|
|
|
|
Abstract: "As an aspiring student in forest ecology, I was never more distressed about my career choice than when a senior graduate student predicted that I would spend more time learning statistics than any other subject. Twenty years later, the contents of my bookshelf and hard drive provide compelling evidence of the truth of this prediction. Community ecology has collected an eclectic mix of multivariate techniques to quantify the relationships among species and their environment. The challenge has always been to sort through the lot to find the most appropriate approach given the question and the data. While I still sometimes pine for the simplicity of a two-way analysis of variance (ANOVA), Bruce McCune and James Grace's book, Analysis of Ecological Communities satisfies my craving for statistical clarity. They provide an exemplary guide to the convoluted terrain of multivariate data analysis". |
|
Austral Ecology, (2003) 28/4 p.469, reproduced with permission Analysis of
Ecological Communities
From John A Ludwig
Tropical Savannas CRC
CSIRO Sustainable Ecosystems
Atherton, Queensland, Australia |
|
As a user of PC-ORD
since version dot, I was delighted to see Bruce McCune provide another excellent product
through MJM Software; this time a reference book on community analysis. Like my own, now
dated, Statistical Ecology: A Primer on Methods and Computing (J. A. Ludwig &
J. F. Reynolds, 1988, Wiley & Sons, New York), Analysis of Ecological Communities
is aimed at the practical user of community analysis techniques. I feel that the authors
have achieved their aim. I found this book easy to read and understand - the style of
presentation is excellent. For each method covered by a chapter, the authors provide a
background to the technique - what it does and when to use it - followed by how it works
and, importantly, what the key results to report are. The authors also provide simple,
easy-to-follow examples that apply the method. They finish each methods chapter with a
description of some of the major variants known for the method; these are well referenced
so that readers can pursue detail, if interested.
However, before
getting into these methods chapters, the authors provide a number of overview chapters
that cover what is typically in multivariate ecological data sets and how these data are
collected by sampling and measuring. Some of the basic properties of these data, such as
species diversity, are also covered in these early chapters. Why these data need to be
screened for outliers and may need to be transformed for statistical and ecological
reasons are also discussed. I particularly enjoyed chapter 5 on species responses to
environmental gradients and the 'dust bunny' distribution - how ecological community data
is like fluff and lint that accumulates in the corner of a room - you will enjoy reading
and thinking about 'dust bunny' data distributions. In these early chapters, the authors
also nicely cover the ways to estimate the similarity or distance (dissimilarity) between
items in ecological data sets - the first step in most multivariate analyses. They do this
without trying to be comprehensive and highly technical, as this has been done in the
second edition of Numerical Ecology (P. Legendre & L. Legendre, 1998, Elsevier,
Amsterdam). Rather, McCune and Grace use simple examples and illustrations to provide
readers with an intuitive feel for the advantages and disadvantages of various distance
measures. I really appreciated their introduction to the problem, and newer solutions, to
the fact that the meaningfulness of distance measures between items declines as data
become increasingly heterogeneous.
Chapters 10-22 cover
classification and ordination topics and methods. I could briefly describe the contents of
each chapter; however, the authors have already done this in their overview chapters on
classification and ordination, chapters 10 and 13, respectively. Rather, I will note a
couple of things that I found most interesting in these chapters from my own personal
experiences with these methods. I liked the authors' arguments for scaling clustering
dendrograms by a function that provides the amount of information lost at each step in a
hierarchical cluster analysis. If this function is rescaled from 0 to 100% information,
one obtains a good feeling for where along the clustering path most information is being
lost. This provides a useful guide as to where to best 'cut the stems' in the dendrogram
to form groups. Like many others using non-metric multidimensional scaling as an
ordination technique, I have found results intuitively pleasing, that is, the positioning
of samples in ordination space provided useful ecological interpretations. However, I also
felt frustrated by the fact that in some cases quite different ordinations were obtained
by different starting configurations. Which one to choose? In version 4 of PC-ORD, McCune
provides an 'autopilot' to assist the user make many runs with different starting
configurations. What I appreciate in chapter 16 is that McCune fully describes his
autopilot procedure and his rationale for selecting the best starting configuration, and
hence, final ordination.
Chapters 23-28 cover
methods for comparing multivariate groups, including the standard parametric multivariate
analysis of variance and discriminant analysis techniques and newer non-parametric Mantel
tests and multiresponse permutation procedures. Of course, application of these methods
requires that groups have been defined, either by classification and ordination
techniques, or by some other ecological criteria such as soil type. In these multivariate
group comparison chapters, I found the authors' discussion of 'when to use it' most
useful.
Dean Urban and James
Grace contribute the final two chapters on classification-regression trees and structural
equation modelling. I have always wanted to learn more about how classification and
regression trees apply to community ecology problems, and Urban provides this in chapter
29. Structural equation modelling is a relatively new box of tools that was developed in
the 1970s in the fields of econometrics, sociology and psychology. Grace and others have
only recently applied these tools to community ecology. In chapter 30, Grace provides the
terminology and background needed to appreciate that structural equation modelling is a
suite of multivariate procedures for addressing questions and testing hypotheses about
multiple causal relationships in ecological data - a most enlightening chapter.
I finish by
recommending this book to all those interested in learning about how to better analyse
ecological community data. It would also make a good textbook. At US$35, this 300-page,
large-format book is good value. |
|
Ecology: Vol. 84, No. 5, pp. 1344-1345. Analyzing Communities
by Stephen B. Cox
Texas Tech University
Department of Environmental Toxicology
Lubbock, Texas |
|
Community ecologists are interested in the way species co-vary in space
and time, or, as McCune and Grace describe, who
lives with whom and why?
As a result, they deal with data that are notoriously difficult to approach from a
statistical point of view. The data are multivariate in nature, i.e., the
inter-relationships between species, between environmental measurements, and between
species and environmental measurements, must be taken into account. To make matters worse,
these inter-relationships frequently are non-linear. In addition, data for each species
consist of presence/absence, counts, or semi-quantitative (e.g., percent cover, cover
class) measurements that do not adhere to the distributional assumptions required for in
many statistical tests. Unfortunately, most biostatistics courses do not adequately train
ecologists to deal with such data.
McCune and Grace provide an excellent resource for ecologists struggling
with the difficult attributes of ecological data. Although there are a number of books
that address the topic of multivariate analysis of ecological communities, this book is
unique in that it successfully combines an extensive breadth of topics with descriptions
that should be generally accessible to any ecologist with some basic training in
statistics. It is also an excellent resource for finding ecological literature that
describes and/or utilizes the techniques being presented. The authors' stated goal is to
provide a book with practical utility for ecologists, and, in my opinion, they definitely
have succeeded.
Overall, the book has the feel of an encyclopedia, and is arranged into
six parts. Part one includes a unique treatment of what the authors call the
complete community data set.
This is a discussion that expands the notion of a site × species presence/absence or
abundance matrix to include information on the environment and individual species' traits,
all within the context of matrix algebra (the basics of which are covered in a short
appendix). It also covers community sampling strategies, a surprisingly broad treatment of
species diversity, environmental gradients, and measures of distance.
Part two contains information on good data practice. Many ecologists
will be tempted to completely over-look this relatively small section of the book (this
topic is mostly ignored in other books on ecological data analysis); however, do not do
so! In terms of avoiding frustration and wasted time, the insight provided in this section
of the book is more important than any information on new and exciting analysis
techniques! I was especially pleased to see the authors note the fact that almost all
analyses have to be repeated (because of some mistake or another). The book provides
practical advice on how to make this repetitive process more efficient.
Parts three through six delve into the statistical techniques. Part
three covers methods for finding groups, part four covers ordination, part five covers
multivariate comparisons of groups, and part six covers structural models. Each technique
is covered in a chapter broken into several sections: background, when to use it, how it
works, what to report (which gives advice on the information that should be reported when
describing results), examples, and variations (which provides information on useful
variants of particular tests). Particular attention is paid to describing the
relationships between tests.
The book also emphasizes nonparametric techniques that, because of the
increasing availability of inexpensive computing power, are becoming practical for all
ecologists. These techniques are based on permutation or optimization routines and are
particularly well suited to the data that community ecologists frequently collect.
Examples include non-metric multidimensional scaling, multi-response permutation
procedures, and classification and regression trees.
Because of the nature of the questions they ask, community ecologists
would be wise to take the time to become proficient analysts as well as insightful
ecologists. This book would be a useful addition to graduate courses in community ecology,
and would also provide an excellent starting place for practicing ecologists when
confronted with a new analytical problem. The prose does tend to be a bit terse at times,
and you may notice some minor lack of attention to detail. However, the emphasis is on
providing practical advice that is useful to any ecologist. Think of it as the
multivariate statistics textbook that is by
ecologists, for ecologists.
© Copyright by Ecological Society of America 2003 |
|
Journal of Experimental Marine Biology and Ecology 289 (2003) 303-305 by Marti J. Anderson
Department of Statistics
Tamaki Campus
University of Auckland
Auckland, New Zealand |
|
This book provides a general overview of some traditional and many more
recently developed multivariate (and some univariate) methods for the analysis of
ecological data. The book is large in size and soft-bound, resembling a manual for a
computer program. Although published by MjM Software Design, the producers of the now
widely-used computer program PC-ORD (McCune and Mefford, 1999), the authors are emphatic
that the book is not intended to be a manual. Instead, they state in their preface that
the book expands the logic behind choosing one technique over the other and
explains the assumptions implicit in that decision.
The contents of the book are organized into six sections or parts: (1)
an overview of community matrices, sampling issues, diversity, species distributions and
distance measures; (2) data screening, adjustments, documentation and transformations; (3)
methods for finding groups, including cluster analysis and TWINSPAN; (4) ordination
methods, including principal component analysis (PCA), non-metric multi-dimensional
scaling (NMS), correspondence analysis (CA), detrended correspondence analysis (DCA) and
canonical correspondence analysis (CCA); (5) methods for comparing groups, including
Mantel tests, discriminant analysis, NPMANOVA, ANOSIM, indicator species analysis (ISA)
and multi-response permutation procedures and, finally, (6) structural equation modeling
(SEM) and classification and regression trees (CART). There is also a succinct appendix on
elementary matrix algebra for reference.
The style of the book is fairly relaxed, generally targeting an audience
of graduate students, scientists and practitioners in ecology. This makes the material
quite accessible, on the one hand, but also means that some ideas are presented as if they
are simple truths when in reality our current understanding is far from
conclusive. Unfortunately, the notation used to describe the various methods is equally
fairly lax symbols used for matrices and variables, (e.g. X, Y, A,
D, p, S, etc.) have different meanings and definitions in different
equations within chapters and between chapters. Also, because the book serves as a general
overview and is not intended to be comprehensive, it suffers a little in places by being
(perhaps by necessity) quite vague in its descriptions of certain methods. One also finds
the odd minor mistake (e.g., in their short description of principal coordinate analysis,
they state: At least one of the resulting eigenvalues is always negative (p.
121), which is not true it depends on the distance measure chosen). It is also
surprising that some published techniques obviously having a similar philosophical
approach to that espoused by the authors (i.e., the use of Bray-Curtis or Sørensen
dissimilarity measure and non-parametric or rank-based methods) are not even mentioned,
such as SIMPER (Clarke, 1993), BIOENV (Clarke and Ainsworth, 1983) and, in the chapter on
diversity, no mention of the notion of taxonomic distinctness (Warwick and Clarke, 1998).
All of this is, however, fairly richly compensated for by the authors
providing many excellent and relevant references as entry points into the literature on
virtually all topics that they do choose to discuss. It is also compensated for by the
book having particularly excellent chapters and added information on some of the more
salient methods available today, such as NMS, CCA, NPMANOVA and ISA. In fact, the
authors chapter on NMS, which is perhaps the most important and useful ordination
method for ecology, is, in my opinion, one of the most complete and accessible treatments
of the topic currently available. The authors also refreshingly provide some more complete
information on certain not-so-recent methods which are, perhaps through accidents of
history, not currently used very widely in ecology, but perhaps should be (e.g., CART,
polar ordination, SEM). These chapters really stretch the imagination and raise all sorts
of interesting possibilities for future research and investigations into the potential
uses of these interesting methods in ecology.
There are many gold nuggets of creativity in the book as well. It was
especially fascinating to read about the authors so-called dust-bunny
distribution, which is also featured on the cover of the book. This suggested shape for
multivariate distributions of counts of species abundances is something that has not
appeared in the literature to date, to my knowledge. Although no statistical basis for it
is given, it does have intuitive appeal and deserves special mention and further study.
In addition, the authors do, in general, provide good sound
methodological advice all the way through, thereby substantially fulfilling their promise
offered in the preface. Individual chapters for each method are very usefully broken down
into: Background, When to use it, How it works, What to report
and Examples. The majority of examples given are from northern hemisphere
terrestrial plant and forest communities, but this does not in any way detract from the
books generality.
In short, this is an exciting new book on a rich variety of topics, old
and new, that provides an important update to current literature on multivariate analysis
for ecology. Whether used independently or as a companion to PC-ORD, students and
practitioners alike can learn a lot from this useful and accessible text.
References
Clarke, K.R. 1993. Non-parametric
multivariate analyses of changes in community structure. Aust. J. Ecol. 18, 117-43.
Clarke, K.R., Ainsworth, M.
1983. A method of linking multivariate community structure to environmental
variables. Mar. Ecol. Prog. Ser. 92, 205-219.
McCune, B., Mefford, M.J. 1999. PC-ORD. Multivariate analysis of
ecological data. Version 4.0. MjM Software, Gleneden Beach, Oregon, USA.
Warwick, R.M., Clarke, K.R. 1998. Taxonomic distinctness and
environmental assessment. J. Appl. Ecol. 35: 532-543. |
|
Folia Geobotanica 40/4 2005, p. 448 by C. Neal Stewart Jr.(ed.) |
|
This book was written by authors as an output of a community analysis
course. According to the authors, it is a textbook and reference book. Multivariate
techniques are nowadays the main methods used in the community studies. The present volume
provides a good overview of the methods and gives following specifics fore each method:
background, when to use this method, how it words, what should be reported and the most
important part, examples. This guide to multivariate tools covers methods of community
ecology from the beginning. Sampling designs, description of basic characteristics,
grouping and ordination methods are followed by chapters introducing methods based on
comparing groups of datasets, classification trees and path analysis. The structure used
in this book makes it very useful and practical for a wide group of users. The book is not
merely a cookbook of methods or list of equations and definitions, which makes it user
friendly. The book can be recommended to many ecologists, teachers and
students bookshelves, and is not going to be covered by dust there. |
|
Analysis of
Ecological Communities Testimonials
Michael Jennings, Ecologist
U.S. Geological Survey and
Bren School of Environmental Science and Management
University of California Santa Barbara
Many thanks for Analysis
of Ecological Communities.
It rocks!
Henning Adsersen, Asociate Professor
Ecological Department of the Botanical Institute
University of Copenhagen.
Please don't be confused that I order a second copy of Analysis of Ecological
Communities. It is a very good book and I can't have my own specimen in peace because my
students want to consult it all the time. |
|
|