. “Matrix-arithmetical relations in the dimensioning
organized by the Bolyai Janos Mathematical Society in of dams and in the study of the operation of large
Gyor, Hungary, September 1971. The purpose of the con- lakes,” by I. Zsuffa
ference was to bring together mathematicians, economists,
and engineers to discuss applications of operations research G. G. K.
to inventory control and water storage. Specific papers
appearing in the publication are:
1. “The hurst phenomenon and the range of sums of
random variables,” by G. N. Alexander
Mathematics and Statistics: Essays in Honour of
2. “Investigation on the water demand distribution
Harald Bergstrom, edited by Peter Jagers and
concerning the irrigation systems in the Tisza
Lennart Rade. Department of Mathematics,
Valley,” by A. Barasz, L. David, and M. Kropf
Chalmers Institute of Technology and the Univer-
3. “A multi-stage stochastic inventory model and its
sity of Gbteborg, 1973, 121 pp., 35 skr.
application,” by A. Chikan and Gy. Meszena
4. “Opimizing the exploitation of a multi-purpose Harald Bergstrom is professor in mathematical statistics
dam,” by B. Djordjevic and S. Opricovic at the Chalmers Institute of Technology and the University
5. “Recent results in first emptiness problems of of Goteborg, Sweden. This publication contains a collection
storage,” by J. Gani and J. Matthews of papers contributed in his honour on the occasion of his
6. “Reduction of the on-hand inventory on given level sixty-fifth birthday. Specific papers appearing in the publica-
of reliability,” By L. Gerencser tion are:
7. “Zur diskreten Roberts-Approximation,” by H. J. 1. “On the equivalence principle in fluctuation theory,”
Girlich by E. Sparre Andersen
8. “Some statistical results for inventory models with 2. “Asymptotic power and expected sample size of
state dependent leadtimes,” by C. M. Harris SPR-tests in one parameter exponential families,”
9. “The stationary solution of multi-product inventory by S. Holm
models,” by D. Hochstadter 3. “A limit theorem for sums of random numbers of
10. “The error term in limit theorems for storage i. i.d. random variables,” by P. Jagers
models,” by D. P. Kennedy 4. “A canonical representation of symmetrically distri-
11. “On the operating characteristic ‘service level’,” buted random measures,” by 0. Kallenberg
by H. Klemm
5. “A characterization of the multi-variate negative
12. “Some results concerning reliability-type inventory binomial distribution,” by E. Lukacs
models,” by Z. Laszlo
6. “A note on logarithmic concave measures,” by A.
13. “Lagerhaltungsmodell fur nichtstationare Nachfr-
Prekopa
age,” by Q. Mann
7. “Note on spherical harmonics,” by M. Rao
14. ‘(On the hydrological dimensioning of Csaszarviz,”
8. “On a random transformation of a point process to a
by G. Papp
Poisson process,” by M. Rudemo
15. “Allowing for the effect of planning in stochastic
9. “Incomplete contingency tables: conditions for the
models: A controlled-variability process,” by I. A.
existence of unique MLE,” by I. R. Savage
Pappas
10. “Summing variances in L,,” by V. V. Yurinskii
16. “Recent research on the ruin problem of collective
risk theory,” by N. U. Prabhu G. G. K.
Issued in three parts per year. Annual subscription 24.50. All enquiries should be addressed
to The Secretary, Royal Statistical Society, 21 Bentinck Street, London WlM 6AR, U.K.
JOURNAL OF THE
AMERICAN STATISTICAL ASSOCIATION
Volume 69 June 1974 Number 346
Stochastic models for single neuron firing A cluster analysismethod for grouping
trains: a survey (Invited Paper) meansin the analysisof variance
S. E. Fienberg A. I. Scott and M. Knott
Biometrics is published quarterly. Its objects are to describe and exemplify the use of mathe-
matical and statistical methods in biological and related sciences in a form assimilable by ex-
perimenters. Inquiries should be addressedto: Biometrics Business Office, Department of Ex-
perimental Statistics, North Carolina State University, Raleigh, North Carolina 27607, U.S.A.
OPERATIONS RESEARCH
Volume 22, Number 4, July-August 1974
CONTENTS
EDWARD P. C. KAo-Modeling the Move- SHANTANU V. TEMBE AND RONALD W.
ment of Coronary Patients within a Hos- WOLFF-The Optimal Order of Service
pital by Semi-Markov Processes in Tandem Queues
PercentagePoints of the Distribution of The Atom Probe and Markov Chain Statis-
the t Statistic when the Parent is tics of Clustering
Student’s t . . . . . . . . Karen K. Yuen and Charles A. Johnson and
V. K. Murthy Jerome H. Klotz
STATISTICAL THEORY
AND METHOD ABSTRACTS
A Journal of the lnfernafional Statistical Institute
The aim of this journal, first published in July 1959, is to give complete coverage
of published papers concerning Statistical Theory and Method. These abstracts
provide valuable information on new developments in many fields; for example,
frequency distributions, estimation and testing problems, experimental design, time-
series analysis, queueing theory, reliability and acceptance inspection. Approximately
1,500 abstracts are published each year, concerning journals from all parts of the
world. Abstracts are also prepared from collections of papers such as reports
of conferences, symposia and seminars; technical reports of experiment and research
stations are covered.
A scheme of classification has been developed for the abstracts that is flexible and
facilitates the transfer of code numbers to punched cards. The pages of the journal
are colour tinted according to the main sections of the classification.
The abstracts are in the English Language up to 400 words-the UNESCO recom-
mendation for the “long” abstract service.
LONGMANGROUPLTD JOURNALSDIVISION
43/45 Annandale Street, Edinburgh EH7 4AT Scotland
1
PREPARATION OF MANUSCRIPTS
Manuscripts should be submitted to the office and n, etc. Subscripts or superscripts should be
of the Editor-Elect, Dr. William H. Lawton, clearly below or above the line. Bars above
MSDD, Bldg. 56, Kodak Park Division, groups of letters (e.g., log Z) and underlined
Rochester, NY 14650. Each manuscript should be letters (e.g., g) are difficult to print and should
typewritten, double spaced, with wide margins be avoided. Symbols are automatically italicized
at sides, top, and bottom. Each manuscript by the printer and should not be underlined on
should include a brief abstract of less than 150 manuscripts. Boldface letters may be indicated
words and key words listed following the abstract by underlining with a wavy line on the manu-
for publication in TECHNOMETRICS. The script; boldface subscripts and superscripts are
original should be submitted with three additional not available. Complicated exponcntials should be
copies on paper that will take corrections. Dittoed represented with the symbol exp particularly
or mimeographed papers are acceptable only if when appearing in the text, that is,
completely legible. Footnotes should be avoided
and replaced by remarks in the text, or placed in cxp [a2 + b2] should be used in place of e(““+b2)
an appendix. References should be numbered and
enclosed in square brackets (e.g. [l] and listed In writing square roots the fractional exponent is
in alphabetical order in the list of references. Only preferable to the radical sign. Fractions in the
references actually cited should be listed. In the body of the text (and when possible in displayed
reference list, each reference should contain, in expressions) and fractions occurring in the
the order indicated, the name and init’ials of the numerators or denominators of fractions are
author followed by those of the co-authors, date preferably written with the solidus; thus
of publication, title of reference, source, volume
number and page. References to books should
(a + b)/(c + d) rather than s.
include publisher’s name and location. Tables,
figures, charts, and diagrams contained in the
manuscript must be submitted in camera-ready Authors will ordinarily receive only galley
condition. proofs. Authors and coauthors will each receive
As far as possible, formulas should be type- four free copies of the issue in which their articles
written and symbols not available on a typewriter appear. Costs for reprints and covers can be
carefully inserted in ink. Authors arc asked to furnished on request.
keep in mind the typographical difficulties of Every article shall include adequate justifica-
complicated mathematical formulas. The dif- tion of the application of the technique preferably
ference between capital and lower-case letters by means of a numerical example of application to
should be clearly shown; care should be taken to a real problem in the physical, chemical, or
avoid confusion between such pairs as zero and engineering sciences. All papers must contain a
the letter 0, the numeral 1 and the letter I, short but clear summary of contents and con-
numeral 1 used as superscript and prime (‘), clusions. Technical derivations should be placed
alpha and a, kappa and Ic, mu and U, nu and v, eta in a appendix.
CONTENTS